Академический Документы
Профессиональный Документы
Культура Документы
NEUROETHOLOGY
BioNB4240
BY CARL D. HOPKINS
Fall 2011
Cover. The motion-sensing neuron, H1, in the lobula plate of the fly (above left), responds to movement of visual stimuli
from rear to front in the horizontal plane. The Hassenstein/Reichardt model of a generalized motion detector (upper right)
in the eye is based on correlation of two visual inputs separated in space. The figures below show the response of the H1
cell (lower left) and the model (lower right) to optomotor inputs at different rotational velocities. LP and HP refer to low pass
and high pass filters; M is the multiplier step. Adapted from:: Borst, A., V.L. Flanagin, and H. Sompolinsky, Adaptation
without parameter change: Dynamic gain control in motion detection. Proc Natl Acad Sci U S A, 2005. 102(17): p. 6172-6.
The motion sensing model for the visual system was an early example of computation in a neural circuit.
Catania, K. C. (1999) A nose that looks like a hand and acts like an eye: the
unusual mechanosensory system of the star-nosed mole. J Comp Physiol A 185:
367-372 ........................................................................................................................ 27-32
Cator, L. J., Arthur, B. J., Harrington, L. C., and Hoy, R. R. (2009) Harmonic
Convergence in the Love Songs of the Dengue Vector Mosquito. Science 323:
1077-1079 .................................................................................................................. 99-102
Hiryu, S., et al (2008) On-board telemetry of emitted sounds from free-flying bats:
compensation for velocity and distance stabilizes echo frequency and amplitude.
J Comp Physiol A 194: 841-851............................................................................... 115-126
Hiryu, S., Hagino, T., Riquimaroux, H., and Watanabe, Y. (2007) Echo-intensity
compensation in echolocating bats (Pipistrellus abramus) during flight measured
by a telemetry microphone. J. Acoust. Soc. Am. 121: 1749-1757 .......................... 127-136
Konishi, M. (1993) Listening with Two Ears. Scientific American pp. 66-73 ............ 145-152
Harmening, W. M., Orlowski, J., Ben-Shahar, O., and Wagner, H. (2011) Overt
attention toward orented objects in free-viewing barn owls. PNAS 108: 8461-
8466 ......................................................................................................................... 153-158
3
Table of Contents
Suthers, R. A. (2004) How birds sing and why it matters. In Natures Music: the
science of birdsong. Marler, P. and Slabbekoorn, H. eds. Elsevier. New York.
Chapter 9 pp. 272-295 ............................................................................................ 247-270
Nottebohm, F. (2005) The Neural Basis of Birdsong. PloS Biology 3: 759-761 ...... 285-288
4
Table of Contents
OMalley, D. M., Kao, Y., and Fetcho, J. R. (1996) Imaging the Functional
Organization of Zebrafish Hindbrain Segments during Escape Behaviors. Neuron
17: 1145-1155 ......................................................................................................... 325-336
Kinkhabwala, A., et al (2011) A structural and functional ground plan for neurons
in the hindbrain of zebrafish. PNAS 108: 1164-1169 .............................................. 349-354
Crapse, T. B., and Sommer, M. A. (2008) Corollary discharge across the animal
kingdom. Nature Reviews/Neuroscience 9: 587-600 .............................................. 369-382
Requarth, T., and Sawtell, N. B. (2011) Neural mechanisms for filtering self-
generated sensory signals in cerebellum-like circuits. Current Opinion in
Neurobiology 21: 602-608 ....................................................................................... 439-446
5
Table of Contents
Bell, C. C., Han, V., and Sawtell, N. B. (2008) Cerebellum-Like Structures and
Their Implications for Cerebellar Function. Annu. Rev. Neurosci. 31: 1-24 ............ 455-480
Borst, A., and Euler, T. (2011) Seeing Things in Motion: Models, Circuits, and
Mechanisms. Neuron 71: 974-994 .......................................................................... 515-536
Passaglia, C., et al (1997) Deciphering a neural code for vision. Proc. Natl. Acad.
Sci. 94: 12649-12654 .............................................................................................. 587-592
6
Greenspan, R. J. (2007) An Introduction to Nervous Systems.
Cold Spring Harbor Press, Cold Spring Harbor, NY. Chapter 7
T E R
W hen a fly makes its way to a fallen, rotting peach, or to the rim of a vat of
crushed grapes, what is it looking for? Perhaps food, but adult flies are not big
eaters. They have already spent their carefree larval youth eating nonstop, day and
night. As adults, they may occasionally snack on yeast, but they have more important
business to attend to-finding a mate. So when a male fly heads for a rotten peach,
he is looking for females. When a female heads for that same peach, she is either
looking for males or a suitable place to lay her eggs once having mated. The next time
you see a bruised piece of fruit lying on the ground, be aware that you are looking
at an active singles scene.
In the world of fruit flies, females generally have the upper hand when it comes to
mate choice. Males must prove themselves worthy before a female will consent. Flies
are certainly not monogamous in the sense of forming some kind of lasting affiliation
between mates, but a female may mate only once or twice in her life, because she will
fertilize many eggs with sperm from a single mating. Therefore, the choice matters.
7
108 CHAPTER 7
L O V E O N THE FLY 109
Female fruit flies produce a mixture of secreted chemicals on the surface of their
external cuticle that acts as a perfume for males (Fig. 7.1). These pheromones are
attractive to males and all% them to determine if this other fruit fly is of the same
species and the opposite sex. Detection of pheromones takes place on chemoreceptor
sensory cells found on the male's forelegs, antennae, and mouthparts (Figs. 7.2 and
7.3). These sensory cells have chemoreceptor proteins and associated ion channels that
respond to pheromones. Mutant males that lack these proteins have a hard time court-
ing females in the dark, where chemosensory cues are essential. (In broad daylight, the
visual perception of a female arouses a male's interest enough to begin courting.)
The fly's chemoreceptors are of the same family as the opsin proteins in pho-
toreceptor cells (see Chapter 2). They are proteins with seven transmembrane seg-
ments, but instead of absorbing quanta of light, they bind pheromones. The binding
triggers a conformational change in the protein so that it interacts with a G protein
and initiates a cascade of enzymatic reactions. Ultimately, the cascade leads to the
opening of ion channels in the membrane of the chemoreceptor cells. Figure 7.2. Chemosensory organs on the fly's
If the male is in the right mood (i.e., if he is old enough but not too old, the head: antennae, maxillary palps, and labial palps.
humidity is right, and he has not been cooped up with other males-where he would
have been exposed to a high dose o f a a l e pheromones, some of which inhibit other
7,l 1-HD
Figure 7.1. Differences in pheromones between male (top) and female (bottom) fruit flies. (Left)
Chemical profiles from extracts of male and female cuticular hydrocarbons detected by gas chro- Figure 7.3. Chemoreceptor (arrows pointing to blue
matography. (Right) Chemical structures of principal peaks from profiles. Arrows indicate key dou- areas) differences between male (left) and female
ble bonds. (right) forelegs.
8
LOVE O N THE F L Y 111
ly and usually follows by vibrating that wing to produce a courtship song). Only one
side of the bilaterally symmetrical brain needs to develop as male, and only the dorsal
part at that. Sexually mixed flies do not perform fully normal courtship, but they will
display these initial steps. (Some investigators have interpreted this finding as confir-
mation of their suspicion that males use very few neurons when pursuing females.)
Later steps in courtship (Fig. 7.5) require male development in other parts of the
nervous system: Production of the song requires male neurons in the thoracic gan-
Figurt ,.-I.
Chemosensory circuitry, necessary for male courtship, that differs between males (left)
and females (right). A representative sample of neurons (light green) is shown from antennae and
palps projecting into the antennal lobes (AL), whose neurons then project to the lateral protocere-
brum (LPR) and to the mushroom bodies (MB).
males from courting them), he will sHte up to a female and tap her abdomen with his
foreleg. The stimulation of his various chemoreceptors by the female's pheromones
will then (usually) trigger his performance of the fruit fly courtship ritual. The signals
from the male's chemoreceptors are relayed through the antennal lobes and into the
major lobes of the brain, the protocerebrum (Fig. 7.4). When neurons in the lateral
protocerebrum are sufficiently excited, the male begins to court vigorously. In males
in which these cells have been genetically engineered to be unresponsive, courtship
will not begin, and conversely, if these cells are genetically engineered to be more Dorsal
hyperexcitable, the male will start courting at the drop of a hat. ~rotocerebrum
9
112 CHAPTER 7
glion, licking of the female's genitals requires male brain tissue in the midbrain,
attempted copulation requires male neurons in the antenna1 lobes and the abdomi-
nal ganglion, and copulatbn requires additional male neurons in the abdominal gan-
glion. Much more of the circuitry for courtship is not sex-specific, however, but,
instead, simply makes use of circuits common to many behaviors. These common
circuits are present in both sexes and have been incorporated into the courtship rou-
tine in males. Courtship is likely to involve a substantial portion of the fly's nervous
system.
A gene called fruitless is essential for much of this male-specific development. Figure 7.7. The bl muscle used in the fly's courtship song.
The gene encodes multiple forms of a transcription factor, one of which is male spe-
cific. In males that are missing the male-specific product, courtship does not occur.
Conversely, when this male-specific product of the fruitless gene is expressed in a The female "hears" the song through her antennae. The tip of the antenna con-
developing female brain, the adult female will perform male-like courtship behavior tains a feather-like structure called the arista that is induced to vibrate by the sound
that is never otherwise seen in females. The male-specific product of the fruitless waves from the male courtship song. These vibrations stimulate stretch receptors in
gene is expressed in most of the brain regions shown to undergo sex-specific devel- the hinge at the base of the antenna, where it attaches to the head, and these stretch
opment for courtship (see Fig. 7.4). receptors convey impulses into the region of her brain required for mechanosensa-
tion (Fig. 7.8). The stretch receptor cells contain mechanosensitive channels in their
membranes, similar to those in jellyfish (Chapter 3) and microorganisms (Chapter 1).
The Love Sow of D. melunogaster
One of the major functions of courtship is for the male to advertise himself to a Flies Have Rhythm
prospective mate. The male does this by extending one wing and vibrating it to pro-
duce a courtship song. This song does not sound very musical to our ears (it resem- One of the ways that females recognize males of the various species is through the
bles a monotonous Morse code of rapid bumps), but it is music to the female (Fig. males' courtship song, and a key characteristic of the song that varies between
7.6). When recordings of simulated courtship song are played to potentially recep- species is its rhythmic timing. The interval between pulses (known as the interpulse
tive females, before placing them in the company of a male, the preexposure makes
them more receptive, that is, it takes them less time to begin copulating.
When flies that are part male and part female (see above) contain male neurons
in the dorsal brain, they will perform the initial stages of courtship and extend one
wing, but they will not produce a proper courtship song unless they also have the
necessary male neurons in their thoracic ganglion (see Fig. 7.5). The inference is that
they will extend their wing if a signal is sent from the brain down into the thorax (sim-
ilar to the signal sent to initiatk flight; see Chapter 6), but that proper control of the
wing's vibration requires fine-tuning from local neurons in the thoracic ganglion, as
is also seen in flight control. In fact, one of the flight-steering muscles ( b l ) beats in
time with the courtship song, firing once with every pulse (Fig. 7.7), and may also be
the trigger for each pulse.
'1
Johnston's
organ
Figure 7.6. Pulses of courtship song produced by male wing vibrations. Each pulse consists of two Figure 7.8. Stretch receptors (Johnston's organ) inside the fly's antenna. The arista is on the right
to three cycles separated by an interval of 32-38 msec. The fly's head would be on the left.
10
114 CHAPTER 7 LOVE O N THE FLY 115
IPI (msec) ,
per S
7 = 54
7 = 43
sec
sec
P
Idelay = 39 sec
heat pulse
0 5 10 15
Time (min)
I Figure 7.10. Blocking action potentials interrupts the song-rhythm pacemaker. Transient suppression
2 4 6 of action potentials was achieved in a male fly whose neurons contain a temperature-sensitive
Time (min)
mutation in the voltage-sensitivesodium channel. (Solid blue line) Actual singing; (dashed line [long
Figure 7.9. Song rhythms parallel circadian rhythms in mutants. Rhythmic oscillation of interpulse dashes]) normal sinusoidal oscillation in the interpulse interval (IPI) up until the moment (blue dash
interval (!PI) in the fly's courtship song in normal males (per') and period mutant males: short day and inset) when a 30-sec pulse of high temperature is delivered. During the 30-sec pulse, singing
0 0
(pe?), long day @eS), and arrhythmic (per ). a e a s u r e s the period of the song rhythm (per has no stops and no action potentials can occur. When the male resumes singing, several minutes later, the
discernible rhythm). (Dashed lines) Sinusoidal waves that best fit the data. period of the oscillation is normal (55sec), but the phase of oscillation of his song is delayed by 39
sec (shaded box in inset). (Dashed line [short dashes]) Phase-shifted sinusoidal oscillation due to
blockage of action potentials. When males not harboring temperature-sensitive sodium channels
are similarly treated, the temperature pulse does not delay the phase of their song. When they
resume singing, they are "in time," as if they have been keeping the beat all along.
interval, or IPI) averages 34 msec in Drosophila melanogaster. But closer inspection
reveals that this interval actually oscillates in a very regular way. Over the course of
approximately 55 seconds, the IPI will gradually increase to 38 msec and then grad-
ually decrease to 32 msec, following a regular, sinusoidal wave (Fig. 7.9). Sequencing in Behavior
The physiology underlying this rhythm is not well understood, but i t shows a
remarkable dependence on the same gene (period) that regulates circadian rhythms. Orienting, tapping, wing extension, and singing the courtship song comprise.the first
Males mutant for alleles of the period gene that produce short-day (19-hr), long-day steps in a larger sequence of courtship actions. If the female does not run away or
(28-hr), and arrhythmic circadian periods sing courtship songs with correspondingly kick the male in his face, he will continue to engage in these early steps. Eventually,
abnormal rhythms. The short-day flies sing with a 46-second rhythm, the long-day if she permits it, he will go on to the next steps: licking the female's genitals, mount-
flies sing with a 96-second rhythm, and the arrhythmic flies have no discernible ing her, and copulating. These actions constitute a stereotypical behavioral sequence
rhythm (Fig. 7.9). that is usually performed in order. It is exceedingly rare for a male to skip the early
Male flies generally sing their courtship song in bouts of 30 seconds to 1 minute at steps and immediately attempt to mount the female. However, when flies are genet-
a time. -When a male pauses in his singing, however, he continues to maintain the same ically engineered to be mutant for fruitless in a central part of the brain called the
rhythm ko that when he resumes, the oscillation is right on the beat as if he had been median bundle, they violate the sequencing rule and skip directly to attempting cop-
humming it all along. This humming-like behavior requires ongoing firing activity of ulation (Fig. 7.11). The experiment suggests that neurons in the median bundle of
neurons, as shown by its ability to be interrupted and delayed when action potentials males normally inhibit the performance of the late steps in courtship.
are suppressed (Fig. 7.1 0). The suppression of action potentials is achieved by means
of a temperature-sensitive mutation in the voltage-gated sodium channel of neurons, Motor Routines and Beyond
so that at normal temperature the neurons fire and at higher temperature they become
silent. (In the experiment, the fly's entire nervous system was mutant, so it is not known Each step in courtship consists of a routine analogous to those described earlier in
which neurons were specifically responsible for maintaining the rhythm.) simpler organisms, such as swimming in the jellyfish (Chapter 3) or flying in the fly
11
116 CHAPTER 7 L O V E O N T H E FLY 117
next. Examples of this kind of analysis were presented in the context of geographical
differences in temperature compensation or light sensitivity due to the variations in
the period gene (Chapter 5).
Pheromone Evolution
The fly's chemical signature varies considerably from place to place. Within D.
melanogaster, the ratios of the different hydrocarbons on the cuticle surface (Fig. 7.1)
differ worldwide (Fig. 7.12), and also as locally as between adjacent vineyards near
Wiirzberg, Germany. Differences between species. can be even greater, but they do
not always reflect phylogenetic distance. In other words, D. melanogaster
lobe lobe
pheromones show greater divergence from those of the closely related species D.
Figure 7.1 1. The median bundle, the region of the fly brain that causes male flies to skip steps in simulans and Drosophila erecta than from those of the more distant species
the courtship sequence when mutant for fruitless.
Drosophila affinis, Drosophila elegans, and Drosophila serrata. Such deviations from
expectation are usually indicative of strong selective pressure, suggesting that
pheromonal signatures play a major role in speciation.
Fruit flies can be confused when mischievous humans paint them with foreign
(Chapter 6), but the physiological basis for these steps is far less understood. It is the
pheromones. Such treatment is sufficient to induce a Drosophila mauritiana male to
sequencing of routines such as these that makes behavioral repertoires elaborate and
take the unprecedented step of courting a D. melangoaster female, which he would
sophisticated. The stereotypical nature of the sequence is a product of natural selec-
never do otherwise. He will even court a dummy fly, if the dummy is painted with
tion and has been co-selected in the species along with the female's response to it.
the appropriate pheromones.
Understanding the neurobiological basis for the sequencing of behaviors represents
Within the D. melanogaster species, some of the geographically diverse strains
a major unanswered question, and its significance may reach far beyond the stereo-
will not court one another. In one case, their reproductive isolation correlates with
typical courtship of fruit flies. There is serious speculation that the origin of thinking
lies in the sequencing of motor routines. In this perspective, the evolutionary leap to
thought came with the ability to string together sequences in the brain such as those
controlling motor routines, but without the actual performance. Thus, a behavior as
apparently stereotypical as the fruit fly's courtship, with its orderly progression
through different actions and its capacity to maintain rhythmic time in the absence
of motor output, may have much to tell us about how brains accomplish more inno-
vative and creative tasks, such as thinking things through or telling a story.
Courtship Evolving
In the fly world, choosing the right partner is not a matter of whether the relationship
will last; one-night stands are the norm. Instead, mate choice-particularly, picking
the right species-means the difference between having offspring or not. As men-
tioned above, olfactory cues are one means of species discrimination and song is
another. Variations in these signals and in the response to them may provide raw
material for species divergence. In trying to infer evolutionary mechanisms, one
source of clues is to examine the divergence in gene function from one species to the
12
118 CHAPTER 7 L O V E O N THE FLY 119
the only D. simulans feature of their song. The average IPI remains unchanged, show-
variation in a gene for one of the enzymes involved in pheromone synthesis, desat2,
ing that it is not affected by the period gene. Because both species have the same 24-
that sets the position of the double bond (see Fig. 7.1). A variant form found in Africa
hour circadian rhythm, this too remains unchanged, indicating that selection on the
and the Caribbean produzs much less of the hydrocarbon 7,Il-HD that is other-
period gene has been confined to its song rhythm properties, allowing it to retain its
wise predominant in D. melanogaster females (Fig. 7.1). The allelic difference
24-hour timekeeping properties. It may be significant that the segment of the D. sim-
between this form of the desat2 gene and the more distributed form is a deletion of
ulans period gene that confers this song characteristic is approximately the same as
16 base pairs from the promoter region that prevents transcription of the mRNA for
the segment affecting temperature compensation among geographically dispersed
this enzyme.
strains of D. melanogaster (see Chapter 5). It is not unusual to find that some portions
To test directly for the role of this gene in reproductive isolation, flies were genet-
of a gene are freer to vary than others, presumably because the functions they affect
ically engineered so that the only difference between them was in the desat2 gene.
are also freer to vary.
These two laboratory strains showed the expected difference in hydrocarbon ratios
Song is not the only aspect of courtship affected by rhythms. Different species
and a mating preference for their own kind. That is, the pheromonal difference was
have distinct preferences for the time of day when they are most likely to mate. D.
capable of producing a certain degree of reproductive isolation, although not nearly
melanogaster tends to prefer late in the day and even seems to mate quite well in dark-
as complete as the original geographical strains' isolation. The incompleteness of
ness, whereas D. pseudoobscura has two different peak mating times, once around
effect is actually consistent with genetic mapping studies between the two geo-
dusk, a few hours later than D. melanogaster, and again later in the night (the rogues!).
graphical strains, showing that multiple genes on different chromosomes are collec-
Using a similar approach to the one described above, D. melanogaster males were
tively responsible for the full-blown isolation between the strains.
engineered to contain a period gene from D. pseudoobscura. As you might expect by
now, the new strain suddenly developed a predilection for mating rather later than
Sorrg Evolution their counterparts who carried an engineered D. melanogaster gene (Fig. 7.1 3).
Is a preference for time of day in mating important for segregating different
As a song characteristic, the rhythmic oscillation in IPI accounts for a substantial
species? The D. pseudoobscura-engineeredD. melanogaster flies offer a partial test
amount of the species specificity of the song. D. simulans, as its name implies, is very
case. They differ only in their period genes. If males and females of both types (nor-
closely related to D. melanogaster, so much so that they can form viable and some-
mal and D. pseudoobscura-engineered D. melanogaster) are mixed together, will the
times fertile offspring. This makes the need for correct recognition all the more press-
engineered flies prefer each other, and likewise, will the nonengineered flies prefer
ing (from an evolutionary standpoint). The average IPI of D. simulans is approxi-
each other? So it seems.
mately 50 msec, in contrast to 35 msec in D melanogaster, and the rhythmic
Because we already know that the period gene also affects courtship song, one
oscillation in D. simulans is 35 seconds, in contrast to 55 seconds in D. melanogaster.
might imagine that the observed mating preference was not related exclusively to
The importance of the two different parameters can be tested by playing synthetic
time of day. By removing the wings from both sets of males, however, any contribut-
courtship songs to virgin females of either species. These females are then placed
ing effect of courtship song as influenced by the period gene was ruled out. (Wing-
with males of the same species and timed for how much more quickly they mate
less males can still court successfully, although it takes them longer.)
than if no song had been played. When synthetic songs are played to D. simulans
What would happen if these two strains were left alone in the laboratory for hun-
females, their own species' rhythm is as effective in stimulating them to mate faster
dreds of generations? Their temporal separation could lead to the buildup of other
with D. simulans males as their species' average IPI. Likewise, D. melanogaster
differences between them by mutation, and an incipient speciation process could be
females care just as much about the rhythm of their own species' song as its average
initiated. Thus, even though they share the same culture bottles, sympatric speciation
IPI, so both have to be right to enhance their mating. Needless to say, a song that has
could occur.
the correct range of IPls, but is arrhythmic, is not very effective with these very dis-
Song variation among species goes beyond merely rhythm and IPI. Some species
criminating females.
differ in the notes (i.e., the pulses) themselves. D. melanogaster pulses have only two
Having already seen that mutations of the period gene in D. melanogaster alter
or three peaks that are of relatively small amplitude (Fig. 7.14). D. virilis song pulses
song rhythm as well as circadian rhythm, it stands to reason that variation in the peri-
have more peaks and they are of greater amplitude. A mutant in D. melanogaster
o d gene between D. simulans and D, melanogaster might affect the species-specific
produces an abnormal song whose pulses resemble those of D. virilis. The affected
song rhythm. This turns out to be true. When D. melanogaster males are genetically
gene is known as dissonance, and it encodes a protein involved in RNA processing.
engineered to have the D. simulans version of the period gene instead of their usual
(The dissonance gene is by no means dedicated exclusively to courtship song. The
version, they sing a song that has the rhythmic oscillation of D. simulans. But this is
13
120 CHAPTER 7 L O V E ON T H E F L Y 121
D. melanogaster
dissonance mutant
0.1J , , ,o.lJ , I , 4 o I
Figure 7.14. Courtship song pulses of normal Drosophila melanogaster (top), mutants of D.
0 3 6 9 12 15 18 21 0 3 6 9 12 15 18 21
melanogaster in the dissonance gene (middle), and normal Drosophila virilis (bottom). The disso-
I
Circadian time Circadian time nance mutants produce polycyclic pulses that resemble pulses in D. virilis.
Figure 7.13. Time of day for mating preference in Drosophila melanogaster and Drosophila
pseudoobscura (left) and genetically e n g i i r e d D. melanogaster (right). Flies were placed in con-
stant darkness so that their circadian rhythms would be governed by their internal molecular clock.
Graphs show proportion of pairs mating at various times of day. (Right) Strains are genetically engi- from the egg onward are capable of performing the full repertoire of courtship
neered D. rnelanogaster with either their same species' normal period gene (upper right) or the D. toward a female on their first try. Their behavior may not be quantitatively normal in
pseudoobscura period gene (lower right). Genetically engineered flies carrying the D. pseudoobscu- all respects, but it is qualitatively normal. That does not mean that courtship behav-
ra period gene show a peak of mating during the night that is absent in normal D. melanogaster and ior is entirely impervious to experience (see Chapter 8), but it suggests that, like fly-
oresent in normal D. pseudoobscura.
ing, there has been selection pressure for fruit flies to "know" how to court as part
of their normal developmental maturation.
If experience is not essential, then what role, if any, do neuromodulators play in
protein is expressed all over the nervous system, and other alleles of dissonance affect a behavior like courtship? Does a male register the presence of a female with the
visual sensitivity as well as song. The mechanisms by which it affects either song or same internal alarm system (dopamine) that he uses when he begins tracking toward
vision are not understood.) a goal during flight (see Chapter 6)?Apparently it is not the same, because male flies
Genetic engineering was brought to bear once again and the normal dissonance that are depleted of dopamine (after being consistently fed with an inhibitor of its
gene from D. virilis was substituted into a strain of D. melanogaster in place of its own synthesis) will still court females. The salience of a female to a male is thus not depen-
gene. The song in these flies clearly resembles that of the dissonance mutant but not dent on the same neuronal circuitry, a fact that may reflect the species' selection his-
that of the normal D. melanogaster (Fig. 7.14). Reminiscent of the period story, the tory. The same selection pressures that have produced males capable of courting at
result suggests that selection for subtle variation in this gene may account at least par- first blush have also produced a perceptual response in them that does not need any
tially for the species difference in courtship song. The variation must necessarily be reinforcement from the neuromodulatory system.
subtle so that other functions of the gene, such as those required for visual sensitivi- The interplay between "hard-wired" and "plastic" behavior is a recurrent theme
ty, are left intact. in evolution. Most behaviors are a mixture of the two, whereas some are more heav-
ily weighted one way or the other. What is flexible in one species may be stereotyp-
ical in another, as seen in the previous discussion of sensitization in Aplysia and its
Neuromodulators and "Instincts"
relatives (see Chapter 4). In all cases, the behavioral differences reflect variations in
Behaviors such as fly courtship are sometimes referred to as "hard wired," reflecting the underlying neurons-in their anatomy, physiology, or both. Variation in genes
the idea that they are innate or instinctual. In fact, male flies raised in total isolation such as period or dissonance ultimately produce their behavioral effects by acting on
14
122 CHAPTER 7
one or more of these properties. At present, we still have only a faint picture of very
few steps in the mechanisms through which such genes affect the nervous system and
--.
alter behavior. .-
Further Reading
Greenspan R.J. and Ferveur 1.-F. 2000. Courtship in Drosophila. Annu. Rev. Cenet. 34: 205-232.
Manoli D.S., Meissner C.W., and Baker B.S. 2006. Blueprints for behavior: Genetic specification of
neural circuitry for innate behaviors. Trends Neurosci. 29: 444-451.
Peixoto A.A. 2002. Evolutionary behavioral genetics in Drosophila. Adv. Cenet. 47: 117-1 50.
15
16
Review articles
17
Review articles
18
Review articles
19
Review articles
20
Review articles
a morphologically distinct structure, the ``barrel,'' in cortex processing in the somatosensory systems is that a genetic
for each whisker on the side of the face (Fig. 2), as well as change or environmental event during development altered
equivalent ``barreloids'' for each whisker in the thalamus and the number of whiskers or rays on the face. Information
brainstem. When they examined those individuals with one about this change, most likely in the form of a neural activity
or two extra or less whiskers, they found the same changes pattern, traveled from the face to the brainstem, then the
in the number of barrels in cortex and barreloids at each thalamus, and then to one or two levels of cortex, altering the
subcortical level.(51) Similarly, when star-nosed moles have development at each level to produce a matching pattern.
an extra appendage, or one less, on their tactile nose, a series Similar alterations occur in the visual system. Siamese cats
of separate bands in cortex always matches the number of result from a mutation that changes coat color and eye color by
appendages (Fig. 5) .(52) Van der Loos and Dorfle(50) con- reducing pigmentation in parts of the body with normal body
cluded from their observations on mice that the cortical temperatures. The reduced pigment in the retina has the un-
array of barrels is ``slaved'' to the peripheral array of whiskers explained developmental consequence of misdirecting most
``as a result of a cascaded induction over three synaptic of the axons from the temporal half of the retina that normally
stations'' (brain stem, thalamus, and primary somatosen- project to the ipsilateral lateral geniculate nucleus to the con-
sory cortex). In star-nosed moles, an altered pattern of bands tralateral lateral geniculate nucleus (Fig. 6).(53) This causes
is also obvious in the second somatosensory area, S2, so the end of the A1 layer, now abnormally innervated from the
the cascade of induction occurs over at least four levels of contralateral eye, to fuse with the end of the A layer with normal
processing. inputs from the contralateral eye. This results in a continuous
To us, the most parsimonious interpretation of the matching representation that has been enlarged by including an extra
changes in the number of whiskers or rays on the nose and the 208 of contralateral retina.(54) The enlarged representation
number of related neural structure at three or four levels of is preserved in a single enlarged pattern in the projection to
21
Review articles
cortex in some Siamese cats,(55) but in others the A1 layer marked by high concentrations of the metabolic enzyme,
inputs to cortex are suppressed.(56) cytochrome oxidase (the CO-blobs). However, ocular dom-
Related alterations in the extent of geniculate and corti- inance (OD) bands and CO-blobs are not found in most
cal maps of the retina have been reported for two different mammals.(32,33,59,60) Such distributions lead to the conclusion
mutant phenotypes of Belgian sheepdogs. In one type, most that the common ancestors of cats and primates had neither
of the retinal ganglion cells of both eyes project to the lateral CO-blobs or OD bands in cortex.(16,25) Thus, these highly
geniculate nucleus of one hemisphere and not the other.(57) similar patterns of modular organization evolved independen-
In the other phenotype, retinal axons of each eye project only tly. In addition, alternating OD bands have even been created
to the ipsilateral hemisphere.(58) Nevertheless, topological in the optic tectum of frogs, normally with input from only
maps of the altered inputs were found in visual cortex. Thus, the contralateral eye, by artificially providing an extra eye to
when the pattern of inputs from the retina is abnormal, the equally innervate the structure.(5) Other examples include the
organizations of the relay nuclei in the thalamus and their independent evolution of the SA and RA modules in S1 of cats
sensory areas in the cortex adjust to accommodate the and monkeys. The independent evolution of similar types of
abnormal pattern. Such adjustments strongly suggest a role modules in sensory nuclei and areas likely reflects similar
for activity patterns that are altered with the misdirected expressions of basic self-organizing factors that have the
inputs. potential to be revealed in all sensory representations. Com-
puter models indicate that alternating bands can result from
Some types of modular organization have activity-based competition that is balanced, while an im-
evolved more than once balance creates a dot (blob) and surround pattern.(61) The
Primary visual cortex in cats and monkeys is characterized variability that exists across species in the expression of
by a pattern of alternating ocular dominance bands (OD) activity-based modules may relate to factors that alter the
or columns, and a distribution of dot-like modules that are period of susceptibility.(62)
22
Review articles
23
Review articles
neighboring auditory and somatosensory areas into the depriv- We have argued here and elsewhere(49) that neural activity
ed visual region.(72) patterns likely create the detailed order within representations,
Quite dramatic alterations in activity patterns have been the general proportions devoted to sensory surfaces, the varia-
produced by allowing retinal axons to grow into the auditory tions in representational order, the septa between adjoining
thalamus after removing auditory inputs in developing mam- representations of separate regions in the receptor sheet,
mals so that auditory cortex becomes activated by visual various types of modules within areas, and possibly congruent
stimuli.(73) This manipulation changes the course of develop- borders between representations. The commonly expressed
ment of auditory cortex to create neurons with orientated visual view is that ``neurons that fire together wire together''.(76,77)
receptive fields that resemble those normally found in visual According to this model of development, neurons that are acti-
cortex. Thus, auditory cortex has the potential to create some vated at the same time maintain and enlarge shared connec-
of the basic features of a visual representation when it is in- tions, while neurons that are activated at slightly different
duced to receive visual information. These and related results times reduce or lose shared connections, apparently to the
seem to indicate that alterations in the nature of activity patterns extent that isolating septa sometimes form between differently
in the sensory inputs from the thalamus have an organizing activated populations of neurons. Co-activations can depend
influence on cortex. not only on sensory stimuli and on the spatial arrangement of
Related experiments also show that visual maps can be receptors, but also on the transduction and response proper-
induced to form in non-visual cortex. When the portion of ties of receptors and neurons. The spontaneous activity within
posterior neocortex that normally develops into visual cor- brain structures can also have an important role as adjacent
tex is largely or completely removed before innervation by neurons respond together.
the visual thalamus, the visual inputs from the thalamus
grow instead to remaining portions of cortex to form visual Conclusions
maps.(74) Position-dependent differences in gene expression and the re-
sulting gradients of signaling molecules appear to have a critical
Implications of normal and induced features of role in the positioning of cortical and subcortical representa-
sensory maps for theories of map formation tions of sensory surfaces. They likely guide the formation of
Many of the features of sensory maps that we have outlined overall patterns of connections between such representations,
here are most easily explained by the assumption that informa- and shape at least a crude pattern of internal organization. We
tion is transported from the receptor sheet to sequentially have reviewed some of the evidence that seems to indicate that
inform and guide the development of each subsequent station information from the receptor sheet is passed on from level to
in the processing sequence. This is possible since representa- level in sensory systems to modify the crude patterns of
tions do develop according to position in the processing se- organization into detailed and often modular representations.
quence. While it seems possible that factors intrinsic to an area Part of the evidence comes from experimental manipulations
of cortex could code for all the major organizational features of the sensory environment during development that alter
of a sensory representation, it seems more likely that major developmental outcomes. These experiments have been more
aspects of internal organization are determined by the nature extensively reviewed elsewhere. Here we concentrated on
of the sensory input. For example, in mice with an extra whisker describing variations in the internal organizations of sensory
on the face or moles with an extra ray on the nose, this change maps within and across species that seem most parsimo-
in the parcellation of afferents results in matching changes in at niously explained by assuming that information from the sen-
least three or four sensory maps in the brain. It seems extre- sory sheet guides the construction of sensory maps. This
mely unlikely that a number of genes have mutated in consort information would most likely transfer from level to level via
so that independent intrinsic guidance factors in the face and at patterns of action potentials, but the transfer of chemical signals
each of several levels of the nervous system have all been by axon transport or the release of chemical signals in axon
changed to precisely match modular components in a single or arbors as a result of non-spiking neural activity are other
a few generations. Instead, it seems more reasonable to pro- possibilities. While further comparative studies of variations
pose that information from the sensory sheet instructs the in map structure may usefully suggest classes of mechan-
formation of central sensory representations in a cascading isms, the roles of such proposed mechanisms in the course
fashion. The nature of this instruction could be chemical, as of development need to be evaluated with experimental
neurons do effectively transport substances, but neurons are manipulations.
specialized to send information in discharge patterns. Neural
activity patterns have the potential to alter gene expression
and other cellular processes that mediate neural growth and References
1. Kaas JH. The evolution of isocortex. Brain Behav Evol 1995;46:187196.
synapse formation, as well as axon retractions and synaptic 2. Sperry RW. Chemoaffinity in the orderly growth of nerve fiber patterns
loss.(75) and connections. Proc Natl Acad Sci USA 1963;50:703710.
24
Review articles
3. Hubel DH, Wiesel TN. The period of susceptibility to the physiological 29. LeVay S, McConnell SK. ON and OFF layers in the lateral geniculate
effects of unilateral eye closure in kittens. J Physiol 1970;206:419436. nucleus of the minK. Nature 1982;300:350351.
4. Shatz CJ. Emergence of order in visual system development. Proc Natl 30. Stryker M, Zahs K. ON and OFF sublaminae in the lateral geniculate
Acad Sci USA 1996;93:602608. nucleus of the ferret. J Neurosci 1983;3:19431951.
5. Constantine-Paton M. The retinotectal hookup: The process of neural 31. Sherman SM, Wilson JR, Kaas JH, Webb SV. X- and Y-cells in the dorsal
mapping. In: Subtelny S, editor. Developmental Order: Its Origin and lateral geniculate nucleus of the owl monkey (Aotus trivirgatus). Science
Regulation, New York: Alan R. Liss; 1982, pp 317349. 1976;192:475477.
6. Crowley JC, Katz LC. Early development of ocular dominance columns. 32. Casagrande VA, Kaas JH. The afferent, intrinsic, and efferent connec-
Science 2000;290:13211324. tions of primary visual cortex in primates, In: Peters A, Rockland K,
7. Chiaia NL, Fish SE, Bauer WR, Bennett-Clarke CA, Rhoades RW. Post- editors. Cerebral Cortex, Volume 10, Primary Visual Cortex in Primates.
natal blockade of cortical activity by tetrodotoxin does not disrupt the New York: Plenum Press; 1994, pp 201259.
formation of vibrissa-related patterns in the rat's somatosensory cortex. 33. Horton JC, Hocking DR. Anatomical demonstration of ocular dominance
Dev Brain Res 1992;66:244250. columns in striate cortex of the squirrel monkey. J Neurosci 1996;16:
8. Kaas JH. What, if anything, is S-I? The organization of the ``first somato- 55105522.
sensory area'' of cortex. Physiol Rev 1983;63:206231. 34. Shatz CJ, Lindstrom SH, Wiesel TN. The distribution of afferents re-
9. Sur M, Weller R, Kaas JH. Physiological and anatomical evidence for a presenting the right and left eyes in the cat's visual cortex. Brain Res
discontinuous representation of the trunk in S-1 of tree shrews. J Comp 1977;131:103116.
Neurol 1981;201:135147. 35. Lyon D, Jain N, Kaas JH. Cortical connections of striate and extrastriate
10. Sur M, Nelson RJ, Kaas JH. Representations of the body surface in visual areas in the tree shrew. J Comp Neurol 1998;401:109128.
cortical areas 3b and 1 of squirrel monkeys: Comparisons with other 36. Krubitzer L. The organization of neocortex in mammals: Are species dif-
primates. J Comp Neurol 1982;211:177192. ferences really so different? Trends in Neurosciences 1995;18:408417.
11. Sur M, Nelson RJ, Kaas JH. The representation of the body surface in 37. Sur M, Weller R, Kaas JH. Modular segregation of functional cell classes
somatic koniocortex in the prosimian (Galago senegalensis). J Comp within the postcentral somatosensory cortex of monkeys. Science 1981;
Neurol 1980;180:381402. 212:10591061.
12. Florence SL, Wall JT, Kaas JH. Central projections from the skin of the 38. Sretavan D, Dykes RW. The organization of two cutaneous submodalities
hand in squirrel monkeys. J Comp Neurol 1991;311:563578. in the forearm region of area 3b of cat somatosensory cortex. J Comp
13. Calford MB, Graydon ML, Huerta MF, Kaas JH, Pettigrew JD. Altered Neurol 1983;213:381398.
somatotopy in the brain of a flying mammal. Nature 1985;313:477479. 39. Woolsey TA, Welker C, Schwartz RH. Comparative anatomical studies of
14. Lane RH, Allman JM, Kaas JH, Miezin FM. The visuotopic organization of the Sm1 face cortex with special reference to the occurrence of ``barrels''
the superior colliculus of the owl monkey (Aotus trivirgatus) and the bush in layer IV. J Comp Neurol 1975;164:7994.
baby (Galago senegalensis). Brain Res 1973;60:335349. 40. Dawson DR, Killackey HP. The organization and mutability of the forepaw
15. Kaas JH, Harting JK, Guillery RW. Representation of the complete retina and hindpaw representations in the somatosensory cortex of neonatal
in the contralateral superior colliculus of some mammals. Brain Res rat. J Comp Neurol 1987;256:246256.
1973;65:343346. 41. Lu SM, Lin RCS. Thalamic afferents of the rat barrel cortex: A light and
16. Kaas JH, Preuss TM. Archontan affinities as reflected in the visual system. electronmicroscope study using Phaseolus valgaris leucoagglutinin as
In: Szalay F, Novacek M, McKenna M, editors. Mammal Phylogeny, New an anterograde tracer. Somatosens Mot Res 1993;10:116.
York: Springer-Verlag; 1993, pp 115128. 42. Henderson TA, Jacquin MF. What makes subcortical barrels? Requisite
17. Payne BR. Representation of the ipsilateral visual field in the transition trigeminal circuitry and developmental mechanisms. In: Jones EG,
zone between areas 17 and 18 of the cat's cerebral cortex. Vis Neurosci Diamond IT, editors. Cerebral Cortex, Volume 11, The Barrel Cortex of
1990;4:445474. Rodents. New York: Plenum Press; 1995, pp 123187.
18. Bosking WH, Kretz R, Pucak ML, Fitzpatrick D. Functional specificity of 43. Catania KC, Kaas JH. The organization of the somatosensory cortex of
callosal connections in tree shrew striate cortex. J Neurosci 2000;20: the star-nosed mole. J Comp Neurol 1995;351:549567.
23462359. 44. Catania KC, Kaas JH. The unusual nose and brain of the star-nosed
19. White LE, Bosking WH, Williams SM, Fitzpatrick D. Maps of central visual mole. BioScience 1996;46:578586.
space in ferret V1 and V2 lack matching inputs from the two eyes. 45. Catania KC. A nose that looks like a hand and acts like an eye: The
J Neurosci 1999;19:70897099. unusual mechanosensory system of the star-nosed mole. J Comp Physiol
20. Allman JM, Kaas JH. The organization of the second visual area (VII) in 1999;185:367372.
the owl monkey: A second order transformation of the visual field. Brain 46. Jain N, Catania KC, Kaas JH. A histologically visible representation of the
Res 1974;76:247265. fingers and palm in primate area 3b and its immutability following long-
21. Rosa MGD, Casagrande VA, Preuss TM, Kaas JH. Visual field represen- term deafferentations. Cerebral Cortex 1998;8:227236.
tation in striate and pre-striate cortices of a prosimian primate (Galago 47. Jain N, Qi HX, Catania KC, Kaas JH. Anatomical correlates of the face
gainetti). J Neurophysiol 1997;77:31933217. and oral cavity representations in somatosensory area 3b of monkeys.
22. Pinon MC, Gattass R, Sousa APB. Area V4 in cebus monkey: Extent and J Comp Neurol 2001;429:455468.
visuotopic organization. Cerebral Cortex 1998;8:685701. 48. Kaas JH, Guillery RW, Allman JM. Some principles of organization in
23. Fukuchi-Shimogori T, Grove EA. Neocortex patterning by the secreted the dorsal lateral geniculate nucleus. Brain Behav Evol 1972;6:253
signaling molecule FGF8. Science 2001;294:10711074. 299.
24. Purves D, Riddel DR, LaMantia AS. Iterated patterns of brain circuitry (or 49. Kaas JH. Organizing principles of sensory representations. In: Novartis
how the cortex gets its spots). Trends Neurosci 1992;15:362368. Foundation Symposium 228, Evolutionary Developmental Biology of the
25. Preuss TM. Taking the measure of diversity: Comparative alternatives to Cerebral Cortex. New York: John Wiley & Sons; 2000, pp 188205.
the Model-Animal paradigm in cortical neuroscience. Brain Behav Evol 50. Van der Loos H, Dorfl J. Does the skin tell the somatosensory cortex how
2000;55:287299. to construct a map of the periphery? Neurosci Lett 1978;7:2330.
26. Conway JL, Schiller PH. Laminar organization of tree shrew dorsal lateral 51. Welker E, Van der Loos H. Quantitative correlation between barrel-field
geniculate nucleus. J Neurophysiol 1983;50:13301342. size and the sensory innervation of the whisker pad: A comparative study
27. Kretz R, Rager G, Norton T. Laminar organization of ON and OFF regions in six strains of mice bred for different patterns of mystacial vibrissae.
and ocular dominance in the striate cortex of the tree shrew (Tupaia J Neurosci 1986a;6:33553373.
belangeri). J Comp Neurol 1986;251:135145. 52. Catania KC, Kaas JH. The mole nose instructs the brain. Somatosem.
28. Thurlow GA, Bowling DB, Cooper RM. ON and OFF activity gradients in Motor Res 1997;14:5658.
the lateral geniculate nucleus of the cat: A combined 14C2-deoxyglucose 53. Guillery RW. Why do albinos and other hypopigmented mutants lack
and D,L-2-amino-4-phosphonobutyric acid study. Visual Neurosci 1993; normal binocular vision, and what else is abnormal in their central visual
10:10271033. pathways. Eye 1996;10:217221.
25
Review articles
54. Guillery RW, Kaas JH. A study of normal and congenitally abnormal 66. Catania KC, Kaas JH. A somatosensory fovea in the star-nosed mole.
retinogeniculate terminations in cats. J Comp Neurol 1971;143:71100. J Comp Neurol 1997a;387:215233.
55. Hubel DH, Wiesel TN. Aberrant visual projections in the Siamese cat. 67. Guillery RW. Competition in the development of visual pathways. In:
J Physiol 1971;218:3362. Parnavelas JG, Stern CD, Stirling RV, editors. The Making of the Nervous
56. Kaas JH, Guillery RW. The transfer of abnormal visual field representa- System. Oxford University Press; 1988, pp 356379.
tions from the dorsal lateral geniculate nucleus to the visual cortex in 68. Zheng D, Purves D. Effects of increased neural activity on brain growth.
Siamese cats. Brain Res 1973;59:6195. Proc Natl Acad Sci USA 1995;92:18021806.
57. Hogan D, Garraghty PE, Williams RW. Asymmetric connections, dupli- 69. Casagrande VA, Wiencken-Barger AE. Developmental plasticity in the
cate layers, and a vertically inverted map in the primary visual system. mammalian visual system. In: Kaas JH, editor. The Mutable Brain.
J Neurosci 1999;19:RC38. Amsterdam, The Netherlands: Harwood Academic Publishers; 2001,
58. Hogan D, Garraghty PE, Williams RW. Lamination and visual topography pp 148.
in the lateral geniculate nucleus of normal and achiasmic dogs. Eur 70. Chapman B, Styker MP. Development of orientation selectivity in ferret
J Anat 1996;1:311. visual cortex and effects of deprivation. J Neurosci 1993;13:5251
59. Florence SL, Kaas JH. Ocular dominance columns in area 17 of Old 5262.
World macaque and Talapoin monkeys: Complete reconstructions and 71. Chapman B, Godecke I, Bonhoeffer T. Development of orientation pre-
quantitative analyses. Visual Neurosci 1992;8:449462. ference in the mammalian visual cortex. J Neurobiol 1999;41:18
60. Preuss TM, Kaas JH. Cytochrome oxidase ``blobs'' and other character- 24.
istics of primary visual cortex in a lemuriform primate, Cheirogaleus 72. Rauschecker JP. Compensatory plasticity and sensory substitution in the
medius. Brain Behav Evol 1996;47:103112. cerebral cortex. TINS 1995;18:3643.
61. Tanaka S. Theory of ocular dominance column formation. Biological 73. Angelucci A, Sharma J, Sur M. Modifiability of neocortical connections
Cybernetics 1991;64:263272. and functions during development. In: Kaas JH, editor. The Mutable
62. Hensch TK, Fagiolini M, Mataga N, Stryker MP, Baekkeskov S, Kash SF. Brain. Amsterdam, The Netherlands: Harwood Academic Publishers;
Local circuit control of experience-dependent plasticity in developing 2001, pp 351392.
visual cortex. Science 1998;282:15041508. 74. Huffman KJ, Molnar Z, VanDellen A, Kahn DM, Blakemore C, Krubitzer L.
63. Lee KJ, Woolsey TA. A proportional relationship between peripheral Formation of cortical fields on a reduced cortical sheet. J Neurosci
innervation density and neuron number in the somatosensory system of 1999;19:99399952.
the mouse. Brain Res 1975;99:349353. 75. Cowan WM, Fawcett JW, O'Leary DD. Regressive events in neurogen-
64. Welker E, Van der Loos H. Is the areal extent of sensory cerebral cortex esis. Science 1984;225:12581265.
determined by peripheral innervation density? Exp Brain Res 1986b;63: 76. Shatz CJ. Impulse activity and the patterning of connectivity during CNS
650654. development. Neuron 1990;5:745756.
65. Azzopardi P, Cowey A. Preferential representation of the fovea in primary 77. Katz LC, Shatz CJ. Synaptic activity and the construction of cortical
visual cortex. Nature 1993;361:719721. circuits. Science 1996;274:11331138.
26
J Comp Physiol A (1999) 185: 367372 Springer-Verlag 1999
REVIEW
K. C. Catania
Abstract The star-nosed mole (Condylura cristata) has a is unique among mammals and raises several basic
snout surrounded by 22 fleshy and mobile appendages. questions about star-nosed mole biology. What is the
This unusual structure is not an olfactory organ, as star and how is it used by the mole to explore its dark
might be assumed from its location, nor is it used to underground environment? What kinds of sensory re-
manipulate objects as might be guessed from its ap- ceptors are associated with the star? How is information
pearance. Rather, the star is devoted to the sense of from the star represented in the neocortex? Such ques-
touch, and for this purpose the appendages are covered tions have been the subject of a number of recent studies
with thousands of small mechanoreceptive Eimers or- exploring the sensory biology of this species and its
gans. Recent behavioral studies find that the star acts relatives (Catania et al. 1993; Catania 1995a, b, c, 1996;
much like a tactile eye, having a small behavioral focus, Catania and Kaas 1995, 1996, 1997a, b). Each aspect of
or fovea at the center used for detailed explorations the sensory system, including the moles behavior, the
of objects of interest. The peripheral and central nervous structure of the nose and sensory receptors, and the
systems of the mole reflect these behavioral specializa- physiology and anatomy of the cortex, has provided
tions, such that the small behavioral focus on the nose is some important insight into these basic questions. But
more densely innervated in the periphery, and has a the greatest insights come from combining these studies.
greatly enlarged representation in the somatosensory Unusual specializations in the cortex are explained by
cortex. This somatosensory representation of the tactile behavioral observations. Anatomical comparisons
fovea is not correlated with anatomical parameters (in- between the star and its cortical representation reveal
nervation density) as found in other species, but rather is unexpected parallels between the moles somatosensory
highly correlated with patterns of behavior. The many system and the primate visual system. These results
surprising parallels between the somatosensory system provide an example of how a specialized species can
of the mole, and the visual systems of other mammals, reveal general principles of brain organization.
suggest a convergent and perhaps common organization
for highly developed sensory systems.
Star-nosed mole behavior reveals a tactile fovea
Key words Cortical magnification Somatosensory
cortex Development Evolution Behavior Star-nosed moles live in wetlands in the eastern United
States and Canada. They build extensive tunnel systems
with their heavily clawed forelimbs, and seldom come to
Introduction the surface. Moles are insectivores with a high rate of
metabolism, and they must find and eat a large number
The star-nosed mole is renowned for its unusual nose, of invertebrate prey each day. As would be predicted
which consists of 22 mobile, fleshy appendages that from their fossorial lifestyle, they have small eyes and a
surround the nostrils. This peculiar nasal specialization tiny optic nerve. Thus a major challenge to their survival
is finding sufficient quantities of often small prey in their
dark underground tunnels. The star seems to have
evolved mainly for this purpose, and may be the most
K.C. Catania sensitive touch organ among mammals.
Department of Psychology, Vanderbilt University,
301 Wilson Hall, 111 21st Ave South,
When moles search for food or explore their envi-
Nashville, TN 37240, USA ronment, the star is in constant motion and is repeatedly
Tel.: +1-615-322-7491; Fax: +1-615-343-8449 touched to the substrate or objects of interest. Each
27
368
touch consists of raising the nose upward while the nasal first encountered by the larger peripheral rays which
rays swing backward, then swinging the rays forward together make up most of the surface area of the star.
while the star is brought into contact with the substrate. This stereotyped sequence of movements is very rapid.
This behavior is very rapid; star-nosed moles may touch Moles can touch a small prey item with the peripheral
ten or more dierent places each second. Similar be- rays, shift the star for several additional touches with ray
havior in other moles without a star has been de- 11, and take the prey into the mouth, all in about 400 ms
scribed as tapping the ground with the nose (Nagorsen (Catania and Kaas 1997b).
1996). When a prey item such as an earthworm is The division of the star into peripheral touch and
encountered, the rays are repeatedly touched to the prey central touch seems analogous to the retina in the visual
as it is bitten, torn up, and eaten. When prey is missed by system of other mammals. In primates, for example,
the star, by even a millimeter or two, the mole proceeds photoreceptors in the peripheral retina typically detect
past without apparent notice (whether searching in wa- potentially important stimuli first, and then a saccade
ter or soil), arguing against the hypothesis that the star brings the image onto the higher resolution, denser re-
functions to detect prey from a distance through elec- gion of photoreceptors of the retinal fovea. Does the
troreception (Gould et al. 1993). anatomy of the nasal rays also reflect the dierent roles
With the naked eye, the details of foraging and prey they play in behavior?
encounters are difficult to discern, but slow motion
videotaped behavior of moles locating small prey reveal
a stereotyped sequence of nose movements and a be- The peripheral anatomy and sensory receptors of the star
havioral focus or fovea on the star (Fig. 1). Ray 11,
the center-most ventral ray (Fig. 2) is preferentially used Although the star is a specialization of the distal portion
to explore prey items, acting in a manner analogous to of the nose, it is obviously not an olfactory structure. At
the fovea in the visual system of other mammals. For first glance, one might imagine the nose acting as an
example, whenever a prey item is first contacted with the extra hand. But the rays do not contain muscles or bones
peripheral rays (rays 1 through 10) the nose is then and are not used to manipulate objects or capture prey.
shifted so that further detailed explorations are made They are controlled through tendons by a complex series
with ray number 11 (Fig. 1A). No food item is eaten of muscles that attach to the skull, and their role seems
without first being explored with ray 11. But since ray 11 to be purely mechanosensory. For this purpose, they are
is one of the smallest rays on the nose, food is usually composed almost entirely of a series of small mechano-
sensory organs called Eimers organs. Under the
scanning electron microscope, these organs are visible in
Fig. 1AC The use of the nose while foraging for food. A A
a honeycomb pattern of epidermal domes on the surface
schematic representation of the nose with the rays numbered from 1 to of each nasal ray (Fig. 2). Eimers organs are very sen-
11, shown in relation to a prey item (gray oval). The first touch made sitive to light touch (Catania and Kaas 1995, 1997a) and
contact with rays 4 and 5 as the mole discovered the food item. The they are found on the snout of every member of the
second touch was centered on the 11th central pair of rays, after which family Talpidae that has been examined (Catania
the food was eaten. For brevity, the prey is shown changing position
relative to the star but in reality, the star is moved relative to the 1995b). However, the star-nosed mole has many more
prey. This sequence of movements is characteristic of prey encounters, Eimers organs than any other species and their distri-
which often include multiple touches with the 11th pair of rays. B A bution across the nasal rays is unique: while most moles
flow diagram illustrating the progression of touches from the lateral have at most a few thousand Eimers organs surround-
rays to the central 11th ray before the food is eaten. C An example of
the distribution of touches across the nose for ten consecutive preys
ing their nostrils, the star-nosed mole has over 25 000 on
encounters, showing the preferential use of the 11th and surrounding the star. Each organ is supplied by a number of primary
rays. B and C from Catania and Kaas 1997b aerents, thus the star is very densely innervated. There
28
369
Terminal Swellings of
2
A. 1
B. C. Free Nerve Endings
Epidermis
5
N
6 Merkel Cell
Neurite Complex
Dermis
Encapsulated
Corpuscle
11 7
10
Myelinated 10m
Fibers
8
9 Eimer's Organ
Fig. 2AC The nose and sensory organs of the star-nosed mole. A characterized, recordings from the cortex show the star to
The nose is composed of 22 appendages, or rays that surround the be highly responsive to several dierent features of fine
nostrils (N). There are 11 symmetric pairs, numbered from 1 to 11 on
each side. The rays are used by the mole to explore its dark tactile stimulation (Catania and Kaas 1995). The super-
underground environment through touch. The rays can be moved ficial nerve terminals in each Eimers organs are ideally
through tendons that connect to facial muscles, and they are positioned to be stimulated by pressure to the top of the
repeatedly touched to the substrate as the mole moves through its cell column when the rays are touched against an object.
system of tunnels. B Each ray is covered with many hundreds of
domed sensory organs called Eimers organs. These are apparent in
The Eimers organs have the same basic structure
a honeycomb distribution on the skin surface at high magnification. across the dierent rays on the star. Ray 11, the be-
The rays are essentially composed of these sensory organs which are havioral focus of the nose, does not have a higher den-
densely innervated by a branch of the infraorbital nerve. C Each sity of organs, or more organs than other rays, as might
Eimers organ is a specialization of the epidermis that contains an be guessed from its important role in search behaviors.
array of sensory receptors, including an encapsulated corpuscle, a
merkel cell-neurite complex, and a series of intra-epidermal nerve In fact, because it is relatively short and has little caudal
endings that terminate in swellings at the apex of the central cell surface, it has fewer Eimers organs than almost any
column. Plate A from Catania and Kaas 1995 other ray. Ray 11 has about 900 Eimers organs on its
surface while some of the lateral rays have well over
are over 100 000 myelinated fibers innervating the rela- 1500 (Catania and Kaas 1997b).
tively small star, and a substantial cross section of each Rather than having more sensory organs, a dierent
ray is taken up by a large nerve branch (Catania 1995a). way in which a skin surface may be more sensitive to
The internal structure of the star-nosed moles Ei- mechanoreceptive input is by increasing its innervation
mers organs is illustrated in Fig. 2C. Each organ is a density and decreasing each aerents receptive field. To
roughly 40-lm swelling of the epidermis that contains a test for this possibility, the number of Eimers organs
central column of keratinocytes. At the bottom of each and the number of myelinated fibers innervating each
organ, in the connective tissue of the dermis, there is a ray were counted and compared (Fig. 3). Rays 1
single encapsulated corpuscle. Just above this, among through 9 each had about 4 fibers per Eimers organ,
the keratinocytes that form the base of the central cell while rays 10 and 11 had significantly higher innervation
column, there is a single merkel cell-neurite complex. densities of 5.6 and 7.1 fibers per organ, respectively.
Within the cell column, a series of nerve fibers branch Thus there is a peripheral specialization of the rays most
from three myelinated fibers in the dermis and ascend used for exploratory behaviors, such that ray 11 has a
along the margins of the column to terminate in a cir- much higher innervation density than more lateral rays.
cular arrangement of swellings just below the outer layer Previous studies of the organization of the somato-
of epidermis. This concise geometric arrangement of sensory cortex, where touch information in mammals is
terminal nerve swellings is entirely enclosed and encap- processed, have found a correlation between peripheral
sulated by a single circular keratinocyte at the apex of innervation density and the size of the corresponding
the cell column and the overlying outer layer of epi- representation or map of the sensory surface in cortex.
dermis is tightly sealed by many desmosomal adhesions To investigate this relationship in moles, we examined
(Catania 1996). the organization of the moles somatosensory cortex and
While the response properties of the separate nerve related these areas to the anatomy of the star (Catania
terminals at the apex of Eimers organ have not been and Kaas 1997b).
29
370
Average Fibers per Eimer's Organ 8 Fibers per Organ Anatomical proportions
7
6
5
4
3
2
1
0
1 2 3 4 5 6 7 8 9 10 11
Ray Number
Fig. 3 The ratio of myelinated fibers to Eimers organs from the 11 Cortical proportions
rays of the nose (from four moles). While rays 1 through 9 have an
average of roughly 4 fibers per Eimers organ, the 10th and 11th have
significantly higher innervation densities of 5.6 and 7.1 fibers per
organ, respectively (from Catania and Kaas 1997b) Fig. 4 Cortical magnification in star-nosed moles. The upper drawing
shows the anatomical proportions of the various body parts. The
lower drawing of a moleunculus shows the relative size of each body
part as it is represented in somatosensory cortex. As would be
The representation of the star in cerebral cortex predicted from its high innervation density and behavioral impor-
tance, the nose dominates the cortical representation. The forelimb
To explore the organization of the somatosensory cortex also has a large cortical representation, probably reflecting its
important sensory role in the excavation of tunnels, rather than in
in star-nosed moles, neuronal responses were recorded locating prey (from Catania and Kaas 1996)
using microelectrodes in anesthetized moles, while the
nose and body were gently stimulated with small probes
(Catania et al. 1993; Catania and Kaas 1995). A large of stripes was a mirror image of the first, and corre-
area of cortex responded to stimulation of the star, while sponded to the second somatosensory area, S2 (Catania
smaller areas responded to the rest of the body and the and Kaas 1995). S1 and S2 are somatosensory areas
limbs (Fig. 4). After mapping the locations of receptive found in all mammals (Kaas 1987, 1995), including ro-
fields from the star and body of the mole onto the so- dents, but there is not a visible representation of each
matosensory areas, the moles were perfused and their whisker representation in S2 of rodents. The third set of
cortex was flattened, sectioned parallel to the cortical stripes also responds to tactile stimulation of the nose,
surface, and stained for the metabolic enzyme cyto- but it has not yet been fully mapped because of its small
chrome oxidase, which often reveals cortical subdivi- size and lateral location in the brain. The presence of
sions. two and perhaps three visible maps of the star in dif-
In the area of cortex that contained an electrophysi- ferent somatosensory areas of the mole is unique, and
ological map of the star, a set of dark stripes was found may be a specialization related to the very high inner-
in brain sections processed for cytochrome oxidase vation density and large number of mechanoreceptors in
(Fig. 5). Each stripe corresponded precisely in location the star.
to an area that responded to a single nasal ray in the Another unusual but not entirely unexpected finding
microelectrode recordings. Thus the representation of was a hugely enlarged cortical representation of the 11th
the star in somatosensory cortex is visible in brain sec- nasal ray. This was seen in electrophysiological record-
tions, much as the representation of whiskers in rodents ings (Catania and Kaas 1995, 1996), and is also strik-
is visible as a set of cortical barrels in their cortex ingly clear in the cytochrome oxidase pattern of stripes
(Woolsey and Van der Loos 1970). However, dierences visible in brain sections (Fig. 5). This ray often takes up
between the barrel cortex in rodents and the star rep- over 25% of the entire S1 cortical representation of the
resentation in moles were immediately apparent. nose, despite its relatively small size on the star. Be-
First, multiple maps of the star were seen in sections haviorally important sensory surfaces often have greatly
of the moles cortex. As in all mammals, each half of the enlarged cortical representations relative to their ana-
body is represented in the opposite cerebral hemisphere, tomical size. But these enlargements are generally at-
but in each mole hemisphere there were at least two tributed to their higher innervation densities, rather than
clearly visible sets of 11 stripes representing the contra- specific influences of behavior. For example, a direct
lateral star. In some favorable cases, a smaller third set linear correlation has been found between the size of
of stripes was also apparent (Catania and Kaas 1995, cortical barrels in rodent somatosensory cortex and the
1996). The most prominent set of stripes corresponded number of aerents innervating the corresponding
to the primary somatosensory cortex, S1. The second set whisker on the face (Welker and Van der Loos 1986). To
30
371
b
4
A. 3 5 Fig. 5AC The cortical representation of the star. A A rotated view
of the left side of the nose (dorsal is to the left, lateral is up) oriented to
match the nose representation in the primary somatosensory cortex
6 (S1) of the right hemisphere (below). B A tangential section through
layer 4 of the primary somatosensory nose representation in cortex,
7 processed for the metabolic enzyme cytochrome oxidase. The pattern
of 11 rays on the nose is mirrored in the brain as a series of 11 dark
stripes. Note, however, the 11th subdivision in the brain is greatly
2 enlarged relative to the size of the 11th ray on the nose. The
disproportionate representation of the 11th ray, and to a lesser extent
8 the adjacent rays, reflects their behavioral importance (Fig. 1). C The
average area of cortex in S1 per primary aerent for each nasal ray.
The representations of the rays in cortex are not proportional to their
respective innervation densities. Rather, the 11th and surrounding
1 9 rays have greater areas of cortex per aerent. The pattern of aerent
10 magnification is very similar to the distribution of touches across the
11 nose during feeding behaviors (Fig. 1C), suggesting a possible role for
behavior in shaping these cortical areas during development. Plate C
from Catania and Kaas 1997b
B.
8 9 innervation density did not account for the size of the
6 7 10 11th subdivision in somatosensory cortex (Catania
5 1995c; Catania and Kaas 1997b). Ray 11 contained
4 about 7% of the Eimers organs on the star, received
about 11% of the nerve fibers innervating the star
3 (accounting for the higher innervation density per or-
gan), but took up about 25% of the cortical represen-
2 tation of the star. Thus ray 11 is specialized both in the
periphery and in the cortex, by a higher innervation
1 11 density and a larger cortical representation per aerent,
respectively.
These findings are dierent from the results of similar
investigations in rodents (Welker and Van der Loos
1986), but nevertheless may be a common condition in
mammalian somatosensory cortex that is simply difficult
C to quantify in most species where the areas of cortical
150 S1 Cortex per Fiber representations and corresponding innervation densities
(Afferent Magnification) of skin surfaces are difficult to measure. This preferential
Average Area of Cortex
31
372
vealing a general aspect of the cortical representations of Catania KC (1995a) The structure and innervation of the sensory
important inputs across dierent sensory systems. organs on the snout of the star-nosed mole. J Comp Neurol 351:
549567
All of these findings raise a number of additional Catania KC (1995b) A comparison of the Eimers organs of three
questions about star-nosed mole sensory biology. For North American moles: the hairy-tailed mole (Parascalops
example, why have multiple cortical representations of breweri), the star-nosed mole (Condylura cristata) and the
the nose, rather than one large representation? Does eastern mole (Scalopus aquaticus). J Comp Neurol 354: 150160
Catania KC (1995c) Magnified cortex in star-nosed moles. Nature
adding new representations of the nose increase com- (Lond) 375: 453454
putational efficiency through parallel processing of dif- Catania KC (1996) Ultrastructure of the Eimers organs of the star-
ferent aspects of touch? These representations are also nosed mole. J Comp Neurol 365: 343354
topographically interconnected both in the same hemi- Catania KC, Kaas JH (1995) The organization of the somatosen-
sphere and across hemispheres to the representation of sory cortex of the star-nosed mole. J Comp Neurol 351:
536548
the contralateral star. The ability to identify each re- Catania KC, Kaas JH (1996) The unusual nose and brain of the
spective ray representation in multiple maps will allow star-nosed mole. Bio Science Rep 46: 578586
future studies to determine the circuitry of this system in Catania KC, Kaas JH (1997a) The organization of somatosensory
detail (with the aid of anatomical tracers) to determine cortex and distribution of corticospinal neurons in the eastern
mole (Scalopus aquaticus). J Comp Neurol 378: 337353
the degree of parallel or hierarchical organization of the Catania KC, Kaas JH (1997b) Somatosensory fovea in the star-
cortical representation. nosed mole: behavioral use of the star in relation to innervation
Another area of future research includes the func- patterns and cortical representation. J Comp Neurol 387:
tioning of Eimers organs. The concise geometric ar- 215233
Catania KC, Kaas JH, Northcutt RG, Beck PD (1993) Nose stars
rangement of nerve terminals at the apex of each organ and brain stripes. Nature (Lond) 364: 493
may be used to detect the surface features of objects in the Drasdo N (1977) The neural representation of visual space. Nature
10- to 100-lm size range. If so, the sensory world of the (Lond) 266: 554556
star-nosed mole could include a new realm of perceptions Gould E, McShea W, Grand T (1993) Function of the star in the
not usually considered salient to mechanosensation, such star-nosed mole, Condylura cristata. J Mammal 74: 108106
Kaas JH (1987) The organization and evolution of neocortex. In:
as microscopic textures and surface features previously Wise SP (ed) Higher brain functions. Wiley, New York, pp 347
considered impossible to detect by mammalian touch. 378
However, the most striking result of these studies is the Kaas JH (1995) The evolution of isocortex. Brain Behav Evol 46:
similarity between the patterns of behavior in moles 187196
Malpeli JG, Baker FH (1975) The representation of the visual field
(Fig. 1C) and the pattern of aerent magnification of the in the lateral geniculate nucleus of Macaca mulatta. J Comp
rays (Fig. 5C). The sizes of the cortical modules repre- Neurol 161: 569594
senting the rays are not correlated with the anatomical Myerson J, Manis PB, Miezin FM, Allman JM (1977) Magnifica-
parameters traditionally assumed to be driving these tion in striate cortex and retinal ganglion cell layer of owl
monkey: a quantitative comparison. Science 198: 855857
specializations, but rather are highly correlated with the Nagorsen DW (1996) Opossums shrews and moles of British
patterns of mole behavior (see Catania and Kaas 1997b Columbia. UBC Press, Vancouver, British Columbia
for more details). Here we may be able to examine the role Perry VH, Cowey A (1985) The ganglion cell and cone distributions
of behavior in brain development to see if it is in fact in the monkeys retina: implications for central magnification
driving the enlargement of cortical areas, or whether these factors. Vision Res 25: 17951810
Silveira LCL, Picanco-Dinz CW, Sampaio LFS, Oswaldo-Cruz
behavioral and anatomical parameters are somehow in- E (1989) Retinal ganglion cell distribution in the cebus monkey:
dependently matched to one another during development. a comparison with the cortical magnification factors. Vision
Res 29: 14711483
Acknowledgements Thanks to Jon Kaas and Glenn Northcutt for Wassle H, Grunert U, Rohrenbeck J, Boycott BB (1989) Cortical
their support, guidance, and many fruitful collaborations in the magnification factor and the ganglion cell density of the primate
course of these studies. Neeraj Jain, Christine Collins and Melanie retina. Nature (Lond) 341: 643646
Catania provided helpful comments on the manuscript. Special Wassle H, Grunert U, Rohrenbeck J, Boycott BB (1990) Retinal
thanks to Hrathkus and Bill Catania for their help capturing star- ganglion cell density and cortical magnification factor in the
nosed moles. primate. Vision Res 30: 18971911
All experiments and animal care procedures were approved by Welker E, Van der Loos H (1986) Quantitative correlation between
the Vanderbilt University Animal Care Committee and follow the barrel-field size and the sensory innervation of the whiskerpad:
National Institute of Health guide for the care and use of labora- a comparative study in six strains of mice bred for dierent
tory animals. patterns of mystacial vibrissae. J Neurosci 6: 33553373
Woolsey TA, Van der Loos H (1970) The structural organization
of layer IV in the somatosensory region (SI) of mouse cerebral
References cortex: the description of a cortical field composed of discrete
cytoarchitectonic units. Brain Res 17: 205242
Azzopardi P, Cowey A (1993) Preferential representation of the
fovea in primary visual cortex. Nature (Lond) 361: 719721
32
Bullock, T.H., Orkand, R., and Grinnell, A. (1977) Introduction
to Nervous Systems. W. H. Freeman &Co. San Francisco. Chapter 7
Integration at the Intermediate Levels. pp. 242-290.
33
242
into ad.lplively patterned ("coordil1<1ted") second means is to vary the frequency of
Chapter 7
sire.llTIS of impulses in the output chan- impulses in each unit, since either the
Integrdlion at the iniermediat(' Levels
nels. Again the connectivity and dynamic aver,'ge tension in a series of twitches or
properties of the network determine its the teta nic tension is a function of fre -
Figure 7.1 outpul, but not merely by passive filter - quency. You can observe frequency and
Rf(rrlilm(ll/ ill /I muscle wilh ing. Often there are intrinsic rhythms of recruitment;control phenomena by plac-
IIlru molor III1i/S.
spontaneous impulse bursts, and these ing a stethoscope over your eyelid and
Silent mOlor result in the generation of specified con- listen ing to the twitches of its muscle
stellations and sequences of activity in while you control them willfully. Re-
populations of units and finally in the cruitment may be important in verte-
effectors. Let us look first at the final brate skeletal muscle only at low ten-
link. sions.
Since there are many motor units in
most vertebrate skeletal muscles, oper-
II. NERVOUS CONTROL IN ating these unit s out of phase with each
r
EFFECTORS: DIVERSITY OF other can give a smooth overall contrac-
PERIPHERAL INTEGRATION tion even when the frequency in each is
so low that it contracts in a series of
The best known vertebrate skeletal twitches. Another possible utility of
muscles consist of muscle fibers whose having many units is that there could be
cell membrane is capable of producing a rotation of activity during low or
propagated all-or-none impulses like medium work loads, sa..that units could
those of nerve fibers. Each muscle fiber is rest. But this old notion is apparently not
innervated, in general, by only one axon confirmed in the best studied materials.
with one terminal at the speci"lized Other muscle fibers in vertebrates arel
end-plate. A motor nerve impulse arriv- incapable of producing propagated
Muscle fibers '1. inactive
ing there gives rise to an end-plate or Mtion potentials. They are innervated
Low ~sjon junctional potential, like an e.p.s.p., not by a single end-plate, but by numer-
Axons with
impuls.!;. trains which is usually suprathreshold for the ous spatially distributed motor nerve
initiation of a propagated muscle action terminals. Such multiterminally in-
potential. Each motor neuron innervates nervated muscle fibers (Fig. 7.2) respond
many muscle fibers. The motor neuron electrically with junctional potentials
and its muscle fibers comprise a motor only, but since these occur at many sites,
unit. Each muscle contains from one to eJectrotonically spreading depolarization
several thousands of motor units. An can activate the whole fiber. Muscle
impulse in one unit, or a synchronous fibers of this type m"y be mixed with
volley in many, causes a brief contraction others, as in some frog muscles. They are
to summate if they overlap, or to fuse usually innervated by small diameter
into a smooth contraction called a teta- motor axons.
j
nus if the frequency is high. When there Since individual vertebrate muscles
are many motor units per muscle, a are excited by many axons, we think of
means of grading the strength of con- them as being driven by pools of motor
traction of the muscle is to vary the neurons. There is a diversity of s ize
number of active un its. This method of within the pool of motor neurons. Gen-
~-v--
High tension control is called recruitment (Fig. 7.1). A er<llly, the smaller ones have lower
34
243
Fig ure 1.2 bers of the muscle fi bers it supplies
T!lI'''s of s1 riR /fd /H1/S(lf illlUrvRlio ,l, Se<:tiol'l II
because diffe rences in the facili talion Nervous Con'rol In Effec,o rs:
properties of the synapses allow trans- O ivers il y of Peri pheral In. egra Uo n
mission to be effecti ve "I different fre-
""""""""""""""" ; """""""'""""',:
Unil erminal innervatio n
q uencies of arriving nerve impu lses in
different fibers. There are also differ-
ences in excitati on-cont raction coupling
threshold in some muscles.
Innerva tion of all th ese muscles is of
the mul tilerminal type. At norma l fre -
Mul1i ' lIrminal innllrvalion quencies of "rriv ing nerve impulses, the
junctions commonly ex hibit the t ime-
de pendent properlies of facilitation or
antifacilitation. Contract ion strength can
";::Z""""""\C""",,z ',,,n, ;:;$';,
Dinou ronal innerva,i on
be controlled not only by average fre -
quency of moior axon impu lses, but, in
some muscles, also by deta iled tempora l
35
244
impulses at high frequen cy for quick
A SLOW movements and produces la rge junc-
tional potent ia ls in the muscle fi ber (Fig.
7.3,8), .md these elicit spike-shaped loca l
response potentials and rapid cont rac-
tions. Anotller axon is a "slow." or tonic,
nerve fiber that normally carries long
trains of impulses at low frequency;
these impulses decrement severely as
they approach the nerve ending, and
B FAST
prod uce smaller and more facilitating
0.25 9 I - ltm\\\~~I~II'lI~\I'I\I\I\I\\"\I'I.
~
~
innervating axon may end upon a large
fraction of all th e fibe rs of the muscle,
the fast axon preferentially reaching the
2 sec fa st (short sa rcome re) and intermediate
C
muscle fibe rs; the slow axon, the s low
ISO (long sarcomere) and intermediate mus-
cle fibe rs (Fig. 7.4). Th e presynaptic
2 endings of the same axon upon muscle
~
.,
0 _
0-
o 100
"0
_ -
0
fibe rs of d ifferent sarcomere length and
hence contraction properties differ in
effectiveness, in arousi ng post junctional
,"
0 -
-=
oE
potentia ls, and in degree of faci litation.
An extr.lord inary heterogeneity of mus-
'~a 50 cles and actions and thei r smooth grada -
~
0
tion ca n now be understood, even in
such animals as arthropods, in which
U
only a few axons s upply a whole muscle.
o ~O--:2C--:.--:,C--~8-:-:I~O--:I2;--:I';'--:'~6 ~.3 Th is heteroge neity is the resu lt of the
di ffer ent combinat ions of phasic, tonic,
Excita tory spaci ng interval (msec)
Fig ure 7.3 and inhib itory axons, th e different de-
Differences .llllong mu sc le s in response 10 stimulus inlerv"L A. A mu S( le grees of polyneu ronal innerva tion, th e
fiber in the c!.1\V. cl 05cr musc le o f " cTab res po nd s to repetit ive sl imu l,l!io n d iversity of type of terminal of each axon
of " "slow" ~ )(on with 1.ITge detion potentials (upper Ir",e) hav ing slow upon dirfel'ent muscle fi bers, the diver-
(,lci lit,ltio n, and wi th gr,u(u,llly SU lllllldling smooth tetani c contTdc! ;o n
s ity of type of muscle nbe r, and the
(lower I rd CC shows tenSio n). B. T he Sdme Illu scle fib er exc ited vi .. a " foist"
,,"o n shows 5111.111, qllickly pl,l\cduing action pote ntia ls and small, cloni c . diversity of rise and fa ll times of fa cilita-
( :::; nonfuse d) co nlrdclions. IAtwood, 1967. ] C. Pd ttern -se nsilive (fi lled cir- ti on.
cles) ,1lId p,lttcrn-i n sensi livc (open c ircles) mu sc le of (ray fish. At a con- Inhi bitor axons exert two distinct
st,lnt ,,,~n,, freq ue 'lcy o f 30 shocks/ sec (== 33.3 msec {ntervals), trains o f effect s, presynapti c and postsynaptic
~ hock s Me d elivered ei ther ,1 t uniform intervals (extreme right) or at alt er-
(Fig. 7.5). Postsynaptic inhibition causes
ndtely short (d b5d ss~ v~ l ucs) ~n d long illter\'~ls; respon se plo tted as a per_
centd ge of the e\'en ly sp~C(!d cO llt ra ction. Refrdctori ness red uces responses hyperpolariz..ll ion and increased con-
at the shortest int erv.ll s. IRipley .I nd Wiersmd, 1953. 1 ductance of the muscle fi ber membrane;
36
245
Fast conlraction
\LliLJ:LLLL1.LLLL.'.LLJ..LlllLLllJJ.J
More
facilitat ion
1
Figure 7.5
Less
fac ilitalion
j Di~grtHlwmli(
of a (msla(e~tI
re"restllialioll of
Hcitatory a"d i"/Iihitory imlervaliml
mllsdt fiber.
III ~
Nerve
/termina I S~
Excitatory Inhibitory
m ))) ) ) ) )6
inhib.
37
246
Chapter 7
,
Integration at Ihe Intermediate levels
,
~
D F
Fig ure 7 .6
The ,tn"tomical distribution of ,1XOnS to the dista l thor"cic limb muscles of several groups of
Mabcostrac ... BI.lC k lines rep resent motor excitor " xons; colored lines represent inhibitory axons.
Br,Kke!s in dicate tht: distribution ofaxons behveen nerve blllldies. A. Br,lChyu,,,.,.u. AIlDmur" . C.
Stom"lopod,,- D. Palinu,a. E. Astaell'''. F. Stenopodide,' (N"I",,!i.,). The boxes rep,esent the muscles
in the three di stal segments o f the legs, .IS indicated by thei r COllllllOn IMllle5: Il. ,1Ccessory nexor; b,
bender; (, closer; f, exten so r; f. nexor; 0, opener; r, rol,l\or; ~. stre tcher. IWiersll1.l and Ripley, 1952.)
presynaptic inhibition redu ces mi niature leased with facilitation, which varies
e.p.s.p.'s and the amount of excitatory among junctions.
transmitter released per impulse. Both An additional degree of complexity of
depend on the arrival time of the inhibi- muscle innervation in arthropods can be
tory impulse relative to the excitatory, appreciated if one examines the innerva-
but in different ways. They are widely tion of the muscle of a whole limb, such
different in importance. Inhibition in as the walking leg of a crayfish. Several
some muscles is mainly postsynaptic; in muscles may be innervated by a single
others, presyna ptic. In some mu scles the motor axon (Fig. 7.6). The overlapping
response to a slow excitor is inhi bited but nonidentical innervation patterns
presytl<1ptically, and the response to a still .1110w independent action. For
fast excitor in the same muscle is 'Iess example, the opener and stretcher mus-
inh ibited . In the crayfi sh claw opener cles are innervated by the same excitor
muscle, inh ibit ion cau sed by 10 impulses axon but by different inh ibitors. Never-
per second is 90% presynaptic, but a t 40 theless, the whole limb is innervated by
impulses per second it is only 50% pre- only 12 excitatory and J inhibitory motor
synaptic. Inhibitory trans mitter is re- fibers, and this fact of s mall numbers
38
247
Figure ??
makes it seem possi ble that we will be all y stimulated by a fract ion of the im- Various Iyl"'s of ,!f,c/ors.
ab le to understand fully ,l t the level of pi ngi ng world that is large but circum-
s ingle nervous units the normal control scribed (the excita tory receptive fi eld,
of movement in arthropods. ERF) and which overlaps substantially , ,
:,
~
- :
The important lesson from the with that of cont iguous ch,lllneis. T his
, , -,
exa mples chosen is that even ,lt th e last applies in some systems to topographic
lin k, the nervous control of effectors, oV('flap and field s, s uch as areas of the , , ... ,
/ '\
th ere is diversity in the principles em- skin, retina, a nd basilar membrane --,
Fly wing muscles
ployed, some cases ex hi biti ng a sub- (which distributes sound frequencies to
stantial degree of in tegration in the aud itory afferent s). In some systems the
periphery. Wags have sa id crabs can overlap involves ambiguity in modality
think (evaluate) in their legs. We h,we (Chapter 6, p. 214).
not exha usted the diversity: fast insect The third principle relevan t to the Snail heart
flight muscles, a va riety of smooth mus- present section states that sensory input
cles (Box 7.1), electric organs, glands, typically exhibits divergence and con-
cilia, chromatophores, a nd other effec- vergence in th e central nervous system,
tors manifest additional principl es of se- the former d istributing information to
ries and parallel control (Fig. 7.7). analyzers concerned with different as-
pects of the sti mu lus world, the la tter Bladder. ureter. u reth ra
permitt ing resolution of ambigu ities,
III. ANALYSIS OF sha rpening of field s, and abstraction of
0 'VIJ\iIfWWVl M!l!\ilflfV\f\']
S ENSORY INPUT, special features. Generally, divergence
PARALLEL AND SERIES predominates in sensory pathways, so rlJlit/lJlJlflJVWW\IWIJV\I)
PROCESSING that even though each neuron still re- ~'1NlfWV\N\I\I\fV\.I\i\iUvt,'j
ce ives synapses from ma ny different
The general principles of sensory recep- presynaptic cells, there are increasing e""""'VWINVlIVWWVW]
tion are common to systems as different nu mbers of cells at each higher neu ral Elect ric organ
as vision and taste and to inputs that level enroute to the sensory cortex . One
c~,
rarely re,lch consciousness, such as those characteristic exam ple is the first- to
from pressure receptors and che mo- second-order neuron stage in the audi-
receptors in the grea t Mteries, as well as tory system, already mentioned (p. 108).
those that do. Foremost is the principle The value of such wide divergence is that
of parallel cha nnels: many receptors in- different postsynaptic cells, receiving Salivary gland
dependently sample the sti mulu s world in put from subtly different, overlapping,
a nd send their imp ulse-coded reports to presynapti c populations, can perform
the centra l nervous system in parallel d ifferent types of information-process ing
afferent fiber s. The basic independence operations in parallel, more fully ex-
may in some organs be abridged by tracting all available in for m'ltion. Thus a
superimposed in fluence fro m the centra l visua l signal ca n be analyzed by diffe rent
nervous system (centrifugal or efferent cell populations for brigh tness, shape,
infl uence) or from neighboring receptors color, distance, and movement; an audi-
(lateral inhibition). tory signal, for frequen cy, rate and di-
Inti mate ly rela ted is the second rection of change of frequen cy, duration,
principle, that of overlapping receptive in tens ity, presence of overtones and
fields: neighboring receptors are gener- temporal patterns, localiza tion of the
Chromatophore
39
248
Schelnatic represenldtio n of the types of autono mic innervation of s mooth mu scle. All th e smooth -
muscle fibers are In terconnected by l ow-re5 i 5 t~ n ce electrotonic junc tio ns, shown here as bridges.
Some receive direct nerve endings (ddrk shading): othe r.; (light shading) do not, but Me clo~
enough to the forego ing to show junction potentidls th~t h,we spread elect rotonical1y. The mo st
remote (unsh ..ded) show no junction potentials; they n",y be elicited by electroton ic sprNd of
aClion potentials from the preceding d"ss and by tr.. nsmitter rele.. sed fro m the varicose nerve fibers
at some disl"nce. [BurnstO( k and IWJ~ama, 1971.1
40
249
source in space, etc. After bu ilding up mainly phasic or mainly tonic char.lcter.
5lIon III
s harply specillc response requirements, For each of the four combinations of AnJiY5iJ of ~n'iOry Inpul:
these elements can again converge to these properties there are two types, I'Molliel ~nd Series ProcesSi ng
combine their specificities. distingu ished by the sign of their COIl-
Ani mals abstract from their total sen centric orga niza tion. There are ON-
sory input specia l qua lities of that input center-OfF-surround units and Ihe con- Overlap 01
receptive lields
before formulating a motor command. A verse, OFF-center-ON-surround units.
frog jumps and snaps at any sma ll, dark, Units of the former sorl are excited by
moving object wit hi n range if it is in the the ON of a sma ll spot of light in the
mood. A male stick leback fi sh attempts centra l zone of the ERF or by the OFF of
to CQmt with any oval, silvery, red- an annular illu mi nation of the s urrou nd-
bottomed object of su itable size, whether ing zone. Units of the latter sort arc
it be a female stickleback or a crude converse in character. The table lists all
mode l. A few ChM.lclers of Ihe whole types of uni ts in the order of their axon
constellation of visual inputs associated diameter, the largest first. T he largest
with the female stickleback seem to be and fa stest un its, those in the brisk-
the only relevant ones to the ma le (sec transient group, are also ca ll ed
Chapter 8). How might fil tering of this V-neuro ns; the next la rgest un its, those
quality (recogni tion) go on in the in the brisk-s ustained group, arc also
Convergence
nervous system? called X-neurons. (All the rest, both
We could discuss mtering networks concentric sluggish and nonconcentric,
for any sensory modality, but space does are coIiectively ca lled W.neurons, but
not permit a survey. Instead we will this term embrilces too heterogeneous a
concent rate on visua l filteri ng. Since set to be useful. )
notions of moda lity, submodality, For ma ny years the cat was descri bed
labeled lines, and tempora l coding have as havi ng esse ntially only two types of
been dealt with in Chapter 6, we can go optic nerve fibers, ON-center and OFF-
directly to a consideration of relatively center. In contrast, the rabbit, ground
complex network functions. squirrel, and gray sq uirrel were de-
The histological structure, cell types, scribed as more frog-l ike in having
and connectivity of the vertebrate retina several com mon nonconcenl ric and more
have already been descri bed (pp. 121- complex types-for example, those pre-
126), together with some of the electro- ferring movement, some specillc for
physiological properties expressi ng the direction of movement, some even for
coupling fun ctions of the connections. orientation. It now appea rs that the cal
At least thirteen fun ctiona lly distinct has about the same v.uiely of Iloncon-
types of optic nerve IIbers carry infor- centric units, including th ese complex
mation to Ihe brain from the eye in the types, but in much small er propo rtion,
ca t. Ta ble 7.1 gives ,I recent classificat ion. 8% compared 1034% in the rllbbi t. In the
Note that 92% have exci tatory receptive cal, "loca l edge detectors" are fi ve ti mes
fie lds (ERF's) that are concentrically as com mon as "direction selective" units;
organized. These arc either brisk or in the rabbit the lalter predominate,
sluggish, referring to the promptness and except in the viSual st reak (the equiva-
vigor of their responses, and either lent of the are.' centralis, for high resolu -
tr.lnsienl or sus tai ned, referri ng to thei r tion) .
41
250
All of these optic nerve fi be r Iypes types are known o nly in parI. In the ca l,
Clupler 1
Intcgr.ltio n at the I nter l11ed l~ 11!' leve ls bespeak transa ctions tha t process the X- a nd V-fibers go ma inly to the dors.l l
information encoded by rods and cones. late ral geniculate, e.lch to its priVate class
La tcra l interact ions, convergence, and of geniculate ne urons; these in turn
highly specified connect ivity (see project 10 the vi sual cortex, X-fibers to
Chapler 3) among receptor, horizontal, the so callea simple cell s and Y-fibers to
bipola r, amacrine, and ga ng lion cells are the complex cells. W-fi bcrs go to the
a ll indicated. This is usua lly referred to midbr.l in, largely to the tectum. It sho uld
as early extraction of fea tures to dis- be pointed ou t here that the re are a t least
tinguish it from further changes in the s ix cent ral targets of optic nerve fibers, of
mea ning of cell discha rge at laler stages whi ch the dorsa l later.ll genicu la te and
in the visual pathways. It also evidences th e tectum are on ly two. Evidence sug-
paralle l processing for centra l destina- gests that they have d ifferent functions.
tions of q uite differe nt fu nction. The T he retinas of frogs. lizards. and
c>l1lr.11 targets of these o ptic nerve fiber pigeons have even fewer concentrica lly
Tabte 7. 1
Recepti ve fi eld types of 960 cal relina l ganglion cell s, Latencies are the ranges of a nti-
dromic conduc tio n ti mes (rom the optic tract stimulus site, and can be t,lke n as pro-
portional 10 Ihe reciproca l of axon diameter.
42
251
organized ganglion cells and many more with a n OFF-sensitive surrounding ex-
Seclion III
movement-specific cells. This is prob- citatory field, as in the cat; they are more Analysis of Sensory Input:
ably not a phylogenetic trend; it may sensitive to blue light than to white or I'arallel and Series I'rocessi ng
have some relation to habit of life and other colors.
the roles of vision. frogs are known best Types 1 and 2 are fine, unmyelinated
from ~onee ring work by ~et vi n, axons in the optic nerve and by far the
Maturana, NrccU11iJcI~s, who most numerous. Types 3, 4, and 5 are
SeCI-Aim:u! h5f more Itah.lf<frk1ilOs than myelinated. Types Ito 4 go to the optic
flashes of diffuse fields or focused tectum of the mesencephalon, type 5 to
beams, and from the quantitative work the dorsal [ateral geniculate of the
of the Griissers (Fig. 7.8). In contrast to diencephalon. Histo[ogica [ types of
the cat, in which all optic nerve fibers are ganglion cells with distinctive dendrite
myelinated, the vast majority in frogs are branching patterns are probably asso-
very fine and unmyelinated. The five ciated with these fiber types (see
types of optic nerve afferenls may be Fig. 3.4).
summarized in the following way. (a) Types 1 and 2 do not respond to gen-
Type 1 fibers, called "susta ined-edge eral illumination, and their responses to
detectors," respond to a sharply focused l movement are not influenced by the
edge of an ob ject, light or dark, moving I [evel of ill um ination over a range of
or recently having moved into the 1 -30l intensity of mOTe than a hundredfold.
ERF. (b) Type 2 fibers, called "convex- Given their req uirements they di scharge
edge detectors" (Fig. 7.9) respond only to at a rate tha t encodes contrast of the
a sma ll object, darker than the back- moving object (but not light [evel) and
ground, that moves into or has recently rate of motion of the object minus that of
moved into the field; they must detect background objects in the IRF (but not
not only the change of position in the ca. relative motion). The discharge is
3 ERF but also tha t there is little or no ambiguous for certain severely con-
change of position in the ca. 15 sur- stra ined combinations of object size,
rounding inhibitory receptive field (IRF). contrast, speed, and amplitude of move-
(c) Types 3 fibers, called "changing con-I ment.
trast detectors:' respond to any edges in Change of position is as good a stim-
motion, large 01' small, dark-an-light, or ulus .15 visible movement; th us a good
light-an-dark, if the contrast and rate are res ponse is elicited by briefly illuminat-
adequate and the edge is not too fuzzy. ing a sta tic scene (no response) and th en,
These fibers give a weak response to during the dark period between flashes,
nonmoving tempor.l[ changes in light; moving an object within the ERF. The
they aTe classical ON-OFF units, but "memory" of the position of objects
much prefer motion. (d) Type 4 fibers, during the first flash lasts through at
c.llled "dimming detectors" respond to least one second of darkness (Fig. 7.10).
any dimming or darkening, whether An additional and remarkable property
caused by motion or not; these are OFF of type 2 units is erasability. When such
fibers but not like the concentric OFF- a unit fires upon motion of a sufficiently
center units of the C.1t. (e) Type 5 fibers convex edge toward the ERF center, it
respond to any brightening; these are ON continues to fire at a [ower rate for some
fibers, but not concentrically organized, seconds after the motion stops; but it
43
252
_ Excitatory synapses
Fi g ure 7.S
Types of ganglion cells and their optic nerve fibers in the frog. Above. The four types found in the
optk tectum, distinguished by thei r responses to four kinds of tests. ER f, excitatory receptive field;
s.g.r., stra tum griscum ce ntrdle of the tectum; . g.s., stratum griseum superficiale. Below. Diagram of
the connections between the elements of the retina converging on a ganglion cell; presumably differ-
ences in the details of these connect ion s and their transfer functions account for the types. II. ama-
crine cells; B, bipolar cclls; G, ganglion cells; fi , horizontal cells; R, receptor cells. [G rilsser and
GUsseT-Cornehls, 1972.]
44
253
Type 1
tlbor
Flash Flash
No
Light 0 response
inion sily
8
or to illuminating or moving
a natural-looking image.
-'
~
8 Stationary obloct
even if th e illumination is
1p::..J '~. 8
very taint. but
Change in position
during dark
Figure 7.10
"Movement neurons" mdY respond to chJngc of
pos iti on, with ~ forgetting time.
or to reversed contrast.
Figure 1.11
Abstrdctlon e.. rly in ,m .1(ferellt pJthway. 1m
pulses recorded from .. frog oplk nerve fibe r
in the roof of the midbr.. in. [Based on d .....
of M"t ur"nil al aI. , 1960. )
45
254
promptly (eases if the light is turned off tectu m. We may speculate that the phase
Chapter 7
Integr~lion at the Intermediate levels
briefly, and does not resume after the locking of its circadi,'11 rhythm with the
light is back on. environment.,1 photoperiod depends on
Notice that units of type 2 respond to a hypoth.1lamic center, the suprach ias-
any sma ll, dark, moving object, espe- matic nucleus, .'s has been shown in rats.
(i,llly if the motion is jerky. A hungry Each of these four central structures re-
frog will jump at .my such object. ceives its own optic input, consisting
Normally any object having these char- la rgely of a distinct mix of the ganglion
acteristics in the environment of a frog cell types.
will be a bug, an edible object. Since As activity proceeds through first-,
there are many detectors of type 2 spread second-, third-, and nth-order neurons in
over the visual field, the activity of one a sensory pathway, the me.1ning of the
or morc fibers of that type can both impulse activity changes. We have been
signal the presence of a bug and give its considering differences in meaning in
spatia l coordinates, hence both activate parallel neurona l pathways as a result of
and steer a jump and a tongue flick. divergence; we should note the differ-
Tadpoles may lack some of the Iypes ences in meaning in successive neurons
more important for the metamorphosed, as a result of convergence. When the
hunting frog. Type 4 units in the frog's criteria for firing include .1 significant
optic nerve signal general dimming, fraction of the complex features of .,
perhaps the approach of a predator or stimulus that releases norma l behavior,
any large object. we m.,y speak of recognition cells; th is is
The known visually oriented behavior a matter of degree. Higher-order ce ll s
of frogs in a natural environment in~ may add dimensions to the criteria, such
cludes finding food and avoiding ob- as novelty or familiarity. A novelty un it
stacles and large, threatening objects. deep in the frog tectum may fire in
The fi ltering processes that occur in the response to a small, dark, moving object
retina in frogs can abstract the relevant anywhere in a large (30) field, but will
aspects of the whole visual input signal soon cease, only to resume if th e same or
for those behaviors. Studies by Ingle and another object wiggles in a fresh part of
by Ewert, us ing single-unit recording, the field. A familiarity unit fires in
ablation, and stimulation suggest that response to a similar object, but, instead
central processing builds on the retinal of ceasing, continues, even maintaining a
filtering by separately analyzing optic low rate of discharge ("muttering") for
input for different behavioral meanings many seconds if the object stops moving.
in different structures. The frog's attrac- It now ignores fresh, moving objects
tion to blue depends on the dorsal within its la rge (30) field. It will fla re up
geniculate; its avoidance of large ob- if "its" object s lowly moves about within
stacles in jumping and its retreat from Ihe field, but will lose it and go silent if
large threatening objects may depend on the object jumps too far-to a fresh part
distinct, overlapping parts of the pretec- of the field!
tum, and its approach to food upon the In the auditory sphere too, we know of
46
255
units of a wide r,lllge of co mplexity of
, d,
Secllo n IV
criteri;!., In bats, for ins tance, a series is
Elemen' MY Ne uronal Nelworks:
found leading to ce lls that do not fire in Eme rgen t P,o!,(, rties of Ci rcuitry
rcsponse to any pure tone at any
intensity but on ly to frcq uc ncy-
modulated to nes with a certain range,
rate, a l.l d direct ion of modulation, like "
f/ ., ga
the bat's echo-ranging cry. In squ irrel
monkeys, cells arc found that s tro ngly
pre fer a certain one out of some 20 Figure 1.11
Efferent fibers ' 0 <l sense o rgd n. Axons from the
tape-recorded sounds chosen from the
brain to the ret in~ in the pigeon. Golgi prepdfa-
35 or so natural vocalizations in the tion. ig, displaced gang lion cell; f, n~t ~macr ine
species' reperto ire. cell; gil, ganglion celt I~yer; ir, inner nurleM
Some workers distinguish between I~ yer; ip, inn er plexiform I~ yer; 5, Snl,l 11 pM,Isol
recognition of natural sti muli and feature amacrine cell. [Maturana and Frenk, 1965.J
extraction; thc form er is potentially more
complex a nd may result from con-
vergence of cells of th e laller type. The fibers to m uscle receptor orga ns in cray-
term "fea ture" in this context means fish are illustrated in Fig ure 2.75. The
lim ited aspects of a complete natural y-effcrents to muscl e spindles a rc trea ted
I
stimulus, such as duratio n or frequen cy below (p. 267 el seq. ). T he mammalian
modulation . Our information is still too coch lea (Fig. 2.80,C) and the avian retina
lim ited to decide whcthcr some sub- (Fig. 7.11) arc also well-known exam ples.
systems in some a nimals work differ- but the fu nctio nal Significance of the
ently from o thers in a fundamental effercnts is no l yel adeq uately under-
sense. It is oft e n s upposed, fo r example, stood.
that some subsystems fu nnel s uccessive
featu re detectors down to a s ingle recog-
nition unit, like a " bug detector," IV. ElEMENTARY NEURONAL
whereas o thers never quite converge the NETWORKS, EMERGENT
set of re leva nt feat ure detccto rs. We d is- PROPERTIES OF CIRCU ITRY
cussed the samc problc m from ano ther
di rection o n pp. 238-240. By whatever In Chapter 3 we introduced some well-
means, complex natura l stimulus recog- stud ied examples of connectivity and
niti on mu st occur widely in nervous sys ccrtain general prin ciples of circuits of
tems-sometim es early in the pathw.1YS, neuron-like units. Here we extend the
sometim es late. di scussion to emphas ize th e physio log-
M odulation of input by ce ntral in- ica l consequences. Three kinds of ,Hrays
fluence via centrifugal (efferent) fibers is will be chosen; these may be c.lllcd
a potentia lly important part of the active " networks," fo llowing th e usage in th e
fil tering in many sense o rga ns, as was literature o n neural mode ling, me,m ing
noted on pages 170 and 225. Inh ibito ry any assemblage of con nected ne uro ns .
47
256
A. Mutually Exdtatory or also receive less excitatory Feed back. Just
Ch3pter 7
Positive-feedback Networks as the whole network was able to rUIl
Integration at the Intermediate levels
away to maximum activity, it now ru ns
If two or morc neurons are capable of away negatively to a min imum state.
exciti ng each other, then input tha t ex- Once the adaptation or fa tigue has worn
ceeds th e threshold of one will likewise a way, a n~ cycle C,lll begin. Networks
excite the others (Fig. 7.12A), Should the of cells, each of which may not be capa-
others exceed threshold as it result, they ble of rhyth mic bursts of activi ty, can
feed excitation back to th e first, a nd a produce bursts of more-or-less syn-
runaway process ensues. The whole chronous activity in all mem bers.
Inetwork may come into a state of max- Networks of this kind have been
imum activity a nd, without lim iting demonst rated in severa l cases and may
/ processes, might stay in that condition. be widespread. Inspiratory interneurons
Most neurons have relatively long-term in the medulla are probably so con-
sel f-inhibitory processes, such as adapta- nected, perhaps leading to their rhythm ic
tion, accommodation, or fati gue. As the bursting. Certain cells in the brai n in
neurons in the positive-fe ed back net- several gastropod molluscs have been
work begin to fatigue, one or more of found to be positively coupled. They
them will decrease in frequency. They produce synchronous rhythm ic bursts ')
then excite the others less, and hence that act as triggers in the control of
feeding and other activities. D eca pod
crustacean hearts al'e controlled by t,ri\
ga nglia containing only nine cells. -; ~ ~
baweerr- some of these is known at
least to aid in the build up of the heart-
beat.
B, Mutually Inhibitory or
Negative-feedback Networks
48
257
Vest>bular l'Iuclel
VIii n(trve roolS Secllon tV
Eleillelltuy Nr urorrdl Ne tworks:
Emergent Properlie5 of CIrc uit ry
Figure 7.13
A reciproedlly inh ibi ting pair of neuron s. The gi.1l1 t cells of Mduthner in the medull~ 0' nsh, sc he
'll~ticdlly shown, with the inhib itory colldlerdl (/I) Indicd tcd only on the left, the indirect VIII nerve
afferenls (8) only on the lefl, ~rrd the dired VJJJ netve dffercrrts (e) CO.lllrlg orrly from the right,
~lthough .III the5e components ~re ru lly bildleul. I Rct~l.lff dnd Fontdlne, 1960. 1
the Cdse of the paired giant Mauthner's den "jump" of some fish when the
fibers of teleost fish and tailed amphibia aquari um glass is struck. The vibrations
(Fig. 7.13). T he cell bodies and large excite the sensory endings in the vest ib-
dendrites ofl hese neurons are situated in ular apparatus of the ear. Although these
the medulla, where they collect input, cells have the largest axons in the body,
including particu larly Ihat fro m Ihe and although each has thousands of
eighth crania l nerve (sec also Fig. 2.17), input terminals and must be often bom-
The axons cross over before descending barded with very many impulses per
the spinal cord to synapse on the motor second, they produce on ly the occasional
neurons of the longi tudinal musculature. output necessary for start le reactions.
Each axon has a bra nch that ends in an Another system of one-to-one alterna-
inhili,ilQry: 5 na se on til contra I ral tion is found in the neurons driving the
Maulhner's cell axon hil , nca r the tymbal muscles in some cicad.l s. A pace-
spike- initiating site. (This is an electricill maker interneuron firing about 200 im-
synapse. ) An impulse in one cell pre- pul ses per second drives two mola l'
vents a simu ltaneous one in the other. neurons, which each fire at halF this rale
Each descending impulse activates, in exact alternation and precisely phased
nearly synchronously, an extensive lon- with respect 10 the pacemaker. The
gitudinal body wa ll musculature on one alternating clicks of the two tymba ls
side, causing a twi tch-li ke curvature of during song double the sound frequency
the posterior body region or tail. This poss ible if they were synchronous.
startle reaction begins the familiar, sud- Reciprocal inhibition Cdn lead to
49
,
258
C hapter 1
Thresho ld 20 mV I
[tlt cgr~tlon at th e [nlermedl.t!e le ... el~
O>---~
'--IlIII----!IIIUIIIIII- lIlIlIlIllI--IIa8111f--IIIII-~IIII'BIIIIllHlIlIlI----III!IIIiIIIIIII--II~1
I -IIIllllIIII 1l1li
'-.
'- --+1111111111 1I11 1 1 111l---111111111 lI ;mllll- 11IIlI1I1I 1I1"1111U1l 111111111 1111 RIll--
3 I I I I I I
Single I.p.s. p. l a tephsse La ler
triggers paUern intens ifies pha se resets
c
Figu re 1.14
Rccip rOC.t1 il1hibilion .. nd pastinhibitory rebound provide ncx ible mechanisms (or ~cnerJling bursts,
A. POSlinhibitory rebound in a model neuro n. A neuron 11t.11 is not spont.lneously ,lcli"e receive s dn
inhibitory input and produces spike o utput by rebound . lJ. Two such neurons, reci procally in hibi.
tory. c;on giW! a long 5eries o f altern ating burst s to ~ brief input bArr~ge from ~ third neuron. C. A
single inp ut impu lse has d iffe rent effects .according to t he phase of the .altern ation when It url\"!'s.
[perke l and M ulloney, 1974.1
1
alternating bursts of act ivity in otherwise com mon neurona l property of post-
nonbursting cells. The networks shown inhibitory rebound (Fig. 7.14,A) is
in Figure 7.14, B and C cons ist of only invoked in this model, and causes bursts
two interconnected cells, but they could that suppress the ot her cell. As the re-
represen t two populat ions. Input to the bound burst in the first cell s lows down,
network is inhibitory in this exam ple, the second is dis inhibited. Released and
.1nd reaches only one of the cells. Th e rebound ing in its tum, the second fires
50
259
and inhi bits th e first. As the seco nd ce ll drastic increase of acti vity in one li ne
Seellon IV
s lows, th e fir st recovers, and the whole would ord ina rily be drowned in the Etrmt' nluy Ne uron,,] Ndworks:
cycle repc.l ts. background activity of the ot he rs; but EIlIt'rgt'nl t'rop"'rlies of C ircuitry
Whether dependent on rebound or reciprocal inh ibi tion in the same cell
not, some such si mple mecha nis m for might all ow one line to dom inate if its
alterna ting burst acli vity, though difficult activity were to rise above some level,
to demonstrate, see ms to operate in and thus serve to d irect allention or to
f.wo rable mate rials, like the lobsler switch control. Still a nother possible use
stomatog.lstric ganglion. We believe it to of such circuits, with only slight cha nges
be an important mechanism for many in the coupling functions, might be as an
kinds of rhythmic and alternati ng be- ala rm system. Time- a nd load-sharing /,1
havior. Ins piratory and expi ratory inter- delay lines. null detection, and filt ering !
neurons in the mam malia n medulla are othe r theo retically availa ble con- I
probably inhibit each other. Locomotory seq uences of such ne tworks.
syste ms in ma ny a nimals may involve An array of inhibitory cross-con-
r~i p rocal inhibition between pace- nections can be thought of as a n arelM
makers for antagon istic muscle sets. in which there is competiti on be-
Ikciproca ll y inhi biting networks can tween par.ll1el streams of impulses.
pe rform a variety of fu nct ions, according Both the nonlinearity of dependence on
to the pa rt icular trans fer functions of the act ivity levels a nd the cri tical inflections
' synapses and the input and output con- in the input -output functions wou ld
nections. T he following examples are all ow one strea m to win control. A
theoretica l and q ualita tive; proof that spectrum of properties is possible from
reciprocal inhi bition is the mech.1 nism democratic to oligarchical to dictatorial.
that operates in living an imals is incom- The reliability of performance of
plete. G ive n certain properties, recipro- networks is considered on p. 233.
cally inhibiting networks can act as
gates, switching ra pid ly fro m control of
the output by one input line to another. C. La teral Inhibition Networks
This can recur at a steady r.lIe, for ,1
steady-s tate input, providing a pace- Th is term differs from the precedi ng
maker in which no s ingle cell is the head ing in directing a tte ntion more to
essential clement, as discussed above. l'l yers or arrays than to alternate cells or
With cert ain dynam ic properties the groups. Many neural tissues consist of
same simple circuit C,ln act as a n layers of similar neurons that inh ibit
intensity-lo-Ii me con verier that m,l Y be e.lch other eithe r direct ly or ind irect ly.
useful in comparing the strength of two The inhibitory connections spread from
inputs by their relative dura tion of con- a ny part icular cell to make conta ct wit h
trol of some downstre,lm system. Slight neighboring and more distant members
changes could provide sensory sca nning of the layer, but wilh decreas ing de ns ity,
by periodically or irregularly sampling aSwe saw in Chapter 3 (p. 11 1). Hence the
e,lch of a number of input li nes a nd effect iveness of inhibition decreases with
giving them control, in turn, of some dista nce.
later elements. In a cell w ith a large Some effects of latera l inhibition a re
number of input li nes converging on it, a ill ustrated schematically in Figure 7.15.
51
,
260
-
Inhibitory sensory nOl,l ro ns
Interneu ron s
5 /0
O..;'-------~,,?--~J:0"''----
0..,/
+ $ -
,s'---------<<E;"-
0'~ - a.."~" s
__(0,:'-----
'"
o,:s'--------~C~"-'-<'~-(o,:S'--C+~1
"Yo"v/'"
~ +"E'--_ _ _ _~O"-'-'-"J
O,S'-" - S + E+ I
(0':'-'--"-''-'
;<,',).
0;s'-+:':.!E'--_ _ _ _~<::~"~,-,-:'~,1os + E + 21
O"S'-'+"E"--____~cr:c"-'-'-'~(OS + E + 21
0;;;;:0..,,/"
"S'-+:':.!E'---_ _ _ _~C:"-'-
0- ~ ''--- ', (O~'-"'-'--"
5 + E + 21
Re ceptors 0 Second-order
Inhibitory sensory neurons
Interneurons
" Darker" "Brighter"
~----.
Stimuh.l! Physiological
in tensity ellect
A B c o
1
52
261
Flgu.~ 1. IS I/llriHg r~SO')
0'
l .ltl'r.ll in hibillon. A. A p.lUl'rn unifolmly gr.y ,"r.lS with shup f>dgn. rrpr~rnlinR .I "isu.1 fit'ld
_n by .I n eye. (M.lch b.nds .In illusion in ",hlch .Ipp.onmtly d.lrk er .nd lightl'r lunds .Irt' _ n El tm~nl.lry
51"<1Ion IV
Neuronal Ntlworks:
.I round t'.tch t'dgt', Mt' not evidl'nt to us in this conrrguration.) B. The sllmulus ploul'd <IS inlensity Emt'rgtnt Propt'Tl ies of Circu itry
(horizontlll) .Ivinsl lpo1 lial e>:tt'n t (,-ert ic", IL emphllsizi ng the unifonn ily of the pnysiclll intt'Mily
wit hin ('IIC"h IIrt'll. C. A network of rtc~ptOfS ",nd S('cond ~rd('r neurons with rt'ciprocal Inhibitory
C"o nntelions vi. inll'rneurons (brokr n linrs). Spont.l nrous act i\~ l y in the replors (5) is .I ugmenlOO by
ucildtion (E ) d ue 10 light. Thl' network CdU$e5 Ihe sl"<ond ~rder nl'u.o ns 10 show 5 dC"ti vity dU g-
1111'ntl'd by E . nd/ or rt'd ucl'd by inhibition (I ) in single or doubll' (11) dose. D. Th(' output, ('(j Ui"d'
len l to our $ensdlion, piatti'd .IS dMker or ligh ter th.n tht' bolCkground dut' to S.
First consider Ihe behavior of one spon- stimulus edge. In terms of the fir ing
taneously active cell in the network frequencies or Ihe cells in the network,
while a stimu lus is moved about its input the stimulus edge has been enhanced;
field . As the stimulus approa ches, it first there has been a spatial di ffere ntiation of
excites neighbors of the recorded ce ll. the input signal. A second layer of
Si nce they inh ibit Ihat cell, it res ponds neurons cou ld be so constructed tha t it
by a decrease in firing frequency. If the lVould detect the edge only. Latera l in-
stimulus passes directly over the re- hibition probably explains our psycho-
corded cell, it is excited 10 fire above its physical illusion known as "Mach
normal ra te, but as the stimulus moves bands" (Fig. 7.15,0), and perhaps oper-
on il is again depressed by ils neighbors. ates widely to en hance the sensitivity to
Any cell in Ihe network may be char- cont rasts.
acterized as having a recept ive fi eld that Since the lateral spredd and ex trd
has an excita tory center and an annular synapse take time, there is a delay in the
inhibitory s urround. Th is is compa r.lble inhibi tion. Th is confers a tem poral prop-
to the ON-center ganglion cells of the cat erty of freq uency se lect ion or fil tering
ret ina. that acts to attenu.l te ra pid changes as a
If, ins te.ld or looking at the response function of dista nce rrom the center and
made by a single cell in the s patial array to prolong the exaggerat ion of the pri-
or Figure 7.15 to a moving sti mulus, we mary response at the center, again fa vor-
examine the ou tput of a whole line of ing low frequenci es.
cells in the array while one hal f of the It Cdn also be seen that the inhibition
array is stimulated more than the other exerted on a given cell by those sur-
half, lVe see an abstracting function of rounding it can be redu ccd by stilllulation
the network. Cells in either uniformly of a slightly more distant popu lation.
stimula ted half of the fi eld all in hi bit T he di stant popu latio n red uces the lcvel
each other symmetrically, but th ose at of activity of the nearer one, wh ich
the edge do not. Ce lls on the strongly "d isi nhibils" the center.
stim ulated side of the edge are inhibited These phenomena were first described
wcak ly by their nCighbors across the by Hartline and his .lssocia tes in a se ries
edge whi le they strongly inhibit those of elegant ex periments on the compound
!lame neighbors. T he result is especially eye of Lil/IIIIIIs. The same latera l inter-
high .llld low firi ng frequencies CI t the action exist, however, in virtua lly "II
53
,
262
sensory syslems in both vertebra tes a nd known <l nd presu mably main ly che mica l.
Chapter 1
Integr.lHo n at the Inlerm~d l~te l eve ls invertebrates and may o peT,l le at the All the fu nctiona lly established connec-
earliest stage in the pathw,lY a nd / or at tions can be an.l torn ica ll y justified by
later stages, even in the cortex (Fig. 7.16). fibers vi sualized by inj ection of Procion
We h,wc seen in Chapter 3 how lateral yell ow. How mu ch spontaneity there is
inh ibition might result from recurrent cannot be sf'ated, but it must be con-
collalera ls of the axon e nding on neigh- s ide rable. Among these junctions, .,
bori ng cells (Fig. 7.17). We saw holV such v,uiety of integrative input-output prop-
inhibition ca n be complicated in the erties are found. It seems like ly tha t if we
cerebellar cortex by th e inhibitory influ- could unrave l the web of influences,
ence of recurrent coJlalerals of Purkinje there wou ld be bot h excitatory a nd in-
cell axons, not only on neighboring hibitory reciprocity, latera l effects, a nd
Purkin je cells bu t on basket cells, mixtures of spontaneous rhythmicity
thereby disinhi bit ing Purki njes in a cer- with imposed bu rst-sha ping effects. The
tain geome tric pattern. Re nshaw cell norlllal activity is largely a rhythm ic
laler,l] inhibition via molor neu ron re- se ries of bu rsts of repeatable but la bile
curren t collaleraJs is trea ted on page 269 s pike pattern; the known connections
(Fig. 7.24). ex plain much of the detail of the bursts.
Alth ough many of the connections of
this network are known (nearly all;
D. Mixed Networks hence far more in proportion than for
a ny ot he r known system of some com-
If both excitatory and in hibitory connec- plexit y), it is difficult to assign causes to
tions exist in homogeneously in a set of th e bursting phenomenon itself. Does
cell s, the variety of possible outputs th e pos itive feedba ck, by itself, cause
expand s to the degree that it is useless to one group of cell s to burst, or is the
make (l priori or genera lized, uncon- reci proc.ll inhibition relationship with
strained models. A recently studied real a not her group necessary, or even s uffi-
exa mple is, however, worth describing cient? In theory, ei ther mechanism could
(Fig. 7.18). T he stomatogastric ganglion prod uce bursting, but physiological evi-
of crustacea ns controls the stomach, one dence suggests that both kinds of con-
part of which contains the gastri c mill nect ions exisl. Perha ps both mechanisms
used to grind food. The gangli on con- oper.lle synergistica lly to make th e
ta ins 30 cells, almost all of which a re whole system more stable.
identifiabl e ,md consta nt in con nections
a nd influe nce. Te n a re motor neurons to
the g.lstric mill part and 14 to the pyloric E. Connections Ensuring
part of the stomach, but both groups also Synchrony of Activity
have direct infl uence upon each other.
Furthermore, two interneurons are pres- Although exactly synchronous activity of
ent with connections to both sets of neu rons is not known to be req ui red in
motor neurons. There are 123 known m.1 ny insta nces, it is importa nt in a few,
inhibitory connections and only 6 ex- and these a re interesti ng in s howing the
ci tatory junctions. Twenty-nine of the fl ex ibility of design with whi ch cells ca n
junctions are electrotonic, the rest un- be connected, among th emselves .l nd to
54
",g ill
263
0
:g 20
.,.30
'~ 40
- 20 "'.
~ SO 20 30 50 100 -"
i ~~1iI ,, - 60
Lower
- 80 aUditory
30
~ 40 m ce nter
.~ - '00
50 20 30 50 100 oE
A Vi b rat ion freq uency (Hz) ."
,-
~
0.2 0.5 2 5 10 20
0> 0
-
00_
.. t
< !
Fig u re 7.16
Sh~ rjX'n i ng by suppressing sensitivity o n e.lch
<
~
- 20
-" V
side of th e best frequency, with converging
inpu t. A. Th resholds for sensMion as the e nd -
point; vibr<ltion felt by 2 or 3 fi ngers. 8. Th resh-
olds for si ngle neuron firi ng as the end poi nt;
un it respo nses to sou nd at lower and higher ~ udi
- 60
- 80
Higher
audito ry
cen ter
..\1
\
H
tory centers. It should be noted t h ~ 1 such narrow - 100
cu rves at h ighe r cen ters are no t common; units
0.2 0.5 2 5 10 20
wi th m.my types of cu rves are found. {Von
Ilckesy, 1967.)
B Sou nd Irequency (k Hz)
~i
f igure 7.11
Recun en! (011,,11'. ,,15, The li ne fibers are the array of colliller.. ls of th ree pyra middl
cells in the corte .. of the killen. IScheibel and Scheibel. 1970d.)
55
264
"" ~~~
LGN
ifjirgii"1'. k':: :: . -".,If:;'; 'I:;:: }ff
j ' " . ," "' " .
PNI ~ I , I I , I , , I I I ! , , I I I I I I I I I I I , I
d-l VN : ' . . 'r .:!",': . .. ... ..... I.'
!. ' ~,. - .~
"
GASTRIC MILL
'"
REGION
~ ____ r,
: E I
I i ibe.s I
: (2) leG's
of 0
o
.... n ___ J0
PYLORIC REGION
A
Inl 1
gm l .2,3a gm 4 c7
56
265
the periphery, to ,lccomplish precise com n1.1nd or p.l ccmaker neurons ,1rC
Sectio n IV
ti ming. These can be rcg,lrded as s pecia l usually located in the medull'l 01' mid- Elfnlent.lrY Neurona l Ne two rks:
cases of the gene r,\l problem of ass uring brain and act th rough interne urons and Emergent Pro perties of Circ uitry
precise lim ing of sequences of activ ity. motor neurons on the electrocytes. In all
Among the best exa mples of systems C,lSes, these neurons are electrica lly
requ iring nearly exact synchrony are the coupled to each other via gap junctions
electric organ discharges of electric fi sh. (see p. 333). The electrotonic connections
(Ot her exa mples are the oculomotor may be between celi bod ies, or den-
neurons and the neurons innervating drites, or via presyn,l ptic fibers that form
sound-producing muscles and wings in electrical synapses on Ill.lny of the cells.
certa in ani ma ls.) The electric organ celis T hese synapses, by equa lizing the level
(electrocytes) are in most cases com- of depolarization between different celi s,
posed of modified muscle cells thai are are both excitatory and inh ibitory, but
oriented in seri es so th,ll Iheir depola ri- ensure synchronous acti vity of all cou-
zalions ca n be sum med. It is necessary in pled cells. The simplest exampl e of such
Ihese orga ns Iha t all the elect rocytes be coupling is in th e electric catfish,
activated nearly sim uli,lneously, and in Mn/npierllrJIs, in which there are two
some species wit hin 0: 1 Insec. Th e electromotor neurons, one on ei th er s ide
57
266
of the /lrst spina l segment, lightly sponse to IMturaJ stimuli . What can we
Clupl t'r 7
Integrdtlon .at Ihe Inlerp.edi.alt' uvels
coupled to each other. In ot her fi sh, the re say is the s implest ne rvously mediated
may be as many as 20- 50 p.lccma hr response?
cells coupled to each other, all of which The verlebrate stretch reflex, while
electrically excite internuncial relay cells specialized for simplicity and speed
that are also electrotonica lly coupled. rather thaI!, be ing primitive, is a good
G iven a synchron ized command, how starting poi nt because it is mono-
aTC electrocytes at di fferen t distances synaptic-that is, no interneurons a rc
from the brain activated synchronously? interpolated between afferent fib ers and
Thi s is accomplished, in most cases, motor neurons (Fig. 7.19). Skeletal mus-
either by systematically shortening the cles contain numerous sense organs,
lengt h of branches to the progressively called muscle s pind les, that a re se nsitive
mOTe dista nt electrocyles, or by altering to s tre tch in the ax is of the muscle fibers
their diameter so that the s lower con- (Fig. 7.20). Passive stretch, as by the
duction velocities ofaxons inne rva ting action of gravity or of other muscles,
the nearer electrocytes compe nsa te for causes a train of im pulses to arise in the
Ihe shorter distance. In several s pecies term inals of a sensory axon in a muscle
the gra ded delay is built into th e electro- spindle, and these are conducted via the
cytes. Dista nt ones a re innervated nea r dorsal root to the spinal cord, the re 10 be
their principal surfa ce, nearer ones at the distributed in axon collaterals to the
end of long, slow-conducti ng stalks of dorsal and ventral horns of the same
the elect rocyte. segment on both s ides of the cord, 10
The ex istence of these mechanis ms for nearby segments up a';;'d down the cord,
building compensating delays into neu- a nd to the dorsa l colum ns ending in
ra l ci rcuits opens the poss ibility tha t nuclei in Ihe medulla. Of these destina-
similar refinements of st ructure may be tions one is the large alpha (a ) motor
involved in other systems that are sensi- neurons of Ih e sa me muscle, wh ich a re
tive to the precise tim ing of inputs, such excited. This excitation is distribut ed by
as the parts of the auditory system re- the axon branches of the motor neuron
sponsible for sound localization or the to a group of mu scle fibers, usually
motor systems controll ing speech, eye between WO and 1000, called a molor
movements, midd le ear muscles, and the unit. Conlrdction of the molar unit tends
li ke. to ca ncel the s tretch. This proprioceptive
refl ex acts as a tonic muscle-length servo
(like a gyrocom pass), tending to main-
V. STIMULUS-TRIGGERED tain the length aga inst any change in
REACTIONS, THE load either way. Gravity is an import.lnt
ORGANIZATION OF norma l stimulu s; th is re Aex arc is the
REFLEX ES principal a nt igravity circuit.
AI th e sa me ti me, coll alcrals of the
With some of the principles of effector spindle afferents fil'c interneurons thai in
control. of sensory input, a nd of ele- turn exci te synergistic motor neurons
mentary circuits in mind, we can now a nd inhibit motor neurons of a ntag-
turn to the lowest levels of motor re- onistic muscles on the sa me side whi le
58
FIgure 7.20
267
Til, mll"multiaH /IlI/SlI, sp;/Jdl,.
1111
'-H -t-7 raxons
J
Molor
Type Aa
Nuclea r affe ren t
c hain axon
Figure! 7.19
fibe r
T he monosYn.lpli<: slrttch rrAu p.ilhlV"Y in IllAmm.. ts, si m pl ifi~ by
oml .. ing Ihr ol her dtslindlions of Ihr s.am(' ..ffrrr ni n('U ron, Ihr oIher Type All
Annulo al/eren!
inputs 10 Ihe s.amt mo lo r nturon, ils oulpul rt.'Currenl coll.. ler~ ls, .md splrlll aKon
Ihe molor conlrol of Ihe spindle!. IGlrd uer, 1963.[ endings
Nuclear bag
Flower -'.-,...,~ fiber
s pray
endings
motor neurons of the homologous mus-
cle and its synergists on the contralater.ll ..... Sill S' orgjHl ,x(;I,d by j1fl$S illt 51ft/ell , Exl ralusal
side are inhi bited and antagonists ex- 0' IIdi vII/ioli 01 ;/$ 01011 ;III,i,Jsir ml/Sllt skelelal
fibtr5 by ga"",111 m% r axo/IS. m uscle fiber
cited. The phenomenon of reciproca l
innervation contributes to coordination
by prevent ing antagonists from working
against e.lcn other. noncontractile enlclrgement filled with
Given this iength- maint.lining re Aex. nucle i, and it is to this zone that the
how ca n the organ is m walk or volun- stretch receptor fibe rs come (Fig. 7.20).
tari ly ch.l nge muscle length? If higher The end regions of the s pind le fibers are
cen ters were simply to comma nd "" - contractile and are innervated by the
motor neurons to greater or less activity. y-efferents. W hen the y-effe rents in-
the stretch renex wou ld quickly cancel crease their activity. the spindle fibers
the effect. Somehow. the set point shorten, the cen tral region is st retched,
(SoHwert) must be changed. This c.lIl be and the stretch receptor fi bers are ex-
done by adjusti ng tension in the muscle ci ted. It is as if the muscle had been
fi bers within the spi ndle. ca lled in tra- passively s tretched. T he opposite effects
fu sal fibers. obtain if the efferents have reduced
There is another set of motor neurons activity or the main muscle has increased
in th e s pinal eOI'd that also se nd th eir activity.
e(ferents to th e mu scle. These are the Th e set point of th e postur.ll renex ca n
gamma (y) effcrents. emana ting fro m be affected by tens ion changes in the
small motor neurons (Fig. 7.21). T hey spi ndle muscle fi bers. When these con-
innerv.lte the muscle fi bers in the muscle tract they stretch the muscle receptors
spindle and alter its sensitivity. In the but do not change the lengt h of the
centr.ll region of the spindle fi bers is a muscle. T he resulting increase in stretch
59
>68
From brain
Chapter 7 2
Integr~tion at the Interm~dlale levels Spinal
cord I
axon
Muscle
tiber
Figure 7.21
The yloop servomechanism. Commands (rom the brain may operate by contracting the mu~le fibers
of the spindle via the y-efferents in th e venlral root (2). This excites the stretch reflex (3) Jnd hence
the main muscle. There are also direct conllections (not shown) from the brain to the Ill,,;n motor
neurons. [MNton, 1972. J
60
269
fun ction, restraining th e motor neuron
Seello ll V
from causi ng the muscle fibers to COll- Stimu lus-Triggered RNc tion5:
tract too violently. T he Organization of HeOu t'S
O ne more automatic s ubsystem is
important in hel ping to determine motor figure 7.23
Spimflt IIfrsrl5 I"rdorr 't(flrlor~.
neuron activity loca ll y; this is thc
j Pll llorr , 1965.J
Rens haw cell nega tive-feedback loop,
acting on motor neurons in the s pi nal
cord (Fig. 7.24). Situated in the ventral
roots there are branches of a-motor
axons called recurrent collalera ls because
they turn back and reenler the ventra l
horn to synapse with small interneurolls,
Go igi
tendon
na med for Renshaw, who discovered
recepto r them physiologically. These cells fire at
high frequency when excited by motor
neuron output and have the effect of
Figure 7.22 inhibiting the same and neighboring
The tendon recep tor for st retch. Anothe r se nse
motor neurons. The roles of this inh ibi-
org~ n, besides the spindle, is in th e tendon,
which differs by being stretched (excited) when tion may include preventing motor neu-
the muscle con lr~cts as well .1$ when it is rons from excessive activity, focusing
lo ..ded. T he sign of its innucnce is such as to activity upon certain cells and perhaps
preven t overconl r.l(lion. su ppressing phasic responses more than
ton ic ones. Muscle con tracted
7.23) . These receptors (Colgi tendon motor neurons, but on interneurons; the
organ receptors) have no monosyna ptic reflex pathw.1Y is polysynaptic. The
endings, but, via inlerneurons, they send input excites flexor action and at the
(a) inh ibitory input to motor neurons of same tim e inhibits ext ensor motor neu-
the muscle they innerv.1 tc and to its rons. If the stimulu s is quite strong its Acceterated
discharge
synergists, (b) excita tory input to molar effects may s pread even to the conlra-
neurons of antagon ist muscles, and (c) lateral limb, where they are opposite in
opposite inputs to motor ncurons of the sign. The crossed extcns ion refle x pre- Musc te contrac' ed
correspond ing muscles on the opposite pares one limb 10 bear the extra weight TENDON RECEPTORS
side of the body. T hese inputs might be s hifted to it when the painfully sti m- ""IN SERIES'
said to perform a tension- regulating ulated one flexes . Similar refl ex con-
61
270
+eo
Motor neuron
+<0
0
- 40
- eo
,-
.s,
+ 80
'~
~n +<0
0
0
D
, E -40
- eo
- 120
Motor
~"'"W" -60~
-70 _ __ :_p_s_p,
-80 I a!
0 30
5
" msec
15 20 25
Figure 7.24
The Renshaw type of inhibitory interneuron. Axon collaterals from motor neurons activate
Renshaw cells to high frequency discharge, which set~ up summa ting inhibitory postsyfl<lp-
tic potentials in neighboring motor neurons of the same pool. [Eccles, 1964.)
ncctions are seen for cutaneous touch, Sherrington sta ted the concept of the
pressure, and temperature receptors. reflex, one of the truly great and fruitful
We have seen now some qui te genera l abstractions in biology, with these
contrasts between flexion ,lnd ex tension words: "The Ullil reae/iOIl hi l1erVOU5 integra-
reflexes, nociceptive .1nd proprioceptive, liOlI is the reflex. because every refl ex is an
pain and postural, cutaneous and integrative reaction and no nervous
muscle-afferent, phasic and Ionic re- action short of a reflex is a complete act
flexes . These are overlapping but not of integration" (1906, emphasis his).
synonymous d ichotomies. Anothet use- Many authors have pointed out the
ful one, referring to their roles in causing limitations of the concept or criticized it
normal behavior is the distinction be- as art ificia l, and Sherri ngton, as clearly as
tween elementa l and tuning reflexes; the anyone else, emphasized the non-
fonner cause the basic sequences, th e existence of a discrete circuit insu lated
laller adjust to momentary conditions. from others. Nevertheless, the funda-
62
271
Nociceptor the pi nna, swallowing, stepping, s neez-
/ in skin Sec tio n V
ing, sal iv.ltion, blink, accommodation, Stimu lus-Tri ggered RCd cllo ns,
and tonic neck refl exes, placi ng, hop- T he O rga nlutio n or Hrncxe5
ping, and righting reflexes. T hey arc
re.ldy- mad e, un leamed, adapti ve move-
ments, prompt and coordinated . The
coordination is not observed just withi n
each refl ex, which we might ex plai n as a
fi xed pattern, but is observed even when
conflicting refl exes are sti mulated simul-
taneously, for th ere is al most in va riably a
resolution of the pote nt ia lly maladaptive
conflict or intermed iate action in fa vor of
adaptive selection .lnlOng them. Cooper-
ative interplay is also marked between
sim ple segmental refl exes, long spinal
intersegmental refl exes of foreli mbs and
hind limbs, and coord ination of body,
Figure 7.25
li mbs, neck, hea d, and eye refl exes, as in
Nociceptors .. nd the polysy n.. ptic p.. thw.. y. An visually guided wa lk ing.
other !OCt of inputs impinging on th e motor neu- T hough the present account of re-
ron comc fro m pain and sirn ila r receptors. T he flexology necessa rily draws mainly from
sign of their aclion is generally excitatory to the ma mmalian literature because lower
ipsilate ral Aexor a nd con tra lateral exten so r mus
cles ~nd inhi bito ry to th e ips il ate ral extensors
forms have been less studied in this
.m d con tral ateral nexors. respect, we believe the princi ples
en unciated are probably gene ra I. Jm pl ici t
in the conce pt of refl exes-si nce there is
almost invariably more than one refl ex
mental usefu lness of recognizing this util izing a given muscle, and hence more
category of responses has been amply than one cent ra l mechanism converging
proved by th e ins ight that experi ments on the sa me motor neurons- is the con-
based upon it have provided. The refl ex cept of the fin al common path. The very
is a useful abstraction, but we qualify the ex press ion em phasizes the int egrative
bro.."ld sta tement that it is the unit of all function.
nervous integration: (a) the arousal func- The propert ies of refl exes (see Box 7.2)
tions of the reticular activa ting system in and the ru les by which they are used
ma mmals cannot be resolved into re- ma ke the best case for their reali ty and
fl exes, nor can (b) mere sensing, (c) importance. We have already learned
autochthonous action (arising from ma ny of these rules in the exam ples
within), or (d ) many instincts. detailed above. Let li S look furth er at the
Familiar examples of the phenomena ways they com bine.
under consideration are the stretch Separate reflexes may be either com-
refl ex, the fl exion, crossed extension, patible or incom patible. The fortner
sha ke and scratch (dog) refl exes, the combine adaptively into compound re~
refl exes of micturition and defecat ion, flexe s. For exa mple, a grav ity refl ex that
63
,
272
keeps a fish u prigh t adds to a ligh t reflex the nervous system d oes not release
Chap ter 7
that keeps the dorsal side toward the them both, One o r the other is sup-
In tegration at the Intermediate Levels
light, so that if light comes from the side, pressed, at each moment. Incom pa ti ble
cert.lin fish tilt to a degree graded reflexes do not add algebraically or com -
according to the ligh t intensity, the bine li nearly; instead, s witches or pat-
strength of gravity, and a central evalua- terned contro l ins ure normally adaptive
tion thai multiplies each in put by a interaction. This may show either in-
weight ing faclor. The weighting depends hibition or facilitat ion . It is as though the
on time of day, temperature, hunger, and system were preorganized for useful
other inputs, such as chemical signal s m ovements.
associated with food, m echa nica l inputs The tonic nec k reflexes and a number
that ind icate a 5ubslr,liul1l, and visual of related postural reflexes d ue to
input su fficiently imaged and "under- vestibular and proprioceptive input
stood" to represent a substratum. interact with each other and with phasic
Incompatible reflexes are those that movements as though add ing a bias o r
cannot be accomplished at the same "tuning" the res ponse to the conditions
time. If the hind leg reflex to scratch the o f th e m oment. For example, if a load is
back is elicited on one side in a dog and li fted by wrist flex ion, mo re wor k c.1I\ b e
at the same time the sti m ul us for an done (stretch reflexes faci lit.1ted) wi th
extensor thrust of the same sid e is given, the head bent down or turned away from
Box 7.2 Properties of Reflexes The ti ming of inhibition is coord inated with that of
excit<1tion, as is cleMly seen in the alternating reflexes,
WIMt are the properties of reflexes that mani fest s haking, stepping, .lnd scratching. Even sim ple flexion
integration? (a) The th reshold st imulus is very much and crossed extension reflexes show ch.lfacteri stic
de pendent on conditions. (b) Above the th reshold, tem poral patterning. (h) Irradiation wi th increasing
gradation of response does not closely correspond with intensity of stimulus occurs in some reflexes-for
grJd,ltion of st imulus, (c) If the stimulus is repetit ive, example, the protective flexion reflex. At th reshold <1
there is USUJlly a poor correspondence between its response lll<1y involve a lim ited part of a synergic
rh ythm and that of the reflex response. (d) Single muscle group <1cross one joint, but with irrJd i,ltion it
afferen t impulses are U511,11ly not adequate; temporJl may spre.ld to other joints of the same append,lge, to
summation is usually necessary to el icit a response. (e) other appendages and segmcnt.llievels, to the he,ld and
A depressed excitability typically follows a reflex and is neck. The spre<1d is generally salt,ltory and is confi ned
oft en qui te long. (f) Afterdisch.uge, or the prolong.dion strictly to certain Jines or muscle groups. However,
of the motor neuron activity after the cessation of the apparently uninvolved muscles m,lY in fact be involved
stimulus, is ,1 prom inent fea tu re of nl.lny reflexes, as as objects of inhibition. The possible movements Me
though the mecha nis m were org,lIlized to com plete a thus cirnllTIscribed in a chM<1cteristic p<1ltern. These
certai n movement in a controlled W<1y. (g) Spa ti.l lly and properties help to define reflexes, to bring out thei r
temporally patterned con trol of severa l muscles is integrative nature, and to emphasize th e centr.ll deter-
probably involved in .111 reflexes. mination of detai ls of form and timing.
64
273
tha t Mill, because of the tonic neck re- to exterlld l influen ce; a progr.1Il1 might
5~dlon VI
fle xes; if the Io.ld is met by wrist ex- have the same, but the te rm ,' pplies also C~nt '~ lI y 5c:ored 8~h~y lor:
tension, the opposite head movements to cases tha t merely ca ll for reactions, I'.! tln ning in 5p~c~ .and Time
e nh'lIlce ou tput. leaving the tim ing to effectors, trdns-
Refl exes Me normally \'/Oven into a n ducers, loops, .lnd largely peripher.l l
integra ted f" brie, without s harp li nes. events.
They Me gr"ded in amplitude by influ - Instead of a s imple d ichotomy we may
ences descend ing from highe r cente rs disti nguish several poss ible mecha nisms
and comb ined unde r the ru les of ,l (Fig. 7.26). In A, we have a s imple reflex, .
hie ra rchy. We ca nnot help wondering such as a n eye blink, a sWdllow, or a
whelher the sa me ci rcuits Ihal a re re- cough, triggered by a stimulus. Sla rtle
fl ex ly t riggered from Ihe periphery might responses media ted by gia nt ribers
be centrally triggered in patterned belong in this category (see Box 7.3).
sequences.
Hav ing erected this edifice of plausible
Even here the t('mpora l pa Hern 0
messages to various muscles is deter-
I
ass umptions and conclus ions from mined by central pathways and integra-
studies of reflexes, we should now raise tive junctions. In B, the re is sensory
the questi on as to wh.,t evidence th ere is feedback from proprioce ptors ea rly
for the centr"l origin of patterned im- enough to determin e a rh yt hm of recur-
pul se discharge. re nce; a chain refl ex aCCOli nts for fre-
quency, phasing, and amplitude. The
cxtreme case, in E, is purely centra lly
VI. CENTI!AllY SCORED timed, without any immediate feedback.
BEHAVIOR: PATTERNING IN C a nd 0 are combi ned mechanis ms in
SPACE AND TIME which the central s ponta neity ca n deter-
mine the bas ic rhythm but feed back may
One way to sta te Ihe fu nction of the alter either the rhythm (C) or only the
n('rvous system is that it formulales details of the expression of the rhythm
appropriately pa tte rned messages to (D).
drive the effectors. A core qu('stion in the Probably all five mechanis ms a re com-
study of intermedia te level integrAtion is: ilion, though us ually there has not been
" How is this doner' T he patterni ng in enough analysis to be s ure which class
tim(' may be treated as a more serious a n activity belongs to. In this section we
issue than tha i in space. Theoret ica lly it are concerned particul.uly with some
could a rise in either or both of two W.1YS: examples of C, 0 and E. In each
by followi ng (a) timing cues from exa mple, there is permiss ive or essentia l
peripheral se nsc organs or (b) liming input from sense org,lIlS that may start or
cues frOIll central pace makers or patt el'll stop the whole pattern or innu ence the
generators. We may call the second overall "central exci tatory state," to lise
mecha nism ,I central scor(' to empha size ShNrington's phrase. This is what lVe
its potential complexity of detail, its mean by spontan eit y, not that there is
mod ifiability from outs ide on any given independence of the envi ronment for
occas ion, and its reali ty as a slored permissive conditions, bu t only for trig-
progr.un dPdft from external inputs. A gering the s uccession of ,1(tions. S pon-
score hdS ti mi ng built in, though subject taneous rhyth ms ca n be in nue nced in
65
A B c o E
Fig ure 7.26
Five mechanisms of pattern fOTmulat ion. The three levels of neurons are understood to represent
brJnch ing cha ins in whose functions integrat ive p roperties may a lter the actua l impulses ,md distr ib_
ute them spat ially as well as temporilily to the effectors (bollo"I). A and B are shown with receptors;
C, 0, and E, with spontaneous p<lcema kers giving simple or group dischMges. Band C have proprio -
cept ive feedback acting on the trigger neuron; D, o nly on the shaping of the pattern. [Bullock, 1961a.)
Striking among the behaviors tha t are merely trigge red results in rapid, neJrly synchronous, widespread mus-
by environmenta l st imuli, but not further guided by cle act ivity. Behaviora lly, g iant fibers ordina ri ly fire
either environmental or propriocept ive inputs, arc the only with rather special input requirements; these often
responses med iated by giant fibers . Giant fibers are have the ch'Hacteristics we call startle.
Found in many invertebrate .m imals and fish (see The giant axons of sq uid arc the best known of all
ChJpter 10). T heir phyletiC d istri bution is scattered Jnd nerve fibers. We may briefly review their funct ional
their st ructures are diverse; hence they a rc likely ana tomy (see <1lso Ch,lpter 10, p. 434) and behaviora l
correl<1 ted only in func tion, not through evolutionJry or role. The muscles of the mantle of the squ id a re doubly
developmental relationships. BecJuse of this diversity, innerv,lted. Many small mo tor axons From the stellate
the follOWing generalizat ions l1lily have exceptions. g,l1lglion excite rela ti vely few muscle fibers each, and
Beciluse of their size they conduct impulses TJpid ly. these slnJIl neurons control the slower movements
Perhaps related to size is the fact that they can have ,1 involved in respir,l\ion and ordinary swim mi ng. The
large divergence rJtio-that is, one giant fiber can excite single giant fibe r in each of the 6 or 8 stellar nerves on
mJny other neu rons or muscle fibers . Moreover, the e,1(h s ide together innervate most or all of the muscle
curren t available for electricJI trJnsmission of the fibers of the m'11l tle. EJch giant fiber has In,my cell
act ion potenti<l l to downstre,ml neurons is large; elec- bodies in a lobe of the g,lnglion and is therefore a fused
trica l trans mission may be common in these systems. In syncytium of many Illotor neurons. "The" giant fiber of
all of the adequ,ltely studied C<1ses gi<1nt fiber ac tion the squid is the I<lst, longest, ,md I,lrgest of t he 6 or 8 on
66
275
67
276
occurrence and Frequency both by tonic phase of one element of the oscillatory
input and by higher central levels (e.g., pattern relative to another. In severa l of
by changes of mood). the known centra lly driven beh.wiors,
we can relate the functions of peripheral
A. Central Rhythms feedback to one or more of these pM-
and Reflex Modulation ticular parameters.
The peripheral or reflex hypothesis fails Lomsl Flight . One of the most decisive
to account adequately for respiratory studies on central rhythms is tha t of
control of vertebrates. A crucial test, total Wilson on wingbeat control in grass-
deafferentation, leaves a functioning hoppers. These ani ma ls will sometimes
central system. Similar tests have shown fly when the g.1nglia of the head and
that many other rhythmic control sys- abdomen are removed, so that the pat-
tems have built-in central scores. The tern generator must be present within
motor neuron discharge seq uence, which the thoracic segments. Th e normal motor
withdraws the mantle and closes the score has been analyzed in such detail
valves in a clam, Mya, occurs after de- that the temporal sequence and phasi ng
afferentation. The copu latory move- of every motor neuron impulse during
ments of a praying mantis, the flight Aight is known. This pattern of motor
rhythm of insects, the walki ng patterns neuron discharge can be recognized in
of insects and amph ibia, respiratory the central stumps of the thoracic nerves,
movements in insects, beating of the even after all those"'11erves have been
swimmerets in decapod crustacea, strid- severed. Isolated thoracic nerve cord
ulatory singing in crickets and grass- preparations are not spontaneously ac-
hoppers, side-la-side alternation of tive in the flight rhythm, but random
longitud inal muscle activity in sharks, electrical st im ulation of the nerve cord
heartbe.1t and gastric mill contro l in ca n elicit nearly the same pattern as is
crustacea, and the swimmi ng beal of found in intact animals during Aight. The
jellyfish have been shown with varying output pattern of deafferented prepara-
degrees of rigor to be centrally con- tions is defi cient in one major respect; it
trolled. is low in frequency. The decre.1sed fre-
A few of these examples deserve more quency can be ascribed to lack of input
detailed discussion. But first we should from four stretch receptor cells, one in
poi nt out that proprioceptive feedback the hi nge of each wing. These stretch
functions have also been demonstrated receptors discharge when the wing is
in nearly all of th em, and in the dis- elevated. During flight they fi re one to a
cussion of centrally patterned motor few impulses toward the end of the
output we should attend to the role of upstroke in each wingbeat cycle (Fig.
reflexes as well. In studying cases of 7.27). The number of impulses in the
oscillatory behavior, as in all considera- burst is correlated with wingbeat ampli-
tions of oscillatory phenomena, three tude, the timing of the burst with wing-
measures always merit notice: frequency beat phase, and the burst repetition rate
(or the reciprocal of frequency, th c pe- with wingbeat frequency -all the meas-
riod) of the osci lla tion, amplitude, and ures of a n oscillation. This information,
68
277
Sec tlo n VI
Cenerall y Scored Behav ior:
l'.IU ern ing in Space and T ime
I I I I I I I I I , I ' I I I I I I I I I I I I I I I I I , I ' I I I I I I I ,I
o '00 '00 300 400
Tima (msec)
upon enteri ng the eNS, would be ade- and ampli tude. Th e ganglion integrates
q uate to trigger and control the events of and smooths the input over several
th e next cycle. Appa rently, however, it wingbeat cycles, thereby almost bu t not
does not even signi fi can tly affect the next quite los ing the phasic informa ti on. The
cycle. In spite of th is rich detail of in- fil tering process is analogous to th e
formation a bout wing position, there is smoothing and integration in a re-
li tt le input/output correlation except a sistance-capacitance electrica l net work.
relatively long-termed one relati ng aver- In su m, locust fli ght reaffercnce primar-
age input frequency in th e stretch re- ily plays a role in con trolling the average
ceptofs to average wingbeat freq uency excitation of the motor p.l t1ern genera tor,
69
278
thus affecting wing beat frequency and to produce th e swimmeret beat com-
Chapter 7
integrdti OI1 at the intennedidte levels
power. It has a very we<lk effect on mand, and the output of a deafferented
wingbeat phase. or isolated preparation is normal in both
An impo rt.l llt lesson to be Je.1!"ned frequency a nd segmental phasing (Fig.
from this exam ple is thai, even though 7.29). The known proprioceptive reflexes
one can demonstrate that proprioceptive cannot even modulate these parameters.
input in a rhythm ic syste m fits the re- They do, however, modulate th e force of
qu irements for a refl ex feedback model, each st roke, affe cting the velocity a nd
that input may not be necessary for th e amplitude by influencing the number
normal pa ttern, and significant infor- and repetition rate of motor impulses
mation parameters for the peripheral during each cycle-in other words, the
hypothesis may not even be used in the magnitude of the motor discharge.
normal operation of intact anima ls. They
may be discarded in a filtering process.
B. Command Cells
Swimming ill Slzarks. When many fi shes A superficially quite different category of
swim, the longitudinal musculature of centra lly scored pattern is that called up
the two sides contracts in alternate by com nl.1nd cells, actually just the e nd
metachronal waves. If a shark is curar- of a spectrum of mechanisms that occur
ized to the point of total paralysis, motor in all degrees. Command units, first
output in the segmental nerves may still discovered in crayfish, are now known in
result when the anima l is stimulated, but many arthropods, annelids, molluscs,
since no movement occurs there can be and vertebrates (see Box 7.3, p. 274 ).
no correlated proprioceptive feedback . In Com mand units, or redundant clusters of
th is circumstance, bursts of impulses still similar units, may turn out to be q uite
issue alternately through contralateral general. The concept and the term come
nerves, but at unusually low frequency. from the observation that certain single
Proprioceptive input may, as in locusts, un its, upon sti mulation, are capab le of
be necessary for ton ic excitation of cen- causing actions rese mbl ing major pieces
tral state. But in sharks, compMison of of normal behavior. Some c<luse static
the outputs on the same side in different postu re, others phasic sequences. In
segments shows them all to be syn- higher invertebrates, in which they Me
chronous. The metachronicity is lost. On prob<lbly all potentially identifiable, non-
the basis of presently available evidence, identical s ubsets of com mand cells may
it appears that proprioceptive feed back act together to determine the form of the
is necessary for the phasi ng of outputs in behavior. Th ey are presumed to trigger
proper segmental seq uence as well as for or release the action, not to instruct the
the ma intenance of normal frequency motor neurons in the te mporal patte rn of
(see Fig. 10.63). their firing; that pattern is alre.1dy pre-
formu lated by other neurons.
Crayfisl! find Lobster Swimmerei Beal. The The best-known command fi bers in
abdomina l appendages of decapod cf<lyfish are not of extraordinary size, are
crustaceans also have a metachronal not motor neurons themselves, and in
rhythmic beat. The completely isola ted general do not even synapse d irectly on
abdominal nerve cord can be stim ulated motor neurons. They are high er-order
70
279
interneurons that run through several CM10
CJ
62 / 1
ganglia or even the whole length of the , 60
neuraxis. They excite whole motor 0", ,
- r 63 ,
' '75
,; ;
systems, small or large, controlling pos- L. __ ,. ~; 1 74
66 ,, /69 , 71 '>' 72
Each is uniquely characterized by posi- --.J ~- ,
tion in the nerve cord, axon diameter, 68 70
output function, and other properties.
The pattern of output does not depend A
much upon command fiber frequen cy or
pattern, and the output may continue
well after the stimulation of the com-
mand fiber has ceased.
The command fibers are obviously
labeled lines. The temporal pattern of
activity in each is relatively unimportant.
What is important is which command
fibers are active, since they drive diverse
behaviors. Each command fiber con-
trolling posture of the crayfish abdomen
seems to have unique but overlapping
output fields. Perhaps the severa l fibers
commanding swimmeret beat are also
unique and produce somewhat different
B
outputs (Fig. 7.29). Whether they do or
not, swimmeret movements can be
differentially controlled by combining
the action of fibers driving the OScillatory
mechanis m with those affecting posture.
Command fibers m.1Y turn motor
system s on or off or bias them for pur-
poses of steering or orientation.
Some giant fibers (see Box 7.3) are
command units- namely, those that are
not motor neurons but interneurons re-
ceiving from many small cells. They are
specialized for prompt, synchronous, and
brief action s that do not continue after Figure 1.28
the giant fiber stops firing. local sites in A. A comm and fibe r for posture. This fiber, called CMlO, travels in area
the hypothalamus of mammals, where 75 of th e circ um eso phageai conne cti ve of the cray fi sh. Wh en stimul ated at
stimulation rather reliably triggers char- a minimum of 20 sho ch per sec, it release s the "defensive postu re"
shown in H, invol ving stereotyped positions o f all appendage s. This is one
acteristic behavior, may represent a of the stati c, postura l respo nses trigge red by anyone of a small numbe r
cluster of cells acting as a kind of com- (3-5) of sp ecifi c co mmand neu rons; othe rs are dynam ic, rh ythmi c move-
mand cente r (pp. 316-319). ment s, such as swimmere! beat ing (e). [l'art A, Wiersma, 1958.J
71
280
Command neuron
Ch"pter 7
Inte gration at the Inte rmed iate Levels
What sets off a command cell? In tion arise? Cou ld it be due to interactions
general, we do not knolV, but an in- between the motor neurons? Several
formed guess might be that it usually sorts of motor neuron interaction a re
requires a number of backgrou nd condi- known . Mutual excitation and reciprocal
tions plus some triggering input. This inhi bition between motor neurons in the
means it is highly integrative, acts as a stomatogastric ganglion of crabs and
recognit ion un it (p. 236) to detect these lobsters are respons ible, at least in part,
criteria, a nd is therefore a decision unit for the way in which the output pattern
(p. 238) in a significant sense. The input of that ganglion rein forces a nd ad justs a
it requires may come fro m sense organs, spon taneous rhythm or command fi ber
but in some cases it seems likely that response (p. 279). Some motor neurons
centrally a risi ng changes in mood or exciting the same muscle in insects are
re.ldiness might do the same th ing (see electrotonically coupled and tend to fire
vacuum activity, p. 315, central spon- together. The sa me is true for contra-
ta neity, p. 325). laterally paired inhi bitory motor neurons
in nervati ng the a bdominal postural
muscles in crayfi sh . Motor neu rons in-
C. Hierarchical Structuring nervati ng th e same flight m uscle in cer-
of Molor Systems tain flies inh ibit each other. Synergistic
motor neurons in vertebrate spina l cords
If rhythmic motor output can be driven are a lso inhibitorily lin ked, throug h
by nonoscillatory impu lse trains in the known short-.lxon inte rne urons, the Ren-
command fibers, where does the oscilla- shaw cells (p. 269).
72
281
If interaction between motor neurons upon identical or overlapping pools of
Section VI
were genera lly responsible for thei r motor neurons. Thus the motor neurons C~"'tr.l.lly Sco<('d Bth~Yior'
rhyth mic activity, one might expect that Me, to use Sherringlon's phr,1se, the r~tt (' rnlng In S p~c ... ~nd Timt
ant idromic stim ulation of sets of motor "final com mon paths" transm itting to the
neurons could reset or modify the output muscles signa l piltt em s th"t are pro-
of the whole network. In genefil!, this is duced at a higher level.
not true. At a more detailed level of If we turn from a preoccupat ion with
analys is one wou ld expect to find th.lt the genesis of rhythm to the question of
intracellular stimu lati on of one motor how alternative motor patterns that in-
neuron would give rise to synaptic po- volve higher level switching are selected
tentials in functionally related ones, but and programmed, we have to dea l
aga in this is not generally true, or else mai nly with conce ptual models that
the effects are quite weak. Current seem compatible with principles of
thought on the matt er is that motor physiologica l organization.
neuron interactions are usually in- The rather wide ly accepted notion
adequate to account for the patterns of today is that actions commanded by
output in motor systems. higher centers are not specified in detail
This concl usion leaves us look ing for by those centers, nor primarily deter-
a process, or even a structure, that mined by peripheral stimuli, but trig-
mediates between the com mands and the gered in a preprogrammed language
motor d ischarge itself. The search ison for ca lling up combinations of elementary
interneurona l pacemaker cells or net- acts in a hierarchy of levels. The highest
works th"t produce rhythmicity(Fig. 7.29}. command, as well as the lower-level
Cons istent with the notion that there instructions to still lower levels, may
is a hierMchy in molor systems, with simply specify the sequence, strength,
input co mmands driving osci ll ators that " nd duration of the nex t lower compo-
drive motor neurons, is the fact that the nents, finally elicit ing moveme nts as
same set of motor ne urons ca n be used though the adequate periphera l stimuli
in more tha n one behavioral pattern. had occurred that can reflexly elicit
Frogs swim or jump with synchronous them . Ana lysis of six ga its of horses
ou tput to homologous cont ralateral s hows that they could be produced
muscles, bu t alterna te them d uring simply by ca lling up components eq ual
walking. Insects use some of the same to certain loca l spinal and long spinal
muscles to move the wings and legs, but reflexes in formulat ed sequences and
they do so according 10 different durations, given a few fixed rules
synergistic/ antagonist ic relationships, (Easton, 1972). Whether the horse
depend ing upon whether the com mand actually works this W,lY, we ca nnot tell,
says "walk," "jump," "sing," or "fly." but the !'ul es are simple, and the model
We are pu shed into thinking th"t even in might work. Of course, superimposed on
the lower ga nglia or s pina l cord there is wha tever "c" lIing up" the bra in initiates,
a mu ltiplicity of p"Uern gener.ltors or proprioceptive input as well as visual
oscillators that can each be turned on or and s01llesthelic ]'eafference (sensory
off or be modul.1ted in frequen cy or .1mp- input c"used by one's own actions) wi ll
lilude by command input, and thai be important in shaping and correcti ng
these pattern generators each converge the centrally patterned prograJll, by
73
2.2
!'DI---==-
l
Comma nd
+.-----...----''-<G)I----==- L""".
Ton ic
Depressors
Ph asic
(campanlfo rm
E)tci lalory se nsi ll a)
Inhibitory
A
Feeding
pacemaker
Com po nen ts 01
intri nsic
program
Positive leedback
c~' ..'''~~
ConUllcUon loop sustaining
S tress ing 01
J
mechano receptor,
Ret ra cti on 01
An tagonist ~j,,.;..,
+ .,.... buccal mass
~----~~----
Delayed nega tive l eedback loop
ending mechanoreceptor bu rst
74
283
s imple adjustments of strengt h and D. Centrall y Scored Pattern
Se(tlo n VI
ph.1Se of pdrticular components. by Sensory Tape Cen trall y Scored Dehulor:
Motor patterning systems have been Patterning in Sp.ce a nd Time
so incompletely s tudied that we kn ow
velY little of th eir actua l mechanisms. Another possible mecha nism of central
Parti,ll models with some s upporting programming, besides the motor score,
evidence ca n be made in many cases. has been suggested by Hoyle. He calls it
(Examples are s hown in Figure 7.30 and a sensory la pe, or 10 use the phrase of
in the figure in Box 7.3, p. 275; see also p. the ethologists, a sensory templa te (sec
333.) Since it has been so diffi cul t to p. 309). Sensory tapes or templates have
ma ke an ana lys is of even relatively sim - not been demonstrated, though strongly
ple motor syste ms in terms of individual inferred for the control of some bird
neuron activities, perha ps we shou ld ex- song. T he idea is worth more discuss ion.
pect 10 fin d eventua lly Ihat presen lly Suppose the CNS contained an instruc-
un known concepts of neural function are tion tha t sa id, " Produce a motor output
involved. that results in a s peci fi ed feedback from
75
,
264
-
Compare
di fference
Genera! Specillc
d river driver
Propriocep tive
afference
Spec ilic
driver
Othe r
"","U
Xp".,
2nd- o rder
driver
Extenso r
Jo int
A
76
285
proprioceptors or other sense organs." tion of a sensory template n~ ally redu ces
Seell a n VI
This ins tructi on would not lead inevit- to reflex modulation of a motor score. Celltrd ll y Score.! Beha vior:
ably to a sing le s tereotyped motor out- r dtt ern ing in Sp.lce ~ nd Ti me
put, as from a molar score ge nerator, but
it could guide motor out put to achieve a E. Coordinated Movement to Gross
goal (Fig. 7.31). Ou tput might at first give Stimulation of the Brain
in correct results, but feedb<1Ck cou ld
modulate th e output on successive cycles A clear progression is ev ident if one
of loop operation . Consistent with th is compa res the responses to crude elec-
notion of central progra mming by com- trica l stimulation of s tructures at succes-
parison of sensory feedback with a sive neural levels. Ventral roots give a
centrally slored goa l pattern is the fa ct segmental, loca l contraction closely re-
that diverse motor oUlputs may be asso- lated to the duration and strenglh of
ciated wit h a pparently identical leg stimula tion. Lower motor centers in the
movements during walk ing in insects. In cord or bra in stem give little more,
some cases the flexors and extensors though the d istribution m.1Y be mOfe
alternate. In others, one muscle contracts functional (e.g. flexion of certain joints).
tonically whil e the other oscil lates. The The responses relevant to this section are
resulting movemcnt is thc same. the quite normal actions involving
T he only reasonably strong case of a sequences of 'movements, s uch as ca n be
sensory template is found in bird song elicited from the hypothalamus of
control (see Fig. 8.18). In the invertebrate mammals (p. 471). These are so natural as
cases in wh ich a sensory principle may 10 suggest a centra l pattern, si nce the
be operat ive, it seems to be supe r- stimuli are like lightning bol ts. A pocket
imposed upon a motor score type of mouse may s tuff invisible seeds inlo its
central generator. Insect fli ght is basically cheek pouches at a high rate, a cal may
programmed by .1 motor score, but ex - arch its back, hiss, unshealh its claws,
teroceptive as well as proprioceptive erect its hair. The value for our purposes
inputs can modify that score for pur- is the same whether we assume that the
poses of s tability, s teeri ng, or com pensa- stimuli trigger motor patterns or sensory
tion for inherent error or damage to bod y " hall ucinations": the patterns are cenlral
paris. Perhaps in these systems the no- and need only an adequate trigger.
77
286
Box 7. 4 Chronology and Background of Ideas on Ihe Ph ysiology of Ihe Nervous System
We present here i\ selection of highlights in the history but it a lso secretes phlegm OWd cools the blood. Even in
o f ide,IS, from the ('.Hlies! till1es up to 1929. The inte r- the Gold en Age the Greeks- d id no l easily fo llow his
pretation of the brain in terms of cells is highlighted in exam ple. The lack of .lutopsies de l'lyed progress.
Box 3. 1, p. 102. The roots of bra in chemistry, membrane
340 B.C. Aristotle syste molt k.llly pursued compar,l-
bio physics, ph,umacology, sensory and psycho-
live an.t to my and in his 19 books set a high wolter mark
physiology, .md be havioral a nalys is arc not attempted
of natura l knowledge thilt l.lsted until the RenaisS.lnce,
he re.
but he did litt le to change ideas on the bra in.
The redder is urged to look furthe r, fo r more balance
a nd adcquille representation. Some useful, more-or- Iess 300 8.(;. Herophilus and the greilt school of
conde nsed accounts aTC by Nordcns kiold (1935), Alexandria in Egypt dissected many cadavers Jnd really
Dampier ( 1948), Singe r (1959), Brazier (1961). Sirks and founded anatomy. He distinguis hed sensory and mo tor
Zirkle (1964), Gard ner (1965), Clarke and O'Millley nerves and sho wed tha t they connect from spinal cord
(1968), and McHe nry (1969). Specia l as pects are dealt to periphery.
with in FeiHing (1 970), Brazier (1961), and Swazey and
250 8.C . Erisistrat us postulated a mechanis m of the
Worden (1976).
brain function: blood and two kinds of air arc carried in
the veins, arteries, and nef\'es; air is changed to vita l
1700 B.C. An Egyptian document, transl,l ted cen-
spi rits in the hear t and these to animal spirits in the
turies later ,lnd published ,lS The Edwin Smi th Surgicdl
brain ventricles, whence they go via the nerves to
Papyrus, includes 13 Cdse descriptions of head injuries.
distend and sho rte n the muscles. -.
Aphasia, paralySiS, and seizures were descri bed, and
suggested the fun ct ions of the brain. Nevertheless, 200 6 .C. Galen culminated the classic period, writing
diseolse continued to be generally oltiributed to suprd- more than 400 works that were dennitive for a
naturoll influences olnd whims of the gods. mille nium. By now much of the naked-eye anatomy of
the nervous system had been d iscovered, including
800 B.C. Homer's works and the flowering of Greek
most of the cranial nerves. Among the few advances in
iHt and intellectua l life led slowly and incomple te ly to
the unde rstanding of function were t he descriptions of
the idea thilt the world is knowilble.
symptoms fo llowing section OInd hemisection of the
500 B.C. Alcmaeon performed the first recorded spi na l cord in lower mammol ls.
d issection of ,l human body. He p,l id some ,ltiention to
400- 800 The D.uk Ages lasted more than 12
the brain and discovered the optic nerves. His teache r,
generat ions. Greek knowledge WolS forgotten in Europe.
Pythagoras, tilught th.t t the brain is conc::erned wi th
Men did not ask to unde rstolnd themse lves or nature but
reasoning. In the next century more dissections and
to be told the supra natura l or religious meanings of
similar speculations were made by others.
things.
400 8.C. Hippocr,ltes of Cos countered the mys tics
600-1200 Islam s pread from Asia minor to Spain,
and the entrenched supranaturalists to introduce ra-
carrying Gree k knowl edge unknown in the Christian
tiona l medicine. This requ ired the systematic accumu-
wo rld. Jewish tfolders introduced Arabic transla tions to
lation of cl inic::al experience, and his desc::ription of
Europe. Long-lost Latin versions of Greek writings were
e pilepsy went unsurpassed un til the work of H ughlings
rediscovered in monastery storeroo ms. Men d id nol ye t
Jackson. However, Hippocratic teaching was dominil ted
ask abou t nature but Jbout thei r heritage.
by the idea tha t fun ction derives from the combination
of four humors; blood, phlegm, ,lnd black and yellow ]200- 1300 This interest in book learning and the
bile. The brilin is the org.tn of intelligence .md d re,lms, new wave of ideas induced schol.lSticis m, which in turn
78
287
79
288
80
289
81
,
290
Chapler 1
Inleg ra tion al the Inter mediate Levels
SUGGESTED READI NGS
82
Neuron, Vol. 16, 701716, April, 1996, Copyright 1996 by Cell Press
83
Neuron
702
84
Review
703
85
Neuron
704
86
Review
705
many fronts. First, whole-cell recordings suggest that neurons the axon is the spike initiation zone and sug-
the neuronal time constant may be much larger (and gested that a major function of dendritic sodium action
thus, the Rm much higher) than previously assumed from potentials is to convey information about the state of
microelectrode recordings (Spruston et al., 1994; Major the soma to the dendrites. This centrifugal, antidromic,
et al., 1994). This discrepancy is attributed to the shunt or backpropagating role for the action potential has also
introduced by the somatic leak induced by the micro- been proposed in olfactory bulb neurons (Rall and Shep-
electrode impalement. Second, another significant re- herd, 1968), Purkinje neurons (Llinas and Nicholson,
sult has been the study of cable models of branched 1971), and CA1 pyramidal neurons (Richardson et al.,
dendritic trees that cannot be approximated by equiva- 1987) and constitutes a blatant contradiction to Cajals
lent cylinder models (Stratford et al., 1989; Major et al., axofugal law.
1993; Major, 1993). This work has confirmed the impor- Similar experiments with dual somatic and dendritic
tance of the shunt created by the electrode and has patch recordings have been carried out in other neuron
suggested the impossibility of good voltage-clamp con- types. While results from Purkinje cells indicate a so-
trol of dendrites from the soma. Finally, two new ap- matic initiation zone for sodium action potentials (Stuart
proaches have been used to estimate the length con- and Hausser, 1994), in CA1 pyramidal cells the spike
stant of the main dendrites: the measurement of voltage initiation was somatic but shifted to proximal dendrites
attenuation between dendritic and somatic electrodes with increased synaptic stimulation (Spruston et al.,
for injected DC pulses of current (Sugimori and Llinas, 1995). Finally, many neurons in substantia nigra had
1984, Soc. Neurosci., abstract; Stuart and Hausser, dendritic initiation; interestingly, this was shown to be
1994) and the use of the voltage ratio between somatic due to the axon emerging from the dendrite (Hausser
and dendritic calcium plateaus, as recorded at the soma et al., 1995).
(Reuveni et al., 1993; Yuste et al., 1994). Taken together, these studies lend support to the idea
In spite of this, the basic cable properties of dendrites that in pyramidal and Purkinje cells the spike initiation
are unknown, and it is even possible that they may not zone is primarily axonal/somatic. However, several im-
be constant throughout the dendritic tree. Still, the past portant caveats preclude a definite answer. First, like in
few years have seen an explosion of modeling studies CA1 neurons, the location may depend on the intensity
of neurons, triggered by the availability of computer of the stimulation. Second, these recent studies do not
programs that implement cable equations in multicom- address the role of calcium spikes in this process, due
partmental models (reviewed by De Schutter, 1992). Re- perhaps to the young age of the cells and the experimen-
sults from different modeling approaches based on tal conditions at which experiments were performed.
careful fitting of experimental data converge on similar Because calcium spikes are prominent in recordings
values for Cm (0.72 mF/cm), Rm (50200 kVcm2 ), and Ri from mature animals, both in vivo (Pockberger, 1991;
(200400 Vcm), suggesting that these values may be Hirsch et al., 1995) or in vitro (Amitai et al., 1993; Sak-
close to the real ones (e.g., Stratford et al., 1989; Sprus- mann, 1995, Soc. Neurosci., abstract), their existence
ton et al., 1994; Rapp et al., 1994; Major et al., 1994). may change the balance of excitation between dendrites
Where Is the Spike Initiation Zone? and soma. Third, dendrites in vivo may sustain their
In the classical work from the fifties, it was concluded own spike-generating zones, isolated from the soma,
that motoneurons had two thresholds for spike initiation, because of the electrical shunt produced in dendrites by
a lower one (z10 mV) in the axonal initial segment (IS) their constant bombardment with excitatory or inhibitory
and a higher one (z30 mV) in the somadendrites synaptic inputs (Bernarder et al., 1991; see below) and
(Coombs et al., 1957). This conclusion was reexamined because significant changes in the length constant
in CA1 pyramidal neurons in vitro with a combination of could be produced by intrinsic electrophysiological
extracellular, intrasomatic, and intradendritic microelec- states of the neuron (see below). In fact, stimulus-
trode recordings. In response to synaptic stimulation at evoked calcium spikes in vivo are more frequent during
threshold, the site of spike initiation was located at or epochs of strong synaptic barrages (Hirsch et al., 1995).
near the soma (Richardson et al., 1987), but shifted to the Inhibition Can Regulate Dendritic Function
proximal dendrites when synaptic strength is increased Approximately 20% of the synaptic contacts in the brain
(Turner et al., 1991). Therefore, the spike initiation zone are inhibitory (Peters et al., 1991); moreover, electron
for CA1 neurons is dependent on the stimulation par- microscopic studies indicate that the location of inhibi-
adigm. tory inputs in dendrites can be very specific (Peters
Experiments with patch electrodes have recently ad- et al., 1991; Buhl et al., 1994). The influence of those
dressed this question in neocortical pyramidal neurons. inhibitory inputs on the physiology of dendrites was
Under somatic voltage clamp, Regehr et al. (1993) found demonstrated by the inhibition of dendritic spikes in
unclamped sodium spikes triggered by synaptic stimu- Purkinje cell by stimulation of inhibitory interneurons
lation and suggested that the spike initiation zone was (Llinas et al., 1968).
dendritic (Regehr et al., 1993). In contrast, Stuart and The role of inhibition in controlling dendritic excitation
Sakmann (1994) used dual recordings from soma and has been recently explored in pyramidal and Purkinje
proximal apical dendrites to show that, after synaptic neurons with calcium imaging and dendritic recordings
stimulation, dendritic sodium spikes were always back- (Callaway et al., 1995; Kim et al., 1995). Callaway et al.
propagated from the soma (Figure 3). The authors then (1995) have studied the effect of inhibition on calcium
patched the axon and showed that, in response to so- accumulations of Purkinje neurons and find that IPSPs
matic current injections, the axon always fired ahead of reduce the dendritic depolarizations and calcium accu-
the soma. They proposed that in neocortical pyramidal mulations produced by climbing fiber activation. This
87
Neuron
706
effect is graded and can be restricted to individual den- CA1 neurons (Jaffe et al., 1992), where, in response to
dritic branches. In intradendritic recordings from neo- a train of spikes generated at the soma by an intracellular
cortical pyramidal neurons, Kim et al. (1995) have shown current injection, the authors detected a difference in
that IPSPs can reduce, and eventually block, the den- calcium and sodium dynamics between proximal and
dritic sodium and calcium spikes generated by current distal dendrites. While calcium and sodium accumulated
injections. They also report that the IPSP timing controls in proximal dendrites during the entire spike train, the
the timing of dendritic spiking, producing phase-ad- accumulations in distal dendrites were limited to the
vance or phase-delay effects. These two studies demon- first few spikes, indicating that the sodium and calcium
strate that inhibition can play a fundamental role in the channels responsible for the distal accumulations had
control of dendritic spiking. Besides its traditional role turned off. Hence, sustained firing of the cell produced
of blocking excitatory inputs, inhibition may also control higher calcium accumulations in proximal dendrites (Re-
temporal aspects of the dendritic response. gehr et al., 1989). A similar result was found in neocorti-
The mechanisms by which inhibitory inputs produce cal intrinsically bursting neurons, where a train of anti-
these effects are likely to be multiple. The increases in dromic spikes generated a dissociation in the calcium
chloride permeability produced by g-aminobutyric acid dynamics of the proximal versus distal apical dendrite
(GABA) could lead to a hyperpolarization of the neuron (Yuste et al., 1994). Again, while calcium accumulated
or to a shunt of depolarizing current produced by the in the proximal compartment of the cell during the entire
concomitant decrease in Rin. Also, an inhibition of the train, the distal apical dendrite only increased its calcium
calcium current by GABA independent from changes in at the beginning of the train. It was hypothesized that
chloride permeability has been described (Dunlap and calcium-activated hyperpolarizing conductances medi-
Fischbach, 1978; Sugimori and Llinas, 1985, Soc. Neu- ated the switching off of the distal dendrite.
rosci., abstract) and could be mediated by cisternal or- More direct evidence for an activity-dependent
ganelles (Benedeczky et al., 1994). switching off of CA1 dendrites was obtained recently
Dendrites Can Be Turned off by Spikes with intradendritic recordings (Spruston et al., 1995; Cal-
One noteworthy recent result has been the discovery laway and Ross, 1995). In both studies, the first spike
that dendritic function can be intrinsically modulated by of an antidromic spike train propagated throughout the
activity-dependent mechanisms. entire dendritic tree, while subsequent spikes were fil-
The first evidence came from imaging experiments in tered to different degrees. In most distal dendrites, a
88
Review
707
gradual decrement in the amplitude of the action poten- two-photon microscopy (Denk et al., 1990), which en-
tials during the train was observed. This activity-depen- ables imaging of spines in brain slices for extended
dent filtering was triggered by action potentials and had periods of time (Figure 4). New results in CA1 and Pur-
a slow time course that produced frequency depen- kinje neurons show that individual spines are activated
dency, i.e., faster firing rates resulted in more attenua- under subthreshold synaptic stimulation, but that once
tion (Callaway and Ross, 1995). Interestingly, all-or-none the cell fires an action potential, every spine in the cell
filtering was seen at some branch points, producing becomes activated (Yuste and Denk, 1995; Denk et al.,
spike invasion failures (Spruston et al., 1995). Finally, 1995). These findings therefore show that spines behave
the recovery from this activity-dependent inactivation as chemical compartments that could constitute the
of a dendrite was accelerated by hyperpolarization minimal functional compartment of the dendrite.
(Spruston et al., 1995). What function might spines serve? The restriction of
Thus, although the mechanisms responsible are still calcium accumulations during subthreshold synaptic
unknown, the intrinsic properties of neurons can turn stimulation to a single spine indicates that spines com-
off certain dendrites in an activity-dependent fashion. partmentalize biochemical changes (Gamble and Koch,
This produces a situation where the specific location of 1987; Wickens, 1988; Koch and Zador, 1993). This com-
the inputs matters, as opposed to a dendritic tree where, partmentalization could segregate plastic changes in
like a computer network, the physical location of the synaptic transmission (Rall, 1974; Lisman, 1989). In fact,
connection is irrelevant. For instance, synapses located electrophysiological experiments in vivo have implied
in distal dendrites may only be sensitive to the first few that individual spines may be specifically potentiated
spikes from a train of action potentials produced by because granule cell spines contacted by the contralat-
the soma, while those located more proximally may be eral perforant pathway, but interspersed among spines
sensitive to every action potential. This filtering would contacted by the ipsilateral pathway, are not potentiated
make distal synapses sensitive only to changes in the together with the rest of the spines (Levy and Steward,
rate of output of the cell, a mechanism that might be 1979; Levy and Desmond, 1985). In agreement with this,
used to implement synaptic algorithms gated by the supralinear calcium accumulations occurred in hippo-
derivative of postsynaptic activity, similar to the delta campal spines when synaptic and spike activation oc-
learning rules explored in artificial neural networks. curred simultaneously, indicating that spines may be
What Is the Minimal Functional Compartment able to function as coincidence detectors (Yuste and
of the Dendrite? Denk, 1995). An implication from this argument is that
A central question in dendritic integration is whether smooth neurons, with few or no spines, cannot compart-
the dendritic tree works as one unit, or whether it has mentalize plastic changes in synaptic transmission like
different functional compartments. We can define a spiny neurons (Wickens, 1988), if plasticity is not pro-
functional compartment as the morphological unit of duced by highly localized, submembrane calcium con-
the cell that carries out a determined functional role. centrations. In preliminary experiments from CA1 stel-
Functional compartments do not necessarily have to late cells, changes of synaptic efficacy in one input do
be electrical units, since chemical compartmentalization influence a second distant input, suggesting that spines
may be of fundamental importance to the function of are necessary to compartmentalize plasticity (S. J. Red-
the neuron. Moreover, certain computations may be im- man, personal communication).
plemented chemically, rather than electrically (Sobel Interaction among Dendritic Compartments
and Tank, 1994). Even if spines are the smallest functional compartment
In 1891, Cajal first described the existence of dendritic of a dendrite, a more pertinent question is what the
spines (Ramon y Cajal, 1891b) and speculated that their typical functional compartments are under the condi-
function was to increase the receptive surface of the tions in which dendrites are integrating inputs. In fact,
dendrites and enable intimate contacts between axons there is clear evidence that dendritic compartments of
and dendrites (Ramon y Cajal, 1904). The idea that a larger scale than a dendritic spine may be functional
spines were sites of synaptic contacts was confirmed units. Whereas anatomical evidence suggests that sev-
with electron microscopy (DeRobertis and Bennett, eral functional areas exist in dendrites, physiological
1955; Palay, 1956). Hence, anatomically, spines are a data show that different dendritic compartments can
basic subunit of the dendrite. Still, no functional informa- interact to generate intrinsic functional dynamics in the
tion existed about dendritic spines until the advent of neuron.
calcium imaging techniques. Initial experiments in hip- While studying the development of synaptic connec-
pocampal pyramidal neurons showed that individual tions in cilliary ganglion cells, Purves and Hume found
spines could have different calcium dynamics than their that developing neurons were innervated by a similar
parent dendrites, behaving as independent calcium number of axons, but in adult neurons the number of
compartments (Muller and Connor, 1991; Guthrie et al., innervating axons precisely matched the number of pri-
1991). In contrast, experiments imaging spines in cul- mary dendrites (Purves and Hume, 1981; Hume and
tured hippocampal neurons and Purkinje cells in slices Purves, 1981). They concluded that cilliary ganglion cells
suggested that spines did not become activated inde- had a number of separate spatial dendritic domains,
pendently, but rather as part of a branchlet, composed each one receiving input from a single axon. This implied
of a small segment of a dendrite with all its spines (Mur- that dendrites could serve to isolate inputs from one
phy et al., 1994; Eilers et al., 1995). The branchlet then another and suggested that a single axon would only
would constitute the minimal compartment of the den- innervate one dendrite. Nevertheless, in a follow-up
drite. Recently, this issue has been reexamined with study with electron microscopy, single dendrites were
89
Neuron
708
shown to be innervated by multiple axons, although a spike (2), which activates a calcium-dependent potas-
spatial segregation of competing inputs was still present sium current (3) that deinactivates a somatic low thresh-
(Forehand and Purves, 1984). Altogether, these results old calcium spike (4), which finally triggers again a so-
suggested the existence of different dendritic compart- matic sodium spike (1).
ments dominated by different inputs. A similar relation- A second case in which the interaction between den-
ship between dendritic domains and functional com- dritic and somatic compartments produces intrinsic
partments was suggested by reconstructions of neurons electrophysiological responses is the generation of
from the tecta of three-eyed frogs (Katz and Con- spike bursting by CA3 neurons. There is evidence that
stantine-Paton, 1988). In these animals, the tectum is high threshold calcium conductances, present in den-
divided into interdigitating stripes, alternatively inner- drites from CA3 pyramidal neurons, are necessary for
vated by each eye. In some tectal neurons individual burst responses (Wong and Prince, 1978). While single
dendritic branches were confined to the stripe of a single compartment models of CA3 pyramidal neurons have
eye, although the entire dendritic tree occupied more been unable to reproduce adequately the electrophysi-
than one stripe. This result suggested that dendritic ology, the nonuniform distribution of calcium and so-
branches were functional units, because they received dium channels into at least two compartments can gen-
erate burst responses (Traub et al., 1991; Pinsky and
only one type of input.
Rinzel, 1994; Rhodes and Gray, 1994). Consistent with
The first demonstration of a functional interaction by
this, imaging of CA3 neurons during bursts and single
separate compartments in a neuron came from work in
spike firing shows that normalized calcium accumula-
inferior olive cells (Llinas and Yarom, 1981). From the
tions are larger in dendrites than in the soma during
mismatch in the sign of the spikes between intra- and bursting (Strowbridge and Tank, 1994, Soc. Neurosci.,
extracellular recordings, the authors deduced that so- abstract). Thus, the sequence of events during a burst
dium and low threshold calcium conductances were would be as follows: somatic sodium spike (1) that trig-
located in the soma, while high threshold calcium con- gers a dendritic calcium spike (2) that depolarizes the
ductances were dendritic. The spatial separation of the soma producing the envelope of the burst and triggering
conductances into somatic and dendritic compartments additional somatic sodium spikes (3). Finally, a recent
generated an intrinsic oscillatory dynamics of these neu- simulation study has extended these ideas by sug-
rons, following this sequence of events: a somatic so- gesting that even the detailed dendritic morphology of
dium spike (1) triggers a dendritic high threshold calcium neocortical pyramidal neurons could influence many of
90
Review
709
their intrinsic electrophysiological properties (T. J. Sej- junctions in adult neocortex (Parnavelas et al., 1994,
nowski, personal communication). 1995, Soc. Neurosci., abstracts) and nonpyramidal neu-
Dendrites Can Serve as the Output Side rons in hippocampus (Gulyas et al., submitted; Acsady
of the Neuron et al., submitted). Finally, dendritic gap junctions among
Dendrites in many invertebrate nervous systems serve interneurons may be necessary to generate hippocam-
as both input and output side of the neuron (Bullock, pal oscillations in the presence of synaptic blockers
1977; Watson and Burrows, 1988), and this motivated (Michelson and Wong, 1994; Traub, 1996).
the use of the term neurite instead of dendrite to ex-
press the lack of specialization among the processes. Future Issues: The Time
Since Cajal, however, dendrites in vertebrate systems of Dendritic Computation
have been considered the input side of the neuron. Nev- In this last section, we review experimental and theoreti-
ertheless, the existence of dendrodendritic synaptic in- cal findings that address more speculative issues of
teraction and dendrodendritic gap junctions are clear dendritic integration. A review covering many theoretical
contradictions of Cajals law and may constitute general results has been recently published (Mel, 1994).
forms of neuronal communications present throughout Direct Measurements of Cable Parameters
the vertebrate central nervous system. Simultaneous somatic and dendritic recordings can be
Dendrodendritic synapses have been clearly docu- used to measure the cable attenuation produced in test
mented in mammalian olfactory bulb and retina. In the voltage pulses (Stuart and Hausser, 1994; Stuart and
bulb, the antidromic excitation of mitral cells is followed Sakmann, 1994; Sugimori and Llinas, 1984, Soc. Neu-
by a long-lasting inhibition (Phillips et al., 1963). In a rosci., abstract). This approach could be used to esti-
computational and electrophysiological study, it was mate Ri for the dendritic segment between the two elec-
proposed that secondary dendrites of mitral cells ex- trodes and establish the basic cable parameters of the
cited granule cells, which would then mediate the inhibi- main dendritic shafts. The difficulty in obtaining whole-
tion of mitral cells (Rall and Shepherd, 1968). Because cell recordings from small dendritic branches, however,
granule cells lack axons, their output had to be dendritic. makes this strategy less optimal when trying to estimate
Finally, reciprocal dendrodendritic synapses between the values of cable parameters in secondary and tertiary
mitral and granule cells were found with electron micros- dendrites, which are target regions for the majority of
copy (Rall et al., 1966). The circuitry of the retina also the inputs (Larkman, 1991).
provides, in the amacrine cell, one clear example of a As an alternative approach, voltage-sensitive dyes
neuron in which dendritic spines and axonal terminals (VSDs) have been used to measure voltage transients
are not morphologically segregated to different pro- in small processes from cultured invertebrate neurons
cesses (Dowling, 1968; Dubin, 1970). Finally, a different (Grinvald et al., 1981). In a recent study, cable equations
form of dendrodendritic synaptic interactions is the den- were fitted to measured spatiotemporal voltage pat-
dritic release of neurotransmitters by substantia nigra terns, and tentative values for Rm (z22 kVcm2 ) and Ri
neurons (Cheramy et al., 1981; Llinas and Greenfield, (z250 Vcm) were concluded (Fromherz and Muller,
1987). This release is calcium sensitive and may require 1994). These values, incidentally, are not that different
exocytosis from subcisternal systems. from those found in numerical simulation studies (see
An issue that has reemerged in the last decade is above).
whether mammalian neurons are coupled via dendritic It should be possible to apply optical recording with
gap junctions, as was first established in olivary neurons VSDs to cultured mammalian neurons, or even to neu-
(Sotelo et al., 1974; Llinas et al., 1974). The existence of rons in a slice, and estimate Ri and Rm more directly.
dendrodendritic gap junctions was suggested to explain Efforts in that direction have already started, using tradi-
dye coupling in neocortical and hippocampal slices tional dyes (Borst et al., 1995, Soc. Neurosci., abstract)
(Gutnick and Prince, 1981; MacVicar and Dudek, 1981). or injectable VSDs that may obviate poor signal-to-
This coupling is particularly dramatic in developing cor- background ratio when recording optically from den-
tex (Connors et al., 1983; Lo Turco and Kriegstein, 1991; drites in a slice (Antic and Zecevic, 1995; Kogan et al.,
Peinado et al., 1993), where it can lead to coactivation 1996). Also, ratio imaging may enable accurate quantita-
of large groups of neurons (Yuste et al., 1992, 1995). tive measurements of dendritic voltages (Montana et al.,
These functional results indicate that in development 1989; Gonzalez and Tsien, 1995). Optical approaches
dendrites can serve as the output side of the neuron. seem ideally suited to investigate questions such as
In the adult brain, the extent of dendritic gap-junctional whether cable parameters are constant in the dendritic
coupling remains a matter of controversy. While inferior tree, or what the behavior of branch points is during the
olive neurons (Sotelo et al., 1974), granule cells in the propagation of voltage signals.
olfactory bulb (Reyher et al., 1991), and many neuronal What Is the Role of the Active
types in the retina (Vaney, 1991) have dendrodendritic Conductances in Dendrites?
gap junctions, in adult pyramidal neurons dye neuronal Although it is clear that there are active conductances
coupling practically disappears (Connors et al., 1983; throughout the dendritic trees of both pyramidal and
Peinado et al., 1993; cf. Michelson and Wong, 1994). Purkinje cells, the role of these conductances remains a
Evidence for gap junctions in adult cortex and hippo- mystery. Because they are voltage sensitive, they could
campus from electron microscopy is scant (Sloper, amplify synaptic inputs and help relay distal EPSPs to
1972; Powell, 1981; Kosaka, 1983) and could be due to the soma (Lorente de No and Coundouris, 1959; Shep-
the difficulty of detecting them. Nevertheless, recent herd et al., 1985). In fact, many recent studies demon-
studies suggest that there are, in fact, dendritic gap strate that synaptic inputs are indeed amplified by the
91
Neuron
710
dendritic tree. For example, in neocortical neurons, de- Purkinje cells, which might be specialized for different
polarization of the soma increases the amplitude and computations. For instance, timing information (when)
duration of EPSPs (Stafstrom et al., 1985; Thompson et could be provided by climbing fibers and preserved in
al., 1988; Sutor and Hablitz, 1989; Hirsch and Gilbert, the complex spike, while other types of information
1991; Cauller and Connors, 1994). This subthreshold (what) might be conveyed by parallel fibers and plateau
amplification is gradual and has been attributed to either potentials (Llinas and Sugimori, 1992).
sodium, calcium, or NMDA receptor conductances. A Pyramidal neurons also have segregated dendritic
recent report has suggested that it may preferentially fields for specific inputs. In hippocampus, CA3 pyrami-
occur near the soma (Stuart et al., 1995, Soc. Neurosci., dal neurons are contacted by mossy fiber terminals se-
abstract). In addition, EPSPs can become amplified into lectively at the base of apical dendrites (Claiborne et
the suprathreshold regime, evoking dendritic spikes. In al., 1986), while different populations of interneurons
fact, in Purkinje cells, climbing fiber input triggers den- contact specific areas of dendritic trees of pyramidal
dritic spikes (Llinas and Nicholson, 1971), and a similar neurons (Buhl et al., 1994). In neocortex, feedback axo-
situation occurs in neocortex with strong layer 1 inputs nal projections make specific contacts with the tips of
to apical dendrites (Cauller and Connors, 1992). Active apical dendrites, located in layer 1 (Rockland and Virga,
conductances in dendrites may act regeneratively under 1989), while stellate cells can project specifically to the
some conditions and in a subthreshold regime under base of the apical dendrite (Kisvarday et al., 1987). In
others, thus providing dendrites with two different fact, the laminar structure of the cortex is a direct reflec-
modes of operation. tion of specific input/output relations, and particular lay-
Another possible function for active conductances is ers can be thought of as having a functional identity
the linearization of synaptic inputs, i.e, the progressive (Gilbert and Wiesel, 1979; Zeki, 1993), which could be
changes in electrical properties induced to avoid satura- preserved in the dendritic tree. Nevertheless, it is still
tion by synchronous inputs (Bernarder et al., 1994). This unknown whether neurons can distinguish between in-
idea builds upon research proposing that active conduc- puts coming from different layers. Also, the reason be-
hind the morphological differences between apical and
tances control the gain of synaptic inputs (Laurent and
basal dendritic trees is still mysterious. Electrophysio-
Burrows, 1989). Finally, as explained above, dendritic
logical and imaging data indicates that distal apical den-
voltage-activated conductances may also regulate the
drites are a separate functional compartment from the
intrinsic temporal dynamics of neurons (Llinas, 1988).
rest of the cell (Yuste et al., 1994), and data from awake
What Is the Role of Antidromic Spikes?
behaving monkeys suggests that they could mediate
In many neurons, action potentials originating at the
specific behavioral responses (Cauller and Kulics, 1991).
soma can invade the dendritic tree (see above) and, in
Do Dendrites Translate Spatial
CA1 neurons, it even reaches every spine in the cell
Patterns of Inputs?
(Yuste and Denk, 1995). What functional significance
The use of the spatial properties of the dendritic tree to
could this antidromic or backpropagating spike have?
filter directional differences in input was proposed by
The temporal coincidence of synaptic inputs with post- Rall (1964). In his model with a passive dendritic tree,
synaptic spikes (i.e., the output of the neuron) could excitatory inputs located in a line would preferentially
trigger a change in synaptic efficacy, analogous to the excite the soma if they were activated in an ordered
Hebb rule (Hebb, 1949). Thus, the spike might be the sequence, with distal inputs excited ahead of proximal
temporal signal used for coincidence detection by the ones. This resulted from the time delays associated with
spines. This idea has been corroborated in experiments the transmission of EPSPs to the soma and suggested
showing that action potential blockade by local perfu- that dendrites could serve to detect spatiotemporal pat-
sion of TTX blocks long-term potentiation (LTP) (Scharf- terns of inputs. A similar idea, incorporating inhibitory
man and Sarvey, 1985), although LTP persists under inputs, was used to explain direction selectivity by the
sodium (but not calcium) action potential blockade with retina (Koch et al., 1983). In this model, the topographic
intracellular QX-314 (Gustafsson et al., 1987). Further- map of visual inputs into a dendritic tree produced a
more, recent results show that the temporal coincidence direction-selective response if excitatory inputs were
of synaptic inputs and spikes produces cooperative cal- distal to inhibitory ones. Although recent results suggest
cium accumulations in dendritic spines (Yuste and Denk, that the extent of the dendritic tree is the same as the
1995). Studies of LTP in identified connections could extent of the receptive field of rabbit ganglion cells (Yang
help resolve this issue (Markram and Sakmann, 1995, and Masland, 1992), the location of excitatory and inhibi-
IBRO, abstract). tory inputs into ganglion cell dendrites does not seem
Do Different Dendritic Compartment to follow any spatial rule (T. Famiglietti, personal com-
Carry out Different Computations? munication).
Anatomical results suggest the possibility of a relation- Do Dendrites Read Temporal Codes?
ship between dendritic morphology and functional com- A closely related issue is whether dendrites are built to
partments (see above). In agreement with this, Purkinje detect temporal input patterns. This issue is actively
cells have clear functional differences in the action of debated because of the increased awareness that pre-
climbing and parallel fibers (Llinas and Sugimori, 1992), cise timing may be used by the nervous system to code
inputs that impinge onto different areas of the dendritic information (Abeles et al., 1993; Softky and Koch, 1992;
tree (Ramon y Cajal, 1904), although the exact location Hopfield, 1995). In cerebellum, recent results show that
of the parallel fiber input may not matter (De Schutter submillisecond timing differences from the olive are pre-
and Bower, 1994). Therefore, it is possible to speak of served by the olivocerebellar projection by subtle ad-
two functional compartments in the dendritic tree of justments in the conduction velocity of climbing fiber
92
Review
711
axons (Sugihara et al., 1993). This striking finding sug- An example of this behavioral approach is the study
gests that precise timing is essential for cerebellar com- of the optomotor response in the fly (Borst and Egelhaaf,
putations. In neocortex, however, it is controversial 1992). Using calcium imaging, the visual field was shown
whether precise timing is utilized (Schadlen and New- to map topographically into the dendritic field of the
some, 1994). On the one hand, repeated temporal firing neuron. This topographic arrangement of inputs could
patterns can be detected in cortical neurons in vivo enable local amplification among inputs coming from
(Abeles et al., 1993), and millisecond temporal syn- the same area of the visual space. A different strategy
chrony has been proposed to underlie the solution of appears to be employed by dendrites from auditory in-
the binding problem by primary visual cortex (Gray et terneurons in the female cricket (Sobel and Tank, 1994;
al., 1989). In addition, the firing of neocortical neurons Figure 5). With in vivo intracellular recording and calcium
in vitro appears surprisingly reliable in response to injec- imaging from identified neurons, it was demonstrated
tions of noisy current waveforms (Mainen and Sejnow- that the dendrite becomes progressively inhibited in re-
ski, 1995). On the other hand, the behavior of awake sponse to loud chirps. This inhibition is mediated by
monkeys performing perceptual tasks can be predicted a calcium-activated hyperpolarizing current and could
from the average firing rate of MT neurons (Zohary et contribute to the forward masking of auditory stimuli
al., 1994; Schadlen and Newsome, 1994). In addition, (Pollack, 1988). An additional example comes from the
the time constants of many of the biophysical events in recent study of collision-sensitive neurons in the visual
neocortical neurons appear too slow for implementation system of the locus (Hatsopoulos et al., 1995). In this
of millisecond timing detection, while, for example, neu- system, the dendritic tree of a single neuron is thought
rons from the cochlear nucleus, specialized in temporal to carry out the multiplication of two independent input
detection, have dendritic morphologies and receptors signals that enable the neuron to calculate the time of
that enable fast transmission (Rubel and Parks, 1988; collision of an approaching object. Finally, the role of
Raman and Trussell, 1992). active conductances have been investigated in visual
Experimental evidence of temporal selectivity by den- interneurons in the fly, by comparing the responses of
dritic trees is scarce. An intriguing study of neurons in a type of neuron with dendritic sodium action potentials
the torus semicircularis in the electric fish has recently with another type that lacks them (Haag and Borst,
shown two populations of neurons with differential fre- 1996). In this study, a correlation was found between
quency sensitivity to sensory stimulation (Rose and Call, synaptic responses to higher temporal frequencies and
1993). The cell type that responds better to lower fre- active dendrites, suggesting that dendritic action poten-
quencies has dendritic spines, while the type that re- tials might be used for amplification of transient inputs.
sponds better to higher frequencies lacks them. The A similar research program studying dendrites from
authors have suggested that spines serve as frequency vertebrate neurons in vivo is feasible. Intracellular re-
filters. cordings from mammalian dendrites can be carried out
Do Dendrites Implement Logical Elements? in anesthetized animals (Llinas and Nicholson, 1971;
Imaginative proposals have suggested that dendrites Houchin, 1973; Pockberger, 1991; B. W. Connors, per-
are small logical elements (Koch et al., 1983; Shepherd sonal communication), while intracellular recordings
and Brayton, 1987; Rall and Segev, 1987). The basic from awake behaving monkeys have been achieved
idea is that summation or shunting occurs when excit- (Matsumura et al., 1988; Glenn et al., 1988). In addition,
atory and inhibitory inputs interact in a dendritic tree. If two-photon microscopy can be used to image individual
there is a nonlinear threshold mechanism afterward, neurons in anesthetized rats (Denk et al., 1994). More-
such as an action potential, these interactions can be over, the study of nonmammalian vertebrate systems
translated into a binary decision. For example, an AND in which the basic cellular machinery may be highly
gate would result if two excitatory inputs are simultane- homologous to that of mammalian neurons, but in which
ously active, an OR gate would result if there is satura- experimental access to behaving preparations is easier,
tion by either one, and an AND/NOT gate would result should speed up these investigations (Connors and
if there is a response to an excitatory input, provided Kriegstein, 1986; Heiligenberg, 1991; Fetcho and OMal-
that a shunting inhibitory one is not active. Although ley, 1995). It appears to us that the combination of the
intriguing, there is at present no experimental evidence new generation of electrophysiological and imaging
to support or contradict these ideas. techniques with the biological relevance of studies in
Studying Dendrites during Behavior: behaving animals holds many fruitful experimental ave-
A Lesson from Invertebrate Work nues for those who wonder, like Cajal did a century
Perhaps the major obstacle in the study of the function ago, what are the functions hidden behind the intriguing
of mammalian dendrites is our lack of understanding of morphologies of the dendrites.
the particular computational tasks that these cells are
solving. This problem has been tackled in invertebrate
systems, where dendrites can be recorded optically or Acknowledgments
electrically during behavioral tasks (Borst and Egelhaaf,
1994). Although conclusions reached from invertebrate We thank A. Borst, B. W. Connors, T. Freund, S. J. Redman, W. N.
Ross, T. J. Sejnowski, and R. D. Traub for sharing their results before
work on dendrites may not be applicable to understand
publication, A. Borst, S. Cash, B. W. Connors, W. Denk, S. R. Golob,
mammalian neurons, the general approach of identifying R. Llinas, W. Regehr, P. A. Rhodes, I. Segev, and K. Svoboda for
the dendritic computation by studying the dendrites dur- helpful comments, and A. Gelperin, J. Lichtman, Z. Mainem, V. Mur-
ing the performance of a behavior may be key toward thy, G. M. Shepherd, L. Trussell, and R. O. L. Wong for their help.
revealing the logic behind the design of dendrites. R. Y. was supported by the Office of Naval Research.
93
Neuron
712
Figure 5. In Vivo Calcium Imaging from a Cricket Auditory Interneuron during a Behavioral Response
Courtesy of E. Sobel, Harvard University. Pseudocolor fluorescence images depicting increases in [Ca2 1]i in an omega neuron from a live
female cricket, in response to a simulated male song. The neurites that receive input from auditory fibers are located in the lower left quadrant
of the cell. The cell was filled with the calcium indicator fura-2 using intracellular iontophoretic injection. Violet/blue corresponds to resting
[Ca2 1]i, while red corresponds to high [Ca21]i. The four images were taken with a cooled CCD camera over a period of 20 s, and they are
displayed in the following order: top left, bottom left, top right, bottom right. Scale bar is 50 mm.
References Borst, A., and Egelhaaf, M. (1992). In vivo imaging of calcium accu-
mulation in fly interneurons as elicited by visual motion stimulation.
Abeles, M., Bergman, H., Margalit, E., and Vaadia, E. (1993). Spatio- Proc. Natl. Acad. Sci. USA 89, 41394143.
temporal firing patterns in the frontal cortex of behaving monkeys.
Borst, A., and Egelhaaf, M. (1994). Dendritic processing of synaptic
J. Neurophysiol. 70, 16291638.
information by sensory interneurons. Trends Neurosci. 17, 257273.
Allbritton, N.L., Meyer, T., and Stryer, L. (1992). Range of messenger
Brock, L.G., Coombs, J.S., and Eccles, J.C. (1952). The recording
action of calcium ion and inositol 1,4,5-trisphosphate. Science 258,
of potentials from motoneurones with an intracellular electrode. J.
18121815.
Physiol. 117, 431460.
Amitai, Y., Friedman, A., Connors, B.W., and Gutnick, M.J. (1993).
Buhl, E.H., Halasy, K., and Somogyi, P. (1994). Diverse sources
Regenerative activity in apical dendrites of pyramidal cells in neo-
of hippocampal unitary inhibitory postynaptic potentials and the
cortex. Cerebral Cortex 3, 2638.
number of synaptic release sites. Nature 368, 823828.
Andersen, P., Holmquist, B., and Voorhoeve, P.E. (1966). Excitatory
synapses on hippocampal apical dendrites activated by entorhinal Bullock, T.H. (1977). Introduction to Nervous Systems (San Fran-
stimulation. Acta Physiol. Scand. 66, 461472. cisco: Freeman).
Antic, S., and Zecevic, D. (1995). Optical signals from neurons with Callaway, J.C., and Ross, W.N. (1995). Frequency-dependent propa-
internally applied voltage-sensitive dyes. J. Neurosci. 15, 1392 gation of sodium action potentials in dendrites of hippocampal CA1
1405. pyramidal neurons. J. Neurophysiol. 74, 19.
Benedeczky, I., Molnar, E., and Somogyi, P. (1994). The cisternal Callaway, J.C., Lasser-Ross, N., and Ross, W.N. (1995). IPSPs
organelle as a Ca21-storing compartment associated with GABAer- strongly inhibit climbing fiberactivated [Ca2 1] in the dendrites of
gic synapses in the axon initial segment of hippocampal pyramidal cerebellar Purkinje neurons. J. Neurosci. 15, 27772787.
neurones. Exp. Brain Res. 101, 216230. Cauller, L.J., and Connors, B.W. (1992). Functions of very distal
Bernarder, O., Douglas, R.D., Martin, K.A., and Koch, C. (1991). dendrites: experimental and computational studies of layer I syn-
Synaptic background activity influences spatiotemporal integration apses on neocortical pyramidal cells. In Single Neuron Computation,
in single pyramidal cells. Proc. Natl. Acad. Sci. USA 88, 11569 T. McKenna, ed. (San Diego: Academic Press), pp. 199229.
11573. Cauller, L.J., and Connors, B.W. (1994). Synaptic physiology of hori-
Bernarder, O., Koch, C., and Douglas, R.J. (1994). Amplification and zontal afferents to layer 1 in slices of rat S1 neocortex. J. Neurosci.
linearization of distal synaptic input to cortical pyramidal cells. J. 14, 751762.
Neurophysiol. 72, 27432753. Cauller, L.J., and Kulics, A.T. (1991). The neural basis of the behav-
Bernardo, L.S., Masukawa, L.M., and Prince, D.A. (1982). Electro- iourally relevant N1 component of the somatosensory-evoked po-
physiology of isolated hippocampal pyramidal dendrites. J. Neu- tentials of S1 cortex of awake monkeys: evidence that backward
rosci. 2, 16141622. cortical projection signals conscious touch sensation. Exp. Brain
Blanton, M.G., Lo Turco, J.J., and Kriegstein, A.R. (1989). Whole- Res. 84, 607619.
cell recording from neurons in slices of reptilian and mammalian Cheramy, A., Leviel, V., and Glowinski, J. (1981). Dendritic release
cerebral cortex. J. Neurosci. Meth. 30, 203210. of dopamine in the substantia nigra. Nature 289, 537542.
94
Review
713
Claiborne, B.J., Amaral, D.G., and Cowan, W.M. (1986). A light and Fatt, P. (1957). Sequence of events in synaptic activation of a moto-
electron microscopic analysis of the mossy fibers of the rat dentate neurone. J. Neurophysiol. 20, 6180.
gyrus. J. Comp. Neurol. 246, 435458. Fetcho, J.R., and OMalley, D.M. (1995). Visualization of active neural
Connor, J.A. (1986). Digital imaging of free calcium changes and of circuitry in the spinal chord of intact zebrafish. J. Neurophys. 73,
spatial gradients in growing processes in single, mammalian central 399406.
nervous system cells. Proc. Natl. Acad. Sci. USA 83, 61796183. Fine, A., Amos, W.B., Durbin, R.M., and McNaughton, P.A. (1988).
Connors, B.W., and Kriegstein, A.R. (1986). Cellular physiology of the Confocal microscopy: applications in neurobiology. Trends Neu-
turtle visual cortex: distinctive properties of pyramidal and stellate rosci. 11, 345351.
neurons. J. Neurosci. 6, 164177. Forehand, C.J., and Purves, D. (1984). Regional innervation of rabbit
Connors, B.W., Bernardo, L.S., and Prince, D.A. (1983). Coupling ciliary ganglion cells by the terminals of preganglionic axons. J.
between neurons of the developing rat neocortex. J. Neurosci. 3, Neurosci. 4, 112.
773782.
Fromherz, P., and Muller, C.O. (1994). Cable properties of a straight
Coombs, S.J., Eccles, J.C., and Fatt, P. (1955a). Excitatory synaptic neurite of a leech neuron probed by a voltage-sensitive dye. Proc.
actions in motoneurons. J. Physiol. 130, 374395. Natl. Acad. Sci. USA 91, 46044608.
Coombs, S.J., Eccles, J.C., and Fatt, P. (1955b). The specific ionic Fuortes, M.G.F., Frank, K., and Becker, M.C. (1957). Steps in the
conductances and the ionic movements across the motoneuronal production of the motoneuron spikes. J. Gen. Physiol. 40, 735752.
membrane that produce the inhibitory post-synaptic potential. J.
Gamble, E., and Koch, C. (1987). The dynamics of free calcium
Physiol. 130, 396413.
in dendritic spines in response to repetitive input. Science 236,
Coombs, S.J., Curtis, D.R., and Eccles, J.C. (1957). The interpreta- 13111315.
tion of spike potentials of motorneurones. J. Physiol. 139, 198231.
Gilbert, C., and Wiesel, T.N. (1979). Morphology and intracortical
Deiters, O. (1865). Untersuchungen uber Gehirn und Ruckenmark projections of functionally characterised neurons in the cat visual
des Menschen und der Saugethiere. (Braunschweig). cortex. Nature 280, 120125.
Denk, W., Strickler, J.H., and Webb, W.W. (1990). Two-photon laser Glenn, L.L., Whitney, J.F., Rewitzer, J.S., Salamone, J.A., and Mari-
scanning fluorescence microscopy. Science 248, 7376. ash, S.A. (1988). Method for stable intracellular recordings of spinal
Denk, W., Delaney, K.R., Gelperin, A., Kleinfeld, D., Strowbridge, a-motoneurons during treadmill walking in awake, intact cats. Brain
B.W., Tank, D.W., and Yuste, R. (1994). Anatomical and functional Res. 439, 396401.
imaging of neurons using 2-photon laser scanning microscopy. J.
Goeppert-Mayer, M. (1931). Ueber Elementareakte mit zwei Quan-
Neurosci. Meth. 54, 151162.
tensprungen. Ann. Physik 9, 273283.
Denk, W., Sugimori, M., and Llinas, R. (1995). Two types of calcium
Gonzalez, J.E., and Tsien, R.Y. (1995). Voltage sensing by fluores-
response limited to single spines in cerebellar Purkinje cells. Proc.
cence resonacne energy transfer in single cells. Biophys. J. 69,
Natl. Acad. Sci. USA 92, 82798282.
12721280.
DeRobertis, E.D.P., and Bennett, H.S. (1955). Some features of the
Gray, C.M., Konig, P., Engel, A.K., and Singer, W. (1989). Oscillatory
submicroscopic morphology of synapses in frog and earthworm. J.
responses in cat visual cortex exhibit inter-columnar synchroniza-
Biophys. Biochem. Cytol. 1, 4758.
tion which reflects global stimulus properties. Nature 338, 334337.
Deschenes, M. (1981). Dendritic spikes induced in fast pyramidal
Grinvald, A., Ross, W.N., and Farber, I. (1981). Simultaneous optical
tract neurons by thalamic stimulation. Exp. Brain Res. 43, 304308.
measurements of electrical activity from mutiple sites on processes
De Schutter, E. (1992). A consumer guide to neuronal modeling of cultured neurons. Proc. Natl. Acad. Sci. USA 78, 31453249.
software. Trends. Neurosci. 15, 462464.
Gustafsson, B., Wigstrom, H., Abraham, W.C., and Huang Y.-Y.
De Schutter, E., and Bower, J.M. (1994). Simulated responses of (1987). Long-term potentiation in the hippocampus using depolariz-
cerebellar Purkinje cells are independent of the dendritic location ing current pulses as the conditioning stimulus to single volley syn-
of granule cell synaptic inputs. Proc. Natl. Acad. Sci. USA 91, 4736 aptic potentials. J. Neurosci. 7, 774780.
4740.
Guthrie, P.B., Segal, M., and Kater, S.B. (1991). Independent regula-
Dodt, H.-U., and Zieglgansberger, W. (1990). Visualizing unstained
tion of calcium revealed by imaging dendritic spines. Nature 354,
neurons in living brain slices by infrared DIC. Brain Res. 537,
7680.
333336.
Gutnick, M.J., and Prince, D.A. (1981). Dye coupling and possible
Dodt, H.-U., and Zieglgansberger, W. (1994). Infrared videomicros-
electrotonic coupling in the guinea pig neocortical slice. Science
copy: a new look at neuronal structure and function. Trends Neu-
211, 6770.
rosci. 17, 453458.
Haag, J., and Borst, A. (1996). Amplification of high-frequency syn-
Dowling, J.E. (1968). Synaptic organization of the frog retina: an
aptic inputs by active dendritic membrane processes. Nature 379,
electron microscopic analysis comparing the retinas of frogs and
639641.
primates. Proc. R. Soc. Lond. (B) 170, 205228.
Hatsopoulos, N., Gabbiani, F., and Laurent, G. (1995). Elementary
Dubin, M.W. (1970). The inner plexiform layer of the vertebrate retina:
computation of object approach by a wide-field visual neuron. Sci-
a quantitative and comparative electron microscope analysis. J.
ence 270, 10001003.
Comp. Neurol. 140, 479506.
Hausser, M., Stuart, G., Racca, C., and Sakmann, B. (1995). Axonal
Dunlap, K., and Fischbach, G.D. (1978). Neurotransmitters decrease
initiation and active dendritic propagation of action potentials in
the calcium component of sensory neuron action potentials. Nature
substantia nigra neurons. Neuron 15, 637647.
276, 837838.
Hebb, D.O. (1949). The Organization of Behaviour (New York: Wiley).
Eccles, J.C. (1964). The Physiology of Synapses (Berlin: Springer),
pg. 104. Heiligenberg, W. (1991). Neural Nets in Electric Fish: Computational
Eccles, J.C., Libet, B., and Young, R.R. (1958). The behavior of Neuroscience Series (Cambridge, Massachusetts: MIT Press).
chromatolysed motoneurons studied by intracellular recording. J. Hermann, L. (1905). Beitrage zur Physiologie und Physiks des
Physiol. (Lond.) 143, 1140. Nerven. Pflugers Arch. 109, 95120.
Edwards, F.A., Konnerth, A., Sakmann, B., and Takahashi, T. (1989). Hillman, D., Chen, S., Aung, T.T., Cherksey, B., Sugimori, M., and
A thin slice preparation for patch recordings from neurones of mam- Llinas, R.R. (1991). Localization of P-type calcium channels in the
malian central nervous system. Pflugers Arch. 414, 600612. central nervous system. Proc. Natl. Acad. Sci. USA 88, 70767080.
Eilers, J., Augustine, G.J., and Konnerth, A. (1995). Subthreshold Hirsch, J.A., and Gilbert, C.D. (1991). Synaptic physiology of hori-
synaptic Ca2 1 signalling in fine dendrites and spines of cerebellar zontal connections in the cats visual cortex. J. Neurosci. 11, 1800
Purkinje neurons. Nature 373, 155158. 1809.
95
Neuron
714
Hirsch, J.A., Alonso, J.-M., and Reid, R.C. (1995). Visually evoked mammalian neurons: insights into central nervous system function.
calcium action potentials in cat striate cortex. Nature 378, 612616. Science 242, 16541664.
His, W. (1893). Uber den Aufbau unseres Nervensystems. Berl. Klin. Llinas, R., and Greenfield, S.A. (1987). On-line visualization of den-
Wochenschr. 40, 41. dritic release of acetylcholinesterase from mammalian substantia
nigra neurons. Proc. Natl. Acad. Sci. USA 84, 30473050.
Hodgkin, A.L., and Keynes, R.D. (1957). Movements of labelled cal-
cium in squid giant axons. J. Physiol. 138, 253281. Llinas, R., and Hess, R. (1976). Tetrodotoxin-resistant dendritic
spikes in avian Purkinje cells. Proc. Natl. Acad. Sci. USA 73, 2520
Hodgkin, A.L., and Rushton, W.A.H. (1946). The electrical constants
2523.
of a crustacean nerve fiber. Proc. R. Soc. Lond. (B) 133, 444479.
Llinas, R., and Nicholson, C. (1971). Electroresponsive properties of
Hopfield, J.J. (1995). Pattern recognition computation using action
dendrites and somata in alligator Purkinje cells. J. Neurophysiol. 34,
potential timing for stimulus representation. Nature 376, 3336.
532551.
Houchin, J. (1973). Procion Yellow electrodes for intracellular re-
Llinas, R., and Sugimori, M. (1980). Electrophysiological properties
cording and staining of neurons in the somatosensory cortex of the
of in vitro Purkinje cell dendrites in mammalian cerebellar slices. J.
rat. J. Physiol. 232, 6769.
Physiol. 305, 197213.
Huguenard, J.R., Hamill, O.P., and Prince, D.A. (1989). Sodium chan-
Llinas, R., and Sugimori, M. (1992). The electrophysiology of the
nels in dendrites of rat cortical pyramidal neurons. Proc. Natl. Acad.
cerebellar Purkinje cells revisited. In The Cerebellum Revisited, R.
Sci. USA 86, 24732477.
Llinas and C. Sotelo, eds. (New York: Springer-Verlag), pp. 167181.
Hume, R.I., and Purves, D. (1981). Geometry of neonatal neurons Llinas, R., and Yarom, Y. (1981). Properties and distribution of ionic
and the regulation of synaptic elimination. Nature 293, 469471. conductances generating electroresponsiveness of mammalian in-
Hyde, I.H. (1921). A micro-electrode and unicellular stimulation. Biol. ferior olivary neurones in vitro. J. Physiol. (Lond.) 315, 569584.
Bull. 40, 130133. Llinas, R., Nicholson, C., Freeman, J.A., and Hillman, D.E. (1968).
Jaffe, D.B., Johnston, D., Lasser-Ross, N., Lisman, J.E., Miyakawa, Dendritic spikes and their inhibition in alligator Purkinje cells. Sci-
H., and Ross, W.N. (1992). The spread of Na spikes determines the ence 160, 11321135.
pattern of dendritic Ca entry into hippocampal neurons. Nature 357, Llinas, R., Baker, R., and Sotelo, C. (1974). Electrotonic coupling
244246. between neurons in cat inferior olive. J. Neurophysiol. 37, 560571.
Katz, L.C., and Constantine-Paton, M. (1988). Relationships between Lorente de No, R., and Coundouris, G.A. (1959). Decremental con-
segregated afferents and postsynaptic neurons in the optic tectum duction in peripheral nerve integration of stimuli in the neuron. Proc.
of three-eyed frogs. J. Neurosci. 8, 31603180. Natl. Acad. Sci. USA 45, 592617.
Kim, H.G., and Connors, B.W. (1993). Apical dendrites of the neocor- Lo Turco, J.J., and Kriegstein, A.R. (1991). Clusters of coupled neu-
tex: correlation between sodium- and calcium-dependent spiking roblasts in embryonic neocortex. Science 252, 563566.
and pyramidal cell morphology. J. Neurosci. 13, 53015311. MacVicar, B.A., and Dudek, F.E. (1981). Electrotonic coupling be-
Kim, H.G., Beierlein, M., and Connors, B.W. (1995). Inhibitory control tween pyramidal cells: a direct demonstration in rat hippocampal
of excitable dendrites in neocortex. J. Neurophysiol. 74, 18101815. slices. Science 213, 782785.
Kisvarday, Z.F., Martin, K.A.C., Friedlander, M.J., and Somogyi, P. Magee, J.C., and Johnston, D. (1995a). Characterization of single
(1987). Evidence for interlaminar inhibitory circuits in the striate cor- voltage-gated Na1 and Ca21 channels in apical dendrites of rat CA1
tex of the cat. J. Comp. Neurol. 260, 119. pyramidal neurons. J. Physiol. 487, 6790.
Koch, C., and Zador, A. (1993). The function of dendritic spines: Magee, J.C., and Johnston, D. (1995b). Synaptic activation of volt-
devices subserving biochemical rather than electrical compartmen- age-gated channels in the dendrites of hippocampal pyramidal neu-
talization. J. Neurosci. 13, 413422. rons. Science 268, 301304.
Koch, C., Poggio, T., and Torre, V. (1983). Nonlinear interactions in Mainen, Z.F., and Sejnowski, T.J. (1995). Reliability of spike timing
a dendritic tree: localization, timing, and role in information pro- in neocortical neurons. Science 268, 15031506.
cessing. Proc. Natl. Acad. Sci. USA 80, 27992802. Major, G. (1993). Solutions for transients in arbitrarily branching
Kogan, A., Ross, W.N., Zecevic, D., and Lasser-Ross, N. (1996). cables. III. Voltage clamp problems. Biophys. J. 65, 469491.
Optical recording from cerebellar Purkinje cells using intracellularly Major, G., Evans, J.D., and Jack, J.B. (1993). Solutions for transients
injected voltage-sensitive dyes. Brain Res., in press. in arbitrary branching cables: I. Voltage recording with a somatic
Kolliker A. (1889). Handbuch der Gewebelehre (Leipzig). shunt. Biophys. J. 65, 423449.
Major, G., Larkman, A.U., Jonas, P., Sakmann, B., and Jack, J.J.
Kosaka, T. (1983). Neuronal gap junctions in the polymorph layer of
(1994). Detailed passive cable models of whole-cell recorded CA3
the rat dentate gyrus. Brain Res. 277, 347351.
pyramidal neurons in rat hippocampal slices. J. Neurosci. 14, 4613
Larkman, A.U. (1991). Dendritic morphology of pyramidal neurons 4638.
of the visual cortex of the rat. III. Spine distributions. J. Comp.
Markram, H., and Sakmann, B. (1994). Calcium transients in apical
Neurol. 306, 332343.
dendrites evoked by single sub-threshold excitatory post-synaptic
Laurent, G., and Burrows, M. (1989). Intersegmental interneurons potentials via low voltage-activated calcium channels. Proc. Natl.
can control the gain of reflexes in adjacent segments of the locust Acad. Sci. USA 91, 52075211.
by their action on nonspiking local interneurons. J. Neurosci. 9,
Matsumura, M., Cope, T., and Fetz, E.E. (1988). Sustained excitatory
30303039.
synaptic input to motor cortex neurons in awake animals revealed
Levy, W.B., and Desmond, N.L. (1985). Associative potentiation/ by intracellular recording of membrane potentials. Exp. Brain Res.
depression in the hippocampal dentate gyrus. In Electrical Activity 70, 463469.
of the Archicortex, G. Buzsaki and C.H. Vanderwolf, eds. (Budapest: Mel, B.W. (1994). Information processing in dendritic trees. Neural
Akademiai Kiado), pp. 359373. Comp. 6, 10311085.
Levy, W.B., and Steward, O. (1979). Synapses as associative mem- Michelson, H.B., and Wong, R.K.S. (1994). Synchronization of inhibi-
ory elements in the hippocampal formation. Brain Res. 175, 233245. tory neurons in the guinea pig hippocampus in vitro. J. Physiol. 477,
Ling, G., and Gerard, R.W. (1949). The normal membrane potential 3545.
of frog sartorius fibers. J. Cell Comp. Physiol. 34, 383396. Minski, M. (1961). Microscopy apparatus. US Patent 1961, 3013467.
Lisman, J. (1989). A mechanism for the Hebb and anti-Hebb pro- Montana, V., Farkas, D.L., and Loew, L.M. (1989). Dual-wavelength
cesses underlying learning and memory. Proc. Natl. Acad. Sci. USA ratiometric fluorescence measurements of membrane potential.
86, 95749578. Biochemistry 28, 45364539.
Llinas, R. (1988). The intrinsic electrophysiological properties of Muller, W., and Connor, J.A. (1991). Dendritic spines as individual
96
Review
715
neuronal compartments for synaptic Ca21 responses. Nature 354, to glutamate and kainate in neurons of the avian cochlear nucleus.
7376. Neuron 9, 173186.
Murphy, T.H., Baraban, J.M., Gil Wier, W., and Blatter, L.A. (1994). Ramon, P. (1890). Investigaciones de histologa comparada en los
Visualization of quantal synaptic transmission by dendritic calcium centros opticos de los vertebrados. PhD thesis, University of Madrid.
imaging. Nature 263, 529532. Ramon y Cajal, S. (1889). Conexion general de los elementos ner-
Nunez, A., Garca-Austt, E., and Buno, W. (1990). In vivo electrophys- viosos. Medicina Practica.
iological analysis of Lucifer Yellowcoupled hippocampal pyramids. Ramon y Cajal, S. (1891a). Significacion fisiologica de las expansi-
Exp. Neurol. 108, 7682. ones protoplasmicas y nerviosas de la sustancia gris. Rev. Ciencias
Nunez, A., Amzica, F., and Steriade, M. (1993). Electrophysiology of Med. 22, 23.
cat association cortical cells in vivo: intrinsic properties and synaptic Ramon y Cajal, S. (1891b). Sur la structure de lecorce cerebrale de
responses. J. Neurophys. 70, 418430. quelques mamiferes. Cellule 7.
Palay, S.L. (1956). Synapses in the central nervous system. J. Bio- Ramon y Cajal, S. (1897). Las leyes de la morfologa y dinamismo
phys. Biochem. Cytol. 2, 193201. de las celulas nerviosas. Revista Trim. Microgr. 1.
Peinado, A., Yuste, R., and Katz, L.C. (1993). Extensive dye coupling Ramon y Cajal, S. (1904). La Textura del Sistema Nerviosa del Hom-
between rat neocortical neurons during the period of circuit forma- bre y los Vertebrados (Madrid: Moya).
tion. Neuron 10, 103114. Ramon y Cajal, S. (1923). Recuerdos de mi Vida: Historia de mi
Peters, A., Palay, S.L., and Webster, H.D. (1991). The Fine Structure Labor Cientfica (Madrid: Alianza Editorial).
of the Nervous System (New York: Oxford University Press). Rapp, M., Segev, I., and Yarom, Y. (1994). Physiology, morphology
Phillips, C.G., Powel, T.P.S., and Shepherd, G.M. (1963). Responses and detailed passive models of guinea-pig Purkinje cells. J. Physiol.
of mitral cells to stimulation of the lateral olfactory tract in the rabbit. (Lond.) 474, 101118.
J. Physiol. 168, 6588. Regehr, W.G., and Tank, D.W. (1992). Calcium concentration dynam-
Pinsky, P.F., and Rinzel, J. (1994). Intrinsic and network rhythmogen- ics produced by synaptic activation of CA1 hippocampal pyramidal
esis in a reduced Traub model for CA3 neurons. J. Comput. Neurosci. cells. J. Neurosci. 12, 42024223.
1, 3960. Regehr, W.G., Connor, J.A., and Tank, D.W. (1989). Optical imaging
Pockberger, H. (1991). Electrophysiological and morphological of calcium accumulation in hippocampal pyramidal cells during syn-
properties of rat motor cortex neurons in vivo. Brain Res. 539, aptic activation. Nature 341, 533536.
181190. Regehr, W.G., Konnerth, A., and Armstrong, C.M. (1992). Sodium
Pollack, G.S. (1988). Selective attention in an insect auditory neuron. action potentials in the dendrites of cerebellar Purkinje cells. Proc.
J. Neurosci. 8, 26352639. Natl. Acad. Sci. USA 89, 54925496.
Regehr, W.G., Kehoe, J., Ascher, P., and Armstrong, C. (1993). Syn-
Powell, T.P.S. (1981). Certain aspects of the intrinsic organization
aptically triggered action potentials in dendrites. Neuron 11,
of the cerebral cortex. In Brain Mechanisms and Perceptual Aware-
145151.
ness, O. Pompeiano and C. Ajmone Marsan, eds. (New York: Raven
Press), pp. 119. Renshaw, B. (1942). Effects of presynaptic volleys on spread of
impulses over the soma of the motoneurons. J. Neurophysiol. 5,
Purpura, D.P. (1967). Comparative physiology of dendrites. In The
235243.
Neurosciences: A Study Program, G.C. Quarton, T. Melnechuk, and
F.O. Schmitt, eds. (New York: Rockefeller University Press), pp. Reuveni, I., Friedman, A., Amitai, Y., and Gutnick, M.J. (1993). Step-
372875. wise repolarization from Ca21 plateaus in neocortical pyramidal
cells: evidence for nonhomogeneous distribution of HVA Ca21 chan-
Purpura, D.P., and Shofer, R.J. (1964). Cortical intracellular poten-
nels in dendrites. J. Neurosci. 13, 46094621.
tials during augmenting and recruiting responses. I. Effects on in-
jected hyperpolarizing currents on evoked membrane potential Reyher, C.H.K., Lubke, J., Larsen, W.J., Hendrix, G.M., Shipley, M.T.,
changes. J. Neurophysiol. 27, 117132. and Baumgarten, H.G. (1991). Olfactory bulb granule cell aggre-
gates: morphological evidence for interperikaryal electrotronic cou-
Purves, D., and Hume, R.I. (1981). The relation of postsynaptic geom-
pling via gap junctions. J. Neurosci. 11, 14851495.
etry to the number of presynaptic axons that innervate autonomic
ganglion cells. J. Neurosci. 1, 441452. Rhodes, P.A., and Gray, C.M. (1994). Simulations of intrinsically
bursting neocortical pyramidal neurons. Neural Comp. 6, 10861110.
Rall, W. (1959). Branching dendritic trees and motoneuron mem-
Richardson, T.L., Turner, R.W., and Miller, J.J. (1987). Action-poten-
brane resistivity. Exp. Neurol. 1, 491527.
tial discharge in hippocampal CA1 pyramidal neurons. J. Neurophys.
Rall, W. (1964). Theoretical significance of dendritic trees for neu- 58, 981996.
ronal input-output relations. In Neural Theory and Modeling, R.F.
Rockland, K.S., and Virga, A. (1989). Terminal arbors of individual
Reiss, ed. (Stanford: Stanford University Press), pp. 7397.
feedback axons projecting from V2 to V1 in the macaque monkey:
Rall, W. (1974). Dendritic spines, synaptic potency and neuronal a study using immunocytochemistry of anterogradely transported
plasticity. In Cellular Mechanisms Subserving Changes in Neuronal Phaseolus vulgaris leucoagglutinin. J. Comp. Neurol. 285, 5472.
Activity, C.D. Woody, K.A. Brown, T.J. Crow, and J.D. Knispel, eds.
Rose, G.J., and Call, S.J. (1993). Temporal filtering properties of
(Los Angeles: Brain Information Services), pp. 1321.
midbrain neurons in an electric fish: implications for the function of
Rall, W. (1995). The Theoretical Foundation of Dendritic Function, dendritic spines. J. Neurosci. 13, 11781189.
I. Segev, J. Rinzel, and G.M. Shepherd, eds. (Cambridge, Massachu-
Ross, W.N., and Werman, R. (1987). Mapping calcium transients in
setts: MIT Press).
the dendrites of Purkinje cells from guinea-pig cerebellum in vitro.
Rall, W., and Segev, I. (1987). Functional possibilities for synapses J. Physiol. 389, 319336.
on dendrites and on dendritic spines. In Synaptic Function, G.E.
Ross, W.N., Lewenstein Stockbridge, L., and Stockbridge, N.L.
Edelman, W.F. Gall, and W.M. Cowan, eds. (New York: Wiley), pp.
(1986). Regional properties of calcium entry in barnacle neurons
605637.
determined with Arsenazo III and a photodiode array. J. Neurosci.
Rall, W., and Shepherd, G.M. (1968). Theoretical reconstruction of 6, 11481159.
field potentials and dendro-dendritic synaptic interactions in olfac- Rubel, E.W., and Parks, T.N. (1988). Organization and development
tory bulb. J. Neurophys. 31, 884915. of the avian brain-stem auditory system. In Auditory Function, G.M.
Rall, W., Shepherd, G.M., Reese, T.S., and Brightman, M.W. (1966). Edelman, W.E. Gall, and W.M. Cowan, eds. (New York: John Wiley
Dendrodenritic synaptic pathways for inhibition in the olfactory bulb. and Sons), pp. 393.
Exp. Neurol. 14, 4456. Sakmann, B., and Neher, E. (1983). Single Channel Recording (New
Raman, I.M., and Trussell, L.O. (1992). The kinetics of the response York: Plenum).
97
Neuron
716
Schadlen, M.N., and Newsome, W.T. (1994). Noise, neural codes Traub, R.D. (1996). Model of sustained population bursts in electri-
and cortical organization. Curr. Opin. Neurobiol. 4, 569579. cally coupled inerneurons containing active dendritic conductances.
Scharfman, H.E., and Sarvey, J.M. (1985). Postsynaptic firing during J. Comput. Neurosci., in press.
repetitive stimulation is required for long-term potentiation in hippo- Traub, R.D., Wong, R.K., Miles, R., and Michelson, H. (1991). A model
campus. Brain Res. 331, 267274. of a CA3 hippocampal pyramidal neuron incorporating voltage-
clamp data on intrinsic conductances. J. Neurophysiol. 66, 635650.
Schwartzkroin, P.A., and Slawsky, M. (1977). Probable calcium
spikes in hippocampus neurons. Brain Res. 135, 157161. Tsien, R.Y. (1989). Fluorescent probes of cell signaling. Annu. Rev.
Neurosci. 12, 227253.
Shelton, D.P. (1985). Membrane resistivity estimated for the Purkinje
Turner, R.W., Meyers, E.R., Richardson, D.L., and Barker, J.L. (1991).
neuron by means of a passive computer model. Neuroscience 14,
The site for initiation of action potential discharge over the somato-
111131.
sensory axis of rat hippocampal CA1 pyramidal neurons. J. Neu-
Shepherd, G.M., and Brayton, R.K. (1987). Logic operations are rosci. 11, 22702280.
properties of computer-simulated interactions between excitable
Vaney, D.I. (1991). Many diverse types of retinal neurons show tracer
dendritic spines. Neuroscience 21, 151165.
coupling when injected with biocytin or neurobiotin. Neurosci. Lett.
Shepherd, G.M., Brayton, R.K., Miller, J.P., Segev, I., Rinzel, J., and 125, 187190.
Rall, W. (1985). Signal enhancement in distal cortical dendrites by
Van Gehuchten, A. (1891). La moelle epiniere et le cervelet. Cellule
means of interactions between active dendritic spines. Proc. Natl.
7.
Acad. Sci. USA 82, 21922195.
Watson, A.H.D., and Burrows, M. (1988). Distribution and morphol-
Sloper, J.J. (1972). Gap junctions between dendrites in the primate ogy of synapses on nonspiking local interneurons in the thoracic
neocortex. Brain Res. 44, 641646. nervous system of the locust. J. Comp. Neurol. 272, 605616.
Sobel, E.C., and Tank, D. W. (1994). In vivo Ca21 dynamics in a cricket Westenbroek, R.E., Ahlijanian, M.K., and Catterall, W.A. (1990). Clus-
auditory neuron: an example of chemical computation. Science 263, tering of L-type Ca2 1 channnels at the base of major dendrites in
823826. hippocampal pyramidal neurons. Nature 347, 281284.
Softky, W.R., and Koch, C. (1992). The highly irregular firing of corti- Westenbroek, R.E., Hell, J.W., Warner, C., Dubel, S.J., Snutch, T.P.,
cal cells is inconsistent with temporal integration of random EPSPs. and Catterall, W.A. (1992). Biochemical properties and subcellular
J. Neurosci. 13, 334350. distribution of an N-type calcium channel a1 subunit. Neuron 9,
Sotelo, C., Llinas, R., and Baker, R. (1974). Structural study of inferior 10991115.
olivary nucleus of the cat: morphological correlates of electrotonic Wickens, J. (1988). Electrically coupled but chemically isolated syn-
coupling. J. Neurophysiol. 37, 541559. apses: dendritic spines and calcium in a rule for synaptic modifica-
tion. Prog. Neurobiol. 31, 507528.
Spencer, W.A., and Kandel, E.R. (1961). Electrophysiology of hippo-
campal neurons: IV. Fast prepotentials. J. Neurophysiol. 24, Wilson, T., and Sheppard, C. (1984). Theory and Practice of Scanning
272285. Optical Microscopy (New York: Academic Press).
Wong, R.K.S., and Prince, D.A. (1978). Participation of calcium
Spruston, N., Jaffe, D.B., and Johnston, D. (1994). Dendritic attenua-
spikes during intrinsic burst firing in hippocampal neurons. Brain
tion of synaptic potentials and currents: the role of passive mem-
Res. 159, 385390.
brane properties. Trends Neurosci. 17, 161166.
Wong, R.K.S., and Steward, M. (1992). Different firing patterns gen-
Spruston, N., Schiller, Y., Stuart, G., and Sakmann, B. (1995). Activ-
erated in dendrites and somata of CA1 pyramidal neurones in
ity-dependent action potential invasion and calcium influx into hip-
guinea-pig hippocampus. J. Physiol. 457, 675687.
pocampal CA1 dendrites. Science 286, 297300.
Wong, R.K.S., Prince, D.A., and Basbaum, A.I. (1979). Intradendritic
Stafstrom, C.E., Schwindt, P.C., Chabb, M.C., and Crill, W.E. (1985).
recordings from hippocampal neurons. Proc. Natl. Acad. Sci. USA
Properties of persistent sodium and calcium conductances of layer
76, 986990.
5 neurons from cat sensory motor cortex in vivo. J. Neurophysiol.
Wu, K., Sachs, L., Carlin, R.K., and Siekevitz P. (1986). Characteris-
55, 153170.
tics of a Ca21/calmodulin-dependent binding of the Ca2 1 channel
Stratford, K., Mason, A., Larkman, A., Major, G., and Jack, J.J. (1989). antagonists nitrendipine, to a postsynaptic density fraction isolated
The modelling of pyramidal neurons in the visual cortex. In The from canine cerebral cortex. Mol. Brain Res. 1, 176184.
Computing Neuron, R. Durbin, C. Miall, and G. Mitchinson, eds.
Yamamoto, C. (1972). Activation of hippocampal neurons by mossy
(Workingham, England: Addison-Wesley), pp. 296322.
fiber stimulation in thin brain sections in vitro. Exp. Brain Res. 14,
Stuart, G.J., and Hausser, M. (1994). Initiation and spread of sodium 423435.
action potentials in cerebellar Purkinje cells. Neuron 13, 703712. Yang, G., and Masland, R.H. (1992). Direct visualization of the den-
Stuart, G.J., and Sakmann, B. (1994). Active propagation of somatic dritic and receptive fields of directionally selective retinal ganglion
action potentials into neocortical pyramidal cell dendrites. Nature cells. Science 258, 19491952.
367, 6972. Yuste, R., and Denk, W. (1995). Dendritic spines as basic units of
Stuart, G.J., Dodt, H.-U., and Sakmann, B. (1993). Patch clamp re- synaptic integration. Nature 375, 682684.
cording from the soma and dendrites of neurons in brain slices using Yuste, R., Peinado, A., and Katz, L.C. (1992). Neuronal domains in
infrared video microscopy. Pflugers Arch. 423, 511518. developing neocortex. Science 257, 665669.
Sugihara, I., Lang, E.J., and Llinas, R. (1993). Uniform olivocerebellar Yuste, R., Gutnick, M.J., Saar, D., Delaney, K.D., and Tank, D.W.
conduction time underlies Purkinje cell complex spike synchronicity (1994). Calcium accumulations in dendrites from neocortical neu-
in the rat cerebellum. J Physiol. (Lond.) 470, 243271. rons: an apical band and evidence for functional compartments.
Neuron 13, 2343.
Sutor, B., and Hablitz, J.J. (1989). EPSPs in rat neocortical neurons
in vitro. II. Involvement of N-methyl-D-aspartate receptors in the Yuste, R., Nelson, D., Rubin, W., and Katz, L.C. (1995). Neuronal
generation of EPSPs. J. Neurophysiol. 61, 621634. domains in developing neocortex: mechanisms of coactivation.
Neuron 14, 717.
Tank, D.W., Sugimori, M., Connor, J.A., and Llinas, R.R. (1988). Spa-
tially resolved calcium dynamics of mammalian Purkinje cells in Zeki, S. (1993). The modularity of the brain. In A Vision of the Brain,
cerebellar slice. Science 242, 773777. S. Zeki, ed. (Oxford: Blackwell), pp. 197296.
Zohary, E., Shadlen, M.N., and Newsome, W. T. (1994). Correlated
Thompson, A.M., Girdlestone, D., and West, D.C. (1988). Voltage-
neuronal discharge rate and its implications for psychophysical per-
dependent currents prolong single-axon postsynaptic potentials in
formance. Nature 370, 140143.
layer III pyramidal neurons in rat neocortical slices. J. Neurophysiol.
60, 18961907.
98
REPORTS
References and Notes 14. S. Ge et al., Nature 439, 589 (2006). 30. We thank D. Ginty, S. Synder, and members of Ming and
1. G. Kempermann, H. van Praag, F. H. Gage, Prog. Brain 15. S. G. Jin, C. Guo, G. P. Pfeifer, PLoS Genet. 4, e1000013 Song laboratories for help and critical comments and
Res. 127, 35 (2000). (2008). L. Liu and Y. Cai for technical support. This work was
2. G. L. Ming, H. Song, Annu. Rev. Neurosci. 28, 223 16. T. Aid, A. Kazantseva, M. Piirsoo, K. Palm, T. Timmusk, supported by NIH, McKnight, and NARSAD (to H.S.) and
(2005). J. Neurosci. Res. 85, 525 (2007). by NIH, March of Dimes, and Johns Hopkins Brain
3. T. M. Madsen et al., Biol. Psychiatry 47, 1043 (2000). 17. K. Y. Alam et al., J. Biol. Chem. 271, 30263 (1996). Science Institute (to G-l.M.). R.A.F. is an investigator with
4. R. Jaenisch, A. Bird, Nat. Genet. 33 (suppl.), 245 (2003). 18. Y. Zhang, F. Madiai, K. V. Hackshaw, Biochim. Biophys. the Howard Hughes Medical Institute.
5. Materials and methods and supporting data are available Acta 1521, 45 (2001).
on Science Online. 19. N. Cervoni, M. Szyf, J. Biol. Chem. 276, 40778 (2001).
20. M. Gehring et al., Cell 124, 495 (2006). Supporting Online Material
6. J. E. Ploski, S. S. Newton, R. S. Duman, J. Neurochem. 99,
21. R. Mtivier et al., Nature 452, 45 (2008).
www.sciencemag.org/cgi/content/full/1166859/DC1
1122 (2006).
7. H. J. Jung et al., Oncogene 26, 7517 (2007). 22. R. Holliday, J. Theor. Biol. 200, 339 (1999). Materials and Methods
8. H. Tran et al., Science 296, 530 (2002). 23. K. Martinowich et al., Science 302, 890 (2003). SOM Text
9. M. C. Hollander, A. J. Fornace Jr., Oncogene 21, 6228 24. I. C. Weaver et al., Nat. Neurosci. 7, 847 (2004). Figs. S1 to S19
(2002). 25. F. D. Lubin, T. L. Roth, J. D. Sweatt, J. Neurosci. 28, Tables S1 and S2
References
10. B. Lu, A. F. Ferrandino, R. A. Flavell, Nat. Immunol. 5, 38 10576 (2008).
(2004). 26. K. Garbett et al., Neurobiol. Dis. 30, 303 (2008).
11. G. Barreto et al., Nature 445, 671 (2007). 27. V. M. Porterfield, H. Piontkivska, E. M. Mintz, BMC 6 October 2008; accepted 16 December 2008
12. V. Ramirez-Amaya et al., J. Neurosci. 25, 1761 (2005). Neurosci. 8, 98 (2007). Published online 1 January 2009;
13. S. W. Flavell, M. E. Greenberg, Annu. Rev. Neurosci. 31, 28. D. Hevroni et al., J. Mol. Neurosci. 10, 75 (1998). 10.1126/science.1166859
563 (2008). 29. M. Majdan, C. J. Shatz, Nat. Neurosci. 9, 650 (2006). Include this information when citing this paper.
M
osquito-borne diseases such as ma- ulate their 300- to 500-Hz wing beat frequencies monic (F1) = 1238.3 T 31.0 (SEM) Hz] matched
laria, yellow fever, and dengue con- to match each other (18). Thus, acoustically me- the females third harmonic [F0 = 430.6 T 10.8,
tinue to afflict millions, even after diated mate attraction involves active modulation F2 = 1356.2 T 29.2 (SEM) Hz] (Fig. 1, A to C).
decades of work to control vector populations. by both sexes, creating a duet. The period of synchronization lasted an average
Despite this effort, basic aspects of mosquito bi- We show that males and females of the den- of 9.71 T 1.05 (SEM) s with the synchroniza-
ology are not fully understood, including mating gue and yellow fever vector A. aegypti also mod- tion frequency averaging 1354.5 T 31.5 (SEM)
behavior, an important target for vector control. ulate their flight tones when brought within a few Hz. A. aegypti do not shift their flight tones in
We describe investigations in Aedes aegypti that centimeters of each other. This modulation, how- the absence of acoustic stimulation, as tested both
require revision of the current understanding of ever, does not match the fundamental wing beat by deafening the mosquitoes [and stimulating
mosquito mating behavior. Since Johnston (1) frequency of around 400 Hz (female) or around with tones, Fisher exact test, males P = 0.02,
first suggested in 1855 that mosquitoes could 600 Hz (male) but a shared harmonic frequency females P = 0.04 (21)] and flying intact control
perceive sound, over 14 studies have been pub- of around 1200 Hz (Fig. 1). Consistent with this, subjects in silence [Fishers exact test, males P =
lished on sound production and hearing in A. a neurophysiological examination of the ears of 0.02, females P = 0.04 (21)].
aegypti (217) (table S1). The buzz of a flying A. aegypti shows response in both males and fe- The presence of the fundamental frequency
female mosquito acts as a mating signal, attracting males up to 2000 Hz (Fig. 2). These results are tone was not necessary for harmonic matching.
males. Typically, the behaviorally salient frequen- unexpected because over 5 decades of behavioral We stimulated tethered mosquitoes with electron-
cy component of flight tone is the fundamental and physiological studies had concluded that male ically generated pure sinusoidal tones as well as
frequency of wing beat, which is between 300 to mosquito ears (antennae and associated Johnstons with harmonic combinations of pure tones lack-
600 Hz depending on species (8). However, mate organ) are tuned to 300 to 800 Hz and deaf to ing the fundamental frequency (Fig. 1, D to F).
attraction is not simply a matter of a male pas- frequencies above 800 Hz (8, 19). The present The intensity of the pure tones was set at a particle
sively hearing and homing in on a 400-Hz tone. study also directly addresses the issue of audi- velocity of 0.024 mm/s corresponding to 54 dB
For example, males and females of the nonblood- tory competence in female mosquitoes. Acoustic sound pressure level (relative to 20 mPa) when
feeding mosquito Toxorhynchites brevipalpis mod- duetting behavior in the nonvector mosquito played through an ear bud speaker positioned 1.5
1
T. brevipalpis (18) would seem to imply active cm in front of the test mosquito. This intensity
Department of Entomology, Cornell University, Ithaca, NY audition in both sexes, and laser vibrometry is well within the response range of A. aegyptis
14853, USA. 2Department of Neurobiology and Behavior,
Cornell University, Ithaca, NY 14853, USA. studies of the Johnstons organ in that species Johnstons organ, as measured by Doppler
*These authors contributed equally to this work.
and A. aegypti (16, 17) indicate that they re- vibrometry (17). Stimuli were played in 10- to 15-s
To whom correspondence should be addressed. E-mail: spond mechanically to salient sounds. Moreover, bursts with 5- to 20-s recovery periods. Play-
rrh3@cornell.edu female frog-biting mosquitoes are reported to be back experiments with pure-tone combinations
100
REPORTS
A B 3. P. Belton, in Insect Flight: Dispersal and Migration,
1200 Hz W. Danthanarayana, Ed. (Springer-Verlag, Berlin, 1986),
SD F1 pp. 6070.
F0 F2 4. P. Belton, Can. J. Zool. 67, 2625 (1989).
F3 5. P. Belton, J. Am. Mosq. Control Assoc. 10, 297 (1994).
SD F4 6. P. Belton, R. A. Costello, Entomol. Exp. Appl. 26, 105 (1979).
0.2 mV
102
J Comp Physiol A (2003) 189: 245256
DOI 10.1007/s00359-003-0406-2
K AR L V ON FR I SCH LE C TU R E
G. Neuweiler
Received: 9 September 2002 / Revised: 17 February 2003 / Accepted: 18 February 2003 / Published online: 28 March 2003
Springer-Verlag 2003
103
246
conserved and is uniform in all mammals including bats lemniscus (VNLL and INLL) are unusually large and
and man. This is not the case for the neuronal aspects of conspicuously organized (Fig. 1): the VNLL is dieren-
audition. tiated into a multipolar and a columnar section. Colum-
nar neurons receive tonotopically arranged inputs via
large calyx-synapses on their cell bodies. This unique
Adaptations to echolocation columnar organization of neurons with calyx-synapses is
also found in echolocating dolphins. The broadly tuned
Neuroanatomical specializations in echolocating bats neurons of the columnar VNLL process specically fre-
quency-modulated (FM) components of echolocation
In echolocating bats, the lower brainstem nuclei of the signals. Columnar neurons are intensity invariant and
ascending auditory pathway, including the inferior colli- respond with one single spike at constant latencies over a
culus of the midbrain, are hypertrophied and some of wide range of intensities and frequencies (Fig. 1). Thus,
them are dierentially structured (Covey and Casseday they are ideally suited as time markers. Except for the few
1995). First, the anteroventral cochlear nucleus (AVCN) cases mentioned above the specic auditory implications
is extremely large. In Pteronotus, there is a unique group for echolocation of these neuroanatomical specialities are
of large multipolar cells in a marginal zone. Second, un- not yet known.
like in other mammals, there is an additional, direct input
from the posterior ventral cochlear nucleus (PVCN) to
the superior olivary complex (SOC). Third, in non- Echo suppression in non-echolocating mammals
echolocating mammals, the medial superior olive (MSO)
receives identical inputs from ipsi- and contralateral In natural situations a human subject or an animal will
AVCN. This is considered a prerequisite for processing not only listen to a sound source but also to its multiple
interaural time dierences. In echolocating bats, howev- reections (clutter) from reverberant surfaces. In most
er, the ipsilateral input is sparse, and in one bat species, experimental studies the natural clutter is simplied into
Pteronotus parnellii, the large majority of MSO-neurons a singular lagging reection following a leading direct
only receives monaural input from the contralateral sound (Litovsky et al. 1999). The auditory system en-
AVCN (Grothe and Neuweiler 2000). Finally, the mon- sures that reections fuse with the sound source to one
aural ventral and intermediate nuclei of the lateral percept. Such perceptual fusion occurs for delays up to
104
247
5 ms between leading and lagging stimulus. When the to echoes returning from objects up to a distance of
delay increases to 810 ms for brief sounds, e.g. for 10 m, and will exclude echoes from targets further away.
clicks, a human listener begins to hear two separate This limitation nicely matches the foraging behaviour
signals. This critical delay is called echo threshold. Ap- of horseshoe bats which preferably utilize a sit-and-wait
parently, for brief delays perception of echoes as distinct strategy when foraging. From a vantage point within
signals is suppressed, and the leading signal dominates vegetation horseshoe bats initiate pursuit of ying in-
the heard sound image. It is hypothesized that a neural sects passing by at distances of not more than about 7 m
gate closes after the rst elicited spike, and reopens after (Neuweiler et al. 1987).
a few milliseconds (echo threshold). The gate could be Unfortunately, recordings in the ascending auditory
implemented by self- or lateral inhibition (Litkovsky pathway of bats that actively emit echolocation signals
et al. 1999). In cats (Litovsky and Yin 1998), rabbits have been discontinued. Thus, the few IC units reported
(Fitzpatrick et al. 1995) and barn owls (Keller and by Schuller (1979) are the only evidence that auditory
Tagahashi 1996) a majority of single units from the units sensitized for echoes by a corollary motor com-
inferior colliculus (IC) showed lag suppression within mand may exist. These sparse neural data are, however,
variable time delays. In these samples the median time corroborated by comparative behavioural studies
delay for lag suppression matches the behaviourally (Roverud and Grinnell 1985; Roverud 1993). Echolo-
measured echo thresholds. However, since in newborn cating bats that were trained for a distance-discrimina-
cats lag suppression is present in the IC, but not tion task failed when they were jammed by intense
behaviourally, it is also suggested that cortical events playbacks of the species echolocation sounds. Distance
may be involved (Litovsky 1998). discrimination could only be disrupted by species-specic
Echo suppression builds up even more when trains of echolocation sounds or signals carrying specic elements
lead-lag signal pairs are heard, and echo thresholds rise of the echolocation sound. Jamming was time-locked to
by several milliseconds. Echo thresholds additionally the beginning of vocalizations, and lost its eectiveness
increase when each leading signal is succeeded not by beyond echo delays of more than 30 ms. These results
one but by several echoes of various time delays (Yost from dierent bat species strongly suggest that a time
and Guzman 1996). window of enhanced echo processing locked to sound
Since echolocating bats listen to trains of lead signals emission indeed exists.
(vocalized echolocation sound) followed each by several
highly correlated but non-identical lag signals (echoes),
they should be heavily aicted by echo suppression Echo sensitivity by combination sensitive neurons
mechanisms as described in humans and a few experi-
mental mammals. A bat would be neuronally deaf to n contrast to the tiny sample of vocalization-triggered
echoes returning within about 2.5 ms, i.e. reected from auditory echo units, another type of echo-sensitive
objects up to at least 40 cm away. neuron has been extensively described in large portions
105
248
of the IC and higher auditory centres: so called com- FM echo-sensitive neurons should be active when the
bination-sensitive neurons. These neurons preferably or bat closes in on a target, for instance when pursuing
exclusively answer to a combination of two auditory prey at distances shorter than about 2 m. Due to its
stimuli (Fig. 3a), e.g. an initial loud FM signal that specic best delays each FM/FM unit will respond only
corresponds to an emitted echolocation signal, and to an echo returning from the appropriate distance;
after a specic delay a fainter second FM signal that hence, these neurons have been also called range-
corresponds to an echo (FM/FM neurons). The so- nding neurons (ONeill and Suga 1982).
called best delay between rst and second signal to According to a recent study in the nuclei of the lateral
which the neuron responds best is specic for each lemniscus (Portfors and Wenstrup 2001), combination
neuron and ranges between 0 and 20 ms with a ma- sensitivity seems to be generated in the IC where about
jority of best delays below 10 ms (Fig. 3b, c; Portfors 75% of units recorded are combination sensitive; of these
and Wenstrup 1999). In terms of a foraging, echolo- 70% are facilitated and 30% inhibited by the rst signal.
cating bat these data indicate that most of these FM/ Echo-sensitive neurons often respond to a rst sig-
nal mimicking specic parameters of an emitted echo-
location sound only after long latencies of ca. 13 ms
Fig. 3ac Combination-sensitivity in the auditory cortex of compared to ca. 7 ms for the second signal (echo).
horseshoe bats. a A FM1/FM2 neuron that only responds to a These long latencies to the rst stimulus may result
combination of the FM elements of the rst harmonic (FM1 marks from inhibition initiated by the beginning of the rst
the emitted echolocation sound) and the second harmonic (FM2 stimulus. These ndings support the hypothesis that
marks the returning echo) with a characteristic delay (3 ms). Upper
and lower graph: CF constant-frequency component. The neuron echo-sensitive neurons are coincidence detectors (Olsen
was stimulated by a CF/FM signal; however, it only responded to and Suga 1991; Portfors and Wenstrup 2001). Coinci-
the FM component. b Delay-tuning curves of ve dierent units in dence is most probably achieved by rebound from in-
the FM/FM area of the auditory cortex. c Arrangement of best hibition elicited by the emitted sound. Duration of
delays of FM/FM units in a rostro-caudal chronotopic order. Bold
line: units with best delays between 2 and 3 ms, corresponding to inhibition varies and corresponds to the best delay of
target distances of 3552 cm, are overrepresented. After Schuller the units. Combination sensitivity disappears when in-
et al. (1991) hibition is blocked in the inferior colliculus.
106
249
Combination-sensitive neurons that are inhibited will direction. Electrical stimulation of cortical FM/FM
be silent during sound emission; however, by rebound neurons enhances the auditory responses of subcortical
from inhibition they may feed additional excitation into combination-sensitive neurons matched in best delay to
facilitated FM/FM neurons and thus enhance echo- the activated cortical neurons, and sharpens delay tuning
sensitivity (Portfors and Wenstrup 2001). without shifting their best delays, whereas it suppresses
FM/FM or other echo-sensitive neurons belong to the auditory responses and shifts the best delays of un-
the broad class of combination-sensitive neurons that matched subcortical delay-tuned neurons (Yan and Suga
have been also recorded in higher auditory centres of 1996).
non-echolocating mammals and of birds. Combination- An earlier study (Kossl and Vater 1989) has demon-
sensitive neurons respond best to specic temporal re- strated that norepinephrine markedly diminishes latency
lationships between spectrally separate sound elements, jitters in units of cochlear nuclei. This eect was inter-
and are considered to be coders of species-specic preted as a more precise coding of echo timings in alert
communication signals. A neuron is classied as com- bats due to the attentive eect of norepinephrinergic
bination sensitive when its response to the combined innervation from locus coeruleus.
stimuli is dierent by at least 20% from the algebraic Thus, corticofugal modulation activated by echoes,
sum of responses to each separate element of the and general modulatory mechanisms of attention such
combined stimulus (Portfors and Wenstrup 1999). as norepinephrinergic or cholinergic inputs from brain
Apparently, combination sensitivity is a general fea- stem nuclei primarily serve to focus attention and per-
ture of auditory reception in all mammals and birds, and ceptive power to a sector of the outer world that is of
not specic to echolocating animals. In evolution, echo momentary behavioural interest. Auditory sensitivity to
sensitivity, and hence, echolocation may have been es- echoes will be enhanced, and frequency and time lters
tablished more by quantitative than by qualitative will be quickly shaped to actual requirements. Top
modications: combination-sensitive neurons are found down modulation may prove to be a powerful way to
in very large aggregations in midbrain and higher au- adjust the auditory system to actual demands of audi-
ditory centres of bats, whereas in non-echolocating tory cognition in echolocation. As in vision, it might also
mammals and birds combination sensitivity is far less provide learned cognitive templates stored in memory
frequently recorded and is narrowly conned to specic from previous experiences in echolocation. Therefore,
subregions of auditory brain centres. Transformation of studies in topdown modulation in the neural auditory
combination sensitivity from coding communication pathway (Yan and Suga 1996) will become a central
sounds to echo coding also implies that temporal rela- issue in echolocation research.
tionships are shortened from hundreds to only a few Echolocation is a neuronal and not a cochlear
milliseconds, and neuronal spectral lters are tuned to achievement. Mechanisms involved in generating echo-
specic echolocation sound elements. sensitivity are still under discussion and include:
It is unlikely that combination sensitivity in echolo-
cating bats is achieved by the reversal of echo-suppres- 1. Time windows of echo-sensitivity triggered by vocal
sion mechanisms in non-echolocating mammals from an command centers (Fig. 2).
inhibitory to a facilitatory mode, since combination 2. Combination sensitivity (Fig. 3) derived from coding
sensitivity and echo suppression coexist in the auditory species-specic communication signals.
system of non-echolocating mammals. 3. Cortical topdown modulation.
Locking time windows of enhanced audition to neu-
ronal commands of sound emission would be the perfect In my opinion, echolocation phylogenetically was an
way to generate echo sensitivity. Only one neural study adaptation of already existing mechanisms in auditory
(Schuller 1979) supports such a mechanism. Apparently, neuronal processing. In pre-bats, the transition from
listening to the tight time relationship between heard neural combination circuits for sound communication to
emitted sound and returning echoes is good enough to those for echolocation was probably elicited by noctur-
generate echo sensitivity by modifying combinatory nal pursuits of small ying insects. The capacity of ight
neuronal circuits common to neuronal coding of species- in turn may have evolved from small insectivorous and
specic communication signals in all mammals and in arboreal mammals catching nocturnal insects while
birds (Margoliash and Fortune 1992; Gehr et al. 2000). running along branches and hopping from twig to twig
in trees (see also Simmons and Stein 1980; Padian 1985).
However, in another line of thinking it is supposed that
Topdown modulation echolocation evolved in cave dwelling animals.
107
250
Yet, ight combined with echolocation was, and still is, FM echolocation
so advantageous to a nocturnal mammal that bats
radiated into many ecological niches. Microchiroptera Most bat species pursue insects on the wing. During the
adapted echolocation systems, wing shapes and ight approach to and capture of prey these bats emit
styles (Rayner 1991) to all possible food sources. Some sequences of brief (several milliseconds) downward FM
species pursue ying insects at high speeds, others pick sounds. FM signals serve as good time-markers in the
up insect prey from the ground, tree bark and twigs, and ascending auditory pathway, and hence are thought to
still others search for frogs, lizards, mice and birds. guarantee a precise distance perception in echolocation.
Among the neotropical phyllostomids there is a large
group of bats species specialized for visiting nocturnal
Constant-frequency/frequency-modulated echolocation
owers for pollen and nectar, or for collecting fruits. The
infamous vampire bats can only subsist on blood from
The Old World families of horseshoe bats (rhinolophids)
mammals and birds. There is no other mammalian order
and hipposiderids, and one neotropical species, Pteron-
that has tapped such a rich variety of food resources.
otus parnellii, emit a combined echolocation sound
Bat species may be dierentiated and identied by the
consisting of a rather long (860 ms) constant frequency
frequency range and the time structure of their echolo-
(CF) invariably terminated by a brief FM element. In
cation calls. These species-specic echolocation systems
many species the echolocation sounds start with a brief
may be grouped under three broad categories (Fig. 4;
frequency upward-modulated element that is especially
Neuweiler 1984).
pronounced in commuting ights of the bats (Neuweiler
et al. 1987). Thus, the correct signature for these echo-
location systems should be FM/CF/FM. However, since
Fig. 4 Echolocation systems in bats. CF/FM: horseshoe bats, no specic function could be assigned to the initial FM
hipposiderids, and the neotropic moustached bat emit multihar- component this designation has not been endorsed.
monic echolocation signals consisting of a long pure tone CF/FM echolocation sounds consist of two or more
terminated by a brief FM component. CF/FM bats often forage harmonics. In the moustached bat and in horseshoe bats
within or close to dense vegetation FM. Most insectivorous bats
foraging on the wing emit brief, downward-FM echolocation
the second harmonic of the CF component (CF2 in
signals when approaching and catching a prey (FM). They often Fig. 4) is the most intense element of the echolocation
emit longer and only shallowly modulated signal when they search signal to which audition is specically adapted.
for prey (not shown) Click-like: gleaning bats and some ower CF/FM bats preferably, but in no way exclusively,
visiting bats emit very brief signals over a broad frequency band forage wing-beating insects in dense vegetation. They
(click-like). These echolocation sounds are far less intense than
those of FM and CF/FM bats possess an auditory fovea in the cochlea and compensate
108
251
Doppler shifts of complete echo signals in order to keep reection of the impinging sound waves when the
audition of CF2 echoes within an auditory fovea (see moving wing passes through a position perpendicular to
below). the incoming sound waves. Behavioural experiments
testied (Link et al. 1986) that horseshoe bats are indeed
specialized to wing beat detection, and deaf to non-ying
Click-like echolocation insects. Insects sitting or moving on a substrate do not
elicit attacks of horseshoe bats even at close distances.
A number of bat species that pick up prey from sub- Apparently, the long CF component of the echolocation
strates or visit owers and fruiting vegetation emit faint sound serves as a carrier for glints induced by uttering
and very brief signals of submillisecond durations. The prey. However, the modulation depths of such glints
ultra-brief sounds consist of several shallowly FM har- amounts to less than 0.25% of the carrier frequency, for
monics that add up to a click-like, broadband signal instance to maximally 185 Hz on a CF component of
covering a frequency band of up to 80100 kHz. Based 75 kHz emitted by a rufous horseshoe bat.
on behavioural experiments it has been suggested that In order to detect even very tiny glints CF/FM bats
broad-band signals may serve to dierentiate textures of have implemented a very narrow cochlear frequency
targets by the interference pattern (colouration) reected lter tuned to the species-specic and even individual
in broad band echoes (Schmidt 1988; Neuweiler and frequency of the CF2 echo component (Fig. 5). We have
Schmidt 1993). called this individually tuned frequency lter an auditory
In ethoecological studies of neotropical and European fovea.
bat communities the correlation between foraging niches
and type of echolocation signals has been subdivided in a
much more detailed way (Schnitzler and Kalko 1998; Mechanisms of auditory foveae
Kalko and Schnitzler 1998).
Structural and functional specialization that produce the
extremely narrow frequency lter of the auditory fovea
Perfect adaptation by auditory specialization: have been most thoroughly studied in Pteronotus
a case study in CF/FM bats parnellii (Henson and Henson 1991; Russell and Kossl
1999; for a review see Kossl and Vater 1995) because in
Auditory adaptations to the three categories of echolo- the moustached bat the cochlea is experimentally more
cation have not been systematically studied with one accessible than in horseshoe bats. The functional prin-
notable and striking exception, the CF/FM bats, i.e. the ciples of an auditory fovea discovered in the moustached
moustached bat (Pteronotus parnellii) and the rufous bat are assumed to apply with moderate modications
horseshoe bat (Rhinolophus rouxi). Over the last decades to all CF/FM bats (rhinolophids and hipposiderids).
these two bat species have become case studies of A complete and precise cochlear frequency map in
sophisticated adaptation of an auditory system to a moustached bats (Fig. 6c) has been obtained on the
specic foraging strategy. basis of an inner hair cell frequency map produced by
dye-labelling auditory neurons characterized by their
best frequency (Kossl and Vater 1985).
Foraging behaviour The foveal frequency lter in the inner ear of the
moustached bat is housed in the extended, large basal
The foraging behaviour of the rufous horseshoe bat cochlear turn and consists of two parts (Fig. 6a): (1) a
exemplies foraging strategies in CF/FM bats (Neuwe- basal sparsely innervated (SI) zone characterized by
iler et al. 1987). When these bats leave their caves in the structural specializations of the basilar and tectorial
evening they immediately take cover within bushes and membrane. In the SI region the basilar membrane is
commute under the canopy of trees and bushes into the tuned from 72 to 62 kHz. The frequency of about
jungle. After a brief period of foraging on the wing the 62 kHz (CF2 frequency), is represented at the transition
bats spend the night foraging in a sit-and-wait strategy. of the SI zone to (2) an adjacent long section of the
Each bat occupies its own foraging area and suspends basilar membrane (CF2 zone) where the narrow fre-
itself from slender twigs. The bat scans the surrounding quency band of the CF2 echo and its glints induced by
by emitting continuous sequences of CF/FM signals wing-beating insects are represented in a widely
throughout the night. Whenever an echolocation sound expanded way.
hits a wing-beating insect the bat takes o for a catch In the SI zone, the thickness of the basilar membrane
and returns with its prey to the very same spot or nearby is markedly increased by longitudinal bres that provide
twigs. The horseshoe bats detect the ying insects mechanical coupling over its length. The tectorial
against the dense echo clutter reverberated from foliage membrane is also highly modied, and shows a beam-
by echo-glints imposed on the long pure tone component like structure (club-shaped in cross-sections) that may
(CF) of the echo by each wing beat of the target. Echo- easily vibrate since it is only loosely attached to the
glints are produced by frequency Doppler shifts induced spiral limbus. Both structural specializations in the
by the speed of the insect wing, and by an intense mirror basilar and tectorial membrane abruptly disappear at
109
252
Fig. 6ac Auditory fovea: the specialized basal turn of the cochlea
in moustached bats. a A camera lucida drawing of the basal turn.
Dashed lines with arrows demarcate the specialized zones SI
(sparsely innervated) and CF2 (fovea). Grey band: spiral ligament,
black band basilar membrane; radial bres: innervation of organ of
Fig. 5 The auditory fovea lter in the moustached bat. The Corti. b Structural specializations for a resonant system in the SI
audiogram is derived from best thresholds of cochlear nucleus zone. Note the abrupt changes in thickness of basilar and tectorial
units. EOAE: the frequency at which an otoacoustic emission may membrane, and in the limbal attachment of the tectorial membrane
be evoked or may occur spontaneously. EOAE frequency is at the transition between SI and CF2 zones of the basal cochlear
identical with the frequency of the resonator in the cochlea. EOAE turn. c Frequency place map of the basilar membrane (BM) and
frequency is slightly dierent in each specimen. In the audiogram tectorial membrane (TM) of the complete cochlea. Note the widely
EOAE frequency is normalized to 62.0 kHz (dotted line). After expanded representation (fovea) of the narrow frequency band
Kossl (1994) around the CF2 echolocation frequency which corresponds to that
of the evoked or spontaneous otoacoustic emission (CF2 echo,
EOAE). From Russell and Kossl (1999)
110
253
occur when the ear is stimulated with two closely spaced operate with a basilar membrane tuned from 72 to
frequencies f1 and f2. The most pronounced distortion is 62 kHz and an overlaying tectorial membrane that only
2f1f2. DPOAEs are ltered by the tectorial membrane resonates at about 62 Hz?
and, therefore, provide a convenient and noninvasive Measurements of basilar membrane vibrations in the
way of measuring the frequency characteristics of SI zone by noninvasive laser interferometry (Fig. 7;
tectorial membranes (Kossl and Vater 1996b). In non- Russell and Kossl 1999) disclosed that (1) each location
echolocating mammals and in non-specialized regions of on the basilar membrane is indeed tuned to a charac-
the bat cochlea there is a mismatch of the mechanical teristic frequency between 72 kHz at the base of SI zone
tuning between basilar and tectorial membrane by about and 62 kHz at the transition from SI to CF2 zone; and
a quarter of an octave at each location (Kossl and Vater (2) when the ear was acoustically stimulated by sounds
1996b). It is assumed that this slight mismatch contrib- of about 62 kHz the SI zone resonated over its full
utes to the sharpening of cochlear frequency lters by length.
reducing low-frequency tails in the tuning curves. These pioneering experiments show prove that the SI
DPOAE measurements in FM bats disclosed the same zone in the cochlea of the moustached bat acts as a res-
frequency-tuning pattern as in non-echolocating mam- onator tuned to CF2. The resonator consists of a tectorial
mals. In CF/FM bats, however, DPOAE-recordings membrane as a driving element, and the longitudinal
showed a functionally highly specialized tectorial mem- coupling of the mechanics of the basilar membrane. Both
brane (Fig. 6b, c; Kossl and Vater 1996a). Throughout mechanical components are interconnected by the non-
the SI zone of the cochlea in P. parnellii the tectorial
membrane is only tuned to the frequency of the CF2
echolocation sound around 62 kHz. This indicates that a
specialized tectorial membrane resonance plays an inte-
gral role in enhancing cochlear tuning to CF2 beyond
values normally encountered in non-CF/FM bats and
non-echolocating mammals.
111
254
112
255
nator systems in horseshoe bats (Henson et al. 1985; According to this concept the prominent mechanical
Kossl 1994). discontinuity and its resulting standing wave may have
occurred accidentally in one of the forbears of CF/FM
bats. The bats may have made the best of it by imple-
Where do the resonators in the cochleae of CF/FM bats menting into their echolocation system the accidental
come from? sharp-frequency ltering that results from the resonating
wave. In this case, function would follow structure and
The literature states that in inner ears resonance must the bats would have added a CF component to their
not occur. Where, then, do the unique resonators in the echolocation call with a frequency dened by the audi-
basal turns of CF/FM bats phylogenetically come from? tory foveal frequency. Vocalization would adapt to an
In the New World, P. parnellii is the only CF/FM bat auditory accident.
so far known. A comparative study in three mormoopid If this scenario were correct, this would be a re-
species closely related to the moustached bats disclosed markable process in which an accidental non-functional
no structural or functional intermediate stages between mishap would be turned into an evolutionary benecial
FM echolocation in related species and the CF/FM and highly sophisticated specialization.
system in the moustached bat (Kossl et al. 1999). The In whatever way the auditory foveae with its narrow
only distinct dierence between them is the large basal frequency ltering may have evolved, the cochlear
turn of the cochlea with its mechanical specializations resonators have made CF/FM bats specialists for de-
described above. tecting uttering targets. In a highly cluttering envi-
Molecular data suggest that the moustached bat only ronment this specialization may be an advantage.
recently evolved about 49 million years ago (Kossl et al. However, such sophisticated specializations are evolu-
1999). In rhinolophids and hipposiderids CF/FM echo- tionary deadlocks. If, for instance, dense vegetations
location is phylogenetically much older, and this may be disappear, the advantage of uttering target detection
the reason why there are about 150 CF/FM species in disappears as well, and species that predominantly rely
the Old World and only one in the neotropics. In any on uttering target detection by CF/FM echolocation
case, the auditory foveae in P. parnelli and in horseshoe may become extinct. Even without anthropogenic im-
bats are one of the most striking examples of convergent pact ecosystems and ecological niches change continu-
evolution. ously. Hence, species with sophisticated adaptations to
It challenges our imagination to conceive evolution- specic niches will disappear and new species answering
ary driving forces initiating mechanical modications to new challenges will evolve. As long as there is time
that turned the basal turn of the cochlea not only into a enough for newly evolving species to adapt to new
place of high-frequency representation but also into environments the turnover of specialized species is a
a sharply tuned resonator. One idea postulates that the natural process and a driving force of biodiversity.
CF resonators are not the result of specic evolutionary Obviously, specialized animals naturally have a limited
driving forces but were invented by an accident which time of existence.
may consist of a change in ontogenetic programs con-
trolling the development of the basilar and tectorial Acknowledgements I thank Prof. Manfred Kossl, Frankfurt for
critically reading the manuscript, and S. Peisker for preparing the
membranes and produce a strong mechanical disconti- gures. I am also grateful to an anonymous reviewer who helped to
nuity. The consequence could be abrupt changes in clarify the text.
acoustical impedance, acoustical reections and resonant
oscillations which could produce a frequency-specic
cochlear insensitivity. This speculation is based on a References
study of OAE-generating mechanisms in other mammals
(Shera and Guinan 1999). OAEs are not only measurable Covey E, Casseday JH (1995) The lower brainstem auditory
in the form of distortion products but may be also elic- pathways. In: Popper AN, Fay RR (eds) Hearing in bats.
ited by distinct stimulus frequencies or they may even Springer handbook of auditory research. Springer, Berlin
Heidelberg New York, pp 235295
occur spontaneously. Spontaneous or frequency driven Faulstich M, Kossl M (1997) Thalamonal alters cochlear me-
OAEs mainly result from reections of travelling waves chanics in the mustached bat. Proc 20th Midwinter Res Meet
at acoustical impedance irregularities occurring haphaz- Assoc Res Otolaryngol, p 11
ardly along the cochlear partition. Coherent reections Fitzpatrick DC, Kuwada S, Batra R, Trahiotis (1995) Neural re-
from such mechanical discontinuities in the cochlea may sponses to simple, simulated echoes in the auditory brainstem of
the unanesthetized rabbit. J Neurophysiol 74:24692486
create narrow-band cochlear standing waves showing up Gehr DD, Komiya H, Eggermont JJ (2000) Neuronal responses in
as OAEs that are recordable in the ear canal. This applies cat primary auditory cortex to natural and altered species-
to the resonator in P. parnellii: exactly such a very specic sounds. Hear Res 150:2742
prominent and punctuate mechanical discontinuity at the Grinnell AD (1963) The neurophysiology of audition in bats:
temporal parameters. J Physiol (Lond) 167:6796
transition from the SI to the CF2 zone, and an intense Grothe B, Neuweiler G (2000) The function of the medial superior
standing wave are the hallmarks of the resonator in the olive in small mammals: temporal receptive elds in auditory
basal cochlear turn of moustached bats. analysis. J Comp Physiol A 186:413423
113
256
Haplea S, Covey E, Casseday JH (1994) Frequency tuning and Neuweiler G, Schmidt S (1993) Audition in echolocating bats. Curr
response latencies at three levels in the brainstem of the echo- Opin Neurobiol 3:563569
locating bat, Eptesicus fuscus. J Comp Physiol A 174:671683 Neuweiler G, Metzner W, Heilmann U, Rubsamen R, Eckrich M,
Henson MM, Henson OW (1991) Specializations for sharp tuning Costa HH (1987) Foraging behaviour and echolocation in the
in the mustached bat: the tectorial membrane and the spiral rufous horseshoe bat of Sri Lanka. Behav Ecol Sociobiol 20:53
limbus. Hear Res 35:237258 67
Henson OW, Schuller G, Vater M (1985) A comparative study of Olsen JF, Suga N (1991) Combination-sensitive neurons in the
the physiological properties of the inner ear in Doppler shift medial geniculate body of the mustached bat: encoding target
compensating bats (Rhinolophus rouxi and Pteronotus parnellii). range information. J Neurophysiol 65:12751296
J Comp Physiol A 157:587597 ONeill WE, Suga N (1982) Encoding of target range and its
Kalko EKV, Schnitzler HU (1998) How echolocating bats representation in the auditory cortex of the mustached bat. J
approach and acquire food. In: Kunz TH, Racey PA (eds) Neurosci 2:1731
Bat biology and conservation. Smithonian Institution Press, Padian K (1985) The origin and aerodynamics of ight in extinct
Washington, DC, pp 197204 vertebrates. Palaeontology 28:413433
Keller CH, Takahashi TT (1996) Response to simulated echoes by Portfors CV, Wenstrup JJ (1999) Delay tuned neurons in the in-
neurons in the barn owls auditory space map. J Comp Physiol ferior colliculus of the mustached bat: implications for analyses
178:499512 of target distance. J Neurophysiol 82:13261338
Kossl M (1994) Evidence for a mechanical lter in the cochlea of Portfors CV, Wenstrup JJ (2001) Responses to combinations of
the constant-frequency bats, Rhinolophus rouxi and Pteronotus tones in the nuclei of the lateral lemniscus. JARO 2:104117
parnellii. Hear Res 72:7380 Rauschecker JP, Tian B, Hauser M (1995) Processing of complex
Kossl M, Vater M (1985) The cochlear frequency map of the sounds in the macaque nonprimary auditory cortex. Science
mustached bat, Pteronotus parnellii. J Comp Physiol A 157:687 268:111114
697 Rayner JMV (1991) Complexity in a coupled system: ight, echo-
Kossl M, Vater M (1989) Noradrenaline enhances temporal audi- location and evolution in bats. In: Schmidt-Kittler, Vogel K
tory contrast and neuronal timing in the cochlear nucleus of the (eds) Constructional morphology and evolution. Springer,
mustached bat. J Neurosci 9:41694178 Berlin Heidelberg New York, pp 173191
Kossl M, Vater M (1995) Cochlear structure and function in bats. Roverud RC (1993) Neural computations for sound pattern
In: Popper AN, Fay RR (eds) Hearing by bats. Springer, Berlin recognition: evidence for summation of an array of frequency
Heidelberg New York, pp 191234 lters in an echolocating bat. J Neurosci 13:23062312
Kossl M, Vater M (1996a) Further studies on the mechanics of the Roverud RC, Grinnell AD (1985) Echolocation sound features
cochlear partition in the mustached bat. II. A second cochlear processed to provide distance information in the CF/FM bat,
frequency map derived from acoustic distortion products. Hear Noctilio albiventris: evidence for a gated time window utilizing
Res 94:7886 both CF and FM components. J Comp Physiol A 156:457469
Kossl M, Vater M (1996b) A tectorial membrane fovea in the Russell IJ, Kossl M (1999) Micromechanical responses to tones in
cochlea of the mustached bat. Naturwissenschaften 83:8991 the auditory fovea of the greater mustached bats cochlea. J
Kossl M, Vater M (2000) Consequences of outer hair cell damage Neurophysiol 82:676686
for otoacoustic emissions and audiovocal feedback in the Schmidt S (1988) Evidence for a spectral basis of texture perception
mustached bat. J Assoc Res Otolaryngol 1:300314 in bat sonar. Nature 331:617619
Kossl M, Mayer F, Frank G, Faulstich M, Russell IJ (1999) Schnitzler HU, Kalko EKV (1998) How echolocating bats search
Evolutionary adaptations of cochlear function in Jamaican and nd food. In: Kunz TH, Racey PA (eds) Bat biology and
mormoopid bats. J Comp Physiol A 185:217228 conservation. Smithonian Institution Press, Washington, DC,
Link A, Marimuthu G, Neuweiler G (1986) Movement as a specic pp 183196
stimulus for prey catching behaviour in rhinolophid and Schuller G (1979) Vocalization inuences auditory processing in
hipposiderid bats. J Comp Physiol A 159:403413 collicular neurons of the CF-FM bat, Rhinolophus ferrumequi-
Litovsky RY (1998) Physiological studies on the precedence eect num. J Comp Physiol A 132:3946
in the inferior colliculus of the kitten. J Acoust Soc Am Schuller G, ONeill WE, Radtke-Schuller S (1991) Facilitation and
103:31393152 delay sensitivity of auditory cortex neurons in CF-FM bats,
Litovsky RY, Yin TCT (1998) Physiological studies of the prece- Rhinolophus rouxi and Pteronotus parnellii. Eur J Neurosci
dence eect in the inferior colliculus of the cat. I. Correlates of 3:11651181
psychophysics. J Neurophysiol 80:12851301 Shera CA, Guinan JJ (1999) Evoked otoacoustic emissions arise by
Litovsky RY, Colburn HS, Yost WA, Guzman SJ (1999) The two fundamentally dierent mechanisms: a taxonomy for
precedence eect. J Acoust Soc Am 106:16331653 mammalian OAEs. J Acoust Soc Am 105:782798
Ma X, Suga N (2001) Corticofugal modulation of duration-tuned Simmons JA, Stein RA (1980) Acoustic imaging in bat sonar:
neurons in the midbrain auditory nucleus in bats. Proc Natl echolocation signals and the evolution of echolocation. J Comp
Acad Sci USA 98:1406014065 Physiol A 135:6184
Margoliash D, Fortune ES (1992) Temporal and harmonic com- Yan J, Suga N (1996) Corticofugal modulation of time-domain
bination-sensitive neurons in the zebra nchs HVc. J Neurosci processing of biosonar information in bats. Science 273:1100
12:43094326 1103
Neuweiler G (1984) Foraging, echolocation and audition in bats. Yost WA, Guzman SJ (1996) Auditory processing of sound
Naturwissenschaften 71:446455 sources: Is there an echo in here? Curr Direct Psychol Sci 5:125
Neuweiler G (2000) Biology of bats. Oxford University Press, New 131
York
114
J Comp Physiol A (2008) 194:841851
DOI 10.1007/s00359-008-0355-x
ORIGINAL PAPER
Received: 21 April 2008 / Revised: 16 July 2008 / Accepted: 20 July 2008 / Published online: 29 July 2008
Springer-Verlag 2008
Abstract To understand complex sensorymotor behavior optimization of successive frequency modulated echoes for
related to object perception by echolocating bats, precise target range estimation to control approach and landing.
measurements are needed for echoes that bats actually lis-
ten to during Xight. Recordings of echolocation broadcasts Keywords Doppler-shift compensation Echo-intensity
were made from Xying bats with a miniature light-weight compensation Rhinolophus ferrumequinum nippon
microphone and radio transmitter (Telemike) set at the CFFM bats
position of the bats ears and carried during Xights to a
landing point on a wall. Telemike recordings conWrm that Abbreviations
Xying horseshoe bats (Rhinolophus ferrumequinum nippon) BF Best frequency
adjust the frequency of their sonar broadcasts to compen- CF Constant frequency
sate for echo Doppler shifts. Returning constant frequency CM Cochlear microphonic
echoes were maintained at the bats reference frequency DSC Doppler-shift compensation
83 Hz during Xight, indicating that the bats compensated FM Frequency modulated
for frequency changes with an accuracy equivalent to that IPI Interpulse interval
at rest. The Xying bats simultaneously compensate for RF Resting frequency
increases in echo amplitude as target range becomes
shorter. Flying bats thus receive echoes with both stabilized
frequencies and stabilized amplitudes. Although it is widely Introduction
understood that Doppler-shift frequency compensation
facilitates detection of Xuttering insects, approaches to a Echolocating animals perceive objects in their environment
landing do not involve Xuttering objects. Combined fre- with great accuracy by comparing emitted pulses to return-
quency and amplitude compensation may instead be for ing echoes (GriYn 1958). Among echolocating bats, the
Rhinolophidae, Hipposideridae and Mormoopidae all emit
biosonar pulses consisting of a constant frequency (CF)
S. Hiryu Y. Shiori T. Hosokawa H. Riquimaroux Y. Watanabe component followed, by a downward-sweeping frequency
Faculty of Engineering, Doshisha University, modulated (FM) component. Often the CF component also
Kyotanabe 610-0321, Japan is preceded by, a short, upward-sweeping FM component
(GriYn 1958; Neuweiler 2000; Thomas et al. 2003). In
H. Riquimaroux Y. Watanabe
Bio-navigation Research Center, CFFM bats, the strongest part of the sound is the second
Doshisha University, Kyotanabe 610-0321, Japan harmonic (CF2 and FM2). When a CFFM bat approaches a
target, the frequency of returning echoes is Doppler-shifted
Present Address: upward depending on the relative velocity between bat and
S. Hiryu (&) H. Riquimaroux Y. Watanabe
Faculty of Life and Medical Sciences, target. Schnitzler (1968) demonstrated that greater horse-
Doshisha University, Kyotanabe 610-0321, Japan shoe bats (Rhinolophus ferrumequinum) decreased the fre-
e-mail: shiryu@mail.doshisha.ac.jp quency of the CF component (CF2) of emitted pulses to
123
115
842 J Comp Physiol A (2008) 194:841851
compensate for the Doppler shift of echoes induced by et al. 1992; Riquimaroux and Watanabe 2000; Hiryu et al.
approaching Xight, thus regulating echoes to a stable 2007). When bats exhibit compensation behaviors, they are
received frequency regardless of Xight velocity. In subse- assumed to regulate their call parameters (behavioral
quent studies (Schuller et al. 1974; Simmons 1974), sta- response) in relation to the echoes returning from objects
tionary greater horseshoe bats changed the frequency of the they are attending to (acoustic stimuli). In this study, we
CF component of their pulses when these bats were experimentally investigated the relation between echoloca-
exposed to artiWcial Doppler-shifted echoes. Further studies tion calls and returning echoes during compensation behav-
revealed that CFFM bats regulated CF echoes to a very iors which were induced from natural Xight.
narrow frequency range that corresponded to the narrowly
tuned frequency range in which the bats hearing is most
sensitive (Suga and Jen 1976; Ostwald 1984). Doppler- Methods
shift compensation (DSC) is an important behavioral
adaptation of echolocation in these taxa and the sensory Animals
motor interactions of DSC behavior are considered to help
our understanding of audiovocal control in mammalian Four adult Japanese horseshoe bats (R. ferrumequinum
auditory system (Suga 1984; Gaioni et al. 1990; Riquima- nippon, all males; mass 2023 g) were used in this study. They
roux et al. 1991; Metzner 1993; Smotherman and Metzner were captured from a natural cave in Hyogo prefecture in
2003). Japan under license and in compliance with current Japa-
Other studies have reported that some species of bats nese laws. The animals were housed in a temperature- and
decrease the amplitude of their emitted pulses when they humidity-controlled colony room (3 2 2 m) at Dosh-
approach prey or Xy towards an obstacle (GriYn 1958; Jen isha University in Kyoto, Japan. In the colony room, the
and Kamada 1982; Vogler and Neuweiler 1983; Kick and bats were allowed to Xy freely and given access to food and
Simmons 1984; Kobler et al. 1985; Hartley et al. 1989; water. The day and night cycle of the room was controlled
Gaioni et al. 1990; Tian and Schnitzler 1997; Boonman and at 12 h dark and 12 h with light. Once every three days, the
Jones 2002). For example, when the CFFM mustached bat bats were released for an hour in a large Xight chamber (see
(Pteronotus parnellii) is attached to a pendulum and swung below for details) for exercise, and they were weighted to
toward a large Wxed target, the bat signiWcantly decreased check their condition.
the amplitude of its emitted pulse during the forward swing
(Kobler et al. 1985; Gaioni et al. 1990). This behavior rep- Recording procedure
resents another compensation mechanism, echo-intensity
compensation, in which pulse amplitude is adjusted in The experiments were conducted in a Xight chamber 8 (L)
relation to the distance to a target, resulting in stabilization 3 (W) 2 m (H) under long wavelength infrared lighting
of echo amplitudes during the bats approach (Kobler et al. with red Wlters (>650 nm) to avoid visual eVects. The
1985; Gaioni et al. 1990). Frequency-adjusting DSC chamber was made of steel plates to minimize interference
responses have also been recorded while swinging bats on a from external electromagnetic waves used by FM radio sta-
pendulum that carries the recording microphone (Henson tions. Each of the four bats was trained to Xy from one end
et al. 1982; Kobler et al. 1985; Gaioni et al. 1990), but oth- of the Xight chamber directly toward the 3 2 m wall at the
erwise frequency and amplitude compensations by Xying other end, and to land on a landing mesh [1 (W) 1 m (H)]
bats have thus far only been estimated from call parame- was attached in the center of the wall. Echolocation sounds
ters recorded from bats Xying at a distance from the micro- emitted during each Xight were recorded using a custom-
phone. To better understand complex sensorymotor made telemetry microphone (Telemike) mounted on the
behavior and its role in object perception by echolocating bat. The recording procedure of the Telemike was the same
bats, more precise measurements are needed for the fea- as in a previous study (Hiryu et al. 2007). The Telemike
tures of echoes that the bats actually listen to during Xight. consisted of a 1/8-inch omni-directional condenser micro-
We report here direct recordings of free-Xying CFFM phone (Knowles, Model FG-3329, Itasca, IL, USA), a min-
horseshoe bats, Rhinolophus ferrumequinum nippon, made iature custom-designed FM transmitter unit, a 1.5-V
without the variability inherent in recordings made with hearing-aid battery (Sony, Type SR421SW, Tokyo, Japan)
remote microphones. To obtain unimpeded measurements and a transmitting antenna. The Telemike was attached to
of the dual compensation behaviors for frequency and the back of the bat with a piece of double-sided adhesive
amplitude, we employed a miniature light-weight micro- tape, with the microphone pointed forward and positioned
phone and radio transmitter (Telemike) attached to the approximately 1 cm above the noseleaf on the bats face.
bat with the microphone at the position of the bats ears and Because the Telemike weighed less than 0.6 g, including
carried by the Xying bat (Henson et al. 1987; Lancaster the battery, it was light enough to be carried by Xying bats.
123
116
J Comp Physiol A (2008) 194:841851 843
Experimental sessions were less than an hour each, and the and bats Xight velocity were determined in conjunction
bats did not exhibit any fatigue during the experiments as with the frequency, amplitude and other parameters of the
veriWed by continuous visual inspection and vigor of Xight bats echolocation sounds. Due to the size of the Xight
(see below for details). The removal of the Telemike from room and the placement of the video cameras, 3D recon-
the back of a bat after each experiment was facilitated by struction of the bats Xight path was limited to positions
the use of a parting agent to avoid skin irritation. The within 56 m of the target wall.
Telemikes transmitter produced radio signals with a carrier
frequency between 100 and 105 MHz that were received by Sound analysis
a wire antenna consisting of several back-and-forth loops
attached to the ceiling of the Xight chamber. The received The acoustic characteristics of the Xying bats broadcast
signals were demodulated to recover the bats ultrasonic sounds and reXected echoes were analyzed from spectro-
broadcasts using a custom-made FM receiver. The signals grams of the Telemike recordings using custom Matlab
from the receiver were then high-pass Wltered at 20 kHz routines on a personal computer. As illustration of the
(NF Corporation, Model 3625, Yokohama, Japan), digi- method, Fig. 1a shows typical echolocation sounds
tized by a DAT recorder (SONY, Model SIR-1000W, recorded by the Telemike when the Xying bat approached
Tokyo, Japan, 16-bit, 384 kHz), and stored as Wles on the the target wall. Each pulse or echo was extracted from the
hard disk of a personal computer. Prior to recording, the recording, and then the second harmonic component of the
output of the Telemike system was calibrated using a loud- pulse-echo pair was analyzed. The spectrogram exhibited a
speaker (Pioneer, Type PT-R7, Tokyo, Japan) and a Brel peak in energy at the CF2 component in each sound, and
and Kajer Model 4138 (1/8-in) condenser microphone. this energy maximum was measured in each pulse and each
The total frequency response of the Telemike system was
Xat to within 4 dB between 20 and 100 kHz. Since echo- A
location sounds were recorded directly by the microphone Amplitude [V]
behind the bat so as not to interfere with bats Xight path. 50ms
The video cameras recorded 125 frames per second, and B
three-dimensional coordinates of the Xying bats were
reconstructed from these video images using motion analy- 40
sis software (DITECT, Dipp-Motion 2D v 2.1). Prior to
recording bat Xights, a three-dimensional (3D) reference 30
frame with known coordinates was positioned in the center
of the Xight chamber and brieXy recorded by the two video
ED [ms]
20
cameras. The analysis software calibrated the 3D Xight path
reconstruction system using the cameras stereo view of the
10
reference frame. Based on a direct linear transformation Sound data
3D coordinate
technique from the reference frames coordinates, succes-
sive positions of the Xying bat as well as locations of other 0
5 4 3 2 1 0
objects were reconstructed from video-scene coordinates Target distance [m]
measured oV the pair of 2D video images. The control sig- Fig. 1 Representative echolocation sounds of R. ferrumequinum
nal that triggered and synchronized the frames of the video nippon, recorded with the Telemike during Xight. a Amplitude patterns
cameras with each other was digitally stored using the DAT of recorded sound and the corresponding sonogram. b Changes in the
recorder so that Xight coordinates could then be synchronized echo delay from the target wall determined from sound data (solid circle)
and three-dimensional coordinate data of the bats location (line). The
with Telemike sound recordings. Using three-dimensional sonogram indicates typical pulse-echo pairs in the second harmonic
coordinate data, the Xight trajectory of the bat, the dis- portion recorded by the Telemike during Xight. Echoes were observed
tance from the target wall (referred to as the target distance) following the second harmonic component of an emitted pulse
123
117
844 J Comp Physiol A (2008) 194:841851
echo to quantify changes in intensity of echoes in relation with 79 Xights per bat. From the video track reconstruc-
to the pulses. For each Xight session, the measured energy tions, the Xight speed of the bat usually reached a maximum
values were normalized relative to the maximum energy of 34 m/s during its Xight. The bats produced trains of echo-
across all that sessions pulses. The CF2 frequency was location sounds while Xying along the room, during landing,
determined from the frequency at the same peak energy and then while stationary after landing on the mesh attached
location in the spectrogram with a frequency resolution of to the wall. During Telemike-recorded Xights, R. ferrumequi-
46 Hz using a fast Fourier transform over 8,192 sample num nippon used compound echolocation signals, each
points identiWed from the spectrogram. consisting of a CF component with the 2nd harmonic around
The representative spectrogram segment in Fig. 1b 6870 kHz being strongest, plus an accompanying initial
shows both pulses and echoes from the target wall. The short upward FM sweep (28 kHz, ending at 6870 kHz) and
accompanying graph in Fig. 1b compares the values of a terminal short downward FM sweep (beginning at 6870
echo delay (ED) determined from spectrograms (data- kHz and extending 812 kHz lower) (Fig. 1a). The resting
points) with the delay expected for diVerent target distances frequency (RF; the CF2 frequency of pulses emitted by bats
determined from the 3D video reconstruction of one land- at rest) was measured for the stationary bat on the landing
ing Xight as the bat approached the target wall (line). The mesh just after each Xight using the Telemike. Table 1
delay of successive echoes from the target wall (t) was shows the mean RFs of the signals emitted by the four
calculated from the target distance (d) using the formula R. ferrumequinum nippon in the 33 Xights, together with
t = 2d/c, where c is the sound velocity in air (344 m/s). By number of sounds contributing to the averages and their stan-
comparing the EDs obtained from the acoustic data with the dard deviations (SD). RFs estimated for each Xight were fur-
expected delays for the target wall from the video recon- ther averaged across Xights to obtain an overall mean RF for
struction, the echoes from the target were identiWed in the each bat (Table 1). Overall mean values of RF ranged from
spectrograms for frequency and amplitude analysis. Statis- 68.86 to 69.23 kHz, with SDs from 0.07 to 0.1 kHz.
tical analysis of data was conducted using either Students Mean RFs were signiWcantly diVerent across the four bats
t test or the Wilcoxon rank sum test. (ANOVA, P < 0.05), while the RF showed no signiWcant
intra-individual changes within a day for each bat.
Figure 1a illustrates the temporal pattern of broadcasts
Results emitted by a bat during a representative Xight. The sounds
were emitted not at a constant rate but in groups of double
Echolocation sounds of bats during Xight and triple pulses (two or three successive pulses) having
shorter interpulse intervals (IPIs) within the group and
Four individual horseshoe bats made repeated Xights carry- longer IPIs between groups. These clusters of sounds have
ing the Telemike. These Xights yielded complete simulta- been called strobe groups (Moss and Surlykke 2001). Just
neous video and acoustic recordings for the type of analysis before landing the bats usually emitted four or more (<10)
shown in Fig. 1 from a total of 33 Xights from the four bats, multiple pulses in a single larger strobe group.
1 8/28/08 69.17 0.03 (74) 8/28/06 69.14 0.02 (63) 9/9/06 68.98 0.05 (71) 10/24/06 68.97 0.02 (95)
2 8/28/06 69.18 0.03 (81) 10/3/06 69.18 0.05 (198) 9/9/06 69.02 0.06 (31) 10/24/06 68.96 0.03 (79)
3 8/28/06 69.16 0.03 (51) 10/12/06 69.16 0.02 (24) 10/12/06 69.03 0.06 (48) 10/24/06 68.99 0.03 (72)
4 8/31/06 69.22 0.03 (81) 10/12/06 69.16 0.02 (24) 10/24/06 68.96 0.07 (106) 11/16/06 68.91 0.04 (83)
5 8/31/06 69.20 0.03 (65) 10/12/06 69.20 0.04 (124) 10/24/06 68.99 0.06 (64) 11/30/06 68.80 0.04 (172)
6 8/31/06 69.18 0.04 (38) 10/12/06 69.22 0.04 (111) 11/16/06 68.93 0.07 (52) 11/30/06 68.80 0.04 (114)
7 8/31/06 69.19 0.03 (51) 12/8/06 68.96 0.06 (90) 11/16/06 68.96 0.06 (63) 11/30/06 68.80 0.05 (162)
8 9/5/06 69.31 0.03 (75) 11/16/06 68.96 0.07 (53) 11/30/06 68.81 0.04 (95)
9 9/5/06 69.30 0.04 (209) 11/16/06 68.98 0.05 (13)
Total RF 9 7 9 8
(kHz) 69.23 0.07 (725) 69.15 0.10 (610) 68.97 0.07 (503) 68.86 0.09 (872)
Values are means SD. The RFs are signiWcantly diVerent among four bats (KruskalWallis; P < 0.05). There is no signiWcant intra-individual
diVerence of RFs within a day for each bat. Number of pulses given in parentheses
123
118
J Comp Physiol A (2008) 194:841851 845
As is the case for other subspecies of Rhinolophus combined CFFM sounds ranged from 30 ms when the
ferrumequinum (Gustafson and Schnitzler 1979; Tian and bat was 5 m from the wall to 10 ms at the moment of
Schnitzler 1997; Hiryu et al. 2005), R. ferrumequinum landing. Decreases in duration were achieved by shortening
nippon modiWed its emission pattern as it came nearer to the CF component because the durations of the initial/ter-
the landing place. The bats started to decrease the IPIs and minal FM components remained approximately Wxed at 1
the pulse durations when they were approximately 23 m ms (Fig. 1a).
from the target wall. As illustrated in Fig. 2, during Xight As the bat Xew along the room, the IPIs were always
the IPIs alternated between shorter and longer intervals as a longer than the ED from the target wall, even when the bat
reXection of the organization of the pulse train into strobe produced pulses with shorter IPIs (ED shown by dashed
groups. The long intervals in this alternating pattern gradu- lines in Fig. 2). Consequently, the bat always waited for
ally shortened from 80 to 60 ms, while the intervening echoes to return from the wall before it emitted the next
short intervals shortened from 30 to 40 ms. When these sound. This not to say that pulses did not overlap echoes,
alternations are taken into account, however, on average, however. Although the bat consistently shortened the dura-
the IPIs did not shorten appreciably while the bat was in tion of the CF portion of the pulses as it approached the tar-
Xight until it had approached to within 2 m from the wall get wall, the relation between pulse duration and the ED
(Fig. 2). Alternations in IPIs persisted until the bat had (compare plots in Fig. 2) indicates that the echo from the
approached to within about 1 m from the wall, at which target wall deeply overlapped the corresponding emitted
point the IPIs abruptly shortened to 20 ms and then con- pulse (up to 10 ms overlap duration) when the bat Xew to
tinued to shorted to 12 ms when landing occurred. Alter- within about 34 m of the wall. The short FM components
nating IPIs were less frequent during this terminal-stage of the pulses did not overlap with the corresponding FM
landing buzz, when the pulse emission rate reached components of echoes because they were only about 1 ms
7080 pulses/s. As shown in Fig. 2, pulse durations for the in duration, but the CF components overlapped very much.
123
119
846 J Comp Physiol A (2008) 194:841851
Frequency [kHz]
ing events, respectively. Bottom 64
magniWed view of pulse-echo
pairs before landing. Echoes
were observed following an
74
emitted pulse, and the frequency
of the echo received was shifted 72
slightly upward from that of the
70
pulse
68
66
64
quences of the bats DSC responses that occurred As an essential aspect of DSC, successive pulses emitted
throughout the entire sequence of Xights and landings by the bat during Xight changed up or down in CF2 fre-
shown in Fig. 3. The CF2 frequencies of echoes that the bat quency to keep the overall frequency proWle roughly Xat
listened to during its Xight, including the landings and the (Fig. 4a). Figure 4d shows how CF2 frequency changed
U-turns, were stabilized by the bats DSC to a frequency between successive pulses. The maximum pulse-to-pulse
near 69 kHz. increment in CF2 varied from 281 to 422 Hz among four
For quantitative analysis, we identiWed echoes returning bats, and the maximum decrement varied from 234 to
from the target wall by measuring EDs from the spectro- 375 Hz. The mean increment and decrement were 61.0
grams (Methods), and then we measured the CF2 fre- and 83.2 Hz, respectively.
quencies of pulse-echo pairs emitted by the bat while it
intended to land. The spectrogram in Fig. 4a illustrates the Parallel compensation for frequency and amplitude
bats DSC response to the target wall during one Xight. In
this typical example, the bat smoothly varied the pulse CF2 In addition to changing the emitted CF2, the bats gradually
frequency as it turned and approached the wall for landing. decreased the pulse amplitude as they approached the target
Because echoes from the wall are clearly visible in this wall (Fig. 5a). This decrease began at a target distance
Telemike recording, the resulting stability of echo CF2 fre- between 2 and 3 m before landing. In the search phase, the
quency is evident. As shown by the graph in Fig. 4b, the bat maximum pulse amplitude was approximately 130 dB SPL
adjusted the pulse CF2 frequency depending on its Xight (re 20 Pa) peak to peak, measured at the Telemike above
speed toward the wall so that the CF2 frequency of the ech- the bats head. Overall, the pulse amplitude was reduced
oes from the target wall was maintained near the RF. As progressively by approximately 40 dB before landing
further conWrmation of a result that has been shown in (Fig. 5b). The reduction rate was 6.38.2 dB (average 7.6
many previous studies (Schnitzler 1968; Schuller et al. dB) per halving of the target distance for the four bats in the
1974; Gaioni et al. 1990), Xying bats adjusted the CF2 fre- last 2 m of distance. As a consequence of decreasing pulse
quency of echoes to a value slightly higher than the RF amplitude, the amplitude of echoes returning from the tar-
measured separately (Methods) (Students t test, P < 0.05). get wall to the bat changed by 35.6 8.1 dB (mean
The estimated undercompensation (the frequency diVerence SD). Fully 50% of the observed echoes from the target wall
between the echo CF2 and the RF of each bat) was 56.4 were stabilized at amplitudes between 30.1 and 41.2 dB
83.2 Hz (mean SD) for all Xight sessions for four bats. weaker than the pulse (Fig. 5c). Thus, during Xight towards
Fully 50% of the observed echoes from the target wall were the wall, the bats compensated for increases in echo ampli-
maintained between 10 and 110 Hz above the RF; for tude as the target range became shorter by decreasing the
these echoes, the mean value was 59.3 Hz (Fig. 4c). The amplitude of their pulses. This behavioral adaptation is
frequency regulation thus was within 0.1% of the RF of interpreted as echo-intensity compensation (Kobler et al.
R. ferrumequinum nippon (f/RF = 0.0593/69 kHz), indi- 1985; Gaioni et al. 1990).
cating that the compensation depth achieved by free-Xying Figure 6a shows the combined frequency (DSC) and
horseshoe bats approached very close to 100%. amplitude compensation during a single landing Xight.
123
120
J Comp Physiol A (2008) 194:841851 847
Frequency [kHz]
the landing Xight. L indicates the 71
landing point. b CF2 frequencies
of pulse-echo pairs as a function 69
of target distance. Open and sol-
id circles indicate the CF2 fre-
67
quency of emitted pulse and
returning echo from the target
wall. A dashed line indicates the 65
resting frequency (RF). A solid
line indicates echo CF2 fre- 63
quency which, when DSC was
absent, was estimated from the
velocity of the bat. c The box B C
71.5 200
plot of CF2 frequency of the
returning echo from the target
wall during the landing Xight. 71.0
Data were taken from all Xight
sessions for all four bats (33 70.5
0
-400 -200 0 200 400 -400 -200 0 200 400
Frequency change of successive pulses [Hz]
Using these two pulse compensation mechanisms in all analyzed Xights of four bats (blue data-points). The bats
tandem (blue data-points), R. ferrumequinum nippon received received echoes from their intended target (i.e. the landing
echoes with both stabilized frequencies and stabilized wall) that were kept within a certain limited amplitude
amplitudes (red data-points). Figure 6b shows the distribu- range (35.6 8.1 dB of the maximum pulse amplitude).
tions of pulse CF2 frequencies and pulse amplitudes from As a result of their combined compensation responses, the
123
121
848 J Comp Physiol A (2008) 194:841851
B
Fig. 5 Amplitude change in pulse-echo pairs by bats intending to land
on the target wall. a Three-dimensional spatio-temporal reconstruc- Pulse
0
tions of echolocation behavior during the landing Xight. Coordinate
grids (1 m2) show the dimensions of the Xight chamber (8 3 2 m).
Relative sound pressure (dB)
Echolocation pulses recorded using the Telemike are shown alongside -10
the Xight trajectory. b Changes in the sound pressure of pulse-echo
pairs. The values were normalized relative to the maximum pulse -20
amplitude of each Xight. c Echo attenuation during the landing Xight.
Data were pooled from all Xight sessions for all four bats
-30
123
122
J Comp Physiol A (2008) 194:841851 849
tor emitted pulses. Here, we have recorded not only the employed the RF measured using the Telemike on the bat
pulses but also the returning echoes that the bats actually just after each landing trial. The immediacy of the RF mea-
listen to during Xight. By this means, we demonstrated that surements following Xights may result in the small fre-
the CF2 frequency of returning echoes was compensated to quency gap observed. We suppose that plasticity in the
within 83 Hz during DSC induced by natural Xight. Fre- auditory system relevant to DSC function may enable such
quency control of vocalization from pulse-to-pulse on aver- Wne regulation of echo CF2 frequency because there are
age was 61.0 Hz for increments and 83.2 Hz for both short- and long-term intra-individual variations of RF.
decrement (Fig. 4d). Because the mean of the SDs of RF in In a previous study using the same landing-Xight proce-
our four horseshoe bats was 90 Hz (Table 1), which dure with the Telemike, we found patterns of alternation in
corresponds to the SDs of RF obtained previously from pulse CF2 frequencies between higher and lower values in
R. ferrumequinum (e.g. 30120 Hz in Schuller et al. 1974), Taiwanese leaf-nosed bats (Hipposideros terasensis)
the Xying bats compensated echo CF2 frequency with an (Hiryu et al. 2005). These frequency diVerences were
accuracy of regulation equivalent to bats at rest. approximately equivalent to DSC changes required for tar-
The Telemike recordings showed that the echo CF2 fre- gets located in diVerent directionslower emitted frequen-
quency reaching to the bat was slightly but signiWcantly cies for the higher Doppler shifts from a target located to
higher than the RF itself (Fig. 4c). In previous DSC studies, the bats front, and higher emitted frequencies for the lower
the estimated echo CF2 frequency also was found to be Doppler shifts from a target oV to the side. The occurrence
maintained at a frequency that was slightly above the RF of such large diVerences in the DSC response suggest that
(the frequency of compensation during active Xight has the bats could shift their attention between echoes from the
been called the reference frequency). This diVerence front to the side and back again while DSC is taking place.
between resting and Xying bats indicates the occurrence of By rapidly focussing and shifting their DSC responses on
systematic slight undercompensation by Xying CFFM bats diVerent echoes (front and side target echoes), the bats may
during DSC (Schnitzler 1968; Schuller et al. 1974). The segregate echoes from diVerent targets and process them as
compensated frequency diVerence between resting and multiple auditory streams. Such streaming may enabling
Xying bats varies among bat species and studies [e.g. the bats to perceive several targets in real time. Some of the
50300 Hz for R. ferrumequinum at an RF of 8181.4 kHz recordings of R. ferrumequinum nippon in this study appear
(Schuller et al. 1974); 110250 Hz for R. rouxi at an RF of to show alternating patterns in pulse CF2 frequency (low
7279 kHz (Neumann and Schuller 1991) and 50250 for and high) as well as duration and IPI (Fig. 2), but we do not
Pteronotus parnellii parnellii at an RF of 61 kHz (Gaioni yet have strong evidence for periodic attention change as in
et al. 1990)]. With regard to the frequency diVerence H. terasensis. One possible reason is that, during Xight, R.
between resting and Xying bats, physiological studies have ferrumequinum emit long pulses (from 30 to 10 ms). In
detected frequency diVerences between a preferred fre- our measurements of the CF component, we recognized
quency in the auditory system and the RF. For example, the only an intense front target echo for each pulse which con-
best frequency (BF) at the lowest threshold of cochlear spicuously overlapped with a large portion of the corre-
microphonic (CM) audiogram in mustached bats is 200 Hz sponding emitted pulse. In contrast, the pulse duration of H.
above the RF (Henson et al. 1982). The BFs of the common terasensis is shorter, ranging from only 5 to 10 ms. For
type of audiovocal neurons in the paralemniscal tegmen- each pulse, two separate echoes are typically seen in the
tum, which are relevant to motor control for DSC, were not Telemike data of H. terasensis while the bat is Xying
found for frequency values between the RF and 150 Hz toward the target wall for landing; one is the echo from the
above (Metzner 1993). Frequency diVerences found in bats front target wall and the other is the echo from the side wall
from diVerent geographic locations further indicate that (or the Xoor) of the Xight chamber. These separate echoes
undercompensation in Xying bats is real. In R. ferrumequi- will have very diVerent Doppler shifts and thus may attract
num nippon, the RF is 6569 KHz, which is considerably the bats attention at diVerent times, causing a shift in the
lower than the 80 kHz RF of European subspecies, and bats attention between echoes and corresponding shifts in
the frequency of the lowest threshold of behavioral audio- the DSC response itself. There also may be diVerences in
gram was reported to be 0.6 KHz higher than the RF the echolocation strategy of these two CFFM bat species
(Taniguchi 1985). The Telemike recordings obtained here in the context of DSC behavior.
showed the gap between RF and the compensation value
for Xying bats to be only about 60 Hz. RFs of CFFM bats Optimization of successive echoes in the task of target
change daily (Schuller et al. 1974; Suga et al. 1987; Hiryu range estimation by echo-intensity compensation
et al. 2006) and also instantaneously because of increase in
body temperature (HuVman and Henson 1993a, b) and Echo-intensity compensation allows the bat to receive ech-
Xight activities (Henson et al. 1990). For this reason, we oes from targets (i.e. the target wall) with amplitudes that
123
123
850 J Comp Physiol A (2008) 194:841851
are kept within a certain optimal sound pressure range bats thus bring a novel adaptive perspective to biomimetic
throughout the Xight. Receiving echoes with a stable ampli- sonar design.
tude, and perhaps thus also signal-to-noise ratio, could help
the bats to sustain consistent analysis of successive echoes Acknowledgments We thank Prof. James A. Simmons for careful
reading of the manuscript and valuable comments. We also thank T.
without distortions of perception caused by amplitude
Hagino, M. Fukuda, E. Fujioka, M. Omura and Y. Osawa for analysis
changes. The auditory system of CFFM bats is precisely and technical support during this experiment; N. Urano for assistance
tuned to a narrow frequency band (Neuweiler 1970; in capturing bats in the Weld. The experiments complied with the Prin-
Ostwald 1984; Suga 1984; Riquimaroux et al. 1991). Bats ciples of Animal Care, publication no. 8623, revised in 1985, of the
National Institutes of Health, and the procedures were approved by the
can detect and analyze small frequency diVerence induced
animal care committee of Doshisha University. This work was partly
by the Xuttering of insect wings in this band (Schnitzler and supported by a grant to the Research Center for Advanced Science and
Henson 1980), and they adjust echoes to fall within this Technology (RCAST) at Doshisha University from the Ministry of
band. It is widely understood that Doppler-shift frequency Education, Culture, Sports, Science, and Technology (MEXT) of Japan:
Special Research Grants for the Development of Characteristic Education
compensation maintains a constant carrier frequency in the from the Promotion and Mutual Aid Corporation for Private Schools of
echoes from Xuttering insects which facilitates detection of Japan and the Innovative Cluster Creation Project.
insect prey by CFFM bats. The very narrow zone of fre-
quencies to which CF2 is adjusted form an acoustic fovea
where Xutter-detecting mechanisms are present. However, References
the occurrence of DSC by bats approaching the landing
Au WWL, Benoit-Bird KJ (2003) Automatic gain control in the echo-
wall suggests that DSC may be important not only for location system of dolphins. Nature 423:861863
detection of Xuttering insects, but also for perceiving sur- Boonman A, Jones G (2002) Intensity control during target approach in
rounding objects especially for target range estimation by echolocating bats; stereotypical sensorymotor behaviour in Dau-
the FM component attached to the CF component. Physio- bentons bats, Myotis daubentonii. J Exp Biol 205:28652874
Edamatsu H, Suga N (1993) DiVerences in response properties of neu-
logical experiments have shown that the delay tuning of rons between two delay-tuned areas in the auditory cortex of the
FMFM neurons in mustached bats is aVected by echo mustached bat. J Neurophysiol 69:17001712
amplitude, suggesting that amplitude compensation acts to Gaioni SJ, Riquimaroux H, Suga N (1990) Biosonar behavior of
stabilize range estimation (Edamatsu and Suga 1993). mustached bats swung on a pendulum prior to cortical ablation.
J Neurophysiol 64:18011817
Combined frequency and amplitude compensation may be GriYn DR (1958) Listening in the dark. Yale University Press,
used for optimization of successive echoes in the task of New Haven
target range estimation to control approach and landing. Gustafson Y, Schnitzler HU (1979) Echolocation and obstacle avoid-
These combined compensations thus may serve more for ance in the hipposiderid bat Assellia tridens. J Comp Physiol A
131:161167
improving target range accuracy with the FM components Hartley DJ, Campbell KA, Suthers RA (1989) The acoustic behavior
of echoes than for any involvement with velocity percep- of the Wsh-catching bat, Noctilio leporinus, during pre-capture.
tion itself. J Acoust Soc Am 86:827
Adjustments in pulse amplitude occurs not only in Henson OW Jr, Pollak GD, Kobler JB, Henson MM, Goldman LJ
(1982) Cochlear microphonic potentials elicited by biosonar sig-
CFFM bats, but also in FM bats (Hiryu et al. 2007) and nals in Xying bats, Pteronotus p. parnellii. Hear Res 7:127147
echolocating dolphins (Au and Benoit-Bird 2003; Li et al. Henson OW Jr, Bishop AL, Keating AW, Kobler JB, Henson MM,
2006). In these wideband echolocating species, perception Wilson BS, Hansen R (1987) Bisonar imaging of insects by Pter-
of objects is based on target range from ED, not perception onotus p. parnellii, the mustached bat. Nat Geor Res 3:82101
Henson OW, Koplas PA, Keating AW, HuVman RF, Henson MM
of Xutter from echo Doppler-shifts. We suggest that com- (1990) Cochlear resonance in the mustached bat: behavioral adap-
pensation mechanisms may underlie the fundamental tations. Hear Res 50:259273
design of biosonar systems. In contrast to the compensation Hiryu S, Katsura K, Lin LK, Riquimaroux H, Watanabe Y (2005)
behavior of biosonar in animals, man-made sonar systems Doppler-shift compensation in the Taiwanese leaf-nosed bat
(Hipposideros terasensis) recorded with a telemetry microphone
generally are designed to transmit sonar sounds with a Wxed system during Xight. J Acoust Soc Am 118:39273933
frequency and intensity, and to obtain target information by Hiryu S, Katsura K, Nagato T, Yamazaki H, Lin LK, Watanabe Y,
measuring deviations in frequency and amplitude in con- Riquimaroux H (2006) Intra-individual variation in the vocalized
junction with ED. CFFM bats adjust their call frequency frequency of the Taiwanese leaf-nosed bat, Hipposideros terasen-
sis, inXuenced by conspeciWc colony members. J Comp Physiol A
and amplitude together to maintain the echo frequency and 192:807815
amplitude within a stabilized range. As a result, CF/FM Hiryu S, Hagino T, Riquimaroux H, Watanabe Y (2007) Echo-inten-
biosonar systems do not need to provide wide-frequency sity compensation in echolocating bats (Pipistrellus abramus)
and dynamic receiving ranges; instead they analyze the during Xight measured by a telemetry microphone. J Acoust Soc
Am 121:17491757
echo frequency in a narrow range (the acoustic fovea) in HuVman RF, Henson OW Jr (1993a) Labile cochlear tuning in the
the auditory system, which may greatly reduce computa- mustached bat. I. Concomitant shifts in biosonar emission fre-
tional eVorts in the brain. Information gathered from Xying quency. J Comp Physiol A 171:725734
123
124
J Comp Physiol A (2008) 194:841851 851
HuVman RF, Henson OW Jr (1993b) Labile cochlear tuning in the Riquimaroux H, Gaioni SJ, Suga N (1991) Cortical computational
mustached bat. II. Concomitant shifts in neural tuning. J Comp maps control auditory perception. Science 251:565568
Physiol A 171:735748 Schnitzler HU (1968) Die Ultraschallortungslaute der Hufeisen-Fle-
Jen PH, Kamada T (1982) Analysis of orientation signals emitted by dermuse (Chiroptera-Rhinolophidae) in verschiedenen Orient-
the CFFM bat, Pteronotus p. parnellii and the FM bat, Eptesicus ierungssituationen [The ultrasonic sounds of horseshoe bats
fuscus during avoidance of moving and stationary obstacles. (Chiroptera-Rhinolophidae) in diVerent orientation situations].
J Comp Physiol A 148:389398 Z Vergl Physiol 57:376408
Kick SA, Simmons JA (1984) Automatic gain control in the bats sonar Schnitzler HU, Henson OW Jr (1980) Performance of airborne animal
receiver and the neuroethology of echolocation. J Neurosci sonar system, I. Microchiroptera. In: Busnel R-G, James FF (eds)
4:27252737 Animal sonar systems. Plenum Press, New York, pp 109181
Kobler JB, Wilson BS, Henson OW Jr, Bishop AL (1985) Echo inten- Schuller G, Beuter K, Schnitzler HU (1974) Response to frequency
sity compensation by echolocating bats. Hear Res 20:99108 shifted artiWcial echoes in the bat Rhinolophus ferrumequinum.
Lancaster WC, Keating AW, Henson OW Jr (1992) Ultrasonic vocal- J Comp Physiol A 89:275286
izations of Xying bats monitored by radiotelemetry. J Exp Biol Simmons JA (1974) Response of the Doppler echolocation system
173:4358 in the bat, Rhinolophus ferrumequinum. J Acoust Soc Am
Li S, Wang D, Wang K, Akamatsu T (2006) Sonar gain control in 56:672682
echolocating Wnless porpoises (Neophocaena phocaenoides) in an Smotherman M, Metzner W (2003) Fine control of call frequency by
open water. J Acoust Soc Am 120(4):18031806 horseshoe bats. J Comp Physiol A 189:435446
Metzner W (1993) An audio-vocal interface in echolocating horseshoe Suga N (1984) The extent to which biosonar information is represented
bats. J Neurosci 13:18991915 in the bat auditory cortex. In: Edelman GM, Gall WE, Cowan
Metzner W, Zhang S, Smotherman M (2002) Doppler-shift compensa- WM (eds) Dynamic aspects of neocortical function. Wiley,
tion behavior in horseshoe bats revisited: auditory feedback con- New York, pp 315373
trols both a decrease and an increase in call frequency. J Exp Biol Suga N, Jen PH (1976) Disproportionate tonotopic representation for
205:16071616 processing CFFM sonar signals in the mustache bat auditory
Moss CF, Surlykke A (2001) Auditory scene analysis by echolocation cortex. Science 194:542544
in bats. J Acoust Soc Am 110:22072226 Suga N, Niwa H, Taniguchi I, Margoliash D (1987) The personalized
Neumann I, Schuller G (1991) Spectral and temporal gating mecha- auditory cortex of the mustached bat: adaptation for echolocation.
nisms enhance the clutter rejection in the echolocating bat, J Neurophysiol 58:643654
Rhinolophus rouxi. J Comp Physiol A 169:109116 Taniguchi I (1985) Echolocation sounds and hearing of the greater
Neuweiler G (1970) Neurophysilogische Untersuchungen zum Japanese horseshoe bat (Rhinolophus ferrumequinum nippon).
Echoortungssystem der groben Hufeisennase Rhinolophus J Comp Physiol A 156:185188
ferrum equinum Schreber, 1774. Z Vergl Physiol 67:273306 Thomas JA, Moss CF, Vater M (2003) Echolocation in bats and
Neuweiler G (2000) The biology of bats. Oxford University Press, dolphins. University of Chicago Press, Chicago
New York Tian B, Schnitzler HU (1997) Echolocation signals of the greater
Ostwald J (1984) Tonotopical organization and pure tone response horseshoe bat (Rhinolophus ferrumequinum) in transfer Xight and
characteristics of single units in the auditory cortex of the greater during landing. J Acoust Soc Am 101:23472364
horseshoe bat. J Comp Physiol A 155:821834 Vogler B, Neuweiler G (1983) Echolocation in the noctule (Nyctalus
Riquimaroux H, Watanabe Y (2000) Characteristics of bat sonar noctula) and horseshoe bat (Rhinolophus ferrumequinum).
sounds recorded by a telemetry system and a Wxed ground micro- J Comp Physiol A 152:421432
phone. Seventh Western PaciWc Regional Acoustics Conference
(WESTPRACVII):233238
123
125
126
Echo-intensity compensation in echolocating bats (Pipistrellus
abramus) during flight measured by a telemetry microphone
Shizuko Hiryua
Research Center for Intelligent Information Science, Doshisha University, Kyotanabe 610-0321, Japan
Tomotaka Hagino
Department of Electrical Engineering, Doshisha University, Kyotanabe 610-0321, Japan
Hiroshi Riquimaroux
Department of Intelligent Information Engineering and Sciences, Doshisha University, Kyotanabe 610-0321,
Japan and Bio-navigation Research Center, Doshisha University, Kyotanabe 6100321, Japan
Yoshiaki Watanabe
Department of Electrical Engineering, Doshisha University, Kyotanabe 610-0321, Japan
and Bio-navigation Research Center, Doshisha University, Kyotanabe 610-0321, Japan
I. INTRODUCTION range in which the bats can hear best. This behavior is
termed Doppler-shift compensation DSC and is an im-
Bats Microchiroptera possess a highly developed sonar portant behavioral adaptation for echolocation in CF-FM
system. As biosonar animals, they are capable of echoloca- bats Schnitzler, 1968; Schuller et al., 1974; Simmons, 1974;
tion that recognizes the physical attributes of their environ- Gustafson and Schnitzler, 1979; Trappe and Schnitzler, 1982;
ment with great accuracy by comparing emitted pulses to
Gaioni et al., 1990; Lancaster et al., 1992; Keating et al.,
returning echoes Griffin, 1958. Echolocation pulses can
1994.
differ markedly in structure, consisting of either only short
During flight, bats decrease the intensity of their emitted
downward frequency-modulated FM pulses or long
pulses when they approach a prey item or an obstacle e.g.,
constant-frequency CF pulses followed by a short FM com-
Griffin, 1958; Jen and Kamada, 1982; Vogler and Neuweiler,
ponent. Such variation in the structure of echolocation pulses
is thought to reflect the adaptations of different species to the 1983; Kick and Simmons, 1984; Hartley et al., 1989; Tian
constraints imposed by their foraging behavior Simmons and Schnitzler, 1997; Boonman and Jones, 2002. Kobler and
and Stein, 1980; Neuweiler, 1984. colleagues 1985 have shown that a bat attached to a pen-
Rhinolophids, Hipposiderids, and a Mormoopid the dulum and swung toward a large fixed target decreases the
mustached bat, Pteronotus parnellii all use a compound intensity of its emitted pulse during the forward swing and
CF-FM pulse. As a bat approaches a target, the frequency of increases the intensity during the backward swing. This be-
the returning echo is Doppler-shifted depending on the ve- havior indicates the possibility of another compensation
locity of the bat in relation to the target. These bats compen- mechanism, in which pulse intensity is adjusted in relation to
sate for the Doppler shifts by changing their pulse frequen- the distance to a target, resulting in maintenance of echo
cies so that the echo frequencies remain constant and are intensity within the optimal sensitivity range echo-intensity
precisely analyzed within the narrowly tuned frequency compensation. Subsequently, reduction of pulse intensity
has been quantitatively investigated in bats given a flight task
Hartley et al., 1989; Waters and Jones, 1995; Tian and
a
Electronic mail: shiryu@mail.doshisha.ac.jp Schnitzler, 1997; Boonman and Jones, 2002 and using a
J. Acoust. Soc. Am. 121 3, March 2007 0001-4966/2007/1213/1749/9/$23.00 2007 Acoustical Society of America 1749
127
pendulum device or a moving-target apparatus to elicit and gated adjustments of call parameters in the context of echo-
measure changes in pulse intensity Gaioni et al., 1990; Hart- intensity compensation.
ley, 1992b. Adjustment of pulse intensity by bats has been
discussed in a number of studies as being related to echo- II. MATERIALS AND METHODS
intensity compensation e.g., Kick and Simmons, 1984; Hart-
A. Subjects
ley et al., 1989; Hartley, 1992a; b; Simmons et al., 1992;
Waters and Jones, 1995; Tian and Schnitzler, 1997; Boon- Three adult Japanese house bats P. abramus were used
man and Jones, 2002. Recently, Au and Benoit-Bird 2003 in this study. The animals were captured from a large colony
demonstrated that free-ranging dolphins in the wild changed roosting in bridge girders near the campus of Doshisha Uni-
the amplitude of emitted echolocation signals depending on versity, Japan. Pipistrellus abramus is a member of the Ves-
the range of the target to compensate for propagating loss in pertilionidae and is commonly found in Japan. Although
sound. Therefore, echo-intensity compensation may be effec- closely related pipistrelle bats such as P. pipistrellus and P.
tive in the biosonar systems of both bats and dolphins. How- kuhli have been studied extensively e.g., Schnitzler et al.,
ever, accurate measurement of the intensity of directional 1987; Kalko, 1995; Waters and Jones, 1995; Holderied and
pulses emitted by biosonar animals is still considered diffi- Helversen, 2003, the echolocation characteristics of P. abra-
cult using the traditional microphone system. For quantita- mus are not well known. The body mass of P. abramus
tive analysis of the pulse intensity which the flying bats ac- ranges from 5 to 8 g, and the wingspan measures approxi-
tually emitted, recorded pulses required several corrections mately 10 cm. Echolocation pulses are emitted through the
for compensating the directional properties of the recording mouth. The bats were kept in a rearing cage of 0.9 m L
microphone and the emitted pulse, and also traveling loss of 0.9 m W 0.6 m H, and were allowed free access to
sound in the air depending on the distance between the bat food mealworms and water. On alternate days, the bats
and the microphone e.g., Hartley et al., 1989; Tian and were allowed to fly freely for several hours in a large flight
Schnitzler, 1997; Boonman and Jones, 2002. Given these room. The experiments complied with the Principles of Ani-
constraints, the pulse intensity measured by the traditional mal Care, publication no. 86-23, revised 1985, of the Na-
microphone potentially confounded interpretation of a bats tional Institutes of Health, and with current Japanese laws.
adjustment of pulse intensity in relation to its spatiotemporal
behavior during echolocation. Furthermore, measuring the B. General experimental procedure
intensity of echoes as well as emitted pulses is necessary to
All experiments were conducted in a flight chamber
reveal any echo-intensity compensation behavior employed
measuring 8 m L 3 m W 2 m H under long wave-
by the animals. However, the relationship between the pulse
length lighting with red filters 650 nm to avoid optical
and returning echo intensities has not been experimentally
effects Hope and Bhatnagar, 1979; Ghose and Moss, 2003.
investigated in detail for bats during flight.
The chamber was made of steel plates to minimize interfer-
In this study, we examined echolocation behavior in
ence from external electromagnetic waves. All internal walls
Japanese house bats Pipistrellus abramus, an FM bat given were painted in black. The bats were released at one end of
a flight landing task in a laboratory. Echolocation sounds the flight chamber and allowed to fly freely to the opposite
were recorded by a telemetry microphone Telemike end where a landing mesh 1 m W 0.7 m H was at-
mounted on the head of the bat Henson et al., 1987; Lan- tached to the wall 1.8 m above the floor. This wall is referred
caster et al., 1992; Riquimaroux and Watanabe, 2000; Hiryu to as the target wall during the landing flight task. Flight
et al., 2005. Since well-studied FM bat species are small behavior was recorded as a flying bat approached the target
e.g., P. abramus weighs 5 8 g in relation to the weight of wall for landing. Recording was conducted for three flight
telemetry microphones, sound recording using telemetry mi- sessions of each bat a total of nine flight sessions.
crophones has never been used to investigate echolocation in
flying FM bats. We developed a small Telemike that was
C. Telemike system
light enough to be carried by the animals, and combined the
Telemike with a high-speed video camera system in order to Echolocation sounds were recorded by a custom-made
examine the relationship between the echolocation sounds telemetry microphone Telemike mounted on the back of the
and the spatiotemporal behavior of bats during flight. bat Fig. 1. The Telemike consists of a 1/8-inch omnidirec-
Bats are assumed to adjust their call parameters by feed- tional condenser microphone Knowles, FG-3329, Illinois,
back control in response to the attended returning echoes. USA, an FM transmitter unit, a hearing aid battery of 1.5 V
Therefore, it is particularly important to investigate the be- Sony, SR421SW, Tokyo, Japan, and a transmitting antenna.
havioral response call parameters of bats associated with The Telemike was attached to the back of a bat with a piece
their corresponding acoustic stimuli echoes under the spa- of double-sided glue tape, with the microphone positioned
tiotemporal conditions in which the pulse-echo pairs were approximately 1 cm above the mouth. Because the Telemike
produced. Since P. abramus emit short downward FM pulses weighed less than 0.6 g, including the battery, it was light
with a mean duration of approximately 1 ms, the echoes ar- enough to be carried by bats weighing about 5 g. The bats
riving from targets could be recorded by the microphone did not exhibit any fatigue during the experiments. Removal
above the head without overlapping with the outgoing of the Telemike from the back of a bat after each experiment
pulses. We analyzed the signal characteristics of pulse-echo was facilitated by use of a parting agent to avoid skin irrita-
pairs to which the flying bats actually listened, and investi- tion.
1750 J. Acoust. Soc. Am., Vol. 121, No. 3, March 2007 Hiryu et al.: Echo-intensity compensation in echolocating bats
128
FIG. 1. Japanese house bat P. abramus with an onboard microphone
Telemike mounted on its back. The Telemike consists of a 1/8-inch con-
denser microphone, FM transmitter, battery, and transmitting wire antenna
weight: 0.6 g, including the battery.
D. Three-dimensional reconstruction of
spatiotemporal echolocation behavior
The flight behavior was recorded using a dual digital
high-speed video camera system NIPPON ROPER Co.,
Ltd., CR Imager model 2000s, Chiba, Japan. Cameras were FIG. 2. Pulse-echo pairs recorded with the Telemike while a bat approached
placed at a corner of the flight chamber and did not interfere the target wall. A Sonograms at a target distance of 4 m top and 1 m
bottom. Arrows indicate the echoes returning from the target wall see
with the bats flight path. The frame rate was 125 per second. text. B Changes in echo delay determined from sound data as a function
Three-dimensional coordinates of the flying bats were recon- of target distance during the landing flight for two bats solid circles. Circle
structed from these video images using a commercial motion size indicates relative variation in echo intensity where bigger circles des-
ignate sounds of greater intensity relative to smaller circles. The three lines
analysis software DITECT, Dipp-Motion 2D ver. 2.1. Prior
represent echo delays between the flying bat and three targetsthe target
to recording bat flights, a reference frame with known coor- wall, floor, and ceiling of the chambercalculated using three-dimensional
dinates was positioned in the center of the flight chamber, coordinate data of the bat.
then recorded by two video cameras. The analysis software
calibrated the reconstruction system with the coordinate data calculated from the distance between the flying bat and an
of the reference frame. Based on a direct linear transforma- object d using the formula t = 2d / c, where c is the sound
tion technique, the position of the flying bat or other object velocity in the air.
was reconstructed from two-dimensional coordinate data in
the video images. The signal triggering the video cameras
E. Sound analysis
was digitally stored using a DAT recorder so that flight co-
ordinates could be synchronized with sound data. Using The acoustical characteristics of echolocation sounds
three-dimensional coordinate data, the flight trajectory of the were analyzed from the sonogram using a custom program of
bat and the distance from the target wall target distance or Matlab on a personal computer. Figure 2A shows sono-
other objects could be determined. The echo delay t was grams of typical pulse-echo pairs recorded by the Telemike
J. Acoust. Soc. Am., Vol. 121, No. 3, March 2007 Hiryu et al.: Echo-intensity compensation in echolocating bats 1751
129
TABLE I. Signal characteristics of the echolocation pulse from P. abramus at rest. iFM: initial FM; tFM: terminal FM; BW: bandwidth.
Bat A 100 0.90 0.14 79.3 4.71 42.7 2.73 36.6 5.72 54.5 5.91 97120 112
Bat B 100 1.20 0.11 84.9 4.81 42.6 2.65 42.3 5.28 56.4 9.29 100117 112
Bat C 100 1.57 0.30 83.6 4.60 46.0 2.86 37.6 5.87 57.4 8.57 99117 109
when the flying bat approached the target wall. Near the the echo delays shown in sound data solid circles, the ech-
center of the chamber 4 m from the target wall, a number oes from each surrounding wall of the flight chamber could
of echoes referred to as an echo train were usually observed be identified in the observed sonogram. For example, the
with different echo delays the time difference between each arrows in Fig. 2A indicate echoes from the target wall.
pulse-echo pair and intensities following an emitted pulse. Echoes from the floor and ceiling of the chamber consis-
The second and higher harmonic components of the return- tently showed intense sound pressure, as well as echoes from
ing echo were most often attenuated beyond 3 4 m from the the target wall directly in the flight path of the bat.
target wall. On the other hand, when the bat was within a Prior to flight recording, pulses of the bat in a stationary
target distance of 1 and 2 m, the number of echoes reaching position were recorded using the Telemike for quantitative
the flying bat decreased. The echoes were clearly separated analysis. The acoustical parameters of the echolocation
from each other, and the second harmonic component ap- pulses were analyzed using the custom program of Matlab
peared in the sonogram. In this study, only the fundamental described above. Signal bandwidth and duration were deter-
component of the pulse-echo pair was analyzed. Each pulse mined from the sonogram at 25 dB relative to the peak
or echo was extracted from the sonogram, and the echo delay intensity of the pulse.
was measured for all echoes observed by the Telemike. The
sound pressure level of the pulse was calculated from the III. RESULTS
peak-to-peak amplitude voltage of the observed pulse in the
A. General echolocation behavior of P. abramus
time domain. Simultaneously, the sonogram exhibited a peak
in energy at around 50 60 kHz in each sound of P. abramus Pipistrellus abramus emits a short downward FM pulse
during flight. Therefore, the spectral energy at the peak en- with maximum energy at the fundamental component. The
ergy portion of each pulse and echo was measured in the mean duration at rest is 1.22 0.34 ms n = 300, and the fun-
displayed sonogram using the custom program of Matlab so damental frequency was modulated from approximately 83
that changes in intensity of a weak echo could be quantita- to 44 kHz by the three bats Table I. The sound pressure
tively evaluated in relation to the pulse intensity. The maxi- level SPL of an emitted pulse at rest ranged from approxi-
mum magnitude of the spectral energy in each sound is re- mately 100 to 120 dB peak-to-peak re 20 Pa, with an
ferred to as the sound intensity in this study. average of 111 dB at the microphone above the bats head.
Typical changes in the echo delay determined from The sonogram exhibited a peak in energy at 56.0 8.12 kHz,
sound data are shown in Fig. 2B as a function of target which was 11 14 kHz higher than the terminal frequency.
distance during landing as the bat approached the target wall. For the landing experiment, a total of nine flight sessions
The size of the solid circles indicates the relative variation in were recorded for three bats. In flight, the pulse duration was
echo intensity where bigger circles designate sounds of elongated to 3 4 ms, and then decreased to 0.5 ms before
greater intensity relative to smaller circles. In this flight ex- landing the target wall. Figure 3 shows the envelope and
periment, the Telemike allowed us to observe echoes with an frequency structure of a typical echolocation pulse emitted
echo delay in the range of approximately 0.5 30 ms, corre- by a P. abramus in flight, recorded by the Telemike at a
sponding to targets 0.1 5 m away from the bat. The three target distance of 3 m. The frequency of the fundamental
lines shown in Fig. 2B represent echo delays between the component was modulated exponentially from 100 to
flying bat and three different objectsthe target wall, floor, 40 kHz, and the bandwidth of the emitted pulse was ex-
and ceiling of the chambercalculated from the three- tended when in flight. A prominent CF-like portion a shal-
dimensional coordinate data of the flying bat. By comparing low sweep portion at the end of the pulse; see Fig. 3 was
1752 J. Acoust. Soc. Am., Vol. 121, No. 3, March 2007 Hiryu et al.: Echo-intensity compensation in echolocating bats
130
FIG. 4. Three-dimensional spatiotemporal reconstruc-
tions of echolocation behavior during landing flights for
three bats. Coordinate grids 1 m2 show the dimen-
sions of the flight chamber 8 3 2 m. Echolocation
pulses recorded by the Telemike are placed alongside
the flight trajectory. The arrow indicates the flight di-
rection of the bat. The amplitude of emitted pulses de-
creased as the bat approached the target wall.
observed beyond a target distance of approximately 2 m. pulse intensity during the search phase. Data were taken
Flight trajectories for the three bats during landing ap- from three flight sessions of each bat. Intensities of all ech-
proaches are illustrated in three-dimensional images in Fig. oes that could be extracted from the displayed sonograms are
4. The echolocation pulses recorded with the Telemike are cumulatively plotted with open circles in Fig. 5A.
superimposed on the flight trajectory to indicate the spa- The pulse intensity in bats intending to land started to
tiotemporal characteristics of echolocation behavior. The decrease at a distance of between 1 and 2 m, and the reduc-
maximum flight speed for a direct approach to the target wall tion was approximately 30 dB before landing. On the other
was approximately 4 m / s at 3 4 m from the target wall. hand, the observed echo intensities were mainly between 40
Several common features were observed in the behavior of to 50 dB weaker than the pulse but remained almost constant
the three bats, including a marked decrease in interpulse in- as the target distance decreased open circles in Fig. 5A.
terval and pulse amplitude as bats approached the target wall
The three bats decreased the intensity of their emitted pulses
approach phase, which usually started at a target distance
logarithmically with target distance Fig. 5B. The reduc-
of between 1 and 2 m. The pulse emission rate increased
tion rate within 2 m of the target distance was 5.6 7.5 dB
from 1020 pulses per second at 3 4 m from the target wall
per halving of the target distance for the three bats. On av-
to 130140 pulses per second immediately prior to landing.
The SPL of each pulse was calculated from the peak-to-peak erage, the reduction rate of pulse intensity was 6.5 dB
amplitude voltage of the observed pulse in the time domain 21.6 dB decrease per decade of the target distance.
by the Telemike. The maximum SPL peak-to-peak at the Figure 5C shows the distribution of all observed echo
microphone above the bats head was approximately 130 dB intensities relative to the average of the pulse intensity dur-
during flight, which was almost 20 dB higher than when the ing the search phase for nine total flight sessions of three
bat was at rest. Before the bats started the approach phase, bats. The distribution showed a single peak, with a mean
the sound pressure level of emitted pulse was almost con- of 42.6 5.5 dB, which corresponds to approximately
stant at 130 dB peak-to-peak search phase. 80 90 dB SPL peak-to-peak. Variation in observed echo
intensity may have been due to variation in the distance or
direction of the object from which the echo returned, and/or
B. Changes in sound intensity of the pulse-echo pair the direction of the emitted pulse.
Figure 5A shows changes in intensity in the peak en- The change in intensity of a pulse-echo pair for one
ergy portions of the pulse and echo as a function of target flight session by a bat is shown in Fig. 6. The intensity of the
distance for the three bats, normalized to the average of the echo returning from the target wall to the bat asterisks in
J. Acoust. Soc. Am., Vol. 121, No. 3, March 2007 Hiryu et al.: Echo-intensity compensation in echolocating bats 1753
131
FIG. 5. Changes in intensity in the peak energy portion the sonogram exhibited a peak in energy at about 50 60 kHz during flight, which was approximately
10 kHz higher than the terminal frequency of the pulse-echo pairs, normalized to the average of intensity of the pulse emitted during the search phase. A
Sound intensity of the pulse solid circles and echo open circles as a function of target distance for three bats. Data were taken from three flight sessions
of each of three bats. B Decrease in pulse intensity for a total of nine flights of three bats. The solid line indicates the correlation line. Pulse intensity
decreased by 6.5 dB on halving the target distance 21.6 dB decrease per decade of the target distance. C Distribution of observed echo intensity relative
to the average of the pulse intensity during the search phase during the landing approach for nine flight sessions of three bats.
IV. DISCUSSION
A. Echo-intensity compensation during flight
Bats are thought to adjust their pulse intensity in re-
sponse to the intensity of the returning echoes that they at-
tend to during echolocation. In this experiment, bats were
expected to attend to the echo from the target wall. The echo
intensity from the target wall asterisks in Fig. 6 recorded
directly by the Telemike was maintained at a constant level
within a target distance of 2 m, whereas the pulse intensity
considerably decreased. This indicates that the bat adjusted
FIG. 6. Sound intensity of pulse-echo pairs as a function of target distance its pulse intensity depending on target distance such that the
for one flight session by Bat A. Asterisks indicate echoes from the target intensity of the echo returning from its destination was con-
wall, determined from sound and three-dimensional coordinate data see
stant. In other words, this finding suggests that the echo from
Methods. The intensity of the echo returning from the target wall indicates
a nearly constant intensity, while the bat considerably decreased the pulse the target destination could be a prominent stimulus to ad-
intensity as it approached the target wall. justment of pulse intensity by bats and that a bat could rec-
1754 J. Acoust. Soc. Am., Vol. 121, No. 3, March 2007 Hiryu et al.: Echo-intensity compensation in echolocating bats
132
ognize an echo from its target the target wall among a a certain optimal intensity range, which may facilitate con-
number of echoes reaching it e.g., Fig. 2A. sistent and precise analysis of target information by the au-
A number of studies have attempted to measure pulse ditory system as seen in DSC Kick and Simmons, 1984.
intensity for bats during flight e.g., Griffin, 1958; Jen and These compensation behaviors associated with adjustments
Kamada, 1982; Vogler and Neuweiler, 1983; Kick and Sim- of pulse frequency and intensity may be a fundamental strat-
mons, 1984; Hartley et al., 1989; Waters and Jones, 1995; egy employed by the animals for streamlining their echolo-
Tian and Schnitzler, 1997; Boonman and Jones, 2002; Hold- cations, and will contribute to inspire the design of future
eried and Helversen, 2003. These studies reported a de- artificial sonar systems or echo-sensing devices.
crease in pulse intensity as a bat approached its target, using
traditional fixed microphones that did not interfere with the B. Multiple echoes reaching a bat during flight
flight path of the bat. In several studies, the reduction rate of
We observed that a number of echoes from surrounding
pulse intensity as a function of distance to the target has been
targets reached the flying bats with different echo delays and
quantitatively estimated e.g., Hartley et al., 1989; Waters
different intensities after a pulse emission Fig. 2. When a
and Jones, 1995; Tian and Schnitzler, 1997; Boonman and
bat was flying, the frequency of echoes received was
Jones, 2002. For example, Tian and Schnitzler 1997 esti-
Doppler-shifted depending on flight speed. A frequency
mated the intensity of emitted pulses in Rhinolophus fer-
sweep was observed with a slight difference among echoes
rumequinum during a landing approach, applying careful
from different targets Fig. 2A, which may be due to dif-
corrections to directional sounds recorded by the fixed mi-
ferences in the extent of the Doppler shift according to the
crophone. Similarly, Noctilio leporinus decreased its pulse
direction or relative velocity of the targets in relation to the
intensity while capturing a small target mealworm at a rate
flight path of the bat. We suggest that the difference in the
of about 6 dB on halving the target distance Hartley, 1989.
frequency slope between returning echoes may provide FM
Although the reduction of emitted pulses by flying bats ap-
bats in flight with direction and velocity information on mul-
proaching a target varied slightly among the studies e.g.,
tiple targets.
Hartley et al., 1989; Waters and Jones, 1995; Boonman and
What specific behavioral strategies do echolocating bats
Jones, 2002, these studies suggest that the bats are supposed
use for processing multiple echoes? Bats are supposed to
to adjust their pulse intensity in response to the intensity of
adjust their call parameters by feedback control in response
returning echoes.
to information from previous echoes. Analysis of the emitted
We found that P. abramus decreased their pulse intensity pulse frequency and interpulse interval shows that bats may
at a rate of 6.5 dB on halving the target distance and exhib- change the focus of their attention between echoes during
ited nearly constant echo intensity from the target wall, flight Hiryu et al., 2005. By periodically focusing on dif-
which should be attended to by a bat on its landing flight. ferent echoes, bats may process multiple auditory streams to
This finding provides direct evidence of echo-intensity com- perceive several targets in real time. In this flight experiment,
pensation by echolocating bats during flight to maintain the the bats occasionally approached the wall attached with a
intensity of the resulting echo within the range necessary for landing mesh referred to as the target wall in case of the
optimal signal processing. The leaf-nosed bat Hipposideros landing scenario and then returned to the starting point with-
terasensis, which emits a CF-FM pulse, decreased its pulse out landing. Interestingly, the bats during the U-turn did not
intensity by approximately 6 dB on halving the target dis- decrease the intensity of emitted pulses as approached that
tance using the same landing flight task as used here Hiryu, wall, as seen when they were landing Fig. 7. In addition,
2005. Recently, echo-intensity compensation was demon- the observed intensity of echoes from that wall did not ap-
strated for the echolocation system in dolphins Au and pear constant, showing a decrease of approximately 20 dB
Benoit-Bird, 2003. Free-ranging dolphins decreased the am- within a distance of 1 m before U-turn point marked with U
plitude of their echolocation signals as they approached the in Fig. 7B. This suggests that the bats making a U-turn did
target by 6 dB per halving of distance. These results suggest not keep their attention on that wall as seen in case of land-
that adjusting pulse intensity may be a common strategy of ing scenario. It is likely that a bat may divert its attention to
biosonar animals approaching a target. other targets depending on its flight direction in order to
The systems employed to compensate for variation in avoid colliding with surrounding walls while making a
intensity and frequency of echoes are unique to biosonar ani- U-turn.
mals. Extant artificial sonar systems are generally designed In this study, we focused on temporal changes in the
to emit a pulse with a fixed frequency to measure the Dop- sound intensity of pulse-echo pairs in the context of echo-
pler shift in the retuning echo in order to detect the velocity intensity compensation. Further investigation of pulse-echo
of the target. In contrast, CF-FM bats adjust their call fre- pairs to which a bat actually listens may lead to other adjust-
quency to maintain an echo frequency within the range they ment mechanisms in echolocation systems for multiple tar-
can hear best Schnitzler, 1968; Suga, 1984. Consequently, gets in the immediate surroundings.
the auditory system in the mustached bats can detect a shift
in echo frequency with a high degree of accuracy, as small as
ACKNOWLEDGMENTS
50 Hz which corresponds to less than 0.1% of the CF2 fre-
quency of the pulse emitted by the mustached bats This work was partly supported by a grant to the Re-
61 kHz Suga, 1984; Riquimaroux et al., 1991. Echo- search Center for Advanced Science and Technology
intensity compensation allows bats to receive echoes within RCAST at Doshisha University from the Ministry of Edu-
J. Acoust. Soc. Am., Vol. 121, No. 3, March 2007 Hiryu et al.: Echo-intensity compensation in echolocating bats 1755
133
echolocating bats. II. The acoustic behavior of the big brown bat, Eptesi-
cus fuscus, when tracking moving prey, J. Acoust. Soc. Am. 91, 1133
1149.
Hartley, D. J., Campbell, K. A., and Suthers, R. A. 1989. The acoustic
behavior of the fish-catching bat, Noctilio leporinus, during pre capture,
J. Acoust. Soc. Am. 86, 827.
Henson, O. W., Jr., Bishop, A. L., Keating, A. W., Kobler, J. B., Henson, M.
M., Wilson, B. S., and Hansen, R. 1987. Bisonar imaging of insects by
Pteronotus p. parnellii, the mustached bat, Natl. Geograph. Res. 3, 82
101.
Hiryu, S. 2005. A study on biosonar engineering based on spatio-
temporal measurement of bats behavior, Ph.D dissertation, Doshisha
University, Kyotanabe, Japan.
Hiryu, S., Katsura, K., Lin, L. K., Riquimaroux, H., and Watanabe, Y.
2005. Doppler-shift compensation in the Taiwanese leaf-nosed bat
Hipposideros terasensis recorded with a telemetry microphone system
during flight, J. Acoust. Soc. Am. 118, 39273933.
Holderied, M. W., and Helversen, O. V. 2003. Echolocation range and
wingbeat period match in aerial-hawking bats, Proc. R. Soc. London, Ser.
B 270, 22932299.
Hope, G. M., and Bhatnagar, K. P. 1979. Electrical response of bat retina
to spectral stimulation: comparison of four microchiropteran species, Ex-
perientia 35, 11891191.
Jen, P. H., and Kamada, T. 1982. Analysis of orientation signals emitted
by the CF-FM bat, Pteronotus p. parnellii and the FM bat, Eptesicus
fuscus during avoidance of moving and stationary obstacles, J. Comp.
Physiol. A 148, 389398.
Kalko, E. 1995. Insect pursuit, prey capture and echolocation in pipist-
relle bats Microchiroptera, Anim. Behav. 50, 861880.
Keating, A. W., Henson, O. W., Jr., Henson, M. M., Lancaster, W. C., and
Xie, D. H. 1994. Doppler-shift compensation by the mustached bat:
quantitative data, J. Exp. Biol. 188, 115129.
FIG. 7. Three-dimensional spatiotemporal reconstruction for a U-turn flight Kick, S. A., and Simmons, J. A. 1984. Automatic gain control in the bats
A, and sound intensity of the pulse solid circles and echo open circles sonar receiver and the neuroethology of echolocation, J. Neurosci. 4,
as a function of the distance to the wall with a landing mesh referred to as 27252737.
target wall in case of the landing flight while a bat making U-turn B. Kobler, J. B., Wilson, B. S., Henson, O. W., Jr., and Bishop, A. L. 1985.
Asterisk indicates the echo from that wall. The bat made a U-turn at a Echo intensity compensation by echolocating bats, Hear. Res. 20, 99
distance of 0.2 m marked U from that wall. The bat during the U-turn did 108.
not decrease the intensity of emitted pulses as approached that wall, as seen Lancaster, W. C., Keating, A. W., and Henson, O. W., Jr. 1992. Ultrasonic
when they were landing. vocalizations of flying bats monitored by radiotelemetry, J. Exp. Biol.
173, 4358.
Neuweiler, G. 1984. Foraging, echolocation and audition in bats, Natur-
wiss. 71, 446455.
cation, Culture, Sports, Science, and Technology MEXT of Riquimaroux, H., Gaioni, S. J., and Suga, N. 1991. Cortical computa-
Japan: Special Research Grants for the Development of tional maps control auditory perception, Science 251, 565568.
Characteristic Education from the Promotion and Mutual Aid Riquimaroux, H., and Watanabe, Y. 2000. Characteristics of bat sonar
Corporation for Private Schools of Japan and the Innovative sounds recorded by a telemetry system and a fixed ground microphone,
The seventh western pacific regional acoustics conference (WEST-
Cluster Creation Project. PRACVII), 233238.
Schnitzler, H. U. 1968. Die Ultraschallortungslaute der Hufeisen-
Au, W. W. L., and Benoit-Bird, K. J. 2003. Automatic gain control in the Fledermause Chiroptera-Rhinolophidae in verschiedenen Orientierungs-
echolocation system of dolphins, Nature London 423, 861863. situationen The Ultrasonic sounds of horseshoe bats Chiroptera-
Boonman, A., and Jones, G. 2002. Intensity control during target ap- Rhinolophidae in different orientation situations, Zeitschrift fur
proach in echolocating bats; stereotypical sensori-motor behaviour in Vergleichende Physiologie 57, 376408.
Daubentons bats, Myotis daubentonii, J. Exp. Biol. 205, 28652874. Schnitzler, H. U., Kalko, E., Miller, L., and Surlykke, A. 1987. The
Gaioni, S. J., Riquimaroux, H., and Suga, N. 1990. Biosonar behavior of echolocation and hunting behavior of the bat, Pipistrellus kuhli, J. Comp.
mustached bats swung on a pendulum prior to cortical ablation, J. Neu- Physiol. A 161, 267274.
rophysiol. 64, 18011817. Schuller, G., Beuter, K., and Schnitzler, H. U. 1974. Response to fre-
Ghose, K., and Moss, C. F. 2003. The sonar beam pattern of a flying bat quency shifted artificial echoes in the bat Rhinolophus ferrumequinum, J.
as it tracks tethered insects, J. Acoust. Soc. Am. 114, 11201131. Comp. Physiol. A 89, 275286.
Griffin, D. R. 1958. Listening in the Dark Yale University, New Haven, Simmons, J. A. 1974. Response of the Doppler echolocation system in
CT. the bat, Rhinolophus ferrumequinum, J. Acoust. Soc. Am. 56, 672682.
Gustafson, Y., and Schnitzler, H. U. 1979. Echolocation and obstacle Simmons, J. A., Moffat, A. J., and Masters, W. M. 1992. Sonar gain
avoidance in the hipposiderid bat Assellia tridens, J. Comp. Physiol. A control and echo detection thresholds in the echolocating bat, Eptesicus
131, 161167. fuscus, J. Acoust. Soc. Am. 91, 11501163.
Hartley, D. J. 1992a. Stabilization of perceived echo amplitudes in Simmons, J. A., and Stein, R. A. 1980. Acoustic imaging in bat sonar:
echolocating bats. I. Echo detection and automatic gain control in the big echolocation signals and the evolution of echolocation, J. Comp. Physiol.
brown bat, Eptesicus fuscus, and the fishing bat, Noctilio leporinus, J. A 135, 6184.
Acoust. Soc. Am. 91, 11201132. Suga, N. 1984. The extent to which biosonar information is represented
Hartley, D. J. 1992b. Stabilization of perceived echo amplitudes in in the bat auditory cortex, in Dynamic aspects of neocortical function,
1756 J. Acoust. Soc. Am., Vol. 121, No. 3, March 2007 Hiryu et al.: Echo-intensity compensation in echolocating bats
134
edited by G. M. Edelman, W. E. Gall, and W. E. Cowan Wiley, New insect-catching horseshoe bats, Naturwiss. 69, 193194.
York, pp. 315373. Vogler, B., and Neuweiler, G. 1983. Echolocation in the noctule Nycta-
Tian, B., and Schnitzler, H. U. 1997. Echolocation signals of the greater lus noctula and horseshoe bat Rhinolophus ferrumequinum, J. Comp.
horseshoe bat Rhinolophus ferrumequinum in transfer flight and during Physiol. A, 152, 421432.
landing, J. Acoust. Soc. Am. 101, 23472364. Waters, E., and Jones, G. 1995. Echolocation call structure and intensity
Trappe, M., and Schnitzler, H. U. 1982. Doppler-shift compensation in in five species of insectivorous bats, J. Exp. Biol. 198, 475489.
J. Acoust. Soc. Am., Vol. 121, No. 3, March 2007 Hiryu et al.: Echo-intensity compensation in echolocating bats 1757
135
136
Biosonar and Neural
Computation in Bats
Bats extract remarkably detailed information about their
surroundings from biosonar signals. Neurons in their auditory
systems are highly specialized for performing this task
by Nobuo Suga
I
t used to be a common misconcep echolocating abilities of bats. The tecting targets larger than the wave
tion that bats' use of sound puls well-defined characteristics of a bat's length of the signal, because the re
es to navigate and locate prey is a auditory world make the animal ideal flected sound energy is highly con
crude system, the acoustic equivalent for elucidating the information proc centrated at a particular frequency. It
of feeling one's way in the dark with eSSing that goes on in its auditory is also ideal for measuring Doppler
a cane. But biosonar has since been system. Similar mechanisms are un shifts. The CF pulse is not appropriate,
shown to be anything but crude: an doubtedly shared by other animals. however, for locating a target precisely
echolocating bat can pursue and cap or discerning its details. A larger num
T
ture a fleeing moth with a facility and here are some 800 species of Mi ber of frequencies is needed to obtain
success rate that would be the envy of crochiropteran bats in the world more information about target fea
any military aerospace engineer. today, all of which are presumed tures. Bats broaden their frequency
In addition to providing information to echolocate. These species live in bandwidth by producing harmonics
about how far away a target is, bat diverse habitats and vary greatly in and by emitting FM bursts that sweep
sonar can relay some remarkable de behavior and physical characteristics. over a wide frequency range. FM puls
tails. Doppler shifts-changes in the Their biosonar pulses also differ, even es also contain more information
frequency of the echo relative to the among species within the same ge about time and so are used to com
original signal-convey information nus. Nevertheless, these pulses can pute echo delays and thereby deter
not only about the relative velocity of be classified into three types: constant mine the distance to a target.
a flying insect but also about its wing frequency (CF ), frequency modulated Certain bat species control the ener
beat. The amplitude of the echo, com (FM) and combined CF-FM. CF puls gy in each harmonic depending on the
bined with the delay, indicates the size es consist of a single frequency, or distance to a target. If the target is far
of the target. The amplitudes of the tone. FM pulses sweep downward and away, they amplify the lower harmon
component frequencies correspond to sound like chirps. Combined CF -FM ics, which are less attenuated by the
the size of various features of the pulses consist of a long, constant tone air. But if the target is nearby, they
target. Differences between the ears followed by a downward chirp, iiiiiiu. enhance the higher harmonics to ob
in intensity and arrival time of sound In many bats the tones are not pure tain finer details of the target. When
give the azimuth of the target, where but rather consist of a fundamental, closing in on prey, Microchiropterans
as the interference pattern of sound or first, harmonic and several higher shorten the duration of pulses and
waves reflected within the structure of harmonics (multiples of the funda increase the rate of pulse emission, up
the outer ear gives the elevation. mental frequency). to 200 per second in FM bats and up to
The complex neural computations Most bat species emit only one type 100 per second in CF-FM bats. This
needed to extract this information oc of pulse. The little broWn bat, Myotis adjustment occurs not only because
cur within a brain the size of a large lucifuqus, is an "FM" bat; it emits FM bats need to characterize the prey in
pearl. For the past 27 years my col pulses lasting between .5 and three greater detail but also because when
leagues and I have been exploring the milliseconds and sweeping downward the distance between a bat and its
neural mechanisms that underlie the by about one octave. The mustached prey is small, the angular position of
bat, Pteronotus parnellii, is a "CF-FM" the prey changes more rapidly, and so
bat; it emits long CF pulses lasting the bat needs to emit more signals to
NORUO SUGA has been professor of between five and 30 milliseconds fol track the prey accurately.
biology at Washington University in Saint lowed by a short FM sweep lasting The hunting strategies and behavior
Louis, Mo. , since 1976. Suga was born in between two and four milliseconds. of a bat species are directly related to
Japan and attended the Tokyo Metropol Several species change their pulses, the characteristics of its biosonar. The
itan UniverSity, where he received his depending on the situation. The fish
B.A. in 1958 and his Ph.D. in biology in
catching bat, Noctilio leporinus, for ex
1963. He then went to Harvard Universi
ample, emits CF and CF - FM pulses
ty as a research associate, where he first
studied the auditory system of bats with
while cruising in flight but emits FM MUSTACHED BAT sweeps in for a mid
Donald R. Griffin. pulses while hunting prey. flight drink from a pond. This species'
A long CF pulse is excellent for de- biosonar has been studied extensively.
zation of its auditory system. Since bat describe findings for this species. pler-shifted echo at 63 kilohertz from
biosonar was established by Donald R . a stationary object, it reduces the fre
Griffin and Robert Galambos four dec ying mustached bat detects the quency of emitted pulses by about 1.8
ades ago, neuroethologists have stud relative velocity of objects by kilohertz, so that subsequent echoes
ied the auditory system in several bat the Doppler shift in the echoes. are stabilized at a "reference" fre
species but most of all in the little When a bat flies toward a stationary quency of around 61.2 kilohertz.
brown bat, the mustached bat and the object, the pulses that strike and are These bats turn out to be special
horseshoe bat, Rhinolophus ferrum reflected by it become compressed, or ized to analyze tiny differences in fre
equinum. Each of these bats produces Doppler-shifted. The echo received by quencies near the reference frequen
distinctive biosonar pulses. the bat is therefore uniformly higher cy. Hence, Doppler-shift compensa
The auditory mechanisms of the in frequency than the emitted pulse. tion brings the echo CF2 into the range
mustached bat have been the most When the animal flies toward a fly at which the bat can most easily detect
thoroughly examined. The biosonar ing insect, the insect's beating wings ripples from beating insect wings.
and peripheral auditory system of this introduce oscillating frequency shifts, This specialization begins in the
bat were first described in 1964 and which are superposed on the overall bat's ear. Of particular interest is the
1972, respectively, by Alvin Novick Doppler shift, like small surface rip inner ear, or cochlea, which contains
and his co-workers at Yale University. ples on an ocean wave. the basilar membrane, a thin, elongat
In 1972 I began research on the pe Certain bats, such as the mustached ed sheet curled up like a snail. When
ripheral auditory system of the mus bat and horseshoe bat, can detect rip sound waves vibrate the eardrum, this
tached bat with James A. Simmons, ples from insect wings against the vibration is conducted to the basilar
now at Brown University, and on the echoes associated with stationary ob membrane, stimulating tiny hair cells
central auditory system with Philip jects, such as walls and vegetation. on the membrane. The excitation is
H.-S. Jen, now at the University of Mis How do they do it? Part of the answer transmitted, via the spiral ganglion
souri at Columbia. Subsequently, To lies in a trick called Doppler-shift cells, along the auditory nerve fibers
shiki Manabe and Kazuro Kujirai, now compensation, first seen by Hans-Ul to the brain.
at Yokohama City University, William rich Schnitzler of the University of The neural signal produced at the
E . O'Neill, now at the University of Tiibingen. A mustached bat at rest cochlea must contain all the infor
Rochester, and nearly two dozen oth emits a fundamental tone of around mation vital to the bat. The physical
ers have joined in studying this bat's 30.5 kilohertz, along with three higher properties of an acoustic signal-am
central auditory system and its neural harmonics; the "resting" frequency of plitude, time and frequency-must in
138
1990 SCIENTIFIC AMERICAN, INC
MUSTACHED BAT lITILE BROWN BAT
N CF. N 120
I-
,u'
\
c:r:
UJ
120 FM.\ \ \\
UJ
J:
J: CF, 90
0 0
90 \
:;z
:;z
FM,\ \ \\ \\
>-
u
z
UJ
=>
60
CF,
CF,
FM, \
\ \ \\,'
>-
u
z
UJ
=>
60
.\
ECHO
\ ..\
0
UJ
30
I FM,\ \\ 0
UJ
30
,'. , '. ". \ , .. \',
c:r: PULSE ECHO c:r:
LL. 0 LL. 0 I i
0 20 40 60 80 100 0 20 40 60 80 100 120
TIME (MILLISECONDS) TIME (MILLISECONDS)
" --,,,
X - ....
\
""tt!f? ',j :
HUNTS IN VEGETATION
BIOSONAR PULSE of the mustached bat consists of a long, target, the bat emits shorter pulses at a higher rate, while
constant-frequency ( CF ) component followed by a short, fre keeping the same tones. The little brown bat emits only FM
quency-modulated (FM ) component. Each pulse contains four chirps. When nearing a target, it emits shorter, lower chirps at
harmonics (indicated by subscripts). When closing in on a a faster rate. Each species emits pulses suited to its behavior.
turn be translated into neural activity. hertz) and CF3 (92-kilohertz) signals. nerve impulses from single neurons,
Amplitude is expressed by the rate at In the mustached bat the great sen my colleagues and I studied the re
which the auditory nerve fibers dis sitivity and sharp tuning of the au sponses of neurons in the mustached
charge impulses: the greater the am ditory periphery to the CF2 frequency bat's auditory system as we stimulat
plitude, the higher the discharge rate. are combined with Doppler-shift com ed the animal with biosonar signals.
The duration of signals and the inter pensation to proffer three advantages. What we discovered was an extraordi
vals between them are mimicked by First, the auditory periphery is exqui narily developed system for process
the pattern of the nerve impulses. The sitely sensitive to the CF2 echo (near ing that information. In particular, we
frequency of the signal is expressed 61 kilohertz) but is insensitive to the found that different processing tasks
by location on the basilar membrane: bat's emitted CF2 pulse (near 59 kilo are parceled out among several ana
high frequencies vibrate the portion hertz) during Doppler-shift compen tomically distinct areas of the audi
nearest the eardrum, whereas lower sation; hence, masking of the echo by tory cortex. One region contains neu
ones stimulate portions farther in. the emitted pulse is minimal. Second, rons that respond only to certain fre
A certain portion of the mustached the sharply tuned neurons are well quencies and amplitudes of echoes. A
bat's basilar membrane is unusually able to detect the signal even if it is second region responds only to fre
thick. This thickness is related to ex embedded in background noise. Third, quency differences between pulses
treme sensitivity to frequencies of be the array of sharply tuned neurons and echoes. A third region is sensitive
tween 61.0 and 61.5 kilohertz (the CF2 has a high likelihood of picking up the to the time interval between pulses
of the Doppler shift-compensated echo from the beating wings of a fly and echoes.
echoes) as well as insensitivity to fre ing insect as the echo sweeps up and By far the largest of the specialized
quencies of around 59.5 kilohertz (the down in frequency. regions in the mustached bat's audi
CF2 of the Doppler shift-compensat These advantages enable the mus tory cortex is the one that processes
ing pulses). In other words, the mem tached bat to hunt insects success Doppler-shifted CF2 signals. This re
brane is strongly stimulated by the fully, even in dense vegetation. Dop gion, called the DSCF area, represents
echoes but poorly stimulated by the pler-shift compensation and frequen only a narrow sliver of the frequen
animal's own vocalizations. cy tuning do not exist in FM bats, such cy range, between 60.6 and 62.3 kilo
The frequency selectivity of the spi as the little brown bat, which hunt in hertz (when the bat's resting frequen
ral ganglion cells is extremely high open air. cy is 61.00 kilohertz). Yet it occupies
within the key range of 61.0 to 61.5 30 percent of the primary auditory
O
kilohertz. They are tuned to single nce the auditory signal has cortex. The exact frequencies overrep
frequencies. That is, each neuron has a been coded into nerve Signals, resented differ among individual bats
"best" frequency (the frequency that it must be further analyzed to according to their resting frequencies.
evokes the largest response), which extract such information as the veloci In other words, each bat's auditory
differs slightly from that of its neigh ty or distance of prey. This process system is personalized.
bors. Indeed, these neurons are so occurs in the central auditory system. Similar overrepresentation is found
sharply tuned to their best frequen From the cochlea, signals are process in the brain wherever the signal being
cies that they can detect shifts as small ed sequentially, beginning at the coch processed is critical to an animal's
as .01 percent. Flying insects can eas lear nucleus and proceeding to the behavior. For example, in cats and
ily evoke frequency shifts an order of lateral lemniscus, inferior colliculus, monkeys the visual cortex overrepre
magnitude greater. The auditory pe medial geniculate body and finally to sents the fovea, the area of the retina
riphery is also highly tuned to analyze the auditory cortex. where visual acuity is highest. the pri
frequency shifts near CF1 (30-kilo- Using slender electrodes to record mate somatosensory cortex overrep-
'
inhibitory signals from adjacent neu
:" SUBTENDED
,
- -
-
ANGLE PLUS
rons enhance the selectivity of a neu
DISTANCESI"
ron to a particular stimulus.
SUBTENDED ANGLE
I :
T
he auditory cortex of the mus
tached bat is about 900 microns, ,
or some 40 to 50 neurons, thick.
When we inserted a recording elec
trode into the cortex, we found that
all of the neurons perpendicular to
the surface are tuned to an identical
frequency and amplitude. Hence, the
DSCF area has a "columnar organiza
tion." Such columnar organization was
first discovered in 1959 in the somato
sensory cortex of monkeys by Vernon
B. Mountcastle of Johns Hopkins Uni
versity and later in the visual cortex of
cats by David H. Hubel and Torsten N.
Wiesel [see "Brain Mechanisms of Vi
sion," by David H. Hubel and Torsten
N. Wiesel; SCIENTIFIC AMERICAN, Sep
tember, 19791.
When we inserted an electrode tan
gentially to the cortical surface of the
DIVERSE INFORMATION can be extracted from biosonar signals. The time delay and
DSCF area, we found that the preferred
Doppler shift of the echo indicate the distance and relative speed of the target. Rap
frequency and amplitude gradually
id flutters betray the presence of beating insect wings. Echo amplitude depends on
change, indicating the existence of fre the relative size (subtended angle) and distance of the target. lnteraural time and
quency-versus-amplitude coordinates amplitude differences convey the azimuth of the target. Interference patterns of
along the surface of the DSCF area. sound waves reflected within the structure of the outer ear indicate the elevation.
One can (crudely) picture the area as a
bicycle wheel: as one moves outward
along a spoke, the best frequency of
the neurons increases; as one moves ----FOR-WARDSWING--> <--BACKWARDSWING----
circularly from one spoke to the next,
the best amplitude changes.
DOPPLER SHIFT BY
What is the function of the DSCF 62
./ PENDULUM SWING
area? Neurons in the area respond
purely to the amplitude and frequen
UJ
:I:
cy of the echo CF2 , regardless of the o
-' 6 1 :.
:: .::: .:: .....------------------------ - .-':;.- 1 : :.::.-:..:.: : .: !
.
- . .
.. . .....: :: :
52
. ..... . ..
frequency of the emitted pulse. DSCF e
. , -.
.. . .:- :. .
RESTING FREQUENCY
neurons, then, presumably are related
>
: :.
to the acuity of frequency and ampli u
z .. . .
/.::.\. ...
UJ ...
::>
tude discrimination, as well as to the .
.
;::
detection of changes in frequency and 8 60 -: ..... :.:.
c:r:
SOUNDS EMITTED
amplitude that would be evoked by u..
BY A BAT DURING ./ :6.;::
flying insects. FIVE SWINGS
According to recent experiments by
Stephen J. Gaioni, Hiroshi Rikimaru 59--------------------L----- --------------------
0 1 2 3
and me, if the DSCF area is destroyed,
TIME (SECONDS)
a bat can no longer discriminate tiny
differences in frequency-only large DOPPLER-SlDFT COMPENSATION is demonstrated by placing a mustached bat on a
ones. The animal requires twice as pendulum. During the forward swing the animal lowers the frequency of its emitted
much time to carry out Doppler-shift pulse (red) such that the echo stays at a "reference" frequency. The animal does not
compensation and performs the task compensate for Doppler shift during the backward swing. O'Dell W. Henson, Jr.,
only half as well. From this we spec- of the University of North Carolina at Chapel Hill first performed the experiment.
1990
SCIENTIFIC AMERICAN June 140 63
1990 SCIENTIFIC AMERICAN, INC
C1
Vl
...J
UJ
co 400
HORSESHOE BAT id
' !
0
UJ
Cl ...J::J
0 MUSTACHED BAT
_UJ
...J
300 U
60
::::>
0:: Cl Vl
0 Vl
...J
UJ
I- 200 o
U
<{
I
Vl
u.. UJCl
O::Z 40
I::::>
100 1-0
Vl
BASILAR MEMBRANE
0
HIGH FREQUENCY -> LOW FREQUENCY 0 25 SO 75 100 67
FREQUENCY (KILOHERTZ)
BASIlAR MEMBRANE is vibrated at the outer part by high in horseshoe bat auditory neurons tuned to 83 kilohertz, the
frequencies and at the inner part by lower frequencies. The CF2 harmonic of that species. Because the little brown bat
vibrations excite hair cells, which in turn excite spiral ganglion produces only an FM pulse, it has no such neurons. The graph
cells. The middle graph shows the high selectivity in the mus on the right shows response thresholds of 12 mustached bat
tached bat of neurons tuned to the three higher CF harmon neurons; for each neuron the threshold plummets at a par
ics, particularly at 6 1 kilohertz ( CF2) but also at 30 kilohertz ticular frequency. The plunge is very steep for neurons that
( CF 1) and 92 kilohertz ( CF 3)' High selectivity is also found respond to frequencies near 6 1 kilohertz, the CF2 harmonic.
N ECHO
I-
0::
UJ 90 FM, FM-FM AREA
I ---................
0
...J 60
:;;2 PULSE
>-
30 FM,
u
Z
NERVE
UJ 0
::::>
0 RESPONSE
UJ
0::
u..
0 10
20 30
TIME (MILLISECONDS)
18
L<,: 1
N
I-
0::
f&><s 4 6 14
'
90 8
(MILLISECO NDS)
10
ECHO
ECHO DELAY
UJ
I
0
...J 60
\ FM,
:;;2 PULSE
>-
30
FM,
u MEDULLA
NERVE
Z 0
UJ
RESPONSE
::::>
0
UJ
0::
u..
0 10
20 30
CF, FRE
QU ENCY
CF/CF AREA
(KILOHERrz
)
2
TIME (MILLISECONDS) "'-"7l :2'. 2 9.5
30.0
AUDITORY NERVE 30.5
8. 7
N
I-
0::
UJ
90 ECHO CF, \.
ACTUAL
I 60
0 SIZE -2
...J
:;;2 PULSE CF,
30
>-
u
Z
UJ
::::>
0
0
....K RVE RESPONSE
UJ
0::
u..
0 10 20 30
TIME (MILLISECONDS)
COMPUTATIONAL MAPS in the auditory cortex of mustached each neuron responds to a specific echo delay and amplitude.
bats represent echo delay (or distance) and Doppler shift (or In the CF / CF area (tan), neurons along the blue lines respond
relative velocity). In the FM-FM area (green), neurons along to a specific CF 1 combined with varying CF2' Neurons along the
each black line respond to a specific echo delay. The top graph black lines respond to Doppler shifts corresponding to a spe
(right) shows the delay-tuning curves of six FM-FM neurons; cific relative target velocity. The bottom graph (right) shows
O
ne important function that we slightly higher to account for Doppler ry pathway, my graduate student John
understand well is that of per shift in the echo.) F. Olsen, now at Stanford University,
ceiving the relative velocity of When the pulse, echo, CF tone or FM found that neurons in certain regions
a target. To do this, the bat's brain sound is delivered alone, these neu of the inferior colliculus are tuned to
must compute the Doppler shift be rons respond very weakly. But when single frequencies: CFI, CF2 or CF3
tween the emitted pulse and the echo. the pulse and echo are combined, they They send signals to a certain region
In other words, there must be neurons show a remarkably strong response, of the medial geniculate body, where
that "examine" the frequency relation becoming sensitive to paired signals the signals are integrated, so that neu
between the two sounds. that are as much as 6,300 times weak rons there respond to the combina
We have found such neurons in a er than the smallest unpaired signals tion of CFI frequencies with specific
part of the auditory cortex that we call capable of evoking a response. CF2 or CF3 frequencies. These neurons
the CF /CF area. There are two types of The functional organization of the then project to the CF /CF area of the
s
CF /CF neurons: CF,/CF2 and CF,/CF3 , CF /CF area supports our conclusion auditory cortex.
each of which forms a distinct region. that these neurons are primarily de
These neurons are sensitive to the voted to processing velocity infor o far I have described auditory
mation. Each column of neurons re processing of only the CF com
sponds best to a particular combina ponents of biosonar. But the
tion of two frequencies, for example, mustached bat also produces an FM
TARGET RANGE (CENTIMETERS) 29.60 and 61.20 kilohertz. Neurons at sound at the end of the CF component.
50 100 150 200 250 a slightly different location respond to What is the purpose of this FM sound?
90 '------Tr---
a different combination, say, 30.05 The FM Signal provides the primary
and 61.10 kilohertz. cue for measuring the time interval
:::; Further measurements showed that between a pulse and echo-and hence
u:j 70 the best combination of frequencies the distance to a target. A one-milli
COw
O....J varies in a regular way along the sur second echo delay corresponds to a
Ww
0 0:: face of the cortex. The preferred CFI target distance of 17.3 centimeters (at
::>
frequency increases along one axis, an air temperature of 25 degrees Cel
g 50
o
::co..
and the CF2 and CF3 frequencies in sius). Simmons found that several spe
lflo crease along the axis at right angles to cies of bats can detect a difference
o:: z it. In other words, the CF /CF region is in distance of between 12 and 17 mil
6 30
Vl organized by a frequency-versus-fre limeters, which means they can dis
quency coordinate system, in which a criminate a difference in echo delay
specific location represents a particu of between 69 and 98 millionths of
10 lar relative target velocity. a second!
0 3 6 9 12 15 Because the harmonics are simply As in the CF /CF area, the processed
ECHO DELAY (MILLISECONDS)
H
ow do pathways in the audito delays as strongly as do the FM -FM liculus is a huge nucleus protruding
ry system give rise to neurons neurons in the auditory cortex and between the cerebrum and cerebel
that are sensitive to pulse-echo that they respond somewhat to un lum. Nerve fibers ascend from the
delays? As with neurons that respond paired pulses and echoes. Significant ventrolateral portion of the colliculus
to combinations of frequencies, those ly, these neurons always take longer to to its dorsomedial portion. Impulses
that respond to pulse-echo delays are respond to the pulse FMj than they do travel a distance of about two millime
first found in the medial geniculate to the echo FM2 , FM3 or FM4 What is ters along these nerve fibers, exciting
body. To find out how these neurons more, the difference in response laten 100 or more neurons along the way,
are connected to the rest of the au cy is the same as the preferred pulse and introduce delays of up to eight
ditory pathway, Olsen injected them echo delay for each neuron. milliseconds.
with horseradish perOxidase, which These observations indicate an in Recall, however, that delays in the
diffuses along nerve pathways. The teresting mechanism for the computa FM-FM area of the auditory cortex are
substance made its way to two distinct tion of echo delays. When a bat hears as long as 18 milliseconds. Hence, the
groups of cells in the inferior collicu its own emitted pulse FMj, the neural delay lines in the inferior colliculus do
Ius as well as to the FM-FM area of the response is delayed as it travels to not account for the full range of de
auditory cortex. ward the medial geniculate body. But lays. The data suggest that additional
Clearly, the two groups of collicular the response to the returning echo is delays are created by inhibitory syn
s
neurons, one group tuned to the pulse not delayed. The delayed response to apses in the medial geniculate body.
FMj and the other to higher harmon the pulse then arrives at a particular
ics in the echo FM, converge on a sin- neuron in the medial geniculate body o far I have discussed how the
mustached bat extracts such in
.J __
formation as the velocity and
ECHO (FM" FM, OR FM.) range of a target. But in doing so, I
:> have avoided the question of why the
bat emits a variety of harmonics. Why
is one harmonic not enough? Techni
cally it should be, but the animal has
to cope with another problem. Bats
live in colonies with hundreds of oth
i j'7"
er bats, and somehow they must be
able to use echolocation without get
0 31
.... 0 00000
rl;'r'VE
NEURONS
I .... ill! ting confused. Several mechanisms, in
cluding binaural hearing, help to cope
with this air-traffic controller's night
INTER-
mare. One important mechanism de
pends on the first harmonic
oro
NEURONS The first harmonic is the weakest
component of the emitted pulse, con
DELAY
LINES
I
CF SELECTIVITY FM SELECTIVITY
iii i iii
PARALLEL PATIIWAYS process different streams of biosonar organized (yellow); the large area (pink) at the center repre
information (left). Various CF and FM harmonics excite differ sents 60.6 to 62.3 kilohertz. CF and FM signals are integrat
ent parts of the basilar membrane, and signals are sent to the ed in the medial geniculate body, giving rise to neurons that
auditory cortex via several subcortical nuclei. Higher up in the respond to specific combinations of CF or FM signals. These
auditory pathway, the neurons become more narrowly selec combination-sensitive neurons project axons to the CF jCF or
tive for frequency and amplitude. In the primary auditory cor FM-FM areas of the auditory cortex, creating maps that corre
tex, frequencies from 10 to 100 kilohertz are tonotopically spond to the relative target velocity (tan) or distance (green).
harmonics that are delayed or Dop Robert R. Capranica of Cornell Uni sory cortex and visual cortex in mon
pler-shifted, can then stimulate FM versity and Albert S. Feng of the Uni keys and cats. Our work on bats' audi
FM and CF /CF neurons. In this way, the versity of Illinois at Urbana-Cham tory cortex, then, contributes not only
neural processing of biosonar signals paign have found that frogs process to the understanding of hearing but
is shielded from the cacophony of complex sounds, such as mating calls, also to the understanding of sensory
echoes generated by the colony. by way of neurons that respond to systems in general.
The suppression of the first har combinations of two essential signal
monic confers another important ad elements. Daniel Margoliash of the Uni
vantage. To avoid being devoured by versity of Chicago has observed simi RJRlliER RFADING
PERFORMANCE OF AIRBORNE ANIMAL So
bats, many species of moths have "bat lar neurons in songbirds. The process
NAR SYSTEMS. H. -U. Schnitzler and O. W.
detectors," or auditory receptors that ing of biologically crucial sounds by Henson, Jr. , in Animal Sonar Systems.
are highly sensitive to sounds of be combination-sensitive neurons, then, Edited by Rene-Guy Busnel and James
tween 15 and 40 kilohertz but relative appears to be an important neural F. Fish. Plenum Press, 1980.
ly insensitive to higher frequencies. By mechanism shared among diverse THE ExTENT TO WHICH BIOSONAR iNFOR
suppressing the first harmOniC, which types of animals. MATION Is REPRESENTED IN THE BAT Au
It seems likely that the human audi DITORY CORTEX. Nobuo Suga in Dynam
is between 24 and 31 kilohertz, a mus
tached bat can approach moths close tory system also uses such neural ar
ic Aspects of Neocortical Function. Edit
ed by Gerald M. Edelman et al. John
Wiley & Sons, Inc., 1984.
ly without alerting them. rays to process speech sounds. Even
though the processing of speech is iNHIBmoN AND LEvEL-TOLERANT FRE
H
earing is as critical to bats as unique and predominantly a function QUENCY TuNING IN THE AUDITORY COR
seeing is to visually oriented of the neocortex, the subcortical audi TEX OF THE MUSTACHED BAT. Nobuo
animals, and so it is not sur tory regions, with their combination Suga and Koichi Tsuzuki in Journal of
pnsmg to find evidence of elegant sensitive neurons, may analyze the Neurophysiology, Vol. 53, No. 4, pages
1109-1145; April, 1985.
neural computations in bats' auditory sounds of speech to a greater extent
AUDITORY FUNCTION: NEUROBIOLOGICAL
cortex. But sound is also biologically than has been generally thought. The
BASES OF HEARING . Edited by Gerald M.
important for many other animals. Do work pioneered by Mountcastle, Hubel Edelman, W. E. Gall and W. M. Cowan.
they have similar neural mechanisms? and Wiesel has shown that similar John Wiley & Sons, Inc. , 1988.
It turns out that many do. mechanisms underlie the somatosen-
by Masakazu Konishi
W
hy do people have two ears? we hear one signala phantomorigi- sitive video camera in a totally dark
We can, after all, make sense of nating from somewhere inside or out- room, I was impressed by the speed and
sounds quite well with a single side the head. If the stimuli fed to the accuracy with which it turned its head
ear. One task, however, requires input ears are equally intense (equally loud) toward a noise. I concluded that the
from both organs: pinpointing the ex- and are conveyed simultaneously, we head-turning response might help un-
act direction from which a sound, such perceive one sound arising from the cover whether such animals use binau-
as the cry of a baby or the growl of a middle of the head. If the volume is low- ral fusion in locating sound. If they did,
dog, is emanating. In a process called ered in just one ear or if delivery to that studies of their brain could help eluci-
binaural fusion, the brain compares in- ear is delayed, the source seems to move date how such fusion is accomplished.
formation received from each ear and in the direction of the opposite ear. As I had anticipated, the head-turning
then translates the dierences into a This much has long been known. response did prove extremely useful to
unied perception of a single sound is- What is less clear is how the brain man- me and my postdoctoral fellows, partic-
suing from a specic region of space. ages to detect variances in timing and ularly after I established a laboratory at
Extensive research has shown that the intensity and how it combines the re- Caltech in 1975. In some of our earliest
spatial cues extracted by the human sulting information into a unied spa- research there, Eric I. Knudsen, now at
brain are dierences in the arrival time tial perception. My colleagues and I at Stanford University, and I obtained in-
and the intensity, or force, of sound the California Institute of Technology direct evidence that barn owls, like hu-
waves reaching the ears from a giv- have been exploring this question for mans, must merge information from the
en spot. Dierences arise because of more than 15 years by studying the be- two ears to locate a sound. When one
the distance between the ears. When a havior and brain of the barn owl (Tyto ear was plugged, the animals turned
sound comes from a point directly in alba). Recently we have uncovered al- the head in response to noise from a
front of us, the waves reach both ears most every step of the computational loudspeaker, but they did not center
at the same time and exert equal force process in these animals. (The only oth- on the speaker [see The Hearing of the
on the receptive surfaces that relay in- er sensory system that is as completely Barn Owl, by Eric I. Knudsen; SCIENTIF-
formation to the brain. But if a sound dened belongs to a sh.) We nd that IC AMERICAN, December 1981].
emanates from, say, left of center, the the owl brain combines aural signals In the early 1980s Andrew Moise
waves will reach the right ear slightly relating to location not all at once but and I additionally showed that the barn
after the left. They will also be some- through an amazing series of steps. In- owl extracts directional information
what less intense at the right because, formation about timing and intensity is from disparities in the timing and the
as they travel to the far ear, some frac- processed separately in parallel path- intensity of signals reaching the two
tion of the waves will be absorbed or ways that converge only late in those earstechnically called interaural time
deected by the head. pathways. It is highly probable that hu- dierences and interaural intensity dif-
The brains use of disparities in tim- mans and other mammals achieve bin- ferences. As part of that eort, we mea-
ing and intensity becomes especially ob- aural fusion in much the same manner. sured the dierences that arose as we
vious when tones are delivered sepa- moved a speaker across the surface of
I
rately to each ear through a headset. In- rst thought of examining the neu- an imaginary globe around an owls
stead of perceiving two distinct signals, ral basis of sound location in owls head. Microphones we had placed in the
in 1963, when I heard Roger S. ears relayed the signals reaching each
Payne, now at the Whale Conservation ear to a device that measured arrival
Institute in Lincoln, Mass., report that time and volume. When we eased the
MASAKAZU KONISHI has been Bing the barn owl can catch a mouse readily speaker from the midline of the face
Professor of Behavioral Biology at the
California Institute of Technology since
in darkness, solely by relying on acous- (zero angle) 90 degrees to the left or
1980. He earned a doctorate in zoology tic cues. I had recently earned a doc-
from the University of California, Berke- torate in zoology and wanted to know
ley, in 1963. Three years later he joined more about how animals identify the
BARN OWL PINPOINTS PREY in the dark
the faculty of Princeton University, where position of a sound source, but I had yet by listening. It determines the appro-
he studied hearing and vocalization in to choose a species to study. Three years
songbirds as well as sound localization
priate trajectory in which to y by com-
later, at Princeton University, I observed paring dierences in the timing and the
in owls. Konishi moved to Caltech as a
the exquisite aural abilities of barn owls intensity of sounds reaching its two ears.
professor in 1975. In his free time, he
enjoys hiking, skiing and visiting En- for myself after I obtained three of them An infrared strobe ashing ve times
gland to attend sheep-dog competitions. from a bird-watcher. When I watched per second caught this barn owl in ac-
one of the owls through an infrared-sen- tion in the authors laboratory.
T
o learn how the brain carries out
globe around the owls head. Dierences in timing locate the sound in the horizon-
binaural fusion, we had to exam-
tal plane (a); the dierence increases 42 microseconds every 20 degrees a sound
source moves (b). Dierences in intensity locate the sound vertically (c ). Sound ine the brain itself. Our research
from above eye level is more intense in the right ear, by the decibel levels shown plan built on work Knudsen and I had
(d ); from below eye level, it is more intense in the left ear. Dierences vary with completed several years earlier. We had
frequency; they were measured for six kilohertz in this case. Combining the two identied cells that are now known
graphs (e ) denes each location in space. When an owl is exposed to a particular to be critical to sound location. Called
pair of dierences, it quickly turns its head in a predictable direction (photograph). space-specic neurons, they react only
S
results were simply fed upward? uch clues led us to seek further While Takahashi was conducting his
Moise and I intended to answer evidence of parallel processing. mapping research, W. E. Sullivan and I
these questions by carrying out experi- Joined by Terry T. Takahashi, explored the ways magnocellular and
ments in which we would deliver sounds now at the University of Oregon, we be- angular nuclei extract timing and inten-
through earphones. But rst we had to gan by examining the functioning of sity information from signals arriving
be certain that signals able to excite the lowest way stations in the brain from the auditory nerve. To understand
particular space-specic neurons truly the cochlear nuclei. Each cerebral hemi- our discoveries, one needs to be aware
mimicked the interaural time and in- sphere has two: the magnocellular nu- that most sounds in nature are made up
tensity dierences that caused the neu- cleus and the angular nucleus. In owls, of several waves, each having a dierent
rons to re under more natural condi- as in other birds, each ber of the audi- frequency. When the waves reach a re-
tionsnamely, when a sound emanated tory nervethat is, each signal-convey- ceptive surface in the ear, known as the
from a spot in the neurons receptive ing axon projecting from a neuron in basilar membrane, the membrane be-
eld. A series of tests gave us the en- the eardivides into two branches af- gins to vibrate, but not uniformly. Dif-
couragement we needed. In these stud- ter leaving the ear. One branch enters ferent parts of the membrane vibrate
ies, we issued sounds through the ear- the magnocellular nucleus; the other maximally in response to particular fre-
phones and monitored the response of enters the angular nucleus. quencies. In turn, neurons that are con-
individual neurons by again holding a We wondered how the space-specic nected to the maximally vibrating areas
microelectrode on or near the cells. As neurons would behave if we prevented (and thus are tuned to specic fre-
we hoped, we found that cells respond- nerve cells from ring in one of the two quencies) become excited. These neu-
ed to specic combinations of signals. cochlear nuclei. We therefore injected a rons propagate impulses along the au-
Further, the sets of timing and intensity minute amount of a local anesthetic ditory nerve to the brain.
dierences that triggered strong ring into either the magnocellular or angu- We and others nd that the intensity
by space-specic neurons corresponded lar nucleus. The results were dramatic: of a sound wave of a given frequency is
exactly to the combinations that caused the drug in the magnocellular nucleus conveyed to the brain from the ear by
an owl to turn its head toward a spot in altered the response of space-specic the ring rate of auditory neurons tuned
the neurons receptive eld. This con- neurons to interaural time dierences to that frequency. This much makes
W
c NONCOINCIDENCE e then proceeded to explore
WEAK FIRING
higher regions, pursuing how
the brain handles timing data
in particular. Other studies, which will
be discussed, addressed intensity. We
learned that when phase-locked im-
pulses induced by sound waves of a
single frequency (a pure tone) leave the
d PHASE AMBIGUITY magnocellular nucleus on each side of
FIRST IMPULSE
the brain, they travel to a second way
station: the laminar nucleus. Impulses
FIRST IMPULSE from each ear are transmitted to the
nucleus on both the opposite and the
same side of the head. The laminar nu-
SOUND WAVE OF A SINGLE FREQUENCY causes neurons sensitive to it to re trains
cleus is, therefore, the rst place where
of impulses at a particular phase angle (a). Coincidence detectors in the owls brain the information from both ears comes
re most strongly when impulses generated at the same phase angle reach the de- together in one place.
tectors simultaneously ( far right in b). Detectors can also re, but more weakly, The general problem of how the
when impulse trains reaching them are slightly asynchronous (c). In what is called brain combines timing data has been a
phase ambiguity, peak ring can occur if a sound to one ear is delayed or advanced subject of speculation for decades. The
by a full cycle from another delivery time that yields coincidence (d ). late Lloyd A. Jeress put forth a rea-
M
Maryland, and I demonstrated in the that situation, and at every 360-degree y colleagues and I understand
barn owl that nerve bers from magno- dierence, coincidence detectors will re- less about the operation of the
cellular neurons serve as delay lines and peatedly be hit by a series of synchron- intensity pathway that converg-
neurons of the laminar nucleus serve as ous impulses and will re maximally. es with the time pathway in the shell.
coincidence detectors. Thus, the same cell can react to more But we have made good progress. Unlike
than one time dierence. the magnocellular nucleus, which pro-
B
ut the owls detection circuit, like Fortunately for the owl, some mech- jects up only one stage, to the laminar
those of mammals that have been anism resolves such phase ambigui- nucleus, the intensity-detecting angular
examined, diers somewhat from ty at higher stages, thereby prevent- nucleus projects directly to many high-
the Jeress model. Neurons of the lam- ing confusion. How this resolution is er stations (except the external nucle-
inar nucleus respond most strongly to achieved remains obscure. Another mys- us). Among them is the posterior later-
coincidence brought about by partic- tery engages us as well: the owl can de- al lemniscal nucleus.
ular time dierences. Yet they also re- tect interaural time dierences as short The posterior lemniscal nucleus on
spond, albeit less strongly, to signals as 10 microseconds (10 millionths of a one side of the head receives direct in-
that miss perfect coincidence. The num- second). Yet a single impulse persists put only from the angular nucleus on
ber of impulses declines gradually as considerably longer than that, on the the opposite side. It nonetheless man-
the interaural time dierence increas- order of 1,000 microseconds. We are ages to discern intensity dierences be-
es or decreases from the value that seeking an explanation for this appar- tween the two ears. Indeed, it is the low-
produces coincidencethat is, until the ent paradox. est station in the brain to do so. The
waves reaching one ear are 180 degrees The rest of the pathway for time de- lemniscal area can detect such dier-
(a full half cycle) out of phase from the tection is more straightforward. After ences because its twin (in the opposite
position that would bring about coin- a coincidence detector in the laminar cerebral hemisphere) sends it informa-
cidence. At that point, ring virtually nucleus on one side of the brain de- tion from the other angular nucleus. In
ceases. (The neurons also respond, at termines the interaural time dierence essence, neurons of the lemniscal nu-
an intermediate level, to signals deliv- produced by a sound of a given frequen- cleus on one side of the head receive ex-
ered to just one ear.) cy, it simply passes the result upward citatory signals from the ear on the op-
In a way, then, coincidence detec- to higher stations, including to the core posite side, and they receive inhibitory
tors, by virtue of the delay lines feeding region of the midbrain auditory area signals from the ear on the same side.
them, can be said to be maximally sen- on the opposite side of the head. Con- The balance between excitatory and in-
hibitory signals determines the rate at In the shell, most neurons respond veal nothing on their own re impulses
which the lemniscal neurons re. strongly to both interaural intensity and collectively in a particular pattern.
We have observed, too, that neurons interaural timing dierences generated
T
of the posterior lemniscal nucleus vary by sounds within a narrow range of fre- ogether our neurological explora-
systematically in the intensity dierenc- quencies. This station does not provide tions have elucidated much of the
es that cause them to re most strong- the owl with sufcient information to algorithm, or step-by-step proto-
ly. Georey A. Manley and Christine ensure accurate sound location, howev- col, by which the owl brain achieves
Kppl of the Technical University of er, because phase ambiguity persists. binaural fusion. Presumably, we humans
Munich showed in my laboratory that The ambiguity disappears only at the follow essentially the same algorithm
neurons at the bottom of the left nucle- level of the external nucleus, home of (although some of the processing sta-
us respond maximally when sound is the space-specic neurons. These neu- tions might dier). Recall, for example,
much louder in the left ear and that rons are broadly tuned to frequency, re- that several lines of evidence suggest
those at the top of the nucleus re ceiving timing and intensity data from mammals rely on delay lines and coin-
most strongly when sound is louder in many frequency channels. This converg- cidence detection in locating sounds.
the right ear. Similarly, neurons at the ence somehow supplies the input need- We can extrapolate even further. The
bottom of the right posterior nucleus ed for the brain to select the correct co- only other neural algorithm for a sen-
respond most strongly when sound is ordinates of a sound source. The selec- sory task that has been deciphered in
much louder in the right ear, and those tivity of space-specic neurons, then, equal detail is one followed by electrici-
at the top of the nucleus prefer louder results from the parallel processing of ty-emitting sh of the genus Eigenman-
sound in the left ear. This arrangement time and intensity data and from the nia. Walter F. Heiligenberg of the Uni-
clearly enables space-specic neurons combination of the results in the shell versity of California at San Diego and
to determine that a noise is coming and in the external nucleus itself. his associates have worked out the rules
from above or below eye level. The pro- We have not yet resolved the number enabling members of this species to de-
cess by which space-specic neurons of space-specic neurons that must re termine whether their electric waves are
convert signals from the posterior lem- in order for an owl to turn its head to- of higher or lower frequency than those
niscal nucleus into vertical coordinates ward a sound source. Nevertheless, we of other Eigenmannia in the immediate
remains to be established, however. know that individual neurons can carry vicinity. (In response, a sh might alter
The next higher station is the later- the needed spatial data. This fact belies the frequency of the wave it emits.) Ei-
al shell of the midbrain auditory area; the view of some researchers that single genmannia rely on parallel pathways to
neurons from the posterior lemniscal neurons cannot represent such complex process separate sensory information.
nucleus on each side of the brain send information and that perceptions arise Also, relevant information is processed
signals to the shell in both hemispheres. only when whole groups of cells that re- in steps; the parallel pathways converge
CORE
ANTERIOR LATERAL POSTERIOR LATERAL Detection and relaying of interaural intensity differences
LEMNISCAL NUCLEUS LEMNISCAL NUCLEUS
TIME INTENSITY
at a high station; and neurons at the top to principles drawn from electronics, nonspeech sounds. By the same token,
of the hierarchy respond selectively to physics and chemistry. The economy of we can ask whether the owl separately
precise combinations of cues. The sh the biological circuit suggests that nat- processes signals for sound location
algorithm is thus remarkably similar to ural principles may help engineers build and other acoustic information. Some
that of the barn owl, even though the analog chips that consume less energy brain stations that participate in spatial
problems that are solved, the sensory and take up less space than usual. orientation may also take part in other
systems involved, the sites of processing My laboratorys research into the owl sensory activities, such as making owls
in the brain and the species are dier- brain is by no means nished. Beyond selectively attuned to the calls of mates
ent. The similarities suggest that brains lling in some of the gaps in our knowl- and chicks. How does the owl, using one
follow certain general rules for informa- edge of binaural fusion, we hope to be- set of neurons, sort out the algorithms
tion processing that are common to dif- gin addressing other problems. For ex- for dierent sensory tasks? By solving
ferent sensory systems and species. ample, Alvin M. Liberman of Haskins such riddles for the owl, we should be-
Carver A. Mead, here at Caltech, thinks Laboratories in New Haven, Conn., has gin to answer some of the big questions
the owl algorithm may also teach some- proposed that the human brain pro- that relate to more complex brains and,
thing to designers of analog silicon cesses speech sounds separately from perhaps, to all brains.
chips, otherwise known as VLSI ( Very
Large Scale Integrated) circuits. In 1988
he and John Lazzaro, then his graduate
student, constructed an owl chip that FURTHER READING
reproduces the steps through which the
A NEURAL MAP OF AUDITORY SPACE IN THE Auditory Function: Neurobiological Bases of
barn owl measures interaural time dif- OWL. E. I. Knudsen and M. Konishi in Sci- Hearing. Edited by G. M. Edelman, W. E.
ferences. The model, about 73 square ence , Vol. 200, pages 795797; May 19, Gall and W. M. Cowan. John Wiley & Sons,
millimeters in area, contains only 64 1978. 1988.
auditory nerve bers in each ear (many NEURONAL AND BEHAVIORAL SENSITIVITY A CIRCUIT FOR DETECTION OF INTERAURAL
fewer than truly exist) and some 21,000 TO BINAURAL TIME DIFFERENCES IN THE TIME DIFFERENCES IN THE BRAIN STEM OF
delay lines. (It also has 200,000 tran- OWL. Andrew Moiseff and Masakazu Ko- THE BARN OWL. C. E. Carr and M. Konishi
nishi in Journal of Neuroscience, Vol. 1, in Journal of Neuroscience, Vol. 10, No.
sistors, mainly to regulate the delay
No. 1, pages 4048; January 1981. 10, pages 32273246; October 1990.
lines.) Even in its pared-down version, NEUROPHYSIOLOGICAL AND ANATOMICAL THE NEURAL ALGORITHM FOR SOUND LO-
the electronic nervous system takes up SUBSTRATES OF SOUND LOCALIZATION IN CALIZATION IN THE OWL. M. Konishi in
much more space and energy than does THE OWL. M. Konishi, T. T. Takahashi, H. The Harvey Lectures, Series 86, pages 47
the biological system. Historically, engi- Wagner, W. E. Sullivan and C. E. Carr in 64; 1992.
neers have constructed chips according
Edited by Thomas D. Albright, Salk Institute for Biological Studies, La Jolla, CA, and approved April 7, 2011 (received for review February 23, 2011)
Visual saliency based on orientation contrast is a perceptual product external camera xed to owls heads (6). Second, barn owls are
attributed to the functional organization of the mammalian brain. known to make conspicuous peering movements (17) and can also
We examined this visual phenomenon in barn owls by mounting covertly shift attention toward interesting targets (18), meaning
a wireless video microcamera on the owls heads and confronting that their vision is likely to incorporate attention mechanisms.
them with visual scenes that contained one differently oriented tar- Third, much is known about barn owls visual system (19, 20), and
get among similarly oriented distracters. Without being conned by the neural circuits underlying visual perception in this species
any particular task, the owls looked signicantly longer, more often, have been studied in some detail (2127). With this in mind, we
and earlier at the target, thus exhibiting visual search strategies so asked whether barn owls exhibit similar visual search behavior as
far demonstrated in similar conditions only in primates. Given the humans. The data presented here indeed show such similarities.
considerable differences in phylogeny and the structure of visual
pathways between owls and humans, these ndings suggest that Results
orientation saliency has computational optimality in a wide variety In our barn owl experiments, the setup and procedures were
of ecological contexts, and thus constitutes a universal building chosen to resemble the classical visual search studies performed
block for efcient visual information processing in general. with humans (1, 12), with a specic focus on saliency due to ori-
entation. Two barn owls (subjects HB and WH) were trained
feature search | gaze map | pop-out | visual behavior | avian vision to carry the OwlCam, a head-mounted wireless microcamera
(Fig. 1A). In a typical experimental trial (Fig. 1B), the owl was
and recent orientation-saliency behavior found in archer sh (16). The authors declare no conict of interest.
NEUROSCIENCE
In the present work we studied visual search in free-viewing This article is a PNAS Direct Submission.
barn owls, which, as we argue, may become a model animal with 1
Present address: School of Optometry, University of California, Berkeley, CA 94720.
several important advantages for exploring this visual attentive 2
To whom correspondence should be addressed. E-mail: wolf@bio2.rwth-aachen.de.
behavior in animals. First, barn owls eye movements are either This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
absent or very small, allowing the study of overt attention with an 1073/pnas.1101582108/-/DCSupplemental.
153
Fig. 1. (A) OwlCam (indicated by the white arrow) attached to the head of a barn owl. Only the antenna and the frontal part of the camera unit are visible.
The total weight of the setup, including the battery, is 5.5 g. (B) Fixation spot calibration procedure. Single xation images (Bottom) are binarized into target
and background regions (n = 1). When several binarized frames are accumulated into a single normalized map, a 2D probability density function of target
locations within the camera frame begins to form (n = 10). A marked xation spot emerges after several thousand xation images are processed (n = 10,000).
(C) Fixation map of one of our owls (WH) after calibration. Absolute occurrences of targets within each pixel (in camera frame coordinates) are color-coded
(compare Inset). In this example, 10,662 xations were accumulated to yield a peak probability at pixel 322/339 of the image (horizontal/vertical coordinates).
The 1D probability functions along both axes are given as well (bright lines).
already xated earlier. An example of a typical xation-saccade iv) Back xations, directed at locations where no bars were
behavior is shown in Movie S1. In a total of 97 experimental visible within the entire camera eld of view. Such xations
trials, 120 min (217,309 frames) of OwlCam video material was were directed at the walls, ceiling, or door of the ex-
recorded and analyzed from the two owls. To present data cor- periment room.
responding to both owls in a simple and clear fashion, in what For simplicity of presentation, we also designate all nonback
follows we present all results using the notation HB/WH. xations (i.e., the sum of target, control, and frontal xations) as
In 28/69 experimental trials, a total of 45/75 min (82,090/135,219 scene xations.
frames) of video material was recorded. The average length of Out of all 985/1,236 xations recorded, 347/264 (35%/21%)
each trial was 97.7/65.3 s (2,931.8/1,959.7 frames). The experi- were back xations. Such xations occurred due to the owls
mental video material was separated into segments of image mo- overall state of alertness (or lack thereof) and its natural visual
tion and nonmotion (17), corresponding to the phases of head scanning behavior. Naturally, these xations did not show useful
movement (sacchades) and nonmovement (xations). Note that visual content that could be used to examine xations on target or
by such a denition, an equal number of saccades and xations distracter items, and thus they were excluded from further analysis.
must occur, because each saccade is bracketed by two xations and The remaining scene xations were used for our main analy-
vice versa. A total number of 985/1,236 xations were analyzed, ses, as described next. Out of all scene xations, 7%/8% were
which lasted a total of 71,673/120,567 frames and averaged 72.76/ outside of the region covered by the 25 bar items.
97.55 frames per xation. Thus, the owls spent 87%/92% of the For stimulus presentation, we divided the scene into two
total recording time on xations, with an average xation time of compartments: the total region of 5 5 items and a central
subarray of 3 3 items. Targets were placed only within the
2.43/3.25 s. The average duration of a saccade was 0.35/0.28 s.
Because of the static nature of xation segments, only one central 3 3 subarray of items, to avoid possible margin effects
due to the fact that items outside the central subarray did not
frame was used to represent the content of each xation. For our
have neighbors on all sides. Interestingly, increased saliency for
analysis, we extracted the middle frame of each xation, on
the bordering items was not observed. Moreover, both owls had a
which the xation spot of 25 pixels in diameter was marked for tendency to focus their overt attention to the central 3 3 sub-
further analysis. First, xations were classied into four classes array, where indeed 76%/62% of their scene xations were di-
based on their visual content (Fig. 2): rected. This bias toward the central subarray was conrmed in a
i) Target xations, in which the target item appeared within number of control trials in which no target bar was present and all
or immediately at the border of the xation spot mark bars were oriented similarly, where 79%/60% of all scene xations
(with up to 1 pixel tolerance). were directed to the central subarray. Thus, the expectations for
ii) Control xations, in which the control item appeared within randomly hitting one of the central items would be 0.76/9 = 0.08
or immediately at the border of the xation spot mark (1 and 0.62/9 = 0.07. With this in mind, the proportion of target and
pixel tolerance). The position of the control item was de- control xations out of the total number of scene xations was
ned by mirroring the target position about the center of the calculated and compared for each experimental trial separately.
3 3 array, and thus was trial-dependent and changed its The mean proportion of target xations was 0.21/0.16, whereas
position according to the target position in each trial. that of control xations was 0.07/0.08 (Fig. 3A). More quantita-
iii) Frontal xations, directed toward the stimulus scene while tively, the mean proportion of xations directed to the target was
being neither target nor control xations. Frontal xations more than twice as high as those directed to the control bar.
include cases in which bars other than the target and the To demonstrate the difference in the number of xations di-
control items were looked at, along with cases in which the rected at the target and the control, we also counted the number
gaze was directed toward the stimulus but the animal x- of xations for each category in each trial. For example, in the
ated at no specic bar. In general, frontal xations were trial shown in Fig. 2B, the owl looked at the target three times,
the most frequent of all xations. whereas the xation spot appeared at the control item two times.
154
Fig. 2. (A) Classes of xations. The xation classes were determined by the
content of the frame in the xation spot (marked by the yellow circle) and
away from it. In target xation, the owl was looking toward the stimulus
array, and the target item (in this case, rotated by 45 compared with all
other items) appeared within or immediately at the border of the xation
spot. In control xation, the owl was looking toward the stimulus array, and
the control item appeared within (or immediately at the border of) the Fig. 3. Target advantage over control in terms of number and total xation
xation spot. The position of the control item was dened by mirroring time. (A) Mean proportion of xations on target (green, lled) and control
the target position about the center of the 3 3 array. In frontal xation, items (white, open) out of all frontal xations for both owl subjects (HB and
the owl was looking toward the stimulus array, but neither the target WH). The difference between target and control conditions was highly sig-
nor the control items appeared within the xation spot. This was the most nicant in both cases (P < 0.0001, Wilcoxon matched-pair signed-rank test).
frequent type of xation. In back xation, none of the foregoing held. In the Error bars are SEM. (B) Normalized cumulative occurrences of xations di-
demonstration case, the owl simply looked at the door of the experiment rected to the target (green line) and to the control (dashed line) plotted
room. (B) An exemplary stimulus scene as reconstructed from many xation against the number of xations. In both owls, the right shift of the target
frames (Materials and Methods). The walls of the experiment room are graphs generally indicates more xations. (C) Mean proportion of time
visible in the far end and along the sides. The reconstructed scan path of the spent on xating the target and control items out of all frontal xations/
owl during this experimental trial is denoted by circles (loci of xations) differences between target and control conditions (color-coded as in A),
connected by straight lines (saccades). Note how the owl repeatedly shifted again highly signicant for both owls (P < 0.002). (D) Normalized cumulative
its gaze between stimulus objects and often returned to specic locations/ occurrences of frames directed to the target and to the control plotted
items. Regions in which target and control xations were registered are against xation time given in the frames for both owls. A right shift denotes
highlighted. Back xations are not shown. longer xations.
A Wilcoxon matched-pair signed-rank test including all trials many as 16/10 times per trial. In summary, our analyses clearly
revealed a highly signicant difference in the target and control show that the target exhibited increased saliency for both owls.
xations per trial (P < 0.001 for both owls). To demonstrate this We aimed to conrm the signicant difference of target se-
perceptual advantage of the target over the control in another lection for oriented objects in the time domain by computing the
relative time spent on each of the frontal, target, and control
way, we also plotted the data as cumulative probability dis-
xations out of all scene xations. The duration of each frontal,
tributions (Fig. 3B). These curves demonstrated a rightward shift
target, and control xation was determined directly from the
of the target distribution compared with the control distribution, video recordings, accumulated by category, and divided by the
suggesting that on average, target xations were more numerous total time spent on all three categories. The mean proportion of
NEUROSCIENCE
per trial. For example, Fig. 3B shows that in the control item was time spent on target items was 0.40/0.34 and that spent on control
not xated in 36%/38% of the trials, whereas the target was not items was 0.20/0.22 (Fig. 3C), whereas the mean proportion of
xated in only 7%/13% of the trials. The same graphs also show frontal viewing without xation of either the target or the control
that the control item was never xated more than 6/5 times per bar was 0.40/0.34. Notably, the difference between target xation
trial, whereas the target was xated much more frequently, as time and control xation time was highly signicant for the in-
Harmening et al. PNAS | May 17, 2011 | vol. 108 | no. 20 | 8463
155
dividual trials as well. The average time per trial spent on target rank test). The faster gazing toward the target than toward the
and control items yielded 548/254.17 frames on target items and control also becomes obvious in the cumulative probability plots
148.25/137.88 frames on control items, a highly signicant dif- for both owls shown in Fig. 4B. The curves reecting the number
ference (P < 0.01/< 0.001, Wilcoxon matched-pair signed-rank of saccades to the target are shifted to the left compared with the
test). The difference in viewing time between target and control control curves. In other words, the target is reached with the rst
is also obvious in the cumulative probability plots, assembled in saccade in 25%/19% of the trials, compared with 0%/7% for the
a way analogous to those shown in Fig. 3B. The curves reecting control. Likewise, after 10 saccades, the target was reached in
the xations at target are shifted to the right compared with the 86%/84% of the trials, whereas the control was reached in only
control curve (Fig. 3D). In >95% of all trials, the owls looked at 36%/49%. These last values indicate that the control was not
the control for <500/400 frames, compared with 1,000/800 frames reached at all in a considerable number of trials. In summary, our
for the target. Overall, the owls xated on the target bars more data show that the target was reached after a lower number of
often and longer than on the control bars. saccades compared with the control bar.
Whether or not the target bar was more salient than other During the training phase, both articial objects and food
items also may be reected in how fast it drew the owls view. We items were scattered on the oor (Materials and Methods). Be-
counted the number of saccades from the onset of a stimulus cause food items should be the most salient visual objects pre-
(lights switched on) until the target or control item was rst sented in an experimental room, it is interesting to examine how
looked at. For example, in the trial shown in Fig. 2B, the owl rst they drew the owls attention compared with our oriented targets
hit the target with its fth saccade, whereas it hit the control with and distracters. On repeating the foregoing analysis, the mean
its third saccade. When all data were averaged, the mean number number of saccades to food was 2.5/3.17, compared with the
of saccades until the target was looked at was 3.92/3.17, and that median number of saccades of 3/3 (Fig. 4A). No signicant dif-
until the control was looked at was 15.5/5.93 (Fig. 4A). Although ferences were found compared with the mean number of sac-
both differences were statistically signicant (P = 0.0022/0.0129), cades to our differently oriented target (P = 0.73/0.59). Thus, the
the distribution of the number of saccades until rst hit was most salient visual items that we observed at all timesfood
itemswere looked at after the same number of saccades as the
positively skewed for both target and control items; that is, there
target item that was dened by a different orientation.
were generally more observations below the arithmetical average.
Specically, in 67% of all cases, owl HB looked at the target item Discussion
after one (i.e., the rst xation was at the target; n = 7), two (n =
Our study of visual search in barn owls demonstrates that the
6), or three (n = 4) saccades, suggesting that the median numbers free-viewing animals looked longer, more often, and earlier at
of saccades might be more informative than the averages in this differently oriented targets than at a control item, in a manner
case. Indeed, the median number of saccades until the target item resembling visual search in humans. The expression of orientation
was rst looked at was 1.5/2, compared with 8.5/4 for the control saliency in visual search, demonstrated here in a bird species, raises
item. The difference between the target and control conditions intriguing questions and has important implications regarding the
was highly signicant for owl HB and signicant for owl WH (P < neural machinery that might be responsible for the observed be-
0.01; n = 17/P = 0.014; n = 38, Wilcoxon matched-pair signed- havior, the evolutionary relationship between birds and primates,
and the role of orientation-based saliency in efcient visual infor-
mation processing.
The predatory barn owl, with its specialization for hunting in
low-light conditions (30) needs to catch approximately two food
items (mainly mice) each day to survive and more than 20 a day
to feed its offspring. The selective pressure on these birds is es-
pecially high if weather conditions are unfavorable due to rain or
snow. Indeed, in central Europe, 60% of barn owl yearlings do
not survive their rst winter (31). Under such high selective
pressure, it would be to the animals advantage to exploit every
possible cue available to nd its prey. Indeed, barn owls are known
to be effective hunters (32), and thus exploiting even minute visual
cues is likely to be an intrinsic part of their visual behavior.
The evolution of different forms of saliency may be related to
the high selective pressure experienced by this bird. The orien-
tation saliency reported here could help the owl detect prey more
easily and more quickly. Pigeons are able to group bars of similar
orientation and discriminate the resulting gure from bars with a
different orientation (4). Pigeons also can detect odd objects
in a scene and even discriminate letters and faces (5, 7). Thus,
birds seem to have the neural machinery necessary for complex
Fig. 4. Target advantage over control in terms of the number of saccades scene analysis.
before rst hit. (A) Mean number of saccades until target (green, lled) and In primates, orientation-based saliency is facilitated by certain
control (white, open) were rst looked at. Black bars indicate the mean neural circuitries, particularly those creating orientation selec-
number of saccades until food items were looked at in training. The dif- tivity (33, 34). Long-range lateral connections found in the pri-
ferences between target and control conditions were highly signicant for mary visual cortex (35, 36) have been shown to be important as
owl HB (**P < 0.0001, Wilcoxon matched-pair signed-rank test), and signif- well (37). Orientation sensitivity in the barn owls visual Wulst is
icant for owl WH (*P = 0.014, Wilcoxon matched-pair signed-rank test). Error
very similar to that seen in the visual cortex (19, 20). The func-
bars are SEM. (B) Normalized cumulative occurrences of saccades until the
target (green line) and the control (dashed line) were rst looked at plotted
tion of the horizontal long-range connections in mammals may
against the number of saccades. Both owls looked at the target much faster, be accomplished in birds through the interconnectivity of many
thereby causing a left shift in the cumulative plot. Note that in many trials, telencephalic nuclei (see ref. 38 for a review). Moreover, within
the owls never looked at the control item, and thus its curve does not the visual Wulst, organizational complexity increases as with
converge to the 1.0 asymptote. increasing latency of neuronal responses, indicating a hierarchy
156
of processing (39). Furthermore, attentional mechanisms were esthesia at an earlier time. Care and treatment of the owls was carried out in
found in the barn owl in cross-modal experiments that success- accordance with the guidelines for animal experimentation as approved by
fully used spatial attention to modulate sound localization (18). local authorities (Landesprsidium fr Natur, Umwelt und Verbraucherschutz
Nordrhein Westfalen, Recklinghausen, Germany) and in compliance with
Cross-modal competition in barn owls also was found to occur in
the National Institutes of Healths guideline for the use and care of labora-
intermediate and deep layers of the optic tectum, a structure tory animals.
known to be involved in gaze control and attention (21).
The fact that two such distant species as humans and barn owls, Experimental Setup and Procedure. Experiments and training were performed
whose brain structures are substantially different, exhibit similar in a large room (4.2 m long 3.2 m wide 3.2 m high), in which the owls were
visual search characteristics has profound implications. In this allowed to move and y freely. Moderate illumination was provided by
sense, our research is similar to a recent study on archer sh, ceiling-mounted tungsten lights that were switchable from outside. To ach-
which have been shown to exhibit orientation-based saliency ieve sound attenuation, the walls, ceiling, and oor were covered with planar
similar to humans (16). Unlike Mokeichev et al. (16), however, and pyramidal foam. A wooden perch placed 1.75 m above the oor close to
one short wall of the room served as a resting post just before and in be-
who explored orientation-based saliency using a rapid forced
tween experiments. A retractable curtain made from thick black cardboard
choice procedure, our experiment was based on free-viewing vi- was placed in front of the perching position, such that the animals view to
sual search, reminiscent of the conditions under which this be- the oor could be blocked until the experiment was started. The owls were
havior is tested in humans. In both cases, bottom-up mechanisms trained to y toward food items presented on the oor and to return to the
are likely to play the main role in the observed behavior (although perch after a successful strike, with the captured prey as a reward. During
the effects of top-down inuence, and of some implicit unspecied training, ights normally occurred after the experimenter left the room.
task, cannot be excluded), and in both cases the behavioral simi- Training trials also were interleaved irregularly between experimental trials
larities in the reported ndings suggest that visual processes, such at a ratio of about 1:5, to ensure high motivation and active viewing be-
as orientation-based visual search, may not necessarily require the havior of the owls in experiments, where no specic task was given.
elaborate cortical structures typically seen in humans. Unveiling Between experimental trials, the following procedure was performed. The
curtain was moved into place to block the owls view of the oor. The ex-
the neural mechanisms that facilitate these processes in animals
perimenter then entered the room and placed 24 distracter items on the
like the barn owl may provide important insight into saliency pro- oor to cover the virtual intersections of a sparsely arranged and randomly
cessing in other organisms as well. jittered 5 5 orthogonal grid (Fig. 2). The visual items were identical rectan-
In humans, classical visual search experiments are also used to gular bar-like shapes (150 50 mm) cut from thick yellow cardboard. One
discriminate between preattentive, pop-out, parallel processes additional item, dened as the target item, was differently oriented and
and serial attentive processes by measuring how target detection slanted by 45 relative to the dominant orientation of the distracters. The
time varies with the number of distracters. Pop-out also has been target item was placed quasi-randomly and counterbalanced in one out of
demonstrated in pigeons (40). Owing to the slower response time nine possible positions at the area of a concentrically arranged 3 3 grid. In
of the owls in our free-viewing paradigm, the difculty in spec- this way, the target item never appeared at the immediate edge of the whole
stimulus array, and possible margin confounds were avoided. The experi-
ifying a specic task, and the indirect way in which the subjects
menter left the room, lights were switched off, and the curtain was retracted
responses must be measured, exploring true pop-out in barn owls to allow a free view onto the stimulus array. The beginning of a trial was
is more challenging. However, comparing our visual search results dened as the time when the lights were switched back on and the owl started
for the differently oriented targets and the food items reveals that to visually inspect the room. The owl was allowed to look around freely for
the both were looked at after approximately the same number of a maximum of 3 min, after which the trial ended. Usually, the owls would y
head saccades. This comparable performance to the most de- toward one of the visual items after a shorter period of inspection and, upon
sirable target may indicate a perceptual popping-out of the dif- entrance of the experimenter, retreat to the perching position. Approximately
ferently oriented target against the distracter array, which draws 515 consecutive trials were performed with one owl per day.
the animals attention equally effectively and may serve as a rst
indication of the existence of the pop-out effect in barn owls. To OwlCam and Video Analysis. During all experimental and training sessions, the
owls carried a head-mounted lightweight wireless camera device, the Owl-
conrm this hypothesis and the equivalence of visual behavior
Cam (SI Materials and Methods). The OwlCam consisted of a miniature
between barn owls and humans in general will require an exten- complementary metal-oxide semiconductor active-pixel sensor and optics
sive and more challenging examination of visual search in barn unit, a 900-MHz video broadcasting unit, a rechargeable lithium-polymer
owls. This will be done by confronting the animals with search battery, and a custom-built attachment unit (Fig. S1). While maintaining
stimuli presented on a monitor and by recording from brain high rigidity at a total weight of 5.5 g, the OwlCam delivered a black-and-
structures putatively involved in visual search (21). Such experi- white video signal at 30 frames per second with an effective vertical reso-
ments may shed light on whether theories hypothesized for human lution of 380 scan lines. The video signal was digitalized online and stored
visual search (10, 11, 13) can model this visual process in the barn in a 640 480 pixel video format for further processing.
owl as well, or whether nature has found a different solution. Using a custom-written algorithm, the raw video material was later di-
Regardless of how this question is resolved, however, the fact that vided into frame segments of image motion and nonmotion (6). Segments of
consecutive frames in which no image motion occurred were dened as
species as distant as humans and barn owls exhibit striking simi-
xation intervals, and the middle frame of each interval was extracted and
larities in a fundamental visual behavior like orientation-based used as the xation frame representing the whole interval. All subsequent
visual search suggests that orientation saliency has computational processing steps were based on these xation frames.
optimality in a wide variety of contexts and provides a universal Each OwlCam was calibrated with respect to the relative geometric ar-
building block for efcient visual information processing. rangement of camera eld of view and the owls gaze, to localize the owls
functional xation point in camera frame coordinates (Results). In several
Materials and Methods calibration trials, a xation map was constructed for each owl and OwlCam
Animals. The experimental animals were two adult American barn owls (T. pair separately. Note that xation maps are valid only for a specic owl
alba pratincola; subjects WH and HB) that were taken from the breeding OwlCam pair, because of idiosyncratic differences of the owls head post
stock of the Department of Zoology at Rheinisch-Westfaelische Technische position, camera layout, and prealignment procedure. Based on this map,
Hochschule Aachen. The birds were hand-raised and tame. The wingspan of a single xation spot relative to the camera frame coordinates was revealed,
barn owls is 1.1 m (41). During the phase of experimentation, the owls as described in detail elsewhere (6). In brief, during calibration, interesting
NEUROSCIENCE
body weight was maintained at 90% of their free feeding weight (415 g bright targets scattered on a dark oor were presented (Fig. 1B). The owl
and 470 g). Water was given ad libitum, and food (dead chicken) was given typically scanned the environment by xating one target and then making
only in the experimental room or as a reward directly after an experiment. a saccade to another target, and so on. As more individual xation frames
Training and experiments were performed on ve or six days per week. For were overlaid and averaged, a distinct circular-shaped xation spot emerged
each owl, a small aluminum headpost, to which the OwlCam was later at- for each owl (Fig. 1C). The resultant xation map reects the probability to
tached (see below), had been xed to the skull on the forehead under an- encounter a bright target in camera frame coordinates. The xation spot
Harmening et al. PNAS | May 17, 2011 | vol. 108 | no. 20 | 8465
157
itself indicates the image coordinates at which the owl would observe ego-centered and perspective-limited representations of the owls view at
a bright target most often. Such a bright spot did indeed occur in the xation a given time. To study the owls viewing behavior as a consequence of the
map of each owl, at camera coordinates (335,322) for owl HB and (322,339) global visual stimulus, the complete visual scene had to be taken into ac-
for owl WH. The diameter of the xation spot was calculated as the mean count. However, to a good approximation, the xation frames could be
width of the probability function at half height and was 26.71 pixels for owl considered limited-view images of the same exterior setting observed from
HB and 22.41 pixels for WH (corresponding to 2.5 and 2.1 of visual angle, a specic vantage point at different viewing angles, resulting from the owls
respectively). In quantitative terms, calibration targets appeared within the head movements while keeping its body relatively xed. Thus, a full-scene
xation spot in 94% (n = 9,804) of all xations for owl HB and in 96% (n = reconstruction was achieved by spatial transformation and alignment of the
10,662) of all xations for owl WH. Once determined, the xation spot of individual xation frames to build a panoramic view of the entire scene as
each owlOwlCam pair was used in the analysis of all video data. Note that would be observed by a wide-angle observer at the vantage point (44). By
the size of the xation spot does not necessarily represent the actual size of mapping the coordinates of the spatially transformed loci of xations in
the animals retinal area of preferred xation, because it is linked to the size each single xation frame to the corresponding coordinates in the full-scene
of the calibration targets used. However, the calibration targets were set to panoramic image, global scan paths were created and analyzed. The owls
have similar size as the bar objects used in our main experiments. After cal- viewing behavior was then studied with respect to three main criteria: (i)
ibration, the xation spot could be marked in each recorded xation frame relative and absolute gaze time spent at a specic location, (ii) relative and
of the main experimental trials to serve as an estimate of where the owl was absolute number of xations directed onto such locations, and (iii) number
looking relative to the camera coordinates. This conclusion is possible with of head saccades performed until the locations were rst looked at.
the barn owl, which virtually lacks eye movements (42, 43). Thus, the view of
a properly aligned and xed head camera is in register with the animals gaze ACKNOWLEDGMENTS. O.B.-S. thanks the generous support of the Zlotowski
at all times. center, the Frankel Fund, and the Paul Ivanier Robotics Center at Ben-Gurion
Given the xed relationship between the OwlCam and the owls gaze, the University. This study was funded by Deutsche Forschungsgemeinschaft
individual xation frames collected during our experimental trials were Grant WA606/17-1.
1. Yarbus AL (1967) Eye Movements and Vision (Plenum, New York). 23. Zahar Y, Reches A, Gutfreund Y (2009) Multisensory enhancement in the optic tectum
2. Einhuser W, Kruse W, Hoffmann K-P, Knig P (2006) Differences of monkey and of the barn owl: Spike count and spike timing. J Neurophysiol 101:23802394.
human overt attention under natural conditions. Vision Res 46:11941209. 24. Harmening WM, Gbbels K, Wagner H (2007) Vernier acuity in barn owls. Vision Res
3. Mazer JA, Gallant JL (2003) Goal-related activity in V4 during free viewing visual 47:10201026.
search: Evidence for a ventral stream visual salience map. Neuron 40:12411250. 25. Harmening WM, Nikolay P, Orlowski J, Wagner H (2009) Spatial contrast sensitivity
4. Cook RG, Cavoto KK, Cavoto BR (1996) Mechanisms of multidimensional grouping, and grating acuity of barn owls. J Vis 9:112.
fusion, and search in avian texture discrimination. Anim Learn Behav 24:150167. 26. Nieder A, Wagner H (1999) Perception and neuronal coding of subjective contours in
5. Blough DS (1977) Visual search in the pigeon: Hunt and peck method. Science 196: the owl. Nat Neurosci 2:660663.
10131014. 27. Wagner H, Frost B (1993) Disparity-sensitive cells in the owl have a characteristic
6. Ohayon S, Harmening WM, Wagner H, Rivlin E (2008) Through a barn owls eyes: disparity. Nature 364:796798.
Interactions between scene content and visual attention. Biol Cybern 98:115132. 28. Wathey JC, Pettigrew JD (1989) Quantitative analysis of the retinal ganglion cell layer
7. Dittrich L, Rose J, Buschmann J-UF, Bourdonnais M, Gntrkn O (2010) Peck tracking: and optic nerve of the barn owl Tyto alba. Brain Behav Evol 33:279292.
A method for localizing critical features within complex pictures for pigeons. Anim 29. Oehme H (1964) Vergleichende untersuchungen an greifvogelaugen. Z Morphol
Cogn 13:133143. Oekol Tiere 53:618635.
8. Krummenacher J, Mller HJ, Deubel H, Wolfe JM, Humphreys GW (2010) Editorial: 30. van der Willigen RF, Harmening WM, Vossen S, Wagner H (2010) Disparity sensitivity
Visual search and selective attention. Vision Res 50:13011303. in man and owl: Psychophysical evidence for equivalent perception of shape-from-
9. Wolfe JM, Horowitz TS (2004) What attributes guide the deployment of visual
stereo. J Vis 10:111.
attention and how do they do it? Nat Rev Neurosci 5:495501.
31. Mebs T, Scherzinger W (2000) Die Eulen Europas: Biologie, Kennzeichen, Bestnde
10. Wolfe JM (1994) A revised model of visual search. Psychon Bull Rev 1:202238.
(Franckh-Kosmos, Stuttgart).
11. Hochstein S, Ahissar M (2002) View from the top: Hierarchies and reverse hierarchies
32. Ilany A, Eilam D (2007) Wait before running for your life: Defensive tactics of spiny
in the visual system. Neuron 36:791804.
mice (Acomys cahirinus) in evading barn owl (Tyto alba) attack. Behav Ecol Sociobiol
12. Treisman AM, Gelade G (1980) A feature-integration theory of attention. Cognit
62:923933.
Psychol 12:97136.
33. Supr H, Spekreijse H, Lamme VA (2001) Two distinct modes of sensory processing
13. Lee TS, Yang CF, Romero RD, Mumford D (2002) Neural activity in early visual cortex
observed in monkey primary visual cortex (V1). Nat Neurosci 4:304310.
reects behavioral experience and higher-order perceptual saliency. Nat Neurosci 5:
34. Kastner S, Nothdurft HC, Pigarev IN (1997) Neuronal correlates of pop-out in cat
589597.
striate cortex. Vision Res 37:371376.
14. Hegd J, Felleman DJ (2003) How selective are V1 cells for pop-out stimuli? J Neurosci
35. Hubel DH, Wiesel TN (1977) Ferrier Lecture: Functional architecture of macaque
23:99689980.
15. Nothdurft H-C, Pigarev IN, Kastner S (2009) Overt and covert visual search in primates: monkey visual cortex. Proc R Soc Lond B Biol Sci 198:159.
Reaction times and gaze shift strategies. J Integr Neurosci 8:137174. 36. Rockland KS, Lund JS (1982) Widespread periodic intrinsic connections in the tree
16. Mokeichev A, Segev R, Ben-Shahar O (2010) Orientation saliency without visual cortex shrew visual cortex. Science 215:15321534.
and target selection in archer sh. Proc Natl Acad Sci USA 107:1672616731. 37. Ben-Shahar O, Zucker S (2004) Geometrical computations explain projection patterns
17. Ohayon S, van der Willigen RF, Wagner H, Katsman I, Rivlin E (2006) On the barn owls of long-range horizontal connections in visual cortex. Neural Comput 16:445476.
visual pre-attack behavior, I: Structure of head movements and motion patterns. J 38. Jarvis ED, et al.; Avian Brain Nomenclature Consortium (2005) Avian brains and a new
Comp Physiol A Neuroethol Sens Neural Behav Physiol 192:927940. understanding of vertebrate brain evolution. Nat Rev Neurosci 6:151159.
18. Johnen A, Wagner H, Gaese BH (2001) Spatial attention modulates sound localization 39. Nieder A, Wagner H (2001) Hierarchical processing of horizontal disparity information
in barn owls. J Neurophysiol 85:10091012. in the visual forebrain of behaving owls. J Neurosci 21:45144522.
19. Liu GB, Pettigrew JD (2003) Orientation mosaic in barn owls visual Wulst revealed by 40. Blough D (1992) Cognitive Aspects of Stimulus Control, eds Honig WK, Fetterman G
optical imaging: Comparison with cat and monkey striate and extra-striate areas. (Psychology Press, Lawrence Erlbaum Associates, Inc., Hillsdale, NJ), p 456.
Brain Res 961:153158. 41. Bachmann T, et al. (2007) Morphometric characterisation of wing feathers of the barn
20. Pettigrew JD, Konishi M (1976) Neurons selective for orientation and binocular owl Tyto alba pratincola and the pigeon Columba livia. Front Zool 4:23.
disparity in the visual Wulst of the barn owl (Tyto alba). Science 193:675678. 42. Steinbach MJ, Money KE (1973) Eye movements of the owl. Vision Res 13:889891.
21. Mysore SP, Asadollahi A, Knudsen EI (2010) Global inhibition and stimulus 43. du Lac S, Knudsen EI (1990) Neural maps of head movement vector and speed in the
competition in the owl optic tectum. J Neurosci 30:17271738. optic tectum of the barn owl. J Neurophysiol 63:131146.
22. Reches A, Gutfreund Y (2008) Stimulus-specic adaptations in the gaze control system 44. DAngelo P (2010) Hugin. Available at http://hugin.sourceforge.net/. Accessed
of the barn owl. J Neurosci 28:15231533. March 16, 2010.
158
The Neural Basis
of Visually Guided Behavior
by Jorg-Peter Ewert
A nimals
ft
see things and then act on the
basis of what they see. What
innate behavioral functions can be mea
sured in successive experiments for some
and tries, through successive turning
movements, to keep the object fixated in
chain of events connects some key time without being significantly affected the center of its visual field. The degree
stimulus with a specific fixed pattern of by accumulated experience. of orienting activity is measured by
responses? In recent years workers in counting the number of turning re
Ta
several laboratories have sought by many oads respond to small objects, such as sponses per minute.
different means to analyze the nerve piece of white cardboard moved The angular size of the stimulus-the
mechanisms by which animals interpret over a black background, with a series of angle it subtends-influences the orient
sensory. signals and select the most ap prey-catching reactions. First there is ing activity [see illustration on page 36].
propriate response. The most effective orientation toward the prey, then binoc Of a variety of square objects toads pre
way to understand the neural basis of be ular fixation, then snapping, gulping and fer those with an edge length of four to
havior appears to be to apply a broad mouth-cleaning. Two basic processes are eight degrees. (The absolute size of such
spectrum of experimental techniques: to required to produce the overall orienting stimuli is five to 10 millimeters. Experi
combine ethological studies of an ani reaction: the identification of a stimulus ments where the distance between ani
mal's behavior with experiments involv and the location of it in space. The iden mal and stimulus is varied show that it is
ing brain anatomy and brain-cell stimu tification process determines the type of the absolute-not the angular-size that
lation and the recording of individual behavior. I t is dependent on specific fea counts; in prey-catching behavior toads
nerve-cell activity. tures of the stimulus such as its angular display "size constancy.") The toads turn
For the past six years, in my labora size, the orientation of the boundaries be away from objects larger than 30 degrees
tory first at the Technical University of tween light and dark, its angular veloc on a side, exhibiting the avoidance re
Darmstadt and then at the University of ity, its contrast with the background and sponse. More particular information is
Kassel, we have been taking this broad so on. A detection process then localizes obtained by substituting bars of various
approach to learn about two kinds of the stimulus and, together with the result lengths for the square stimuli. As a two
visually controlled behavior in the toad: of the identification process, determines by-two-degree stimulus is elongated
orienting (prey-catching) behavior and the motor response, which can be either along the horizontal axis the orienting
avoidance (escape) behavior. There are to turn toward the stimulus if it is iden activity increases until a saturation level
several good reasons for working with tified as prey or to avoid it if it appears is reached; wormlike objects turn out to
the toad. Amphibians are vertebrates, to be an enemy. In what follows I shall be particularly attractive to toads. In
so that what we learn at their relatively attempt to analyze the neurophysiologi contrast, the response decreases as a
low level of behavioral integration con cal basis of signal identification, localiza small stimulus is extended vertically, or
tributes to our understanding of more tion and the triggering of the associated perpendicularly to the direction of move
complex vertebrate functioning. Toads instinctive actions. ment.
in particular have a limited and easily To begin one must analyze quantita Other experiments indicate that toads
surveyed behavioral repertory. In re tively the key stimuli for orienting and discriminate prey from enemy objects
sponse to specific stimuli one can repeat avoidance behavior. This is done by through analysis of the visual stimulus in
edly elicit predictable reactions, such as changing various characteristics of a vi terms of point or edge configurations,
snapping at prey, fleeing from an enemy, sual stimulus in an ordered way. The toad also taking into consideration the direc
clasping during courtship and making is placed in a cylindrical glass compart tion of movement. A horizontal chain
particular wiping motions after tactile ment where it observes a small square of consisting of several two-by-two-degree
stimulation. (The fickle European frog, black cardboard moving against a white units moving along the same path signi
in contrast, undergoes short-term chang background at a constant angular veloc fies prey. One such unit moving alone
es in motivation and is not suitable for ity, describing a circle around the animal constitutes a prey stimulus just above the
behavioral experiments.) Finally, the at a distance of seven centimeters. The response threshold. When the horizontal
toad is not easily conditioned, so that its toad interprets such a stimulus as prey chain is supplied with a separate vertical
34
159
1974 SCIENTIFIC AMERICAN, INC
d
b e
c f
BEHAVIORAL PATTERNS characteristic of the toad Bulo bulo to the brain. An electric current applied to the optic tectum, a visu.
are illustrated. The actions are commonly elicited in the animal by al center in the brain, elicits a prey.catching sequence: orienting,
the sight of visual objects. These drawings, however, are hased on or turning (a), snapping (b) and mouth.cleaning (c). Electrical
photographs of toads whose brains were being stimulated electri stimulation, instead, of a site in the left or right thalamus brings a
cally as part of the author's investigation of the neural bases of vi "plantingdown" defensive posture (d, e) and stimulation of anoth
sually guided behavior. The electrode on the toad's bead penetrates er part of the thalamus brings a crouching avoidance response (f).
35
160
1974 SCIENTIFIC AMERICAN, INC
extension (making it in effect an L
shaped structure moving on its long
side), it loses efficiency as a prey-catching
stimulus. The inhibitory effect of the ver
tical extension depends on its distance
from the horizontal element. If a second
vertical extension is introduced, in effect
making the stimulus a shallow U-shaped
structure, the total configuration signifies
"enemy." The ethological interpretation
is that it symbolizes a "swarm," and in
the toad's brain inhibitory interactions
first restrain prey-catching behavior and
then induce escape behavior.
For constant form and angular veloc
ity the behavioral activity generally in
1'\
(TURNS TOWARD) ./
v background are normally more attractive
as prey than black objects on white; the
latter, on the other hand, are more ef
I
Il.
<Jl
W sponse. The common critical feature for
<Jl
20 key stimuli representing both prey and
Il.
enemy is movement, and the two kinds
VERBTAIRCAL HEIGHT (DEGRE S) HEIGHT (DEGRE S)
ill
a:
Z
a:
::J
0
1
l"-t'-
2 4 8 16 32 2 4 8 16 32
of stimulus are differentiated primarily
on the basis of their form: extension of
the object in the horizontal direction of
the movement generally means prey,
40 whereas extension perpendicular to the
V
f-- +-- direction of the movement signifies "not
V VV
t--""'"
prey" or "enemy."
20
B
II
<{
LENGTH (DEGRE S) LENGTH (DEGRE S)
1 2 4 8 16 32 2 4 8 16 32 in the fascinating research of Jerome Y.
Lettvin and his colleagues at the Massa
chusetts Institute of Technology, and
was later investigated quantitatively by
a:
O.-J. Grusser and his co-workers at the
!
d Free University of Berlin. To ask the
2
DISTANCE (DEGRE S)
4 8
d
16 32
question is to open the "black box" of the
toad's brain, or at least to examine the
brain functions that participate in trans
forming input from visual stimuli into
relevant behavioral patterns. At this
point I shall describe neurophysiological
BEHA VIORAL RESPONSES of the toad to objects of various shapes and sizes were quan findings concerning whether it is in the
tified. Small black objects were moved across the visual field at seven centimeters' distance retina of the toad's eye that the key
and the orienting response was determined for normal toads (left) and those whose thala stimuli "prey" and "enemy" are encoded.
mus had been removed (right). Prey-catching responses (turning toward the object) were In the toad retina there are three types
elicited most effectively in normal toads by squares with sides subtending four to eight de
of ganglion cells that send their fibers by
grees; the toads turned away from larger squares. Vertical bars were ineffective as prey ob
way of the optic nerve to the structure
jects-and increasingly ineffective with increasing height. Horizontal (wormlike) bars were
called the optic tectum in the midbrain.
increasingly effective as prey ohjects with increasing length, up to a limit. Double bars (a
One can record the action potentials, or
horizontal bar plus a vertical extension) were less attractive, the effect varying with distance
between bars; the ratio of their effect to that of a single bar is shown (bottom). In toads nerve signals, from the ends of these fi
lacking the thalamus the orienting response becomes "disinhibited." The animal tends to bers by introducing a microelectrode into
orient toward a target without discrimination, even if the target normally signals "danger." the tectum. John E. Dowling, then at
36
161
1974 SCIENTIFIC AMERICAN, INC
a <E--RECEPTIVE FIELD----;::;' c 40 r---,----.---.---.,---.
r'vL:RINE GA"NGLION-CEL
SQUARE o 0 L-
Z
8w
1
l.-__l.-__-L-__-'-_-'-_-'
__
\!!CAENLGLION
40r---.---,----r-----.
I-@
II-
b en
w
R E C E P T I V E FIELDS ?\
::::l
II
:;
w
O
20f----t..o''''--+----+--l---l
o
II
en
CGEALNGALXlON I N H I B I
F ELD (I F)T O R Y w
II:
40
--II
--@
--.,.,...
20
-
r MICRO
o
1 2 4 8 16 32
ELE T DE B
II
I
ADOUBLE@ dt II:
..:
e::.
NEURAL RESPONSES of the toad to the same objects were mea the strength of the inhibitory surrounds are different for each of
sured with recording electrodes. Tbe electrodes recorded impulses three classes of ganglion cells (b). For the square object and the
at the terminals in the optic tectum of fibers from individual gan vertical bar (which the ganglion cell "confuses"), maximum activi
glion cells, the cells in the retina of the eye on which signals from ty is elicited when the size of the object matches the excitatory-field
the receptor cells converge via intermediate cells (a). Each gan size of each type of ganglion cell (c)_ Horizontal length does not
glion cell has an excitatory receptive field surrounded by an in much affect these cells' response. Vertical extension of a horizontal
hibitory receptive field. The diameter of the excitatory fields and bar has less effect on these cells than on behavior (opposite page)_
the Johns Hopkins University School of istics, including in particular the diame tion of each cell type is therefore elicited
Medicine, showed through electron mi ter of their excitatory receptive fields: by objects of different sizes [see illustra
crography that in the frog (or toad) retina about four degrees for the so-called Class tion above J. Extending a small square
each ganglion cell is connected to a num II ganglion cells, about eight degrees for horizontally (making it a "worm") does
ber of receptor cells by bipolar and ama Class III cells and from 12 to 15 degrees not bring about any change in nerve-cell
crine cells. Each ganglion cell is thus fed for Class IV cells. (Class I cells have been activation; this is in sharp contrast to the
information from a particular part of the identified in frogs but not in toads.) previously noted effect of extension on
animal's visual field. Lateral connections With microelectrodes we measure the the behavioral response. The depen
established by horizontal and amacrine rate of ganglion-cell discharge to see how dence of neuronal activation on the size
cells play a role in determining the prop it changes when objects (corresponding of the stimulus is instead primarily a
erties of this receptive field. In toads as to those in the behavioral experiments function of extension perpendicular to
well as frogs the field consists of a cen described above) are moved through the the direction of movement. Indeed, the
tral circular excitatory receptive field im receptive field of the cell. The impulse discharge frequency is almost exactly the
mediately surrounded by an inhibitory frequency increases with the length of same in response to a narrow vertical bar
receptive field. The movement of an ob the side of a square object until the as it is to a square with the same height
ject through the excitatory field elicits a length about equals the diameter of the as the bar. A retinal ganglion cell "con
ganglion-cell discharge, which is inhibit excitatory field; then it decreases as the fuses" the two stimuli-but the toad does
ed if another object is simultaneously object becomes large enough to stimulate not: the square excites behavioral activ
moving through the inhibitory field. The part of the surrounding inhibitory field. ity and the bar inhibits it. When the ob
three ganglion-cell types in the toad (as In accordance with the different sizes of ject size is held constant, however, the
in the frog) differ in several character- the excitatory fields the maximum activa- dependence of the discharge rate on con-
37
162
1974 SCIENTIFIC AMERICAN, INC
trast between stimulus and background background contrast or from larger size? 27 degrees that are activated exclusively
and on angular velocity is the same as it The differentiation can be made only if by moving objects. These neurons prob
is in the behavioral experiments. separate groups of cells receive different ably represent a localization system. This
In summary, it is clear that the first inputs from different optic-nerve fibers. supposition is reinforced by experiments
important operations on the visual input In fact they do. The fibers of the optic in which we stimulate the tectum of free
from a prey stimulus or a threatening one nerve pass from each eye through the ly moving toads with trains of impulses
are performed by the toad retina. For optic chiasm to the opposite side of the delivered by means of an implanted elec
any particular prey or enemy stimulus brain, ending in various parts of the fore trode. Stimulation of a given region of
the behavioral response to velocity and brain and midbrain. Two of these desti the tectum always causes toads to turn
background contrast seems to depend on nations are of particular interest in our toward a particular part of the visual
information processing in the retina. The work. One, to which most optic-nerve field. Presumably the neurons we are
size-dependent excitatory and inhibitory fibers project, is in the surface layers of thus activating have a direct connection
processes, however, which were noted the optic tectum in the midbrain. The with the animal's motor system, since (in
the behavioral experiments and which other is in the thalamus and the pretectal contrast to the natural orienting move
play an essential role in pattern discrimi region of the diencephalon. ments made in response to a prey object)
CHIOPTIASMC MEDIAN
10 @
@@ @
-----'-----0'1...:
OPTIC OFSURFTECTUMACE
TECTUM
SPINAL CORD
BACK AVOIDANCE ORIENTIN(
__ H
I
NEUROPHYSIOLOGICAL EXPERIMENTS yield data on the ulation (c) of various parts of the tectum (letters) causes the toad
functions of different parts of the toad's visual system. Fibers of the to turn to corresponding parts of the visual field. On the other
retinal ganglion cells project primarily to the optic tectum and to hand stimulation (black disks) of the thalamus, which partially un
the thalamus (a). The visual field of each eye is mapped (b), on a derlies the tectum, causes the opposite action: avoidance, or turn
one-to-one basis (numbers), on the dorsal surface of the opposite ing away. As a recording electrode penetrates below the surface of
side of the tectum_ By the same token experimental electrical stim- the tectum it encounters successive populations of cells with differ-
38
163
1974 SCIENTIFIC AMERICAN, INC
location of a large object, since whenever in effect to raise the level of visual alert- They are activated respectively by four
the three types are excited simultaneous ness. distinct stimulus situations: (1) move
ly the object must be at the fixation The optic tectum also comprises a ment of enemy objects extended perpen
point. neuronal system that processes behav dicularly to the direction of motion, ex
In natural situations the behavior of iorally relevant aspects of moving stimuli citatory receptive field of about 46 de
toads can be influenced by sensory mo [see illustrations on next page]. The cells grees; (2) movement of an object toward
dalities other than vision. If, for exam have excitatory receptive fields about 27 the toad, field about 90 degrees; (3) large
ple, a beetle crosses the field of vision, degrees in diameter. Those designated stationary objects, field about 45 de
the toad's orienting reaction can be ei Type I tectal neurons are activated main grees; (4) stimulation of the balance sen
ther accelerated or retarded by simul ly if the stimulus surface of an object sors in the toad's ear by tilting. In gen
taneous vibratory and tactile stimuli. moved through the receptive field is ex eral these thalamic neurons are activated
Such results can be obtained in experi tended in the direction of movement; principally in situations that tend to call
ments if prey models are presented to extension perpendicular to direction of for evasive movements-turning away
gether with acoustic or tactile stimuli. movement does not have the same effect. from an enemy, sidestepping or compen
The area for producing such changes in Other cells, the Type II tectaI neurons, sating for tilting of the body. Brain
behavioral activity seems to be in the differ from Type I neurons in that their stimulation experiments support our feel
subtectal region, where multisensory in discharge rate actually diminishes with ing that the thalamic-pretectal region is
tegration is achieved. In the area below surface extension perpendicular to direc one in which reactions can be assembled
the third ventricle of the midbrain there tion of movement. The response of these that lead to protective movements. Elec
are large-field neurons with fields similar neurons constitutes the key stimulus trical stimulation of various sites in the
to those of the large-field tectal cells. "prey." That is, they can presumably be region elicits the following reactions:
These subtectal neurons receive addi considered the trigger system for the closing of the eyelids, ducking, turning
tional inputs from neurons excited by prey-catching response. away, panicky springing away or tilting
tactile and vibratory stimuli. The "mech of the body.
R'C'''OR5
tum are evoked by small wormlike prey.
e Large objects extended perpendicularly
d
TECTAL
FIELD CELSMALLS L\ '
"' tectal region, both directly through ret
2\!! AMACRINES
inal inputs and indirectly by way of the
\ ..:
optic tectum. These thalamic-pretectal
TECTALL A RGE
z
i=
II
II ; j=:
a::
\ illIlIlIlIlI!l
can also activate avoidance behavior [see
3 illustmtion on page 41].
SM''TECTAL
The existence of the postulated con
'-'''''' cms
nections between the structures in the
"RG,"TECTAL,"," cm5
5
ways. One way is by direct electrical
stimulation. Thalamic neurons that are
sensitive to movement can also be acti
I I I SUBTECTAL LARGE
vated by stimulation of points in the
VIBRATITACTINEURONS
ONLESENSIAND TIVE
the response of Type II neurons in the
7
tectum to moving objects can be inhibit
8 ed by the stimulation of cells in the thala
mus. The other way is by surgical opera
tion: if the optic tectum is removed,
11
orienting movements are lost-and so are
avoidance reactions, which is evidence
for pathways from the tectum to the
thalamus. If the thalamic-pretectal re
ent receptive fields (d). There are small-field cells with fields a little larger than those of
gion is removed without damage to the
ganglion cells and, lower down, three kinds of large-field cells, each with different coverage
(color, horizontal hatching and vertical hatching) _ A drawing based on a stained brain sec tectum, then avoidance behavior is lost
tion indicates the layers at which each of these is found. The final drawing (e) relates the and the orienting response is dramatical
various cell populations and shows another layer of large-field cells that receive inputs from ly freed from inhibition even in the pres
visual cells above them and also from cells that respond to tactile or vibratory stimuli. ence of enemy objects; this may be
39
164
1974 SCIENTIFIC AMERICAN, INC
T Y P E I T E C T A L NEURON THALAMIC NEURON
EJl "'II'I II 'I
I 1I\\\\\\Wlll,
8
-
I
8
l<E- :
I
,.
evidence for the existence of inhibitory moved, the disinhibition extends to the II tectal cells to moving stimuli shows a
pathways from the thalamus to the optic entire visual field on the opposite side; similar "disinhibition" effect after tha
tectum. In toads lacking the thalamic small lesions in the thalamic-pretectal re lamic-pretectal removal [see illustrations
pretectal region every moving stimulus gion affect only local small parts of the on page 36 and below].
elicits the orienting movements; the cau visual field. Quantitative experiments
T suggest
tionary thalamic-pretectal system, which with toads lacking the thalamic-pretectal he findings I have described so far
ordinarily allows orientation toward the region make it clear that these animals the following sequence of
stimulus only in behaviorally appropriate cannot discriminate between stimuli that events: On the basis of retinal ganglion
situations, is missing. If one lateral half are behaviorally relevant and those that cell input, the optic tectum tells the toad
of the thalamic-pretectal region is re- are irrelevant. The response of the Type where in the visual field a stimulus is
- -8
1 -8
TRIGGER UNITS for the entire preycatching response seem to be tectal neurons) by stimuli that in behavioral experiments are irrel
the Type II tectal nenrons. In the normal toad (left) the cells are evant for prey.catching (vertical bar). After removal of the thala
most activated by wormlike objects (horizontal bar). They are less mus, however (right), their response to those irrelevant stimuli is
activated {and the decrease is greater than in the case of Type I greatly increased, suggesting that the thalamic signal is inhibitory.
40
165
1974 SCIENTIFIC AMERICAN, INC
situated, how large it is, how strongly it these observations? One can speculate cells. With the approach of winter and
contrasts with the background and how that for toads, which are active at twi the period of hibernation, toads stop
fast it is moving. The connections from light, biologically important prey stimuli catching prey. What makes them stop?
the tectum to structures in the thalamic that appear in the lower half of the visual One mechanism may be an inversion
pretectal region enable the toad to dis field are paler then their background; of ganglion-cell response characteristics,
cern the significance to its behavior of those in the upper part of the field, how brought about by signals from the brain
the visual signals. The basic filtering ever, are for the most part relatively to the retina, such that the stimulus
process for the prey-enemy differentia dark, or at least just as often dark as background contrast relation is out of
tion can be conceived of as passage pale. Each of these contrast relations phase with the real world, making prey
through a series of "window discrimi could be reflected in the sensitivity char objects less visible. For the toad, in other
nators" [see illustration on next page], acteristics of the Class II retinal ganglion words, identical objects appear to be dif-
each stage of which analyzes a particular
aspect of the object in question. Each
retinal ganglion cell acts as a vertical
window that codes extension perpendic
ular to the direction of movement. The
retinal analysis is repeated and amplified
OPTIC THALAMUS
in the thalamic-pretectal region, where a
TECTUM
neuron pool acts as another vertical win
dow, this one with a certain minimum
response threshold. Extension in a hori
zontal direction is coded primarily by
Type I tectaI cells, which constitute a
ORIENTING AVOIDANCE
horizontal window. Type II tectal cells
perform a summation, with signals arriv
ing from the thalamic-pretectal region
ELSTIECTRODE
MULUS
having an inhibitory effect and those
RECORDING
from the Type I cells having an excita
tory effect. The resultant signal acts as
the trigger stimulus for the orienting
movement. The triggering of avoidance
THALAMUS OPTIC
TECTUM
behavior is probably achieved through
the activation of still another pool of
thalamic-pretectal neurons, the activa
tion being proportional to an additive
function of inputs from two of the win
dow-discriminator pools.
One of the remarkable aspects of this
system is a degree of plasticity, or
changeability. During the summer
months white prey objects moved against
ELSTIECTRODE
MULUS
black backgrounds elicit orienting be
havior much more effectively than do
black objects against white. In fall and
winter the situation is reversed, and at
OPTIC
the same time the overall prey-catching
activity of the toads decreases. Recently
our recording electrodes revealed that
the activation of single Class II ganglion
cells in the retina exhibits this seasonal
TECTUM THALAMUS
I VI1ISIUALI11STI11MULUS
shift in white-black preference. In win
ter neurons with receptive fields in the
d c
lower part of the visual field are more
I :
I II
strongly activated by black objects than
by white ones; in the upper field the
situation is reversed. In summer, how
ever, the neurons whose receptive field
is in the upper half of the visual field are
CONNECTIONS between the optic tectum and the thalamus were indicated by preceding
activated primarily by black stimuli,
experiments: signals from the retina excite botb tectum and thalamus; subsequent impulses
whereas neurons receptive to the lower
from Type I tectal neurons further excite cells in the thalamus, whereas signals from the
half of the visual field become more thalamus inhibit activity in Type II tectal neurons. Two different kinds of motor activity
strongly activated by white stimuli and are thereupon initiated by the two structures (top). Confirmatory evidence was obtained by
remain so until in the fall black stimuli electrical stimulation. Stimulation of the tectum (middle) elicits impulses (a) from cells in
again become dominant. the thalamus that ordinarily respond to visual stimuli (b). Stimulation of the thalamus (bot
What is the biological significance of tom) inhibits (e) impulses normally elicited (d) in Type II neurons by moving objects.
41
166
1974 SCIENTIFIC AMERICAN, INC
ferent when they are seen in winter than ments we believe the sequence of events the rest of the prey-catching sequence is
when they are seen in summer, a re controlling a natural orientation response activated, quite independently of the re
minder that an organism's picture of the is about as follows: A pattern is formed sult, or even of the short-term benefit to
environment is a product of its brain. by a natural stimulus on a portion of the the animal, of such activation. For ex
In contrast to the plasticity of the retina that is outside the fixation region; ample, if an experimental prey object is
systems for the filtering and storage of the retinal locus has a corresponding pro removed at the instant when it is fixated
information and for pattern recognition, jection locus in the optic tectum. If the by the toad, the entire normal prey
brain mechanisms involved in instinctive filtering process described above has catching routine nevertheless proceeds.
actions are quite inflexible and do not identified the object as prey, then the The toad snaps, gulps and wipes its
adapt easily to changes in the stimulus appropriate neuronal system is activated. mouth in spite of the "situational vac
situation. For each instinctive action in A value corresponding to the distance uum." The sequence is similar in its in
the behavioral repertory there is a pre between the prey's locus on the retina eVitability to what happens when the
programmed "printed" neuronal circuit and the fixation point is transferred to triggering region of the tectum is stimu
that coordinates the appropriate motor the toad's motor system. The result is lated with an electrode.
act-even if it becomes inappropriate! If orientation: a turning movement such As for avoidance behavior, the results
such a circuit is triggered (either natural that the retinal representation of the prey of thalamic stimulation indicate that it is
ly or by electrode stimulation), the innate is brought to the fixation point. That trig controlled by a single master program.
reaction proceeds automatically. Prey gers a locus in the optic tectum that cor The response consists in a firm planting
catching behavior is a good example. On responds to the fixation point. As soon as of the extremities on one side of the
the basis of brain-stimulation experi- this triggering reaches a threshold value toad's body and a gathering together of
the limbs on the opposite side. With the
a
STIMULUS
t
b -
toad in this stationary, poised position
the additional behavior patterns for cor
>t<
recting tilting of the body or making the
4m T that
he evidence I have reviewed shows
in a lower vertebrate the neuro
O
T( EYCPTEUIM) O but are intimately connected with one
another. In the course of evolution the
centers for two of these processes, visual
localization and instinctive action, have
apparently remained in about their orig
inal positions. They occupy the same
c d
I areas of the brain, the tectum and the
thalamus, in monkeys and cats as they do
. .
in toads. The organization of these parts
of the brain, to which both neurophysi
ological and ethological methods have
provided investigative access, shows re
markable constancy in all classes of ver
tebrates. That is not the case, however,
.} jV.
for stimulus identification. In toads this
process takes place primarily in the tha
lamic-pretectal region and also in the
O-w+- O-+-
ever, underwent further evolution, cor
responding to the importance of pattern
recognition in the evolution of their be
havior. A new substrate developed for
two associated but highly specialized
processes, filtering and storage of infor
IDENTIFICATION OF AN OBJECT AS PREY OR ENEMY is symbolized as a series of mation: the visual cortex. From the in
operations by "window discriminators" (a). A ganglion cell in the retina codes vertical ex vestigations of Gerald E. Schneider at
tension (perpendicular to direction of movement), in effect responding to as much of a vi
M.I.T. we learn that in this case ontog
sual object as appears in a vertical window; extension beyond the window has an inhibitory
eny reflects phylogeny. In newborn ham
effect. Cells in the thalamus do the same thing. In the tectum Type I cells code horizontal
sters subcortical pathways between the
extension (in the direction of movement). Type II tectal cells sum the excitatory signal
from Type I cells and the inhibitory signal from the thalamus, and the resultant signal trig.
tectum and the thalamus are implicated
gers an orienting movement. At each stage the cell discharge depends on the relation be in pattern discrimination. In adult ani
tween the object and the window that senses its extension either in or perpendicular to the mals, on the other hand, pattern discrim
direction of movement (b, c, d). (The celldischarge patterns shown here are schematic.) ination takes place in the cortex.
42
167
1974 SCIENTIFIC AMERICAN, INC
News: Fo r fi e l d-e ffe c t de v i c e s,
EASTMAN 14080 is now ready to
ship. At 18 I'm spacing, a typical lot
shows a threshold of only 1.5 volts
rms for 60-Hz sine wave (or 2 volts
dc), an operating temperature range
of 0 to 75C, and a dielectric anisot
ropy of +11.9 at 25C. At 5 V rms,
tum-on time at room temperature is
about 100 ms; tum-off time, about
350 ms; contrast ratio, about 50: 1.
For dynamic scat tering de vices,
EASTMAN 14099 is equally ready,
with the same 0 to 75C operating
temperature range. It aligns itself ho
meotropically on cle a n electrodes
without additional magic-an impor
tant advantage. Typical threshold is
4 volts rms, 5 volts dc.
More details from George Grau at
Organic Chemical Sales, Kodak,
Rochester, N.Y. 14650 (716-325-
2000, ext. 57288).
We make them.
Behind the mystery and glamour: Those when water sits in contact with a bar of tation, all the molecules strive for paral
who could reasonably be expected to soap. In a nematic liquid crystal there are lelism to the wall. But ions that have been
phone Mr. Grau are probably interested no tiers, but the molecules stay parallel. incorporated in the mixture migrate, col
in making display devices ranging from Possibilities exist for controlling their liding with them to knock them every
mass-marketed consumer goods to some alignment. And this alignment controls which way. The resulting optical inhomo
of the more ambitious reaches of the en what they do to light passing through. And geneity scatters light. These turbid areas
gineering imaginaHon. that's what the action is mostly about in can be made to look either darker or
Liquid-crystal technology has blazed up industry circles. lighter than the clear areas, depending on
after smoldering quietly for most of the Field effect, one approach used to cre directions of illumination and viewing.
time since 1888. Organic chemists get to ate a visible pattern in a thin layer of ne Appearance of the pattern is more sensi
collaborate with electronic engineers matic material between patterned, trans tive to angle of view than with field effect.
through the medium of the marketplace, parent electrodes, amounts to electrically Power drain is more, because of the tur
if not personally. For now, Kodak has set tuned birefringence. (We hope you didn't bulence to be maintained. But you don't
up its booth on the chemical side of the cut the lecture on birefringence in Physics have to find (or license from somebody)
street. 1.) Contrast between "on" and "off" is a way to align the molecules. You just use
The present generation of electronic attained through various arrangements of EASTMAN 14099.
engineers catch on fast to subjects they light polarizers and quarter-wave retarda Say, if you've read this far you are
didn't necessarily concentrate on in school. tion plates. Where and when the field is off, probably interested enough to ask Dept.
In liquid-crystal work, one deals with the the molecules must line up parallel to each 412-L, Kodak, Rochester, N.Y. 14650 for
different forms and degrees of orderedness other and to the cell walls. This is called Eastman Organic Chemical Bulletin, Vol.
among molecules, ranging between the "homogeneous" alignment. Surface treat 45, No.2 (1973), where four scientists of
randomness of an ordinary isotropic liq ment to make it happen is protected by the Kodak Research Laboratories, who
uid and the periodic architecture when it patents or trade secrecy. write more like scientists than engineers,
freezes to a crystal. Only certain com D y n a m i c s c a t t e r i n g is t h e o t h e r will take you farther into this than you'd
pounds assume this mesomorphic state. approach. Here, i n the "off" state, the want to go for just recreation. It has a
Molecules of such compounds have a gen alignment can be either homogeneous or 76-item bibliography. For our 3,281-item
erally elongated shape. perpendicular to cell walls. The latter is Liquid Crystal Bibliography (Kodak Pub
The most highly ordered kind of liquid "homeotropic" in the lingo of the art. lication 11-193) on microfiche, make that
crystal, called smectic, where the mole Either way, light passes straight through, Dept. 454 and send $25* (plus applicable
cules tier up in layers and the layers can and the layer looks clear. Where field is state and local taxes).
only slide against each other, can be seen applied, then regardless of previous orien- Price subject to change without notice.
43
168
1974 SCIENTIFIC AMERICAN, INC
The Journal of Neuroscience, May 25, 2011 31(21):77537762 7753
Cellular/Molecular
The retina contains ganglion cells (RGCs) that respond selectively to objects moving in particular directions. Individual members of a
group of ON-OFF direction-selective RGCs (ooDSGCs) detect stimuli moving in one of four directions: ventral, dorsal, nasal, or temporal.
Despite this physiological diversity, little is known about subtype-specific differences in structure, molecular identity, and projections. To
seek such differences, we characterized mouse transgenic lines that selectively mark ooDSGCs preferring ventral or nasal motion as well
as a line that marks both ventral- and dorsal-preferring subsets. We then used the lines to identify cell surface molecules, including
Cadherin 6, CollagenXXV1, and Matrix metalloprotease 17, that are selectively expressed by distinct subsets of ooDSGCs. We also
identify a neuropeptide, CART (cocaine- and amphetamine-regulated transcript), that distinguishes all ooDSGCs from other RGCs.
Together, this panel of endogenous and transgenic markers distinguishes the four ooDSGC subsets. Patterns of molecular diversification
occur before eye opening and are therefore experience independent. They may help to explain how the four subsets obtain distinct inputs.
We also demonstrate differences among subsets in their dendritic patterns within the retina and their axonal projections to the brain.
Differences in projections indicate that information about motion in different directions is sent to different destinations.
Figure 1. Transgenic markers of ooDSGCs that prefer different directions. A, Responses of a BD-RGC to a white rectangle moving across the receptive field center in eight different directions at 575 m/s.
Average responses are displayed in a polar plot and surrounding traces show raster plots for seven repeats. Note ON and OFF phases of spiking in rasters. V, Ventral; D, dorsal; N, nasal; T, temporal. B, Preferred
directions of BD-RGCs recorded from 60% eccentricity. Arrow length indicates the extent of direction selectivity, calculated as in Kim et al. (2010). C, Preferred directions of W9-RGCs (black arrows) and
DRD4-RGCs(grayarrows).D,E,RetinawholemountshowingBD-RGCs(red)andDRD4-RGCs(green),intripletransgenicmouse(DRD4-GFPFSTL4-CreERROSA-CAGS-STOP-tdTomato).BlueboxinDshows
region enlarged in E. RGCs express either green or red fluorescent proteins, but not both, demonstrating that DRD4- and BD-RGCs are distinct populations.
label all four subtypes with 90% accuracy. Subtype identity is overlapping subsets of RGCs in the W3, W7, and W9 lines, presumably
specified molecularly before eye opening, suggesting that differ- due to effects of sequences near the site of transgene integration in the
ences among ooDSGCs arise independent of visual experience. genome (Feng et al., 2000).
We also document differences in dendritic arbors and axonal To mark cdh6-expressing cells, CreER was inserted into the initiation
codon of the cdh6 gene by homologous recombination in embryonic stem
projections that distinguish the ooDSGCs selective for ventral
cells. The targeting vector was generated by lambda phage-mediated recom-
motion from those selective for nasal motion. In addition to pro- bineering (Chan et al., 2007). Chimeric mice with the targeted embryonic
viding reagents for mechanistic studies of ooDSGC subtype di- stem cells were generated and mated to obtain germ line transmission. Mice
versification and synaptic specificity, our results provide support were mated to reporter lines as described above for FSTL4-CreER.
for the idea that molecular markers can be used to distinguish Thy1-YFP-H mice were generated and characterized by Feng et al.
closely related neuronal subtypes. (2000). Dopamine receptor D4-GFP (DRD4-GFP) mice were obtained
from Mutant Mouse Regional Resource Center-University of North Car-
Materials and Methods olina (www.mmrrc.org).
Mice. FSTL4-CreER transgenic mice, used to label bistratified dendrite- Electrophysiology. Methods for electrophysiological analysis have been
retinal ganglion cells (BD-RGCs), were described by Kim et al. (2010). described by Kim et al. (2008, 2010). Briefly, dark-adapted retinas were
Briefly, the transgene was generated by insertion of a tamoxifen- isolated in Ringers solution and targeted for cell-attached recording with
responsive cre recombinase (CreER) cDNA at the initiation codon of the patch microelectrodes. Light stimuli were delivered from a computer-
Fstl4 coding sequence in a bacterial artificial chromosome. These animals driven video projector through a custom-made substage lens. Receptive
were crossed to mice that express a fluorescent protein following Cre- field centers were determined with small flashing spots, then direction
mediated excision of stop sequences. The following three reporter lines selectivity was assessed with stimuli moving through the receptive field
were used: Thy1-STOP-YFP mice line 15 (Buffelli et al., 2003), and mice center in eight different directions.
that conditionally express tdTomato (Madisen et al., 2010) (obtained Histology. Methods for histological analysis of retina and brain are
from Jackson Laboratories, catalog #007909) or green fluorescent pro- described by Yamagata et al. (2006) and Kim et al. (2010). Antibodies
tein (GFP) (Kim et al., 2009) (kindly provided by S. Dymecki, Harvard used were as follows: rabbit anti-GFP, goat anti-choline acetyltrans-
Medical School, Boston, MA) from the Rosa26 locus. Tamoxifen (100 ferase, mouse anti-Brn3a (all from Millipore), goat anti-vesicular acetyl-
g, Sigma) was injected intraperitoneally into double transgenics at choline transporter (Promega), rabbit anti-CART (Phoenix), and rabbit
postnatal day (P) 0 1 to activate CreER and thereby initiate expres- anti-MMP17 (Epitomics). Secondary antibodies were from Invitrogen
sion of reporter. or Jackson ImmunoResearch.
The W9 mouse line was generated in parallel with the W3 and W7 lines Cell isolation and expression profiling. To isolate BD-RGCs, retinas
described by Kim et al. (2010). In this transgene, Thy1 regulatory ele- from P6 mice were dissociated using papain (Worthington). The cell
ments drive expression of YFP. YFP was expressed in distinct and non- suspension was incubated with anti-Thy1 antibodies conjugated to mag-
170
Kay et al. Diversity of Direction-Selective RGCs J. Neurosci., May 25, 2011 31(21):77537762 7755
netic beads (Miltenyi Biotec), and passed over a magnetic column, ac-
cording to the manufacturers instructions. This RGC-enriched cell
fraction was then passed through a MoFlo cytometer (Dako) to select
YFP-positive cells, which were sorted directly into RNA-stabilizing lysis
buffer from the PicoPure RNA Isolation Kit (MDS).
To isolate starburst amacrines, we used a Thy1 transgenic line in which
all starbursts and a subset of RGCs are labeled with the Kusabira Orange
fluorescent protein (OFP; line Thy1-OFP3) (J. Livet and J. R. Sanes,
unpublished observation). Purification of these cells used the same pro-
cedure as above, except that the Thy1-negative fraction (cells not re-
tained on the magnetic column) was used for sorting. We confirmed that
all OFP-positive cells recovered in this manner were starbursts by plating
the sorted cells and staining for starburst and RGC markers.
For microarray hybridization, RNA was isolated using the PicoPure
kit. Two rounds of amplification were performed with the MessageAm-
pII system (Ambion/Applied Biosystems), the second resulting in biotin-
labeled samples. These were hybridized to Affymetrix Mouse 430 2.0
arrays according to the manufacturers instructions. We hybridized two
replicates per transgenic line, each representing a sample of 200 cells
collected from different litters. Data were analyzed, and cell-type-specific
genes identified using Resolver (Rosetta) and dChip software (Li and
Wong, 2001).
Results
Transgenic lines marking ooDSGCs that prefer ventral or
nasal motion
We recently generated a transgenic line in which a group of
ooDSGCs, which we call BD-RGCs, are labeled with a fluorescent
protein (Kim et al., 2010). To begin this study, we assessed the
directional preference of BD-RGCs in adult retina (eccentricity of
60% measured from the optic nerve head to the retinal periph-
ery). Each cell was stimulated with bars moving in eight direc-
tions. Over 90% of BD-RGCs responded best to bars moving
from dorsal to ventral on the retina (i.e., to ventral motion) (Fig.
1 A, B).
We also recorded from RGCs in a transgenic line called W9,
which was generated in parallel with the W3 and W7 lines de-
scribed previously (Kim et al., 2010). The ventral retina of the W9
line contained a nearly homogeneous population of labeled bis-
tratified RGCs. These cells were ooDSGCs that responded pref-
erentially to nasal motion (Fig. 1C). This preference was similar
to that reported previously for ooDSGCs labeled in a transgenic
line generated by the GENSAT project and called DRD4-GFP
(Huberman et al., 2009). We obtained this line and confirmed
that DRD4-RGCs prefer nasal motion (Fig. 1C).
These results imply that BD-RGCs are distinct from W9 and
DRD4-RGCs. To test this idea, we generated BDDRD4 double- Figure 2. Relationship between structure and function of ooDSGCs. A, B, Morphology of W9
RGCs does not correlate with their preferred direction. Confocal stack z-projection showing two
transgenic mice in which BD-RGCs were labeled with a red fluo-
W9 cells that were injected with Lucifer yellow following recording. Scale bar, 100 m. C,
rophore and could thus be distinguished from GFP-positive DRD-4-RGC filled with Lucifer yellow following recording. Arrow indicates preferred direction,
DRD4-RGCs. As expected, few fluorescent cells in these retinas which is distinct from the orientation of its dendritic asymmetry. D, Sketch of part of a whole
(1%) were both RFP and GFP positive (Fig. 1 D, E). We also mounted retina (see inset at bottom left) showing dendritic asymmetry of the BD-RGCs. Arrows
expect that W9-RGCs and DRD4-RGCs are overlapping subsets, originate from the somas and point in the direction of dendritic asymmetry. Length of arrow is
but because they are labeled with indistinguishable fluorophores proportional to degree of dendritic asymmetry. Dot, Optic disc. E, Micrograph of two BD-RGCs
(YFP and GFP), we could not test this idea directly. from the retina. Scale bar: (in E) C, E, 100 m. F, Polar plot summarizing the dendritic asym-
Based on these results, we used the BD, W9, and DRD4 lines to metry of BD-RGCs from a retina similar to that shown in D and E. G, Relationship between
compare the structural and functional properties of ooDSGCs dendritic asymmetry and direction selectivity of 22 BD-RGCs. The preferred direction is plotted
selective for ventral and nasal motion. We used both DRD4 and relative to the direction of the dendritic arbor (dot). Black lines indicate dorsal-preferring cells
from the dorsal margin of the retina.
W9 lines in most studies but, unless otherwise noted, present
results from DRD4 here.
varied among cells but in no case was as dramatic as that docu-
Different patterns of asymmetry in cells preferring ventral mented previously for OFF-DSGCs (Kim et al., 2008). Consistent
and nasal motion with previous reports (Oyster et al., 1993; Huberman et al., 2009),
BD-, DRD4-, and W9-RGCs all had bistratified dendritic arbors there was no obvious relationship between the structural asym-
that, when viewed en face, were generally displaced asymmetri- metry and directional preference of W9- or DRD4-RGCs. For
cally around the somata (Fig. 2 AC,E). The degree of asymmetry example, the W9-RGC in Figure 2 A and the DRD4-RGC shown
171
7756 J. Neurosci., May 25, 2011 31(21):77537762 Kay et al. Diversity of Direction-Selective RGCs
Figure 3. Molecular markers for subsets of ooDSGCs. A, B, In situ hybridization for Cdh6 RNA (red) combined with anti-GFP antibody staining (green) reveals expression of Cdh6 in BD- but not
DRD4-RGCs. Yellow arrows in AH indicate double-labeled cells. Red arrows in AH indicate marker-positive cells that do not express GFP. C, D, In situ hybridization for Col25a1 RNA (red) shows
expression in GFP-labeled (green) BD-RGCs but not DRD4-RGCs. E, F, In situ hybridization for Mmp17 RNA (red) shows expression in GFP-labeled (green) DRD4- but not BD-RGCs. G, H, CART is a marker
of both BD- and DRD4-RGCs as shown by immunostaining with anti-CART (red) and anti-GFP (green). Laminae marked in H (INL, inner nuclear layer; IPL, inner plexiform layer) apply to AH, and K.
Scale bars: 10 m. I, J, Quantification of the fraction of RGCs expressing Cdh6, Col25a1, or Mmp17 at P14 (I ) and at P7 (J ) (n 44 for BD-RGCs, 104 for DRD4-RGCs, 1198 for all RGCs). K, MMP17
immunoreactivity in section from P14 retina. Antibody labels the putative starburst layers of the IPL, suggesting immunoreactivity in the dendrites of starburst amacrines and/or ooDSGCs. A subset
of RGC cell bodies (arrows) as well as putative starburst amacrines (arrowheads) are labeled in the GCL and INL. Scale bar, 25 m.
in Figure 2C had dendrites displaced dorsonasally and dorso- parallel, we isolated and profiled several other RGC and amacrine
temporally from their somata, respectively, but preferred na- subtypes (Kay, 2011), including starburst amacrines, which pro-
sal motion. vide the main inhibitory input to ooDSGCs. We filtered the mi-
In striking contrast, most of the asymmetric BD-RGCs had croarray data to identify genes expressed at several fold higher
dendrites that were displaced vertically from the soma. In most of levels in BD-RGCs and/or starburst amacrines compared with
the retina, the predominant direction was ventral, but a small other cell types in the dataset, then used in situ hybridization
number of BD-RGCs near the dorsal pole of the retina had arbors (ISH) to ask they whether they were expressed by BD-, DRD4-, or
directed dorsally (Fig. 2 DF ). Together with the physiological W9-RGCs.
results presented above, this finding suggested a correlation be- Two genes, cadherin 6 (Cdh6 ) and collagen 25a1 (Col25a1),
tween the structure and function of BD-RGCs. To test this rela- were expressed by BD-RGCs but not by DRD4-RGCs (Fig. 3A
tionship stringently, we recorded from 13 BD-RGCs in central D,I ). Cdh6, a type II cadherin, was previously shown to be ex-
retina and 3 BD-RGCs at the dorsal pole, then imaged their den- pressed in mouse retina (Honjo et al., 2000); Col25a1 encodes a
drites following recording. In 15, including all 3 of the dorsal transmembrane collagen that is expressed in the nervous system
cells, the preferred direction of the cell, determined physiologi- and accumulates in amyloid plaques in Alzheimers disease
cally, corresponded to the direction in which the dendritic arbor (Hashimoto et al., 2002). Cdh6 and Col25a1 were also expressed
was displaced from the soma (Fig. 2G); the 16th had a symmet- by starburst amacrines but not by other retinal cells.
rical dendritic arbor. Thus, not only the majority of BD-RGCs We were also interested in identifying markers of ooDSGCs
that prefer ventral motion but also the minority in dorsal retina that prefer nasal motion. The most obvious candidate was Drd4,
that prefer dorsal motion exhibit an association of structural and since regulatory elements from this gene drive expression of GFP
functional asymmetries. in the DRD4 transgenic line. In some cases, however, including
the line that labels BD-RGCs, transgenes are expressed in cells
Distinct molecular signatures of RGCs that prefer vertical and that do not express the endogenous gene, because of omission of
nasal motion critical regulatory elements from the transgene or to influences at
To identify genes selectively expressed in BD-RGCs cells, we the genomic site of integration (Feng et al., 2000; Haverkamp et
isolated them based on their fluorescence and profiled gene ex- al., 2009; Kim et al., 2010). Indeed, consistent with data from the
pression using microarrays (see Experimental procedures). In rat (Klitten et al., 2008), in situ hybridization showed that Drd4
172
Kay et al. Diversity of Direction-Selective RGCs J. Neurosci., May 25, 2011 31(21):77537762 7757
Discussion
In many respects, ooDSGCs appear to be a single major RGC
subtype: they are similar in dendritic morphology (the main cri-
terion by which RGCs are classified), in response properties and
in the biophysical mechanisms that underlie their direction selec-
tivity (Demb, 2007; Zhou and Lee, 2008). Yet in one respect they
are readily divisible: they can be grouped into four distinct sets,
each responsive to one of the four cardinal directions (Oyster and
Barlow, 1967). Thus, ooDSGCs are more properly viewed as
comprising four closely related subtypes that, despite many com-
mon features, are likely to be molecularly distinct and might send
their information to different central targets. However, despite
intensive study of ooDSGCs in the aggregate, differences among
them have not been explored until very recently (Huberman et
al., 2009). Here, we used a set of four transgenic lines to mark
ooDSGCs with preferences for ventral (BD), nasal (DRD4 and
W9), and vertical (dorsal and ventral; Cdh6) motion. By compar-
ing the lines, we identified molecular and structural differences
among ooDSGC subsets and documented differences in their
central targets.
176
Kay et al. Diversity of Direction-Selective RGCs J. Neurosci., May 25, 2011 31(21):77537762 7761
177
7762 J. Neurosci., May 25, 2011 31(21):77537762 Kay et al. Diversity of Direction-Selective RGCs
178
J. exp. Biol. 146, 255-275 (1989) 255
Printed in Great Britain The Company of Biologists Limited 1989
BY WALTER HEILIGENBERG
Scripps Institution of Oceanography, Neurobiology Unit, University of
California at San Diego, La Jolla, CA 92093, USA
Summary
Studies of the electrosensory system of gymnotiform fish have revealed
principles of neuronal coding and processing of information which also character-
ize more advanced systems, such as vision and audition in higher vertebrates.
1. Animals may have different classes of receptors adapted to code different
variables within a given modality, and the separation of their central projections
provides the basis for independent initial processing of these variables by higher-
order neurones.
2. These separate pathways, however, eventually converge at the level of still
higher-order neurones which are adapted to 'recognize' particular spatial and
temporal constellations, or patterns, of the stimulus variables conveyed by these
pathways.
3. As different stimulus patterns may control different forms of behavioural
responses, corresponding neuronal structures can be identified which are adapted
to recognize specific patterns. Neurones at an early level of pattern discrimination
may still show very general response properties, whereas neurones closer to the
ultimate control of a given behaviour show more specific response properties.
These latter are less sensitive to stimulus features which are irrelevant to the
control of the behaviour, and they code relevant features more purely and with
higher acuity than do lower-level neurones.
4. The acuity of stimulus discrimination displayed by some high-order neurones
may rival that observed at the behavioural level. This high sensitivity is achieved
through pooling and integration of information supplied by large populations of
less-sensitive receptors and lower-order neurones.
Introduction
Animals are naturally adapted to detect and to process very specific stimulus
patterns in their environment, and studies of their ecology and ethology help us to
identify such stimulus patterns as well as natural behavioural responses which they
evoke and control. These responses, in turn, provide crucial assays for the
experimental dissection of stimulus patterns and for the identification of their
relevant components. Moreover, experimental manipulation of these components
and their configuration within patterns enable us to pin down the computational
179
256 W. HEILIGENBERG
rules which underlie the evaluation of patterns by the animal's brain. In the course
of this analysis, we will then gain confidence in our conceptual models of
perception to the extent that we are able to predict behavioural consequences that
are due to specific alterations in stimulus patterns.
After behavioural experiments have identified computational rules of percep-
tion, their specific neuronal implementations can be determined by physiological
and anatomical approaches. In view of the overwhelming structural and functional
complexities of brain mechanisms one would prefer to study neuronal mechanisms
in the relatively simple systems found in invertebrates and lower vertebrates, such
as fish and amphibians. The small size of their brains as well as the more modest
number of structural components and interconnections facilitate such an analysis
at the single-cell level.
The electrosensory system of certain orders of fish appears to have a relatively
simple structural and functional organization, but it also displays sophisticated
features, such as complex receptive fields, ordered central representations and
modulations of sensory processing by recurrent descending pathways, which are
known in more 'advanced' systems, such as mammalian vision and hearing. The
basic neuronal principles of sensory information processing appear to be the same
across different modalities and classes of organisms. The greater 'transparency' of
simple systems, however, makes the study of these principles more feasible, and
their successful exploration in simpler systems should ultimately enhance our
understanding of more complex systems.
What makes the electrosensory system particularly appealing are behavioural
responses, such as the jamming avoidance response (JAR), which remain intact in
physiological preparations. Studies at the single-cell level can thus be combined
with simultaneous behavioural assays to test the significance of stimulus patterns at
both levels as well as to monitor the overall intactness of the system.
The following review summarizes our knowledge of coding and central
processing in the electrosensory system of gymnotiform fish and points out
commonalities with other systems, particularly the auditory system of the barn owl
(Konishi et al. 1988). Various aspects of the anatomy and physiology of the
electrosensory system have recently been presented in great detail (Bullock &
Heiligenberg, 1986; Bastian, 1986a; Carr & Maler, 1986; Heiligenberg, 1986) so
that references can largely be limited to more recent work. Issues of information
processing have been addressed separately in later publications (Heiligenberg,
1987,1988). This presentation will start with a description of stimulus coding at the
periphery and then focus on central mechanisms of information processing, with
particular emphasis on the jamming avoidance response (JAR) of Eigenmannia.
The summary diagram in Fig. 1 shows the flow of information controlling this
behaviour and may serve as a reference throughout this presentation.
180
Processing of electrosensory information 257
LOCATION STRUCTURE FUNCTION
Electric organ (EO) discharges at frequency
Tail EO
of the PN.
\
Pacemaker nucleus (PN) drives each discharge
Hindbrain PN cycle of EO by single command pulse.
Diencephalon
t,
PPN
Neurones in the prepacemaker nucleus (PPN) are excited
by Df<0 and inhibited by Df >0. Their excitation
raises the frequency of the PN.
I i
Body surface
tf
I> T
P- and T-type receptors code local amplitude and
phase, respectively.
Fig. 1. The flow of information controlling the jamming avoidance response (JAR) in
Eigenmannia. The fish generates a continuous, nearly sinusoidal electric organ
discharge (EOD) with a fundamental frequency of a few hundred hertz. The electric
signal caused by the interference of a neighbour's EODs with the animal's own EODs
is modulated in its instantaneous amplitude and phase at the difference frequency, Df,
of the two EODs. The animal minimizes the effect of jamming by a frequency too close
to its own by lowering its frequency in response to small positive Dfs (i.e. if the
neighbour's EOD frequency is slightly higher than its own) and by raising its frequency
in response to small negative Dfs. The necessary information about the sign of Df is
contained in the pattern of amplitude and phase modulations distributed over the
animal's body surface (see details in Heiligenberg, 1986, 1988).
181
258 W. HEILIGENBERG
H.-H
12
oJ i 0J
0
H.-H,
Fig. 2. Objects which differ electrically from the surrounding water distort the
animal's electric organ discharge (EOD) signal. (A) The nearly sinusoidal EOD of a
fish is recorded at two points, A and B, on its body surface. A small perturbation at A
could affect the local EOD amplitude, | S \A, as well as the EOD phase, or the timing of
the zerocrossing, HAHB, with reference to that of an unperturbed EOD at point B.
Sampling of amplitude and phase values at successive EOD cycles, t l t t2, ..., yields a
string of paired values which can be plotted in a two-dimensional amplitude-phase
plane to form a graph (B). This graph is coded by the activity of two classes of
'tuberous' electroreceptors (C), P-type receptors which fire intermittently, and T-type
receptors which fire one spike on each EOD cycle, phase-locked to the zerocrossing of
the signal. The probability, pA, of firing of P-type receptors at A codes the local
amplitude, ] S \A, whereas the difference, T^TB, in the timing of action potentials of
T-type receptors at A and B codes the differential phase, H^Hs. The EOD
distortions have been exaggerated for demonstration purposes. Realistic values are
shown in Figs 3 and 4.
electric field generated by their own electric organ discharges (EODs). Electrical
measurements in gymnotiform fish with continuous, nearly sinusoidal EODs show
that the amplitude as well as the phase (i.e. the timing of zerocrossings) of the
signal perceived by the animal are affected by the ohmic and capacitive loads of a
nearby object (Rose & Heiligenberg, 1986a). The spatial pattern of alterations of
amplitude and phase, monitored by large arrays of electroreceptors on the body
surface, represents the electric image of the object (Figs 2, 3).
Amplitude and phase of the signal at the animal's body surface are also
modulated when EODs of a neighbour interfere with the animal's own EODs. A
simultaneous evaluation of amplitude and phase modulations by the central^
182
Processing of electrosensory information 259
nervous system enables the animal to determine the sign of the frequency
difference between the two interfering EODs and to shift its own EOD frequency
away from that of its neighbour. In this way, the animal avoids jamming of
electrosensory inputs by frequencies too close to its own (Fig. 4). This jamming
avoidance response (JAR) has become a powerful tool for analysing central
mechanisms for the processing of amplitude and phase information.
Amplitude and phase are coded by separate types of 'tuberous' electroreceptors
distributed over the body surface. P-type receptors fire intermittently and
modulate their rate of firing in accordance with modulations in the local amplitude
of the signal. T-type receptors fire one spike in each EOD cycle, at a fixed phase in
reference to the zerocrossing of the signal. Separate primary afferents relay
responses of P- and T-type receptors to different targets in the electrosensory
lateral line lobe (ELL) which contains three somatotopically organized represen-
tations of the animal's body surface. Each representation receives the same
primary afferent information via axon collaterals, and each has distinct strata of
neurones for separate processing of amplitude and phase information.
In addition to tuberous electroreceptors, which are tuned to the fundamental
frequency of the animal's own EODs, electric fish also have 'ampullary' electrore-
ceptors which are sensitive to low-frequency signals. Such signals originate from
various animate and inanimate sources which the fish passes in its environment
(Kalmijn, 1987). Ampullary receptors project to a separate, single map of the body
surface in the ELL.
Gymnotiform fish with wave-type EODs thus have separate classes of electrore-
ceptors coding different aspects of the electrical environment. This separation of
channels is maintained at the first-order processing station, the ELL of the
hindbrain. A similar separation is seen in gymnotiform fish with pulse-type
discharges, although the organization and functional significance of their ampli-
tude- and phase-coding systems have not yet been explored in detail.
Two classes of electroreceptors, sensitive to different frequency ranges, are also
found in the African mormyriform fish which evolved an electrosensory system
independently from the gymnotiforms. 'Small pores', or ampullary receptors,
respond to low-frequency signals, and two types of tuberous electroreceptors are
tuned to the higher frequency range of the animal's EODs. In contrast to the
situation in gymnotiforms, however, these two types of tuberous receptors cannot
be considered as phase and amplitude coders. Instead, one type, called Knollenor-
gans, is adapted to code EODs of neighbours (Hopkins, 1986), whereas the second
type, called Mormyromasts, is adapted to code alterations of the animal's own
EOD field. These two types project to different central targets, and their
information is gated differently by a corollary signal of the electric organ
pacemaker (see Bell, 1989).
183
260 W. HEILIGENBERG
A R closed
l 2-0-
<
-30 -10 0
R open R open
2-0- A 00
" no C
2-0-
E
>
D Eigenmannia
a
< 2-0- VA
1-9-
-20 - 1 0 6 0 10 20 -10 0
Phase (/)
Fig. 3
nervous system appear to be even more selective in their response properties. The
P- and T-type receptors of gymnotiform fish, for example, project to different
classes of first-order neurones in the electrosensory lateral line lobe (ELL) of the
hindbrain (Maler et al. 1981; Mathieson et al. 1987). P-receptors, whose firing rate
is modulated in accordance with variations of signal amplitude, contact 'basilar'
pyramidal cells (or E-units) directly, and 'non-basilar' pyramidal cells (or I-units)
indirectly via inhibitory interneurones. As a consequence of these connections and
additional network properties, such as descending recurrent inputs (Bastian,
1986a,b; Bastian & Bratton, 1988), E-units are excited by a rise in signa|
184
Processing of electrosensory information 261
Fig. 3. Modulations in EOD amplitude and phase depend upon the electrical qualities
of the object. The inset in the upper left is a top view of a fish temporarily immobilized
by injection of a curare-like drug. Since this drug also silences the animal's EOD, a
substitute signal, S, of similar frequency (500Hz), amplitude and field geometry has
been provided by a signal generator and electrodes placed inside the mouth and at the
tip of the tail. The EOD-like signal is recorded by a pair of differential electrodes (+)
at the side of the animal's head. An object is moved sinusoidally at a rate of 0-5 Hz back
and forth between the locations A and B, approximately 1 cm away from and parallel
to the animal's body surface. In A-D the local peak-to-peak amplitude of 5 is plotted
on the ordinate, and its phase (i.e. its zerocrossing), measured in microseconds and in
reference to the cycle of the generator, is plotted on the abscissa. Since the frequency
of 5 is 500 Hz, a full cycle (i.e. a phase value of 2K) measures 2 ms. A signal amplitude
of 2mVcm"' and a phase defined as zero are recorded with the object in position B,
which is distant from the animal. In A-C, the object is a thin, 1-5 cm wide, vertically
suspended strip of metal foil, insulated on the side facing away from the animal. This
foil is connected via a capacitor, C, or an electrical short-circuit (R closed) to a distant
point in the water. In A dots mark every tenth cycle of the EOD-like signal, S, as the
object travels from A to B. Arrows indicate the direction of motion in the
amplitude-phase plane. Although the physical motion from A to B is, theoretically,
symmetrical to the motion from B to A, slight turbulence-related asymmetries in the
motion cause differences in the respective sections of the graph. Different natural
objects were chosen in D, a leaf of the water plant Vallisneria, approximately 1-5 cm
wide; a bulbous leaf stem of the water hyacinth, Eichhomia, approximately 2cm in
diameter; and a conspecific fish, with its EOD silenced by MS 222 anaesthesia. (From
Rose & Heiligenberg, 1986a.)
amplitude, whereas I-units are excited by a fall in signal amplitude. Both cell types
thus respond to changes (i.e. the temporal derivative) of signal amplitude and, in
this regard, resemble 'on' and 'off elements known in visual systems. Their
somatotopic organization within the ELL ensures that the animal can monitor the
spatial order of changes in signal amplitude over its entire body surface.
Although P-type receptor afferents fire action potentials within a preferred
phase range of the nearly sinusoidal wave form of the EOD and thus, at least
statistically, convey information about the phase of the signal, their representa-
tives in the ELL fire with almost constant probability at any phase of the EOD.
This is particularly the case for the I-units which, owing to the intervening
inhibitory interneurones, are synaptically even further removed from the recep-
tors. The E- and I-units in the ELL thus discard the small amount of phase
information that is left in the firing of their primary afferents and are totally
devoted to the coding of amplitude information.
T-receptor afferents code phase with some jitter as the timing of their action
potentials with reference to the EOD cycle shows a standard deviation of the order
of 30/is. Several T-receptor afferents from a given location on the body surface,
however, converge upon a single spherical cell in the ELL (Fig. 5). As a
consequence of their electrotonic synapses and the electrical properties of the
spherical cell and its spike initiation zone, the action potentials relayed by the
Spherical cell phase-lock to the EOD cycle with less jitter than do individual
185
262 W. HEILIGENBERG
0 J ,-
0 2n
Beat cycle Beat cycle
Fig. 4. Modulations in EOD amplitude and phase are also induced by the interfering
EODs of a neighbour, and the temporal and spatial patterns of these modulations
reflect the sign of the difference between the interfering frequencies. Following the
notation introduced in Fig. 2, the diagrams in the top row show the modulation in
amplitude, j S |^ (left) and phase, H^HB, (right), at a point A on the body surface
which experiences a mixture of the animal's own signal and that of its neighbour. The
phase reference point, B, is assumed not to be exposed to the neighbour's signal, and
the amplitude at point A measured in the absence of the neighbour's signal is defined as
1. Amplitude and phase modulations are recorded over one full beat cycle, indicated
schematically underneath the abscissae. (Note that with the animal's EOD frequency
being 500 Hz and that of its neighbour being 501 Hz, the beat frequency would be 1 Hz,
i.e. the beat cycle would be 1 s long and would cover 500 EOD cycles.) A joint plot of
the amplitude and phase modulations in the amplitude-phase plane yields a circle, and
the sense of rotation of this circle reflects the sign of the frequency difference: the
counterclockwise sense indicated implies that the neighbour has a higher frequency.
The lower row of diagrams shows the neuronal representation of the events in the top
row. Amplitude is coded by the probability, p^, of local P-type receptor firing, whereas
phase is coded by the difference in the timing of action potentials, T^Tg, of T-type
receptors located at points A and B on the body surface. The values plotted for p^ are
the mean numbers of spikes recorded from a single primary afferent per successive
EODs in the beat cycle. Data from 25 beat cycles were averaged. Similarly, the values
plotted for T^TB are means from 25 beat cycles. (From Heiligenberg & Partridge,
1981.)
186
Processing of electrosensory information 263
studies, has meanwhile been confirmed by intracellular recordings from spherical
cells in Eigenmannia (M. Kawasaki & W. Heiligenberg, unpublished obser-
vation). The same phenomenon has been observed in the giant cells of lamina 6 of
the torus which receive converging inputs from spherical cells (Carr et al. 1986a).
Although some T-afferents may respond to an increase in signal amplitude by
slightly advancing the timing of their action potential within the EOD cycle, the
timing of the action potentials of spherical cells is rather insensitive to modulations
of signal amplitude. Spherical cells thus provide phase information which is not
only less jittery but also purer than phase information provided by T-receptors.
A particular form of signal filtering is employed in the processing of primary
afferent information in the mormyrids. The Knollenorgans have lower thresholds
than the Mormyromasts and are, therefore, suitable for the detection of EODs of
distant neighbours. In addition, by firing only a single spike in response to an EOD
pulse, they only code the occurrence of this event but not its intensity. Afferent
information from Knollenorgans is gated by an inhibition generated by a corollary
discharge of the animal's electric organ pacemaker system, so that action
potentials caused by the animal's own EODs are not relayed to higher-order
neurones in the midbrain. Central information provided by Knollenorgans,
therefore, only represents the electrical presence of neighbours (for more details
see Bell, 1989).
The higher threshold of Mormyromasts makes this type of receptor less sensitive
to the EODs of neighbours and, by firing a burst of spikes which increases with the
amplitude of the signal, they code local intensity of the animal's own EOD.
Moreover, their input to higher-order neurones is facilitated by a corollary
discharge of the electric-organ pacemaker system so that the central nervous
system is informed preferentially about local intensities of the animal's own signal.
The Mormyromast system is thus dedicated to the analysis of electric images coded
by alterations of the animal's own EOD field (for more details see Bell, 1989).
Other sensory systems similarly show separate channels which are adapted to
code different aspects within their modality, and a further separation of these
aspects may be achieved by separate central pathways. Well-known examples are
the magno- and parvocellular systems in mammalian vision.
Although the availability of different classes of receptors facilitates separate
coding of different stimulus aspects, their presence is not always necessary to
achieve such a separation at a central level. The auditory system of the barn owl
presents an example. Primary auditory afferents are tuned to individual frequency
bands and respond to modulations in signal amplitude by modulating their rate of
firing accordingly. Although these afferents are unable to fire an action potential
on each cycle of a carrier signal of sufficiently high frequency, they are still able to
code its phase by firing their action potentials at a preferred moment within the
carrier cycle. Primary afferents send axon collaterals to two nuclei, the nucleus
angularis, which processes amplitude information, and the nucleus magnocellular-
is, which processes phase information. Neurones in the nucleus angularis
modulate their rate of firing in accordance with amplitude modulations of the
187
264 W. HEILIGENBERG
TECTUM
CD
V \ Lamina 3
Q
t Lamina 5
TORUS
^- Lamina 6
c Lamina 7
t Lamina 8
T-afferent P-afferent
(phase coder) 6 (amplitude coder)
RECEPTORS
o
Fig. 5
auditory signal within their frequency band, but the timing of their spikes shows
almost no correlation with the phase of the carrier. Neurones in the nucleus
magnocellularis, however, fire action potentials which appear to be even more
tightly locked to the phase of the carrier than are those of primary afferents, but
their rate of firing does not vary with the amplitude of the carrier. Differences in
dendritic and synaptic organization within these two nuclei must account for this
separation of amplitude information and phase information, which from then on
are processed separately up to the level of the midbrain, much as in the case of the
electrosensory system of gymnotiform fish (see Takahashi, 1989).
188
Processing of electrosensory information 265
Fig. 5. The somatotopic representation of electrosensory information in the hindbrain
and midbrain. Four maps of the body surface are found on each side of the hindbrain in
the electrosensory lateral line lobe (ELL). Although the most medial map only
receives input from ampullary receptors, the remaining three maps receive identical
input through collaterals of primary afferents from tuberous electroreceptors,
P-afferents as well T-afferents. The organization of first-order neurones in the
tuberous pathway of the ELL is shown on the right. T-afferents from small receptive
fields on the body surface converge via electrotonic synapses upon spherical cells
which, in turn, project in somatotopic order to lamina 6 of the torus semicircularis of
the midbrain. P-afferents form excitatory synapses on the basilar dendrites of basilar-
pyramidal cells (E-units) and, via inhibitory interneurones, also contact non-basilar
pyramidal cells (I-units). As a consequence of these connections, E-units are excited by
a rise in stimulus amplitude, whereas I-units are excited by a fall. These amplitude-
coding cells project in somatotopic order to various laminae of the torus semicircularis
above and below lamina 6. Note that the torus as well as the next-order station, the
tectum opticum, both contain only a single map of the body surface. (From
Heiligenberg, 1988.)
189
266 W. HEILIGENBERG
with the difference in the timing of their two inputs. These small cells, therefore,
only code differential phase, and the timing of their spikes is, effectively, no longer
related to the individual timing of their inputs. The system discards information
about the absolute timing of signals at individual sites on the body surface and only
retains the more relevant information about differences in timing. This difference
is now coded by the rate of firing of neurones dedicated to the comparison of
inputs from two distinct sites on the body surface. A small neurone within the
location of lamina 6 representing a specific site, A, on the body surface, for
example, will raise its rate of firing if the signal in A experiences a small phase lead
with reference to the signal in some specific area, B, and the same neurone will
lower its rate of firing if area A experiences a phase lag instead. A nearby neurone
of this kind also represents area A, but it may have a different phase reference
area, C, on the body surface (Fig. 6).
Similar and, in some cases, more elaborate transformations of information
occur at still higher levels of the nervous system in connection with the analysis of
complex stimulus patterns. The differential phase information, for example,
provided by the small neurones in lamina 6 for a given area on the body surface is
combined with amplitude information from the same area in deeper laminae of the
torus and tectum.
190
Processing of electrosensory information 267
Giant
cells
Small
cells
I H.-Hr
Torus
Lamina 6
Ho
H fl
Spherical
cells I B(r c i D(t. ELL
Fig. 6. Lamina 6 of the torus semicircularis contains small cells which code differential
phase between signals arriving from different points, A,B,C,D, on the body surface.
Spherical cells of the ELL (see Fig. 5), coding signal phase, H^,... ,H D , in their
respective receptive fields, project in somatotopic order to the giant cells of lamina 6 as
well as to the dendrites of small cells in their vicinity. (Although several spherical cells
may converge on a single giant cell, only single-cell connections have been drawn here
for simplicity.) Giant cells, in turn, send axonal collaterals across wide areas of lamina
6 and contact the somata of small cells. The size of their synapses appears to preclude
that more than one giant cell can contact any given small cell. Small cells in the location
of lamina 6 representing point A on the body surface modulate their rate of firing in
accordance with the small temporal disparities, H^-H*, between action potentials
arriving from area A at their dendrites and from specific reference areas, B,C,D, on
their soma. (From Heiligenberg, 1987, and based upon data from Carr et al. 19866.)
difference and allows the fish to shift its own frequency in the appropriate
direction. If the neighbour's EOD frequency is lower than the animal's own, for
example, large areas of the body surface will predominantly experience rises in
local stimulus amplitude paired with phase leads and falls in local stimulus
amplitude paired with phase lags. Higher-order neurones in the deeper laminae of
the torus and tectum recognize one or the other of these combinations by
apparently gating amplitude inputs by phase inputs (Heiligenberg & Rose, 1986).
It is essential for this form of evaluation that amplitude and phase modulations be
compared in spatial register, and this ordered comparison is facilitated by the
laminar organization of the torus semicircularis of the midbrain. While lamina 6
processes phase information, laminae 5 and 7 receive somatotopically ordered
191
268 W. HEILIGENBERG
Phase discriminator
Spike rate
0
AM coder ( HB
Sign-selective neurone d
o o
Fig. 7. The gating of amplitude information by phase information. Sign-selective
neurones of the midbrain respond to particular combinations of changes in local
amplitude, | S \A, in a given area A on the body surface and values of differential phase,
HAHB, between this area and some reference area, B. Sign-selective neurones,
therefore, can discriminate the sense of rotation of graphs in the amplitude-phase
plane. The unit on the left is excited by a clockwise rotation centred at zero, whereas
the unit on the right is excited by a counterclockwise rotation. The AM coders
represent higher-order E- and I-type neurones which are excited by a rise and fall,
respectively, in local signal amplitude. Phase discriminators occur first in the form of
small cells in lamina 6 of the torus (see Fig. 6). Higher-order phase discriminators are
found in deeper laminae of the torus and in the tectum. The synaptic organization
shown here is inferred from responses of sign-selective neurones to independent
manipulations of amplitude and phase inputs. (From Heiligenberg & Rose, 1986.)
amplitude information. Since all laminae are stacked parallel to each other as well
as in somatotopic register, local perpendicular connections between laminae
provide connections necessary for the joint computation of local amplitude and
phase information. As a consequence of this ordered arrangement, very simple
rules for interlaminar connections should suffice to provide the necessary ordered
inputs of amplitude and phase information for higher-order neurones (Heiligen-
berg, 1987) (Fig. 7).
192
Processing of electrosensory information 269
behaviourally defined selectivity for a stimulus pattern gradually emerges at the
neuronal level through several levels of integration of sensory information.
Eigenmannia discriminates the sign of Df by simultaneous evaluation of
modulations in local signal amplitude and differential phase. Accordingly, the
lowest-order neurones sensitive to the sign of Df are found in the torus
semicircularis, where the amplitude- and phase-coding pathways converge. Since
individual 'sign-selective' neurones of the torus, however, receive information
only from limited receptive fields on the body surface, their responses depend
significantly upon the orientation of the interfering EOD field, and a change in the
orientation may even reverse a neurone's sign preference (Rose & Heiligenberg,
1986a). The activity of a single neurone, therefore, does not reflect the sign of Df
unambiguously. One can show, however, that the pooled activity of a large
population of such neurones, with receptive fields distributed over a wide area of
the body surface, should always yield reliable information about the sign of Df
(Heiligenberg & Rose, 1986). The population of sign-selective cells in the torus
can indeed be compared with a parliament in which members cast conflicting
votes, and a behavioural decision reflects the opinion of the majority. One can
even show that a minority of members, due to deficiencies in the transfer of phase
information, always casts the 'wrong' vote (Heiligenberg, 1987). The sign-selective
neurones of the torus differ widely with regard to the magnitude of Df at which
they respond best, although the majority prefers magnitudes of Df in the
behaviourally optimal range between 2 and 8 Hz.
The nucleus electrosensorius complex of the diencephalon receives direct and,
via the tectum opticum, indirect input from the torus. Sign-selective cells at this
level appear to obtain converging information from larger populations of sign-
selective cells of the torus and tectum and, as a likely consequence, discriminate
the sign of Df more reliably. Although their degree of sign discrimination may still
depend upon the orientation of the interfering EOD field, their sign preference
never reverses (Keller, 1988). Two additional features result from the assumed
convergence of sensory information from torus to nucleus electrosensorius.
Neurones in the latter structure are more sensitive to weak interfering EOD fields,
some of them almost reaching behavioural threshold, and the rate of their firing is
less strongly locked to the temporal modulation pattern of amplitude and phase
generated by the interfering EODs. The neurones of the nucleus electrosensorius,
however, do not yet fully reflect the dynamic properties of the JAR as at least
some of them prefer magnitudes of Df outside the behaviourally optimal range
2-8 Hz. Moreover, a neurone responding most vigorously to positive Dfs in the
range 2-8 Hz may not be inhibited evenly across all values of negative Dfs. It may
actually be more active for negative Dfs between 2 and 8 Hz than for positive
Dfs outside the range 2-8Hz. The firing rate of such a neurone, therefore, still
does not reveal the sign of Df unambiguously for all magnitudes of Df (Keller,
1988).
These minor differences between behavioural and neuronal response properties
vanish at a still higher level of integration, in the diencephalic prepacemaker
193
270 W. HEILIGENBERG
nucleus which innervates the final motor command structure, the electric-organ
pacemaker nucleus in the medulla of the hindbrain (Kawasaki et al. 1988a). Small
neurones of the prepacemaker raise their rate of firing in response to negative Dfs
(i.e. when the neighbour's EOD frequency is lower), and the animal raises its own
EOD frequency so that a larger separation of the interfering frequencies is
achieved. The same neurones lower their rate of firing in response to positive Dfs
which cause a lowering of the animal's EOD frequency. The firing rate of any
single neurone thus unambiguously reflects the sign of Df, and the strength of the
neuronal response is directly related to the strength of the behavioural response.
These sign-selective neurones are rather homogeneous in their response charac-
teristics, which closely resemble the behavioural response. In particular, they
show practically the same dependence upon the magnitude of Df as does the JAR,
many of them are as sensitive to weak interfering fields as is the behavioural
response of the intact animal, and the response of many neurones varies as little
with the orientation of the interfering EOD field as does the intact behaviour.
Finally, the firing rate of these neurones is no longer modulated at the difference
frequency, Df, with the exception of those very small magnitudes of Df which also
cause corresponding weak modulations in the animal's own EOD frequency (Rose
et al. 1988). The small cells of the prepacemaker nucleus, therefore, code a
behavioural effort, namely to change the frequency of the pacemaker, and their
activity is no longer affected by aspects of the stimulus regime which are irrelevant
to this particular behavioural response.
Although torus and tectum are structured somatotopically, no obvious somato-
topic order is found at the next higher level, the nucleus electrosensorius which,
instead, reveals a motor map. Whereas stimulation of one portion of this nucleus
raises the pacemaker frequency, stimulation of a more ventral and rostral portion
lowers the pacemaker frequency. Both structures are essential for the JAR, since
their selective, bilateral lesion abolishes the animal's ability either to raise or to
lower its frequency, respectively (Keller & Heiligenberg, 1989).
194
Processing of electrosensory information 271
availability of multiple maps receiving identical information thus offers the
opportunity for processing this information under different aspects in separate,
specialized networks. It appears plausible that additional maps originated as
'accidental' reduplications of existing maps and were then progressively adapted
for specific tasks in sensory processing. An evolutionary scenario of this kind was
proposed by AUman et al. (1981) for multiple visual representations in mammals.
Multiple presentations are found at all levels of the nervous system, from
sensory processing to motor control. Neurones which respond very selectively to a
specific stimulus pattern and appear to be dedicated to the control of a particular
kind of behaviour emerge rather late within the neuronal hierarchy. Lower-order
neurones, even at a level as high as the tectum opticum, still have rather general
response characteristics. Electrosensory neurones of the tectum of gymnotiform
fish respond to certain combinations of phase and amplitude modulations which
may be generated by moving objects as well as by interfering EODs of neighbours
(Rose & Heiligenberg, 1986a; Heiligenberg & Rose, 1987). The firing of such a
neurone, therefore, tells us little about the world outside, unless we monitor the
activity of this neurone for a period much longer than the behavioural response
latency of the animal. Simultaneous recordings from many such neurones and
parallel evaluation of the spatial and temporal order of their activity, however,
would allow us to discriminate between moving objects and interfering EODs
almost immediately. These tectal neurones send axonal collaterals to various
targets in the diencephalon, midbrain and hindbrain so that their information can
be processed by different algorithms in separate higher-order structures. One of
these structures, the nucleus electrosensorius, receives additional inputs from the
torus semicircularis and participates in the control of the JAR (Keller &
Heiligenberg, 1989). Other structures in the reticular formation are apparently
adapted to monitor the motion of objects and to control movements of the body.
Various behavioural systems thus share rather general spatial and temporal
information provided by the tectum and are able to extract specific patterns for
their own control.
195
272 W. HEILIGENBERG
pattern is restricted to smaller regions of the body and thus to a smaller set of
electroreceptors (Rose & Heiligenberg, 1985). This extreme temporal resolution
is surprising in view of the considerable jitter observed in the phase-coding system.
As was mentioned previously, T-type afferents record the timing of the
zerocrossing of a signal by firing an action potential at a specific latency, with a
standard deviation of approximately 30/is. Through convergence at the level of
the ELL and within lamina 6 of the torus, this jitter is reduced by a factor of
approximately three at the level of lamina 6 (Carr et al. 1986a). Higher-order units
of the torus can resolve differential phase values as small as 10 pts,, and averaging of
their activity over periods much longer than the latency of the JAR reveals
thresholds for their phase discrimination close to that measured behaviourally
(Rose & Heiligenberg, 19866). Converging inputs from large numbers of such
neurones at the level of the nucleus electrosensorius and further convergence from
this nucleus to the prepacemaker nucleus should, therefore, yield single neurones
with a temporal resolution close to that of the intact behaviour. Neurones of such
sensitivity are indeed found in the prepacemaker nucleus, and, much as is true for
the intact behaviour, they require sensory inputs from sufficiently large areas of
the body surface to achieve this high sensitivity (Kawasaki et al. 19886).
196
Processing of electrosensory information 273
specific types of synaptic connections and, thereby, to establish the rules for
connections within networks, rather time-consuming ultrastructural studies of
labelled neurones are necessary (Carr et al. 1986i>; Mathieson et al. 1987). In some
instances, in vivo experiments can be complemented by in vitro experiments on
structures isolated in a slice chamber. This approach, allowing for better control of
ionic conditions and the presence of agonists or antagonists of certain transmitters,
has been very fruitful in the study of the ELL (Mathieson & Maler, 1988) and the
medullary pacemaker (Dye, 1988) by revealing the functional significance of
certain transmitters. These findings could then be tested in vivo by applying
specific agonists or antagonists to particular locations within these structures
(Shumway & Maler, 1989; Dye etal. 1988). The structural and functional
organization of the ELL has now been identified in great detail, and the response
properties of its neurones can, at least qualitatively, be explained on the basis of
their specific connections and the action of known transmitters. A quantitative
theory of the operation of the ELL, however, will not only require more detailed
studies of its dynamic properties but also network simulations, preferably
performed on large computers structured in parallel.
References
ALLMAN, J. M., BAKER, J. F., NEWSOME, W. T. & PETERSEN, S. E. (1981). Visual topography
and function. In Cortical Sensory Origination, vol. 2 (ed. C. N. Woolsey), pp. 171-185.
Qifton, New Jersey: Humana Press.
BALDI, P. & HEIUGENBERG, W. (1988). How sensory maps could enhance resolution through
ordered arrangements of broadly tuned receivers. Biol. Cybernetics 59, 313-318.
197
274 W. HEILIGENBERG
198
Processing of electrosensory information 275
MALER, L., SAS, E. & RODGERS, J. (1981). The cytology of the posterior lateral line lobe of high-
frequency weakly electric fish (Gymnotidae): Dendritic differentiation and synaptic
specificity in a simple cortex. J. comp. Neurol. 195, 87-140.
MATHIESON, W. B., HEIUGENBERG, W. & MALER, L. (1987). Ultrastructural studies of
physiologically identified electrosensory afferent synapses in the gymnotiform fish,
Eigenmannia. J. comp. Neurol. 255, 526-537.
MATHIESON, W. B. & MALER, L. (1988). Morphological and electrophysiological properties of a
novel in vitro preparation: the electrosensory lateral line lobe brain slice. /. comp. Physiol. A
163, 489-506.
MITTELSTAEDT, H. & EGGERT, T. (1988). How to transform topographically ordered spatial
information into motor commands? In Visuomotor Coordination: Amphibians, Comparisons,
Models and Robots (ed. M. A. Arbib & J. P. Ewert), pp. 569-585. New York: Plenum Press.
ROSE, G. J. & HEIUGENBERG, W. (1985). Temporal hyperacuity in the electric sense of fish.
Nature, Lond. 318, 178-180.
ROSE, G. J. & HEIUGENBERG, W. (1986a). Neural coding of difference frequencies in the
midbrain of the electric fish Eigenmannia: reading the sense of rotation in an amplitude-phase
plane. J. comp. Physiol. A 158, 613-624.
ROSE, G. J. & HEILIGENBERG, W. (1986b). Limits of phase and amplitude sensitivity in the torus
semicircularis of Eigenmannia. J. comp. Physiol. A 159, 813-822.
ROSE, G. J., KAWASAKI, M. & HEILIGENBERG, W. (1988). 'Recognition units' at the top of a
neuronal hierarchy? J. comp. Physiol. A 162, 759-772.
SHUMWAY, C. A. (1989a). Multiple electrosensory maps in the medulla of weakly electric
gymnotiform fish. I. Physiological differences. J. Neuroscience (in press).
SHUMWAY, C. A. (19896). Multiple electrosensory maps in the medulla of weakly electric
gymnotiform fish. II. Anatomical differences. /. Neuroscience (in press).
SHUMWAY, C. A. & MALER, L. (1989). GABAergic inhibition shapes temporal and spatial
response properties of pyramidal cells in the electrosensory lateral line lobe of gymnotoid fish.
/. comp. Physiol. A 164, 391-407.
TAKAHASHI, T. (1989). The neural coding of auditory space. J. exp. Biol. 146, 307-322.
199
200
The Journal of Experimental Biology 202, 13111318 (1999) 1311
Printed in Great Britain The Company of Biologists Limited 1999
JEB2088
*e-mail: mfriedman@hms.harvard.edu
Summary
Mormyrid electric fish communicate using pulse-type appears to take place in the nucleus exterolateralis pars
electric organ discharges (EODs). The fine temporal anterior (ELa), where tightly phase-locked inputs from the
structure of the waveforms of EODs varies widely hindbrain drive a direct, excitatory input through a long
throughout the 200 or more species of mormyrids. These axonal delay line and also drive an indirect, inhibitory
signals carry information about the species, the sex and input with negligible delay through the ELa large cell.
even the individual identity of the signaller. Behavioral These two inputs converge on ELa small cells, where they
experiments have shown that some species of fish are are hypothesized to interact in a delay-line/blanking
capable of using this information. Of the four known types model. This initial temporal analysis is further refined in
of electroreceptors in mormyrid fish, the knollenorgan the nucleus exterolateralis pars posterior, where units
electroreceptor is the one most likely to be involved in the tuned to ranges of pulse durations have been identified
detection of conspecific EOD signals. Here, we review some physiologically.
recent advances in understanding how the central
knollenorgan pathway might be analyzing the temporal Key words: time-coding, knollenorgan, mormyrid, electric organ
structure of the EOD waveform. Fine temporal analysis discharge, waveform.
Introduction
Electric communication signals retain their fine temporal 1981). These experiments suggested that the salient feature of
structure as they are transmitted from signaller to receiver in the EOD is its temporal waveform structure.
the aquatic environment, unlike auditory signals, which are Mormyrid fish have four different kinds of electroreceptors
corrupted by echoes and reverberation (Hopkins, 1986b). This used for active and passive electrolocation and communication.
enables weakly electric fish to communicate useful information Among these, knollenorgan receptors appear the most likely to
even with their brief electric organ discharges (EODs). EODs detect and to respond to communication signals from other
from different species of African mormyrid electric fish vary fish. First, they have a high sensitivity and they are broadly
widely in waveform (14 phases), duration (60 s to 20 ms) tuned to the spectrum of the species own EOD (Bass and
and polarity. The EODs of many species show sexual Hopkins, 1980, 1984). Furthermore, if parts of the
differences in waveform (Kramer, 1997; Hopkins, 1999). It is knollenorgan pathway in the central nervous system (CNS) are
possible in some cases to recognize individuals on the basis of lesioned, some communication behaviors are abolished
their EODs (Crawford, 1992; Friedman and Hopkins, 1996). (Moller and Szabo, 1981). The motor areas that generate the
Mormyrids use EOD waveforms to discriminate the species fishs own EOD send a strong inhibitory input to the primary
and sex of the signaller (Hopkins, 1981; Hopkins and Bass, sensory areas in the hindbrain (Zipser and Bennett, 1976; Bell
1981; Graff and Kramer, 1992). One behavioral study and Grant, 1989). Thus, the primary afferents from
addressed the mechanism by which they perform this task. knollenorgan receptors respond to the fishs own discharge, but
Males of the species referred to as Brienomyrus brachyistius this response never gets past the hindbrain electrosensory
(tp) (Hopkins and Bass, 1981) or Brienomyrus sp. 2 (Alves- lateral line lobe (ELL), making the fish deaf to its own
Gomes and Hopkins, 1997) exhibit a sex difference in the EOD discharges. Together, these characteristics imply that the
waveform. Males responded with courtship displays (rasps or knollenorgan pathway is specialized for detecting and
bursts of EODs delivered at a high rate) when presented with processing EOD signals generated by neighboring, conspecific
square pulses of a duration that matched female EODs, but they fish.
failed to respond to shorter or longer square pulses or to pulses Knollenorgan electroreceptors are scattered over the body
derived from a phase-shifted female EOD (Hopkins and Bass, surface. For a stimulus consisting of a brief pulse of electrical
201
1312 M. XU-FRIEDMAN AND C. D. HOPKINS
B Normal
A polarity
Reverse
polarity
EOD stimulus
0.5 ms
C D
OB
ELa ll
Tel ELp Ipsi
MV
Contra
OT
VA Valvula
ELa
IG SPE
ELp
1 Aff
ELL
nPLL Knollenorgan
1 mm NELL
Fig. 1. Summary of the knollenorgan pathway. (A) Drawing of a mormyrid fish, Brienomyrus brachyistius. (B) Response of a knollenorgan
(upper trace) to an electric organ discharge (EOD) stimulus (lower trace). The response is displayed as a post-stimulus time histogram. For the
upward histogram, the EOD stimulus shown was presented to the knollenorgan multiple times. For the downward histogram, the inverted EOD
stimulus was presented. The downward histogram is equivalent to the responses of knollenorgans on the opposite side of the body. Thus,
knollenorgans on one side of the body respond to the upward edge of the stimulus (with some delay) while knollenorgans on the opposite side
respond to the downward edge. Bar, 25 spikes per bin. (C) Overview of the brain of Brienomyrus brachyistius. On the right, the valvula
cerebelli has been removed. ELa, nucleus exterolateralis pars anterior; ELL, electrosensory lateral line lobe; ELp, nucleus exterolateralis pars
posterior; nPLL, posterior lateral line nerve; OB, olfactory bulb; OT, optic tectum; Tel, telencephalon; VA, valvula cerebelli. (D) Overview of
the known connections of the knollenorgan pathway. Contra, contralateral; IG, isthmic granule nucleus; Ipsi, ipsilateral; ll, lateral lemniscus;
MV, nucleus medialis ventralis; NELL, nucleus of the ELL; SPE, subpreminential nucleus; 1 Aff, primary afferent. Modified from Amagai
et al. (1998) and Friedman and Hopkins (1998).
current, either a square pulse or an EOD, knollenorgans on one line lobe (NELL) (Fig. 1D; Bell and Russell, 1978; Szabo et
side of the body respond to the onset of current with a phase- al., 1983; Hopkins et al., 1993). The terminals onto the NELL
locked spike, and those on the opposite side of the body cells are large and electrotonic (Mugnaini and Maler, 1987b).
respond to the offset (Fig. 1B; Hopkins and Bass, 1981; The NELL somata are large and adendritic (Szabo and
Hopkins, 1986a). Therefore, fish could analyze the temporal Ravaille, 1976) and are strongly immunoreactive for calretinin
structure of the stimulus by comparing the temporal pattern of (Friedman and Kawasaki, 1997). These anatomical and
spikes arriving from knollenorgan electroreceptors all over the neurochemical characteristics are commonly associated with
body. In the simplest case, they could measure the duration of neural pathways specialized for the preservation of fine
the pulse by measuring the time difference between the two temporal information, such as the auditory azimuthal
sides of the body (Hopkins and Bass, 1981). This comparison localization pathways in birds and mammals (Konishi, 1991)
must occur in the CNS, and here we describe some anatomical and the electrosensory phase-coding systems in the electric fish
and physiological characteristics of the knollenorgan pathway Eigenmannia and Gymnarchus (Carr, 1986; Kawasaki, 1997).
that could provide the neural basis for this task. The NELL sends a thick axon bilaterally up the lateral
lemniscus to end in two nuclei in the torus semicircularis: the
nucleus medialis ventralis (MV) and the nucleus exterolateralis
Specializations for temporal analysis pars anterior (ELa) (Fig. 1D; Szabo et al., 1983). The axon
Knollenorgan electroreceptors project roughly somato- branch to the MV is thin, and the terminals there are small en
topically to the ipsilateral nucleus of the electrosensory lateral passant boutons (Amagai et al., 1998; Friedman and Hopkins,
202
Knollenorgan pathway in mormyrids 1313
1998). In contrast, the axon branch to the ELa is thick and direct, but delayed, excitatory input from a NELL cell. Also,
heavily myelinated, and the terminals it makes are large and the delays to different small cells may vary because NELL
electrotonic. Thus, the ELa shows anatomical specializations cells show a range of axonal lengths.
typically associated with the preservation of fine temporal Like knollenorgan electroreceptors, both NELL axons and
information, but the MV does not. In addition, the midbrain is ELa large cells respond to electrosensory stimuli with a tightly
the first stage in the knollenorgan pathway where information phase-locked spike (Amagai et al., 1998; Friedman and
from all parts of the body can be compared. Therefore, the ELa Hopkins, 1998). NELL axons and ELa large cells are not
is more likely to be the site where temporal analysis takes place segregated according to their receptive field locations because
(Szabo et al., 1983), while the MV may perform some other a single recording electrode advanced through the ELa runs
function, such as spatial analysis. across cells at different depths that respond to either one edge
There are only two cell types in the ELa, large cells and of the stimulus or the other (Amagai et al., 1998).
small cells (also called interstitial and granule cells by Mugnaini and Maler (1987a) first suggested that inhibition
Mugnaini and Maler, 1987a). They are easily differentiated by could be used in a temporal blanking model. Refining this
size (approximately 10 m versus approximately 6 m in soma model to include subsequent discoveries about physiological
diameter). The NELL axon terminates onto ELa large cells and responses and axonal arborizations, we show how the
small cells, with mixed chemical and electrical synapses, anatomical specializations within the ELa could be responsible
suggesting that synaptic delays are extremely short (Mugnaini for fine temporal analysis (Friedman and Hopkins, 1998). In
and Maler, 1987a). ELa large cells and small cells both appear this model, a small cell receives one input from a large cell
to be adendritic, according to intracellular dye-fills of ELa with its receptive field on one side of the body (Fig. 3A). It
large cells with biocytin and Lucifer Yellow and retrograde also receives a second input from a NELL cell with a receptive
labelling of small cells with biotinylated dextrans (Amagai et field on the opposite side of the body, but with some delay
al., 1998; Friedman and Hopkins, 1998), although Mugnaini between the large cell and NELL inputs. For a sufficiently long
and Maler (1987a) found dendritic arborizations on ELa large negative-going pulse, the excitatory NELL input will arrive
cells using Golgi staining. In any case, the terminals from first, and the small cell will respond. However, if the pulse is
NELL axons all appear to be directly onto the large cell and shorter than the axonal delay, then the inhibitory large cell
small cell somata, making it possible to identify completely input will arrive first, and the NELL input will be suppressed.
each postsynaptic cell contacted by an intracellularly labelled In other words, a small cell responds if the pulse is longer than
NELL axon. Reconstructions show that NELL axons terminate some threshold duration, set by the length of the NELL axonal
on 12 large cells, then wind extensively and terminate on delay. Fig. 3B illustrates the predicted response probability as
small cells throughout the ELa (Fig. 2A) (Friedman and a function of square pulse duration and polarity. For a positive-
Hopkins, 1998). going pulse, the inhibitory large cell input would always arrive
ELa large cells project entirely within the ELa (Fig. 2B), first, suppressing responses to the excitatory input for the
terminating with large calyceal terminals that envelop their duration of the inhibitory postsynaptic potential (IPSP), which
postsynaptic small cells. The large cells are probably inhibitory is presumably long enough to block all behaviorally relevant
because they are immunopositive for glutamic acid stimuli. Thus, the ELa small cell should respond only to one
decarboxylase, the enzyme that makes the neurotransmitter polarity of the stimulus. If different NELL axons have different
-aminobutyric acid (GABA) (Mugnaini and Maler, 1987a). axonal delays, then different small cells would be tuned to
The large cell axons project fairly directly across the nucleus, discriminate between stimuli of different duration. By
ending on more restricted bands or patches of small cells combining the inputs of many small cells, each specialized for
(Friedman and Hopkins, 1998). a different delay and for different patches of the body surface
Linear reconstructions (Fig. 2C) show that, when a NELL and polarities of stimulation, there should be sufficient
axon enters the ELa, it first contacts a large cell and then travels information to make waveform duration discriminations.
for approximately 1 mm, where it may contact a second large This delay line-blanking model is supported by some
cell; it then travels for 34 mm with few terminals, before preliminary data. Recordings from the ELa, putatively from
branching widely over a large number of small cells (Friedman small cells, have as predicted two different kinds of synaptic
and Hopkins, 1998). The distance between the first large cell activity, which can be tied to different edges of the stimulus
terminal and the last small cell terminal may be as great as and with different latencies (Friedman and Hopkins, 1998).
7 mm, even though the ELa is only 1 mm in diameter. This These synaptic potentials are consistent with direct inhibitory
difference in axonal delay translates into a NELL cell input from large cells and delayed excitation from NELL
activating its first large cell 230460 s earlier than its last axons. However, data are available from only a few small cells
small cell, using a first estimate of conduction velocity of because their small size and adendritic somata limit successful
15 m s1 for the NELL axons (Enger et al., 1976). In contrast, penetrations.
the large cell axon runs only approximately 1 mm from soma
to terminals (equivalent to 60 s), so it relays signals quickly
to its postsynaptic small cells. Thus, the small cell appears to Further processing of temporal information
receive an indirect, inhibitory input from a large cell, and a ELa small cells project exclusively to the neighboring
203
1314 M. XU-FRIEDMAN AND C. D. HOPKINS
A B
1
2
3
4
100 m 5
6
7
100 m
0 1 2 3 4 5 6 7
Relative axon length (mm)
Fig. 2. Anatomy of cell types in the nucleus exterolateralis pars anterior (ELa). (A) Reconstruction of a nucleus of the electrosensory lateral
line lobe (NELL) cell axonal arborization within the ELa. Lateral is left, anterior is up. The cell was filled with biocytin, and the reconstruction
is made from 15 sections of 50 m thickness. Terminals onto small cells are green, and terminals onto large cells are red. The thickness of the
axon denotes its dorsal-to-ventral extent. The outline of the nucleus is taken from the largest section encompassing the entire reconstruction.
(B) Reconstruction of an ELa large cell, from seven sections of 50 m thickness. The soma is red, and the terminals onto small cells are green.
(C) Linear reconstructions of four NELL axonal arborizations and one ELa large cell. Modified from Friedman and Hopkins (1998).
midbrain nucleus exterolateralis pars posterior (ELp) They project straight across the nucleus, preserving their
(Hauged-Carr, 1979). The small cell axons within the ELp topographical organization (Friedman and Hopkins, 1998).
are thin, making small en passant boutons along their length. The ELp has two cell types, identified from extracellular
204
Knollenorgan pathway in mormyrids 1315
A
NELL ELa large cell
t t
+
-
Fig. 3. Small cell model. (A) Small cell inputs. Knollenorgans
in one patch of the body surface (red) respond to the upward + ELa small cell
edge of a square pulse, and contribute signals to a nucleus of tdelay
NELL
the electrosensory lateral line lobe (NELL) axon that
terminates on a nucleus exterolateralis pars anterior (ELa)
large cell, which in turn makes an inhibitory contact onto a
small cell. On the opposite side of the body (green patch), the
knollenorgans respond to the downward edge of the stimulus B
and contribute signals to a NELL axon (green) that terminates
directly on the small cell, but with a relative delay tdelay. This 100% response probability
is only one example of possible receptive field organization.
(B) Proposed small cell responses. The abscissa represents the
duration and polarity of a square pulse stimulus, and the
ordinate is the response probability. For stimuli of one
polarity, small cells respond to stimuli longer than some
threshold duration. The threshold depends on the delay line,
which is different for different cells. Modified from Friedman
and Hopkins (1998). Pulse duration (t t )
recordings (Amagai, 1998). Type I cells respond at a shorter the outputs from the ELa in a manner less dependent on precise
latency (79 ms), with lower jitter and with higher response spike times, although more detailed physiological studies are
probability, whereas type II cells respond at longer latency necessary.
(1220 ms), with higher jitter and with lower response Individual type I cells project widely throughout the
probability (Amagai, 1998). Most interestingly, type I cells midbrain, with terminals in several areas: in two clusters in the
respond to square pulses, provided that the pulse is longer than ELp, one near the soma and one several hundred micrometers
some threshold duration (long pass; Fig. 4A). The thresholds distant, in the medial ventral nucleus (MV), in the ipsilateral
of type I cells range from 0.02 to 0.2 ms. This response profile and contralateral isthmic granule nucleus (IG) and in the
is similar to that predicted for the ELa small cells, supporting subpreminential nucleus (SPE) (Fig. 4C). The terminals
the delay line-blanking model described above. Type II cells within the ELp imply the presence of complex local circuitry
respond to square pulses, provided that the pulse duration is and, indeed, intracellular recordings show several phases of
within a restricted range of durations (band-pass; Fig. 4B). excitation and inhibition (M. A. Friedman and C. D. Hopkins,
The best response varied between 0.1 and 10 ms for different unpublished results). The areas outside the ELp are unexplored
cells. The responses of type II cells could therefore be used by physiologically, so the role of the wide terminal field of
the fish to make decisions about the duration of a signallers individual ELp cells is not known, but it implies that the
EOD. Important physiological issues left to resolve are the information that the type I cell carries is significant.
complex effects of stimulus amplitude and geometry on these Anatomical studies have shown that the IG projects to the
responses (Amagai, 1998) and the mechanism that could valvula cerebelli (Finger et al., 1981) and that the MV projects
underlie discrimination of pulse durations as long as 10 ms, to the optic tectum (OT; Wulliman and Northcutt, 1990). The
since one based on axonal delay lines seems unlikely. OT (the homolog of the superior colliculus) plays a role in
Of these two physiological types, only the anatomy of type spatial analysis of signals from many sensory modalities in
I cells has been reconstructed, as shown in Fig. 4C (M. A. vertebrates (Bastian, 1982; Knudsen, 1982; Bartels et al., 1990;
Friedman and C. D. Hopkins, unpublished results). They have Stein and Meredith, 1993), suggesting that the MV may be
large (150 m), spiny dendritic arborizations that stretch involved in spatial analysis of knollenorgan information.
perpendicular to the thin, incoming small cell axons. This
morphology contrasts strongly with the anatomical
specializations described in the previous section for NELL and Comparative considerations
ELa cells. This change in cellular structure is correlated with When we compare the known connections of the
the increase in response jitter in the ELp, suggesting that the knollenorgan pathway with other octavolateral pathways, such
anatomical specializations evident in the NELL and the ELa as the ampullary/mormyromast system and the auditory
allow fine temporal discrimination to be made in the ELa, after system, several parallels become evident (Fig. 5). In all three
which spike times need not be so precise. The dendritic systems, there is a direct pathway from the receptor, through
arborizations in the ELp are more likely to function to integrate the hindbrain, to a distinct toral nucleus and finally to the optic
205
1316 M. XU-FRIEDMAN AND C. D. HOPKINS
Fig. 4. Physiology and anatomy of the nucleus exterolateralis pars posterior (ELp). (A) Normalized responses of type I cells, showing long-
pass tuning. (B) Normalized responses of type II cells, showing band-pass tuning. (C) Anatomical reconstruction of a type I cell and
terminals in other midbrain nuclei (outlined in bold). The cell was processed for biocytin staining and reconstructed from 26 sections of 50 m
thickness. Numbers next to bold outlines of other brain nuclei indicate their depths relative to the ELp soma. Terminals are found in the
subpreminential nucleus (SPE), the ipsi- and contralateral isthmic granule nucleus (IG) (enlarged in the lower right inset), the nucleus
medialis ventralis (MV) and the ELp, both close to the soma (enlarged in the lower left inset, in red) and in a distinct cluster 400 m ventral to
the soma. The cells have widely branching dendritic arborizations (lower left inset, in green). ELa, nucleus exterolateralis pars anterior; ll,
lateral lemniscus; mm, mesomesencephalic tract; OT, optic tectum; tp, toropreminential tract. A and B are modified from Amagai (1998).
206
Knollenorgan pathway in mormyrids 1317
OT OT OT computational task that the ELa is performing. In the avian
brainstem, it is two ears being compared, whereas in the ELa,
all parts of the body surface are being compared. In this sense,
the phase-coding pathways of South American electric fish are
ELa MV L MD
more similar to those of the mormyrid. In Eigenmannia, the
spherical cell afferent from the ELL ends on a giant cell and
ELp nearby small cells in layer VI of the torus semicircularis (Carr
et al., 1986; Heiligenberg, 1991). In this case, the giant cell
IG SPE PE IG VPE distributes the signal widely throughout layer VI, analogously
to the NELL axon in mormyrids. The major departure of the
mormyrid system is that it appears to use an inhibitory
mechanism in measuring time differences, whereas the barn
NELL ELL dzD owl and Eigenmannia use only excitatory synapses.
207
1318 M. XU-FRIEDMAN AND C. D. HOPKINS
Carr, C. E. (1986). Time coding in electric fish and barn owls. Brain Hopkins, C. D., Harned, G. D. and Schmid, U. (1993). Anatomical
Behav. Evol. 28, 122133. projections of the time coding pathway of the mormyrid electric
Carr, C. E. and Konishi, M. (1990). A circuit for detection of fish studied using fluorescent-labelled dextrans. Soc. Neurosci.
interaural time differences in the brain stem of the barn owl. J. Abstr. 19, 376.
Neurosci. 10, 32273246. Kawasaki, M. (1997). Sensory hyperacuity in the jamming avoidance
Carr, C. E., Maler, L. and Taylor, B. (1986). A time-comparison response of weakly electric fish. Curr. Opin. Neurobiol. 7,
circuit in the electric fish midbrain. II. Functional morphology. J. 473479.
Neurosci. 6, 13721383. Knudsen, E. I. (1982). Auditory and visual maps of space in the optic
Crawford, J. D. (1992). Individual and sex specificity in the electric tectum of the owl. J. Neurosci. 2, 11771194.
organ discharges of breeding mormyrid fish (Pollimyrus isidori). J. Konishi, M. (1991). Deciphering the brains codes. Neural Computat.
Exp. Biol. 164, 79102. 3, 118.
Enger, P. S., Libouban, S. and Szabo, T. (1976). Fast conducting Kozloski, J. and Crawford, J. D. (1998). Functional neuroanatomy
electrosensory pathway in the mormyrid fish, Gnathonemus of auditory pathways in the sound producing fish Pollimyrus. J.
petersii. Neurosci. Lett. 2, 133136. Comp. Neurol. (in press).
Finger, T. E., Bell, C. C. and Russell, C. J. (1981). Electrosensory Kramer, B. (1997). Electric organ discharges and their relation to sex
pathways to the valvula cerebelli in mormyrid fish. Exp. Brain Res. in mormyrid fishes. Naturwissenschaften 84, 119121.
42, 2333. Moller, P. and Szabo, T. (1981). Lesions in the nucleus
Friedman, M. A. and Hopkins, C. D. (1996). Tracking mormyrid mesencephali exterolateralis: Effects on electrocommunication in
electric fish in the field using individual differences in electric the mormyrid fish Gnathonemus petersii (Mormyriformes). J.
organ discharges. Anim. Behav. 51, 391407. Comp. Physiol. A 144, 327333.
Friedman, M. A. and Hopkins, C. D. (1998). Neural substrates for Mugnaini, E. and Maler, L. (1987a). Cytology and
species recognition in the time-coding electrosensory pathway of immunocytochemistry of the nucleus exterolateralis anterior of the
mormyrid electric fish. J. Neurosci. 18, 11711185. mormyrid brain: Possible role of GABAergic synapses in temporal
Friedman, M. A. and Kawasaki, M. (1997). Calretinin-like analysis. Anat. Embryol. 176, 313336.
immunoreactivity in mormyrid and gymnarchid electrosensory and Mugnaini, E. and Maler, L. (1987b). Cytology and
electromotor systems. J. Comp. Neurol. 387, 341357. immunocytochemistry of the nucleus of the lateral line lobe in the
Graff, C. and Kramer, B. (1992). Trained weakly-electric fishes electric fish Gnathonemus petersii (Mormyridae): Evidence
Pollimyrus isidori and Gnathonemus petersii (Mormyridae, suggesting that GABAergic synapses mediate an inhibitory
Teleostei) discriminate between waveforms of electric pulse corollary discharge. Synapse 1, 3256.
discharges. Ethology 90, 279292. Stein, B. E. and Meredith, M. A. (1993). The Merging of the Senses.
Hauged-Carr, F. (1979). The mesencephalic exterolateral Cambridge, MA: MIT Press.
posterior nucleus of the mormyrid fish Brienomyrus niger: Efferent Szabo, T. and Ravaille, M. (1976). Synaptic structure of the
connections studied by the HRP method. Brain Res. 178, 179184. lateral line lobe nucleus in mormyrid fish. Neurosci. Lett. 2,
Heiligenberg, W. (1991). Neural Nets in Electric Fish. Cambridge, 127131.
MA: MIT Press. Szabo, T., Ravaille, M., Libouban, S. and Enger, P. S. (1983). The
Hopkins, C. D. (1981). On the diversity of electric signals in a mormyrid rhombencephalon. I. Light and EM investigations on the
community of mormyrid electric fish in West Africa. Am. Zool. 21, structure and connections of the lateral line lobe nucleus with HRP
211222. labelling. Brain Res. 266, 119.
Hopkins, C. D. (1986a). Behavior of Mormyridae. In Wulliman, M. F. and Northcutt, R. G. (1990). Visual and
Electroreception (ed. T. H. Bullock and W. Heiligenberg), pp. electrosensory circuits of the diencephalon in mormyrids: An
527576. New York: John Wiley & Sons. evolutionary perspective. J. Comp. Neurol. 297, 537552.
Hopkins, C. D. (1986b). Temporal structure of non-propagated Young, S. R. and Rubel, E. W. (1983). Frequency-specific
electric communication signals. Brain Behav. Evol. 28, 4359. projections of individual neurons in chick brainstem auditory
Hopkins, C. D. (1999). Design features for electric communication. nuclei. J. Neurosci. 3, 13731378.
J. Exp. Biol. 202, 12171228. Zipser, B. and Bennett, M. V. L. (1976). Interaction of
Hopkins, C. D. and Bass, A. H. (1981). Temporal coding of species electrosensory and electromotor signals in lateral line lobe of a
recognition signals in an electric fish. Science 212, 8587. mormyrid fish. J. Neurophysiol. 39, 713721.
208
The Journal of Experimental Biology 207, 1073-1084 1073
Published by The Company of Biologists 2004
doi:10.1242/jeb.00851
Summary
Like all mormyrid fish, Brienomyrus brachyistius of the excitatory neurotransmitter L-glutamate (L-Glu)
produces an electric organ discharge (EOD) with a into DP led to acceleration-like output patterns, while in
constant waveform and variable sequence of pulse PCN it led to scallop-like output patterns. Iontophoresis
intervals (SPI). Periodic bursts fall into two display of the inhibitory neurotransmitter -amino-butyric acid
categories termed scallops and accelerations, with a (GABA) into DP and PCN led to an elongation of
third category termed rasps that appears to combine the intervals, as did iontophoresis of L-Glu into VP.
two. The medullary EOD command nucleus (CN) receives Iontophoresis of the GABAA receptor blocker bicuculline
excitatory input from the midbrain precommand nucleus methiodide (BMI) into DP and PCN induced repetitive
(PCN) and the thalamic dorsal posterior nucleus (DP), bursting behavior and eliminated differences in the effects
both of which are regulated by a recurrent inhibitory of L-Glu iontophoresis in the two nuclei. These results
projection from the ventroposterior nucleus of the torus support our three hypotheses, suggesting that production
semicircularis (VP). We tested the following hypotheses: of different communication behaviors may be regulated
(1) PCN and DP are responsible for generating different by spatially distinct groups of neurons, and recurrent
burst types (scallops and accelerations, respectively), (2) inhibition and disinhibition may play an active role in
differences in the strength of recurrent inhibition are driving and shaping such behaviors.
related to physiological differences between PCN and
DP and (3) recurrent inhibition regulates the resting Key words: mormyrid, electric fish, Brienomyrus brachyistius,
electromotor rhythm, while disinhibition releases PCN electric organ discharge, electromotor, central pattern generator,
and DP, allowing them to generate bursts. Iontophoresis pacemaker, disinhibition, iontophoresis.
Introduction
The electric signaling behavior of mormyrid fish is A recent quantitative analysis of bursts in Brienomyrus
characterized by two components: the waveform of each brachyistius has revealed three modal display categories based
electric organ discharge (EOD) and the sequence of pulse on variation in the temporal patterning of EOD output (Fig.1B;
intervals (SPI). While the EOD is constant and signals the Carlson and Hopkins, unpublished observations). Scallops
senders identity (Carlson et al., 2000; Freedman et al., 1989; are stereotyped pulse sequences in which intervals suddenly
Friedman and Hopkins, 1996; Hopkins, 1981), the SPI is drop to 1020ms and then immediately return to baseline
variable and appears to play a role in signaling motivation or intervals of 100300ms. Accelerations are graded decreases
behavioral state (Carlson, 2002a; Hopkins, 1986; Kramer, in interval, typically to values of 2060ms. Accelerations are
1993). The SPI is characterized by a relatively slow baseline less stereotyped than scallops, and minimum intervals for
rhythm, with EOD intervals typically ranging from accelerations may be maintained over several EOD cycles with
approximately 100ms to 300ms (Carlson, 2002a; Teyssedre a high degree of regularity. Subjectively, rasps appear to
and Boudinot, 1987). This baseline rhythm may be periodically combine an initial scallop-like onset with an acceleration-like
interrupted by a variety of bursts and cessations in the termination (Fig.1B), which is supported by the quantitative
discharge (Carlson, 2002a; Hopkins, 1986). Such displays characteristics of the three displays (Carlson and Hopkins,
occur in specific behavioral contexts such as courtship and unpublished observations). Thus, rasps in this species probably
aggression, suggesting that they play an important role in result from a combination of two distinct displays.
social behavior (Bell et al., 1974; Bratton and Kramer, 1989; Recent anatomical and physiological studies on the
Kramer, 1974, 1976; Kramer and Bauer, 1976; Moller et al., electromotor system of mormyrids have suggested a closed-
1989; Scheffel and Kramer, 1997, 2000). loop circuit that may function as a relatively simple central
209
1074 B. A. Carlson and C. D. Hopkins
pattern generator (CPG) for regulating electromotor output Emde et al., 2000), apparently via a projection from the dorsal
(Carlson, 2002b, 2003; von der Emde et al., 2000). Fig.1A subdivision of the ventroposterior nucleus (VPd) in the torus
illustrates the functional connectivity of this network. Each semicircularis (Bell et al., 1983; Carlson, 2002b, 2003; Carlson
EOD is initiated in the medullary command nucleus (CN), and Hopkins, 2001). This recurrent inhibition probably provides
which activates spinal electromotor neurons (EMNs) indirectly a rate-limiting factor to the activity of DP and PCN neurons that
through a projection to the medullary relay nucleus (MRN; Bell may be responsible for producing rhythmic resting electromotor
et al., 1983; Grant et al., 1986). CN integrates excitatory input output (Carlson, 2003; von der Emde et al., 2000). A few large
from two distinct nuclei, the precommand nucleus (PCN) in the neurons at the ventral edge of VP also project to DP, PCN and
mesencephalon, and the dorsal posterior nucleus (DP) in the CN (Bell et al., 1983; Carlson, 2002b), although the functional
thalamus (Bell et al., 1983; Carlson, 2002b, 2003; von der Emde role of these neurons has not yet been explored.
et al., 2000). DP and PCN both receive inhibitory feedback DP and PCN neurons in B. brachyistius show a wide
from the electric organ corollary discharge pathway (von der diversity of firing patterns, and correlations between single unit
activity and burst production suggest that
distinct neuronal populations are responsible for
A C3
generating scallops and accelerations (Carlson,
2003). In distantly related gymnotiform electric
Val fish, the central posterior and prepacemaker
EGp
TM nuclei appear analogous to DP and PCN
EL (Carlson, 2002b), and the two are responsible
for generating distinct electrical behaviors
MCA Tel (Metzner, 1999). Based on these two lines of
ELL evidence, we hypothesized that DP and PCN are
DP
OB likewise responsible for driving different
VPd PCN
electrical behaviors in B. brachyistius.
BCA MRN Preliminary experiments using extracellular
CN electrical stimulation support this hypothesis,
IL suggesting that accelerations are generated
To 1 mm by DP, while scallops are generated by
EMN
PCN (Carlson and Hopkins, unpublished
observations). We tested this hypothesis using
B
Scallop Acceleration Rasp
iontophoresis of the excitatory neurotransmitter
250 250 250 L-glutamate (L-Glu) to stimulate DP and PCN
EOD interval (ms)
210
Stereotyped temporal pattern generation in mormyrids 1075
inhibitory input. Second, we used iontophoresis of L-Glu in VP Novato, CA, USA) and broken to a composite diameter of
to test whether this also causes increases in EOD interval. approximately 10m, resulting in individual barrel diameters
Finally, we used iontophoresis of the GABAA receptor blocker of ~23m. Each barrel was filled with one of the following
bicuculline methiodide (BMI) in DP and PCN to block solutions: (1) 3moll1 NaCl for recording local field
inhibitory input and determine whether eliminating recurrent potentials; (2) 2% alcian blue (Sigma Chemical Co.) in
inhibition drives decreases in EOD interval. Walpole acetate buffer (pH=4.0) for marking electrode
Differences in the effects of DP and PCN on the SPI are locations; (3) 0.1moll1 L-Glu (pH=8.0, adjusted with NaOH)
likely to be caused by differences in their physiology, which for excitatory iontophoresis; (4) 0.5moll1 GABA (pH=3.5,
may in turn relate to differences in the strength of recurrent adjusted with HCl) for inhibitory iontophoresis or (5)
inhibition from VPd neurons (Carlson, 2003). To test this 20mmoll1 BMI in 165mmoll1 NaCl (pH=3.2) for blocking
hypothesis, we compared the effects of L-Glu iontophoresis GABAA receptors.
in DP and PCN before and after BMI iontophoresis. If the Electromotor output was monitored by placing a silver
observed differences resulting from stimulating the two wire against the caudal peduncle with a reference several
nuclei with L-Glu are due to variation in inhibitory feedback, centimeters away. Although the electric organ is silenced by
then blocking this inhibition should eliminate these flaxedil, the EOD command can be recorded as a three-spike
differences. potential resulting from the synchronous activation of EMN
(Bennett et al., 1967). The first negative peak in the EMN
volley was defined as the reference time for EOD output (t0),
Materials and methods which in a natural situation precedes the EOD by 45ms. At
Animals the start of each experiment, either DP, PCN or VP was
We used a total of 27 Brienomyrus brachyistius (Gill 1862), localized initially through landmarks on the dorsal surface of
ranging in size from 8.0g to 51.0g in body mass and 7.9cm the brain and then more precisely by recording characteristic
to 18.2cm in total length. Fish were either wild-caught field potentials that were phase-locked to the EMN volley (see
or laboratory-bred. They were housed in 280-liter group Carlson, 2002b). Field potentials and EMN output were band-
aquaria at a temperature of 2527C and conductivity of pass filtered from 10Hz to 5000Hz, amplified 10000 on a
150200Scm1 on a 12h:12h light:dark cycle and fed live differential AC amplifier (A-M Systems, Inc., Everett, WA,
black worms daily. All procedures were in accordance with the USA; model 1700) and monitored on a digital oscilloscope
guidelines established by the National Institutes of Health and (Tektronix, Inc., Beaverton, OR, USA; model 5223).
were approved by the Cornell University Institutional Animal Iontophoretic currents were provided by a separate amplifier
Care and Use Committee. (A-M Systems, Inc.; Neuroprobe model 1600).
After locating a given nucleus, the horizontal position and
Surgery depth of the electrode were adjusted using a microelectrode
Surgical procedures were identical to those described drive (Burleigh Instruments, Inc., Fishers, NY, USA;
previously (Carlson, 2002b, 2003). Animals were anesthetized Inchworm 6000) that was held by a micromanipulator
in a solution of 500mgl1 tricaine methanesulfonate (MS-222; (Newport Co., Fountain Valley, CA, USA; model 462-XY-M).
Sigma Chemical Co., St Louis, MO, USA) and then respirated The position of the electrode was adjusted in 50m steps
under a solution of 160mgl1 MS-222 during the surgery. Fish in all three dimensions and the location where pulsed
were placed on a horizontal platform with lateral supports and iontophoresis of L-Glu (0.5A, 500ms pulses at 0.25Hz)
completely immersed in aquarium water except for the dorsal resulted in the strongest modulation in the SPI was used for all
surface of the head. A flap of skin was removed from the head subsequent iontophoretic injection experiments in that nucleus
and the underlying tissue was scraped away to expose the dorsal (Fig.2).
surface of the skull. Lidocaine (100200l of a 2% solution; For all experiments, the EMN signal was sent to a Schmitt
Radix Laboratories, Inc., Eau Claire, WI, USA) was used as a Trigger, which was output to an event timer that recorded the
local anesthetic. A metal post was affixed to the skull using time of t0 using a clock rate of 1MHz (Tucker-Davis
superglue, and a small rectangular portion of the skull and Technologies, Alachua, FL, USA; model ET1). Data on EOD
meninges was removed to expose the dorsal surface of the times of occurrence were saved using custom-made software.
midbrain and caudal forebrain. A reference electrode was then For experiments involving L-Glu or GABA iontophoresis, 20s
placed in the dorsal musculature at the posterior end of the skull. of data were recorded before iontophoresis, followed by 20s
The fish were then immobilized and electrically silenced with of iontophoresis (1.0A for L-Glu, +1.0A for GABA) and
an intramuscular injection of flaxedil (gallamine triethiodide; an additional 20s of recording. In most cases, opposite polarity
100300l of a 3mgml1 solution; Sigma Chemical Co.), and current (+1.0A for L-Glu, 1.0A for GABA) was tested as
the respiration was switched to freshwater for recovery. a control, and this resulted in no observable modulation in the
SPI. For experiments involving BMI iontophoresis, 1min of
Experimental procedure data were recorded before iontophoresis, followed by 4min of
Triple-barrel electrodes were pulled using a Sutter Flaming iontophoresis (+100nA), followed by 1min of recovery. The
Brown Micropipette Puller model P-87 (Sutter Instrument Co., longer duration of BMI iontophoresis compared with L-Glu
211
1076 B. A. Carlson and C. D. Hopkins
and GABA iontophoresis was chosen based on results from basis for this difference is unclear but is probably related to
previous studies in other systems in which the effects of BMI differences in the pharmacological effects of a receptor blocker
iontophoresis occurred after relatively long latencies and compared with naturally occurring neurotransmitters. In some
persisted for several minutes after termination (Fujita and experiments, the effects of L-Glu iontophoresis in DP and PCN
Konishi, 1991; Heiligenberg et al., 1996). The physiological before and after BMI iontophoresis were determined. The
A PCN tc L tc L
+100
Dorsal
PGc
PGc +50
0 m
PGm PGm
Ventral
50
nLR IL nLR IL
Hyp Hyp 100
B pc pc 2s
v
v FR +100
FR
Dorsal
DP
+50
0 m
Ventral 50
CP
CP 100
2s
C ll
L L ll +100
MV
Dorsal
MV
VP +50
tt tt 0 m
Ventral
50
nLR
nLR 100
IL IL
2s
L-Glu
Fig.2. Localization of iontophoresis sites. (A) The precommand nucleus (PCN). (B) The dorsal posterior nucleus (DP). (C) The ventroposterior
nucleus (VP). The first column shows photomicrographs of transverse sections with retrogradely labeled neurons against a background of
cresyl violet counterstain, taken from the anatomical study by Carlson (2002b). In A and B, the label results from an injection of neurobiotin
into the command nucleus (CN), while in C it results from an injection of neurobiotin into PCN. The second column shows photomicrographs
of transverse sections with alcian blue staining against a background of neutral red counterstain, at the same approximate rostro-caudal
locations as the first column. The small blue dots provide a precise marker of electrode location, which was always accurately placed into one
of the three nuclei. In the third column, continuous voltage traces of electromotor neuron (EMN) activity are shown, with each spike
corresponding to a single EMN volley. Iontophoretic injections of L-glutamate (L-Glu; 500ms pulses of 500nA) occurred during the times
represented by horizontal green lines below each series of traces. Each example is taken from the site shown in the second column, with 0m
corresponding to the exact location of the alcian blue marker. The effects of L-Glu iontophoresis at 50m and 100m dorsal and ventral to
these sites are also shown. CP, central posterior nucleus of the thalamus; FR, fasciculus retroflexus; Hyp, hypothalamus; IL, inferior lobe of the
hypothalamus; L, lateral nucleus of the torus semicircularis; ll, lateral lemniscus; MV, medioventral nucleus of the torus semicircularis; nLR,
nucleus of the lateral recess; pc, posterior commissure; PGc, caudal subdivision of the preglomerular nucleus; PGm, medial subdivision of the
preglomerular nucleus; tc, tectocerebellar tract; tt, toro-praeeminential tract; v, ventricle. Scale bars, 200m in A, 50m in B and 200m in C.
212
Stereotyped temporal pattern generation in mormyrids 1077
procedure for L-Glu iontophoresis in these cases was identical regions. In general, L-Glu iontophoresis only led to changes in
to the normal L-Glu iontophoresis procedure, and L-Glu EMN interval within a range of depths of 50150m from the
iontophoresis was always performed within 2min after alcian blue mark (Fig.2).
terminating BMI iontophoresis. Statistica 6.1 (StatSoft, Inc., In three different fish, doseresponse curves were
Tulsa, OK, USA) was used for all statistical analyses of data constructed by measuring the effects of L-Glu iontophoresis
on the SPI. on median EMN intervals with varying levels of current
magnitude (from 100nA to 900nA in steps of 200nA) in
Histology all three nuclei (Fig.3A). In both DP and PCN, increasing
At the end of each experiment, we marked the location of levels of current led to greater shortening of EMN intervals,
the electrode by iontophoretic injection of alcian blue, using a with the response beginning to saturate at approximately
500ms, 150V pulse (Grass Medical Instruments, Quincy, 500nA and showing complete saturation at 700nA to
MA, USA; model S88 stimulator). After completing the 900nA (Fig.3B). Similarly, in VP, increasing levels of
experiments, fish were placed back under general anesthesia current led to a greater elongation of EMN intervals, with the
(160mgl1 MS-222) and then perfused transcardially with response saturating at 700nA to 900nA (Fig.3B). Thus, for
Hickmans ringer solution (6.48gl1 NaCl, 0.15gl1 KCl, all experiments using L-Glu iontophoresis, we used current
0.29gl1 CaCl2, 0.12gl1 MgSO4, 0.084gl1 NaHCO3, magnitudes of 1.0A, which was well above the level of
0.06gl1 NaH2PO4) followed by ice-cold 4% saturation for all three nuclei and therefore provided maximal
paraformaldehyde/1% glutaraldehyde in 0.1moll1 phosphate stimulation of each nucleus.
buffer (PB; pH=7.2) for fixation. The brains were removed and 20s injections of L-Glu into the three nuclei led to
postfixed overnight and then transferred to 0.1moll1 PB for characteristic modulations in the SPI (Fig.4). In both DP and
storage. Brains were transferred to a solution of 30% sucrose PCN, there was a marked, maintained decrease in EMN
in 0.1moll1 PB on the night prior to sectioning. Transverse interval that persisted throughout the duration of the stimulus,
sections were cut on a freezing microtome at 50m, mounted while in VP there was a complete cessation of activity for the
on chrom-alum-subbed slides, counterstained with neutral red, whole period of stimulation and usually for many additional
dehydrated in a graded alcohol series and coverslipped with seconds after terminating the current. There was a highly
Permount (Sigma Chemical Co.). significant decrease in EMN intervals during L-Glu
iontophoresis in DP and PCN and a highly significant increase
in EMN intervals during L-Glu iontophoresis in VP (Table1).
Results Although stimulation of both DP and PCN led to a
Glutamate iontophoresis shortening of EMN intervals, the responses of the two
Brief pulses of L-Glu in DP (N=25 fish) and PCN (N=24 fish) nuclei were typically quite different. Stimulation of DP
led to a shortening of EMN intervals, while in VP (N=20 fish) typically resulted in a smooth decrease in interval to values
it typically led to complete cessations (Fig.2). Alcian blue of ~2050 ms that were maintained throughout the period of
marking of the locations where iontophoresis caused the stimulation with a high degree of regularity (Figs4,5).
greatest response led to very restricted, bright blue dots in the Stimulation of PCN also led to a decrease in baseline
center of DP or PCN or in the region around VP (Fig.2), intervals, but this baseline was typically not as low or regular
verifying that the electrodes were located in the desired as during DP stimulation (Table1) and was punctuated by the
213
1078 B. A. Carlson and C. D. Hopkins
DP PCN VP
500
400
300
200
100
0
500
EMN interval (ms)
400
300
200
100
0
500
400
300
200
100
0
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Time (s)
Fig.4. Three representative examples of the effects of L-glutamate (L-Glu) iontophoresis on electromotor neuron (EMN) output in the dorsal
posterior (DP), precommand (PCN) and ventroposterior nuclei (VP). Each row corresponds to a single fish. The first column shows the effects
of L-Glu iontophoresis in DP, the second column shows the effects of L-Glu iontophoresis in PCN, and the third column shows the effects of L-
Glu iontophoresis in VP. In each case, iontophoretic currents consisted of 1A for 20s, which is indicated by the horizontal bar beneath each
column.
214
Stereotyped temporal pattern generation in mormyrids 1079
Table 1. Changes in median EMN interval in response to iontophoresis of L-glutamate (L-Glu) for 20s using a current magnitude
of 1.0A in the dorsal posterior (DP), precommand (PCN) and ventroposterior (VP) nuclei
Median EMN interval (mean S.E.M.)
Nucleus (N) Pre-stimulation L-Glu stimulation Post-stimulation F P
DP (25) 381.1648.868 37.6602.6131 420.6762.789 29.617 <0.000001
PCN (24) 296.7032.242 67.2295.9719 339.8148.942 31.032 <0.000001
VP (20) 298.5636.302 213713744.8 275.4131.000 31.678 <0.000001
F-statistics and P-values from a repeated measures analysis of variance (ANOVA) for each nucleus are shown. N is number of fish.
215
1080 B. A. Carlson and C. D. Hopkins
DP PCN VP
1000
800
600
400
200
0
1000
EMN interval (ms)
800
600
400
200
0
1000
800
600
400
200
0
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Time (s)
Fig.7. Three representative examples of the effects of -amino-butyric acid (GABA) iontophoresis on electromotor neuron (EMN) output in the
dorsal posterior (DP), precommand (PCN) and ventroposterior nuclei (VP). Each row corresponds to a single fish. In each case, iontophoretic
currents consisted of +1A for 20s, which is indicated by the horizontal bar beneath each column.
Glu stimulation in PCN (von der Emde et al., 2000) and Neurons in PCN appear to be surrounded by GABAergic
recordings of single unit activity in DP and PCN (Carlson, inhibitory terminals (Niso et al., 1989), and single units in both
2003; von der Emde et al., 2000). Furthermore, electromotor DP and PCN receive recurrent inhibitory feedback via the
output resulting from stimulating the two nuclei was corollary discharge pathway (Carlson, 2003; von der Emde et
significantly different. DP stimulation resulted in smooth, al., 2000). Anatomically, DP and PCN receive a dense
maintained accelerations to intervals of 2050ms. PCN projection from VPd, which in turn receives input from the
stimulation led to a more modest decrease in baseline intervals corollary discharge pathway (Carlson, 2002b). These three
that was interrupted by the repeated production of transient, lines of evidence suggest that VPd provides recurrent,
scallop-like bursts. This supports the hypothesis that these two GABAergic inhibition to DP and PCN. The results of the
nuclei are responsible for generating distinct communication current study support this hypothesis: iontophoresis of GABA
displays. Recent studies of single unit activity in DP and PCN into DP and PCN induces a significant elongation of
have shown that units with low baseline firing rates experience EOD intervals, as does stimulation of VPd using L-Glu
an increase in activity during accelerations, while units with iontophoresis. In most cases, stimulation of VPd led to
high baseline firing rates experience an increase in activity complete cessations in the discharge, suggesting that cessations
during scallops (Carlson, 2003). In light of the findings of the may result from increases in the activity of VPd neurons and
current study, this suggests that neurons within DP may therefore stronger recurrent inhibition.
correspond to the former, while neurons within PCN may We hypothesized that this recurrent inhibition is responsible
correspond to the latter. for regulating DP and PCN activity, and thereby maintaining
Table 2. Changes in median EMN interval in response to iontophoresis of -amino-butyric acid (GABA) for 20s using a current
magnitude of +1.0A in the dorsal posterior (DP), precommand (PCN) and ventroposterior (VP) nuclei
Median EMN interval (mean S.E.M.)
Nucleus (N) Pre-stimulation GABA stimulation Post-stimulation F P
DP (15) 260.2439.315 318.4242.941 223.0236.025 15.566 <0.00003
PCN (17) 284.6744.115 448.0471.982 255.1240.670 17.185 <0.00001
VP (14) 237.8235.102 243.6030.572 246.9436.056 0.8172 >0.45
F-statistics and P-values from a repeated measures analysis of variance (ANOVA) for each nucleus are shown. N is number of fish.
216
Stereotyped temporal pattern generation in mormyrids 1081
200
100
0
500
DP
400
300
200
100
0
0 10 20 30 40 50 60 0 50 100 150 200 250 300 350 0 10 20 30 40 50 60
Time (s)
Number of bursts
Scallop Scallop Scallop
0 0 0
0 3 0 3 0 3
250 250 250 40
20
DP
Fig.8. (A) One example each of the effects of bicuculline methiodide (BMI) iontophoresis in the dorsal posterior (DP) and precommand nuclei
(PCN), and the effects of L-glutamate (L-Glu) iontophoresis immediately before and immediately after BMI iontophoresis. L-Glu iontophoretic
currents consisted of 1A for 20s, while BMI iontophoretic currents consisted of +100nA for 4min, each of which is shown as a horizontal
bar beneath each column. (B) Excerpts from A of examples of different burst types occurring during BMI iontophoresis in DP and PCN.
(C) Number of different burst types induced by BMI iontophoresis in DP and PCN. Values shown are means S.E.M. There were no significant
differences in the numbers of any of the three burst types.
the irregular, baseline rhythm of 100300ms EOD intervals. neurons, which produce a stereotyped burst of action potentials
During burst displays, the activity of VPd neurons decreases starting within a few milliseconds of EOD production
(Carlson, 2003), so we hypothesized that disinhibition plays a (Carlson, 2003; von der Emde et al., 2000). Across neurons,
role in disrupting the baseline rhythm by releasing DP and there is wide variation in the duration of these bursts. If
PCN from their normal rate-limiting factor, thereby allowing different subsets of VPd neurons project to DP and PCN, this
them to drive burst displays. In support of this hypothesis, variation could lead to differences in the baseline activity of
application of the GABAA receptor blocker BMI to DP and DP and PCN neurons and therefore different effects on
PCN caused the SPI to go from a resting rhythm to repetitive electromotor output when stimulated. In support of this
bursting. It is unclear how the activity of VPd neurons is hypothesis, the effects of L-Glu iontophoresis in the two nuclei
modulated in a natural situation, although the tectum were identical following BMI iontophoresis. Furthermore, if
mesencephali has a strong projection to VPd (Carlson, 2002b; recurrent inhibition to DP and PCN is separated into different
Wullimann and Northcutt, 1990) and retrograde labeling pathways, this may provide a means for differentially
suggests that VPd may also receive inputs from hypothalamic activating the two nuclei and generating distinct behaviors.
and preoptic areas (Carlson, 2002b). Application of GABA to Although we suggest that disinhibition plays a role in
VP did not elicit any changes in EMN activity, suggesting that driving burst displays in a natural situation, we were able to
a reduction in VPd activity is mediated by a different elicit such displays solely through the application of L-Glu to
neurotransmitter or through neuromodulatory inputs. DP and PCN, which presumably did not affect recurrent
One potential source of physiological differences between inhibition. Most likely, the increased excitatory input caused
DP and PCN may be variation in recurrent inhibition from VPd by L-Glu application counterbalanced the ongoing recurrent
217
1082 B. A. Carlson and C. D. Hopkins
2 (Bullock et al., 1983). EOD production in both groups of fish
interval (ms) A DP (N=7) is controlled by a ventral midline nucleus in the medulla (CN
CV in EMN
B
interval (ms)
C
interval (ms)
800
intrinsic dendrites.
Stimulating CP in gymnotiforms leads to rises (smooth,
400 graded increases in frequency), while stimulating PPN leads to
* chirps (transient, intense bursts; Metzner, 1999). This is
7 strikingly similar to electromotor output patterns induced by
Pre-BMI Post-BMI
stimulation in DP and PCN, respectively, with the former
Fig.9. The effects of L-glutamate (L-Glu) iontophoresis in the dorsal driving accelerations (smooth, graded increases in frequency)
posterior (DP) and precommand nuclei (PCN) before and after and the latter driving scallops (transient, intense bursts). Thus,
bicuculline methiodide (BMI) iontophoresis on the coefficient of convergence in anatomical substrates appears to be directly
variation (CV) in electromotor neuron (EMN) interval (A), the linked to convergence in the behaviors they control. In
minimum EMN interval (B), and the maximum EMN interval (C). gymnotiforms, the different electrical behaviors resulting from
Values shown are means S.E.M. Asterisks represent statistically activation of CP and PPN are related to differences in the
significant differences (MannWhitney U test; P<0.05).
location of synapses in PN and differences in the glutamate
receptor subtypes found at these synapses (Metzner, 1999). It
inhibition, resulting in increases in EMN activity. The different is possible that similar differences play a role in the differential
effects of L-Glu stimulation in DP and PCN probably relate to effects of DP and PCN on electromotor behavior in mormyrids,
the different strengths of recurrent inhibition that counteract but the findings of the current study suggest that these
the stimulatory effects of L-Glu to varying degrees. Application differences may be due solely to the effects of variation in
of BMI to these nuclei, by contrast, leads to a maintained recurrent inhibition.
blockage of recurrent inhibition but no excitatory input. Not
surprisingly, this also leads to increases in EMN activity. Diversity in mormyrid electric signaling behavior
Unlike stimulation with L-Glu, however, there were no Every species of mormyrid that has been studied produces
observable differences in the output patterns caused by BMI acceleration-like displays that appear to play a role in
stimulation in DP and PCN, which can be explained by the fact aggression, but there is wide diversity across species in the
that BMI application also removed the source of physiological other types of displays that may be produced (Carlson, 2002a).
differences between the two nuclei, which was not the case In Gnathonemus petersii, agonistic encounters are often
with L-Glu stimulation. In a natural situation, disinhibition is accompanied by repetitive pulse pairs, with EOD intervals
typified by a modest, temporary reduction in inhibitory input alternating between 1516ms and 89ms (Bauer, 1972; Bell
from the baseline level rather than a complete, maintained et al., 1974). Such displays have never been observed in B.
removal (Carlson, 2003). Unlike BMI application, this would brachyistius or any other species (Carlson, 2002a), although
not eliminate differences between DP and PCN but would the number of species studied is relatively small. By contrast,
provide a transient excitatory effect similar to the effects of scallops have never been described for G. petersii, although
L-Glu stimulation, causing the two nuclei to drive different they have been described for B. niger (Serrier and Moller,
behaviors. 1989). This suggests the hypothesis that pulse pairs may be
driven by PCN in species that do not produce scallops.
Convergence in the central control of electromotor behavior There is also wide diversity in the characteristics of certain
Gymnotiform electric fish from South America have displays between species. For example, scallops in B. niger and
electromotor and electrosensory systems that share many B. brachyistius are quite similar in their basic structure but
striking similarities with those of mormyrids (Hopkins, differ in minimum interval (Carlson, 2002a; Carlson and
1995), despite overwhelming evidence that their electrogenic Hopkins, unpublished observations; Serrier and Moller, 1989).
and electrosensory capabilities have evolved independently Such differences are possibly related to differences in the
218
Stereotyped temporal pattern generation in mormyrids 1083
morphology and physiology of PCN neurons. Field recordings electric organ (Bennett, 1971), while the latter is determined
from various Brienomyrus species in Gabon reveal the by patterns of activity in CN (Grant et al., 1986). Thus,
production of rasps that differ dramatically from those in B. unraveling the mechanisms involved in regulating the SPI
brachyistius and do not appear to result from combining a breaks down to a problem of understanding the generation of
scallop and an acceleration (Hopkins, 1983; Hopkins and Bass, spike times in CN. As this and other recent studies have shown
1981). Scallops have not been described for these species, so (Carlson, 2002b, 2003; von der Emde et al., 2000), this relative
it is possible that PCN controls rasp production in this group. simplicity makes the mormyrid electromotor network an
The rich diversity of mormyrids and the signals they produce excellent model system for studying the mechanisms of
provide a rare opportunity for studying the evolution of neural generating stereotyped temporal patterns in vertebrate
circuits that govern communication behavior. communication. The many similarities between this system
and the gymnotiform electromotor network, as well as with
Motor networks and behavior CPGs in general, suggest that insights gained into the
Much of our understanding of the mechanisms underlying functioning of this network are likely to be instructive towards
stereotyped motor output comes from relatively simple general issues in the motor control of behavior.
networks involved in rhythmic behaviors such as locomotion,
digestion, respiration and heartbeat (Marder and Bucher, This research was supported in part by grants from the
2001). The stereotyped, rhythmic output of these central National Institute of Mental Health (Grant MH37972) and
pattern generators (CPGs) results from a combination of the National Science Foundation (0108372). B.A.C. was
several cellular and molecular specializations, many of which supported by a National Science Foundation Predoctoral
are shared across different networks. The findings of the Fellowship and a National Institute of Mental Health
current study, as well as other recent studies (Carlson, 2002b, Predoctoral Training Grant (MH15793). Thanks to M.
2003; von der Emde et al., 2000), reveal several similarities Kawasaki for advice on the alcian blue staining procedure.
between these networks and the mormyrid electromotor Thanks to A. H. Bass and two anonymous reviewers for their
system. For instance, recurrent inhibition plays an important helpful comments on earlier versions of the manuscript.
role in establishing rhythmic motor output in many different
networks (Friesen and Stent, 1978) as well as in regulating
electromotor output in mormyrids. Similarly, disinhibition References
Apfelbach, R. (1972). Electrically elicited vocalizations in the gibbon
serves a permissive function in activating stereotyped motor Hylobates lar (Hylobatidae), and their behavioral significance. Zeit.
output in several motor systems (Faumont et al., 1998; Noga Tierpsychol. 30, 420-430.
et al., 1988; Wang and Bieger, 1991), which also seems to be Bauer, R. (1972). High electrical discharge frequency during aggressive
behavior in a mormyrid fish (Gnathonemus petersii). Experientia 28, 669-
the case for generating burst displays in mormyrids. 670.
Extracellular stimulation of restricted brain regions can Bell, C. C., Libouban, S. and Szabo, T. (1983). Pathways of the electric organ
elicit the production of semi-natural, species-specific discharge command and its corollary discharges in mormyrid fish. J. Comp.
Neurol. 216, 327-338.
communication signals in a variety of vertebrate species Bell, C. C., Myers, J. P. and Russell, C. J. (1974). Electric organ discharge
(Apfelbach, 1972; Demski and Gerald, 1972; Fine and Perini, patterns during dominance-related behavioral displays in Gnathonemus
1994; Fu and Brudzynski, 1994; Goodson and Bass, 2000; petersii (Mormyridae). J. Comp. Physiol. 92, 201-228.
Bennett, M. V. L. (1971). Electric organs. In Fish Physiology (ed. W. S. Hoar
Jrgens and Richter, 1986; Phillips and Youngren, 1973; and D. J. Randall), pp. 347-491. London: Academic Press.
Schmidt, 1966; Schuller and Radtkeschuller, 1990; Seller and Bennett, M. V. L., Pappas, G., Aljure, E. and Nakajima, Y. (1967).
Armitage, 1983; Valentine et al., 2002; Williams and Vicario, Physiology and ultrastructure of electrotonic junctions. II. Spinal and
medullary electromotor nuclei in mormyrid fish. J. Neurophysiol. 30, 180-
1993). While it is clear that stereotyped signal production may 208.
be controlled by spatially distinct groups of neurons, there is Bratton, B. O. and Kramer, B. (1989). Patterns of the electric organ
often insufficient information about anatomical circuitry and discharge during courtship and spawning in the mormyrid fish, Pollimyrus
isidori. Behav. Ecol. Sociobiol. 24, 349-368.
how it relates to patterns of neuronal activity to gain insight Bullock, T. H., Bodznick, D. A. and Northcutt, R. G. (1983). The
into the network and cellular mechanisms involved in signal phylogenetic distribution of electroreception: evidence for convergent
generation. Part of the reason for this is that many vertebrate evolution of a primitive vertebrate sense modality. Brain Res. Rev. 6, 25-46.
Carlson, B. A. (2002a). Electric signaling behavior and the mechanisms of
communication signals are relatively complex, involving electric organ discharge production in mormyrid fish. J. Physiol. Paris 96,
several features that vary semi-independently over time. As a 405-419.
result, the neural substrates underlying the generation of these Carlson, B. A. (2002b). Neuroanatomy of the mormyrid electromotor control
system. J. Comp. Neurol. 454, 440-455.
signals are similarly complex, making it difficult to formulate Carlson, B. A. (2003). Single-unit activity patterns in nuclei that control the
hypotheses that directly link the activity patterns of individual electromotor command nucleus during spontaneous electric signal
neurons to specific signal characteristics. production in the mormyrid Brienomyrus brachyistius. J. Neurosci. 23,
10128-10136.
By contrast, electric signaling behavior in mormyrids is Carlson, B. A., Hopkins, C. D. and Thomas, P. (2000). Androgen correlates
relatively simple, consisting of two distinct components: a of socially induced changes in the electric organ discharge waveform of a
stereotyped EOD waveform and a variable pattern of EOD mormyrid fish. Horm. Behav. 38, 177-186.
Demski, L. and Gerald, J. (1972). Sound production evoked by electrical
production (the SPI). The characteristics of the former are stimulation of the brain in toadfish (Opsanus beta). Anim. Behav. 20, 507-
controlled by the morphophysiological characteristics of the 513.
219
1084 B. A. Carlson and C. D. Hopkins
Dye, J. C. and Meyer, J. H. (1986). Central control of the electric organ Kramer, B. and Bauer, R. (1976). Agonistic behavior and electric signaling
discharge in weakly electric fish. In Electroreception (ed. T. H. Bullock and in a mormyrid fish, Gnathonemus petersii. Behav. Ecol. Sociobiol. 1, 45-
W. Heiligenberg), pp. 71-102. New York: John Wiley & Sons. 61.
Faumont, S., Simmers, J. and Meyrand, P. (1998). Activation of a lobster Marder, E. and Bucher, D. (2001). Central pattern generators and the control
motor rhythm-generating network by disinhibition of permissive of rhythmic movements. Curr. Biol. 11, R986-R996.
modulatory inputs. J. Neurophysiol. 80, 2776-2780. Metzner, W. (1999). Neural circuitry for communication and jamming
Fine, M. L. and Perini, M. A. (1994). Sound production evoked by electrical avoidance in gymnotiform electric fish. J. Exp. Biol. 202, 1365-1375.
stimulation of the forebrain in the oyster toadfish. J. Comp. Physiol. A 174, Moller, P., Serrier, J. and Bowling, D. (1989). Electric organ dishcharge
173-185. displays during social encounter in the weakly electric fish Brienomyrus
Freedman, E. G., Olyarchuk, J., Marchaterre, M. A. and Bass, A. H. (1989). niger L. (Mormyridae). Ethology 82, 177-191.
A temporal analysis of testosterone-induced changes in electric organs and Niso, R., Serrier, J. and Grant, K. (1989). Mesencephalic control of the
electric organ discharges of mormyrid fishes. J. Neurobiol. 20, 619-634. bulbar electromotor network in the mormyrid Gnathonemus petersii. Eur.
Friedman, M. A. and Hopkins, C. D. (1996). Tracking individual mormyrid J. Neurosci. Suppl. 2, 176.
electric fish in the field using electric organ discharge waveforms. Anim. Noga, B., Kettler, J. and Jordan, L. (1988). Locomotion produced in
Behav. 51, 391-407. mesencephalic cats by injections of putative transmitter substances and
Friesen, W. and Stent, G. (1978). Neural circuits for generating rhythmic antagonists into the medial reticular formation and the pontomedullary
movements. Annu. Rev. Biophys. Bioeng. 7, 37-61. locomotor strip. J. Neurosci. 8, 2074-2086.
Fu, X. and Brudzynski, S. (1994). High-frequency ultrasonic vocalization Phillips, R. and Youngren, O. (1973). Electrical stimulation of brain as a tool
induced by intracerebral glutamate in rats. Pharmacol. Biochem. Behav. 49, for study of animal communication behavior evoked in mallard ducks
835-841. (Anas platyrhynchos). Brain Behav. Evol. 8, 253-286.
Fujita, I. and Konishi, M. (1991). The role of GABAergic inhibition in Scheffel, A. and Kramer, B. (1997). Electrocommunication and social
processing of interaural time difference in the owls auditory system. J. behaviour in Marcusenius senegalensis (Mormyridae, Teleostei). Ethology
Neurosci. 11, 722-739. 103, 404-420.
Goodson, J. L. and Bass, A. H. (2000). Rhythmic midbrain-evoked Scheffel, A. and Kramer, B. (2000). Electric signals in the social behavior
vocalization is inhibited by vasoactive intestinal polypeptide in the teleost of sympatric elephantfish (Mormyridae, Teleostei) from the upper Zambezi
Porichthys notatus. Brain Res. 865, 107-111. river. Naturwissenschaften 87, 142-147.
Grant, K., Bell, C. C., Clausse, S. and Ravaille, M. (1986). Morphology and Schmidt, R. (1966). Central mechanisms of frog calling. Behaviour 26, 251-
physiology of the brainstem nuclei controlling the electric organ discharge 285.
in mormyrid fish. J. Comp. Neurol. 245, 514-530. Schuller, G. and Radtkeschuller, S. (1990). Neural control of vocalization
Heiligenberg, W., Metzner, W., Wong, C. J. and Keller, C. H. (1996). in bats mapping of brainstem areas with electrical microstimulation
Motor control of the jamming avoidance response of Apteronotus eliciting species-specific echolocation calls in the Rufous horseshoe bat.
leptorhynchus: evolutionary changes of a behavior and its neuronal Exp. Brain Res. 79, 192-206.
substrate. J. Comp. Physiol. A 179, 653-674. Seller, T. and Armitage, S. (1983). Diencephalic sites from which calling can
Hopkins, C. D. (1981). On the diversity of electric signals in a community of be evoked with small currents in Japanese quail. Behav. Brain Res. 9, 305-
mormyrid electric fish in west Africa. Am. Zool. 21, 211-222. 314.
Hopkins, C. D. (1983). Neuroethology of species recognition in Serrier, J. and Moller, P. (1989). Patterns of electric organ discharge activity
electroreception. In Advances in Vertebrate Neuroethology (ed. J. P. Ewert, in the weakly electric fish Brienomyrus niger L. (Mormyridae). J. Exp. Biol.
R. R. Capranica and D. J. Ingle), pp. 871-881. New York: Plenum. 48, 235-244.
Hopkins, C. D. (1986). Behavior of Mormyridae. In Electroreception (ed. T. Teyssedre, C. and Boudinot, M. (1987). Rhythmicity as an intrinsic property
H. Bullock and W. Heiligenberg), pp. 527-576. New York: John Wiley & of the mormyrids electromotor command system. Physiol. Behav. 41, 201-
Sons. 207.
Hopkins, C. D. (1995). Convergent designs for electrogenesis and Valentine, D., Sinha, S. and Moss, C. (2002). Orienting responses and
electroreception. Curr. Opin. Neurobiol. 5, 769-777. vocalizations produced by microstimulation in the superior colliculus of the
Hopkins, C. D. and Bass, A. H. (1981). Temporal coding of species echolocating bat, Eptesicus fuscus. J. Comp. Physiol. A 188, 89-108.
recognition signals in an electric fish. Science 212, 85-87. von der Emde, G., Sena, L. G., Niso, R. and Grant, K. (2000). The midbrain
Jrgens, U. and Richter, K. (1986). Glutamate-induced vocalization in the precommand nucleus of the mormyrid electromotor network. J. Neurosci.
squirrel monkey. Brain Res. 373, 349-358. 20, 5483-5495.
Kramer, B. (1974). Electric organ discharge interaction during interspecific Wang, Y. and Bieger, D. (1991). Role of solitarial GABAergic mechanisms
agonistic behavior in freely swimming mormyrid fish: a method to evaluate in control of swallowing. Am. J. Physiol. 261, R639-R646.
two or more simultaneous time series of events with a digital analyzer. J. Williams, H. and Vicario, D. (1993). Temporal patterning of song
Comp. Physiol. A 93, 203-235. production: participation of nucleus uvaeformis of the thalamus. J.
Kramer, B. (1976). Electric signaling during aggressive behavior in Neurobiol. 24, 903-912.
Mormyrus rume (Mormyridae Teleostei). Naturwissenschaften 63, 48-49. Wullimann, M. F. and Northcutt, R. G. (1990). Visual and electrosensory
Kramer, B. (1993). Electrocommunication in weakly electric fish: review of circuits of the diencephalon in mormyrids: an evolutionary perspective. J.
signals sent and received. J. Comp. Physiol. A 173, 719-722. Comp. Neurol. 297, 537-552.
220
Brain Evolution Triggers Increased Diversification of Electric Fishes
Bruce A. Carlson, et al.
Science 332, 583 (2011);
DOI: 10.1126/science.1201524
If you wish to distribute this article to others, you can order high-quality copies for your
colleagues, clients, or customers by clicking here.
Permission to republish or repurpose articles or portions of articles can be obtained by
following the guidelines here.
Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by the
American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. Copyright
2011 by the American Association for the Advancement of Science; all rights reserved. The title Science is a
registered trademark of AAAS.
221
REPORTS
Space Center. Support was provided by NSF grants and current records from 20042005 have SOM Text
OCE-0424953, OCE-0425361, and OCE-0647948; been deposited in the Marine Geoscience Data System: Figs. S1 to S6
Woods Hole Oceanographic Institution (WHOI) RIDGE 2000 Data Portal. The authors declare no Table S1
grants from the Deep Ocean Exploration Institute competing financial interests. References
and the Ocean Venture Fund; a National Defense Movies S1 and S2
Science and Engineering Graduate Fellowship to D.K.A.; Supporting Online Material
and the WHOI Jannasch Chair for Excellence in www.sciencemag.org/cgi/content/full/332/6029/580/DC1 30 November 2010; accepted 28 March 2011
Oceanography to L.S.M. Larval fluxes, chemical fluxes, Materials and Methods 10.1126/science.1201066
Mormyrinae
cess these signals in a well-defined sensory path- Paramormyrops hopkinsi
clade A
Paramormyrops gabonensis
way devoted solely to the analysis of electric Paramormyrops sp. SZA
communication signals (1012) (fig. S1). origin of Marcusenius ntemensis
Mormyrids generate electric signals to com- ELa/ELp Boulengeromyrus knoepffleri
Ivindomyrus opdenboschi
municate and to actively sense their environment origin of electrocyte Ivindomyrus marchei
Mormyridae
(11). These signals have evolved more rapidly Stomatorhinus ivindoensis
stalk flexibility Marcusenius moorii
than body shape, size, and trophic ecology, sug- Isichthys henryi
gesting that electric communication behavior has Mormyrops zanclirostris
origin of Myomyrus macrops
played a key role in the radiation of mormyrids Petrocephalus simus
electrocyte
(3). Further, playback experiments in a few spe- Petrocephalus sullivani
stalks Petrocephalus valentini
cies suggest that these signals are critical for
Petrocephalinae
Petrocephalus sauvagii
species recognition during mate choice (1316). Petrocephalus pulsivertens
Electric signals are generated by an electric organ Petrocephalus christyi
Petrocephalus mbossou
in the tail, which consists of electrically excitable Petrocephalus grandoculis
cells called electrocytes (17). Electrocyte stalks Petrocephalus zakoni
origin of Petrocephalus binotatus
evolved with the origin of mormyrids, and devel- Petrocephalus balayi
opmental flexibility in stalk morphology arose ELa/ELp Petrocephalus odzalaensis
Petrocephalus microphthalmus (Gabon)
with the origin of the subfamily Mormyrinae (2). Petrocephalus microphthalmus (Congo)
This evolutionary change in the Mormyrinae es- Gymnarchus niloticus (Gymnarchidae)
1
Fig. 1. Inferred tree of phylogenetic relationships among mormyrid species and morphs. The phylogeny
Department of Biology, Washington University in St. Louis, St.
Louis, MO 63130, USA. 2Department of Biological Sciences,
was estimated by Bayesian analysis of cytb sequences (values at nodes are posterior probabilities). A
University of Idaho, Moscow, ID 83844, USA. 3Human Biology sequence from the closest outgroup to the Mormyridae (Gymnarchus niloticus) was used to root the tree.
Division, Fred Hutchinson Cancer Research Center, Seattle, WA Green branches represent a small exterolateral nucleus (EL) and magenta branches represent an enlarged
98109, USA. EL divided into anterior and posterior subdivisions (ELa/ELp); we reconstructed ancestral states using
*To whom correspondence should be addressed. E-mail: parsimony (see text). Gray outline represents electric organs with electrocyte stalks, and black outline
carlson.bruce@wustl.edu represents electric organs with developmentally labile stalks, based on a previous study (2).
1 cm
D Myomyrus macrops
C
Fig. 2. Anatomy of the exterolateral nucleus (EL). 5
EL volume/brain mass (mm3/g)
ELa/ELp
(A) Midbrain portion of 50-mm horizontal sections EL
from the clade A species B. brachyistius and the 4
P. microphthalmus
Paramormyrops
Stomatorhinus
Gnathonemus
P. soudanensis
P. pulsivertens
P. grandoculis
Marcusenius
Brienomyrus
Ivindomyrus
Mormyrops
Pollimyrus
P. valentini
P. sullivani
Myomyrus
Isichthys
P. zakoni
P. balayi
P. simus
nucleus; ll, lateral lemniscus; tel, telencephalon; OT, lenorgans, as found in all clade A species. (B) P.
optic tectum; MD, mediodorsal nucleus; val, valvula microphthalmus, the sole petrocephaline species
cerebellum. (C) Bar graph showing the median T with an ELa/ELp, also has a broad knollenorgan dis-
range of normalized EL volumes across all taxa tribution. (C) P. soudanensis has three knollenorgan
studied (table S2 gives sample sizes). clusters. (D) M. macrops has an intermediate pat-
clade A tern, with a single cluster and a low density of knol-
Mormyrinae Petrocephalinae lenorgans throughout the body.
experimental
pause duration (s)
12
control 6 Petrocephalinae (2) impeded rapid diversification
through signal divergence in the P. microphthalmus
Change in
9 NS p<0.01
4
lineage.
6
2 Evolution of communication systems can have
3 profound effects on species radiation. Evolution-
0 0 ary change in the structure of the anuran inner ear
may have fostered increased rates of speciation
-3 -2
kingsleyae (N=26)
curvifrons (N=1)
moorii (N=2)
walkeri (N=2)
brachyistius (N=2)
marchei (N=5)
brachyistius (N=2)
balayi (N=1)
simus (N=7)
sullivani (N=4)
microphthalmus (N=10)
Paramormyrops
Paramormyrops
Marcusenius
Stomatorhinus
Brienomyrus
Ivindomyrus
Brienomyrus
Petrocephalus
Petrocephalus
Petrocephalus
Petrocephalus
0.4 450
400 which brain evolution directly promotes diversi-
0.3
350 fication. We reveal that evolutionary change in
Dimension 2
226
www.sciencemag.org/cgi/content/full/332/6029/583/DC1
Bruce A. Carlson,* Saad M. Hasan, Michael Hollmann, Derek B. Miller, Luke J. Harmon,
Matthew E. Arnegard
227
Materials and Methods
Phylogenetic Reconstruction
A sound phylogenetic framework is required to understand patterns of brain
evolution and how they relate to the diversification of signals and species. Mormyrids
have already been well studied phylogenetically (2-6, 8, 31-37). Our purpose in the
present analysis was not to extend this phylogenetic literature. Rather, our aim was to
estimate a single phylogeny, including all relevant species in the best-sampled
communities, for making a formal comparison of rates of signal divergence and species
diversification between mormyrid lineages (see below). Thus, we used published cytb
sequences from the mormyrid assemblage of the Ivindo River basin near Makokou,
Gabon (2, 3, 34) (aligned sequences provided by J. P. Sullivan) and from the
petrocephaline assemblage of Odzala National Park in the Lkoli River basin, Republic
of the Congo (5, 6) (aligned sequences provided by S. Lavou). In the case of the Odzala
petrocephaline assemblage, we only considered the most common cytb haplotype for
every species, each of which appears to be exclusively monophyletic with respect to all
other sympatric species (5, 6). To these focal species we added published cytb sequences
for the outgroup species, Gymnarchus niloticus (3) (aquarium specimen), and a specimen
of Myomyrus macrops collected by J. P. Sullivan from the Ubangi River, Central African
Republic (2). Table S1 lists the specimen numbers and GenBank accession numbers for
all specimens used to construct the phylogeny.
We conducted a Bayesian phylogenetic analysis on the matrix of aligned cytb
sequences using MrBayes ver. 3.1.2 (38, 39) and the GTR+ model of sequence
evolution: i.e., a general time reversible model with nucleotide substitution rate variation
assumed to follow a gamma distribution. The Metropolis-coupled, Markov chain Monte
Carlo (MCMCMC) process included four chains, three heated and one cold. Starting
from random trees, we simultaneously performed two independent Bayesian analyses
(runs) for 2,000,000 generations each. We were confident that stationarity had been
reached by this stopping point because the standard deviation of split frequencies had
dropped to 0.006. During the parallel runs, we sampled parameter values and trees every
100 generations. Log-likelihood values for the sampled trees stabilized by 100,000
generations into each run. Therefore, we only used the last 1,900,000 generations in both
runs to estimate the 50% majority-rule consensus tree and clade credibility values (i.e.,
posterior probabilities, PP). The consensus tree (Fig. 1) was computed using all 38,000
trees pooled across the two Bayesian runs. Although the monophyly of the
Petrocephalinae appears rather weakly supported in the consensus tree (PP = 0.52; see
Fig. 1), the monophyly of this group and its sister-clade relationship to the Mormyrinae
are strongly supported by morphological synapomorphies (21) and phylogenetic analysis
using whole mitogenome sequences (40). The phylogeny provides robust support for the
monophyly of clade A, the focus of the current study (PP = 1.00; see Fig. 1), in
agreement with several existing phylogenies based on multiple molecular markers (2-6,
32, 34, 37).
Next, we pruned the Ivindo River Petrocephalus microphthalmus specimen from the
tree, leaving the P. microphthalmus specimen collected in Odzala (this is the only species
with more than one representative in our phylogeny; see Fig. 1). Using the penalized
likelihood routine in the program r8s ver. 1.71 (41, 42), we converted the consensus
228
Bayesian tree to ultrametric form. We selected a smoothing parameter (log10 = 2.2)
using cross-validation and ran the penalized likelihood procedure, checking the stability
of the solution using the checkgradient command in r8s. All analyses of signal
divergence and species diversification (below) were conducted on this ultrametric tree.
When taxa were not included in an analysis, we simply pruned them from this ultrametric
tree, which served as the single phylogenetic framework for subsequent tests of signal
divergence and species diversification rates.
Brain Histology
We obtained fixed brains in one of three ways: (1) laboratory specimens after
anesthesia in 300 mg/l MS-222, each fish was perfused through the heart with Hickmans
Ringer, followed by 4% paraformaldehyde in 0.1 M phosphate buffer (e.g. Brienomyrus
brachyistius in Fig. 2A and Petrocephalus soudanensis in Fig. 2B); (2) freshly caught
field specimens after anesthesia in 300 mg/l MS-222, the skull was opened, followed by
immersion fixation in 4% paraformaldehyde in 0.1 M phosphate buffer (e.g.
Petrocephalus microphthalmus in Fig. 2A); (3) specimens donated by the Cornell
University Museum of Vertebrates brains were removed from curated specimens,
which had been fixed in 10% phosphate-buffered formalin for approximately two weeks
and subsequently stored in 70% ethanol (e.g. Myomyrus macrops in Fig. 2B). In each
case, we post-fixed brains in 4% paraformaldehyde in 0.1 M phosphate buffer for several
days, then embedded them in gelatin and post-fixed them overnight before slicing. We
obtained 50 m horizontal sections with a vibrating microtome. Sections were mounted
on chrom-alum subbed slides, stained with cresyl violet, dehydrated in a graded alcohol
series, cleared with xylene, and then coverslipped (43).
The exterolateral nucleus (EL) was clearly identified in each brain based on
topology (10, 12, 44-50), which is largely considered the single-most important criterion
for establishing the homology of brain regions (51-53). Previous anatomical studies of
EL in three clade A species describe separate anterior and posterior subdivisions that are
clearly divisible based on cytology; these are referred to as ELa and ELp, respectively
(10, 12, 44-50). As described in the main text, some of the species we studied fit this
previous description, while other species have just a single, small EL lacking any
subdivisions (Fig. 2); we refer to these two phenotypes as ELa/ELp and EL, respectively.
Evoked potential recordings from the ELa/ELp of P. microphthalmus reveal 2-3 ms
latency responses to electrosensory stimulation that are blocked at short delays following
the electric discharge motor command, exactly as described for species in clade A (48,
50, 54-56). This further satisfies the following additional criteria for establishing
homology (51, 57): physiological properties, connectivity (input from knollenorgans via
the nucleus of the electrosensory lateral line lobe, or nELL; see fig. S1), and function
(detection of electric signals generated by other fish). For each brain, we delineated the
borders of EL or ELa/ELp in each section based on cytology (12, 44-47). We then
determined the total volume of EL or ELa/ELp by summing cross-sectional areas,
multiplying by section thickness, and taking the mean of the left and right sides (no
lateral asymmetries were detected). We normalized this value by total brain mass,
determined after post-fixing and before embedding. No apparent differences in the
architecture or normalized volumes of these brain regions were observed between the
229
brains of closely-related specimens obtained through different fixation methods. A full
list of species for which we obtained EL anatomy data is provided in table S2.
Electroreceptor Histology
Curated specimens donated by the Cornell University Museum of Vertebrates had
been fixed in 10% phosphate-buffered formalin for approximately two weeks before
storage in 70% ethanol. Laboratory specimens were fixed in 70% ethanol. Mormyrids
have three distinct types of electroreceptors: ampullary organs used for passive
electrolocation, mormyromasts used for active electrolocation, and knollenorgans used
for electric communication (18, 20). There is a semi-transparent layer of skin that covers
the body where these receptors are located. In fixed specimens, this skin layer turns white
and is easily removed. We photographed specimens before removing this layer in pieces.
We then placed each piece in a solution of 0.05% toluidine blue for 10 minutes, followed
by three 5-minute washes in 70% ethanol. The three different types of electroreceptors
could then be visually identified and distinguished under a microscope (fig. S2A). We
mapped the location of knollenorgans in each piece of skin onto the original photograph,
and then outlined the specimen (Fig. 3; fig. S2B, C). A full list of species for which we
obtained data on knollenorgan distributions is provided in table S2.
230
a temporally distorted signal waveform with a frequency spectrum and total energy
equivalent to the original waveform (13, 60). Stimulus trains consisted of 10 bursts of 10
pulses each, with an intra-burst interval of 30 ms, inter-burst interval of 10 s, and peak-to-
peak intensity of 145 mV/cm. For control stimulation, all 100 pulses were an identical
conspecific waveform. Experimental stimuli were the same, except that all 10 pulses in
the 9th burst were a phase-shifted version of the same conspecific waveform (fig. S3C).
Each playback used a stimulus waveform recorded from a different conspecific (61).
In species that responded to stimulation with increases in discharge rate, we
determined the maximum discharge rates in response to each of the 10 bursts after
Gaussian smoothing (fig. S4A). In species that responded to stimulation by pausing, we
determined the duration of pauses in response to each of the 10 bursts (fig. S4B).
Discrimination was assessed as the change in maximum discharge rate or change in pause
duration from the 8th to 9th bursts (Fig. 4A).
231
a pooled within-species variance-covariance matrix (3). We then plotted D2 against
patristic distances taken from the ultrametric tree for all clade A pair-wise comparisons
and all other pair-wise comparisons outside this clade. Patristic distances were scaled to a
maximum of 1.0. This plot allowed us to visualize the pattern of accumulation of signal
differences across distances in the phylogenetic tree without assuming any model of trait
evolution a priori (3). For clarity, we refer to D2 as signal waveform distance and
patristic distance as phylogenetic distance in Fig. 4C. We also computed D2 by
calculating separate pooled variance-covariance matrices for all clade A species and all
other species. This resulted in the same pattern of more rapid signal evolution in clade A
(data not shown).
We used Brownie ver. 2.1.1 to statistically compare rates of accumulation of signal
waveform disparity between clade A and all outgroup mormyrids in a Brownian motion
framework by means of maximum likelihood (24). The Brownian motion rate parameter
(2) describes the rate with which trait variance among species has increased over
evolutionary time. Using a censored rate test, we compared a model that fit a single 2
across the entire ultrametric tree of mormyrid relationships to a two-rate model allowing
clade A and all other mormyrids to differ in 2. This comparison was made separately for
each MDS-derived axis of signal variation. Given the moderate number of taxa under
consideration, we made each comparison using the small-sample-size-corrected version
of the Akaike Information Criterion (AICc) and a likelihood ratio test evaluated with a
parametric bootstrapping procedure (5,000 pseudoreplicates). These model evaluation
approaches avoid the inflated Type 1 errors that stem from using the 2 distribution with
small sample sizes (24, 68).
232
Fig. S1. Dorsal view of the brain of the clade A species Brienomyrus brachyistius,
highlighting the knollenorgan electrosensory pathway. Knollenorgan (KO) primary
afferent fibers project ipsilaterally to the nucleus of the electrosensory lateral line lobe
(nELL) in the hindbrain via the anterior (not shown) and posterior lateral line nerves
(nPLL). Neurons in the nELL project bilaterally to the anterior exterolateral nucleus
(ELa) in the midbrain, which in turn projects ipsilaterally to the adjacent posterior
exterolateral nucleus (ELp).
233
Fig. S2. Mapping of knollenorgan electroreceptor locations. (A) Portion of skin from the
clade A species Brienomyrus brachyistius stained with toluidine blue. Knollenorgans can
easily be distinguished from ampullary and mormyromast electroreceptors. (B) Maps of
knollenorgan locations from three species in clade A. (C) Maps of knollenorgan locations
from three species outside of clade A. Knollenorgan locations are indicated by red dots.
Scale bars = 1 cm.
234
Fig. S3. Behavioral discrimination experiments. (A) Set-up for field playback
experiments. Each fish was placed in a rectangular PVC enclosure with stimulus
electrodes spanning the length of both sides of the inside of the enclosure (red) and
recording electrodes at each end (blue). Stimuli were generated on a laptop and delivered
through the digital-to-analog converter (DAC) of a USB-connected portable processor.
Stimuli were isolated from ground prior to delivery (SIU). The output of the fish was
amplified (AMP), then digitized by the analog-to-digital converter (ADC) of the
processor. (B) Examples of stimulus waveforms used in experiments for three species.
Conspecific signals were recorded in the field. These waveforms were temporally
distorted by phase-shifting them by 90. (C) Two types of stimulus trains were used.
Control stimulus trains consisted of 10 bursts of 10 pulses each; each pulse was an
identical conspecific waveform. Experimental stimulus trains were the same except that
the 9th burst of pulses consisted of phase-shifted versions of the same conspecific
waveform.
235
Fig. S4. Habituation of behavioral responses to electrosensory stimulation. (A) In
Paramormyops kingsleyae, stimulation resulted in bursts of electrical output, quantified
as the maximum discharge rate after Gaussian smoothing. This response habituates
throughout the course of control stimulus trains. (B) In all four petrocephaline species,
Petrocephalus balayi (N = 1), P. microphthalmus (N = 10), P. simus (N = 7), and P.
sullivani (N = 4), stimulation elicited pauses in electrical output, quantified as pause
duration. This response also habituates throughout the course of control stimulus trains.
Values show the mean s.e.m. Statistical significance of the decrease in response with
repeated stimulus presentations was assessed using repeated-measures ANOVA.
236
Fig. S5. Electric signal waveforms used for analyzing rates of signal evolution. (A)
Waveforms from the Ivindo mormyrid assemblage of Gabon (3). (B) Waveforms from
the petrocephaline assemblage of Odzala National Park in the Republic of the Congo (5,
6). Species from clade A are shown in red, all other species are shown in blue. In each
case, waveforms are amplitude-normalized and plotted head-positive up. Multiple
waveforms from different individuals of the same species are superimposed and aligned
to the head-positive peak (except for Paramormyrops sp. TEN, for which waveforms
are aligned to the head-negative peak). The left and right columns show waveforms at
two different timescales (1 ms and 0.1 ms scale bars, respectively). The longest
waveforms are shown only in the left column.
237
Fig. S6. Cross-correlation method used to measure waveform similarity. (A) Example
showing the cross-correlation of two conspecific waveforms recorded from different
individuals of Paramormyrops curvifrons. The cross-correlation coefficients (bottom)
reveal the correlation between the two waveforms as a function of the relative delay (lag)
between them. The maximum of the absolute values of this function (C = 0.9587)
provides a scalar measure of waveform similarity. (B) Example showing the cross-
correlation of two heterospecific waveforms (P. curvifrons and Paramormyrops sp.
magnostipes type 1). Here the waveforms are less similar (C = 0.3996).
12
238
Table S1. Cytochrome b (cytb) sequences used for estimating mormyrid phylogeny,
showing corresponding specimen information. Museum catalog numbers preceded by
CU are specimens housed in the Cornell University Museum of Vertebrates, Ithaca,
NY. The single museum catalog number preceded by AMNH is a specimen housed in
the American Museum of Natural History, New York, NY.
Museum catalog GenBank
Taxon no., specimen no. Collection site accession no.
Paramormyrops sp. type 1 CU 78326, #2297 Loa Loa Rapids, Ivindo River, Gabon AF477452
Paramormyrops sp. type 2 CU 78344, #2295 Loa Loa Rapids, Ivindo River, Gabon AF477455
Paramormyrops sp. TEN CU 80809, #2011 Bial Stream, Ivindo R. basin, Gabon AF477453
Paramormyrops curvifrons CU 81661, #2050 Loa Loa Rapids, Ivindo River, Gabon AF477469
Paramormyrops longicaudatus CU 78355, #2289 Loa Loa Rapids, Ivindo River, Gabon AF201576
Paramormyrops kingsleyae CU 80816, #2116 Bial Stream, Ivindo R. basin, Gabon AF477466
Paramormyrops sp. SN9 CU 89360, #5552 Loa Loa Rapids, Ivindo River, Gabon FJ830628
Paramormyrops hopkinsi CU 78352, #2285 Loa Loa Rapids, Ivindo River, Gabon AF201575
Paramormyrops gabonensis CU 79702, #2048 Loa Loa Rapids, Ivindo River, Gabon AF201603
Paramormyrops sp. SZA CU 80848, #2008 Bial Stream, Ivindo R. basin, Gabon AF477475
Marcusenius ntemensis CU 79706, #2186 Loa Loa Rapids, Ivindo River, Gabon AF201593
Boulengeromyrus knoepffleri CU 79692, #2248 Loa Loa Rapids, Ivindo River, Gabon AF201573
Ivindomyrus opdenboschi CU 84642, #2106 Loa Loa Rapids, Ivindo River, Gabon DQ166689
Ivindomyrus marchei CU 81642, #2183 Loa Loa Rapids, Ivindo River, Gabon DQ166677
Stomatorhinus ivindoensis CU 70703, #2074 Bial Stream, Ivindo R. basin, Gabon AF201612
Marcusenius moorii CU 79697, #2013 Bal Creek, Ivindo R. basin, Gabon AF201595
Isichthys henryi CU 79705, #2179 Loa Loa Rapids, Ivindo River, Gabon AF201590
Makokou region, Ivindo River,
Mormyrops zanclirostris CU 79707, #2210 AF201599
Gabon
AMNH 228166, Ubangi River, Congo R. basin,
Myomyrus macrops AF201602
#2524 Central African Republic
Petrocephalus simus CU 79701, #2035 Bal Creek, Ivindo R. basin, Gabon AF201604
Petrocephalus sullivani CU 79700, #2038 Bal Creek, Ivindo R. basin, Gabon AF201606
Petrocephalus valentini CU 88058, #5175 Lkoli River, Congo R. basin, Congo EU770181
Petrocephalus sauvagii CU 87864, #5206 Lkoli River, Congo R. basin, Congo EU770160
Petrocephalus pulsivertens CU 88097, #5263 Lkoli River, Congo R. basin, Congo EU770174
Petrocephalus christyi CU 88095, #5261 Lkoli River, Congo R. basin, Congo EU770183
Petrocephalus mbossou CU 92389, #6183 Lkoli River, Congo R. basin, Congo EU770163
Petrocephalus grandoculis CU 92385, #6181 Lkoli River, Congo R. basin, Congo EU770155
Petrocephalus zakoni CU 92391, #6132 Lkoli River, Congo R. basin, Congo EU770170
Petrocephalus binotatus CU 88064, #5001 Lkoli River, Congo R. basin, Congo EU770164
Petrocephalus balayi CU 88111, #5314 Lkoli River, Congo R. basin, Congo EU770192
Petrocephalus odzalaensis CU 87852, #5148 Lkoli River, Congo R. basin, Congo EU770156
Petrocephalus microphthalmus CU 82208, #2199 Loa Loa Rapids, Ivindo River, Gabon EU770185
Petrocephalus microphthalmus CU 87940, #5092 Lkoli River, Congo R. basin, Congo EU770188
Gymnarchus niloticus CU 80334, no spec. # Aquarium import AF201586
239
Table S2. Specimens used for studying brain anatomy and knollenorgan distributions.
Taxonomic assignments are based on our cytochrome b (cytb) phylogeny (Fig. 1) as well
as several published studies (2-6, 32, 34, 37, 58).
Sub- EL anatomy Pattern of EL Knollenorgan
family Genus Species1 sample size organization2 distribution3
Brevimyrus niger B4
Brienomyrus brachyistius 5 ELa/ELp B
numenius 1 ELa/ELp
Campylomormyrus rhynchophorus B4
tamandua 1 ELa/ELp B
Gnathonemus petersii 1 ELa/ELp B
Isichthys henryi 2 ELa/ELp
Ivindomyrus marchei 2 ELa/ELp
2 ELa/ELp B4
Clade A
moorii
Mormyrinae
Marcusenius
senegalensis B4
nigricans 1 ELa/ELp B4
Mormyrops
zanclirostris 2 ELa/ELp B
Mormyrus caballus B4
gabonensis 2 ELa/ELp
longicaudatus 1 ELa/ELp
Paramormyrops
sp. BON 1 ELa/ELp
sp. VAD 1 ELa/ELp
Pollimyrus adspersus 2 ELa/ELp B
Stomatorhinus walkeri 3 ELa/ELp
macrops 1 EL C-B
Myomyrus
pharao 1 EL C-B
balayi 1 EL C4
binotatus C4
christyi C4
grandoculis 1 EL C
Petrocephalinae
microphthalmus 3 ELa/ELp B
odzalaensis C4
Petrocephalus pulsivertens 1 EL C
sauvagii C4
simus 4 EL C
soudanensis 4 EL C
sullivani 2 EL C
valentini 1 EL C
zakoni 3 EL B
1. Undescribed species are referred to by established 3-letter cheironyms (34, 59).
2. EL = relatively small exterolateral nucleus; ELa/ELp = relatively large exterolateral nucleus
divided into anterior and posterior subdivisions.
3. B = broad distribution of knollenorgan receptors throughout the head and trunk; C = distinct
clusters of knollenorgan receptors on the head; C-B = intermediate pattern with one cluster of
knollenorgan receptors on the head as well as a broad distribution of knollenorgan receptors at
relatively low density.
4. Knollenorgan distributions based on published studies (6, 20, 21). These previous studies used visual
examination of intact specimens rather than the staining method employed in the current study.
240
Table S3. Model comparisons testing for a shift in diversification rate in clade A relative
to other mormyrids. We set the relative extinction fractions () to different values, and
then compared a model with constant net diversification rate (r) across the tree to one
with different net diversification rates for clade A (rA) and the rest of the tree (rother).
Model Parameter estimates rA/rother lnL Delta p-value
=0
Single rate r=0.0079 -95.1
Two rate rother=0.0032, rA=0.0100 3.12 -84.0 22.2 <0.001
= 0.5
Single rate r=0.0062 -92.7
Two rate rother=0.0025, rA=0.0083 3.32 -84.1 17.1 <0.001
= 0.9
Single rate r=0.0030 -90.1
Two rate rother=0.0010, rA=0.0042 4.20 -85.7 8.8 0.003
= 0.99
Single rate r=0.0005 -90.5
Two rate rother=0.00014, rA=0.00073 5.21 -88.1 4.8 0.03
15
241
References and Notes
1. C. J. Hoskin, M. Higgie, Speciation via species interactions: The divergence of mating
traits within species. Ecol. Lett. 13, 409 (2010).
2. J. P. Sullivan, S. Lavou, C. D. Hopkins, Molecular systematics of the African electric
fishes (Mormyroidea: teleostei) and a model for the evolution of their electric
organs. J. Exp. Biol. 203, 665 (2000).
3. M. E. Arnegard et al., Sexual signal evolution outpaces ecological divergence during
electric fish species radiation. Am. Nat. 176, 335 (2010).
4. S. Lavou, J. P. Sullivan, C. D. Hopkins, Phylogenetic utility of the first two introns of
the S7 ribosomal protein gene in African electric fishes (Mormyroidea: Teleostei)
and congruence with other molecular markers. Biol. J. Linn. Soc. Lond. 78, 273
(2003).
5. S. Lavou, M. E. Arnegard, J. P. Sullivan, C. D. Hopkins, Petrocephalus of Odzala
offer insights into evolutionary patterns of signal diversification in the
Mormyridae, a family of weakly electrogenic fishes from Africa. J. Physiol. Paris
102, 322 (2008).
6. S. Lavou, J. P. Sullivan, African weakly electric fishes of the genus Petrocephalus
(Osteoglossomorpha: Mormyridae) of Odzala National Park, Republic of the
Congo (Lkoli River, Congo River basin) with description of five new species. M.
E. Arnegard, Zootaxa 2600, 1 (2010).
7. W. N. Eschmeyer, J. D. Fong, Catalog of Fishes (California Academy of Sciences, San
Francisco, 2010).
8. P. G. D. Feulner, F. Kirschbaum, V. Mamonekene, V. Ketmaier, R. Tiedemann,
Adaptive radiation in African weakly electric fish (Teleostei: Mormyridae:
Campylomormyrus): A combined molecular and morphological approach. J. Evol.
Biol. 20, 403 (2007).
9. C. D. Hopkins, On the diversity of electric signals in a community of mormyrid
electric fish in west Africa. Am. Zool. 21, 211 (1981).
10. M. A. Xu-Friedman, C. D. Hopkins, Central mechanisms of temporal analysis in the
knollenorgan pathway of mormyrid electric fish. J. Exp. Biol. 202, 1311 (1999).
11. B. A. Carlson, in Communication in Fishes, F. Ladich, S. P. Collin, P. Moller, B. G.
Kapoor, Eds. (Science Publishers, Enfield, NH, 2006), vol. 2, pp. 805-848.
12. C. C. Bell, T. Szabo, in Electroreception, T. H. Bullock, W. Heiligenberg, Eds. (John
Wiley & Sons, New York, 1986), pp. 375-421.
13. C. D. Hopkins, A. H. Bass, Temporal coding of species recognition signals in an
electric fish. Science 212, 85 (1981).
16
242
14. P. G. D. Feulner, M. Plath, J. Engelmann, F. Kirschbaum, R. Tiedemann, Electrifying
love: Electric fish use species-specific discharge for mate recognition. Biol. Lett.
5, 225 (2009).
15. P. Machnik, B. Kramer, Female choice by electric pulse duration: Attractiveness of
the males communication signal assessed by female bulldog fish, Marcusenius
pongolensis (Mormyridae, Teleostei). J. Exp. Biol. 211, 1969 (2008).
16. M. E. Arnegard, B. S. Jackson, C. D. Hopkins, Time-domain signal divergence and
discrimination without receptor modification in sympatric morphs of electric
fishes. J. Exp. Biol. 209, 2182 (2006).
17. A. H. Bass, in Electroreception, T. H. Bullock, W. Heiligenberg, Eds. (Wiley, New
York, 1986), pp. 1370.
18. H. H. Zakon, in Electroreception, T. H. Bullock, W. Heiligenberg, Eds. (Wiley, New
York, 1986), pp. 103156.
19. A. H. Bass, C. D. Hopkins, Comparative aspects of brain organization of an African
wave electric fish, Gymnarchus niloticus. J. Morphol. 174, 313 (1982).
20. W. Harder, Die Beziehungen zwischen Elektrorezeptoren, Elektrischem Organ,
Seitenlinienorganen und Nervensystem bei den Mormyridae (Teleostei, Pisces).
Z. Vgl. Physiol. 59, 272 (1968).
21. S. Lavou, C. D. Hopkins, A. K. Toham, The Petrocephalus (Pisces,
Osteoglossomorpha, Mormyridae) of Gabon, Central Africa, with the description
of a new species. Zoosystema 26, 511 (2004).
22. N. Post, G. von der Emde, The novelty response in an electric fish: Response
properties and habituation. Physiol. Behav. 68, 115 (1999).
23. P. Moller, J. Serrier, D. Bowling, electric organ discharge displays during social
encounter in the weakly electric fish Brienomyrus niger L. (Mormyridae).
Ethology 82, 177 (1989).
24. B. C. OMeara, C. An, M. J. Sanderson, P. C. Wainwright, Testing for different rates
of continuous trait evolution using likelihood. Evolution 60, 922 (2006).
25. D. L. Rabosky, S. C. Donnellan, A. L. Talaba, I. J. Lovette, Exceptional among-
lineage variation in diversification rates during the radiation of Australias most
diverse vertebrate clade. Proc. Biol. Sci. 274, 2915 (2007).
26. M. J. Ryan, Neuroanatomy influences speciation rates among anurans. Proc. Natl.
Acad. Sci. U.S.A. 83, 1379 (1986).
27. O. Seehausen et al., Speciation through sensory drive in cichlid fish. Nature 455, 620
(2008).
28. Y. Terai et al., Divergent selection on opsins drives incipient speciation in Lake
Victoria cichlids. PLoS Biol. 4, e433 (2006).
17
243
29. J. B. Sylvester et al., Brain diversity evolves via differences in patterning. Proc. Natl.
Acad. Sci. U.S.A. 107, 9718 (2010).
30. C. A. Shumway, Habitat complexity, brain, and behavior. Brain Behav. Evol. 72, 123
(2008).
31. J. Alves-Gomes, C. D. Hopkins, Molecular insights into the phylogeny of
mormyriform fishes and the evolution of their electric organs. Brain Behav. Evol.
49, 324 (1997).
32. S. Lavou, R. Bigorne, G. Lecointre, J.-F. Agnse, Phylogenetic relationships of
mormyrid electric fishes (Mormyridae; Teleostei) inferred from cytochrome b
sequences. Mol. Phylogenet. Evol. 14, 1 (2000).
33. S. Lavou, J. P. Sullivan, M. E. Arnegard, C. D. Hopkins, Differentiation of
morphology, genetics and electric signals in a region of sympatry between sister
species of African electric fish (Mormyridae). J. Evol. Biol. 21, 1030 (2008).
34. J. P. Sullivan, S. Lavou, C. D. Hopkins, Discovery and phylogenetic analysis of a
riverine species flock of African electric fishes (Mormyridae: Teleostei).
Evolution 56, 597 (2002).
35. J. P. Sullivan, S. Lavou, M. E. Arnegard, C. D. Hopkins, AFLPs resolve phylogeny
and reveal mitochondrial introgression within a species flock of African electric
fish (Mormyroidea: Teleostei). Evolution 58, 825 (2004).
36. P. G. D. Feulner, F. Kirschbaum, R. Tiedemann, Adaptive radiation in the Congo
River: An ecological speciation scenario for African weakly electric fish
(Teleostei; Mormyridae; Campylomormyrus). J. Physiol. Paris 102, 340 (2008).
37. M. E. Arnegard, D. J. Zwickl, Y. Lu, H. H. Zakon, Old gene duplication facilitates
origin and diversification of an innovative communication systemtwice. Proc.
Natl. Acad. Sci. U.S.A. 107, 22172 (2010).
38. J. P. Huelsenbeck, F. Ronquist, MRBAYES: Bayesian inference of phylogenetic
trees. Bioinformatics 17, 754 (2001).
39. F. Ronquist, J. P. Huelsenbeck, MrBayes 3: Bayesian phylogenetic inference under
mixed models. Bioinformatics 19, 1572 (2003).
40. S. Lavou et al., Remarkable morphological stasis in an extant vertebrate despite tens
of millions of years of divergence. Proc. Biol. Sci. 278, 1003 (2011).
41. M. J. Sanderson, Estimating absolute rates of molecular evolution and divergence
times: A penalized likelihood approach. Mol. Biol. Evol. 19, 101 (2002).
42. M. J. Sanderson, r8s: Inferring absolute rates of molecular evolution and divergence
times in the absence of a molecular clock. Bioinformatics 19, 301 (2003).
43. B. A. Carlson, Neuroanatomy of the mormyrid electromotor control system. J. Comp.
Neurol. 454, 440 (2002).
18
244
44. F. Hauged-Carr, The mesencephalic exterolateral posterior nucleus of the
mormyrid fish Brienomyrus niger: Efferent connections studied by the HRP
method. Brain Res. 178, 179 (1979).
45. E. Mugnaini, L. Maler, Cytology and immunocytochemistry of the nucleus
extrolateralis anterior of the mormyrid brain: Possible role of GABAergic
synapses in temporal analysis. Anat. Embryol. (Berl.) 176, 313 (1987).
46. M. A. Friedman, M. Kawasaki, Calretinin-like immunoreactivity in mormyrid and
gymnarchid electrosensory and electromotor systems. J. Comp. Neurol. 387, 341
(1997).
47. M. A. Friedman, C. D. Hopkins, Neural substrates for species recognition in the time-
coding electrosensory pathway of mormyrid electric fish. J. Neurosci. 18, 1171
(1998).
48. B. A. Carlson, Temporal-pattern recognition by single neurons in a sensory pathway
devoted to social communication behavior. J. Neurosci. 29, 9417 (2009).
49. S. Amagai, M. A. Friedman, C. D. Hopkins, Time coding in the midbrain of
mormyrid electric fish. I. Physiology and anatomy of cells in the nucleus
exterolateralis pars anterior. J. Comp. Physiol. A Neuroethol. Sens. Neural Behav.
Physiol. 182, 115 (1998).
50. S. Amagai, Time coding in the midbrain of mormyrid electric fish. II. Stimulus
selectivity in the nucleus exterolateralis pars posterior. J. Comp. Physiol. A
Neuroethol. Sens. Neural Behav. Physiol. 182, 131 (1998).
51. G. F. Striedter, Principles of Brain Evolution (Sinauer, Sunderland, MA, 2005).
52. R. Nieuwenhuys, Comparative neuroanatomy: Place, principles, practice and
programme. Eur. J. Morphol. 32, 142 (1994).
53. L. Puelles, L. Medina, Field homology as a way to reconcile genetic and
developmental variability with adult homology. Brain Res. Bull. 57, 243 (2002).
54. P. S. Enger, S. Libouban, T. Szabo, Fast conducting electrosensory pathway in the
mormyrid fish, Gnathonemus petersii. Neurosci. Lett. 2, 133 (1976).
55. C. J. Russell, C. C. Bell, Neuronal responses to electrosensory input in mormyrid
valvula cerebelli. J. Neurophysiol. 41, 1495 (1978).
56. T. Szabo, P. S. Enger, S. Libouban, Electrosensory systems in the mormyrid fish,
Gnathonemus petersii: Special emphasis on the fast conducting pathway. J.
Physiol. (Paris) 75, 409 (1979).
57. C. B. G. Campbell, W. Hodos, The concept of homology and the evolution of the
nervous system. Brain Behav. Evol. 3, 353 (1970).
58. C. D. Hopkins, S. Lavou, J. P. Sullivan, in The Fresh and Brackish Water Fishes of
Lower Guinea, West-Central Africa, M. L. J. Stiassny, G. G. Teugels, C. D.
19
245
Hopkins, Eds. (IRD, Publications scientifiques du Musum, MRAC, Paris, 2007),
vol. 1.
59. M. E. Arnegard, S. M. Bogdanowicz, C. D. Hopkins, Multiple cases of striking
genetic similarity between alternate electric fish signal morphs in sympatry.
Evolution 59, 324 (2005).
60. W. Heiligenberg, R. A. Altes, Phase sensitivity in electroreception. Science 199, 1001
(1978).
61. D. Kroodsma, Suggested experimental designs for song playbacks. Anim. Behav. 37,
600 (1989).
62. C. D. Hopkins, Design features for electric communication. J. Exp. Biol. 202, 1217
(1999).
63. C. Chatfield, The Analysis of Time Series: An Introduction. (Chapman and Hall, New
York, 2004).
64. C. D. Hopkins, Temporal structure of non-propagated electric communication signals.
Brain Behav. Evol. 28, 43 (1986).
65. J. B. Kruskal, M. Wish, Multidimensional Scaling. (Sage, Beverly Hills, CA, 1978).
66. P. Mahalanobis, On the generalized distance in statistics. Proc. Natl. Inst. Sci. India
2, 49 (1936).
67. N. Eldredge et al., The dynamics of evolutionary stasis. Paleobiology 31, (suppl), 133
(2005).
68. S. A. Price et al., Functional innovations and morphological diversification in
parrotfish. Evolution 64, 3057 (2010).
69. D. L. Rabosky, Extinction rates should not be estimated from molecular phylogenies.
Evolution 64, 1816 (2010).
70. M. E. Alfaro et al., Nine exceptional radiations plus high turnover explain species
diversity in jawed vertebrates. Proc. Natl. Acad. Sci. U.S.A. 106, 13410 (2009).
Acknowledgments: We thank C. D. Hopkins and J. R. Gallant for help with field work,
J. P. Friel (Cornell University Museum of Vertebrates) for providing specimens,
and S. Lavou and J. P. Sullivan for providing cytb sequences (GenBank
accession numbers provided in table S1). Supported by NSF IOS-0818390
(B.A.C.); L.J.H. was supported by NSF DEB-0919499.
20
246
Suthers, Roderick (2004) How birds sing and why it matters. In Nature's Music: the
science of birdsong. Marler, P. and Slabbekoorn, H. eds. Elsevier. New York. Chapter 9
pp. 272-295.
How birds sing and why it matters 273
I
How birds sing and why it matters
RODERICK A. SUTHERS 'I
1 I
! MT
!
I
INTRODUCTION briefly to indicate the variety of avian vocal I
systems, and to provide a perspective from which I
1 I
Songbirds have both an esthetic and a scientific to focus on the oscine songbirds. Readers 1
impact on our lives. Birdsong adds beauty and interested in additional information on songbirds Figure 9.1 Examples of variation in syringeal anatomy. The tracheal parrot syrinx has two syringeal
vitality to our environment, The possibility of should consult reviews of this subject (Nowicki muscles and a pair of lateral tympaniform membranes. The bronchial syrinx of the oilbird has one pair of
its absence due to in c r eas ing levels of & Marler 1988; Suthers 1997, 1999a, b; Gaunt syringeal muscles and a pair of medial and lateral tympaniform membranes in each bronchus. Songbirds
environmental toxins, so eloquently described & Nowicki 1998; Doupe 81Kuhl 1999; Suthers have several pairs of syringeal muscles in their tracheobronchial syrinx. Tr, trachea; ST, sternotrachealis
by Rachel Carson (1962) in her goundbreaking ef ale 1999; Goller & Larsen 2002)- muscle; SY SUP, superficial syringeal muscle; SY PROF, deep syringeal muscle; SY VALV, pneumatic
valve; LTM, lateral tympaniform me~brane;MTM, medial tympaniform membrane; BCI, first bronchial
book, Silent Spring, was a potent factor in cartilage; SY, syringeal muscle; ML, medial labium; LL, lateral labium; SYR, muscles of syrinx. (Oilbird
mobilizing public opinion to become better modified after Suthers & Hector 1985; parrot and canary modified after King 1989).
stewards of our environment. Scientifically, the
highly developed vocal communication of this AN ORGAN FOR SINGING
The Respiratory Rhythm direction through the syrinx and into the
group, sharing as it does a number of parallels
with speech, has made songbirds especially suited The avian vocal organ, the syrinx, is located The timing of vocalization and the tempo of esophagus. After each coo there is a short
for the study of complex, learned vocal deep in the chest in an air sac that is connected song begins with the respiratory cycle (Vicario expiration followed by a brief inspiration (Gaunt
communication at every level, from its molecular to other air sacs and the lungs. It is present in all 1991). Expiratory muscles compress the bellows- et al. 1982).
biology to its evolution. birds except vultures, but its exact location and like air sacs in the thorax and abdomen, increasing
This chapter is about how birdsong is structure varies considerably (King 1989). In respiratory pressure so air flows out through the
produced. How do birds generate such a variety some species, including doves, pigeons and syrinx and trachea (Hartley 1990). During song, The Source of Sound
of sounds and what vocal gymnastics are required parrots, the syrinx consists of modified rings of a small inspiration called a 'mini-breath' (Calder It is now generally agreed that vocalizations are
to produce them?The vocal system is the interface cartilage near the lower end of trachea 1970) is usually taken after each syllable, except generated by airflow-induced oscillation of
between the bird's brain and his song. Knowing (Fig. 9.1). at very high syllable repetition rates (Hartley & elements in the wall of the syrinx that convert
how it functions, and coming to appreciate its In other species, such as penguins, many owls, Suthers 1989). Inspiratory muscles expand the some of the air's kinetic energy into acoustic
limitations as well as its capabilities, is important nightjars, and some cuckoos, the syrinx thorax and abdomen, reducing pressure in the energy. These oscillations are presumably
in understanding vocal communication. Birdsong exists as two semisyrinxes located in the primary air sacs and reversing the direction of airflow sustained by interaction between Bernoulli forces
is the product of the carefully coordinated activity bronchi (Fig. 9.1). In many birds, however, through the syrinx (Wild et al. 1998). and the inertia of air in the vocal tract, in much
of many different muscles. Some of these are including but by no means limited to the oscine Vocalization occurs during expiratory airflow, the same way as are oscillations of human vocal
associated with the vocal organ but many belong songbirds, the syrinx is at the junction where aside from rare exceptions when it occurs during folds (Titze 1994; Gardner et al. 200 1). An
to other muscle groups such as the respiratory the two primary bronchi join to form the trachea. inspiration, as in the 'wah' sounds of doves (Gaunt alternative mechanism for sound production
system and the vocal tract. Together they convert This tracheobronchial syrinx includes modified I et al. 1982), and certain syllables in the songs of based on the principle of an aerodynamic whistle
nerve impulses from the brain into song. cartilages from both the upper end of each some zebra finches (Goller & Daley 2001). Most was put forth (Gaunt et al. 1982) to explain
Less is known about the vocal mechanisms of bronchus and the lower end of the trachea birds sing with their beaks open, but doves and pure tone vocalizations in birds such as doves.
non-songbirds, and they will be mentioned only (Fig. 9.1). pigeons coo with their beaks and nares closed. However, this aerodynamic whistle hypothesis
Sound is produced during airflow in an expiratory is not supported by the observation of
247
Suthers, Roderick (2004) How birds sing and why it matters. In Nature's Music: the
science of birdsong. Marler, P. and Slabbekoorn, H. eds. Elsevier. New York. Chapter 9
pp. 272-295.
How birds sing and why it matters 273
I
How birds sing and why it matters
RODERICK A. SUTHERS 'I
1 I
! MT
!
I
INTRODUCTION briefly to indicate the variety of avian vocal I
systems, and to provide a perspective from which I
1 I
Songbirds have both an esthetic and a scientific to focus on the oscine songbirds. Readers 1
impact on our lives. Birdsong adds beauty and interested in additional information on songbirds Figure 9.1 Examples of variation in syringeal anatomy. The tracheal parrot syrinx has two syringeal
vitality to our environment, The possibility of should consult reviews of this subject (Nowicki muscles and a pair of lateral tympaniform membranes. The bronchial syrinx of the oilbird has one pair of
its absence due to in c r eas ing levels of & Marler 1988; Suthers 1997, 1999a, b; Gaunt syringeal muscles and a pair of medial and lateral tympaniform membranes in each bronchus. Songbirds
environmental toxins, so eloquently described & Nowicki 1998; Doupe 81Kuhl 1999; Suthers have several pairs of syringeal muscles in their tracheobronchial syrinx. Tr, trachea; ST, sternotrachealis
by Rachel Carson (1962) in her goundbreaking ef ale 1999; Goller & Larsen 2002)- muscle; SY SUP, superficial syringeal muscle; SY PROF, deep syringeal muscle; SY VALV, pneumatic
valve; LTM, lateral tympaniform me~brane;MTM, medial tympaniform membrane; BCI, first bronchial
book, Silent Spring, was a potent factor in cartilage; SY, syringeal muscle; ML, medial labium; LL, lateral labium; SYR, muscles of syrinx. (Oilbird
mobilizing public opinion to become better modified after Suthers & Hector 1985; parrot and canary modified after King 1989).
stewards of our environment. Scientifically, the
highly developed vocal communication of this AN ORGAN FOR SINGING
The Respiratory Rhythm direction through the syrinx and into the
group, sharing as it does a number of parallels
with speech, has made songbirds especially suited The avian vocal organ, the syrinx, is located The timing of vocalization and the tempo of esophagus. After each coo there is a short
for the study of complex, learned vocal deep in the chest in an air sac that is connected song begins with the respiratory cycle (Vicario expiration followed by a brief inspiration (Gaunt
communication at every level, from its molecular to other air sacs and the lungs. It is present in all 1991). Expiratory muscles compress the bellows- et al. 1982).
biology to its evolution. birds except vultures, but its exact location and like air sacs in the thorax and abdomen, increasing
This chapter is about how birdsong is structure varies considerably (King 1989). In respiratory pressure so air flows out through the
produced. How do birds generate such a variety some species, including doves, pigeons and syrinx and trachea (Hartley 1990). During song, The Source of Sound
of sounds and what vocal gymnastics are required parrots, the syrinx consists of modified rings of a small inspiration called a 'mini-breath' (Calder It is now generally agreed that vocalizations are
to produce them?The vocal system is the interface cartilage near the lower end of trachea 1970) is usually taken after each syllable, except generated by airflow-induced oscillation of
between the bird's brain and his song. Knowing (Fig. 9.1). at very high syllable repetition rates (Hartley & elements in the wall of the syrinx that convert
how it functions, and coming to appreciate its In other species, such as penguins, many owls, Suthers 1989). Inspiratory muscles expand the some of the air's kinetic energy into acoustic
limitations as well as its capabilities, is important nightjars, and some cuckoos, the syrinx thorax and abdomen, reducing pressure in the energy. These oscillations are presumably
in understanding vocal communication. Birdsong exists as two semisyrinxes located in the primary air sacs and reversing the direction of airflow sustained by interaction between Bernoulli forces
is the product of the carefully coordinated activity bronchi (Fig. 9.1). In many birds, however, through the syrinx (Wild et al. 1998). and the inertia of air in the vocal tract, in much
of many different muscles. Some of these are including but by no means limited to the oscine Vocalization occurs during expiratory airflow, the same way as are oscillations of human vocal
associated with the vocal organ but many belong songbirds, the syrinx is at the junction where aside from rare exceptions when it occurs during folds (Titze 1994; Gardner et al. 200 1). An
to other muscle groups such as the respiratory the two primary bronchi join to form the trachea. inspiration, as in the 'wah' sounds of doves (Gaunt alternative mechanism for sound production
system and the vocal tract. Together they convert This tracheobronchial syrinx includes modified I et al. 1982), and certain syllables in the songs of based on the principle of an aerodynamic whistle
nerve impulses from the brain into song. cartilages from both the upper end of each some zebra finches (Goller & Daley 2001). Most was put forth (Gaunt et al. 1982) to explain
Less is known about the vocal mechanisms of bronchus and the lower end of the trachea birds sing with their beaks open, but doves and pure tone vocalizations in birds such as doves.
non-songbirds, and they will be mentioned only (Fig. 9.1). pigeons coo with their beaks and nares closed. However, this aerodynamic whistle hypothesis
Sound is produced during airflow in an expiratory is not supported by the observation of
248
274 Nature's Music How birds sing and why it matters 275
I
L
tympaniform membrane vibration in pigeons the syrinx to control vocalization receives input
from.the brain by the hypoglossal nerve. In sohe
( BOX 36
(Goller & Larsen 1997a), or by experiments with
collared doves vocalizing in light gas mixture species, such as swiftlets, there are no intrinsic
(Ballintijn & ten Cate 1998). In both collared syringeal muscles, meaning muscles having both I AVIAN SONAR: CLICKING IN THE DARK I
Two groups of birds, some Aerodramus Asian swiftlets (Medway 1967; Griffin & Suthers 1970; Griffin & Thompson 1982; Coles
doves and ringdoves, there is evidence that the their origin and insertion on the syrinx (Box et al. 1987) and the neotropical oilbird (Griffin 1953; Konishi & Knudsen 1979) make clicking sounds with their syrinx that enable
sound generated in their syrinx contains 36, p. 275). In these birds, syringeal function them to navigate in the dark by echolocation. Although swiftlets and oilbirds have quite different lifestyles, their sonar allows them
during vocalization is controlled by muscles that to nest in the relative safety of dark caves. Swiftlets are diurnal insectivores and spend the day on the wing catching insects which
prominent harmonics, which must be filtered they locate visually. Oilbirds are nocturnal frugivores with excellent night-time vision. They spend the daytime resting on ledges
out by the vocal tract to produce the tonal have at least one of their attachments outside in their nesting cave and use vision, supplemented by echolocation on dark nights, to feed on the fruits of tropical trees.
the syrinx. Other birds, such as parrots, have Clicks make good sonar signals because they contain energy over a wide range of frequencies and have a short duration,
properties of the coo (Beckers et al. 2003b). making it easier to hear faint echoes reflected by surrounding objects (CD2 #64). Furthermore, clicks can be produced by an
The anatomical structures in the syrinx that one or more pairs of specialized syringeal muscles. anatomically simple syrinx. Although details of click production differ between oilbirds (Suthers & Hector 1985) and swiftlets
produce the sound differ with the syringeal Doves have two pairs of extrinsic syringeal (Suthers & Hector 1982), in both it involves rapid, synchronous opening and closing of both sides of the syrinx. Each cycle
typically produces a pair of clicks (A). The first click occurs when the syringeal valve, located at the labia and tympaniform
anatomy. Goller and Larsen used a fiberoptic muscles (see Fig. 10.5, p. 310). One, the membranes, briefly vibrates as it narrows the syringeal lumen just before closing it completely. After a short, silent interval while
endoscope to make direct observations of sternotrachealis, reduces the tension across the the valves are closed, a second click occurs as the valve begins to open, again allowing airflow to vibrate the membranes until
syrinx and allows the lateral tympaniform they are withdrawn from the syringeal air stream.
syringeal motion during vocalizations elicited Swiftlets have no intrinsic syringeal muscles so they operate their syringeal valves by using extrinsic muscles in sequence
by stimulating vocal control regions in the brain membranes to fold into the tracheal lumen where to first close the valve by pulling the tracheobronchial syrinx towards'the bronchi (accomplishedby the sternotrachealis muscles)
of anesthetized birds (Box 37, p. 277). In the airflow causes them to vibrate. Contraction of and then opening the valve by stretching bronchi (through contraction of the tracheolateralis muscles; Suthers & Hector 1982).
The mechanism is similar in oilbirds with two important differences. Oilbirds have a bronchial syrinx (Muller 1841; Garrod 1873;
tracheal syrinx of pigeons they showed that the the other muscle, the tracheolateralis, abducts see p. 273; Fig. 9.1), in which each syringeal%alve (i.e. semisyrinx) is in the primary bronchus and the valve is opened by a
lateral tympaniform membranes (LTMs), on each the tympaniform membranes out of the lumen syringeal muscle that lies along the surface of each bronchus and attaches to a bronchial cartilage bordering the lateral tympanic
membrane (Suthers & Hector 1985). One of the beauties of this simple mechanism for producing sonar clicks is that two sonar
side of the trachea bulge toward each other into and terminates phonation (Goller & Larsen pulses are generated during each cycle of muscle contraction, thus doubling the number of echoes produced.
the tracheal lumen during vocalization where 1997a). The role of these muscles, if any, in Why do some birds have a bronchial syrinx? Oilbirds may provide a clue. A puzzling feature of their syrinx is that the two
frequency modulation of coos is uncertain. Gaunt semisyrinxes are placed at different positions along the bronchus, with the distance from the trachea varying on the right and left
they form a slit-like aperture and vibrate during sides, and between individuals. By inserting a plug in either the left or right bronchus to seal it off from the trachea, we showed
respiratory airflow (Goller & Larsen, 1997a, b; (1988) reported that during a coo both muscles that each bronchus had a quarter wave resonance, like that of a stopped tube, allowing frequencies with a wavelength 4 times
Larsen & Goller 1999). The LTMs (Fig. 9.1) are simultaneously active in ring doves, and the length of the bronchial tube to pass while filtering out other frequencies (Suthers 1994; 6). The distribution of sound energy
as a function of frequency thus depends on the length of the section of rigid bronchus between the trachea and semisyrinx.
also can be seen to fold into the tracheal lumen Beckers et al. (2003~)demonstrated that gradual Because this length differs between the left and right bronchus and varies between individuals, the vocalizations of each oilbird
and vibrate during vocalization in the tracheal frequency modulation in the coo, but not abrupt contain a pair of frequency bands of high intensity sound, or formants that are unique to that individual. Detectable in some
echolocating clicks (Suthers & Hector 1988), these formants are prominent in many of the bird's long duration broadband 'social'
syrinx of the cockatiel (Larsen & Goller 1999, frequency jumps, closely parallels pressure changes vocalizations, and might provide a way for individuals to recognize each other in the dark (CD2 #65). Interestingly, most bronchial
2002). in the interclavicularair sac, suggesting expiratory syrinxes occur in nocturnal species (Suthers 1994).
In songbirds, the relatively well-developed muscles may be important regulators of sound Roderick A. Suthers
Figs. 9.1 & 9.2). However, endoscopic tympaniform membranes or the glottis (Beckers 16
observations of the songbird syrinx during et al. 2 0 0 3 ~ ) .
Songbirds have the most complex syiingeal 8
vocalization reveal that it is the connective tissue
forming the internal and external labia at the musculature of all. They have 5 pairs of muscles 0
anterior end of each bronchus that is adducted that attach to the syrinx and a sikth pair that
acts on it indirectly via its insertion on the trachea RB = right bronchus formant
into the syringeal lumen and vibrates during 4 Click
LB = left bronchus formant
vocalization (Larsen & Goller 1999). Surgical (King 1989). In general, more muscles provide T = tracheal formant '
removal of both MTMs has only a minor effect morgdegrees of freedom in configuring the syrinx 3 Closed' 16
on the vocalization, demonstrating that whatever
u
to produce different sounds (Gaunt 1983). The
their function, they are not essential for song complexity of song is roughly correlated with
production. the number of syringeal muscles, but this T
relationship is not a simple one. Parrots, for 0.5 s
example, succeed in producing complex
Syringeal Muscles Have Separate Functions vocalizations with only two pairs of syringeal
The muscular apparatus capable of acting on muscles (Fig. 9.1A) and the ~Lstralianlyreb&d,
249
274 Nature's Music How birds sing and why it matters 275
I
L
tympaniform membrane vibration in pigeons the syrinx to control vocalization receives input
from.the brain by the hypoglossal nerve. In sohe
( BOX 36
(Goller & Larsen 1997a), or by experiments with
collared doves vocalizing in light gas mixture species, such as swiftlets, there are no intrinsic
(Ballintijn & ten Cate 1998). In both collared syringeal muscles, meaning muscles having both I AVIAN SONAR: CLICKING IN THE DARK I
Two groups of birds, some Aerodramus Asian swiftlets (Medway 1967; Griffin & Suthers 1970; Griffin & Thompson 1982; Coles
doves and ringdoves, there is evidence that the their origin and insertion on the syrinx (Box et al. 1987) and the neotropical oilbird (Griffin 1953; Konishi & Knudsen 1979) make clicking sounds with their syrinx that enable
sound generated in their syrinx contains 36, p. 275). In these birds, syringeal function them to navigate in the dark by echolocation. Although swiftlets and oilbirds have quite different lifestyles, their sonar allows them
during vocalization is controlled by muscles that to nest in the relative safety of dark caves. Swiftlets are diurnal insectivores and spend the day on the wing catching insects which
prominent harmonics, which must be filtered they locate visually. Oilbirds are nocturnal frugivores with excellent night-time vision. They spend the daytime resting on ledges
out by the vocal tract to produce the tonal have at least one of their attachments outside in their nesting cave and use vision, supplemented by echolocation on dark nights, to feed on the fruits of tropical trees.
the syrinx. Other birds, such as parrots, have Clicks make good sonar signals because they contain energy over a wide range of frequencies and have a short duration,
properties of the coo (Beckers et al. 2003b). making it easier to hear faint echoes reflected by surrounding objects (CD2 #64). Furthermore, clicks can be produced by an
The anatomical structures in the syrinx that one or more pairs of specialized syringeal muscles. anatomically simple syrinx. Although details of click production differ between oilbirds (Suthers & Hector 1985) and swiftlets
produce the sound differ with the syringeal Doves have two pairs of extrinsic syringeal (Suthers & Hector 1982), in both it involves rapid, synchronous opening and closing of both sides of the syrinx. Each cycle
typically produces a pair of clicks (A). The first click occurs when the syringeal valve, located at the labia and tympaniform
anatomy. Goller and Larsen used a fiberoptic muscles (see Fig. 10.5, p. 310). One, the membranes, briefly vibrates as it narrows the syringeal lumen just before closing it completely. After a short, silent interval while
endoscope to make direct observations of sternotrachealis, reduces the tension across the the valves are closed, a second click occurs as the valve begins to open, again allowing airflow to vibrate the membranes until
syrinx and allows the lateral tympaniform they are withdrawn from the syringeal air stream.
syringeal motion during vocalizations elicited Swiftlets have no intrinsic syringeal muscles so they operate their syringeal valves by using extrinsic muscles in sequence
by stimulating vocal control regions in the brain membranes to fold into the tracheal lumen where to first close the valve by pulling the tracheobronchial syrinx towards'the bronchi (accomplishedby the sternotrachealis muscles)
of anesthetized birds (Box 37, p. 277). In the airflow causes them to vibrate. Contraction of and then opening the valve by stretching bronchi (through contraction of the tracheolateralis muscles; Suthers & Hector 1982).
The mechanism is similar in oilbirds with two important differences. Oilbirds have a bronchial syrinx (Muller 1841; Garrod 1873;
tracheal syrinx of pigeons they showed that the the other muscle, the tracheolateralis, abducts see p. 273; Fig. 9.1), in which each syringeal%alve (i.e. semisyrinx) is in the primary bronchus and the valve is opened by a
lateral tympaniform membranes (LTMs), on each the tympaniform membranes out of the lumen syringeal muscle that lies along the surface of each bronchus and attaches to a bronchial cartilage bordering the lateral tympanic
membrane (Suthers & Hector 1985). One of the beauties of this simple mechanism for producing sonar clicks is that two sonar
side of the trachea bulge toward each other into and terminates phonation (Goller & Larsen pulses are generated during each cycle of muscle contraction, thus doubling the number of echoes produced.
the tracheal lumen during vocalization where 1997a). The role of these muscles, if any, in Why do some birds have a bronchial syrinx? Oilbirds may provide a clue. A puzzling feature of their syrinx is that the two
frequency modulation of coos is uncertain. Gaunt semisyrinxes are placed at different positions along the bronchus, with the distance from the trachea varying on the right and left
they form a slit-like aperture and vibrate during sides, and between individuals. By inserting a plug in either the left or right bronchus to seal it off from the trachea, we showed
respiratory airflow (Goller & Larsen, 1997a, b; (1988) reported that during a coo both muscles that each bronchus had a quarter wave resonance, like that of a stopped tube, allowing frequencies with a wavelength 4 times
Larsen & Goller 1999). The LTMs (Fig. 9.1) are simultaneously active in ring doves, and the length of the bronchial tube to pass while filtering out other frequencies (Suthers 1994; 6). The distribution of sound energy
as a function of frequency thus depends on the length of the section of rigid bronchus between the trachea and semisyrinx.
also can be seen to fold into the tracheal lumen Beckers et al. (2003~)demonstrated that gradual Because this length differs between the left and right bronchus and varies between individuals, the vocalizations of each oilbird
and vibrate during vocalization in the tracheal frequency modulation in the coo, but not abrupt contain a pair of frequency bands of high intensity sound, or formants that are unique to that individual. Detectable in some
echolocating clicks (Suthers & Hector 1988), these formants are prominent in many of the bird's long duration broadband 'social'
syrinx of the cockatiel (Larsen & Goller 1999, frequency jumps, closely parallels pressure changes vocalizations, and might provide a way for individuals to recognize each other in the dark (CD2 #65). Interestingly, most bronchial
2002). in the interclavicularair sac, suggesting expiratory syrinxes occur in nocturnal species (Suthers 1994).
In songbirds, the relatively well-developed muscles may be important regulators of sound Roderick A. Suthers
Figs. 9.1 & 9.2). However, endoscopic tympaniform membranes or the glottis (Beckers 16
observations of the songbird syrinx during et al. 2 0 0 3 ~ ) .
Songbirds have the most complex syiingeal 8
vocalization reveal that it is the connective tissue
forming the internal and external labia at the musculature of all. They have 5 pairs of muscles 0
anterior end of each bronchus that is adducted that attach to the syrinx and a sikth pair that
acts on it indirectly via its insertion on the trachea RB = right bronchus formant
into the syringeal lumen and vibrates during 4 Click
LB = left bronchus formant
vocalization (Larsen & Goller 1999). Surgical (King 1989). In general, more muscles provide T = tracheal formant '
removal of both MTMs has only a minor effect morgdegrees of freedom in configuring the syrinx 3 Closed' 16
on the vocalization, demonstrating that whatever
u
to produce different sounds (Gaunt 1983). The
their function, they are not essential for song complexity of song is roughly correlated with
production. the number of syringeal muscles, but this T
relationship is not a simple one. Parrots, for 0.5 s
example, succeed in producing complex
Syringeal Muscles Have Separate Functions vocalizations with only two pairs of syringeal
The muscular apparatus capable of acting on muscles (Fig. 9.1A) and the ~Lstralianlyreb&d,
250
276 Nature's Music How birds sing and why it matters 277
known for its vocal versatility and now generally contraction of these muscles, perhaps assisted
considered to be a songbird (Higgins et al. 200 l), by other muscles (Larsen & Goller 2002),
has but 3 pairs. The remainder of this chapter terminates or prevents phonation on the same
focuses on the oscine songbirds.
The function of syringeal muscles in songbirds
side of the syrinx by moving the labia still closer
together and closing the slit, stopping airflow
I RECORDING FROM THE SYRINX OF A SINGING BIRD
Much can be learned about song production by monitoring respiratory dynamics and the activity of syringeal and
I
has been studied by recording their electrical and arresting labial oscillation. respiratory muscles during song. The bird is anesthetized and a tiny thermistor bead is placed in the lumen of
activity in the form of electromyograms (EMGs), The large ventral syringeal muscles, the each primary bronchus, just below the syrinx (Suthers 1990; Suthers et al. 1994). The thermistors are connected
syringealis ventralis, appear to be involved in to a circuit that maintains them at a constant temperature of about 60C. Air flowing through the bronchus during
while measuring the rate of airflow through each either inspiration or expiration will conduct heat away from the thermistor and the additional current required to
side of the syrinx, and monitoring respiratory regulating sound frequency (Fig. 9.2). The maintain its temperature provides a measure of the rate of airflow through each side of the syrinx.
pressure as the bird moves about freely in its amplitude of their EMG is correlated in a positive The respiratory pressure driving airflow through the syrinx is measured with a miniature piezoresistive pressure
cage and sings spontaneously in front of a way with the fundamental frequency of the transducer mounted on the bird's back andAattachedto a small tube inserted into one of the air sacs. The bird's
microphone (Suthers 1990, 1997; Suthers et al. vocalization (Goller & Suthers 1995b, 1996a). respiratory system is different from that of mammals. Birds have several air sacs that act like bellows pumping
They presumably control fundamental frequency air through the lungs. Air sacs are connected to the lungs and to each other by complex passages and openings.
1994, 1996, 1999; Goller & Suthers 1996a, b; A midline interclavicular air sac connects the two sides of the respiratory system and contains the syrinx, so
Box 37, p. 277). Additional insights into syringed by varying the tension or elasticity of the respiratory pressure should be the same in both bronchi. By measuring the respiratory pressure, inspirations and
function have been obtained by using an oscillating labia. Upward-sweeping frequency expirations can be distinguished. The movement and pokition of the labia on each side of the syrinx can be
endoscope to make direct observations on the modulation (FM) is accompanied by a monitored by using bronchial airflow and respiratory pressure to calculate changes in syringeal resistance. A
motion of syringeal structures in response to corresponding increase in ventral muscle activity, positive air sac pressure with no airflow indicates the labial valve on that side of the syrinx is closed, whereas a
which is also high during high-pitched notes high flow rate with low pressure indicates the labia are drawn back out of the air stream.
electrical stimulation of individual muscles in The function of various syringeal and respiratory muscles can be deduced by inserting a pair of very fine hair-
anesthetized birds (Larsen & Goller 2002). with a high fundamental frequency but little like stainless steel wires into them and recording the relationship of their electromyograms (EMG) while monitoring
Alth~ugh~different muscles may have more than FM. syringeal airflow, respiratory pressure and the acoustic properties of the song at a microphone in front of the bird
one role in sound production and interact in There is great variety in the frequency (Goller & Suthers 1996b). All of these physiological signals are carried by fine wires to a velco tab on the bird's
complex ways that are still poorly understood, structure, or 'syllable morphology' of birdsongs. back where they connect to other wires that travel up through the top of the cage to recording instruments. Birds
Producing such a variety of sounds would seem instrumented in this way are free to move around in their cages and sing spontaneously. At the end of an
two of the most basic aspects of song, its timing experiment, which may last a week or more, the bird is again anesthetized, the apparatus is removed and he is
and its frequency, are primarily controlled by to require very complicated patterns of syringeal returned to the aviary.
the dorsal and ventral syringeal muscles, muscle activity to generate a different vocal Observing syringeal motion directly, Goller and Larsen did some elegant experiments using an angiofiberscope
respectively, acting in conjunction with the gesture for each syllable type, although some to view the mechanical action of the syrinx in response to electrical stimulation of the song control system in the
muscles that mechanically ventilate the respiratory theoretical models of song production suggest brain of anesthetized birds (Goller & Larsen 1997b; Larsen & Goller 2002). A 1.4 mm diameter optical fiber was
the vocal gestures required may be much simpler inserted, either down the trachea to view labial movements, or into the air sac containing the syrinx to see the
system during breathing. muscles on its external surface. In some experiments they electrically stimulated individual muscles, and in
The dorsal muscles, the syringealis dorsalis than previously assumed (Gardner et al. 2001; others (Larsen & Goller 1999) they used a laser optical system to detect the vibration of sound-producing
and the tracheobronchialis dorsalis, play a key Laje et al. 2002; Suthers & Margoliash 2002). structures. Although the vocalizations produced by this method are not normal song, theyeprovide valuable
role in operating a pneumatic valve at the upper information about how sound is produced.
end of each bronchus that controls the timing Sensory Feedback: Fine-tuning the Song Roderick A. Suthers
of phonation on each side of the syrinx (Goller Motor Program i
& Suthers 1995b, 199613). Each valve is formed Location of the transducers in the respiratory system Measuring Syringeal Function
by the medial and lateral labia that are also the To achieve or maintain the optimum respiratory used to study song production. FR, thermistor During Song
structures generating sound (Fig. 9.2). When pressure and the appropriate configuration of measuring airflow through the right side of the syrinx;
FL, thermistor measuring airflow through the left Inspiratory muscles
these dorsal muscles are relaxed during quiet the syrinx during song, the bird needs to monitor
side of the syrinx; P, cannula in the air sac to measure
respiration, and during mini-breaths between these variables as it sings. Using sensory feedback respiratory pressure. EMG electrodes may be placed
syllables, the labia are abducted out of the air to achieve this, it could then adjust the ongoing in syringeal, thoracic inspiratory andlor abdominal
stream. In this position they do not oscillate motor program being sent from the brain to the expiratory muscles. lnspiratory muscles enlarge the
and the syringeal lumen is open, presenting muscles to correct any deviations from the chest, reducing air sac pressure and drawing air
intended vocal gesture.-Such deviations might in. Expiratory muscles compress the chest and force
minimal resistance to airflow. Contraction of
air out through the syrinx.
the dorsal muscles initiates sound production occur as a result of changes in posture or ongoing
by adducting the labia into the lumen where physical activity. We know that auditory feedback
they form a slit and oscillate, as air flows across is essential during song learning (Konishi 1965a),
them (Goller & Suthers 1995b; Goller & Larsen and continues to play a role in the long-term
1997b; Larsen & Goller 1999). Stronger maintenance of adult crystallized song (Nordeen
251
276 Nature's Music How birds sing and why it matters 277
known for its vocal versatility and now generally contraction of these muscles, perhaps assisted
considered to be a songbird (Higgins et al. 200 l), by other muscles (Larsen & Goller 2002),
has but 3 pairs. The remainder of this chapter terminates or prevents phonation on the same
focuses on the oscine songbirds.
The function of syringeal muscles in songbirds
side of the syrinx by moving the labia still closer
together and closing the slit, stopping airflow
I RECORDING FROM THE SYRINX OF A SINGING BIRD
Much can be learned about song production by monitoring respiratory dynamics and the activity of syringeal and
I
has been studied by recording their electrical and arresting labial oscillation. respiratory muscles during song. The bird is anesthetized and a tiny thermistor bead is placed in the lumen of
activity in the form of electromyograms (EMGs), The large ventral syringeal muscles, the each primary bronchus, just below the syrinx (Suthers 1990; Suthers et al. 1994). The thermistors are connected
syringealis ventralis, appear to be involved in to a circuit that maintains them at a constant temperature of about 60C. Air flowing through the bronchus during
while measuring the rate of airflow through each either inspiration or expiration will conduct heat away from the thermistor and the additional current required to
side of the syrinx, and monitoring respiratory regulating sound frequency (Fig. 9.2). The maintain its temperature provides a measure of the rate of airflow through each side of the syrinx.
pressure as the bird moves about freely in its amplitude of their EMG is correlated in a positive The respiratory pressure driving airflow through the syrinx is measured with a miniature piezoresistive pressure
cage and sings spontaneously in front of a way with the fundamental frequency of the transducer mounted on the bird's back andAattachedto a small tube inserted into one of the air sacs. The bird's
microphone (Suthers 1990, 1997; Suthers et al. vocalization (Goller & Suthers 1995b, 1996a). respiratory system is different from that of mammals. Birds have several air sacs that act like bellows pumping
They presumably control fundamental frequency air through the lungs. Air sacs are connected to the lungs and to each other by complex passages and openings.
1994, 1996, 1999; Goller & Suthers 1996a, b; A midline interclavicular air sac connects the two sides of the respiratory system and contains the syrinx, so
Box 37, p. 277). Additional insights into syringed by varying the tension or elasticity of the respiratory pressure should be the same in both bronchi. By measuring the respiratory pressure, inspirations and
function have been obtained by using an oscillating labia. Upward-sweeping frequency expirations can be distinguished. The movement and pokition of the labia on each side of the syrinx can be
endoscope to make direct observations on the modulation (FM) is accompanied by a monitored by using bronchial airflow and respiratory pressure to calculate changes in syringeal resistance. A
motion of syringeal structures in response to corresponding increase in ventral muscle activity, positive air sac pressure with no airflow indicates the labial valve on that side of the syrinx is closed, whereas a
which is also high during high-pitched notes high flow rate with low pressure indicates the labia are drawn back out of the air stream.
electrical stimulation of individual muscles in The function of various syringeal and respiratory muscles can be deduced by inserting a pair of very fine hair-
anesthetized birds (Larsen & Goller 2002). with a high fundamental frequency but little like stainless steel wires into them and recording the relationship of their electromyograms (EMG) while monitoring
Alth~ugh~different muscles may have more than FM. syringeal airflow, respiratory pressure and the acoustic properties of the song at a microphone in front of the bird
one role in sound production and interact in There is great variety in the frequency (Goller & Suthers 1996b). All of these physiological signals are carried by fine wires to a velco tab on the bird's
complex ways that are still poorly understood, structure, or 'syllable morphology' of birdsongs. back where they connect to other wires that travel up through the top of the cage to recording instruments. Birds
Producing such a variety of sounds would seem instrumented in this way are free to move around in their cages and sing spontaneously. At the end of an
two of the most basic aspects of song, its timing experiment, which may last a week or more, the bird is again anesthetized, the apparatus is removed and he is
and its frequency, are primarily controlled by to require very complicated patterns of syringeal returned to the aviary.
the dorsal and ventral syringeal muscles, muscle activity to generate a different vocal Observing syringeal motion directly, Goller and Larsen did some elegant experiments using an angiofiberscope
respectively, acting in conjunction with the gesture for each syllable type, although some to view the mechanical action of the syrinx in response to electrical stimulation of the song control system in the
muscles that mechanically ventilate the respiratory theoretical models of song production suggest brain of anesthetized birds (Goller & Larsen 1997b; Larsen & Goller 2002). A 1.4 mm diameter optical fiber was
the vocal gestures required may be much simpler inserted, either down the trachea to view labial movements, or into the air sac containing the syrinx to see the
system during breathing. muscles on its external surface. In some experiments they electrically stimulated individual muscles, and in
The dorsal muscles, the syringealis dorsalis than previously assumed (Gardner et al. 2001; others (Larsen & Goller 1999) they used a laser optical system to detect the vibration of sound-producing
and the tracheobronchialis dorsalis, play a key Laje et al. 2002; Suthers & Margoliash 2002). structures. Although the vocalizations produced by this method are not normal song, theyeprovide valuable
role in operating a pneumatic valve at the upper information about how sound is produced.
end of each bronchus that controls the timing Sensory Feedback: Fine-tuning the Song Roderick A. Suthers
of phonation on each side of the syrinx (Goller Motor Program i
& Suthers 1995b, 199613). Each valve is formed Location of the transducers in the respiratory system Measuring Syringeal Function
by the medial and lateral labia that are also the To achieve or maintain the optimum respiratory used to study song production. FR, thermistor During Song
structures generating sound (Fig. 9.2). When pressure and the appropriate configuration of measuring airflow through the right side of the syrinx;
FL, thermistor measuring airflow through the left Inspiratory muscles
these dorsal muscles are relaxed during quiet the syrinx during song, the bird needs to monitor
side of the syrinx; P, cannula in the air sac to measure
respiration, and during mini-breaths between these variables as it sings. Using sensory feedback respiratory pressure. EMG electrodes may be placed
syllables, the labia are abducted out of the air to achieve this, it could then adjust the ongoing in syringeal, thoracic inspiratory andlor abdominal
stream. In this position they do not oscillate motor program being sent from the brain to the expiratory muscles. lnspiratory muscles enlarge the
and the syringeal lumen is open, presenting muscles to correct any deviations from the chest, reducing air sac pressure and drawing air
intended vocal gesture.-Such deviations might in. Expiratory muscles compress the chest and force
minimal resistance to airflow. Contraction of
air out through the syrinx.
the dorsal muscles initiates sound production occur as a result of changes in posture or ongoing
by adducting the labia into the lumen where physical activity. We know that auditory feedback
they form a slit and oscillate, as air flows across is essential during song learning (Konishi 1965a),
them (Goller & Suthers 1995b; Goller & Larsen and continues to play a role in the long-term
1997b; Larsen & Goller 1999). Stronger maintenance of adult crystallized song (Nordeen
252
II How birds sing and why it matters 279
I
Songbird Syrinx
Trachea
I SONG IS DISRUPTED BY DELAYED FEEDBACK AND THEN RECOVERS
Songbirds use auditory feedback to both leain and maintain their songs, and the dynamics of this process may
be investigated by manipulating the auditory feedback heard by singing birds. Deafening adult zebra finches
results in a complete loss of auditory feedback and, after roughly four weeks, a marked degradation in song
structure (Nordeen & Nordeen 1992). However, it is difficult, though not impossible (Woolley & Rubel 2002), to
1 restore auditory feedback to deafened birds. In contrast, by using a computer-controlled system to continuously
monitor a bird's vocalizations, specific types of artificial auditory feedback may be generated in real-time, played
back to the bird as he sings, and started and stopped at any moment (Leonardo & Konishi 1999). If the artificial
feedback signal is a 100 ms delay of the bird's song (ABCD) as he sings it, then when the bird sings syllable B
he will hear B+A, with the normal feedback for B superimposed with the delayed feedback for A. Two to four
weeks of exposure to delayed auditory feedback caused a dramatic loss of the spectral and temporal stereotypy
seen in crystallized song, consisting of stuttering, deletion, and distortion of song syllables and the creation of
new ones (CD2 #66). Stuttering sometimes increased the length of a song bout of 2-6 s to more than 60 s.
Restoration of normal feedback enabled the recovery of the original songs of each bird. All of the changes in song
structure gradually became more infrequent, and were eventually replaced by the temporal and spectral organization
characteristic of the original song. A complete recovery took 2-4 months. Thus adult zebra finches appear to
retain a great deal of potential song plasticity, even though they do not normally modify their songs after
I
Bronchus crystallization. Furthermore, despite destabilization of the behavior, a memory of the original song persists, and
can be used to correct the vocal output upon restoration of normal auditory feedback. This indicates that the song
Figure 9.2 A songbird syrinx showing its bipartite structure with 2 sound generators and multiple pairs of is maintained by an active control process which requires both a memorized song model and the auditory
muscles. On the left is a frontal section through a brown thrasher syrinx showing the location of microbead feedback generated during singing. Elucidating the neural mechanisms underlying this feedback control process
thermistors used to sense airflow through each side. The medial and lateral labia form pneumatic valves remains an area of vigorous research.
at the cranial end of each bronchus and oscillate to produce sound. On the right is a ventrolateral view of
the same syrinx depicting the syringeal muscles. vS, ventral syringeal muscle; vTB, ventral tracheobronchialis
muscle; dS, dorsal syringeal muscle; dTB dorsal tracheobronchialis muscle. For other abbreviations see
Figure 9.1. (Modified after Goller & Suthers 1996a).
Normal Song
& Nordeen 1992; Okanoya & Yamaguchi 1997; Similar recordings from syringed muscles indicate
Leonardo & Konishi 1999; Box 38, p. 279). that they also adjust their activity to compensate
Non-auditory feedback from mechano- for unexpected perturbations in subsyringeal
receptors or proprioceptors that may sense the pressure during phonation (Suthers & Wild
tension can provide a rapid response to correct 2000). The sensory receptors that provide the
a parameter that is too high or too low. Air feedback underlying these responses have not
pressure or the relative position of structures in been identified, but likely include
the respiratory and vocal system are also mechanoreceptors or proprioceptors in the
important, even in adult crystallized song. For
example, if the respiratory pressure is momentarily
muscles, air sacs or walls of the respiratory and
vocal systems.
Decrystallized Song with Dropped Syllables and Stuttering I
increased experimentally during a syllable by
injecting a small, randomly timed puff of air Lateral independence of Motor Control
into the air sac of a singing cardinal, it elicits a
compensatory partial relaxation of the abdominal Each side of the syrinx is capable of acting
expiratory muscles that returns the pressure to independently, since it is innervated separately
near its original level (Suthers et al. 2002). Since by the tracheosyringeal branch of the hypoglossal
this muscle response does not require extensive nerve on the same side, which in turn receives
I I
0 Time (s) 4
processing at higher levels of the brain, it can most of its input from the same side of the brain
occur rapidly - often within the same syllable (Nottebohm & Nottebohm 1976; Nottebohm
during which the pressure perturbation occurred. 1977; Vicario & Nottebohm 1988; Wild et al.
253
II How birds sing and why it matters 279
I
Songbird Syrinx
Trachea
I SONG IS DISRUPTED BY DELAYED FEEDBACK AND THEN RECOVERS
Songbirds use auditory feedback to both leain and maintain their songs, and the dynamics of this process may
be investigated by manipulating the auditory feedback heard by singing birds. Deafening adult zebra finches
results in a complete loss of auditory feedback and, after roughly four weeks, a marked degradation in song
structure (Nordeen & Nordeen 1992). However, it is difficult, though not impossible (Woolley & Rubel 2002), to
1 restore auditory feedback to deafened birds. In contrast, by using a computer-controlled system to continuously
monitor a bird's vocalizations, specific types of artificial auditory feedback may be generated in real-time, played
back to the bird as he sings, and started and stopped at any moment (Leonardo & Konishi 1999). If the artificial
feedback signal is a 100 ms delay of the bird's song (ABCD) as he sings it, then when the bird sings syllable B
he will hear B+A, with the normal feedback for B superimposed with the delayed feedback for A. Two to four
weeks of exposure to delayed auditory feedback caused a dramatic loss of the spectral and temporal stereotypy
seen in crystallized song, consisting of stuttering, deletion, and distortion of song syllables and the creation of
new ones (CD2 #66). Stuttering sometimes increased the length of a song bout of 2-6 s to more than 60 s.
Restoration of normal feedback enabled the recovery of the original songs of each bird. All of the changes in song
structure gradually became more infrequent, and were eventually replaced by the temporal and spectral organization
characteristic of the original song. A complete recovery took 2-4 months. Thus adult zebra finches appear to
retain a great deal of potential song plasticity, even though they do not normally modify their songs after
I
Bronchus crystallization. Furthermore, despite destabilization of the behavior, a memory of the original song persists, and
can be used to correct the vocal output upon restoration of normal auditory feedback. This indicates that the song
Figure 9.2 A songbird syrinx showing its bipartite structure with 2 sound generators and multiple pairs of is maintained by an active control process which requires both a memorized song model and the auditory
muscles. On the left is a frontal section through a brown thrasher syrinx showing the location of microbead feedback generated during singing. Elucidating the neural mechanisms underlying this feedback control process
thermistors used to sense airflow through each side. The medial and lateral labia form pneumatic valves remains an area of vigorous research.
at the cranial end of each bronchus and oscillate to produce sound. On the right is a ventrolateral view of
the same syrinx depicting the syringeal muscles. vS, ventral syringeal muscle; vTB, ventral tracheobronchialis
muscle; dS, dorsal syringeal muscle; dTB dorsal tracheobronchialis muscle. For other abbreviations see
Figure 9.1. (Modified after Goller & Suthers 1996a).
Normal Song
& Nordeen 1992; Okanoya & Yamaguchi 1997; Similar recordings from syringed muscles indicate
Leonardo & Konishi 1999; Box 38, p. 279). that they also adjust their activity to compensate
Non-auditory feedback from mechano- for unexpected perturbations in subsyringeal
receptors or proprioceptors that may sense the pressure during phonation (Suthers & Wild
tension can provide a rapid response to correct 2000). The sensory receptors that provide the
a parameter that is too high or too low. Air feedback underlying these responses have not
pressure or the relative position of structures in been identified, but likely include
the respiratory and vocal system are also mechanoreceptors or proprioceptors in the
important, even in adult crystallized song. For
example, if the respiratory pressure is momentarily
muscles, air sacs or walls of the respiratory and
vocal systems.
Decrystallized Song with Dropped Syllables and Stuttering I
increased experimentally during a syllable by
injecting a small, randomly timed puff of air Lateral independence of Motor Control
into the air sac of a singing cardinal, it elicits a
compensatory partial relaxation of the abdominal Each side of the syrinx is capable of acting
expiratory muscles that returns the pressure to independently, since it is innervated separately
near its original level (Suthers et al. 2002). Since by the tracheosyringeal branch of the hypoglossal
this muscle response does not require extensive nerve on the same side, which in turn receives
I I
0 Time (s) 4
processing at higher levels of the brain, it can most of its input from the same side of the brain
occur rapidly - often within the same syllable (Nottebohm & Nottebohm 1976; Nottebohm
during which the pressure perturbation occurred. 1977; Vicario & Nottebohm 1988; Wild et al.
254
280 Nature's Music How birds sing and why it matters 28 1
2000). The implications of this independence types have not yet been identified in songbirds, for that syllable (Fig. 9.3). Many syllables include labia to the other. To the human ear, and when
of right and left sides of the syrinx for vocal they have an important role in controlling many extended upward or downward frequency sweeps viewed as time-frequency sonograms, there is
communication are far reaching, for it gives rhythmic behaviors in other animals (Getting that sometimes exceed 2 octaves. Cardinals are usually no obvious interruption or discontinuity
son g birds the possibilit y of two, virtually 1989; Pearson 1993). unusual in that females also sing, though their in the frequency sweep at the moment of change
independent sound sources. Although both right Whatever the neural mechanism underlying song is less stereotyped and harmonics are more from one side to the other.
and left sides are subjected to similar respiratory syllable production, syllables rather than whole prominent compared to the songs of males This is an extraordinary feat of virtuosity.
pressures generated by the respiratory muscles songs, seem to represent functional units of song (Lemon & Scott 1966; Dittus & Lemon 1969; Consider the sequence of actions that a cardinal
and both share the same supra-syringeal vocal production. Birds normally stop singing after Yamaguchi 1998). must execute with precision to produce even an
tract, the dorsal syringeal muscles can switch finishing a syllable, as opposed to in the middle Except during a rapid trill, each syllable is acoustically simple note lasting less than half a
sound production from one side of the syrinx to of a syllable. If one interrupts the song of a followed by a mini-breath that replenishes second. A single broadband FM sweep from high
the other very rapidly, silencing one side by closing zebra finch by flashing a strobe light, these stops respiratory air. Since both inspiration and to low begins with: (i) closure of the left syrinx;
it while holding the labia on the opposite side are almost always between syllables (Cynx 1990; expiration are active processes, each syllable- (ii) expiratory muscle contraction; (3) opening
in a phonatory position. Likewise, the ventral Franz & Goller 2002). The stereotypy of syllables mini-breath cycle is accompanied in sequence the right syrinx and configuring it to produce
syringeal muscles can independently vary the is reflected in the stereotypy of the pattern of by contraction first of the abdominal expiratory the first portion of the sweep; (iv) closing the
fundamental frequency generated on each side. expiratory pressure and airflow that accompanies muscles (Hartley 1990), followed immediately ?right syrinx while (v) opening the left syrinx
Songbirds have taken full advantage of this lateral each syllable (Allan & Suthers 1994; Suthers et by the thoracic inspiratory muscles (Wild et al. and configuring it to continue the sweep; (vi)
independence to greatly increase the variety and al. 1996; Franz & Goller 2002). These respiratory 1998). During trills with high syllable repetition closing the left syrinx to terminate phonation;
complexity of their songs (Greenewalt 1968; Stein parameters depend on both the syringeal muscle rates there is not enough time for a mini-breath (vii) relaxing expiratory muscles, and (viii)
1968; Suthers 1999a). activity influencing the aperture of the labial between syllables. Instead, expiratory muscles contracting inspiratory muscles; (ix) opening both
valve in each side of the syrinx and on the activity continue to contract and maintain a positive sides of the syrinx for an inspiratory mini-breath
of the expiratory muscles that compress the air pressure below the syrinx during the entire trill. to replace the air used to produce the syllable.
The Stereotypy of Adult Song sacs and generate the positive subsyringeal One side of the syrinx is kept closed and the Depending on the repetition rate of the syllable,
Most adult songbirds sing a stereotyped repertoire pressure that drives airflow. Each syllable type other side produces the trill by repetitively this entire sequence of events may take place in
in which reiterations of the same syllable type has a relatively invariant pattern of pressure and opening and closing so that each puff of air scores of milliseconds and be repeated with
vary little in their acoustic properties. This airflow produced by its vocal gesture. When zebra generates a syllable (Fig. 9.3). precision up to 16 times per second to produce
acoustic stereotypy, together with the fact that finches copy a syllable they typically also copy To our ears, the long FM sweeps of a cardinal a phrase in the song!
there is, at least sometimes, no immediate change the pattern of respiratory pressure that the tutor sound smooth and continuous. It was therefore
in the vocalization if auditory feedback is removed used to produce that syllable (Franz & Goller surprising when closer examination revealed that
by deafening, is part of the evidence suggesting 2002; Box 42, p. 322). each side of the syrinx contributes to a different LEARNING TO SING
that the pattern of syringeal and respiratory part of the same syllable. Fundamental frequencies
muscle activity required to produce a particular below about 3.5 kHz are consistently sung on Juvenile songbirds must meet many
syllable is stored in the brain as a 'motor program' CARDINAL SONG: A CASE STUDY the left side, and those above this frequency are developmental challenges on their way to
(konishi 1985; Vu et al. 1994). According to sung on the right side (Suthers & Goller 1996, adulthood. Not the least of these is learning to
this hypothesis, the syllables of crystallized song Even relatively simple songs like that of the 1997; Suthers 1997, 1999a; Suthers et al. 1999; sing their species' song. Males must become expert
might be produced by activating the appropriate northern cardinal require skillful coordination Fig. 9.3). The two sides of the cardinal syrinx vocalists before they are a year old if they are to
network of neurons in the brain; these are of syringeal and respiratory muscles. An adult are thus specialized to operate over different have a chance of reproducing during their second
presumably wired together in such a way that male cardinal assembles its songs from a repertoire frequency ranges, facilitating the production of year. How does a young cardinal, of either sex,
they automatically generate the basic pattern of of between 8 and 21 different syllables (Halkin extended FM sweeps that cross the boundary go about learning to execute this 'checklist' of
coordinated muscle contractions, the vocal & Linville 1999). A song, a few seconds long, between these registers, and thus contain actions that are necessary to produce even a simple
gesture, needed to produce that syllable. Auditory typically consists of from one to several of these sequential contributions from each side. A syllable wideband frequency sweep (see Chapter 3)?
or somatosensory feedback may not be required, syllables, each repeated several times to form a that sweeps from 2to 7 kHz, for example, begins
at least in the short term, to produce the basic series of phrases (CD1 #67).There are distinctive in the left syrinx and is switched to the right
vocal gesture, but it could adjust or modify it to patterns of airflow and pressure associated with syrinx midway in the upward sweep. This EcaZ Learning
correct for errors that might otherwise occur if, each syllable type. These patterns reflect the fact sequence of lateralized production is reversed in Cardinals are age-limited, closed-ended learners,
for example, conditions vary in the peripheral that each is produced by a stereotyped motor syllables that sweep from high to low frequencies- and their adult song repertoire is learned during
vocal or respiratory system. Although central pattern coordinating the respiratory and syringeal Most impressively the song generally contains their first year (Lemon & Scott 1966; Dittus &
pattern-generating circuits for different syllable muscles to produce the particular vocal gesture . no hint of this switch from one set of oscillating Lemon 1969; Halkin & Linville 1999). Shortly
255
280 Nature's Music How birds sing and why it matters 28 1
2000). The implications of this independence types have not yet been identified in songbirds, for that syllable (Fig. 9.3). Many syllables include labia to the other. To the human ear, and when
of right and left sides of the syrinx for vocal they have an important role in controlling many extended upward or downward frequency sweeps viewed as time-frequency sonograms, there is
communication are far reaching, for it gives rhythmic behaviors in other animals (Getting that sometimes exceed 2 octaves. Cardinals are usually no obvious interruption or discontinuity
son g birds the possibilit y of two, virtually 1989; Pearson 1993). unusual in that females also sing, though their in the frequency sweep at the moment of change
independent sound sources. Although both right Whatever the neural mechanism underlying song is less stereotyped and harmonics are more from one side to the other.
and left sides are subjected to similar respiratory syllable production, syllables rather than whole prominent compared to the songs of males This is an extraordinary feat of virtuosity.
pressures generated by the respiratory muscles songs, seem to represent functional units of song (Lemon & Scott 1966; Dittus & Lemon 1969; Consider the sequence of actions that a cardinal
and both share the same supra-syringeal vocal production. Birds normally stop singing after Yamaguchi 1998). must execute with precision to produce even an
tract, the dorsal syringeal muscles can switch finishing a syllable, as opposed to in the middle Except during a rapid trill, each syllable is acoustically simple note lasting less than half a
sound production from one side of the syrinx to of a syllable. If one interrupts the song of a followed by a mini-breath that replenishes second. A single broadband FM sweep from high
the other very rapidly, silencing one side by closing zebra finch by flashing a strobe light, these stops respiratory air. Since both inspiration and to low begins with: (i) closure of the left syrinx;
it while holding the labia on the opposite side are almost always between syllables (Cynx 1990; expiration are active processes, each syllable- (ii) expiratory muscle contraction; (3) opening
in a phonatory position. Likewise, the ventral Franz & Goller 2002). The stereotypy of syllables mini-breath cycle is accompanied in sequence the right syrinx and configuring it to produce
syringeal muscles can independently vary the is reflected in the stereotypy of the pattern of by contraction first of the abdominal expiratory the first portion of the sweep; (iv) closing the
fundamental frequency generated on each side. expiratory pressure and airflow that accompanies muscles (Hartley 1990), followed immediately ?right syrinx while (v) opening the left syrinx
Songbirds have taken full advantage of this lateral each syllable (Allan & Suthers 1994; Suthers et by the thoracic inspiratory muscles (Wild et al. and configuring it to continue the sweep; (vi)
independence to greatly increase the variety and al. 1996; Franz & Goller 2002). These respiratory 1998). During trills with high syllable repetition closing the left syrinx to terminate phonation;
complexity of their songs (Greenewalt 1968; Stein parameters depend on both the syringeal muscle rates there is not enough time for a mini-breath (vii) relaxing expiratory muscles, and (viii)
1968; Suthers 1999a). activity influencing the aperture of the labial between syllables. Instead, expiratory muscles contracting inspiratory muscles; (ix) opening both
valve in each side of the syrinx and on the activity continue to contract and maintain a positive sides of the syrinx for an inspiratory mini-breath
of the expiratory muscles that compress the air pressure below the syrinx during the entire trill. to replace the air used to produce the syllable.
The Stereotypy of Adult Song sacs and generate the positive subsyringeal One side of the syrinx is kept closed and the Depending on the repetition rate of the syllable,
Most adult songbirds sing a stereotyped repertoire pressure that drives airflow. Each syllable type other side produces the trill by repetitively this entire sequence of events may take place in
in which reiterations of the same syllable type has a relatively invariant pattern of pressure and opening and closing so that each puff of air scores of milliseconds and be repeated with
vary little in their acoustic properties. This airflow produced by its vocal gesture. When zebra generates a syllable (Fig. 9.3). precision up to 16 times per second to produce
acoustic stereotypy, together with the fact that finches copy a syllable they typically also copy To our ears, the long FM sweeps of a cardinal a phrase in the song!
there is, at least sometimes, no immediate change the pattern of respiratory pressure that the tutor sound smooth and continuous. It was therefore
in the vocalization if auditory feedback is removed used to produce that syllable (Franz & Goller surprising when closer examination revealed that
by deafening, is part of the evidence suggesting 2002; Box 42, p. 322). each side of the syrinx contributes to a different LEARNING TO SING
that the pattern of syringeal and respiratory part of the same syllable. Fundamental frequencies
muscle activity required to produce a particular below about 3.5 kHz are consistently sung on Juvenile songbirds must meet many
syllable is stored in the brain as a 'motor program' CARDINAL SONG: A CASE STUDY the left side, and those above this frequency are developmental challenges on their way to
(konishi 1985; Vu et al. 1994). According to sung on the right side (Suthers & Goller 1996, adulthood. Not the least of these is learning to
this hypothesis, the syllables of crystallized song Even relatively simple songs like that of the 1997; Suthers 1997, 1999a; Suthers et al. 1999; sing their species' song. Males must become expert
might be produced by activating the appropriate northern cardinal require skillful coordination Fig. 9.3). The two sides of the cardinal syrinx vocalists before they are a year old if they are to
network of neurons in the brain; these are of syringeal and respiratory muscles. An adult are thus specialized to operate over different have a chance of reproducing during their second
presumably wired together in such a way that male cardinal assembles its songs from a repertoire frequency ranges, facilitating the production of year. How does a young cardinal, of either sex,
they automatically generate the basic pattern of of between 8 and 21 different syllables (Halkin extended FM sweeps that cross the boundary go about learning to execute this 'checklist' of
coordinated muscle contractions, the vocal & Linville 1999). A song, a few seconds long, between these registers, and thus contain actions that are necessary to produce even a simple
gesture, needed to produce that syllable. Auditory typically consists of from one to several of these sequential contributions from each side. A syllable wideband frequency sweep (see Chapter 3)?
or somatosensory feedback may not be required, syllables, each repeated several times to form a that sweeps from 2to 7 kHz, for example, begins
at least in the short term, to produce the basic series of phrases (CD1 #67).There are distinctive in the left syrinx and is switched to the right
vocal gesture, but it could adjust or modify it to patterns of airflow and pressure associated with syrinx midway in the upward sweep. This EcaZ Learning
correct for errors that might otherwise occur if, each syllable type. These patterns reflect the fact sequence of lateralized production is reversed in Cardinals are age-limited, closed-ended learners,
for example, conditions vary in the peripheral that each is produced by a stereotyped motor syllables that sweep from high to low frequencies- and their adult song repertoire is learned during
vocal or respiratory system. Although central pattern coordinating the respiratory and syringeal Most impressively the song generally contains their first year (Lemon & Scott 1966; Dittus &
pattern-generating circuits for different syllable muscles to produce the particular vocal gesture . no hint of this switch from one set of oscillating Lemon 1969; Halkin & Linville 1999). Shortly
256
282 Nature's Music
How birds sing and why it matters 283
Song of an Adult Cardinal after fledging, juveniles begin to produce muscle actions underlying the motor patterns
sequences of low intensity, highly variable notes that are required to produce adult song based
having little resemblance to adult song. Gradually on an auditory 'template' in the brain, which
during the autumn, and then again early in the may be either innate or shaped by experience
following spring, these vocal ramblings become early in life (Marler 1997). Experiments on zebra
louder and some notes begin to recur in a finches, in which this process of sensorimotor
recognizably similar form. This early stage of integration was temporarily disrupted at different
vocal development, 'subsong,' (Fig. 9.4) is slowly stages of song development by using botulinum
transformed into the next stage, 'plastic song,' toxin to reversibly block transmission of nerve
as a recognizable, though initially poorly impulses to the syringeal muscles, suggest that
controlled, syllable morphology develops in addition to the sensitive period for song
(CDI # 067). Eventually, when the bird is about memorization, there is another one for
10 months old plastic song becomes sensorimotor integration during the latter part
indistinguishable from adult song. Instead of of song development (Pytte & Suthers 2000).
being plastic or malleable, the song becomes +
0.5
(B) Time (s)
Figure 9.3 Phrases from songs of an adult northern cardinal. (A) Two long, downward sweeping FM
syllables followed by faster upward FM syllables. In each syllable, frequencies below about 3.5 or 4.0 kHz
are sung on the left side (L) as indicated by airflow through the left syrinx; higher frequencies are produced
on the right side (R). Absence of flow through either side during a positive respiratory pressure indicates
that the labial valve is closed, preventing both airflow and sound production on that side. Note the mini-
breaths between each syllable and the stereotypical patterns of airflow and pressure that are characteristic 0.5
of each syllable type. (9)Another syllable with a low frequency fundamental produced entirely on the left Time (s)
side (L). This is followed by a trill using pulsatile expiration instead of mini-breaths. Expiratory pressure
remains positive during the entire phrase. FL and FR show the rate of airflow through the left and right Figure 9.4 Subsong from a 92-day-old juvenile cardinal. Arrows indicate periods of expiration- without
syrinx, respectively, P is the pressure in the cranial thoracic air sac. Horizontal lines indicate zero airflow sound production. Abbreviations as in Figure 9.3. In those marked B, both sides contribute.
and ambient air pressure. Shaded portions of the airflow are inspiratory mini-breaths.
257
282 Nature's Music
How birds sing and why it matters 283
Song of an Adult Cardinal after fledging, juveniles begin to produce muscle actions underlying the motor patterns
sequences of low intensity, highly variable notes that are required to produce adult song based
having little resemblance to adult song. Gradually on an auditory 'template' in the brain, which
during the autumn, and then again early in the may be either innate or shaped by experience
following spring, these vocal ramblings become early in life (Marler 1997). Experiments on zebra
louder and some notes begin to recur in a finches, in which this process of sensorimotor
recognizably similar form. This early stage of integration was temporarily disrupted at different
vocal development, 'subsong,' (Fig. 9.4) is slowly stages of song development by using botulinum
transformed into the next stage, 'plastic song,' toxin to reversibly block transmission of nerve
as a recognizable, though initially poorly impulses to the syringeal muscles, suggest that
controlled, syllable morphology develops in addition to the sensitive period for song
(CDI # 067). Eventually, when the bird is about memorization, there is another one for
10 months old plastic song becomes sensorimotor integration during the latter part
indistinguishable from adult song. Instead of of song development (Pytte & Suthers 2000).
being plastic or malleable, the song becomes +
0.5
(B) Time (s)
Figure 9.3 Phrases from songs of an adult northern cardinal. (A) Two long, downward sweeping FM
syllables followed by faster upward FM syllables. In each syllable, frequencies below about 3.5 or 4.0 kHz
are sung on the left side (L) as indicated by airflow through the left syrinx; higher frequencies are produced
on the right side (R). Absence of flow through either side during a positive respiratory pressure indicates
that the labial valve is closed, preventing both airflow and sound production on that side. Note the mini-
breaths between each syllable and the stereotypical patterns of airflow and pressure that are characteristic 0.5
of each syllable type. (9)Another syllable with a low frequency fundamental produced entirely on the left Time (s)
side (L). This is followed by a trill using pulsatile expiration instead of mini-breaths. Expiratory pressure
remains positive during the entire phrase. FL and FR show the rate of airflow through the left and right Figure 9.4 Subsong from a 92-day-old juvenile cardinal. Arrows indicate periods of expiration- without
syrinx, respectively, P is the pressure in the cranial thoracic air sac. Horizontal lines indicate zero airflow sound production. Abbreviations as in Figure 9.3. In those marked B, both sides contribute.
and ambient air pressure. Shaded portions of the airflow are inspiratory mini-breaths.
258
284 Nature's Music How birds sing and why it matters 285
development. During early subsong and even or even the duration of individual notes might unclear how often this limit is actually reached,
or achieve independent control of their two
into plastic song, expiration sometimes continues be limited by the volume of air that can be though this may sometimes occur (see Chapter
syringeal sound sources? Are there bottlenecks
in the process underlying motor development into or through the silent intervals between notes exhaled to produce them. This possible limitation 11).
(Fig. 9.4, arrows). is particularly acute in small birds that sing big The syllable repetition rate that requires a
and coordination that limit song acquisition?
As subsong progresses, there is a tendency to songs. Canaries are small birds that can sing in bird to switch from mini-breaths to pulsatile
Does the progression through successive stages
alternate sides and to produce a pair of notes, a seemingly continuous fashion for up to about expiration is probably determined by the physical
of juvenile song correspond to the mastery of
one from each side, during a single expiration. a minute. During a single song they may produce properties of respiratory structures - particularly
specific hurdles in vocal production? To better
This foreshadows the adult motor pattern of scores or hundreds of syllables, yet their resting the mass and elasticity of the thorax and abdomen
understand the motor development of song, we
using the two sides in sequence to make a single respiratory tidal volume is only a fraction of a - that must oscillate at the syllable repetition
studied its production in juvenile male cardinals
continuous frequency sweep. But in early juvenile milliliter! Calder (1970) showed that the depth rate to drive ventilation. Small birds can sustain
at various stages of vocal learning (Suthers &
song the pair of notes within a single expiration of the thorax increases slightly between most a mini-breath respiratory pattern at higher syllable
Goller 1998a, b).
are not joined, nor is there any clear, consistent notes in a canary song. From this he hypothesized repetition rates than can large birds. The 'cut-
Even young. nestlings may produce different
coordinated sequence in their frequency patterns. that the bird takes a mini-breath between syllables, off' repetition rate for mini-breaths for an 18 g
sounds simultaneously on each side of the syrinx.
The linking of a high frequency sound from the to replenish its air supply. Measurements of canary is about 30 syllables per second compared
Two-voice elements in the begging calls of young
chaffinches indicate that separate sounds are being right with a lower frequency contribution from respiratory pressure and tracheal airflow during . to about 16 syllables per second for a 35-40 g
the left side of the syrinx to form a coordinated, song showed that each mini-breath replaces about cardinal. Pulsatile expiration is not subject to
generated on each side of the syrinx in nestlings
seamless frequency sweep occurs in the latter the same volume of air that is exhaled to produce this limit since it requires little movement of
only a few days old (Nottebohm 1971b;
Wilkinson 1980). It is not clear, however, if these part of song development. J
the syllable (Hartley & Suthers 1989). As a result, the body wall, and the labia, whose valve-like
two-voice vocalizations are simply a passive there is probably little or no net change in action gates airflow, have little mass.
byproduct of poorly controlled airfl.ow through respiratory volume during mini-breath song. Some zebra finches attempt to get the best of
both sides of the syrinx, or if they are under PUSHING THE ACOUSTIC The mini-breath respiratory pattern is not both worlds by vocalizing during some mini-
active control of the syringeal muscles. ENVELOPE unique to songbirds. It appears to be used by breaths between 'normal' expiratory syllables
In cardinals, the ability to control which side many species, including non-songbirds as diverse (Goller & Daley 2001). These inspiratory syllables
of the syrinx produces sound is present in In adult song, the respiratory and syringeal as parrots and oilbirds (Suthers 2001). In our have a distinctive high fundamental frequency
fledglings singing subsong when they are less muscles act together like experienced partners own species, shortened inspirations, analogous and are copied by young zebra finches exposed
than 6 weeks old. Juveniles at this age are already in a dance choreographed by the brain, but the t o mini-breaths, are also present during to them in tutor songs. It is not known if
capable of unilateral sound production by closing dance steps differ between species. The possession conversational speech (Hixon 1973). In all these inspiratory syllables have special perceptual
one or the other side of their syrinx, as do adults. of two parallel sound sources, each with its own cases brief inspirations permit a continuity of significance in communication or why more birds
Sometimes both sides of the syrinx produce sound pneumatic valve and neural control, gives oscine vocal communication that is free from periodic don't use them. It is possible that they may be
at the same time, but often a note is generated songbirds an additional degree of freedom absent interruptions by long inspirations. energetically expensive to produce since the labia
only on the left or the right side, the silent side in species that have a tracheal syrinx or that lack must remain in the air stream during inspiration,
being closed to airflow (Fig. 9.4). Unlike adult independent control of the &o sides of the Increasing Syllable Repetition Rate resulting in a higher resistance to airflow and
song, each note is quite variable and the frequency bronchotracheal or bronchial syrinx. Studies of Although mini-breaths allow a bird to sing longer presumably requiring more effort from the
structure of successive left and right notes is not how different species use the two sides of their songs, they limit the song's tempo to rates of inspiratory muscles.
coordinated or linked (Suthers & Goller 1998a). syrinx during song suggest that the distinctive syllable repetition having silent intervals long
When adult cardinals sing, the timing of styles of singing among various songbird taxa enough to accommodate an inspiration. Some
birds get around this constraint by switching to Increasing Spectral Diversity
expiratory airflow is carefully coordinated with are achieved by using their duplex vocal organ
vocalization. Each syllable typically begins as soon in different ways. These may optimize particular a different respiratory technique involving Bandwidth: Left and Right Voral Registers
as the syringeal valve opens and ends when the acoustic effects at the cost of sacrificing virtuosity pulsatile expiration, described above for cardinals, Songbirds use the two sides of their syrinx in
valve closes (Fig. 9.3). As a result, nearly all of in other aspects of sound production (Suthers which does not require the reversal of airflow different ways to extend the bandwidth of their
the exhaled air contributes to phonation and 1997, 1999a, b; Suthers & Goller 1997). between syllables. Pulsatile expiration can almost song or to produce syllables that vary widely in
very little is 'wasted' between syllables, even double the maximum syllable repetition rate that their tonal quality. The range of frequencies a
du;ing the fast pulsatile trills. The respiratory- is possible with mini-breaths. This increase in song encompasses, its bandwidth, is increased
Longer Song Bouts Versus Faster Tempos syllable repetition rate is achieved at the cost of by having each side of the syrinx specialized to
syringeal coordination necessary for this efficient
;se of air during song is not present at the Increasing Song Dlnration depleting the volume of respiratory air available cover a different frequency band. In all species
beginning of juvenile song, and emerges only Respiratory air is an essential, but potentially for sound production. This limits the length of studied, the fundamental bandwidth of the right
gradually during the earl y stages of song limited, resource for vocalization. Sustained song phrases based on pulsatile expiration, but it is side is shifted upward compared to that of the
259
284 Nature's Music How birds sing and why it matters 285
development. During early subsong and even or even the duration of individual notes might unclear how often this limit is actually reached,
or achieve independent control of their two
into plastic song, expiration sometimes continues be limited by the volume of air that can be though this may sometimes occur (see Chapter
syringeal sound sources? Are there bottlenecks
in the process underlying motor development into or through the silent intervals between notes exhaled to produce them. This possible limitation 11).
(Fig. 9.4, arrows). is particularly acute in small birds that sing big The syllable repetition rate that requires a
and coordination that limit song acquisition?
As subsong progresses, there is a tendency to songs. Canaries are small birds that can sing in bird to switch from mini-breaths to pulsatile
Does the progression through successive stages
alternate sides and to produce a pair of notes, a seemingly continuous fashion for up to about expiration is probably determined by the physical
of juvenile song correspond to the mastery of
one from each side, during a single expiration. a minute. During a single song they may produce properties of respiratory structures - particularly
specific hurdles in vocal production? To better
This foreshadows the adult motor pattern of scores or hundreds of syllables, yet their resting the mass and elasticity of the thorax and abdomen
understand the motor development of song, we
using the two sides in sequence to make a single respiratory tidal volume is only a fraction of a - that must oscillate at the syllable repetition
studied its production in juvenile male cardinals
continuous frequency sweep. But in early juvenile milliliter! Calder (1970) showed that the depth rate to drive ventilation. Small birds can sustain
at various stages of vocal learning (Suthers &
song the pair of notes within a single expiration of the thorax increases slightly between most a mini-breath respiratory pattern at higher syllable
Goller 1998a, b).
are not joined, nor is there any clear, consistent notes in a canary song. From this he hypothesized repetition rates than can large birds. The 'cut-
Even young. nestlings may produce different
coordinated sequence in their frequency patterns. that the bird takes a mini-breath between syllables, off' repetition rate for mini-breaths for an 18 g
sounds simultaneously on each side of the syrinx.
The linking of a high frequency sound from the to replenish its air supply. Measurements of canary is about 30 syllables per second compared
Two-voice elements in the begging calls of young
chaffinches indicate that separate sounds are being right with a lower frequency contribution from respiratory pressure and tracheal airflow during . to about 16 syllables per second for a 35-40 g
the left side of the syrinx to form a coordinated, song showed that each mini-breath replaces about cardinal. Pulsatile expiration is not subject to
generated on each side of the syrinx in nestlings
seamless frequency sweep occurs in the latter the same volume of air that is exhaled to produce this limit since it requires little movement of
only a few days old (Nottebohm 1971b;
Wilkinson 1980). It is not clear, however, if these part of song development. J
the syllable (Hartley & Suthers 1989). As a result, the body wall, and the labia, whose valve-like
two-voice vocalizations are simply a passive there is probably little or no net change in action gates airflow, have little mass.
byproduct of poorly controlled airfl.ow through respiratory volume during mini-breath song. Some zebra finches attempt to get the best of
both sides of the syrinx, or if they are under PUSHING THE ACOUSTIC The mini-breath respiratory pattern is not both worlds by vocalizing during some mini-
active control of the syringeal muscles. ENVELOPE unique to songbirds. It appears to be used by breaths between 'normal' expiratory syllables
In cardinals, the ability to control which side many species, including non-songbirds as diverse (Goller & Daley 2001). These inspiratory syllables
of the syrinx produces sound is present in In adult song, the respiratory and syringeal as parrots and oilbirds (Suthers 2001). In our have a distinctive high fundamental frequency
fledglings singing subsong when they are less muscles act together like experienced partners own species, shortened inspirations, analogous and are copied by young zebra finches exposed
than 6 weeks old. Juveniles at this age are already in a dance choreographed by the brain, but the t o mini-breaths, are also present during to them in tutor songs. It is not known if
capable of unilateral sound production by closing dance steps differ between species. The possession conversational speech (Hixon 1973). In all these inspiratory syllables have special perceptual
one or the other side of their syrinx, as do adults. of two parallel sound sources, each with its own cases brief inspirations permit a continuity of significance in communication or why more birds
Sometimes both sides of the syrinx produce sound pneumatic valve and neural control, gives oscine vocal communication that is free from periodic don't use them. It is possible that they may be
at the same time, but often a note is generated songbirds an additional degree of freedom absent interruptions by long inspirations. energetically expensive to produce since the labia
only on the left or the right side, the silent side in species that have a tracheal syrinx or that lack must remain in the air stream during inspiration,
being closed to airflow (Fig. 9.4). Unlike adult independent control of the &o sides of the Increasing Syllable Repetition Rate resulting in a higher resistance to airflow and
song, each note is quite variable and the frequency bronchotracheal or bronchial syrinx. Studies of Although mini-breaths allow a bird to sing longer presumably requiring more effort from the
structure of successive left and right notes is not how different species use the two sides of their songs, they limit the song's tempo to rates of inspiratory muscles.
coordinated or linked (Suthers & Goller 1998a). syrinx during song suggest that the distinctive syllable repetition having silent intervals long
When adult cardinals sing, the timing of styles of singing among various songbird taxa enough to accommodate an inspiration. Some
birds get around this constraint by switching to Increasing Spectral Diversity
expiratory airflow is carefully coordinated with are achieved by using their duplex vocal organ
vocalization. Each syllable typically begins as soon in different ways. These may optimize particular a different respiratory technique involving Bandwidth: Left and Right Voral Registers
as the syringeal valve opens and ends when the acoustic effects at the cost of sacrificing virtuosity pulsatile expiration, described above for cardinals, Songbirds use the two sides of their syrinx in
valve closes (Fig. 9.3). As a result, nearly all of in other aspects of sound production (Suthers which does not require the reversal of airflow different ways to extend the bandwidth of their
the exhaled air contributes to phonation and 1997, 1999a, b; Suthers & Goller 1997). between syllables. Pulsatile expiration can almost song or to produce syllables that vary widely in
very little is 'wasted' between syllables, even double the maximum syllable repetition rate that their tonal quality. The range of frequencies a
du;ing the fast pulsatile trills. The respiratory- is possible with mini-breaths. This increase in song encompasses, its bandwidth, is increased
Longer Song Bouts Versus Faster Tempos syllable repetition rate is achieved at the cost of by having each side of the syrinx specialized to
syringeal coordination necessary for this efficient
;se of air during song is not present at the Increasing Song Dlnration depleting the volume of respiratory air available cover a different frequency band. In all species
beginning of juvenile song, and emerges only Respiratory air is an essential, but potentially for sound production. This limits the length of studied, the fundamental bandwidth of the right
gradually during the earl y stages of song limited, resource for vocalization. Sustained song phrases based on pulsatile expiration, but it is side is shifted upward compared to that of the
260
386 Nature's Music How birds sing and why it matters 287
left (Suthers 1999a), but there is often substantial by as much as 40 dB (Okanoya & Dooling 1985; prominent in songs of some mimic thrushes such capacity of the two sides of the syrinx to function
overlap in the frequencies produced by each side. Gleich et al. 1994a, b; Gleich & Klump 1994). as grey catbirds and brown thrashers (Suthers independentl y during sound production,
Cardinals are unique, among the small sample Their strong left syringeal dominance may well 1992; Suthers et al. 1994, 1996; Goller & Suthers inharmonic vocalizations can also result from
of species studied, for their limited overlap be the result of being partially deaf to most of 1995a, b, 1996a, by 1999). T h e separate, the nonlinear interaction between the sounds
between frequencies produced on each side and the sounds from the right syrinx. simultaneous 'voices' that form the distinctive generated on each side. The 'dee' syllable of the
their habit of combining sound from the two In the family FringilZidae, which includes inharmonic elements in these songs (Fig. 9.5; black-capped chickadee's song consists of a series
sides to produce a single FM note. The anatomical sparrows, cardinals and Darwin's finches, there CD I # 68) tend to differ less from each other in of overlapping frequency components that differ
or physiological basis for these separate left-right is an inverse relationship between syllable their fundamental frequency than when each from those produced by either side of the syrinx
vocal registers is not known. In many songbirds bandwidth and repetition rate - rapid trills have side sings alone, suggesting that simultaneous alone. It is not the sum of the sounds produced
there are anatomical asymmetries between the narrower bandwidths (Podos 1996; see Chapter two-voice phonation comes with constraints on on each side afier unilateral syringed denervation.
two primary bronchi that might include subtle 11). It is suggested that syllable bandwidth might the degree of frequency separation between the It appears instead to be composed of the sum
differences in the labia or muscles on each side, be constrained at high repetition rates by the two sides that is possible when they operate and difference heterodyne frequencies generated
as yet unidentified. Whatever its mechanism, time required to change the shape and tuning of simultaneously. This limitation might be due to by nonlinear interactions between the sound
the increased overall bandwidth available for the mouth and throat to track the rapidly biomechanical interactions between the two produced by each side of the syrinx (Nowicki &
vocalization increases the possibilities for spectral changing fundamental frequency (Nowicki et halves of the syrinx or to a need to keep both Capranica 1986a, b). The coupling between left
diversity in songs. al. 1992; Podos 1997, 2001). In many cases fundamentals within the frequency pass band and right sound generators might be mediated
T h e influence on song of lateralized beak opening is correlated with sound frequency of the vocal tract, with which both sides connect. by air pressure fluctations associated with sound
specialization for particular frequency ranges is and appears to be a factor in tuning resonances Whereas two-voice syllables depend on the or transmitted mechanically through the syringed
evident in different strains of canaries. One of of the vocal tract by altering its effective length
these is the Waterschlager canary, which was and suppressing harmonics. Podos argues that Brown Thrasher Song
inbred for its distinctive low-pitched song. beak size and syllable repetition rate have evolved
Another strain is the outbred domestic canary. together in Darwin's finches. He suggests that
Nottebohm and Nottebohm (1976) conducted the relatively low syllable repetition rate of finches
a series of experiments on Waterschlagers in which with large beaks is due to the difficulty of moving
either the left or right side of the syrinx had larger mandibles rapidly to quickly change the
been denervated by cutting the tracheosyringeal tuning of the vocal tract filter (Podos 2001).
nerve on one side. They demonstrated that this The role of vocal tract resonance and beak
strain produces about 90% of its syllable movements in birdsong is covered in Chapter
repertoire in its left syrinx. As a result of the 11 (Nowicki 1987; Westneat et al. 1993;
strong lateral dominance of the left side, most Moriyama & Okanoya 1996; Suthers & Goller
of this bird's song is below 3 kHz, except for an 1997; Suthers 1999a; Suthers et al. 1999; Hoese
occasional right side syllable that has a median et al. 2000; Williams 2001).
frequency range between 2.5 and 4.2 kHz
(Nottebohm & Nottebohm 1976). This left Inharmonic Sounds
dominance is in contrast to song of the outbred The songbird syrinx is well suited to achieve
domestic strain in which each side of the syrinx dissonant sounds by singing two overlapping
contributes about an equal proportion of the notes that are different in pitch but not
repertoire, and some syllables include notes from harmonicall y related to each other. T h e
both sides (Suthers et al. 200 1). In these domestic widespread occurrence of these 'two-voice' notes
birds, syllables frequently have fundamental and their likely origin from separate sides of the
frequencies as high as 5 or 6 kHz, and a single syrinx was pointed out by Greenewalt (1968)
bilaterally generated syllable may have a and Stein (1968). Two-voice syllables require
bandwidth of 6 or 7 kHz compared to 3 or 4 airflow through both sides of the syrinx, while
kHz in the case of unilaterally produced syllables. different motor programs to the muscles on each 0.2
side generate unrelated sounds that can be Time (s)
Subsequent studies show that Waterschlager
canaries have an inherited auditory defect that independently modulated in frequency and Figure 9.5 Brown thrasher song illustrating inharmonic, two-voice syllables (CDI #68). V is the time
decreases their sensitivity to sounds above 2 kHz amplitude. Two voice syllables are especially waveform of vocalization. Abbreviations as in Figure 9.3. Modified from Suthers et al. 1994.
261
386 Nature's Music How birds sing and why it matters 287
left (Suthers 1999a), but there is often substantial by as much as 40 dB (Okanoya & Dooling 1985; prominent in songs of some mimic thrushes such capacity of the two sides of the syrinx to function
overlap in the frequencies produced by each side. Gleich et al. 1994a, b; Gleich & Klump 1994). as grey catbirds and brown thrashers (Suthers independentl y during sound production,
Cardinals are unique, among the small sample Their strong left syringeal dominance may well 1992; Suthers et al. 1994, 1996; Goller & Suthers inharmonic vocalizations can also result from
of species studied, for their limited overlap be the result of being partially deaf to most of 1995a, b, 1996a, by 1999). T h e separate, the nonlinear interaction between the sounds
between frequencies produced on each side and the sounds from the right syrinx. simultaneous 'voices' that form the distinctive generated on each side. The 'dee' syllable of the
their habit of combining sound from the two In the family FringilZidae, which includes inharmonic elements in these songs (Fig. 9.5; black-capped chickadee's song consists of a series
sides to produce a single FM note. The anatomical sparrows, cardinals and Darwin's finches, there CD I # 68) tend to differ less from each other in of overlapping frequency components that differ
or physiological basis for these separate left-right is an inverse relationship between syllable their fundamental frequency than when each from those produced by either side of the syrinx
vocal registers is not known. In many songbirds bandwidth and repetition rate - rapid trills have side sings alone, suggesting that simultaneous alone. It is not the sum of the sounds produced
there are anatomical asymmetries between the narrower bandwidths (Podos 1996; see Chapter two-voice phonation comes with constraints on on each side afier unilateral syringed denervation.
two primary bronchi that might include subtle 11). It is suggested that syllable bandwidth might the degree of frequency separation between the It appears instead to be composed of the sum
differences in the labia or muscles on each side, be constrained at high repetition rates by the two sides that is possible when they operate and difference heterodyne frequencies generated
as yet unidentified. Whatever its mechanism, time required to change the shape and tuning of simultaneously. This limitation might be due to by nonlinear interactions between the sound
the increased overall bandwidth available for the mouth and throat to track the rapidly biomechanical interactions between the two produced by each side of the syrinx (Nowicki &
vocalization increases the possibilities for spectral changing fundamental frequency (Nowicki et halves of the syrinx or to a need to keep both Capranica 1986a, b). The coupling between left
diversity in songs. al. 1992; Podos 1997, 2001). In many cases fundamentals within the frequency pass band and right sound generators might be mediated
T h e influence on song of lateralized beak opening is correlated with sound frequency of the vocal tract, with which both sides connect. by air pressure fluctations associated with sound
specialization for particular frequency ranges is and appears to be a factor in tuning resonances Whereas two-voice syllables depend on the or transmitted mechanically through the syringed
evident in different strains of canaries. One of of the vocal tract by altering its effective length
these is the Waterschlager canary, which was and suppressing harmonics. Podos argues that Brown Thrasher Song
inbred for its distinctive low-pitched song. beak size and syllable repetition rate have evolved
Another strain is the outbred domestic canary. together in Darwin's finches. He suggests that
Nottebohm and Nottebohm (1976) conducted the relatively low syllable repetition rate of finches
a series of experiments on Waterschlagers in which with large beaks is due to the difficulty of moving
either the left or right side of the syrinx had larger mandibles rapidly to quickly change the
been denervated by cutting the tracheosyringeal tuning of the vocal tract filter (Podos 2001).
nerve on one side. They demonstrated that this The role of vocal tract resonance and beak
strain produces about 90% of its syllable movements in birdsong is covered in Chapter
repertoire in its left syrinx. As a result of the 11 (Nowicki 1987; Westneat et al. 1993;
strong lateral dominance of the left side, most Moriyama & Okanoya 1996; Suthers & Goller
of this bird's song is below 3 kHz, except for an 1997; Suthers 1999a; Suthers et al. 1999; Hoese
occasional right side syllable that has a median et al. 2000; Williams 2001).
frequency range between 2.5 and 4.2 kHz
(Nottebohm & Nottebohm 1976). This left Inharmonic Sounds
dominance is in contrast to song of the outbred The songbird syrinx is well suited to achieve
domestic strain in which each side of the syrinx dissonant sounds by singing two overlapping
contributes about an equal proportion of the notes that are different in pitch but not
repertoire, and some syllables include notes from harmonicall y related to each other. T h e
both sides (Suthers et al. 200 1). In these domestic widespread occurrence of these 'two-voice' notes
birds, syllables frequently have fundamental and their likely origin from separate sides of the
frequencies as high as 5 or 6 kHz, and a single syrinx was pointed out by Greenewalt (1968)
bilaterally generated syllable may have a and Stein (1968). Two-voice syllables require
bandwidth of 6 or 7 kHz compared to 3 or 4 airflow through both sides of the syrinx, while
kHz in the case of unilaterally produced syllables. different motor programs to the muscles on each 0.2
side generate unrelated sounds that can be Time (s)
Subsequent studies show that Waterschlager
canaries have an inherited auditory defect that independently modulated in frequency and Figure 9.5 Brown thrasher song illustrating inharmonic, two-voice syllables (CDI #68). V is the time
decreases their sensitivity to sounds above 2 kHz amplitude. Two voice syllables are especially waveform of vocalization. Abbreviations as in Figure 9.3. Modified from Suthers et al. 1994.
262
I
How birds sing and why it matters 289
288 Nature's Music
i Brown-headed Cowbird Song
tissue. Similar nonlinear bilateral coupling has the only way, or are they simply independent
not yet been noted in other species, perhaps modes of singing, each with the potential to
because bilateral interaction is less prominent generate a wide range of song types? To what
in the larger syrinx of most other birds studied. extent have the evolutionary forces shaping the
Other kinds of nonlinear phenomena may acoustic properties of a species' song influenced
also account for the abrupt transitions from the evolution of production mechanisms, and
periodic to chaotic sounds, as well as period vice versa? Here I approach from a different
doubling and other acoustic effects that are viewpoint, some of the issues considered by ten
present in some songs, such assthoseof the zebra Cate (see Chapter 1O), and by Podos and Nowicki
finch (Fee et al. 1998). The realization that non- (see Chapter 1 1).
linear vocalizations may have a role in social Vocal mimics, such as mockingbirds, provide
communication has led to an increased interest an opportunity to investigate these questions.
in their production (Fitch et al. 2002). When a mimic copies another species' song does
it also use the same mechanism employed by
Spectrdl Contvart that species to produce the song, or does it invent
The duplicate sound sources in the oscine syrinx a different way to produce the song?The northern
facilitate the introduction into song of abrupt mockingbird is renowned for its ability to mimic
changes in fundamental frequency. The song of other species (Baylis 1982), but the vocal
the brown-headed cowbird, for example, begins mechanism it uses to accomplish this is unknown.
with 2 or 3 expirations each containing a 'cluster' When a mockingbird imitates a cardinal's song,
of notes, rapidly produced using pulsatile for example, does it do so by joining sounds
expiration (Allan & Suthers 1994; Fig. 9.6; CDl produced sequentially on each side of the syrinx
# 69). Successive notes are produced on opposite to generate extended FM sweeps as the cardinal
sides of the syrinx. In each case, the first note is does, or does it produce an unbroken sweep on
sung on the left side and is immediately followed, one or both sides together?
or may partially overlap, the next note, which is Zollinger and I (2004) tutored hand-reared
sung on the right side at a higher frequency. juvenile mockingbirds with recorded songs of
The frequency of these alternating notes increases other species. We found that when a mockingbird
in a staggered step-wise sequence until the end sings a cardinal song it usually mimics the vocal
of the expiration. Abrupt frequency steps between mechanism of the cardinal, switching between
successive notes that follow each other with little sides of the syrinx in mid-frequency sweep, but
or no silent interval between them are possible the switch is seldom as seamless or the frequency
if separate generators are involved. While one sweep as smooth as when the cardinal sings it. 0.2
side of the syrinx is producing a note, the muscles Time (s)
When the mockingbird's motor pattern differed
of the other closed, silent side can adjust the from that of the cardinal, the mockingbird's Figure 9.6 Song of a brown-headed cowbird (CD1 #69). Each introductory note cluster is produced
tension on its labia so that it starts the next note vocalization also differed from that of the model. during a single expiratory phase by pulsatile expiration that alternates between sides, permitting abrupt
on pitch without a slurred FM between the end In a similar experiment, juvenile mockingbirds frequency changes between successive notes. The last expiration produces the final whistle and is always
of one note and the beginning of the next, were tutored with recorded cowbird songs. The on the right side. Abbreviations as in Figure 9.3.
something considered again below. cowbird's song covers an exceptionally wide
frequency range that often extends from a few sides, beginning with the lowest frequency note were tutored, not with natural songs, but with
hundred Hz for the lowest introductory notes on the left. computer synthesized cardinal-like FM sweeps
PERFORMANCE CONSTRAINTS: to 10 kHz or more in the final whistle. Although To test the possibility that the taped tutor or cowbird-like note clusters. When as an adult,
INSIGHTS FROM A VOCAL MIMIC mockingbirds were unable to reproduce the songs of cardinals or cowbirds contained some the mockingbird mimicked the synthesized
highest or lowest frequencies of this tutor, they subtle cues such as brief pauses or discontinuities, cardinal-like FM sweeps, it sang the high
Have these species-specific patterns of song mimicked both the sound and the cowbird's vocal other than frequency, that might influence when frequency portion on its right side and the low
production evolved because they are the best production mechanism for other notes in the the mockingbird switches from one side of the frequency portion on its left, switching sides at
way to generate that species' style of song, perhaps note clusters, singing alternate notes on opposite syrinx to the other, some juvenile mockingbirds 3 or 4 kHz, during both upward and downward
263
I
How birds sing and why it matters 289
288 Nature's Music
i Brown-headed Cowbird Song
tissue. Similar nonlinear bilateral coupling has the only way, or are they simply independent
not yet been noted in other species, perhaps modes of singing, each with the potential to
because bilateral interaction is less prominent generate a wide range of song types? To what
in the larger syrinx of most other birds studied. extent have the evolutionary forces shaping the
Other kinds of nonlinear phenomena may acoustic properties of a species' song influenced
also account for the abrupt transitions from the evolution of production mechanisms, and
periodic to chaotic sounds, as well as period vice versa? Here I approach from a different
doubling and other acoustic effects that are viewpoint, some of the issues considered by ten
present in some songs, such assthoseof the zebra Cate (see Chapter 1O), and by Podos and Nowicki
finch (Fee et al. 1998). The realization that non- (see Chapter 1 1).
linear vocalizations may have a role in social Vocal mimics, such as mockingbirds, provide
communication has led to an increased interest an opportunity to investigate these questions.
in their production (Fitch et al. 2002). When a mimic copies another species' song does
it also use the same mechanism employed by
Spectrdl Contvart that species to produce the song, or does it invent
The duplicate sound sources in the oscine syrinx a different way to produce the song?The northern
facilitate the introduction into song of abrupt mockingbird is renowned for its ability to mimic
changes in fundamental frequency. The song of other species (Baylis 1982), but the vocal
the brown-headed cowbird, for example, begins mechanism it uses to accomplish this is unknown.
with 2 or 3 expirations each containing a 'cluster' When a mockingbird imitates a cardinal's song,
of notes, rapidly produced using pulsatile for example, does it do so by joining sounds
expiration (Allan & Suthers 1994; Fig. 9.6; CDl produced sequentially on each side of the syrinx
# 69). Successive notes are produced on opposite to generate extended FM sweeps as the cardinal
sides of the syrinx. In each case, the first note is does, or does it produce an unbroken sweep on
sung on the left side and is immediately followed, one or both sides together?
or may partially overlap, the next note, which is Zollinger and I (2004) tutored hand-reared
sung on the right side at a higher frequency. juvenile mockingbirds with recorded songs of
The frequency of these alternating notes increases other species. We found that when a mockingbird
in a staggered step-wise sequence until the end sings a cardinal song it usually mimics the vocal
of the expiration. Abrupt frequency steps between mechanism of the cardinal, switching between
successive notes that follow each other with little sides of the syrinx in mid-frequency sweep, but
or no silent interval between them are possible the switch is seldom as seamless or the frequency
if separate generators are involved. While one sweep as smooth as when the cardinal sings it. 0.2
side of the syrinx is producing a note, the muscles Time (s)
When the mockingbird's motor pattern differed
of the other closed, silent side can adjust the from that of the cardinal, the mockingbird's Figure 9.6 Song of a brown-headed cowbird (CD1 #69). Each introductory note cluster is produced
tension on its labia so that it starts the next note vocalization also differed from that of the model. during a single expiratory phase by pulsatile expiration that alternates between sides, permitting abrupt
on pitch without a slurred FM between the end In a similar experiment, juvenile mockingbirds frequency changes between successive notes. The last expiration produces the final whistle and is always
of one note and the beginning of the next, were tutored with recorded cowbird songs. The on the right side. Abbreviations as in Figure 9.3.
something considered again below. cowbird's song covers an exceptionally wide
frequency range that often extends from a few sides, beginning with the lowest frequency note were tutored, not with natural songs, but with
hundred Hz for the lowest introductory notes on the left. computer synthesized cardinal-like FM sweeps
PERFORMANCE CONSTRAINTS: to 10 kHz or more in the final whistle. Although To test the possibility that the taped tutor or cowbird-like note clusters. When as an adult,
INSIGHTS FROM A VOCAL MIMIC mockingbirds were unable to reproduce the songs of cardinals or cowbirds contained some the mockingbird mimicked the synthesized
highest or lowest frequencies of this tutor, they subtle cues such as brief pauses or discontinuities, cardinal-like FM sweeps, it sang the high
Have these species-specific patterns of song mimicked both the sound and the cowbird's vocal other than frequency, that might influence when frequency portion on its right side and the low
production evolved because they are the best production mechanism for other notes in the the mockingbird switches from one side of the frequency portion on its left, switching sides at
way to generate that species' style of song, perhaps note clusters, singing alternate notes on opposite syrinx to the other, some juvenile mockingbirds 3 or 4 kHz, during both upward and downward
264
I
Nature's Music
290
sweeping sounds (Fig. 9.7; CDl #70). When The ability to achieve abrupt step-wise frequency
I How birds sing and why it matters
265
I
Nature's Music
290
sweeping sounds (Fig. 9.7; CDl #70). When The ability to achieve abrupt step-wise frequency
I How birds sing and why it matters
266
How birds sing and why it matters 293
292 Nature's Music
Imitation of Stepped Tones in brown thrashers. It is not yet known whether on its syllable type. Some syllables may be almost
Reveals Motor Constraints females pay attention to these particular aspects a pure tone repeated a few times a second, others
Mockingbird righttleft of cardinal or thrasher song. However, in brown- include upward or downward frequency sweeps
2-tone imitations headed cowbirds and domestic canaries, there is of varying complexity sung at various repetition
some evidence suggesting that parts of their songs rates which may exceed 30 per second.
requiring a high degree of motor coordination Researchers at the University of Paris (Vallet
f--- 2 kHz
may have a special perceptual significance to & Kreutzer 1995; Vallet et al. 1998) found that
prospective mates. certain kinds of 'type A' syllables are more effective
Male brown-headed cowbirds have a vocal than others in attracting the interest of a
repertoire that includes several different song prospective mate (Box 4, p. 58). They played
types. In the eastern subspecies, each song begins different phrases, each consisting of a single
with 2 or 3 note clusters followed by a whistle syllable type, to sexually receptive female canaries
(Fig. 9.6). West, King, and their colleagues and recorded their CSD responses. The female
Slurring with showed that some songs are more effective than invites the male to mate with her by arching her
righttright imitation others in eliciting copulati,onsolicitation displays ,back, raising her tail, and fluttering her partially
6l
(CSD) from female cowbirds (King &West 1977, spread wings. Vallet and his colleagues found
1983; West et al. 1979, 1981; see Chapter 14). that wide-band syllables that were more complex
A male's breeding success is correlated with the than simple frequency modulated sweeps,
potency of his songs. Experimental manipulation composed of two or more notes, and sung at a
of songs indicates that the note clusters are more repetition rate of 16 per second or higher evoked
effective than the final whistle in eliciting CSDs. significantly more female displays than other
A high-frequency note, designated the syllables. Similar type 'A' syllables recorded from
'interphrase unit' at the end of the last note wild canaries also elicit CSDs from domestic
cluster appears to be particularly important. Songs canaries when played back to them in the
of dominant males were more effective than those laboratory (Leitner et al. 200 I), indicating they
of other males. The inclusion of stereotyped note are not an artifact of domestication.
0.5
Time (sec)
clusters in juvenile plastic song triples the Studies of how domestic canaries produce their
effectiveness of juvenile song in evoking a female songs suggest that 'A' syllables may be relatively
(B)
copulatory response (west & ~ i n 1988a). g difficult to sing (Suthers et al. 200 1). Domestic
Figure 9.8 (A) A mockingbird's imitations of pairs of computer-synthesized tones showing that the context Further experiments with computer-edited a canaries sing some syllables in their left syrinx,
as well as the frequency of a note influences the side on which it is produced. (A) When the 2 kHz tone songs are needed to identify the specific temporal others in the right, and still' other multi-note
is preceded by a tone at a slightly higher frequency, the first tone is sung on the right and the 2 kHz tone or acoustic aspects of note clusters that are most syllables contain contributions from both sides
is sung on the left side. If the preceding tone is at a lower frequency, it is sung on the left, and the 2 kHz important in stimulating a female. But it already of the syrinx. The wide-band 'A' syllables of three
tone is now sung on the right side. (B) Switching sides of production enhances the spectral contrast seems clear that a male cowbird's mating success birds studied, included contributions from both
between successive notes. When singing the second pair of synthesized tones, the mockingbird produced
is influenced by his bilateral motor skills in singing sides of the syrinx (Fig. 9.9; C D l #71). The
both on the same side. As a result, the two tones are joined by an FM sweep. R and L indicate side of
production. Tutor tones and syringeal airflow and pressure are not shown. (A modified after Zollinger and
stereotyped patterns of notes that are rapidly syllables that are most effective in eliciting CSDs
Suthers 2004). produced, with alternating steps in frequency. from a female canary thus require the male to
The domestic canary provides another example produce stereotyped notes from each side of his
in which the motor mechanisms of song of a possible relationship between vocal motor syrinx at note repetition rates at least twice the
but are not necessarily correlated with the level
production have been studied, an aspect of singing dexterity and female choice. Male domestic syllable rate. Furthermore, behavioral playback
of motor skill, and are unlikely to reflect the
that seems likely to require special vocal motor canaries have a song repertoire of about 20-30 experiments (Vallet et al. 1998) suggest that the
difficulty that is required to execute the vocal
skills is coordination between the two halves of different syllables. Individual songs contain a effectiveness of 'A' syllables in stimulating a female
performance. A male's quality might be reflected
the syrinx. Examples are the acoustically seamless subset of these syllables, repeated to form a is greater if the silent interval between the notes
not only in his stamina for sustained singing,
switch between right and left sides during-wide sequence of phrases. Each syllable type is always within each syllable is short (between 16 and 23
but also in his ability to flawlessly and repeatedly
band frequency sweeps by cardinals or the sung in the same way and at the same repetition ms, as opposed to 23-30 ms). This is consistent
sing syllables or phrases that require special motor
concurrent production of independently rate, but there are large differences in the spectral with the hypothesis that bilateral coordination
skills (Suthers & Goller 1997; Nowicki et al.
modulated sounds forming two-voice syllables and temporal properties of a phrase, depending involving very rapid left-right switching between
1998a; ten Cate et al. 2002). Among the species
267
How birds sing and why it matters 293
292 Nature's Music
Imitation of Stepped Tones in brown thrashers. It is not yet known whether on its syllable type. Some syllables may be almost
Reveals Motor Constraints females pay attention to these particular aspects a pure tone repeated a few times a second, others
Mockingbird righttleft of cardinal or thrasher song. However, in brown- include upward or downward frequency sweeps
2-tone imitations headed cowbirds and domestic canaries, there is of varying complexity sung at various repetition
some evidence suggesting that parts of their songs rates which may exceed 30 per second.
requiring a high degree of motor coordination Researchers at the University of Paris (Vallet
f--- 2 kHz
may have a special perceptual significance to & Kreutzer 1995; Vallet et al. 1998) found that
prospective mates. certain kinds of 'type A' syllables are more effective
Male brown-headed cowbirds have a vocal than others in attracting the interest of a
repertoire that includes several different song prospective mate (Box 4, p. 58). They played
types. In the eastern subspecies, each song begins different phrases, each consisting of a single
with 2 or 3 note clusters followed by a whistle syllable type, to sexually receptive female canaries
(Fig. 9.6). West, King, and their colleagues and recorded their CSD responses. The female
Slurring with showed that some songs are more effective than invites the male to mate with her by arching her
righttright imitation others in eliciting copulati,onsolicitation displays ,back, raising her tail, and fluttering her partially
6l
(CSD) from female cowbirds (King &West 1977, spread wings. Vallet and his colleagues found
1983; West et al. 1979, 1981; see Chapter 14). that wide-band syllables that were more complex
A male's breeding success is correlated with the than simple frequency modulated sweeps,
potency of his songs. Experimental manipulation composed of two or more notes, and sung at a
of songs indicates that the note clusters are more repetition rate of 16 per second or higher evoked
effective than the final whistle in eliciting CSDs. significantly more female displays than other
A high-frequency note, designated the syllables. Similar type 'A' syllables recorded from
'interphrase unit' at the end of the last note wild canaries also elicit CSDs from domestic
cluster appears to be particularly important. Songs canaries when played back to them in the
of dominant males were more effective than those laboratory (Leitner et al. 200 I), indicating they
of other males. The inclusion of stereotyped note are not an artifact of domestication.
0.5
Time (sec)
clusters in juvenile plastic song triples the Studies of how domestic canaries produce their
effectiveness of juvenile song in evoking a female songs suggest that 'A' syllables may be relatively
(B)
copulatory response (west & ~ i n 1988a). g difficult to sing (Suthers et al. 200 1). Domestic
Figure 9.8 (A) A mockingbird's imitations of pairs of computer-synthesized tones showing that the context Further experiments with computer-edited a canaries sing some syllables in their left syrinx,
as well as the frequency of a note influences the side on which it is produced. (A) When the 2 kHz tone songs are needed to identify the specific temporal others in the right, and still' other multi-note
is preceded by a tone at a slightly higher frequency, the first tone is sung on the right and the 2 kHz tone or acoustic aspects of note clusters that are most syllables contain contributions from both sides
is sung on the left side. If the preceding tone is at a lower frequency, it is sung on the left, and the 2 kHz important in stimulating a female. But it already of the syrinx. The wide-band 'A' syllables of three
tone is now sung on the right side. (B) Switching sides of production enhances the spectral contrast seems clear that a male cowbird's mating success birds studied, included contributions from both
between successive notes. When singing the second pair of synthesized tones, the mockingbird produced
is influenced by his bilateral motor skills in singing sides of the syrinx (Fig. 9.9; C D l #71). The
both on the same side. As a result, the two tones are joined by an FM sweep. R and L indicate side of
production. Tutor tones and syringeal airflow and pressure are not shown. (A modified after Zollinger and
stereotyped patterns of notes that are rapidly syllables that are most effective in eliciting CSDs
Suthers 2004). produced, with alternating steps in frequency. from a female canary thus require the male to
The domestic canary provides another example produce stereotyped notes from each side of his
in which the motor mechanisms of song of a possible relationship between vocal motor syrinx at note repetition rates at least twice the
but are not necessarily correlated with the level
production have been studied, an aspect of singing dexterity and female choice. Male domestic syllable rate. Furthermore, behavioral playback
of motor skill, and are unlikely to reflect the
that seems likely to require special vocal motor canaries have a song repertoire of about 20-30 experiments (Vallet et al. 1998) suggest that the
difficulty that is required to execute the vocal
skills is coordination between the two halves of different syllables. Individual songs contain a effectiveness of 'A' syllables in stimulating a female
performance. A male's quality might be reflected
the syrinx. Examples are the acoustically seamless subset of these syllables, repeated to form a is greater if the silent interval between the notes
not only in his stamina for sustained singing,
switch between right and left sides during-wide sequence of phrases. Each syllable type is always within each syllable is short (between 16 and 23
but also in his ability to flawlessly and repeatedly
band frequency sweeps by cardinals or the sung in the same way and at the same repetition ms, as opposed to 23-30 ms). This is consistent
sing syllables or phrases that require special motor
concurrent production of independently rate, but there are large differences in the spectral with the hypothesis that bilateral coordination
skills (Suthers & Goller 1997; Nowicki et al.
modulated sounds forming two-voice syllables and temporal properties of a phrase, depending involving very rapid left-right switching between
1998a; ten Cate et al. 2002). Among the species
268
294 Nature's Music How birds sing and whv it matters 295
'Sexy' Syllables of a Domestic Canary Song vocal tract or cranial muscle groups involved in mate selection, and it is not clear if there is a
161s singing. It may be that males who are not in top correlation between the presence of 'A' syllables
1 R 181s
physical condition have difficulty producing these and reproductive success. Whatever the case, a
phrases. If so, the 'A' phrase may be an honest better understanding of how birds sing promises
indicator of male fitness (see Chapter 2). to be a valuable tool in deciphering the
Another possibility is that the presence of 'A' communicative functions and evolution of song.
phrases in song may indicate a male's
,-h reproductive state - that he is ready to breed.
Field studies of wild canaries show that the CONCLUSIONS
I syllable repetition rate of their repertoire, but
I not repertoire size, changes seasonally. X syllables Twenty years ago, in a review on avian sound
sung during the breeding season are replaced production, Brackenbury (1982) lamented that,
during the winter with syllables at lower "many ideas about the functioning of the
repetition rates (Leitner et al. 2001). Seasonal passerine syrinx are based on guesswork." In the
changes in steroid levels can affect syringeal .past decade new techniques for recording and
muscles and vocal behavior (DeVoogd 199 1; observing syringeal function during song have
! DeVoogd et al. 1991; Beani et al. 1995; Hartley yielded significant progress toward removing some
l et al. 1997; Tramontin et al. 2000). The presence of this guesswork and advancing our
I
I of 'A' phrases in a male's song might indicate understanding how birds sing, though the
his readiness to mate. Domestic canaries sing number of species sampled is still very small.
I longer strings of 'A' phrases in the presence of There is every reason to believe the next decade
other male or female canaries than when they will be equally productive. As we learn more
are singing alone, even though the number and about how the avian vocal system works, and its
duration of songs is similar in both situations acoustic possibilities and limitations, we will gain
(Kreutzer et al. 1999). new insights into the neural organization,
These two possible functions of 'A' syllables behavioral significance, and evolution of a
in intersexual communication are not mutually behavioral system that is unique in the animal
exclusive. Other factors may also play a role in kingdom.
0.1
Time (s)
Figure 9.9 A segment of a domestic canary song containing two phrases of type 'A' syllables (CDI #74).
Each syllable is composed of two notes. The first note sweeps upward on the right side for about 20 ms
and is followed by an approximately 10 ms downward sweep at a lower frequency on the left side. To be
maximally effective in eliciting copulation displays from a receptive female, syllables must be complex and
sung at repetition rates greater than 16 per second. Abbreviations as in Figure 9.3.
sides of the syrinx may have special perceptual the male's health or fitness. Of all the syllables
importance in the vocal communication of birds. in a canary's repertoire, 'A' phrases may demand
The kind of information females get from 'A' the greatest precision in bilateral motor
syllables remains to be determined. O n e coordination within the syrinx, and also between
possibility is that they provide an indication of the syrinx and the respiratory muscles and other
269
294 Nature's Music How birds sing and whv it matters 295
'Sexy' Syllables of a Domestic Canary Song vocal tract or cranial muscle groups involved in mate selection, and it is not clear if there is a
161s singing. It may be that males who are not in top correlation between the presence of 'A' syllables
1 R 181s
physical condition have difficulty producing these and reproductive success. Whatever the case, a
phrases. If so, the 'A' phrase may be an honest better understanding of how birds sing promises
indicator of male fitness (see Chapter 2). to be a valuable tool in deciphering the
Another possibility is that the presence of 'A' communicative functions and evolution of song.
phrases in song may indicate a male's
,-h reproductive state - that he is ready to breed.
Field studies of wild canaries show that the CONCLUSIONS
I syllable repetition rate of their repertoire, but
I not repertoire size, changes seasonally. X syllables Twenty years ago, in a review on avian sound
sung during the breeding season are replaced production, Brackenbury (1982) lamented that,
during the winter with syllables at lower "many ideas about the functioning of the
repetition rates (Leitner et al. 2001). Seasonal passerine syrinx are based on guesswork." In the
changes in steroid levels can affect syringeal .past decade new techniques for recording and
muscles and vocal behavior (DeVoogd 199 1; observing syringeal function during song have
! DeVoogd et al. 1991; Beani et al. 1995; Hartley yielded significant progress toward removing some
l et al. 1997; Tramontin et al. 2000). The presence of this guesswork and advancing our
I
I of 'A' phrases in a male's song might indicate understanding how birds sing, though the
his readiness to mate. Domestic canaries sing number of species sampled is still very small.
I longer strings of 'A' phrases in the presence of There is every reason to believe the next decade
other male or female canaries than when they will be equally productive. As we learn more
are singing alone, even though the number and about how the avian vocal system works, and its
duration of songs is similar in both situations acoustic possibilities and limitations, we will gain
(Kreutzer et al. 1999). new insights into the neural organization,
These two possible functions of 'A' syllables behavioral significance, and evolution of a
in intersexual communication are not mutually behavioral system that is unique in the animal
exclusive. Other factors may also play a role in kingdom.
0.1
Time (s)
Figure 9.9 A segment of a domestic canary song containing two phrases of type 'A' syllables (CDI #74).
Each syllable is composed of two notes. The first note sweeps upward on the right side for about 20 ms
and is followed by an approximately 10 ms downward sweep at a lower frequency on the left side. To be
maximally effective in eliciting copulation displays from a receptive female, syllables must be complex and
sung at repetition rates greater than 16 per second. Abbreviations as in Figure 9.3.
sides of the syrinx may have special perceptual the male's health or fitness. Of all the syllables
importance in the vocal communication of birds. in a canary's repertoire, 'A' phrases may demand
The kind of information females get from 'A' the greatest precision in bilateral motor
syllables remains to be determined. O n e coordination within the syrinx, and also between
possibility is that they provide an indication of the syrinx and the respiratory muscles and other
270
Fiete, I.R. and Seung, H.S. (2009) Birdsong Learning. In: Encyclopedia of Neuroscience.
Squire, L., Editor. pp. 227-239. Elsevier, New York.
Birdsong Learning 227
Birdsong Learning
I R Fiete, University of California, Santa Barbara, and juvenile has begun to practice its song. Sensorimotor
Center for Neural Systems, Pasadena, CA, USA learning is a slow process of self-generated trial and
H S Seung, Massachusetts Institute of Technology, error and can proceed successfully in complete audi-
Boston MA, USA tory and social isolation after template acquisition.
2009 Published by Elsevier Ltd. However, auditory feedback of the birds own voca-
lizations is crucial during sensorimotor learning
because a bird deafened at this stage cannot learn to
reproduce its mental template of the tutor song.
Introduction This article focuses on the motor and sensorimotor
Birdsong has been compared to human speech aspects of song production, in particular the forma-
because both are examples of animal vocalizations tion of neural sequences to drive song, the learning of
that are at least partially learned. Indeed, some of the motor map, and the reasons song may be encoded
the neural pathways underlying these vocal behaviors the way it is by songbird premotor neurons. The aim
may be similar in songbirds and humans. More gen- is to illustrate how theoretical work has contributed
erally, birdsong can be compared to any motor behav- to the understanding of the song system and highlight
ior that involves sequential control and learning and the resulting predictions for experiment. Auditory
is an ideal playground for experimental and theoreti- neural coding and dynamics and the physics (acous-
cal explorations of the neural basis for motor learning tics and mechanics) of songbird vocal production are
and control. beyond the scope of the article.
The neurons involved in song production and learn-
ing are clustered in a small number of discrete nuclei Basic Functional Anatomy of
with well-defined boundaries, fairly well-characterized
the Song Pathway
connectivity, and increasingly well-studied neuro-
physiology. The various species of oscine songbirds In the oscine vocal premotor circuit, high-level vocal
display a vast spectrum of complexity in syllable center nucleus (HVC) projects to nucleus robustus
sequencing (syntax), improvisation, and vocal acqui- archistriatalis (RA). RA drives the motor neurons
sition. For example, song can be as simple as repeti- that innervate the syringeal and respiratory organs
tions of a set of syllables arranged in fixed order for song production (Figure 1). Lesioning any of
(e.g., zebra finches), or it can be intricately impro- these nuclei immediately disables song production,
vised, with syllable ordering governed by probabilistic while nuclei upstream of HVC are not necessary for
transitions between different syllables (e.g., domesti- singing in adult zebra finches. Therefore, HVC origi-
cated bengalese finches). Mockingbirds can acquire nates the top-level premotor drive for adult zebra
novel songs and learn novel nonbird sounds through- finch song. Although the circuitry between the pre-
out life. Yet the basic neural song circuit is similar in motor nuclei is feedforward, connectivity within
all oscines. This diversity of behavior subserved by an HVC and within RA is extensively recurrent. In
underlying similarity of structure makes birdsong addition to the premotor circuit, the song system
an intriguing system for future comparative studies contains the anterior forebrain pathway (AFP),
of sophisticated motor sequencing and control. which connects HVC to RA through an indirect
Models have so far focused on the zebra finch, a path. Lesion studies reveal that the AFP is crucial
songbird with one of the simplest of oscine song reper- for the sensorimotor phase of song learning and
toires, because most recordings of neural activity have long-term song maintenance but not for song produc-
been obtained from these birds. The male zebra finch tion. The AFP is an avian homolog of the mammalian
learns one song motif with fixed syllable order and thalamocorticalbasal ganglia pathway.
sings repetitions of that motif throughout its life. Song Based on anatomical, neurophysiological, and
acquisition takes place in a sensory phase, when the lesion data, it is believed that HVC generates the
juvenile bird acquires a mental template of the song spatiotemporal premotor drive for sequential motor
of a possibly unrelated adult tutor, and an overlapp- activation in the form of sequential neural activity.
ing sensorimotor phase, when the juvenile begins The HVC activity is abstract, in the sense that it
to vocalize and tries to match its vocalizations with encodes only temporal ordering, not song features
its mental template. In the laboratory the sensory (Figures 2(a) and 2(b)). The synapses from HVC to
and sensorimotor phases can be separated: template RA and from RA to the motor neurons convert the
acquisition is fast and can be completed and the tutor sequence of HVC activity patterns into the sequence
bird removed from the juveniles presence before the of motor commands that produces song (Figure 2).
271
228 Birdsong Learning
HVC
UVA Cortex
RA LMAN
Syrinx nXIIts X
DLM
Motor
Premotor Excitatory b
AFP Inhibitory
a
Figure 1 The neural pathways underlying song learning and song production. (a) Song learning and production involves a discrete set
of nuclei. (b) The same areas, unfolded to better illustrate their connectivity. Lines terminating in arrows designate excitatory or putative
excitatory connectivity. Lines terminating in circles designate inhibitory synapses. Blue nuclei are part of the premotor pathway. Green
nuclei are part of the anterior forebrain pathway (AFP) and are important for sensorimotor song learning but not song production. HVC,
high-level vocal center nucleus; LMAN, lateral magnocellular nucleus of the anterior nidopallium; NIf, interfacial nucleus; RA, nucleus
robustus archistriatalis.
According to this perspective, there are two types of different from HVCRA activity: these neurons fire
learning. First, HVC acquires the capability of gener- tonically throughout song, with high rates and rela-
ating a long sequence of neural activity patterns tively little modulation in firing frequency across song
(Figures 2(f) and 2(g)). Second, the circuitry down- (Figure 2(b)).
stream from HVC acquires the capability of mapping In the adult zebra finch, deafening causes no imme-
HVC activity patterns to the correct sound-producing diate effect on song production, suggesting that audi-
patterns of activity in RA. The HVCRA synapses tory feedback is not important for sequence generation.
show extensive structural plasticity, with synaptic Also, lesion studies indicate that input from the higher
growth followed by massive retraction, precisely interfacial nucleus to HVC is not necessary for singing
during the sensorimotor learning period. Electro- in zebra finches. What network connectivity and neural
physiological synaptic maturation accompanies such dynamics underlie the sequence generation dynamics
structural remodeling. Therefore, motor map learning observed in the top-level premotor area, and how
from HVC to the vocal outputs is likely to involve might that connectivity become established as hatchl-
plasticity in the HVCRA synapses. Other possible ings mature?
loci of plasticity for sensorimotor learning, not ruled
out by experiments, include the recurrent RARA The Synaptic Chain Model
synapses or the RAmotor neuron connections. The synaptic chain model is an old idea for how
networks of neurons might produce sequential pat-
terns of activity. (The idea was already old when it
Sequence Generation
was critiqued by Lashley in 1951 under the name
The high-level premotor area HVC displays clear associative chaining model.) In the models simplest
sequential patterns of neural activation. In the awake version, the neurons in a network are divided into
singing zebra finch, each recorded RA-projecting groups, and the groups are ordered in a sequence.
HVC neuron (also called an HVCRA neuron) fires at From each group of neurons, there are excitatory
most one brief burst of spikes (lasting 6 ms) per synaptic connections to the neurons in the next
second-long song motif (Figure 2(a)). These neurons group of the sequence. The groups and their synaptic
are otherwise inactive during singing. Different RA- connections are like the links of a chain.
projecting HVC neurons fire their single bursts at If the first group in the sequence is activated, it
different fixed times in the motif. Thus as a popula- sends synaptic drive to the second group, which
tion, the HVCRA neurons cover time during song. then becomes active. The second group in turn excites
In addition to RA-projecting neurons, HVC con- the third group to become active, and so on down the
tains inhibitory interneurons which ramify within chain. (An alternative mechanism for the propagation
HVC and X-projecting neurons which send their out- of sequential activity in songbird HVC is based on
puts to the AFP. Inhibitory interneuron activity is very the removal of lateral inhibition between groups of
272
Birdsong Learning 229
273
230 Birdsong Learning
274
Birdsong Learning 231
the bidirectional model can also produce other kinds no doubt messier than in these idealized models,
of activity patterns that the unidirectional model can- the predictions should be phrased probabilistically.
not (Figures 3(c) and 3(d)), which could be used as the Suppose A and B are two RA-projecting HVC neurons
basis of experimental tests that distinguish between that are activated in succession. In the unidirectional
the two models. model, the probability of a connection from A to B is
In these tests, activity in the HVC network would expected to be higher than that of a connection from
be initiated by some means, and the subsequent B to A or of a connection between two neurons chosen
response to stimulation would be observed. If the last at random. (Alternatively, the predicted asymmetry of
group in the chain is stimulated (Figures 3(c) and 3(d), coupling between A and B might be reflected in syn-
blue electrode), there would be no subsequent activity aptic strength rather than in the presence or absence of
in the unidirectional model. In the bidirectional model, connections.) In the bidirectional model, the probabil-
the activity would propagate backward along the chain ity of a connection from A to B is equal to the proba-
and would result in a sequence of motor commands bility of a connection from B to A.
that is reversed from normal song. Another experiment Testing this kind of prediction could be done by
would be to stimulate an intermediate group in the comparing physiological and neuroanatomical mea-
chain (Figures 3(c) and 3(d), green electrode). Accord- surements. Optical imaging could be used to find two
ing to the bidirectional model, subsequent activity neurons A and B that are activated in succession. Then
would simultaneously propagate both forward and optical stimulation could be used to see whether and
backward along the chain. (Note that the playback how they are connected to each other. Alternatively, a
will not continue from the forward to the backward completely neuroanatomical approach would be to
direction, or from the backward to the forward direc- reconstruct the entire connectivity of HVC using serial
tion, because of the existence of a refractory period for section electron microscopy, look for a chain within
activation. This point is well known in the theory of this connectivity, and determine its directionality.
excitable media.) In the unidirectional model, activity
would propagate only forward.
The role of inhibition The synaptic chain models
While such experiments may sound straight-
above were based on synaptic excitation, without
forward, they are actually technically demanding. So
reference to the role of synaptic inhibition. Inhibition
far no topographic organization of RA-projecting
is often included in such models to globally dampen
HVC neurons has been detected. In other words, the
activity in a temporally nonspecific way and can
location of a neuron in HVC does not seem to be
enhance the stability of sequential activity in the
related to the time at which it is activated during a
excitatory neurons. To implement such a global inhi-
song motif. If coactive neurons were co-located in
bition effect, each inhibitory interneuron receives
HVC, it would be relatively straightforward to acti-
excitatory input from a large random set of projection
vate a hypothetical group in the synaptic chain models
neurons, and each projection neuron in return
by stimulation through a microelectrode. Instead, the
receives synapses from the interneurons. If the burst
neurons of a hypothetical group appear to be scattered
times of the projection neuron population were
throughout HVC. Therefore, a group would be best
distributed across the entire song motif, then each
identified through optical imaging of activity, and best
interneuron would receive roughly continuous drive
stimulated through optical means also.
and would spike continuously.
So far, no one has attempted to precisely stimulate a
Consistent with this picture, the temporal firing
group of HVC neurons. However, cruder perturbation
patterns of inhibitory interneurons in HVC are
experiments have been performed. For example, HVC
much less specific than those of the HVCRA neurons:
activity during song has been perturbed by stimulating
an inhibitory interneuron is typically active through-
RA, which leads to retrograde effects on HVC. Stimu-
out most of the motif. This suggests that the connec-
lation leads to skips in motif sequencing, but the
tivity of inhibitory neurons is also much less specific
subsequent vocalization is a recognizable fragment of
than that of excitatory neurons.
the motif sung in the forward direction, beginning at a
different position within the motif. This is more consis-
Sequence Learning
tent with the unidirectional model than with the bidi-
rectional model. But more-conclusive tests will require Hand in glove with the question of what intrinsic
precisely controlled stimulation of activity in HVC. neural properties and network architectures could
A more direct way of testing the synaptic chain underlie the stable propagation of activity in HVC is
models is to check their predictions concerning the the challenge of understanding how such architectures
synaptic connectivity of HVC. Since the real HVC is could emerge from initially less-structured networks.
275
232 Birdsong Learning
Intuitively, associative (Hebbian) learning rules not local to the individual synapses, they are local
such as spike-timing-dependent plasticity (STDP) to single neurons. Remarkably, these local neural
seem ideally suited to produce the kinds of asymmet- bounds nevertheless serve to uniformly and globally
ric connectivity characteristic of synaptic chains: if distribute activity throughout the network in both
neuron A fires before B, the connection from A to space and time.
B is strengthened, but that from B to A is weakened. If The qualitative nature of these synaptic bounds,
a nave network is trained with this rule for synaptic and how they are triggered and enforced, is where
plasticity on a set of input patterns presented sequen- the two models diverge. The model of Jin and Jun
tially, one would expect network weights to encode assumes that the formation of a finite number of
the temporal correlations in the training patterns strong connections from a neuron triggers axon
and later to autonomously reproduce the same long, remodeling and elimination of weak synapses. Fiete
stable chains of neural activity when stimulated. et al. propose two separate schemes based on neural
But contrary to intuition, past modeling studies resource constraints: one assumes that the production
show that associative learning rules are not sufficient rate of synapse-building resources is limited,
to configure networks to autonomously generate long so potentiation in one synapse triggers a uniform
neural sequences. Networks trained in this way usu- heterosynaptic weakening of all other synapses at
ally require finely tuned excitatory and inhibitory that neuron. The other scheme assumes a cap on the
connectivity that is continuously renormalized to bal- total resources a neuron can devote to synaptic main-
ance network activity during learning, as well as tenance; therefore all synapses at a given neuron
additional pacemaker inputs to propagate activity undergo uniform weakening whenever STDP drives
from one time step to the next. The situation is even their summed synaptic strength to exceed the cap.
worse if the network does not have access to training Both models make predictions for the rules gov-
inputs that are themselves sequential. In such cases, erning synaptic plasticity within HVC and for the
the best response even with fine-tuning of parameters statistics of the resulting connectivity within HVC.
is the formation of extremely short chains (fewer than The Fiete et al. model produces several sequences
five steps) using thousands of neurons. of different lengths, with a definite distribution of
The reason for this is that associative learning chain lengths. Determining whether this distribution
rules do not balance activity equitably across space matches syllable length distributions in birdsong could
(neurons) and time; instead, they tend to amplify indicate whether songbird syllables attain their specific
inequities. For example, a neuron that by chance lengths largely due to bottom-up (low-level network)
receives strong synapses from its inputs will fire properties or top-down (behavioral, evolutionary, or
often. Because of its frequent firing, it will tend to respiratory) constraints.
rapidly form strong outgoing synaptic connections as
well. As a result, activity tends to accumulate around
Sensorimotor Learning of the Motor Map
a few overly active neural hubs, producing epileptic
or synchronous burstlike network states rather than The main goal-directed aspect of learning in zebra
long, orderly chains. Empirically, these problems finches takes place during the sensorimotor phase,
remain unsolved with the imposition of local con- when the juvenile learns to match, through auditory
straints on the maximum strengths of individual feedback and template comparison, its own string of
synapses. For these reasons, the question of how vocalizations with its memorized mental copy of the
plasticity rules that are local in both space and time tutor song. Because the goal-directed aspect of song
can impose global constraints on activity balance acquisition is so striking and because natural goal-
over all space and all time is nontrivial. directed learning behaviors that are easy to quantify
Two independent models provide a conceptual are rare, several models have focused on this aspect of
framework for how associative learning rules must song learning. Yet the microscopic synaptic plasticity
be augmented to allow initially unstructured networks rules that underlie such learning remain experimen-
to self-organize and generate stable, long sequences tally undiscovered and theoretically debated. Predic-
of neural activation. Both models use STDP as the tions from existing theoretical or computational
primary activity-dependent mechanism for synaptic models can serve to greatly narrow, in a principled
change. But crucially, in addition to STDP, both mod- way, the dauntingly large number of possible plastic-
els rely strongly on competitively enforced synaptic ity paradigms to investigate in experiment, and they
bounds that are not computed on the level of single are now specific enough to be tested in detail in
synapses but are triggered by neuronwide constraints. slice preparations of the song system. (Experimental
Although the rules governing synaptic remodeling are testing of the proposed paradigms is contingent on
276
Birdsong Learning 233
277
234 Birdsong Learning
278
Birdsong Learning 235
delivers to the premotor network (Figure 4(b)). The A recent model of song learning by Fiete and colla-
associational rule also includes a subtraction term for borators returns to a reinforcement learning frame-
synaptic weakening, bringing it closer to reinforce- work and attempts to do two things: (1) propose that,
ment learning rules, but the specific form used is not and illustrate how, spiking RA neurons might perform
proven in the literature to guarantee improved perfor- reinforcement learning by using rapidly time-varying
mance. As a result, there are no general assurances of LMAN synaptic inputs that perturb RA membrane
convergence or stability, even in networks of simplified voltages (rather than using static synaptic pertur-
rate-based neurons. The model requires fine-tuning by bations of the plastic HVCRA weights; Figure 4(c))
hand of several homeostatic mechanisms to maintain and (2) show as a proof-of-principle that although it is
stability during learning. The authors demonstrated widely thought to be slow, reinforcement learning
through numerical simulation in rate-based networks can be fast enough to explain song acquisition in
that under particular parameter settings, with a small a full-scale model of the birdsong network with spik-
set of RA assemblies, HVCRA connections, and a ing, conductance-based neurons, delayed feedback,
few syllable-long time steps, the network can learn to scalar reinforcement, and purely random independent
match a desired output. perturbations.
A strength of this model is that, unlike others in The model assumes that LMAN perturbs the pre-
the literature, it stresses the importance of lateral motor pathway by injecting independent random
connections within RA, which, when subject to the currents into different RA neurons (Figure 4(c))
same plasticity rules as HVCRA synapses, form through ordinary glutamatergic synaptic transmission
assemblies that are capable of pattern completion (Figures 5, 6(d), and (e)). This is consistent with recent
even with partially correct inputs, to generate good stimulation and pharmacological inactivation experi-
output syllables. The AFPs only role in the model is ments in LMAN, which show that LMAN helps drive
to generate and convey a spatially uniform reinforce- song variability through glutamatergic synaptic projec-
ment signal to RA (Figure 4(b)). However, in the tions to RA. The model requires at most as many
zebra finch, LMAN activity is spatially inhomoge- independent LMAN inputs as there are RA neurons.
neous (neural activity is not strongly correlated across The LMAN inputs are assumed to be time varying
the nucleus), and connections to RA are not highly throughout song. Reinforcement is assumed to be com-
divergent, which implies that different RA neurons puted and delivered from elsewhere in the brain, with
are likely to receive fairly uncorrelated inputs from learning following the high-level, actorcritic schema.
LMAN rather than a uniform signal. As formula- Under these conditions, the derivation of the rules for
ted, the model cannot explain why, in experiments, how activity should modify synaptic weights, and the
LMAN lesions reduce song variability. Experiments rules themselves, are quite different from weight
fail to reveal qualitative changes in LMAN activity perturbation-like algorithms (Figures 6(d)6(f)): each
when an adult birds auditory feedback is altered, HVCRA synapse must somehow deduce an instruc-
which suggests that LMAN may not primarily convey tion for change based on time-varying fluctuations
an error or match signal. (If LMAN does generate in the conductance of its postsynaptic RA neuron
an error signal, but bases it on learned predictions rather than simply decide whether to retain or discard
about auditory feedback, it is possible that prolonged direct static perturbations of its own weight through
rather than transient exposure to distorted auditory correlation with the reinforcement signal. The resulting
feedback is necessary to produce changes in LMAN rule is guaranteed to robustly (without parameter tun-
activity.) ing) improve song performance, even in recurrent net-
Subsequent informal proposals, not grounded in works of conductance-based spiking neurons and with
numerical simulation or modeling, have espoused delayed reinforcement.
more information-rich primary roles for LMAN in The model is able to produce learning in a biologi-
song learning. These include suggestions that LMAN cally realistic network model of the song system,
may provide detailed instructive or supervisory error with spiking neurons and an impoverished, delayed
signals to RA. Unfortunately, these suggestions lack reinforcement signal (Figures 6(a) and 6(b)). Learning
mechanistic proposals or models of how the AFP takes place online, and activity in the network evolves
might correctly compute and generate these informa- on the millisecond timescale, driven by 56 ms bursts
tion-rich, time-varying supervisory signals, leaving in the HVC inputs. Within 2000 iterations, the net-
the bulk of the computational burden of song learn- work output resembles the tutor song, which is a
ing unsolved and at the door of the AFP. In reality it recording of an actual zebra finch song (Figure 6(c)).
is possible that the AFP supplies such a supervisory Fiete et al. have shown that learning time even with
signal to RA, but no experimental or computational random LMAN input should scale well in the full-
studies exist yet to suggest that it does. sized premotor network and may converge in far
279
236 Birdsong Learning
HVC
mV
60
48
mV
RA
56
arb units
Motor
5 kHz
10 30 50 70
a Time (ms) b 75 ms
400
300
Error
200
100
0
0 500 1000 1500 2000
c Iterations
HVC
Empiric LMAN
synapse
R
RA LMAN
d e
HVC
LMAN
R
f W > 0 W < 0 W = 0 W = 0
Figure 6 Empiric synapses: results and predictions. (a) Activity of two model high-level vocal center nucleus (HVC) neurons (top), two
representative nucleus robustus archistriatalis (RA) neurons (middle), and one of the motor output pools (bottom; black: tutor, blue:
network response) during song learning. (b) Spectrograms of the tutor song and the output of the model song network before and after
1200 iterations of learning, together with the respective sound pressure waves. (c) Mismatch between the network output and the tutor
song decreases relatively rapidly. Learning has reached an approximate asymptote by 1000 iterations. (d,f) Prediction for bidirectional
synaptic plasticity from the empiric synapse model. (d) Possible experimental design for testing the empiric synapse rule in the birdsong
pathway involves focused stimulation in HVC, intracellular recording in RA, and fiber bundle stimulation of the lateral magnocellular
nucleus of the anterior nidopallium (LMAN)RA axons, together with control of the reinforcement signal (R). (e) Magnified schematic of a
single RA neuron and its inputs. Each RA neuron receives a few empiric synapses from LMAN and several from different HVC neurons; in
addition, RA receives a reinforcement signal. Blue (gray): active (quiescent) premotor inputs; green: empiric input; red: reinforcement
signal. (f) Protocol for induction of bidirectional synaptic plasticity in active HVCRA synapses: Simultaneous stimulation of an HVC
neuron and the LMAN axon bundle, followed by administration of reinforcement, should lead to potentiation of all stimulated HVC
synapses.
fewer iterations than sung by the typical zebra finch The model above is consistent with a large body
(100 000 renditions) during song acquisition. of experimental data in the song system and provides
The models central prediction is a specific paradigm concrete predictions for synaptic plasticity. How-
for the induction of bidirectional, heterosynaptic activ- ever, LMANs role in the model is to provide purely
ity-dependent synaptic plasticity in the premotor vocal random exploration. This assumption does not con-
pathway (Figure 6(f)) that should be possible to test tradict the data but is a limited view of what the
in slice experiments. AFP, a complex neural pathway, might be capable
280
Birdsong Learning 237
of. More-sophisticated strategies for AFP-delivered when the tutor song has opposing characteristics
perturbation may speed up learning. This avenue is (e.g., pitch or amplitude), at those two times. The
wide open for discovery by theory and experiment. synapses will be pulled in opposing directions to best
More generally, models of song learning based match the tutor song (Figure 7(c)) at each time.
on the AFP as a source of random or directed pertur- Because the learning task is realizable, there is some
bations to the premotor system could be used as value of synaptic strength that achieves the desired
comparative tools for studying the role of the basal output, and in the end this value will be reached.
ganglia thalamocortical loop in mammalian sensori- But the conflicting demands on synapse strength at
motor learning. the two times in the motif make approach to the
solution with a hill-climbing rule slower, even if the
parameter controlling learning step size is optimized to
Premotor Codes and Representations
produce the fastest possible learning without network
Up to 30 000 neurons are devoted to song production instability.
and learning in the zebra finch. Together, these neu- Yet answers to why questions, however compel-
rons drive only eight to ten muscles that produce ling, risk being just-so stories unless they generate
song. What are some of the functional requirements specific testable predictions. The work described
and costs that determine why so many neurons are above argued that unary codes may exist in HVC
required for song production? because of their utility for fast song learning. How-
The ultrasparse or unary coding of song by HVCRA ever, single-unit HVCRA activity and unary codes
neurons (Figures 2(a)) exacts a large energetic and have been observed only in adults. Thus, the central
space cost in terms of the number of neurons required prediction which if false would invalidate the
to encode a given set of patterns. To produce a sequence hypothesis is that activity in HVCRA neurons,
of 100 steps (L) with 200 neurons per time, unary even if nonsparse initially (early in the sensory
coding requires 20 000 neurons. At the other extreme, period), should be unary near the onset of senso-
a combinatorial code with a downstream readout capa- rimotor learning in juvenile zebra finches instead of
ble of distinguishing between different combinatorial emerging gradually after or toward the end of senso-
patterns would require far fewer neurons, scaling as log rimotor learning.
(L) instead of L; intermediate possibilities would The analysis leads to an additional prediction. To
require intermediate numbers of neurons. What are accommodate a larger song repertoire, HVC could
some of the properties of a unary code that might offset change in two a priori equally likely ways: (1) keep
its capacity costs and explain why it exists in HVC? HVC volume (number of neurons) fixed but increase
A possible explanation for ultrasparse HVC the number of bursts per neuron or (2) increase
sequences could be mechanistic: recurrent networks HVC volume but maintain a unary code. Energetics
cannot store many dense patterns because of increas- might favor the first possibility. But according to the
ing interference between patterns. Studies show that learning speed analysis, if zebra finches are under
sparsely active attractor networks can store more pressure to learn their songs quickly, then unary cod-
patterns. Nevertheless, in these studies, sparse is ing should be conserved, even at the cost of a larger
far denser than unary, so the reason for ultrasparse HVC. This suggests that coding in HVC across zebra
coding must lie elsewhere. finches and related songbirds with somewhat larger
Because oscine birdsong is learned, a natural place repertoires (measured by number of unique syllables)
to investigate the role of unary HVC codes is in should remain unary while HVC size scales with rep-
learning. There are actually two distinct roles that ertoire size. Indeed, existing experimental findings
unary codes could play in song learning. corroborate these predictions: HVC size scales with
First, an analytical and numerical study by Fiete repertoire size after correcting for overall brain size,
et al. shows that unary coding within HVC enables while RA, LMAN, or other song nuclei do not display
optimally fast learning of the feedforward motor similarly consistent relationships.
map from abstract sequences in HVC to sequences Second, analysis and simulation by Fiete et al. has
of muscle activation. If each HVC neuron were to fire shown that if the HVC code is unary and if, during
twice instead of once, at random times in a song a song rendition, performance is continuously evalu-
motif of fixed length, the fastest possible speed with ated and a reinforcement signal continuously delivered
which the inputoutput map can be learned is twice (albeit delayed), then the time taken to learn a song
as slow (Figures 7(a)7(d)). There is an intuitive is independent of the length of the learned song.
explanation of the underlying mathematical reason Because the HVC code is unary in zebra finches, this
for this effect. Suppose an HVC neuron and its prediction can be used in experiments (where learning
corresponding HVCRA synapses are active twice, time is monitored for birds trained on tutor songs of
281
238 Birdsong Learning
HVC1
HVC2
HVC3
a
b Time
HVC1 RA
HVC2 RA
HVC3 RA
Wj
Wi
Random nonunary representation
d Unary representation
HVC1
HVC2
e
HVC1 RA
HVC2 RA
g
Figure 7 Unary coding in high-level vocal center nucleus (HVC) helps speed the acquisition of the motor map. (a) The activity of
three hypothetical HVC neurons. The first two are each active only once in the song motif, but the third is active twice. (b) Hypothetical
pitch of the tutor song (black) and the pupil network (gray). (c) Direction in which synapses should change to reduce error between tutor
and pupil pitch. Synapses from HVC neuron 3 should be strengthened at the first and weakened at the second activity burst to improve the
tutorpupil match. Such conflicting demands on the synapses of neurons that are active at two or more random times in a motif cause a
slowdown in learning speed. (d) Contours of iso-error in the learning surface. Learning a feedforward map in a network with unary coding
in the top layer is like learning on an isotropic cost surface and can be fast. Denser coding in the input layer produces correlations and
makes the learning surface anisotropic. To keep the error from diverging along the steep directions, the learning rate must be kept low.
As a result, best-case learning is slower than in the isotropic case. HVC activity tuned to acoustic features may be helpful for generalizable
learning. (e) If HVC neurons fired multiple bursts at selected points when the acoustic features of tutor song are similar, rather than at
random times, there would be no interference in the learning update, (f). If HVC neurons acquired such a tuning to features in the tutor
song, it could be easy for the bird to quickly reproduce a specific heard sound by activating the requisite sound-tuned HVC neuron. RA,
nucleus robustus archistriatalis.
different lengths) as an indirect probe of whether the of song, then the time taken to learn a tutor song
elusive reinforcement signal in the song pathway is should scale linearly with its length. Similarly, if the
delivered online or in batch mode. If the reinforcement HVC code at each time consisted of a random pattern
signal is evaluated and delivered only once, at the end of activations, with half of all neurons active per time
282
Birdsong Learning 239
(dense code), learning time should scale with song See also: Bird Song Systems: Evolution; Birdsong
length (even if reinforcement is delivered online). Learning: Evolutionary, Behavioral, and Hormonal Issues;
Do other songbirds also display unary HVC repre- Birdsong: The Neurobiology of Avian Vocal Learning;
sentations of song? At present, single-unit recordings Hebbian Plasticity; Learning, Action, Inference and
Neuromodulation; Motor Sequences; Signal Production
of RA-projecting HVC neurons are available only
and Amplification in Birds; Spike-Timing-Dependent
from zebra finches. Learning with a pure time code Plasticity Models; Spiking Neuron Models; Synaptic
is nongeneralizable: like one piece of magnetic tape, Mechanisms of Learning; Vocal Communication in Birds.
each pattern is used to produce song at only one
specific time. An entirely different set of HVC neu-
rons and HVCRA weights must be used and trained Further Reading
to produce the same sound if it recurs elsewhere in the
motif. Therefore, songbirds that rapidly acquire and Brainard MS and Doupe AJ (2000) Interruption of a basal ganglia
forebrain circuit prevents plasticity of learned vocalizations.
imitate new songs (albeit after slowly acquiring a Nature 404(6779): 762766.
first song) may use a different encoding strategy Doya K and Sejnowski TH (1995) A novel reinforcement model of
(e.g., tuning codes based on acoustic features) from birdsong vocalization learning. In: Tesauro G, Touretzky DS,
a time-code (Figures 7(e)7(g)). In addition, song- and Leen TK (eds.) Advances in Neural Information Processing
birds with vastly larger repertoires, regardless of Systems, pp. 101108. Cambridge, MA: MIT Press.
Fiete IR, Burger L, Senn W, and Hahnloser RHR (2007) Enforcing
the speed of song acquisition, may display different global coding constraints by synaptic competition a model for
HVC codes because of capacity constraints. the formation of long ultrasparse sequences in the songbird.
Submitted, 2007. (See also Society for Neuroscience Abstracts,
2005.)
Conclusions Fiete IR, Fee MS, and Seung HS (2007) Model of birdsong learning
based on gradient estimation by dynamic perturbation of neural
Advances in the acquisition of behavioral and anato- conductances. Journal of Neurophysiology 98: 20382057.
mical data and neurophysiological recordings from Fiete IR, Hahnloser RH, Fee MS, and Seung HS (2004) Temporal
awake, singing birds are rapidly turning the birdsong sparseness of the premotor drive is important for rapid learning
in a neural network model of birdsong. Journal of Neurophysi-
circuit into an ideal system for unraveling the
ology 92(4): 22742282.
mechanisms underlying motor control and learning. Jin DZ, Ramazanoglu F, and Seung HS (2007) Intrinsic bursting
Theoretical models, by quantitatively validating or enhances the robustness of a neural network model of sequence
eliminating candidate explanations of network func- generation by avian brain area HVC. Journal of Computational
tion, have helped lay a groundwork for understand- Neuroscience 23(3): 283299.
ing the dynamical and coding principles that lead to Jun JK and Jin DZ (2007) Development of neural circuitry for
precise temporal sequences through spontaneous activity, axon
such functionality. Theoretical models also provide a remodeling, and synaptic plasticity. PLoS ONE 2(8): e723.
functionally motivated set of predictions about neural Li M and Greenside H (2006) Stable propagation of a burst
activity, connectivity, coding, and plasticity that nar- through a one-dimensional homogeneous excitatory chain
row the choices of future experiments and are pres- model of songbird nucleus HVC. Physical Review. E, Statistical,
ently ripe for experimental testing. Future experiments Nonlinear, and Soft Matter Physics 74(1.1): 011918.
Troyer TW and Doupe AJ (2000) An associational model of
and modeling in songbird species with more-flexible birdsong sensorimotor learning I. Efference copy and the
learning behaviors should greatly enhance understand- learning of song syllables. Journal of Neurophysiology 84(3):
ing of generalizable motor learning and control. 12041223.
283
284
Open access, freely available online
Primer
T
here is a tradition in biology of using specic animal
models to study generalizable basic properties of a
system. For example, the giant axon of squid was used
for the pioneering work on nerve transmission; the fruit y
(Drosophila) has played a key role in researchers discovering
the role of homeobox genes in embryogenesis; the sea slug
(Aplysia) is used to study the molecular biology of learning;
and the round worm (Caenorhabditis elegans) is used to study
programmed cell death. Basic insights gained from these four
systems apply widely to other multicellular animals. Here, I
will review basic discoveries made by studying birdsong that
have helped answer more general questions in vertebrate
neuroscience.
285
the right and left sides of the brain can operate, to some RA was conrmed by recording from HVC and RA while the
extent, independently, each responsible for a different array bird sang [33], but the manner in which the sounds of song
of sounds. In birds such as the canary, the chafnch, and the were represented in HVC remained unclear. This issue was
white-crowned sparrow a majority of the sounds of song are resolved by recording from individual HVC neurons that
produced by the left syringeal half, under the control of left, projected to nucleus RA. These neurons, it was shown, red
uncrossed pathways. This phenomenon has been referred to very sparsely and at narrowly dened times, each neuron
as left hypoglossal or left hemispheric dominance [13]. ring always during the same six-millisecond window while
the bird produced its single learned song [34]. The inference
Adult Variation, Neurogenesis, and Neuronal that these neuronsand the PDP of which they are part
Replacement carried the learned pattern of song seems inescapable. Since
The song system of birds is sexually dimorphic: it is better these are the very HVC neurons that are replaced when birds
developed in males, which usually sing more and produce modify their song [31], it follows that the replacement cells
a more complex repertoire than females. For example, the learn their score.
nucleus HVC of canaries is three times larger in males than But what, then, is the role of the AFP in song learning? It
in females; in zebra nches it is eight times larger [14]. In was known that the AFP was necessary for the acquisition but
seasonal singers such as the canary and song sparrow, song not for the production of learned song [35]. It was known,
system nuclei such as HVC are signicantly larger in the too, that the variable song typical of juvenile songbirds
spring than in late summer, after breeding stops [15,16]. became very stereotyped after bilateral lesions of the lateral
Cells in many of the song control nuclei are androgen and magnocellular nucleus of the nidopallium (LMAN) (part
estrogen sensitive [17]. The nucleus HVC of adult female of the AFP), from which it was inferred that LMAN played
canaries treated with physiological doses of testosterone a crucial role in fostering circuit plasticity necessary for
doubles in volume, and these birds start to sing in a male-like learning [36]. But the mechanism for this effect remained
manner [18]. Initially such changes were thought to result unknown. Two recent independent studies now show how
solely from growth of dendritic trees and synapse formation this effect comes about [37,38]. In this issue of PLoS Biology,
[19], but subsequently it was found that new neurons were lveczky and colleagues show that the LMAN neurons
added, too. These new neurons, as in embryos, are born that project to RA re in a quasi-random pattern when
in the wall of the forebrains lateral ventricle [20,21]. variable song is produced in juvenile birds. Thus, while
Interestingly, the addition of new neurons to nucleus HVC the HVCRA projection carries the learned song, the
occurred also in male and female adult canaries that had LMANRA projection carries the jitter that induces the
received no hormonal treatment [22]. variability in motor output necessary for the imitation of a
Earlier claims of neurogenesis in the adult mammalian model. This jitter, presumably, is imposed on the ring of the
brain [23,24] had met resistance [25]. Nucleus HVC yielded same RA neurons that receive the more orderly output from
the rst unambiguous example of adult neurogenesis, in HVC. When the LMAN neurons are silent (or absent), the
that individual cells labeled with a cell birth marker provided HVCRA pathway produces a stereotyped pattern; when the
neurophysiological recordings that were unmistakably LMANRA neurons are ring, song is more variable. This is
neuronal [26]. We now know that the recruitment of new a most elegant breakthrough. However, it is not clear exactly
HVC neurons is part of a process of constant replacement how this pathway functions in song learning. One possibility
[27,28]. This replacement is particularly active in canaries is that birds trying to imitate a model succeed by retaining,
during seasonal changes in the song repertoire [29]. Male from the variability generated, those patterns that more
canaries develop a new song repertoire each year. New closely approximate the model and discard the rest, thus,
neurons are constantly added, as well, to many regions of the over a period of time, achieving a perfect imitation. A second
adult avian telencephalon, where they are probably involved possibility is that the variable mismatch between a model and
in a variety of brain functions. There is no evidence of the attempted imitation drives output modication, so that
neuronal addition to other parts of the adult songbird brain patterns that had not occurred before now rst appear. Both
[22]. We now know that adult neurogenesis and neuronal mechanisms would depend on auditory feedback. It is the
replacement are probably common to all vertebrates [30]. rst time we are so close to a mechanism for vocal learning.
The song system of birds helped change the way in which we
think of brain circuits and their potential for rejuvenation Open Questions
and repair. Just as important, the discovery of neuronal The stage is set for many more insights. An unsolved
replacement has raised basic questions about the brain question is the extent to which the very pathways that
variables that set limits to learning [31]. produce learned song may also partake in the perception
of song. On a different front, why is it that HVCRA
Neurophysiology Offers Insights on the Mechanisms neurons are periodically replaced? It was thought that
for Vocal Learning changes in dendritic conguration, dendritic spines, and
From early on, the song system drew the attention of synaptic number and efciency provided all the plasticity
neurophysiologists. It was known that lesions of HVC and needed to change circuit conguration and explain how
the robust nucleus of the arcopallium (RA) affected the new information was acquired and remembered. But if so,
organization of song differently, the former being more why replace whole neurons? Is it possible that dendritic
devastating than the latter [6]. Likewise, stimulation of and synaptic changes underlying learning are less easy to
HVC during song interrupted and reset the song program, achieve in older neurons? If so, are older neurons replaced
something that did not happen if the stimulating electrode to reinstate a level of plasticity necessary for learning? Or
was in RA [32]. This hierarchical relation between HVC and might it be that in some cases the whole neuron, rather than
286
the synapse, is the unit of learning? In this latter scenario, 12. Nottebohm F (1999) The anatomy and timing of vocal learning in birds.
In: Hauser MD, Konishi M, editors. The design of animal communication.
changes associated with learning would be committed Cambridge (Massachusetts): MIT Press. pp. 63110.
as permanent gene expression changes akin to those 13. Nottebohm F (1977) Asymmetries in neural control of vocalization in the
characterizing cellular differentiation. Such a change would canary. In: Harnad S, editor. Lateralization in the nervous system. New
York: Academic Press. pp. 2344.
be a very stable way to encode learning, but it would have a 14. Nottebohm F, Arnold AP (1976) Sexual dimorphism in vocal control areas
major drawback: the more learning that occurred, the fewer of the songbird brain. Science 194: 211213.
15. Nottebohm F (1981) A brain for all seasons: Cyclical anatomical changes in
neuronal pupils would remain. Thus, we are left to wonder song control nuclei of the canary brain. Science 214: 13681370.
whether neuronal replacement takes place to make up for 16. Brenowitz EA, Nalls B, Wingeld JC, Kroodsma DE (1991) Seasonal
the lost plasticity of aging neurons, or whether it takes place changes in avian song nuclei without seasonal changes in song repertoire. J
Neurosci 11: 13671374.
as part of a normal recycling of memory space and of the 17. Arnold AP, Nottebohm F, Pfaff DW (1976) Hormone concentrating cells
memories it holds. in vocal control and other areas of the brain of the zebra nch (Poephila
During the early 1970s, before the song system was guttata). J Comp Neurol 165: 487512.
18. Nottebohm F (1980) Testosterone triggers growth of brain vocal control
discovered, it was widely believed that the learning of any nuclei in adult female canaries. Brain Res 192: 89107.
one skill had a wide representation in the vertebrate brain 19. DeVoogd TJ, Nottebohm F (1981) Gonadal hormones induce dendritic
growth in the adult brain. Science 214: 202204.
[39]. The discovery of discrete brain regions devoted to song 20. Goldman SA, Nottebohm F (1983) Neuronal production, migration and
learning and execution in the bird brain helped change differentiation in a vocal control nucleus of the adult female canary brain.
that view. It was also widely believed that the brains of male Proc Natl Acad Sci U S A 80: 23902394.
21. Alvarez-Buylla A, Nottebohm F (1988) Migration of young neurons in adult
and female vertebrates were virtually identical, with small avian brain. Nature 335: 353354.
allowances for the levels of circulating hormones. The song 22. Nottebohm F (1985) Neuronal replacement in adulthood. Ann N Y Acad
system changed that, too. And it was widely believed that Sci 457: 143161.
23. Altman J (1963) Autoradiographic investigation of cell proliferation in the
the anatomy of adult brains was set, but we now know that brains of rats and cats. Anat Rec 145: 573591.
the volume of brain structures can change seasonally and in 24. Kaplan MS, Hinds JW (1977) Neurogenesis in the adult rat. Electron
microscopic analysis of light radioautographs. Science 197: 10921094.
response to blood hormone levels. Most importantly, it was 25. Rakic P (1985) Limits of neurogenesis in primates. Science 227: 154156.
widely held that though a straggling few neurons might still 26. Paton JA, Nottebohm F (1984) Neurons generated in adult brain are
be added after birth to late-developing parts of the brain, recruited into functional circuits. Science 225: 10461048.
27. Kirn JR, Nottebohm F (1993) Direct evidence for loss and replacement of
brain cells, once lost, could not be replaced. Again, work projection neurons in adult canary brain. J Neurosci 13: 16541663.
on the song system changed the prevailing view. It may well 28. Scharff C, Kirn J, Grossman J, Macklis, Nottebohm F (2000) Targeted
be that our best understanding of how complex skills are neuronal death affects neuronal replacement and vocal behavior in adult
songbirds. Neuron 25: 481492.
acquired and how broken circuits can be xed will come not 29. Kirn JR, OLoughlin B, Kasparian S, Nottebohm F (1994) Cell death and
from humans, or other primates, but from the way birds learn neuronal recruitment in the high vocal center of adult male canaries are
temporally related to changes in song. Proc Natl Acad Sci U S A 19: 7844
their song. 7848.
30. Gross CG (2000) Neurogenesis in the adult brain: Death of a dogma. Nat
References Rev 1: 6772.
1. Thorpe WH (1958) The learning of song patterns by birds, with special 31. Nottebohm F (2002) Why are some neurons replaced in adult brain? J
reference to the song of the chafnch, Fringilla coelebs. Ibis 100: 535570. Neurosci 22: 624628.
2. Marler P (1970) A comparative approach to vocal learning: Song learning 32. Vu ET, Mazurek MK, Kuo YC (1994) Identication of a forebrain motor
in white-crowned sparrows. J Comp Physiol Psychol 71: 125. programming network for the learned song of zebra nches. J Neurosci 14:
3. Konishi M (1965) The role of auditory feedback in the control of 69246934.
vocalization in the white-crowned sparrow. Z Tierpsychol 22: 770783. 33. Yu AC, Margoliash D (1996) Temporal hierarchical control of singing in
4. Marler P, Tamura M (1964) Culturally transmitted patterns of vocal birds. Science 273: 18711875.
behavior in sparrows. Science 146: 14831486. 34. Hahnloser RHR, Kozhevnikov AA, Fee MS (2002) An ultra-sparse code
5. Liu WC, Gardner TJ, Nottebohm F (2004) Juvenile zebra nches can use underlies the generation of neural sequences in a songbird. Nature 419:
multiple strategies to learn the same song. Proc Natl Acad Sci U S A 101: 6570.
1817718182. 35. Bottjer SW, Miesner EA, Arnold AP (1984) Forebrain lesions disrupt
6. Nottebohm F, Stokes TM, Leonard CM (1976) Central control of song in development but not maintenance of song in passerine birds. Science 224:
the canary, Serinus canaria. J Comp Neurol 165: 457486. 901903.
7. Vates GE, Vicario DS, Nottebohm F (1997) Reafferent thalamo-cortical 36. Scharff C, Nottebohm F (1991) A comparative study of the behavioral
loops in the song system of oscine songbirds. J Comp Neurol 380: 275290. decits following lesions of various parts of the zebra nch song system:
8. Paton JA, Manogue KR, Nottebohm F (1981) Bilateral organization of the Implications for vocal learning. J Neurosci 11: 28962913.
vocal control pathway in the budgerigar, Melopsittacus undulates. J Neurosci 37. Kao HM, Doupe AJ, Brainard MS (2005) Contributions of an avian basal
1: 12791288. ganglia-forebrain circuit to real-time modulation of song. Nature 433:
9. Jarvis ED, Ribeiro S, da Silva ML, Ventura D, Vielliard, et al. (2000) 638643.
Behaviorally driven gene expression reveals song nuclei in hummingbird 38. lveczky BP, Andalman AS, Fee MS (2005) Vocal experimentation in the
brain. Nature 406: 628632. juvenile songbird requires a basal ganglia circuit. PLoS Biol 3: e153. DOI:
10. Bottjer SW, Johnson F (1997) Circuits, hormones, and learning: Vocal 10.1371/journal.pbio.0030153
behavior in songbirds. J Neurobiol 33: 602618. 39. Lashley KS (1950) In search of the engram. In: Physiological mechanisms in
11. Luo M, Perkel DJ (1999) A GABAergic, strongly inhibitory projection to a animal behaviour, Symp Soc Exp Biol IV. Cambridge: Cambridge University
thalamic nucleus in the zebra nch song system. J Neurosci 19: 67006711. Press. pp. 454482.
287
288
letters to nature
theless retain a significant amount of plasticity in the brain.
The song of zebra finches consists of three levels of organization:
Decrystallization of adult syllables, which are individual sound components of the song
separated by silent intervals; motifs, which are sequences of
birdsong by perturbation syllables; and bouts, which are sequences of motifs3. The spectral
structure of the syllables is unstable in the juvenile bird. Similarly,
of auditory feedback motifs and bouts are organized differently from song to song.
However, as the song gradually assumes its adult form, these
Anthony Leonardo & Masakazu Konishi different levels of organization become highly stereotyped, the
Computation and Neural Systems Program and Division of Biology, MC 216-76, variability of the spectral structure of the syllables becomes
California Institute of Technology, Pasadena, California 91125, USA extremely small, and the bird sings these syllables in a highly
.........................................................................................................................
predictable order. At this stage, the song is referred to as crystallized.
Young birds learn to sing by using auditory feedback to compare Some birds, like the zebra finch, maintain their crystallized song
their own vocalizations to a memorized or innate song pattern; throughout adulthood and are called age-limited learners. Open-
if they are deafened as juveniles, they will not develop normal ended learners like canaries can, in contrast, learn new songs in
songs1,2. The completion of song development is called crystal- adulthood4.
lization. After this stage, song shows little variation in its tem- The stability of song in age-limited learners was previously
poral or spectral properties. However, the mechanisms underlying thought to be maintained without auditory feedback1,2. Recent
this stability are largely unknown. Here we present evidence that reports, however, show that deafening these birds after crystal-
auditory feedback is actively used in adulthood to maintain the lization causes changes in song, suggesting that some auditory
stability of song structure. We found that perturbing auditory feedback is important for song maintenance throughout life57.
feedback during singing in adult zebra finches caused their song to Six to eight weeks after deafening, the song of adult zebra finches
deteriorate slowly. This decrystallization consisted of a marked deteriorates, and shows addition and deletion of syllables, abnormal
loss of the spectral and temporal stereotypy seen in crystallized repetition of syllables (stuttering), and modified syllable sequences.
song, including stuttering, creation, deletion and distortion By opening the auditory feedback loop, deafening shows how well
of song syllables. After normal feedback was restored, these the song pattern generator can maintain its original output without
deviations gradually disappeared and the original song was this signal. However, to learn how the song control system works,
recovered. Thus, adult birds that do not learn new songs never- it is necessary to manipulate auditory feedback without disabling
Detect singing
Generate feedback
Generate feedback
Figure 1 Protocols for constructing the feedback signals. In the adaptive protocol, monitored the birds song in 50-ms bins and detected the production of a single
the computer alternated between recording vocalizations and playing the last target syllable. In both protocols, the bird heard the superposition of his own
vocalization back to the bird. In the syllable-triggered protocol, the computer vocalizations and the computer-generated feedback.
466
1999 Macmillan Magazines Ltd 289
NATURE | VOL 399 | 3 JUNE 1999 | www.nature.com
letters to nature
either the auditory or vocal motor system. The ability to modify vocalizations and playing back the last vocalization to the bird. The
feedback signals dynamically allows particular spectral and tem- position of the feedback varied, depending on the exact timing of
poral components of the song to be chosen for manipulation. the birds song, and the feedback changed as the song changed (see
Furthermore, the effects of restoring normal auditory feedback Methods). To investigate the effects of feedback on a more local
after exposure to abnormal feedback can be observed. We show level, we designed a second syllable-triggered protocol (n 2 birds)
here how these manipulations produce dramatic changes in the in which we perturbed the feedback of only a single target syllable.
adult song. The computer recognized and played back a stored copy of this
We developed a computer-controlled system to perturb auditory syllable each time the bird produced it. The timing of the feedback
feedback and designed two feedback paradigms (Fig. 1). In both signal was fairly constant across different song deliveries.
methods, the computer detected singing and then played a feedback We recorded the songs of five adult male zebra finches in separate,
signal to the bird. Because the playback signals were delivered sound-attenuated chambers for several weeks before beginning the
through an overhead speaker, the birds heard a superposition of experiment. The variability of their songs was well within that found
natural and artificial sounds. In the first paradigm, the adaptive in crystallized songs. The birds were then placed in the feedback
protocol (n 3 birds), a computer alternated between recording system. After 14 months in this environment, four of the five birds
Motif 1 = Motif 2
i i A B C D E i A B C D E
9,000
Frequency (Hz)
0
0 2
Time (s)
Stuttering
A B B B B B B B B i A B
9,000
Frequency (Hz)
b A
0
0 3
Time (s)
i A B C D E F G i A B B C D E i A B H i A
9,000
Frequency (Hz)
0
0 4
Time (s)
Figure 2 Normal and decrystallized birdsong spectrograms. a, Two motifs of a decrystallized song of the bird shown in a. The syllable sequences now vary from
normal, crystallized zebra finch song. Each syllable has a stable and well defined motif to motif, indicating a loss of the fundamental temporal stereotypy which
spectral structure. The ordering of the song syllables is identical in the two motifs. characterizes crystallized zebra finch song. Motifs one and three contain new
The syllable sequence A B C D E represents the baseline motif for this bird. The song syllables (F, G, H) which were not present in the baseline motif shown in a.
is represent introductory notes, which occur in variable numbers at the Motif two contains a repetition of syllable B. After the feedback was removed, the
beginning of motifs. b, The decrystallized song of the bird shown in a, with a sequences shown in b and c gradually disappeared and only the baseline song
motif containing stuttering of the song syllable B. c, Another example of the was produced.
of old syllables from the song. Many zebra finch syllables contain
a sets of frequencies that are integer multiples of a common
fundamental frequency; such sets are called harmonic stacks (Fig.
2a, syllable D). The same three birds also showed spectral distortion
in their song syllables, including wobbles in the harmonic structure
of simple notes and the production of two superimposed harmonic
stacks, indicating a loss of precise control over the vocal organ (the
syrinx)8,9. There was considerable variability in the magnitude and
0
0 120
time course of the changes between different birds, but significant
Time (ms) changes were generally seen within six weeks. Finally, all the changes
in song structure described above could occur within different
10,000
motifs in the same bout. This is significant because one of the
hallmarks of crystallized song is its robust temporal sterotypyan
identical syllable sequence is maintained within all the motifs of a
bout. Decrystallized song, in contrast, lacks this stereotypy (Fig. 2c).
Two birds received feedback in the syllable-triggered protocol. In
one of the birds, significant changes in the spectrum of the target
Frequency (Hz)
468
1999 Macmillan Magazines Ltd 291
NATURE | VOL 399 | 3 JUNE 1999 | www.nature.com
letters to nature
0.8
0
0 540
Time (days of recovery)
Conditional entropy (bits)
1.5
0.5
0 540
Time (days of recovery)
Figure 4 Time course of decrystallization and recovery. a, The probability of one calculated from the syllable-to-syllable transition probabilities of the Markov
bird singing his baseline motif as a function of time. The shaded area represents chain for the song (Prob{A}*Prob{B|A}*Prob{C|B}*Prob{D|C}*Prob{E|D}). The two
feedback interval. Circles represent the probability calculated from the entire curves are identical to within 4% error. b, The conditional entropy of the song,
sequence of syllables (Prob {A B C D E}); asterisks represent the probability which is a measure of song variability, for the same bird.
the conditional entropy of the Markov chain can be used as an the temporal pattern of song organization and the spectral structure
estimate of the variability of the song on a particular day (Fig. 4b)12. of individual syllables. This finding is not consistent with the
The conditional entropy measures the uncertainty in observing classical depiction of song development in which a dynamic learn-
syllable B, given that syllable A was just produced. If each syllable is ing period in youth ends in a static maintenance period in adult-
followed only by a single other syllable (only one type of syllable hood. Thus, the distinction between age-limited and open-ended
sequence is produced), then the conditional entropy will be low, learners may not be as sharp as these names would indicate.
.........................................................................................................................
whereas if any syllable can follow any other syllable (many different
sequences are produced), the conditional entropy will be high. Methods
We examined the data for all three adaptive protocol birds using Birds were housed in custom-designed plexiglass cages and were paired with a
this method and found that birds in a decrystallized state had a female who lived in a separate partition of the cage. The playback speakers
significantly higher conditional entropy than they did in their output was calibrated to be approximately the same as the sound level of the
baseline state. Taken together, the time course of the baseline birds vocalizations (8090 dB SPL in the birds ear). At the start of the baseline
motif probability and the conditional entropy shows that the bird song recordings, the birds ranged from 130 to 300 days old (mean age was 200
in Fig. 4 went from being a low-variance singer, with essentially a days; zebrafinches reach adulthood at 90 days). The singing rates of the
single motif, to being a high-variance singer who produced a large different birds ranged from tens to thousands of motifs per night. However,
number of different motifs. there was no apparent correlation of age or singing volume with the magnitude
After the removal of the artificial feedback, the temporal and of the effects we observed. For the two birds who did not sing frequently, we
spectral variability in the songs of all three adaptive protocol birds collected no baseline data during the feedback period.
slowly decreased over the following weeks and months. The songs Adaptive feedback perturbation. Microphone data were sampled at 40 kHz,
eventually recrystallized and became stable again. Both the prob- after being low-pass filtered (10 kHz cutoff, 7-pole anti-aliasing filter). As
ability of the baseline motif and the conditional entropy of the song is shown in Fig. 1, the computer continuously acquired data from the
returned to their baseline levels. Stuttering, abnormal sequencing microphone in segments of 100 ms. Each of these bins of data was passed
of syllables and modified spectral organization gradually became through a software-based infinite-impulse response (IIR) filter (the trigger
infrequent and were replaced by the temporal and spectral organi- filter) that was used to detect song vocalizations while avoiding noise artifacts
zations characteristic of the original song. A complete recovery took (such as pecking, wing flaps and low amplitude calls). The playback of sound
about 24 months. The syllable-triggered bird made a partial occurred when the output of the trigger filter exceeded a root mean square
recovery by 8 months after the cessation of feedback. The slow threshold. The artificial feedback was thus produced with a 100-ms delay, and
progression of these changes suggests that auditory feedback does coincided with a silent interval in the birds song or with the following syllable,
not exert a great deal of instantaneous control over the production depending on the timing of the song with respect to the bin borders created by
of song, but instead has a cumulative effect on song maintenance. the computer. The alternation between recording vocalizations and playing
The recrystallized songs appeared to remain stable indefinitely. We sounds prevented a positive feedback loop from developing. The playback
tracked one bird for a year after his recovery and saw no departures stimulus was constructed from the 100 ms of sound that had caused the trigger
from the baseline song structure. event, and was narrowband-filtered before being played back to the bird. A
Our results demonstrate that zebra finches need auditory feed- narrowband filter was chosen to make the playback sounds difficult to localize.
back to maintain their songs in adulthood. This species, which does For one bird a wideband IIR filter was used. No significant differences in results
not modify its song or learn new songs after crystallization, appears were observed between this bird and the two narrowband birds. Delayed
to retain a great deal of plasticity in its auditoryvocal control feedback was used instead of other interfering stimuli such as white noise so as
system. This plasticity is sufficient to produce modifications in both to replicate the structure of syllables that these birds normally hear. The effects
achieved by using a series of IIR filters in conjunction with each other to Humans and monkeys have similar abilities to discriminate the
perform a logical operation (for example, power in band X and not in band Y). difference in frequency between two mechanical vibrations
The triggering was based on a small segment of the time-varying spectrum of applied sequentially to the fingertips13. A key component of
the target syllable which was unique to that syllable. An original copy of the this sensory task is that the second stimulus is compared with
crystallized trigger syllable was used as the playback stimulus for the duration the trace left by the first (base) stimulus, which must involve
of the experiment. Typical zebra finch syllables are 80150 ms in length. The working memory. Where and how is this trace held in the brain?
triggering resolution was 50 ms, which was short enough to ensure that the This question was investigated by recording from single neurons
feedback always overlapped the trigger syllable itself (and, partially, the in the prefrontal cortex of monkeys while they performed the
following syllable). The other syllables of the song received no feedback. somatosensory discrimination task. Here we describe neurons in
Spectral analysis. We calculated the time-frequency spectrogram for each the inferior convexity of the prefrontal cortex whose discharge
song with a sliding window (58 ms) in which each time point consisted of rates varied, during the delay period between the two stimuli, as a
the direct multitaper estimate of the power spectrum (with a time-bandwidth monotonic function of the base stimulus frequency. We describe
product NW of 3 or 4) (ref. 13). The data shown in Fig. 2 were analysed in this this as monotonic stimulus encoding, and we suggest that the
manner. A harmonic analysis was then used to determine the location and result may generalize: monotonic stimulus encoding may be the
magnitude of the jumps in the discrete spectrum of each syllable by calculating
the F-spectrum14 of each of the multitapered spectral estimates. This analysis
revealed additional statistically significant harmonic frequencies in the target a
syllable after the feedback was presented to the bird (P , 0:01) (Fig. 3).
Syllable classification. For each syllable, we extracted the length and a
500 ms
number of time-varying parameters (envelope, peak frequency, pitch, good-
ness-of-pitch and Wiener entropy13) based on the spectral analysis described PD KD Base Comparison KU PB
above. A modified K-means clustering algorithm15 was then used to partition
the syllables produced on a given day into subsets. These subsets were labelled
by the experimenter (syllable A, syllable B and so on; labelling was done blind to
the day on which the data were acquired). For each day of data, approximately b Set A c Set B
1,000 syllables were analysed. The standard deviations, which are shown as
y
34 42
x=
Comparison (Hz)
Comparison (Hz)
470
1999 Macmillan Magazines Ltd 293
NATURE | VOL 399 | 3 JUNE 1999 | www.nature.com
294
letters to nature
................................................................. however, that alteration of auditory feedback by deafening or by the
Interruption of a basal ganglia playback of masking sounds can lead to gradual deterioration of
the songs of adult zebra nches1,5. This suggests the alternative
forebrain circuit prevents hypothesis that the neural mechanisms involved in the evaluation of
auditory feedback remain active in adult birds, but that normally
plasticity of learned vocalizations there is no impetus for change in song because it matches the stored
song target (Fig. 1b). According to this model, the changes in song
Michael S. Brainard & Allison J. Doupe following disruption of auditory feedback reect an active process,
in which the mismatch between feedback and the stored song target
Keck Center for Integrative Neuroscience, Departments of Physiology generates a large but aberrant instructive signal, which drives non-
and Psychiatry, University of California San Francisco, San Francisco, adaptive changes in song (Fig. 1c).
California 94143-0444, USA This interpretation raises the question of where in the brain
.............................................................................................................................................. auditory feedback of the bird's song is evaluated. One candidate is
Birdsong, like speech, is a learned vocal behaviour that relies
greatly on hearing; in both songbirds1 and humans2 the removal of a Learning
auditory feedback by deafening leads to a gradual deterioration of mismatch between evaluation
of auditory
adult vocal production. Here we investigate the neural mechan-
developing song
and stored target auditory feedback
feedback
isms that contribute to the processing of auditory feedback during
the maintenance of song in adult zebra nches. We show that the large
deleterious effects on song production that normally follow instructive signal
learn to produce and stabilize their own song (Fig. 1ad). During
normal sensorimotor learning, young birds use auditory feedback large
instructive signal X
to match their own vocalizations to the previously memorized tutor is interrupted
song7. To carry out this matching process, the nervous system must
rst encode auditory feedback from the bird's own song, and then stable
song
song
motor song
compare that feedback with the stored song target (tutor song). This nuclei
comparison is presumed to generate an instructive signal that
gradually drives adaptive changes in the motor pathway for song Figure 1 Hypothetical ow of auditory feedback through the song system. a, During
(Fig. 1a). At the end of sensorimotor learning, the bird's song sensorimotor learning, auditory feedback from song is compared with a previously
resembles that of the tutor and in many species is retained relatively memorized song target. Differences between the bird's own song and the target song
unchanged throughout adulthood. result in an instructive signal that drives adaptive modication of song. b, After song
The unchanging nature of adult birdsong might reect a learning, the bird's own song closely approximates the target song. Consequently, there is
stabilization of synapses in the motor pathway for song production, little drive for further changes. c, Removal of feedback leads to generation of an aberrant
such that they become independent of auditory experience once instructive signal and non-adaptive changes to song. d, Interruption of the instructive
sensorimotor learning is complete. Recent experiments have shown, signal removes impetus for changes to song, even when feedback is altered.
NATURE | VOL 404 | 13 APRIL 2000 | www.nature.com 2000 Macmillan Magazines Ltd 296 763
letters to nature
measures of song stability to two birds that were recorded imme- prevented by lesions of LMAN but not by sham lesions (Fig. 4b).
diately before and after cutting the tracheosyringeal nerve (Fig. 2a), Thus, lesions of LMAN block deafening-induced vocal plasticity in
which is essential for much of the structure of individual syllables, adult zebra nches, not only of syllable structure, but also of the
but not for the timing of patterned expiration during song19. The overall temporal pattern of song.
measures indicated a complete deterioration of syllable structure The results are consistent with the hypothesis outlined above
(Fig. 4a, `nerve cut'), while the overall temporal pattern of song was (Fig. 1c), that disruption of auditory feedback in adult birds leads to
well preserved (Fig. 4b, `nerve cut'), demonstrating that changes to the generation of an instructive signal that actively drives changes in
these two aspects of song were dissociable. song, and that this signal either arises within or is conveyed through
Despite this dissociability, changes to the temporal pattern of the AFP. This possibility is supported both by the song-selective
song for the experimental birds paralleled changes to syllable properties of AFP neurons, and by the recent nding that this circuit
structure; deafening caused a signicant deterioration of the tem- is active during singing, even in deaf birds15. It is critical to our
poral pattern of song relative to controls, and this deterioration was experiment that, in deafened birds, song production is initially
6 6
pre
4 4
2 2
0 0
a i i a a a a b c? d e a c? a b c d e a b c d e
8kHz 8kHz
6 6
4 182 d post 4
2 2
0 0
6 6
pre
4 4
2 2
0 0
e? f? g? i bc d e f bc
8kHz 8kHz
6 6
36 d post
4 4
2 2
0 0
i bc d e f bc
8kHz 8kHz
6 6
321 d post
4 4
2 2
0 0
500ms
Figure 3 Changes to song following deafening in intact and lesioned birds. Shown are syllables and eventual deterioration of song into a series of short, noisy notes. In contrast,
spectrograms illustrating changes to song for 2 deafened birds (A and C) and their 2 songs of the lesioned brothers (B and D) recognizably retained all original syllables and
brothers, which additionally received LMAN lesions (B and D). Each bird initially sang a sequencing. The largest change for these birds was loss of repetition of one syllable (e in
xed repertoire of syllables (labelled with letters) in a stereotyped sequence or `motif' panel D). Songs for all groups of birds also speeded up, as previously observed in
(boxes). One of the smaller effects of deafening is shown in A: gradual loss or deterioration controls30 and following LMAN lesions21. This is apparent from the time scale bars
of some syllables (for example, c) and introduction of abnormal sequences (for example, (500 ms).
repeated a's). C, One of the larger effects of deafening: relatively rapid loss of identiable
2.5
(average score)
Zebra nches (Taeniopygia guttata) were raised in individual breeding cages with their
0.6
parents and siblings until approximately 60 days of age when they were transferred to all-
male group cages. Preliminary experiments suggested that there were greater effects of
0.4
deafening on young adult birds (90140 days of age) than on older birds (.200 days of
age). For this study, baseline recordings and experimental manipulations were therefore
0.2
restricted to birds that were 98140 days of age, except for one pair of brothers (one
lesioned and one intact) that were deafened at 204 days of age. These two birds were
0
included in Fig. 4, but not in statistical analyses. Where possible, we studied pairs of
Random brothers that were housed together, experienced manipulations at the same ages, and had
-0.2
similarity similar songs by virtue of exposure to each other and the same adult tutor. There was a
signicant correlation between syllable stability (see below) for deafened birds and syllable
-0.4
stability for their lesioned brothers, indicating that the identity of brothers indeed
control deaf lesioned sham nerve cut accounted for some of the variability in song stability (see also Fig. 4, where data that
& deaf & deaf
derive from brothers are connected by lines).
Figure 4 Summary of song stability following deafening of intact or LMAN lesioned birds.
a, Syllable stability. Each point corresponds to an individual bird and shows the average Lesions
similarity (see Methods) between syllables from baseline songs and those from songs Birds were deafened by bilateral cochlear removal7. For some birds, LMAN was electro-
recorded at least six months later, except for nerve-cut birds, where song changes were lytically lesioned 15 days before deafening. Lesions were stereotaxically targeted at
assessed within 1 week. Lines connect points corresponding to brothers. Bars show LMAN, and evaluated in tissue labelled with an antibody to CGRP (Fig. 2)29. For all but
one of the nominally lesioned birds, the percentage of LMAN that was removed bilaterally
means and standard errors for each group. b, Temporal stability. Each point shows the ranged from 70% to 100%. The remaining bird had only a 37% lesion, but it included the
temporal similarity (see Methods) between baseline songs and songs recorded after the region where RA projecting axons leave LMAN, and almost all CGRP labelling in RA was
indicated manipulations. eliminated. `Sham' lesions were entirely anterior and dorsal to LMAN.
NATURE | VOL 404 | 13 APRIL 2000 | www.nature.com 2000 Macmillan Magazines Ltd 298 765
letters to nature
Analysis of changes to song 22. Johnson, F. & Bottjer, S. W. Afferent inuences on cell death and birth during development of a
cortical nucleus necessary for learned vocal behavior in zebra nches. Development 120, 1324 (1994).
To characterize changes to the structure of individual syllables, we used a subjective
23. Akutagawa, E. & Konishi, M. Two separate areas of the brain differentially guide the development of a
scoring procedure similar to that employed in previous studies of birdsong1620. Observers
song control nucleus in the zebra nch. Proc. Natl Acad. Sci. USA 91, 1241312417 (1994).
who were blind to the experimental manipulation of each bird scored the similarity
24. Kirn, J. R. & Nottebohm, F. Direct evidence for loss and replacement of projection neurons in adult
between spectrographic representations of syllables from songs recorded before any canary brain. J. Neurosci. 13, 16541663 (1993).
manipulation and those from later songs. For each baseline syllable that was initially 25. Graybiel, A. M., Aosaki, T., Flaherty, A. W. & Kimura, M. The basal ganglia and adaptive motor
present in the repertoire of a bird (512 distinct syllables per bird), observers identied the control. Science 265, 18261831 (1994).
most similar syllable present in songs recorded at later dates. Observers additionally 26. Knowlton, B. J., Mangels, J. A. & Squire, L. R. A neostriatal habit learning system in humans. Science
judged the degree of similarity between each baseline syllable and its best match on a scale 273, 13991402 (1996).
of 0 (no similarity) to 3 (identical). As a measure of the maximum expected deterioration 27. Nakamura, K., Sakai, K. & Hikosaka, O. Effects of local inactivation of monkey medial frontal cortex
of syllables, we determined the random similarity between syllables of songs from in learning of sequential procedures. J. Neurophys. 82, 10631068 (1999).
unrelated birds (n = 12 pairs of songs). This random similarity averaged 0.9, indicating 28. Klein, D., Zatorre, R. J., Milner, B., Meyer, E. & Evans, A. C. Left putaminal activation when speaking a
some generic similarity between the syllables of unrelated zebra nches. Scores were second language: evidence from PET. Neuroreport 5, 22952297 (1994).
averaged across 3 observers (mean r2 for pairwise correlation of observers' scores, 0.81). 29. Bottjer, S. W., Roselinsky, H. & Tran, N. B. Sex differences in neuropeptide staining of song-control
All songs were recorded from birds isolated in sound attenuating boxes and were nuclei in zebra nch brains. Brain Behav. Evol. 50, 284303 (1997).
therefore `undirected'. Signicance of differences between groups was based on a criterion 30. Arnold, A. P. The effects of castration on song development in zebra nches (Poephila guttata). J. Exp.
of P , 0.05 using analysis of variance (ANOVA) and post-hoc Fischer's Protected Least Zool. 191, 261278 (1975).
Signicant Difference tests.
To assess temporal stability, songs were represented by the timing of syllable production,
while the spectral structure of individual syllables was ignored. For each bird, the timing Acknowledgements
pattern representing the most common motif before any experimental manipulation was We thank M. Stryker, M. Churchland, T. Troyer and K. T. Moortgat for comments on the
identied. We then searched songs from later recording sessions for the pattern of syllables manuscript, and A. Arteseros, G. Carrillo and A. Tam for technical assistance. This work
that provided the closest temporal match to the initial motif. At each offset (in increments was supported by a Burroughs Wellcome Fund fellowship of the Life Sciences Research
of 5 ms) between motif and song, we quantied the temporal overlap as the average of the Foundation (M.S.B.), and by the John Merck Fund, the EJLB Foundation and the National
percentage overlap of syllables and the percentage overlap of intervals. For each bird, the Institutes of Health (A.J.D.).
maximal overlap was calculated for 10 songs and then averaged to provide a measure
(`xcorr') of how well the temporal pattern of song was conserved following experimental Correspondence and requests for materials should be addressed to M.S.B. (e-mail:
manipulation. Because the delivery of some songs speeded up `proportionately' over time msb@phy.ucsf.edu).
(that is, with no apparent change in the relative durations of notes and intervals), we
allowed for proportional changes in the temporal pattern of the song (ranging from 75%
to 110% of initial song duration) in searching for the maximal overlap. To control partially
for varying complexity of different birds' motifs, we calculated the maximal overlaps
between each bird's motif and randomly selected songs from unrelated birds (`rand'). The
temporal stability of song was then expressed as a normalized value ranging from 0
.................................................................
(indicating no more preservation of temporal pattern than random) to 1 (indicating
perfect preservation of temporal pattern). The baseline motif for one pair of brothers had a SHATTERPROOF MADS-box genes
control seed dispersal in Arabidopsis
very simple temporal pattern (3 syllables) and consequently showed a high degree of
temporal similarity to all songs (including those from unrelated birds). This pair of birds
was therefore excluded from further analysis of changes in temporal pattern.
Sarah J. Liljegren*, Gary S. Ditta*, Yuval Eshed, Beth Savidge*,
Received 12 December 1999; accepted 1 February 2000.
John L. Bowman & Martin F. Yanofsky*
1. Nordeen, K. W. & Nordeen, E. J. Auditory feedback is necessary for the maintenance of stereotyped
song in adult zebra nches. Behav. Neural Biol. 57, 5866 (1992).
* Section of Cell and Developmental Biology, University of California at San
2. Cowie, R. & Douglas-Cowie, E. Postlingually Acquired Deafness: Speech Deterioration and the Wider
Consequences 1304 (Mouton de Gruyter, Berlin, 1992).
Diego, La Jolla, California 92093-0116, USA
3. Marler, P. A comparative approach to vocal learning: song development in white-crowned sparrows. Section of Plant Biology, University of California at Davis, Davis,
J. Comp. Physiol. Psychol. 71, 125 (1970). California 95616, USA
4. Doupe, A. J. & Kuhl, P. K. Birdsong and human speech: common themes and mechanisms. Annu. Rev.
..............................................................................................................................................
Neurosci. 22, 567631 (1999).
5. Leonardo, A. & Konishi, M. Decrystallization of adult birdsong by perturbation of auditory feedback. The fruit, which mediates the maturation and dispersal of seeds, is
Nature 399, 466470 (1999). a complex structure unique to owering plants. Seed dispersal in
6. Houde, J. F. & Jordan, M. I. Sensorimotor adaptation in speech production. Science 279, 12131216
(1998).
plants such as Arabidopsis occurs by a process called fruit
7. Konishi, M. The role of auditory feedback in the control of vocalization in the white-crowned sparrow. dehiscence, or pod shatter. Few studies13 have focused on identi-
Z. Tierpsychol. 22, 770783 (1965). fying genes that regulate this process, in spite of the agronomic
8. Bottjer, S. W. & Johnson, F. Circuits, hormones, and learning: vocal behavior in songbirds. J. Neurobiol. value of controlling seed dispersal in crop plants such as canola4,5.
33, 602618 (1997).
9. Nottebohm, F., Stokes, T. M. & Leonard, C. M. Central control of song in the canary, Serinus canarius.
Here we show that the closely related SHATTERPROOF (SHP1)
J. Comp. Neurol. 165, 457486 (1976). and SHATTERPROOF2 (SHP2) MADS-box genes are required for
10. Bottjer, S. W., Miesner, E. A. & Arnold, A. P. Forebrain lesions disrupt development but not fruit dehiscence in Arabidopsis. Moreover, SHP1 and SHP2 are
maintenance of song in passerine birds. Science 224, 901903 (1984).
functionally redundant, as neither single mutant displays a novel
11. Sohrabji, F., Nordeen, E. J. & Nordeen, K. W. Selective impairment of song learning following lesions
of a forebrain nucleus in the juvenile zebra nch. Behav. Neural Biol. 53, 5163 (1990). phenotype. Our studies of shp1 shp2 fruit, and of plants con-
12. Scharff, C. & Nottebohm, F. A comparative study of the behavioral decits following lesions of various stitutively expressing SHP1 and SHP2, show that these two genes
parts of the zebra nch song system: implications for vocal learning. J. Neurosci. 11, 28962913 control dehiscence zone differentiation and promote the lignica-
(1991).
13. Doupe, A. J. Song- and order-selective neurons in the songbird anterior forebrain and their emergence
tion of adjacent cells. Our results indicate that further analysis of
during vocal development. J. Neurosci. 17, 11471167 (1997). the molecular events underlying fruit dehiscence may allow
14. Solis, M. M. & Doupe, A. J. Contributions of tutor and bird's own song experience to neural selectivity genetic manipulation of pod shatter in crop plants.
in the songbird anterior forebrain. J. Neurosci. 19, 45594584 (1999). The MADS-box gene family encodes transcriptional regulators
15. Hessler, N. A. & Doupe, A. J. Singing-related neural activity in a dorsal forebrain-basal ganglia circuit
of adult zebra nches. J. Neurosci. 19, 1046110481 (1999).
involved in diverse aspects of plant development, and phylogenetic
16. Nordeen, K. W. & Nordeen, E. J. Long-term maintenance of song in adult zebra nches is not affected and functional studies show extensive redundancy between family
by lesions of a forebrain region involved in song learning. Behav. Neural Biol. 59, 7982 (1993). members69. SHP1 and SHP2 (previously known as AGL1 and
17. Vu, E. T., Mazurek, M. E. & Kuo, Y. C. Identication of a forebrain motor programming network for
AGL5) were two strong candidates for functional redundancy, as
the learned song of zebra nches. J. Neurosci. 14, 69246934 (1994).
18. Yu, A. C. & Margoliash, D. Temporal hierarchical control of singing in birds. Science 273, 18711875 (1996). they share 87% identity at the amino-acid sequence level and show
19. Vicario, D. S. Contributions of syringeal muscles to respiration and vocalization in the zebra nch. almost identical expression patterns in developing Arabidopsis fruit
J. Neurobiol. 22, 6373 (1991). (refs 1012; and C. Ferrandiz, Y.E., J.L.B. and M.F.Y., unpublished
20. Morrison, R. G. & Nottebohm, F. Role of a telencephalic nucleus in the delayed song learning of
socially isolated zebra nches. J. Neurobiol. 24, 10451064 (1993).
results).
21. Williams, H. & Mehta, N. Changes in adult zebra nch song require a forebrain nucleus that is not
necessary for song production. J. Neurobiol. 39, 1428 (1999). Present address: Calgene, Davis, California 95616, USA.
300
J Comp Physiol A (1986) 159: 827-840 Journal of
Comparative Sensory,
Neural,
and
P h y s i o l o g y A B....lo~,
Physiology
9 Springer-Verlag 1986
Summary. 1. Eight large interneurons descending direction inhibited the response to target move-
in the dragonfly (Aeshna umbrosa, Anax junius) ment (Fig. 8).
ventral nerve cord from the brain to the thoracic 6. These neurons mediate, in part, the visual
ganglia were identified anatomically with intracel- control of flight orientation. I propose that they
lular dye injection (Fig. 3). All eight were strictly convey turning signals to the wing motor in re-
visual and responded only to movements of small sponse to objects moving relative to the animal.
patterns, such as black squares, 'targets', moving
on a white background.
2. The target interneurons all projected from
the protocerebrum at least as far as the metatho-
racic ganglion. Within the protocerebrum they ar- Introduction
borized in the posterodorsal neuropil region, near Descending interneurons in insects receive input
the base of the circumesophageal connectives from the sensory structures of the head and pro-
(Fig. 3). vide input to thoracic motor centers to initiate or
3. The receptive fields of six of the cells were guide behavior (Tanouye and Wyman 1980; Blon-
large, including most of the forward hemisphere deau 1981; Pearson and Robertson 1981; Bacon
of vision. For five of these, spiking responses were and M6hl 1983; Bicker and Pearson 1983; Rei-
often restricted to a much smaller region within chert and Rowell 1985). From studies of descend-
the receptive field, with stimulation of other areas ing interneurons in a variety of insects, such as
yielding only subthreshold responses (Figs. 4 and grasshoppers (Catton and Chakraborty 1969;
5, Table 1). Rowell 1971; Bacon and Tyrer 1978; Simmons
4. The pattern of selectivity for target size var- 1980; Rowell and Pearson 1983; Kien and Altman
ied, with some neurons responding only to small 1984), flies (Strausfeld and Bacon 1983), and
targets, some showing consistent responses over moths (Olberg 1983), two generalizations emerge:
a wide range of target sizes, and one preferring 1) Many descending interneurons are multimodal
larger targets (Fig. 6, Table 1). and therefore appear relatively nonspecific to the
5. Five of the interneurons were directionally experimenter (but see Olberg 198t b; Reichert et al.
selective. Movement in the antipreferred direction 1985) and 2) single descending interneurons usual-
elicited hyperpolarizing responses in two of them. ly do not elicit movements by themselves. The only
Movements of large patterns, such as a checker- published exceptions to the second generalization
board pattern covering the forward hemisphere, are the descending interneurons which elicit song
elicited opposite directional responses, i.e., hyper- in the cricket (Bentley 1977) and the giant fibers
polarizations in the preferred target direction and which initiate escape in flies (Tanouye and Wyman
subthreshold depolarizations in the antipreferred 1980; Tanouye and King 1983).
direction (Fig. 7). A large pattern moving in any If descending interneurons appear relatively
Abbreviations: D I T Dorsal intermediate tract; D M T Dorsal nonspecific in their sensory reponses, how is the
median tract; M D T Median dorsal tract; V N C Ventral nerve precision which characterizes many insect behav-
cord; D C M D descending contralateral movement detector iors (e.g. Collett and Land 1978) achieved? Kien
301
828 R.M. Olberg: Identified visual target interneurons in the dragonfly
302
R.M. Olberg: Identified visual target interneurons in the dragonfly 829
within the cord and to determine its tract through the ganglion.
I photographed good sections within the connective and gangli-
on under phase contrast with Polaroid 667 Film and marked
the fluorescing axonal cross sections.
To determine the three dimensional arrangement of den-
dritic arborizations I sectioned the brain sagittally (30 gm) and
reconstructed the neurons from the sections with a drawing
tube.
In the adult dragonfly, the brain and optic lobes bend up-
ward to lie in a plane that is nearly at right angles to the
plane of the rest of the nervous system. Dorsal tracts in the
subesophageal and thoracic ganglia are therefore continuous
with and homologous to posterior tracts in the brain. For sim-
plicity I will refer here to the posterior surface of the brain
as its dorsal surface.
Results
M o s t l a r g e d i a m e t e r a x o n s in t h e a d u l t d r a g o n f l y
thoracic nerve cord were visual. The eight inter-
neurons described here have been grouped together
because they shared certain characteristics, both
a n a t o m i c a l a n d p h y s i o l o g i c a l . T h e y all d e s c e n d e d
from the brain, and their axons were among the
l a r g e s t in t h e v e n t r a l n e r v e c o r d ( V N C ) . T h e y all
responded selectively to small pattern (target) mo-
tion and not to the movement of large textured
Fig. 2. Identification of target interneuron axons in cross sec- backgrounds. Each was recorded and dye-injected
tion of prothoracic ganglion. Top : Double exposure, using epi- three or more times and showed consistent sensory
fluorescent illumination and phase contrast to show individual- responses and anatomy.
ly stained axon (MDT 1, arrow) with background detail. Bot- O f t h e e i g h t t a r g e t i n t e r n e u r o n s , five s h o w e d
tom: Enlargement of area outlined above showing the locations
of the target interneurons. Two arrows from DIT3 indicate directional preferences. At least one target inter-
observed variation in the axon location. There was some varia- n e u r o n in e a c h c o n n e c t i v e w a s s e l e c t i v e f o r e a c h
tion in the order of MDT2, 3, and 4, but order shown here d i r e c t i o n , left, r i g h t , u p a n d d o w n . ( D i r e c t i o n a l
was most typical. (Scale bar 50 gin) p r e f e r e n c e s will b e p r e s e n t e d h e r e o n l y f o r t h e in-
terneurons whose axons extend down the right
Control experiments showed that these movements did not elicit c o n n e c t i v e . T h e i r s y m m e t r i c a l p a r t n e r s in t h e left
responses in the identified neurons except for the neuron c o n n e c t i v e s h o w e d t h e e x p e c t e d r e v e r s a l in p r e f e r -
(DMTI) whose receptive field was in the backwards direction. e n c e s t o left a n d r i g h t . ) S e v e n o f t h e i n t e r n e u r o n s
One series of experiments (Fig. 8) required simultaneous h a d r e c e p t i v e f i e l d c e n t e r s s o m e w h e r e in t h e f o r -
movements of a small black target and a larger black and white
ward visual hemisphere, and the eighth (DMT1)
striped grating pattern. The grating pattern consisted of a series
of equally spaced black stripes, printed with a Kodak .copying l o o k e d b a c k w a r d ( F i g . 3).
machine on a transparent acetate sheet. I used an XY recorder
(MFE) to move the grating back and forth about 5 cm in front
of the projection screen. The target was projected through the Anatomical identification
acetate sheet onto the screen. The black target was visible even
on the dark stripes as a darker black square. Injection of Lucifer Yellow into the axons within
the cervical connective between the subesophageal
Anatomy. After recording from an axon, I injected Lucifer Yel- a n d p r o t h o r a c i c g a n g l i a u s u a l l y e n a b l e d m e t o see
low (5-10 nA) for 7 to 30 min. After allowing an additional and draw the soma and major dendritic arboriza-
diffusion time of 3045 rain, I dissected out the head and thor-
tions within the brain. The dye-filled axon profiles
acic ganglia. These I fixed in 4% formalin in a phosphate buffer
solution (10min) followed by 4% formalin in methanol w e r e a l w a y s v i s i b l e in c r o s s s e c t i o n s o f t h e c e r v i c a l
(45 min). I dehydrated the ganglia in propanol and cleared them c o n n e c t i v e a n d t h e p r o t h o r a c i c g a n g l i o n ( F i g . 2,
in methyl salicylate. Each filled neuron was drawn using a top).
Nikon drawing tube on a fluorescence microscope (Olympus I have grouped and named the target interneu-
BHS) using a fluorescent pencil and UV light source to illumi-
nate the drawing. The ganglia were then embedded in soft Spurr rons based on the longitudinal tract through which
resin for sectioning. The cervical connectives and prothoracic t h e y p a s s in t h e p r o t h o r a c i c g a n g l i o n . T a r g e t i n t e r -
ganglion were sectioned transversely (20 gin) to locate the axon neuron axons always traveled through one of three
303
Fig. 3. Anatomical identification of the giant descending target interneurons. Camera lucida drawings (center) of Lucifer Yellow
stained neuron profiles in the brain, subesophageal ganglion and prothoracic ganglion viewed from dorsal aspect. Reconstruction
of serial sagittal sections (upper right) gives view of the same neuron looking from the right side of the brain (dorsal is left,
ventral right). Arrows in cross-sections of the cervical connectives (left top) at the anterior margin of the prothoracic ganglion
304
and cross-sections of the prothoracic ganglion (left bottom) show positions of the axons, retouched with white ink. (Scale bars
50 ~tm). Diagram of projection screen (lower right) indicates location and direction of greatest response to target movement.
(MDT3 and DIT3 were not directionally selective. DMT1 gave no responses to target movement on the screen - see text.)
In this and remaining figures, upward and rightward on screen mean upward and rightward with respect to the animal. (H,
Horizon; Z, line directly above the animal)
305
306
307
834 R.M. Olberg: Identified visual target interneurons in the dragonfly
Table 1. Anatomical and response characteristics of the target interneurons. Where there were discrepancies, the most commonly
observed responses are shown. Soma locations are with respect to axons recorded in the right connective
Number of recordings 5 3 3 4 5 8 3 6
Soma location in brain LT RV LT RV LV RV LD LD
Left (L), Right (R),
Top (T), Dorsal (D)
Ventral (V)
Directional preference U L ND D R L ND -
Left (L), Right (R),
U p (U), Down (D)
Nondirectional (ND)
Spontaneous spike rate 0 0 0 0 0 0 0 0
Max. number of spikes to 7 10 4 20 20 8 17 8
target motion
Light on response E E E 0 E E E E
EPSP(E), No response (0)
Light off response E E E S E E S E
EPSP(E), Spike (S)
Air puff response 0 0 0 0 E 0 0 0
EPSP(E), No response (0)
Greater response to = = B B B = B -
white (W) or black (B) target?
Preferred target size A S L S S A A -
Small (S), Large (L), All (A)
Range in number of quadrants 8-10 7-10 7-11 4 7 10-12 10-12 12 -
where EPSPs were elicited
R a n g e i n number of quadrants 0- 7 1- 5 2- 5 4 1 11 1- 3 8 I0 -
where spikes were elicited
longitudinal tracts through the prothoracic gangli- mushroom bodies. All of their axons projected
on, the median dorsal tract (MDT), dorsal inter- down the ipsilateral connective. The soma of DIT1
mediate tract (DIT) or the dorsal median tract was also ventral, but lateral to those above. The
(DMT). The tract chosen by a given interneuron soma of DMT1 was located against the mid-dorsal
was consistent, and even the position of the axon surface of the brain.
within the tract showed little variation (Fig. 2, bot- All of the target interneurons arborized in the
tom). postero-dorsal region of neuropil very near the
The eight target interneurons descended from point where their axons left the brain via the
the protocerebrum of the brain and projected at circumesophageal connective (Fig. 3). None of
least as far as the metathoracic ganglion. I could these interneurons made visible associations with
discern very little branching below the prothoracic the mushroom bodies, central body or protocere-
ganglion, due to the distance from the recording bral bridge.
site. The soma location and branching pattern Within the protocerebrum the branching pat-
within the brain and axon locations within the cer- tern of the neurons was quite consistent from ani-
vical connective and prothoracic ganglion were mal to animal, making each neuron easily recog-
used to identify the cells anatomically (Fig. 3). nizable by its profile alone. Their large size, consis-
Within the brain the processes of the target tent sensory responses, and consistent locations
interneurons were restricted to the protocerebrum within the longitudinal tracts of the prothoracic
(Fig. 3). The somata of three of them ( M D T I , ganglia provided further evidence that the neurons
MDT3, and DIT3) were located very near one an- were uniquely identified (Fig. 3).
other in an antero-dorsal group of somata just me- There was no o b v i o u s correlation between the
dial to the globuli cell cluster. All three of these profiles of the neurons within the brain and their
cells sent their axons down the connective contra- receptive field centers or directional preferences.
lateral to their somata. The somata of DIT2, Figure 3 includes the target locations and move-
M D T 2 and M D T 4 were located very near one an- ment directions on the projection screen to which
other, in a thin cortex of ventral cell bodies even each neuron responded maximally. Notice that
with or slightly medial to the alpha lobe of the D M T I , whose soma and axon were not close to
308
R.M. Olberg: Identified visual target interneurons in the dragonfly 835
Fig. 4. Response of DIT2 to target movement on various screen locations. Outlines indicate the entire screen divided into 45 ~ 45 ~
quadrants. Target (4 ~ black square) moved right, left, down, and up with movements centered across each quadrant. Bars under
traces show 200 ms stimulus in indicated direction. Only one spike (s) was elicited by target movements (spike clipped to show
synaptic potentials). Note subthreshold EPSPs on most of screen and IPSPs to antipreferred (rightward) target movement. (Move-
ments 45~ ms; Scale bar 5 mV; Cell 6-28-1)
any of the others, was the only cell whose receptive brush. None of these stimuli elicited spiking in the
field was not in the forward direction. target interneurons. Table 1 summarizes some ana-
tomical and sensory properties of the target inter-
neurons.
Electrophysiological recordings In most target interneuron recordings com-
U p o n penetrating an interneuron, I presented a pound EPSPs and IPSPs were visible even though
preliminary set of sensory stimuli to determine the recording site in the axon was a long distance
whether it was a target interneuron. I first moved (ca. 2 mm) from, and separated by a ganglion from
a white card (7.5 12.5 cm) in various locations the probable site of spike initiation in the brain.
and directions around the animal ca. 3 cm from Such long length constants were due to the large
the eye. The 3 mm black dot at its center was an diameter of the axons. I did not see PSPs in other,
effective stimulus, producing spiking in all of the smaller axons descending from the brain recorded
target interneurons. A control card without a dot during the study. Furthermore, when the largest
never elicited spiking from the forward looking tar- diameter fibers, M D T I and D M T I , were pene-
get interneurons (the backward looking DMT1 trated below the prothoracic ganglion, PSPs were
could not be tested in this way). A card covered no longer visible.
with black and white stripes also failed to elicit
spikes when moved at a variety of velocities.
To test the neuron for mechanosensory re-
Receptive fields
sponses, I blew air onto the animal from the front To determine the receptive field size and location,
and sides, moved each of its wings up and down I moved a 4 ~ or 16 ~ target up, down, left and right
and stroked its thorax and abdomen with a soft in each of the twelve 45 ~ x 45 ~ quadrants compris-
309
836 R.M. Olberg: Identified visual target interneurons in the dragonfly
-Z
-H
9 9 t~
Preferred Preferred
-Z -Z
-H -H
9 4 9
Antipreferred Antipreferred
Fig. 5. Variation in responsiveness and apparent receptive field size. Recordings of DITI in two different animals (left and right).
Top screens show responses to rightward (preferred) target movement across entire 180 ~ screen along horizontal lines indicated.
Bottom screens show responses to three leftward target movements. In this and remaining figures, spikes are clipped to show
PSP activity. (Scale bars 200 ms, 5 mV)
ing the projection screen (Fig. 4). The receptive tern of compound EPSPs in the two rcordings,
field area in which this target motion evoked however, was almost identical. The peaks of depo-
EPSPs was much greater than the area in which larization of the left cell in Fig. 5 corresponded
movements elicited spiking (Fig. 4). To quantify to the highest spike densities in the right cell. What
this I counted the number of screen quadrants in appeared to vary was the overall level of excitabili-
which I saw PSPs to target motion and the number ty. Even antipreferred motion (lower screens of
in which I saw spikes (Table 1). For all but MDT4 Fig. 5) elicited 9 spikes in the second DIT1, while
and DMT1 movements nearly anywhere on the it elicited none from the first. Similar variability
screen (10 to 12 quadrants) elicited EPSPs. occurred for M D T I and MDT2; the variability
The receptive field area in which target move- was less extreme in MDT3, DIT2, and DIT3 (Ta-
ment elicited spiking varied greatly across prepara- ble 1). For all of these neurons, variability between
tions for a given cell type. Figure 5 shows traces preparations was much greater than variability
recorded from DIT1 in two different preparations. over time within a given preparation. The spiking
Rightward, preferred-direction movements elicited receptive field area of M D T 4 was quite consistent
a total of only 4 spikes within a very restricted among preparations. It was always confined to a
region in the first DITI, but the same stimuli eli- ca. 30 ~ wide vertical strip along the midline above
cited 67 spikes in the second. The underlying pat- the horizon.
310
R.M. Olberg: Identified visual target interneurons in the dragonfly 837
DIT1 DIT2
32 ~
Fig. 6. Target size selectivity in DIT1 and DIT2. Square black targets of indicated sizes moved in preferred direction (right
for DIT1, left for DIT2) during the period indicated by the stimulus marker below bottom traces. Checkerboard pattern (8 ~
black and 8~ white squares) covered entire screen. (All movements 45~ Scale bar 5 mV; DIT1 cell 9-5-2; DIT2 cell 9-2-1)
I Directional selectivity
Fig. 7. Opposing directional preference in responses of target Five of the target interneurons showed clear direc-
interneuron to target and checkerboard pattern movement. Pre-
ferred direction for target movement in this cell (MDT4) was
tional selectivity (Fig. 3); movement in their pre-
downward (trace 3). Downward movement of large pattern ferred direction elicited spikes, and movement in
showed hyperpolarizing response (trace 1) and upward pattern the antipreferred direction usually did not (Fig. 7).
movement showed depolarizing response (trace 2). (Movement For two neurons (MDT2, DIT2), compound
45~ ms; Scale bar 5 mV; Cell 7-12-a2) IPSPs indicated that target movement in the anti-
preferred direction was inhibitory (Fig. 4).
The directionally selective target interneurons
Selectivity for target size
(MDT1, MDT2, MDT4, DIT1, DIT2) also
All of the target interneurons responded to object showed direction-dependent subthreshold re-
movement but not to movement of a large textured sponses to movement of the large checkerboard
background. This visual discrimination occurred pattern. For these cells, the preferred direction for
over a range of velocities from 45~ to 225~ the large pattern movement was opposite to that
When moving black squares of various sizes for the small target (Fig. 7, cf. 2nd and 3rd traces).
were presented to these interneurons, their re- Checkerboard pattern movements very rarely eli-
sponses fell into one of three types (Table 1). Some cited any spikes from target interneurons, but these
cells (DIT1, MDT2, MDT4) consistently preferred invariably resulted from pattern movement in the
311
838 R.M. Olberg: Identified visual target interneurons in the dragonfly
312
R.M. Olberg: Identified visual target interneurons in the dragonfly 839
although background motion in the antipreferred mation they carry down the cord. Second is their
direction elicits a depolarizing response, this depo- direct influence on the flight system. High fre-
larization is in fact inhibitory. This suggests the quency (200 Hz), intracellular stimulation of some
presence of at least two different ionic mechanisms, individual target interneurons produces move-
one hyperpolarizing and one slightly depolarizing, ments of from two to four wings, even when the
underlying inhibitory inputs to the target inter- animal is not flying (Olberg 1978; Olberg in prepa-
neurons. The nature of the inhibition underlying ration). Their selective visual responses and their
directional and size selectivity will be considered influence on the flight musculature imply that the
in more detail in a later paper. target interneurons control, in part, oriented flight
responses to objects whose images move across the
dragonfly's retina.
Comparison with other descending
The moving objects of greatest importance to
interneurons in insects
a flying dragonfly are flying insect prey, flying pre-
In its size, structure, and location within the central dators, and other dragonflies. Six of the target in-
nervous system, MDT1 bears a resemblance to the terneurons are dominated by visual input from a
orthopteran Descending Contralateral Movement forward direction above the horizon. This is the
Detector (DCMD, for review see Rowell 1971). direction in which prey are viewed during the ap-
The soma size and location, the profile within the proach, the dragonfly sweeping up from below its
brain, and the axon size and location within the intended prey (Mayer 1957). Potential predators
nerve cord are all quite similar (cf. O'Shea et al. (birds) also usually approach from above. Thus
1974). This suggests that MDT1 and DCMD may both prey and predators would elicit a response from
be homologous giant neurons. Like MDTI, various combinations of the target interneurons.
DCMD responds maximally to small pattern It seems likely that the activity of the target inter-
movement and is inhibited by large pattern move- neurons governs the first turning response of the
ment (Palka 1967; Pinter 1977, 1979). flying dragonfly to the moving prey or predator.
There are differences, however, in the visual
properties of DCMD and MDT1. Whereas Acknowledgements. I am indebted to David Wohlers for teach-
DCMD's visual input is strictly from the contralat- ing me embedding and sectioning techniques, and to Dawn
Tamarkin for her technical help. I thank Robert Pinter for
eral eye, MDT1 receives binocular input, strongly help with some of the experiments and with light level measure-
weighted for input directly in front of the animal. ments, and Andrea Worthington for her critical review of the
MDT1 is selective for upward pattern movement. manuscript. This work was supported by National Science
DCMD is not directionally selective. Foundation Grant No. BNS-84-06254, by a Cottrell College
Science Grant from Research Corporation, and by grants from
In general, the large, descending interneurons the Union College Faculty Research Fund.
that have been described in insects have not been
as selective as the dragonfly target interneurons
in their visual properties. Although directionally References
selective descending visual interneurons have been Bacon J, M6hl B (1983) The tritocerebral commissure giant
found in grasshoppers (Kien 1974, 1975; Reichert (TCG) wind-sensitive interneurone in the locust. I. Its activi-
et al. 1985) and in the moth Manduca (Rind 1983), ty in straight flight. J Comp Physiol 150:439-452
Bacon J, Tyrer M (1978) The tritocerebral commissure giant
these have responded to wide field movement. The (TCG): a bimodal interneurone in the locust, Schistocerca
target interneurons described here represent an ad- gregaria. J Comp Physiol 126 : 317-325
ditional, inhibitory step in the processing of the Bentley D (1977) Control of cricket song patterns by descending
visual signal. A target moving in the preferred di- interneurons. J Comp Physiol 3[16:19-38
Bicker G, Pearson K G (1983) Initiation of flight by an identi-
rection is only excitatory if it does not exceed a fied wind sensitive neurone (TCG) in the locust. J Exp Biol
certain dimension. Beyond that point, inhibition 104:289-293
outweighs the excitation produced by the target. Blondeau J (1981) Electrically evoked course control in the
fly Calliphora erythrocephala. J Exp Biol 92:143-153
Catton WT, Chakraborty A (1969) Single neurone responses
Role of the target interneurons in behavior to visual and mechanical stimuli in the thoracic nerve cord
of the locust. J Insect Physiol 15 : 245-258
There are two reasons to assign an important be- Collett TS, Land M F (1978) How hoverflies compute intercep-
havioral role to the target interneurons. First is tion courses. J Comp Physiol 125:191-204
the size of their axons. The eight axons represent Kien J (1974) Sensory integration in the locust optomotor sys-
tem. II. Behavioral analysis. Vision Res 14:1255-1268
more than 10% of the total cross-sectional area Kien J (1975) Neuronal mechanisms subserving directional se-
of each connective. Their large size implies a selec- lectivity in the locust optomotor system. J Comp Physiol
tive advantage to rapid transmission of the infor- 102:337-355
313
840 R.M. Olberg: Identified visual target interneurons in the dragonfly
Kien J, Altman JS (1984) Descending interneurones from the Pinter RB (1979) Inhibition and excitation in the locust DCMD
brain and suboesophageal ganglia and their role in the con- receptive field: spatial frequency, temporal and spatial char-
trol of locust behaviour. J Insect Physiol 30: 59-72 acteristics. J Exp Biol 80:191-216
Mayer G (1957) Bewegungsweisen der Odonatengattung Reichert H, Rowell CHF (1985) Integration of nouphaselocked
Aeshna. Osterreich Arbeit Jb Stadt Linz 4: 211-219 exteroceptive information in the control of rhythmic flight
Olberg RM (1978) Visual and multimodal interneurons in dra- in the locust. J Neurophysiol 53:1201-1218
gonflies. PhD Dissertation, Univ. of Washington, Seattle Reichert H, Rowell CHF, Griss C (1985) Course correction
Olberg RM (1981a) Object- and self-movement detectors in circuitry translates feature detection into behavioural action
the ventral nerve cord of the dragonfly. J Comp Physiol in locusts. Nature 315:142-144
141 : 327-334 Rind FC (1983) A directionally sensitive motion detecting neu-
Olberg RM (1981b) Parallel encoding of direction of wind, rone in the brain of a moth. J Exp Biol 102:253-271
head, abdomen, and visual pattern movement by single in- Rowell CHF (1971) The orthopteran descending movement de-
terneurons in the dragonfly. J Comp Physiol 142:27-41 tector (DCMD) neurones: a characterisation and review.
Olberg RM (1983) Pheromone-triggered flip-flopping inter- Z Vergl Physiol 73:167-194
neurons in the ventral nerve cord of the silkworm moth, Rowell CHF, Pearson KG (1983) Ocellar input to the flight
Bombyx mori. J Comp Physiol 152:297-307 motor system of the locust: structure and function. J Exp
O'Shea M, Rowell CHF, Williams JLD (1974) The anatomy Biol 103:265-288
of a locust visual interneurone; the descending contralateral Simmons P (1980) A locust wind and ocellar brain neuron.
movement detector. J Exp Biol 80:191-216 J Exp Biol 85:281-294
Palka J (1967) An inhibitory process influencing visual re- Strausfeld N J, Bacon JP (1983) Multimodal convergence in the
sponses in a fibre of the ventral nerve cord of locusts. J central nervous system of insects. In: Horn E (ed) Multimo-
Insect Physiol 13:235-248 dal convergence in sensory systems. Fortschr Zool 28, Gus-
Pearson KG, Robertson RM (1981) Interneurons coactivating tav Fischer, Stuttgart, pp 47-76
hindleg flexor and extensor motoneurons in the locust. J Tanouye MA, Wyman RJ (1980) Motor outputs of giant nerve
Comp Physiol 144: 391-400 fiber in Drosophila. J Neurophysiol 44 :405-421
Pinter RB (1977) Visual discrimination between small objects Tanouye MA, King DG (1983) Giant fibre activation of direct
and large textured backgrounds. Nature 270:429-431 flight muscles in Drosophila. J Exp Biol 105:241-251
314
J Comp Physiol A
DOI 10.1007/s00359-007-0223-0
ORIGINAL PAPER
123
315
J Comp Physiol A
Flight cage
123
316
J Comp Physiol A
nylon monoWlament Wsh line stretched between the ends of Head orientation
a Y-shaped fork of bamboo. This more rigid mounting gave
us better control of the beads movements. We believe that One of our goals in this project was to determine whether
the Wsh line was not visible to the foraging dragonXies, the dragonXy uses its dorsal fovea to Wxate the prey during
because they occasionally collided with it in Xight. We pursuit. When viewed from the side the front and back of
found no diVerence in the response of the dragonXies to the dragonXys head appear to be bounded by nearly straight
two diVerent methods of bead presentation. lines, and we used those lines to estimate the direction of
We found that the animals became habituated to our view of the center of the fovea. To do this we viewed iso-
experimental conditions and after several capture attempts lated Erythemis heads from directly above using a dissect-
stopped taking oV after our artiWcial prey object. In a few ing microscope and axial illumination. We rotated the
cases a small insect was applied to the bead; this provided a head, mounted on a goniometer for angle measurements,
reward to the foraging dragonXy and reduced the behav- and observed the pseudopupil until the center of the fovea
ioral habituation to the bead stimulus. Because of this was directed straight up. We identiWed the center of the
habituation, we usually released the dragonXies after fovea as the region with largest pseudopupil; the entire
23 days of videotaping and replaced them with fresh ani- fovea was about 22 wide, extending about 11 forward
mals. and backward from the center along the dorsal midline.
We then photographed the head from the side and mea-
Video sured the angles of the front and back of the head (Fig. 1).
In three animals we found that the center of the fovea was
High-speed video clips were captured with a Redlake approximately 34 (33.5, 34, 36) forward from the line
Motion Pro 2000 (Redlake MASD LLC, Tucson, AZ, of the back of the head and 19 (17.3, 19.5, 20.3) for-
USA) camera with an Elicar 90 mm macro lens, which pro- ward from the line of the front of the head. Given the 3
vided a 12 cm Weld at 1 m. We placed the camera from 0.5 variability in our angle measurements and also our uncer-
to 1.5 m away from the perching animals. To aid in our tainty in assigning a precise line to the front or back of the
analysis of head position, we designed and oriented the head, especially in images somewhat blurred by motion,
perches to encourage the dragonXies to perch either aligned we estimate that our measurements are only accurate to
with and towards the camera or at right angles to the axis of within about 5.
view. Because dragonXy foraging Xights are invariably
upward, we positioned the camera with the dragonXy at the
bottom of the cameras Weld of view. The camera was con- Results
nected to an Acme KB-108-1 Weld computer running
Microsoft Windows 2000 and Redlake MIDAS software In this study we obtained high-speed (500 frames/s) video
for capture at 500 frames per second and 1,280 1,024 clips of 128 pursuit Xights by Erythemis simplicicollis
pixel resolution. females. Forty-eight of these were views directly from the
front (n = 33) or directly from the side (n = 15), and we
Analysis analyzed these 48 clips for head position before and, if pos-
sible, during pursuit Xights. We estimate that these 48 clips
Video clips were digitized frame-by-frame by hand, using represent about 25 diVerent individual dragonXies. The 48
the Redlake MIDAS analysis software. The coordinates clips include data from both means of bead presentation;
were then exported to Microsoft Excel spreadsheets for the results from the two methods were indistinguishable
analysis. For each clip we digitized the bead position, cen- from one another. In this report we use the term prey to
ter of the head and two other points on the head. In side refer to the stimulus, a white bead. The bead was usually
shots, we also digitized the point at the tip of the tail. presented within about 15 cm of the animal and sometimes
Twelve of the clips were digitized a second time by a diVer- as close as 5 cm, resulting in short pursuit-Xight durations,
ent person with the results that were always within a few typically less than 200 ms (mean 168 ms; range
pixels on one another. 62480 ms). These Xight durations were similar to the aver-
To calibrate approximate distances we used average age duration of 184 ms measured in naturally foraging
dimensions of Erythemis females, 6 mm head width or dragonXies (Olberg et al. 2000).
41 mm body length. Because we found about 3% variability In this high-speed video study we conWrmed the conclu-
in the length of the animals we measured and because the sion of an earlier Weld study (Olberg et al. 2000) that drag-
animal was not always oriented precisely head-on or at right onXies use an interception strategy, i.e., the dragonXys
angles to the camera-viewing axis, we estimate that our dis- Xight course is aimed at a point in front of its intended prey.
tance calibrations are only accurate within about 5%. In almost all cases, when the prey maintained consistent
123
317
J Comp Physiol A
Xight speeds and direction, the dragonXys Xight path was (0.916) occurred with a latency of only 28 ms, very close to
nearly a straight line to the point of interception (Fig. 2ac). the 29 ms average wing response latency measurement.
Straight Xights such as those shown in Fig. 2ac do not
arise from a ballistic strategy in which the dragonXy Xies Head orientation
along a predetermined interception Xight track. Flight
tracks in which the prey changed direction (Fig. 2df) During pursuit, the dragonXys head was precisely oriented
showed that the dragonXy reacts to changes in the preys with respect to the prey. In 18 cases we were able to quan-
trajectory with changes in its own Xight path (Fig. 2df). tify the dragonXys head angle during the pursuit Xight.
Seven of these were viewed from head-on allowing us to
Behavioral response latencies determine the direction of view of the visual midline
(Fig. 4a), and 11 of these were from the side, allowing us to
Circuitous Xight tracks such as those in Fig. 2df allowed determine the direction viewed by the dorsal fovea
us to estimate the latency of the dragonXys reaction to (Fig. 4bc). With only one exception, we found that the
changes in the direction of its prey. In some Xights, we dragonXy oriented its head during Xight to maintain its tar-
could distinguish exaggerated wing strokes before prey get on or very near the crosshairs formed by the intersec-
induced turns (Fig. 3a, lower inset). The latency measured tion of the visual midline with the dorsal foveal band.
from prey deviation to wing correction ranged from 26 to Two features of the top trace in Fig. 4a suggest that the
42 ms (mean = 29 ms, s = 6.4, n = 6). We measured an rotation of the head to Wxate the prey on midline includes a
average latency of 38 ms (s = 5.1, n = 16) to an observable predictive component to prey tracking. Firstly the head
course correction after abrupt changes in prey direction. rotation at takeoV continues along the same slope (i.e.,
The annotated Xight track shown in Fig. 3 illustrates a angular velocity) that the prey moved about 26 ms earlier.
wing-response latency of 28 ms resulting in a visible turn in Thus the eye traces the path that the prey would have taken
the Xight track 36 ms after the prey turn. had it not reversed direction. The change of direction of the
A second method conWrms and reWnes our behavioral head rotation lags behind the change of direction of the
latency measurements. Rather than determining behavioral prey by about 26 ms. Secondly, as the prey swings back to
latency by inspection of wing movements or by Xight path the left during the later part of the trace, the head angle fol-
deviation, which requires an accurate assessment of the cue lows the prey angle with no obvious time lag.
used by the dragonXy, we examined the x-coordinate of the The histograms of Fig. 5 show that during most of the
dragonXy and prey locations over time. For example, in pursuit the dragonXy is kept within about 10 of midline
Fig. 3b a plot of the x component of the two paths illustrates and 10 of the foveal center. Examples with the prey mov-
graphically the dragonXy position lagging behind the prey ing forward (Fig. 5c) and backward (Fig. 5d) above the
position. The highest correlation between these two paths dragonXy reveal no obvious bias in the foveal tracking of
123
318
J Comp Physiol A
the prey. There may be a slight diVerence in tracking instance the dragonXy was viewed from head-on as it Xew
between cases in which the dragonXy Xies forward and precisely sideways in pursuit of the prey. For most of this
cases when it Xies backward in pursuit. The median angle pursuit the dragonXy kept its head nearly in level, rather
between prey and fovea during forward Xight is 4.3, than rotating the head to Wxate the prey on midline
indicating that the center of fovea is aimed behind the bead (Fig. 6b). We could not ascertain whether the preys image
relative to Xight direction. The median angle between prey was maintained on the lateral region of the dorsal fovea, but
and fovea during backward Xight is 3.1, indicating that the the head position as viewed from the front was consistent
center of fovea is again aimed somewhat behind the bead with this possibility.
relative to Xight direction. These slight skews notwithstand-
ing in the accumulated histograms (Fig. 5a, b), as well as in
each of the histograms of diVerent pursuit scenarios Discussion
(Fig. 5cf), the bead is kept within 10 of the center of the
crosshairs at least 85% of the time. Three principal points emerge from this high-speed video
Of the 18 clips analyzed for head orientation during study of dragonXy prey interception. (1) The dragonXy
Xight, only one (Fig. 6) did not follow the pattern estab- steers an interception course, leading its prey, (2) during
lished by all the others, suggesting that other visual pursuit pursuit the dragonXys head rotates freely with respect to
strategies are available to the foraging dragonXy. In this the rest of the body to Wxate the preys image on the cross-
123
319
J Comp Physiol A
Fig. 4 Fixation of the prey image on midline and dorsal fovea. Open to maintain the prey image near visual midline (clips 7-22i, 7-8b).
circles indicate angle from the dragonXy head to the prey. Filled dia- b, c Head orients to maintain the prey image near the center of the
monds indicate direction of view of midline (a) or dorsal fovea (b, c). fovea when the prey moves forward (a clips 7-29c, 6-29a) or backward
Dashed line indicates 0 ms, the time of takeoV. a Head orients in Xight (b clips 7-29a, 7-22e) relative to the dragonXy.
hairs formed by the intersection of dorsal foveal band with to keep the target Wxated. These results then require us to
visual midline, (3) the behavioral latency in response to the modify our earlier hypothesis (Olberg et al. 2000) for the
movement of the prey image is remarkably fast, typically control of interception Xight. We had hypothesized that the
less than 30 ms. In this report we have focused on relatively dragonXy would stabilize its head, and thus its eyes, on the
short pursuit Xights (usually 515 cm), because we could visual environment, responding with compensatory Xight
accurately measure head orientation only with close-up corrections to any drift of the preys image on the retina. In
camera views. However, short Xights such as these are not this way the preys image would be stabilized on the eye and
uncommon in foraging Erythemis females. They are not the Xight would describe an interception course. In a land-
restricted to such short forays, however; we also observed mark study of head and body positions and orientations of
them taking oV after prey at a distance of 0.5 m or more. blowXies in free Xight, van Hateren and Schilstra (1999)
Points (1) and (2) above imply that the dragonXy is not showed that head is rotationally stabilized relative to the
usually Xying in the direction viewed by the midline dorsal visual environment with rapid head saccades accompanying
fovea. Instead, the dragonXys Xight course is directed at a course changes. However, in the blowXy study Xies were
point in front of the preys movement, while the head rotates not actively orienting to an object, as were the dragonXies.
Fig. 5 Histograms showing distribution of midline or fovea angles prey angle is behind the center of the fovea. c, d Subsets of the side-
relative to the angle to the prey. Visual angle is subtracted from prey ways clips with target moving forward (dotted arrow, c) or backward
angle in each frame, and histogram of each clip is normalized to 100%. (dotted arrow, d) relative to the animal show no obvious bias. e, f Sub-
Normalized histograms are averaged. a Angle to prey is centered on vi- sets of the sideways clips with the animal Xying forward (dotted arrow,
sual midline in head-on clips. b Angle to prey is centered on fovea in e) or backward (dotted arrow, f) show no obvious bias
the accumulated sideways clips. Positive values in bf indicate that the
123
320
J Comp Physiol A
Target
image
moves
?? TSDNs
123
321
J Comp Physiol A
Eye specialization Sauseng et al. 2003). This movement usually results in the
objects image passing across the dorsal foveal band. The
Our results provide evidence for the role of the dorsal rocking head motion includes a signiWcant translational
foveal band in prey interception. It has been known for a component (Olberg and Seaman, unpublished) that could
long time that the dorsal ommatidia of the dragonXy show a allow the dragonXy to extract distance information based
remarkable degree of specialization, not only in their spec- on motion parallax, a possibility discussed in detail else-
tral sensitivity (Labhart and Nilsson 1995), but also in their where (Olberg et al. 2005). The pre-Xight head movements
large lens diameters and small inter-ommatidial angles also include Wxation of the prey image on the foveal mid-
(Sherk 1978; Horridge 1978), properties that result in a line, several examples of which can be seen in Fig. 4.
high acuity fovea (see Horridge 2005 for review). Various
authors (Sherk 1978; Sauseng et al. 2003) have suggested Response latencies
that this region could be important in prey capture, since
dragonXies invariably attack their prey from below, placing The higher spatial and temporal resolution in this study
the prey on the upward-facing retina. Sherk (1978) also allowed us to reWne our estimates of the latency between
suggested that the precise alignment of these facets and detecting and responding to object image motion. In an
their underlying rhabdomes could endow them with sensi- early study (Olberg et al. 2000), based exclusively on eval-
tivity to the pattern of polarization in the sky above, which uating dragonXy and prey Xight tracks, we estimated this
could be useful in orientation and navigation. These sug- latency to be between 33 and 50 ms, somewhat higher than
gested functions are not mutually exclusive. the average latency of 29 ms that we report in this study.
There are many examples of specialized eye regions The remarkable speed of this visual reXex may account for
serving behavioral functions in insects. One of the best- the impressive success rates of foraging dragonXies. Baird
known examples is the male-speciWc acute zone of the Xy and May (1997) measured a success rate of 76% in foraging
eye. This region is located on visual midline and about 20 Pachydiplax longipennis, and we found an astonishing
30 above the horizon (Land and Eckert 1985). Male combined capture rate of 97% in short Xights of Erythemis
houseXies pursuing females keep the image of the female simplicicollis and Leucorrhinia intacta (Olberg et al. 2000).
on this acute zone at least 80% of the time (Wagner 1986). Perhaps not surprisingly, the visual reXexes of dragon-
Furthermore, male-speciWc neurons in the lobula of the Xies foraging in outdoor sunlight are much faster than
optic lobe are directionally selective for small target move- would be predicted from electrophysiological studies in the
ment, precisely the properties appropriate for controlling laboratory. The shortest visual latency measured in
tracking behavior (Gilbert and Strausfeld 1991). descending interneurons in the cervical connectives was
Although there are some parallels between the tracking 45 ms. The descending interneurons synapse onto thoracic
Xight in male Xies mediated by the male-speciWc acute zone pre-motor or motor neurons, which must then excite mus-
and the prey pursuit Xight in dragonXies mediated by the cle, increasing appreciably the estimate for latency based
dorsal fovea, there are also fundamental diVerences. on lab measurements.
Although tethered Xies rapidly rotate their heads in all
planes to follow moving gratings (Gilbert et al. 1995), dur- The role of prediction in prey interception
ing tracking Xights the males body appears to remain
mostly in line with the head, which is directed towards its Whether the goal is to catch a ball, avoid a moving projectile,
quarry, another Xy. DragonXies show no tendency to align or grab a Xying insect out of the air, it is important to be able
their bodies towards their prey. The head is rotated to place to predict where a moving object will be some time in the
the preys image on the foveal midline, but the orientation future. This ability has been investigated in a variety of verte-
of the rest of the body is highly variable. The direction of brate animals and especially in humans (see Regan et al. 1998
Xight relative to the body can as easily be backwards or for review). The head movement plotted in the top trace of
sideways as forward, but it is always in a direction that Fig. 4a suggests that the dragonXy is able to predict the future
leads the prey for interception (Olberg et al. 2000). In this position of its prey. The Wrst head rotation in this trace shows
way the dragonXy pursuit is more similar to the interception a simple tracking response to the preys movement with a
Xights of some hoverXies (Collett and Land 1978) than to visual latency of about 26 ms. However, after the prey is again
the tracking Xights of houseXies. Wxated in Xight (at about 60 ms) the image is held on midline
The dragonXys dorsal fovea is probably also used for with no obvious response latency. This might suggest that the
estimating the distance, and therefore the size of a potential head movement is now governed by the angular velocity,
prey item (Olberg et al. 2005). Perched dragonXies typi- rather than the position, of the moving image. Responding to
cally bob their heads upward in response to an object pass- the velocity of the preys image would allow the dragonXy to
ing above them (Kirmse and Lssig 1971; Miller 1995; predict its future position (Adelman et al. 2003).
123
322
J Comp Physiol A
The predictive ability of the dragonXys visual response Horridge A (2005) The spatial resolutions of the apposition compound
is also seen in its choice of initial takeoV direction. The eye and its neuro-sensory feature detectors: observation versus
theory. J Insect Physiol 51:243266
Xight tracks in Fig. 2ac, show that the takeoV direction is Kirmse W, Lssig P (1971) Strukturanalogie zwischen dem system der
quite an accurate prediction of the line that the dragonXy horizontalen blickbewegungen der augen beim menschen und
should take to intercept its prey. Like the prediction con- dem system der blickbewegungen des kopfes bei insekten mit Wx-
trolling head orientation, determination of the best line for ationsreaktionen. Biol Zbl 90:175193
Labhart T, Nilsson D-E (1995) The dorsal eye of the dragonXy Sympe-
interception is based on the angular velocity of the preys trum: specializations for prey detection against the blue sky. J
image (Olberg and Henry, unpublished). The interplay Comp Physiol A 176:437453
between target Wxation and prediction of future target posi- Land MF, Collett TS (1974) Chasing behavior of houseXies (Fannia
tion underlies a far more sophisticated mechanism for prey canicularis): a description and analysis. J Comp Physiol A 89:
331357
interception by dragonXies than we previously proposed Land MF, Eckert H (1985) Maps of the acute zones of Xy eyes. J Comp
(Olberg et al. 2000). Physiol A 156:525538
Miller PL (1995) Visually controlled head movements in perched An-
Acknowledgments This study was supported in part by NSF RUI isopteran dragonXies. Odonatologica 24:301310
0211467 grant to RMO, and by Union College summer research fel- Olberg RM (1981) Object- and self-movement detectors in the ventral
lowships to RCS and MIC. All experiments complied with the Princi- nerve cord of the dragonXy. J Comp Physiol A 141:327334
ples of animal care, publication no. 8623, revised 1985, of the Olberg RM (1986) IdentiWed target-selective visual interneurons descend-
National Institute of Health. ing from the dragonXy brain. J Comp Physiol A 159:827840
Olberg RM, Worthington AH, Venator KR (2000) Prey pursuit and
interception in dragonXies. J Comp Physiol A 186:15562
Olberg RM, Worthington AH, Fox JL, Bessette CE, Loosemore MP
References (2005) Prey size selection and distance estimation in foraging
adult dragonXies. J Comp Physiol A 191:791797
Adelman TL, Bialek W, Olberg RM (2003) The information content of Regan D, Gray R, Portfors CV, Hamstra SJ, Vincent A, Hong XH,
receptive Welds. Neuron 40:823-833-1500 Kohly R, Beverley K (1998) Catching, hitting, and collision
Baird JM, May ML (1997) Foraging behavior of Pachydiplax longi- avoidance. In: Harris LR, Jenkin M (eds) Vision and action. Cam-
pennis (Odonata: Libellulidae). J Insect Behavior 10: 655678 bridge University Press, Cambridge, pp 181214
Collett TS, Land MF (1978) How hoverXies compute interception Rowe RJ (1987) The dragonXies of New Zealand. Auckland Univ
courses. J Comp Physiol A 125: 191204 Press, New Zealand
Frye MA, Olberg RM (1995) Visual receptive Weld properties of fea- Sauseng M, Pabst M-A, Kral K (2003) The dragonXy Libellula quad-
ture detecting neurons in the dragonXy. J Comp Physiol A 177: rimaculata (Odonata: Libellulidae) makes optimal use of the dor-
569576 sal fovea of the compound eyes during perching. Eur J Entomol
Gilbert C, Strausfeld NJ (1991) The functional organization of male- 100: 475479
speciWc visual neurons in Xies. J Comp Physiol 169:395411 Sherk TE (1978) Development of the compound eyes of dragonXies
Gilbert C, Gronenberg W, Strausfeld NJ (1995) Oculomotor control in (Odonata) III. Adult compound eyes. J Exp Zool 203: 6180
Calliphorid Xies: head movements during activation and inhibi- van Hateren JH, Schilstra C (1999) BlowXy Xight and optic Xow II.
tion of neck motor neurons corroborate neuroanatomical predic- Head movements during Xight. J Exp Biol 202:149
tions. J Comp Neurol 361:285297 Wagner H (1986) Flight performance and visual control of Xight of the
Horridge GA (1978) The separation of visual axes in apposition com- free-Xying houseXy (Musca domestica L.) II. Pursuit of targets.
pound eyes. Phil Trans R Soc Lond B 285:159 Phil Trans R Soc Lond B 312:553579
123
323
324
Neuron, Vol. 17, 11451155, December, 1996, Copyright 1996 by Cell Press
325
Neuron
1146
326
Functional Organization of Zebrafish Hindbrain
1147
327
Neuron
1148
328
Functional Organization of Zebrafish Hindbrain
1149
The small increase due to a single antidromic spike cell during escapes could be explained by the calcium
was comparable to the smallest increases observed in influx associated with a single action potential. They do
the Mauthner cell during escapes elicited by a tap. The not, however, rule out the possibility that some of the
small tap-evoked increases occurred in early trials in a fluorescence responses are a result of subthreshold
series of escapes elicited at 2 min intervals. However, postsynaptic potentials. To determine whether we could
the Mauthner cell responses were often larger, ranging detect fluorescence increases during large, subthresh-
from 30% to 70% in most experiments. This was some- old input to the Mauthner cell, we varied the voltage
what surprising because the evidence, including data applied to the piezoelectric crystal driving the head tap-
from larval and adult fish, indicates that the Mauthner per and examined the fluorescence changes in the
cell in teleost fishes fires only a single action potential Mauthner cell at stimulus strengths near threshold for
each time it triggers an escape behavior (Zottoli, 1977; eliciting an escape (i.e., a strength where the Mauthner
Featherstone et al., 1991). However, our data do not al- fires only 50% of the time). Figure 4 shows the responses
low us to rule out the possibility that the larger calcium of a Mauthner cell to a series of stimuli of different
responses are a consequence of multiple action poten- strength, including several near the threshold for es-
tials. capes. The calcium response only occurred when an
Although the increase in the size of the responses in escape was produced. In cases where the fish showed
the Mauthner cell with repetitive escapes is an interest- no evidence of producing the escape, no calcium re-
ing phenomenon that may reflect plasticity in the size sponse resulted. This included trials in which the stimu-
of the calcium change evoked by a single spike or an lus strength was the same for escape and nonescape
increase in the number of spikes fired (OMalley and trials. Thus, in the absence of escapes, there is little if
Fetcho, 1996, Soc. Neurosci., abstract), we were mainly any fluorescence change in the soma, even when one
concerned with whether or not the cell was responding would expect there to be a large postsynaptic potential
during different sensory stimuli. Thus, the large size of in the cell. This strengthens the link of the somatic cal-
the responses typically seen was helpful in determining cium responses to both the behavioral response and to
whether or not the cell had been activated. There was the firing of action potentials by the Mauthner cell. It
no evidence that these larger calcium responses were also supports other evidence that the calcium increases
a consequence of damage because they recovered rap- in the soma seen with this approach are associated with
idly and completely and could be observed reliably in the firing of action potentials rather than subthreshold
conjunction with escapes for up to 132 trials spread postsynaptic potentials (ODonovan et al., 1993; McClel-
over as long as 2 days. Thus, the cells remained healthy lan et al., 1994; Fetcho and OMalley, 1995).
and could be monitored physiologically for an extended
time, as long as the illumination was minimized. Activation Pattern of Segmental Homologs
After using the well-studied Mauthner cell to establish
Calcium Responses at Threshold the reliability of the technique as an indicator of neuronal
The experiments with antidromic stimulation indicate firing, we went on to study the responses of MiD2cm
that the smallest fluorescence increases in the Mauthner and MiD3cm, for which there were no prior physiological
329
Neuron
1150
Figure 5. Differential Activation of the Mauthner Cell and Its Homologs during Escapes
Each row shows a trial consisting of a sequence of images during which an escape was elicited by an ipsilateral touch to the head or tail.
The color scale (applies to all figures) represents fluorescence intensity (blue, lowest; red, highest).
(A) Response of MiD3cm during an escape elicited by a head tap. MiD3cm is shown at relatively high magnification to illustrate clearly the
fluorescence increase. The tap was applied during the third frame. The escape event transiently causes the cell to move out of the frame,
which provides a convenient record of the behavioral event. This movement frame is omitted from subsequent rows.
(B) Simultaneous imaging of both the Mauthner cell (left) and MiD3cm (far right) during a head stimulus. Both cells respond.
(C) Same field as in (B), but with a tail stimulus. The Mauthner cell responds but not MiD3cm.
(D) Simultaneous imaging of the Mauthner cell (left) and MiD2cm (right) during a head stimulus. Both cells respond.
(E) Same field as in (D), but with a tail stimulus. Only the Mauthner cell responds. Arrows mark the first frame after the escape in (B)(E).
Images were acquired at 400 ms intervals.
Scale bars: 10 mm in (A), 15 mm in (B)(E). The size of the fluorescence increases in each trial were as follows: 41% (A); 62% in Mauthner
cell, 21% in MiD3cm (B); 53% in Mauthner cell, 2% in MiD3cm (not significant, see Figure 2) (C); 21% in Mauthner cell, 22% in MiD2cm (D);
67% in Mauthner cell, none in MiD2cm (E).
data. We compared the responses of the Mauthner cell outlasted the movement artifact; they returned nearly
and its homologs during escapes elicited by one of the to baseline in 46 s, which is typical for calcium signals
two tappers positioned to stimulate either the head or in neuronal somata (Regehr et al., 1989; Lev-Ram et al.,
the tail. The Mauthner cell and its serial homologs 1992; Yuste et al., 1994). In marked contrast with the
(MiD2cm and MiD3cm) all responded to a touch on the response of this group during head taps, only the
ipsilateral side of the head at a strength above threshold Mauthner cell responded to ipsilateral tail stimuli even
for eliciting an escape (Figures 5A, 5B, and 5D). These when the stimuli were well above the threshold for es-
responses consisted of fluorescence increases ranging capes (Figures 5C and 5E).
from 9% to 41% in the homologs and 21% to 110% If the homologs are indeed playing a role in determin-
in the Mauthner cells. The fluorescence changes far ing the magnitude of the escape response, then their
330
Functional Organization of Zebrafish Hindbrain
1151
trial-to-trial behavior should consistently show a differ- series of trials, within the limits of the detection system
ential response to head and tail stimulation. Addressing (Figure 7B).
this issue requires not only that the in vivo recordings If MiD2cm and MiD3cm contribute to the escape along
be relatively stable, but that successive trials be directly with the Mauthner cell, then their threshold should be
observed without signal averaging. High efficiency op- similar to that of the Mauthner cell, which is known to
tics and the large fluorescence signals of calcium indica- fire an action potential only in conjunction with escapes
tors were important for this because they allow the use (Zottoli, 1977). The relative thresholds for activating
of minimal excitation light, providing stable recording these cells were determined by varying the strength of
conditions. This allowed the recording of neuronal re- the stimulus applied to the head or tail. When the head
sponses over a series of successive trials in which head stimulus was near or at threshold (50% probability of
and tail stimuli were alternately applied to elicit escapes response) for an escape, the Mauthner cell and its homo-
(Figure 6). Throughout these trials, head stimulation acti- logs responded together whenever an escape occurred
vated all of the cells (Mauthner cell, MiD2cm, and (e.g., Figure 7). The Mauthner cell showed a very clear
MiD3cm), whereas tail stimulation activated only the all or none response with large fluorescence increases
Mauthner cell; such differences were observed in 17 during escapes and none in response to threshold stim-
fish, with no exceptions to the differential response of uli that did not lead to an escape. In some animals, as
MiD2cm and MiD3cm to head and tail stimuli that elicited in Figure 7, the homologs of the Mauthner cell also had
escapes. a clear threshold, responding only in conjunction with
escapes. In other fish, the homologs showed a weak
Timing and Threshold of Homolog Responses response to head stimuli that did not elicit an escape,
These results suggest that the homologs of the Mauth- but even in these cases, they produced a larger increase
ner cell are involved in the production of the more vigor- in fluorescence in conjunction with escape events.
ous escapes produced by stimulation of the head. Be- MiD2cm and MiD3cm did not respond to tail stimuli at
cause the escape behavior is a very fast event (the initial any strength, even when those stimuli were three times
C-bend is completed 2025 ms after the stimulus onset; threshold for eliciting an escape.
Eaton and Farley, 1975), the segmental homologs must
be activated close in time to the Mauthner cell if they Discussion
are to contribute to the initial escape bend. To examine
the time course of activation, the laser scanning micro- A role for the Mauthner cell homologs in escapes would
scope was used in a one-dimensional imaging mode, help to explain both the normal adaptive variability of
i.e., a single line across these cells was repetitively escapes as well as the production of escapes in fish
scanned. This provided a much faster acquisition rate (2 without Mauthner cells, thus unifying a series of obser-
ms per line) than the 400 ms needed for two-dimensional vations made over the past 15 years. Until now, it had
images. These experiments showed that MiD3cm and not been possible to obtain functional data from the
the Mauthner cell are activated within 30 ms of one identified Mauthner cell homologs (Metcalfe et al., 1986;
another (Figure 7A). While the movement artifact pre- Foreman and Eaton, 1993). This is a consequence of
vented a more exact determination of the response la- the smaller size of MiD2cm and MiD3cm, which makes
tency, both cells were synchronously activated over a it difficult both to find them and to record their electrical
331
Neuron
1152
Figure 7. Responses of the Mauthner Cell and MiD3cm near the Threshold for Escapes
(A) The fluorescence of a line (black line in top panel) that passed through both the Mauthner cell (left) and MiD3cm (right) was monitored
repeatedly as shown in the bottom panel. Consecutive lines were acquired at 2 ms intervals (vertical scale bar, 200 ms) and are displayed
from top to bottom. An escape elicited by a tap to the ipsilateral side of the head is indicated by the arrows. The trace was briefly disrupted
by the escape movement, but following the movement, the fluorescence of the vertical bands representing the two cells had increased,
indicating that both cells had responded rapidly, within 30 ms of one another.
(B) Quantification of responses from the cells in (A) over several trials in which the head was tapped at a strength near threshold for escape
(escapes occurred in 67% of the trials). The points show the average intensity of successive sets of 15 lines (i.e. 30 ms per data point) plotted
over time. Successive trials, 2 min apart are separated by dashed lines. The starting baseline in each trial was normalized to 100 as in Figure
3. Repeated stimuli at a constant stimulus strength near threshold led to escapes in trials 2, 4, 5 and 6, but not in 1 and 3. When an escape
occurred, both the Mauthner cell and MiD3cm responded, but neither responded when there was no escape.
Horizontal scale bar, 20 mm. Size of increases in (A): 70% in Mauthner cell; 34% in MiD3cm.
activity. Our data indicate that the Mauthner cell and its (Figure 3) or after direct firing of single action potentials
serial homologs (MiD2cm and MiD3cm) in the hindbrain in cultured bullfrog neurons (OMalley, 1994). In addition,
do indeed function together during escapes. They re- we are aware of no imaging studies reporting detectable
spond together during the large escape turns produced somatic calcium signals due to subthreshold EPSPs.
by lateral stimulation on the head, but in response to All of these observations support the conclusion that
stimulation of the tail, only the Mauthner cell is activated. MiD2cm and MiD3cm are firing in response to head
This is exactly the pattern of activation predicted by stimuli that elicit escapes.
behavioral studies. It suggests a population code, with The differential activation we observed among the
the extent of the bend related to the extent of activation Mauthner cell, MiD2cm, and MiD3cm most likely arises
of a population of serially homologous hindbrain cells from differences in the sensory input to this set of hind-
(Eaton et al., 1991). We examined only the extremes brain neurons. Based upon previous work, the most
of the escape behavior by using stimuli that elicit the likely possibility is that rostral stimulation leads to acti-
weakest and strongest escapes. Intermediate forms of vation of the auditory system, whereas caudal stimula-
the escape might be produced not only by variations in tion activates cutaneous and/or lateral line inputs (Eaton
which cells are activated, but also by changes in the et al., 1984). All of these are known to excite the Mauth-
degree of activation of individual cells. ner cell in goldfish, consistent with its responses to both
The large size of the somatic calcium signals seen head and tail stimulation (Faber et al., 1989). Previously,
during sensory activation of MiD2cm and MiD3cm sug- there was no information about the inputs to MiD2cm
gests that these cells are firing one or more action poten- and MiD3cm. Our work shows that they respond only
tials. These responses are unlikely to result from sub- to head stimulation, suggesting that they receive little
threshold excitation. In the case of the Mauthner cell, or no input from the caudal cutaneous or lateral line
there is usually no detectable calcium increase from system. The similar thresholds and conjoint activation
stimuli that are just below threshold (Figures 4 and 7), of MiD2cm and MiD3cm with the Mauthner cell in re-
even though such stimuli lead to a large synaptic input sponse to head stimuli are consistent with their sharing
(Faber et al., 1991). This is not unexpected, as action common rostral sensory inputs, most likely from the
potentials should produce considerably larger calcium auditory system. Such differences in sensory input to
signals than subthreshold inputs because action poten- escape neurons is not unprecedented, as they are well
tials will open high threshold calcium channels and documented in crayfish. The crayfish system has obvi-
NMDA-gated channels (if the NMDA receptors and glu- ous parallels with zebrafish in that rostral and caudal
tamate are present). The somatic calcium signals ob- stimuli also elicit different forms of escape by differential
served in MiD2cm and MiD3cm after sensory stimulation activation of giant neurons (Wine and Krasne, 1972).
(Figures 6 and 7) are typically larger than those produced Our observations that MiD2cm and MiD3cm are co-
either by single antidromic spikes in the Mauthner cell activated with the Mauthner cell during escapes and
332
Functional Organization of Zebrafish Hindbrain
1153
that they share a similar axonal trajectory are consistent caudal (postanal) spinal cord of posthatching larval zebrafish (Danio
with their playing a role in the generation of escapes. rerio, usually within the first 2 weeks after hatching) that were anes-
thetized with 0.02% 3-aminobenzoic acid ethyl ester (Fetcho and
Although this evidence confirms previous predictions,
OMalley, 1995). In most cases, this injection was into ventral cord
it is still correlative, and nothing is known about the to selectively label the Mauthner cell and its homologs without dis-
output connections of MiD2cm and MiD3cm. A more rupting more dorsal sensory pathways. After injection, the fish were
causal relationship could be established by lesioning allowed to recover and were maintained in 10% Hanks solution. We
the cells in the live fish and observing the effects on only studied fish that exhibited no obvious disruptions of swimming
performance. Our data lead to the prediction that le- or escape behaviors following the injection. Twelve or more hours
sioning MiD2cm and MiD3cm should reduce the perfor- later, the fish were briefly anesthetized, embedded on their backs
in soft agar on a cover glass in a petri dish (Eaton et al., 1984) and
mance of escapes elicited by head taps, but not those
then rinsed with 10% Hanks solution to allow recovery from the
from tail taps. The expectation is that such lesions anesthetic. Confocal images were obtained by looking into the head
should make the C-bend to a head stimulus more like of the intact fish using a Zeiss IM35 inverted microscope with a
that to a tail stimulus. We have preliminary data that 503 Leitz 1.0 NA objective and a Biorad MRC 600 laser-scanning
neurons in intact zebrafish can be deleted by using opti- confocal imaging system (Hernandez-Cruz et al., 1990; OMalley,
cal methods, so it should eventually be possible to eval- 1994; Fetcho and OMalley, 1995). The transparency of the fish al-
lowed us to not only see neurons inside the living animal but also
uate this prediction.
allowed easy monitoring of the viability of the fish by observing the
The escape system is particularly attractive for studies heart beat and blood flow. To confirm the identity of the cells studied
of how activity in neuronal populations determines ver- physiologically, stacks of images showing the morphology in suc-
tebrate behaviors because the behavior is produced by cessive confocal sections were acquired. Signal averaging, usually
a population of neurons, but the number of neurons and seven frames, was used when acquiring this morphological data.
synapses in the circuit is manageably small (Faber et Maximum projections were made from stacks of these sections. The
al., 1989). The ability to image the activity of any of image stacks were also reconstructed in three-dimensions using the
VolVis program (Sobierajski et al., 1995), allowing us to examine the
the array of hindbrain neurons and to delete them and
details of the dendritic morphology and axonal projections of each
monitor behavioral changes offers the prospect of a cell.
more complete understanding of how a population of Escapes were elicited by an abrupt touch produced by the dis-
neurons in a vertebrate determines the form of a behav- placement of a small glass probe attached to a piezoelectric crystal.
ior. These approaches should prove useful for future The voltage applied to the crystal could be varied to change the
studies of neural circuits not only in the normal animal, excursion of the glass probe. Two probes were used: one to stimu-
but also in the many behavioral mutants that have been late the head and the other the tail. All stimuli were ipsilateral to
the reticulospinal neurons, in the region of the otolith for the head
generated by large scale mutagenesis of zebrafish (Mul-
stimulus and postanally for the tail stimulus. The tail stimulus was
lins and Nusslein-Volhard, 1993; Driever et al., 1995). located rostral to the calcium green injection site so as to activate
Our work has several implications concerning the sensory neurons whose pathways to the brain were presumably not
functional organization of hindbrain segments. Hind- damaged by the injection. The responses of the Mauthner cell to
brain segments are thought to have arisen from the tail stimuli indicated that ascending sensory pathways were indeed
duplication of an ancestral segment, with subsequent intact. The stimuli produced an escape movement during which the
evolutionary divergence of the segments (Metcalfe et cell(s) moved briefly out of the plane of section, but then returned
rapidly because the agar controlled the resting position of the fish.
al., 1986). This would lead to the similar structural organi- The escape attempt could also be observed visually (with protective
zation in successive segments observed in vertebrates. goggles to block the laser light) through a dissecting scope mounted
Our observations indicate that another consequence of above the inverted microscope. Free swimming larval zebrafish gen-
this duplication is the production of a series of function- erate a variety of movements subsequent to the escape (Eaton and
ally related neurons in successive segments. The serially Farley, 1975). Although the initial escape event is very obvious in
homologous neurons studied retain a similar functional the agar, we could not distinguish subtle differences in movement
following escapes. However, the responses we observed were
role in escapes, but they are not complete functional
clearly linked to the initial escape movement and not subsequent
clones of one another. Instead, there is variability in the behavior, because they occurred whether or not the fish continued
sensory inputs that drive the cells and this is associated to move after the initial escape event.
with variability in the resulting escape behavior. The The fluorescence intensity of neurons during escapes was moni-
duplication of segments may have permitted an expan- tored either by collecting a sequence of images of a cell or group
sion of the behavioral repertoire of the animal by produc- of cells (usually at 400 ms intervals) or by repeatedly scanning a
ing segmental groups of neurons involved in particular single line through the cells at 2 ms intervals. To assure that an
increase in the brightness of the cell was not the result of movement
behaviors. A subsequent evolutionary divergence of the
to a brighter plane, we collected a series of optical sections spanning
inputs and outputs of these cells would allow for behav- the cell(s) (this was done for each block of trials). We then focused
ioral diversification, much as gene duplication and diver- at the brightest focal plane immediately prior to each trial. In experi-
gence has led to diversification at the molecular level. ments with multiple cells in the field, we picked a focal plane where
This organization of the hindbrain into serial sets of the cells of interest were at or near their maximal fluorescence. A
functionally related neurons is likely to be a very general cell was deemed to have responded only if its fluorescence response
one because the hindbrain has changed relatively little exceeded its highest resting value at any plane in the reference
image stack. For physiological experiments, the confocal aperture
during the evolution of vertebrates, as illustrated by the
was usually set fully open to maximize light collection. This yielded
very similar organization of escape circuits in fish and an optical section of 45 mm. These dynamic calcium images were
startle circuits in mammals (Cruce and Newman, 1984; acquired without signal averaging or other processing. For illustra-
Lingenhohl and Friauf, 1994; Butler and Hodos, 1996). tion purposes, the images are linearly contrast enhanced and
smoothed with a 3 3 3 or 5 3 5 low pass filter to reduce pixel to
Experimental Procedures pixel noise. All quantitation, however, was performed on the raw
data. The percent fluorescence increases associated with each set
A 50% solution of calcium green dextran (10,000 MW) in 10% Hanks of images are given in the figure captions.
solution was pressure injected via a glass microelectrode into the The fluorescence responses can be converted to absolute levels
333
Neuron
1154
of free calcium by knowing the resting calcium level, the K D of the Faber, D.S., Korn, H., and Lin, J.-W. (1991). Role of medullary net-
indicator, and the dynamic range. The latter two numbers were works and postsynaptic membrane properties in regulating
unknown for calcium green dextran in living neurons. We determined Mauthner cell responsiveness to sensory excitation. Brain Behav.
them by using cultured bullfrog sympathetic neurons whose acces- Evol. 37, 286297.
sibility and large size relative to zebrafish neurons allowed the stable Featherstone, D., Drewes, C.D., and Coats, J.R. (1991). Noninvasive
whole cell patch recordings needed for such determinations. The detection of electrical events during the startle response in larval
cultured cells were patch clamped and filled with free calcium set Medaka. J. Exp. Biol. 158, 583589.
at fixed levels, using 10 mM BAPTA. The cells were then flooded
Fetcho, J.R., and Faber, D.S. (1988). Identification of motoneurons
with calcium via repetitive voltage pulses under voltage clamp and
and interneurons in the spinal network for escapes initiated by the
the maximal fluorescence increase that could be obtained at the
Mauthner cell in goldfish. J. Neurosci. 8, 41924213.
differing resting calcium levels determined. This allowed the intra-
cellular K D of calcium green dextran to be determined. The average Fetcho, J.R., and OMalley, D.M. (1995). Visualization of active neural
value obtained, 250 nM, was quite close to the in vitro value (252 circuitry in the spinal cord of intact zebrafish. J. Neurophysiol. 73,
nM; Molecular Probes, Eugene, Oregon). The maximum dynamic 399406.
range obtainable (upon raising calcium from subnanomolar levels Foreman, M.B., and Eaton, R.C. (1993). The direction change con-
to indicator-saturating levels) was only 8.7-fold, considerably less cept for reticulospinal control of goldfish escape. J. Neurosci. 13,
than the range obtainable in intracellular solution in cuvettes (15- 41014113.
fold). Based on these numbers, and by assuming a resting calcium Fraser, S., Keynes, R., and Lumsden, A. (1990). Segmentation in the
level (e.g., 100 nM), the fluorescence responses may be equated to chick embryo hindbrain is defined by cell lineage restrictions. Nature
changes in free calcium. Since the actual resting calcium levels are 344, 431435.
not known, the calcium responses in the paper are expressed in
Guthrie, S. (1995). The status of the neural segment. Trends Neu-
terms of relative fluorescence increases. An estimate of the sizes
rosci. 18, 7479.
of the increases based upon the calibration data is provided in the
Results. Hanneman, E., Trevarrow, B., Metcalfe, W.K., Kimmel, C.B., and
Westerfield, M. (1988). Segmental pattern of development of the
hindbrain and spinal cord of the zebrafish embryo. Development
Acknowledgments
103, 4958.
Correspondence should be addressed to J. R. F. We thank B. J. Hernandez-Cruz, A., Sala, F., and Adams, P.R. (1990). Subcellular
Burbach for technical assistance and P. R. Adams for advice during calcium transients visualized by confocal microscopy in a voltage
the course of the study. Supported by National Institute of Neurolog- clamped vertebrate neuron. Science 247, 858862.
ical Disorders and Stroke grants NS26539 (to J. R. F.) and NS09113 Kimmel, C.B., Powell, S.L., and Metcalfe, W.K. (1982). Brain neurons
(to D. M. O.), The Sloan Foundation (to J. R. F.), and the Howard which project to the spinal cord in young larvae of the zebrafish. J.
Hughes Medical Institute. Comp. Neurol. 205, 112127.
The costs of publication of this article were defrayed in part by Lee, R.K.K., and Eaton, R.C. (1991). Identifiable reticulospinal neu-
the payment of page charges. This article must therefore be hereby rons of the adult zebrafish, Brachydanio rerio. J. Comp. Neurol. 304,
marked advertisement in accordance with 18 USC Section 1734 3452.
solely to indicate this fact.
Lev-Ram, V., Miyakawa, H., Lasser-Ross, N., and Ross, W.N. (1992).
Calcium transients in cerebellar Purkinje neurons evoked by intra-
Received October 1, 1996; revised October 30, 1996.
cellular stimulation. J. Neurophysiol. 68, 11671177.
References Lingenhohl, K., and Friauf, E. (1994). Giant neurons in the rat reticular
formation: a sensorimotor interface in the elementary acoustic star-
Butler, A.B., and Hodos, W. (1996). Comparative Vertebrate Neuro- tle circuit. J. Neurosci. 14, 11761194.
anatomy: Evolution and Adaptation (New York: Wiley-Liss). Lumsden, A., and Keynes, R. (1989). Segmental patterns of neuronal
Clarke, J.D.W., and Lumsden, A. (1993). Segmental repetition of development in the chick hindbrain. Nature 337, 424428.
neuronal phenotype sets in the chick embryo hindbrain. Develop- McClellan, A.D., McPherson, D., and ODonovan, M.J. (1994). Com-
ment 118, 151162. bined retrograde labeling and calcium imaging in spinal cord and
Cruce, W.L.R., and Newman, D.B. (1984). Evolution of motor sys- brainstem neurons of the lamprey. Brain Res. 663, 6168.
tems: the reticulospinal pathways. Am. Zool. 24, 733753. Metcalfe, W.K., Mendelson, B., and Kimmel, C.B. (1986). Segmental
Driever, W., Stemple, D., Schier, A., and Solnica-Krezel, L. (1995). homologies among reticulospinal neurons in the hindbrain of the
Zebrafish: genetic tools for studying vertebrate development. zebrafish larva. J. Comp. Neurol. 251, 147159.
Trends Genet. 10, 152159. Morton, D.W., and Chiel, H.J. (1994). Neural architectures for adap-
Eaton, R.C., and Emberley, D.S. (1991). How stimulus direction de- tive behavior. Trends Neurosci. 17, 413420.
termines the trajectory of the Mauthner initiated escape response. Mullins, M.C., and Nusslein-Volhard, C. (1993). Mutational ap-
J. Exp. Biol. 161, 469487. proaches to studying embryonic pattern formation in the zebrafish.
Eaton, R.C., and Farley, R.D. (1975). Mauthner neuron field potential Curr. Opin. Genet. Dev. 3, 648654.
in newly hatched larvae of the zebra fish. J. Neurophysiol. 38, Nissanov, J., Eaton, R.C., and DiDomenico, R. (1990). The motor
502512. output of the Mauthner cell, a reticulospinal command neuron. Brain
Eaton, R.C., Lavender, W.A., and Wieland, C.M. (1982). Alternative Res. 517, 8898.
neural pathways initiate fast start responses following lesions of the ODonovan, M.J., Ho, S., Sholomenko, G., and Yee, W. (1993). Real-
Mauthner neuron in goldfish. J. Comp. Physiol. 145, 485496. time imaging of neurons retrogradely and anterogradely labelled
Eaton, R.C., Nissanov, J., and Wieland, C.M. (1984). Differential acti- with calcium sensitive dyes. J. Neurosci. Meth. 46, 91106.
vation of Mauthner and non-Mauthner startle circuits in the zebra- OMalley, D.M. (1994). Calcium permeability of the neuronal nuclear
fish: implications for functional substitution. J. Comp. Physiol. 155, envelope: evaluation using confocal volumes and intracellular perfu-
813820. sion. J. Neurosci. 14, 57415758.
Eaton, R.C., DiDomenico, R., and Nissanov, J. (1991). Role of the Regehr, W.G., Connor, J.A., and Tank, D.W. (1989). Optical imaging
Mauthner cell in sensorimotor integration by the brain stem escape of calcium accumulation in hippocampal pyramidal cells during syn-
network. Brain Behav. Evol. 37, 272285. aptic activation. Nature 341, 533536.
Faber, D.S., Fetcho, J.R., and Korn, H. (1989). Neuronal networks Rubenstein, J.L., Martinez, S., Shimamura, K., and Puelles, L. (1994).
underlying the escape response in goldfish: general implications for The embryonic vertebrate forebrain: the prosomeric model. Science
motor control. Ann. NY Acad. Sci. 563, 1133. 266, 578580.
334
Functional Organization of Zebrafish Hindbrain
1155
Sobierajski, L.M., Avila, R.S., OMalley, D.M., Wang, S., and Kauf-
mann, A.E. (1995). Visualization of calcium activity in nerve cells.
IEEE Comp. Graphics Appl. 15, 5561.
Trevarrow, B., Marks, D.L., and Kimmel, C.B. (1990). Organization
of hindbrain segments in the zebrafish embryo. Neuron 4, 669679.
Wine, J.J., and Krasne, F.B. (1972). The organization of escape
behavior in the crayfish. J. Exp. Biol. 56, 118.
Yuste, R., Gutnick, M.J., Saar, D., Delaney, K.R., and Tank, D.W.
(1994). Ca2 1 accumulations in dendrites of neocortical pyramidal
neurons: an apical band and evidence for two functional compart-
ments. Neuron 13, 2343.
Zottoli, S.J. (1977). Correlation of the startle reflex and Mauthner
cell auditory responses in unrestrained goldfish. J. Exp. Biol. 66,
243254.
335
336
Neuron, Vol. 23, 325335, June, 1999, Copyright 1999 by Cell Press
Katharine S. Liu and Joseph R. Fetcho* extremely fast escape response, also known as the fast-
Department of Neurobiology and Behavior start response, is essential to the survival of the animal.
State University of New York at Stony Brook As reproducible as the behavior is, it is also modifiable.
Stony Brook, New York 11794 Responses are directional, with the magnitude of the
turn depending upon the location of the stimulus (Eaton
et al., 1984; Eaton and Emberley, 1991). Thus, the cir-
cuitry that mediates successful escape responses pro-
Summary vides not only short latency and high speed but also
proper timing and direction of movement.
Segmentation of the vertebrate brain is most obvious The Mauthner cells are thought to be involved in the
in the hindbrain, where successive segments con- initiation and laterality of the escape response. There is
tain repeated neuronal types. One such set of three a one-to-one correspondence between the Mauthner
repeated reticulospinal neuronsthe Mauthner cell, cell firing and subsequent expression of the fast-start
MiD2cm, and MiD3cmis thought to produce different response (Zottoli, 1977; Eaton et al., 1981), supporting
forms of the escape response that fish use to avoid the cells role in the initiation of escape. The laterality
predators. We used laser ablations in larval zebrafish of the initial turn of the escape response is thought to
to test the hypothesis that these segmental hindbrain be determined by which of the two Mauthner cells is
cells form a functional group. Killing all three cells activated (Eaton and Kimmel, 1980). A large sensory
eliminated short-latency, high-performance escape input on one side of the fish leads to activation of the
responses to both head- and tail-directed stimuli. Kill- Mauthner cell on that side. Because the Mauthner axon
ing just the Mauthner cell affected escapes from tail- crosses the midline to excite axial motoneurons on the
directed but not from head-directed stimuli. These opposite side, activation of the Mauthner cell causes a
results reveal the contributions of one set of reticulo- turn away from the stimulus.
spinal neurons to behavior and support the idea that The existence of a bilateral pair of Mauthner cells
serially repeated hindbrain neurons form functional might explain how lateral directionality of escape is me-
groups. diated but is not sufficient to account for the varied
directionality of observed turns. Direct electrical stimu-
lation of the Mauthner cell in unrestrained goldfish re-
Introduction sults in weaker and less variable responses than sensory-
evoked responses (Nissanov et al., 1990), suggesting
A striking feature of the vertebrate brain is its division that the Mauthner cell cannot, by itself, produce all forms
into segments that are recognizable at all levels of orga- of escape. Lesion experiments also support the involve-
nization, from the gross anatomical to the molecular. ment of other cells in escapes. In goldfish, electrolytic
This segmentation is most evident in the hindbrain, removal of the Mauthner cell yielded no effect on the
which contains reticulospinal neurons whose axons behavior, other than a small increase in latency to re-
project into spinal cord to influence spinal circuits in- sponse (Eaton et al., 1982). These observations led re-
volved in motor control. Successive hindbrain segments searchers to postulate the existence of parallel pathways,
contain morphologically similar reticulospinal neurons acting along with the Mauthner cell during escapes (Ea-
that form serially repeated arrays of cells. Even though ton and DiDomenico, 1985; Metcalfe et al., 1986; Fore-
serially homologous neurons are present in the hind- man and Eaton, 1993). The obvious candidates for these
brain of diverse vertebrates, little is known about their pathways were the Mauthner-like segmental homologs
contributions to behavior. MiD2cm and MiD3cm. Foreman and Eaton (1993) sug-
In larval zebrafish, where the small number of neurons gested that differential activity of the cells in the
allows the identification of individual cells, one can rec- Mauthner array (Mauthner, MiD2cm, and MiD3cm) con-
ognize single, morphologically similar cells in succes- trols the magnitude of the escape. Activity in all three
sive hindbrain segments (Kimmel et al., 1982; Metcalfe cells would excite more inter- and motoneurons, leading
et al., 1986). One such set includes three bilateral pairs to a larger turn, whereas activity in only one or two cells
of neurons in adjoining segmentsthe Mauthner cell in would excite fewer downstream neurons, leading to a
hindbrain segment 4 and two serial homologs, MiD2cm smaller turn.
and MiD3cm, located in successive segments caudal The proposal of Foreman and Eaton (1993) was sup-
to the Mauthner cell. The neurons in this set are thought ported by our previous functional imaging studies in
to generate an escape behavior used by fish to avoid zebrafish, which showed that sensory stimuli known to
predators. When presented with a threatening stimulus, elicit different forms of escape behavior also produce
a fish responds with a characteristic fast turn away from different patterns of activation in the Mauthner array
the stimulus that allows the animal to swim off in the (OMalley et al., 1996). These studies took advantage of
direction opposite the threat. This highly reproducible, the transparency of zebrafish larvae and used a calcium
indicator to image the activity of the Mauthner cell,
MiD2cm, and MiD3cm during escape responses. As pre-
* To whom correspondence should be addressed (e-mail: jfetcho@ dicted in the Foreman and Eaton model, all three homo-
neurobio.sunysb.edu). logs were active during escapes elicited by stimuli to
337
Neuron
326
the head, while only the Mauthner cell was active during calcium imaging of neurons in zebrafish (Fetcho et al.,
escapes elicited by stimuli to the tail. These correlations 1998) and set out to determine if this phototoxicity could
between neural activity and behavior suggested that be used as a reliable method of killing single cells in
cells in the Mauthner array might generate different zebrafish larvae. Targeted cells were labeled by retro-
forms of escape movements but offered no causal links grade uptake after injection of a fluorescent indicator,
between the neurons and the behavior. calcium green dextran (CGD), and identified by position
The transparency of zebrafish larvae offered us the and morphology (Kimmel et al., 1982; Metcalfe et al.,
opportunity to conclusively link neurons and behavior 1986) on a confocal microscope. The laser was then
by examining the consequences of laser ablation of neu- focused and maintained at highest intensity on the cen-
rons in the Mauthner array. Toward this end, we devel- ter of a labeled cell. This led to an immediately noticeable
oped a noninvasive approach for killing single cells that decrease in fluorescence of the cell that was likely due
takes advantage of the relative phototoxicity of the dyes to bleaching of the indicator. A day later, the fluorescent
we use for calcium imaging. The targets we chose for cell body was no longer evident in confocal images,
ablation were the Mauthner cell alone and the entire suggesting cell death.
array, as these sets represent the extremes of activity In the current work, a major concern was to identify an
patterns observed in the imaging studies. Based upon accessible and reliable indicator that a cell was indeed
the functional imaging work, our predictions were that successfully killed by this illumination. Because larval
killing all three cells in the Mauthner array would remove zebrafish are transparent, we can view the effects of
the ability of the fish to produce high-performance es- the photoablation directly in confocal optical sections.
cape turns to both head and tail stimuli, whereas ablat- The most dramatic examples of the effectiveness of
ing just the Mauthner cell would have a greater effect on the approach were seen in lesions of the Mauthner cell
turns in response to tail-directed than to head-directed (Figures 1A and 1B) because its large axon was easily
stimuli. The results from killing all three cells in the array identifiable, even without the cell body. In these cases,
matched our predictions. The Mauthner lesions, while optical sections 24 hr after ablations revealed a swollen,
partially supporting our predictions, produced unex- truncated axon stump of the Mauthner cell leading to-
pected results that allow us to resolve paradoxical ob- ward the original position of the soma (Figure 1B). Sur-
servations from previous studies. Our results provide vival of the Mauthner axon after somatic lesions has
direct support for the idea that repeated hindbrain neu- been reported before (Eaton et al., 1982; DiDomenico et
rons form functionally related groups. al., 1988). We found that a swollen, abruptly terminating
Mauthner axon stump was always associated with the
Results loss of a fluorescent soma. In about 25% of the ablation
attempts, the Mauthner cell body was fluorescent after
Specificity and Effectiveness of Lesions 24 hr, with an intensity that was higher than that immedi-
The indicator dyes typically used for visualizing cell ac- ately after the attempted ablation. In these cases, the
tivity have adverse affects on the cells when exposed to axons of the cells always looked normal and the abla-
intense light. We observed such effects in our previous tions were obviously unsuccessful. We suspect that the
338
Behavioral Roles of Hindbrain Neurons in Zebrafish
327
fluorescence intensity recovered because dye that was the latter. Cells were targeted with the confocal laser
bleached out of the soma during the attempt to ablate until Texas red fluorescence was reduced to back-
the cell was replaced by dye from the intact axon. The ground levels. While these cells were still visible with
one-to-one correlation between swollen axon stumps CGD shortly following the exposure, they were no longer
and loss of fluorescence in the soma supports the con- visible with either dye after 24 hr and beyond. Cells not
clusion that the loss of fluorescence (after 24 hr) is due subjected to intense illumination remain fluorescent for
to somatic death. weeks. These results suggest that the absence of fluo-
This conclusion is also supported by experiments in rescence is not due simply to photobleaching but to cell
which we viewed cells under differential interference death. Thus, morphological evidence of degeneration
contrast (DIC) optics shortly after exposure to the laser. in fluorescence and DIC as well as double-labeling ex-
Targeted cells showed clear indicators of necrosisthe periments all indicate that the loss of fluorescence after
formation of large vacuoles and a more pronounced 24 hr is a reliable indicator of cell death.
outline as the cell membrane begins to pull away from The amount of illumination time required to kill a cell
its neighbors. These effects were always associated varied with cell size and location. The relatively small
with a loss of fluorescence the next day, again support- and exposed motoneurons in the spinal cord (Figures
ing the conclusion that this loss is a reliable indicator 1C and 1D) required less than 1 min, while the much
of cell death. larger cells of the Mauthner array deep in the hindbrain
To demonstrate further that the loss of fluorescence (Figures 1A and 1B) required 1012 min. In the spinal
was not simply due to photobleaching, we simultane- cord, the success rate was close to 100%. In the hind-
ously labeled spinal neurons with Texas red (lex 5 568) brain, where the cells are much deeper and occasionally
and CGD (lex 5 488) so that cells could be lesioned by occluded by pigment, the success rate dropped to about
excitation of the former and their status monitored with 75%. It is possible that some targeted neurons that were
339
Neuron
328
still fluorescent after 24 hr were damaged, but we did to ablation procedures previously used for behavioral
not use these animals in our studies. studies in vertebrate systems.
Laser illuminating unlabeled fish for the same time
periods and in the same regions of the brain as labeled Behavioral Experiments
fish did not produce the behavioral effects observed In response to a pulse of water on one side, fish perform
after illuminating labeled cells. This indicates that the a large, fast turn away from the stimulus, followed by
lesions were not a direct effect of the laser light, but a counterbend and further high-speed swimming. To
depended on interactions of the light with the dye in the expedite analysis of this escape response, we wrote a
labeled cells. program in Labview that automatically tracked the fish
These photoablations are highly specific. This is best and generated plots of several kinematic parameters for
demonstrated in clusters of spinal motoneurons. A sin- each trial. We focused our analysis on the performance
gle cell can be targeted in the midst of near neighbors of the initial turn in the escape because previous data
(Figure 1C). Neighboring cells retain their fluorescence suggested that the Mauthner array is important for gen-
both immediately and 1 day after lesioning (Figure 1D). erating the high-performance movements during this
In addition, in calcium imaging studies similar to OMal- turn. Several features of the initial turn were studied: the
ley et al. (1996), we found that neurons that respond latency to its initiation (time from the contact of the pulse
during escapes retain that ability after lesioning of adja- of water to the beginning of movement), the maximum
cent cells (data not shown). Thus, this technique is effec- angle of the turn, its peak angular velocity, and its dura-
tive, minimally invasive, and highly specific compared tion. Two classes of responsesnonescape turns that
340
Behavioral Roles of Hindbrain Neurons in Zebrafish
329
341
Neuron
330
342
Behavioral Roles of Hindbrain Neurons in Zebrafish
331
was 23.28/ms; that to tail-directed stimuli was 21.8 ms. ms) and tail-elicited responses (13.5 ms), with only small
In the array-lesioned group, there was an asymmetry changes on the intact side. This produced a significant
between the two sides in response to head stimuli prior asymmetry between the two sides after the lesion (Fig-
to lesions (likely due to the labeling injection), but the ure 6, duration; Table 1). Photoablation of the Mauthner
other prelesion results showed good symmetry between cell alone resulted in a smaller but significant (p 5
sides before the lesion. Killing all three homologs led 0.0048) increase in turn duration for tail-elicited re-
to a large, significant decrease in angular velocity (to sponses but had no effect on head-elicited responses.
138148/ms) on the lesioned side for both head- and Again, the pattern of these changes was similar to those
tail-elicited responses. There were smaller decreases seen for latency and angular velocity.
in performance on the intact side after lesion that we
attribute to nonspecific effects. However, the large Discussion
change on the lesioned side led to the introduction of
an obvious and significant asymmetry between the two We set out to evaluate the role of the Mauthner array in
sides after the lesion (Figure 6, angular velocity; Table generating escape responses by examining the behav-
1). Photoablation of the Mauthner cell alone resulted in ioral consequences of killing these cells. The Foreman
a significant (p 5 0.0175) decrease in angular velocity and Eaton model (1993) proposed that the array cells
for tail-elicited responses but had no effect on head- contribute differentially to control the strength of the
elicited responses. Thus, killing all three cells dramati- motor output. The Mauthner cell alone mediates tail-
cally slowed the turn in response to both head and elicited responses, resulting in a shallower bend, while
tail stimuli, whereas killing the Mauthner cell alone only all three homologs are involved in head-elicited re-
slowed tail responses. sponses, resulting in a deeper bend. Using calcium im-
aging in larval zebrafish, OMalley et al. (1996) reported
Effects of Lesions on Duration of Initial Turn that the activity patterns of the Mauthner array match
The previous sections reveal that after array lesions, the this model. Our lesion results now provide a causal link
fish turn more slowly, but to about the same extent. between these neurons and behavior. Lesions of the
Thus, one might expect that they should be taking a Mauthner array resulted in the elimination of high-per-
significantly longer time to complete the turn. This is formance escape responses to both head- and tail-
indeed what we found (Figure 6, duration). Before le- directed stimuli, as measured by latency to response,
sions, there was excellent symmetry in the duration of angular velocity, and turn duration. Lesions of the
the turns to both sides. Head turns averaged 8.2 ms; Mauthner cell alone had significant, similarly detrimental
tail turns, 7.3 ms. Killing all three homologs doubled the effects on tail-elicited responses, while head-elicited
duration of turns on the lesioned side for both head- (17 responses appeared unchanged. Since we have not
343
Neuron
332
done separate lesions of MiD2cm and MiD3cm, we do cell is to insure a powerful bend by overriding other
not yet know their individual contributions to escapes. motor programs (Svoboda and Fetcho, 1996). It is possi-
However, the greater impact of the array lesions support ble that MiD2cm and MiD3cm can, by themselves, pro-
the idea that MiD2cm and/or MiD3cm, like the Mauthner duce a high-performance escape from rest, when axial
cell, contributes to escape behavior. motor activity is minimal, but could not do so without
These results resolve an apparent paradox in earlier the Mauthner cell if there was substantial conflicting motor
Mauthner lesion studies. Substantial physiological evi- activity. If this explanation is correct, then Mauthner cell
dence links activity of the Mauthner cell to escape be- lesions should produce impairments in escapes elicited
havior (Zottoli, 1977; Eaton et al., 1981; Nissanov et al., by head stimuli during swimming.
1990). It was therefore surprising when lesioning studies The Foreman and Eaton model and our previous cal-
reported few and inconsistent behavioral changes after cium imaging studies, which showed that the Mauthner
killing the Mauthner cell (Eaton and Kimmel, 1980; Kim- cell alone is active in escapes to tail stimuli, led us to
mel et al., 1980; Eaton et al., 1982). After irradiating predict that ablating the Mauthner cell should remove
zebrafish early in development, Kimmel et al. (1980) all high-performance tail-elicited responses. However,
found that those larvae that lacked a Mauthner cell also even in tail-elicited escapes, we did occasionally ob-
exhibited decreased fast-start performances on the le- serve high-performance responses from the lesioned
sioned side. However, irradiation likely led to the re- side after killing the Mauthner cell. One possible expla-
moval of other cells, some of which might be involved nation for the remaining high-performance turns was
in escape behavior. In the most cell-specific lesions, that they occurred when the stimulus was more rostral
Eaton et al. (1982) electrolytically killed the Mauthner than usual, thereby activating one or both of the re-
cell in adult goldfish and found no effects except a small maining homologs. However, there was no correlation
but significant increase in latency. Since the nature of between these high-performance responses and the
their stimulus did not allow them to determine where the rostrocaudal position of the stimulus. Another possibil-
fish would perceive the stimulus along the rostrocaudal ity is that MiD2cm and/or MiD3cm takes over after
axis, head- and tail-elicited responses were pooled to- Mauthner lesions. The fact that these high-performance
gether in their analyses. We now know that lesioning responses are completely eliminated in animals where
the Mauthner cell alone has no effect on head-elicited the entire array has been lesioned supports a role for
escapes but does substantially impair tail-elicited re- MiD2cm and/or MiD3cm in the production of these re-
sponses. These impairments included large effects on sponses after Mauthner lesions. Functional substitution
the latency to respond and smaller, though still signifi- by MiD2cm and MiD3cm could be revealed by calcium
cant, effects on angular velocity and duration of the imaging if these cells, which do not respond to tail-
initial turn. If head- and tail-elicited escapes were pooled directed stimuli in intact fish, do respond to tail-directed
together following Mauthner cell lesions, the unchanged stimuli after Mauthner lesions.
head responses would mask the smaller deficits in angu- Our observations indicate that zebrafish can produce
lar velocity and duration and reduce the large increases turns of a given amplitude via multiple pathways. While
in latency. This may explain why, in their pooled data, all short-latency, high-performance escapes are re-
Eaton et al. (1982) found only changes in latency follow- moved by lesioning the Mauthner array, the animals
ing Mauthner cell lesions. are still capable of sensing the stimulus, calculating its
Our observations demonstrate that the Mauthner cell direction, and responding with an appropriate amplitude
does indeed play an important role in generating escape turn, albeit much more slowly and at a very long latency.
responses, particularly those elicited by caudal stimuli. One can compare the Mauthner array circuit to a reflex
While clarifying the role of the Mauthner cell in escape pathway, allowing quick and decisive retreat from dan-
behavior, our observations raise new and interesting ger. After its removal, fish can still turn away from stimuli,
questions about the relative contributions of the three but they most likely do so via other slower, polysynaptic
homologs in the Mauthner array. Based on the Foreman pathways that require more processing and hence a
and Eaton model, we would predict that the removal of longer latency to respond. Although these fish can re-
any of the three homologs should result in a decrease spond with a turn of appropriate amplitude, these slow
in head-elicited escape performance. However, removal turns do not even begin until after a normal escape
of the Mauthner cell had no discernible effect on head- would be complete. Such fish would likely be doomed
elicited responses. This is surprising, given that activa- in a predatory attack.
tion of the Mauthner cell is sufficient to produce a strong Our experiments show that the serially repeated retic-
bend (Nissanov et al., 1990). These observations high- ulospinal neurons in the Mauthner array form a function-
light the difficulties in making predictions about le- ally related group. These experiments were possible
sioning experiments in complex systems, where several because zebrafish have relatively few reticulospinal neu-
neurons potentially contribute to a behavior (Eaton and rons and they form sets of repeated cells with individu-
DiDomenico, 1985). ally identifiable members. However, the overall segmen-
In this particular case, a likely explanation for the ap- tal organization of hindbrain neurons, so obvious in
parent lack of effects of Mauthner lesions on head-elic- zebrafish, is evident in vertebrates generally (Clarke and
ited responses is the behavioral context in which the Lumsden, 1993). Many of these hindbrain neurons are
escapes were elicited. The escapes were produced while part of a reticulospinal system that is strikingly similar in
the animal was still, instead of while swimming. This fish and mammals (Fetcho, 1992; Lingenhohl and Friauf,
allowed us more easily to control stimulus direction and 1994). Thus, it seems very likely that segmental func-
strength but ignored some of the adaptive relevance of tional groups similar to those in zebrafish also exist in
the fast-start response. A key function of the Mauthner other vertebrates.
344
Behavioral Roles of Hindbrain Neurons in Zebrafish
333
Laser ablations of single cells, used first to study de- needle tip was kept about 0.5 mm from the fish at the time the
velopment (Sulston and White, 1980; Eisen et al., 1989), stimulus was given. A successful escape trial was recorded from
the beginning of the trigger to the picospritzer, as indicated by a
have proved fruitful for linking neurons to behavior in
stimulus indicatora deflection of a piezoelectric crystal that was
invertebrate systems (Selverston and Miller, 1980; Chal- wired in parallel with the picospritzer. Typically, the water left the
fie et al., 1985; Bargmann and Horvitz, 1991). Our lesion needle tip about 1014 ms after the crystal deflection and hit the
studies of the Mauthner array in zebrafish extend this fish about 2 ms after that.
approach to the analysis of the behavioral contribution The water pulses were generated by a pressure pulse of 17 psi
of identified neurons in a vertebrate system. Zebrafish and 5 ms duration. The pressure was occasionally adjusted in order
to achieve a consistent stimulus (as monitored by the speed of exit
have many other small groups of identifiable neurons in
of the dye) but adjustments did not vary more than 61 psi. These
spinal cord and brain whose behavioral roles could be settings were determined to be the lowest level that can reliably
analyzed with this approach. Cell-specific ablations, produce an escape response without the water stream disturbing
when combined with the functional imaging at single- the animals movements.
cell resolution and the genetic tools also available in Squirting the colored water in front of the fish, so that the animal
zebrafish (Granato et al., 1996; Fetcho and Liu, 1998), might see but not feel the stimulus, did not elicit a response. There-
fore, we believe the animals are not responding to the visual aspects
should make zebrafish a powerful model for exploring
of the stimulus. The most likely sensory modalities mediating the
the neural basis of vertebrate behavior. response are the somatosensory system, the lateral line, or the
ear. If the response is visual, our latency measurements should be
Experimental Procedures calculated from the time the dye exits the needle, instead of from
the time the water made contact with the fish. This would have
Larval fish were obtained from laboratory breeding stock. Zygotes added only about 2 ms to our latency data. This small change would
were kept in 10% Hanks buffer at 28.58C until the fish had developed not have altered our conclusions about the consequences of lesions,
to the point where they could swim in the water column (about 4 which had large affects on latency.
days). At this stage, they were moved to room temperature in order Stimuli to the head were directed at the ear. Stimuli to the tail
to acclimate the fish to the temperature at which the experiments were directed caudal to the anal pore but well rostral to the injection
were done (268C). site. The order of stimulus presentation was to the left side of the
head and then the right, followed by the left side of the tail and then
Retrograde Labeling the right. All of the fish in the group were tested for the same
Four- to five-day-old larval zebrafish were anesthetized with 0.02% quadrant before moving on to the next trial/quadrant. Therefore,
3-aminobenzoic acid ethyl ester (MS222). The targeted cells were each fish rested between trials for an average of around 20 min. Also,
retrogradely labeled by pressure injection via a glass microelectrode each fish was not presented with a stimulus in the same quadrant for
of a 50% solution of CGD (10,000 MW; Molecular Probes, Eugene, well over 1 hr. Such delays should prevent habituation, fatigue, etc.
OR) in 10% Hanks into the spinal cord (Fetcho and OMalley, 1995). On occasion, a fish either failed to respond to the stimulus, the
Injections were targeted to the ventral cord to selectively label stimulus missed the fish, the fish gave a premature response, or
Mauthner and its homologs without disrupting more dorsal sensory the fish turned slightly on its side. Only responses that occurred
pathways. Since any injection would disrupt local circuits, injections after the stimulus was deployed and in which the fish remained
were also targeted to the most caudal part of the tail and all test upright throughout the initial bend were analyzed digitally. A mini-
stimuli were given rostral to the injection site. After injection, the mum of 20 trials were collected per fish (5 trials per quadrant).
fish were allowed to recover in 10% Hanks solution containing food In pilot experiments, we elicited escapes by using a glass probe
(cultured paramecia). Nine to ten hours later, the cell labeling was driven by a piezoelectric device to abruptly touch the fish on the
verified under the confocal microscope. Between four to six of the head or tail. This produced more robust differences in the form of
best-labeled animals were retained in 10% Hanks containing para- the escapes to head and tail stimuli than the squirts, but because
mecia for behavioral trials. the taps were more traumatic to the fish we elected to use squirts
to obtain the many trials needed for the lesion experiments.
Confocal Microscopy
Photoablation of Cells
The fish were briefly anesthetized in MS222, placed on a cover glass
After prelesion behavioral testing and overnight recovery, the fish
in a petri dish, embedded on their backs in a thin layer of 1.2% agar
were mounted in agar as described above and the target neurons
(Eaton et al., 1984), and screened with confocal microscopy for the
either Mauthner alone or all three array cellswere reidentified by
desired labeling. Confocal images were obtained by looking into
their position and morphology (Kimmel et al., 1982; Metcalfe et al.,
the head of the intact fish using a Zeiss inverted microscope with
1986). To lesion a targeted cell, the confocal microscope was fo-
a 633 water objective and a Zeiss laser-scanning confocal imaging
cused at the highest zoom (Zeiss 633 water 0.9 NA) on the brightest
system (LSM 510). Mauthner cells and their homologs were identi-
point of the labeled cell, and the cell was exposed to the laser (lex 5
fied by their highly characteristic morphology and position (Kimmel
488) at maximum intensity. To kill the Mauthner cell, exposure for
et al., 1982; Metcalfe et al., 1986). The potential for photo-induced 10 min was usually necessary. The smaller homologs, probably
damage to the cells during screening was reduced by minimizing because they contain less dye, required longer exposures of up to
the illumination used and increasing the sensitivity of the system 12 min. We checked the region of the illumination to make sure
(by opening the confocal aperture and increasing gain in the photo- there was no evidence of general tissue damage immediately after
multiplier). the ablation and 2 days later. Exposure for longer than 20 min could
lead to tissue damage, so we limited the exposure to well below
High-Speed Recording of Behavior that duration. Our attempts to use the minimal effective exposure
Labeled fish were placed into individual petri dishes (3.5 cm) filled time might explain why not all of our lesion attempts were success-
with 10% Hanks buffer at a depth of 34 mm. Escape responses ful. After exposure to the laser, the fish were transferred from agar
were recorded with a high-speed camera that captures images digi- into a small petri dish of 10% Hanks buffer containing paramecia
tally at 1,000 frames/s (EG&G Reticon, Sunnyvale, CA). At the same and allowed to recover overnight. Following postlesion behavioral
time, a slow-speed video camera recorded the entire procedure. In trials, the animals were remounted for confocal microscopy in order
this way, we could verify later which trials were retained digitally to verify the success of the lesions and check for peripheral damage.
and the history of each trial. Escape responses were elicited by a
measured pulse of water, delivered from a picospritzer (General Data Analysis
Valve, Fairfield, NJ) through a syringe and 27 gauge needle cut blunt Several kinematic parameters of the initial turn of the escape were
and bent to about 1008. In order to monitor the stimulus, fast green selected for analysis: the latency to its initiation (time from the con-
dye was added to the water at a concentration of 0.05 mg/ml. The tact of the pulse of water to the beginning of the bending movement),
345
Neuron
334
the maximum angle of the turn, its peak angular velocity, and its References
duration (time from the beginning of movement to the maximum
angle). The movements of the fish were analyzed for these kinematic Bargmann, C.I., and Horvitz, H.R. (1991). Chemosensory neurons
parameters by using a specialized program written in Labview (Na- with overlapping functions direct chemotaxis to multiple chemicals
tional Instruments, Austin TX). The analysis was automated; the in C. elegans. Neuron 7, 792742.
image of the fish was thresholded and the binary silhouette of Chalfie, M.J., Sulston, J.E., White, J.G., Southgate, E., Thomson,
the fish was used to determine, via fitting routines, the location J.N., and Brenner, S. (1985). The neural circuit for touch sensitivity
of the rostral midline, which does not bend much during the turns. in Caenorhabditis elegans. J. Neurosci. 5, 956964.
The midline from each successive frame was plotted to give a repre- Clarke, J.D., and Lumsden, A. (1993). Segmental repetition of neu-
sentation of the animals movements (Figure 3, top). The program ronal phenotype sets in the chick embryo hindbrain. Development
calculated the angle between the position of the midline in succes- 118, 151162.
sive frames and its original position (Figure 3, middle) and also
DiDomenico, R., Nissanov, J., and Eaton, R.C. (1988). Lateralization
provided other kinematic data such as the duration of the turn and
and adaptation of a continuously variable behavior following lesions
the angular velocity, which was obtained by differentiation of the
of a reticulospinal command neuron. Brain Res. 473, 1528.
curves for turn angle (Figure 3, bottom).
These parameters were then statistically compared using analysis Eaton, R.C., and DiDomenico, R. (1985). Command and the neural
of variance (SuperAnova, Abacus Concepts, Berkeley CA) with re- causation of behavior: a theoretical analysis of the necessity and
peated measures. We used statistical contrasts in the ANOVA to sufficiency paradigm. Brain Behav. Evol. 27, 132164.
perform specific comparisons that we identified as the important Eaton, R.C., and Emberley, D.S. (1991). How stimulus direction de-
ones prior to the experiment. These included comparisons of prele- termines the trajectory of the Mauthner-initiated escape response
sion versus postlesion, intact side versus lesioned side, head versus in a teleost fish. J. Exp. Biol. 161, 469487.
tail, and lesions of the array versus lesions of the Mauthner cell Eaton, R.C., and Kimmel, C.B. (1980). Directional sensitivity of the
alone. Mauthner cell system to vibrational stimulation in zebrafish larvae.
Not every trial collected was used in our statistical analysis. Those J. Comp. Physiol. [A] 140, 337342.
trials that were omitted fell into two classes: nonescape or scoot Eaton, R.C., Lavender, W.A., and Wieland, C.M. (1981). Identification
responses, and escape responses in which the fish turned toward of Mauthner-initiated response patterns in goldfish: evidence from
the stimulus. These classes and the justification for their omission simultaneous cinematography and electrophysiology. J. Comp.
are discussed below. The data presented comprise three trials prele- Physiol. [A] 144, 521531.
sion and three postlesion (out of five collected in each case) from
Eaton, R.C., Lavender, W.A., and Wieland, C.M. (1982). Alternative
each quadrant of eleven fish: six fish with array lesions and five with
neural pathways initiate fast-start responses following lesions of
just Mauthner lesions.
the Mauthner neuron in goldfish. J. Comp. Physiol. [A] 145, 485496.
In a small percentage (17%) of trials, the fish moves away from
the stimulus with a movement that is sufficiently different from an Eaton, R.C., Nissanov, J., and Wieland, C.M. (1984). Differential acti-
escape response that it can be distinguished visually during the vation of Mauthner and non-Mauthner startle circuits in the zebra-
behavior. These responses consist of a very weak turn away from fish: implications for functional substitution. J. Comp. Physiol. [A]
the stimulus, followed by a somewhat larger turn in the same direc- 155, 813820.
tion, with no evidence of the initial turn/counter turn sequence typi- Eisen, J.S., Pike, S.H., and Debu, B. (1989). The growth cones of
cal for escapes. This appears visually as a weak scoot away from identified motoneurons in embryonic zebrafish select appropriate
the stimulus. We analyzed the kinematics of these responses to pathways in the absence of specific cellular interactions. Neuron 2,
verify that they were easily distinguishable from escapes. They con- 10971104.
sisted of an initial very small (about 108) and slow (less than 48/ms) Fetcho, J.R. (1992). The spinal motor system in early vertebrates
turn away, followed after a long (more than 20 ms) pause by another and some of its evolutionary changes. Brain Behav. Evol. 40, 8297.
turn in the same direction. This is very different from an escape,
Fetcho, J.R., and Liu, K.S. (1998). Zebrafish as a model system for
which consists of a large, rapid, continuous turn away from the
studying neuronal circuits and behavior. In Neural Mechanisms for
stimulus, followed by a robust counter turn. The scoot responses
Generating Locomotor Activity, O. Kiehn, R.M. Harris-Warrick, L.M.
were seen with the same frequency in prelesion trials and postlesion
Jordan, H. Hultborn, and N. Kudo, eds. Ann. NY Acad. Sci. 860,
trials and were, therefore, not a result of cell lesion. Since the scoots
333345.
were qualitatively different movements from escapes, we did not
include them in our analysis. Fetcho, J.R., and OMalley, D.M. (1995). Visualization of active neural
On occasion, an animal turned toward the direction of the stimu- circuitry in the spinal cord of intact zebrafish. J. Neurophysiol. 73,
lus. This inappropriate response was almost never seen in intact 399406.
fish but the frequency increased in fish that had been injected (4%) Fetcho, J.R., Cox, K.J.A., and OMalley, D.M. (1998). Monitoring
and increased even more in lesioned fish (11%). There was no differ- activity in neuronal populations with single-cell resolution in a be-
ence in the frequency of these turns between fish in which the entire having vertebrate. Histochem. J. 30, 153167.
array was lesioned and those in which only Mauthner was lesioned. Foreman, M.B., and Eaton, R.C. (1993). The direction change con-
Turns toward the stimulus exhibited a short latency and high angular cept for reticulospinal control of goldfish escape. J. Neurosci. 13,
velocity that is characteristic of responses of intact fish. After le- 41044113.
sioning, these turns occurred only when the stimulus was presented Granato, M., van Eeden, F.J., Schach, U., Trowe, T., Brand, M.,
on the lesioned side and not when it was presented on the intact Furutani-Seiki, M., Haffter, P., Hammerschmidt, M., Heisenberg,
side. For these reasons, we believe that these turns were the result C.P., Jiang, Y.J., et al. (1996). Genes controlling and mediating loco-
of the Mauthner array cells on the remaining, intact side being inap- motion behavior of the zebrafish embryo and larva. Development
propriately activated. Increased activation of the intact side follow- 123, 399413.
ing unilateral Mauthner cell lesion has been reported previously in
Kimmel, C.B., Eaton, R.C., and Powell, S.L. (1980). Decreased fast-
goldfish (DiDomenico et al., 1988). Because these inappropriate
start performance of zebrafish larvae lacking Mauthner neurons. J.
responses were most likely mediated by the cells on the unlesioned
Comp. Physiol. [A] 140, 343350.
side, they were not included in our analysis.
Kimmel, C.B., Powell, S.L., and Metcalfe, W.K. (1982). Brain neurons
Acknowledgments which project to the spinal cord in young larvae of the zebrafish. J.
Comp. Neurol. 205, 112127.
The authors would like to thank Melina Hale for many helpful com- Lingenhohl, K., and Friauf, E. (1994). Giant neurons in the rat reticular
ments on the manuscript. The work reported here was supported formation: a sensorimotor interface in the elementary acoustic star-
by NIH grant NS26539 (J. R. F.) and a postdoctoral fellowship from tle circuit? J. Neurosci. 14, 11761194.
the Helen Hay Whitney Foundation (K. S. L.). Metcalfe, W.K., Mendelson, B., and Kimmel, C.B. (1986). Segmental
homologies among reticulospinal neurons in the hindbrain of the
Received April 15, 1999; revised May 17, 1999. zebrafish larva. J. Comp. Neurol. 251, 147159.
346
Behavioral Roles of Hindbrain Neurons in Zebrafish
335
347
348
A structural and functional ground plan for neurons in
the hindbrain of zebrash
Amina Kinkhabwalaa,b, Michael Rileya, Minoru Koyamaa, Joost Monena, Chie Satouc,d, Yukiko Kimurac,
Shin-ichi Higashijimac,d, and Joseph Fetchoa,1
a
Department of Neurobiology and Behavior, Cornell University, Ithaca, NY 14853; bPrinceton Neuroscience Institute, Princeton University, Princeton, NJ 08544;
c
Okazaki Institute for Integrative Bioscience, National Institute for Physiological Sciences, National Institutes of Natural Sciences, Okazaki, Aichi 444-8787,
Japan; and dDepartment of Physiological Sciences, Graduate University for Advanced Studies (SOKENDAI), Okazaki, Aichi 444-8585, Japan
Edited by Lynn T. Landmesser, Case Western Reserve University, Cleveland, OH, and approved December 9, 2010 (received for review August 16, 2010)
The vertebrate hindbrain contains various sensory-motor net- when they are freely swimming and most of the major adult
works controlling movements of the eyes, jaw, head, and body. behaviors driven by hindbraine.g., swimming, feeding, and eye
Here we show that stripes of neurons with shared neurotransmit- movementsare functional. This suggests that the diverse net-
ter phenotype that extend throughout the hindbrain of young works in hindbrain might share an orderly functional patterning of
zebrash reect a broad underlying structural and functional neurons. We now show that the transmitter stripes reect an
patterning. The neurotransmitter stripes contain cell types with underlying order in the hindbrain in which stripes contain cell
shared gross morphologies and transcription factor markers. types, stacked in order by age as well as structural and functional
Neurons within a stripe are stacked systematically by extent and properties. This pattern is tied to behavior because neurons are
location of axonal projections, input resistance, and age, and are
recruited along the axis of a stripe as the speed of a motor be-
havior increases. We conclude that the diverse networks in the
recruited along the axis of the stripe during behavior. The
hindbrain have at their foundation a common structural and
implication of this pattern is that the many networks in hindbrain
functional plan.
are constructed from a series of neuronal components organized
into stripes that are ordered from top to bottom according to Results
a neurons age, structural and functional properties, and behav-
Hindbrain Transmitter Stripes and Transcription Factor Patterning.
ioral roles. This simple organization probably forms a foundation
We examined the transmitter stripe patterning in more than 20
for the construction of the networks underlying the many behav- larval, posthatching sh from each of two BAC transgenic lines
iors produced by the hindbrain. expressing green (GFP) and red (DsRed) uorescent proteins
driven by promoter regions of glyt2 and vglut2.1 (vglut2) to mark
interneuron | locomotion | recruitment | topography glycinergic and glutamatergic neurons, respectively (24, 25). Stripes
were clearly evident in the hindbrain of dual transgenic lines (vglut:
DsRed glyt2:GFP), as shown in the example from a 4-d post-
T he hindbrain contains a diverse set of sensory-motor networks
that control movements required for vision, respiration, mas-
tication, and locomotion in all vertebrates (1, 2). Most often these
fertilization (dpf) sh in Fig. 1A (n = 16). In cross-sections, these
stripes were organized in an interleaved manner from medial to
different networks are studied separately from one another, per- lateral on each side of the brain in all hindbrain segments (rhom-
haps because the behaviors are distinct, and the regional differ- bomeres), with a glutamatergic stripe located medially followed by
entiation of hindbrain suggests that its several networks might alternating glycinergic and glutamatergic stripes that extend pre-
have little in common. Thus, we have strong data for the hindbrain dominately dorsoventrally and rostrocaudally. This organization
control of eye movements, respiration, and locomotion (310), but was present in all of the sh. Although the glycinergic and gluta-
fewer unifying principles of structural and functional organization matergic neurons were segregated from one another in a columnar
that apply across the different networks. pattern, there were other unlabeled neurons, both scattered within
Structurally, the hindbrain is divided into segments, called stripes and in large contiguous areas between stripes that are
rhombomeres, which differ in the expression of homeotic genes, probably neurons with other transmitter phenotypes (cholinergic
in the morphological differentiation of neurons, and in their and GABAergic, e.g.) based upon prior in situ staining (23).
sensory inputs and motor outputs (2, 11). Though there are clear We next used transgenic lines and immunostaining to in-
distinctions among rhombomeres, there are indications from vestigate the relationship between the transmitter stripes and the
previous developmental work using in situ staining for tran- transcription factors alx (called chx10 in mammals), dbx1b, en-
scription factors and backlling of hindbrain neurons that there grailed-1, and barhl2. The transcription factors were expressed
may be structural patterns that cross rhombomere boundaries in stripe-like patterns resembling the patterning of transmitter
(1216). Prior work has also revealed parallels in the deve- stripes. Neurons expressing the alx (chx10) transcription factor
lopment of hindbrain and spinal cord, with the hindbrain sharing were clustered medially and overlapped the most medial gluta-
features of the now-classic transcription factor code that directs matergic stripe (Fig. 1B). Three-dimensional colocalization
development in spinal cord (1722). Though these studies did revealed that most, if not all, medial glutamatergic stripe neu-
not explore function because they were performed during early rons express the alx transcription factor (Fig. 1B, panel 3; n = 2
development, they raised the possibility of a broader structural- sh). Immunostaining for alx protein in the alx:GFP transgenic
functional patterning that spans rhomobomeres and may un- line conrmed that this line reliably marked the alx positive
derlie the organization of circuits for different behaviors.
Here we show that there is indeed a broad structural and
functional patterning of neurons in the hindbrain of young Author contributions: A.K. and J.F. designed research; A.K., M.R., M.K., and J.M. per-
formed research; C.S., Y.K., and S.-i.H. contributed new reagents/analytic tools; A.K.,
zebrash. The work was initially prompted by a striking patterning
M.R., M.K., J.M., and J.F. analyzed data; and A.K. and J.F. wrote the paper.
observed in earlier work in which we used in situ staining for
The authors declare no conict of interest.
markers of neurotransmitter phenotype to reveal putative glyci-
nergic, GABAergic, and glutamatergic neurons in the hindbrain This article is a PNAS Direct Submission.
(23). We found that neurons of the same transmitter phenotype Freely available online through the PNAS open access option.
were clustered together into stripes when viewed in cross-sections, 1
To whom correspondence should be addressed. E-mail: jrf49@cornell.edu.
and that these extended as columns throughout much of the This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
rostrocaudal axis of the hindbrain. The pattern is evident in sh 1073/pnas.1012185108/-/DCSupplemental.
349
glutamatergic domain and, more lateral, most likely, glycinergic
one. Neurons expressing the barhl2 transcription factor were
located in a band at the lateral edge of the hindbrain (Fig. 1E;
n = 10), overlapping a portion of a lateral crescent-like stripe of
glutamatergic neurons throughout hindbrain (Fig. 1E, panel 3).
In summary, neurons expressing particular transcription factors
are clustered into bands that align with neurotransmitter stripes
and, depending on the transcription factor, can be coextensive
with a transmitter stripe (alx, engrailed-1), overlap multiple
transmitter stripes (dbx1b), or overlap only a restricted, spatially
segregated portion of a stripe (barhl2).
We examined how the distribution of neurons expressing these
transcription factors changes from spinal cord (where they are also
expressed) into the hindbrain by examining optical cross-sections
at different rostrocaudal locations from confocal image stacks
from live sh in which neurons expressing two transcription factors
(alx and barhl2) were labeled in different colors. A dorsoventral
segregation of the different transcription factors in spinal cord
gradually changed at its rostral end into a mediolateral segregation
in hindbrain, indicating a topological transformation in expression
domains between the two regions (Fig. 1F).
NEUROSCIENCE
vglut:DsRed dbx1b:GFP transgenic sh shows that dbx colocalizes with
both the middle glutamatergic stripe as well as a more lateral stripe in
a region known to contain glycinergic neurons. (E) Cross-sections from
rhombomere 8 in a 4-dpf vglut:DsRed barhl2:GFP sh indicate that barhl2
staining overlaps the lateral portion of the most lateral glutamatergic
populations. (F) Cross-sections of the transformation of transcription factor
stripes from hindbrain (F1) to spinal cord (F5) in a cross of barhl2GFP and alx:
DsRed lines. White regions in panel 4 in BE indicate the location of the
transcription factor expression relative to the neurotransmitter stripes. All
cross-sections are maximum-intensity projections of 20- to 50-m volumes
from a confocal image stack of the entire hindbrain. (Scale bars, 20 m.)
Kinkhabwala et al. PNAS | January 18, 2011 | vol. 108 | no. 3 | 1165
350
axonal projections (13 of 15 neurons in rhombomeres 38; Fig.
2B) that sometimes projected beyond rostral hindbrain. All eight
neurons labeled in the middle (putative dbx1b) glycinergic stripe
(rhombomeres 68) had contralaterally extending axons with
both ascending and descending branches within hindbrain, often
of similar length (Fig. 2C). All seven neurons reconstructed from
the third, lateral glycinergic stripe (Fig. 2D) had both ipsilateral
and contralateral axonal projections that distinguished them
from those belonging to other stripes; there was, however, vari-
ation in the extent of ascending or descending projections on the
two sides of the hindbrain.
Although these 63 reconstructed single cells from four of the
transmitter stripes represent only a small fraction of the neurons
in hindbrain, they reveal a general pattern consistent with the
idea that these hindbrain stripes contain neurons with different
morphological features. This conclusion is also supported by
a series of backlling experiments designed to more specically
examine projection patterns within these transmitter stripes (26).
351
brain in R78 across the dorsoventral extent of the alx stripe.
Neurons positioned very dorsally had small amplitude, relatively
long duration action potentials, a depolarized resting membrane
potential, and very high input resistances in both preparations,
suggesting that they were very young, in accord with the Kaede
photoconversion data (see above). Interneurons above 70% of
the way up the stripe showed persistent activity that was not
correlated with behaviors such as swimming. Consequently, we
focused our input resistance and functional analysis on the
ventral 70% of the stripe, which included older neurons that
were already clearly incorporated into networks based upon their
rhythmic activation during swimming (see below).
Neurons with the lowest input resistance values were consis-
tently located in the most ventral stripe regions at 5 dpf, both in
the preparations with massive exposure of the brain (black dots
on Fig. 4D) and those with reduced exposure (gray squares on
Fig. 4D). There was a signicant correlation between the position
of a neuron along the stripe axis and its input resistance within
both preparations with either massive or reduced exposure of the
brain (Pearson correlation: P < 0.01 for reduced exposure, P <
Fig. 4. Structural and functional patterning within the alx stripe. (A) Patch
0.05 for massive exposure, P < 0.05 for both together). The input
recording of an alx:GFP+ neuron in hindbrain segment 7 of a 5-dpf sh,
resistance was lowest at the bottom of the stripe and increased in
while simultaneously recording motor activity in a ventral root. (A, panel 1)
Cross-section view of the patched cell (red) within the alx stripe (green). (A,
more dorsal locations in both cases, although the measurements
panel 2) Recording from the alx cell on the left is shown below the simul-
from the more exposed preparation were more scattered, probably
taneous ventral root recording. (B) Variation in axonal projections into spinal because of damage during exposure.
cord of hindbrain alx neurons. (B, panel 1) An example of a reconstruction of The presence of an orderly pattern of structural and physio-
a single alx neuron (red) within the alx:GFP transgenic line (lateral view). (A, logical properties along the stripe axis by age raised the question
panel 2) Lateral views of 3D reconstructions of neurons from different of whether the neurons were recruited during behavior in an
dorsoventral locations in the medial glutamatergic stripe in rhombomere 7, orderly way along a stripe. To explore the involvement of neu-
with the most dorsal one at the top (normalized dorsoventral positions: 0.51, rons at different dorsoventral locations in the alx stripe at dif-
0.28, and 0.16 from ventral edge of stripe, rostral to the left). (C) A plot of ferent frequencies/speeds of swimming in a minimally invasive
the total 3D axonal length of 10 labeled alx neurons (two of them overlap, way, we used calcium imaging with Oregon Green BAPTA-1
shown with white circle) vs. their dorsal ventral location in the stripe. More dextran (Invitrogen) electroporated into the neurons in the alx:
ventral neurons have systematically longer axonal length (P < 0.0001). Red, DsRed transgenic line. We recorded from ventral roots to
green, and blue points correspond with the three neurons in B. (D) Plot of monitor the frequency of swimming produced by light or elec-
the input resistance of a neuron vs. its dorsoventral location in a stripe. trical stimulation and determined, based upon the calcium re-
More-dorsal neurons have systematically higher input resistances in both
sponse, the lowest swimming frequency at which a neuron was
more- (black dots, P < 0.05) and less-exposed brain preparations (gray
activated (25). We subsequently collected confocal image stacks
squares, P < 0.01). See text for further details. (E) Examples of calcium im-
aging of alx neurons from different locations in the stripe during different
through the hindbrain in the region of labeled neurons so that we
frequencies of swimming in 5-dpf sh. (E, panels 13) Cross-sections images
could measure the position of the imaged cells relative to the axis
of alx:DsRed sh with locations of the neurons (dots) shown relative to the of the stripe.
NEUROSCIENCE
stripes. (E, panels 46) Calcium responses of the neurons on the left in two We collected many trials (averaging about 100) over a range
example trials at swimming frequencies near those when the neurons are of frequencies of swimming for each neuron to determine its
rst recruited. (F) Plot of the minimum swimming frequency at which minimum recruitment frequency. Some examples of calcium
a neuron responds vs. the dorsoventral location of the neuron, including responses of neurons at different dorsoventral locations are
both alx:DsRed+ alx:DsRed neurons in the region of the alx stripe (P < 0.05 shown in Fig. 4E. In these examples, the top neuron exhibited
for each correlation). Neurons from Fig. 4B are shown in color. (Scale bars, a uorescence increase in a bout with a peak frequency of 24 Hz,
200 m.) but not in one at 22 Hz; the middle neuron at 28Hz, but not
26Hz; and the bottom cell at 31 but not 29 Hz. We obtained
recruitment patterns and positions for 29 neurons from 20 sh.
region (rhombomeres 68) of 6-dpf sh by reconstructing them Eleven of the neurons expressed the alx transcription factor and
in three dimensions and measuring the total length of their axon. 18 did not, although the 18 non-alx neurons were in the general
A plot of location vs. axonal length for these neurons showed region of the alx stripe, so their position along the stripe could be
a signicant (P < 0.0001) correlation between their total axonal measured. A plot of recruitment frequency vs. position for both
length and their location (Fig. 4C), with the length increasing alx and non-alx neurons, in Fig. 4F, showed a consistent pattern
systematically, and nearly linearly, from dorsal to ventral along in which more-dorsal, younger cells were recruited at the lowest
the stripe axis. Tracking the growth of axons from dorsal and swimming frequencies and increasingly more-ventral neurons
ventral neurons in the alx stripe as they descend into spinal cord were recruited at increasingly higher swimming frequencies (P <
indicates that the axons of ventral neurons grow along cord faster 0.05 for both). Thus, the neurons are recruited from dorsal to
after their extension than those of more dorsal cells, suggesting ventral along the axis of the stripe as the frequency of swimming
that the differences in axonal extent are not just a result of dorsal increases, indicating a systematic relationship between position,
neurons having less time to grow, and raising the possibility that age, and recruitment that maps onto the axis of the alx stripe and
the differences might persist later in life. extends to other non-alx neurons in hindbrain as well.
We also explored whether physiological properties might vary
systematically along the axis of a stripe. Using hyperpolarizing Discussion
current steps while patch recording from single alx-positive Our experiments reveal that the striking organization into trans-
neurons in ve dpf alx:GFP transgenic sh, followed by labeling mitter stripes in hindbrain reects a broad patterning of neu-
and 3D reconstruction, we examined whether the input re- rons by cell type, morphology, age, projections, cellular properties,
sistance of alx neurons varied systematically with location in and activity patterns (summarized in Fig. 5). The sh we studied
a stripe. We used both an exposed-brain preparation as well as are freely swimming animals with functional motor networks for
a more intact preparation and sampled neurons from the hind- swimming, escape, jaw, and eye movements, all located in a hind-
Kinkhabwala et al. PNAS | January 18, 2011 | vol. 108 | no. 3 | 1167
352
relative to the midline parallel the patterns in zebrash (1214,
16, 34, 35). Morphological data from backlling studies in em-
bryonic chicks indicates that neurons with different morpholog-
ical features occupy systematic locations relative to the midline
(15), consistent with an early orderly disposition of cell types.
Though these studies provide structural indications of a pat-
terning like that in sh, there is no information about the func-
tional organization at early stages to allow comparison with the
zebrash patterning. We would predict, however, that networks
are set up when the patterning is simple, with the later func-
tion of the neurons mapping systematically onto their age as
in zebrash.
The organization of hindbrain has clear similarities to the
Fig. 5. Summary of the structural and functional rules governing hindbrain patterning in spinal cord, suggesting that the two regions giving
patterning. (A) Interleaved neurotransmitter stripes overlap with transcrip- rise to all of the motor output from the nervous system have
tion factor expression patterns. Different neuronal morphologies (Right) are elements of the same structural and functional network plan (17
associated (by matching colors here) with the different stripes. (B) Within an 22, 25, 27, 28). Though there are undoubtedly changes with
individual stripe (alx stripe, Left), the somata of neurons are ordered along growth, nervous system plasticity, and divergence in the networks
the axis of the stripe by age, the location of their processes in the neuropil,
for specic tasks, the basic features of structural and functional
and their total axonal length. This corresponds to systematic changes in in-
put resistance and recruitment along the axis of the stripe.
organization that we describe may lie at the foundation of the
construction of the many sensory-motor circuits throughout the
hindbrain, which produces a much broader range of sensory-
brain with a strikingly regular organization. The transmitter stripes motor outputs than spinal cord. The challenge now is to un-
extend throughout the hindbrain and across the well-studied derstand how many different networks in hindbrain are con-
rhombomeres (2, 11, 13, 15, 29), through regions containing many structed from this basic ground plan and how the early pattern is
different networks. The implication is that the many neural circuits reorganized as the brain continues to grow into its adult form.
in hindbrain are built from a set of neuronal types, which might be We explore one of these specialized hindbrain networks, and its
viewed as circuit components, with particular morphologies, neu- relationship to the patterning, in the study by Koyama et al. (26).
rotransmitters, and transcription factor phenotypes. The stripes Experimental Procedures
represent these different neuronal types. Within a given stripe, the
Fish Care. All experiments were performed on zebrash (Danio rerio) be-
neurons are arranged in an orderly way by structural and functional
tween 1 and 8 dpf obtained from a laboratory stock of wild-type and
properties (and, importantly, age) so that those with more exten-
transgenic adults. All procedures conform to the National Institutes of
sive projections and higher thresholds for activation, which are Health guidelines regarding animal experimentation and were approved by
involved in the fastest movements, are at the bottom of a stripe, Cornell Universitys Institutional Animal Care and Use Committee.
with the neurons involved in increasingly rened movements
stacked in order above them. We infer that hindbrain networks are Immunostaining. Standard whole-mount antibody staining procedures were
constructed in an orderly manner by drawing cells from organized used as described previously (36, 37).
stripes, much like selecting components from a set of parts where
the parts are arranged in order according to their structural or Stochastic Labeling. Stochastic labeling was performed as described previously
functional properties. (23).
Though the larval sh has nearly all of the same motor
behaviors as an adult, the brain of the adult is considerably larger Transgenic Lines. The transgenic lines we used included ones described in
and the neurons are clustered into nuclei located among the ber prior studies (24, 25, 38), as well as two new ones with the promoters dbx1b
tractsan arrangement much different from the simple pat- and barhl2. These new lines, Tg(dbx1b:GFP) and Tg(barhl2:GFP), were con-
terning in the young sh. This raises the question of the re- structed with the BACs zK17G17 and zC15L16 using a previously described
lationship between the networks in the young sh and those in method (38). The detailed approaches for generating these lines are de-
the adult. A wholesale reorganization of the connectivity of the scribed elsewhere.
networks with growth seems unlikely, because the patterns of
motor output in the young sh parallel those in adults (e.g., the Confocal Imaging. Confocal imaging was performed as described previously
alternation and rostrocaudal delays characteristic of swimming (28, 39).
are present throughout life) (3032). A reconstruction of the
networks would mean dispensing with a motor circuitry that al- Colocalization. Colocalization was assessed using the colocalization add-on in
ready worked. The more likely possibility is that the essentials of the Imaris software package (Bitplane).
the larval pattern of network organization and recruitment are
retained even as the neurons disperse to reorganize into nuclei. Neuronal Tracing and Location Measurements. Neurons were traced using the
lament reconstruction feature provided in Imaris.
This would imply that circuits are set up early when brain pat-
terning is simple, but that the somata then migrate with growth
Photoconversion and Analysis. Huc:Kaede and alx:Kaede transgenic embryos
to form nuclei, obscuring a simple underlying network pattern-
(38, 40) were illuminated with UV light from a mercury bulb source for 10
ing. If so, we predict that if we could track neurons as the brain
40 s within their chorion. Immediately afterward, photoconversion was
differentiated further, the recruitment order and the basic pat- conrmed by observation of the presence of red expression and an absence
terns of connectivity established early would be retained later in of green expression. Photoconverted sh were then kept at 28.5 C in
life, after migration, and be related to the age of the neuron, a light-tight container until the day of imaging.
which corresponded to its location in the younger sh. This
migration may occur to minimize the extent of axonal and den- Targeted Whole-Cell Patching of alx Neurons in an Exposed Brain Preparation for
dritic arbors in a much larger brain, given the evidence for Input Resistance Measurements. Larvae were anesthetized in 3-aminobenzoic
placement of neurons to minimize wiring in other systems (33). acid ethyl ester (0.02% in HBSS) and then immobilized using -bungarotoxin
This early organization in zebrash probably extends to the (Sigma-Aldrich; 0.1% in HBSS). To measure input resistance of alx neurons in
hindbrain of other vertebrates. Longitudinal bands of some of hindbrain, we initially used a dissection procedure similar to one described
the transcription factors we studied, and others, extend through previously (41). We later switched to a less-extensive exposure of the brain,
the hindbrain of frogs, chicks, and mice, and their locations leaving more of the head, including the eyes, intact to minimize potential
353
damage to the neurons. After the electrophysiological measurements, the The following day, we imaged the calcium indicator in hindbrain on a Zeiss
preparation was xed with 4% formaldehyde with the pipette in place to LSM 510 inverted confocal microscope while recording from a ventral root
avoid the movement of the cell body during the retraction of pipette. Then a z using methods similar to those applied previously (25, 27).
stack was acquired with a confocal microscope (LSM 510 META; Zeiss) for the Analysis of calcium imaging and ventral root recordings was performed
measurement of the soma position relative to the alx stripe. using custom-written MATLAB software with an approach similar to McLean
et al. (25).
Image stacks were then reconstructed in 3D using Imaris software (Bit-
In Vivo Whole-Cell Recordings in the Hindbrain. Whole-cell recordings were
plane). Neurons that were active during the experiment were identied. The
done in current clamp mode in 5-dpf alx:GFP nacre transgenic larvae using
length of the left and right stripes (at the position of the cell body) and the
methods similar to those described previously (25, 30, 41).
location of the neuron were each measured three times and averaged to
obtain the location of the neurons in the stripe.
Calcium Imaging. Transgenic 4-dpf alx:DsRed Casper sh (42) were rst More details of methods are provided in SI Experimental Procedures.
anesthetized using 3-aminobenzoic acid ethyl ester (MS-222, 0.02% in HBSS)
and embedded in low-melting-point agarose (1.6% in HBSS; Sigma). A patch ACKNOWLEDGMENTS. We thank David McLean for comments on the
electrode (510 M resistance) was lled with 20% Oregon Green BAPTA-1 manuscript and Lindsay Heller for maintaining the many lines of sh. This
(10,000 Mr; Invitrogen/Molecular Probes), and the indicator was electro- work was supported by National Science Foundation Integrative Graduate
porated along the dorsoventral axis of the alx stripe in caudal hindbrain by Education and Research Traineeship Grant 0333366 (to A.K.), National
Institutes of Health Training Grant T32 GM007469 (to A.K. and M.R), the
using a single-cell Axoporator (4 V, 20-ms duration square pulse; Molecular Ministry of Education, Science, Technology, Sports and Culture of Japan
Devices). Larvae were then removed from the agarose, placed in a Petri dish (S.-i.H.), the Japan Society for the Promotion of Science (M.K.), and National
containing HBSS, and stored in an incubator (28.5 C) overnight. Institutes of Health Grant NS26539 (to J.F.).
1. Garcia-Campmany L, Stam FJ, Goulding M (2010) From circuits to behaviour: Motor 23. Higashijima S, Schaefer M, Fetcho JR (2004) Neurotransmitter properties of spinal
networks in vertebrates. Curr Opin Neurobiol 20:116125. interneurons in embryonic and larval zebrash. J Comp Neurol 480:118.
2. Tmpel S, Wiedemann LM, Krumlauf R (2009) Hox genes and segmentation of the 24. Bae YK, et al. (2009) Anatomy of zebrash cerebellum and screen for mutations
vertebrate hindbrain. Curr Top Dev Biol 88:103137. affecting its development. Dev Biol 330:406426.
3. Sparks DL (2002) The brainstem control of saccadic eye movements. Nat Rev Neurosci 25. McLean DL, Fan J, Higashijima S, Hale ME, Fetcho JR (2007) A topographic map of
3:952964. recruitment in spinal cord. Nature 446:7175.
4. Smith JC, Abdala AP, Rybak IA, Paton JF (2009) Structural and functional architecture 26. Koyama M, et al. (2010) Mapping a sensory-motor network onto a structural and
of respiratory networks in the mammalian brainstem. Philos Trans R Soc Lond B Biol
functional ground plan in the hindbrain. Proc Natl Acad Sci USA, in press.
Sci 364:25772587.
27. McLean DL, Fetcho JR (2009) Spinal interneurons differentiate sequentially from
5. Grillner S, et al. (1995) Neural networks that co-ordinate locomotion and body
those driving the fastest swimming movements in larval zebrash to those driving the
orientation in lamprey. Trends Neurosci 18:270279.
6. Smith JC, Ellenberger HH, Ballanyi K, Richter DW, Feldman JL (1991) Pre-Btzinger slowest ones. J Neurosci 29:1356613577.
complex: A brainstem region that may generate respiratory rhythm in mammals. Science 28. McLean DL, Masino MA, Koh IY, Lindquist WB, Fetcho JR (2008) Continuous shifts in
254:726729. the active set of spinal interneurons during changes in locomotor speed. Nat Neurosci
7. Roberts A, Li WC, Soffe SR, Wolf E (2008) Origin of excitatory drive to a spinal 11:14191429.
locomotor network. Brain Res Brain Res Rev 57:2228. 29. Heyman I, Kent A, Lumsden A (1993) Cellular morphology and extracellular space at
8. Aksay E, et al. (2007) Functional dissection of circuitry in a neural integrator. Nat rhombomere boundaries in the chick embryo hindbrain. Dev Dyn 198:241253.
Neurosci 10:494504. 30. Masino MA, Fetcho JR (2005) Fictive swimming motor patterns in wild type and
9. Dubuc R, et al. (2008) Initiation of locomotion in lampreys. Brain Res Brain Res Rev 57: mutant larval zebrash. J Neurophysiol 93:31773188.
172182. 31. Liu DW, Westereld M (1988) Function of identied motoneurones and co-ordination
10. Faber DS, Fetcho JR, Korn H (1989) Neuronal networks underlying the escape response of primary and secondary motor systems during zebra sh swimming. J Physiol 403:
in goldsh. General implications for motor control. Ann N Y Acad Sci 563:1133. 7389.
11. Moens CB, Prince VE (2002) Constructing the hindbrain: Insights from the zebrash. 32. Fetcho JR, Svoboda KR (1993) Fictive swimming elicited by electrical stimulation of the
Dev Dyn 224:117.
midbrain in goldsh. J Neurophysiol 70:765780.
NEUROSCIENCE
12. Moreno N, Bachy I, Rtaux S, Gonzlez A (2005) LIM-homeodomain genes as territory
33. Chen BL, Hall DH, Chklovskii DB (2006) Wiring optimization can relate neuronal
markers in the brainstem of adult and developing Xenopus laevis. J Comp Neurol 485:
structure and function. Proc Natl Acad Sci USA 103:47234728.
240254.
34. Passini MA, Raymond PA, Schechter N (1998) Vsx-2, a gene encoding a paired-type
13. Schubert FR, Dietrich S, Mootoosamy RC, Chapman SC, Lumsden A (2001) Lbx1 marks
homeodomain, is expressed in the retina, hindbrain, and spinal cord during goldsh
a subset of interneurons in chick hindbrain and spinal cord. Mech Dev 101:181185.
14. Cepeda-Nieto AC, Pfaff SL, Varela-Echavarra A (2005) Homeodomain transcription embryogenesis. Brain Res Dev Brain Res 109:129135.
factors in the development of subsets of hindbrain reticulospinal neurons. Mol Cell 35. Storm R, et al. (2009) The bHLH transcription factor Olig3 marks the dorsal
Neurosci 28:3041. neuroepithelium of the hindbrain and is essential for the development of brainstem
15. Clarke JD, Lumsden A (1993) Segmental repetition of neuronal phenotype sets in the nuclei. Development 136:295305.
chick embryo hindbrain. Development 118:151162. 36. Higashijima S, Masino MA, Mandel G, Fetcho JR (2004) Engrailed-1 expression marks
16. Gray PA (2008) Transcription factors and the genetic organization of brain stem a primitive class of inhibitory spinal interneuron. J Neurosci 24:58275839.
respiratory neurons. J Appl Physiol 104:15131521. 37. McLean DL, Fetcho JR (2004) Ontogeny and innervation patterns of dopaminergic,
17. Briscoe J, Pierani A, Jessell TM, Ericson J (2000) A homeodomain protein code species noradrenergic, and serotonergic neurons in larval zebrash. J Comp Neurol 480:
progenitor cell identity and neuronal fate in the ventral neural tube. Cell 101: 3856.
435445. 38. Kimura Y, Okamura Y, Higashijima S (2006) alx, a zebrash homolog of Chx10, marks
18. Goulding M (2009) Circuits controlling vertebrate locomotion: Moving in a new ipsilateral descending excitatory interneurons that participate in the regulation of
direction. Nat Rev Neurosci 10:507518.
spinal locomotor circuits. J Neurosci 26:56845697.
19. Ericson J, et al. (1997) Pax6 controls progenitor cell identity and neuronal fate in
39. Hale ME, Ritter DA, Fetcho JR (2001) A confocal study of spinal interneurons in living
response to graded Shh signaling. Cell 90:169180.
larval zebrash. J Comp Neurol 437:116.
20. Osumi N, et al. (1997) Pax-6 is involved in the specication of hindbrain motor neuron
40. Sato T, Takahoko M, Okamoto H (2006) HuC:Kaede, a useful tool to label neural
subtype. Development 124:29612972.
21. Placzek M, Yamada T, Tessier-Lavigne M, Jessell T, Dodd J (1991) Control of morphologies in networks in vivo. Genesis 44:136142.
dorsoventral pattern in vertebrate neural development: Induction and polarizing 41. Drapeau P, Ali DW, Buss RR, Saint-Amant L (1999) In vivo recording from identiable
properties of the oor plate. Development 2 (Suppl 2):105122. neurons of the locomotor network in the developing zebrash. J Neurosci Methods
22. Takahashi M, Osumi N (2002) Pax6 regulates specication of ventral neurone 88:113.
subtypes in the hindbrain by establishing progenitor domains. Development 129: 42. White RM, et al. (2008) Transparent adult zebrash as a tool for in vivo transplantation
13271338. analysis. Cell Stem Cell 2:183189.
Kinkhabwala et al. PNAS | January 18, 2011 | vol. 108 | no. 3 | 1169
354
Mapping a sensory-motor network onto a structural
and functional ground plan in the hindbrain
Minoru Koyamaa, Amina Kinkhabwalaa,b, Chie Satouc,d, Shin-ichi Higashijimac,d, and Joseph Fetchoa,1
a
Department of Neurobiology and Behavior, Cornell University, Ithaca, NY 14853; bPrinceton Neuroscience Institute, Princeton University, Princeton, NJ 08544;
c
Okazaki Institute for Integrative Bioscience, National Institute for Physiological Sciences, National Institutes of Natural Sciences, Okazaki, Aichi 444-8787,
Japan; and dDepartment of Physiological Sciences, Graduate University for Advanced Studies (SOKENDAI), Okazaki, Aichi 444-8585, Japan
Edited by Lynn T. Landmesser, Case Western Reserve University, Cleveland, OH, and approved December 9, 2010 (received for review August 17, 2010)
The hindbrain of larval zebrash contains a relatively simple uorescent dyes (Alexa 647 or Texas Red) targeted to regions of
ground plan in which the neurons throughout it are arranged the M-cell on the basis of the innervation patterns of neurons in
into stripes that represent broad neuronal classes that differ in the network known from goldsh, which is diagrammed in Fig. 1
transmitter identity, morphology, and transcription factor expres- (5). These backlls were performed in a transgenic line, Tg(glyt2:
sion. Within the stripes, neurons are stacked continuously accord- GFP), in which the neurons in glycinergic stripes can be visual-
ing to age as well as structural and functional properties, such as ized by GFP expression (Fig. 1 A1A3). Here we organize these
axonal extent, input resistance, and the speed at which they are backlling data from sensory processing to motor output com-
recruited during movements. Here we address the question of ponents of the escape circuit.
how particular networks among the many different sensory-motor A balance between direct sensory excitation of the M-cell and
networks in hindbrain arise from such an orderly plan. We use
a feed-forward inhibition regulates Mauthner ring and the initi-
a combination of transgenic lines and pairwise patch recording to
ation of the escape response (Fig. 1A4) (5, 9, 10). To localize po-
tential glycinergic feedforward inhibitory neurons, we electro-
identify excitatory and inhibitory interneurons in the hindbrain
porated a red dye onto the cell body of one of the two M-cells in the
network for escape behaviors initiated by the Mauthner cell. We
line with GFP labeled glycinergic neurons [Tg(glyt2:GFP)] and
map this network onto the ground plan to show that an individual
looked for contralateral neurons that were both red (backlled)
hindbrain network is built by drawing components in predictable and green (glycinergic/GFP) (Fig. 1B1; n = 8 sh). Nearly all of the
ways from the underlying broad patterning of cell types stacked colabeled neurons were located in the most lateral of the three
within stripes according to their age and structural and functional glycine stripes (Fig. 1 B2 and B3; 64 of 66 cells) in hindbrain seg-
properties. Many different specialized hindbrain networks may ment (rhombomere) 4 (Fig. 1 B2B4), the segment that contains
arise similarly from a simple early patterning. the M-cell body. The colabeled neurons were largely located in the
ventral portion of the stripe (Fig. 1B3; 57 of 64 cells).
development | locomotion | neuronal circuit | Mauthner neuron Another less-well-studied interneuron thought to play a role in
sensory processing in the network is the spiral ber neuron (Fig.
1 A4 and C) (11, 12). To identify potential spiral ber neurons in
T he vertebrate hindbrain contains many different sensory-
motor networks that control movements of structures in the
body and the head. Our recent work shows that despite the di-
zebrash, we electroporated Alexa 647 or Texas Red into the
axon cap of the M-cell in Tg(GlyT2:GFP) sh and screened for
verse networks it contains, there is a remarkably orderly pat- contralateral GFP-negative neurons that were lled with the red
terning of neurons that extends throughout the hindbrain in dye (Fig. 1C1; n = 4 sh). Almost all of these were located in
young zebrash, across the well-studied segments, or rhombo- rhombomere 3 between the medial and middle of the three
meres (14). This pattern consists of a series of neurotransmitter glycinergic stripes, at their ventral ends (Fig. 1 C2 and C3; 67 of
stripes that represent broad neuronal classes that differ in 68 cells). The middle of the three glutamatergic stripes is located
transmitter identity, morphology, and transcription factor ex- in this region between the medial and the middle glycinergic
pression. Within these stripes, neurons are stacked continuously stripes, suggesting that the spiral ber neurons are located in the
according to age as well as structural and functional character- ventral part of the middle glutamatergic stripe.
istics, such as axonal extent, input resistance, and the speed at The outputs of the M-cell in hindbrain are relayed to other
which they are recruited during movements. neurons via an interneuron called a cranial relay neuron that is
The presence of such an orderly array of neurons suggests that monosynaptically activated by the M-cell (1315). To locate
networks are built by drawing components from particular stripes cranial relay (also called T-reticular) neurons in zebrash, we
that contain cell types with the required structure and transmitter backlled from the medial longitudinal fasciculus in rhombo-
phenotype, and from the particular position within a stripe oc- mere 8, where the axons of these neurons run, and looked for
cupied by neurons with the appropriate functional properties, red, GFP-negative neurons caudal to rhombomere 6 on the
much like selecting parts from a catalog. Here we test this pos- contralateral side in the Tg(glyt2:GFP) line (Fig. 1D1). Most of
sibility through a study of the hindbrain network for the escape the backlled neurons were located ventrally near the middle
behavior initiated by the Mauthner cell (M-cell) (58). We con- and the lateral glycinergic stripes (Fig. 1 D2D8; 48 of 51 cells).
clusively identify neurons in the hindbrain escape network of the The M-cell res single action potentials in goldsh because
M-cell in zebrash and map each cell type onto the arrangement multiple ring is blocked by feedback inhibition (Fig. 1A4) (15).
into the stripes that we describe in the previous work. Our work We located potential feedback inhibitory neurons in zebrash
reveals that this network is built from neurons in predictable
locations that are consistent with the idea that the stripe pat-
terning in hindbrain represents an orderly array of neurons from Author contributions: M.K., A.K., and J.F. designed research; M.K. and A.K. performed
which network components are drawn. Many different hindbrain research; C.S. and S.-i.H. contributed new reagents/analytic tools; M.K., A.K., and J.F.
analyzed data; and M.K. and J.F. wrote the paper.
networks probably arise in a similar way by drawing neurons from
a shared structurally and functionally ordered set of parts. The authors declare no conict of interest.
This article is a PNAS Direct Submission.
Results Freely available online through the PNAS open access option.
Retrograde Labeling of Interneurons. We set out to identify the 1
To whom correspondence should be addressed. E-mail: jrf49@cornell.edu.
locations of neuronal candidates for a role in the Mauthner es- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
cape network in zebrash by backlling them by injections of 1073/pnas.1012189108/-/DCSupplemental.
355
larvae by electroporating red dye into the axon cap (initial seg-
ment) of the M-cell in the transgenic line containing GFP-
labeled glycinergic neurons and then searching for the ipsilateral
glycinergic neurons that were both red and green (Fig. 1E1; n =
4 sh). All of the colabeled glycinergic neurons ipsilateral to the
backll were located ventrally among the most medial glycinergic
neurons (Fig. 1 E2 and E3; 15 neurons) and were in rhombomere
5, caudal to the M-cell (Fig. 1E4).
This backlling served to identify candidate neurons in the M-
cell network for patch recording, so we could both ll them with
dye to examine their structure relative to neurons in the associated
stripes and explore their connectivity and pharmacology in pair-
wise patch recording experiments to verify their roles in the
network and map rmly identied neurons onto the hindbrain
patterning.
NEUROSCIENCE
directly apposed to the cell body of the M-cell. This neuron also
had both an ipsilateral descending projection as well as a pro-
Fig. 1. Retrograde labeling of prospective interneurons in the Mauthner
jection to the contralateral hindbrain that approached the other
circuit. (A) Location of the M-cell relative to glycinergic neurons in the hind-
M-cell and then extended further rostrocaudally from there.
brain. (A1) Location of images on right. (A2) Dorsal view of a Tg(glyt2:GFP) line
at 5 dpf with reticulospinal neurons backlled from the spinal cord with Texas
To represent the location of all of the feedforward glycinergic
Red. (A3) Cross-section view of the M-cell and the three glycinergic stripes at
neurons identied physiologically, we registered the confocal im-
rhombomere 4. White arrowheads mark the glycinergic stripes. (A4) Circuit age stacks from different sh onto a template image stack of the Tg
diagram of the Mauthner circuit from goldsh. Excitatory neurons, white; (glyt2:GFP) line. As shown in Fig. 2 B1 and B2, all of the func-
inhibitory neurons, black. (B) Retrograde labeling of potential feedforward tionally identied feedforward inhibitory neurons were located in
inhibitory neurons. (B1) Experimental diagram. (B2) Cross-section view of the the ventral part of the lateral glycinergic stripe at rhombomere 4.
backlled neurons at rhombomere 4 in the Tg(glyt2:GFP) line at 5 dpf. Cola- To conrm that the glycinergic neurons in this region actually
beled voxels are orange. Arrowheads mark glycinergic stripes. (B3) Cross- receive sensory inputs, we used dye electroporations into auditory
section from registrations showing the glycinergic neurons from four example and facial nuclei and found that they innervate the region con-
sh, with different colored spheres for each. (B4) Top-down view of neurons in taining cell bodies of the identied glycinergic neurons, although
B3. White dashed line marks the M-cell. (C) Retrograde labeling of prospective their afferent terminations were somewhat separate within this
spiral ber neurons. (C1) Experimental diagram. (C2) Cross-section at rhom- region. We then tested whether the glycinergic neurons receive
bomere 3 in the Tg(glyt2:GFP) line at 5 dpf. Voxels that only contain the far red excitatory input from the auditory nerve by stimulating the audi-
dye used to backll are shown in red. (C3) Cross-section of the registered tory nerve while doing pairwise patch recordings (n = 6 pairs; Fig.
reconstructions for three different sh, as in B3. (C4) Top-down view of neu-
2A4) from the M-cell and the interneuron. The stimulus evoked an
rons in C3. M-cell marked by white dashed line. (D) Retrograde labeling of
action potential in these glycinergic neurons at strengths at which
potential cranial relay neurons. (D1) Experimental diagram. (D2D7) Back-
lled neurons from four different sh shown in registered cross sections in
the M-cell only showed subthreshold PSPs (4/6), indicating an
rhombomere 7/8. Arrowheads mark medial and middle glycinergic stripes. input pattern consistent with a feedforward identity.
(D8) Top-down view of the neurons in D2D7. White lines mark corresponding We reconstructed all of the recorded neurons to examine their
cross sections. (E) Retrograde labeling of prospective feedback inhibitory ipsilateral and contralateral projections and then quantied the
neurons. (E1) Experimental diagram. (E2) Cross-section of glycinergic neurons extent of these projections in rostral and caudal directions (Fig.
at rhombomere 5 in the Tg(glyt2:GFP) line at 5 dpf. Orange indicates colab- 2C). All of the neurons had ipsilateral descending and/or as-
eling with GFP and the far red dye. White arrowhead marks medial glycinergic cending projections and axons that crossed to the contralateral
neurons. (E3) Cross-section of the registered reconstructions from four sh. hindbrain, where they formed relatively long ascending and/or
(E4) Top-down view of the same neurons. r/c, rostral and caudal; d/v, dorsal and descending branches (n = 17). All of these neurons also had an
ventral; m/l, medial and lateral in this and later gures. (Scale bars, 20 m.) axon close to the contralateral M-cell, suggesting bilateral M-cell
Koyama et al. PNAS | January 18, 2011 | vol. 108 | no. 3 | 1171
356
Fig. 2. Feedforward inhibitory neurons. (A1) Pairwise whole-cell recording
of a GFP-positive feedforward inhibitory neuron and the ipsilateral M-cell
from a Tg(glyt2:GFP) relaxed line at 4 dpf. (A2) Firing the feedforward
neuron led to PSPs (average of ve traces) in the M-cell that inverted below
threshold (black) and were blocked by strychnine (1 M, gray). (A3) Response
of the feedforward inhibitory neuron to single spikes in the M-cell. (A4)
Eighth nerve stimulation led to ring of the interneuron and an EPSP in the
M-cell. (A5) Top-down view of the feedforward inhibitory neuron (green) Fig. 3. Spiral ber neurons. (A1) Pairwise recording of a GFP-negative spiral
and the M-cell (red) from A2 relative to the glycinergic neurons (blue). (A6) ber neuron at rhombomere 3 and the contralateral M-cell in the Tg(glyt2:
Cross-section of the same neurons, in rhombomere 4. Arrowheads mark GFP) relaxed line at 4 dpf (Left). (A2) Firing the spiral ber led to EPSPs (black)
glycinergic stripes. (B1 and B2) Location in top-down (B1) and cross-section in the M-cell. The slow component of the EPSPs was blocked by the gluta-
views (B2) of 12 registered feedforward inhibitory neurons (orange spheres) matergic blockers 10 M NBQX and 100 M D-AP5 (gray). (A3) PSP in the spiral
identied by pairwise recordings. Arrowheads mark the glycinergic stripes. ber neuron in response to ring the M-cell. (A4 and A5) Top-down and cross-
(C) Morphology of feedforward inhibitory neurons (16 cells). One cell is section views of the spiral ber neuron (green) and the M-cell (red) from A2
excluded owing to poor lling. (C1) Box-and-whisker plots showing the ex- and A3 relative to the glycinergic neurons (blue). Arrowheads mark the gly-
tent of axonal projections in four directions: ipsilateral ascending (ia), ipsi- cinergic stripes. (B1 and B2) Summary of the locations of eight spiral ber
lateral descending (id), contralateral ascending (ca), and contralateral neurons (blue) in the glycinergic line in rhombomeres 4 and 5, laid out as in Fig.
descending (cd). The whiskers extend to the most extreme data point, which 2B. (C1) Cross-section view of the backlled spiral ber neurons (red) in the
is no more than 1.5 times the interquartile range from the box. Asterisk transgenic glutamatergic line [Tg(vglut2.1:GFP)] at rhombomere 3. Arrow-
marks the line for the neuron in C2. (C2) Example reconstruction; dotted line heads mark the glutamatergic stripes. (C2) Cross-section of the backlled spiral
marks the midline. (Scale bars, 20 m.) ber neurons (white) in the Tg(vglut2.1:RG dbx1b:Cre) line. Arrowhead
indicates the dbx1b-positive middle glutamatergic stripe (green). (D1 and D2)
Projections of the spiral ber neurons shown as in Fig. 2C. Asterisk on D1 marks
connections as in goldsh. The inhibitory inputs to the contra- a point on the line for the neuron in D2. (Scale bars, 20 m.)
lateral M-cell were conrmed by pairwise patch recordings from
an interneuron and a contralateral M-cell in separate experi-
ments (n = 3 pairs). We conclude that these are indeed feed- cell only led to subthreshold PSPs in SFNs (Fig. 3A3; n = 10
forward glycinergic neurons; they are located ventrally among pairs); the pathway mediating these PSPs and its functional role
the oldest cells in the lateral glycinergic stripe, and they have the are unknown.
ipsilateral and contralateral projections characteristic of neurons The neuron shown in Fig. 3 A2 and A3, like all physiologically
stochastically labeled from many different locations in this stripe identied spiral ber neurons, was located in rhombomere 3
in our previous work (2). between the medial and the middle glycinergic stripes, where the
middle glutamatergic stripe is located (Fig. 3 A4, A5, B1, and B2;
Spiral Fiber Neurons. Paired patch recordings from the M-cell and n = 8). To conrm their location in the middle glutamatergic
contralateral GFP-negative cells in the Tg(glyt2:GFP) in the re- stripe, we backlled the neurons in transgenic sh [Tg(vglut2a:
gion containing putative spiral ber neurons led to the identi- GFP)] in which the glutamatergic neurons are labeled with GFP,
cation of 10 GFP-negative neurons that had excitatory con- and found that the neurons were indeed in the middle gluta-
nections with the contralateral M-cell, characteristic of spiral ber matergic stripe (Fig. 3C1). We also tested whether the neurons
neurons (Fig. 3A1) (11). The majority of these neurons produced were in the dbx1b transcription factor stripe (dbx1b marks the
biphasic excitatory postsynaptic potentials (EPSPs) in the con- middle glutamatergic and glycinergic stripes) by backlling the
tralateral M-cell (Fig. 3A2; n = 9 of 10 pairs), with a short latency spiral ber neurons in sh from a cross of a dbx1b:cre line with
(peak to peak, 0.24 0.03 ms, n = 10 pairs). The later component a line in which the vesicular glutamate transporter drives ex-
in the EPSP was blocked by glutamatergic blockers [Fig. 3A2; pression of oxed red uorescent protein, with GFP down-
10 M 2,3-dihydroxy-6-nitro-7-sulfamoyl-benzo[f]quinoxaline-2,3- stream. This cross produces sh in which the glutamatergic
dione (NBQX) and 50 M D-()-2-Amino-5-phosphonopentanoic neurons that are also dbx1b positive are green and the other
acid (D-AP5), n = 4 pairs), suggesting this slower component was glutamatergic neurons are red. The spiral ber neurons were in
glutamatergic. Injection of negative current pulses into one cell led the ventral part of the dbx1b-positive middle glutamatergic stripe
to a hyperpolarizing response in the other neuron (n = 8 pairs), (Fig. 3C2; n = 4 sh, 75 of 77 cells). For reasons we do not
and the fastest early component of the PSP was resistant to high- understand, although they were in the glutamatergic stripe, they
frequency ring over 100 Hz (n = 6 pairs). Both of these observa- were not stained red in the glutamatergic transgenic line. The
tions indicate that the fast component was probably an electrical pharmacology of their synaptic connections, however, supported
synapse mediated by gap junctions. An action potential in the M- the conclusion that they are glutamatergic.
357
Reconstructions of all of the patched neurons (n = 8) showed transmission and the glycinergic neurons in turn feedback to in-
that they had only contralateral and descending axonal projec- hibit the M-cell. Consistent with this possibility, the recorded pu-
tions running directly into the axon cap of the contralateral M- tative cranial relay neurons innervated the region where the
cell, with a characteristic axonal elaboration inside the cap (Fig. feedback glycinergic neurons were located (Fig. 4 A4A6; n = 10).
3 D1 and D2). In summary, the evidence shows that the spiral These cranial relay neurons also exhibited extensive inner-
ber neurons are located ventrally in the middle glutamatergic vation of the contralateral hindbrain (Fig. 4A4), including the
stripes and have contralateral projections like those of neurons regions where cranial motor nuclei (fth, seventh, and 10th
in the adjacent middle glycinergic stripe. motor nuclei) and pectoral n motoneurons are known to be
located. We did several experiments to examine these connec-
Cranial Relay Neurons. We conducted pairwise patch recordings tions. Paired recordings of an M-cell and a cranial relay neuron in
between the M-cell and GFP-negative neurons in the Tg(glyt2: the Tg(islet1:GFP) line in which motoneurons are GFP labeled
GFP) line in the region of putative cranial relay neurons and found (n = 3) showed that cranial relay neurons contributing to the
10 GFP-negative neurons that red in response to the M-cells feedback inhibition of M-cell projected to the cranial motor nu-
activation and whose activation led to (polysynaptic) IPSPs in the clei. Projections to the region of the pectoral n motoneurons
M-cell (Fig. 4 A1A3).These neurons red an action potential in were conrmed morphologically by single-cell electroporation of
response to the activation of the M-cell (Fig. 4A2; n = 10 pairs) cranial relay neurons in sh with n motoneurons backlled
with a short latency (onset of EPSP: 0.15 0.09 ms; peak of action (n = 3). Finally, in the most direct approach, an excitatory con-
potential: 0.70 0.31 ms, n = 10 pairs). A single action potential in nection to the fth motor nuclei was observed electrophysiolog-
a contralateral neuron in this class led to a longer latency, pre- ically by paired recordings of cranial relay neurons and moto-
sumably polysynaptic, IPSP in the M-cell (1.60 0.28 ms, n = 7 neurons in the fth motor nuclei in the Tg(islet1:GFP) line (n =
pairs; Fig. 4A3). This IPSP was blocked by a cholinergic blocker 7). The somata of the recorded neurons in this class were located
(Fig. 4A3, Right, mecamylamine, 100 M; n = 5 pairs), as well as below the middle or lateral glycinergic stripes, which contain
a glycinergic blocker (strychnine, 1 M; n = 3 pairs), suggesting neurons with commissural projections like the cranial relay neu-
that this class may excite glycinergic neurons via cholinergic rons (Fig. 4 B1 and B2). Compared with their extensive contra-
lateral innervation of motor pools, these relay neurons had only
short or no ipsilateral axonal/dendritic processes (Fig. 4 A4, C1,
and C2).
NEUROSCIENCE
produced by a single feedback-inhibitory neuron was smaller than
the IPSP produced by ring a cranial relay neuron (Figs. 5A3 vs.
4A3), consistent with the possibility that a single cranial relay
neuron activates multiple feedback-inhibitory cells.
All of the electrophysiologically identied feedback inhibitory
neurons were located among the most ventral and medial gly-
cinergic populations in rhombomere 5 (Fig. 5 B1 and B2; n = 9).
Because the stripe-like structure at rhombomere 5 in the glyci-
nergic line was not always very crisp, we also examined the stripe
location of the neurons by immunostaining for engrailed-1,
which clearly marks the medial glycinergic stripe in other
rhombomeres (2). The feedback-inhibitory neurons identied by
Fig. 4. Cranial relay neurons. (A1) Pairwise recordings of a GFP-negative backlling were in the ventral part of the engrailed-1positive
cranial relay neuron at rhombomere 7/8 and the contralateral M-cell in the stripe (Fig. 5C1; n = 14). The engrailed labeling of these cells
Tg(glyt2:GFP) relaxed line at 4 dpf. (A2) Firing the M-cell led to an action was dim (probably owing to down-regulation with age), but in
potential in the cranial relay neuron. (A3) Firing the cranial relay neuron single-neuron lling experiments we found engrailed-positive
with current injection led to a long-latency IPSP in the M-cell (black) that was staining of this cell type (Fig. 5C2).
blocked by application of the cholinergic blocker mecamylamine (100 M, Six of nine of these neurons had only ipsilateral ascending
gray). (A4) Top-down view of the recorded cranial relay neuron (green) and projections directly to the M-cells cell body, ending at its initial
the M-cell (red) relative to glycinergic neurons (blue). (A5) Cross-section of
segment. The other three cells had only ipsilateral and primarily
the cranial relay neuron at rhombomere 8. Arrowheads mark glycinergic
ascending projections contacting the proximal part of the ventral
stripes. (A6) Close up of axonal terminations from the cranial relay neuron in
region of the white rectangle in A4 showing apposition (arrowheads) to
dendrite of the M-cell (Fig. 5 D1 and D2).We also identied four
glycinergic neurons in the region. (B1 and B2) Locations of 10 cranial relay additional glycinergic neurons in the same region that had in-
neurons at rhombomeres 7/8 (red) identied with pairwise recordings, in hibitory input to the ipsilateral M-cell but did not reach ring
top-down (B1) and cross-section (B2) views. (C1 and C2) Summary of the threshold in response to M cells activation and so were excluded
morphology of seven cranial relay neurons, laid out as in Fig. 2C. Xs mark from the more detailed analysis. These also had only ipsilateral
neurons whose whole axonal extent was not captured after recording. Box- ascending projections directly to the axon cap. These several lines
and-whisker plots are from seven electroporated cells marked with blue of morphological and electrophysiological evidence show that
diamonds whose whole axonal extent was captured. Asterisk marks a point feedback-inhibitory neurons are located ventrally among the
on the line shown for the neuron in C2. (Scale bars, 20 m.) oldest cells in the medial glycinergic stripe and have the ipsilateral
Koyama et al. PNAS | January 18, 2011 | vol. 108 | no. 3 | 1173
358
the neurons located in stripes that correspond to their morpho-
logical and neurotransmitter phenotypes, they all lie at the ventral
end of the stripes, exactly where we would predict the fastest motor
behaviors to map on the basis of the continuous map of swimming
speed from slow to fast as one moves from dorsal to ventral along
a stripe, as we described in our previous article (2). Here the be-
havior is not a rhythmic one but is the fastest motor behavior
produced by the sh, and the neurons participating in it occupy
the low input resistance, high threshold, fast end of the stripes,
which also contain the oldest neurons. The Mauthner network
thus draws neurons by stripe and position just as expected given
their particular functional roles.
The hindbrain of vertebrates contains a variety of motor net-
works controlling sensory-motor aspects of movements of the
eyes, jaws, gills, and diaphragm (5, 16, 17). If the other networks
follow the same general patterning as the Mauthner network, we
predict that those neurons contributing to faster (more coarse)
sensory-motor behaviors will be drawn from early differentiating
cells located ventrally in those stripes that give rise to neurons
with the appropriate morphological and transmitter phenotype.
For example, a behavior that involves a range of movement
speeds, such as eye movements, which include very fast, large
saccadic movements as well as slow pursuit movements, would be
expected to draw from a range of dorsoventral locations (and
ages) in the stripes (16). Those neurons driving the fast saccadic
eye movements would be expected to have ventral (older)
locations in stripes, with slower eye movements driven by more
dorsal (younger) cells.
We do not yet know how the several other hindbrain networks
in zebrash or in any other animals map onto the overall order
Fig. 5. Feedback-inhibitory neurons. (A1) Pairwise recordings from a we describe because physiologically dening even one network
feedback-inhibitory neuron at rhombomere 5 and the ipsilateral M-cell in
the Tg(glyt2:GFP) relaxed line at 4 dpf. (A2) Firing the M-cell led to an
action potential in the feedback neuron. (A3) Firing the feedback inhibitory
neuron led to an IPSP in the M-cell that was blocked with 1 M strychnine
(gray line). (A5 and A6) Top-down (A5) and cross-section (A6, at rhombomere
5) views of the recorded feedback-inhibitory neuron (green) and the M-cell
(red) relative to the glycinergic neurons (blue). Arrowhead marks the most
medial glycinergic neurons. (B1 and B2) Summary of the position of nine
recorded feedback-inhibitory neurons (pink) in top-down (B1) and cross-
section (B2) views at rhombomere 5. Arrowhead marks the medial glycinergic
stripe. (C1) Hemicross-section of backlled feedback-inhibitory neurons (red)
in the Tg(glyt2:GFP) at 5 dpf, immunostained forengrailed-1 (En1, blue). (C2)
Single electroporated feedback neuron (FB) stained for engrailed (En1), and
the merged image. (D1 and D2) Morphology of feedback-inhibitory neurons
(n = 7 cells) formatted as in Fig. 2 C1 and C2. Black Xs mark two cells whose
whole axonal extent was not captured. Asterisk marks the line for the neuron
in D2. (Scale bars, 2 m in C2, 20 um in other panels.)
359
and its connectivity is a major challenge. Nonetheless, morpho- cholinergic transmission from the M-cell. Therefore, we primarily used the
logical and transcription factor expression studies suggest that relaxed mutants, in which a mutation in the skeletal muscle dihydropyridine
the ground plan in zebrash is likely to be shared across the receptor b1 gene, cacnb1, eliminates muscle contraction. We also used
vertebrate subphylum (1724). The extent to which this plan formamide treatment to paralyze sh in some experiments (34, 35). These
presages age-related networks in other vertebrates that are akin two preparations produced similar results, so we pooled the data.
to those in zebrash remains to be determined. In humans, faster
behaviors such as saccadic eye movements and startle responses Neuronal Labeling. Four-day-old larvae were anesthetized and embedded in
develop before slower ones, consistent with a patterning like that 1.7% agarose. Then, 20% Alexa Fluor Dextran, 10,000 MW, anionic, lysine
in zebrash (25, 26). Large Mauthner-like hindbrain neurons in xable (Molecular Probes) was applied in extracellular solution through a patch
the mammalian startle response differentiate early (27, 28), micropipette by electroporating the dye into regions of interest with 20-s trains
of 20-V, 1-ms pulses at 50 Hz. Confocal images were acquired at 5 dpf.
suggesting there might be similarities in patterning in mammals
at the neuronal as well as behavioral level. We expect that other
Immunohistochemistry. Standard whole-mount antibody staining procedures
vertebrate networks, like that of the M-cell, will draw compo-
were used (1, 36). A rabbit anti-Enhb1 antibody (a gift from A. Joyner, Sloan-
nents in predictable ways from the orderly array of parts that
Kettering Institute, New York, NY; 1:2501:500) was detected with goat anti-
underlies the construction of networks in hindbrain.
rabbit antibody conjugated with HRP (Invitrogen; 1:200) and the Alexa Fluor
647-Tyramide system (Invitrogen).
Experimental Procedures
Fish Care. All experiments were performed on 4- to 5-dpf zebrash (Dan-
Image Analysis. To summarize the location of the recorded or backlled
iorerio) obtained from a laboratory stock of wild-type, transgenic, and
neurons from multiple experiments, we registered volumes to a template
mutant adults. All procedures conform to the US National Institutes of
image of the Tg(glyt2:GFP) line by using custom MATLAB (MathWorks) soft-
Health guidelines regarding animal experimentation and were approved by
ware to integrate the functionalityof Imaris, including ImarisXT (Bitplane), with
Cornell Universitys Institutional Animal Care and Use Committee.
the coregistration/normalization functionality of SPM8 (http://www.l.ion.ucl.
ac.uk/spm). The quality of the registration was carefully examined by checking
Transgenic Lines and Mutants. The transgenic lines Tg(glyt2:GFP (29) and Tg
the alignment of stripes, and unsatisfactory registrations were discarded from
(vglut2a:GFP) (30) were described previously. The transgenic lines Tg(vglut2a:
the nal visualization. Volumetric colocalization was done in Imaris.
loxPDsRed-GFP) and Tg(dbx1b:Cre) were constructed using bacterial articial
chromosomes (DKEY-145P24 and zK17G17, respectively) (31). Details will be Details of the above methods are provided in SI Experimental Procedures.
described elsewhere. The mutant sh relaxed (red ts25a) was obtained from
the Max Planck Institute in Tubingen, Germany (32). ACKNOWLEDGMENTS. This work was supported by Japan Society for Pro-
motion of Science Postdoctoral Fellowships for Research Abroad (to M.K.),
Uehara Memorial Foundation Postdoctoral Fellowships (to M.K.), National
Electrophysiology. Patch-clamp recordings were performed as described Science Foundation Integrative Graduate Education and Research Trainee-
previously, with minor modications. We initially used cholinergic blockers ship 0333366 (to A.K.), National Institutes of Health (NIH) Training Grant T32
-bungarotoxin or d-tubocurarine to paralyze sh by blocking transmission GM007469 (to A.K.), the Ministry of Education, Science, Technology, Sports
at the neuromuscular junction (33), but this treatment also blocked the and Culture of Japan (S.-i.H.), and NIH Grant NS26539 (to J.F.).
1. Higashijima S, Mandel G, Fetcho JR (2004) Distribution of prospective glutamatergic, 20. Moreno N, Bachy I, Rtaux S, Gonzlez A (2005) LIM-homeodomain genes as territory
glycinergic, and GABAergic neurons in embryonic and larval zebrash. J Comp Neurol markers in the brainstem of adult and developing Xenopus laevis. J Comp Neurol 485:
480:118. 240254.
2. Kinkhabwala A, et al. (2011) A structural and functional ground plan for neurons in 21. Sieber MA, et al. (2007) Lbx1 acts as a selector gene in the fate determination of somato-
the hindbrain of zebrash. Proc Natl Acad Sci USA 108:11641169. sensory and viscerosensory relay neurons in the hindbrain. J Neurosci 27:49024909.
3. Hale ME, Kheirbek MA, Schriefer JE, Prince VE (2004) Hox gene misexpression and 22. Ericson J, et al. (1997) Pax6 controls progenitor cell identity and neuronal fate in
cell-specic lesions reveal functionality of homeotically transformed neurons. J
response to graded Shh signaling. Cell 90:169180.
NEUROSCIENCE
Neurosci 24:30703076.
23. Placzek M, Yamada T, Tessier-Lavigne M, Jessell T, Dodd J (1991) Control of
4. Moens CB, Cordes SP, Giorgianni MW, Barsh GS, Kimmel CB (1998) Equivalence in the
genetic control of hindbrain segmentation in sh and mouse. Development 125:381391. dorsoventral pattern in vertebrate neural development: induction and polarizing
5. Faber DS, Fetcho JR, Korn H (1989) Neuronal networks underlying the escape response properties of the oor plate. Development 2(Suppl 2):105122.
in goldsh. General implications for motor control. Ann N Y Acad Sci 563:1133. 24. Takahashi M, Osumi N (2002) Pax6 regulates specication of ventral neurone subtypes
6. Liu KS, Fetcho JR (1999) Laser ablations reveal functional relationships of segmental in the hindbrain by establishing progenitor domains. Development 129:13271338.
hindbrain neurons in zebrash. Neuron 23:325335. 25. Luna B, Velanova K, Geier CF (2008) Development of eye-movement control. Brain
7. Eaton RC, Bombardieri RA, Meyer DL (1977) The Mauthner-initiated startle response Cogn 68:293308.
in teleost sh. J Exp Biol 66:6581. 26. de Vries JI, Visser GH, Prechtl HF (1982) The emergence of fetal behaviour.
8. Kimmel CB, Sessions SK, Kimmel RJ (1981) Morphogenesis and synaptogenesis of the I. Qualitative aspects. Early Hum Dev 7:301322.
zebrash Mauthner neuron. J Comp Neurol 198:101120. 27. Lingenhhl K, Friauf E (1994) Giant neurons in the rat reticular formation: A sensori-
9. Zottoli SJ, Faber DS (1980) An identiable class of statoacoustic interneurons with motor interface in the elementary acoustic startle circuit? J Neurosci 14:11761194.
bilateral projections in the goldsh medulla. Neuroscience 5:12871302. 28. Altman J, Bayer SA (1980) Development of the brain stem in the rat. IV. Thymidine-
10. Nakayama H, Oda Y (2004) Common sensory inputs and differential excitability of radiographic study of the time of origin of neurons in the pontine region. J Comp
segmentally homologous reticulospinal neurons in the hindbrain. J Neurosci 24:31993209.
Neurol 194:905929.
11. Scott JW, Zottoli SJ, Beatty NP, Korn H (1994) Origin and function of spiral bers
29. McLean DL, Fan J, Higashijima S, Hale ME, Fetcho JR (2007) A topographic map of
projecting to the goldsh Mauthner cell. J Comp Neurol 339:7690.
recruitment in spinal cord. Nature 446:7175.
12. Lorent K, Liu KS, Fetcho JR, Granato M (2001) The zebrash space cadet gene controls
30. Bae YK, et al. (2009) Anatomy of zebrash cerebellum and screen for mutations
axonal pathnding of neurons that modulate fast turning movements. Development
128:21312142. affecting its development. Dev Biol 330:406426.
13. Kimmel CB, Metcalfe WK, Schabtach E (1985) T reticular interneurons: A class of 31. Kimura Y, Okamura Y, Higashijima S (2006) alx, a zebrash homolog of Chx10, marks
serially repeating cells in the zebrash hindbrain. J Comp Neurol 233:365376. ipsilateral descending excitatory interneurons that participate in the regulation of
14. Hackett JT, Faber DS (1983) Mauthner axon networks mediating supraspinal spinal locomotor circuits. J Neurosci 26:56845697.
components of the startle response in the goldsh. Neuroscience 8:317331. 32. Granato M, et al. (1996) Genes controlling and mediating locomotion behavior of the
15. Hackett JT, Faber DS (1983) Relay neurons mediate collateral inhibition of the zebrash embryo and larva. Development 123:399413.
goldsh Mauthner cell. Brain Res 264:302306. 33. Drapeau P, Ali DW, Buss RR, Saint-Amant L (1999) In vivo recording from identiable
16. Sparks DL (2002) The brainstem control of saccadic eye movements. Nat Rev Neurosci neurons of the locomotor network in the developing zebrash. J Neurosci Methods 88:113.
3:952964. 34. Wang M, Wen H, Brehm P (2008) Function of neuromuscular synapses in the zebrash
17. Gray PA (2008) Transcription factors and the genetic organization of brain stem choline-acetyltransferase mutant bajan. J Neurophysiol 100:19952004.
respiratory neurons. J Appl Physiol 104:15131521. 35. Ono F, Higashijima S, Shcherbatko A, Fetcho JR, Brehm P (2001) Paralytic zebrash
18. Clarke JD, Lumsden A (1993) Segmental repetition of neuronal phenotype sets in the
lacking acetylcholine receptors fail to localize rapsyn clusters to the synapse.
chick embryo hindbrain. Development 118:151162.
J Neurosci 21:54395448.
19. Cepeda-Nieto AC, Pfaff SL, Varela-Echavarra A (2005) Homeodomain transcription
36. McLean DL, Fetcho JR (2004) Ontogeny and innervation patterns of dopaminergic,
factors in the development of subsets of hindbrain reticulospinal neurons. Mol Cell
Neurosci 28:3041. noradrenergic, and serotonergic neurons in larval zebrash. J Comp Neurol 480:3856.
Koyama et al. PNAS | January 18, 2011 | vol. 108 | no. 3 | 1175
360
613
The Journal of Experimental Biology 213, 613-620
2010. Published by The Company of Biologists Ltd
doi:10.1242/jeb.037085
SUMMARY
Despite the key function of the Mauthner cells (M-cells) in initiating escape responses and thereby promoting survival, there are
multiple examples of M-cell loss across the teleost phylogeny. Only a few studies have directly considered the behavioral
consequences of naturally occurring M-cell variation across species. We chose to examine this issue in pufferfishes, as previous
research suggested that there might be variability in M-cell anatomy in this group of fish. We characterized the M-cell anatomy
and fast-start responses of two pufferfish species, Tetraodon nigroviridis and Diodon holocanthus. T. nigroviridis showed robust
fast-starts to both tactile and acoustic startling stimuli. These fast-starts occurred with a latency typical of M-cell initiation in other
fish, and retrograde labeling of spinal-projection neurons revealed that T. nigroviridis does have M-cells. By contrast, D.
holocanthus only rarely exhibited fast-start-like behavior, and these responses were at a substantially longer latency and were
much less extensive than those of T. nigroviridis. Using three complementary anatomical techniques we were unable to identify
obvious M-cell candidates in D. holocanthus. These results provide a clear correlation between M-cell presence or absence and
dramatic differences in fast-start behavior. The rich diversity within the pufferfish clade should allow future studies investigating
the factors that contribute to this correlated anatomical and behavioral variation.
Key words: evolution, fish, behavior, neuroanatomy, escape.
INTRODUCTION the onset of the response is delayed (DiDomenico et al., 1988; Eaton
The spectacular array of behaviors exhibited by animals has been et al., 1982; Liu and Fetcho, 1999; Zottoli et al., 1999).
generated through the processes of evolution. Behavioral Although considered a key example of a highly conserved
modifications are ultimately produced by differences in neural identifiable neuron, M-cells exhibit substantial diversity within
circuits, but what specific circuitry changes contribute to behavioral teleosts (Bierman et al., 2009; Stefanelli, 1980; Zottoli, 1978a). M-
evolution? What are the mechanisms by which evolution shapes the cells can be identified in the majority of teleosts, but many species
structure of these circuits? What constraints dictate how circuits can across the teleost phylogeny lack obvious M-cells or possess M-
respond to specific selective pressures? Probing the causes and cells with altered anatomy (Stefanelli, 1980; Zottoli, 1978a).
consequences of nervous system evolution requires a tractable neural Variable M-cell morphologies include a substantially smaller size,
circuit and a group of animals that exhibit variation in this circuit smaller axon diameter or differences in the structure of the axon
(for review, see Carr et al., 2001; Katz and Harris-Warrick, 1999; cap, which is a collection of inhibitory and excitatory fibers that
Nishikawa, 1997; Rose, 2004; Tierney, 1996; Wright, 2000). The surround the initial segment and axon hillock regions of the M-cell
circuitry that generates escape behavior in teleost fishes presents (Bierman et al., 2009; Stefanelli, 1980; Zottoli, 1978a). In some
such a system. cases, the discovery of small or missing M-cells has been associated
The escape response, also known as the fast-start or startle with specific types of life history variation, including bottom
response, is initiated by a pair of large, identifiable neurons in the dwelling and the use of crypsis or camouflage as an anti-predator
hindbrain: the Mauthner or M-cells (reviewed by Eaton et al., 2001; strategy (e.g. toadfish, flounder, lumpfish), or the extreme
Korn and Faber, 2005; Zottoli and Faber, 2000). Fast-starts in most modification of caudal fin anatomy (e.g. seahorse, pipefish, ocean
teleost fish consist of a C-type fast-start (C-start), the first stage of sunfish) (Marshall, 1971; Stefanelli, 1980; Uchihashi et al., 1960;
which is characterized by a rapid unilateral contraction of trunk Zottoli, 1978a).
musculature leading to head and tail movement, which causes the Given the diverse anatomy of M-cells among fishes, an
fish to bend into a C-shape. Stage 1 is typically followed by outstanding question is how does this natural variation affect fast-
subsequent movements, including a tail stroke that results in a start behavior? Only a single published study has directly examined
forward propulsion of the center of mass (stage 2), and either gliding the fast-start behavior of fish with missing M-cells (Hale, 2000).
or a burst swim (stage 3) (Domenici and Blake, 1997; Eaton et al., Lumpfish larvae exhibit significantly delayed C-starts when
1981; Foreman and Eaton, 1993). M-cell activity precedes the C- compared to larval zebrafish (Hale, 2000), which is consistent with
start, and electrical stimulation of Mauthner axons can elicit a C- predictions from M-cell ablation studies (Eaton et al., 1982; Zottoli
start (Nissanov et al., 1990; Zottoli, 1977). However, M-cells are et al., 1999). However, this study compared fish species across vast
not necessary for the production of a C-start. When M-cells in phylogenetic scales, and there may be a host of other factors that
goldfish or zebrafish are ablated, a C-start can be elicited, although contribute to the observed behavioral differences.
Pufferfishes, which comprise two sister families within the Order a minimum of 3-min intervals. The tactile stimulus consisted of a
Tetraodontiformes, offer a useful system for further exploring the sudden touch to the body using an opaque plastic rod (20cm1cm).
behavioral consequences of M-cell loss. Pufferfishes exhibit a suite The responses to tactile stimuli were not analyzed quantitatively
of adaptations in anti-predator defenses, including the capacity to because it was difficult to determine when the tactile stimulus actually
harbor the neurotoxin tetrodotoxin and inflation behavior. These made contact with the fish; however, in both species the behavior
anti-predator innovations might be predicted to alter selection on evoked by this stimulus was qualitatively similar to that evoked by
an ancestral anti-predator behavior, the fast-start. In support of this the acoustic stimulus (see Results).
hypothesis, the only two species of pufferfish that have been Each fish received two acoustic trials, with the exception of one
previously examined do not perform a fast-start response to a tactile D. holocanthus individual that had three trials. The acoustic stimulus
stimulus (Brainerd and Patek, 1998). In addition, anatomical was provided by the impact of a rubber mallet (340g) on the side
evidence suggests that some pufferfish lack M-cells (Otsuka, 1964; of the platform holding the tank. The mallet was suspended next to
Zottoli, 1978b). However, in several other pufferfish species, M- the platform and was raised to the same height for each trial in order
cells are evident but apparently reduced in size relative to other to provide a consistent stimulus.
teleosts (Stefanelli, 1980; Uchihashi et al., 1960). We quantified several parameters of fast-start behavior from
Several questions arise from this prior work, which could provide acoustic stimulus trials. We measured some components that are
insight into M-cell and fast-start evolution. From these previous known to be M-cell mediated, as defined by the fact that they are
studies it was unclear whether the lack of M-cells was associated consistently affected by experimental M-cell ablation (latency and
with the absence of fast-start behavior within a single species. In probability of the response) (Eaton et al., 1982; Liu and Fetcho,
addition, no pufferfish were previously shown to exhibit fast-starts; 1999; Zottoli et al., 1999). In addition, we measured components
therefore the function of the M-cells identified in some species was that probably involve the contribution of additional hindbrain
uncertain. Here, we examined fast-start responses using high-speed neurons (stage 1 duration and angle and peak angular velocity)
video and M-cell anatomy using both retrograde labeling and serial (Foreman and Eaton, 1993). The probability of a fast-start was
plastic sections in two species from the two pufferfish families: determined for each species by dividing the total number of trials
Tetraodon nigroviridis Marion de Proc (green spotted puffer, evoking a fast-start by the total number of trials (T. nigroviridis, 12
Family Tetraodontidae) and Diodon holocanthus Linnaeus trials: six fish with two trials each; D. holocanthus, 13 trials: five
(porcupine puffer or balloonfish, Family Diodontidae). fish with two trials each and one fish with three trials). For additional
quantification of behavior, imported video files were compiled into
MATERIALS AND METHODS image stacks in ImageJ Software (NIH, Bethesda, MD, USA).
Animals Latency was determined by counting the number of 2ms frames
Choice of species from the visible impact of the mallet with the apparatus to when
Tetraodon nigroviridis is closely related to, and often confused with, the fish commenced axial movement of the head or tail. The end
T. fluviatilis, which was previously found to have identifiable, but of stage 1 was delineated by the fish beginning to straighten its tail
small, M-cells (Stefanelli, 1980; Tagliani, 1905). Diodon and, in some cases, to turn its head in the opposite direction
holocanthus does not exhibit a fast-start in response to tactile (Domenici and Blake, 1997; Hale et al., 2002). Although the
stimulation (Brainerd and Patek, 1998), but its M-cell anatomy has transition from stage 1 was not as rapid for D. holocanthus (see
not been previously described. Results), it was nevertheless an obvious change in movement. To
quantify the amount and direction of movement during the C-start,
Housing conditions we measured the turning angle of each fish during the response
Fish were obtained through the aquarium trade (T. nigroviridis: (Foreman and Eaton, 1993; Zottoli et al., 1999). Using ImageJ, a
Aquariumfish.net; D. holocanthus: Sea Dwelling Creatures and line was extended along the midline of the anterior portion of the
Saltwaterfish.com) and were housed at both the Marine Biological fish and the change in angle of this line at the end of stage 1, and
Laboratory (MBL) and the Fred Hutchinson Cancer Research 10ms and 70ms after movement onset was measured [escape
Center (FHCRC). Tetraodon nigroviridis were housed in brackish trajectory angle (Foreman and Eaton, 1993)]. These values were
water at 15p.p.t. and D. holocanthus were housed in either filtered also used to calculate the peak angular velocity at 4ms intervals
seawater (MBL) or artificial sea water (FHCRC; Instant Ocean, during stage 1. The minimum straight-line distance between the
Aquarium Systems, Mentor, OH, USA). Fish were fed freeze-dried position of each fish before the onset of movement and after 70ms
or frozen krill every other day (Hikari, Hayward, CA, USA). All was also measured (Zottoli et al., 1999). To do this, an identifiable
animals were treated in accordance with the Institutional Animal point on the fish (on the midline directly in between the pectoral
Care and Use Committees at the MBL (protocol #07-07E) and fins) was tracked rather than the exact center of mass, which was
FHCRC (protocol #1781). not determined in this study.
Statistical analysis was performed using R statistical software
Behavior (http://www.r-project.org). Fishers exact test was used to compare
The startle responses of six T. nigroviridis and six D. holocanthus the proportion of trials in which individuals of each species exhibited
individuals were recorded using high-speed videography at 500 frames a C-start. Owing to the small number of D. holocanthus fast-starts
per second. Two different high-speed cameras were used over the that were recorded (see Results: three responses from two animals),
course of the experiment: Fastcam-PCI (Photron, San Diego, CA, we did not perform statistical tests on other behavioral measures.
USA) and 1200hs (The Cooke Corporation, Romulus, MI, USA). An
acrylic aquarium (22cm14cm13cm) was positioned on top of a Neuroanatomy
platform, and a mirror that was placed above the tank at a 45 deg. Retrograde tracing
angle enabled the top view to be recorded with the camera positioned The spinal cords of eleven T. nigroviridis (3.95cm standard length)
horizontally. Fish were presented with four or five trials of tactile and and eight D. holocanthus (5.88.1cm standard length) were cut and
acoustic stimuli, which were randomly interleaved and presented at back-filled with biotin dextran amine (BDA) to identify all cells
that project from the brain into the spinal cord, including the M- A B
cells. Fish were anesthetized in 0.03% MS-222 and were maintained
under anesthesia during the surgery. An incision was made in the
skin, and muscle and bone were removed in order to expose the
spinal cord. The spinal cord was cut using iridectomy scissors, and
a solution of biotin dextran amine (Invitrogen, Carlsbad, CA, USA)
that had resolidified onto a thin wire was applied onto the cut surface C
of the cord. The spinal cord was cut 14mm behind the base of the *
cerebellum (pufferfishes have extremely short spinal cords). There
was no difference in the extent of labeling in preparations that had
D *
tracer placement varying within this range (data not shown).
Following tracer application, the incision was closed with sutures
and Vetbond (3M Animal Care Products, St Paul, MN, USA). After
23days, fish were anesthetized in 0.03% MS-222 and perfused
with fixative [4% formaldehyde, 1.25% glutaraldehyde in 0.1moll1 Fig.1. Comparison of fast-start behavior in T. nigroviridis and D.
phosphate buffer, pH7.4 (PB)]. Brains were removed, post-fixed holocanthus. Photographs of (A) T. nigroviridis (size range: 3.95cm) and
(B) D. holocanthus (size range: 5.88.1cm). Silhouettes of (C) T.
for 4h, equilibrated in 30% sucrose in PB, embedded in OCT nigroviridis and (D) D. holocanthus traced from high-speed videos of
(Sakura Finetek, Torrance, CA, USA), frozen, and sectioned at representative responses to startling acoustic stimulus presentation, as
50m using a cryostat. Brains were sectioned in horizontal, sagittal viewed from above. Silhouettes are at 2ms intervals. The onset of the
or transverse planes. Free-floating sections were then processed to stimulus is indicated by an arrowhead. The asterisk indicates the latency of
detect BDA using a Vector ABC kit (Vector Laboratories, the response, determined by the first occurrence of lateral head and/or tail
Burlingame, CA, USA) and DAB as a chromagen with nickel movement. Scale bars, 2cm.
intensification. Sections were processed as follows: 10min in 3%
hydrogen peroxide/10% methanol in PB, 5min in PB, 2 5min in in 0.15moll1 sodium cacodylate for 4h, washed in buffer,
PB with 0.2% Triton X-100 (PBT), 3h in ABC reagent in PBT, 2 dehydrated in ethanol, cleared in propylene oxide and embedded in
5min in PBT, 5min in PB, 3 5min in 0.05moll1 Tris, pH8.4, Epon. Serial transverse sections (12m) were mounted on slides,
15min in DAB presoak (0.4% DAB, 0.4% nickel ammonium sulfate stained with Toluidine Blue and observed with a conventional light
in Tris), DAB reaction (DAB presoak with 0.015% H2O2), then 3 microscope.
5min in cold 0.05moll1 Tris, pH8.4. Sections were then transferred
to 0.1moll1 PB, mounted on gelatin-subbed slides, counterstained RESULTS
with Cresyl Violet, dehydrated and a coverslip was placed on top. Behavior
All chemicals were from Sigma (St Louis, MO, USA) unless Six T. nigroviridis individuals were presented with abrupt tactile
otherwise indicated. and auditory stimuli in order to elicit startle responses. All fish
readily produced fast-starts to both types of stimuli. An example of
Silver stain and plastic sections a fish exhibiting a fast-start to the auditory stimulus is shown in
Two T. nigroviridis (3.9 and 4.1cm standard length) brains and one Fig.1. These responses were C-type fast-starts (C-starts), in which
D. holocanthus (8.1cm standard length) brain were prepared for the fish assumed a C-shape during the first stage of the response.
silver staining. Fish were anesthetized in 0.03% MS-222. Brains C-starts had a rapid onset of approximately 11ms after the auditory
were removed and immersed in AFA fixative (9 parts 80% ethanol, stimulus (Table1). Superficially similar fast-start behavior was
5 parts formalin, 5 parts glacial acetic acid), then dehydrated, cleared elicited following a tactile stimulus, although the latency of this
in methyl salicylate, embedded in paraffin, sectioned at 15m, and response could not be determined (see Materials and Methods).
stained with Morses modification of Bodians silver technique. Thus, T. nigroviridis exhibits robust fast-start behavior with a short
One T. nigroviridis (3.2cm standard length) and one D. latency that is suggestive of M-cell initiation (Eaton et al., 1977).
holocanthus (8cm standard length) brain were embedded in plastic Following stage 1, fish exhibited a propulsive tail stroke (stage 2)
for thin sectioning. Fish were initially anesthetized in 0.03% MS- followed by either gliding or a burst swim (stage 3). The first stage
222 and were then maintained in 0.012% MS-222 during perfusion consisted of a variable amount of turning (up to 180 deg.), evidenced
with 2.5% glutaraldehyde and 1% paraformaldehyde in 0.15moll1 by the large variation in the angle traveled during stage 1 (Table1).
sodium cacodylate (pH7.2). The brains were removed and placed Diodon holocanthus, by contrast, did not show robust fast-start
in fresh fixative for 12h, washed in buffer, post-fixed in 2% OsO4 responses. Six fish were tested with both tactile and auditory stimuli.
E F
of zebrafish and goldfish (Lee and Eaton, 1991; Lee et al., 1993). (Fig.4) and could be traced after crossing the midline to the
The M-cells were substantially larger than all other reticulospinal Mauthner cell body on the opposite side of the brain in Toluidine-
cells and exhibited a characteristic anatomy, including two large Blue-stained plastic sections (Fig.3). The axon cap did not have the
principal dendrites and large axons that crossed the midline (Figs2, features attributed to the composite axon cap of the goldfish and
3). M-cells were clearly distinguishable in silver-stained paraffin appears more similar to a simple axon cap (Bierman et al., 2009).
sections (data not shown) and Toluidine-Blue-stained plastic sections By contrast, we did not identify any obvious M-cell candidates
(Fig.3). The M-axons were readily identified in the spinal cord in D. holocanthus. In eight brains, we were unable to distinguish
any large retrogradely labeled neurons in the fourth segment of
reticulospinal cells at the caudal portion of the fifth motor nucleus
(Fig.2). Careful examination of sections of the spinal cord of D.
A holocanthus from brains that were BDA labeled and Nissl stained
(N8), silver stained (N1) or Toluidine-Blue stained (N1) did not
reveal any axons that were distinctly larger than other axons (Fig.4
and data not shown).
The classification of a putative M-cell based solely on anatomical
location or size would potentially miss a cell that was substantially
modified compared to other fish. Therefore, we expanded our search
to account for this possibility. If M-cells were reduced in size but
still present in a homologous position, we would nevertheless expect
to be able to identify them anatomically because M-cells are
characteristically more dorsal than other cells in the fourth
reticulospinal segment (see Fig.2). However, we did not see any
smaller, retrogradely labeled cells in this anatomical location that
fitted this description. We also evaluated whether large M-cells
might indeed be present in the homologous location but that they
B do not project far enough down the spinal cord to be labeled by a
spinal backfill. However, no large cell bodies were distinguishable
in this region in D. holocanthus in Nissl-stained sections, silver-
stained paraffin sections, or Toluidine-Blue-stained thin plastic
sections (data not shown).
DISCUSSION
Mauthner cells have a long history of study in comparative
neurobiology (Bierman et al., 2009; Stefanelli, 1980; Zottoli,
1978b), although little work has been devoted to studying the
behavioral consequences of naturally occurring M-cell loss (Hale,
2000). Our results provide the first correlation between the presence
or absence of M-cells and variable fast-start behavior in related
species. These findings support and expand classic studies of M-
C cell ablations in the laboratory (Eaton et al., 1982; Liu and Fetcho,
1999) by examining the behavioral consequences of M-cell loss in
an evolutionary context. To our knowledge, this work represents
the only published demonstration that naturally occurring M-cell
loss in adults is correlated with dramatically altered fast-start
behavior.
Using both retrograde labeling of axonal projections and
conventional anatomical techniques, we identified distinctive M-
cells in T. nigroviridis but were unable to distinguish M-cells in D.
holocanthus. We cannot entirely exclude the possibility that D.
holocanthus may have atypical M-cells that were missed by our
analysis, although we were careful to consider liberal definitions of
M-cells in our search. The M-cell was thought to be absent in adult
Xenopus laevis (Stefanelli, 1951) and freshwater eels (Zottoli,
Fig.3. M-cells and the crossing of the M-axons in T. nigroviridis. (AC) 1978b), but subsequently the M-cell was identified in both (Meredith
1m cross sections stained with Toluidine Blue. (A)M-cells can be seen on and Roberts, 1987; Will, 1991). Future work, including
either side of the midline (arrows). Dorsal is up and the midline is in the developmental analysis or the use of molecular markers (Flores et
center of the photograph. (B)Higher power micrograph of the left M-cell. al., 2008), will be required to definitively confirm that these fish
The beginning portion of the lateral dendrite can be seen to the left. The lack M-cells. Our current results suggest that D. holocanthus either
axon hillock and initial segment can be seen to the right. Dorsal is up and
the midline is toward the right. (C)The M-axons can be traced caudally
lacks M-cells or possesses highly atypical M-cells. Therefore, we
where they cross (decussation is designated by an arrow). Dorsal is up and can confidently conclude that we have identified the presence of
the midline of the spinal cord is in the center of the image. Scale bars, M-cells in T. nigroviridis and the absence of conserved M-cells in
50m. D. holocanthus.
A B
*
*
Fig.4. Comparison of spinally projecting axons in T. nigroviridis and D. holocanthus at the level of the obex. (A,B)1m cross sections of the fasciculus
longitudinalis medialis at the level of the obex, stained with Toluidine Blue, and shown in grayscale. (A)T. nigroviridis spinal cord in which the M-axons are
easily identified (asterisks). (B)D. holocanthus spinal cord in which the M-axons could not be identified. Note that no pair of axons stands out on the basis of
size or amount of myelination. Dorsal is up and the midline of the spinal cord is in the center of both images. The position of the midline is indicated with a
dashed line. Scale bar, 50m.
Differences in M-cell neuroanatomy are correlated with protracted stage 1 duration compared with T. nigroviridis, and the
distinctive fast-start behavior in these species. Tetraodon nigroviridis peak angular velocity of these responses was dramatically reduced.
performed robust C-type fast-start responses that were similar to In addition, D. holocanthus essentially did not move away from its
M-cell mediated C-starts from other teleosts, including goldfish initial starting position (i.e. no stage 2) and did not exhibit subsequent
(Foreman and Eaton, 1993). However, under identical stimulus stages of movement (i.e. no stage 3) following the weak C-start
conditions, D. holocanthus only rarely exhibited weak responses at other than slow relaxation of the C-shape. It is unclear that this
a substantially longer latency. Diodon holocanthus may have response would actually be adaptive for predator avoidance and
different stimulus requirements than T. nigroviridis or the whether it can be considered a true fast-start.
behavioral context of the animal before the stimulus is delivered These substantial kinematic differences suggest that additional
might have an influence on the responsiveness of this species (Zottoli factors beyond the lack of identifiable M-cells probably contribute
et al., 1995). However, in pilot experiments, these fish did not exhibit to the distinct behaviors of D. holocanthus and T. nigroviridis. Besides
any rapid trunk movement following presentation of a variety of latency differences, larval lumpfish do not show kinematic differences
acoustic, tactile and visual stimuli other than the weak C-start in their fast-start performance (Hale, 2000). Similarly, M-cell ablations
following an acoustic stimulus. Diodon holocanthus also did not alone do not affect angular, velocity, or distance measures in adult
exhibit fast-starts to a tactile stimulus in a previous study (Brainerd goldfish (Eaton et al., 1982; Zottoli et al., 1999). The altered
and Patek, 1998). kinematics of D. holocanthus fast-starts could result from additional
Several aspects of the behavior of D. holocanthus are reminiscent neuroanatomical and/or morphological factors. The reduced angle and
of the fast-start responses of other species of fish with missing M- protracted stage 1 duration in D. holocanthus suggests that there may
cells. Compared with T. nigroviridis, D. holocanthus exhibited a be reduced ipsilateral muscle activation (Domenici and Blake, 1997;
longer latency of movement onset and a reduced probability of Foreman and Eaton, 1993). In addition, D. holocanthus may lack
performing a C-start. Experimental M-cell ablations result in fast- contralateral muscular activity that typically facilitates a propulsive
start responses that occur at a longer latency (Eaton et al., 1982; tail stroke (Domenici and Blake, 1997). These differences imply that
Kohashi and Oda, 2008; Liu and Fetcho, 1999; Zottoli et al., 1999), other aspects of fast-start circuitry might differ between D.
and yield a reduced probability of generating a fast-start (Zottoli holocanthus and T. nigroviridis. In larval zebrafish, slight kinematic
et al., 1999). In addition, the delayed latency of onset is similar to differences in behavior that are apparent after M-cell ablations are
another species that naturally lacks M-cells (larval lumpfish: Hale, enhanced following additional ablations of the M-cell homologues
2000). Although consistent with missing M-cells, a delayed latency in adjacent segments (Liu and Fetcho, 1999). It will be interesting to
alone is not sufficient evidence to indicate the lack of an M-cell. determine if pufferfish possess the M-cell homologues that are found
The sea robin has small M-cells and small M-axons (which in both goldfish and zebrafish (Kimmel, 1982; Kohashi and Oda, 2008;
although small can nevertheless be distinguished, unlike D. Lee and Eaton, 1991; Lee et al., 1993), and whether these cells also
holocanthus) and exhibits delayed, weak C-starts at a reduced show anatomical differences between pufferfish species. In addition
probability compared with goldfish (S.J.Z., unpublished) (Zottoli to variation in neural circuitry, kinematic differences between D.
et al., 1992). holocanthus and T. nigroviridis fast-starts could result from
Aside from a reduced probability and longer latency, the behavior morphological differences, including musculature, body shape and
of D. holocanthus is substantially different from fast-start behavior body stiffness. The relative contribution of neural versus
that typically results from experimental M-cell ablations. Diodon morphological factors to altered fast-start responses can be clarified
holocanthus did not bend as deeply during the C-starts and had a in the future using electromyography.
Our findings raise interesting evolutionary questions inflation behavior, whereas D. holocanthus did so frequently
concerning both the loss of M-cells in D. holocanthus relative to during routine handling.
ancestral teleosts and the variation in M-cells and behavior Although at this stage we cannot pinpoint the key factors that
between the two pufferfish species. The absence of M-cells and contribute to the species differences, our findings provide a clear
robust fast-start behavior in D. holocanthus relative to other foothold into variation in M-cells and correlated fast-start behavior
teleosts could result from a relaxation of selection for the characters in the pufferfish phylogeny. In the future, we can take
maintenance of fast-start behavior in the presence of a multitude advantage of the diversity within pufferfish to examine the
of pufferfish-specific anti-predator adaptations. In addition, the evolutionary history of and ultimately the mechanisms contributing
morphological changes that are associated with the gain of to the gain or loss of fast-start behavior and circuitry.
inflation include loss of hypaxial musculature (Brainerd, 1994),
which may preclude a robust fast-start, leading to a loss of M-
ACKNOWLEDGEMENTS
cells. The outcome of relaxed selection could vary among species,
We are grateful to Janice Simmons, Scott Lindell and Shaun McCann for advice
which might explain the differences in M-cell presence between and assistance with fish husbandry; to Gwyneth Card, Michael Dickinson and
T. nigroviridis and D. holocanthus. Jason Ritt for assistance with high-speed cameras; and to Tina Wong, Nancy
Piatczyc and Farah Musani for technical assistance. This work was supported by
Alternatively, there may be direct selection for the loss of a a Grass Fellowship and a Helen Hay Whitney Postdoctoral Fellowship to A.K.G.,
fast-start response and a functional M-cell in D. holocanthus. There a Burroughs Wellcome Career Award in the Biomedical Sciences to C.L.P. and a
are obvious tradeoffs between performing a fast-start and Williams College Howard Hughes Medical Institute and Essel Foundation grant to
S.J.Z.
generating inflation behavior. Both of these behaviors are used to
avoid predation, but a fish cannot perform both fast-start and
REFERENCES
inflation simultaneously because these behaviors require opposing Alfaro, M. E., Santini, F. and Brock, C. D. (2007). Do reefs drive diversification in
states of trunk muscle activation (contraction in fast-start and marine teleosts? Evidence from pufferfishes and their allies (Order
relaxation in inflation). At a neural level, there must be a decision Tetraodontiformes). Evolution 61, 2104-2126.
Bierman, H. S., Zottoli, S. J. and Hale, M. E. (2009). Evolution of the Mauthner axon
to generate one of the two behaviors in response to a perceived cap. Brain Behav. Evol. 73, 174-187.
threat. If inflation is consistently a better strategy for D. Brainerd, E. L. (1994). Pufferfish inflation: functional morphology of postcranial
structures in Diodon holocanthus (Tetraodontiformes). J. Morphol. 220, 243-261.
holocanthus, the chance of making the wrong choice (the fast- Brainerd, E. L. and Patek, S. N. (1998). Vertebral column morphology, C-start
start) might be selected against. However, potential tradeoffs might curvature, and the evolution of mechanical defenses in Tetraodontiform fishes.
Copeia 4, 971-984.
be alleviated to some extent if these behaviors are utilized in Carr, C. E., Soares, D., Parameshwaran, S. and Perney, T. (2001). Evolution and
somewhat different stimulus contexts, for example, initial predator development of time coding circuits. Curr. Opin. Neurobiol. 11, 727-733.
Chen, C.-Y. and Chou, H.-N. (1998). Detection of tetrodotoxin by high performance
detection versus capture. Systematic field observations will be liquid chromatography in lined-moon shell and puffer fish. Acta Zool. Taiwanica 9,
required to assess the relative use of these behaviors in different 41-48.
DiDomenico, R., Nissanov, J. and Eaton, R. C. (1988). Lateralization and adaptation
species. of a continuously variable behavior following lesions of a retiuclospinal command
If the loss of M-cells in D. holocanthus is due to selection, T. neuron. Brain Res. 473, 15-28.
nigroviridis must have experienced a distinct selective regime. Domenici, P. and Blake, R. (1997). The kinematics and performance of fish fast-start
swimming. J. Exp. Biol. 200, 1165-1178.
These species are from two different pufferfish families, thought Eaton, R. C. and Hackett, J. T. (1984). The role of the Mauthner cell in fast-starts
to have diverged at least 50 million years ago (Alfaro et al., 2007), involving escape in teleost fishes. In Neural Mechanisms of Startle Behavior (ed. R.
C. Eaton), pp. 213-266. New York: Plenum Press.
and they exhibit multiple differences in morphology, coloration, Eaton, R. C., Bombardieri, R. A. and Meyer, D. L. (1977). The Mauthner-initiated
life history and habitat. At this point, data are lacking from a startle response in teleost fish. J. Exp. Biol. 66, 65-81.
Eaton, R. C., Lavender, W. A. and Wieland, C. M. (1981). Identification of Mauthner
sufficient number of species to determine whether the presence initiated response patterns in goldfish: evidence from simultaneous cinematography
or absence of M-cells represents the ancestral pufferfish state, so and electrophysiology. J. Comp. Physiol. A 144, 521-531.
Eaton, R. C., Lavender, W. A. and Wieland, C. M. (1982). Alternative neural
we will discuss the findings in regard to known differences between pathways initiate fast-start responses following lesions of the Mauthner neuron in
the two species. As described above, morphological differences goldfish. J. Comp. Physiol. A 145, 485-496.
Eaton, R. C., Lee, R. K. and Foreman, M. B. (2001). The Mauthner cell and other
between these species may impact fast-start performance, and identified neurons of the brainstem escape network of fish. Prog. Neurobiol. 63, 467-
thereby selection. In addition, because fast-start performance is 485.
Flores, C. E., Ene, S. and Pereda, A. E. (2008). An immunohistochemical marker for
important in predator escape (Eaton and Hackett, 1984; Walker goldfish Mauthner cells. J. Neurosci. Methods 175, 64-69.
et al., 2005), an obvious difference to consider between these Foreman, M. B. and Eaton, R. C. (1993). The direction change concept for
reticulospinal control of goldfish escape. J. Neurosci. 13, 4101-4113.
species is predation pressure. Unfortunately, little is known about Hale, M. E. (2000). Startle responses of fish without Mauthner neurons: Escape
the life histories of these species. Diodon holocanthus is a marine behavior of the lumpfish (Cyclopterus lumpus). Biol. Bull. 199, 180-182.
species, inhabiting mostly inshore and reef habitats throughout the Hale, M. E., Long, J. H., Jr, McHenry, M. J. and Westneat, M. W. (2002). Evolution
of behavior and neural control of the fast-start escape response. Evolution 56, 993-
tropics. T. nigroviridis is a smaller euryhaline species that resides 1007.
in estuaries and freshwater streams in the Indo-Pacific. Large bony Katz, P. S. and Harris-Warrick, R. M. (1999). The evolution of neuronal circuits
underlying species-specific behavior. Curr. Opin. Neurobiol. 9, 628-633.
fish and sharks are known predators of D. holocanthus (Leis, 2001; Kimmel, C. B. (1982). Reticulospinal and vestibulospinal neurons in the young larva of
Randall, 1967). Predators of T. nigroviridis are not well a teleost fish, Brachydanio rerio. In Progress in Brain Research: Anatomy of
Descending Pathways to the Spinal Cord, vol. 57 (ed. H. G. J. M. Kuypers and G. F.
characterized, but could include both birds and fish, given their Martin), pp. 1-24. New York: Elsevier Biomedical.
size and habitat. Divergence in several characteristics supports the Kohashi, T. and Oda, Y. (2008). Initiation of Mauthner- or non-Mauthner-mediated fast
escape evoked by different modes of sensory input. J. Neurosci. 28, 10641-10653.
idea that these species may experience different predation regimes. Korn, H. and Faber, D. S. (2005). The Mauthner cell half a century later: a
Although both species are known to contain tetrodotoxin (Chen neurobiological model for decision-making? Neuron 47, 13-28.
Lee, R. K. and Eaton, R. C. (1991). Identifiable reticulospinal neurons of the adult
and Chou, 1998; Mahmud et al., 2003), T. nigroviridis has a zebrafish, Brachydanio rerio. J. Comp. Neurol. 304, 34-52.
potentially aposematic coloration pattern (Fig.2), whereas D. Lee, R. K., Eaton, R. C. and Zottoli, S. J. (1993). Segmental arrangement of
reticulospinal neurons in the goldfish hindbrain. J. Comp. Neurol. 329, 539-556.
holocanthus exhibits mottled, disruptive coloration, most likely Leis, J. M. (2001). Family Diodontidae. In FAO Species Identification Guide for Fishery
for camouflage. In addition, our anecdotal laboratory observations Purposes, vol. Bony Fishes part 4 (ed. K. E. Carpenter and V. H. Niem), pp. 3958-
3965.
suggest that these fish employ inflation behavior to different Liu, K. S. and Fetcho, J. R. (1999). Laser ablations reveal functional relationships of
extents. Tetraodon nigroviridis was rarely observed to perform segmental hindbrain neurons in zebrafish. Neuron 23, 325-335.
Receptor (or sensor) An animal that never moved could possess a relatively only kind of input that a stationary animal would expe-
A sensory end organ that simple sensory system. It would need receptors to detect rience. Exafference contrasts with reafference, which
detects changes in the external environmental changes, and a nervous system capable refers to those inputs that inevitably result from an ani-
world or the internal viscera. of interpreting the information. The animal would be mals own movements. Reafference is action-dependent
Effector
limited by two main factors: the range of energy changes and thus would be absent in immobile animals.
An organ that becomes active it could detect and the sophistication of the subsequent Sensory receptors are indifferent to the cause of
in response to a nerve signal. analyses it could perform. Within these limits, its their activation, so in principle they would convey both
sensory receptors should reliably detect near (for exam- exafference and reafference equally, and downstream
Afferent
ple, involving physical contact) and distant (for example, processing would proceed identically for both. However,
A neuronal projection that
conveys information to a
involving light) environmental occurrences. von Holst and Mittelstaedt2 pointed out that this would
structure. The term is often Once an animal moves, however, the situation lead to a manifold sensory and interpretative problem:
used in reference to sensory changes drastically. Movements contribute to form- reafference would be confused for exafference. This
channels. ing a sense of space, as discussed by Poincar1, and to potential confusion would be particularly significant for
improved, discriminative sensory perception. But move- the special abilities of some animals. For example, bats
ments have a potential downside as well: they introduce emit sounds that reflect back to them, but the reaffer-
ambiguity about the source of sensory input. A newly ent echoes are mingled with exafferent noise from other
mobile animal would be unable to determine whether a sources (including nearly identical biosonar probes from
disturbance registered by its sensory receptors was the other bats). Furthermore, informational ambiguity is
result of a change in the environment or simply a conse- only one facet of the underlying problem. Motor action
quence of its own movement. As it walked, it might not can be quite violent to receptors, many of which are
know whether sudden activation of its skin receptors was located on or near effectors (for example, muscles), and
due to a predators paw or to an inanimate obstacle in so movements could desensitize the primary afferents of
Department of Neuroscience, its path, or whether detected changes in light and shade a particular sensory channel. The animal would thus be
Center for the Neural Basis
were caused by movements of external objects or by unresponsive to independently occurring inputs that
of Cognition, and Center
for Neuroscience at the movements of its own photosensory organs. followed jarring motor acts.
University of Pittsburgh, Because the sensory problems that are introduced
University of Pittsburgh, The cause of sensory input: self or other? by movements are potentially devastating, any ancestral
Pittsburgh, Pennsylvania The fundamental distinction concerning the origin species that failed to solve them would have faced an
15260, USA.
Correspondence to T.B.C.
of sensory input was discussed by von Holtz and evolutionary disadvantage. Those species that survived
e-mail: tbc6@pitt.edu Mittelstaedt2. They termed input that results from occur- seem to have overcome the problem with a remarkably
doi:10.1038/nrn2457 rences in the environment as exafference; this is the uniform mechanism: the animals nervous systems keep
Sensory processing stream track of their movement commands and inform the Holtz and Mittelstaedt were addressing in invertebrates
The series of neuronal areas sensory processing stream about movements that are immi- and for the general analysis of sensory processing that
that are involved in analysing nent. In the terminology of von Holtz and Mittelstaedt, takes place close to the motor output. However, it has
the information acquired by the animal sends an efference copy signal that is, a become apparent that the decussation from motor to sen-
sense organs.
copy of the efferent motor command issued to an effector sory areas might occur at any number of levels of motor
Efferent to the sensory pathway. Thus, the combination of the control, some of which are remote from the final effector
A neuronal projection that input received from the sensory receptors and the motor stage (FIG.1b). In studies on fish, Sperry3 coined the term
conveys information away from copy that indicates how a movement might influence corollary discharge (CD) to denote motor-related sig-
a structure. The term is often
the input signal allows a moving animal to resolve the nals that influence sensory processing, but his concep-
used when referring to motor
commands.
confusion in sensory input. tion was less specific as to where the branch from motor
FIGURE1a shows the basic circuit in which a sensor to sensory pathways should emerge. In this Review we
Decussation and a series of steps along a sensory pathway convey the compare motor-to-sensory circuits across different spe-
The point where an axon or a exafferent and reafferent sensory signals to a motor path- cies and different levels of the nervous system, and so we
pathway crosses another.
way that impinges on the muscles to produce the appro- use what seems to be the more general of the two terms:
Phylogeny priate movement. A branch from the motor pathway to corollary discharge. At a mechanistic level CD can adopt
The evolutionary development the sensory pathway provides the efference copy. The one of multiple forms depending on how it is used: it can
or history of a group of general strategy involves the coincident production of a facilitate, inhibit or otherwise modulate its target.
organisms, often depicted in
motor command destined for an effector and a motor- Our main goal in this Review is to compare corol-
family trees.
command copy destined for a sensory structure. On lary systems across the animal kingdom. What is evi-
receipt of the copy signal, the sensory structure adjusts dent from experiments that have been performed over
appropriately to minimize, eliminate or compensate for the past few decades is that multiple types of CD have
the sensory consequences of the movement. evolved, and that each is particularly well suited to the
The term efference copy implies an actual copy of the problems faced by the species. Two recent reviews have
motor command (the efference) that targets the muscles; summarized a few of them4,5. Here we provide a broader
this term seemed appropriate for the questions that von overview of the varieties of CD and place emphasis on
two themes. First, we focus on critical functional dif-
ferences, illustrated by the impact of CD on recipient
a Efference copy
Motor output sensory structures. Animals that occupy diverse niches
Action show differences in their use of CD that, we propose, can
Higher Lower Motor neurons Effector
be summarized with a functional taxonomy. Second, we
consider the general similarities in CD circuits among
animals. We find that CD circuits conform to a com-
Efference copy Reafference mon neuronal plan, with only minor modifications
in motor source and sensory termination. By classi-
fying each circuit we attempt not only to summarize
recent findings, but also to distill from them the core
Higher Lower Sensory neurons Sensor circuits for CD that they exemplify. To help illustrate
Exafference
Sensory channel our points, throughout this article we use FIG.1b and
its colour-coded conventions as a template onto which
each animals particular pathway is represented. We
b Corollary discharge close by discussing some implications of our current
Motor output
understanding of CD and considering several questions
Action
Higher Lower Motor neurons Effector that remain unanswered.
A taxonomy for CD
Reafference
As is made evident in this Review, CD lends itself to a
Corollary discharge taxonomic classification scheme (FIG.2). Overall, CD can
be dichotomized into lower- and higher-order functional
(rather than phylogenetic) categories based on its opera-
Higher Lower Sensory neurons Sensor tional impact on the nervous system. The operational
Exafference impact is the way in which the CD signal influences the
Sensory channel
recipient structure, with the aim of achieving sensori
Figure 1 | Efference copy versus corollary discharge. a | A schematic of a motor harmony. The lower- and higher-order categories
sensorimotor circuit composed of a sensory pathway (shownNature in orange) and| Neuroscience
Reviews a motor are subject to further subdivisions that represent the
pathway (shown in brown). Each pathway consists of a number of tiers that represent
more specific functions of the signal.
the complexity of the processing that has been performed and the distance from the
Lower-order CD signalling is used for functions such
periphery. A branch from the motor pathway to the sensory pathway (shown in blue)
provides the efference copy. b | Corollary discharge. Motor-to-sensory signals are not as reflex inhibition and sensory filtration, both of which
confined to exact copies that target early tiers of the sensory channel. They can arise are examples of the control of sensation by the CNS. This
from almost all levels of the motor pathway and can target any tier of the sensory type of CD serves a sentinel function by intervening at
processing stream. These signals are known as corollary discharges (shown as thick various points along a sensorimotor pathway to regulate
arrows). Schematic inspired by von Holst and Mittelstaedt2. the sensory information that enters the system (in the
Teleception movement. This example illustrates how CD signals that mechanosensitive hair cells of the lateral line system, a
Sensory reception that is originate from a lower motor-control area target sen- network of mechanoreceptors that line the cephalic and
specialized for the detection of sory neurons positioned at an early stage of the sensory trunk regions17,18 and detect disturbances in the water
distant external stimuli, such as processing stream (FIG.4c). In terms of function, this column19. As with the cockroach and the crayfish, these
light, sound and smell.
circuit is almost identical to one that has been identified hairs are stimulated by locomotion. To avoid reafferent
in the cockroach, another organism with a behaviourally saturation, CDs of swimming commands directly inhibit
important hair-cell/escape-response circuit16. In both of the lateral-line hair cells20. In this case the CD signals act
these animals, CD signals prevent reafferent saturation directly on a sensor.
and inappropriate escape responses. The preceding examples illustrate how CD of loco-
Fish face similar problems to the preceding inverte- motion modulates pathways that are associated with
brate examples. The dogfish, for example, swims by gen- contact and teleceptive mechanosensation. Next we assess
erating rhythmic sinuous movements of its torso. These the role of CD in regulating sensory pathways associated
movements induce water turbulence that displaces the with two other teleceptive senses: audition and vision.
a b
Withdrawal Withdrawal
muscles muscles
Oral veil
Tactile
afferents
Front Back
Brain
WCNs WCNs
Interneurons
Feeding CPGs
Motor
neurons
Buccal
CDIs CDIs ganglion
Forward Backward
movement movement
Inhibitory Excitatory
c Motor output
Action
Higher Lower Motor neurons Effector
Corollary Reafference
discharge
Box 1 | Vestibular and proprioceptive signals that is optimal for lower-motor to sensory-neuron con-
tact), the cricket CDI is well suited to perform selective
Corollary discharge (CD) signals have a role in regulating the sensation of position, filtration operations that are coupled to motor activity.
velocity and acceleration that is mediated in part by the vestibular system and Intracellular recordings revealed that the CDI fires
proprioception. The vestibular system, which is composed of the semicircular canals
nearly simultaneously with wing motor-neuron bursts
and otolith organs of the inner ear, helps to detect head motion and plays a critical part
and exerts a phase-locked inhibitory influence on the
in maintaining balance. It also drives the vestibulocollic reflex. In response to a sudden
displacement of the head, this reflex quickly restabilizes the head through activation of interneurons of the auditory system25. By rhythmically
the neck musculature. There are times, however, when a vestibulocollic response is imposing and lifting a blockade in phase with the song
maladaptive, such as when an animal makes a willful attempt to move its head. It has cycle, the cricket can chirp and remain receptive to the
been demonstrated in monkeys that CD of head-movement commands suppresses auditory environment.
neurons of the vestibular nucleus during voluntary head movement, thus preventing Marmosets encounter the same problem as crickets:
the inappropriate reflex response80. These CDs permit differential processing of active in principle, the sounds that they make should affect their
and passive head movements by the vestibular nuclei, another case of CD acting in the hearing26. A protective mechanism is observable in the
service of reflex inhibition. marmoset primary auditory cortex, where many neurons
Proprioceptors are found throughout the bodies of invertebrate and vertebrate
are suppressed during self-vocalizations27,28. Suppression
organisms and provide continual information about body position and orientation.
begins ~200 ms before vocalization and continues for
These receptors operate in static or dynamic contexts, signalling when parts of the
body are at rest or are moving from one configuration to the next. It is in the latter its duration. Both the predictive nature of the effect and
context that CD has a role. Inputs from proprioceptors must be regulated during active the dependence on self-vocalization implicate CD as the
movement81 for two main reasons. If the proprioceptors are too sensitive, they could be cause of the suppression. In primates there are reciprocal
triggered by self-movements and initiate inappropriate reflexes. If they are not connections between the primary auditory cortex and
sensitive enough, they could fail to protect the limb (for example, if an unexpected motor regions of the cerebral cortex2931, and activity
obstacle is struck). One type of proprioceptor, the muscle spindle, is regulated directly in the frontal cortex can precede vocalization by up to
through projections from gamma motor neurons that receive CD of the movement 1s (ref. 32). Because the suppression begins ~200 ms
commands sent to muscle-innervating alpha motor neurons82. More generally, CD before the vocalization, the effect is most likely medi-
signals influence proprioceptive processing at multiple points in the neuraxis, ranging
ated by direct, rather than indirect, inhibition from one
from the spinal cord83 to the cortex84.
of these areas. As such, this could be a case in which
CD interconnects motor and sensory areas that occupy
comparable tiers of a sensorimotor pathway.
Sound-producing and -receiving organs allow As with the sense of hearing, the sense of vision could
organisms to exploit the physical characteristics of the theoretically be severely compromised by an animals
surrounding environment by inducing and detecting behaviour. However, audition and vision differ with
pressure fluctuations. Among the many purposes of respect to the nature of the motor interference. In audi-
sound for animals is communication between conspe- tion, problems arise from reafferential acoustic noise.
cifics21,22. A serious problem that is faced by organisms In vision, problems arise not from autogenic energy
that are equipped with organs of sound production, (energy originating from the animal itself for example,
however, is the substantial input that their sonic emis- the sound that results from a vocalization) but from the
sions impart to their auditory sensors. Following a sonic actions of the animal. If the eyes or the head move,
event, the auditory circuits could be overwhelmed; this the retinas go along for the ride. The retinas contain
would render the animal temporarily deaf to indepen sheets of photoreceptors that are passively responsive
dently incoming sounds. CD mechanisms devoted to to patterns of impinging photons. The moving of the
sensory filtration have evolved concurrently with the retinas called a gaze shift enables an animal to
motor and sensory systems that are involved in acoustic sample a new part of visual space but, as we shall see,
communication. The solution for this problem seems every gaze shift has its costs.
to be ubiquitous, as evidenced by the examples of the A gaze shift can be achieved with a saccadic (quick
cricket and the marmoset. Despite these animals occu- and discrete) eye movement and/or with a saccadic
pying very different niches, they both adopted the same head movement. Saccadic gaze shifts cause, at least in
solution: tight coordination between their auditory and principle, two serious reafferential problems. The first
vocal systems by phasic CD that minimizes auditory is that the speed of the movement could transiently blur
reafference. the image. The second is the general displacement of the
Crickets (see FIG.4b) chirp by rubbing their forewings visual scene from before to after the movement, which
together, a process that is known as stridulation23. They could cause the scene to seem like it is jumping from one
hear through a tympanate membrane on their forelegs place to another.
that is only millimeters from the site of stridulation24. The transient blur that occurs during saccades is
Tympanate membrane
A crickets auditory system must therefore deal with minimized by a process called saccadic suppression33.
A thin membrane that detects
sound (also known as the ear a steady barrage of acoustic signals generated by the In this process, a CD of the gaze shift lowers the gain of
drum). chirping. To prevent desensitization and ensure maximal target visual structures to reduce information transfer
attunement to environmental events, the cricket selec- between visual areas33. Saccadic suppression mecha-
Gain tively screens the signals that enter the system. A CDI nisms have been identified in a host of different spe-
An inputoutput ratio that
defines a neurons
that performs this sensory filtration operation has been cies, such as pigeons, chickens, locusts, monkeys and
responsiveness to incoming identified25 (FIG.4b). Positioned at the anatomical inter- cats3438. Although the suppressive systems of these
signals. face of the auditory and motor systems (an arrangement species differ with respect to the anatomical structures
a Sensory filtration b
MoGs ANs
Prothorax From
tympanum
CDIs
SGs LGs Mesothorax MNs
CPG
CDIs To wings
PADIs
SAs SIs
Metathorax
c
Motor output
Action
Higher Lower Motor neurons Effector
Corollary Reafference
discharge
Figure 4 | Corollary discharge used for sensory filtration. a | The illustration on the left is of a crayfish species,
Procambarus clarkii. The crayfish escapes from potential threats by producing a rapid tail-flipNature
response. Mechanoreceptors
Reviews | Neuroscience
lining its abdomen detect events in the water column and elicit the escape behaviour. Corollary discharge (CD) signals of
movement commands protect the afferents to the mechanosensory escape system from maladaptive activation and
desensitization. The schematic on the right depicts the circuitry that is involved in producing the crayfishs tail-flip
response. Mechanical information enters the system through sensory afferents (SAs) and reaches the lateral giants (LGs) by
way of sensory interneurons (SIs). The LGs communicate with segmental giants (SGs) and movement generators (MoGs)
that activate flexor muscles in the abdomen. CD interneurons (CDIs) are activated by the SGs and convey signals to
primary afferent depolarization interneurons (PADIs) that inhibit the SAs. This transient inhibition briefly silences the
mechanosensory pathway and prevents tail-flip-induced reafference from generating further tail flips. Part c shows a
circuit that represents an example of CD signals that originate from a lower motor cortical area and target sensory
neurons at an early stage of the processing stream. b | The illustration on the left is of a cricket species, Gryllus bimaculatus.
Crickets communicate with one another by chirping. Chirps are generated by rubbing their forewings together, a process
that is known as stridulation. CDs of the wingbeats phasically inhibit components of the auditory system and prevent
saturation and desensitization. The schematic of the cricket thorax on the right depicts how CD coordinates the song
and auditory systems. The activity of the central pattern generator (CPG) drives the wing motor neurons (MNs) that
produce song (musical notes). Signals are routed concurrently from the CPGs to the CDIs. Collaterals of the CDIs synapse
on to the axon terminals of auditory primary afferents and on to the cell bodies of auditory interneurons (ANs) in the
prothorax. CPG-induced CDI activity rhythmically imposes and lifts an inhibitory gate on the auditory systems and
restricts auditory traffic from the tympanum. Part c shows how, for the system in part b, the point of contact between the
motor and sensory systems is at the lower levels of the idealized sensorimotor circuit.
ing RF requires the appropriate CD to calculate the new responses of the ELL from effects of the electric-organ
RF. At least some of this CD arises from a pathway that CD by pharmacologically decoupling the electric-
ascends from the midbrain40,41 to the frontal eye field, a organ motor command centre from the electric organ52.
cortical structure that is involved in visual processing This preparation effectively silenced the electric organ:
and eye-movement control42. Interestingly, this same no electrical currents were produced. Nevertheless,
CD pathway assists in motor planning (discussed in the the motor command and CD were still produced by the
sensorimotor learning and planning section). motor command centre. ELL neurons responded to
Similar perceptual issues hold for animals for which the CD in a manner that was dependent on recent expe-
vision is a low priority. Rats explore the world largely rience. If no external signals (that is, electrical current
by whisking tactile objects of interest43. How does the reafference) preceded the motor command, no response
whisking rat construct a stable representation of the was elicited by the CD; however, if external electrical
world from the variable and volatile inputs that flow stimuli were paired with motor commands, responses
in from the vibrissae ? One clue to this puzzle was that were opposite in sign and equal in duration to the
the identification of neural activity in the rat barrel stimulus interval gradually emerged. If external stimuli
cortex that was modulated by the act of whisking and were removed, these responses were extinguished. The
was correlated with whisking amplitude44. The activity CD of the ampullary system is effectively a plastic photo
persisted during inactivation of the facial motor nerve negative of the expected reafference, and it varies as a
to rule out proprioception. Because the signal was of function of recent sensory experience. A modifiable
central origin, a CD seems to have been at work. Such a CD is evolutionarily advantageous because mormyrids
signal could provide motor context to the barrel cortex navigate diverse aquatic environments that differ subtly
and assist in the interpretation of vibrissal inputs, a step in conductivity and resistivity. Interestingly, the
towards constructing a stable representation. Motor mormyrid has two additional electroreceptor subtypes
pathways are known to target the barrel cortex from the (mormyromasts and knollenorgans) that are suscep-
motor cortex, the superior colliculus and other brain- tible to electric-organ discharge (EOD) feedback and,
stem areas45,46. Overall, the vibrissal system comprises a accordingly, compensatory CD systems are in place
web of sensorimotor loops that span much of the neur to accommodate them50,51. These systems differ from
axis. Any of these loops could convey CD signals to the ampullary system in that they require CDs that
the barrel cortex and thus mediate the effects observed amplify (mormyromast) or inhibit (knollenorgan) the
in the above study44 and the assembly of coherent respective electroreceptor during EOD. As the mormy-
information from whisking events. romast is specialized for electrolocation and the knollen
It is particularly intriguing to consider active sensing organ is specialized for conspecific communication,
systems (those in which energy generated by an animal CDs for amplification and suppression, respectively, are
is used to probe the environment). Animals that are appropriate. As the command centre and the ELL
equipped with these exotic signalling modes exercise are located in the hindbrain, the CD is shown in FIG.5Ab
considerable control over the properties of the emitted as linking lower motor and lower sensory levels.
signal (most of which are species- and modality-specific) Outside of aquatic environments, electrical probes
and rely on complex predictive mechanisms for carrying have little utility. However, air is an excellent carrier
out detailed analysis of the echo (the returned carrier of acoustic probes, a fact that is exploited by bats (see
signal)47. Two common modes of active sensing are used FIG.5B). Bats explore the environment by emitting high-
by fish and bats, and CD is essential in both. intensity, high-frequency beams of sound and then com-
We first consider the mormyrid, a weakly electric paring spatiotemporal aspects of the returning echo with
fish (see FIG.5Aa). The mormyrid uses self-generated the emitted sound53,54. Like the mormyrid, this active
electrical currents for communication and object iden- process of emission and comparison enables the bat to
tification. These currents flow from the mormyrids assemble an image of the external world5557 and thus
electric organ into the surrounding water column and successfully navigate and predate.
are analysed on their return by arrays of electroreceptors. Neurons that are sensitive to specific emissionecho
A number of neural processes have evolved that allow intervals have been identified in the inferior colliculus
the mormyrid to both extract the maximum information (FIG.5Ba) and, together with CD, have a key role in the
content from the return signal and avoid interference neural circuitry for auditory-scene analysis54,58. The
arising from reafference48. echo-sensitive neurons are thought to be primed by CDs
Mormyrids sense the electromagnetic environment of vocalizations that open time windows of echo analy-
using electroreceptors (FIG.5Aa) . Primary afferents sis. If the echo returns within a specific time frame, then
from these electroreceptors converge on layers of a neurons that have been primed for analysis within that
Whisking
medullary structure known as the electrosensory lat- particular window will respond optimally and convey
The act of tactile exploration in
which a whisker is rhythmically eral line lobe (ELL)49. Separate mechanisms involving information to perceptual structures for further analysis.
swept across an object. CD accommodate and protect the highly variegated Such signals convey information that is related to the
electroreceptor systems50,51. For the ampullary system, distance of approaching obstacles and the dimensions
Vibrissae an adaptable CD mechanism was found to be at work52. of environmental objects. These signals could arise from
Specialized long hairs located
near the mouth of most
Each time a motor command exits the electric-organ cortical or subcortical areas, so in FIG. 5Bb we represent
mammals that are used for command centre, a CD of the motor command also the putative CD signals as emerging from both higher
tactile exploration. exits, targeting the ELL. One study dissociated reafferent and lower motor levels.
Aa Mormyrid brain
Cerebrum
ELL
Motor Tectum
EOCD
command
Electric
organ Command
nucleus
Electrosensory
input
Ab Motor output
Action
Higher Lower Motor neurons Effector
Corollary
discharge Reafference
Ba
Priming Higher-order
Vocalization
corollary perceptual
centre
discharge centre
Vocalization Emissionecho
motor Inferior interval-specific
command colliculus auditory signals
Echo-sensitive
neurons
Vocalization Echo
Bb Motor output
Action
Higher Lower Motor neurons Effector
Corollary
discharge Reafference
Figure 5 | Corollary discharge used for sensory analysis and stability. Aa | The illustration on the
Nature left is|of
Reviews the
Neuroscience
mormyrid species Gnathonemus petersii. The mormyrid generates electrical signals to probe the aquatic environment.
Multiple types of corollary discharge (CD) of electrolocation commands allow the fish to gate, amplify or predict the
return signal. The schematic of the mormyrid brain on the right illustrates pathways of the electrosensory system. Electric
organ CD (EOCD; shown in blue) from the electric-organ motor command centre (which generates the motor command,
shown in purple) reaches cellular networks of the electrosensory lateral line lobe (ELL), following which a host of
interactions with electrosensory input (shown in orange) occur. Ab | The CD circuit that connects the command centre
and the ELL is represented as a link between lower motor and lower sensory levels. Ba | The illustration on the left is of the
bat species Rhinolophus rouxi. During high-speed flight, this bat uses sound to hunt. It compares a CD of the sonic probe to
the measured echo to interpret the acoustic input. The schematic on the right shows how CD is used in this system. CD
(shown in blue) represents the efferent motor command and innervates the inferior colliculus, where it is compared with
the echo input. Differences between the CD and the input (shown as dashed lines) are analysed by higher-order centres
to estimate the size, location and speed of the object that caused the echo. Bb | The CD signals could arise from any
number of subcortical and cortical structures, corresponding to multiple pathways emerging from both higher and lower
motor levels.
In summary, these examples illustrate the diversity as crystallization, and this is the phase in which mature
of sensory systems that are assisted by higher-order CD. song emerges. CD seems to have a key role during the
In all of these different species, CDs aid the construction second phase of vocal learning68,69. The bird is thought
of an internal representation of the environment. We to fine-tune its song during this phase by comparing
now consider motor planning and learning, the other auditory feedback of its own vocalizations with copies
category of higher-order CD, in the saccading primate of tutors songs that are stored in memory. The errors
and vocalizing bird. between the memory and the actual feedback are cor-
rected through an iterative cycle of vocal production,
Sensorimotor learning and planning. Among the most followed by adjustment of the motor programme. The
stereotyped movements generated by primates are sac- bird has to keep track of its own motor commands in
cadic eye movements. As a primate looks around a scene, order to properly adjust and calibrate its subsequent
it must continuously update its internal record of the commands. This is where CD plays a part. Intricate
current saccade to facilitate planning the next. CD has feedback loops that course throughout the avian song
repeatedly been implicated in this process5964. system are thought to convey the auditory, memory and
Electrophysiological recordings from a transiently CD signals70,71 (FIG.6Ba). The feedback loops execute a
inactivated circuit in the brain have shed some light delay function by increasing the distance over which
on how CD is used for saccadic sequencing in rhesus CD that is issued from vocal centres travels before it
monkeys62,63,65 (FIG.6Aa). Neurons that connect the mid- collides with auditory feedback72,73. Evidence for this
brain to the prefrontal cortex transmit bursts of action CD function was recently found in neurons in the high
potentials that encode saccadic parameters. When vocal centre (HVC) that fired with the same latency
this pathway was inactivated by muscimol, a GABAA irrespective of whether the bird was actively singing or
(-aminobutyric acid receptor type A) agonist, behav- passively listening74. As the HVC is a premotor structure,
ioural deficits in the simplest saccadic sequencing task the neurons would be expected to fire earlier for singing
were observed. In this two-step task, the monkey was than for listening. The comparable latencies that were
required to saccade to the locations of two rapidly flashed found for singing and passive listening imply that there is
sequential targets. This particular task was chosen a delay function performed by a CD. Because a major site
because it required the monkey to retrace the flash of comparison is the HVC, which is a higher brain struc-
sequence without the aid of sensory feedback. The monkey ture of the auditory forebrain, in FIG. 6Bb we represent
had only the internal feedback provided by CD to inform the CD as a signal between higher levels.
it of its current eye position. If CD of the first eye move- In these examples, CD is used for the higher-level
ments were impaired by the inactivation, then the mon- functions of planning and learning actions. Accordingly,
key would make incorrect second saccades owing to the on exiting the motor domain, the CD signal impinges on
absence of the feedback signal. The monkey would be higher-level structures that are highly sensory and/or
oblivious to the execution of the first saccade and would executive in nature. As a result, appropriate behaviours
perform the second saccade as if it had never made the can be prepared for the future (planning) and modified
first. The results conformed to this prediction, revealing based on the lessons of the past (learning).
a loss of CD. This is an example in which CD emerges
from a lower-level motor structure and ascends the Conclusions and future directions
neuraxis to impinge on a higher-level executive centre We have reviewed the operational impact of movement
(FIG.6Ab). As discussed above and in BOX2, this pathway on sensation, and the ways in which nervous systems
also has a sensory analysis and stability function. use CD signals to deal with movement-induced sen-
Vocalizations in birds are another prime example sory problems. CD systems have evolved in tandem
of action sequencing. Singing birds are thought to use with the sensory systems of organisms that operate in
updated, internal records of current phonations to diverse environments. The CD and the sensory systems
generate subsequent phonations. Bilateral feedback are both shaped by the problems that are encountered
pathways have been identified in the singing finch that by each organism, resulting in many different uses
might convey such vital information66. These pathways for CD signals. We have summarized these manifest
ascend from midbrain and medullary vocal-control differences with a functional taxonomy. After decades
nuclei up to the forebrain song system. Although their of experimentation, we now know a considerable
functional role in CD has not been tested directly using amount about these CD circuits in both invertebrate
causal methods, they carry signals that are consistent and vertebrate organisms. In general, CD seems to be
with CDs of vocalization motor commands. necessary for the proper function of nearly all sensory
CD is especially important while vocalization is being systems, and it exists in several guises at multiple levels
learned. Male juvenile songbirds (see FIG.6Ba) exhibit a in diverse species (perhaps all species). Some common
stereotyped song that emerges gradually following sev- alities and subtle differences are apparent from the
eral developmental stages67. The first stage is a sensory reviewed examples.
phase in which the juvenile bird listens to the songs of At one level (the lower-order level) CD is a discrimi-
other adult birds known as tutors. The second stage con- natory mechanism that prevents maladaptive responses
sists of sensorimotor training in which the bird begins and sensory saturation by restricting or filtering infor-
to sing spontaneously with syllables of varying intensity mation. Functions such as reflex inhibition and sensory
and motifs of high variability. The final phase is known filtration are tightly controlled by this mode of signalling,
Aa Macaque brain
Visual input
FEF
CD
MD
SC
To saccade-
generating
circuitry
Ab Motor output
Action
Higher Lower Motor neurons Effector
Corollary
discharge Reafference
RA X
To syrinx
Bb Motor output
Action
Higher Lower Motor neurons Effector
Corollary
discharge Reafference
Figure 6 | Corollary discharge used for sensorimotor planning and learning. Aa | The illustration on the left is of the
macaque monkey species Macaca mulatta. The macaque monkey visually explores its arboreal environment with rapid
sequences of saccades. Corollary discharges (CDs) permit it to plan such sequences in rapid succession and enable it to
predict the visual outcome for purposes of perceptual stability. The schematic of the macaqueNature Reviews
brain on | Neuroscience
the right illustrates
the course of the CD (shown in blue): it ascends from the superior colliculus (SC) to the frontal eye field (FEF) by way of the
medial dorsal nucleus of the thalamus (MD). Ab | This pathway is an example in which CD emerges from a lower-level
motor area and targets a higher-level sensory area. Ba | The illustration on the left is of the songbird Poephila guttata (a
finch species). The developing male finch progresses through a series of song-learning stages that conclude with the
appearance of a mature, fully formed song. The schematic of the finch brain on the right depicts the circuitry and nuclei of
the avian song-learning system. Intricate feedback loops that reside in the finch forebrain are involved in the song-
learning process. CD has been proposed to course through some of these pathways and enable sensorimotor comparisons
to occur within appropriate temporal windows. Bb | The major site of comparison is proposed to reside in the forebrain;
this CD pathway would correspond to contact between higher motor and sensory levels. DLM, medial nucleus of the
dorsolateral thalamus; HVC, high vocal centre; L, field L; LMAN, lateral magnocellular nucleus of the anterior neostriatum;
nXIIts, tracheo-syringeal portion of the hypoglossal nerve nucleus; RA, robust nucleus of the archistriatum; Uva, uvaeform
nucleus of the thalamus; X, area X.
which is a type of access control. Computationally, similar (that is, they were all lower-motor-area-to-
the information content of these CD types is probably sensory-area neurons). Although the CD signals of the
minimal, as they generally implement a gain mechanism mormyrid and the bat, two sensory analysis/stability
that modulates a reflex or gates the inflow of sensory cases, emerged from differing motor tiers, their sen-
information at the periphery. As such, these signals seem sory-termination level was comparable (lower sensory
to be one-dimensional in nature: time is the most crucial areas). Similarly, the two illustrative sensorimotor spe-
variable. This point is illustrated best in the nematode cies (a bird and a monkey) had different sites of motor
and gastropod examples of reflex inhibition. Here, CD emergence but common points of sensory termination
intervenes at the precise time of the motor act to prevent (both terminated at higher sensory areas). An implicit
an antagonistic reflex response. Timing is important too assumption in this comparison is that the sensory target
in the sensory-filtration examples. In the cricket, for occupies increasingly higher tiers as one ascends from
instance, CD is synchronized rhythmically with singing lower-order CD through the stages of higher-order CD
to protect the auditory system. In short, each subtype of (compare the nematode with the songbird). This illus-
lower-order signal provides information about when the trates the important point that although Sperrys original
sensory input that is elicited from the motor act should conception of CD matches the general flow of informa-
occur. In other words, it is not so much what the signal tion from motor systems to sensory systems throughout
says that matters, but when it is said. the animal kingdom, it seems to be inappropriately sim-
We explained that lower-order CD circuits are illus- plistic to use a single term to describe the signal. There
trated best in invertebrates. The main characteristic of is no single type of CD rather, there are numerous
these circuits is the presence of a single interneuron, the subtypes that correspond both to anatomical levels of
CDI, which exerts inhibitory actions on various cellular the source and the target and, as we have emphasized,
components and controls interactions between motor to functional utilities.
and sensory circuitries through its suppressive influ- As a next step, one could break down the taxonomic
ences. CDIs have been identified in Pleurobranchaea, subtypes even further. One way to do this would be to
crayfish and crickets. Suppressive influences by an consider underlying mechanisms. Not all instantiations
inhibitory CD are also found in vertebrates, and it of CD are equal in their actions: CD signals can excite,
is tempting to speculate that in some species these inhibit or modulate their targets. Although finer-grained
operations could also be carried out by a type of CDI. elaborations of the taxonomy would be beneficial, the
CD is also present in the animal kingdom as a higher- point of the present Review was to synthesize all of
order mechanism that mediates sophisticated predictive the available data into what we see as the most essential
computations. This type of CD underwrites complex functional taxonomy.
behavioural phenotypes and cognitive operations, such We are still at an early age of CD research. Future
as motor sequencing, sensorimotor learning and per- studies should search for signs of CD in sensory domains
ceptual stabilization. Higher-order CD operates in the that have been neglected, such as olfaction and gusta-
context of internal feedback circuitry, which involves tion. There is evidence that components of the olfactory
multiple components and spans various levels of the neur system are modulated by motor signals during the act of
axis. Unlike CD that is involved with the central control sniffing75. Are these motor signals CDs? When animals
of sensation, which is almost solely time-modulated, masticate, do CDs modulate components of the gusta-
CD that is involved with the central control of action tory system and thereby enhance taste discrimination?
and perception carries information that represents a We should also further test the idea that aspects of cogni-
number of variables. In the monkey, for example, CD tion, such as thinking and decision making, use a type of
in the visuosaccadic system encodes spatial as well as CD for cohesion of self-identity76,77.
temporal information about imminent saccades. The At the systems level, work should be devoted to
songbirds CD carries rich information too, particularly understanding inter-areal circuits that mediate CD in the
about song structure. In general, higher-order CDs are behaving animal. To date, only a few large-scale circuit-
multidimensional and encode more parameters than just level investigations of CD that combined behaviour with
time. As we have described, the exact content and func- physiology have been performed38,40. We know quite a bit
tional role of the signal (whether the CD is the prediction about the anatomical connections that probably convey
or whether it enables the recipient substrate to generate CD for example, a motor area A that projects to a
the prediction) varies from species to species and from sensory area B. But, until we introduce electrodes and
system to system. But, in the end, what is said is just as record from these circuits during self-movements of the
important as when it is said. This architecture seems to subjects, we can only speculate about their role in CD
be exclusive to higher vertebrates and might represent a function.
later phylogenetic advancement that is particular to the At the more reductionist level, researchers should
demands of vertebrate life. attempt to isolate CD mechanisms at ever-finer reso-
In all cases the corollary systems of remarkably lution, towards the cellular and molecular levels. The
diverse species could be easily accommodated into a recent sequencing of various animal genomes is par-
common template that illustrates their underlying simi- ticularly exciting, as CD circuits could be manipulated
larities. For the species that use reflex inhibition and sen- genetically at various stages of an animals life cycle. This
sory filtration, which are both cases of lower-order CD, could shed light on how CD circuits develop, an area that
the motor-to-sensory crossing points were schematically at present is poorly understood.
Another outstanding question concerns the range and covert perceptual changes might accompany these
of functional utilities of a given CD signal. A single behavioural changes. Eye movements in a two-step
CD circuit can mediate many different functions. One task, for example, change if CD about previous move-
example we reviewed was from the visuosaccadic sys- ments are impaired. Comparable experiments could be
tem of the primate40,62. In this system, a CD circuit has performed in other animals, such as rats that have been
a role in sensorimotor planning as well as in sensory trained to whisk an object in a particular sequence. Do
stability (compare FIG. 6A with BOX2). What is the con- such animals, with their CD circuits interrupted, expe-
nection between these two differing processes and the rience a sensory world that fractionates from stable to
CD circuit? Is there more than one CD signal coursing jumpy? How could we determine this? Although it is
through the circuit, or is there a single, multi-purpose challenging to assess what an animal is perceiving, in
CD signal? Does behavioural context play a part in rhesus monkeys, for example, it seems to be possible78,79.
determining what computation is implemented by the Now that the basic layout of many CD circuits has been
CD, and its functional interpretation by the recipient established, it will be exciting to manipulate the circuits
sensory structure? Similar CD complexities are likely while probing animals with careful psychophysical tests
to be present in other sensorimotor systems. Does the to determine how perception is altered.
whisking rat use the same CD signal for sequencing its Whatever is uncovered by future studies of CD,
whisker movements as for constructing a stable tactile it seems safe to say that we can currently recognize at
image of the whisked object? Or does it use different least one fundamental principle. All animals, from the
CD signals for each functional application? These humble nematode to the cognitively advanced primate,
perplexing questions remain to be addressed by fur- require the type of signalling that is enabled by CD. In
ther circuit-level interventions that attempt to disen- addition to the usual flow of information from sensory
tangle the multifarious uses nervous systems have for systems to motor systems, there is extensive signalling
CD signals. in the opposite direction by motor systems reporting
Finally, the ultimate goal is to discover how CD their activities to sensory structures. It is this coordina-
influences perception. Experiments thus far have shown tion between the two systems that makes it possible to
that inactivation of CD pathways can alter behaviour, analyse the world while moving within it.
1. Poincar, H. Science et methode (Flammarion, Paris, 14. Levi, R. & Camhi, J. Wind direction coding in the 28. Eliades, S.J. & Wang, X. Dynamics of auditory-vocal
1897). cockroach escape response: winner does not take all. interaction in monkey auditory cortex. Cereb. Cortex
2. Holst, E.V. & Mittelstaedt, H. The reafference Neuroscience 20, 38143821 (2000). 15, 15101523 (2005).
principle. Naturwissenschaften 37, 464467 (1950). 15. Krasne, F.B. & Bryan, J.S. Habituation: regulation 29. Alexander, G., Newman, J. & Symmes, D. Convergence
3. Sperry, R. Neural basis of the spontaneous optokinetic through presynaptic inhibition. Science 182, of prefrontal and acoustic inputs upon neurons in the
response produced by visual inversion. J. Comp. 590592 (1973). superior temporal gyrus of the awake squirrel monkey.
Physiol. Psychol. 43, 482489 (1950). 16. Delcomyn, F. Corollary discharge to cockroach giant Brain Res. 116, 334338 (1976).
References 2 and 3 are two groundbreaking papers interneurons. Nature 269, 160162 (1977). 30. Hackett, T., Stepniewska, I. & Kaas, J. Prefrontal
that were published independently and nearly 17. Kroese, A.B.A. & van Netten, S.M. in The connections of the parabelt auditory cortex in
simultaneously. They were the first to propose in a Mechanosensory Lateral Line: Neurobiology and macaque monkeys. Brain Res. 817, 4558 (1999).
rigorous manner, and with supporting experimental Evolution (eds Coombs, S., Gorner, P. & Munz, H.) 31. Morel, A. & Kaas, J. Subdivisons and connections of
evidence, that motortosensory feedback has a 265284 (Springer, New York, 1989). auditory cortex in owl monkeys. J. Comp. Neurol. 318,
critical role in regulating animal behaviour. 18. Coombs, S. & Montgomery, J.C. in Comparative 2763 (1992).
4. Cullen, K.E. Sensory signals during active versus Hearing: Fish and Amphibians (eds Fay, F.R. & 32. Gemba, H., Miki, N. & Sasaki, K. Cortical field
passive movement. Curr. Opin. Neurobiol. 14, Popper, A.N.) 319362 (Springer, New York, 1999). potentials preceding vocalization and influences of
698706 (2004). 19. Harris, G.G. & van Bergeijk, W.A. Evidence that the cerebellar hemispherectomy upon them in monkeys.
5. Poulet, J.F. & Hedwig, B. New insights into corollary lateral-line organ responds to near-field displacements Brain Res. 697, 143151 (1995).
discharges mediated by identified neural pathways. of sound sources in water. J.Acoust. Soc. Am. 34, 33. Ross, J., Morrone, M.C., Goldberg, M.E. & Burr,
Trends Neurosci. 30, 1421 (2007). 18311841 (1962). D.C. Changes in visual perception at the time of
6. White, J.G., Southgate, E., Thomson, J.N. & Brenner, S. 20. Roberts, B.L. & Russell, I.J. The activity of lateral line saccades. Trends Neurosci. 24, 113121 (2001).
The structure of the nervous system of the nematode efferent neurones in stationary and swimming dogfish. 34. Marin, G., Letelier, J.C. & Wallman, J. Saccade-related
Caenorhabditis elegans. Philos. Trans. R. Soc. Lond. B J. Exp. Biol. 57, 435448 (1972). responses of centrifugal neurons projecting to the
Biol. Sci. 314, 1340 (1986). 21. Michelsen, A. in The Evolutionary Biology of Hearing chicken retina. Exp. Brain Res. 82, 263270 (1990).
7. Rankin, C.H. Interactions between two antagonistic (eds Webster, D.B., Fay, F.R. & Popper, A.N.) 6177 35. Zaretsky, M. & Rowell, C.H.F. Saccadic suppression
reflexes in the nematode Caenorhabditis elegans. (Springer, New York, 1992). by corollary discharge in the locust. Nature 280,
J. Comp. Physiol. A 169, 5967 (1991). 22. Popper, A.N., Platt, P. & Edds, P. in The Evolutionary 583584 (1979).
8. Chalfie, M. etal. The neural circuit for touch sensitivity Biology of Hearing (eds Webster, D.B., Fay, F.R. & 36. Thiele, A., Henning, P., Kubischik, M. & Hoffmann,
in Caenorhabditis elegans. J. Neurosci. 5, 956964 Popper, A.N.) 4957 (Springer, New York, 1992). K.P. Neural mechanisms of saccadic suppression.
(1985). 23. Hedwig, B. Pulses, patterns, and paths: neurobiology Science 295, 24602462 (2002).
9. Sillar, K.T. & Roberts, A. A neuronal mechanism for of acoustic behavior in crickets. J. Comp. Physiol. A 37. Lee, D. & Malpeli, J.G. Effects of saccades on the
sensory gating during locomotion in a vertebrate. 192, 677689 (2006). activity of neurons in the cat lateral geniculate
Nature 331, 262265 (1988). 24. Hoy, R.R. & Robert, D. Tympanal hearing in insects. nucleus. J. Neurophysiol. 79, 922936 (1998).
10. Davis, W.J., Siegler, M.V.S. & Mpitsos, G.J. Annu. Rev. Entomol. 41, 433450 (1996). 38. Yang, Y., Cao, P., Yang, Y. & Wang, S.R. Corollary
Distributed neuronal oscillators and efference copy in 25. Poulet, J.F.A. & Hedwig, B. The cellular basis of a discharge circuits for saccadic modulation of the
the feeding system of pleurobranchaea. corollary discharge. Science 311, 518522 pigeon visual system. Nature Neurosci. 11, 595602
J. Neurophysiol. 36, 258274 (1973). (2006). (2008).
This was one of the first electrophysiological This report is one of a series of elegant studies 39. von Helmholtz, H. Helmholtzs Treatise on
studies to characterize CD signals at the cellular carried out by the authors in which they homed in Physiological Optics (Optical Society of America, New
level. on, and both anatomically and physiologically York, 1925).
11. Eaton, R. Neural Mechanisms of Startle Behavior identified, a CDI in the cricket auditory system. 40. Sommer, M.A. & Wurtz, R.H. Influence of the
(Plenum, New York, 1984). 26. Agamaite, J. & Wang, X. Quantitative classification of thalamus on spatial visual processing in frontal cortex.
12. Edwards, D., Heitler, W. & Krasne, F. Fifty years of a the vocal repertoire of the common marmoset Nature 444, 374377 (2006).
command neuron: the neurobiology of escape (Callithrix jacchus jacchus). Assoc. Res. Otolaryngol. 41. Sommer, M.A. & Wurtz, R.H. Brain circuits for the
behavior in the crayfish. Trends Neurosci. 22, Abstr. 20, 573 (1997). internal monitoring of movements. Annu. Rev.
153161 (1999). 27. Eliades, S.J. & Wang, X. sensory-motor interaction in Neurosci. 31, 317338 (2008).
13. Hatsopoulos, N., Gabbiani, F. & Laurent, G. the primate auditory cortex during self-initiated 42. Schall, J.D. On the role of frontal eye field in guiding
Elementary computation of object approach by a wide- vocalizations. J. Neurophysiol. 89, 21942207 attention and saccades. Vision Res. 44, 14531467
field visual neuron. Science 270, 10001003 (1995). (2003). (2004).
43. Kleinfeld, D., Ahissar, E. & Diamond, M.E. Active 58. Schuller, G. Vocalization influences auditory processing 73. Troyer, T.W. & Doupe, A.J. An associational model of
sensation: insights from the rodent vibrissa in collicular neurons of the CFFM bat, Rhinolophus birdsong sensorimotor learning II. Temporal
sensorimotor system. Curr. Opin. Neurobiol. 16, ferrumequinum. J. Comp. Physiol. A 132, 3946 hierarchies and the learning of song sequence.
435444 (2006). (1979). J. Neurophysiol. 84, 12241239 (2000).
44. Fee, M.S., Mitra, P.P. & Kleinfeld, D. Central versus 59. Bellebaum, C., Daum, I., Koch, B., Schwarz, M. & 74. Prather, J.F., Peters, S., Nowicki, S. & Mooney, R.
peripheral determinants of patterned spike activity in Hoffmann, K.P. The role of the human thalamus in Precise auditoryvocal mirroring in neurons for
rat vibrissa cortex during whisking. J. Neurophysiol. processing corollary discharge. Brain 128, learned vocal communication. Nature 451, 305310
1997, 11441149 (1997). 11391154 (2005). (2008).
45. Ahissar, E. & Kleinfeld, D. Closed-loop neuronal 60. Bellebaum, C., Hoffmann, K.P., Koch, B., Schwarz, M. 75. Mainland, J. & Sobel, N. The sniff is part of the
computations: focus on vibrissa somatosensation in & Daum, I. Altered processing of corollary discharge in olfactory percept. Chem. Senses 31, 181196 (2006).
rat. Cereb. Cortex 13, 5362 (2003). thalamic lesion patients. Eur. J. Neurosci. 24, 76. Feinberg, I. & Guazzelli, M. Schizophreniaa disorder
46. Kleinfeld, D., Berg, R.W. & OConner, S.M. 23752388 (2006). of the corollary discharge systems that integrate the
Anatomical loops and their electrical dynamics in 61. Guthrie, B.L., Porter, J.D. & Sparks, D.L. Corollary motor systems of thought with the sensory systems of
relation to whisking by rat. Somatosens. Mot. Res. 16, discharge provides accurate eye position information consciousness. Br. J. Psychiatry 174, 196204
6988 (1999). to the oculomotor system. Science 221, 11931195 (1999).
47. Nelson, M.E. & MacIver, M.A. Sensory acquisition in (1983). 77. Ford, J.M. etal. Neurophysiological evidence of
active sensing systems. J. Comp. Physiol. A 192, 62. Sommer, M.A. & Wurtz, R.H. A pathway in primate corollary discharge dysfunction in schizophrenia. Am.
573586 (2006). brain for internal monitoring of movements. Science J. Psychiatry 158, 20692071 (2001).
48. Caputi, A.A. Contributions of electric fish to the 296, 14801482 (2002). 78. Logothetis, N.K. Single units and conscious vision.
understanding of sensory processing by reafferent This was the first study to identify a CD pathway in Philos. Trans. R. Soc. Lond. B Biol. Sci. 353,
systems. J. Physiol. (Paris) 98, 8197 (2004). the primate brain. 18011818 (1998).
49. Meek, J., Grant, K. & Bell, C. Structural organization 63. Sommer, M.A. & Wurtz, R.H. What the brain stem 79. Parker, A.J. & Newsome, W.T. Sense and the single
of the mormyrid electrosensory lateral line lobe. tells the frontal cortex. II. Role of the SCMDFEF neuron: probing the physiology of perception. Annu.
J. Exp. Biol. 202, 12911300 (1999). pathway in corollary discharge. J. Neurophysiol. 91, Rev. Neurosci. 21, 227277 (1998).
50. Mohr, C., Roberts, P.D. & Bell, C.C. The 14031423 (2004). 80. Roy, J.E. & Cullen, K.E. Dissociating self-generated
mormyromast region of the mormyrid electrosensory 64. Tanaka, M. Inactivation of the central thalamus delays from passively applied head motion: neural
lobe. I. Responses to corollary discharge and self-timed saccades. Nature Neurosci. 9, 2022 (2006). mechanisms in the vestibular nuclei. J. Neurosci. 24,
electrosensory stimuli. J. Neurophysiol. 90, 65. Lynch, J., Hoover, J. & Strick, P. Input to the primate 21022111 (2004).
11931210 (2003). frontal eye field from the substantia nigra, superior 81. Rossignol, S., Dubuc, R. & Gossard, J.P. Dynamic
51. Bell, C.C. & Grant, K. Corollary discharge inhibition colliculus, and dentate nucleus demonstrated by sensorimotor interactions in locomotion. Physiol. Rev.
and preservation of temporal information in a sensory transneuronal transport. Exp. Brain Res. 100, 86, 89154 (2006).
nucleus of mormyrid electric fish. J. Neurosci. 9, 181186 (1994). 82. Matthews, P.B.C. Where does sherringtons muscular
10291044 (1989). 66. Striedter, G.F. & Vu, E.T. Bilateral feedback projections sense originate? Muscles, joints, corollary
52. Bell, C.C. An efference copy which is modified by to the forebrain in the premotor network for singing in discharges? Annu. Rev. Neurosci. 5, 189218 (1982).
reafferent input. Science 214, 450453 (1981). zebra finches. J. Neurobiol. 34, 2740 (1998). 83. Seki, K., Perlmutter, S.I. & Fetz, E.E. Sensory input to
This was a pioneering study that unveiled a plastic 67. Catchpole, D.K. & Slater, P.J.B. Bird Song: Biological primate spinal cord is presynaptically inhibited during
CD in the mormyrid that was modifiable by recent Themes and Variations (Cambridge Univ. Press, voluntary movement. Nature Neurosci. 6,
sensory experience. Cambridge, 1995). 13091316 (2003).
53. Moss, C.F. & Sinha, S.R. Neurobiology of 68. Brainard, M. & Doupe, A.J. Auditory feedback in 84. Voss, M., Ingram, J.N., Haggard, P. & Wolpert, D.M.
echolocation in bats. Curr. Opin. Neurobiol. 13, learning and maintenance of vocal behaviour. Nature Sensorimotor attenuation by central motor command
751758 (2003). Rev. Neurosci. 1, 3140 (2000). signals in the absence of movement. Nature Neurosci.
54. Neuweiler, G. Evolutionary aspects of bat 69. Margoliash, D. Evaluating theories of bird song 9, 2627 (2006).
echolocation. J. Comp. Physiol. A 189, 245256 learning: implications for future directions. J. Comp.
(2003). Physiol. A 188, 851866 (2002). Acknowledgements
55. Simmons, J.A. A view of the world through the bats 70. Margoliash, D. Functional organization of forebrain We thank R. H. Wurtz for comments on an earlier version of
ear: the formation of acoustic images in echolocation. pathways for song production and perception. the manuscript. Supported by the Alfred P. Sloan foundation
Cognition 33, 155199 (1989). J. Neurobiol. 33, 671693 (1997). and RO1-EY017592 to M.A.S.
56. Simmons, J.A. & Kick, S.A. Physiological mechanisms 71. Reiner, A., Yamamoto, K. & Karten, H. Organization
for spatial filtering and image enhancement in the and evolution of the avian forebrain. Anat. Rec. A
sonar of bats. Annu. Rev. Physiol. 46, 599614 Discov. Mol. Cell. Evol. Biol. 287, 10801102 (2005). FURTHER INFORMATION
(1984). 72. Troyer, T.W. & Doupe, A.J. An associational model of Marc Sommers homepage: http://cnup.neurobio.pitt.edu/
57. Simmons, J.A. etal. in Hearing by Bats (eds Fay, F.R. birdsong sensorimotor learning i. efference copy and people/peopleDetail.aspx?uid=381
& Popper, A.N.) 146190 (Springer, New York, the learning of song syllables. J. Neurophysiol. 84, All links are active in the online pdf
1995). 12041223 (2000).
R EV IE W
J. F. A. Poulet
Received: 8 December 2004 / Revised: 7 June 2005 / Accepted: 12 June 2005 / Published online: 25 October 2005
Springer-Verlag 2005
Abstract The romantic notion of crickets singing on a fused for external, or exaerent, sensory information of
warm summers evening is quickly dispelled when one the same modality, which could elicit an inappropriate
comes ear to ear with a stridulating male. Remarkably, behavioural response. Thus, the discrimination between
stridulating male crickets are able to hear sounds from self-generated and external sensory stimuli is a funda-
the environment despite generating a 100 db song mental problem in perception and a central question of
(Heiligenberg 1969; Jones and Dambach 1973). This sensory neuroscience.
review summarises recent work examining how they Philosophers and scientists over the centuries have
achieve this feat of sensory processing. While the proposed that responses to self-generated stimuli are
responsiveness of the crickets peripheral auditory sys- modulated by neural signals that feedforward from
tem (tympanic membrane, tympanic nerve, state of the motor to sensory networks during behaviour (Grusser
acoustic spiracle) is maintained during sound produc- 1986). In 1950 two papers furthered modern thinking
tion, central auditory neurons are inhibited by a feed- about these concepts and termed the feedforward signals
forward corollary discharge signal precisely timed to corollary discharges or eerence copies (Sperry
coincide with the auditory neurons maximum response 1950; von Holst and Mittelstaedt 1950). More recently,
to self-generated sound. In this way, the corollary dis- these terms have been rened so that motor to sensory
charge inhibition prevents desensitisation of the crickets discharges indicate corollary discharge signals; whereas
auditory pathway during sound production. an eerence copy is compared with the actual sensory
feedback and represents an accurate negative image of
Keywords Corollary discharge Eerence copy the reaerent information (McCloskey 1981; Bell 1984).
Stridulation Cricket Auditory Presynaptic Neurophysiological evidence for these mechanisms re-
inhibition quires recording neural activity during motor behaviour.
Despite this technical diculty, a growing body of evi-
dence suggests that feedforward signals are widespread
Feedforward signals in the CNS and have been recorded from a number of
sensory systems including visual (Zaretsky and Rowell
To gather sensory cues from the environment, animals 1979; Sommer and Wurtz 2002), vestibular (Roy and
have developed an array of highly sensitive sensory Cullen 2004), electrosensory (Bell 1989), mechanosen-
systems. These systems, however, not only respond to sory (Murphey and Palka 1974; Delcomyn 1977; El
environmental sensory information, but also to stimuli Manira et al. 1996; Blakemore et al. 1998; Li et al. 2002)
generated as a by-product of the animals own behav- and proprioceptive systems (Sillar and Skorupski 1986;
iour. In principle self-generated, or reaerent (von Holst Gossard et al. 1991; Wolf and Burrows 1995).
and Mittelstaedt 1950), information could desensitise
the animals own sensory pathway and/or also be con-
Auditory responsiveness during sound production
383
980
384
981
membranes and the activity of the whole tympanic nerve Huber 1982; Stumpner et al. 1995). (3) Ascending neu-
(Poulet and Hedwig 2001). ron 2 (AN2) also ascends to the brain, it responds best to
To examine the responsiveness of the tympanic higher frequencies (>10 kHz) and is inhibited, or only
membrane, we presented acoustic stimuli to stridulating weakly excited, by 4.5 kHz sound (Wohlers and Huber
crickets and measured the wing movements, sound 1978, 1982; Boyan 1980; Nolen and Hoy 1987), which
produced and any oscillations or displacement with a supports roles in courtship song recognition (Wohlers
laser vibrometer/interferometer. In a resting cricket, the and Huber 1982) and bat detection during ight (Nolen
tympanic membrane would oscillate with constant and Hoy 1983).
amplitude to 4.5 kHz, 90 dB SPL sound pulses. Occa- As expected from the whole nerve recordings, single
sionally, during ventilation, the tympanic membrane auditory aerent neurons responded with bursts of
was slowly displaced in synchrony with the abdominal spikes in response to each syllable (Fig. 2a). Addi-
pumping movements, as seen in other Orthoptera tionally prior to the start of some of the quieter chirps,
(Meyer and Hedwig 1995; Meyer and Elsner 1995). The the axonal arborisations of the aerents were depolar-
displacements were accompanied by a decrease in the ised (arrow, Fig. 2a). The interneurons ON1 (Fig. 2b)
amplitude of the sound-induced oscillations (Fig. 1b). and AN1 (Fig. 2c) responded with bursts of spikes to
Waxing the spiracle shut could produce the same eect. the crickets own song. However, as in the aerents,
Closing the acoustic spiracles could, therefore, control besides the excitation a second input was present in
the amplitude of reaerent auditory input. However, ON1 consisting of small hyperpolarising potentials
during stridulation, the spiracles remained open. (arrow, Fig. 2b). Unlike the other neurons, AN2 was
Therefore, the self-generated auditory signals had full inhibited during the chirps and rarely generated a spike
access to both sides of the tympanic membranes. (Fig. 2d).
During stridulation the tympanic membrane oscillated To characterise the inhibitory inputs, we removed
in response to each self-generated syllable (Fig. 1c). one of the wings and elicited silent stridulation. In the
Further analysis of the response of the tympanic absence of any self-generated or external sounds in the
membrane to computer generated sound pulses, aerent axon we recorded phasic groups of depolaris-
showed that there was no change in its responsiveness ing synaptic inputs termed primary aerent depolari-
during the chirp or chirp interval. Whole nerve sations or PADs (Fig. 3a). Each PAD started during
recordings of the tympanic nerve in the foreleg of the the closing wing movements of the chirp when sound
cricket revealed a similar result: bursts of activity were would normally have been generated. Recordings from
recorded in response to the crickets own song (Fig. 1d), ON1 during silent stridulation revealed clear groups of
and the response to externally generated acoustic IPSPs in phase with each chirp (Fig. 3b). As with the
stimuli presented were not modulated throughout the PADs they started during the closing wing movements
song. Thus, in contrast to most other acoustically of each syllable and therefore must be expressed syn-
communicating animals, the cricket does not take chronous with ON1s spiking response during singing.
advantage of a relatively simple mechanism to reduce Recordings from AN1 and AN2 showed small ampli-
the responsiveness of its peripheral auditory pathway tude hyperpolarising potentials during the chirps
during sound production. How then does the crickets (Fig. 3c, d). The large amplitude IPSPs recorded in
central auditory system copes with the mass of self- AN2 during sonorous singing must therefore have been
generated sensory input during stridulation? a response to reaerent input and correspond to the
4.5 kHz inhibition.
To test what eect the PADs and IPSPs have on
Central auditory responses during stridulation sound processing, we presented silently singing crickets
with a continuous series of sound pulses. In this way, it
The auditory aerents terminate in the auditory neuropil was possible to evoke an auditory response in these
in the anterior ventral prothoracic ganglion. Here they neurons and examine any modication during the PAD
forward their information to the dendritic branches of or IPSP. Auditory aerents clearly responded with a
auditory interneurons (Wohlers and Huber 1982). We train of spikes when presented with the sound pulses
went on to make intracellular recordings and stainings in (Fig. 4a). The frequency of spiking in the aerent cells
the auditory neuropil from the axonal branches of the was consistent throughout the chirp and chirp intervals.
auditory aerents and from the dendrites of the auditory However on closer inspection of the spike amplitude,
interneurons during stridulation. We recorded from those spikes coinciding with the PADs were reduced.
three identied interneurons: (1) the omega neuron 1 The reduction in spike height is a good indication that
(ON1) is an W-shaped, local auditory interneuron that is the PADs have a presynaptic inhibitory function (Bur-
broadly tuned but is most sensitive to the carrier fre- rows and Matheson 1994; Clarac and Cattaert 1996).
quency of male calling song (4.5 kHz) (Casaday and The PADs momentarily shunt the membrane resistance
Hoy 1977; Popov et al. 1978; Wohlers and Huber 1978); and cause a reduction in spike height that results in a
(2) ascending neuron 1 (AN1) is tightly tuned to 4.5 kHz smaller depolarisation of the axon terminal. Therefore,
sound stimuli but possesses an ascending axon that less neurotransmitter is released and the response in the
terminates in the brain (Boyan 1980; Wohlers and postsynaptic cell is reduced. In a resting cricket and
385
982
386
983
Fig. 5 Inhibitory inputs are present during ctive singing with all
motor and sensory neurons cut. a PADs occur in auditory aerent
neurons, b IPSPs occur in ON1 and low amplitude hyperpolarising
potentials are present in c AN1 and d AN2 during the chirps. These
inputs were therefore generated within the nervous system and the
result of a corollary discharge signal. For further details, see Figs. 1
and 2. Used with permission from Poulet and Hedwig (2003a)
387
984
388
985
El Manira A, Tegner J, Grillner S (1996) Locomotor-related pre- Muller-Preuss P, Ploog D (1981) Inhibition of auditory cortical
synaptic modulation of primary aerents in the lamprey. Eur J neurons during phonation. Brain Res 215:6176
Neurosci 9:696705 Murphey RK, Palka J (1974) Eerent control of cricket giant -
Gossard J-P, Cabelguen J-M, Rossignol S (1991) An intracellular bres. Nature 248:249251
study of muscle primary aerents during ctive locomotion in Narins PM (1992) Reduction of tympanic membrane displacement
the cat. J Neurophysiol 65:914926 during vocalization of the arboreal tree frog, Eleutherodactylus
Grusser O-J (1986) Interaction of eerenct and aerent signals in coqui. J Acoust Soc Am 91:35513557
visual perception a history of ideas and experimental para- Nolen TG, Hoy RR (1983) Initiation of behaviour by single neu-
digms. Acta Psychologica 63:321 rons: the role of behavioural context. Science 226:992994
Hedwig B (1986) On the role in stridulation of plurisegmental in- Nolen TG, Hoy RR (1987) Postsynaptic inhibition mediates high-
terneurons of the acridid grasshopper Omocestus viridulus L. II. frequency selectivity in the cricket Teleogryllus oceanicus:
Anatomy and physiology of ascending and T-shaped inter- implications for ight pyhonotaxis behaviour. J Neurosci
neurons. J Comp Physiol A 158:429444 7:20812096
Hedwig B (1990) Modulation of auditory responsiveness in strid- Numminen J, Salmelin R, Hari R (1999) Subjects own speech re-
ulating grasshoppers. J Comp Physiol A 167:847856 duces reactivity of the human auditory cortex. Neurosci Lett
Hedwig B (2000) Control of cricket stridulation by a command 265:119122
neuron: ecacy depends on the behavioural state. J Neuro- Otto D (1978) Anderungen von Gesangsparametern bei der Grille
physiol 83:712722 (Gryllus campestris L.) nach Injektion von Pharmaka ins
Hedwig B, Meyer J (1994) Auditory information processing in Gehirn. Verh Dt Zool Ges 245
stridulating grasshoppers: tympanic membrane vibrations and Pollack GS (1988) Selective attention in an insect auditory neuron.
neurophysiology. J Comp Physiol A 174:121131 J Neurosci 8:26352639
Heiligenberg W (1969) The eect of stimulus chirps on a crickets Popov AV, Markovich AM, Andjan AS (1978) Auditory inter-
chirping. Z Vergl Physiol 65:7097 neurons in the prothoracic ganglion of the cricket, Gryllus
Hennig RM, Weber T, Huber F, Kleindienst H-U, Moore TE, bimaculatus deGeer I. The large segmental auditory neuron
Popov AV (1994) Auditory threshold change in singing cicadas. (LSAN). J Comp Physiol A 126:183192
J Exp Biol 187:4555 Poulet JFA, Hedwig B (2001) Tympanic membrane oscillations and
Hill KG, Boyan GS (1976) Directional hearing in crickets. Nature auditory receptor activity in the stridulating cricket Gryllus
262:390391 bimaculatus. J Exp Biol 204:12811293
von Holst E, Mittelstaedt H (1950) Das Reaerenzprinzip: Wech- Poulet JFA, Hedwig B (2002) A corollary discharge maintains
selwirkungen zwischen Zentralnervensystem und Peripherie. auditory sensitivity during sound production. Nature 418:872
Naturwissenschaften 37:464476 876
Jones MDR, Dambach M (1973) Response to sound in crickets Poulet JFA, Hedwig B (2003a) Corollary discharge inhibition of
without tympanal organs (Gryllus campestris L.). J Comp ascending auditory information in the stridulating cricket.
Physiol A 87:8998 J Neurosci 23:47174725
Kirzinger A, Jurgens U (1991) Vocalization-correlated single-unit Poulet JFA, Hedwig B (2003b) A corollary discharge mechanism
activity in the brain stem of the monkey. Exp Brain Res 84:545 modulates central auditory processing in singing crickets.
560 J Neurophysiol 89:15281540
Kleindienst H-U, Wohlers DW, Larsen ON (1983) Tympanal Roy JE, Cullen KE (2004) Dissociating self-generated from pas-
membrane motion is necessary for hearing in crickets. J Exp sively applied head motion: neural mechanisms in the vestibular
Biol 151:397400 nuclei. J Neurosci 24:21022111
Larsen ON, Michelsen A (1978) Biophysics of the ensiferan ear III. Samson A-H, Pollack GS (2002) Encoding of sound localization
The cricket ear as a four input system. J Comp Physiol A cues by an identied auditory interneuron: eects of stimulus
123:217227 temporal pattern. J Neurophysiol 88:23222328
Larsen ON, Kleindienst K-U, Michelsen A (1989) Biophysical as- Schuller G (1979) Vocalization inuences auditory processing in
pects of sound reception. In: Huber F et al (eds) Cricket collicular neurons of the CF-FM-bat, Rhinolophus ferrumequi-
behaviour and neurobiology. Cornell University Press, Ithaca num. J Comp Physiol A 132:3946
London, pp 364390 Sillar KT, Skorupski P (1986) Central input to primary aerent
Li W-C, Soe SR, Roberts A (2002) Spinal inhibitory neurons that neurons in craysh, Pacifastacus leniusculus, is correlated with
modulate cutaneous sensory pathways during locomotion in a rhythmic motor output of thoracic ganglia. J Neurophysiol
simple vertebrate. J Neurosci 22:1092410934 55:678688
McCasland JS, Konishi M (1981) Interaction between auditory and Sommer MA, Wurtz RH (2002) A pathway in primate brain for
motor activities in an avian song control nucleus. Proc Natl internal monitoring of movements. Science 296:14801482
Acad Sci USA 78:78157819 Sperry RW (1950) Neural basis of the spontaneous optokinetic
McCloskey DI (1981) Corollary discharges: motor commands and response produced by visual inversion. J Comp Physiol Psych
perception. In: Brooks VB (ed) Handbook of physiology. The 43:482489
nervous system. Am Physiol Soc, Bethesda, pp 14151447 Stumpner A, Atkins G, Stout JF (1995) Processing of unilateral
Metzner W (1989) A possible neuronal basis for Doppler-shift com- and bilateral auditory inputs by the ON1 and L1 interneurons
pensation in echolocating horseshoe bats. Nature 341:529532 of the cricket Acheta domesticus and comparison to other
Metzner W (1993) An audio-vocal interface in echolocating cricket species. J Comp Physiol A 177:379388
horseshoe bats. J Neurosci 13:18991915 Suga N, Jen P (1975) Peripheral control of acoustic signals in the
Meyer J, Elsner N (1995) How respiration aects auditory sensi- auditory system of echolocating bats. J Exp Biol 62:277311
tivity in the grasshopper Chorthippus biguttulus (L.). J Comp Suga N, Schlegel P (1972) Neural attenuation of responses to
Physiol A 176:563573 emitted sounds in echolocating bats. Science 177:8284
Meyer J, Hedwig B (1995) The inuence of tracheal pressure Suga N, Shimozawa T (1974) Site of neural attenuation of re-
changes on the responses of the tympanal membrane and sponses to self-vocalized sounds in echolocating bats. Science
auditory receptors in the locust Locusta migratoria L. J Exp 183:12111213
Biol 198:13271339 Wenzel B, Hedwig B (1999) Neurochemical control of cricket
Michelsen A (1994) Directional hearing in crickets and other small stridulation revealed by pharmacological injections into the
animals. Fort der Zool 39:195207 brain. J Exp Biol 202:22032216
Michelsen A (1998) Biophysics of sound localization in insects. In: Wohlers DW, Huber F (1978) Intracellular recording and staining
Hoy RR et al (eds) Comparative hearing: insects. Springer, of cricket auditory interneurons (Gryllus campestris L., Gryllus
Berlin Heidelberg New York, pp 1862 bimaculatus DeGeer). J Comp Physiol A 127:1128
389
986
Wohlers DW, Huber F (1982) Processing of sound signals by six Wolf H, Helversen OV (1986) Switching-o of an auditory
types of neurons in the prothoracic ganglion of the cricket, interneuron during stridulation in the acridid grasshopper
Gryllus campestris L. J Comp Physiol A 146:161173 Chorthippus biguttulus L. J Comp Physiol A 158:861871
Wolf H, Burrows M (1995) Proprioceptive sensory neurons of a Zaretsky M, Rowell CHF (1979) Saccadic suppression by corollary
locust leg receive rhythmic presynaptic inhibition during discharge in the locust. Nature 280:583585
walking. J Neurosci 15:56235636
390
REPORTS
Fig. 4. Compressive strength of 11. H. Ishiguro, B. Rubinsky, Cryobiology 31, 483 (1994).
porous HAP scaffolds. Results 12. J. O. M. Karlsson, Science 296, 655 (2002).
13. G. Gay, M. A. Azouni, Cryst. Growth Des. 2, 135 (2002).
from literature [blue stars (22), 14. A. P. Jackson, J. F. V. Vincent, R. M. Turner, Proc. R. Soc.
red stars (23), inverted green Lond. Ser. B 234, 415 (1988).
triangles (24), black triangle 15. J. Aizenberg et al., Science 309, 275 (2005).
(25), blue circle (26), inverted 16. R. Z. Wang, Z. Suo, A. G. Evans, N. Yao, I. A. Aksay,
blue triangle (27), diamonds J. Mater. Res. 16, 2485 (2001).
(28), cross (29), and red circles 17. F. Song, A. K. Soh, Y. L. Bai, Biomaterials 24, 3623 (2003).
18. E. Saiz, R. M. Cannon, A. P. Tomsia, Acta Mater. 48, 4449
(30)] versus IT porous HAP (2000).
scaffolds. The typical pore sizes 19. L. L. Hench, J. M. Polak, Science 295, 1014 (2002).
of conventional porous HAP 20. T. Dutta Roy, J. L. Simon, J. L. Ricci, E. D. Rekow,
scaffolds are on the order of V. P. Thompson, J. R. Parsons, J. Biomed. Mater. Res. 66A,
100 to 800 mm in order to al- 283 (2003).
21. Y. H. An, in Mechanical Testing of Bone and the
low bone ingrowth. In the IT
Bone-Implant Interface (CRC Press, Boca Raton, FL,
materials exhibiting the greatest 2000), pp. 4163.
strength, the pores are typical- 22. A. Almirall et al., Biomaterials 25, 3671 (2004).
ly 20 by 200 mm wide and 23. R. P. del Real, J. G. C. Wolke, M. Vallet-Regi, J. A. Jansen,
several millimeters long; pre- Biomaterials 23, 3673 (2002).
vious studies have indicated 24. Y. Ota, T. Kasuga, Y. Abe, J. Am. Ceram. Soc. 80, 225
(1997).
A
been able to fabricate HAP-based composites
with stiffness (10 GPa), strength (150 MPa), stant flow of sensory information that is to forward a signal, or corollary discharge,
and work of fracture (220 J/m2) that match can update or fine-tune ongoing motor from motor to sensory regions during behavior
that of compact bone for an equivalent mineral/ activity (1) but can also desensitize the animal_s to counter the expected, self-generated sensory
organic content (around 60/40 vol %). own sensory pathways and/or be confused with feedback (2, 3). A role for corollary discharges
391
518 27 JANUARY 2006 VOL 311 SCIENCE www.sciencemag.org
REPORTS
392
www.sciencemag.org SCIENCE VOL 311 27 JANUARY 2006 519
REPORTS
The morphology of CDI highlights three and also with the timing of inhibitory inputs in ings from CDI_s dendrite and auditory neurons
major structural properties that are crucial for auditory neurons during singing (14). To test revealed the synaptic connectivity. About 60
its function. First, its cell body and extensive whether CDI is part of the singing CPG, we auditory afferent neurons extend from the
dendritic arborization is in the mesothoracic used 100-ms depolarizing current pulses to cricket_s ears in the forelegs into prothoracic
ganglion; therefore, it can receive synaptic input elicit spikes and measured any effect on the ganglion, where they terminate in a median-
from the singing central pattern generator singing motor pattern (Fig. 2C). Elicited bursts ventral auditory neuropil (17). Here they for-
(CPG). Second, the neuron has profuse axonal of spikes in CDI never had an effect on the ward excitatory information onto a small
arborizations that overlap with the auditory timing of the ongoing singing motor pattern number of auditory interneurons such as the
neuropil in the prothoracic ganglion, a pre- (n 0 10) (Fig. 2D). Singing also continued local omega neuron 1 (ON1) (18). Paired re-
requisite to forming direct output synapses with normally when CDI was prevented from spiking cordings from CDI and auditory afferent
auditory neurons. Third, its axon targets wide- by injection of hyperpolarizing current (n 0 axonal arborizations demonstrated that spikes
spread areas of the CNS and could affect other 10). We therefore conclude that CDI is not part in CDI occurred in synchrony with primary
sensory pathways activated during singing. of the singing CPG but instead is driven by it. afferent depolarizations (PADs) during wing-
Recordings from CDI were made from its Because CDI receives excitatory input dur- closing motor activity (Fig. 3A). PADs cause a
dendritic branches during pharmacologically ing wing-closing movements, we tested whether reduction in spike height in cricket auditory
elicited fictive singing, with all thoracic sensory it might also be activated during flying. When afferents (14) and mediate presynaptic inhibi-
and motor nerves cut except for the auditory the crickets generated the flight motor pattern tion in a number of sensory systems (1921).
nerves (3, 15). CDI generates bursts of spikes in (n 0 7), CDI was always inhibited by a barrage When depolarizing current was injected into
Fig. 2. Recordings of CDI and wing motor nerve during fictive singing and during fictive flight. (F) Transition from fictive flight to fictive singing
flight. (A) Spikes in CDI as recorded in the dendrite during fictive chirps. demonstrates the change in activity. (G) CDI does not respond to acoustic
(B) The PSTH and superimposed instantaneous spike rate (n 0 59 chirps, stimulation with 4.8 kHz, 85 dB SPL pulses 21 ms in duration. (H) Paired
759 spikes) with the averaged wing motor nerve activity demonstrate that recording from the dendrite in the mesothoracic ganglion and axon in the
CDI fires bursts of spikes during the wing-closing motor activity of each prothoracic auditory neuropil of the same CDI. (I) Average of paired
chirp, as indicated by the stippled lines. (C) Injection of depolarizing recording from the same CDI shows, in this example, a delay of 2.4 ms
current into CDI elicited bursts of spikes but did not change the ongoing from spike recorded in the dendrite to spike recorded in the axon in the
singing pattern. (D) Phase response curve. Q 0 (duration of ongoing chirp prothoracic ganglion (n 0 181 spikes). CDI, intracellular CDI recording;
period N)/(duration of chirp period N 1) at the phase of stimulation of Wing Nerve, activity of mesothoracic nerve 3A. Vertical scale bars, 20 mV
CDI with a current pulse of 100 ms. Analysis shows no modulation in the [(A), (C), (F), (G), and dendrite in (H)], 10 mV [(E) and axon in (H)];
duration of chirps by CDI stimulation. (E) CDI is rhythmically inhibited horizontal scale bars, 250 ms [(A), (G), (H)], 200 ms [(C) and (E)], 1 s (F).
393
520 27 JANUARY 2006 VOL 311 SCIENCE www.sciencemag.org
REPORTS
thoracic ganglion, and spikes in CDI elicited neurons, both with bilateral outputs, the per- two populations of neurons (25, 26). CDI is a
IPSPs and PADs in both the left- and right-side sisting inhibition was most likely due to spikes rare example of a functionally and anatomical-
ON1s and afferents. Paired recordings of CDI in CDI_s contralateral partner cell. However, ly identified neuron that extends throughout the
in its dendrite and prothoracic axon showed we could not rule out the possibility of paral- entire insect nervous system. Simultaneously,
that CDI spikes take 2.8 ms to reach the lel inhibitory neurons. To examine whether CDI mediates the presynaptic inhibition of
auditory neuropil (n 0 4) (Fig. 2, H and I), the CDIs are the only neurons mediating auditory afferents with PADs and the post-
which leaves less than 1 ms from the spike in the inhibition of the auditory pathway, we synaptic inhibition of an identified auditory
the prothoracic ganglion to the release of IPSPs made a paired recording of CDI and ON1 interneuron with IPSPs. This twofold inhibition
or PADs. We never recorded a failure of an during singing and then cut the contralateral reduces the auditory response to self-generated
IPSP or PAD in response to a CDI spike. prothoracic-to-mesothoracic connective, which sounds and protects the cricket_s auditory
Therefore, both the anatomical evidence of contained the axon of the partner CDI (n 0 2). pathway from desensitization during sound
axonal arborizations of CDI in the auditory When the animal_s singing recovered, ON1 was production, allowing it to remain sensitive to
neuropil and the physiological evidence sug- inhibited by CDI during fictive singing and environmental sounds (6, 12, 14). Thus, even
gest that the connection from CDI to auditory failed to respond to the ongoing auditory in the small nervous system of the cricket, self-
afferents and to ON1 is monosynaptic. stimuli (Fig. 4, C and D). When CDI was hy- generated sensory signals are processed in a
What effect does CDI have on auditory perpolarized and prevented from spiking, ON1 similar way to more complex vertebrate ner-
processing? Paired recordings were obtained responded to the sound pulses both in the chirp vous systems (27, 28). During flight, CDI is
from CDI and ON1 during continuous presen- interval and in the chirp (Fig. 4, E and F). This inhibited and prevented from firing; hence,
Fig. 3. Inhibitory inputs in auditory neurons are elicited by CDI spikes. (A) with IPSPs in ON1. (E) Every spike elicited by depolarizing current injection in
Paired recording of CDI and an auditory afferent during singing. Spikes in CDI CDI elicits an IPSP in ON1. (F) Superimposed traces of CDI and ON1 show a
coincide with PADs in the afferent. (B) Stimulation of CDI with intracellular constant latency (in this example 3.4 ms) from CDI spike to the IPSP. Afferent:
depolarizing current injection. Each CDI spike elicits a PAD in the afferent. (C) intracellular auditory afferent recording; ON1, intracellular ON1 recording.
Superposition of CDI and afferent recording triggered by spikes in CDI. In this Vertical scale bars, CDI: 15 mV (A), 30 mV (B), 20 mV (D), 25 mV (E); ON1:
example, PADs were elicited after a constant delay of 3.4 ms from CDI spike. 15 mV (D), 5 mV (E); afferent: 20 mV (A), 10 mV (B); current: 5 nA; horizontal
(D) Paired recording of CDI and ON1 during singing. Spikes in CDI coincide scale bars, 250 ms [(A) and (D)], 200 ms [(B) and (E)].
394
www.sciencemag.org SCIENCE VOL 311 27 JANUARY 2006 521
REPORTS
Fig. 4. The effect of CDI on References and Notes
sound processing. (A) ON1s 1. K. G. Pearson, Annu. Rev. Neurosci. 16, 265 (1993).
2. R. W. Sperry, J. Comp. Physiol. Psychol. 43, 482 (1950).
response to a continuous se-
3. E. von Holst, H. Mittelstaedt, Naturwissenschaften 37,
quence of 4.8 kHz, 75 dB SPL 464 (1950).
sound pulses, 8 ms in du- 4. M. Zaretsky, C. H. F. Rowell, Nature 280, 583 (1979).
ration with intervals of 7 ms, 5. M. A. Sommer, R. H. Wurtz, Science 296, 1480 (2002).
is completely inhibited dur- 6. J. F. A. Poulet, B. Hedwig, Nature 418, 872 (2002).
ing periodic current injec- 7. K. T. Sillar, P. Skorupski, J. Neurophysiol. 55, 678 (1986).
8. K. T. Sillar, A. Roberts, Nature 331, 262 (1988).
tion in CDI. (B) PSTH and 9. S. J. Blakemore, D. Wolpert, C. D. Frith, Nat. Neurosci. 1,
superimposed instantaneous 635 (1998).
spike frequency of ON1 10. C. C. Bell, Science 214, 450 (1981).
averaged over 36 trials dem- 11. J. E. Roy, K. E. Cullen, J. Neurosci. 24, 2102 (2004).
onstrate that ON1 activity is 12. J. F. A. Poulet, B. Hedwig, J. Neurosci. 23, 4717 (2003).
reduced during CDI stimula- 13. J. F. A. Poulet, B. Hedwig, J. Exp. Biol. 204, 1281 (2001).
14. J. F. A. Poulet, B. Hedwig, J. Neurophysiol. 89, 1528 (2003).
tion. (C) In an animal with the 15. See supporting material on Science Online.
contralateral prothoracic-to- 16. R. M. Hennig, J. Comp. Physiol. A 167, 629 (1990).
mesothoracic connective cut, 17. K. Michel, Z. Morph. Tiere 77, 285 (1974).
ON1 responds with a train of 18. K. Schildberger, D. W. Wohlers, F. Huber, in Cricket Behavior
spikes during the chirp inter- and Neurobiology, F. Huber, T. E. Moore, T. E. Loher, Eds.
(Cornell Univ. Press, Ithaca, NY, 1989), pp. 423458.
I
Causeway, Miami, FL 33149, USA.
dispersal remains one of the fundamental limited adult movement, so the relatively *To whom correspondence should be addressed. E-mail:
challenges to marine ecology and ocean- short, pelagic larval phase represents the pri- rcowen@rsmas.miami.edu
395
522 27 JANUARY 2006 VOL 311 SCIENCE www.sciencemag.org
396
Sensory signals during active versus passive movement
Kathleen E Cullen
Our sensory systems are simultaneously activated as the test passive rather than active sensation. Recent results
result of our own actions and changes in the external world. from several laboratories have, however, yielded major
The ability to distinguish self-generated sensory events from insights into our understanding of how sensory signals are
those that arise externally is thus essential for perceptual processed during movement. In this review, I consider
stability and accurate motor control. Recently, progress has recent advances in this field, focusing on experiments in
been made towards understanding how this distinction is the vestibular system that have provided evidence for the
made. It has been proposed that an internal prediction of the differential processing of reafference early in sensory
consequences of our actions is compared to the actual sensory systems [1,2,3]. Parallels between findings in this sys-
input to cancel the resultant self-generated activation. tem and those emerging from recent studies of the visual
Evidence in support of this hypothesis has been obtained [4,5,6] and somatosensory [7,8,9] systems are also
for early stages of sensory processing in the vestibular, considered. This body of work is discussed in relation
visual and somatosensory systems. These findings have to prominent theories of motor control and addresses the
implications for the sensorymotor transformations that are following questions. What information or cues are avail-
needed to guide behavior. able to achieve perceptual stability? At what level in
sensory processing are active and passive sensory signals
Addresses differentially encoded? What mechanisms underlie dif-
Department of physiology, 3655 Promenade Sir William Osler, ferential sensory processing? Finally, I consider the impli-
Montreal, Quebec H3G 1Y6, Canada
e-mail: kathleen.cullen@mcgill.ca
cations of these results in relation to how the processing of
sensory signals during motion is reflected in the sensory
motor transformations that are needed to guide behavior.
Current Opinion in Neurobiology 2004, 14:698706
397
Sensory signals during active versus passive movement Cullen 699
Figure 1
(a)
First-stage
Higher-order
central
processing
processing
(b)
Electric organ
Negative image EO
E.O. discharge command
AR
EMF exafference Ampullary receptor
ELL
EO AR
(c)
Respiration muscles
Motor command RM
PR
EMF exafference Proprioceptive receptor
DON
AR AR RMPR
Ampullary receptor
(a) Simplified scheme of the reafference principle of von Holst and Mittelstaedt [11]. The motor command is sent to the effector muscle and
in turn sensory activation, resulting from the activation of sensory receptors by an effector, is returned. This reafference is then compared to an
efference copy of the original motor command. Here, reafference is arbitrarily marked + and the efference copy is marked . When the
reafference and efference copy signals are of equal magnitude they cancel, and no sensory information is transmitted to the next levels of
processing. By contrast, a difference between the reafference and the efference copy indicates an externally generated event (i.e. exafference)
that is considered behaviorally relevant and thus is processed further. (b) Primary afferent fibers from electroreceptors project to the cerebellum-like
electrosensory lobe (ELL) of mormyrid electric fish. Descending motor commands result in the generation of electric organ discharge, which
results in the activation of ampullary afferents (i.e. reafference). Corollary discharge signals associated with the motor command that elicits the
electric organ discharge are prominent in the mormyrid electrosensory lobe and are used to reduce the central effects of activity in ampullary
receptors evoked by the electric organ discharge (i.e. exafference) of the fish. Abbreviation: EMF, electromagnetic field. (c) Skate do not generate
electric fields but can detect them. The first-order neurons in this system are the principal cells of the dorsal octavolateral nucleus (DON). These
neurons receive proprioceptor inputs from the parallel fibers inputs of the dorsal granular ridge, which cancel electric magnetic field reafference
in the DON during respiration.
of the expected sensory results of a motor command, ized this original proposal by suggesting that an internal
which they termed the efference copy, is subtracted prediction of the sensory consequence of our actions,
from the sensory signal to eliminate reafferent informa- derived from the motor efference copy, is compared to
tion. More recent behavioral investigations have general- the actual sensory input [1315].
398
700 Motor systems
In some model systems, including the electrosensory tinue to respond selectively to passive head motion when
systems of mormyrid fish [16,17] and elasmobranchii a monkey generates voluntary head movements while
(sharks, skates and rays [18]), the mechanosensory system undergoing passive whole-body rotation.
of the crayfish [19,20], and the auditory system of the
cricket [21], evidence indicates that sensory information There are at least three extravestibular cues that could
arising from self-generated behaviors is selectively sup- contribute to canceling reafference at the level of the
pressed at the level of afferent fibers and/or the central vestibular nuclei. First, inputs from neck proprioceptors;
neurons to which they project. The mechanisms by which second, knowledge of the self-generated motion; and
suppression occurs can differ. Bell and co-workers [16,17] third, neck efference copy signals. Roy and I [3,23]
have provided concrete support for Von Holst and have explored each of these possibilities in the rhesus
Mittelstaedts principle of reafference in their investiga- monkey. First, we found that the activation of neck
tions of the electrosensory system in mormyrid fish. In proprioceptors is not sufficient for suppressing vestibular
this system, a negative image of the predicted reafference reafference: neurons in the vestibular nuclei encode head
is generated and added to neurons at the first stage of velocity similarly during passive rotations of the head
central processing (Figure 1b). As a result, the fish does relative to the body and during passive rotations of the
not respond to the discharge from its own electric organ. head and body together.
In the skate, by contrast, information from proprioceptive
inputs alone can generate a negative image of self-motion Second, higher-order areas, such as parietoinsular vestib-
during respiration [18]. This negative image is then used ular cortex, that are involved in the perception of self-
at the first central stage of electrosensory processing to motion [24] are known to send substantial projections to
remove the modulation of its electroreceptors that results each of the vestibular nuclei [25]. We found, however,
from its own motion (Figure 1c). that knowledge of self-generated head motion alone is not
sufficient to suppress vestibular reafference. Neurons
Vestibular processing during active head respond robustly to head velocity when monkeys drive
movements themselves through space by rotating a steering wheel
The vestibular system provides information about head connected to the motor controller of a vestibular turntable
motion relative to space that is necessary for maintaining [23]. Third, a copy of the command to the neck muscles
posture, computing spatial orientation and perceiving (i.e. a motor efference signal) could be used to cancel
self-motion. The differential processing of actively versus vestibular inputs during active head movements. We
passively generated vestibular stimuli is crucial, however, found, however, that when head-restrained monkeys
for controlling eye, neck and body movements, as well as attempted to move their heads, generating levels of neck
for perceptual stability (reviewed in [1]). This point can torque comparable to those issued during large active
be easily appreciated by considering the simple example head movements, neuronal responses were not modu-
of a vestibularly driven spinal reflex the vestibulo- lated [3].
collic reflex which, in response to head motion, stabi-
lizes the head in space via activation of the neck muscu- Further exploration has provided evidence for Von Holst
lature (Figure 2a). The compensatory head movements and Mittelstaedts principle of reafference in the primate
produced by this reflex are clearly beneficial when the vestibular system [3]. The activity of individual neurons
behavioral goal is to stabilize head position in space. in the vestibular nuclei was recorded in monkeys making
When the behavioral goal is to make an active head active head movements and the correspondence between
movement, however, the vestibular drive to the reflex intended and actual head movement was experimentally
pathway would command an inappropriate head move- controlled. We found that a cancellation signal was gen-
ment to move the head in the direction opposite to the erated only when the activation of neck proprioceptors
intended goal. matched the motor-generated expectation during active
head movements (Figure 3). This mechanism functions
At what level are the vestibular signals that arise from to eliminate selectively self-movements from the subse-
passive versus active head movement first differentially quent computation of orientation and posture control.
encoded? In the alert primate, the vestibular system does These findings are the first to confirm Von Holst and
not seem to distinguish between active and passive head Mittelstaedts proposal at the level of single neurons in a
movements at the level of the vestibular afferents (Figure mammalian system.
2b; [2]), but the differential treatment of vestibular sig-
nals is evident at the next stage of processing. The head- Correlates for the differential processing of active and
velocity-related modulation of one population of vestib- passive vestibular inputs have been identified upstream
ular nuclei neurons, which receive direct inputs from the of the vestibular nuclei. Head direction cells in several
vestibular afferents, is markedly attenuated in response to areas of the classic limbic circuit discharge preferentially
vestibular inputs that result from active head movements when a monkey or rat orients its head in a specific
(Figure 2c; [22,23]). Notably, these same neurons con- preferred direction. These neurons, which are important
399
Sensory signals during active versus passive movement Cullen 701
Figure 2
Horizontal Neck
canal motoneurons
Cerebellum
thalamus/cortex?
100 H 100 H
100 spikes 100 spikes
FR FR
2s
100 100
H H
100 spikes 100 spikes
FR FR
100 ms
100 deg/s
100 H passive 100 H passive
H active H active
50 spikes
50 sp/s 50 spikes
H combined H combined
FR FR
2s
In the vestibular system, second-order neurons distinguish between sensory inputs that result from self-actions and sensory inputs that arise
externally. (a) A subclass of neurons in the vestibular nucleus, named vestibular-only neurons on the basis of their responses in head-restrained
models, receive direct projections from the semicircular canals and in turn project bilaterally to spinal motor neurons to activate the neck
musculature. These neurons probably also send projections to the cerebellum, thalamus and cortex. (b) Activity of an example horizontal canal
afferent during (i) passive head movements, (ii) active head movements and (iii) combined active and passive head-in-space motion. Afferents
reliably encode head-in-space motion in all conditions. (c) Activity of an example VO neuron in the vestibular nucleus during the head movements
described in (b). Neuronal responses to the active component of head-in-space motion are significantly attenuated; by contrast, the neurons
show no attenuation in response to the passive rotation component. Abbreviations: FR, firing rate;H, horizontal head velocity. Afferent responses are
based on data from [2], and (c) is modified from data reported in [23].
for spatial memory and navigation, respond more robustly tion of extrapersonal space. I take this point up again
during active than during passive head rotations [26]. In further below.
addition, neurons in the ventral interparietal area, one of
several vestibular areas identified in the parietotemporal Visual processing during voluntary eye
cortex, have been studied during active and passive head movements
movements. The results show that there are considerable How the visual world is perceived as stable despite
differences in neural responses during active versus pas- movements of the eyes, head and body is an issue that
sive head movements [27]. This differential processing of concerned many eminent scientists of the last century
vestibular inputs most probably reflects the integration of including von Helmholtz, Hering, Mach and Sherrington.
information from an efference copy of motor commands Although targets rapidly jump across the retina as we
and from proprioceptive and vestibular sources required move our eyes to make saccades, we never see the world
for the perception of self-motion and for the representa- move over our retina. Helmholtz [28] made the salient,
400
702 Motor systems
Figure 3
100 100
HS 100
HB active
Passive table
In the vestibular system, an internal model of the sensory consequences of active head motion is used to selectively suppress reafference
at the level of the vestibular nuclei. (a) Activity of an example VO neuron (gray-filled trace) during passive whole-body rotation. In this condition,
only vestibular inputs are available to the central nervous system and there is no motor efference copy signal because the monkey does
not actively move its head. (b) Activity of the same neuron during active head-on-body movements. In this condition, the monkey commands an active
head movement and thus an efference copy signal is theoretically available. In addition, the movement of the head activates vestibular and
proprioceptive afferents as a result of the head-in-space (HS ) and head-on-body (HB ) movement, respectively. A prediction of the activity of the
neuron, based on its response to passive head motion, is superimposed (bold trace). (c) The neuron is recorded as the monkey actively moves its
head; however, the head-in-space velocity (HS ) that is generated by the monkey (broken arrow) is experimentally cancelled by simultaneously
rotating the monkey in the opposite direction (unbroken arrow). Consequently, in this condition the head moves relative to the body (HB ),
but not to space (HS ); as a result, an efference copy signal is generated and the neck proprioceptors are activated, but vestibular afferent input is
greatly reduced. This approach reveals a cancellation signal that is sent to the vestibular nuclei in conditions in which neck proprioceptive inputs
match those expected based on the neck motor command, but not when these inputs are vastly different. Indeed, the marked inhibition in the
response of the neuron shows excellent correspondence to that from the difference in response during passive (a) versus active (b) head
movements. Abbreviation: FR, firing rate. Modified from data reported in [3].
and easily replicated, observation that tapping on the correlates of saccadic suppression are most evident in
canthus of the eye to displace the retinal image, for magnocellular recipient areas including medial temporal
example during a saccadic eye movement, results in an and medial superior temporal cortex of rhesus monkey
illusionary shift of the visual world. How can continually [30].
changing retinal inputs to the visual system thus result
in the perception of a stable visual world during eye Results of studies in humans are remarkably consistent
movements? with these findings. For example, functional magnetic
resonance imaging (fMRI) studies show saccade-related
Our visual sensitivity is reduced during and just before a changes in areas that receive magnocellular input includ-
saccadic eye movement a phenomenon that is referred ing medial temporal cortex, V7 and V4 [31]. In addition,
to as saccadic suppression. Psychophysical data have transcranial stimulation studies in humans have provided
shown that suppression is strongest for stimuli that would evidence that saccadic suppression occurs via extraretinal
preferentially activate the magnocellular dorsal visual mechanisms in the thalamus or primary visual cortex [6].
processing stream, which carries transient, motion-related Note, however, that there might be differences in the
visual information (e.g. see [29]). Neural correlates for strategies adopted across species (e.g. see the work of
saccadic suppression have been now identified at several Olveczky et al. [32] in rabbit).
stages of visual processing. Significant saccade-related
responses can be observed early in visual processing at The mechanisms that underlie saccadic suppression
the level of the dorsal lateral geniculate nucleus of the remain to be fully elucidated. Proprioceptive information
thalamus [4,5]. Reppas et al. [5] found that the most from the stretch receptors in the extraocular muscles
common effect was a biphasic modulation of response could be used for this purpose; however, the relative
strength (weak suppression followed by strong enhance- timing of saccadic suppression and eye movement sug-
ment), which was far more prominent for neurons in the gests that inputs from saccadic planning centers (e.g. an
magnocellular than the parvocellular layer. Moreover, efference copy or corollary discharge signal) are crucial.
401
Sensory signals during active versus passive movement Cullen 703
Somatosensory processing during active ated only when the activation of neck proprioceptors
touch matches the motor-generated expectation, thereby elim-
Touch is also clearly an active process. Fingers or whis- inating vestibular signals that result from active head-on-
kers, like eyes, move as they scan the external world. body movements. Such a mechanism has clear analogies
These movements thus determine the nature of the to that used by the electrosensory system of the mormyrid
sensory input that will be encountered. As pointed out fish to cancel reafference. In addition, recent behavioral
by Gibson [33], the constancy of the tactual world is studies in humans have shown that a similar strategy is
characterized by the stability of an object. For example, used by the somatosensory system.
movement of a skin surface over a corner of an object does
not usually give rise to recognition of the tactile motion. Where is the essential cancellation signal actually com-
puted for each system? Evidence from work in electric
The whiskers (vibrissae) system in rats has proved to be fish indicates that the cerebellum-like electrosensory
an excellent experimental model of active touch. During lobes provide the signal that is used to cancel the sensory
whisking, the responses of first-order neurons in the response to self-generated stimulation [39]. Moreover,
trigeminal ganglion of the vibrissae system cannot be fMRI studies have suggested that the cerebellum has a
inferred from their responses to comparable passive whis- similar role in the suppression of tactile stimulation dur-
ker deflection. Ahissar and co-workers [7] have recently ing self-produced tickle [3638]. More recently, however,
shown that, during intervals of whisking, these neurons the results of a fMRI study of active hand movements has
carry both a reference signal that encodes the current led to the proposal that efference copy information is not
position of the rats vibrissae and a more typical sensory located in the cerebellum per se, but instead is imple-
signal that encodes contact of the vibrissae with an object. mented via interactions between perceptual areas and
Timing differences between the reference and the con- motor areas in a task-dependent way [40]. Identifying
tact signals carried by neurons can be used to compute the the neural representations of efference copy information
position of an object in space. In this system, the position in different tasks promises to be an interesting area of
of vibrissae within a cycle could be encoded primarily investigation.
through direct sensory activation rather than via central
pathways. Nevertheless, there is some evidence for the It remains to be determined whether a comparable frame-
existence of a modulatory efference copy at the cortical work can be used to describe the processing that selec-
level that might influence the amplitude of whisking [34]. tively suppresses reafference across all or even most
Central information of this kind can be used to tune large- modalities. It is now clear that other mechanisms can
scale feedback loops in the vibrissae sensorimotor system contribute to the differential processing of reafference.
to optimize sensory processing [8]. For example, peripheral mechanisms have an essential
role in processing touch during active whisking, as
In primates, correlates for the differential active proces- described above. In addition, a recent study has shown
sing of touch have been found in somatosensory cortex, that during active wrist movements in primates, cutaneous
where neuronal responses are attenuated for self- inputs are presynaptically inhibited at the level of the
produced versus externally produced tactile stimuli spinal cord afferents [9]. The timing of the attenuation
[35]. In humans, self-generated tactile stimulation does suggests that descending motor commands, rather than
not result in the same tickling sensation that arises when peripheral feedback from the movement, generate the
the stimulation is externally produced [3638]. Blakemore inhibition. This strategy is markedly different from that
et al. [37] have shown, however, that when an artificial used by the vestibular system: vestibular afferents do not
delay or trajectory perturbation is introduced between a differentially encode active versus passive head move-
movement and the resultant tactile stimulation, self- ments; instead, reafference is distinguished only at the
generated stimulation is rated as more ticklish. Thus, next stage of processing in the vestibular nuclei.
as in the vestibular system, when the sensory sensation
no longer corresponds to the motor command, the pre- Implications: consequences for motor
dicted image of the afferent signal (in this case a tickle control
stimulus) does not fully cancel the distorted reafference. The differentiation of sensory stimulation that arises from
The analogy between findings in these two systems passive versus active movement is not only crucial for
implies that the vestibular system is effectively ticklish. perceptual stability but is also required to produce accu-
rate neural representations of the environment to guide
Common strategies? behavior accurately. The implications of this are particu-
Studies of the primate vestibular system have shown that larly obvious for the vestibular system, where the second-
a cancellation signal is used at the first level of central order sensory neurons function as both sensory and pre-
processing to suppress vestibular activation resulting from motor neurons. Second-order vestibular neurons, which
active head movements. An investigation by Roy and are differentially sensitive to active and passive head
myself [3] found that this cancellation signal is gener- movements, project to the spinal cord and mediate
402
704 Motor systems
postural reflexes such as the vestibulo-collic reflex cessing of active and passive movement that can be found
(Figure 2a). Accordingly, their reduced sensitivity during at these higher levels of processing.
active head movements is consistent with their functional
role in head stabilization, because their modulation is Conclusions
actually counterproductive during these self-generated Over 50 years ago, Von Holst and Mittelstaedt [11]
movements. The fact that these same neurons continue proposed that the brain generates a sensory expectation
to encode information faithfully about passive head rota- based on the motor command, compares it with the actual
tions that occur during the execution of voluntary move- sensory feedback, and subtracts the self-generated sen-
ments [22,23] is also important. As a person explores the sation. In this way, the nervous system could theoreti-
environment, these neurons will selectively respond to cally differentiate sensory inputs that arise from external
adjust postural tone in response to any head movements sources from those that result from self-generated move-
that the brain does not expect. This selectivity can be ments. Recent work in several systems has provided
crucial, for example, in recovering from tripping over an evidence in support of this hypothesis, as well as evi-
obstacle while walking or running, which requires a dence for other mechanisms that suppress reafference in
robust postural response. the early stages of sensory processing. It remains a
challenge to understand how the differential processing
Vestibular pathways that control postural responses must of sensory inputs in the early stages is used by the
combine vestibular inputs with information from other upstream networks that mediate perceptual stability
sources, such as neck and body proprioceptive informa- and guide behavior.
tion, to generate appropriate motor responses. This can be
easily appreciated by considering two subjects: one whose Acknowledgements
head is pointing forward, and another whose head is I thank J Roy, S Sadeghi, M McCluskey, A Andrei and M Van Horn
for helpful discussions on this topic; and A Andrei for valuable
rotated 908 to one side. To ensure postural stability, contributions to the design and exposition of the illustrations. This
the same vestibular stimulus will need to activate differ- work was supported by a grant from the Canadian Institutes of Health
ent muscle combinations in each subject. In this example, Research (CIHR). The author holds a McGill Dawson Chair and a
CIHR Investigator Award.
combining neck proprioceptive information with head-
referenced vestibular inputs to the brain will provide References and recommended reading
information about motion of the body in space and, Papers of particular interest, published within the annual period of
indeed, there is recent evidence for such a multimodal review, have been highlighted as:
integration of vestibular and proprioceptive inputs. Some of special interest
neurons in the rostral fastigial nucleus of alert primates of outstanding interest
encode movement of the body rather than the head in
space during passively applied rotational [41] and transla- 1. Cullen KE, Roy JE: Signal processing in the vestibular system
during active versus passive head movements.
tional [42] motion. Thus, during passive movements J Neurophysiol 2004, 91:1919-1933.
these neurons encode movement in a body-referenced This review contrasts the responses of two classes of neuron in the
vestibular nuclei during active movement: PVP neurons, which mediate
coordinate system. the vestibulo-ocular reflex (VOR); and VO neurons, which are thought to
mediate the vestibulo-collic reflex (VCR) and to shape vestibular informa-
tion for the computation of spatial orientation by upstream structures. The
To date, coordinate transformations, which have been head-velocity signals carried by VOR interneurons (PVP neurons) are
observed in many brain areas, have been characterized in reduced when the goal is to redirect gaze in space, regardless of whether
head movement is active or passive. By contrast, the vestibular signals
studies designed to test passive rather than active sensa- carried by VCR interneurons (the neurons discussed in the current review)
tion. What do these neurons encode during active move- are reduced in response to active head-on-body movements. The
ments? The rostral fastigial nucleus is a particularly mechanisms that underlie this differential processing of vestibular infor-
mation by PVP and VO neurons use efference copies of gaze- and neck-
interesting example to consider with regard to this ques- movement commands, respectively.
tion. It is thought to be important in vestibulo-spinal 2. Cullen KE, Minor LB: Semicircular canal afferents similarly
control, including the regulation of gait and postural encode active and passive head rotation: implications for the
mechanisms. Thus, the same arguments made above role of vestibular efference. J Neurosci 2002, 22:RC226:1-7.
for the utility of selective gating of vestibular reafference 3. Roy JE, Cullen KE: Dissociating self-generated from passively
applied head motion: neural mechanisms in the vestibular
at the level of the vestibulo-spinal neurons could be nuclei. J Neurosci 2004, 24:2102-2111.
applied to the rostral fastigial nucleus. By extension, a This study shows that vestibular signals that arise from self-generated
similar logic can be applied to areas that have been head movements are inhibited by a mechanism that compares the
internal prediction of the sensory consequences by the brain to the actual
implicated in higher-order perceptual functions. Spatial resultant sensory feedback. Self-generated vestibular inputs are selec-
perception is more accurate in response to active than to tively cancelled early in processing at the level of second-order neurons.
passive rotations [4345]. The absence of a signal related 4. Ramcharan EJ, Gnadt JW, Sherman SM: The effects of saccadic
to motor efference copy, at the level of higher-order eye movements on the activity of geniculate relay neurons in
the monkey. Vis Neurosci 2001, 18:253-258.
vestibular areas such as parietal cortex might prevent
5. Reppas JB, Usrey WM, Reid RC: Saccadic eye movements
the complete updating required for accurate spatial loca- modulate visual responses in the lateral geniculate nucleus.
lization [46], consistent with the differences in the pro- Neuron 2002, 35:961-974.
403
Sensory signals during active versus passive movement Cullen 705
6. Thilo KV, Santoro L, Walsh V, Blakemore C: The site of saccadic that responses to self-generated sound are reduced by a centrally
suppression. Nat Neurosci 2004, 7:13-14. generated corollary discharge.
Transcranial magnetic stimulation of the human occipital cortex and
retina is used to induce illusory visual perceptions (phosphenes). The 22. McCrea RA, Gdowski GT, Boyle R, Belton T: Firing behavior of
authors show that the perception of phosphenes induced by stimulation vestibular neurons during active and passive head
of occipital cortex is immune to saccadic suppression, whereas that of movements: vestibulo-spinal and other non-eye-movement
phosphenes induced by retinal stimulation is not. related neurons. J Neurophysiol 1999, 82:416-428.
7. Szwed M, Bagdasarian K, Ahissar E: Encoding of vibrissal active 23. Roy JE, Cullen KE: Selective processing of vestibular
touch. Neuron 2003, 40:621-630. reafference during self-generated head motion.
During active touch, the response of neurons in the trigeminal ganglion of J Neurosci 2001, 21:2131-2142.
the vibrassae system can be divided into two categories: first, touch-
24. Deutschlander A, Bense S, Stephan T, Schwaiger M, Brandt T,
sensitive neurons that encode either the interval of whisker contact with
Dieterich M: Sensory system interactions during simultaneous
the object or the timing of object contact or detachment; and second,
vestibular and visual stimulation in PET. Hum Brain Mapp 2002,
whisking neurons, which are active during the motor act of whisking
16:92-103.
regardless of whether the object touches an object. Accordingly, timing
differences between the discharges of touch-sensitive neurons and those 25. Akbarian S, Grusser OJ, Guldin WO: Corticofugal connections
of whisking-sensitive neurons encode horizontal object position. The between the cerebral cortex and brainstem vestibular nuclei in
authors propose that locked loops, such as those in the paralemniscal the macaque monkey. J Comp Neurol 1994, 339:421-437.
system [8], could be used to detect timing differences between these cells
and to translate them into a spike count code. 26. Wiener SI, Berthoz A, Zugaro MB: Multisensory processing in
the elaboration of place and head direction responses by
8. Ahissar E, Kleinfeld D: Closed-loop neuronal computations: limbic system neurons. Brain Res Cogn Brain Res 2002,
focus on vibrissa somatosensation in rat. Cereb Cortex 2003, 14:75-90.
13:53-62.
27. Klam F, Graf W: Vestibular signals of posterior parietal cortex
9. Seki K, Perlmutter SI, Fetz EE: Sensory input to primate spinal neurons during active and passive head movements in
cord is presynaptically inhibited during voluntary movement. macaque monkeys. Ann N Y Acad Sci 2003, 1004:271-282.
Nat Neurosci 2003, 6:1309-1316.
This study provides evidence for differential encoding of somatosensory 28. von Helmholtz H: Handbuch der Physiologischen Optik, vol 3, edn
inputs at the afferent level in primates. Cutaneous afferents to the spinal 1. Hamburg: Voss; 1867. Treatise on Physiological Optics (English
cord are presynaptically inhibited during active wrist movements. This Translation), vol 3. Translated by Southall JPC. Rochester: Optical
inhibition occurs preferentially during active movements and precedes Society of America; 1925:44-51.
electromyographic activity, suggesting that descending motor com-
mands, rather than peripheral feedback from the motion itself, have a 29. Burr DC, Morrone MC, Ross J: Selective suppression of the
dominant role in generating the inhibition. In addition, the monosynaptic magnocellular visual pathway during saccadic eye
responses in active first-order interneurons are reduced simultaneously movements. Nature 1994, 371:511-513.
with increased excitability of the relevant afferent terminals.
30. Thiele A, Henning P, Kubischik M, Hoffmann KP: Neural
10. Sherrington CS: The Integrative Action of the Nervous System. mechanisms of saccadic suppression. Science 2002,
New York: Charles Scribners Sons; 1906: 306-307. 295:2460-2462.
11. von Holst E, Mittelstaedt H: Das reafferenzprinzip. 31. Kleiser R, Seitz RJ, Krekelberg B: Neural correlates of saccadic
Naturwissenschaften 1950, 37:464-476. [Translation of title: suppression in humans. Curr Biol 2004, 14:386-390.
The principle of reafference].
32. Olveczky BP, Baccus SA, Meister M: Segregation of object and
12. Sperry R: Neural basis of spontaneous optokinetic response background motion in the retina. Nature 2003, 423:401-408.
produced by visual inversion. J Comp Physiol Psychol 1950, This study describes a subclass of center surround retinal ganglion cells
43:482-489. in rabbit retina that respond optimally when there are differences in center
versus surround motion. In other words, these neurons respond to motion
13. Decety J: Neural representation for action. Rev Neurosci 1996, in the visual field but only if the surround and center move with different
7:285-297. trajectories. Thus, these neurons would not be stimulated by global retinal
motion arising from retinal drift during fixation or more rapid movements
14. Farrer C, Franck N, Paillard J, Jeannerod M: The role of during saccades.
proprioception in action recognition. Conscious Cogn 2003,
12:609-619. 33. Gibson JJ: Observations on active touch. Psychol Rev 1962,
69:477-491.
15. Wolpert DM, Ghahramani Z, Jordan M: An internal model for
sensorimotor integration. Science 1995, 269:1880-1882. 34. Fee MS, Mitra PP, Kleinfeld D: Central versus peripheral
determinants of patterned spike activity in rat vibrissa cortex
16. Bell C: An efference copy which is modified by reafferent input. during whisking. J Neurophysiol 1997, 78:1144-1149.
Science 1981, 214:450-453.
35. Chapman CE: Active versus passive touch: factors influencing
17. Mohr C, Roberts PD, Bell CC: The mormyromast region of the the transmission of somatosensory signals to primary
mormyrid electrosensory lobe. I. Responses to corollary somatosensory cortex. Can J Physiol Pharmacol 1994,
discharge and electrosensory stimuli. J Neurophysiol 2003, 72:558-570.
90:1193-1210.
36. Blakemore SJ, Wolpert DM, Frith CD: Central cancellation of
18. Hjelmstad GO, Parks G, Bodznick D: Motor corollary discharge self-produced tickle sensation. Nat Neurosci 1998, 1:635-640.
activity and sensory responses related to ventilation in the
skate vestibulolateral cerebellum: implications for 37. Blakemore SJ, Wolpert DM, Frith CD: The cerebellum
electrosensory processing. J Exp Biol 1996, 199:673-681. contributes to somatosensory cortical activity during self-
produced tactile stimulation. NeuroImage 1999, 10:448-459.
19. Krasne FB, Bryan JS: Habituation: regulation through
presynaptic inhibition. Science 1973, 182:590-592. 38. Blakemore SJ, Wolpert DM, Frith CD: Spatio-temporal
prediction modulates the perception of self-produced stimuli.
20. Edwards DH, Heitler WJ, Krasne FB: Fifty years of a command J Cogn Neurosci 1999, 11:551-559.
neuron: the neurobiology of escape behavior in the crayfish.
Trends Neurosci 1999, 22:153-161. 39. Bell CC, Han VZ, Sugawara Y, Grant K: Synaptic plasticity in the
mormyrid electrosensory lobe. J Exp Biol 1999, 202:1339-1347.
21. Poulet JF, Hedwig B: Corollary discharge inhibition of
ascending auditory neurons in the stridulating cricket. 40. Leube DT, Knoblich G, Erb M, Grodd W, Bartels M, Kircher TT:
J Neurosci 2003, 23:4717-4725. The neural correlates of perceiving ones own movements.
This article examines how the auditory pathways in crickets process self- NeuroImage 2003, 20:2084-2090.
generated auditory signals during singing. Modulation of neurons during In this study, subjects opened and closed their hand and their movement
pharmacologically elicited sonorous, silent and fictive singing indicates was filmed and projected online onto a screen. The temporal delay
404
706 Motor systems
between movement and its visual feedback was varied, and subjects 43. Blouin J, Gauthier GM, Vercher JL: Failure to update
were asked to report whether there was a delay. fMRI imaging during this the egocentric representation of the visual space
task shows that activity in the posterior superior temporal sulcus (pSTS) is through labyrinthine signal. Brain Cogn 1995,
positively correlated with the magnitude of the temporal delay between a 29:1-22.
self-generated movement and visual feedback. By contrast, the cere-
bellum seems to be involved in providing information related to the 44. Fery YA, Magnac R, Israel I: Commanding the direction of
conscious detection of visual event timing. passive whole-body rotations facilitates egocentric spatial
updating. Cognition 2004, 91:B1-B10.
41. Kleine JF, Guan Y, Kipiani E, Glonti L, Hoshi M, Buttner U: Trunk
position influences vestibular responses of fastigial nucleus 45. Jurgens R, Boss T, Becker W: Estimation of self-turning in the
neurons in the alert monkey. J Neurophysiol 2004, dark: comparison between active and passive rotation.
91:2090-2100. Exp Brain Res 1999, 128:491-504.
42. Shaikh AG, Meng H, Angelaki DE: Multiple reference frames 46. Duhamel JR, Colby CL, Goldberg ME: Ventral intraparietal area
for motion in the primate cerebellum. J Neurosci 2004, of the macaque: congruent visual and somatic response
24:4491-4497. properties. J Neurophysiol 1998, 79:126-136.
405
406
481
407
482 Sensory systems
Figure 1
believed to have evolved their electrosensory system motor command (out to about 120 ms), to the polarity of the
independently [11]. sensory response (increase or decrease in firing frequency;
Figure 2a), and to the amplitude of the sensory response.
The mormyrid ELL The negative image is also specific to the spatial distribu-
Memory-based expectations were first observed in electric tion of the sensory response across the surface of the ELL,
fish of the family Mormyridae in the region of the ELL that as the negative image is only formed if the sensory stimu-
receives afferents from ampullary electroreceptors, and is lus is within the receptive field of a cell and evokes a
responsible for passive electrolocation [12,13]. Ampullary response. During sensory stimulation, the addition of the
afferents respond to the pulsatile EOD with a change in negative image of past or predicted input minimizes the
discharge frequency; but such responses could interfere with central effect of predictable features in concurrent afferent
the sensory function of the ampullary system. Responses of input (compare the effect of command plus stimulus at the
these afferents to the fishs own EOD vary with water beginning and at the end of pairing in Figure 2b).
conductivity and the surrounding environment. Thus,
cancellation or minimizing of the EOD-induced interference The responses to corollary discharge alone disappear
requires an adaptive memory-based mechanism. within a few minutes after turning off the sensory stimulus
(Figure 2b). This disappearance is not a passive decay,
The principal cells of the mormyrid ELL are not only however, but rather an active rematching of the changed
affected by afferent input from the periphery but also by a sensory input that now follows the motor command [14].
centrally originating corollary discharge signal coupled to The negative image can last 30 minutes if rematching is
the motor command that elicits the EOD. Corollary dis- prevented by silencing the EOD motor command during
charge effects are studied in fish in which the synaptic the intervening period.
effect of electromotor neurons on the electric organ has
been blocked by the drug curare. The EOD is blocked in Adaptive corollary discharge effects are also found in
these fish but the motor command that would normally regions of the mormyrid ELL that receive afferent input
elicit an EOD continues to be emitted. Artificial electrical from tuberous electroreceptors and which are responsible
stimuli are delivered to the receptive field of the cell at for active electrolocation [15,16]. The subtraction of pre-
fixed times with regard to the motor command. Responses dicted sensory input in these regions probably enhances
to the corollary discharge are minimal when no electrosen- sensitivity to small changes in EOD-evoked input.
sory stimuli have been given for several minutes but
become prominent after pairing the command with an The gymnotid ELL
effective sensory stimulus for a few minutes (Figure 2). The gymnotid studies have been done in regions of ELL
The responses to corollary discharge alone after pairing are that are responsible for active electrolocation in species
a highly specific negative image of the previously paired with continuous, near sinusoidal EODs. The electric
response to a sensory stimulus [13]. The negative image is organs of these species extend from near the head to the
specific to the timing of the sensory response after the tip of the tail. When the tail is bent, the electroreceptors on
408
Memory-based expectations in electrosensory systems Bell 483
Figure 2
2min
pairing for 5 minutes (bottom). Right: same cell EOD
Command plus
6 minutes later when the responses to stimulus EOD
command alone have returned to a baseline Command alone
level. Same sequence as in (a), except that the after
pairing was with an inhibitory sensory stimulus.
(In ampullary electroreceptors, a simple EOD
Command alone
inversion of the stimulus from a positive to a after
negative voltage inverts the response.) In both 0 160 msec
50ms
cases, the responses to command alone after Current Opinion in Neurobiology
pairing are opposite to the effects of the
previously paired stimuli. (b) Raster display of occurred at time 0. The electrosensory result of the development of a response to
another cell in the ampullary region showing stimulus was delivered immediately afterwards, command alone that is a negative image
the effect of pairing the EOD command with a as indicated by the black line. The response to (a burst-pause) of the previously paired
brief sensory stimulus that evoked a pause- command plus stimulus is reduced after sensory response. (Reproduced with
burst in cell activity. The EOD command several minutes of pairing and this is as a permission from [14].)
one side of the body are more strongly affected by the stimulus to the whole body that mimicked the voltage
EOD, whereas those on the other side are less strongly changes caused by tail movement. Pairing a local elec-
affected. These changes in EOD-evoked afferent trosensory stimulus with the whole body electrosensory
responses caused by tail movement, however, convey no stimulus at a fixed phase resulted in a negative image of the
information about external conductances and would in fact response to a local electrosensory stimulus that was evoked
interfere with their perception. by the whole body stimulus.
409
484 Sensory systems
Figure 3
the negative images: corollary discharge signals associated site of plastic change because all of the effective predictive
with ventilatory motor commands; proprioceptive signals signals are conveyed by parallel fibers and some, such as
associated with ventilation; and whole-body electrosensory proprioceptive signals, are conveyed only by parallel fibers.
stimulation associated with ventilation [20]. This hypothesis was also suggested by the known plasticity
of parallel fiber synapses in the cerebellum itself [2426].
Disappearance of the negative image within a few minutes
after turning off the electrosensory stimulus is again, as in Plasticity with appropriate characteristics has now been
mormyrids and gymnotids, an active rematching of the demonstrated at parallel fiber synapses in all three elec-
change in sensory response. This was shown using passive trosensory structures. Pairing parallel fiber input with
bending of the fin as the predictive stimulus and pairing postsynaptic depolarization causes a reduction in synaptic
such bending with a local electrosensory stimulus. The neg- efficacy at this synapse [20,27,28], a form of plasticity
ative image was still present three hours after turning off the referred to as anti-Hebbian [29]. As a result of such plasticity,
electrosensory stimulus, if rematching was prevented by not the association of parallel fiber activity with principal cell
moving the fin during the intervening three hour period. activation by electrosensory input will reduce synaptic
strength at the paired parallel fiber inputs (relative to other
Sites and mechanisms of the plastic change parallel fiber inputs), resulting in a hyperpolarizing response
responsible for the negative images to the predictive signals conveyed by the paired inputs.
The negative images of predicted input are due, in large
part at least, to plastic changes within the cerebellum-like The synaptic mechanisms responsible for robust depolar-
structures themselves. This was shown by pairing the izing responses to predictive signals following pairing with
predictive signals with intracellular current injection rather hyperpolarizing sensory responses (Figure 2a, right side)
than with sensory stimuli, reasoning that if the intracellular are less clear. Associative synaptic plasticity following pair-
current affects only the recorded cell and if the pairing ing with purely hyperpolarizing stimuli is rare [30], and
results in the generation of a negative image, then changes possible mechanisms for such plasticity are obscure. In
must have taken place at synapses onto the recorded cell. addition, although pairing parallel fiber input with postsy-
Depolarizing intracellular current pulses were paired with naptic hyperpolarization yielded an increase in synaptic
the EOD corollary discharge in mormyrid fish [23] and efficacy in gymnotid and elasmobranch fish, such pairing
with ventilation in elasmobranch fish [20]. After the had no effect in mormyrid fish (C Bell, unpublished data).
pairing, the same predictive signals alone evoked This difference could reflect the fact that the mormyrid
hyperpolarizing responses in both groups of fish. Thus, results were obtained in vitro whereas results from the
plastic change appeared to have taken place at synapses other two groups of fish were obtained in vivo from the
between fibers conveying predictive signals and principal whole animal. Postsynaptic spiking is critical for synaptic
cells. Parallel fiber synapses were hypothesized to be the depression in the mormyrid ELL [31], but spiking,
410
Memory-based expectations in electrosensory systems Bell 485
411
486 Sensory systems
for the generation of negative images [32,36]. A variety of systems: a comparative overview of anatomical and functional
similarities. Auditory Neuroscience 1995, 1:207-231.
different learning rules for parallel fiber plasticity were
5. Bastian J: Pyramidal-cell plasticity in weakly electric fish: a
tested in a model of the mormyrid ELL [32], and the neg- mechanism for attenuating responses to reafferent
ative images resulting from the use of an idealized form of electrosensory inputs. J Comp Physiol 1995, 176:63-78.
the experimentally-determined, asymmetric learning rule 6. Bell CC, Grant K, Serrier J: Corollary discharge effects and sensory
(thin continuous line in Figure 4a) were significantly more processing in the mormyrid electrosensory lobe: I. Field
potentials and cellular activity in associated structures.
accurate and more stable than the negative images result- J Neurophysiol 1992, 68:843-858.
ing from any other learning rule. 7. Hjelmstad GO, Parks G, Bodznick D: Motor corollary discharge
activity and sensory responses related to ventilation in the skate
Conclusions vestibulolateral cerebellum: Implications for electrosensory
processing. J Exp Biol 1996, 199:673-681.
The role of parallel fiber synapses in generating negative
8. Conley RA, Bodznick D: The cerebellar dorsal granular ridge in an
images of predicted sensory input is indicated by the elasmobranch has proprioceptive and electroreceptive
predictive signals that they convey, by the presence of representations and projects homotopically to the medullary
electrosensory nucleus. J Comp Physiol 1994, 174:707-721.
appropriate plasticity at these synapses, and by the model-
ing studies. Thus, the connection between synaptic 9. Sas E, Maler L: The organization of afferent input to the caudal
lobe of the cerebellum of the gymnotid fish Apteronotus
plasticity at the cellular level and a systems level adaptive leptorhynchus. Anat Embryol 1987, 177:55-79.
function is relatively secure for these cerebellum-like 10. Bell C, Bodznick D, Montgomery J, Bastian J: The generation and
structures, as secure perhaps as in any vertebrate system. subtraction of sensory expectations within cerebellum-like
structures. Brain Behav Evol 1997, 50:17-31.
However, much remains to be elucidated about the gener- 11. Bullock TH, Bodznick DA, Northcutt RG: The phylogenetic distribution
of electroreception: evidence for convergent evolution of a primitive
ation of negative images before the full value of these vertebrate sense modality. Brain Res Rev 1983, 6:25-46.
cerebellum-like structures for understanding adaptive 12. Bell CC: An efference copy modified by reafferent input. Science
processing is obtained. Future work must determine the 1981, 214:450-453.
following: first, the cellular mechanisms of plastic change; 13. Bell CC: Properties of a modifiable efference copy in electric fish.
second, the mechanism by which pairing with hyperpolar- J Neurophysiol 1982, 47:1043-1056.
izing sensory stimuli results in depolarizing predictive 14. Bell CC: Duration of plastic change in a modifiable efference
copy. Brain Res 1986, 369:29-36.
responses; third, the possible role of plasticity at inhibitory
15. Bell CC, Grant K: Corollary discharge effects and sensory
synapses; and fourth, how the adaptive functions are processing in the mormyromast regions of the mormyrid
integrated with other known functions of the circuitry such electrosensory lobe: II. cell types and corollary discharge
as lateral inhibition [37,38], gating [39,40], gain control plasticity. J Neurophysiol 1992, 68:859-875.
[41], and attention-like processes [42]. Behavioral demon- 16. Bell CC, Caputi A, Grant K: Physiology and plasticity of
morphologically identified cells in the mormyrid electrosensory
strations of the subtraction of predictable sensory features lobe. J Neurosci 1997, 17:6409-6422.
are also needed. Finally, work on other systems is required 17. Bastian J: Plasticity in an electrosensory system. I. General features
to determine whether knowledge of these electrosensory of dynamic sensory filter. J Neurophysiol 1996, 76:2483-2496.
structures can contribute to an understanding of other 18. Bastian J: Plasticity of feedback inputs in the apteronotid
cerebellum-like structures, such as the dorsal cochlear electrosensory system. J Exp Biol 1999, 202:1327-1337.
This article provides a good summary of both systems-level and cellular-level
nucleus or the cerebellum itself [10,43,44]. studies of negative image formation in the ELL of gymnotid fish. The article
also shows the similarities and differences between plasticity at parallel fiber
synapses in the dorsal molecular layer of the gymnotid ELL and plasticity at
Acknowledgements synapses made by descending fibers from a higher order electrosensory
I thank Joe Bastian, David Bodznick and Kirsty Grant for helpful nucleus (nucleus preeminentialis) in the ventral molecular layer.
discussions. My research was supported by grants from the National
Institutes of Mental Health (MH49792 and MH60996). 19. Bodznick D: The specificity of an adaptive filter that suppresses
unwanted reafference in electrosensory neurons of the skate
medulla. Biol Bull 1993, 185:312-314.
References and recommended reading
Papers of particular interest, published within the annual period of review, 20. Bodznick D, Montgomery JC, Carey M: Adaptive mechanisms in the
have been highlighted as: elasmobranch hindbrain. J Exp Biol 1999, 202:1357-1364.
This article provides a good summary of both systems-level and cellular-level
of special interest studies of negative image formation in the DON of elasmobranch fish. The
of outstanding interest article also shows that in the elasmobranch fish, corollary discharge signals
associated with motor commands, proprioceptive signals, and descending
1. Bullock TH, Heiligenberg W: Electroreception. New York: Wiley; 1986. electrosensory signals from higher centers can all serve as predictive signals
which can elicit negative images of predicted sensory patterns after suitable
2. Turner RW, Maler L, Burrows M (Eds): Electroreception and
periods of association.
Electrocommunication. J Exp Biol 1999, 202:1167-1458.
This entire issue of the Journal of Experimental Biology is devoted to 21. Montgomery JC, Bodznick D: An adaptive filter that cancels
electrosensory systems. The volume is comprehensive and well-edited, like self-induced noise in the electrosensory and lateral line
the book edited by Bullock and Heiligenberg [1] but is focused on recent mechanosensory systems of fish. Neurosci Lett 1994, 174:145-148.
work. It is a good place for the reader to review the many ways in which
electrosensory systems can contribute to our understanding of general 22. Montgomery JC: Noise cancellation in the electrosensory system
issues in sensory biology. of the thornback ray; common mode rejection of input produced
by the animals own ventilatory movement. J Comp Physiol 1984,
3. Moller P: Electric Fishes: History and Behavior. London: Chapman 155:103-111.
and Hall; 1995.
23. Bell CC, Caputi A, Grant K, Serrier J: Storage of a sensory pattern
4. Montgomery JC, Coombs S, Conley RA, Bodznick D: Hindbrain by anti-Hebbian synaptic plasticity in an electric fish. Proc Natl
sensory processing in lateral line, electrosensory, and auditory Acad Sci USA 1993, 90:4650-4654.
412
Memory-based expectations in electrosensory systems Bell 487
24. Ito M, Sakurai M, Tongroach P: Climbing fibre induced depression 34. Wang D, Maler L: In vitro plasticity of the direct feedback pathway
of both mossy fibre responsiveness and glutamate sensitivity of in the electrosensory system of Apteronotus leptorhynchus.
cerebellar Purkinje cells. J Physiol 1982, 324:113-134. J Neurophysiol 1997, 78:1882-1889.
25. Crepel F, Jaillard D: Pairing of pre- and postsynaptic activities in 35. Bastian J: Modulation of calcium-dependent postsynaptic
cerebellar Purkinje cells induces long-term changes in synaptic depression contributes to an adaptive sensory filter.
efficacy in vitro. J Physiol 1991, 432:123-141. J Neurophysiol 1998, 80:3352-3355.
26. Linden DJ, Dickenson MH, Smeyne M, Connor JA: A long-term 36. Nelson ME, Paulin MG: Neural simulations of adaptive reafference
depression of AMPA currents in cultured cerebellar Purkinje suppression in the elasmobranch electrosensory system. J Comp
neurons. Neuron 1991, 7:81-89. Physiol 1995, 177:723-736.
27. Bell CC, Han VZ, Sugawara S, Grant K: Synaptic plasticity in a 37. Bastian J, Court Right J, Crawford J: Commissural neurons of the
cerebellum-like structure depends on temporal order. Nature electrosensory lateral line lobe of Apteronotus leptorhynchus:
1997, 387:278-281. morphological and physiological characteristics. J Comp Physiol
1993, 173:257-274.
28. Bastian J: Plasticity in an electrosensory system III: Contrasting
properties of spatially segregated dendritic inputs. J Neurophysiol 38. Meek J, Hafmans TM, Han VZ, Bell CC, Grant K: Myelinated
1998, 79:1839-1857. dendrites in the mormyrid electrosensory lobe. J Comp Neurol
2001, 431:255-275.
29. Hebb DO: The Organization of Behavior. New York: Wiley; 1949.
39. Bell CC: Mormyromast electroreceptor organs and their afferents
30. Fregnac Y, Burke JP, Smith D, Friedlander ME: Temporal covariance in mormyrid electric fish: II. Intra-axonal recordings show initial
of pre- and postsynaptic activity regulates functional connectivity stages of central processing. J Neurophysiol 1990, 63:303-318.
in the visual cortex. J Neurophysiol 1994, 71:1403-1421.
40. Meyer JC, Bell CC: Behavioral measurements of sensory gating by
31. Han VZ, Grant G, Bell CC: Reversible associative depression and a corollary discharge. J Comp Physiol 1983, 151:401-406.
nonassociative potentiation at a parallel fiber synapse. Neuron
41. Bastian J: Gain control in the electrosensory system mediated by
2000, 27:611-622.
descending inputs to the electrosensory lateral line lobe.
We show associative synaptic depression at parallel fiber synapses in the
J Neurosci 1986, 6:553-562.
mormyrid ELL depends on activation of NMDA receptors and the occurrence
of a postsynaptic dendritic spike. This work also provides evidence that both 42. Berman NJ, Maler L: Neural architecture of the electrosensory
depression and potentiation have a presynaptic locus of expression. lateral line lobe: adaptations for coincidence detection, a sensory
searchlight and frequency-dependent adaptive filtering. J Exp Biol
32. Roberts PD, Bell CC: Computational consequences of temporally 1999, 202:1243-1253.
asymmetric learning rules: II. Sensory image cancellation.
J Comput Neurosci 2000, 9:67-83. 43. Paulin MG: The role of the cerebellum in motor control and
perception. Brain Behav Evol 1993, 41:39-50.
33. Bastian J: Plasticity in an electrosensory system. II. Postsynaptic
events associated with a dynamic sensory filter. J Neurophysiol 44. Devor A: Is the cerebellum like cerebellum-like structures? Brain
1996, 76:2497-2507. Res Rev 2000, 34:149-156.
413
414
Further
Spike TimingDependent
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
ANNUAL
REVIEWS
Click here for quick links to
Annual Reviews content online, Plasticity: A Hebbian
including:
by Cornell University on 10/30/11. For personal use only.
25
415
repeatedly or persistently takes part in ring it,
Contents some growth process or metabolic change takes
place in one or both cells such that As efciency,
INTRODUCTION . . . . . . . . . . . . . . . . . . 26
as one of the cells ring B, is increased (Hebb
CELLULAR MECHANISMS . . . . . . . . 27
1949). This postulate gained strong experimen-
tLTP Window . . . . . . . . . . . . . . . . . . . . . 28
tal support with the nding of long-term po-
tLTD Window. . . . . . . . . . . . . . . . . . . . . 28
tentiation (LTP) of synaptic transmission, ini-
STDP OF INHIBITION . . . . . . . . . . . . . 29
tially discovered in the hippocampus (Bliss &
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
26 Caporale Dan
416
Zhang et al. 1998) (Figure 1a, I). Such spike- a Excitatory to excitatory
timing-dependent plasticity (STDP) (Abbott &
Nelson 2000) has now been observed at excita- I II
tory synapses in a wide variety of neural cir-
cuits (Boettiger & Doupe 2001, Cassenaer & 50
Laurent 2007, Egger et al. 1999, Feldman 2000, 50
Froemke & Dan 2002, Sjostrom et al. 2001,
Tzounopoulos et al. 2004). Compared with the
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
CELLULAR MECHANISMS 25
417
receptor activation and the rise in postsy- magnitude of tLTP (Bi & Poo 1998, Froemke
naptic Ca2+ level (Bi & Poo 1998, Debanne et al. 2006, Magee & Johnston 1997). In the
et al. 1998, Feldman 2000, Magee & neocortex, a similar boosting of the BAP by the
BAP: back-
propagating action Johnston 1997, Markram et al. 1997, Sjostrom preceding EPSP is achieved by voltage-gated
potential et al. 2001, Zhang et al. 1998). Can this simple Na2+ channel activation in the distal dendrites
AP: action potential model for conventional LTP and LTD account (Stuart & Hausser 2001). Such nonlinear in-
for STDP, in particular for its temporal window teractions between the EPSP and BAP at short
VDCC: voltage-
dependent Ca2+ on the time scale of tens of milliseconds? positive intervals could explain the supralinear
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
28 Caporale Dan
418
through NMDA receptors in turn leads to through presynaptic CB1 endocannabinoid
tLTD. This model is supported by the obser- receptors is also required for tLTD for several
vations that tLTD induction requires activation excitatoryexcitatory (Bender et al. 2006;
mGluR:
of VDCCs (Bender et al. 2006, Bi & Poo 1998, Nevian & Sakmann 2006; Sjostrom et al. metabotropic
Froemke et al. 2005, Nevian & Sakmann 2006) 2003) and excitatoryinhibitory connections glutamate receptor
and that pairing EPSPs and BAPs at negative (Tzounopoulos et al. 2007), presumably by
intervals leads to sublinear summation of Ca2+ inhibiting presynaptic transmitter release. In
inux (Koester & Sakmann 1998, Nevian & Figure 2, we have outlined the major signaling
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
inactivation varied with dendritic location, mir- Balanced excitation and inhibition are crucial
roring the location dependence of the tLTD for normal brain functions (Shu et al. 2003) and
window at these synapses (Froemke et al. 2005).
In some other synapses, tLTD induction
does not depend on activation of postsynaptic
NMDA receptors (Bender et al. 2006, Egger
et al. 1999, Nevian & Sakmann 2006, Sjostrom tLTD tLTP
et al. 2003). These studies suggest a model
involving two coincidence detectors, with the CB1-R
NMDA receptor for tLTP and an additional co-
incidence detector for tLTD. In a two-detector
model proposed by Karmarkar & Buonomano
(2002), tLTD induction requires activation of
postsynaptic mGluRs (metabotropic glutamate
receptors) and Ca2+ inux through VDCCs, Glu
eCB
a premise supported by experimental ndings
NMDA-R
in the barrel cortex (Bender et al. 2006, Egger
mGlu-R
et al. 1999, Nevian & Sakmann 2006). Signal-
VDCC
ing through mGluRs can lead to phospholipase
C (PLC) activation, and Ca2+ inux through Ca 2+
419
for regulating experience-dependent develop- (Figure 1b, II). For these synapses, LTD
mental plasticity (Hensch 2005). Although the depends on postsynaptic NMDA receptor
strength of excitatory synapses can be modi- activation, Ca2+ inux, and endocannabinoid
ed through STDP, an important question is signaling, similar to the ndings at excitatory
whether and how correlated pre- and postsy- synapses onto pyramidal neurons (see previous
naptic activity affects inhibitory circuits. Inhi- section). Interestingly, synapses from the same
bition in a network depends on both the excita- presynaptic bers onto excitatory postsynaptic
tory synapses onto inhibitory neurons and the neurons (fusiform principal neurons) exhibit
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
inhibitory synapses themselves. Spike timing a STDP window similar to that of other
dependent plasticity has been studied at both excitatory-excitatory connections (Figure 1a,
of these synapses. I) (Tzounopoulos et al. 2004, 2007). This
by Cornell University on 10/30/11. For personal use only.
30 Caporale Dan
420
depends on postsynaptic Ca2+ inux through between multiple spikes was nonlinear, but the
the l-type Ca2+ channels, and presynaptic ex- specic forms of nonlinearity were different. In
pression was excluded because no change was cortical L2/3, the nonlinear interactions could
observed in the paired pulse ratio. In the hip- be accounted for by a suppression model, in
pocampus (Woodin et al. 2003), the changes in which the efcacy of later spikes in each train for
inhibitory postsynaptic current (IPSC) ampli- synaptic modication is reduced by the preced-
tude are due to changes in the Cl reversal po- ing spikes (Froemke & Dan 2002). This model
tential mediated by modication of the KCC2 accurately predicted the synaptic changes
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
K+ -Cl cotransporter, further indicating that induced by natural spike trains recorded in vivo
the expression is postsynaptic. in response to visual stimulation. In cultured
hippocampal neurons, the pre post pre
by Cornell University on 10/30/11. For personal use only.
421
depolarization. The combined dependence of gion for correlated activation of the associa-
synaptic modication on burst timing and fre- tional/commissural (A/C) bers and the mossy
quency can be accounted for by a model bers (Kobayashi & Poo 2004), although no
in which LTP wins over LTD, and only depression was observed. In both studies, the
the interactions between neighboring spikes width of the temporal window seems to scale
contribute to synaptic modication (Sjostrom with the duration of the spike bursts used in the
et al. 2001). In another study in L2/3 neu- induction protocol, and the changes in synap-
rons in rat visual cortical slices (Froemke et al. tic strength depend on the interburst interval
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
2006), pairing of pre- and postsynaptic bursts rather than the precise timing of individual
at high frequencies also favored LTP regard- spikes. Such burst timingdependent plastic-
less of the pre/post spike timing. However, ity rules may be functionally advantageous for
by Cornell University on 10/30/11. For personal use only.
systematic examination of the dependence of the circuits in which the information relevant
synaptic modication on both the number and for synaptic renement is contained in the tim-
the timing of pre- and postsynaptic spikes led ing of the bursts rather than that of individual
to a modied suppression model (Froemke spikes (Butts & Rokhsar 2001).
et al. 2006), which incorporates short-term Together, the studies described above indi-
depression of the presynaptic input (Zucker cate that the integration across multiple spike
& Regehr 2002) and frequency-dependent pairs for the induction of synaptic modication
attenuation of postsynaptic spikes (Spruston is highly nonlinear. The nature of the nonlin-
et al. 1995). Note that in both models described ear interaction is likely to depend on short-term
above, burst-induced synaptic modication is plasticity of the presynaptic neurons, on the
accounted for by integrating the contributions biophysical properties of the postsynaptic den-
of individual spike pairs. However, in some drites, and on the downstream signaling path-
synapses the learning rule for bursts seems to ways present in different cell types. Further
be completely different from that for individ- characterization of the diversity of integration
ual spikes (Birtoli & Ulrich 2004, Kampa et al. mechanisms for STDP will allow better under-
2006, Pike et al. 1999). standing of circuit remodeling induced by nat-
Although the above studies focused on ural patterns of neuronal activity.
synaptic modications induced by short bursts
lasting for tens of milliseconds, in some cir-
cuits bursts can last for hundreds of milliseconds DEPENDENCE ON DENDRITIC
to several seconds. In the developing retino- LOCATION
geniculate synapse, bursts of retinal ganglion In the central nervous system, each neu-
cells lasting seconds are believed to be critical ron may receive thousands of synaptic in-
for circuit renement (Butts & Rokhsar 2001). puts distributed throughout its dendritic tree.
Temporally overlapping pre- and postsynaptic The processing of each input depends on
bursts (interval within a window of 1 s) re- the dendritic location (Hausser & Mel 2003)
sult in synaptic potentiation, whereas nonover- owing to both the passive cable properties
lapping bursts cause a slight depression (Butts (Rall 1967) and the nonuniform distribution
et al. 2007). The degree of potentiation can be of active conductances (Migliore & Shepherd
predicted by a model in which LTP depends on 2002). Such location-dependent processing and
the interval but not the order between the pre- integration of synaptic inputs are believed
and postsynaptic bursts, and it increases linearly to be essential aspects of neuronal computa-
with the number of spikes in the burst. This is tion. Since a hallmark of STDP is its depen-
reminiscent of the classic correlation-based dence on the BAPs, which are strongly atten-
learning rule for synaptic plasticity (Stent uated along the dendrite (Stuart & Sakmann
1973). A strikingly similar window for burst 1994, Stuart et al. 1997b, Waters et al. 2005),
timing was found in the hippocampal CA3 re- synaptic modication is likely to vary with
32 Caporale Dan
422
Visual cortex Visual cortex Barrel cortex
L2/3 to L2/3 L2/3 to L5 L2/3 to L5
L5 to L5
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
by Cornell University on 10/30/11. For personal use only.
Figure 3
Dependence of STDP
on dendritic location.
dendritic location (Rao & Sejnowski 2001b). gests that distal tLTP requires cooperativity
Recent studies have examined the location de- among inputs.
pendence of both tLTP and tLTD. In L2/3 Two distinct effects have been reported for
of rat visual cortex, the magnitude of tLTP post pre pairing. In L2/3 pyramidal neu-
induced by pre post pairing of single rons, the width of the tLTD window mea-
spikes was smaller at intermediate-distal (100 sured with single spike pairing is broader for
150 m) than at proximal (<50 m) segments intermediate-distal than for proximal inputs
of the apical dendrite (Froemke et al. 2005) (Froemke et al. 2005) (Figure 3, left col-
(Figure 3, left column). This reduction of umn). This difference in width is correlated
tLTP amplitude is likely due to distance- with the window for AP-induced suppression
dependent attenuation of the BAP. In experi- of NMDA receptor activation, which sug-
ments with paired recordings from a L5 and a gests that the suppression plays an important
L2/3 pyramidal neuron or from two L5 neu- role in setting the tLTD window. In L2/3
rons (Sjostrom & Hausser 2006), burst pairing L5 synapses in rat barrel cortex, pairing sin-
at positive intervals led to LTP at the proxi- gle presynaptic spikes with postsynaptic bursts
mal synapses but LTD at the distal synapses at negative intervals leads to LTD at proxi-
(Figure 3, middle column). Similar location mal locations but LTP of distal inputs (Letzkus
dependence was also found among L2/3 to et al. 2006) (Figure 3, right column). This
L5 connections by pairing a single EPSP distal LTP could be explained by the induc-
with a postsynaptic burst at positive intervals tion of dendritic Ca2+ spikes by the later
(Letzkus et al. 2006) (Figure 3, right column). BAPs in the burst (Larkum et al. 1999a, Stuart
BAP boosting by subthreshold local dendritic et al. 1997a), such that the EPSP coincides
depolarization or extracellular stimulation re- with the peak postsynaptic depolarization. Lo-
covered tLTP at distal synapses (Letzkus et al. cal dendritic spikes can also play a prominent
2006, Sjostrom & Hausser 2006), which sug- role in coincidence detection in the neocortex
423
(Larkum et al. 1999b) and in LTP induction in At the cellular level, neuromodulators can
hippocampal CA1 (Golding et al. 2002) and the inuence AP backpropagation by modulat-
amygdala (Humeau & Luthi 2007). ing the activation and inactivation of var-
Comparison across these studies suggests ious active conductances ( Johnston et al.
that the degree of spatial variation of the learn- 1999). For example, agonists to muscarinic
ing rule depends on the dendritic morphology, ACh receptors can reduce spike attenua-
with quantitative changes over short distances tion during high-frequency bursts, probably
(e.g., dendrite of L2/3 neurons) and qualita- through reduction of Na+ channel inactivation
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
tive differences along long dendrites (e.g., apical ( Johnston et al. 1999, Tsubokawa & Ross 1997).
dendrites of L5 pyramids). Although the den- Both -adrenergic and muscarinic ACh recep-
dritic variations of STDP summarized above tor agonists can boost AP backpropagation by
by Cornell University on 10/30/11. For personal use only.
34 Caporale Dan
424
The timing and location of inhibitory in- with synaptic plasticity to mediate learning and
puts can also affect STDP. Somatic inhibition memory.
can prevent AP propagation through hyper- Changes in neuronal excitability have also
polarization and shunting (Miles et al. 1996, been examined in the context of STDP. In hip-
Tsubokawa & Ross 1996), which may pre- pocampal cell cultures (Ganguly et al. 2000)
clude STDP induction. In contrast, inhibitory and neocortical slices (Li et al. 2004), repeated
inputs to the dendrites have a variety of ef- pre post pairing of single spikes leads to
fects, from reducing dendritic depolarization LTP and to an enhancement of excitability and
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
through shunting to facilitating depolarization spike time reliability of the presynaptic neu-
and even spike generation (Gulledge & Stu- rons. Pairings at negative intervals result in
art 2003). An additional layer of complexity LTD and a reduction in presynaptic excitability
by Cornell University on 10/30/11. For personal use only.
is added by the fact that the strength and (Li et al. 2004). Mechanistically, these presy-
distribution of inhibition are developmentally naptic changes require NMDA receptor acti-
regulated (Hensch 2005), predicting that the vation and Ca2+ inux to the postsynaptic neu-
learning rule can vary considerably across de- ron, suggesting the involvement of retrograde
velopmental stages. In hippocampal CA1 pyra- signaling. On the presynaptic side, PKC is nec-
midal neurons, pairing single pre- and postsy- essary for the increase in excitability (Ganguly
naptic spikes at positive intervals leads to tLTP et al. 2000), and both PKC and PKA are re-
in juvenile (p9p14) but not in young (p22 quired for the decrease (Li et al. 2004). Inter-
p28) rats (Meredith et al. 2003). However, in estingly, the changes in excitability can be dis-
young rats tLTP can be rescued by replacing sociated from the changes in synaptic strength
the single postsynaptic spike with a burst or by because presynaptic blockage of PKC and/or
adding GABAA antagonists, suggesting that the PKA abolished the excitability changes with lit-
change in tLTP threshold might be due to a de- tle effect on the synaptic modications.
velopmental enhancement of inhibition in this Activity-dependent changes in intrinsic
circuit. membrane properties can also affect synaptic
integration (Magee & Johnston 2005). A
recent study examined the changes in spatial
PLASTICITY OF NEURONAL summation between two input pathways in
EXCITABILITY AND hippocampal CA1 neurons following STDP in-
SYNAPTIC INTEGRATION duction (Wang et al. 2003). Induction of tLTP
Information processing by neuronal net- in one pathway resulted in an increase in the
works depends not only on the connectivity linearity of spatial summation of the two path-
between neurons, but also on the intrinsic ways, whereas induction of tLTD produced the
conductances in each neuron that deter- opposite effect. The observed changes depend
mine its excitability and synaptic integration. on NMDA receptor activation and may be me-
Changes in neuronal excitability have been re- diated by modications of the Ih channels. In
ported in a variety of invertebrate and verte- another study in hippocampal CA1, LTP induc-
brate neural circuits during associative learn- tion by paired theta bursts causes an increase in
ing (Daoudal & Debanne 2003, Zhang & the linearity of temporal summation between
Linden 2003). At the cellular level, LTP in- the potentiated input and a neighboring input
duction by tetanic stimulation also leads to (Xu et al. 2006); the temporal specicity of this
increases in intrinsic excitability in both the effect varied with dendritic location. For distal
hippocampus and the cerebellum (Aizenman inputs, the increase in linearity is limited to
& Linden 2000, Armano et al. 2000, Bliss & EPSPs arriving within 5 ms of each other,
Gardner-Medwin 1973, Bliss & Lomo 1973). favoring summation of coincident inputs.
These activity-dependent changes in intrinsic In contrast, for proximal inputs the increase
neuronal properties may interact synergistically can be observed for EPSPs arriving within
425
a broader window of 20 ms. Such location- tion at a given orientation with cortical electri-
dependent modulation of synaptic integration cal stimulation leads to changes in the orienta-
may interact with the location dependence of tion map (Schuett et al. 2001). Electrical acti-
the STDP learning rule (see above) to further vation after the arrival of the visual input causes
enrich dendritic processing. expansion of the cortical representation of the
paired orientation, whereas the reverse order
causes a reduction. Whole-cell recordings in
STDP IN VIVO juvenile rat visual cortex showed that pairing
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
Whereas most of the early experiments on visual stimulation with single neuron spiking
STDP were conducted in slices and cell cul- leads to potentiation or depression of the vi-
tures, an increasing number of studies have sual response, depending on the order between
by Cornell University on 10/30/11. For personal use only.
begun to address the functional consequences the visual inputs and the postsynaptic spiking
of STDP in intact nervous systems. Neu- (Meliza & Dan 2006).
ral circuits in vivo exhibit both spontaneous STDP has also been described in other sen-
activity and sensory-evoked responses, mod- sory modalities in vivo. In the somatosensory
ulated by the behavioral states of the ani- cortex of anesthetized rats, pairing subthresh-
mal. Backpropagation of the APs may also be old whisker deections with postsynaptic spik-
more variable in vivo, as the neurons receive ing at negative intervals leads to LTD of the
barrages of excitatory and inhibitory inputs paired whisker ( Jacob et al. 2007). In an olfac-
(Destexhe et al. 2003). These factors could sig- tory circuit of the locusts (-lobe in the mush-
nicantly complicate the rules for synaptic plas- room body), pairing odor-induced synaptic ac-
ticity. How well does the STDP learning rule tivity with postsynaptic spiking results in robust
described in vitro apply to activity-dependent synaptic modications, with a temporal win-
synaptic modication in vivo? dow similar to those for vertebrate excitatory
synapses (Figure 1a, I) (Cassenaer & Laurent
2007).
Electrical Stimulation In the motor system, STDP has been
The rst demonstration of STDP in vivo demonstrated in human subjects. Pairing elec-
came from a study at the retinotectal pro- trical stimulation of a somatosensory afferent
jection in the developing Xenopus (Zhang nerve with transcranial magnetic stimulation
et al. 1998). Repetitive electrical stimulation (TMS) of the motor cortex leads to long-lasting
of the retinal ganglion cells within 20 ms be- changes in the motor-evoked potentials (MEPs)
fore tectal neuron spiking leads to LTP, whereas elicited by TMS (Wolters et al. 2003). The di-
pairings at negative intervals lead to LTD. Both rection and magnitude of the change depend on
LTP and LTD are NMDA receptor depen- the relative timing between the afferent stim-
dent, and the temporal window is similar to the ulation and the TMS within a window of tens
STDP windows measured in vitro (e.g., Bi & of milliseconds, comparable to the STDP win-
Poo 1998, Froemke & Dan 2002, Tzounopou- dows measured in vitro. The potentiation in-
los et al. 2004). In addition to the strength of duced by pairing at positive intervals can be
the retinotectal connection, the amplitude of blocked by NMDA receptor antagonists (Stefan
the tectal visual response can also be modied et al. 2002), and the depression at negative in-
by pairing visual stimulation with postsynaptic tervals is blocked by both NMDA receptor and
spiking (Mu & Poo 2006, Vislay-Meltzer et al. VDCC antagonists (Wolters et al. 2003), con-
2006). sistent with the pharmacological properties of
Plasticity with similar asymmetric windows STDP found in several studies (Bi & Poo 2001).
has also been demonstrated in the mammalian Wolters et al. (2005) also used a similar exper-
visual cortex. Optical imaging in the kitten vi- imental protocol to demonstrate STDP in hu-
sual cortex showed that pairing visual stimula- man somatosensory cortex.
36 Caporale Dan
426
Paired Sensory Stimulation rule. In the Xenopus tadpole, repeated presen-
tation of a moving bar in a given direction se-
Although electrical stimulation affords excel-
lectively potentiated the response to the con-
lent control of spike timing in the study of
ditioned direction, resulting in the emergence
STDP, an important question is whether the
of direction sensitivity in the tectal neurons
temporal requirements of this learning rule can
(Engert et al. 2002). Induction of direction se-
be satised under natural conditions, as spik-
lectivity through STDP has indeed been pre-
ing responses to sensory stimuli are known
dicted in a theoretical study (Rao & Sejnowski
to be highly variable (Shadlen & Newsome
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
427
place elds of hippocampal neurons are known FINAL REMARKS
to be dynamically modied as the animal navi- Over the past decade, the STDP learning rule
gates in a novel environment. During repeated has been demonstrated in a range of species
running of a linear track, the place elds of from insects to humans, and our understanding
both CA1 and CA3 cells are initially symmet- of its cellular mechanisms and functional impli-
rical, but they experience a gradual asymmet- cations has progressed signicantly. However,
ric expansion against the direction of loco- many questions remain unresolved.
motion (Lee et al. 2004; Mehta et al. 1997, Regarding the mechanism, it remains un-
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
2000). Simulation with a simple feedforward clear whether a single model can explain STDP
network model showed that this effect can be at different synapses or whether different neu-
explained by STDP (Blum & Abbott 1996, rons employ distinct molecular machineries
by Cornell University on 10/30/11. For personal use only.
Mehta et al. 2000). In the orientation domain, to achieve similar outcomes. Studies are only
Yu et al. (2006) recently reported a similar beginning to examine whether and how STDP
shift in head-direction tuning curves in thala- depends on several signaling events that have
mic head-direction cells as the animal runs in a been strongly implicated in conventional LTP
circular track. and LTD, including secretion of brain-derived
neurotrophic factor (BDNF) and nitric oxide
Sensory Deprivation (Mu & Poo 2006), activation of CaMKII
STDP may also play a role in other forms (Tzounopoulos et al. 2007) and phosphatases
of experience-dependent plasticity, even if the (Froemke et al. 2005), and modication
sensory inputs do not explicitly involve tim- and insertion/removal of AMPA receptors.
ing on the order of tens of milliseconds. In It would also be interesting to investigate
an experiment measuring the neural activity whether the type of NMDA receptor subunits
during sensory deprivation, rats were chroni- (NR2A/NR2B) and their synaptic location
cally implanted with electrode arrays to mon- play a role in STDP (Sjostrom et al. 2003), as
itor the spiking activity in L4 and L2/3 of has been suggested for LTP/LTD induced by
the barrel cortex during free-moving behav- HFS/LFS (Cull-Candy & Leszkiewicz 2004,
iors (Celikel et al. 2004). Stimulus deprivation Liu et al. 2004). In addition, whereas several
induced by trimming a single whisker, a ma- molecules have been proposed as coincidence
nipulation known to induce whisker map re- detectors at excitatory synapses (Figure 2),
organization, caused an immediate reversal of there is so far no candidate for inhibitory
the ring order and decreased correlation be- synapses. Furthermore, although postsynaptic
tween L4 and L2/3 neurons. Both of these Ca2+ signals are required for STDP in most
changes are known to drive tLTD in barrel cor- cell types, recent imaging experiments showed
tical slices (Feldman 2000), thus providing a that volume-averaged Ca2+ transients in the
plausible explanation for deprivation-induced dendritic spines are poorly correlated with the
LTD of L4 to L2/3 connections (Allen et al. direction of synaptic modication (Nevian &
2003). In addition to the somatosensory system, Sakmann 2006). Perhaps new techniques that
sensory deprivation induces circuit reorganiza- allow measurement of Ca2+ signals at a more
tion in the visual and auditory systems (Buono- microscopic scale (e.g., microdomains) will
mano & Merzenich 1998, Gilbert 1998). It shed new light on the cellular mechanisms of
would be interesting to test whether depriva- STDP.
tion in these modalities (e.g., monocular depri- To understand the functional consequences
vation of visual input) also induces changes in of STDP, an important factor to consider is the
the relative spike timing among neurons that high level of ongoing activity in vivo. Spon-
could cause the observed circuit modications taneous activity can signicantly affect mem-
through STDP. brane potential, conductance, and intracellular
38 Caporale Dan
428
Ca2+ levels, and in some cases it can boost AP tivity in early sensory circuits (Galan et al.
backpropagation in vivo (Waters & Helmchen 2006, Yao et al. 2007) or be replayed in the
2004). These effects will likely modulate the hippocampus during sleep ( Ji & Wilson 2007,
rules for synaptic modication. Furthermore, Louie & Wilson 2001, Nadasdy et al. 1999,
spontaneous postsynaptic spiking reduces the Ribeiro et al. 2004, Wilson & McNaughton
persistence of synaptic potentiation and depres- 1994). These reactivated patterns may serve to
sion (Zhou et al. 2003). An important question consolidate the transient effects of sensory stim-
is how experience-dependent synaptic modi- ulation into long-lasting circuit modications.
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
cations can persist in vivo in the face of the Characterization of neuronal plasticity at the
ongoing network activity. Recent studies have network level during natural behaviors is a cru-
suggested that sensory-evoked activity patterns cial step in understanding the neural basis for
by Cornell University on 10/30/11. For personal use only.
DISCLOSURE STATEMENT
The authors are not aware of any biases that might be perceived as affecting the objectivity of this
review.
LITERATURE CITED
Abbott LF, Nelson SB. 2000. Synaptic plasticity: taming the beast. Nat. Neurosci. 3(Suppl):117883
Aizenman CD, Linden DJ. 2000. Rapid, synaptically driven increases in the intrinsic excitability
of cerebellar deep nuclear neurons. Nat. Neurosci. 3:10911
Allen CB, Celikel T, Feldman DE. 2003. Long-term depression induced by sensory deprivation
during cortical map plasticity in vivo. Nat. Neurosci. 6:29199
Armano S, Rossi P, Taglietti V, DAngelo E. 2000. Long-term potentiation of intrinsic excitability
at the mossy ber-granule cell synapse of rat cerebellum. J. Neurosci. 20:520816
Artola A, Brocher S, Singer W. 1990. Different voltage-dependent thresholds for inducing long-
term depression and long-term potentiation in slices of rat visual cortex. Nature 347:6972
Artola A, Singer W. 1987. Long-term potentiation and NMDA receptors in rat visual cortex.
Nature 330:64952
Artola A, Singer W. 1993. Long-term depression of excitatory synaptic transmission and its rela-
tionship to long-term potentiation. Trends Neurosci. 16:48087
Bear MF, Singer W. 1986. Modulation of visual cortical plasticity by acetylcholine and nora-
drenaline. Nature 320:17276
Bell CC, Han VZ, Sugawara Y, Grant K. 1997. Synaptic plasticity in a cerebellum-like structure
depends on temporal order. Nature 387:27881
Bender VA, Bender KJ, Brasier DJ, Feldman DE. 2006. Two coincidence detectors for spike
timing-dependent plasticity in somatosensory cortex. J. Neurosci. 26:416677
Bi G, Poo M. 2001. Synaptic modication by correlated activity: Hebbs postulate revisited. Annu.
Rev. Neurosci. 24:13966
Bi GQ, Poo MM. 1998. Synaptic modications in cultured hippocampal neurons: dependence on
spike timing, synaptic strength, and postsynaptic cell type. J. Neurosci. 18:1046472
Birtoli B, Ulrich D. 2004. Firing mode-dependent synaptic plasticity in rat neocortical pyramidal
neurons. J. Neurosci. 24:493540
Bissiere S, Humeau Y, Luthi A. 2003. Dopamine gates LTP induction in lateral amygdala by
suppressing feedforward inhibition. Nat. Neurosci. 6:58792
429
Bliss TV, Collingridge GL. 1993. A synaptic model of memory: long-term potentiation in the
hippocampus. Nature 361:3139
Bliss TV, Gardner-Medwin AR. 1973. Long-lasting potentiation of synaptic transmission in the
dentate area of the unanaesthetized rabbit following stimulation of the perforant path. J.
Physiol. 232:35774
Bliss TV, Lomo T. 1973. Long-lasting potentiation of synaptic transmission in the dentate area
of the anaesthetized rabbit following stimulation of the perforant path. J. Physiol. 232:33156
Blum KI, Abbott LF. 1996. A model of spatial map formation in the hippocampus of the rat.
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
Buonomano DV, Merzenich MM. 1998. Cortical plasticity: from synapses to maps. Annu. Rev.
Neurosci. 21:14986
Butts DA, Kanold PO, Shatz CJ. 2007. A burst-based Hebbian learning rule at retinogeniculate
synapses links retinal waves to activity-dependent renement. PLoS Biol. 5:e61
Butts DA, Rokhsar DS. 2001. The information content of spontaneous retinal waves.
J. Neurosci. 21:96173
Cassenaer S, Laurent G. 2007. Hebbian STDP in mushroom bodies facilitates the synchronous
ow of olfactory information in locusts. Nature 448:70913
Celikel T, Szostak VA, Feldman DE. 2004. Modulation of spike timing by sensory deprivation
during induction of cortical map plasticity. Nat. Neurosci. 7:53441
Chapman PF, Kairiss EW, Keenan CL, Brown TH. 1990. Long-term synaptic potentiation in
the amygdala. Synapse 6:27178
Chevaleyre V, Takahashi KA, Castillo PE. 2006. Endocannabinoid-mediated synaptic plasticity
in the CNS. Annu. Rev. Neurosci. 29:3776
Clugnet MC, LeDoux JE. 1990. Synaptic plasticity in fear conditioning circuits: induction of
LTP in the lateral nucleus of the amygdala by stimulation of the medial geniculate body. J.
Neurosci. 10:281824
Couey JJ, Meredith RM, Spijker S, Poorthuis RB, Smit AB, et al. 2007. Distributed network
actions by nicotine increase the threshold for spike-timing-dependent plasticity in prefrontal
cortex. Neuron. 54:7387
Cull-Candy SG, Leszkiewicz DN. 2004. Role of distinct NMDA receptor subtypes at central
synapses. Sci. STKE 2004:re16
Daoudal G, Debanne D. 2003. Long-term plasticity of intrinsic excitability: learning rules and
mechanisms. Learn. Mem. 10:45665
Debanne D, Gahwiler BH, Thompson SM. 1998. Long-term synaptic plasticity between
pairs of individual CA3 pyramidal cells in rat hippocampal slice cultures. J. Physiol.
507(Pt. 1):23747
Destexhe A, Rudolph M, Pare D. 2003. The high-conductance state of neocortical neurons in
vivo. Nat. Rev. Neurosci. 4:73951
De Valois RL, De Valois KK. 1991. Vernier acuity with stationary moving Gabors. Vision Res.
31:161926
Dudek SM, Bear MF. 1993. Bidirectional long-term modication of synaptic effectiveness in the
adult and immature hippocampus. J. Neurosci. 13:291018
Egger V, Feldmeyer D, Sakmann B. 1999. Coincidence detection and changes of synaptic efcacy
in spiny stellate neurons in rat barrel cortex. Nat. Neurosci. 2:1098105
Engert F, Tao HW, Zhang LI, Poo MM. 2002. Moving visual stimuli rapidly induce direction
sensitivity of developing tectal neurons. Nature 419:47075
40 Caporale Dan
430
Feldman DE. 2000. Timing-based LTP and LTD at vertical inputs to layer II/III pyramidal cells
in rat barrel cortex. Neuron 27:4556
Froemke RC, Dan Y. 2002. Spike-timing-dependent synaptic modication induced by natural
spike trains. Nature 416:43338
Froemke RC, Poo MM, Dan Y. 2005. Spike-timing-dependent synaptic plasticity depends on
dendritic location. Nature 434:22125
Froemke RC, Tsay IA, Raad M, Long JD, Dan Y. 2006. Contribution of individual spikes in
burst-induced long-term synaptic modication. J. Neurophysiol. 95:162029
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
Fu YX, Djupsund K, Gao H, Hayden B, Shen K, Dan Y. 2002. Temporal specicity in the cortical
plasticity of visual space representation. Science 296:19992003
Fu YX, Shen Y, Gao H, Dan Y. 2004. Asymmetry in visual cortical circuits underlying motion-
by Cornell University on 10/30/11. For personal use only.
431
Ito M. 2005. Bases and implications of learning in the cerebellumadaptive control and internal
model mechanism. Prog. Brain Res. 148:95109
Jacob V, Brasier DJ, Erchova I, Feldman D, Shulz DE. 2007. Spike timing-dependent synaptic
depression in the in vivo barrel cortex of the rat. J. Neurosci. 27:127184
Ji D, Wilson MA. 2007. Coordinated memory replay in the visual cortex and hippocampus during
sleep. Nat. Neurosci. 10:1007
Johnston D, Hoffman DA, Colbert CM, Magee JC. 1999. Regulation of back-propagating action
potentials in hippocampal neurons. Curr. Opin. Neurobiol. 9:28892
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
Kampa BM, Clements J, Jonas P, Stuart GJ. 2004. Kinetics of Mg2+ unblock of NMDA receptors:
implications for spike-timing dependent synaptic plasticity. J. Physiol. 556:33745
Kampa BM, Letzkus JJ, Stuart GJ. 2006. Requirement of dendritic calcium spikes for induction
by Cornell University on 10/30/11. For personal use only.
42 Caporale Dan
432
Liu L, Wong TP, Pozza MF, Lingenhoehl K, Wang Y, et al. 2004. Role of NMDA receptor
subtypes in governing the direction of hippocampal synaptic plasticity. Science 304:102124
Liu QS, Pu L, Poo MM. 2005. Repeated cocaine exposure in vivo facilitates LTP induction in
midbrain dopamine neurons. Nature 437:102731
Louie K, Wilson MA. 2001. Temporally structured replay of awake hippocampal ensemble activity
during rapid eye movement sleep. Neuron 29:14556
Maejima T, Oka S, Hashimotodani Y, Ohno-Shosaku T, Aiba A, et al. 2005. Synaptically driven
endocannabinoid release requires Ca2+ -assisted metabotropic glutamate receptor subtype 1
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
Magee JC, Johnston D. 2005. Plasticity of dendritic function. Curr. Opin. Neurobiol. 15:33442
Malenka RC, Bear MF. 2004. LTP and LTD: an embarrassment of riches. Neuron 44:521
Markram H, Lubke J, Frotscher M, Sakmann B. 1997. Regulation of synaptic efcacy by coinci-
dence of postsynaptic APs and EPSPs. Science 275:21315
Martin SJ, Grimwood PD, Morris RG. 2000. Synaptic plasticity and memory: an evaluation of
the hypothesis. Annu. Rev. Neurosci. 23:649711
Mayer ML, Westbrook GL, Guthrie PB. 1984. Voltage-dependent block by Mg2+ of NMDA
responses in spinal cord neurones. Nature 309:26163
Mehta MR, Barnes CA, McNaughton BL. 1997. Experience-dependent, asymmetric expansion
of hippocampal place elds. Proc. Natl. Acad. Sci. USA 94:891821
Mehta MR, Quirk MC, Wilson MA. 2000. Experience-dependent asymmetric shape of hippocam-
pal receptive elds. Neuron 25:70715
Meliza CD, Dan Y. 2006. Receptive-eld modication in rat visual cortex induced by paired visual
stimulation and single-cell spiking. Neuron 49:18389
Meredith RM, Floyer-Lea AM, Paulsen O. 2003. Maturation of long-term potentiation induction
rules in rodent hippocampus: role of GABAergic inhibition. J. Neurosci. 23:1114246
Migliore M, Shepherd GM. 2002. Emerging rules for the distributions of active dendritic con-
ductances. Nat. Rev. Neurosci. 3:36270
Miles R, Toth K, Gulyas AI, Hajos N, Freund TF. 1996. Differences between somatic and dendritic
inhibition in the hippocampus. Neuron 16:81523
Mu Y, Poo MM. 2006. Spike timing-dependent LTP/LTD mediates visual experience-dependent
plasticity in a developing retinotectal system. Neuron 50:11525
Mulkey RM, Malenka RC. 1992. Mechanisms underlying induction of homosynaptic long-term
depression in area CA1 of the hippocampus. Neuron 9:96775
Nadasdy Z, Hirase H, Czurko A, Csicsvari J, Buzsaki G. 1999. Replay and time compression of
recurring spike sequences in the hippocampus. J. Neurosci. 19:9497507
Nevian T, Sakmann B. 2004. Single spine Ca2+ signals evoked by coincident EPSPs and back-
propagating action potentials in spiny stellate cells of layer 4 in the juvenile rat somatosensory
barrel cortex. J. Neurosci. 24:168999
Nevian T, Sakmann B. 2006. Spine Ca2+ signaling in spike-timing-dependent plasticity.
J. Neurosci. 26:1100113
Nishida S, Johnston A. 1999. Inuence of motion signals on the perceived position of spatial
pattern. Nature 397:61012
Nishiyama M, Hong K, Mikoshiba K, Poo MM, Kato K. 2000. Calcium stores regulate the polarity
and input specicity of synaptic modication. Nature 408:58488
Nowak L, Bregestovski P, Ascher P, Herbet A, Prochiantz A. 1984. Magnesium gates glutamate-
activated channels in mouse central neurones. Nature 307:46265
433
Pike FG, Meredith RM, Olding AW, Paulsen O. 1999. Rapid report: postsynaptic bursting is
essential for Hebbian induction of associative long-term potentiation at excitatory synapses
in rat hippocampus. J. Physiol. 518(Pt. 2):57176
Pu L, Liu QS, Poo MM. 2006. BDNF-dependent synaptic sensitization in midbrain dopamine
neurons after cocaine withdrawal. Nat. Neurosci. 9:6057
Rall W. 1967. Distinguishing theoretical synaptic potentials computed for different soma-
dendritic distributions of synaptic input. J. Neurophysiol. 30:113868
Ramachandran VS, Anstis SM. 1990. Illusory displacement of equiluminous kinetic edges. Per-
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
ception 19:61116
Rao RP, Sejnowski TJ. 2001a. Predictive learning of temporal sequences in recurrent neocortical
circuits. Novartis Found Symp. 239:20829; discussion 2940
by Cornell University on 10/30/11. For personal use only.
Rao RP, Sejnowski TJ. 2001b. Spike-timing-dependent Hebbian plasticity as temporal difference
learning. Neural Comput. 13:222137
Ribeiro S, Gervasoni D, Soares ES, Zhou Y, Lin SC, et al. 2004. Long-lasting novelty-induced
neuronal reverberation during slow-wave sleep in multiple forebrain areas. PLoS Biol. 2:E24
Rosenmund C, Feltz A, Westbrook GL. 1995. Calcium-dependent inactivation of synaptic NMDA
receptors in hippocampal neurons. J. Neurophysiol. 73:42730
Schuett S, Bonhoeffer T, Hubener M. 2001. Pairing-induced changes of orientation maps in cat
visual cortex. Neuron 32:32537
Seol GH, Ziburkus J, Huang S, Song L, Kim IT, et al. 2007. Neuromodulators control the polarity
of spike-timing-dependent synaptic plasticity. Neuron 55:91929
Shadlen MN, Newsome WT. 1994. Noise, neural codes and cortical organization. Curr. Opin.
Neurobiol. 4:56979
Shouval HZ, Bear MF, Cooper LN. 2002. A unied model of NMDA receptor-dependent bidi-
rectional synaptic plasticity. Proc. Natl. Acad. Sci. USA 99:1083136
Shu Y, Hasenstaub A, Badoual M, Bal T, McCormick DA. 2003. Barrages of synaptic activity
control the gain and sensitivity of cortical neurons. J. Neurosci. 23:10388401
Siegelbaum SA, Kandel ER. 1991. Learning-related synaptic plasticity: LTP and LTD. Curr. Opin.
Neurobiol. 1:11320
Sjostrom PJ, Hausser M. 2006. A cooperative switch determines the sign of synaptic plasticity in
distal dendrites of neocortical pyramidal neurons. Neuron 51:22738
Sjostrom PJ, Turrigiano GG, Nelson SB. 2001. Rate, timing, and cooperativity jointly determine
cortical synaptic plasticity. Neuron 32:114964
Sjostrom PJ, Turrigiano GG, Nelson SB. 2003. Neocortical LTD via coincident activation of
presynaptic NMDA and cannabinoid receptors. Neuron 39:64154
Snowden RJ. 1998. Shifts in perceived position following adaptation to visual motion. Curr. Biol.
8:134345
Softky WR, Koch C. 1993. The highly irregular ring of cortical cells is inconsistent with temporal
integration of random EPSPs. J. Neurosci. 13:33450
Spruston N, Schiller Y, Stuart G, Sakmann B. 1995. Activity-dependent action potential invasion
and calcium inux into hippocampal CA1 dendrites. Science 268:297300
Stanton PK, Sejnowski TJ. 1989. Associative long-term depression in the hippocampus induced
by Hebbian covariance. Nature 339:21518
Stefan K, Kunesch E, Benecke R, Cohen LG, Classen J. 2002. Mechanisms of enhancement of
human motor cortex excitability induced by interventional paired associative stimulation. J.
Physiol. 543:699708
Stent GS. 1973. A physiological mechanism for Hebbs postulate of learning. Proc. Natl. Acad. Sci.
USA 70:9971001
44 Caporale Dan
434
Stuart G, Schiller J, Sakmann B. 1997a. Action potential initiation and propagation in rat neocor-
tical pyramidal neurons. J. Physiol. 505(Pt. 3):61732
Stuart G, Spruston N, Sakmann B, Hausser M. 1997b. Action potential initiation and backprop-
agation in neurons of the mammalian CNS. Trends Neurosci 20:12531
Stuart GJ, Hausser M. 2001. Dendritic coincidence detection of EPSPs and action potentials. Nat.
Neurosci. 4:6371
Stuart GJ, Sakmann B. 1994. Active propagation of somatic action potentials into neocortical
pyramidal cell dendrites. Nature 367:6972
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
Tong G, Shepherd D, Jahr CE. 1995. Synaptic desensitization of NMDA receptors by calcineurin.
Science 267:151012
Tsubokawa H, Ross WN. 1996. IPSPs modulate spike backpropagation and associated [Ca2+ ]i
by Cornell University on 10/30/11. For personal use only.
435
Xu NL, Ye CQ, Poo MM, Zhang XH. 2006. Coincidence detection of synaptic inputs is facilitated
at the distal dendrites after long-term potentiation induction. J. Neurosci. 26:30029
Yang SN, Tang YG, Zucker RS. 1999. Selective induction of LTP and LTD by postsynaptic [Ca2+ ]i
elevation. J. Neurophysiol. 81:78187
Yao H, Dan Y. 2001. Stimulus timing-dependent plasticity in cortical processing of orientation.
Neuron 32:31523
Yao H, Shen Y, Dan Y. 2004. Intracortical mechanism of stimulus-timing-dependent plasticity in
visual cortical orientation tuning. Proc. Natl. Acad. Sci. USA 101:508186
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
Yao H, Shi L, Han F, Gao H, Dan Y. 2007. Rapid learning in cortical coding of visual scenes. Nat.
Neurosci. 10:77278
Yu X, Yoganarasimha D, Knierim JJ. 2006. Backward shift of head direction tuning curves of the
by Cornell University on 10/30/11. For personal use only.
46 Caporale Dan
436
Annual Review of
Neuroscience
Function
Curtis C. Bell, Victor Han, and Nathaniel B. Sawtell p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 1
Spike TimingDependent Plasticity: A Hebbian Learning Rule
by Cornell University on 10/30/11. For personal use only.
437
Brain Circuits for the Internal Monitoring of Movements
Marc A. Sommer and Robert H. Wurtz p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 317
Wnt Signaling in Neural Circuit Assembly
Patricia C. Salinas and Yimin Zou p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 339
Habits, Rituals, and the Evaluative Brain
Ann M. Graybiel p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 359
Mechanisms of Self-Motion Perception
Kenneth H. Britten p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 389
Mechanisms of Face Perception
Doris Y. Tsao and Margaret S. Livingstone p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 411
Annu. Rev. Neurosci. 2008.31:25-46. Downloaded from www.annualreviews.org
Indexes
Errata
vi Contents
438
Available online at www.sciencedirect.com
This review focuses on recent progress in understanding has been made in a number of model systems, from
mechanisms for filtering self-generated sensory signals in primates to crickets [4]. This review focuses on recent
cerebellum-like circuits in fish and mammals. Recent in vitro studies that have contributed to understanding how
studies in weakly electric gymnotid fish have explored the synaptic plasticity contributes to sensory filtering in sen-
interplay among anti-Hebbian plasticity, synaptic dynamics, sory structures with circuitry resembling the cerebellum
and feedforward inhibition in canceling self-generated that are found at the first stages of electrosensory proces-
electrosensory inputs. Studies of the mammalian dorsal sing in fish, auditory and auditory processing in mammals.
cochlear nucleus have revealed multimodal integration and
anti-Hebbian plasticity, suggesting that this circuit may A combination of in vitro, in vivo, and computational
adaptively filter incoming auditory information. In vivo studies in studies conducted over three decades has led to a com-
weakly electric mormryid fish suggest a key role for granule cell pelling account of how anti-Hebbian synaptic plasticity in
coding in sensory filtering. The clear links between synaptic cerebellum-like circuits in fish acts to predict and cancel
plasticity and systems level sensory filtering in cerebellum-like the electrosensory consequences of the animals own
circuits may provide insights into hypothesized adaptive behavior [57]. In what appears to be a remarkable
filtering functions of the cerebellum itself. case of convergent evolution, very similar predictive
mechanisms have been revealed in cerebellum-like cir-
Address cuits in three distinct groups of fish in which electrosen-
Department of Neuroscience and Kavli Institute for Brain Science, sory systems likely evolved independently [8]. These
Columbia University Medical Center, New York, NY 10032, United include the dorsal octavolateral nucleus (DON) in elas-
States
mobranch fish (the group that includes sharks, skates, and
Corresponding author: Sawtell, Nathaniel B (ns2635@columbia.edu) rays), the electrosensory lobe (ELL) of weakly electric
mormyrid fish from Africa and the ELL of weakly electric
gymnotid fish from South America.
Current Opinion in Neurobiology 2011, 21:602608
439
Neural mechanisms for filtering self-generated sensory signals Requarth and Sawtell 603
Figure 1
before plasticity
parallel fiber inputs: individual PF inputs
membrane potential
corollary discharge
proprioception
other sensory modalities
summed PF input
membrane potential
(PF input+ sensory input)
site of anti-Hebbian
sensory input
plasticity
after plasticity
individual PF inputs
membrane potential
sensory input
(external+self-generated signals) filtered output
summed PF input
(negative image)
sensory cancellation
(b) mormyrid ELL (c) gymnotid ELL (d) dorsal cochlear nucleus
post before pre before post before pre before post before pre before post before pre before
pre post pre post pre post pre post
CWC
-100 ms 100 ms
Sensory filtering and synaptic plasticity in cerebellum-like circuits. (a) Left, schematic of a generic cerebellum-like circuit. Principal cells integrate
peripheral sensory input (orange) with predictive signals conveyed by a granule cellparallel fiber system (blue). Right, schematic representation of the
formation of a negative image (blue line in bottom panel) of a predictable temporal pattern of sensory input (orange). Negative images are formed via bi-
directional changes in the strength of parallel fiber synapses (blue) according to an anti-Hebbian learning rule (see text). After plasticity, the addition of
negative images (blue) with sensory input (orange) results in the cancellation of predictable features of the sensory input (black). (b) Circuitry and anti-
Hebbian learning rule at parallel fiber synapses onto GABAergic medium ganglion (MG) cells in the mormyrid ELL. Anti-Hebbian plasticity has been
observed in efferent cells as well, though the learning rule has not been studied in detail. (c) Circuitry and anti-Hebbian learning rule at parallel fiber
synapses onto glutamatergic efferent cells in the gymnotid ELL. Long-term depression is induced by correlated presynaptic and postsynaptic bursts
irrespective of relative timing. (d) Circuitry and learning rules at parallel fiber synapses onto GABAergic cartwheel cells (CWC) and glutamatergic
efferent cells in the DCN. PF synapses onto CWC synapses exhibit timing-dependent anti-Hebbian plasticity. PF synapses onto efferent cells exhibit
timing-dependent Hebbian plasticity that can be converted to anti-Hebbian plasticity by activating cholinergic inputs.
mechanisms that strengthen synapses between neurons synapses. Conversely, predictable decreases in principal
that are active together, the plasticity found in cerebellum- cell firing are opposed by increases in parallel fiber synaptic
like circuits is anti-Hebbian synapses that are active strength. The mechanism is indifferent to the nature of the
together are weakened. In the context of cerebellum-like signals conveyed by parallel fibers as long as they reliably
circuitry, this learning rule provides a simple and powerful predict changes in principal cell output. On the basis of
mechanism for filtering out self-generated signals. these results it was suggested that cerebellum-like circuitry
Increases in principal cell firing that occur together with could operate as an adaptive filter by continually generat-
(i.e., can be predicted by) parallel fiber input are opposed ing and updating sensory predictions based on associations
and eventually canceled by weakening of parallel fiber between sensory and motor signals conveyed by parallel
440
604 Sensory and Motor Systems
fibers and current sensory inputs, then subtracting these long-lasting depression is induced when presynaptic
predictions from the neural response of principal cells. spike bursts are temporally correlated with postsynaptic
Similar adaptive filtering mechanisms have been proposed bursts (within 50 ms) irrespective of temporal order. At
for the mammalian cerebellar cortex in the context of present, the functional significance of timing-dependent
motor control [15,16]. anti-Hebbian plasticity in the mormryid ELL versus
correlation-based anti-Hebbian plasticity in the gymnotid
Cellular mechanisms for negative image ELL is unclear, though it may relate to precisely timed
formation motor corollary discharge signals which are present in the
Anti-Hebbian plasticity has been characterized in some mormyrid ELL but not in the gymnotid ELL.
detail in a brain slice preparation of the mormryid ELL
[17,18] (Figure 1b). Notably, the sign and magnitude of The Harvey-Girard et al. study also revealed another
the changes in parallel fiber strength depend critically on important difference in plasticity mechanisms in morm-
the timing of the excitatory postsynaptic potential (EPSP) ryids and gymnotids. Associative depression in the morm-
evoked by parallel fiber stimulation relative to a postsyn- ryid ELL is reversed by a long-lasting nonassociative
aptic dendritic spike. This was among the first demon- potentiation induced by parallel fiber stimulation alone.
strations of spike timing-dependent plasticity in the No such mechanism for reversing burst-induced depres-
vertebrate brain. Parallel fiber EPSPs that preceded den- sion was found in the gymnotid ELL. The lack of a
dritic spikes by 50 ms or less were depressed, while potentiation mechanism is troubling because in vivo
EPSPs occurring at all other delays were potentiated. negative images are characterized by both decreased
Because potentiation also occurred when parallel fiber responses that oppose predictable increases in principal
EPSPs were evoked at sufficiently high rates (>0.5 Hz) cell output (presumably due to associative depression at
without a postsynaptic spike, the potentiation was termed parallel fiber synapses) as well as increased responses that
nonassociative. The associative synaptic depression oppose predictable decreases in principal cell output.
required activation of NMDA-type glutamate receptors How are increased responses explained given the lack
and changes in postsynaptic calcium. The nonassociative of a mechanism for potentiation in the gymnotid ELL? As
potentiation reversed the depression and vice versa, with the authors point out, a mundane but important possib-
both forms of plasticity appearing to share a presynaptic ility is that potentiation mechanisms exist but were
locus of expression. Modeling studies demonstrated that simply not engaged under the conditions of the exper-
the learning rule observed in vitro could account for the iments. As a case in point, it took many years to discover a
formation of negative images observed in vivo [19]. form of parallel fiber potentiation that reversed parallel
fiber long-term depression found in the mammalian
In vivo studies have revealed negative images with similar cerebellum [22].
properties in the mormryid ELL, the gymnotid ELL, and
the elasmobranch DON. However, until recently, less The authors suggest another, more intriguing, possibility
was known about the cellular mechanisms for plasticity in for increased responses motivated by recent in vitro
the latter two structures. A recent study combining in vivo studies in the gymnotid ELL by Lewis et al. [23,24].
electrophysiology and immunohistochemistry provides Lewis et al. utilized a data-based model of parallel fiber
strong evidence for a role for NMDA receptors in nega- synaptic dynamics [23] and dynamic clamp to explore
tive image formation in the DON [20]. Pharmacological how changes in the balance of excitatory and inhibitory
blockade of NMDA receptors reversibly blocked nega- inputs to ELL principal cells contribute to negative
tive image formation in response to pairing artificial image formation [24]. Lewis et al. found that the effects
electrosensory stimuli with either electrical stimulation of increasing excitatory parallel fiber input rate on prin-
of parallel fibers or natural patterns of parallel fiber cipal cell firing were nonmonotonic, such that higher
activation related to the fishs ventilatory cycle. Venti- input rates actually inhibited principal cell firing. The
latory movements are a significant source of self-gener- shift from net excitation to inhibition was due to the
ated electrosensory input for elasmobranchs and parallel interplay between synaptic dynamics and feedforward
fibers are known to convey proprioceptive and motor inhibition to principal cells via molecular layer inter-
corollary discharge signals related to ventilation. neurons. Harvey-Girard et al. extended this work by
showing that a model incorporating parallel fiber synaptic
A recent in vitro study by Harvey-Girard et al. has pro- dynamics, feedforward inhibition, and associative synap-
vided the first detailed report of synaptic plasticity mech- tic depression could account for negative images observed
anisms in the gymnotid ELL [21] (Figure 1c). As in the in vivo. Although experiments are needed to validate the
mormyrid ELL and the DON, plasticity in the gymnotid model, attempts to understand the complex interactions
ELL is anti-Hebbian and depends on NMDA receptors. among long-term synaptic plasticity, short-term synaptic
Interestingly, plasticity in the gymnotid ELL does not dynamics and inhibitory circuitry are of general import-
depend on the relative timing of single presynaptic and ance. Such interactions are likely to occur in many circuits
postsynaptic spikes as in the mormryid ELL. Instead a where plasticity is present but are, in general, poorly
441
Neural mechanisms for filtering self-generated sensory signals Requarth and Sawtell 605
understood. Cerebellum-like circuits in weakly electric cells [29,30,31]. Using dual recordings to examine inter-
fish provide a system in which to explore the significance actions among cartwheel cells, fusiform cells, and parallel
of multiple sites and forms of synaptic plasticity in the fibers, Roberts and Trussell [31] made the surprising
context of well-defined systems level sensory filtering observation that parallel fiber connectivity was extremely
functions. sparse. Minimal stimulation showed that neighboring
cells were rarely activated by the same parallel fiber
Cerebellum-like circuitry and anti-Hebbian synapses, suggesting that feedforward inhibition may
plasticity in the mammalian auditory system not be the dominant mode of cartwheel cell action.
Recent studies have also provided insights into functional
circuitry and synaptic plasticity in the dorsal cochlear Although in vivo evidence for sensory filtering in the DCN
nucleus (DCN) a cerebellum-like circuit at the first has not yet been obtained, in vitro studies have revealed
stage of auditory processing in mammals(Figure 1d). parallel fiber synaptic plasticity mechanisms remarkably
DCN principal cells integrate auditory nerve input with similar to those in the mormyrid ELL [3234] (Figure 1b
parallel fiber input conveying both auditory and nonau- and d). In both structures plasticity is anti-Hebbian,
ditory information from a variety of brain regions [25]. NMDA receptor-dependent, presynaptically expressed,
These signals could be used to filter incoming auditory and apparently reversed by a timing-independent form
inputs, as in cerebellum-like circuits in fish. One possib- of parallel fiber synaptic potentiation. Anti-Hebbian
ility is that proprioceptive and/or motor corollary dis- plasticity is also found at analogous sites in the two struc-
charge information about the position and movements tures at parallel fiber synapses onto the spiny dendrites
of the head and external ears (conveyed by parallel fibers) of interneurons (cartwheel cells in the DCN and medium
may be needed to properly interpret incoming auditory ganglion cells in mormyrid ELL) that inhibit nearby
nerve input. The DCN is believed to play a role in glutamatergic output neurons. Interestingly, parallel fiber
processing spectral cues for localizing sound sources in synapses onto fusiform cells exhibit a Hebbian form of
elevation. Changes in spectral cues could reflect move- spike timing-dependent plasticity, i.e. presynaptic inputs
ment of an external sound source, movement of the that precede a postsynaptic spike result in long-term
animal itself, or both. Hence, in order to localize sounds, potentiation. The mechanisms underlying this site-specific
auditory information must be combined with information plasticity have been studied in some detail [3335] and
about the position and movements of the animal. Given depend in part on differences in endocannabinoid sig-
its distinctive cerebellum-like circuitry and plasticity (see naling at cartwheel versus fusiform cell synapses.
below), this computation may begin at the first stage of
auditory processing in the DCN. Consistent with this Hebbian plasticity at parallel fiberfusiform cell synapses
possibility, principal cells in the DCN of cats are known is puzzling from the standpoint of a sensory filtering
to respond strongly to somatosensory stimulation, in- function for the DCN. Such plasticity would apparently
cluding movements of the external ear [26]. A recent strengthen the effects of predictable signals, rather than
study demonstrated that brief electrical stimulation of the canceling them as occurs in cerebellum-like structures in
somatosensory dorsal column nucleus (a major input to fish. A recent study provides a possible resolution, show-
DCN granule cells) elicits complex and long-lasting ing that Hebbian long-term potentiation at parallel fiber
alterations in the responses of DCN neurons to sound synapses onto fusiform cells can be converted to anti-
[27]. Hebbian long-term depression if cholinergic neuromodu-
latory inputs are activated [36]. The switch was shown
Another related possibility is that parallel fibers convey to depend on interactions between postsynaptic M1/M3
signals related to the animals own chewing, vocalization, muscarinic acetylcholine receptor activation and endo-
or respiration that could be used to predict and cancel cannabinoid signaling pathways. The authors speculate
their auditory consequences. Consistent with this, tract that neuromodulation could switch the polarity of
tracing, immunohistocehmical, and in vivo electrophysio- plasticity in a context-dependent manner and that both
logical studies of multimodal integration in the DCN of Hebbian and anti-Hebbian plasticity may play function-
rodents suggest that signals related to chewing, respir- ally important roles. The presence of anti-Hebbian
ation, and vocalization may indeed reach DCN principal plasticity at parallel fiber synapses onto both cartwheel
cells via the granule cellparallel fiber system [28]. cell and fusiform cell synapses strongly motivates in vivo
tests of adaptive sensory filtering in the DCN.
Responses to nonauditory inputs in DCN appear to be
mediated in large part by cartwheel cells a network of Significance of granule cell coding for sensory
inhibitory interneurons that receive the large majority of filtering
parallel fiber inputs. Cartwheel cells powerfully inhibit In both cerebellum-like circuits and the cerebellum itself,
fusiform cells, the output cells of the DCN. A number of diverse sensory and motor signals conveyed from various
recent in vitro studies have shed light on the interactions brain regions by mossy fibers are recoded in granule cells
among cartwheel cells as well as their effects on fusiform before being relayed to principal cells or Purkinje cells. Yet
442
606 Sensory and Motor Systems
due to the difficulty of obtaining in vivo recordings from to the combinations of signals encoded by individual
these small, densely packed cells, little is known about granule cells, i.e. responses that were specific to particular
their function. Granule cell circuitry is extremely similar in tail positions and particular times relative to motor com-
cerebellum-like circuits and the cerebellum in terms of mands. Hence, integration of sensory and motor signals in
the granule cells themselves and the inhibitory Golgi cells granule cells appears to provide a neural substrate for
and excitatory unipolar brush cells that shape granule cell generating predictions about the sensory consequences of
responses to mossy fiber inputs. Hence insights regarding motor commands that take proprioceptive context into
the importance of granule cell circuitry for sensory filtering account. Such predictions are likely to be of general use.
in fish may be relevant for understanding functions of Predicting the consequences of a reaching movement, for
cerebellar granule cell circuitry more generally. example, requires both the issued motor commands and
information about the starting position of the hand and
Influential computational theories of cerebellar function the target.
developed independently by Marr and Albus in the 1970s
posit that the large number of cerebellar granule cells Future directions
(around 70 billion in humans, or approximately 75% of the Results described here span multiple levels of analysis and
total number of neurons in the entire brain) serve to hence provide many avenues for future studies from
sparsely recode mossy fiber inputs, such that only a detailed descriptions of molecular plasticity mechanisms,
small fraction of the granule cell population is active to the identification of behavioral correlates for sensory
for a given pattern of mossy fiber input [37,38]. Sparse filtering in cerebellum-like circuits in fish, to in vivo
coding is expected to increase the capacity of downstream studies of sensory filtering in the DCN. Among the most
Purkinje cells to discriminate between different mossy intriguing possibilities is that the relatively detailed
fiber input patterns through associative plasticity at paral- insights into sensory filtering mechanisms available from
lel fiber synapses and is hypothesized to be important for studies of cerebellum-like circuits could provide general
cerebellum-dependent motor learning. One way in which insights into neural mechanisms for predicting sensory
sparse coding could be accomplished is if different mossy events. Prediction is critical for a wide range of sensory,
fiber inputs converged onto individual granule cells and if motor, and cognitive functions. For example, predictions
simultaneous activation of multiple inputs was required about the sensory consequences of motor commands could
to evoke action potentials in granule cells. Granule cell be used not only to cancel self-generated sensory signals,
responses have only begun to be explored in the mam- as described above, but also for guiding rapid movements
malian cerebellum in vivo, and the question of sparse in the face of sensory feedback that is delayed and/or
coding remains open [39]. noisy. Such predictive mechanisms, termed forward
models, are widely discussed in the context of motor
A recent study in the mormryid ELL provided a clear control [41]. Interestingly, the generation of such forward
example of multimodal integration in individual granule models is widely believed to occur in the cerebellum [42
cells, suggestive of sparse coding. The study also linked 44]. This raises the additional exciting possibility that
multimodal responses in granule cells to the ability of understanding mechanisms for sensory filtering in cerebel-
ELL principal cells to generate negative images via anti- lum-like circuits will be a direct source of insight into the
Hebbian plasticity [40]. Using in vivo whole-cell record- function of the cerebellum itself.
ings in awake, paralyzed fish, it was shown that a subset of
granule cells integrate excitatory input from two distinct Acknowledgements
classes of mossy fibers, one conveying proprioceptive This work was supported by a grant from the National Science Foundation
(IOS-1025849), the Kavli Institute for Brain Science, and the Alfred P. Sloan
information about body position and the other corollary Foundation.
discharge signals related to the motor command to dis-
charge the electric organ. Rates of spontaneous action References and recommended reading
potential firing in granule cells were extremely low, and in Papers of particular interest, published within the period of review,
have been highlighted as:
most cells activation of corollary discharge signals alone
did not evoke action potentials. However, if propriocep- of special interest
tive and corollary discharge signals were activated of outstanding interest
together some granule cells did fire, thereby encoding
specific information about the conjunction of a motor 1. Churchland PS: Self-representation in nervous systems.
Science 2002, 296:308-310.
command and a particular body configuration. These
results are suggestive of sparse coding, though sampling 2. Sperry RW: Neural basis of the spontaneous optokinetic
response produced by visual inversion. J Comp Physiol Psychol
from a much larger number of cellse.g., using calcium 1950, 43:482-489.
imagingwould be required to show this directly. 3. von Holst E, Mittelstaedt H: The reafference principle.
Naturwissenschaften 1950, 37:464-476.
The author went on to show that principal neurons in 4. Crapse TB, Sommer MA: Corollary discharge across the animal
ELL were capable of generating negative images specific kingdom. Nat Rev Neurosci 2008, 9:587-600.
443
Neural mechanisms for filtering self-generated sensory signals Requarth and Sawtell 607
5. Bell C, Bodznick D, Montgomery J, Bastian J: The generation and electrosensory reafference suppression. J Exp Biol 2007,
subtraction of sensory expectations within cerebellum-like 210:4437-4447.
structures. Brain Behav Evol 1997, 50:17-31. This in vitro study of the gymnotid ELL uses dynamic clamp and a model
of parallel fiber synaptic dynamics to demonstrate that the net effect of
6. Bell CC: Memory-based expectations in electrosensory parallel fiber activation on principal cell firing shifts from excitation to
systems. Curr Opin Neurobiol 2001, 11:481-487. inhibition as parallel fiber input rate increases. The authors suggest that
this shift could contribute to the formation of negative images appropriate
7. Montgomery JC, Coombs S, Conley RA, Bodznick D: Hindbrain
to cancel both predictable increases and decreases in principal cell firing.
sensory processing in lateral line, electrosensory, and
auditory systems: a comparative overview of anatomical and 25. Oertel D, Young ED: Whats a cerebellar circuit doing in the
functional similarities. Audit Neurosci 1995, 1:207-231. auditory system? Trends Neurosci 2004, 27:104-110.
8. Bell CC: Evolution of cerebellum-like structures. Brain Behav 26. Kanold PO, Young ED: Proprioceptive information from the
Evol 2002, 59:312-326. pinna provides somatosensory input to cat dorsal cochlear
nucleus. J Neurosci 2001, 21:7848-7858.
9. Bell CC, Han V, Sawtell NB: Cerebellum-like structures and their
implications for cerebellar function. Annu Rev Neurosci 2008, 27. Kanold PO, Davis KA, Young ED: Somatosensory context alters
31:1-24. auditory responses in the cochlear nucleus. J Neurophysiol
2010, 105:1063-1070.
10. Bastian J: Plasticity in an electrosensory system. I. General
This study uses in vivo extracellular recordings and electrical stimulation
features of dynamic sensory filter. J Neurophysiol 1996,
of the dorsal column nucleus to explore the interactions between auditory
76:2483-2496.
stimuli and somatosensory input in the DCN of cats. A single stimulus
11. Bell CC: An efference copy modified by reafferent input. pulse evoked large and long-lasting effects on the responses of DCN
Science 1981, 214:450-453. neuron to sound, consisting of both increases and decreases in evoked
firing. The authors provide evidence that inhibition from cartwheel cells
12. Bell CC, Caputi A, Grant K, Serrier J: Storage of a sensory may mediate some of the observed effects.
pattern by anti-Hebbian synaptic plasticity in an electric fish.
Proc Natl Acad Sci 1993, 90:4650-4654. 28. Shore SE, Zhou J: Somatosensory influence on the cochlear
nucleus and beyond. Hear Res 2006, 216217:90-99.
13. Bodznick D, Montgomery JC, Carey M: Adaptive mechanisms in
the elasmobranch hindbrain. J Exp Biol 1999, 202:1357-1364. 29. Mancilla JG, Manis PB: Two distinct types of inhibition
mediated by cartwheel cells in the dorsal cochlear nucleus. J
14. Sawtell NB, Williams A: Transformations of electrosensory Neurophysiol 2009, 102:1287-1295.
encoding associated with an adaptive filter. J Neurosci 2008,
28:1598-1612. 30. Roberts MT, Bender KJ, Trussell LO: Fidelity of complex spike-
mediated synaptic transmission between inhibitory
15. Dean P, Porrill J, Stone JV: Decorrelation control by the interneurons. J Neurosci 2008, 28:9440-9450.
cerebellum achieves oculomotor plant compensation in
simulated vestibulo-ocular reflex. Proc Biol Sci 2002, 269:1895- 31. Roberts MT, Trussell LO: Molecular layer inhibitory
1904. interneurons provide feedforward and lateral inhibition in the
dorsal cochlear nucleus. J Neurophysiol 2010, 104:2462-2473.
16. Fujita M: Adaptive filter model of the cerebellum. Biol Cybern This elegant in vitro study of the mouse DCN uses dual recordings from
1982, 45:195-206. cartwheel and fusiform cells and minimal stimulation of parallel fibers to
explore the detailed interactions between these circuit elements. A
17. Bell CC, Han VZ, Sugawara S, Grant K: Synaptic plasticity in a surprising finding of the study is that parallel fiber connectivity is extre-
cerebellum-like structure depends on temporal order. Nature mely sparse such that neighboring neurons rarely are activated by the
1997, 387:278-281. same parallel fiber input. This finding raises questions about the pre-
valence of true feedforward inhibition in the DCN.
18. Han VZ, Grant G, Bell CC: Reversible associative depression
and nonassociative potentiation at a parallel fiber synapse. 32. Fujino K, Oertel D: Bidirectional synaptic plasticity in the
Neuron 2000, 27:611-622. cerebellum-like mammalian dorsal cochlear nucleus. Proc Natl
Acad Sci U S A 2003, 100:265-270.
19. Roberts PD, Bell CC: Computational consequences of
temporally asymmetric learning rules: II. Sensory image 33. Tzounopoulos T, Kim Y, Oertel D, Trussell LO: Cell-specific, spike
cancellation. J Comput Neurosci 2000, 9:67-83. timing-dependent plasticities in the dorsal cochlear nucleus.
Nat Neurosci 2004, 7:719-725.
20. Zhang Z, Bodznick D: The importance of N-methyl-D-aspartate
(NMDA) receptors in subtraction of electrosensory 34. Tzounopoulos T, Rubio ME, Keen JE, Trussell LO: Coactivation of
reafference in the dorsal nucleus of skates. J Exp Biol 2010, pre- and postsynaptic signaling mechanisms determines cell-
213:2700-2709. specific spike-timing-dependent plasticity. Neuron 2007,
54:291-301.
21. Harvey-Girard E, Lewis J, Maler L: Burst-induced anti-Hebbian
depression acts through short-term synaptic dynamics to 35. Zhao Y, Rubio ME, Tzounopoulos T: Distinct functional and
cancel redundant sensory signals. J Neurosci 2010, 30:6152- anatomical architecture of the endocannabinoid system in the
6169. auditory brainstem. J Neurophysiol 2009, 101:2434-2446.
This in vitro study provides the first characterization of long-term mechan-
isms in the ELL of weakly electric gymnotid fish, complementing previous 36. Zhao Y, Tzounopoulos T: Physiological activation of cholinergic
studious in the mormyrid ELL and the mouse DCN. Plasticity is NMDA inputs controls associative synaptic plasticity via modulation
receptor-dependent and anti-Hebbian as in other cerebellum-like cir- of endocannabinoid signaling. J Neurosci 2011, 31:3158-3168.
cuits, but is not spike timing-dependent. Also in contrast to findings in This in vitro study shows that previously described timing-dependent
other cerebellum-like circuits no mechanism for long-lasting potentiation Hebbian plasticity (i.e. long-term potentiation) at parallel fiber synapses
of parallel fiber synapses was found. The authors develop a model that onto fusiform cells in the mouse dorsal cochlear nucleus can be con-
shows how interactions among parallel fiber synaptic dynamics, feedfor- verted to timing-dependent anti-Hebbian plasticity (i.e. long-term depres-
ward inhibition, and long-term depression could account for bi-directional sion) if cholinergic inputs are activated during pairing. The paper provides
negative images observed in vivo. a striking example of how neuromodulation can alter rules for plasticity
and also strengthens the case for a sensory filtering function for the DCN
22. Jorntell H, Hansel C: Synaptic memories upside down: similar to those described previously in cerebellum-like circuits in fish.
bidirectional plasticity at cerebellar parallel fiber-Purkinje cell
synapses. Neuron 2006, 52:227-238. 37. Albus JS: A theory of cerebellar function. Math Biosci 1971,
10:25-61.
23. Lewis JE, Maler L: Dynamics of electrosensory feedback:
short-term plasticity and inhibition in a parallel fiber pathway. 38. Marr D: A theory of cerebellar cortex. J Physiol (Lond) 1969,
J Neurophysiol 2002, 88:1695-1706. 202:437-471.
24. Lewis JE, Lindner B, Laliberte B, Groothuis S: Control of neuronal 39. Arenz A, Bracey EF, Margrie TW: Sensory representations in
firing by dynamic parallel fiber feedback: implications for cerebellar granule cells. Curr Opin Neurobiol 2009, 19:445-451.
444
608 Sensory and Motor Systems
40. Sawtell NB: Multimodal integration in granule cells as a basis 41. Miall RC, Wolpert DM: Forward models for physiological motor
for associative plasticity and sensory prediction in a control. Neural Netw 1996, 8:1-15.
cerebellum-like circuit. Neuron 2010, 66:573-584.
This study of the mormyrid ELL uses in vivo whole-cell recordings to show 42. Bastian AJ: Learning to predict the future: the cerebellum
that individual granule cells integrate proprioceptive and motor corollary adapts feedforward movement control. Curr Opin Neurobiol
signals conveyed by separate classes of mossy fibers. The study also 2006, 16:645-649.
provides evidence that this multimodal integration allows for the genera-
tion of more specific negative images in principal cells than would be 43. Ebner TJ, Pasalar S: Cerebellum predicts the future motor
possible if proprioceptive and corollary discharge signals were conveyed state. Cerebellum 2008, 7:583-588.
separately. These findings provide the first experimental evidence for
multimodal integration in granule cells and suggest a function for this 44. Miall RC, Weir DJ, Wolpert DM, Stein JF: Is the cerebellum a
integration that is in-line with longstanding MarrAlbus theories of cer- smith predictor. J Motor Behav 1993, 25:203-216.
ebellar function.
445
446
General and Comparative Endocrinology 157 (2008) 259265
Minireview
Seasonal-like growth and regression of the avian song control system: Neural
and behavioral plasticity in adult male Gambels white-crowned sparrows
John Meitzen *,1, Christopher K. Thompson 1
Graduate Program in Neurobiology and Behavior, University of Washington, Box 356515, Seattle, WA 98195-6515, United States
a r t i c l e i n f o a b s t r a c t
Article history: Birdsong is regulated by a series of discrete brain nuclei known as the song control system. In seasonally-
Received 29 January 2008 breeding male songbirds, seasonal changes in steroid sex hormones regulate the structure and electro-
Revised 12 March 2008 physiology of song control system neurons, resulting in dramatic changes in singing behavior. Male song-
Accepted 17 March 2008
birds can be brought into the laboratory, where circulating levels of steroid hormone and photoperiod
Available online 25 March 2008
can be abruptly manipulated, providing controlled conditions under which rapid seasonal-like changes
in behavior and morphology can be carefully studied. In this mini-review, we discuss the steroidal and
Keywords:
cellular mechanisms underlying seasonal-like growth and regression of the song control system in adult
Testosterone
Estrogen
male Gambels white-crowned sparrows (Zonotrichia leucophrys gambelii), and its impact on song behav-
Androgen ior. Specically, we discuss recent advances concerning: (1) the role of androgen and estrogen receptors
Seasonal plasticity in inducing seasonal-like growth of the song control system; (2) how photoperiod modulates the time
Regression course of testosterone-induced growth of the song control system; (3) how bilateral intracerebral infu-
Growth sion of androgen and estrogen receptor antagonists near the song control nucleus HVC prevents sea-
Song control system sonal-like increases in song stereotypy but not song rate; and (4) the steroidal and cellular
Photoperiod mechanisms that mediate rapid regression of the song control system. Throughout this mini-review
we compare data collected from white-crowned sparrows to that from other songbird species. We con-
clude by outlining avenues of future research.
2008 Elsevier Inc. All rights reserved.
1. Introduction in the song control system and song behavior. Gambels white-
crowned sparrows are long-distance migrants that winter along
Plasticity in brain and behavior is fundamental for animals to the west coast and western interior of the contiguous United
react to changing environmental demands. An organisms ability States, and breed in Alaska and Canada, unlike some other white-
to adapt to seasonal changes in the environment is critical to crowned sparrow subspecies such as the non-migratory Nuttalls
reproductive success, and breeding typically happens during the (Farner and Lewis, 1973; Chilton et al., 1995). In the spring, longer
season with the highest probability of successfully rearing off- day lengths induce an increase in plasma testosterone (T) level,
spring. It is thus not surprising that seasonal plasticity in brain which in turn triggers anatomical and electrophysiological changes
and behavior has been found in every vertebrate taxon (Tramontin in the nuclei of the song control system and in white-crowned
and Brenowitz, 2000). These seasonal changes in the environment sparrow song behavior, most notably increases in song rate, song
are often signaled to the brain via changes in hormone levels, stereotypy, and duration (Nottebohm, 1981; Smith et al., 1995;
which can trigger massive restructuring of the neural substrates Tramontin et al., 2000; Brenowitz, 2004; Park et al., 2005; Meitzen
that regulate behavior. In this mini-review we survey recent ad- et al., 2007a). The volumes of the song control nuclei HVC, Area X,
vances in understanding the proximate steroidal and cellular and RA all increase, and the cellular mechanisms underlying these
mechanisms underlying an example of adult seasonal plasticity: increases vary by nucleus. The increase in HVC volume is largely
the seasonal growth and regression of the song control system driven by an increase in neuron number (Fig. 2A), whereas growth
(Fig. 1) and the resulting changes in song (a reproduction-related of RA (Fig. 2B) and Area X (Thompson and Brenowitz, 2005) is dri-
vocal behavior) in the male Gambels white-crowned sparrow. ven by changes in neuron size and density. The morphological
White-crowned sparrows are age-limited learners (Marler and changes that underlie growth of the song control system nuclei
Tamura, 1964) that as adults exhibit substantial seasonal changes have been recently reviewed in detail elsewhere (Brenowitz,
2004), so we will not discuss this issue further.
* Corresponding author. Fax: +1 206 543 5152.
Changes in the song control system and singing behavior can be
E-mail address: jmeitzen@u.washington.edu (J. Meitzen). induced in the laboratory using the appropriate environmental and
1
These authors contributed equally to this work. hormonal cues, allowing carefully controlled experiments. To mi-
0016-6480/$ - see front matter 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.ygcen.2008.03.014
447
260 J. Meitzen, C.K. Thompson / General and Comparative Endocrinology 157 (2008) 259265
2. Mechanisms of growth
Fig. 1. Simplied schematic of the avian song control system showing the distri-
bution of steroid receptors. The nuclei HVC and RA comprise the main descending
motor circuit. RA projects to several brainstem nuclei (including nXIIts) that contain
2.1. Photoperiod increases plasma T level
motoneurons that project to the muscles controlling respiration and the sound
production organ, the syrinx. Both brainstem motoneurons and the syrinx express Seasonal-like plasticity in the song control system is modulated
androgen receptors. HVC, X, DLM, and LMAN comprise the anterior forebrain pat- by changes in photoperiod and circulating steroid sex hormones,
hway (AFP), which is necessary for song learning. HVC, used as a proper name; DLM,
with important contributions coming from both of these factors
the medial portion of the dorsolateral nucleus of the anterior thalamus; LMAN, the
lateral magnocellular nucleus of anterior nidopallium; RA, the robust nucleus of the (Fig. 3). Very early in the breeding season, increasing day length
arcopallium; nXIIts, the tracheosyringeal portion of the hypoglossal nucleus; X, activates the hypothalamicpituitary gonadal axis, which stimu-
Area X, a subdivision of the medial striatum. Nomenclature used here follows Re- lates growth of the testes and elevates circulating levels of T. This
iner et al. (2004).
increase in circulating T primarily regulates song control system
Fig. 2. Schematic, normalized representation of seasonal-like changes in morphology of HVC and RA. These representations are based upon the means of data taken from
Tramontin et al. (2000) and Thompson et al. (2007). We made these data sets congruent using the one time point shared between both sets (long-day photoperiod and
systemic testosterone (LD + T) for 20 days) and then normalized them to the short-day photoperiod (SD) time point. Birds were exposed to at least 10 weeks of SD
photoperiod in both studies to insure photosensitivity. (A) HVC grows within 7 days after the transition to LD + T, driven largely by changes in neuron number. After the
transition from LD + T to SD photoperiod, HVC volume regresses within 12 h, initially driven by a sharp increase in neuron density followed by a slower decrease in neuron
number. (B) RA volume increases signicantly within 20 days after the transition to LD + T, largely driven by an increase neuronal soma area and a decrease in density. After
the transition from LD + T to SD photoperiod, RA volume regresses signicantly by 20 days, with signicant changes in density and soma area by 2 days by continuing to
signicantly change over the entire time course.
448
J. Meitzen, C.K. Thompson / General and Comparative Endocrinology 157 (2008) 259265 261
449
262 J. Meitzen, C.K. Thompson / General and Comparative Endocrinology 157 (2008) 259265
HVC (new neurons in HVC are thought to be RA-projecting, Scotto- In song sparrows (Melospiza melodia), a species closely related to
Lamasse et al., 2007) and increases synthesis of a transsynaptic tro- white-crowned sparrows (Carson and Spicer 2003), the nidopalli-
phic signal that drives changes in intrinsic activity and morphology um surrounding HVC expresses aromatase mRNA, but HVC itself
in RA and Area X. The exact nature of the trophic signal that HVC does not (Soma et al., 2003), similar to other songbird species (For-
provides to RA and Area X is unclear, but it could involve neurotro- lano et al., 2006). Aromatase has been observed in pre-synaptic ter-
phin release and/or changes in activity (further reviewed in Breno- minals within HVC of zebra nches (Peterson et al., 2005).
witz, 2004). Given that increases in Area X and RA nucleus volume Aromatase could thus modulate local concentrations of neuroster-
occur after HVC grows (Tramontin et al., 2000), the trophic signal oids (reviewed in Forlano et al., 2006).
produced by HVC may either take time to be produced and trans-
ported, and/or new RA-projecting neurons may need to be re- 2.6. Changes in song stereotypy and song production may be
cruited to HVC and establish synapses on the increased dendritic differentially regulated
arbors of RA neurons (Hill and DeVoogd, 1991). AR in RA may be
permissive for RA neurons to respond to or amplify the incoming LD photoperiod and systemic T induce dramatic changes in
trophic signal from HVC, as exposing RA neurons to the AR antag- Gambels white-crowned sparrow song behavior. These changes
onist utamide prevents the increase in ring rate and soma size include: an increase in song rate (Smith et al., 1995; Meitzen
induced by LD photoperiod and systemic T in three of ve birds et al., 2007a), an increase in syllable stereotypy (Smith et al.,
(Meitzen et al., 2007a). Further studies of the hypothesized permis- 1995; Tramontin et al., 2000; Meitzen et al., 2007a), and an in-
sive role of AR in RA are required. crease in overall song length and buzz length (Smith et al., 1995).
White-crowned sparrows are closed-ended learners that do not
2.4. Do androgens act directly on nXIIts and the syrinx? learn new syllables seasonally (Marler and Tamura, 1964), unlike
some other species (Brenowitz and Beecher, 2005). While numer-
The steroidal mechanism underlying growth of the brainstem ous studies have shown strong correlations between changes in
motoneurons and of the syrinx (the avian vocal organ) is unclear, song behavior and song control nuclei, few have tested the causal
although it may not depend upon activation of steroid hormone relationship between them. Recent studies have begun to provide
receptors in either HVC or RA. When small intercerebral T implants evidence regarding which seasonal changes in the song control
are unilaterally placed near either HVC or RA, neither nXIIts nor the system mediate particular changes in song behavior. When song
syrinx increase in volume or mass, respectively (Brenowitz and system growth is blocked by infusing AR and ER antagonists near
Lent, 2002). Lesions of HVC prevented the growth of nXIIts (Breno- HVC in white-crowned sparrows exposed to LD photoperiod and
witz and Lent, 2001). Since both nXIIts and the syrinx contain AR systemic T, song stereotypy differs between treatment groups,
(Smith et al., 1996), one reasonable hypothesis is that androgens but song rate does not (Meitzen et al., 2007a). This result suggests
act individually and directly on them to increase growth. Other that activation of sex steroid receptors in HVC is not necessary for
hypotheses include but are not limited to: (1) retrograde support the systemic T and LD photoperiod-induced increase in song rate,
of nXIIts growth from the syrinx, (2) afferent support of nXIIts from and instead mediates changes in song stereotypy, and furthermore,
either or both RA, and the dorsomedial portion of the intercollicu- that increases in song rate are not sufcient for seasonal changes in
lar nucleus (ICo), which both contain AR, and (3) afferent support the song control system. In support of this, deafened white-
of syringeal growth from nXIIts. These hypotheses remain to be crowned sparrows greatly reduce song production, yet their song
tested. control systems still grow and song becomes more stereotyped in
response to LD photoperiod and systemic T (Brenowitz et al.,
2.5. Photoperiod may modulate the growth of the song control system 2007).
450
J. Meitzen, C.K. Thompson / General and Comparative Endocrinology 157 (2008) 259265 263
lease, followed by a decrease in circulating levels of luteinizing anism mediating regression is the same as that underlying the
hormone and follicle stimulating hormone, which ultimately lead growth of RA: changes in soma area and neuron density with no
to regression of the testes. The onset of photorefractoriness is change in neuron number.
determined weeks earlier when individuals enter breeding condi-
tion in response to the initial exposure to LD photoperiod, given 3.4. Neurodegenerative mechanisms of HVC regression and steroidal
that the transition to photorefractoriness is relative to the onset neuroprotection
of LD and largely independent of subsequent changes in photope-
riod (Dawson and Goldsmith, 1984; Bentley et al., 1997; Dawson, In male white-crowned sparrows, the loss of HVC neurons dur-
2001; Dawson et al., 2001; Dawson, 2004). Ultimately, the de- ing song control system regression is mediated by caspase-depen-
crease in circulating T removes steroidal trophic support for an en- dent cell death pathways. Apoptotic-like processes are to be
larged song control system, which then regresses via expected, given that HVC neuron number decreases by nearly
neurodegenerative mechanisms. 30% following a seasonal-like transition to non-breeding condi-
It is not yet known if changes in photoperiod are sufcient to tions (Thompson et al., 2007). HVC cells positive for activated cas-
induce steroid-independent regression of the song control system pase-3 are apparent three days after the transition to non-breeding
in photostimulated males with high levels of circulating T. Despite conditions (accepted for publication, C.K. Thompson). In addition,
the fact that the onset of photorefractoriness is not dependent infusion of a cocktail of caspase inhibitors near HVC in male Gam-
upon a transition to SD photoperiod, we typically expose birds to bels white-crowned sparrows transitioned to non-breeding condi-
SD photoperiod at the same time circulating T is withdrawn to tions seems to prevent a decrease in HVC volume, neuron number,
insure that any stimulatory signal LD photoperiod may contribute neuron spacing, and soma area (accepted for publication, C.K.
to maintaining a breeding-state song control system is removed Thompson).
(Thompson et al., 2007). This manipulation is not the same as a T and/or its metabolites act within HVC to both induce and
transition to photorefractoriness, because male white-crowned maintain its seasonal growth (discussed above). This suggests that
sparrows are probably still photostimulated after just 34 weeks T and/or its metabolites could play a neuroprotective role in rescu-
of LD and high levels of circulating T. Yet the manipulation abruptly ing HVC from the regression induced by the withdrawal of sys-
removes two factors known to promote enlarged song control temic T and exposure to SD photoperiod, not unlike the
system nuclei: high levels of circulating T and LD photoperiod. hormone-mediated neuroprotection that is seen in several in vivo
The purpose of the manipulation is to remove as much exogenous animal models of neurodegenerative insult (Ramsden et al. 2003;
trophic support for the song control system as possible so that it Pike et al. 2006). Consistent with this, we found that direct intrace-
resembles that which is seen in birds held non-breeding conditions. rebral infusion of T near HVC unilaterally in castrated male white-
Regardless, there are many unanswered questions regarding the crowned sparrows transferred to SD photoperiod and systemic T-
steroid-independent contributions of photoperiod to song control withdrawal seems to ameliorate neurodegeneration of the ipsilat-
system plasticity. eral HVC (unpublished data, C.K. Thompson). These results suggest
that T and/or its metabolites directly act on HVC neurons to protect
3.2. Steroidal mechanism of regression them from degenerative mechanisms induced by withdrawal of
circulating sex steroids and photoperiod shift. Seasonal-like rapid
Acute withdrawal of circulating T induces a signicant decrease regression of the song control system therefore may serve as an
in HVC volume within 12 h (Thompson et al., 2007). This decrease excellent model to further elucidate the molecular mechanisms
occurs before the birds experience a change in photoperiod, which that underlie hormone-mediated neuroprotection.
demonstrates that, at the very least, the initial regression of HVC
volume is driven largely by the withdrawal of circulating sex ste- 3.5. Behavior
roids and not by physiological changes driven by photoperiod. This
suggests that regression, like growth, of the song control system in Male white-crowned sparrows stop singing immediately fol-
male white-crowned sparrows is largely driven by changes in cir- lowing the withdrawal of circulating T (Meitzen et al., unpublished
culating levels of sex steroids. data). This observation strongly suggests that T is rapidly cleared
from the blood stream. Though a time course for the clearance of
3.3. Changes in neuronal morphology that underlie song control circulating T following castration is not known in male white-
system regression crowned sparrows, intramuscular injections of highly concentrated
T propionate dissolved in peanut oil into male song sparrows is
In general, long-term changes in HVC volume result from cleared within 90 min (Soma et al., 1999). In male house nches,
changes in neuron number. The initial decrease in HVC volume removal of subcutaneous T pellets results in signicantly reduced
within 12 h of T withdrawal, however, results from an increase in circulating levels of T 24 h later (Deviche et al., 2006). If a similar
neuron density (Fig. 2A, Thompson et al., 2007). HVC essentially time course for clearance of circulating T applies to white-crowned
collapses in on itself as the space between neurons decreases. Over sparrows, it would suggest that abrupt cessation of singing in
the next four days, neuron number and soma area (a measure of males is driven by the rapid withdrawal of T independent of
neuron size) signicantly decrease (Thompson et al., 2007). At this photoperiod.
point, the rate of neuron loss offsets the decrease in neuron spac-
ing, and neuron density signicantly decreases. Thus, the increase
in neuron density lasts only a few days. 4. Future directions
The downstream nuclei of HVC, RA and Area X, regress more
slowly than HVC, taking days to weeks instead of hours to days. There are many unanswered questions about seasonal-like
The volumes of Area X and RA are not signicantly regressed until plasticity of the song control system in male white-crowned spar-
seven and 20 days, respectively, following the transition to non- rows, and many ways that the model presented here could be fur-
breeding conditions (Thompson et al., 2007). RA soma area and ther tested, rened, and expanded. In addition to the questions
neuron density signicantly regress by two days after the transi- raised above, outstanding questions include: (1) what are the
tion to non-breeding conditions and continue to regress for at least downstream molecular cascades that are turned on or off by
20 days (Fig. 2B, Thompson et al., 2007). Thus the neuronal mech- changes in circulating sex steroids? (2) Do the various neuron
451
264 J. Meitzen, C.K. Thompson / General and Comparative Endocrinology 157 (2008) 259265
types in HVC (RA-projecting, Area X-projecting, and interneurons) Dawson, A., Goldsmith, A.R., 1984. Effects of gonadectomy on seasonal-changes in
plasma-LH and prolactin concentrations in male and female starlings (Sturnus
and RA (glutamatergic projection neurons and GABAergic inter-
vulgaris). J. Endocrinol. 100, 213218.
neurons) contribute to song control system growth and regres- Dawson, A., King, V.M., Bentley, G.E., Ball, G.F., 2001. Photoperiodic control of
sion in different ways? (3) How does the absence of seasonality in birds. J. Biol. Rhythms 16, 365380.
metabolites of T contribute to regression of the song control sys- Deviche, P., Martin, R.K., Small, T., Sharp, P.J., 2006. Testosterone induces testicular
development but reduces GnRH-I ber density in the brain of the house nch,
tem? (4) How do rates of proliferation, migration, and incorpora- Carpodacus mexicanus. Gen. Comp. Endocrinol. 147, 167174.
tion of new neurons into HVC change during seasonal-like growth Dloniak, S.M., Deviche, P., 2001. Effects of testosterone and photoperiodic condition
and regression? (5) What other electrophysiological properties, in on song production and vocal control region volumes in adult male dark-eyed
juncos (Junco hyemalis). Horm. Behav. 39, 95105.
addition to the ones described here, change during the growth Farner, D.S., Lewis, R.A., 1973. Annual cycles of white-crowned sparrows. J. Reprod.
and regression of the song control nuclei? (6) What changes occur Fert. Suppl. 19, 3550.
in the syrinx during the transition to breeding conditions and to Forlano, P.M., Schlinger, B.A., Bass, A.H., 2006. Brain aromatase: new lessons from
non-mammalian model systems. Front. Neuroendocrinol. 27, 247274.
non-breeding conditions? (7) How similar are the proximate Fusani, L., Vant Hof, T., Hutchison, J.B., Gahr, M., 2000. Seasonal expression of
mechanisms in Gambels white-crowned sparrow seasonal plas- androgen receptors, estrogen receptors, and aromatase in the canary brain in
ticity to other steroid-sensitive models? These and other ques- relation to circulating androgens and estrogens. J. Neurobiol. 43, 254268.
Gahr, M., Gttinger, H.R., Kroodsma, D.E., 1993. Estrogen receptors in the avian
tions illustrate that study of seasonal plasticity within the avian brain: survey reveals general distribution and forebrain areas unique to
song control system will continue to serve as a useful model for songbirds. J. Comp. Neurol. 327, 112122.
hormone-mediated neural plasticity. Gulledge, C.C., Deviche, P., 1997. Androgen control of vocal control region volumes
in a wild migratory songbird (Junco hyemalis) is region and possibly age
dependent. J. Neurobiol. 32, 391402.
Acknowledgments Hidalgo, A., Barami, K., Iversen, K., Goldman, S.A., 1995. Estrogens and non-
estrogenic ovarian inuences combine to promote the recruitment and
decrease the turnover of new neurons in the adult female canary brain. J.
We thank Eliot A. Brenowitz and David J. Perkel for their men- Neurobiol. 27, 470487.
torship and support. We also thank Eliot A. Brenowitz and two Hill, K.M., DeVoogd, T.J., 1991. Altered daylength affects dendritic structure in a
anonymous reviewers for comments that improved this manu- song-related brain region in red-winged blackbirds. Behav. Neural. Biol. 56,
240250.
script. J.M. and C.K.T. made equal contributions to this work. Grant Johnson, F., Bottjer, S.W., 1993. Hormone-induced changes in identied cell
Sponsor: NIH: MH53032 (E.A.B), MH068530 (D.J.P.); 5 T32 populations of the higher vocal center in male canaries. J. Neurobiol. 24, 400
GM07108 (training grant supporting J.M. and C.K.T.) 418.
Johnson, F., Bottjer, S.W., 1995. Differential estrogen accumulation among
populations of projection neurons in the higher vocal center of male canaries.
References J. Neurobiol. 26, 87108.
Kim, Y.H., Perlman, W.R., Arnold, A.P., 2004. Expression of androgen receptor mRNA
Arnold, A.P., Nottebohm, F., Pfaff, D.W., 1976. Hormone concentrating cells in vocal in zebra nch song system: developmental regulation by estrogen. J. Comp.
control and other areas of the brain of the zebra nch (Poephila guttata). J. Neurol. 469, 535547.
Comp. Neurol. 165, 487511. Marler, P., Tamura, M., 1964. Culturally transmitted patterns of vocal behavior in
Ball, G.F., Auger, C.J., Bernard, D.J., Charlier, T.D., Sartor, J.J., Riters, L.V., Balthazart, J., sparrows. Science 146, 14831486.
2004. Seasonal plasticity in the song control system: multiple brain sites of Meddle, S.L., Maney, D.L., Wingeld, J.C., 1999. Effects of N-methyl-D-aspartate on
steroid hormone action and the importance of variation in song behavior. Ann. luteinizing hormone release and fos-like immunoreactivity in the male white-
NY Acad. Sci. 1016, 586610. crowned sparrow (Zonotrichia leucophrys gambelii). Endocrinology 140, 5922
Bentley, G.E., Goldsmith, A.R., Dawson, A., Glennie, L.M., Talbot, R.T., Sharp, P.J., 5928.
1997. Photorefractoriness in European starlings (Sturnus vulgaris) is not Meitzen, J., Moore, I.T., Lent, K., Brenowitz, E.A., Perkel, D.J., 2007a. Steroid hormones
dependent upon the long-day-induced rise in plasma thyroxine. Gen. Comp. act transsynaptically within the forebrain to regulate neuronal phenotype and
Endocrinol. 107, 428438. song stereotypy. J. Neurosci. 27, 1204512057.
Bentley, G.E., Vant Hof, T.J., Ball, G.F., 1999. Seasonal neuroplasticity in the songbird Meitzen, J., Perkel, D.J., Brenowitz, E.A., 2007b. Seasonal changes in intrinsic
telencephalon: a role for melatonin. Proc. Natl. Acad. Sci. USA 96, 46744679. electrophysiological activity of song control neurons in wild song sparrows. J.
Bernard, D.J., Wilson, F.E., Ball, G.F., 1997. Testis-dependent and -independent Comp. Physiol. A 193, 677683.
effects of photoperiod on volumes of song control nuclei in American tree Metzdorf, R., Gahr, M., Fusani, L., 1999. Distribution of aromatase, estrogen receptor,
sparrows (Spizella arborea). Brain Res. 760, 163169. and androgen receptor mRNA in the forebrain of songbirds and nonsongbirds. J.
Bernard, D.J., Bentley, G.E., Balthazart, J., Turek, F.W., Ball, G.F., 1999. Androgen Comp. Neurol. 407, 115129.
receptor, estrogen receptor alpha, and estrogen receptor beta show distinct Nastiuk, K.L., Clayton, D.F., 1995. The canary androgen receptor mRNA is localized in
patterns of expression in forebrain song control nuclei of European starlings. the song control nuclei of the brain and is rapidly regulated by testosterone. J.
Endocrinology 140, 46334643. Neurobiol. 26, 213224.
Brenowitz, E.A., 2004. Plasticity of the adult avian song control system. Ann. NY Nottebohm, F., 1981. A brain for all seasons: cyclical anatomical changes in song
Acad. Sci. 1016, 560585. control nuclei of the canary brain. Science 214, 13681370.
Brenowitz, E.A., Beecher, M.D., 2005. Song learning in birds: diversity and plasticity, Park, K.H., Meitzen, J., Moore, I.T., Brenowitz, E.A., Perkel, D.J., 2005. Seasonal-like
opportunities and challenges. Trends Neurosci. 28, 127132. plasticity of spontaneous ring rate in a songbird pre-motor nucleus. J.
Brenowitz, E.A., Lent, K., 2001. Afferent input is necessary for seasonal growth and Neurobiol. 64, 181191.
maintenance of adult avian song control circuits. J. Neurosci. 21, 23202329. Peterson, R.S., Yarram, L., Schlinger, B.A., Saldanha, C.J., 2005. Aromatase is pre-
Brenowitz, E.A., Lent, K., 2002. Act locally and think globally: intracerebral synaptic and sexually dimorphic in the adult zebra nch brain. Proc. Biol. Sci.
testosterone implants induce seasonal-like growth of adult avian song control 272, 20892096.
circuits. Proc. Natl. Acad. Sci. USA 99, 1242112426. Pike, C.J., Rosario, E.R., Nguyen, T.V.V., 2006. Androgens, aging, and Alzheimers
Brenowitz, E.A., Lent, K., Rubel, E.W., 2007. Auditory feedback and song production disease. Endocrine 29, 233241.
do not regulate seasonal growth of song control circuits in adult white-crowned Ramsden, M., Shin, T.M., Pike, C.J., 2003. Androgens modulate neuronal vulnerability
sparrows. J. Neurosci. 27, 68106814. to kainate lesion. Neuroscience 122, 573578.
Carson, R.J., Spicer, G.S., 2003. A phylogenetic analysis of the emberizid sparrows Reiner, A., Perkel, D.J., Bruce, L.L., Butler, A.B., Csillag, A., Kuenzel, W., Medina, L.,
based on three mitochondrial genes. Mol. Phylogenet. Evol. 29, 4357. Paxinos, G., Shimizu, T., Striedter, G., Wild, M., Ball, G.F., Durand, S., Guturkun,
Chilton, G., Baker, M.C., Barrentine, C.D., Cunningham, M.A., 1995. White-crowned O., Lee, D.W., Mello, C.V., Powers, A., White, S.A., Hough, G., Kubikova, L.,
sparrow (Zonotrichia leucophrys). In: Poole, A., Gill, F. (Eds.), The Birds of North Smulders, T.V., Wada, K., Dugas-Ford, J., Husband, S., Yamamoto, K., Yu, J., Siang,
America, No. 183. The Academy of Natural Sciences, Philadelphia, PA, and The C., Jarvis, E.D., 2004. Revised nomenclature for avian telencephalon and some
American Ornithologists Union, Washington, DC. related brainstem nuclei. J. Comp. Neurol. 473, 377414.
Dawson, A., 2001. The effects of a single long photoperiod on induction and Riters, L.V., Baillien, M., Eens, M., Pinxten, R., Foidart, A., Ball, G.F., Balthazart, J.,
dissipation of reproductive photorefractoriness in European starlings. Gen. 2001. Seasonal variation in androgen-metabolizing enzymes in the
Comp. Endocrinol. 121, 316324. diencephalon and telencephalon of the male European starling (Sturnus
Dawson, A., 2004. Evidence against a period of relative photorefractoriness during vulgaris). J. Neuroendocrinol. 13, 985997.
the recovery of photosensitivity in common starlings. Gen. Comp. Endocrinol. Saldanha, C.J., Deviche, P.J., Silver, R., 1994. Increased VIP and decreased GnRH
136, 117121. expression in photorefractory dark-eyed juncos (Junco hyemalis). Gen. Comp.
Dawson, A., Goldsmith, A.R., 1983. Plasma prolactin and gonadotrophins during Endocrinol. 93, 128136.
gonadal development and the onset of photorefractoriness in male and female Sartor, J.J., Balthazart, J., Ball, G.F., 2005. Coordinated and dissociated effects of
starlings (Sturnus vulgaris) on articial photoperiods. J. Endocrinol. 97, 253 testosterone on singing behavior and song control nuclei in canaries (Serinus
260. canaria). Horm. Behav. 47, 467476.
452
J. Meitzen, C.K. Thompson / General and Comparative Endocrinology 157 (2008) 259265 265
Schlinger, B.A., Slotow, R.H., Arnold, A.P., 1992. Plasma estrogens and brain Strand, C.R., Deviche, P., 2007. Hormonal and environmental control of song control
aromatase in winter white-crowned sparrows. Ornis Scand. 23, 292297. region growth and new neuron addition in adult male house nches,
Scotto-Lomassese, S., Rochefort, C., Nshdejan, A., Scharff, C., 2007. HVC interneurons Carpodacus mexicanus. Dev. Neurobiol. 67, 827837.
are not renewed in adult male zebra nches. Eur. J. Neurosci. 25, 16631668. Thompson, C.K., Brenowitz, E.A., 2005. Seasonal change in neuron size and spacing
Smith, G.T., Brenowitz, E.A., Wingeld, J.C., Baptista, L.F., 1995. Seasonal changes in but not neuronal recruitment in a basal ganglia nucleus in the avian song
song nuclei and song behavior in Gambels white-crowned sparrows. J. control system. J. Comp. Neurol. 481, 276283.
Neurobiol. 28, 114125. Thompson, C.K., Bentley, G.E., Brenowitz, E.A., 2007. Rapid seasonal-like regression
Smith, G.T., Brenowitz, E.A., Prins, G.S., 1996. Use of PG-21 immunocytochemistry to of the adult avian song control system. Proc. Natl. Acad. Sci. USA 104, 15520
detect androgen receptors in the songbird brain. J. Histochem. Cytochem. 44, 15525.
10751080. Tramontin, A.D., Brenowitz, E.A., 2000. Seasonal plasticity in the adult brain. Trends
Smith, G.T., Brenowitz, E.A., Wingeld, J.C., 1997. Roles of photoperiod and Neurosci. 23, 251258.
testosterone in seasonal plasticity of the avian song control system. J. Tramontin, A.D., Hartman, V.N., Brenowitz, E.A., 2000. Breeding conditions induce
Neurobiol. 32, 426442. rapid and sequential growth in adult avian song circuits: a model of seasonal
Sohrabji, F., Nordeen, K.W., Nordeen, E.J., 1989. Projections of androgen- plasticity in the brain. J. Neurosci. 20, 854861.
accumulating neurons in a nucleus controlling avian song. Brain Res. 488, Tramontin, A.D., Perto, N., Wingeld, J.C., Brenowitz, E.A., 2001. Seasonal growth of
253259. song control nuclei precedes seasonal reproductive development in wild adult
Soma, K.K., Hartman, V.N., Wingeld, J.C., Brenowitz, E.A., 1999. Seasonal changes in song sparrows. Gen. Comp. Endocrinol. 122, 19.
androgen receptor immunoreactivity in the song nucleus HVc of a wild bird. J. Tramontin, A.D., Wingeld, J.C., Brenowitz, E.A., 2003. Androgens and estrogens
Comp. Neurol. 409, 224236. induce seasonal-like growth of song nuclei in the adult songbird brain. J.
Soma, K.K., Schlinger, B.A., Wingeld, J.C., Saldanha, C.J., 2003. Brain aromatase, 5 Neurobiol. 57, 130140.
alpha-reductase, and 5 beta-reductase change seasonally in wild male song Wilson, F.E., 1985. An androgen-independent mechanism maintains
sparrows: relationship to aggressive and sexual behavior. J. Neurobiol. 2003 56, photorefractoriness in male tree sparrows (Spizella arborea). J. Endocrinol.
209221. 107, 137143.
Soma, K.K., Tramontin, A.D., Featherstone, J., Brenowitz, E.A., 2004. Estrogen Wingeld, J.C., Farner, D.S., 1978. Annual cycle of plasma IRLH and steroid-
contributes to seasonal plasticity of the adult avian song control system. J. hormones in feral populations of the white-crowned sparrow, Zonotrichia-
Neurobiol. 58, 413422. leucophrys-gambelii. Biol. Reprod. 19, 10461056.
453
454
Cerebellum-Like Structures
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
ANNUAL
REVIEWS Further
Click here for quick links to
and Their Implications for
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
1
455
parallel bers together with the dendrites and
Contents cell bodies on which the bers terminate. The
parallel bers are numerous and closely packed.
LOCAL CIRCUITRY, GENE
The granule cells that give rise to the paral-
EXPRESSION, AND
lel bers in cerebellum-like structures are mor-
EVOLUTION OF
phologically similar to cerebellar granular cells
CEREBELLUM-LIKE
(Mugnaini et al. 1980a,b) but are usually located
STRUCTURES . . . . . . . . . . . . . . . . . . . 2
in an external granule cell mass rather than in
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
General Features. . . . . . . . . . . . . . . . . . . 2
a granule cell layer beneath the molecular layer
Local Circuitry of Different
as in the cerebellum. Unipolar brush cells and
Cerebellum-Like Structures . . . . . 3
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
2 Bell
Han Sawtell
456
Predictive inputs
Corollary discharge signals
Higher levels of the same modality
Other sensory modalities Granule layer
(e.g. proprioception)
???
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
Molecular layer
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
Figure 1
Schematic drawing showing major features of cerebellum-like sensory structures. Inhibitory stellate cells of
the molecular layer are shown in black. Blue upward arrows indicate afferent input from the periphery
terminating in the sensory input layer. In some cerebellum-like structures the afferent input also terminates
on the smooth proximal portion of the apical dendrites as indicated by the small blue arrowheads.
cells, on proximal apical dendrites of principal This review describes major features of the dif-
cells, or on interneurons that relay the informa- ferent cerebellum-like structures of craniates
tion from the periphery to the principal cells. but is not exhaustive. Recent reviews (Bell 2002,
Some of the interneurons of the deep layers Bell & Maler 2005, Montgomery et al. 1995)
are inhibitory, allowing for a change of sign, and the original papers on individual structures,
whereby excitation in the periphery is con- as provided below, should be consulted for more
verted into inhibition of some principal cells. complete descriptions. Some of the structures
The peripheral input to the deep layers forms a are also much better known than others, which
map of a sensory surface, such as the skin sur- is reected in the level of detail in the following
face, the retina, or the cochlea. descriptions.
Myxinoidea Myxinoids ?
Eptatretidae ?
Petromyzontiformes
Lampetra ?
Elasmobranchii
Chondrichthyes
Holocephalii
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
Chondrosteii
Holosteii
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
Neopterygii Osteoglossomorpha
Gnathostomata
Actinopterygii
Vertebrata
Craniata
Elopomorpha
Teleosteii Clupeomorpha
Euteleostei
Dipneustii
Crossopterygii
Sarcopterygii
Urodela
Tetrapoda Amphibia
Anura
Apoda ? ?
Reptilia
Aves
Mammalia
Figure 2
Distribution of cerebellum-like structures and the cerebellum in different craniate groups. A lled circle means the structure is present
in all or almost all the members of that group. A lled half circle means the structure is present only sporadically in that group. A
question mark means that presence of the structure in that group is controversial. CBM, cerebellum; DCN, dorsal cochlear nucleus;
DON, dorsal octavolateral nucleus; ELL, electrosensory lobe; MON, medial octavolateral nucleus; OTML, marginal layer of the optic
tectum; RLN, rostrolateral nucleus of thalamus.
layer known as the cerebellar crest (Figure 3a Dorsal octavolateral nucleus (DON). The
d ). The parallel bers of the cerebellar crest dorsal octavolateral nucleus (DON) processes
descend from an anterior granule cell mass primary afferent input from electroreceptors
known as the lateral granular mass in elas- and is present in many basal vertebrates with
mobranchs and the eminentia granularis in an electrosense (Figures 2, 3a) Electrorecep-
other sh. The inputs to these granule cells in- tion is a vertebrate sense that may have orig-
clude lateral line primary afferents (Bodznick inated as early as the lateral line or vestibular
& Northcutt 1980), eighth nerve primary affer- senses (Bullock et al. 1983). The Myxinoidea
ents (Puzdrowski & Leonard 1993), input from do not have electroreceptors and do not have
the spinal cord (Schmidt & Bodznick 1987), and a DON (Ronan 1986). Electroreception was
descending input from higher-order lateral line lost during the evolution of neopterygian bony
and acoustic centers (Bell 1981c, McCormick sh, and these sh do not have a DON. Elec-
1997, Tong & Finger 1983). The basilar den- troreception reappeared independently at least
DON: dorsal drites of MON efferent cells are affected by pri- twice during the evolution of the teleost ra-
octavolateral nucleus
mary afferent input. diation: once during the evolution of the two
4 Bell
Han Sawtell
458
a b c d
Many aquatic Mormyrid Gymnotid Most aquatic
vertebrates fish fish vertebrates
DON ELL ELL MON
MON MON MON
p CC
EG
DG R
CB CB
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
CC CB MON
CB
p
EG
DO
N
nAll
ELL
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
ELL
CC CC CC nAll
MON
MON
nVIII
MON
e f g h
Ray finned Some bony Most Most
fish fish mammals vertebrates
OTML RLN DCN Cerebellum
TL
Climbing
Tel CB fibers
N
RL DCN
CB
VCN
TS Opt tr
Figure 3
Cerebellum-like structures in different vertebrate groups. The molecular layer, granule cell mass, and sensory input map are shown in
different colors, as indicated at the bottom of the gure. The climbing ber input to the cerebellum is shown here as a sensory input (see
text). CB, cerebellum; CC, cerebellar crest; DCN, dorsal cochlear nucleus; DGR, dorsal granular ridge; DON, dorsal octavolateral
nucleus; EGp, eminentia granularis posterior; ELL, electrosensory lobe; gran, granular layer; MON, medial octavolateral nucleus; mol,
molecular layer; nAll, anterior lateral line nerve; nVIII, eighth nerve; Opt tr, optic tract; RLn, rostrolateral nucleus; Tel, telencephalon;
TL, torus longitudinalis; TS, torus semicircularis; VCN, ventral cochlear nucleus.
related groups, Mormyriformes and Xeno- and connections. Primary afferent input from
mystinae, and a second time during the evolu- electroreceptors terminates on the basilar den-
tion of the other two related groups, Gymno- drites of efferent cells and inhibitory neurons
tiformes and Siluriformes (Bullock et al. 1983). of the deep layers, as in the MON (Bodznick &
However, the more recently derived electrore- Northcutt 1980, Puzdrowski & Leonard 1993).
ceptors and associated electrosensory central The spine-covered apical dendrites of efferent
structures of teleosts are quite different from cells extend up into the overlying cerebellar
those of other aquatic vertebrates (see elec- crest.
trosensory lobe below). Parallel bers of the DON cerebellar
The DON is located just dorsal to the MON crest arise from the dorsal granular ridge,
and is similar to the MON in its structure which receives proprioceptive input, recurrent
PE PE
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
Corollary
discharge
d e f
DCN Teleost cerebellum Mammalian cerebellum
Parallel fibers
(GCD) Parallel fibers Parallel fibers
To IC To CN
Climbing Climbing
fiber fiber
Auditory
afferents
Figure 4
Local circuits of some cerebellum-like structures, the teleost cerebellum, and the mammalian cerebellum. Granule cells and parallel
bers are in red, afferent input from the periphery is in blue, and the additional inputs to the mormyrid and gymnotid ELLs are in
green. The Purkinje-like cells of the mormyrid ELL and mammalian DCN as well as the Purkinje cells of the teleost and mammalian
cerebellums are black. Excitatory efferent cells are white. IC, inferior colliculus; CN, cerebellar nucleus.
electrosensory input, and corollary discharge Marginal layer of the optic tectum. The op-
input associated with motor commands tic tectum of actinopterygian (ray-nned) shes
(Bodznick & Boord 1986, Conley & Bodznick is distinctive in that its outer layers are cerebel-
1994, Hjelmstad et al. 1996). All three types lum like (Figures 2, 3e) (Meek 1983, Vanegas
of input are active in relation to the shs et al. 1979). The external layer of the optic tec-
respiratory cycle. Electroreceptors in elasmo- tum in these sh is a molecular layer known
branchs are strongly affected by the shs own as the optic tectum marginal layer (OTML).
respiration (Montgomery & Bodznick 1993). The cell bodies of principal cells, the type I
The activity in parallel bers can therefore neurons of Meek (1983), are located below
be used to predict the effect of these cyclic the marginal layer. The type I neurons ex-
OTML: marginal changes on electroreceptive input to the deep tend their spine-covered apical dendrites up
layer of the optic
layers of DON (see Adaptive Processing in into the marginal layer and input from the
tectum
Cerebellum-Like Structures, below). retina maps onto their basilar dendrites and
6 Bell
Han Sawtell
460
the smooth proximal portions of their apical electroreceptors in the skin map onto the deep
dendrites. layers, terminating on the basilar dendrites of
The parallel bers of the marginal layer principal cells or on interneurons (Bell & Maler
ELL: electrosensory
arise from a medially located granule cell mass 2005). The ELL efferent cells of mormyrid lobe
known as the torus longitudinalis. Granule cells (Bell et al. 1997b), gymnotid (Saunders & Bas-
EOD: electric organ
of the torus longitudinalis respond to corol- tian 1984), and silurid sh (McCreery 1977) are discharge
lary discharge signals associated with the motor of two main types: E-cells, which are excited by
EOCD: electric organ
commands that evoke eye movements and re- an increase in peripheral stimulus strength in
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
corollary discharge
spond to visual stimuli as well (Northmore et al. the center of their receptive elds, and I-cells,
1983). Parallel ber activity driven by corol- which are inhibited by such an increase. These
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
lary discharge signals associated with eye move- two functionally distinct cell types are also mor-
ments could predict changes in retinal input to phologically distinct; the E-cells have more ex-
the deep layers, a possible interaction between tensive basilar dendrites.
the two types of input similar to that described Parallel bers of ELLs arise from gran-
above for the DON. ule cells of the eminentia granularis posterior
(EGp), which in mormyrids, at least, also con-
Electrosensory lobe (ELL). Electrorecep- tains Golgi cells and unipolar brush cells similar
tion is present in four groups of teleosts: to the same cell types in the mammalian cere-
Mormyriformes, an order of electric sh from bellum (Campbell et al. 2007). The inputs to
Africa; Gymnotiformes, a superorder of elec- EGp in mormyrid and gymnotid sh include
tric sh from South America; Siluriformes, the proprioceptive signals associated with bending
order of catsh; and Xenomystinae, an African of the body or the ns, recurrent electrosen-
subfamily of the family Notopteridae (Bullock sory input from a higher levels of the system,
& Heiligenberg 1986). All these sh have a and in mormyrids only, a corollary discharge
cerebellum-like electrosensory lobe (ELL) that signal associated with the motor command that
receives primary afferent input from electrore- elicits the electric organ (corollary) discharge
ceptors (Figures 2, 3b,c) (Bell & Russell 1978, (EOCD) (Bastian & Bratton 1990; Bell et al.
Braford 1982, Finger & Tong 1984, Maler et al. 1992; Carr & Maler 1986; Szabo et al. 1979,
1981). 1990). These different inputs to EGp are re-
The Mormyriformes and Gymnotiformes layed to ELL as parallel ber inputs, where they
are electric sh with electric organs as well can predict changes in electroreceptor input to
as electroreceptors. The order Mormyri- the deep layers associated with tail movements,
formes includes the family Mormyridae, all of some other electrosensory input, or the EOD
which have electric organ discharges (EODs) (see Adaptive Processing in Cerebellum-Like
that are brief and pulse like, and the single- Structures).
species family Gymnarchidae, which has a con- The mormyrid (Bell et al. 1981), gym-
tinuous wave-like EOD. The order Gymno- notid (Carr & Maler 1986), and silurid (Tong
tiformes includes some families with wave-like 1982) ELLs receive additional input aside
EODs and other families with pulsatile EODs. from the peripheral and parallel ber inputs.
The ELLs of pulsatile mormyrids and wave They receive direct recurrent input from a
gymnotids have been studied most extensively, higher-order electrosensory nucleus just ros-
although some work has been done on the ELLs tral to ELL, the nucleus preeminentialis dor-
of wave mormyriforms (Kawasaki & Guo 1998, salis (PE) (Figures 4b,c). The deep layers of the
Matsushita & Kawasaki 2005) and pulse gym- mormyrid ELL also receive EOCD input di-
notiforms (Caputi et al. 2002, Schlegel 1973). rectly from an EOD motor commandrelated
The spine-covered apical dendrites of ELL nucleus (Bell & von der Emde 1995). This in-
principal cells extend up into the overlying put is in addition to the EOCD input conveyed
molecular layer. Primary afferent bers from via parallel bers.
ing molecular layer. However, they differ from The cartwheel cell is a second type of princi-
Purkinje cells because they have basilar den- pal cell in the DCN (Cant 1992, Nieuwenhuys
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
drites and do not receive climbing ber input. et al. 1997). These cells are Purkinje-like
The medium ganglion cells are interneurons because they are GABAergic, have extensive
that inhibit both nearby efferent cells and each spine-covered dendrites in the molecular layer,
other (Figure 4b). They are more numerous and inhibit the efferent fusiform cells. The cell
than the efferent cells and have many more den- bodies of cartwheel cells are in the molecular
drites and spines in the molecular layer (Meek layer, and their dendrites are restricted to the
et al. 1996). They must therefore have a central molecular layer.
role in the integration of peripheral and paral- The local circuits of the DCN and the
lel ber inputs in the mormyrid ELL. These mormyrid ELL are very similar to the local
and other differences between the mormyrid circuit of the cerebellar cortex in actinopteryr-
and gymnotid ELLs are consistent with their ian sh where most Purkinje cells are in-
independent evolutionary origins. terneurons that terminate locally on efferent
cells (Figure 4e) (Finger 1978, Meek 1998).
Rostrolateral nucleus of the thalamus. The The parallel bers of the DCN, the mormyrid
rostrolateral nucleus (RLN) (Figures 2, 3f ) of ELL, and the actinopterygian cerebellum pass
the thalamus is a small, cerebellum-like struc- through and excite the dendrites of both
ture found in the thalamus of a few widely efferent cells and Purkinje or Purkinje-like
scattered neopterygian sh (Figure 2) (Butler cells. In all three cases, the Purkinje cells or
& Saidel 1992). The principal cells of RLN Purkinje-like cells inhibit nearby efferent cells
receive topographically organized direct in- (Figures 4b,d,e). The efferent neurons of the
put from the retina on the smooth proximal actinopterygian cerebellum are equivalent to
parts of their apical dendrites. The more dis- the cerebellar nucleus neurons of mammals
tal apical dendrites are covered with spines (Figure 4f ).
and receive parallel ber input from the torus The granule cells of the DCN receive var-
longitudinalis. ious types of input: recurrent auditory input
from the inferior colliculus (Caicedo & Herbert
Dorsal cochlear nucleus. All mammals 1993) and auditory cortex (Weedman & Ryugo
possess a dorsal cochlear nucleus (DCN) 1996); primary vestibular afferent input (Burian
(Figures 2, 3g, 4d ). The DCN is laminated & Gstoettner 1988); input from the pontine nu-
and cerebellum-like in marsupials and euthe- clei (Ohlrogge et al. 2001); somatosensory in-
rian mammals but not in monotremes (Cant put from the dorsal column nuclei (Weinberg
1992, Nieuwenhuys et al. 1997). Fusiform & Rustioni 1987), the trigeminal nuclei (Zhou
cells are the major efferent cell type of the & Shore 2004), and the somatosensory cortex
DCN. Their basilar dendrites are contacted (Wolff & Kunzle 1997); and direct input from
by primary afferent bers from the cochlea, the cochlea via ne unmyelinated Type II affer-
which form a topographic map of the cochlea ents (Brown et al. 1988). DCN granule cells
in the deeper layers below the molecular layer. also receive input from brainstem nuclei as-
The fusiform cells extend their spine-covered sociated with vocalization and respiration that
8 Bell
Han Sawtell
462
may convey corollary discharge signals (Shore puts. The cerebellums of different vertebrates
& Zhou 2006). Proprioceptive input from the can vary markedly, but all the cerebellums that
pinna has particularly strong effects on DCN have been closely examined have a specic in-
granule cells in the cat (Kanold & Young 2001). put from the inferior olive that terminates as
Movements of the animals pinna, head, or body climbing bers. We suggest that the presence
have predictable effects on how the cochlea of a climbing ber is the dening characteris-
responds to an external sound source, and an tic of the cerebellum that distinguishes it from
animals own vocalization and respiration will cerebellum-like structures.
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
have predictable consequences on auditory in- Climbing bers and the peripheral sensory
put. Thus the signals conveyed by the parallel input to cerebellum-like structures are similar
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
bers in the DCN molecular layer could gen- in many respects. Climbing bers signal rather
erate predictions about changes in afferent ac- specic sensory events in most of the cases
tivity from the cochlea that arrive at the deep where the information they convey has been
layers, as in other cerebellum-like structures. identied. Such sensory signals include retinal
slip in a particular direction (Maekawa & Simp-
son 1972), somatosensory stimulation within a
Comparison of the Local Circuitries small region of skin (Ekerot & Jorntell 2001,
of Cerebellum-Like Structures Robertson 1985), and vestibular stimulation
and the Cerebellum with tilt in a particular direction (Barmack &
Many similarities in cell types and local cir- Shojaku 1992). Moreover, the climbing bers of
cuitry between the cerebellum and cerebellum- vertebrates other than mammals do not termi-
like structures have been described in the pre- nate throughout the molecular layer as in mam-
ceding section. The similar cellular elements mals. They terminate instead on smooth, prox-
include the granule cells, the Golgi cells, the imal dendrites at the base of the molecular layer
unipolar brush cells, the parallel bers, the stel- (Nieuwenhuys et al. 1997) in a manner similar
late cells, and the spine-covered molecular layer to that of retinal input onto the smooth, prox-
dendrites of principal cells. imal dendrites of principal cells in the OTML
The most crucial similarity is that between and RLN. This is not to say that the inferior
the two inputs to cerebellum-like structures olive is a simple sensory relay. It is not. But
and the two inputs to cerebellar Purkinje clearly sensory stimuli have a strong inuence
cells. Cerebellum-like structures receive paral- on the inferior olive and on climbing bers, a
lel ber and peripheral input, whereas Purkinje result consistent with the origin of the inferior
cells of the cerebellum receive parallel ber in- olive from the embryos alar or sensory plate.
put and climbing ber input. In both cases, one Devor (2002) has in fact suggested that the in-
input, the parallel bers, conveys a rich variety ferior olive has been interposed between pe-
of information to an entire set of principal cells ripheral sensory structures and the cerebellum
or Purkinje cells. In both cases, a second input to gate sensory signals by motor commands and
peripheral input for cerebellum-like structures by the inferior olives own intrinsic rhythmicity.
and climbing ber input for the cerebellum As noted in the previous section, the parallel
conveys specic information that subdivides the bers of cerebellum-like structures convey in-
set of Purkinje cells that share the same parallel formation that is associated with sensory input
ber input. changes to the deep layers and that can there-
Olivary input to Purkinje cells is more spe- fore predict such changes. The parallel bers
cic than the peripheral input to the deep lay- of the cerebellum similarly convey information
ers of cerebellum-like structures insofar as it is that can predict the occurrence of climbing
conveyed by just a single climbing ber. Effer- ber input. Climbing bers in the occu-
ent cells and Purkinje-like cells in cerebellum- lonodular lobe of the mammalian cerebel-
like structures do not have such single ber in- lum, for example, signal retinal slip (Maekawa
The presence of a climbing ber is perhaps layers of the cerebellum, the ELL, the MON,
the critical difference between the cerebellum and the OTML, but not elsewhere in the brain
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
and cerebellum-like structures. Other differ- ( J. Zhang & C. Bell, unpublished observa-
ences include the presence of basilar dendrites tions). Expression of the GluRdelta2 gene in still
on most principal cells of cerebellum-like struc- other cerebellum-like structures remains to be
tures but not on Purkinje cells; the presence of established.
planar dendritic trees in most Purkinje cells but Some genes are expressed in some of the
not in most principal cells; the presence of cell cerebellum-like structures or the cerebellum in
types in cerebellum-like structures not present the adult but are expressed only in other such
in the cerebellum; and the presence of other in- structures during development. The zebrin II
puts besides parallel bers and climbing bers gene, for example, is expressed only in Purkinje
in cerebellum-like structures not present in the cells in adult mammals, birds, and sh (Hawkes
cerebellum, such as the preeminential input in & Herrup 1995, Lannoo et al. 1991) but is ex-
electroreceptive teleosts (Figures 4b,c). pressed transiently during development in the
MON and in part of the ELL of gymnotid
sh (Lannoo et al. 1992). Similarly, functional
Patterns of Gene Expression N-methyl-D-aspartate (NMDA) receptors are
in Cerebellum-Like Structures present on principal cells of the adult mormyrid
and the Cerebellum and gymnotid ELLs (Grant et al. 1998, Berman
Similarities and differences between the differ- et al. 2001), as well as principal cells of the
ent cerebellum-like structures and the cerebel- adult DCN (Manis & Molitor 1996), but are
lum itself are also revealed in gene expression present on cerebellar Purkinje only during de-
patterns. Some genes are expressed in many velopment (Dupont et al. 1987).
different cerebellar and cerebellum-like struc- The common features in the local circuitry
tures, whereas others are expressed in only a and in the gene expression patterns suggest
few of these structures (Bell 2002). Common the presence of a shared genetic-developmental
patterns of gene expression between cerebellar program in all craniates, a program that
Purkinje cells and cartwheel cells of the DCN once activated can generate a cerebellum or
are particularly prominent, and many muta- cerebellum-like structure. Some ndings from
tions affect both cell types (Berrebi et al. 1990). experimental embryology support this idea.
One gene, the GluRdelta2 gene, may be ex- Thus, ectopic cerebellum-like structures de-
pressed in most if not all cerebellum-like struc- velop in the forebrain or midbrain of a chick
tures and also in the cerebellum, but not in other embryo if beads are coated with broblast
structures. This gene is structurally related to growth factor 8 and placed at those sites in
the ionotropic glutamate receptors but does the embryo (Martinez et al. 1999). Similarly,
not form ion channels (Yuzaki 2003). The gene cerebellar tissue will develop ectopically in the
is necessary for long-term depression (LTD) midbrain and forebrain of a mouse embryo
at the parallel ber to Purkinje cell synapse with a genome that is Otx1+/- and Otx2+/-
(Yawata et al. 2006). In mammals, the GluR- (Drosophila orthodenticle protein, a transcrip-
delta2 gene is expressed in Purkinje cells (Yuzaki tion factor) (Acampora et al. 1997).
10 Bell
Han Sawtell
464
Evolution of Cerebellum-Like climbing bers on putative Purkinje cells
Structures and the Cerebellum could indicate the presence of a cerebellum,
but no efforts to identify climbing bers
The similarities between all the different
have been made in myxinoids and lampreys.
cerebellum-like and cerebellar structures can-
Purkinje cellspecic markers that do not stain
not be explained solely by homology in the
cerebellum-like structures could also help
sense of historical or phylogenetic homology
determine the presence of a cerebellum. Thus
(Butler & Saidel 2000). In this usage of the term,
the nding that the Zebrin II antibody does
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
trosensory signals that could be used to pre- natural candidates for the site of plastic changes.
dict the electrosensory consequences of the an- Negative image formation requires that the
imals own behavior. Direct evidence for the plasticity be anti-Hebbian in character, i.e., cor-
generation of such predictions has been ob- relations between pre- and postsynaptic ac-
tained from in vivo recordings from princi- tivity should decrease synaptic strength, and
pal cells in the mormyrid and gymnotid ELL researchers have obtained evidence for anti-
and elasmobranch DON (Bastian 1996a, Bell Hebbian plasticity at parallel ber synapses with
1981a, Bell et al. 1997b, Bodznick et al. 1999). principal cells in all three classes of sh. Anti-
In each case, pairing articial electrosensory Hebbian plasticity at parallel ber synapses has
stimuli with central predictive signalsa corol- also been shown recently in the DCN of mam-
lary discharge signal at a particular delay after mals (Fujino & Oertel 2003, Tzounopoulos
the EOD motor command in the case of the et al. 2004) but has not yet been connected to
mormyrid ELL (Figure 5a), a proprioceptive systems-level adaptive ltering.
signal at a particular tail angle in the case of the Modeling studies have helped to link the
gymnotid ELL (Figure 5b), and a propriocep- properties of negative image formation with
tive or corollary discharge signal at a particular mechanisms of synaptic plasticity (Nelson &
phase of the ventilatory cycle in the case of the Paulin 1995, Roberts 1999, Roberts & Bell
elasmobranch DON (Figure 5c)results in a 2000). Temporal specicity is a key feature of
change in the response to the predictive sig- negative image formation. In the mormyrid
nals alone that resembles a negative image of ELL, parallel bers convey corollary discharge
the response to the previously paired (and now signals related to the motor command that
predicted) stimulus. The negative images de- drives the EOD. Pairing with electrosensory
velop rapidly over the course of a few minutes stimuli at various delays relative to the motor
of pairing and are specic to the sign as well command results in negative images that are
as to the spatial and temporal patterns of activ- specic to the paired delay (Bell 1982). Re-
ity evoked by the stimulus. On the basis of these sults of modeling studies suggest that tempo-
results investigators suggested that cerebellum- rally specic negative images could be gener-
like circuitry could operate as an adaptive lter ated using an anti-Hebbian learning rule similar
by continually generating and updating sensory to that observed experimentally (see below) to-
predictions on the basis of associations between gether with an array of parallel ber inputs
central signals and current sensory inputs and active at different delays following the mo-
subtracting these predictions from the neural tor command (Roberts 1999, Roberts & Bell
response. Adaptive ltering could thus allow 2000). The mechanisms for generating tem-
external electrosensory signals to be detected porally specic negative images in this model
more easily. are quite similar to those proposed for some
Several lines of evidence conrm that for- forms of cerebellar learning, such as the learn-
mation of negative images is due, at least in ing of adaptively timed responses in classical
large part, to plastic changes occurring within eye-blink conditioning (Medina et al. 2000) or
12 Bell
Han Sawtell
466
a b c
Mormyrid Gymnotid Elasmobranch
ELL ELL DON
EOD Ventilation
Ex In
Command Tail bend alone
alone alone before
before before
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
EOD
0 min
(9 min of C+S pairing) Command
plus
stimulus 1.5 min
Tail bend
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
15 min
EOD
Command 25 min
2 min
alone
after Tail bend
alone
after 2 min Ventilation
4 min alone
after
8 min
0 ms 160 20 0 +20 0 20
0 1
EOD
Command Tail displacement () s
Figure 5
Formation of negative images of predicted sensory responses in three different cerebellum-like structures. (a) Raster display of the
responses of a cell in the ampullary region of the mormyrid ELL. Each dot represents an action potential. The EOD motor command
occurs at time 0. The command alone initially has no effect on the cell. An electrosensory stimulus (vertical black line) that evokes a
pause-burst response is then paired with the command. After several minutes of pairing, the stimulus is turned off and a response to
command alone is revealed, which was not present before the pairing and which is a negative image of the previously paired sensory
response. From Bell 1986. (b) Raster display of responses of cell in the gymnotid ELL. The tail is moved back and forth passively. Each
row of dots shows response to one movement cycle. Initially the tail bend has no effect on the cell. An electrosensory stimulus that
evokes a burst-pause is then delivered in phase with the movement. The electrosensory stimulus is turned off after several minutes of
pairing, which reveals a response to tail bending alone that was not present before the pairing and which is opposite to the previously
paired sensory response. From Bastian 1995. (c) Histogram display of responses of a cell in the elasmobranch DON. Initially the cell
does not respond to the exhalation (Ex)inhalation (In) ventilatory cycle of the sh (top histogram). An electrosensory stimulus that
evokes a burst-pause is then delivered in phase with the ventilatory cycle. The response to ventilation plus the electrosensory stimulus
decreases during 25 min of pairing. Turning off the electrosensory stimulus after pairing reveals the presence of a response to ventilation
alone, which was not present before and which is a negative image of the previously paired sensory response. From Bodznick 1993.
of appropriate phase relations in the vestibular tive to the parallel ber evoked excitatory post-
ocular reex (Raymond & Lisberger 1998). synaptic potential (EPSP) onset. Depression
The cellular properties of anti-Hebbian develops when a postsynaptic dendritic spike
synaptic plasticity have been studied in some occurs within 50 ms of EPSP onset, whereas
detail at synapses between parallel bers and other timing relations yield potentiation or
Purkinje-like medium ganglion cells in an in no effect. Potentiation as measured in vitro is
vitro preparation of the mormyrid ELL (Bell nonassociative and likely depends on simple
et al. 1997c, Han et al. 2000). Synaptic depres- repetition of the parallel ber stimuli at a suf-
sion requires a postsynaptic dendritic spike and ciently high rate, although in vivo experiments
depends on the precise timing of the spike rela- suggest a spike timingdependent component
the DCN is also anti-Hebbian, spike timing (Safo & Regehr 2007, Wang et al. 2000).
dependent, NMDA dependent, and presynap- The mechanisms of synaptic plasticity are
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
tically expressed (Tzounopoulos et al. 2004, clearly not the same in the cerebellum and in
2007). the cerebellum-like structures where it has been
Investigators have observed both similar- studied. Plasticity at parallel ber synapses onto
ities and differences between plasticity in efferent or Purkinje-like cells in the mormyrid
cerebellum-like structures and plasticity in the ELL (Han et al. 2000) and the mammalian
cerebellum itself. The depression of responses DCN (Tzounopoulos et al. 2004) depends on
to signals conveyed by parallel bers following activation of NMDA receptors, but synap-
the pairing of these signals with postsynaptic tic plasticity at parallel ber synapses onto
excitation in cerebellum-like structures is sim- Purkinje cells does not (Ito 2001). However,
ilar to the depression of responses to parallel some aspects of the plasticity mechanisms may
ber stimulation in the mammalian Purkinje be shared as indicated by the presence of
cells following pairing with climbing ber input the GluRdelta2 gene in the cerebellum and in
or with postsynaptic depolarization (Ito 2001). cerebellum-like structures, and by the involve-
Such depression has been linked to the forma- ment of this gene in plasticity at Purkinje cell
tion of negative images of predicted sensory in- synapses (Hirano et al. 1995).
put in cerebellum-like structures and to mo- Adaptive processes in the cerebellum appear
tor learning in the mammalian cerebellum (Ito similar to those in cerebellum-like structures. In
1984). It is of interest in this regard that the cerebellum-like structures, the pairing of par-
timing of stimulus-driven parallel berevoked allel ber signals with excitatory input from
simple spike activity is consistently close to the the periphery results in such signals eliciting
inverse of climbing ber responses in almost all a predictive reduction in principal cell activity.
the systems where this relation has been exam- In the cerebellum, the pairing of parallel ber
ined (Barmack & Shojaku 1992, Ebner et al. signals with climbing ber input likely leads to
2002, Graf et al. 1988, Kobayashi et al. 1998, such signals eliciting a reduction in the ring
Stone & Lisberger 1990). Thus in many sys- of Purkinje cells (but see Steuber et al. 2007 for
tems, simple spike activity is a kind of negative a contrary view). If the climbing bers convey
image of predicted climbing ber activity. Plas- some type of sensory signal, gated through the
ticity at parallel ber synapses may play a role in inferior olive, then the parallel ber signals that
generating the antiphase relation, but it is only are paired with the climbing bers, and which
part of the explanation because the antiphase predict their occurrence, will reduce Purkinje
relation is still present when parallel ber LTD cell activity, as shown by Jirenhed et al. (2007)
is blocked (Goossens et al. 2004). during eye-blink conditioning.
The timing constraints on parallel ber plas- This review focuses on sensory predictions
ticity may be more restrictive in cerebellum- through mechanisms of associative synaptic
like structures than in the cerebellum. LTD in plasticity and with those features of cerebellum-
the cerebellum-like structures where timing re- like structures that are particularly relevant to
lations have been tested occurred only when cerebellar function. Cerebellum-like structures
the postsynaptic spike followed the presynap- are also excellent sites for addressing other
14 Bell
Han Sawtell
468
important issues in neuroscience, which cannot lum then generates a prediction about the sen-
be discussed here because of space constraints. sory consequences of the commanded motor act
These include the roles of recurrent feedback in the current context. In an inverse model, the
Forward model:
from higher to lower levels of the same sen- desired goal of an action together with infor- predicts the future
sory system (Chacron et al. 2003, 2005; Do- mation about the current state are conveyed to state of the system on
iron et al. 2003), the effects of motor commands the cerebellum, which then generates the pre- the basis of the current
on sensory processing (Bell & Grant 1992), the cise motor commands that will yield the desired state and the motor
command
preservation and analysis of temporal informa- goal. Both types of models must be capable of
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
tion (Kawasaki 2005), and the neural process- plastic change or learning to adapt to changes Inverse model:
generates an
ing of spectral cues for sound localization in the in the task or in the system, such as changes in
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
appropriate motor
DCN (Young & Davis 2002). load or initial limb position. command that will
Forward models are particularly important cause a desired change
in generating fast, coordinated movement se- in the state of the
Predictions in the Cerebellum quences. Feedback from peripheral sensory re- system
The many similarities between cerebellum-like ceptors is slow. An appropriate command for
structures and the cerebellum suggest that the one phase of a movement must often be issued
cerebellum too may be involved in generating before peripheral feedback can arrive about the
predictions concerning expected sensory input consequences of a motor command that evoked
or states of the system (Bell et al. 1997a, Devor a previous phase of the movement. A forward
2000), and a variety of experimental, clinical, model that predicts the sensory consequences
and theoretical studies of the cerebellum sup- of a motor command, accounting for all that is
port this hypothesis (Diedrichsen et al. 2007, known about the current state of the system, al-
Nixon 2001, Paulin 2005, Wolpert et al. 1998). lows the next motor command in a sequence to
The probable involvement of the cere- be issued appropriately and in accord with the
bellum in predictive or feedforward control expected consequences of previous commands.
through learning is well recognized (Bastian Such a process allows for the chunking of sepa-
2006, Ito 1984, Miall et al. 1993, Ohyama et al. rate components of a motor sequence and their
2003, Wolpert et al. 1998). Predictive control automatization, as described by Nixon (2001).
allows for prior knowledge to shape an action, Moreover, classic symptoms of cerebellar dam-
as in knowing if a cup is full or empty before age such as decomposition of movement, slow-
picking it up. Several studies indicate that pre- ness, and tremor can all be understood as due
dictive feedforward control is decient in cere- to the absence of predictive forward models and
bellar patients (Morton & Bastian 2006, Smith reliance on peripheral feedback (Bastian 2006,
& Shadmehr 2005). Such patients do not adapt Nixon 2001).
their responses to predictable perturbations, al- What is required in such automatization of
though they respond quite well to sudden un- a sequence of movements is the predicted ef-
predictable perturbations of a movement, indi- fect of the motor command: the sensory con-
cating that feedback control from the periphery sequences or state that results from the action,
is functional. not simply the motor command itself. Recent
Theoreticians have proposed that the cere- experiments by Pasalar et al. (2006) suggest
bellum may act in an adaptive and predictive that the Purkinje cell output from large regions
manner through the generation of two types of the cerebellar hemispheres is indeed more
of models: forward models and inverse mod- tightly coupled with predictions about conse-
els (Wolpert et al. 1998). In a forward model, quences of the movement than with the mo-
copies of a motor command are conveyed to the tor commands themselves (but see Yamamoto
cerebellum together with information about et al. 2007). Pasalar et al. (2006) recorded from
the current state of the system such as posi- Purkinje cells over a wide area of the hemi-
tions and velocities of the limbs. The cerebel- sphere in monkeys that had been trained to
simple spikes depended only on the position, di- sory stimulus, retinal slip (signaled by climbing
rection, and velocity of the movement. Purkinje bers), by the occurrence of another sensory
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
cell activity was phase advanced, that is, predic- stimulus, vestibular input (signaled by mossy
tive of the movement parameters or state of the bers). More broadly, Paulin (1993, 2005) has
arm (T. Ebner, personal communication). suggested that the cerebellum estimates future
Pasalar et al. (2006) took their results as an states of the organism or environment using
argument against an inverse model in the cere- a combination of sensory, motor, and possibly
bellum because Purkinje cell activity had little other types of information.
relation to the motor commands to the mus- In simpler systems, such as the vestibular
cles. Although one could argue that the ac- ocular reex, in which Purkinje cell output is
tivity reects a high-level motor command, in coupled quite directly with motor pathways, the
movement rather than muscle coordinates, the adaptive alteration in Purkinje cell activity after
simpler explanation is that the Purkinje cell ac- pairing with the climbing ber can be viewed as
tivity reects a forward model of expected con- either a prediction about a sensory input or as
sequences, as required for the automatization of a motor command. In more complex systems,
movement sequences. Their experiments sug- where Purkinje cell output is less tightly cou-
gest that not all sensory consequences are pre- pled with motor pathways, as in the tracking
dicted; only those critical for accomplishing the task studied by Pasalar et al. (2006), the hy-
task are predicted. Thus presumed changes in pothesis of Purkinje cell activity as a predic-
touch or muscle receptors associated with force tor of consequences may provide a more useful
changes were not predicted by Purkinje cell ac- perspective.
tivity; only velocity and position of the limb
were predicted.
Examples of what are, in effect, forward DIRECTIONS FOR
models in the cerebellum-like structures of FUTURE RESEARCH
mormyrid and elasmobranch sh are described Our understanding of adaptive processing in
in the previous section, showing that forward cerebellum-like structures is far from complete,
models can indeed be generated within struc- and future work will be useful both for under-
tures such as the cerebellum. In these sys- standing the neural mechanisms of sensory pro-
tems, corollary discharge signals come to elicit cessing and for understanding the cerebellum.
a prediction about the sensory input pattern Promising lines of research are outlined briey
that is expected to follow the motor com- below.
mand. The possibility of such corollary dis-
charge effects in the OTML and DCN was also
mentioned. Activity Patterns in Granule Cells
Cerebellum-like structures can generate How the different types of predictive inputs are
predictions on the basis of other sensory in- combined and represented in the granule cells
puts (Bastian 1996a, Bodznick et al. 1999), not that are associated with cerebellum-like struc-
just on the basis of motor commands, and the tures remains unclear, as is also the case for cere-
cerebellum may do so also. For example, in bellar granule cells.
16 Bell
Han Sawtell
470
Adaptive Filtering in ber inputs in auditory processing. Similarly,
Electrosensory Systems very little is known about adaptive ltering in
the MON or OTML.
Several aspects of adaptive ltering in
cerebellum-like structures require further
investigation, including (a) the behavioral Purkinje-Like Cells
consequences of adaptive ltering; (b) the The functional roles of Purkinje-like cells
effects of adaptive ltering on encoding natu- remain unclear. Recent work has shown
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
ralistic stimuli in the presence of self-generated that dendritic spikes that drive anti-Hebbian
interference; (c) the mechanisms of plasticity plasticity in Purkinje-like MG cells of the
and the presence of plasticity at other sites, mormyrid ELL are strongly regulated by
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
such as inhibitory synapses; and (d ) the possible central signals, suggesting a parallel to su-
generation of more complex expectations such pervised learning mediated by climbing ber
as those based on memories of entire scenes inputs to the cerebellum (Sawtell et al. 2007).
or sequences. The possibility of more complex In addition, the mormyrid ELL, the DCN, and
expectations is suggested by the massive de- the teleost cerebellum all provide excellent op-
scending inputs that cerebellum-like structures portunities for examining interactions between
receive from higher levels of the same sensory Purkinje or Purkinje-like cells and neighboring
systems. efferent cells (analogous to deep cerebellar nu-
clear cells in the mammalian cerebellum).
DISCLOSURE STATEMENT
The authors are not aware of any biases that might be perceived as affecting the objectivity of this
review.
ACKNOWLEDGMENTS
We thank Drs. Neal Barmack, Timothy Ebner, and Johannes Meek for their critical reviews of
the manuscript. The work was supported by grants from the National Institutes of Health (MH
49792 to C.C.B. and NS44961 to V.H.) and the National Science Foundation (IOB 0618212 to
N.B.S.) and by a National Research Service Award (NS049728 to N.B.S.).
LITERATURE CITED
Acampora D, Avantaggiato V, Tuorto F, Simeone A. 1997. Genetic control of brain morphogenesis
through Otx gene dosage requirement. Development 124(18):363950
Barmack NH, Shojaku H. 1992. Vestibularly induced slow oscillations in climbing ber responses
of Purkinje cells in the cerebellar nodulus of the rabbit. Neuroscience 50:15
J. Neurosci. 10:122640
Bastian J, Zakon H. 2005. Plasticity of sense organs and brain. See Bullock et al. 2005,
pp. 195228
Bell CC. 1981a. An efference copy modied by reafferent input. Science 214:45053
Bell CC. 1981b. Central distribution of octavolateral afferents and efferents in a teleost (Mormyri-
dae). J. Comp. Neurol. 195:391414
Bell CC. 1981c. Some central connections of medullary octavolateral centers in a mormyrid sh.
In Hearing and Sound Communication in Fishes, ed. RR Fay, AN Popper, WN Tavolga, pp.
38392. Berlin: Heidelberg, Springer-Verlag
Bell CC. 1982. Properties of a modiable efference copy in electric sh. J. Neurophysiol. 47:104356
Bell CC. 1986. Duration of plastic change in a modiable efference copy. Brain Res. 369:29
36
Bell CC. 2001. Memory-based expectations in electrosensory systems. Curr. Opin. Neurobiol.
11:48187
Bell CC. 2002. Evolution of cerebellum-like structures. Brain Behav. Evol. 59:312
26
Bell CC, Bodznick D, Montgomery J, Bastian J. 1997a. The generation and subtraction
of sensory expectations within cerebellum-like structures. Brain Behav. Evol. 50:17
31
Bell CC, Caputi A, Grant K. 1997b. Physiology and plasticity of morphologically identied cells
in the mormyrid electrosensory lobe. J. Neurosci. 17:640922
Bell CC, Caputi A, Grant K, Serrier J. 1993. Storage of a sensory pattern by anti-Hebbian synaptic
plasticity in an electric sh. Proc. Natl. Acad. Sci. USA 90:465054
Bell CC, Finger TE, Russell CJ. 1981. Central connections of the posterior lateral line lobe in
mormyrid sh. Exp. Brain Res. 42:922
Bell CC, Grant K. 1992. Corollary discharge effects and sensory processing in the mormyromast
regions of the mormyrid electrosensory lobe: II. Cell types and corollary discharge plasticity.
J. Neurophysiol. 68:85975
Bell CC, Grant K, Serrier J. 1992. Corollary discharge effects and sensory processing in the
mormyrid electrosensory lobe: I. Field potentials and cellular activity in associated structures.
J. Neurophysiol. 68:84358
Bell CC, Han VZ, Sugawara S, Grant K. 1997c. Synaptic plasticity in a cerebellum-like structure
depends on temporal order. Nature 387:27881
Bell CC, Maler L. 2005. Central neuroanatomy of electrosensory systems in sh. See Bullock
et al. 2005, pp. 68111
Bell CC, Russell CJ. 1978. Termination of electroreceptor and mechanical lateral line afferents in
the mormyrid acousticolateral area. J. Comp. Neurol. 182:36782
Bell CC, von der Emde G. 1995. Electric organ corollary discharge pathways in mormyrid sh:
II. The medial juxtalobar nucleus. J. Comp. Physiol. A. 177:46379
18 Bell
Han Sawtell
472
Berman N, Dunn RJ, Maler L. 2001. Function of NMDA receptors in a feedback pathway of the
electrosensory system. J. Neurophysiol. 86:161221
Berrebi AS, Morgan JI, Mugnaini E. 1990. The Purkinje cell class may extend beyond the cere-
bellum. J. Neurocytol. 19(5):64354
Bodznick D. 1993. The specicity of an adaptive lter that suppresses unwanted reafference in
electrosensory neurons of the skate medulla. Biol. Bull. 185:31214
Bodznick D, Boord RL. 1986. Electroreception in Chondrichthyes: central anatomy and physi-
ology. See Bullock & Heiligenberg 1986, pp. 22556
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
Bodznick D, Montgomery JC, Carey M. 1999. Adaptive mechanisms in the elasmobranch hind-
brain. J. Exp. Biol. 202:135764
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
Bodznick D, Northcutt RG. 1980. Segregation of electro- and mechanoreceptive inputs to the
elasmobranch medulla. Brain Res. 195:31321
Braford MR. 1982. African, but not Asian, notopterid shes are electroreceptive: evidence from
brain characters. Neurosci. Lett. 32:3539
Brown MC, Berglund AM, Kiang NY, Ryugo DK. 1988. Central trajectories of type II spiral
ganglion neurons. J. Comp. Neurol. 278(4):58190
Bullock TH, Bodznick DA, Northcutt RG. 1983. The phylogenetic distribution of electrorecep-
tion: evidence for convergent evolution of a primitive vertebrate sense modality. Brain Res.
Rev. 6:2546
Bullock TH, Heiligenberg W. 1986. Electroreception. New York: Wiley
Bullock TH, Hopkins CD, Popper AN, Fay RR, eds. 2005. Electroreception. New York: Springer
Burian M, Gstoettner W. 1988. Projection of primary vestibular afferent bres to the cochlear
nucleus in the guinea pig. Neurosci. Lett. 84(1):1317
Butler AB, Saidel WM. 1992. Tectal projection to an unusual nucleus in the diencephalon of a
teleost sh, Pantodon buchholzi. Neurosci. Lett. 145:19396
Butler AB, Saidel WM. 2000. Dening sameness: historical, biological, and generative homology.
BioEssays 22(9):84653
Caicedo A, Herbert H. 1993. Topography of descending projections from the inferior colliculus
to auditory brainstem nuclei in the rat. J. Comp. Neurol. 328(3):37792
Campbell HR, Meek J, Zhang J, Bell CC. 2007. Anatomy of the posterior caudal lobe of the cere-
bellum and the eminentia granularis posterior in a mormyrid sh. J. Comp. Neurol. 502(5):714
35
Cant NB. 1992. The cochlear nucleus: neuronal types and their synaptic organization. In
The Mammalian Auditory Pathway: Neuroanatomy, ed. DB Webster, AN Popper, RR Fay,
pp. 66116. New York: Springer
Caputi AA, Castello ME, Aguilera P, Trujillo-Cenoz O. 2002. Electrolocation and electrocommu-
nication in pulse gymnotids: signal carriers, prereceptor mechanisms and the electrosensory
mosaic. J. Physiol. Paris 96(56):493505
Carr CE, Maler L. 1986. Electroreception in gymnatiform sh: central anatomy and physiology.
See Bullock & Heiligenberg 1986, pp. 31974
Chacron MJ, Doiron B, Maler L, Longtin A, Bastian J. 2003. Non-classical receptive eld mediates
switch in a sensory neurons frequency tuning. Nature 423(6935):7781
Chacron MJ, Maler L, Bastian J. 2005. Feedback and feedforward control of frequency tuning to
naturalistic stimuli. J. Neurosci. 25(23):552132
Conley RA, Bodznick D. 1994. The cerebellar dorsal granular ridge in an elasmobranch has pro-
prioceptive and electroreceptive representations and projects homotopically to the medullary
electrosensory nucleus. J. Comp. Physiol. A 174:70721
Crosby EC. 1969. Comparative aspects of cerebellar morphology. In Neurobiology of Cerebellar
Evolution and Development, ed. R Llinas, pp. 1941. Chicago: Am. Med. Assoc.
Dupont JL, Gardette R, Crepel F. 1987. Postnatal development of the chemosensitivity of rat
cerebellar Purkinje cells to excitatory amino acids. An in vitro study. Brain Res. 431(1):5968
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
Ebner TJ, Johnson MT, Roitman A, Fu Q. 2002. What do complex spikes signal about limb
movements? Ann. N.Y. Acad. Sci. 978:20518
Ekerot CF, Jorntell H. 2001. Parallel bre receptive elds of Purkinje cells and interneurons are
climbing bre-specic. Eur. J. Neurosci. 13:130310
Finger TE. 1978. Efferent neurons of the teleost cerebellum. Brain Res. 153:60814
Finger TE, Tong SL. 1984. Central organization of eighth nerve and mechanosensory lateral line
systems in the brainstem of ictalurid catsh. J. Comp. Neurol. 229:12951
Fujino K, Oertel D. 2003. Bidirectional synaptic plasticity in the cerebellum-like mammalian
dorsal cochlear nucleus. Proc. Natl. Acad. Sci. USA 100(1):26570
Goossens HH, Hoebeek FE, van Alphen AM, Van Der SJ, Stahl JS, et al. 2004. Simple spike and
complex spike activity of occular Purkinje cells during the optokinetic reex in mice lacking
cerebellar long-term depression. Eur. J. Neurosci. 19(3):68797
Graf W, Simpson JI, Leonard CS. 1988. Spatial organization of visual messages of the rabbits
cerebellar occulus. II. Complex and simple spike responses of Purkinje cells. J. Neurophysiol.
60(6):2091121
Grant K, Sugawara S, Gomez L, Han VZ, Bell CC. 1998. The Mormyrid electrosensory lobe in
vitro: physiology and pharmacology of cells and circuits. J. Neurosci. 18:600925
Han VZ, Grant G, Bell CC. 2000. Reversible associative depression and nonassociative potenti-
ation at a parallel ber synapse. Neuron 27:61122
Hawkes R, Herrup K. 1995. Aldolase C/zebrin II and the regionalization of the cerebellum.
J. Mol. Neurosci. 6(3):14758
Hirano T, Kasono K, Araki K, Mishina M. 1995. Suppression of LTD in cultured Purkinje cells
decient in the glutamate receptor d2 subunit. NeuroReport 6:52426
Hjelmstad GO, Parks G, Bodznick D. 1996. Motor corollary discharge activity and sensory re-
sponses related to ventilation in the skate vestibulolateral cerebellum: implications for elec-
trosensory processing. J. Exp. Biol. 199:67381
Ito M. 1984. The Cerebellum and Neural Control. New York: Raven
Ito M. 2001. Cerebellar long-term depression: characterization, signal transduction, and functional
roles. Physiol. Rev. 81(3):114395
Jirenhed DA, Bengtsson F, Hesslow G. 2007. Acquisition, extinction, and reacquisition of a cere-
bellar cortical memory trace. J. Neurosci. 27(10):2493502
Kanold PO, Young ED. 2001. Proprioceptive information from the pinna provides somatosensory
input to cat dorsal cochlear nucleus. J. Neurosci. 21(19):784858
Kawasaki M. 2005. Physiology of tuberous electrosensory systems. See Bullock et al. 2005,
pp. 154194
Kawasaki M, Guo YX. 1998. Parallel projection of amplitude and phase information from the hind-
brain to the midbrain of the African electric sh Gymnarchus niloticus. J. Neurosci. 18(18):7599
611
20 Bell
Han Sawtell
474
Kim JJ, Thompson RF. 1997. Cerebellar circuits and synaptic mechanisms involved in classical
eyeblink conditioning. Trends Neurosci. 20:17781
Kobayashi Y, Kawano K, Takemura A, Inoue Y, Kitama T, et al. 1998. Temporal ring patterns
of Purkinje cells in the cerebellar ventral paraocculus during ocular following responses in
monkeys II. Complex spikes. J. Neurophysiol. 80(2):83248
Lannoo M, Hawkes R. 1997. A search for primitive Purkinje cells: zebrin II expression in sea
lampreys (Petromyzon marinus). Neurosci. Lett. 237:5355
Lannoo MJ, Maler L, Hawkes R. 1992. Zebrin II distinguishes the ampullary organ receptive map
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
from the tuberous organ receptive maps during development in the teleost electrosensory
lateral line lobe. Brain Res. 586:17680
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
Lannoo MJ, Ross L, Maler L, Hawkes R. 1991. Development of the cerebellum and its extracellular
Purkinje cell projection in teleost shes as determined by zebrin II immunocytochemistry.
Prog. Neurobiol. 37:32963
Larsell O. 1967. The Comparative Anatomy and Histology of the Cerebellum from Myxinoids through
Birds. Minneapolis: Univ. Minn. Press
Lisberger SG, Fuchs AF. 1974. Response of occulus Purkinje cells to adequate vestibular stim-
ulation in the alert monkey: xation vs compensatory eye movements. Brain Res. 69:34753
Maekawa K, Simpson JI. 1972. Climbing ber activation of Purkinje cells in the occulus by
impulses transferred through the visual pathway. Brain Res. 39:24551
Maler L, Sas EKB, Rogers J. 1981. The cytology of the posterior lateral line lobe of high frequency
weakly electric sh (Gymnotidae): dendritic differentiation and synaptic specicity in a simple
cortex. J. Comp. Neurol. 195:87140
Manis PB, Molitor SC. 1996. N-methyl-d-aspartate receptors at parallel ber synapses in the
dorsal cochlear nucleus. J. Neurophysiol. 76:163955
Martinez S, Crossley PH, Cobos I, Rubenstein JL, Martin GR. 1999. FGF8 induces formation
of an ectopic isthmic organizer and isthmocerebellar development via a repressive effect on
Otx2 expression. Development 126(6):1189200
Matsushita A, Kawasaki M. 2005. Neuronal sensitivity to microsecond time disparities in the
electrosensory system of Gymnarchus niloticus. J. Neurosci. 25(49):1142432
Matsushita M, Tanami T. 1987. Spinocerebellar projections from the central cervical nucleus in
the cat, as studied by anterograde transport of wheat germ agglutinin-horseradish peroxidase.
J. Comp. Neurol. 266(3):37697
McCormick CA. 1997. Organization and connections of octaval and lateral line centers in the
medulla of a clupeid, Dorosoma cepedianum. Hear Res. 110(12):3960
McCormick CA. 1999. Anatomy of the central auditory pathways of sh and amphibians. In
Comparative Hearing: Fish and Amphibians, ed. RR Fay, AN Popper, pp. 155217. New York:
Springer Verlag
McCreery DB. 1977. Two types of electroreceptive lateral lemniscal neurons of the lateral line
lobe of the catsh Ictalurus nebulosus; connections from the lateral line nerve and steady-state
frequency response characteristics. J. Comp. Physiol. 113:31739
Medina JF, Garcia KS, Nores WL, Taylor NM, Mauk MD. 2000. Timing mechanisms in the
cerebellum: testing predictions of a large-scale computer simulation. J. Neurosci. 20(14):5516
25
Meek J. 1983. Functional anatomy of the tectum mesencephali of the goldsh. An explorative
analysis of the functional implication of the laminar structural organization of the tectum.
Brain Res. Rev. 6:24797
Meek J. 1998. Holosteans and teleosts. In The Central Nervous System of Vertebrates, ed. R Nieuwen-
huys, HJ Ten Donkelaar, C Nicholson, Vol. 15, pp. 759937. Berlin: Springer
sion of ventilatory reafference in the electrosensory system of the little skate Raja erinacea.
J. Exp. Biol. 183:20315
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
Montgomery JC, Bodznick D. 1999. Signals and noise in the elasmobranch electrosensory system.
J. Exp. Biol. 202(Pt. 10):134955
Montgomery JC, Coombs S, Conley RA, Bodznick D. 1995. Hindbrain sensory processing in
lateral line, electrosensory, and auditory systems: a comparative overview of anatomical and
functional similarities. Auditory Neurosc. 1:20731
Morton SM, Bastian AJ. 2006. Cerebellar contributions to locomotor adaptations during splitbelt
treadmill walking. J. Neurosci. 26(36):910716
Mugnaini E, Dino MR, Jaarsma D. 1997. The unipolar brush cells of the mammalian cerebellum
and cochlear nucleus: cytology and microcircuitry. Prog. Brain Res. 114:13150
Mugnaini E, Osen KK, Dahl AL, Friedrich VL Jr, Korte G. 1980a. Fine structure of granule cells
and related interneurons (termed Golgi cells) in the cochlear nuclear complex of cat, rat and
mouse. J. Neurocytol. 9(4):53770
Mugnaini E, Warr WB, Osen KK. 1980b. Distribution and light microscopic features of granule
cells in the cochlear nuclei of cat, rat, and mouse. J. Comp. Neurol. 191(4):581606
Nelson ME, Paulin MG. 1995. Neural simulations of adaptive reafference suppression in the
elasmobranch electrosensory system. J. Comp. Physiol. A 177:72336
Nieuwenhuys R, Ten Donkelaar HJ, Nicholson C. 1997. The Central Nervous System of Vertebrates.
Heidelberg: Springer
Nixon PD, Passingham RE. 2001. Predicting sensory events: the role of the cerebellum in motor
learning. Exp. Brain. Res. 138:25157
Noda H, Warabi T. 1982. Eye position signals in the occulus of the monkey during smooth-
pursuit eye movements. J. Physiol. 324:187202
Northmore DPM, Williams B, Vanegas H. 1983. The teleostean torus longitudinalis: responses
related to eye movements, visuotopic mapping, and functional relations with the optic tectum.
J. Comp. Physiol. A 150:3950
Ohlrogge M, Doucet JR, Ryugo DK. 2001. Projections of the pontine nuclei to the cochlear
nucleus in rats. J. Comp. Neurol. 436(3):290303
Ohyama T, Nores WL, Murphy M, Mauk MD. 2003. What the cerebellum computes. Trends
Neurosci. 26(4):22227
Pasalar S, Roitman AV, Durfee WK, Ebner TJ. 2006. Force eld effects on cerbellar Purkinje cell
discharge with implications for internal models. Nat. Neurosci. 9(11):140411
Paulin MG. 1993. The role of the cerebellum in motor control and perception. Brain Behav. Evol.
41:3950
Paulin MG. 2005. Evolution of the cerebellum as a neuronal machine for Bayesian state estimation.
J. Neural. Eng. 2(3):S21934
Petralia RS, Wang YX, Zhao HM, Wenthold RJ. 1996. Ionotropic and metabotropic glutamate
receptors show unique postsynaptic, presynaptic, and glial localizations in the dorsal cochlear
nucleus. J. Comp. Neurol. 372(3):35683
22 Bell
Han Sawtell
476
Puzdrowski RL, Leonard RB. 1993. The octavolateral systems in the stingray, Dasyatis sabina. I.
Primary projections of the octaval and lateral line nerves. J. Comp. Neurol. 332(1):2137
Raymond JL, Lisberger SG. 1998. Neural learning rules for the vestibulo-ocular reex.
J. Neurosci. 18(21):911229 (Abstr.)
Roberts PD. 1999. Computational consequences of temporally asymmetric learning rules: I. dif-
ferential Hebbian learning. J. Comp. Neurosci. 7:23546
Roberts PD, Bell CC. 2000. Computational consequences of temporally asymmetric learning
rules: II. sensory image cancellation. J. Comput. Neurosci. 9:6783
Annu. Rev. Neurosci. 2008.31:1-24. Downloaded from arjournals.annualreviews.org
Robertson LT. 1985. Somatosensory representation of the climbing ber system in the rostral
intermediate cerebellum. Exp. Brain Res. 61(1):7386
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
Ronan M 1986. Electroreception in cyclostomes. See Bullock & Heiligenberg 1986, pp. 20924
Safo P, Regehr WG. 2008. Timing dependence of the induction of cerebellar LTD. Neurophar-
macology. In press
Saunders J, Bastian J. 1984. The physiology and morphology of two types of electrosensory neurons
in the weakly electric sh Apteronotus leptorhynchus. J. Comp. Physiol. A 154:199209
Sawtell NB, Williams A, Bell CC. 2007. Central control of dendritic spikes shapes the responses of
Purkinje-like cells through spike timing-dependent synaptic plasticity. J. Neurosci. 27(7):1552
65
Schlegel P. 1973. Perception of objects in weakly electric sh Gymnotus carapo as studied in record-
ings from rhombencephalic neurons. Exp. Brain Res. 18:34054
Schmidt AW, Bodznick D. 1987. Afferent and efferent connections of the vestibulolateral cere-
bellum of the little skate, Raja erinacea. Brain Behav. Evol. 30:282302
Shore SE, Zhou J. 2006. Somatosensory inuence on the cochlear nucleus and beyond. Hear. Res.
21617:9099
Smith MA, Shadmehr R. 2005. Intact ability to learn internal models of arm dynamics in Hunt-
ingtons disease but not cerebellar degeneration. J. Neurophysiol. 93(5):280921
Steuber V, Mittman W, Hoebeek FE, Silver RA, De Zeeuw CI, et al. 2007. Cerebellar LTD and
pattern recognition by Purkinje cells. Neuron 54(1):12136
Stone LS, Lisberger SG. 1990. Visual responses of Purkinje cells in the cerebellar occulus during
smooth-pursuit eye movements in monkeys. II. Complex spikes. J. Neurophysiol. 63(5):1262
75
Szabo T, Libouban S, Haugede-Carre F. 1979. Convergence of common and specic sensory
afferents to the cerebellar auricle (auricula cerebelli) in the teleost sh Gnathonemus demon-
strated by HRP method. Brain Res. 168:61922
Szabo T, Libouban S, Denizot JP. 1990. A well dened spinocerebellar system in the weakly
electric teleost sh Gnathonemus petersii. Arch. Ital. Biol. 128:22947
Tong S. 1982. The nucleus praeeminentialis: an electro- and mechanoreceptive center in the
brainstem of the catsh. J. Comp. Physiol. A 145:299309
Tong S, Finger TE. 1983. Central organization of the electrosensory lateral line system in bullhead
carsh Ictalurus nebulosus. J. Comp. Neurol. 217:116
Tzounopoulos T, Kim Y, Oertel D, Trussell LO. 2004. Cell-specic, spike timing-dependent
plasticities in the dorsal cochlear nucleus. Nat. Neurosci. 7(7):71925
Tzounopoulos T, Rubio ME, Keen JE, Trussell LO. 2007. Coactivation of pre- and postsynap-
tic signaling mechanisms determines cell-specic spike-timing-dependent plasticity. Neuron
54(2):291301
Vanegas H, Williams B, Freeman JA. 1979. Responses to stimulation of marginal bers in the
teleostean optic tectum. Exp. Brain Res. 34(2):33549
Wang SS, Denk W, Hausser M. 2000. Coincidence detection in single dendritic spines mediated
by calcium release. Nat. Neurosci. 3:126673
24 Bell
Han Sawtell
478
AR346-FM ARI 20 May 2008 15:1
Annual Review of
Neuroscience
Function
Curtis C. Bell, Victor Han, and Nathaniel B. Sawtell p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 1
by CORNELL UNIVERSITY on 10/26/09. For personal use only.
479
v
AR346-FM ARI 20 May 2008 15:1
Indexes
Errata
480
vi Contents
J. comp. Physiol. 89, 331--357 (1974)
9 by Springer-Verlag 1974
Summary. 1. Chases in which male flies (2'annia canicularis) pursue other flies
were studied by filming such encounters from directly below. Males will start to
chase whenever a second fly comes within 10-15 cm (Fig. 3).
2. Throughout these chases there was a continuous relationship between the
angle (0e) made by the leading fly and the direction of flight of the chasing fly, and
the angular velocity of the chasing fly (w]). This relation was approximately linear,
with a slope of 20 ~ s-1 per degree 0e (Figs. 4-7).
3. The maximum correlation between w/and 0e occurs after a lag of approxi-
mately 30 ms, which represents the total delay in the system (Fig, 8).
4. In the region close to the chasing fly's axis (0e less than about 35 ~ a second
mechanism exists in which the angular velocity of the chasing fly (o)/) is controlled
by the relative angular velocity of the leading fly (we), rather than its relative
position. The ratio of w/to o~e in this region is approximately 0.7.
5. Using the results in 2-4 above, and an empirically determined relation
between the angular and forward velocities of the chasing fly, it was possible to
simulate the flight path of the chasing fly, given that of the leading fly (Fig. 11).
Because these simulations predict correctly the manoeuvres and outcomes of quite
complicated chases, it is concluded that the control system actually used by the
fly is accurately described by conclusions 2-4.
6. The physiological implications of this behaviour, and the possible function
of chasing, are discussed.
Introduction
I t is quite c o m m o n t o see houseflies e n g a g e d in w h a t a p p e a r t o be
s h o r t fast chases, which involve m a n y r a p i d changes of course. I f these
are i n d e e d chases, iu t h e sense t h a t one fly is t r y i n g t o c a t c h t h e o t h e r
t h e n this implies t h a t t h e fly which is doing t h e chasing m u s t possess a n
a c c u r a t e a n d r a p i d s y s t e m for d e t e r m i n i n g t h e course of t h e l e a d i n g fly,
a n d for controlling its own flight p a t h accordingly. This p a p e r p r e s e n t s
a n a c c o u n t of filmed records of such chases, a n d i t is shown t h a t these
are cases of real pursuit, a n d n o t meaningless d i s p l a y s of aerobatics.
B y e x a m i n i n g t h e form of t h e chases it should be possible to d e t e r m i n e
t h e k i n d of control s y s t e m t h e chasing fly uses. There are two basic
possibihties. The s y s t e m m i g h t be " c o n t i n u o u s " in t h e sense t h a t some
22*
481
332 M.F. Land and T. S. Collett
Methods
Filming
There are usually two species of fly found inside houses in England, the common
house fly (Musca domestica) and the lesser house fly (~annia canieularis). The
species are easily distinguished by the sharp upward bend of wing vein 4 iC in
M. domestica compared with the straight vein in F. canicularis (see e.g. Colyer and
Hs,mmond, 1951). The other difference which is important for this study is that
male iv. canicularis tend to congregate round prominent objects such as lampshades,
making horizontal "patrolling" flights near them for long periods. M. domestica on
the other hand do not seem to prefer to fly in any particular parts of a room. Males
of both species engage in chases, of females or more commonly of other males, but
because only _~. canicularis can be relied upon to perform these at a known location,
it has so far only been possible to film this species.
Films were made by positioning the camera (16 mm Bolex with a constant
speed motor operating at 50 frames per second) directly beneath a lampshade
frequented by F. canicularis, and pointing upwards. A reduced shutter sector was
used to give an exposure time of 3 ms, which minimised blur. Inspection of chasing
flights from the side and below shows that most manoeuvres take place in the
horizontal plane, and that vertical components of chases are rather small in com-
parison. Where there are vertical components to the chases these seem to be mainly
in the form of spiralling movements, i.e. ones having a nearly constant upward or
downward component, which will reduce the apparent velocity of the flies along
the course as seen from below, but will not greatly change their angular relations,
which are the important determining features of chasing behaviour. Examination
of the flies which had taken part in filmed chases showed that most were males,
and the chases analysed here are almost certainly interactions between males.
~ilms were taken on hot days (about 25~ in normal room daylight (backgrounds
about 50 cd. era-2).
482
Control of Chasing in Flies 333
/
5Of
Fig. 1. Alternative ways of measuring the error angle 0~, and the angular velocity
of the chasing fly cot= ~Oj~t, at the instants measured on the film, and instants
halfway between them
Analysis
The films were analysed frame by frame using a Specto projector, and the
courses of the flies in the horizontal plane were plotted. Usually it was impossible
to determine the exact direction of the fly's axis, and in subsequent analyses it is
assumed that this is the same as the direction of flight. From these course plots
three kinds of information were extracted.
(i) The angle made by the line of flight of the chasing fly and a line joining the
chasing and leading fly: this is referred to as the error angle (0e, Fig. 1), and it is
assumed that this represents the location of the leading fly on the retina of the
chasing fly at a particular instant. The extent to which flies roll about their long
axes during turns is not known ("banking") so this angle cannot simply be inter-
preted as an angle in the chasing fly's longitudinal plane, but in most instances it
is likely that the major component is in this plane, and for the following analysis
it does not matter whether or not this is true. The error angular velocity and error
angular acceleration (coe and ~e) can be obtained from 0e as indicated in Table 1.
Two methods of obtaining 0e are shown in Fig. 1.
(ii) The angular velocity of the chasing fly (~ol). This was measured as the
change in course angle between two frames of film ((~01) divided by the sampling
time ((~t~ 20 ms). Fig. 1 gives two methods of measuring ~0 I. This angular velocity
is taken as being appropriate to an instant halfway between the two sampling
instants. Angular~acceleration (o~/) can be obtained approximately from the dif-
ference between successive values of a)f divided again by the sampling time.
(iii) The forward velocities of leading and chasing flies (vl and vl). These are
taken as the distances travelled along the flight paths between frames of film,
divided by the sampling time. The errors involved in aU these estimates increase
as the angular velocities of the flies increase, but this is not likely to cause serious
errors at angular velocities in the observed range. There does not appear to be any
"fine structure" to the records described that is not resolved at 50 frames per
second.
483
334 M. 1~. Land and T. S. Collett
Table 1. Symbols
0~ Error angle; the angle between a tangent to the flight path of the chasing
fly and a line joining the two flies at a particular instant.
Error angular velocity; rate of change of 0e, measured as ~Oe/~t i.e. the
change in 0e during the sampling time ~t.
~be Error angular acceleration; measured as ~coe/6t.
Angular velocity of the following (chasing) fly; measured as ~O//~s i.e.
the change in course angle during the sampling time.
c51 Angular acceleration of the chasing fly; measured as ~co//($t.
vl, v! Horizontal forward velocities of the leading and following fly; measured
as the distances travelled along the respective flight paths between
sampling instants, divided by the sampling interval.
~t Sampling intervM used in measurement and reconstruction of flight
paths, usually 20 ms.
d Delay or latency of the response; defined as difference in time between
the measurement of an "input" variable (e.g. 0e) and the time of maxi-
mum correlation between this and an "output" variable (e.g. co]).
sign. con- All anti-clockwise angles, angular velocities and angular accelerations
vention are taken as positive, and vice-versa.
Results
General Description o/the _Flight Pattern o/Fannia canicularis
Male Fannia s p e n d m u c h of t h e i r t i m e m a k i n g c h a r a c t e r i s t i c " p a t r o l -
l i n g " flights a r o u n d p r o m i n e n t objects, a n d i t is in t h e course of such
flights t h a t e n c o u n t e r s w i t h o t h e r flies t a k e place. Colyer a n d H a m m o n d
(1951) give a n e x c e l l e n t d e s c r i p t i o n of this b e h a v i o u r , which t h e y s a y is
c h a r a c t e r i s t i c of t h e genus.
" T h i s flight is m a d e on a series of i r r e g u l a r t r i a n g u l a r or q u a d r i l a t e r a l
courses, a n a l m o s t i m p e r c e p t i b l e h o v e r i n g t a k i n g place a t t h e corners
a n d t h e sides being c o v e r e d in a r a p i d d a r t . W h e n u n d i s t u r b e d a n d alone,
t h e s e flies m a i n t a i n a m o r e or less c o n s t a n t h e i g h t a n d r e g u l a r course,
b u t w h e n m o r e t h a n one decides t o p a t r o l t h e same ' b e a t , ' i t u s u a l l y
h a p p e n s t h a t one d a r t s r a p i d l y t o w a r d s t h e other, a s h a r p f l u r r y a n d
p r o m p t dispersal ensues a n d e v e n t u a l l y one of t h e m r e c o m m e n c e s t h e
patrolling."
Two e x a m p l e s of such p a t r o l l i n g flights are shown in Fig. 2. T h e
s t r a i g h t s e g m e n t s are a b o u t 20 cm long, a n d t h e a v e r a g e v e l o c i t y of t h e
flies is 65 cm s -1, a b o u t half t h e speed seen d u r i n g chases. S o m e t i m e s
t w o or e v e n t h r e e m a l e s m a n a g e t o co-exist in t h e s a m e p a r t of t h e room,
b u t w h e n t h i s occurs t h e flies are f o u n d to be s t r a t i f i e d v e r t i c a l l y , so t h a t
484
Fig. 2. Two examples of "patrolling" flights made beneath a lampshade, which is
indicated by the heavy line. Time interval between points: 20 ms
4 4 ~"',
Fig. 3. Patrolling flight ending in a chase. Chasing fly e, leading fly o. Points are
at 20 ms intervals. Corresponding instants in the two flight paths are numbered
at 100 ms intervals. Dotted line indicates relative positions of the flies at the
probable instant of visual contact
485
336 M.F. Land and T. S. Collett
between the flies when the chasing fly makes its first turn is 9 cm, which
is probably a slight under-estimate because there must be some latent
period. However, if 10 cm is taken as the "visual contact distance,"
then a 6 m m long fly will occupy a visual angle of 3.4 ~ b y 2 ~ which is
comparable with an interommatidial angle of about 2 ~ (in the closely
similar housefly Musca).
Females rarely strayed into the area occupied b y males, although
when t h e y did t h e y were chased and sometimes caught, the two flies
joining, and flying in tandem to a nearby surface to copulate. Un-
fortunately, only one instance of what appeared to be a maleflemale
chase was filmed, and this was so brief t h a t little could be learned from
it. B y far the majority of chases were between males.
486
Control of Chasing in Flies 337
I !
"~ 5cm
Fig. 4. Flight paths of chasing (o) and leading (9 flies during the longest recorded
chase. Points at 20 ms intervals. Corresponding instants on the two paths numbered
at 200 ms intervals
487
338 M.F. Land and T. S. Collett
+lod
Oe
-too
+ 2,50C
O.S-1
0"2 s - - ~
Lof
-2,50C
Fig. 5. Plots of error angle (0e) and the angular velocity of the chasing fly (to1)
during the chase shown in Fig. 4, showing the continuous correspondence between
the two. Numbers on the time axis refer to the numbered points in Fig. 4
path, and only if this procedure fails to start looking for more com-
plicated kinds of responses.
On comparing input and output variables it became clear t h a t there
was a relationship, apparently continuous, between error angle (0e) and
the chasing fly's angular velocity (~ol). This is shown in two forms in
Figs. 5 and 6a. Fig. 5 shows t h a t whenever the leading fly departed
from a position ahead of the chasing fly (0~ = 0) the chasing fly turned
towards it with an angular velocity roughly proportional to the deviation.
Comparing the two plots, there seems to be a delay of between 20 and
40 ms between the two, and this will be examined further below. The
scatter diagrams in Fig. 6 are all based on the assumption t h a t the input
variables maximally affect the output 30 ms later. From Fig. 6a it can
be seen t h a t there is a strong and apparently linear relation between
o)I and 0e. The correlation coefficient is 0.76 and is highly significant.
The relationship can be approximated b y the regression line, which has
a slope of 21 s-1 (the full units are degrees per second per degree 0e, or
simply s-l).
488
Control of Chasing in Flies 339
a)
+I T5,ooo
o.s_ .
b)
.
T +5,000
.tuf . . f , % 4
(1 , ~ o~o~t o_ 9 ,
.
, ,:,i"iI!?'"
, l I
*. ; -:.~'..." 9 .. 9D |
ee (,oe
".:~""~-o
5,00
C)
O.Jf
I+2' 9
~ d)
T+2,500
Wf t z's'l "
Q 9 9
. =;
9 "9
, - , l
~'I
' ,
9 Oe
l
+40
:
~ . . . . . .
. . . ' .:l.i.. ~
. .
I "'=, ~ 1
Fig. 6a--d. Scatter diagrams showing the relationships between co! and Oe, and
% and col for all points measured on Fig. 4 and 3 shorter chases (a and b), and for
only those points for which the error angle (0e) lay between + 3 5 ~ and --35 ~
(c and d). In all cases the delay has been taken as 30 ms, i.e. the ordinate of each
point corresponds to an instant 30 ms later than that of its abscissa. The correlation
coefficients are a) 0.76 (p<0.001), b) 0.09 (n.s.), c) 0.38 (0.01< p < 0.02), d) 0.65
( p < 0.002)
489
340 M.F. Land and T. S, Collett
-2,500
o.s4
oo"
, , ~
ee
Fig. 7. Relation between o~l and 0e under conditions that approximate to a steady
state. The points are the average values of o~! measured during periods of 60 to
100 ms, in which the values of 0e were approximately constant. Closed circles are
from Fig. 4, other symbols are taken from four other chases. The straight line has
a slope of 20 s-1
are the first derivatives of 0 e and ~oi, between which there is a relationship.
The question arises as to whether one of these pairs of relationships is
causal, and the other its consequence. I t seems hkely t h a t the Oe, o~l
relation is the real one, because there is no reason w h y the differentiated
relationship should be " z e r o position error seeking." I n other words, if
o~e was actually being used to control d)1 one would not expect the plot
of 0~ vs. r I to pass t h r o u g h zero, which it does.
There is a problem in accepting the value of 21 s -1 as the value t h a t
the fly actually uses in converting 0~ into angular velocity o~t because
of the u n c e r t a i n t y a b o u t the exact latency of the response. Ideally, one
would like to be able to measure this conversion factor u n d e r steady
state conditions, t h a t is, to measure the angular velocity of the chasing
fly when 0~ is k e p t constant. I n free flight this is n o t strictly possible
because the angular velocity of the chasing fly tends to reduce 0e v e r y
rapidly. H o w e v e r during certain manoeuvres, like the loop between
5 and 6 in Fig. 4, a situation does arise in which 0 e stays a p p r o x i m a t e l y
the same for 3-5 frames. F r o m such situations a reasonable approxi-
m a t i o n to the steady state conversion factor can be obtained b y plotting
the average values of w I during these semi-stable periods against the
average value of 0~ in the period t a k e n as starting 30 ms earlier. Such a
plot is shown in Fig. 7. I t is well fitted b y a straight line whose slope is
20 s -1 (i.e. 20 ~ s -1 per o 0~).
F r o m these considerations, the operation performed b y the pursuing
fly can be defined as the conversion of the position of the retinal image
490
Control of Chasing in Flies 341
of the target (the leading fly) into the angular velocity of the pursuer's
flight path. Or :
to] = k 0e, where k--~ 20 s -1. (1)
This, however, is not the only mechanism involved. An examination
of certain of the turns in Fig. 4, notably the right turn following 2 and
the left turn after 4, shows t h a t the chasing fly begins to turn in the
direction of motion of the leading fly, even before the leading fly has
crossed the chasing fly's axis. For such turns the above expression can-
not be true, and one would expect instead that there would be a closer
relation between error angular velocity (toe) and the angular velocity of
the chasing fly (tot)" Fig. 6b shows t h a t in general there is no such
relation between tot and toe: the correlation coefficient is 0.09, and not
significant. However, on closer examination it can be seen t h a t there is
a cluster of points near the origin t h a t appear to be correlated, and b y
breaking the distribution down with respect to 0~ it turns out t h a t nearly
all these points are derived from situations in which the leading fly is
close to the axis of the chasing fly. There is a highly significant correlation
between tot and toe when 0e is close to =t=30~ and between these values,
but this correlation disappears when 0e is ~= 40 ~ or more. This is shown
in Fig. 6 c and d, which demonstrate the virtual absence of a relationship
between 0e and tot in this central region, and a strong correlation (0.65,
p ~0.001) between to1 and toe- The latter relation is possibly sigmoid,
with the response saturating at values of toe around 1500 ~ s-1, although
on the basis of the present data this can only be a guess. I n any event,
this relation is not badly fitted b y the regression line, which has a slope
of 0.7, and this value will be adopted.
This can be regarded as meaning t h a t when the leading fly is more or
less directly in front, its relative velocity rather than its position determine
the angular velocity, and hence the course of the chasing fly. This can
be expressed:
tot--~ k'r where k' ~--0.7, (2)
provided t h a t 0e lies between q-35 ~
The significance of this mechanism will be examined in more detail
in the discussion section, where an a t t e m p t is made to simulate a chase.
Response Time
Inspection of Fig. 5 indicates a delay (d) of 20-40 ms between the
appearance of the leading fly at a particular angular position 0e, and
the attainment of the appropriate angular velocity tot b y the following
fly. To examine this further the cross-correlation coefficients between
0r ~nd 0)1 were determined for a range of time differences between the
two measurements from - - 3 0 ms (to] leads 0e) to Jr 70 ms. These are
491
342 M.F. Land and T. S. Collett
~ 0
492
Control of Chasing in Flies 343
2 ]
m.s-
9 99
~f ~o~m~o'~ 9
"" .
9 9 o o~
0_1 1 I I I I
0 5~O00~ "1
LOf
Fig. 9. The relation between the forward velocity v t and angular velocity eot of the
chasing fly measured during each interval between points in Fig. 4. The angled
line is an arbitrary attempt to fit the data, and this line is used to adjust course
lengths in the simulations in Figs. 11-13
493
344 M.F. Land and T. S. Collett
Discussion
a) A n Attempt to Simulate a Chase
A valuable test of whether or not the behaviour of the chasing fly is
adequately described b y the relationships given in Eqs.(1) and (2) in
the results is to use these expressions to a t t e m p t to simulate a chase. If
the manoeuvres predicted are similar to those t h a t occur, and the overall
result is the same, then it can be concluded t h a t Eqs. (1) and (2) provide
a sufficient explanation.
Adding the two equations, and incorporating the delay (d) gives:
~o/(t+~) ~ b 0e(t) ~ b' o)~(t) (85~ > 0r > --36 ~ (3)
and this, together with the relation between forward speed (v]) and
angular velocity (w]) shown in Fig. 9, is used to reproduce the flight p a t h
of the chasing fly, given t h a t of the leading fly. I n the simulation~ shown
in Fig. 11 the p a t h of the leading fly is t a k e n directly from the chase
shown in Fig. 4, so t h a t direct comparison can be made.
The technique used is to start with the first three points of the chasing fly's
track as the given initial conditions (t = t, t ~ 2 0 and t-I-40 ms) and then, by
measuring 0e at t ~- 10 ms, to calculate the course change that must be made at
t ~ 40 ms, if the value of the delay (d) is to be 30 ms.
494
Control of Chasing in Flies 345
2 a,
b)
+1o6 +5,000- c)
:S-1
Oe Wf
-lod J
_1~0 o 1 O O !
+ 2,500 QO
0e
~.S-1
OJf
9 -5,000
-2,500
Fig. 10a---e. Instance of a chase in which the two flies execute quite different
manoeuvres, b u t in which the relation between co! and 0e is preserved, a) Record
of ~he chase, b) Plots of 0e and o 1. Marks on the time axis correspond to the num-
bered points on the tracks, and are 200 ms apart, e) Relation between w / a n d 0e
measured 20 ms earlier
Since k O v : r (dOf)/dt, the required change b0! is given b y kOe 9(~t, where
6t, the sampling interval, is in this case 20 ms. Similarly, where the value of 0e is
appropriate, an angle proportional to we can be included b y measuring the dif-
ference between the values of 0e a t t ~ 20 and t = 0, multiplying this b y /r and
adding this to course change already obtained. This is the required change because,
if eel= k' o~e, t h e n (dO/)/dt = k" (dOe)/dt , or (~Ol= k" 30 e. This now gives the direction
23 E. comp. Physiol., Vol. 89
495
346 M.F. Land and T. S. Collett
o2 the line between the points at t + 40, and t + 60 ms and this is then drawn with
a protractor. The length of this line will be vt- ~t and by assuming that (o] is equal
to the course change divided by ~t, the appropriate value of v] can be obtained
from Fig. 9. This then establishes the fourth point in the course (t + 60 ms). By
repeating these operations (which are simpler than they sound) for successive points,
the complete course can be built up. Different values of/c and/c' can be used, and
with slight modifications to the method as described the delay (d) and sampling
time (~t) can be altered. The relative speeds of the two flies can also be changed by
multiplying the values of vI obtained from Fig. 9 by a constant factor. In general
the smaller the sampling time (~t) the better, but for most purposes 20 ms was
found to be short enough not to introduce appreciable distortion.
I n the simulation shown in Fig. l l b , / ~ 2 0 s -1, the delay is 20 ms
and the velocity t e r m has not for the m o m e n t been included. I t can be
seen t h a t there is a reasonably good m a t c h to the actual flight p a t h
shown in Fig. 11 a. E a c h manoeuvre of the " r e a l " pursuing fly is dupli-
cated b y a corresponding manoeuvre of the simulated fly, even t h o u g h
these differ in detail. The principal difference between the real and
simulated fly is t h a t in the former the turns are all rather tighter, and as
a result the simulated pursuer appears to s t r a y rather further from the
leading fly's course t h a n it should. Nevertheless, i~ one compares the
relative positions of the flies at the m a r k e d instants 3, 4 and 5 it can be
seen t h a t the simulated fly is not m u c h further behind t h a n the real fly.
I n the final loop the leading fly does "escape," in b o t h cases, because
the distance separating the flies comes to exceed the visual limit of
10 era. I t is concluded from this t h a t the postulated " b a s i c " mechanism,
in which error angle is converted b y the fly to angular velocity, is able
to explain most features of the observed chase, although capable of
refinement.
Before considering the effects of adding in the error velocity (/c')
term, it is interesting to see w h a t the effects of changing/c and d are,
a n d i m p o r t a n t to show t h a t changing t h e m makes the simulation worse
in the sense t h a t it looks less like the actual course. Fig. 12 shows a
n u m b e r of simulations of the first two manoeuvres of the same course
as iu Figs. 4 and l l a . I n Fig. 12a it is seen t h a t reducing the delay to
zero makes the course intolerably good, so t h a t the trailing fly has caught
up with the leader several times after only 0.4 s, and there is virtually
no overshoot on the first bend. E q u a l l y dramatically, with a 40 ms delay
the fly overcorrects wildly on the first b e n d producing a figure of eight
course which wastes time and lets the leading fly get well in front. I n
Fig. 12b, halving the value of/c from 20 to 10 s -1 (20 ms delay) makes
the first b e n d m u c h too wide, while increasing /c b y a factor of 1.75
clearly makes the b e n d m u c h too tight and causes some overshoot. F r o m
these and similar results it is concluded t h a t realistic simulations of the
flight p a t h require t h a t b o t h the delay and the constant/c s t a y close to
the values a d o p t e d in deriving Fig. 11 b.
496
Control of Chasing in Flies 347
I a) 1!
c)
I
Fig. 11 a--c. Attempts to simulate the chase shown in Fig. 4. The course of the
leading fly is taken from Fig. 4, and the course of the chasing fly is plotted by the
method given in the text. a) Actual chase; b) Simulation using only the 0e, eoI
relationship: k = 2 0 s -1, d = 2 0 m s , e) As b) but incorporating the relationship
between me and co/: k = 2 0 s-1, k ' = 0 . 7 , d = 3 0 ms. The sampling time used in the
simulations was 20 ms in b) and 10 ms in c) although for clarity only every second
point is shown
23*
497
348 M.F. Land and T. S. Collett
k= 2 5 s -1
d = 20ms
Fig. 12. Effects of changing the parameters I: and d on the form of the simulation
of the first two bends in the course shown in l~ig. 4. All courses start in the same
way, and the cross line indicates where they would have reached when the track
of the leading f i t ends. I t can be seen that reasonable approximations to the actual
flight path are only obtained when I~ and d are close to values estimated in the
results
498
Control of Chasing in Flies 349
~ fse
k'2~ i
d=3Oms, k=2Os4,k =0"7 '
Fig. 13. The effect of including the error velocity term (Ic'eOe)on simulation of the
second bend in the chase shown in Fig. 4. The middle plot shows the path without
this term, and the lower left the improvement resulting from its inclusion. The
overshoot is reduced and the turn "anticipated" as in the actual track (upper right)
while the leading fly is still on the left, though moving right. A similar
situation occurs in the left turn after 4, and the right turn into the loop
after 5. This clearly cannot be simulated with only the position error
system, but as shown in Fig. 13 the inclusion of the angular velocity
term makes this possible b y opposing the effect of the position term while
the leading fly is approaching the chasing fly's line of flight, and aug-
menting it once the line of flight has been crossed. Thus an effect t h a t
almost looks like anticipation is explicable.
499
350 M.F. Land and T. S. Collett
v, g -~_ Oe
L Of
xfyf I
...... l . X
Fig. 14. Left: the geometrical relations between the leading (L) and following
fly (F) at an instant during a chase (represented here as two dimensional). Right:
diagram showing the double effect on 0e of changes in the angular velocity of the
following fly (col). The heavy line indicates the straightforward feedback loop in
which changes in course angle of the following fly (01) reduce the error angle (0e).
The thin lines indicate that eoI also affects the relative positions of the two flies at
time t, and hence the angle ~ of the line joining them (LF). In this diagram the
relation between coe and w! and the delay have been ignored
500
Control of Chasing in Flies 351
501
352 M.F. Land and T. S. Collett
c) PhysiologicalIn/erences
If male flies track each other in the way that has been described,
they must have neural mechanisms for converting certain specific kinds
of stimuli into appropriate output patterns. The response time of the
fly is very rapid, about 30 ms, which includes every process from seeing
the stimulus to accelerating to an appropriate velocity. Thus it seems
likely that number of neural operations involved will be small, and that
it is sensible to inquire how the systems might be organised with as few
neurones as possible. The two operations that need explanations are
(i) the conversion of stimulus position into angular velocity, and (ii) the
conversion, in the anterior regions of the eye, of stimulus velocity into
angular velocity. For the moment we shall assume that there exist
neurones passiag to the thorax whose discharge rate specifies the fly's
502
Control of Chasing in Flies 353
velocity ~ ~ position
system o~ system
Fig. 15. Possible scheme by which position and velocity information might be
extracted. The semicircles represent the two eyes, and it is assumed that each
contains an array of small-field movement detectors (represented by arrows). On
the right these feed onto interneurones (a) whose effect on the output neurone (b)
increases in the numbered order shown, from anterior to posterior. On the left the
movement detectors feed onto two units which collect selectively from the clock-
wise and anti-clockwise "on"-direction sets. These units in turn connect to the
outpu~ neurones. For further explanation see text
u l t i m a t e a n g u l a r velocity, a n d c o n c e n t r a t e on h o w t h e i n p u t s to t h e s e
n e u r o n e s m i g h t be a r r a n g e d .
A suggestion as t o h o w t h e first o p e r a t i o n m i g h t be p e r f o r m e d is
given i~ Fig. 15. This assumes t h a t t h e r e are neurones which d e t e c t
m o v e m e n t a n d collect f r o m r e s t r i c t e d regions of t h e eye (a). These feed
o n t o a collector n e u r o n e (b) which could be t h e n e u r o n e t o t h e wing
muscles. T h e (a) neurones m u s t h a v e t h e following properties.
(i) T h e y m u s t r e s p o n d o p t i m a l l y t o quite small objects. D u r i n g t h e
f i l m e d chases t h e average s e p a r a t i o n of t h e t w o flies was 3.7 cm, a n d
t h e r e was no sign of loss of visual c o n t a c t a t distances less t h a n 10 cm.
T h e r e l e v a n t stimulus angles on t h e r e t i n a for a 6 m m long f l y a r e 9.3 ~
a n d 3.4 ~, for t h e longest dimension of t h e stimulus.
(ii) T h e y m u s t be c a p a b l e of r e s p o n d i n g to objects m o v i n g a t a n g u l a r
velocities u p t o a t l e a s t 2 500 ~ s -1. This is a c o n s e r v a t i v e e s t i m a t e b a s e d
on t h e d i s t r i b u t i o n of values of e% (Fig. 6) which a c t u a l l y e x t e n d to
twice t h i s figure. I n t h e case of a 5 ~ o b j e c t passing o m m a t i d i a s p a c e d a t
2 ~ this r e p r e s e n t s a t o t a l "exposure t i m e " of a b o u t 2 . 8 m s p e r ore-
503
354 M.F. Land and T. S. Collett
504
Control of Chasing in Flies 355
units, like the (a) units postulated above, must also meet the condition
t h a t t h e y distinguish small target motion from background motion. I t
could simply be t h a t a small target ahead of the animal is a more powerful
stimulus to these cells t h a n the rest of the field moving in the opposite
(or the same) direction, but this is not yet clear. I t is also possible that
the velocity system (Fig. 15) is distributed over the whole eye, but t h a t
it is only detectable in front, where the position system has its smallest
effect.
Finally, the total response delay raises difficulties. Using available
data, we have McCann and Dill's (1969) estimate of 20 ms as the mini-
m u m latency for visual interneurones. Mulloney (1969) has shown in
Calliphora t h a t the total conduction time from optic lobe stimulation to
the appearance of an action potential in the "take-off jump muscle"
(tergo-trochanteral muscle) is only 3.6 ms. This is agreeably short and
includes one central synapse and a neuro-muscular junction. Heide
(1968) found t h a t the direct (non-fibrillar) flight muscles likely to be
responsible for yaw give a fused tetanus when stimulated at 60 Hz,
which suggests t h a t contraction and relaxation times cannot be much
less t h a n 17 ms (Calliphora). Acceleration time has already been discussed
and could be as short as 2 ms. E v e n using these optimistic estimates, the
total delay should be at least 43 ms, compared with the 30 ms found
here. Either visual latencies are rather shorter than existing measure-
merits suggest, or direct flight muscles operate more rapidly, or both.
d) Function o/Chasing
Male flies of m a n y species will chase any moving object of about the
right velocity and dimensions. Houseflies (Musca and Fannia) will chase
both males and females of their own or other species: for example it is
quite common to see Musca chasing CaUiphora twice their size. I t is
easy also to lure Musca off a wall, or Fannia from their patrolling flights
b y throwing small objects past them (e.g. peas). Provided the missile
does not pass too close, the animal's response is to fly towards it, and to
follow its trajectory for a while. This is thus pursuit and not evasion.
Greenbottles (probably Lucilia sp.) are especially spectacular. They will
sit on a particular leaf, and respond to small (1 cm) stones thrown just
above them b y darting up to ir~tercept, and then flying with the stone
for several metres before returning to the same leaf. One fly repeated
this performance 27 times in succession. Interestingly, a tendency to
chase thrown or flying objects turns out to be an excellent indicator of
sex. We have never seen a female muscid or calliphorid fly chase any-
thing.
Chasing thus seems to be confined to males and the stimuli t h a t
elicit it are not very specific. There are really only two plausible rune-
505
356 M.F. Land and T. S. Collett
We are very grateful to Alan King, Peter Slater and John )/Iaynard Smith
for critically reading the manuscript. This work was supported by a grant from ~he
Science Research Council of the U.K.
References
Autrum, H. : :Die Belichtungspotentiale und das Sehen der Insekten (Untersuchun-
gen an CaUiphora und Dixippus). Z. vergl. Physiol. 32, 176-227 (1950)
Bishop, L. G., Keehn, D. G., h[cCann, G. D. : Studies of motion detection by inter-
neurones of the optic lobes and brain of the flies Calliphora phaenicia and Musca
domestica. J. Neurophysiol. 81, 509-525 (1968)
Collett, T.: Visual neurones for tracking moving targets. Nature (Lond.) 2112,
127-130 (1971)
Collett, T. : Visual neurones in the anterior optic tract of the privet hawk moth.
J. eomp. Physiol. 78, 396-433 (1972)
Colyer, C. N., Hammond, C. O. : Plies of the British Isles. London: Warne 1951
Faust, R.: Untersuehungen zum Halterenproblem. Zool. Jb., Allg. Zool. Physiol.
63, 325-366 (1952)
Fermi, G., Reiehardt, W. : Optomotorische Reaktionen der Fliege Musca domestica.
Kybernetik 2, 15-28 (1963)
Heide, G.: Flugsteuerung durch nicht-fibrill~ire Flugmuskeln bei der Schmeil3fliege
Calliphora. Z. vergl. Physiol. 59, 456-460 (1968)
506
Control of Chasing in Flies 357
507
508
Vol 454 | 21 August 2008 | doi:10.1038/nature07183
LETTERS
Cryptochrome mediates light-dependent
magnetosensitivity in Drosophila
Robert J. Gegear1, Amy Casselman1, Scott Waddell1 & Steven M. Reppert1
Although many animals use the Earths magnetic field for orienta- Although there is good behavioural evidence for the involvement
tion and navigation1,2, the precise biophysical mechanisms under- of short-wavelength photoreceptors in the detection of a geomag-
lying magnetic sensing have been elusive. One theoretical model netic field5,1618, an essential link between Cry and magnetoreception
proposes that geomagnetic fields are perceived by chemical reac- has not been established in any animal. Drosophila are ideally suited
tions involving specialized photoreceptors3. However, the specific to investigate a role for Cry as a magnetoreceptor, because they only
photoreceptor involved in such magnetoreception has not been have the light-sensitive Cry14 in which the action spectrum peaks in
demonstrated conclusively in any animal. Here we show that the the ultraviolet-A range (350400 nm) with a plateau in the near blue
ultraviolet-A/blue-light photoreceptor cryptochrome (Cry) is range (430450 nm)19,20. Notably, flies that lack Cry (cry0)21 or har-
necessary for light-dependent magnetosensitive responses in bour the chemically induced missense cryb mutation22,23 can be used
Drosophila melanogaster. In a binary-choice behavioural assay to evaluate the role of Cry in magnetosensitive responses.
for magnetosensitivity, wild-type flies show significant naive and We initiated our studies by developing a behavioural assay for
trained responses to a magnetic field under full-spectrum light magnetosensitivity in Drosophila (Fig. 1a). In this illuminated appar-
( 300700 nm) but do not respond to the field when wavelengths atus, flies experience a magnetic field generated by an electric coil
in the Cry-sensitive, ultraviolet-A/blue-light part of the spectrum system and display their magnetosensitivity in a binary-choice
(,420 nm) are blocked. Notably, Cry-deficient cry0 and cryb flies T-maze. The two-coil system is ideal for behavioural studies of mag-
do not show either naive or trained responses to a magnetic field netosensitivity, because it produces a magnetic field on one side of
under full-spectrum light. Moreover, Cry-dependent magnetosen- the T-maze, while producing no field on the opposite side. This
sitivity does not require a functioning circadian clock. Our work design eliminates non-magnetic differences such as heat generated
provides, to our knowledge, the first genetic evidence for a Cry- by the electric coils between sides during test sessions24. Flies were
based magnetosensitive system in any animal. tested either for their response to the magnetic field in the naive state
The ability of an animal to detect geomagnetic fields has substan- (naive group) or after a training session pairing the field with sucrose
tial biological relevance as it is used by many invertebrate and verte- reward (trained group).
brate species for orientation and navigation purposes, including Wild-type Canton-S, white-eyed w;Canton-S, Oregon-R and
homing, building activity and long-distance migration2,4. Three gen- Berlin-K strains all developed a learned preference for a magnetic
eral modes of magnetoreception have been proposed5. One mode is field (Fig. 1b). The trained groups in the two Canton-S lines showed
electromagnetic induction by the Earths magnetic field, which may the greatest response to the field (P 5 0.002, one-way analysis of
occur in electrosensitive marine fish, although there is little evidence variance (ANOVA)) and were the only ones to show a naive avoid-
to support such sensing. The two other modes, for which experi- ance of the field (P , 0.0001, one-sample t-test). Thus, Drosophila
mental evidence does exist, are a magnetite-based process68 and consistently show magnetosensitivity that varies in magnitude in a
chemical-based reactions9,10 that are modulated by magnetic fields. strain-dependent manner. The similarity of behavioural responses
One chemical model of magnetoreception proposes that magnetic between red-eyed, wild-type Canton-S flies and white-eyed
information is transmitted to the nervous system through the light- w;Canton-S flies shows that eye colour does not substantially alter
induced product of magnetically sensitive radical-pair reactions in behavioural responses to the magnetic field.
specialized photoreceptors3. Because wild-type Canton-S flies showed the most robust trained
Cry proteins are flavoproteins that have been postulated to gen- and naive responses of the strains tested, we used them to determine
erate magnetosensitive radical pairs that could provide a photoin- whether the magnetic responses we observed were light-dependent.
duced electron transfer reaction for the detection of magnetic fields3. We assayed naive and trained Canton-S flies under different long-
Cry proteins are best known for their roles in the regulation of cir- wavelength pass filters that transmitted wavelengths of light
cadian clocks11,12 and can be categorized into two groups on the basis at .500 nm, .420 nm or .400 nm (Fig. 2a). In contrast to flies
of current phylogenetic and functional relationships13,14. Drosophila- assayed under full-spectrum light (Fig. 1b and Fig. 2a), flies did not
like Cry proteins are sensitive to light in the ultraviolet-A/blue range15 show either naive or trained responses to the field when wave-
and function primarily as photoreceptors that synchronize (entrain) lengths ,420 nm were blocked (Fig. 2b). Because the filter that
circadian clocks. Vertebrate-like Cry proteins, which have also been blocked light ,420 nm also caused a 13% decrease in total irradiance
found in every non-drosophilid insect so far examined14, do not seem (Fig. 2c, red line), we examined whether the filter-induced lack of
to be directly light-sensitive. Instead, vertebrate-like Cry proteins are behavioural responses to the magnetic field was secondary to the
potent repressors of the Clock and Bmal1 (known as Cycle in insects) decrement in irradiance. When Canton-S flies were studied under
transcription factors which, as heterodimers, drive the intracellular full-spectrum light, with a total irradiance level lower than that
transcriptional feedback loop of the circadian clock mechanism in all imposed by the filter (Fig. 2c, blue line), the flies still showed signifi-
animals studied. cant naive (P 5 0.0005, one-sample t-test) and trained responses to
1
Department of Neurobiology, University of Massachusetts Medical School, Worcester, Massachusetts 01605, USA.
1014
509
2008 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 454 | 21 August 2008 LETTERS
the magnetic field (Fig. 2d). Thus, the filter-induced loss of beha-
vioural responses to the magnetic field is due to the loss of short-
wavelength light. a
12 Full spectrum 12 >500
Behavioural responses to the magnet were partially restored when
400420 nm light was included (Fig. 2b), which is consistent with the
8 8
action spectrum of Drosophila Cry tailing into the near blue19, and, as
4 4
a
Train Light source 0 0
300 400 500 600 700 300 400 500 600 700
Wavelength (nm)
b ****
Black box Filter
0.2
Coil
0.1 *
Preference index
12
11 11 9
0
11 11 9
10
0.1
Elevator T-port Training tube
Test
0.2
Full spectrum >500 nm >420 nm >400 nm
c d
Photons s1 cm2 nm1 (1012)
12 0.2 ****
Preference index
0.1
8 11
0
11
4
0.1
b **** 0 0.2
0.2 300 400 500 600 700
Full spectrum
**** e Wavelength (nm) (low intensity)
0.1 *
Photons s1 cm2 nm1 (1011)
Preference index
12
** 20
10 11
0 11
11 11 15
10 9
0.1 10
5
0.2
Canton-S w;Canton-S Oregon-R Berlin-K 0
300 340 380 420
Figure 1 | Behavioural apparatus for magnetosensitivity and behavioural
Wavelength (nm)
responses in different Drosophila strains. a, Behavioural apparatus for
magnetosensitivity. The top diagram (Train) shows the frontal view of the Figure 2 | Short-wavelength light is required for magnetosensitivity in
choice chamber apparatus positioned for training. The chamber apparatus Canton-S flies. a, Irradiance curves for different light conditions. Light
consisted of a training tube, an elevator to transfer flies, and a duel-choice measurements were taken inside the training and test tube. Dashed lines
point (T-port). For training, the apparatus, with training tube only, was denote cutoff points of the blocking filters. b, Wavelength-dependence of the
placed upright in an illuminated black box containing a two-coil system. A magnetic response. Bars show the preference index of the naive (white) or
population of flies (dots) was loaded into the training tube with or without trained (black) groups. Full-spectrum data are from Fig. 1b. Numbers
sucrose reinforcement and a magnetic field. The bottom diagram (Test) represent the groups tested. *, P , 0.05; ****, P , 0.0001. c, Irradiance
shows the frontal view of the choice chamber apparatus positioned for curves depicting full-spectrum light (black line), light .420 nm (red line)
testing. For testing, the apparatus, with tubes attached to the T-port (T- and full-spectrum light with reduced total irradiance (full spectrum, low
maze), was rotated to the horizontal and flies were transferred from the intensity; blue line). d, Canton-S flies still elicited significant responses to the
elevator section to the T-port. Wavelength dependence was examined using magnetic field under full-spectrum, low intensity light. ****, P , 0.0001.
long-wavelength pass filters. b, Drosophila strains vary in their behavioural e, Irradiance values from 300420 nm. Data are expanded scale from full-
response to a magnetic field under full-spectrum light (Fig. 2a). Bars show spectrum pattern in a. The irradiance values in ultraviolet-A/blue light in
the preference index of the naive (white) or trained (black) groups. Numbers our studies (300420 nm) are in line with those reported for Drosophila Cry
represent the groups tested and values are mean 6 s.e.m. *, P , 0.05; **, function using other biological responses19,20; that is, a range of 1011 to 1012
P , 0.01; ****, P , 0.0001. photons s21 cm22 nm21. Values from b and d are mean 6 s.e.m.
1015
510
2008 Macmillan Publishers Limited. All rights reserved
LETTERS NATURE | Vol 454 | 21 August 2008
data are consistent with the hypothesis that Cry can function as a magnetic field after at least 5 days in constant light when the flies
magnetoreceptor in Drosophila. were shown to express arrhythmic locomotor behaviour (Fig. 4a), to
We next used Cry-deficient cry0 mutant flies to examine directly have disrupted Period abundance rhythms (Fig. 4b), and to express
whether Cry is required for magnetosensitive behaviour. We tested constantly low levels of Cry (Fig. 4c). Notably, these arrhythmic flies
two of the newly generated cry0 fly lines, because, in cry0 flies, the continued to show significant naive (P 5 0.004, one-sample t-test)
entire cry coding sequence has been replaced with mini-white1 by
homologous recombination, ensuring that, unlike in the more com- a
0.20
monly used point-mutant cryb flies, there is no possibility of residual
Cry activity21. In addition, the three cry0 fly lines (from cry01 to cry03) ****
0.15
were backcrossed independently into a w1118 background21. Thus, we
Preference index
were able to use the appropriate w1118 control flies to test the contri- 0.10
bution of the cry gene in magnetosensitive behaviour.
15
Control w1118 flies showed a clear naive preference for, rather than 0.05
avoidance of, the magnetic field (Fig. 3a). The difference in the dir-
ection of the naive response to the magnetic field between Canton-S 12
0
12
flies and the w1118 line re-emphasizes the importance of controlling
for genetic background in studies of magnetosensitivity in flies. 0.05 w1118 w1118 w1118
Nonetheless, like Canton-S flies, the naive response of w1118 flies to Full spectrum >500 nm >420 nm
the magnetic field was light dependent; the naive preference for the b
magnetic field was abolished in the absence of ultraviolet-A/blue light 0.20
(,420 nm; Fig. 3a). **
0.15 ***
Homozygous cry02 flies lacking Cry did not show a naive response
Preference index
to the magnet under full-spectrum light, in contrast to the significant
0.10
naive responses manifested by both w1118 and heterozygous cry02/1
10
flies (Fig. 3b). Training control w1118 flies to prefer the magnetic field 0.05
11
under full-spectrum light significantly enhanced their naive pref-
erence for the field (Fig. 3c). In contrast, homozygous cry01 flies 0
10
did not show either a naive preference for the field (like cry02 flies)
or an enhanced preference for the field after training (Fig. 3c). The 0.05 w1118 cry02 cry02/+
loss of the response to the magnetic field in the Cry-deficient flies
c
resembled the behaviour when w1118 flies were deprived of ultra- 0.20 *
violet-A/blue light (Fig. 3a), which is consistent with Cry being the
relevant light sensor. These data using two cry null strains strongly 0.15
Preference index
suggest that both naive and trained responses to the magnetic field in
Drosophila require Cry function. 0.10 10
The Cry-defective cryb mutant flies are also unable to respond to
0.05 10
the magnetic field; the cryb mutation renders CryB essentially non-
functional22,23. Because the genetic background of cryb mutant flies is 0
10
not well defined, we compared behavioural responses to the magnetic 10
field between homozygous cryb flies and heterozygous cryb/Canton-S 0.05
flies. Whereas homozygous cryb flies did not show either naive or w1118 cry01
d 0.20
trained responses to the magnetic field under full-spectrum light,
heterozygous cryb/Canton-S flies showed significant naive
0.15
(P 5 0.0004, one-sample t-test) and trained responses (Fig. 3d); the
trained response in the heterozygotes was less than that of wild-type ****
0.10
Canton-S flies (Fig. 1b) and probably results from differences in
Preference index
a c
LD Cry
2.0
Day
1.0
5 0.5
LL 0
LD LL LD LL cry01
C-S
1
d
Day
0.2
****
5
0.1
Preference index
10
b 0
2.0 Per 10
Relative protein levels
0.1
1.5
1.0 0.2
0.5
LD
LL
0 LD
LL
1 5 9 13 17 21
Zeitgeber time/circadian time (h)
Figure 4 | Constant light disrupts circadian function but not Cry-mediated light-dark cycle (P , 0.01, one-way ANOVA), but not in constant light.
magnetosensitivity in Canton-S flies. a, Mean activity records in 12 h Head extracts were analysed by western blot30 and normalized against
light:12 h dark lighting cycle (LD; top) or constant light (LL; bottom) in a-tubulin. Values are mean 6 s.e.m. from three sets of heads. c, Cry
double-plotted format (n 5 62 for each group). The lighting conditions were abundance is decreased in LL. Values are mean 6 s.e.m. from three sets of
identical to those used for housing flies tested for responses to magnet; light heads collected over 24 h in the light-dark cycle or in constant light. Right,
irradiance, 1.5 3 1015 photons s21 cm22 nm21. For the light-dark cycle, 94% western blot probed for Cry30 showing presence (arrow) in the light-dark
expressed circadian rhythms when released in constant darkness (period, cycle or in constant light in Canton-S (C-S) heads, and absence in cry01
24.6 6 0.03 h). All the constant-light flies were arrhythmic. b, Temporal heads. d, Flies in constant light elicit behavioural responses to the magnetic
profiles of Period (Per) in heads. Protein abundance was rhythmic in the field. Values are mean 6 s.e.m. ****, P , 0.0001.
and trained responses to the magnetic field (Fig. 4d). Thus, the con- Our behavioural assay for magnetosensitivity does not at present
tinuous activation of Cry by light does not disrupt its ability to sense have a pure directional component, and therefore it is difficult to
the magnet, and an intact circadian system is not required for the relate our findings directly to the use of geomagnetic fields for animal
magnetoreception mechanism to operate. orientation and navigation. Nevertheless, it is probable that the res-
There are two other published reports of magnetosensitivity in ponse we have identified is the prototype for the involvement of Cry
adult Drosophila26,27. One describes behavioural evidence that male in chemical-based magnetic sensing. Thus, our findings open new
wild-type Oregon-R flies show a light-dependent magnetic compass avenues of investigation into the cellular and molecular basis of
response in a radial maze whereas female flies did not respond to the chemical-based magnetic sensing in animals. The powerful genetics
magnet27. Additionally, male flies responded in opposite directions of Drosophila will facilitate an understanding of the precise mech-
when tested under either 365 nm or 500 nm light. In our studies, both anism of action of Cry in magnetosensitivity, such as the actual
male and female flies showed a magnetic response. Regardless of involvement of magnetosensitive radical pairs produced by photo-
experimental differences, both the previous study27 and ours dem- induced electron transfer reactions28. Our data further show that the
onstrate that fruitflies can respond to a magnetic field in a wave- biological functions of Drosophila Cry extend beyond those in cir-
length-dependent manner. cadian clocks.
Our results extend substantially the presence of a light-dependent
magnetic sense in Drosophila by showing the necessity of Cry. We METHODS SUMMARY
cannot distinguish unequivocally whether fly Cry functions as the Fly stocks were raised on standard cornmeal/agar medium at 25 uC and 60%
relative humidity under a 12 h light:12 h dark lighting cycle. The w1118;;cry0 flies
actual magnetoreceptor or whether it is an essential component (from cry01 to cry03) were a gift from J. C. Hall and are described in ref. 21. The
downstream of the receptor. Cry is necessary for both the naive w1118 stock used in our experiments was the same stock used to create w1118;;cry0
and trained responses to the magnetic field, consistent with the flies21. The cryb line was a gift from P. Emery22. Our choice apparatus was based
notion that Cry is in the input pathway of magnetic sensing. In on the olfactory conditioning apparatus as described 29. Our two-coil system was
addition, the continued behavioural responses to the magnet in con- based on the double-wrapped coil system described previously24. We adjusted
stant light, in which the known Cry signalling components are being the current flowing through the coils so that the magnetic field intensity was no
constantly degraded and the circadian clock is rendered non-func- more than 5 G in any area along the tube. Coils were positioned 45u to the
tional, is also consistent with an input function. The most compelling horizontal for experiments involving Canton-S, w;Canton-S, Berlin-K and
Oregon-R flies. Coils were positioned parallel to the horizontal for all other
evidence supporting a magnetoreceptor role for Cry is that the Cry- experiments, because it produced a more robust response and eliminated a polar
dependent behavioural responses to the magnetic field require ultra- gradient; that is, there was no horizontal magnetic gradient, as the field was
violet-A/blue light, which matches the action spectrum of Drosophila perpendicular to the T-port tubes. To assess the magnetoresponse of flies, we
Cry19,20. used a simple choice paradigm. Flies were placed in a glass vial containing
1017
512
2008 Macmillan Publishers Limited. All rights reserved
LETTERS NATURE | Vol 454 | 21 August 2008
moistened Whatman paper and starved for 22 h before training. All experiments 17. Ritz, T., Dommer, D. H. & Phillips, J. B. Shedding light on vertebrate
were performed between 8:00 and 12:00 EST. For each population of flies tested magnetoreception. Neuron 34, 503506 (2002).
(100150 flies per group), we calculated a preference index value on the basis of 18. Wiltschko, R. & Wiltschko, W. Magnetoreception. Bioessays 28, 157168 (2006).
the equation: (PM 2 0.5)/[(PM 1 0.5) 2 (2PM 3 0.5)], in which PM is the pro- 19. VanVickle-Chavez, S. J. & Van Gelder, R. N. Action spectrum of Drosophila
cryptochrome. J. Biol. Chem. 282, 1056110566 (2007).
portion of flies on the magnetic field side of the T-port. To test whether flies
20. Helfrich-Forster, C. et al. The extraretinal eyelet of Drosophila: Development,
responded to the experimental magnetic field, we used either a Students t-test to
ultrastructure, and putative circadian function. J. Neurosci. 22, 92559266
compare preference index values between trained and naive groups or a one- (2002).
sample t-test to compare preference index values to zero (that is, preference 21. Dolezelova, E., Dolezel, D. & Hall, J. C. Rhythm defects caused by newly
index values expected with no response to the magnetic field). engineered null mutations in Drosophilas cryptochrome gene. Genetics 177,
329345 (2007).
Full Methods and any associated references are available in the online version of 22. Emery, P. et al. CRY, a Drosophila clock and light-regulated cryptochrome, is a
the paper at www.nature.com/nature. major contributor to circadian rhythm resetting and photosensitivity. Cell 95,
669679 (1998).
Received 25 February; accepted 19 June 2008.
23. Stanewsky, R. et al. The cryb mutation identifies cryptochrome as a circadian
Published online 20 July 2008.
photoreceptor in Drosophila. Cell 95, 681692 (1998).
1. Lohmann, K. J., Lohmann, C. M. F. & Putman, N. F. Magnetic maps in animals: 24. Kirschvink, J. L. Uniform magnetic-fields and double-wrapped coil systems
natures GPS. J. Exp. Biol. 210, 36973705 (2007). improved techniques for the design of bioelectromagnetic experiments.
2. Wiltschko, W. & Wiltschko, R. Magnetic orientation and magnetoreception in Bioelectromagnetics 13, 401411 (1992).
birds and other animals. J. Comp. Physiol. A Neuroethol. Sens. Neural Behav. Physiol. 25. Stanewsky, R. Genetic analysis of the circadian system in Drosophila melanogaster
191, 675693 (2005). and mammals. J. Neurobiol. 54, 111147 (2003).
3. Ritz, T., Adem, S. & Schulten, K. A model for photoreceptor-based 26. Wehner, R. & Labhart, T. Perception of geomagnetic field in fly Drosophila
magnetoreception in birds. Biophys. J. 78, 707718 (2000). melanogaster. Experientia 26, 967968 (1970).
4. Luschi, P. et al. Marine turtles use geomagnetic cues during open-sea homing. 27. Phillips, J. B. & Sayeed, O. Wavelength-dependent effects of light on magnetic
Curr. Biol. 17, 126133 (2007). compass orientation in Drosophila melanogaster. J. Comp. Physiol. A Sens. Neural
5. Johnsen, S. & Lohmann, K. J. The physics and neurobiology of magnetoreception. Behav. Physiol. 172, 303308 (1993).
Nature Rev. Neurosci. 6, 703712 (2005). 28. Maeda, K. et al. Chemical compass model of avian magnetoreception. Nature 453,
6. Kirschvink, J. L. & Gould, J. L. Biogenic magnetite as a basis for magnetic-field 387390 (2008).
detection in animals. Biosystems 13, 181201 (1981). 29. Tully, T. & Quinn, W. G. Classical-conditioning and retention in normal and
7. Walker, M. M. A model for encoding of magnetic field intensity by magnetite- mutant Drosophila melanogaster. J. Comp. Physiol. A Sens. Neural Behav. Physiol. 157,
based magnetoreceptor cells. J. Theor. Biol. 250, 8591 (2008). 263277 (1985).
8. Kirschvink, J. L., Walker, M. M. & Diebel, C. E. Magnetite-based 30. Rush, B. L., Murad, A., Emery, P. & Giebultowicz, J. M. Ectopic CRYPTOCHROME
magnetoreception. Curr. Opin. Neurobiol. 11, 462467 (2001). renders TIM light sensitive in the Drosophila ovary. J. Biol. Rhythms 21, 272278
9. Leask, M. J. M. Physicochemical mechanism for magnetic-field detection by (2006).
migratory birds and homing pigeons. Nature 267, 144145 (1977).
10. Schulten, K., Swenberg, C. E. & Weller, A. Biomagnetic sensory mechanism based Supplementary Information is linked to the online version of the paper at
on magnetic-field modulated coherent electron-spin motion. Z. Phy. Chem. www.nature.com/nature.
(Frankfurt) 111, 15 (1978).
Acknowledgements We thank H. Zhu for the protein work in Fig. 4b, c; L. Foley for
11. Cashmore, A. R. Cryptochromes: Enabling plants and animals to determine
assistance; J. C. Hall for the cry0 flies; P. Emery for the Per and Cry antibodies; and
circadian time. Cell 114, 537543 (2003).
P. Emery, P. Perrat, B. Leung, S. DasGupta, M. Krashes and H. Zhu for discussions.
12. Partch, C. L. & Sancar, A. Photochemistry and photobiology of cryptochrome
This work was supported by grants from the NIH.
blue-light photopigments: The search for a photocycle. Photochem. Photobiol. 81,
12911304 (2005). Author Contributions S.M.R. and R.J.G. conceived the idea of using Drosophila to
13. Zhu, H. et al. The two CRYs of the butterfly. Curr. Biol. 15, R953R954 (2005). study magnetosensitivity; S.W. and R.J.G. conceived the idea of using appetitive
14. Yuan, Q., Metterville, D., Briscoe, A. D. & Reppert, S. M. Insect cryptochromes: conditioning to study magnetoresponses; R.J.G. designed the experimental
Gene duplication and loss define diverse ways to construct insect circadian apparatus; R.J.G., S.W., A.C. and S.M.R. designed the experiments and analysed the
clocks. Mol. Biol. Evol. 24, 948955 (2007). data; R.J.G. performed the experiments with help from A.C.; R.J.G., S.M.R., S.W. and
15. Ozturk, N., Song, S. H., Selby, C. P. & Sancar, A. Animal type 1 cryptochromes A.C wrote the paper.
analysis of the redox state of the flavin cofactor by site-directed mutagenesis. J.
Biol. Chem. 283, 32563263 (2008). Author Information Reprints and permissions information is available at
16. Phillips, J. B. & Borland, S. C. Behavioral evidence for use of a light-dependent www.nature.com/reprints. Correspondence and requests for materials should be
magnetoreception mechanism by a vertebrate. Nature 359, 142144 (1992). addressed to S.M.R. (steven.reppert@umassmed.edu).
1018
513
2008 Macmillan Publishers Limited. All rights reserved
doi:10.1038/nature07183
METHODS
Fly strains. Oregon-R, Berlin-K and w1118 stocks were provided by the
Bloomington Drosophila Stock Center (product number 6326, 4269 and 8522,
respectively).
Behavioural apparatus. The main chamber consisted of a training tube, a centre
elevator section to transfer flies, and a two-tube choice point or T-port for testing
the relative response of flies to a magnetic field (Fig. 1a). The training and T-port
tubes were round-bottom polystyrene tubes and could be removed from the
main section of the apparatus. The magnetic stimulus was delivered to the
training and T-port tubes by placing the choice chamber apparatus inside an
opaque housing box that contained the magnetic coil system. The box was
constructed such that the main chamber could be placed between the two coils
in either an upright position (as shown in Fig. 1a, top panel) or a horizontal
position (as shown in Fig. 1a, bottom panel). The upright position of the cham-
ber apparatus was used for training so that the tube could be placed in the centre
of one coil, and the horizontal position of the apparatus was used to suspend the
two tubes of the T-maze in the same area of the coil during test sessions. In this
way, flies were subjected to the same light conditions and intensity of magnetic
field during training and test sessions.
Each of the two coils was wrapped with two wires; the wires were wrapped in
one direction on one coil and in the opposite direction on the other. Current flow
through the coils produced a magnetic field in one coil (parallel current flow) but
not in the other (opposite current flow). A double-pole, double-throw switch
reversed the current flow in one wire loop but not the other, which allowed us to
move the magnetic field easily from the right to the left side of the T-maze. A DC
power supply with current and voltage controls was used so that we could change
the intensity of the magnetic field produced by the coils. Each coil was mounted
on a plastic track so that it could be positioned directly under the tubes (field
perpendicular to the tubes), at the end of the tubes at a 45u angle (as shown in
Fig. 1a), or at the end of the tubes (field parallel to the tubes). We used a magnetic
field intensity of 5 G for our experiments, because it gave the most consistent
naive response; decreasing to 1 G increased variability to the point that responses
were no longer significant.
The housing box for the test or choice chamber was open on the top so that the
chamber, regardless of position, could be illuminated by one ZooMed Reptisun
10.0 UVB fluorescent tube (F20T12) and one Agrobrite full spectrum fluorescent
grow tube. Wavelengths entering the box were restricted by covering the top of
the box with a long-wavelength filter that transmitted wavelengths of
light .500 nm (Edmund Optics) or .420 nm or .400 nm (E400 and E420 from
Gentex). Irradiance measurements (from an Ocean Optics USB 2000 fibre optic
spectrometer) were taken from inside one arm of the T-maze portion of the
choice chamber apparatus; thus, lighting conditions represent those experienced
by flies while being either trained or assayed for sensitivity to a magnetic field.
Experimental procedure. For the training group, a population of 100150 flies
was loaded into the elevator section of the choice apparatus with an empty
training tube facing one of the coils (Fig. 1a, top panel). Flies were transferred
to the training tube for 2 min and then transferred back to the elevator and held
for a 1-min rest period. The empty training tube was next replaced with a tube
containing sucrose reinforcement and flies were allowed to feed for 2 min in the
presence of a magnetic field. Flies were then transferred back to the elevator and
held for 1 min while the coil system was turned off. During this time, the training
tube was also removed, and two empty tubes were added to form the two arms of
the T-port. The choice chamber was then positioned horizontally in the box
(Fig. 1a, bottom panel). The coil system was turned on, and flies were transferred
to the T-port, in which they were allowed to choose between the sides with or
without a magnetic field. After 2 min, the two arms of the T-port were blocked
and flies from each side were collected into separate empty vials and counted.
For the naive group, a second population of 100150 flies was immediately
loaded into the elevator section of the horizontally placed choice chamber and
the coil system was turned on. After 1 min, flies were transferred directly to the
T-port for 2 min.
Trained and naive groups were tested consecutively and with the magnetic
field on the same side. This was done to control for the possibility that the choice
behaviour of flies reflected a preference for one arm of the T-port and not a
response to the magnetic field. As an extra control for side preferences inde-
pendent of magnetic stimuli, we alternated the side of the T-port containing the
field after each consecutive set of trained and naive flies (that is, trained and naive
with magnet on the left side and then trained and naive with magnet on the
right).
514
2008 Macmillan Publishers Limited. All rights reserved
Neuron
Review
Motion vision provides essential cues for navigation and course control as well as for mate, prey, or predator
detection. Consequently, neurons responding to visual motion in a direction-selective way are found in
almost all species that see. However, directional information is not explicitly encoded at the level of a single
photoreceptor. Rather, it has to be computed from the spatio-temporal excitation level of at least two photo-
receptors. How this computation is done and how this computation is implemented in terms of neural
circuitry and membrane biophysics have remained the focus of intense research over many decades.
Here, we review recent progress made in this area with an emphasis on insects and the vertebrate retina.
Review
A B
x x
t t
C D E
HP HP
+
~dI/dt
.. ~dI/dx
AND NOT
Motion-Detector Models for biological motion detectors actually do not calculate the
Several models have been proposed in the past that calculate spatial and the temporal gradient of the moving image. They
the direction of motion from the brightness changes as captured rather correlate the brightness values measured at two adjacent
by the photoreceptors. image points with each other after one of them has been filtered
The gradient model (Figure 1C) describes the most straightfor- in time (correlation model, Figure 1D). Consequently, their output
ward way to implement a motion detector with the above is not proportional to image motion but rather deviates from it in
mentioned mathematical relationship. Here, the spatial gradient a characteristic way. In fact, this deviation has been the crucial
dI/dx is approximated by the brightness difference dI, of the hint for researchers in motion vision to propose exactly this
pattern, I, sampled at two neighboring image points separated type of model. The first correlation detector was proposed on
by a distance, dx. Both input signals become high-pass filtered, the basis of experimental studies on the optomotor behavior of
approximating the temporal derivative, and then added together. insects (Hassenstein and Reichardt, 1956; Reichardt, 1961;
These two quantities are then divided by each other yielding an 1987). This correlation detector is commonly referred to as the
estimate of the local image velocity (Srinivasan, 1990). This esti- Reichardt detector (van Santen and Sperling, 1985), and has
mate will only depend on the image velocity and not on the spatial also been applied to explain motion detection in different verte-
structure of the moving pattern because the local image contrast brate species including man (for review, see Borst and Egelhaaf,
is expressed in a steeper spatial, as well as in a steeper temporal 1989). Such a detector consists of two mirror-symmetrical
gradient: Dividing them leads to a cancellation of image contrast. subunits. In each subunit, the signals derived from two neigh-
However, as attractive as the gradient model of motion detec- boring inputs are multiplied with each other after one of them
tion might appear, most models that were proposed to account has been shifted in time by a temporal low-pass filter. The final
Review
detector response is given by the difference of the output sig- ardt detector will consist of two parts: a constant DC shift that is
nals. Various elaborations of the basic Reichardt model have DS and, superimposed, a periodic modulation with the local
been proposed to accommodate this motion detection scheme brightness of the pattern. Only when the summed output of an
to perform in a species-specific way. array of Reichardt detectors is considered, these local modula-
Perhaps the simplest correlation-type movement detector has tions will disappear since they are phase-shifted with respect
been proposed by Barlow and Levick to explain their experi- to each other. This also holds true for the Barlow-Levick model.
mental findings on DS ganglion cells in the rabbit retina (Barlow
and Levick, 1965). The Barlow-Levick model (Figure 1E) is DS Cells
almost identical with respect to its layout but with only one Neurons responding differently to visual stimuli moving in oppo-
subunit of the basic Reichardt model. It consists of two input site directions are called DS. Such neurons have long been
lines carrying the brightness signals which are compared after known to exist in the visual system and other parts of the verte-
one of the signals has been delayed. In contrast to the Reichardt brate and invertebrate nervous system.
model, this comparison is accomplished by a specific logical
operation, an AND-NOT or veto gate, suppressing the detectors Insects
activity when the delay line is activated first and, consequently, In invertebrates, the first DS neurons were found in flies, located
both signals arrive simultaneously at the AND-NOT gate. The in a brain structure called the lobula plate. The lobula plate is the
corresponding direction of motion, i.e., from left to right, is, third of a stack of neuropiles of the flys optic lobe, each forming
therefore, the detectors null direction. For motion in the detec- a retinotopic representation of the image as initially formed by
tors preferred direction the veto signal arrives too late to have the compound eye. Starting from the periphery, these are called
an effect. lamina, medulla, and lobula complex, the latter being divided into
Another model which is often applied to human psycho- an anterior lobula and a posterior lobula plate (Figure 2A). As
physics and motion-sensitive neurons in the mammalian cortex a consequence of the retinotopic structure, each neuropile is
is the so-called motion energy model (Adelson and Bergen, built from repetitive columns containing an identical set of
1985). Interestingly, if the Reichardt model is equipped with the neurons first described anatomically by Ramon y Cajal on the
same spatial and temporal filters in its input channels, it assumes basis of Golgi staining (Cajal and Sanchez, 1915). For the fruit
the same specific functional characteristics as the energy model fly Drosophila melanogaster, a large set of columnar neurons
and even is mathematically equivalent (van Santen and Sperling, has been cataloged (Fischbach and Dittrich, 1989). More re-
1985; Adelson and Bergen, 1985). This identity, however, only cently, this set has been complemented by assigning transmitter
holds for the final, fully opponent output signal of both detectors systems to various columnar neurons (e.g., Morante and Des-
and does not pertain to its internal structure. plan, 2008; Raghu and Borst, 2011; Raghu et al., 2011). Each
Despite many differences in detail, all models of motion detec- columnar neuron, whether located in the lamina, medulla, or lob-
tion share the following commonalities: (1) they all have at least ula complex, has distinct arborizations in particular layers of its
two spatially separated input lines that read the brightness levels neuropile and some neurons connecting the lamina with the
of adjacent pixels in the image, (2) they all have some sort medulla or the medulla to the lobula plate. Furthermore, all these
of asymmetry with respect to the temporal filtering of the input cells restrict their arborizations to a small part of their respective
(a temporal derivative in case of the gradient detector, a low- neuropile, mostly respecting the columnar borders. This is
pass filter in one of the input channels of the Reichardt detector, different for the lobula plate, where dendrites of the so-called
a delay line in the Barlow-Levick model), and (3) they all possess lobula plate tangential cells span large parts of the neuropile,
an essential nonlinearity (division in the gradient detector, a multi- apparently collecting signals from local neurons within hundreds
plication in the Reichardt detector, and an AND-NOT gate in the of columns. These tangential cells have been thoroughly ana-
Barlow-Levick model). lyzed, first in the blow fly Calliphora (Hausen, 1982a; 1982b;
They differ, however, in many other aspects that can be used Hengstenberg, 1982; Hengstenberg et al., 1982; Borst and
to discriminate between them experimentally. (1) As a character- Haag, 1996; Haag et al., 1997; 1999) and, more recently, also
istic hallmark, the gradient detector delivers a signal that is in the fruit fly Drosophila (Joesch et al., 2008; Schnell et al., 2010).
proportional to image velocity independent of the local image Although the exact number depends on the species, the
contrast. (2) The output of the Reichardt detector grows quadrat- tangential cells comprise roughly 50 neurons, each of which
ically with image contrast. Furthermore, it displays a maximum at can be uniquely identified on the basis of its anatomy, receptive
a certain image velocity. The optimum velocity is proportional to field, and electrical response properties. All tangential cells
the spatial pattern wavelength such that the maximum response respond to visual motion in a DS way. Among them, the three
is always at the same temporal frequency (image velocity divided cells of the horizontal system, called HS cells, respond most
by pattern wavelength). (3) The Barlow-Levick model is charac- strongly to horizontal image motion: When the pattern moves
terized by a null-direction inhibition. from the front to the back, the cells depolarize (Figure 2B). This
For an experimental analysis, it is also important to make the direction of image motion is their preferred direction. When the
distinction between the response properties of the individual pattern moves from the back to the front, they hyperpolarize.
local motion detector, and those of a spatially integrated de- This direction of image motion is their null direction. When
tector array. When stimulated by a periodic grating moving at a stimulated by a moving bar instead of a grating, their preferred
constant velocity, the local gradient detector will signal a con- and null direction remains the same, no matter whether a white
stant value as well. In contrast, the output signal of a local Reich- bar is moving a black background or a black bar on a white
Review
Review
background (Figure 2C). Another prominent group, the cells of DS circuitry as one of the most investigated and best understood
the vertical system, called VS cells, in general respond most neuronal circuitries in the vertebrate brain.
strongly to vertical image motion; downward is their preferred Types of DS Ganglion Cells
direction and upward is their null direction. However, precise The first type of retinal DS ganglion cells fires both at the leading
mapping of the cells local preferred directions revealed a and the trailing edge of a stimulus moving along the preferred
spatially nonuniform receptive field; the different preferred direc- direction through the receptive field (Barlow and Levick, 1965).
tions in different parts of the flys visual field resemble an optic In other words, a bright spot on a dark background evoked
flow pattern as might be elicited by the fly during certain flight very similar DS responses as a dark spot on a bright background.
maneuvers (Krapp and Hengstenberg, 1996; Krapp et al., Due to this contrast independence, this cell type is referred to as
1998). These large and elaborate receptive fields could be shown ON/OFF DS ganglion cell (for review, see Masland, 2004; Vaney
to result from a combination of direct feed-forward input the et al., 2001). They have a distinct morphology with loopy
tangential cells receive from columnar motion-sensitive ele- dendrites (Figure 3A; Amthor et al., 1984; 1989) ramifying in
ments and lateral synaptic interactions between the various both the ON and the OFF sublamina of the inner plexiform layer
tangential cells within the lobula plate (Borst and Weber, 2011; (IPL) (Figure 3D, red cell). The two arborizations can differ in size
for review, see Borst et al., 2010). and shape (Oyster et al., 1993; Vaney, 1994), suggesting that the
As for the nature of their retinotopic input elements, the lobula ON and the OFF DS circuits work independently. ON/OFF DS
plate tangential cells have been subjected to numerous tests ganglion cells are inhibited by synchronous motion outside their
investigating whether they conform to the Reichardt model in receptive field center and are, thus, sensitive to motion contrast
blow flies, hover flies, and fruit flies. In these experiments, (Chiao and Masland, 2003). As a result of their response proper-
tangential cells were stimulated by periodic gratings moving at ties, ON/OFF ganglion cells are considered to be local motion
a constant velocity (Haag et al., 2004; Joesch et al., 2008; detectors. They display a rather broad tuning in both the
Schnell et al., 2010) or with a dynamic velocity profile (Egelhaaf temporal and spatial frequency domain (see e.g., Figure 2 in
and Reichardt, 1987; Egelhaaf and Borst, 1989; Borst et al., Grzywacz and Amthor, 2007). Nevertheless, they seem to be
2003; Reisenman et al., 2003; Borst et al., 2005; Spavieri et al., tuned to the temporal frequency of the stimulus rather than to
2010). Some studies investigated the local motion response by its velocity, speaking in favor of the Reichardt detector as an
restricting the field of view to a small window through which appropriate description of the underlying mechanism. ON/OFF
the pattern was shown to the fly (Egelhaaf et al., 1989) or by using DS cells can be clustered into four functional subtypes (Oyster
intracellular calcium concentration changes as a readout for and Barlow, 1967), each of which preferring a different motion
local activity in the dendrite (Single and Borst, 1998; Haag direction roughly parallel to the dorsal-ventral (superior, inferior)
et al., 2004). Tangential cells were also stimulated by natural or nasal-temporal (anterior, posterior) axis (Figure 3D, bottom).
images (Dror et al., 2001) or by apparent motion stimuli con- A second type of DS cell responds to only the leading edge of
sisting of spatially displaced sequences of discrete brightness a bright stimuli moving on a dark background and is, therefore,
steps (Egelhaaf and Borst, 1992). All these studies concluded referred to as an ON DS ganglion cell. They are monostratified
that the Reichardt detector accurately describes the behavior (Figure 3B), and their dendritic arborization ramifies in the inner
of these input elements. As an example, the responses of (ON) sublamina of the IPL (Figure 3D, blue cell) (Amthor et al.,
Drosophila HS cells have been measured as a function of pattern 1989; Buhl and Peichl, 1986; He and Masland, 1998). In contrast
contrast (Figure 2D): Although the response does not rise to ON/OFF DS cells, ON DS cells respond best to global motion
quadratically as predicted by a perfect multiplication, it clearly (Wyatt and Daw, 1975) and are tuned to lower temporal frequen-
increases with increasing pattern contrast, thus ruling out a cies (Grzywacz and Amthor, 2007). With respect to their
division of temporal by spatial gradient as specified in the preferred direction, ON DS ganglion cells can be clustered into
gradient detector. When stimulated by a periodic grating drifting three subtypes (Figure 3D, bottom). There is also recent
at different velocities, the response of HS cells displays a velocity evidence for further functional subdivision into transient and sus-
optimum, as predicted by the Reichardt detector (Figure 2E, tained types, each of which has distinct anatomical features
black trace). Furthermore, when the test is repeated with a (Kanjhan and Sivyer, 2010).
grating of twice the spatial wavelength, the optimum velocity is Recently, a third type of DS cell was discovered in transgenic
doubled (Figure 2E, gray trace). When the pattern velocity is mice expressing green fluorescent protein under the control of
divided by the spatial wavelength of the pattern, both curves the junctional adhesion molecule B (JAM-B) promoter exclu-
coincide, revealing a peak at the same temporal frequency sively in a subset of ganglion cells (Kim et al., 2008). JAM-B posi-
of 1 Hz (Figure 2F), exactly as predicted by the Reichardt tive cells have a peculiar morphology: Their asymmetrical
detector. wedge-shaped dendritic arbors are aligned with the dorsal-
ventral axis of the retina and point ventrally (Figure 3C). They
Vertebrate Retina respond best to centripetal motion, i.e., from the soma to the
The first reports of DS neurons in the vertebrate retina appeared dendritic tips, and thus, are directionally tuned to upward motion
in the 1960s (for references see Wyatt and Daw, 1975). In partic- (Figure 3C, bottom)taking into account that the lens inverts the
ular, an elegant series of papers by Barlow, Levick, and co- retinal image. With the exception of very large diameter spots,
workers (e.g., Barlow and Hill, 1963; Barlow et al., 1964; Barlow they fire only at the offset of a light spot and have their dendrites
and Levick, 1965) on DS ganglion cells in the rabbit retina initi- at the distal border of the IPL (Figure 3D, green cell). Neverthe-
ated more than 40 years of research that established the retinal less, they respond to preferred direction motion for both
Review
A B C
contrasts (Kim et al., 2008). Interestingly, ganglion cells with Projections of Retinal DS Ganglion Cells
asymmetrical dendrites but orientation-selective responses, Starburst cells represent a type of amacrine cell (Famiglietti,
reminiscent of mouse JAM-B cells, have been reported in the 1983; Masland and Mills, 1979) that had been suggested to be
rabbit (Amthor et al., 1989). Thus, OFF DS ganglion cells might critical for direction selectivity. When selectively ablated through
also exist in other species. a nifty genetic manipulation, ON and ON/OFF DS ganglion cell
Review
Review
responses became indiscriminate to directional motion (Yoshida physiological recordings, primate ganglion cells that are
et al., 2001). Moreover, this manipulation also resulted in a com- morphologically equivalent to rabbit DS cells have been docu-
plete loss of the optokinetic nystagmus (OKN) (Amthor et al., mented (Dacey, 2004; Yamada et al., 2005). Also starburst
2002; Yoshida et al., 2001). This indicates that one or both of amacrine cells, which are crucial to the DS circuitry, have been
these DS cell types provide signals essential for the control of found (Rodieck, 1989). Furthermore, retrograde tracing data on
eye movement and gaze stabilization (reviewed in Berson, the retinal projections to the AOS are consistent with the pres-
2008; Vaney et al., 2001). It is likely that ON DS cells are the ence of ON DS ganglion cells in primates (Telkes et al., 2000).
main source of visual input for these tasks (Oyster et al., 1972), Direction selectivity has also been studied in several nonmam-
because they prefer global motion, as caused by image slip- malian vertebrates (Vaney et al., 2001; Wyatt and Daw, 1975).
page. Furthermore, their preferred directions correspond to For instance, DS ganglion cells in turtle (Marchiafava, 1979)
the three axes of the semicircular canals in the inner ear have functional properties very similar to those of mammals
(Figure 3D, bottom; see also Simpson et al., 1988b). Instead of (Borg-Graham, 2001). Birds also possess retinal DS cells (for
projecting to the superior colliculus (SC) and the lateral genicu- research on pigeons see Pearlman and Hughes, 1976), but little
late nucleus (LGN), like the majority of other ganglion cell types, is known about the underlying circuitry (e.g., Uchiyama et al.,
ON DS ganglion cells indeed project to the accessory optic 2000). Results from fish suggest how the refinement of retinal
system (AOS), a collection of nuclei that controls eye movement DS circuitry might have progressed during evolution: In ancient
(Figures 3E and 3F, for review, see Berson, 2008). Using trans- fish, such as dogfish, DS ganglion cells were found by recording
genic mice, researchers confirmed that the axonal projections retinal input from pretectal neurons, but no clustering of
of ON DS cells with different preferred direction form discrete preferred directions is evident (Masseck and Hoffmann, 2008).
clusters in the medial terminal nucleus, the primary nucleus of By contrast, recordings from the optic tectum of modern fish,
the AOS (Yonehara et al., 2009), as proposed earlier (Simpson like carp, provide evidence for retinal ON and OFF DS cells,
et al., 1988a). The ON/OFF DS ganglion cells also provide each with three clusters of preferred directions (Damjanovic
some input to the AOS and, therefore, contribute to the control et al., 2009).
of eye movement, possibly for higher velocities. Consistent
with this is also the fact that their preferred directions are roughly Network, Cellular, Subcellular, and Biophysical
aligned with the four directions of apparent movement caused by Mechanisms
eye muscles contractions (Oyster and Barlow, 1967). ON/OFF Having introduced the neurons found in various animal species
DS ganglion cells send collaterals to the SC and the LGN and, that respond to image motion in a DS way, we will now discuss
therefore, may serve other visual functions as well, such as di- what cellular, subcellular, and biophysical mechanisms give
recting attention to moving objects (reviewed in Berson, 2008). rise to this particular response property.
No projections to the AOS were found for the JAM-B positive
OFF DS ganglion cells; they project to the SC and the dorsal Insects
LGN (Kim et al., 2008), but the functional role of these inputs is As outlined above, there is overwhelming evidence that the lob-
not yet understood. Altogether, with the exception of the contri- ula plate tangential cells of flies receive input from arrays of local
bution to the optokinetic system, little is currently known about motion detectors of the Reichardt type. However, the small size
the functional role of retinal direction selectivity for higher visual of the columnar elements in the optic lobe has made it difficult
processing. to determine which of the many cells take part in the neural
Retinal Direction Selectivity across Vertebrate Species circuitry implementing this algorithm. However, this situation
Only recently, with the tremendous increase in transgenic mouse has changed recently, largely due to the application of electro-
diversity, research on DS mechanisms started to shift from physiological recording techniques to Drosophila (Wilson et al.,
rabbits, on which most studies had focused, toward mice. 2004; Joesch et al., 2008; Maimon et al., 2010), in combination
Despite a few minor differences, ON and ON/OFF DS ganglions with the wide armory of genetic tools already available for this
cells are functionally and morphologically very similar in mice organism (for review, see Borst, 2009).
(Sun et al., 2006; Weng et al., 2005) and rabbits. There is First of all, it was demonstrated that Drosophila tangential cells
evidence for retinal direction selectivity in other mammals (for receive excitatory and inhibitory input from local motion sensitive
review see Vaney et al., 2001), and therefore, it is conceivable elements with opposite preferred direction (Joesch et al., 2008).
that this function is largely conserved among mammals. Interest- This was done by injecting depolarizing and hyperpolarizing
ingly, in primates the existence of retinal direction selectivity has current into the tangential cell during motion stimulation in the
not yet been convincingly shown. It is possible that this absence preferred and null direction (Figure 4A): Without current injection,
reflects a sampling bias specific to primates: Compared to the visual stimulation leads to depolarization of the cell during
overwhelming number of, for example, midget ganglion cells, preferred direction motion and hyperpolarization during null
which underlie high acuity vision, DS cells may be too infrequent. direction motion (Figure 4A, middle trace). When depolarizing
Supporting the notion that these cells might have been missed in current is injected, the preferred direction response becomes
(G) Blocking the synaptic output from either L1 or L2 results in selective loss of tangential cell responses to either moving ON-edges (L1-block, in green) or moving
OFF edges (L2-block, two different driver lines, in red). Asterisks indicate the significance level of the difference between the mean values: *p < 0.05, **p < 0.001.
NS is an abbreviation for not significant. (from Joesch et al., 2010).
(H) Splitting of brightness information into ON and OFF pathways, corresponding to L1 and L2, respectively.
Review
smaller and the null direction response larger (top trace). The synthesis of acetylcholine (ACh) (Raghu and Borst, 2011), while
opposite is observed during injection of hyperpolarizing current T5 cells activate the promoter upstream of the gene encoding
(bottom trace). This can be reproduced by simulation of a single the vesicular glutamate transporter (VGluT) (Raghu and Borst,
electrical compartment model that receives two synaptic inputs 2011). However, a conclusive physiological proof that indeed
with reversal potentials above and below the resting potential of T4 cells are cholinergic and T5 cells are glutamatergic, and
the cell: The depolarizing current injection reduces the driving whether they exert excitatory or inhibitory action on the lobula
force for the excitatory input while increasing it for the inhibitory plate tangential cells, is still missing. Furthermore, it is not known
input, and hyperpolarizing current injection does the opposite which cells provide the above mentioned GABAergic input to the
(Figure 4B). These results suggest that the subtraction stage in tangential cells.
the Reichardt detector is localized within the tangential cells Based on the costratification of columnar cells in individual
dendrites. Earlier experiments on blow fly tangential cells arrived layers of the medulla, lobula, and lobula plate as well as
at similar conclusions (Borst and Egelhaaf, 1990; Borst et al., 2-deoxy-glucose labeling, T4 and T5 cells were proposed to
1995; Single et al., 1997). The chemical identity of the transmitter be the target cells of two separate pathways starting from the
systems involved in this push-pull input organization was clari- photoreceptor terminals R1-6 in the lamina (Figure 4C; Fisch-
fied by in vitro studies of blow fly lobula plate tangential cells. bach et al., 1992). Here, the photoreceptor terminals surround
These studies indicated that excitation is mediated by excitatory the dendrites of two large lamina neurons, called L1 and L2,
nicotinic acetylcholine receptors (nAChRs) and inhibition by which contact separate strata in the medulla (Figure 4E). There,
g-aminobutyric acid (GABA) receptors (Egelhaaf et al., 1990; the signals are supposed to be picked up by specific intrinsic and
Brotz and Borst, 1996; Brotz et al., 2001; Single et al., 1997). transmedullary neurons that terminate in the dendritic areas of
Through the use of a Gal4-driver line that leads to expression T4 and T5 cells, respectively. Keeping in mind the limited
in lobula plate tangential cells of two types of labeled reporter evidence for the existence of these pathways to begin with,
genes, excitatory and inhibitory transmitter receptors were found one could only speculate how the signals in these two pathways
to be colocalized on the fine dendritic branches of HS and VS differ and how they might correspond to the Reichardt model.
cells of Drosophila (Raghu et al., 2007; 2009). Thus, direction This situation has changed due to a study where tangential cell
selectivity in the tangential cells results from summation of two responses were recorded in Drosophila while the chemical trans-
inputs with opposite preferred directions. mitter release from L1 or L2 cells (Figure 4F) was genetically
But what neurons represent these excitatory and inhibitory blocked in a cell-specific way (Joesch et al., 2010). While block-
input elements to the lobula plate tangential cells? For a number ing the output from either L1 or L2 led to reduced but still signif-
of reasons, bushy T cells are the prime candidates for providing icant responses to drifting gratings, blocking L1 completely and
input to the lobula plate tangential cells. T4 cells exist in four selectively abolished the response to drifting ON-edges, and
different subtypes per column, with dendrites ramifying in the blocking L2 erased the response to drifting OFF edges (Fig-
most proximal layer of the medulla. Each of the four T4-cell ure 4G). Using a behavioral readout instead of tangential cell
subtypes projects into one out of four different strata of the lob- responses, another study obtained similar results (Clark et al.,
ula plate (Figure 4C). In a similar way, four subtypes per column 2011). These findings demonstrate that in fruit flies, the photore-
are found for T5 cells as well, and they connect the posteriormost ceptor signal from R1-6 is split in the lamina into separate ON
layer of the lobula to one of the four strata of the lobula plate. and OFF pathways, represented by L1 and L2 cells, respectively.
Following extended stimulation by moving gratings, Buchner This is analogous to the vertebrate retina where cone photore-
et al. (1984) found strong 2-deoxy-glucose labeling in one of ceptors contact ON and OFF bipolar cells in parallel (reviewed
the four layers in the lobula plate depending on the particular in Wassle, 2004). However, in the vertebrate retina the split is im-
direction of the motion stimulus (Figure 4D). The direction of plemented by different types of glutamate receptors in ON and
motion which activates a specific stratum, as labeled using the OFF bipolar cells (Nomura et al., 1994) so that light depolarizes
2-deoxy-glucose method, matches the preferred direction of ON bipolar cells and hyperpolarizes OFF bipolar cells. In fruit
those tangential cells extending their dendrite in that stratum. flies, however, the dendritic membrane response to light is iden-
In addition to the lobula plate, 2-deoxy-glucose labeling was tical in L1 and L2 and consists of a transient hyperpolarization at
highest in the most proximal layer of the medulla, where T4 cells the beginning and a rebound excitation at the end of a light pulse.
ramify, and in the posterior most layer of the lobula, where T5 In L2 cells, the selectivity for light decrements seems to originate
cells extend their branches (Buchner et al., 1984). Finally, an in the axon terminal, as suggested by Ca2+ imaging (Reiff et al.,
electron microscopy study in the blow fly has shown unequivo- 2010): Whereas the intracellular calcium concentration is only
cally a chemical synapse between an HS-cell dendrite and a slightly reduced at the onset of light, a large and long lasting
columnar T4 cell (Strausfeld and Lee, 1991). Because of their calcium increase is elicited by light offset. Thus, L2-terminals
small size, however, the visual response properties of T4 and amplify predominantly the off-signal to postsynaptic neurons.
T5 cells have proven very difficult to study. The few successful L1-terminals reveal calcium signals similar to the ones of L2-
recordings showed that T5 cells reveal a fully DS response, terminals but with a stronger decrease of calcium concentration
whereas T4 cells are direction unselective (Douglass and Straus- at light onset (Clark et al., 2011). The easiest way to explain the
feld, 1995; 1996). As to the type of transmitter these cells use, selectivity for light increments in the L1-pathway would be to
recent studies identified T4 cells as among the group of neurons assume a signal inversion by inhibition with subsequent rectifica-
activating the ChAT-promoter, which controls the expression of tion in the neurons postsynaptic to L1. This, however, remains
the enzyme choline-acetyl-transferase (ChAT) involved in the a pure speculation at present.
Review
An interesting question following from the finding about the view is that in the fruit fly, two separate motion detection systems
splitting of the input into ON and OFF pathways concerns the operate in parallel, one analyzing the movement of light incre-
number of motion detector subtypes being at work in the fly brain: ments and the other one the movement of light decrements
Do four different detectors exist, one for each stimulus combina- (Figure 4H). While the exact nature and role of the participating
tion (ON-ON, OFF-OFF, ON-OFF, OFF-ON), or are there only two neurons is still unclear, the splitting of the positive- and nega-
detectors (ON-ON, OFF-OFF)? The most intuitive experiment to tive-going brightness signal into two channels, one for signals
investigate this question is the use of apparent motion stimuli in of the positive, the other for signals of the negative sign, has
which the brightness in two adjacent bars is stepped sequentially interesting consequences for the multiplication as postulated in
from an intermediate level, that is also present in the surround, to the Reichardt detector. Without splitting, the output of such
either a high (ON-Step) or to a low (OFF-Step) level. By applying a putative multiplication neuron would need to increase in
such stimuli to blow flies and fruit flies, various studies con- a supralinear way when both input signals go positive as well
sistently found positive responses to ON-ON and OFF-OFF as when they go negative. Such a mechanism is difficult to
sequences and negative responses to ON-OFF and OFF-ON realize. However, splitting the input into separate channels leads
sequences, either at the level of lobula plate tangential cells or to positive signals only, and while a number of biophysically
in a behavioral assay (Egelhaaf and Borst, 1992; Eichner et al., plausible mechanisms have been proposed to do that (Torre
2011; Tuthill et al., 2011; Clark et al., 2011). While these findings and Poggio, 1978; Srinivasan and Bernard, 1976; Gabbiani
seem to clearly indicate the existence of four detector subtypes, et al., 2002; Hausselt et al., 2007; Enciso et al., 2010), the exact
careful quantitative modeling, including the peripheral filter mechanism active within these neurons presynaptic to the fly
stages, suggests that responses to mixed-brightness steps can LPTCs remains to be determined. In a similar way, the bio-
also be obtained from only two detectors (ON-ON and OFF- physics underlying the temporal filtering as postulated by the
OFF) if some residual information about the average brightness Reichardt detector represents another challenge for future
level is preserved at the motion-detector input. Using a more research.
selective stimulus sequence consisting of brief brightness pulses
instead of steps led to responses to pulse sequences of the same Vertebrate Retina
sign only, ruling out the existence of mixed-sign motion detectors Since the majority of studies focused on ON/OFF DS cells and
in blow flies and fruit flies (Eichner et al., 2011). This conclusion is their circuitry, we will concentrate on those, while only briefly
supported by an earlier study on house flies, Musca domestica, touching upon other types (see Mechanisms in Other Types of
that used sophisticated optics to sequentially stimulate individual Retinal DS Ganglion Cells). The original Barlow-Levick model
photoreceptors within one ommatidium projecting to neigh- (Barlow et al., 1964; reviewed in Masland, 2004) proposed that
boring cartridges in the lamina (Franceschini et al., 1989). While DS ganglion cells receive delayed and/or long lasting inhibition
ON-ON and OFF-OFF sequences along the preferred direction preferentially from interneurons displaced to the null side of their
of the cell led to strong responses in the H1 tangential cell, no dendritic field. Note that the term null/preferred side refer to
responses were detected for mixed-sign sequences, i.e., ON- the positions from which the null/preferred direction stimulus
OFF and OFF-ON. enters the ganglion cells dendritic field. This inhibition would
In contrast to the two-detector model, Clark et al. (2011) advo- be triggered by a stimulus moving in the null direction toward
cate for a model consisting of six detectors, with an asymmetric the cells receptive field center and would cancel out any exci-
distribution of the mixed detectors across the two pathways (L1: tation caused by the stimulus when it eventually enters the
ON-ON, OFF-OFF, OFF-ON; L2: ON-ON, OFF-OFF, ON-OFF), center. Understandably, the original model does not fully
which are nevertheless selective for ON and OFF edges, respec- capture the multitiered organization of the retinal DS circuitry
tively. However, to achieve this selectivity, the model requires as it is known today. The original model did, however, identify
highly specific stimulus conditions as well as model parameters the key properties of any DS circuit (see above). A hallmark of
that are hard to reconcile with previous work. The delay-filter retinal ON/OFF DS cells is their surprising robustness: The
time constant of 10 s, necessary to reproduce edge selectivity retinal cells easily outperform their counterparts in primary visual
in Clarks model, is two to three orders of magnitude larger cortex (V1) in many respectsexcept maybe for directional
than the value derived from all previous studies (e.g., Guo and tuning width (45 in retinal ON/OFF cells versus R 15 in
Reichardt, 1987; Egelhaaf and Reichardt, 1987; Dror et al., V1, reviewed in Grzywacz and Amthor, 2007). The direction of
2001; Safran et al., 2007). This large time constant introduces motion within the ON/OFF DS cells receptive field center is reli-
a long-lasting ringing in response to the onset of motion (A.B., ably detected largely independent of contrast (Merwine et al.,
unpublished data), incompatible with the observed cellular 1998) and velocity (Grzywacz and Amthor, 2007; Oyster et al.,
responses (Egelhaaf and Borst, 1989; Reisenman et al., 2003). 1972; Wyatt and Daw, 1975), even for small movements of a
A further conflict arises in this models prediction of negative few micrometers (Grzywacz et al., 1994). Although their spiking
responses to ON-OFF and OFF-ON pulses, which are clearly frequency peaks at velocities of 30 /s, direction discrimination
absent in the experimental data (Eichner et al., 2011). Neverthe- is constant over a velocity range of more than two orders of
less, while we feel that there is evidence arguing against the magnitude (reviewed in Grzywacz and Amthor, 2007). In the light
six-detector model with mixed channels, definitive clarification of this robustness, it is very likely that the underlying circuitry
of the discrepancies will require further direct investigation. relies on multiple pathways and computational mechanisms
Taken together with the evidence of the pathways leading to generate and enhance DS signals, as we discuss in the
from L1 to T4 and from L2 to T5 cells, respectively, our current following.
Review
A
photoreceptor
bipolar cell
OFF sublamina
SACs
ON sublamina Glu
excitation
ACh
GABA inhibition
ON/OFF DS ganglion cell
(w/preferred direction to the right)
B1 B2 B3 motion
motion
+ -- + -
SAC
50% - +
-- +
C1 D1
C2 D2
E preferred null
+ + + + + + + -- + + -- +
Review
The Circuitry of ON/OFF DS Ganglion Cells glietti, 1991) and transmitter release from SACs is Ca2+-depen-
DS ON/OFF ganglion cells receive excitatory input from bipolar dent (OMalley et al., 1992; Zheng et al., 2004), this indicated
cells but also from the previously mentioned starburst cells that SACs are able to provide DS ganglion cells with directionally
(Figure 5A), which are also known as cholinergic amacrine cells tuned input. In fact, since the dendritic sectors are electrically
(Famiglietti, 1983; Masland and Mills, 1979). Besides ACh, star- isolated from each other, each sector can be thought of as an
burst amacrine cells (SACs) also release GABA (Brecha et al., independent detector for centrifugal motion (Figure 5B3). Around
1988; Masland et al., 1984b; Vaney and Young, 1988) and the same time, the related long-standing question as to whether
provide DS ganglion cells with inhibition as well (Figure 5A). In retinal direction selectivity is computed in the ganglion cells
addition, the DS ganglion cells receive both GABA and glyciner- themselves or presynaptically by interneurons (reviewed in Mas-
gic inhibition from other amacrine cell types (reviewed in Da- land, 2004) was successfully addressed. Patch-clamp studies
cheux et al., 2003). The role of this additional inhibition in the revealed that the synaptic input to ON/OFF DS cells is already
DS circuitry, however, is not yet well understood (see e.g., DS (Borg-Graham, 2001; Fried et al., 2002; Taylor and Vaney,
Neal and Cunningham, 1995). 2002): Preferred direction motion elicits more excitation and
Starburst amacrine cells (Figure 5B1) feature a characteristic less inhibition in the ganglion cells, whereas null direction motion
morphology (Famiglietti, 1983; Tauchi and Masland, 1984; Va- elicits more inhibition and less excitation. This suggested (1) that
ney, 1984) that is well conserved across vertebrate species: both inhibitory and excitatory inputs are DS, (2) that the Barlow-
Their dendritic arbor is composed of 46 sectors, each arising Levick model does not fully capture retinal DS computations,
from a primary dendrite that radiates from the soma before and (3) that the latter are indeed already performed presynapti-
dividing into smaller branches. SACs come in an ON and an cally by interneurons. Note that the latter point does not exclude
OFF variety, which appear to be functionally equivalent. They that postsynaptic, i.e., ganglion cell-intrinsic mechanisms con-
costratify with the respective dendritic subtrees of ON/OFF DS tribute to the overall direction selectivity observed in ganglion
cells; ON SACs also costratify with the ON DS type (Figure 5A) cells (see Mechanisms at the Ganglion Cell Level).
(Famiglietti, 1992). Each SAC dendrite is anatomically and phys- The Role of Inhibition
iologically strongly polarized: Synaptic inputs from both bipolar It was already known for long that blocking GABAA receptors
and amacrine cells cover the whole dendritic length, but outputs abolishes DS responses in the ganglion cells but leaves their
are restricted to the distal part (Figure 5B1) (Famiglietti, 1983). In responsiveness intact (Caldwell et al., 1978; Massey et al.,
addition, some channels and transporters are differentially 1997). This indicated that GABAergic inhibition plays a crucial
distributed along SAC dendrites, which, in combination with role not only in directly cancelling of null direction motion
the morphology, leads to electrical isolation of the sectors from responses in the ganglion cells (Barlow and Levick, 1965), but
each other (Miller and Bloomfield, 1983; Velte and Miller, also in rendering the excitatory input to the ganglion cells DS
1997). As a result, the individual dendritic sectors can be consid- (see The Role of Excitation). Furthermore, in combination with
ered as largely independent processing units (Borg-Graham and the SAC ablation experiments (Amthor et al., 2002; Yoshida
Grzywacz, 1992; Euler et al., 2002; Miller and Bloomfield, 1983; et al., 2001), it suggested that SACs are a major source for
Vaney, 1990). In contrast to most retinal neurons, SACs display this GABAergic inhibition. While a contribution from other
an extensive dendritic overlap (Tauchi and Masland, 1984), GABAergic amacrine cells cannot be excluded, it seems unlikely
which enables them to provide independent neuronal hardware that they fit the role of SACs because they typically are not
for the different functional subtypes of DS cells. numerous enough to provide the different functional subtypes
It has been proposed earlier that SACs are importantly of DS ganglion cells with adequate input (reviewed in Vaney
involved in the DS computation (Borg-Graham and Grzywacz, et al., 2001). This lack of cells may, however, be compensated
1992; Masland et al., 1984a; Vaney et al., 1989), but experimental for by highly localized dendritic processing.
proof for this notion came less than 10 years ago, when it was To distinguish between different mechanistic hypotheses of
shown that massive ablation of SACs results in a selective loss direction selectivity, it is important to differentiate between direc-
of retinal direction selectivity (Amthor et al., 2002; Yoshida tionally tuned (Figure 5B) and spatially offset inhibition (Figure 5D).
et al., 2001; but see He and Masland, 1997). Optical measure- While both originate, at least in part, in SACs, they represent
ments of light-stimulus-evoked Ca2+ concentration changes different aspects of the DS computation. Spatially offset inhibition
(Denk and Detwiler, 1999) in the dendrites of SACs demon- as such can be sufficient to render ganglion cell responses DS
strated that stimuli moving from the soma to the dendritic tips, (Figures 5D and 5E; Borg-Graham and Grzywacz, 1992; Koch
i.e., centrifugal motion, evoked larger Ca2+ responses in the et al., 1983; Barlow and Levick, 1965). Spatially offset inhibition
distal dendrites than motion in the opposite direction, i.e., is a recurrent theme in the retinal DS circuit (Fried et al., 2005).
centripetal motion (Figure 5B2) (Euler et al., 2002). Because It acts not only at the level of the ganglion cell, but possibly also
SAC output synapses are located in the distal dendrites (Fami- at the bipolar cell terminals presynaptic to the DS ganglion cells
(D) Asymmetric synaptic wiring between SACs and DS ganglion cells, determined from large-scale ultrastructural reconstruction of retina that was functionally
characterized with optical recordings (see Briggman et al., 2011, for details). Specific connectivity of synaptic SAC outputs (black skeleton), with varicosities
indicated by black dots (D1). Synapses are color-coded by the preferred direction of the postsynaptic DS ganglion cells (color-coded dashed ellipses indicate
their dendritic trees; the inset shows their preferred motion directions). Locations relative to the SAC somas of output synapses (n = 831) from all 24 reconstructed
SACs (D2). The scale bars represent 50 mm.
(E) Direction selectivity in ganglion cells (blue) result from multiple presynaptic mechanisms: (1) DS in SAC dendrites (see B), resulting in (2) null direction inhibition
via asymmetrical GABAergic input (see D) and (3) preferred direction excitation via symmetrical cholinergic input (see C and The Role of Excitation).
Review
(Figure 5E) and possibly between SACs. The first evidence for gated channels are suited to boost the SACs DS response if
DS ganglion cells receiving spatially offset inhibition from SACs differentially activated in a way that depends on the motion
as a consequence of asymmetrical wiring came from paired direction (e.g., Borg-Graham and Grzywacz, 1992; Hausselt
recordings. These revealed that SACs located on the null side et al., 2007; Tukker et al., 2004). SACs possess a variety of suit-
of a ganglion cell provide significantly stronger inhibition than able voltage-gated Ca2+ (Cohen, 2001) as well as tetrodotoxin-
SACs located on its preferred side (Fried et al., 2002; Lee resistant Na+ channels (OBrien et al., 2008). Recently, the latter
et al., 2010; Wei et al., 2011). The anatomical correlate for shifted into the focus of interest since blocking these Na+ chan-
such asymmetrical wiring has been elusive, since neither the nels led to a reduction in dendritic direction selectivity in SACs
DS cells morphology (Amthor et al., 1984) nor the distribution (Oesch and Taylor, 2010). Aside from the radially asymmetric
of synaptic inputs (Jeon et al., 2002) allowed researchers to distribution of SAC output synapses, two kinds of gradients
predict its preferred direction. While complex synaptic arrange- along SAC dendrites have been proposed to serve as functional
ments between SACs and DS ganglion cells have been described asymmetry for DS detection: a voltage and a Cl concentration
at the ultrastructural level (Dacheux et al., 2003), very recently, gradient. Optical Ca2+ measurements in SACs suggest that the
a new EM technique for recording tissue volumes of sufficient distal dendrite is tonically depolarized relative to the soma
size and resolution (Briggman and Denk, 2006) allowed re- (Hausselt et al., 2007), possibly due to tonic glutamatergic input
searchers to statistically analyze the SAC-DS ganglion cell (Taylor and Wassle, 1995; but see Oesch and Taylor, 2010).
connectivity for the first time (Figure 5D). In their elegant study, Modeling suggests that even a small dendritic voltage gradient
Briggman et al. (2011) revealed two important circuit properties: in combination with voltage-gated channels could generate
(1) The great majority of SAC synapses a DS ganglion cell receives a robust DS signal in SAC dendrites (Hausselt et al., 2007). In
come from SAC dendrites that roughly point in the ganglion cells another model it was proposed, that SACs generate a Cl
null direction, meaning that the respective SAC somata are concentration gradient along their dendrites due to a differential
located preferentially on the ganglion cells null side. (2) SACs distribution of Cl intruders and extruders, and that this results in
provide input to all DS ganglion cells within their dendritic field GABAergic input causing depolarization at the proximal and
but only via dendrites pointing in the null direction of the respec- hyperpolarization at the distal dendrite, respectively (for details
tive ganglion cell. Moreover, Briggman et al. (2011) found that the see Enciso et al., 2010; Gavrikov et al., 2003; Gavrikov et al.,
probability for a synaptic contact is substantially higher when the 2006). According to this model, the asymmetry in the effect of
local angle of the SAC dendrite closely matches the ganglion GABAergic inputs leads to dendritic direction selectivity. Other
cells null direction, suggesting that the synapse forming mecha- than the voltage gradient model, the Cl gradient model requires
nism makes use the local dendritic geometry and/or activity. GABAergic input and therefore does not account for the finding
The retinal DS circuitry does not solely rely on spatially offset that SAC responses remain DS in the presence of GABA
inhibition from SACs but ensures that this inhibition itself is DS receptor blockers (see below).
(Figure 5B). A number of models have been put forward to Ultrastructural (Millar and Morgan, 1987) and functional data
explain the generation of dendritic direction selectivity in SACs (Zheng et al., 2004) indicate that mature SACs form reciprocal
(reviewed in Euler and Hausselt, 2008). While these models differ GABAergic synapses, which have been implicated in the compu-
in the neuron types recruited or in the biophysical mechanisms tation of DS signals (e.g., Munch and Werblin, 2006). If a SAC is
employed, they are not necessarily mutually exclusive. One class excited, it inhibits its neighborthis in turn reduces the neigh-
of models focuses on intrinsic properties of SACs and makes use bors GABA release and in effect enhances the first SACs
of the fact that their dendrites are polarized computational response. Such interaction may sharpen the DS contrast in
subunits (see The Circuitry of ON/OFF DS Ganglion Cells). neighboring SAC dendrites pointing in opposite directions (Lee
Passive models predict that SAC dendrites generate weak DS et al., 2010; Lee and Zhou, 2006). However, since GABA receptor
signals simply by input summation along the dendrite (Tukker antagonists do not abolish dendritic direction selectivity in SACs
et al., 2004). As a result, centrifugal motion (which extends (Euler et al., 2002; Hausselt et al., 2007; Oesch and Taylor, 2010),
from the soma to the dendritic tips) evokes a larger signal in it is unlikely that these interactions are essential for the SACs
the dendritic tips, whereas centripetal motion evokes a larger intrinsic DS mechanism.
signal in the soma (see also Borg-Graham and Grzywacz, The Role of Excitation
1992; Branco et al., 2010). As the SACs output synapses are In addition to inhibition, DS ganglion cells receive DS excitatory
located in the dendritic tips, centrifugal motion would lead to input from bipolar cells (Fried et al., 2005). This tuning could arise
a larger output signal. In this scenario, the location of the output from DS suppression of bipolar cell output by GABAergic ama-
synapses serves as spatial asymmetry and the threshold for crine cells (Figure 5E), which would explain why this excitatory
transmitter release as the essential nonlinearity. The direction DS pathway is eliminated by GABA receptor blockers (see The
selectivity found in passive models seems, however, small and Role of Inhibition). Because ablating SACs abolishes ganglion
too sensitive to stimulus parameters to explain the robust direc- cell DS responses (Amthor et al., 2002; Yoshida et al., 2001), it
tion discrimination observed at the ganglion cell level. Also, is likely that SACs are involved in tuning bipolar cell outputif
passive models predict a larger electrical signal at the soma other amacrine cells were crucial, some residual direction selec-
for centripetal motion which contradicts experimental observa- tivity after ablation would be expected.
tions (Euler et al., 2002; Peters and Masland, 1996). Besides glutamatergic excitation, DS ganglion cells also
To circumvent the deficits of passive models, voltage-gated receive excitatory cholinergic input from SACs (reviewed in Va-
channels have been incorporated as nonlinearities. Voltage- ney et al., 2001). Blocking cholinergic receptors in the presence
Review
of GABA receptor antagonists reduces the responses of DS is likely considering the highly branched morphology of DS
ganglion cells independent of motion direction (Chiao and ganglion cells.
Masland, 2002), suggesting that cholinergic excitation provides Mechanisms at the Ganglion Cell Level
motion-sensitive but not DS excitation (He and Masland, 1997). By using dendritic Ca2+ imaging, it was shown that light stimuli
On the other hand, there is also evidence that this cholinergic can locally initiate spikes in DS ganglion cell dendrites and that
input is DS (Figure 5C, Fried et al., 2005; Lee et al., 2010). these dendritic spikes are independent of the somatic spike
In fact, paired recordings confirmed that SACs provide DS generator (Oesch et al., 2005). The role of these dendritic spikes
ganglion cells with cholinergic input from all sides but also in ON/OFF DS cells was recently studied in a detailed biophys-
showed that this input is modulated by GABA (Lee et al., 2010). ical compartment model (Figure 6, Schachter et al., 2010). The
SACs on the preferred side of a DS cell release ACh unhindered simulation results suggest that the dendritic arbor of DS ganglion
and thereby facilitate the motion response of the ganglion cells is partitioned into separate electrotonic regions (Figure 6B),
cell, whereas the ACh release from SACs on the ganglion cells each of which sums locally inhibitory and excitatory inputs to
null side is suppressed by inhibitory inputs (Figure 5E, Lee et al., decide whether or not a dendritic spike is fired. The dendritic
2010). In other words, in the cholinergic pathway direction selec- spikes not only sharpen the directional tuning of the synaptic
tivity results from DS modulation of otherwise symmetrical input input, but are also needed to relay the decision of the dendritic
to the ganglion cells. This is in stark contrast to the GABAergic regionindependently of the activity in other regionsto the
pathway where the asymmetry is implemented as spatially- soma, where a somatic spike can then be triggered. The model
biased synaptic connectivity. Three findings highlight that the also suggests that synaptic inhibition can prevent local spike
interactions between the cholinergic and the GABAergic path- initiation but is not strong enough to suppress spike propagation.
ways are still not fully understood: (1) The cholinergic pathway Schachter and colleagues (2010) provide a sensible explanation
appears to be more relevant for grating stimuli (Grzywacz et al., for the so-called nondiscriminating zone, an area without DS
1998). (2) EM data suggest that SACs located on the preferred responses located at the preferred side of the ganglion cells
side of a DS ganglion cell make only few synapses with this cell dendritic arbor (Barlow and Levick, 1965): They suggest that
(Briggman et al., 2011), which leaves one wondering how the the distal dendrites of DS ganglion cells are intrinsically DS
cholinergic signals are relayed. Paracrine ACh release is a possi- with their preferred directions oriented radially from the cells
bility, which is supported by the fact that while SACs are the sole center. Such intrinsic direction selectivity would interact with
source of retinal ACh, even ganglion cells that do not costratify the DS synaptic input, enhancing it at the null side while antago-
with SACs possess ACh receptors. (3) In SACs, GABA and ACh nizing it at the preferred side. Taken together, dendritic spike
are differentially released in a Ca2+ level-dependent way, likely initiation and electrotonic isolation of dendritic regions allow
from separate vesicle populations (Lee et al., 2010), adding the ganglion cells to respond reliably to different scales of motion
another level of complexity to the circuitry. within their receptive field center and to sharpen the rather broad
Recent modeling data (Poleg-Polsky and Diamond, 2011; directional tuning of the synaptic inputs (Figure 6C).
Schachter et al., 2010) suggest that the observed direction- Mechanisms in Other Types of Retinal DS Ganglion Cells
dependent difference in excitatory input can be alternatively ex- Although much less is known about the DS mechanisms in the
plained by interactions between excitatory and inhibitory other types of DS ganglion cells, it is likely that the circuitry of
conductances in the ganglion cells, without requiring DS excita- the ON DS cells resembles that of ON/OFF DS cells in many
tion. Such electrotonic interactions are to be expected if a cells respects. For instance, conductance measurements revealed
membrane potential cannot be spatially well controlled, which that ON DS cells receive directionally tuned inhibitory and
Review
excitatory input (Sivyer et al., 2010)similar to the ON/OFF cells. ula plate cells. Thus, the dendritic spanning field seems develop-
There are some circuit differences to be expected because mentally tightly controlled, while the branching topology is not.
unlike their ON/OFF counterparts, ON DS cells respond to global Assuming that synapses are located mostly toward the endings
motion. The recent finding that wide-field amacrine cells affect of the dendritic branch, this would suggest that branch topology
ON DS cell responses via gap junctions (Ackert et al., 2009) has no influence on the wiring of the particular cell as long as the
may be relevant in this context. branches reach their presynaptic input at the proper location.
While JAM-B positive OFF DS cells (Kim et al., 2008) have not Another study directly investigated the effect of sensory experi-
yet been studied in detail, there are a number of findings indi- ence on the anatomy of the lobula plate tangential cell VS1 in
cating that they employ rather different mechanisms than the Drosophila (Scott et al., 2003). The authors compared this cell
ON and ON/OFF types: (1) Because the dendrites of OFF DS from individuals that were raised in a 12 hr/12 hr light-dark cycle
ganglion cells stratify in a different IPL level than the SACs to the same cell in flies that were dark-reared. When the mor-
(Figures 4C and 4D, Kim et al., 2008), a substantial involvement phology of the dendrite as well as the axon terminals was exam-
of SACs is unlikely. (2) The DS tuning strength of OFF DS cells is ined, no conspicuous differences could be observed between
positively correlated with the degree of dendritic asymmetry, the two groups. Thus, visual experience has no influence
suggesting that the cells curious morphology plays a crucial on the anatomy of the tangential cells. A third study investigated
role for DS (Kim et al., 2008). This is in contrast to ON/OFF DS the influence of visual experience on the receptive field proper-
ganglion cells, where the cells morphology does not predict its ties of tangential cells in blow flies (Karmeier et al., 2001). Again,
preferred direction. (3) In the OFF DS cells, the responses to the authors compared neurons from two groups, one raised at
dark moving spots are tuned to higher velocities than those to a 12 hr/12 hr light-dark cycle and the other reared in the dark.
bright moving spots, suggesting different ON and OFF DS mech- The receptive field of the neurons from both groups turned out
anismsagain in contrast to ON/OFF DS ganglion cells. to be indistinguishable. This clearly indicates that, despite the
quite dramatic changes seen in the peripheral parts of the visual
Establishment of DS during Development system, the circuits responsible for direction selectivity in the fly
Another intriguing question is what mechanisms lead to the do not require visual experience to properly develop.
selective asymmetrical wiring in neuronal circuits that compute
the direction of motion during development. Vertebrate Retina
How the retinal DS circuitry arises during development has not
Insects yet been solved (Elstrott and Feller, 2009). This process very
The stereotyped anatomy of the insect optic lobe, which has a likely does not require visual input: DS responses in ganglion
high degree of invariance of identified neurons between cells can be recorded already at eye opening (Masland, 1977)
different individuals, seems to indicate, right from the begin- and attempts at modifying retinal direction selectivity by various
ning, that visual experience does not play a major role in manipulations, such as dark-rearing (Chan and Chiao, 2008;
shaping the circuits in the visual system, including the ones Chen et al., 2009; Elstrott et al., 2008; Yonehara et al., 2009),
responsible for direction selectivity. However, in contrast to blocking of GABAergic inhibition (Wei et al., 2011), or raising
these expectations, early visual experience was demonstrated animals in environments with biased motion direction statistics
to have a rather strong effect on the size of the optic lobes in (Daw and Wyatt, 1974), were unsuccessful. SACs appear very
Drosophila (Barth et al., 1997). Monocular deprivation (done early during development (Feller et al., 1996) and form a network
by painting over one eye) decreased the total volume of the that is dominated by excitatory cholinergic interactions and
lamina, medulla, and lobula plate by up to 6%, and the lamina involved in generating activity waves (reviewed in Masland,
showed a volume difference of up to 30%. The changes in 2005), which have been proposed to participate in the DS
the lamina were largely attributable to changes in the terminals circuitry generation. Recent work, however, suggests that these
of the photoreceptor cell axons. Another study in house flies activity waves are not crucial for setting up the asymmetrical
similarly concluded that flies reared in constant darkness have wiring of the DS circuitry because ganglion cell DS responses
about 20% fewer synapses than the control group (Rybak and were not altered in knock-out mice lacking cholinergic waves
Meinertzhagen, 1997). (Elstrott et al., 2008). Also recent recordings of DS ganglion cells
However, the question is still whether the motion detection in the developing retina showed that while the cells responded to
circuit function is impaired by the changes reported above. passing waves, the direction of wave propagation did not modu-
This question has been addressed by investigating the anatomy late the cells responses (Elstrott and Feller, 2010).
and physiology of the lobula plate tangential cells. Studying their The dendritic architecture of DS ganglion cells and SACs is
anatomy in blow flies, Cuntz and colleagues (2008) quantified the established before the retina becomes light responsive (Wong
dendrites of the same identified neurons from different individ- and Collin, 1989; Wong, 1990; Stacy and Wong, 2003). Two
uals. They measured the branching topology in terms of numbers recent studies investigated the development of the synaptic
and distribution of branch points and branch angles, as well as connectivity between SACs and DS ganglion cells by using
the area covered by the dendrite, their so-called spanning field, two different but equally elegant experimental approaches. Yo-
within the lobula plate. The astonishing outcome was that nehara et al. (2009) used optogenetics to examine the develop-
branching topology varies widely from individual to individual ment of the connectivity between SACs and ON DS ganglion
while the dendritic spanning field is highly invariant and can be cells in transgenic mice, in which (1) upward motion preferring
used as a single parameter to distinguish between different lob- ON DS ganglion cells were fluorescently labeled, and (2) SACs
Review
expressed Cre-recombinase. Via viral transfection, the latter Optic Flow Processing Neurons
allowed for targeted expression of channelrhodopsin (Ch2R), Neurons responding to optic flow stimuli are found in the insect
a light-sensitive cation channel (Nagel et al., 2003). In this way, lobula plate (Krapp and Hengstenberg, 1996) as well as in area
SACs could be directly light-activated before the retina becomes medial superior temporal visual area (MST) of primates (Duffy
light sensitive, which in mice is around postnatal day (P) 10. By and Wurtz, 1997). In both cases, these neurons have extremely
activating SACs at different positions around an ON DS cell, large receptive fields and possess different preferred directions
they probed the synaptic connections between the two cell in different locations of the receptive field. Furthermore, both
types. They revealed that within a 2 day time window (P6P8) classes of neurons receive DS input from upstream cells that
inhibitory connections change from symmetrical to asymmet- have small receptive fields. Whether this input to MST cells
rical, and there is an increase in inhibition from SACs on the also has a push-pull organization with neurons of opposite
null side and a decrease in inhibition from SACs on the preferred preferred directions making excitatory and inhibitory contacts
side. In the second study, Wei et al. (2011) performed paired onto the dendrite of MST cells such as is realized in fly lobula
recordings from SACs and one subtype of ON/OFF DS ganglion plate tangential cells is not known at present. Interestingly, the
cells. For mice at P3P4, they found that the inhibitory SAC input receptive fields match an optic flow occurring during certain
to the ganglion cells was spatially symmetrical. At P14 or later, types of ego-motion such as translation for various heading
however, the inhibitory input the ganglion cell received from directions in case of MST cells and rotation around various
SACs on the null side was strongly increased, whereas the body axes in case of fly lobula plate tangential cells (Krapp
preferred side inhibition had remained constant. Despite minor et al., 1998). For lobula plate tangential cells, it is known that
deviations, the results from the two studies are largely consis- the network connectivity between the various tangential cells is
tent: Retinal DS circuitry maturation takes place rather quickly responsible for the exact spatial lay-out of the receptive field
before bipolar cells connect and the retina becomes light sensi- (Borst and Weber, 2011; for review see Borst et al., 2010). For
tive. That the inhibitory events measured in ganglion cells do not MST neurons, it is unclear whether the layout of their receptive
change in amplitude but become more frequent (Wei et al., 2011) field is due to dendritic sampling of appropriately oriented local
suggests a change in synapse number rather than strength, motion-sensitive input elements or to a connectivity between
pointing at selective synapse formation and/or elimination that the MST neurons themselves.
facilitates the development of asymmetric processing along
the dendrite. It is likely that molecular cues or marker gradients Multistep and Multilevel Computation of DS
along the different retinal axes are involved in this process (El- A push-pull type of input organization, however, has indeed been
strott and Feller, 2009); however, the identity of these markers found for ON/OFF DS ganglion cells of the retina. As outlined
and the underlying guidance mechanisms are unknown. above, the excitatory as well as inhibitory inputs to these gan-
glion cells are already, to some extent, DS as a result of complex
Species Commonalities and Differences presynaptic interactions. Other circuit features such as the
Given the still unclear situation about the cellular and subcellular spatially offset inhibition, together with particular dendritic pro-
mechanisms of direction selectivity, both in the insect optic lobe cessing, seem to significantly enhance direction selectivity at
and in the vertebrate retina, a comparison is rather challenging. the level of the ganglion cell output, as compared to the input
Further complications arise from the fact that it is not obvious signals driving them. In a similar way, direction selectivity is pro-
which DS cell types and circuitries are functionally equivalent duced in the insect optic lobe as a multistep process (Borst and
between insects and vertebrates. Nevertheless, as premature Egelhaaf, 1990; Single et al., 1997; Joesch et al., 2008). In fly lob-
the situation might be, we will draw some conclusions in the ula plate tangential cells, the spatial layout of the input does not
following based on the data available at present. contribute to the direction selectivity: Each part of the receptive
field can be stimulated separately with moving gratings, and the
ON-OFF Input Rectification cell will respond the same way provided it has the same sensi-
Independent of DS, the most conspicuous commonality tivity in both locations. Interestingly, DS ganglion cells behave
between the visual processing in insects and vertebrates relates in a similar way: With the exception of the nondiscriminating
to the splitting of the photoreceptor input into ON and OFF chan- zone (see Mechanisms at the Ganglion Cell Level), motion
nels (Werblin and Dowling, 1969; Franceschini et al., 1989; restricted to different subsections of the receptive field elicits
Joesch et al., 2010). One obvious function of such a split is to similar DS responses. As to direction selectivity of fly neurons
increase the dynamic range for coding light increments and further upstream in the processing chain, mechanisms similar
decrements. Also, it is an economic way of handling metabolic to the ones in ganglion cells and starburst amacrine cells might
costs: If mainly the changes of brightness levels are to be trans- account for direction selectivity, for example in the dendrites of
mitted to downstream circuits, then without such a split, a T4 or T5 cells.
second order neuron would have to maintain a high constant
level of activity in order to signal both light increments and decre- Development of DS Circuitries
ments. In addition, splitting the visual input signal is beneficial for In higher visual centers, such as the amphibian tectum (Engert
subsequent motion detection circuits: Given the requirement for et al., 2002) or ferret area V1 (Li et al., 2008), visual experience
nonlinear processing from at least two displaced input, dealing is not only necessary but also instructive for the development
exclusively with positive signals alleviates the biophysical imple- of DS (reviewed in Elstrott and Feller, 2009). In contrast, at early
mentation significantly. sensory stages like in the vertebrate retina or the insects lobula
Review
plate, the development of DS circuitries appears to be indepen- Barlow, H.B., and Hill, R.M. (1963). Selective sensitivity to direction of
movement in ganglion cells of the rabbit retina. Science 139, 412414.
dent of visual experience and even activity independent to some
extent. This suggests that the development of DS circuitries is Barlow, H.B., and Levick, W.R. (1965). The mechanism of directionally
largely genetically guided by processes that are still hardly selective units in rabbits retina. J. Physiol. 178, 477504.
understoodan interesting and exciting commonality between Barlow, H.B., Hill, R.M., and Levick, W.R. (1964). Rabbit retinal ganglion cells
vertebrate retina and insect DS pathways and a common chal- responding selectively to direction and speed of image motion in the rabbit.
J. Physiol. 173, 377407.
lenge for future comparative research.
Barth, M., Hirsch, H.V.B., Meinertzhagen, I.A., and Heisenberg, M. (1997).
Experience-dependent developmental plasticity in the optic lobe of Drosophila
CONCLUSION melanogaster. J. Neurosci. 17, 14931504.
The last decade has witnessed much progress in our under- Berson, D.M. (2008). Retinal Ganglion Cell Types and Their Central Projec-
tions. In Vision I, The Senses A Comprehensive Reference, R.H. Masland
standing of the cellular and subcellular mechanisms underlying and T.D. Albright, eds. (San Diego: Academic Press), pp. 491519.
direction selectivity. To a large extent, this is due to the applica-
Borg-Graham, L.J. (2001). The computation of directional selectivity in the
tion of advanced optical as well as genetic methods to this retina occurs presynaptic to the ganglion cell. Nat. Neurosci. 4, 176183.
problem. Optical methods are indispensible whenever different
anatomical compartments of a neuron turn out to be electrically Borg-Graham, L.J., and Grzywacz, N.M. (1992). A model of the directional
selectivity circuit in retina: Transformations by neurons singly and in concert.
separated, operating almost in isolation from the rest of the cell, In Single Neuron Computation, T. McKenna, S.F. Zornetzer, and J.L. Davis,
such as the different dendritic branches of a SAC in the verte- eds. (St Louis, MO: Academic Press), pp. 347375.
brate retina (Euler et al., 2002) and the output terminals versus Borst, A. (2009). Drosophilas view on insect vision. Curr. Biol. 19, R36R47.
the dendrite of lamina cells (Reiff et al., 2010) or the dendrite of
lobula plate tangential cells in the fly (Elyada et al., 2009). Look- Borst, A., and Egelhaaf, M. (1989). Principles of visual motion detection.
Trends Neurosci. 12, 297306.
ing at the corresponding circuits at the ultrastructural level
reveals an intriguing complexity, both within the IPL of the retina Borst, A., and Egelhaaf, M. (1990). Direction selectivity of blowfly motion-
sensitive neurons is computed in a two-stage process. Proc. Natl. Acad.
(Briggman et al., 2011) as well as in the columns of the insect Sci. USA 87, 93639367.
optic lobe (Takemura et al., 2008). The above examples demon-
strate that this complexity has to be taken into account when Borst, A., and Haag, J. (1996). The intrinsic electrophysiological characteristics
of fly lobula plate tangential cells: I. Passive membrane properties. J. Comput.
modeling the corresponding circuits (e.g., Poleg-Polsky and Dia- Neurosci. 3, 313336.
mond, 2011; Schachter et al., 2010; Hausselt et al., 2007).
Borst, A., and Weber, F. (2011). Neural action fields for optic flow based navi-
Another amazing fact is how much effort over so many years gation: A simulation study of the fly lobula plate network. PLoS ONE 6, e16303.
had to be invested in this one single problem of direction selec-
tivity in order to achieve the current level of understanding, Borst, A., Egelhaaf, M., and Haag, J. (1995). Mechanisms of dendritic integra-
tion underlying gain control in fly motion-sensitive interneurons. J. Comput.
a problem that, in terms of computation and information pro- Neurosci. 2, 518.
cessing, seems quite modest (telling leftward from rightward),
Borst, A., Reisenman, C., and Haag, J. (2003). Adaptation of response tran-
compared to the complex intellectual capabilities of humans. sients in fly motion vision. II: Model studies. Vision Res. 43, 13091322.
Our hope is that understanding this simple neural computation
of direction selectivity in full detail will provide an important step- Borst, A., Flanagin, V.L., and Sompolinsky, H. (2005). Adaptation without
parameter change: Dynamic gain control in motion detection. Proc. Natl.
ping stone toward our understanding of more complex functions Acad. Sci. USA 102, 61726176.
of the nervous system.
Borst, A., Haag, J., and Reiff, D.F. (2010). Fly motion vision. Annu. Rev. Neuro-
sci. 33, 4970.
ACKNOWLEDGMENTS
Branco, T., Clark, B.A., and Hausser, M. (2010). Dendritic discrimination of
temporal input sequences in cortical neurons. Science 329, 16711675.
T.E. is supported by the DFG (EXC 307); A.B. is supported by the Max-Planck-
Society and by the DFG (SFB 870). Brecha, N., Johnson, D., Peichl, L., and Wassle, H. (1988). Cholinergic
amacrine cells of the rabbit retina contain glutamate decarboxylase and
gamma-aminobutyrate immunoreactivity. Proc. Natl. Acad. Sci. USA 85,
REFERENCES
61876191.
Ackert, J.M., Farajian, R., Volgyi, B., and Bloomfield, S.A. (2009). GABA Briggman, K.L., and Denk, W. (2006). Towards neural circuit reconstruction
blockade unmasks an OFF response in ON direction selective ganglion cells with volume electron microscopy techniques. Curr. Opin. Neurobiol. 16,
in the mammalian retina. J. Physiol. 587, 44814495. 562570.
Adelson, E.H., and Bergen, J.R. (1985). Spatiotemporal energy models for the Briggman, K.L., Helmstaedter, M., and Denk, W. (2011). Wiring specificity in
perception of motion. J. Opt. Soc. Am. A 2, 284299. the direction-selectivity circuit of the retina. Nature 471, 183188.
Amthor, F.R., Oyster, C.W., and Takahashi, E.S. (1984). Morphology of on-off Brotz, T.M., and Borst, A. (1996). Cholinergic and GABAergic receptors on fly
direction-selective ganglion cells in the rabbit retina. Brain Res. 298, 187190. tangential cells and their role in visual motion detection. J. Neurophysiol. 76,
17861799.
Amthor, F.R., Takahashi, E.S., and Oyster, C.W. (1989). Morphologies of rabbit
retinal ganglion cells with complex receptive fields. J. Comp. Neurol. 280, Brotz, T.M., Gundelfinger, E.D., and Borst, A. (2001). Cholinergic and
97121. GABAergic pathways in fly motion vision. BMC Neurosci. 2, 1.
Amthor, F.R., Keyser, K.T., and Dmitrieva, N.A. (2002). Effects of the destruc- Buchner, E., Buchner, S., and Buelthoff, I. (1984). Deoxyglucose mapping of
tion of starburst-cholinergic amacrine cells by the toxin AF64A on rabbit retinal nervous activity induced in Drosophila brain by visual movement. J. Comp.
directional selectivity. Vis. Neurosci. 19, 495509. Physiol. A. 155, 471483.
Review
Buhl, E.H., and Peichl, L. (1986). Morphology of rabbit retinal ganglion cells Egelhaaf, M., and Reichardt, W. (1987). Dynamic response properties of move-
projecting to the medial terminal nucleus of the accessory optic system. ment detectors: Theoretical analysis and electrophysiological investigation in
J. Comp. Neurol. 253, 163174. the visual system of the fly. Biol. Cybern. 56, 6987.
Cajal, S.R., and Sanchez, D. (1915). Contribucion al Conocimiento de los Egelhaaf, M., Borst, A., and Reichardt, W. (1989). Computational structure of
Centros Nerviosos de los Insectos (Madrid: Imprenta de Hijos de Nicholas a biological motion-detection system as revealed by local detector analysis
Moja). in the flys nervous system. J. Opt. Soc. Am. A 6, 10701087.
Caldwell, J.H., Daw, N.W., and Wyatt, H.J. (1978). Effects of picrotoxin and Egelhaaf, M., Borst, A., and Pilz, B. (1990). The role of GABA in detecting visual
strychnine on rabbit retinal ganglion cells: Lateral interactions for cells with motion. Brain Res. 509, 156160.
more complex receptive fields. J. Physiol. 276, 277298.
Eichner, H., Joesch, M., Schnell, B., Reiff, D.F., and Borst, A. (2011). Internal
Chan, Y.C., and Chiao, C.C. (2008). Effect of visual experience on the matura- structure of the fly elementary motion detector. Neuron 70, 11551164.
tion of ON-OFF direction selective ganglion cells in the rabbit retina. Vision
Res. 48, 24662475. Elstrott, J., and Feller, M.B. (2009). Vision and the establishment of direction-
selectivity: A tale of two circuits. Curr. Opin. Neurobiol. 19, 293297.
Chen, M., Weng, S., Deng, Q., Xu, Z., and He, S. (2009). Physiological proper-
ties of direction-selective ganglion cells in early postnatal and adult mouse Elstrott, J., and Feller, M.B. (2010). Direction-selective ganglion cells show
retina. J. Physiol. 587, 819828. symmetric participation in retinal waves during development. J. Neurosci.
30, 1119711201.
Chiao, C.C., and Masland, R.H. (2002). Starburst cells nondirectionally facili-
tate the responses of direction-selective retinal ganglion cells. J. Neurosci. Elstrott, J., Anishchenko, A., Greschner, M., Sher, A., Litke, A.M., Chichilnisky,
22, 1050910513. E.J., and Feller, M.B. (2008). Direction selectivity in the retina is established
independent of visual experience and cholinergic retinal waves. Neuron 58,
Chiao, C.C., and Masland, R.H. (2003). Contextual tuning of direction-selective 499506.
retinal ganglion cells. Nat. Neurosci. 6, 12511252.
Elyada, Y.M., Haag, J., and Borst, A. (2009). Different receptive fields in axons
Clark, D.A., Bursztyn, L., Horowitz, M.A., Schnitzer, M.J., and Clandinin, T.R. and dendrites underlie robust coding in motion-sensitive neurons. Nat. Neuro-
(2011). Defining the computational structure of the motion detector in sci. 12, 327332.
Drosophila. Neuron 70, 11651177.
Enciso, G.A., Rempe, M., Dmitriev, A.V., Gavrikov, K.E., Terman, D., and
Clifford, C.W., and Ibbotson, M.R. (2002). Fundamental mechanisms of visual Mangel, S.C. (2010). A model of direction selectivity in the starburst amacrine
motion detection: Models, cells and functions. Prog. Neurobiol. 68, 409437. cell network. J. Comput. Neurosci. 28, 567578.
Cohen, E.D. (2001). Voltage-gated calcium and sodium currents of starburst Engert, F., Tao, H.W., Zhang, L.I., and Poo, M.M. (2002). Moving visual stimuli
amacrine cells in the rabbit retina. Vis. Neurosci. 18, 799809. rapidly induce direction sensitivity of developing tectal neurons. Nature 419,
470475.
Cuntz, H., Forstner, F., Haag, J., and Borst, A. (2008). The morphological iden-
tity of insect dendrites. PLoS Comput. Biol. 4, e1000251. Euler, T., and Hausselt, S.E. (2008). Direction Selective Cells. In Vision I, The
Senses A Comprehensive Reference, R.H. Masland and T.D. Albright, eds.
Dacey, D.M. (2004). Origins of perception: Retinal ganglion cell diversity and (San Diego: Academic Press), pp. 413422.
the creation of parallel visual pathways. In The Cognitive Neurosciences,
M.S. Gazzaniga, ed. (Cambridge, MA: MIT), pp. 281301. Euler, T., Detwiler, P.B., and Denk, W. (2002). Directionally selective calcium
signals in dendrites of starburst amacrine cells. Nature 418, 845852.
Dacheux, R.F., Chimento, M.F., and Amthor, F.R. (2003). Synaptic input to the
on-off directionally selective ganglion cell in the rabbit retina. J. Comp. Neurol. Famiglietti, E.V., Jr. (1983). Starburst amacrine cells and cholinergic neurons:
456, 267278. Mirror-symmetric on and off amacrine cells of rabbit retina. Brain Res. 261,
138144.
Damjanovic , I., Maximova, E., and Maximov, V. (2009). Receptive field sizes of
direction-selective units in the fish tectum. J. Integr. Neurosci. 8, 7793. Famiglietti, E.V. (1991). Synaptic organization of starburst amacrine cells in
rabbit retina: Analysis of serial thin sections by electron microscopy and
Daw, N.W., and Wyatt, H.J. (1974). Raising rabbits in a moving visual environ- graphic reconstruction. J. Comp. Neurol. 309, 4070.
ment: An attempt to modify directional sensitivity in the retina. J. Physiol. 240,
309330. Famiglietti, E.V. (1992). Dendritic co-stratification of ON and ON-OFF
directionally selective ganglion cells with starburst amacrine cells in rabbit
Denk, W., and Detwiler, P.B. (1999). Optical recording of light-evoked calcium retina. J. Comp. Neurol. 324, 322335.
signals in the functionally intact retina. Proc. Natl. Acad. Sci. USA 96, 7035
7040. Feller, M.B., Wellis, D.P., Stellwagen, D., Werblin, F.S., and Shatz, C.J. (1996).
Requirement for cholinergic synaptic transmission in the propagation of spon-
Douglass, J.K., and Strausfeld, N.J. (1995). Visual motion detection circuits in taneous retinal waves. Science 272, 11821187.
flies: Peripheral motion computation by identified small-field retinotopic
neurons. J. Neurosci. 15, 55965611. Fennema, C.L., and Thompson, W.B. (1979). Velocity determination in scenes
containing several moving objects. Computer Graphics and Image Processing
Douglass, J.K., and Strausfeld, N.J. (1996). Visual motion-detection circuits in 9, 301315.
flies: Parallel direction- and non-direction-sensitive pathways between the
medulla and lobula plate. J. Neurosci. 16, 45514562. Fischbach, K.F., and Dittrich, A.P.M. (1989). The optic lobe of Drosophila
melanogaster. I. A Golgi analysis of wild-type structure. Cell Tissue Res.
Dror, R.O., OCarroll, D.C., and Laughlin, S.B. (2001). Accuracy of velocity esti- 258, 441475.
mation by Reichardt correlators. J. Opt. Soc. Am. A Opt. Image Sci. Vis. 18,
241252. Fischbach, K.F., Ramos, R.G.P., and Bausenwein, B. (1992). The optic lobe of
Drosophila melanogaster: Experimental approaches to an analysis of struc-
Duffy, C.J., and Wurtz, R.H. (1997). Planar directional contributions to optic ture, function and development. Verh Dtsch Zool Ges 85.2, 133148.
flow responses in MST neurons. J. Neurophysiol. 77, 782796.
Franceschini, N., Riehle, A., and Le Nestour, A. (1989). Directionally selective
Egelhaaf, M., and Borst, A. (1989). Transient and steady-state response motion detection by insect neurons. In Facets of Vision, H. Stavenga, ed.
properties of movement detectors. J. Opt. Soc. Am. A 6, 116127. (Berlin, Heidelberg: Springer), pp. 360390.
Egelhaaf, M., and Borst, A. (1992). Are there separate ON and OFF channels in Fried, S.I., Munch, T.A., and Werblin, F.S. (2002). Mechanisms and circuitry
fly motion vision? Vis. Neurosci. 8, 151164. underlying directional selectivity in the retina. Nature 420, 411414.
Review
Fried, S.I., Munch, T.A., and Werblin, F.S. (2005). Directional selectivity is Joesch, M., Schnell, B., Raghu, S.V., Reiff, D.F., and Borst, A. (2010). ON and
formed at multiple levels by laterally offset inhibition in the rabbit retina. Neuron OFF pathways in Drosophila motion vision. Nature 468, 300304.
46, 117127.
Kanjhan, R., and Sivyer, B. (2010). Two types of ON direction-selective
Gabbiani, F., Krapp, H.G., Koch, C., and Laurent, G. (2002). Multiplicative ganglion cells in rabbit retina. Neurosci. Lett. 483, 105109.
computation in a visual neuron sensitive to looming. Nature 420, 320324.
Karmeier, K., Tabor, R., Egelhaaf, M., and Krapp, H.G. (2001). Early visual
Gavrikov, K.E., Dmitriev, A.V., Keyser, K.T., and Mangel, S.C. (2003). Cation experience and the receptive-field organization of optic flow processing
chloride cotransporters mediate neural computation in the retina. Proc. Natl. interneurons in the fly motion pathway. Vis. Neurosci. 18, 18.
Acad. Sci. USA 100, 1604716052.
Kim, I.J., Zhang, Y., Yamagata, M., Meister, M., and Sanes, J.R. (2008).
Gavrikov, K.E., Nilson, J.E., Dmitriev, A.V., Zucker, C.L., and Mangel, S.C. Molecular identification of a retinal cell type that responds to upward motion.
(2006). Dendritic compartmentalization of chloride cotransporters underlies Nature 452, 478482.
directional responses of starburst amacrine cells in retina. Proc. Natl. Acad.
Sci. USA 103, 1879318798. Koch, C., Poggio, T., and Torre, V. (1983). Nonlinear interactions in a dendritic
tree: Localization, timing, and role in information processing. Proc. Natl. Acad.
Grzywacz, N.M., and Amthor, F.R. (2007). Robust directional computation in Sci. USA 80, 27992802.
on-off directionally selective ganglion cells of rabbit retina. Vis. Neurosci. 24,
647661. Krapp, H.G., and Hengstenberg, R. (1996). Estimation of self-motion by optic
flow processing in single visual interneurons. Nature 384, 463466.
Grzywacz, N.M., Amthor, F.R., and Merwine, D.K. (1994). Directional hyper-
acuity in ganglion cells of the rabbit retina. Vis. Neurosci. 11, 10191025. Krapp, H.G., Hengstenberg, B., and Hengstenberg, R. (1998). Dendritic
structure and receptive-field organization of optic flow processing interneu-
Grzywacz, N.M., Amthor, F.R., and Merwine, D.K. (1998). Necessity of acetyl- rons in the fly. J. Neurophysiol. 79, 19021917.
choline for retinal directionally selective responses to drifting gratings in rabbit.
J. Physiol. 512, 575581. Lee, S., and Zhou, Z.J. (2006). The synaptic mechanism of direction selectivity
in distal processes of starburst amacrine cells. Neuron 51, 787799.
Guo, A., and Reichardt, W. (1987). An estimation of the time constant of move-
ment detectors. Naturwissenschaften 74, 9192. Lee, S., Kim, K., and Zhou, Z.J. (2010). Role of ACh-GABA cotransmission in
detecting image motion and motion direction. Neuron 68, 11591172.
Haag, J., Theunissen, F., and Borst, A. (1997). The intrinsic electrophysiolog-
ical characteristics of fly lobula plate tangential cells: II. Active membrane Li, Y., Van Hooser, S.D., Mazurek, M., White, L.E., and Fitzpatrick, D. (2008).
properties. J. Comput. Neurosci. 4, 349369. Experience with moving visual stimuli drives the early development of cortical
direction selectivity. Nature 456, 952956.
Haag, J., Vermeulen, A., and Borst, A. (1999). The intrinsic electrophysiological
characteristics of fly lobula plate tangential cells: III. Visual response proper- Limb, J.O., and Murphy, J.A. (1975). Estimating the velocity of moving images
ties. J. Comput. Neurosci. 7, 213234. in television signals. Computer Graphics and Image Processing 4, 311327.
Haag, J., Denk, W., and Borst, A. (2004). Fly motion vision is based on Reich- Maimon, G., Straw, A.D., and Dickinson, M.H. (2010). Active flight increases
ardt detectors regardless of the signal-to-noise ratio. Proc. Natl. Acad. Sci. the gain of visual motion processing in Drosophila. Nat. Neurosci. 13, 393399.
USA 101, 1633316338.
Marchiafava, P.L. (1979). The responses of retinal ganglion cells to stationary
Hassenstein, B., and Reichardt, W. (1956). Systemtheoretische Analyse der and moving visual stimuli. Vision Res. 19, 12031211.
Zeit-, Reihenfolgen- und Vorzeichenauswertung bei der Bewegungsperzep-
tion des Riisselkiifers Chlorophanus. Z Naturforsch IIb, 513524. Masland, R.H. (1977). Maturation of function in the developing rabbit retina.
J. Comp. Neurol. 175, 275286.
Hausen, K. (1982a). Motion sensitive interneurons in the optomotor system of
the fly. I. The horizontal cells: Structure and signals. Biol. Cybern. 45, 143156. Masland, R.H. (2004). Direction Selectivity in Retinal Ganglion Cells. In The
Visual Neurosciences, L.M. Chalupa and J.S. Werner, eds. (Cambridge, MA:
Hausen, K. (1982b). Motion sensitive interneurons in the optomotor system of The MIT Press), pp. 451462.
the fly. II. The horizontal cells: Receptive field organization and response char-
acteristics. Biol. Cybern. 46, 6779. Masland, R.H. (2005). The many roles of starburst amacrine cells. Trends
Neurosci. 28, 395396.
Hausselt, S.E., Euler, T., Detwiler, P.B., and Denk, W. (2007). A dendrite-auton-
omous mechanism for direction selectivity in retinal starburst amacrine cells. Masland, R.H., and Mills, J.W. (1979). Autoradiographic identification of
PLoS Biol. 5, e185. acetylcholine in the rabbit retina. J. Cell Biol. 83, 159178.
He, S., and Masland, R.H. (1997). Retinal direction selectivity after targeted Masland, R.H., Mills, J.W., and Cassidy, C. (1984a). The functions of acetyl-
laser ablation of starburst amacrine cells. Nature 389, 378382. choline in the rabbit retina. Proc. R. Soc. Lond. B Biol. Sci. 223, 121139.
He, S., and Masland, R.H. (1998). ON direction-selective ganglion cells in the Masland, R.H., Mills, J.W., and Hayden, S.A. (1984b). Acetylcholine-synthe-
rabbit retina: Dendritic morphology and pattern of fasciculation. Vis. Neurosci. sizing amacrine cells: Identification and selective staining by using radioauto-
15, 369375. graphy and fluorescent markers. Proc. R. Soc. Lond. B Biol. Sci. 223, 79100.
Hengstenberg, R. (1982). Common visual response properties of giant vertical Masseck, O.A., and Hoffmann, K.P. (2008). Responses to moving visual stimuli
cells in the lobula plate of the blowfly Calliphora. J. Comp. Physiol. A Neuroe- in pretectal neurons of the small-spotted dogfish (Scyliorhinus canicula).
thol. Sens. Neural Behav. Physiol. 149, 179193. J. Neurophysiol. 99, 200207.
Hengstenberg, R., Hausen, K., and Hengstenberg, B. (1982). The number and Massey, S.C., Linn, D.M., Kittila, C.A., and Mirza, W. (1997). Contributions
structure of giant vertical cells (VS) in the lobula plate of the blowfly Calliphora of GABAA receptors and GABAC receptors to acetylcholine release and
erytrocephala. J. Comp. Physiol. A Neuroethol. Sens. Neural Behav. Physiol. directional selectivity in the rabbit retina. Vis. Neurosci. 14, 939948.
149, 163177.
Merwine, D.K., Grzywacz, N.M., Tjepkes, D.S., and Amthor, F.R. (1998).
Jeon, C.J., Kong, J.H., Strettoi, E., Rockhill, R., Stasheff, S.F., and Masland, Non-monotonic contrast behavior in directionally selective ganglion cells
R.H. (2002). Pattern of synaptic excitation and inhibition upon direction-selec- and evidence for its dependence on their GABAergic input. Vis. Neurosci.
tive retinal ganglion cells. J. Comp. Neurol. 449, 195205. 15, 11291136.
Joesch, M., Plett, J., Borst, A., and Reiff, D.F. (2008). Response properties Millar, T.J., and Morgan, I.G. (1987). Cholinergic amacrine cells in the rabbit
of motion-sensitive visual interneurons in the lobula plate of Drosophila retina synapse onto other cholinergic amacrine cells. Neurosci. Lett. 74,
melanogaster. Curr. Biol. 18, 368374. 281285.
Review
Miller, R.F., and Bloomfield, S.A. (1983). Electroanatomy of a unique amacrine Reiff, D.F., Plett, J., Mank, M., Griesbeck, O., and Borst, A. (2010). Visualizing
cell in the rabbit retina. Proc. Natl. Acad. Sci. USA 80, 30693073. retinotopic half-wave rectified input to the motion detection circuitry of
Drosophila. Nat. Neurosci. 13, 973978.
Morante, J., and Desplan, C. (2008). The color-vision circuit in the medulla of
Drosophila. Curr. Biol. 18, 553565. Reisenman, C., Haag, J., and Borst, A. (2003). Adaptation of response
transients in fly motion vision. I: Experiments. Vision Res. 43, 12911307.
Munch, T.A., and Werblin, F.S. (2006). Symmetric interactions within a homo-
geneous starburst cell network can lead to robust asymmetries in dendrites of Rodieck, R.W. (1989). Starburst amacrine cells of the primate retina. J. Comp.
starburst amacrine cells. J. Neurophysiol. 96, 471477. Neurol. 285, 1837.
Nagel, G., Szellas, T., Huhn, W., Kateriya, S., Adeishvili, N., Berthold, P., Ollig, Rybak, J., and Meinertzhagen, I.A. (1997). The effects of light reversals on
D., Hegemann, P., and Bamberg, E. (2003). Channelrhodopsin-2, a directly photoreceptor synaptogenesis in the fly Musca domestica. Eur. J. Neurosci.
light-gated cation-selective membrane channel. Proc. Natl. Acad. Sci. USA 9, 319333.
100, 1394013945.
Safran, M.N., Flanagin, V.L., Borst, A., and Sompolinsky, H. (2007). Adaptation
Neal, M.J., and Cunningham, J.R. (1995). Baclofen enhancement of acetylcho- and information transmission in fly motion detection. J. Neurophysiol. 98,
line release from amacrine cells in the rabbit retina by reduction of glycinergic 33093320.
inhibition. J. Physiol. 482, 363372.
Schachter, M.J., Oesch, N., Smith, R.G., and Taylor, W.R. (2010). Dendritic
Nomura, A., Shigemoto, R., Nakamura, Y., Okamoto, N., Mizuno, N., and
spikes amplify the synaptic signal to enhance detection of motion in a simula-
Nakanishi, S. (1994). Developmentally regulated postsynaptic localization of
tion of the direction-selective ganglion cell. PLoS Comput. Biol. 6, e1000899.
a metabotropic glutamate receptor in rat rod bipolar cells. Cell 77, 361369.
OBrien, B.J., Caldwell, J.H., Ehring, G.R., Bumsted OBrien, K.M., Luo, S., and Schnell, B., Joesch, M., Forstner, F., Raghu, S.V., Otsuna, H., Ito, K., Borst, A.,
Levinson, S.R. (2008). Tetrodotoxin-resistant voltage-gated sodium channels and Reiff, D.F. (2010). Processing of horizontal optic flow in three visual inter-
Na(v)1.8 and Na(v)1.9 are expressed in the retina. J. Comp. Neurol. 508, neurons of the Drosophila brain. J. Neurophysiol. 103, 16461657.
940951.
Scott, E.K., Reuter, J.E., and Luo, L. (2003). Dendritic development of
OMalley, D.M., Sandell, J.H., and Masland, R.H. (1992). Co-release of Drosophila high order visual system neurons is independent of sensory
acetylcholine and GABA by the starburst amacrine cells. J. Neurosci. 12, experience. BMC Neurosci. 4, 14.
13941408.
Simpson, J.I., Leonard, C.S., and Soodak, R.E. (1988a). The accessory optic
Oesch, N.W., and Taylor, W.R. (2010). Tetrodotoxin-resistant sodium channels system of rabbit. II. Spatial organization of direction selectivity. J. Neurophy-
contribute to directional responses in starburst amacrine cells. PLoS ONE 5, siol. 60, 20552072.
e12447.
Simpson, J.I., Leonard, C.S., and Soodak, R.E. (1988b). The accessory optic
Oesch, N., Euler, T., and Taylor, W.R. (2005). Direction-selective dendritic system. Analyzer of self-motion. Ann. N Y Acad. Sci. 545, 170179.
action potentials in rabbit retina. Neuron 47, 739750.
Single, S., and Borst, A. (1998). Dendritic integration and its role in computing
Oyster, C.W., and Barlow, H.B. (1967). Direction-selective units in rabbit retina: image velocity. Science 281, 18481850.
Distribution of preferred directions. Science 155, 841842.
Single, S., Haag, J., and Borst, A. (1997). Dendritic computation of direction
Oyster, C.W., Takahashi, E., and Collewijn, H. (1972). Direction-selective selectivity and gain control in visual interneurons. J. Neurosci. 17, 60236030.
retinal ganglion cells and control of optokinetic nystagmus in the rabbit. Vision
Res. 12, 183193. Sivyer, B., van Wyk, M., Vaney, D.I., and Taylor, W.R. (2010). Synaptic inputs
and timing underlying the velocity tuning of direction-selective ganglion cells in
Oyster, C.W., Amthor, F.R., and Takahashi, E.S. (1993). Dendritic architecture rabbit retina. J. Physiol. 588, 32433253.
of ON-OFF direction-selective ganglion cells in the rabbit retina. Vision Res.
33, 579608. Spavieri, D.L., Jr., Eichner, H., and Borst, A. (2010). Coding efficiency of fly
motion processing is set by firing rate, not firing precision. PLoS Comput.
Pearlman, A.L., and Hughes, C.P. (1976). Functional role of efferents to the Biol. 6, e1000860.
avian retina. I. Analysis of retinal ganglion cell receptive fields. J. Comp. Neu-
rol. 166, 111122. Srinivasan, M.V. (1990). Generalized gradient schemes for the measurement of
two-dimensional image motion. Biol. Cybern. 63, 421431.
Peters, B.N., and Masland, R.H. (1996). Responses to light of starburst
amacrine cells. J. Neurophysiol. 75, 469480. Srinivasan, M.V., and Bernard, G.D. (1976). A proposed mechanism for
multiplication of neural signals. Biol. Cybern. 21, 227236.
Poleg-Polsky, A., and Diamond, J.S. (2011). Imperfect space clamp permits
electrotonic interactions between inhibitory and excitatory synaptic conduc- Stacy, R.C., and Wong, R.O.L. (2003). Developmental relationship between
tances, distorting voltage clamp recordings. PLoS ONE 6, e19463. cholinergic amacrine cell processes and ganglion cell dendrites of the mouse
retina. J. Comp. Neurol. 456, 154166.
Raghu, S.V., and Borst, A. (2011). Candidate glutamatergic neurons in the
visual system of Drosophila. PLoS ONE 6, e19472.
Strausfeld, N.J., and Lee, J.K. (1991). Neuronal basis for parallel visual pro-
Raghu, S.V., Joesch, M., Borst, A., and Reiff, D.F. (2007). Synaptic organiza- cessing in the fly. Vis. Neurosci. 7, 1333.
tion of lobula plate tangential cells in Drosophila: GABA-receptors and chem-
ical release sites. J. Comp. Neurol. 502, 598610. Sun, W., Deng, Q., Levick, W.R., and He, S. (2006). ON direction-selective
ganglion cells in the mouse retina. J. Physiol. 576, 197202.
Raghu, S.V., Joesch, M., Sigrist, S.J., Borst, A., and Reiff, D.F. (2009). Synaptic
organization of lobula plate tangential cells in Drosophila: Dalpha7 cholinergic Takemura, S.Y., Lu, Z., and Meinertzhagen, I.A. (2008). Synaptic circuits of the
receptors. J. Neurogenet. 23, 200209. Drosophila optic lobe: The input terminals to the medulla. J. Comp. Neurol.
509, 493513.
Raghu, S.V., Reiff, D.F., and Borst, A. (2011). Neurons with cholinergic pheno-
type in the visual system of Drosophila. J. Comp. Neurol. 519, 162176. Tauchi, M., and Masland, R.H. (1984). The shape and arrangement of the
cholinergic neurons in the rabbit retina. Proc. R. Soc. Lond. B Biol. Sci. 223,
Reichardt, W. (1961). Autocorrelation, a principle for the evaluation of sensory 101119.
information by the central nervous system. In Sensory Communication, W.A.
Rosenblith, ed. (New York, London: MlT Press and Wiley), pp. 303317. Taylor, W.R., and Vaney, D.I. (2002). Diverse synaptic mechanisms generate
direction selectivity in the rabbit retina. J. Neurosci. 22, 77127720.
Reichardt, W. (1987). Evaluation of optical motion information by movement
detectors. J. Comp. Physiol. A Neuroethol. Sens. Neural Behav. Physiol. Taylor, W.R., and Wassle, H. (1995). Receptive field properties of starburst
161, 533547. cholinergic amacrine cells in the rabbit retina. Eur. J. Neurosci. 7, 23082321.
Review
Telkes, I., Distler, C., and Hoffmann, K.P. (2000). Retinal ganglion cells projec- Velte, T.J., and Miller, R.F. (1997). Spiking and nonspiking models of starburst
ting to the nucleus of the optic tract and the dorsal terminal nucleus of the amacrine cells in the rabbit retina. Vis. Neurosci. 14, 10731088.
accessory optic system in macaque monkeys. Eur. J. Neurosci. 12, 2367
2375. Wassle, H. (2004). Parallel processing in the mammalian retina. Nat. Rev. Neu-
rosci. 5, 747757.
Torre, V., and Poggio, T. (1978). A synaptic mechanism possibly underlying
directional selectivity to motion. Proc. R. Soc. Lond. B Biol. Sci. 202, 409416. Wei, W., Hamby, A.M., Zhou, K., and Feller, M.B. (2011). Development of
asymmetric inhibition underlying direction selectivity in the retina. Nature
Tukker, J.J., Taylor, W.R., and Smith, R.G. (2004). Direction selectivity in 469, 402406.
a model of the starburst amacrine cell. Vis. Neurosci. 21, 611625.
Weng, S., Sun, W., and He, S. (2005). Identification of ON-OFF direction-selec-
Tuthill, J.C., Chiappe, M.E., and Reiser, M.B. (2011). Neural correlates of tive ganglion cells in the mouse retina. J. Physiol. 562, 915923.
illusory motion perception in Drosophila. Proc. Natl. Acad. Sci. USA 108,
96859690. Werblin, F.S., and Dowling, J.E. (1969). Organization of the retina of the mud-
puppy, Necturus maculosus. II. Intracellular recording. J. Neurophysiol. 32,
Uchiyama, H., Kanaya, T., and Sonohata, S. (2000). Computation of motion 339355.
direction by quail retinal ganglion cells that have a nonconcentric receptive
field. Vis. Neurosci. 17, 263271. Wilson, R.I., Turner, G.C., and Laurent, G. (2004). Transformation of olfactory
representations in the Drosophila antennal lobe. Science 303, 366370.
van Santen, J.P.H., and Sperling, G. (1985). Elaborated Reichardt detectors.
Wong, R.O.L. (1990). Differential growth and remodelling of ganglion cell
J. Opt. Soc. Am. A 2, 300321.
dendrites in the postnatal rabbit retina. J. Comp. Neurol. 294, 109132.
Vaney, D.I. (1984). Coronate amacrine cells in the rabbit retina have the Wong, R.O., and Collin, S.P. (1989). Dendritic maturation of displaced putative
starburst dendritic morphology. Proc. R. Soc. Lond. B Biol. Sci. 220, cholinergic amacrine cells in the rabbit retina. J. Comp. Neurol. 287, 164178.
501508.
Wyatt, H.J., and Daw, N.W. (1975). Directionally sensitive ganglion cells in the
Vaney, D.I. (1990). The mosaic of amacrine cells in the mammalian retina. Prog. rabbit retina: Specificity for stimulus direction, size, and speed. J. Neurophy-
Retin. Eye Res. 9, 49100. siol. 38, 613626.
Vaney, D.I. (1994). Territorial organization of direction-selective ganglion cells Yamada, E.S., Bordt, A.S., and Marshak, D.W. (2005). Wide-field ganglion cells
in rabbit retina. J. Neurosci. 14, 63016316. in macaque retinas. Vis. Neurosci. 22, 383393.
Vaney, D.I., and Young, H.M. (1988). GABA-like immunoreactivity in cholin- Yonehara, K., Ishikane, H., Sakuta, H., Shintani, T., Nakamura-Yonehara, K.,
ergic amacrine cells of the rabbit retina. Brain Res. 438, 369373. Kamiji, N.L., Usui, S., and Noda, M. (2009). Identification of retinal ganglion
cells and their projections involved in central transmission of information about
Vaney, D.I., Collins, S.P., and Young, H.M. (1989). Dendritic relationships upward and downward image motion. PLoS ONE 4, e4320.
netween cholinergic amacrine cells and direction selective retinal ganglion
cells. In Neurobiology of the Inner Retina, R. Weiler and N.N. Osborn, eds. (Ber- Yoshida, K., Watanabe, D., Ishikane, H., Tachibana, M., Pastan, I., and
lin: Springer), pp. 157168. Nakanishi, S. (2001). A key role of starburst amacrine cells in originating retinal
directional selectivity and optokinetic eye movement. Neuron 30, 771780.
Vaney, D.I., He, S., Taylor, W.R., and Levick, W.R. (2001). Direction-Selective
Ganglion Cells in the Retina. In Motion Vision - Computational, Neural, and Zheng, J.J., Lee, S., and Zhou, Z.J. (2004). A developmental switch in the
Ecological Constraints, J.M. Zanker and J. Zeil, eds. (Berlin, Heidelberg, excitability and function of the starburst network in the mammalian retina.
New York: Springer), pp. 1356. Neuron 44, 851864.
536
t e c h n ic a l r e p o r t s
to apply during navigation. Here we describe methods for imaging address these and other questions by providing a full spatio-temporal
the activity of neurons in the CA1 region of the hippocampus with view of neuronal population activity within a microcircuit, by record-
subcellular resolution in behaving mice. Neurons that expressed ing from genetically labeled neuronal subtypes, and by monitoring
the genetically encoded calcium indicator GCaMP3 were imaged subcellular compartments such as dendritic branches and spines.
through a chronic hippocampal window. Head-restrained mice However, it is not currently possible to perform high-resolution func-
performed spatial behaviors in a setup combining a virtual tional imaging of neurons in the hippocampi of awake, navigating mice.
reality system and a custom-built two-photon microscope. To achieve this goal, three obstacles must be overcome: cellular reso-
We optically identified populations of place cells and determined lution imaging in the brain of a mobile mouse; imaging more than a
the correlation between the location of their place fields in the millimeter beneath the cortical surface; and imaging that is compatible
virtual environment and their anatomical location in the local with navigation behavior. Two-photon microscopy10 in combination
circuit. The combination of virtual reality and high-resolution with calcium-sensitive dyes has been used to provide cellular resolution
functional imaging should allow a new generation of studies to functional imaging in awake mobile mice on a spherical treadmill11,12.
investigate neuronal circuit dynamics during behavior. Chronic functional imaging in this preparation has been made possible
using the genetically encoded calcium indicator GCaMP313. In an acute
The location-specific firing of hippocampal place cells1 during anesthetized mouse preparation, non-activity dependent fluorescent
navigation represents a salient neural correlate of spatial informa- probes have been used for subcellular imaging of the hippocampus14. In
tion in the mammalian brain. More generally, place cells and other addition, a virtual reality system has been developed that allows spatial
hippocampal neurons are thought to be involved in the encoding of behaviors to be studied in head-restrained mice15.
episodic memories2, making the behaving animal the most appro- Here, we combined and extended these methods to make possible
priate setting in which to study their activity patterns. The firing high-resolution functional imaging of the hippocampus in mice navi-
properties of hippocampal neurons in behaving rodents have been gating in a virtual reality environment. We developed a preparation for
characterized using extracellular electrode techniques in numerous chronic imaging of CA1 neuronal activity with cellular and subcellular
experimental paradigms3,4. However, electrophysiological methods resolution in awake mice with their heads restrained while they ran on
have limitations in characterizing precise spatio-temporal activity a spherical treadmill. This was combined with a custom two-photon
patterns, targeting genetically labeled subpopulations and monitor- microscope and light-blocking methods designed for imaging in the
ing subcellular compartments. presence of the visual display surrounding the treadmill that defined
For example, it is unclear whether hippocampal neurons with a virtual environment. Using these methods, we optically recorded
similar place fields are spatially organized within the hippocam- from populations of ~80100 GCaMP3-expressing neurons in the
pus. Although some studies have reported organization on spatial CA1 region of the hippocampus while the mouse was running along
scales ranging from tens of micrometers5 to millimeters6,7, a separate a linear track. Subpopulations of the recorded neurons were identi-
study found no organization on either of these scales8. This question fied as place cells. Because our methods could precisely determine the
has been difficult to address because extracellular electrodes can- physical location of the place cells in the hippocampus, we could show
not report the precise anatomical location of the recorded cells69. that for cells separated by more than a few tens of micrometers, there
Indirect methods, such as immediate early gene (IEG) expression5,8, was no strong relationship in our experimental paradigm between
are also problematic because the relationship between gene expression the location of place fields in the virtual reality environment and the
and neuronal spiking is not well established. Tetrode recordings have position of the corresponding place cells in area CA1. Nearby cells
1Department of Molecular Biology and Princeton Neuroscience Institute, Princeton University, Princeton, New Jersey, USA. 2Howard Hughes Medical Institute, Janelia
Farm Research Campus, Ashburn, Virginia, USA. Correspondence should be addressed to D.W.T. (dwtank@princeton.edu) or D.A.D. (ddombeck@princeton.edu).
(less than <35 m separation) showed enhanced correlation, although A window for chronic imaging of CA1 neurons in awake mice
the possibility that this was produced by mixing of optical signals The CA1 region of the hippocampus is more than a millimeter below
from adjacent cells could not be excluded. In addition, we show that the cortical surface and cannot be directly imaged using two-photon
it is possible to record place-related activity in putative hippocampal microscopy17. We carefully removed the overlying cortex by aspira-
interneurons and CA1 apical dendrites. tion (down to the external capsule) and replaced it with a metal can-
nula with a coverslip sealing one of the openings (see Online Methods,
RESULTS Fig. 1e). This created a chronic hippocampal window that allowed
Combining two-photon microscopy and mouse virtual reality direct imaging of the hippocampus.
Our apparatus (Fig. 1a,b) was designed around our previously We used genetically encoded calcium indicators to optically record the
described virtual reality system15. The limbs of a head-restrained activity of CA1 neurons. Hippocampal injections of adeno-associated
mouse rested on a spherical treadmill12. A toroidal screen that sub- virus AAV2/1-synapsin-1-GCaMP3 (a moderate, neuron-specific pro-
tended a mouses visual field surrounded the treadmill and displayed moter18 in a serotype with efficient gene delivery to neurons19) resulted
a computer-generated image of a virtual environment15,16. Ball move- in expression of GCaMP3 in a large population of CA1 neurons. Two-
ments recorded with an optical computer mouse provided information photon imaging of the GCaMP3-expressing area at a shallow depth
on running speed and direction and this was used by the computer through the hippocampal window showed axons in the external capsule
program that implemented the virtual environment to update posi- and alveus that appeared as a dense plexus of fibers running parallel to
tion and view angle. As reported previously15, head-restrained mice the hippocampal surface (Fig. 1f). Inspection of images from the stra-
were trained using operant conditioning to run back and forth along tum pyramidale (~130160 m below the external capsule) revealed the
a 180 cmlong virtual linear track (Fig. 1c). The mice received water densely packed neurons of this cell body layer. We did not see voids of
rewards at the ends of the track after successfully traversing the full fluorescence indicative of unlabeled cells, suggesting that local labeling
2010 Nature America, Inc. All rights reserved.
track length; consecutive rewards at the same end were not available. was nearly ubiquitous (Fig. 1f; note that GCaMP3 is excluded from the
We designed and constructed a two-photon microscope that could fit nucleus, which causes a donut-like appearance of the cell bodies13). All
within the geometric constraints of our virtual reality apparatus with- pyramidal neuron cell bodies in the stratum pyramidale have a promi-
out obstructing the mouses view of the display (Fig. 1a). The physical nent apical dendrite that projects ventrally into the stratum radiatum
dimensions of the design were most severely constrained by the small (deeper into the hippocampus). Imaging perpendicular to the long axis
distance (~13 cm) between the top of the headplate on the mouses head of these dendrites in the stratum radiatum showed their cross-sections
and the bottom of the reflecting mirror in the virtual reality projec- as small dots (Fig. 1f; see Supplementary Movie 1 for a z-series through
tion path. In addition, we designed the microscope to be completely the labeled CA1 region).
shielded from the bright light of the virtual reality projection display so
that the smaller number of photons from the fluorescent probe could be Optical identification of CA1 place cells
detected by the photomultiplier tubes (PMTs) without contamination. We developed a timeline for the sequence of steps necessary to image
We then implemented additional light-blocking measures at the laser hippocampal activity in mice trained to perform the spatial behavior
input port and the hole for the microscope
objective (see Online Methods; Fig. 1a,d,e) so
that the amount of detected background light
a Projector b
AAM
was less than ~5% of the baseline fluorescence L BP
RM FL
level from labeled cells. M PMT
CL M
DM
Axis of
Figure 1 Experimental setup. (a) The Z translation
rotation
experimental apparatus, consisting of a spherical Rubber TL
treadmill, a virtual reality apparatus (projector, tube
reflecting mirror (RM), angular amplification
Toroidal SL
c
mirror (AAM), toroidal screen and optical screen
computer mouse to record ball rotation) and Optical X-Y
computer
a custom two-photon microscope (titanium: mouse Ti:S
sapphire laser (Ti:S), long-pass filter (LP),
X-Y translation View from left end of
galvanometers (X-Y), scan lens (SL), mirror (M), Air LP virtual linear track
tube lens (TL), dichroic mirror (DM), collection
Sliding Top view of virtual linear track
lens (CL), biconcave lens (L), bandpass filter stage
Microscope
(BP), focusing lens (FL), photomultiplier tube
(PMT), sliding stage (used to move microscope d e Opaque
Kwik-Sil
Stainless
180 cm
Skull meta-bond
for treadmill access), X-Y translation (moves cannula
f Axons
in the
treadmill and mouse), Z-translation (objective Cortex Glass
Kwik-Sil external
focus control) and rubber tube (shown in cross- capsule/
section, for light shielding)). (b) Photograph alveus
of experimental setup. (c) Top, view from one
end of the virtual linear track. Bottom, top view Rubber
tube CA1 cell
of the linear track. (d) View of materials used bodies in
stratum
to block background light from entering the Outer metal
pyramidale
ring
microscope objective hole. Hippocampal imaging Inner metal
window can also be seen. (e) Detailed view of ring
CA1 apical
hippocampal imaging window (from boxed region dendrites in
in d). (f) In vivo two-photon images at different Headplate stratum
40 m radiatum
depths through the hippocampal window.
Cell number
position for a subset of the cells labeled in a 6
(right). (e) A plot of mean F/F versus linear
track position for all of the cells labeled in a 3
(right). (f) Place cells are colored according 50%
to the location of their place fields along F/F
2
the virtual linear track. Only place cells with 180 cm
significant place fields during running in the Linear
track
positive direction are shown.
position
2010 Nature America, Inc. All rights reserved.
0 cm
Rewards
10 s 2s
task. Mice were first implanted with a metal
headplate to allow their heads to be restrained
d f 180 cm
9 2
while they were on the spherical treadmill. 1
They were placed on water scheduling for sev- 5
Cell number
6 Position of place
eral days and then trained in the virtual real- 30% mean 8
10 fields along linear
F/F track
ity apparatus. Once the mice were proficient 3 7 9
60 cm 6
at the task (~2 rewards per minute, ~710 d
3
of training) we injected the GCaMP3 virus, 2 4
0 cm
and the next day we implanted the hippo 0 cm 180 cm
Normalized
signal-to-noise ratio for imaging calcium mean F/F
transients. Using the hippocampal imaging
window and viral delivery of GCaMP3, we
could image activity in CA1 neurons repeat- 1
edly over the course of about 34 weeks. 0 cm 180 cm
0
Position of place
13
9 11
fields along
linear track
14
17
15
c Across our 47 time-series datasets, we
16 2
6 350 identified 808 place cells and 881 place fields
Normalized
mean F/F
0
10 10 0 0.5 1.0
Directionality index with two place fields was 73; 18 6 complete
runs through each place field). The calcium
1 1 0 transients that defined these place fields had
0 cm 180 cm 0 cm 180 cm
a mean duration of 1.1 0.9 s (Fig. 4a) and a
Position along linear track Position along linear track
mean peak F/F of 35 29% (Fig. 4b). The
Figure 3 Place cells differ depending on the running direction in the linear track. (a) An example distribution of place fields as a function of
imaging field in which the place cells are colored according to the location of their place fields
position along the track was inhomogeneous
along the virtual linear track. Significant place fields during running in the positive (left panel)
and negative (right panel) directions are shown. Example place cells with different place fields (Fig. 4c), as seen previously31, with more cen-
or no place fields depending on the running direction are highlighted with closed arrowheads or tered between the reward sites and the towers
open arrowheads, respectively. (b) A plot of mean F/F versus linear track position for the positive (tall distal cues located ~60 cm from each end
2010 Nature America, Inc. All rights reserved.
direction place cells labeled in a (left) during running in the positive (left) and negative (right) of the track; see Fig. 1c) than in the middle
directions. (c) Histogram of directionality index for all place fields. of the track. The mean width of the optically
identified place fields was 50 19 cm (Fig. 4d),
and statistically significant fields of neuronal activity (P < 0.05 from which is ~20% larger than the mean width of place fields defined by
bootstrapping, see Online Methods). The spatially restricted activity firing rate in extracellular recordings under similar conditions15 (41
of these neurons and the well-defined shape of the activity fields were 14 cm). To investigate this broadening, we convolved spike train elec-
similar to extracellular recording measurements of place cells in real trophysiology recordings with model calcium transient waveforms
and virtual reality linear tracks15,26,27. Therefore, we concluded that (see Online Methods) and observed a similar increase in the resultant
these neurons are place cells. As well as identifying the place field place field width (~20%; see Supplementary Fig. 3). This result is
for each of the place cells, imaging also reveals their exact relative consistent with the idea that our place fields are defined by underlying
anatomical locations (Fig. 2f and Supplementary Movie 2). spiking activity that is convolved with the time-broadening effects of
Consistent with previous studies15,26, optically identified place cells intracellular calcium dynamics (Supplementary Fig. 3).
had different place fields in the same environment depending on the When optically identified place cells were active, the activity most
direction of running (Figs. 2b and 3). To analyze the directionality of frequently occurred in the place field; however, the place cells are not
activity during place field traversals, we divided the fluorescence time necessarily active during every traversal of the place field. This is similar
series into epochs defined by running direction. The mean fluorescence to previous tetrode study reports of excess activity variance28 in place
versus track position was calculated separately for each direction. A cell cells. Figure 5a shows the activity trace (temporal activity pattern) for
that had a place field during at least one direction of running was called each individual traversal of the mouse in the positive running direction
a place cell; each place cell could therefore have up to two place fields. for a subset of the place cells shown in Figure 2. Although place cells
Many place cells had different place fields depending on the running 2, 3, 6 and 9 are reliably active at the same location during nearly every
direction (Fig. 3a) or had a place field for only one running direction traversal (active during >75% of traversals through the place field and
(Fig. 3a). This high degree of directionality (Fig. 3b) was consistent active for >66% of the time spent in the place field), place cell 10 shows
across all datasets and was quantified for all place fields using a direc- more variable activity. This cell is reliably active at the same location, but
tionality index (0.83 0.25, Fig. 3c; 1 represents high directionality, not during each traversal (active during 38% of traversals through the
0 represents no directionality; see Online Methods for definition). place field and active for 35% of the time spent in the place field). For
Although we analyzed only high reward rate periods in the Results, all of the 881 place fields, the place cells were active during 65 18% of
we also analyzed place fields during different behavioral contexts2830 traversals through the place field (Fig. 5) and active for 53 17% of the
including the low reward rate periods (see Supplementary Fig. 1). time spent traversing through the place field (Fig. 5).
a b c d Minimum
3,000 3,000 100 70 width
Number of place fields
Number of place fields
Number of calcium
Number of calcium
50
transients
2,000 Mean = 50 19 cm
transients
Figure 4 Characterization of place cell calcium transients and place fields. Histograms of place cell transient widths (a) and transient peak F/F (b) are
shown for periods of mouse movement along the virtual track. Histograms of place field position along the linear track (c) and place field widths (d) are
also shown.
Figure 5 Place cell activity variability in place fields. (a) Temporal activity
pattern versus virtual linear track position traces for a subset of the cells
a Cell 2 Cell 3 Cell 6 Cell 9 Cell 10
F/F
slightly less than, the percentage in our whole-cell patch clamp record- 0 cm 180 cm
a 180 cm
b c
fields along linear
0.16
Position of place
180
75 0.045
track
Pairwise neuron-neuron
correlation coefficient
place fields (cm)
120
0 cm 0.005
50
0 180 0.08 0 180
Place cells
60 All neurons
0 0
0 60 120 180 0 60 120 180
Pairwise distance between place cells (m) Pairwise neuron-neuron distance (m)
20 m
Figure 6 Spatial organization of place cells in dorsal CA1. (a) Example images from different fields of view in which the place cells are colored according to
the location of their place fields along the virtual linear track. Each image shows place cells with significant place fields during running in either the positive
or negative direction. (b) Plot of mean distance between place cells in the hippocampus versus the mean distance between their place fields along the track
averaged over all 47 time series. The error bars represent s.e.m. (c) Plot of mean distance between cells in the hippocampus versus the mean correlation
between their temporal activity patterns averaged over all 47 time series for all place cells (gray) and all neurons (black). The error bars represent s.e.m.
the distance between their place fields along (d,e) A two-photon image (e) of a field of view ~50 m dorsal to the stratum pyramidale cell body
the track when the place cells were more layer (dashed line in d). Sparsely distributed cell bodies in e are assumed interneurons. (f) Mean
than about 35 m apart (although there F/F versus linear track position for the interneurons labeled in the image in e.
were some exceptions to this average rule;
Supplementary Fig. 1). The data point that corresponded to the most in all of the fields. Back-propagating action potentials invade the den-
closely neighboring place cells (less than about 35 m apart) was sig- dritic arbors of CA1 pyramidal neurons32,33. Therefore, it is likely
nificantly lower (P < 0.025, two-tailed t-test) than the points for all that the activity in the apical dendrites is due to opening of voltage-
other distances between cells. This could be interpreted as fine-scale gated calcium channels induced by back-propagating action potentials
spatial clustering of place cells with similar place fields, but residual rather than to the smaller calcium transients that are associated with
crosstalk, residual brain motion or a common neuropil signal between synaptic input.
neighboring ROIs could also explain this result (see Discussion). To show that our methods can be used to record from inter
We next studied whether the correlation between the temporal neurons, we adjusted the imaging plane to ~50 m dorsal to the
activity patterns of pairs of neurons varied as a function of the distance stratum pyramidale (Fig. 7d) where we found sparsely distributed
between them in the hippocampus. For all place cells in each imag- cell bodies (Fig. 7e). The neurons in this region of the stratum oriens
ing field, we calculated all pairwise temporal activity pattern correla- have been identified as interneurons, suggesting that these cells
tions and pairwise distances between place cells and then combined are probably basket, oriens lacunosum moleculare or axo-axonic
the results from all 47 datasets. When we plotted the mean temporal interneurons34. As in the stratum pyramidale, we found neurons with
correlation as a function of mean distance between the place cells spatially modulated activity patterns in the stratum oriens (Fig.7f);
(Fig.6c), we found a statistically insignificant relationship (Spearmans some of these had off fields (Fig. 7f), as has been observed with
rank correlation coefficient of 0.25, P = 0.5) for cells spaced more extracellular recording35,36. We imaged interneurons in four fields of
than about 35 m apart. Again, the data point that corresponded to view (~23 interneurons per field) in two mice and found spatially
the near-neighbor place cells was statistically different (P < 0.035, modulated interneuron activity patterns in all of the fields.
two-tailed t-test) from all but one of the other data points. When we
repeated this calculation for all neurons, including both place cells DISCUSSION
and non-place cells, the overall correlation was lower than for the Our method relies on several combined technologies. We developed
place cells alone and there was a statistically significant relationship surgical methods to implant a hippocampal window that allowed
between mean temporal correlation and distance between the neurons chronic subcellular resolution imaging in the CA1 region of the hip-
(Spearmans rank correlation coefficient of 0.95, P < 105; Fig. 6c), pocampus in behaving mice. The combination of this window with
regardless of whether the data point for the most closely neighboring a custom two-photon microscope and background light suppression
neurons was included. methods allowed imaging in mice that were interacting with a visual
virtual reality system. Using the genetically encoded calcium indicator
Imaging activity in dendrites and putative interneurons GCaMP3, our methods allowed us to study the spatially modulated
To test whether our imaging method had the required resolution activity patterns of pyramidal neurons in the stratum pyramidale,
and signal-to-noise ratio to record activity patterns from dendrites, putative interneurons in the stratum oriens and apical dendrites in the
we acquired time series ~75 m ventral to the stratum pyramidale stratum radiatum. By imaging the activity of populations of ~80100
(Fig.7). In this plane, we could identify apical dendrites from the neurons in the stratum pyramidale in mice that had been trained to
overlying pyramidal neurons (individual bright spots in Fig. 7b). When navigate along a virtual linear track, we identified place cells that had
we plotted the mean activity versus linear track position, we found characteristics very similar to those of spiking rate-defined place cells
well-defined place fields in many of the apical dendrites (Fig. 7c). in both real and virtual environments15,26,27,37.
We imaged dendrites in four fields of view (~50 dendrites per field) Tetrodes are the most common method used to record the acti
in three mice and saw spatially modulated dendritic activity patterns vity of CA1 place cells in behaving mice. Tetrode array recordings
combined with spike-sorting procedures can be used to identify hippocampal dynamics and place cell properties that were measured
around five single units per tetrode38,39. Although sub-millisecond in our preparation compared to those of control mice.
temporal resolution is a major advantage of this method compared to Miniaturized head-mounted microscopes47,48 may allow hippo
imaging, tetrode methods are limited by low spatial resolution (>100s campal imaging in freely moving animals. Such experiments would
of micrometers) and sparse sampling within the microcircuitry38,40. benefit from the natural array of inputs, as opposed to the lack of
Our imaging method has several advantages over tetrode methods. vestibular input and potentially altered gait of our mice. Our methods,
For example, it is possible to report the precise anatomical position however, have the advantages of not requiring a miniaturized micro-
of functionally identified neurons within the microcircuitry. Imaging scope, easy combination with electrophysiology, and the potential to
also allows researchers to make functional recording from subcel- manipulate specific environmental cues using virtual reality in ways
lular compartments, to identify all neurons (even silent cells), and to that would be difficult or impossible in real environments.
take advantage of the growing number of available genetic tools41. It We used an ICA/PCA algorithm to identify individual neurons on
should be possible to image subcellular dynamics such as signal trans- the basis of both their spatial and temporal characteristics21. However,
duction42 and structural plasticity43 in the context of learning and this method could not identify silent neurons and occasionally missed
memory during behavioral paradigms. In addition, optical methods active cells, making it hard to estimate the number of potentially active
might allow researchers to identify the functional properties of spe- neurons. Therefore, it was difficult to quantify the exact fraction of
cific neurons in a large population, followed by either the subsequent neurons that were place cells. Based simply on the morphology of
reconstruction of the underlying connections 44 or modification of neurons in the fields of view, we estimate that our fields contained
their activity45. Finally, imaging methods can allow the unambiguous ~80100 neurons, meaning that ~1520% of the neurons were place
identification and recording of the same neurons over many weeks. cells, slightly less than previous estimates15. In addition to the uncer-
Although this was not the focus of this research, to address the techni- tain number of potentially active neurons, this slight difference could
2010 Nature America, Inc. All rights reserved.
cal feasibility of such studies we have imaged the same region of CA1 also be due to differences in the recording methods and differences
over multiple days (Supplementary Fig. 5). in the definitions of place fields.
Here we assumed that calcium transients could be used as a proxy The ICA/PCA algorithm was successful in limiting the crosstalk
for spiking activity. Sodium action potentials generate calcium tran- between neighboring ROIs. However, when studying the spatiotempo-
sients in GCaMP3-expressing hippocampal pyramidal neurons in ral organization of neurons in CA1, we could not rule out the possibil-
brain slices. Calcium transient amplitude increases linearly with the ity of residual crosstalk, residual brain motion or a common neuropil
number of evoked action potentials, until saturation at more than signal between the most closely neighboring cells (less than about
about 10 action potentials13. There is a quantitatively similar rela- 35 m). This is one possible explanation for the statistically outlying
tionship between spiking activity and calcium transient amplitude data points in Figure 6b,c that correspond to the most closely neigh-
for neocortical neurons in brain slices and in vivo in both anesthe- boring cells. An alternative explanation for these outlying data points
tized and awake mice13. It is therefore reasonable to assume that the is that they represent small clusters of functionally similar neurons,
spiking activity of the CA1 neurons in vivo generated calcium tran- as recent studies have suggested5,9.
sients similar to those seen in slices13. Using these previous calcium For place cells that were further separated (more than about
transient measurements13 and our extracellular place cell recordings, 35 m), the distance between their place fields along the track was
we calculated the spike-induced calcium transients that would be statistically unrelated to the distance between their positions in the
expected in our imaging experiments (Supplementary Fig. 3). The hippocampus. A few previous electrode recording studies found
characteristics of the optically defined place fields are consistent with anatomically organized clusters of functionally similar hippocampal
and can be fully explained by calcium transients induced by sodium neurons on the 0.61 mm spatial scale6,7, whereas a separate study
action potentials. Therefore, the contribution of other sources to our of CA1 place cell location concluded that there was a lack of ana-
calcium transients, such as calcium influx due to synaptic input or tomical spatial organization 8. These previous methods provided
dendritic calcium spikes, is probably small. only an indirect measure of the spatial organization of place cells
It is unlikely that our methods allow us to detect single action in the CA1 microcircuitry. By contrast, using our methods, we were
potentials13, to determine firing rates or to reliably count the number able to directly measure the spatial organization. Although possible
of spikes. Although these limitations probably do not pose a problem crosstalk concerns meant that we could not unambiguously measure
for identifying place fields in place cells owing to the marked increase the organization down to the finest scale, it was possible to conclude
in spiking rate in the place field and the ability to average over many that place cells are anatomically distributed down to a scale of at
traversals, it is still possible that place cells with low activity levels least ~35 m.
may not be detected. For place cells separated by at least this scale, there was also no
Cortical excavation was used to expose the hippocampus for acute relationship between the correlation between their temporal activity
electrophysiology experiments in anesthetized cats nearly 50 years patterns and the physical distance between them. For all neurons,
ago46. More recently, cortical excavation was combined with the use we found a statistically significant decrease in temporal correlation
of a polystyrene tube filled with agarose and sealed with a cover- between the neurons the further they were from each other in the
slip to facilitate two-photon imaging of dendritic spines in an acute, physical space of the hippocampus (Fig. 6). Although this decrease
anesthetized mouse preparation14. Here, we used these methods as a is significant, both the overall correlation and the rate at which the
starting point to develop a chronic hippocampal window designed for correlation decreases as a function of distance are nearly an order of
imaging in behaving mice. The stainless steel cannula and coverslip magnitude smaller than has been observed in the motor cortex of
directly bonded to the external capsule surface with Kwik-Sil formed behaving mice11.
a rigid support structure that minimized brain motion during animal Finally, we note that spatiotemporal organization can occur in many
movements and allowed repeated imaging of the hippocampus for forms, most of which were not examined here. The methods outlined
weeks. The removal of the cortex overlying the hippocampus did in this research should allow future studies to search more thoroughly
not detectably alter the mouses performance of the task, or the for micro-organization within the hippocampus.
Methods 18. Thiel, G., Greengard, P. & Sudhof, T.C. Characterization of tissue-specific
transcription by the human synapsin I gene promoter. Proc. Natl. Acad. Sci. USA
Methods and any associated references are available in the online
88, 34313435 (1991).
version of the paper at http://www.nature.com/natureneuroscience/. 19. Burger, C. et al. Recombinant AAV viral vectors pseudotyped with viral capsids from
serotypes 1, 2, and 5 display differential efficiency and cell tropism after delivery to
Note: Supplementary information is available on the Nature Neuroscience website. different regions of the central nervous system. Mol. Ther. 10, 302317 (2004).
20. Ji, N., Milkie, D.E. & Betzig, E. Adaptive optics via pupil segmentation for high-
Acknowledgments resolution imaging in biological tissues. Nat. Methods 7, 141147 (2010).
This work was supported by Princeton University, the National Institutes of Health 21. Mukamel, E.A., Nimmerjahn, A. & Schnitzer, M.J. Automated analysis of cellular
signals from large-scale calcium imaging data. Neuron 63, 747760 (2009).
(grant 5R01MH083686-03), The Howard Hughes Medical Institute, The Helen Hay
22. Reidl, J., Starke, J., Omer, D.B., Grinvald, A. & Spors, H. Independent component
Whitney Foundation and The Patterson Trust. We thank F. Collman for the brain analysis of high-resolution imaging data identifies distinct functional domains.
motion correction algorithm. Neuroimage 34, 94108 (2007).
23. Kerr, J.N., Greenberg, D. & Helmchen, F. Imaging input and output of neocortical
AUTHOR CONTRIBUTIONS networks in vivo. Proc. Natl. Acad. Sci. USA 102, 1406314068 (2005).
D.A.D. and D.W.T. designed the research. D.A.D. performed the imaging 24. Sato, T.R., Gray, N.W., Mainen, Z.F. & Svoboda, K. The functional microarchitecture
experiments and developed the chronic hippocampal window system and surgery/ of the mouse barrel cortex. PLoS Biol. 5, e189 (2007).
training sequences. D.W.T. designed and implemented the combined two-photon 25. Greenberg, D.S., Houweling, A.R. & Kerr, J.N. Population imaging of ongoing neuronal
microscope and virtual reality instrumentation. D.A.D. and C.D.H. performed activity in the visual cortex of awake rats. Nat. Neurosci. 11, 749751 (2008).
26. McNaughton, B.L., Barnes, C.A. & OKeefe, J. The contributions of position,
extracellular recording and optimized virtual reality training procedures. L.T. and
direction, and velocity to single unit activity in the hippocampus of freely-moving
L.L.L. provided AAV2/1-synapsin-1-GCaMP3. D.A.D. analyzed the data. D.A.D. rats. Exp. Brain Res. 52, 4149 (1983).
and D.W.T. wrote the paper. 27. Nakazawa, K. et al. Hippocampal CA3 NMDA receptors are crucial for memory
acquisition of one-time experience. Neuron 38, 305315 (2003).
COMPETING FINANCIAL INTERESTS 28. Fenton, A.A. & Muller, R.U. Place cell discharge is extremely variable during
The authors declare no competing financial interests. individual passes of the rat through the firing field. Proc. Natl. Acad. Sci. USA 95,
31823187 (1998).
2010 Nature America, Inc. All rights reserved.
Published online at http://www.nature.com/natureneuroscience/. 29. Frank, L.M., Brown, E.N. & Wilson, M. Trajectory encoding in the hippocampus
Reprints and permissions information is available online at http://www.nature.com/ and entorhinal cortex. Neuron 27, 169178 (2000).
30. Griffin, A.L., Eichenbaum, H. & Hasselmo, M.E. Spatial representations of
reprintsandpermissions/.
hippocampal CA1 neurons are modulated by behavioral context in a hippocampus-
dependent memory task. J. Neurosci. 27, 24162423 (2007).
1. OKeefe, J. & Dostrovsky, J. The hippocampus as a spatial map. Preliminary evidence 31. Gupta, A.S., van der Meer, M.A., Touretzky, D.S. & Redish, A.D. Hippocampal replay
from unit activity in the freely-moving rat. Brain Res. 34, 171175 (1971). is not a simple function of experience. Neuron 65, 695705 (2010).
2. Ergorul, C. & Eichenbaum, H. The hippocampus and memory for what, where, 32. Regehr, W.G., Connor, J.A. & Tank, D.W. Optical imaging of calcium accumulation in
and when. Learn. Mem. 11, 397405 (2004). hippocampal pyramidal cells during synaptic activation. Nature 341, 533536 (1989).
3. OKeefe, J. Hippocampal neurophysiology in the behaving animal. in The 33. Spruston, N., Schiller, Y., Stuart, G. & Sakmann, B. Activity-dependent action
Hippocampus Book (ed. P. Andersen) 475548 (Oxford University Press, Oxford, potential invasion and calcium influx into hippocampal CA1 dendrites. Science
2007). 268, 297300 (1995).
4. Leutgeb, S., Leutgeb, J.K., Moser, M.B. & Moser, E.I. Place cells, spatial maps and 34. Klausberger, T. et al. Brain state and cell typespecific firing of hippocampal
the population code for memory. Curr. Opin. Neurobiol. 15, 738746 (2005). interneurons in vivo. Nature 421, 844848 (2003).
5. Nakamura, N.H. et al. Hippocampal cells encode places by forming small anatomical 35. Ego-Stengel, V. & Wilson, M.A. Spatial selectivity and theta phase precession in
clusters. Neuroscience 166, 9941007 (2010). CA1 interneurons. Hippocampus 17, 161174 (2007).
6. Eichenbaum, H., Wiener, S.I., Shapiro, M.L. & Cohen, N.J. The organization of 36. Wilent, W.B. & Nitz, D.A. Discrete place fields of hippocampal formation
spatial coding in the hippocampus: a study of neural ensemble activity. J. Neurosci. interneurons. J. Neurophysiol. 97, 41524161 (2007).
9, 27642775 (1989). 37. Schnee, A. Rats in Virtual Reality: The Development of an Advanced Method to
7. Hampson, R.E., Simeral, J.D. & Deadwyler, S.A. Distribution of spatial and Study Animal Behaviour (Eberhard-Karls-University, Tbingen, 2008).
nonspatial information in dorsal hippocampus. Nature 402, 610614 (1999). 38. Buzski, G. Large-scale recording of neuronal ensembles. Nat. Neurosci. 7,
8. Redish, A.D. et al. Independence of firing correlates of anatomically proximate 446451 (2004).
hippocampal pyramidal cells. J. Neurosci. 21, RC134 (2001). 39. Wilson, M.A. & McNaughton, B.L. Dynamics of the hippocampal ensemble code
9. Takahashi, S. & Sakurai, Y. Sub-millisecond firing synchrony of closely neighboring for space. Science 261, 10551058 (1993).
pyramidal neurons in hippocampal CA1 of rats during delayed non-matching to 40. Henze, D.A. et al. Intracellular features predicted by extracellular recordings in the
sample task. Front. Neural Circuits 3, 9 (2009). hippocampus in vivo. J. Neurophysiol. 84, 390400 (2000).
10. Denk, W., Strickler, J.H. & Webb, W.W. Two-photon laser-scanning fluorescence 41. Luo, L., Callaway, E.M. & Svoboda, K. Genetic dissection of neural circuits. Neuron
microscopy. Science 248, 7376 (1990). 57, 634660 (2008).
11. Dombeck, D.A., Graziano, M.S. & Tank, D.W. Functional clustering of neurons in 42. Miyawaki, A. Visualization of the spatial and temporal dynamics of intracellular
motor cortex determined by cellular resolution imaging in awake behaving mice. signaling. Dev. Cell 4, 295305 (2003).
J. Neurosci. 29, 1375113760 (2009). 43. Holtmaat, A. & Svoboda, K. Experience-dependent structural synaptic plasticity in
12. Dombeck, D.A., Khabbaz, A.N., Collman, F., Adelman, T.L. & Tank, D.W. Imaging the mammalian brain. Nat. Rev. Neurosci. 10, 647658 (2009).
large-scale neural activity with cellular resolution in awake, mobile mice. Neuron 44. Denk, W. & Horstmann, H. Serial block-face scanning electron microscopy to
56, 4357 (2007). reconstruct three-dimensional tissue nanostructure. PLoS Biol. 2, e329 (2004).
13. Tian, L. et al. Imaging neural activity in worms, flies and mice with improved GCaMP 45. Boyden, E.S., Zhang, F., Bamberg, E., Nagel, G. & Deisseroth, K. Millisecond-
calcium indicators. Nat. Methods 6, 875881 (2009). timescale, genetically targeted optical control of neural activity. Nat. Neurosci. 8,
14. Mizrahi, A., Crowley, J.C., Shtoyerman, E. & Katz, L.C. High-resolution in vivo 12631268 (2005).
imaging of hippocampal dendrites and spines. J. Neurosci. 24, 31473151 46. Kandel, E.R., Spencer, W.A. & Brinley, F.J. Jr. Electrophysiology of hippocampal
(2004). neurons. I. Sequential invasion and synaptic organization. J. Neurophysiol. 24,
15. Harvey, C.D., Collman, F., Dombeck, D.A. & Tank, D.W. Intracellular dynamics of 225242 (1961).
hippocampal place cells during virtual navigation. Nature 461, 941946 (2009). 47. Helmchen, F., Fee, M.S., Tank, D.W. & Denk, W. A miniature head-mounted two-
16. Hlscher, C., Schnee, A., Dahmen, H., Setia, L. & Mallot, H.A. Rats are able to photon microscope. high-resolution brain imaging in freely moving animals. Neuron
navigate in virtual environments. J. Exp. Biol. 208, 561569 (2005). 31, 903912 (2001).
17. Denk, W. & Svoboda, K. Photon upmanship: why multiphoton imaging is more than 48. Sawinski, J. et al. Visually evoked activity in cortical cells imaged in freely moving
a gimmick. Neuron 18, 351357 (1997). animals. Proc. Natl. Acad. Sci. USA 106, 1955719562 (2009).
to the top of the headplate and centered around the craniotomy. The flexibility top-most layers of the external capsule were peeled away so that the remaining
of the thin rubber allowed movement of the microscope objective with respect external capsule was never physically touched. To reduce animal movement-
to the craniotomy while maintaining the light-tight connections at both ends of induced brain motion that could interfere with time-series image acquisition,
the rubber tube. It was also possible for stray light (especially longer wavelength the surface of the external capsule was allowed to dry until tacky, a small drop of
red light) to enter the microscope by traveling through the animal and subse- uncured Kwik-Sil was applied to the surface, and then the cannula was inserted
quently through the exposed skull between the craniotomy and the inner edge (with a few hundred micrometers left above the skull surface) and cemented
of the metal ring that was attached to the animals skull. This area was therefore to the skull using opaque Meta-bond. When cured, the small drop of Kwik-Sil
covered with opaque dental cement (Meta-bond, Parkell, made opaque by adding had the effect of bonding the cannula to the external capsule and reducing in-
India Ink at ~5% vol/vol) so that the only remaining source of stray light through focal plane (X-Y) brain motion. In some cases, the cannula did not fit through
the microscope objective hole was through the hippocampal window itself the craniotomy and a hand drill was used to slightly widen the hole. After this
(Fig. 1e). The light through this window proved to be inconsequential in the hippocampal window implantation surgery, the mice typically woke up after
green imaging channel (<~5% of the baseline fluorescence level from labeled ~10 min and were walking around the cage within ~15 min. The next day, train-
cells). Though red fluorophores were not used in this research, we found that a ing in the virtual linear track resumed as before the surgery. Approximately
red light blocking filter (for example, 500/40, Chroma) in front of the projector 7 days after the window surgery, imaging experiments began. The experiments
was needed to reduce the amount of stray light in the red imaging channel to took place every 23 days for up to ~34 weeks in the same mouse (up to
<~5% of the baseline fluorescence level from labeled cells. ~45 weeks after the window surgery). GCaMP3 expression reached a somewhat
The Ti:sapphire excitation laser (Chameleon Ultra II, Coherent) was oper- steady state level ~14 days after injection. A small fraction of nuclear-filled
ated at 920 nm (~3050 mW average power at the sample). Green GCaMP3 GCaMP3 neurons with altered physiology have been described in the cortex13.
fluorescence was isolated using a bandpass filter (Semrock, 542/50) and detected A similar small fraction of nuclear-filled neurons were observed here in CA1
using a GaAsP PMT (1077P-40, Hamamatsu). ScanImage (v3.6)49 was used for (starting ~23 weeks after injection). These neurons often generated large, long-
microscope control and acquisition. Images (256 64 pixels, ~200 100 m duration calcium transients that rarely defined a place field.
field of view) were acquired at 15.6 Hz. During the ~913 min of time-series The cannula was composed of a 1.5-mm segment of a 2.77 mm outer diameter
acquisition at a single location, photo-bleaching was observed at a mean rate of and 2.31 mm inner diameter thin-walled stainless steel tube (Small Parts Inc.).
~12% per minute. A 2.5-mm diameter round coverglass (Erie Scientific, Thermo Scientific) was
Our spherical treadmill and virtual reality system has been described12,15. We cemented to one end of the tubing using UV curable adhesive (Norland Products).
used a Digidata (Axon Instruments, 1440A) data acquisition system to record
(Clampex 10.2) and synchronize position in the linear track, reward timing and Does cortical excavation alter hippocampal dynamics and place cell
two-photon image frame timing (using the command signal to the slow galva- properties? To image the CA1 region of the hippocampus with cellular resolu-
nometer). Instead of the MX-1000 computer mouse used previously, here we used tion, it was necessary to unilaterally remove part of the overlying cortex (includ-
an MX-518 (Logitech) mouse to record ball rotation. The ball movements mea ing parietal cortex and parts of visual and hindlimb sensory cortex). This raises
sured using the LED based MX-518 mouse were less sensitive to the exact distance the concern that cellular and network properties within the hippocampus may
between the sensor and ball surface than the MX-1000. In addition, the MX-518 be altered either by direct mechanical trauma to the hippocampus or by altering
mouse can be adjusted between three gain settings, the lowest of which allowed upstream inputs to the hippocampus due to the cortical lesion (although the
measurements of larger ball velocity without saturation than the MX-1000. excavated cortical regions do not provide strong direct projections to the hippo
campal region). Regarding mechanical trauma, our surgical procedures allow the
Mouse training, hippocampal window and AAV injections. All experiments removal of the overlying cortex without the need to physically touch the surface
were performed in compliance with the Guide for the Care and Use of Laboratory of the hippocampus. The cortex and the top-most axons in the external capsule
Animals (http://www.nap.edu/readingroom/books/labrats/). Specific protocols were peeled away from the remaining external capsule without applying direct
were approved by the Princeton University Institutional Animal Care and Use pressure or inadvertent mechanical trauma to the hippocampus. Furthermore, the
Committee. Imaging experiments were performed on five male C57BL/6 mice external capsule itself provides a protective barrier to the underlying CA1 region
(postnatal day ~50). First, mice were anesthetized (Isoflurane, ~1%), a headplate of the hippocampus. Therefore, overt signs of damage, such as vesiculated den-
was attached to the skull using opaque metabond and the stereotactic coordi- drites or reduced or aberrant neuronal activity observed via calcium transients,
nates overlying the hippocampus were marked (1.8 mm lateral, 2.0 mm caudal were not observed. Although overt signs of damage were easily avoided, more
to bregma). Any exposed skull was covered with a silicone elastomer (Kwik-Sil, subtle effects due to alterations of the upstream circuits projecting to the hippo
World Precision Instruments). Anesthesia was then removed and the mouse campus were still possible. However, the properties of CA1 place cells measured
genetically encoded indicators, we used GCaMP3 (ref. 13). lined above. This process was repeated 1,000 times and the P-value was defined
as the ratio of the number of times out of 1,000 that the random shuffled trace
Data analysis. Analysis was performed using ImageJ (1.40 g) and custom scripts generated a place field that met the above criteria.
written in MatLab (version 7). All data in the text and figures are presented A directionality index for mean F/F in a place field in positive and negative
as mean s.d., except in the plots shown in Figure 6b,c and Supplementary running directions (F+, F) was defined as |F+ F|/(F+ + F). A directionality
Figure 2, where the error bars represent s.e.m. index of 0 indicates identical activity in both directions, whereas an index of 1
Time-series datasets were motion corrected using whole-frame cross- indicates activity in one direction only.
correlation. Although our previously published HMM line-by-line motion Note that while very little movement along the width of the virtual linear track
correction algorithm12 would probably provide superior correction, our imaging was possible, this position as well as the view angle were not controlled for in
frame rate (~15 Hz) was fast enough that in-frame distortions were minimal using our analysis. These factors could contribute to increased variability of the place
whole-frame correction. In addition, it was more straightforward to apply the fields. Also note that during time-series acquisition, the mice rarely stopped in
ICA/PCA cell ROI selection algorithm to the whole-frame corrected time-series the middle (between reward zones) of the track (Supplementary Fig. 1): only
than to the HMM-corrected time-series in which blank lines made automatic ROI 5 3% of the time per time series was the virtual velocity <0.05 cm s1 in the
detection difficult. The mean frame-to-frame in-plane (X-Y) Euclidean distance middle of the track.
brain motion during all time periods (the mice ran almost continuously during Correlation values were defined as the Pearsons correlation coefficient.
our time-series acquisitions) was 1.8 1.3 m; the motion during periods of Correlation versus distance between neuron and place field distance versus
mouse movement along the virtual track used to define place fields (see below) distance between neuron plots were generated from imaging fields that had at
was 1.5 1.2 m. Out of plane (Z) motion (measured as described12) was least two place cells and any two cells that had overlapping ROIs were excluded
measured in untrained mice running on a spherical treadmill; during periods of from the analysis. The position of a place field along the virtual track was defined
running the Z-motion was 0.7 0.2 m. as the position within the field with the maximal F/F value. For the mean place
ROIs were defined as described previously21 (mu = 0.5, 150 principal com- field width calculation, only fields in which neither field edge was at the end of
ponents, 100 independent components, s.d. threshold = 1.25, 60 pixels < area the track were included. This avoided the inclusion of fields that were artificially
of ROI < 400 pixels). Manual inspection revealed that the ROIs nearly always narrow due to clipping at one end of the track.
defined single cell regions. F/F versus time traces were generated for each ROI.
Slow time-scale changes in the fluorescence traces were removed by examining Electrophysiology. Extracellular place cell recordings were performed in four mice
the distribution of fluorescence in a ~15-s interval around each sample time with hippocampal windows using methods similar to those described15. All pro-
point and subtracting the 8% percentile value. The baseline-subtracted neuron cedures for these mice were identical to those used in imaging experiments (water
fluorescence traces were then subjected to analysis of the ratio of positive- to restriction, training, virus injection, surgeries, and so on) except that a small hole
negative-going transients of various amplitudes and durations as described12. We (~0.5 mm) was drilled in the center of the window coverslip before implantation.
used this analysis to identify significant transients with <5% false positive error After implantation, this small hole allowed electrode access to the hippocampus
rates and generated the significant transient-only traces (Figs. 2b,c and 5a) that through the cannula. A tungsten metal electrode (3 M impedance at 1 kHz, FHC)
were used for all subsequent analysis11. Note that the significant transient only or glass microelectrode (resistance ~2.5 M, filled with 0.5 M NaCl) recorded
traces are not discontinuous traces; the significant transients are left untouched, CA1 extracellular currents that were amplified (DAM80, WPI or AxoClamp
but the time points between the significant transients are all set to 0. 2B) and filtered (3 Hz10 kHz, Brown-Lee Model 440) before being digitized at
Place fields were identified and defined as follows. Long running periods of 20 kHz with a Digidata (Axon Instruments, 1440A). Note that this preparation
mouse movement along the virtual track in which the virtual velocity was >8.3 was intended to mimic the conditions of the imaging experiments, but did not
cm s1 and the run length was >53 cm (straight run without changing direction or allow long-duration extracellular recordings; the hole in the coverslip to allow elec-
hitting the end of the track) were identified. These periods were first categorized trode access probably caused instability due to reduced stabilizing pressure. While
based on the direction of running (positive or negative direction). Positive and these recordings (~35 min duration per cell) allowed clear identification of place
negative long running direction periods were then further subdivided into two cells and phase precession, the reduced number of passes through the place field
categories based on the animals current performance of the task. Segments of compared to longer recordings led to noisier place field maps and phase preces-
time between two rewards in which long running periods of only one direction sion plots (Supplementary Fig. 4). To reduce low frequency (<~50Hz) electrical
occurred were defined as high reward rate periods, all other long running periods noise observed during treadmill rotation, the ball was coated with unscented fabric
were defined as low reward rate periods. Timeseries datasets were only included softener (Ultra Downy, Procter and Gamble) and allowed to dry overnight for
if a mean of at least 10 long running segments during high reward rate periods in electrical recordings. In other experiments not reported here, treating the ball with
548
Vol 461 | 15 October 2009 | doi:10.1038/nature08499
ARTICLES
Intracellular dynamics of hippocampal
place cells during virtual navigation
Christopher D. Harvey1,2,3, Forrest Collman1,2,3, Daniel A. Dombeck1,2,3 & David W. Tank1,2,3
Hippocampal place cells encode spatial information in rate and temporal codes. To examine the mechanisms underlying
hippocampal coding, here we measured the intracellular dynamics of place cells by combining in vivo whole-cell recordings
with a virtual-reality system. Head-restrained mice, running on a spherical treadmill, interacted with a computer-generated
visual environment to perform spatial behaviours. Robust place-cell activity was present during movement along a virtual
linear track. From whole-cell recordings, we identified three subthreshold signatures of place fields: an asymmetric ramp-like
depolarization of the baseline membrane potential, an increase in the amplitude of intracellular theta oscillations, and a
phase precession of the intracellular theta oscillation relative to the extracellularly recorded theta rhythm. These
intracellular dynamics underlie the primary features of place-cell rate and temporal codes. The virtual-reality system
developed here will enable new experimental approaches to study the neural circuits underlying navigation.
Hippocampal place cells encode spatial information during navigation spherical treadmill with its head held fixed in space using a head plate.
using rate and temporal codes1,2. The rate code refers to a selective The mouse is surrounded by a toroidal screen that covers a wide area to
increase in firing rate at a specific location in a local environment3, accommodate a rodents large field of view20. An image is projected
and the temporal code includes the precise timing of spikes relative to onto the screen from a digital light processing projector by an angular
the hippocampal theta rhythm (phase precession)46. To explain the amplification mirror21 (Fig. 2a, Supplementary Fig. 1 and Methods).
origins of these codes, theoretical network and cellular models have To control the virtual-reality system, we developed custom software
been proposed4,716. These models make differing predictions about the using the open source Quake2 video game engine. The visual display
subthreshold membrane potential dynamics of place cells (Fig. 1ae, was updated on the basis of the movements of the animal, measured as
gi), reflecting differences in the proposed mechanisms underlying rotations of the spherical treadmill using an optical computer mouse
hippocampal coding. The predicted intracellular dynamics include (Methods).
steady oscillations at theta frequencies that reflect a global hippocampal We addressed whether head-restrained mice can perform visually
theta rhythm79,12 (Fig. 1e), modulation of the amplitude and frequency guided spatial behaviours in a virtual environment. We trained
of membrane potential theta oscillations4,10,11,1315 (Fig. 1bd), and water-scheduled mice using operant conditioning to run along a
ramps of depolarization of the baseline membrane potential710,12,13 virtual linear track (180 cm long) that had proximal and distal walls
(Fig. 1de). To account for phase precession, these models also make with varying patterns for location cues (Fig. 2b). Mice were able to
differing predictions about the relationship between intracellular theta turn around at any position along the tracks length. Small water
oscillations and the local field potential (LFP) theta rhythm. Models rewards were given for running between reward zones located at
predict either that intracellular theta is phase-locked to the LFP theta opposite ends of the track; consecutive rewards were not available
rhythm with phase precession resulting from a ramp of depolari- at a single reward site. After several training sessions, mice ran large
zation79,12, or that intracellular theta in the place field is at a higher total distances with high peak running speeds (session 4: total dis-
frequency than the LFP rhythm4,10,11,1315 (Fig. 1gi). tance 5 217 6 97 m per 40 min, peak speed 5 41 6 17 cm s21 over a
Because the models of hippocampal coding make differing predic- 2 s period, mean 6 s.d.). Individual mice received rewards at increas-
tions of subthreshold membrane potential dynamics, it is possible to ing rates over time (Fig. 2c, d). Also, the average distance travelled
distinguish between these models by intracellular measurements between rewards decreased across sessions (Fig. 2e), consistent with
from place cells during spatial behaviours. However, intracellular learning of the task. After ten training sessions, mice ran 281 6 53 cm
recording methods require a level of mechanical stability that is dif- between rewards, on average, approaching an ideal performance of
ficult to obtain in freely moving animals17,18. Previously, head 180 cm per reward (that is, the distance between reward zones). These
restraint on a spherical treadmill has been used to facilitate optical data indicate that head-restrained mice can perform visually guided
imaging at cellular resolution in awake, mobile mice19. Furthermore, spatial behaviours in a virtual-reality environment.
a previous study has provided evidence that body-tethered rats can
navigate through virtual environments20. We reasoned that these Place cells in a virtual environment
could be combined to facilitate whole-cell recordings to show the Although our behavioural results indicate that mice have a spatial
intracellular dynamics of place cells, and thus distinguish between understanding of the virtual environment, the activation of naviga-
models of hippocampal coding. tion circuits during these behaviours would provide further evidence.
To assess the function of the hippocampal place-cell circuitry, we
Spatial behaviour in a virtual environment performed acute extracellular recordings in the dorsal hippocampus
The visual virtual-reality system for head-restrained mice we from CA1 pyramidal neurons (Fig. 3a). We recorded during beha-
developed is shown in Fig. 2a. A mouse runs on top of an air-supported viour along the virtual linear track from mice that had been trained
1
Princeton Neuroscience Institute, 2Lewis-Sigler Institute for Integrative Genomics, 3Department of Molecular Biology, Princeton University, Princeton, New Jersey 08544, USA.
941
549
2009 Macmillan Publishers Limited. All rights reserved
ARTICLES NATURE | Vol 461 | 15 October 2009
Position (cm)
Time
RM 80
Optical 160
Position (cm)
c mouse VR
computer 80
Air
d 0
0 150
b Time (s)
d
3
Rewards per
e
minute
2
1
View from left end
f Measured subthreshold Vm 0
2 6 10
e Session
Distance run
5 mV
10
0.2 s View from right end
Number of spikes
mice), during which time mice ran long distances at high speeds in
the virtual environment (27 6 5 m per min; total distance range 125
50
458 m). We did not detect any major motion-induced artefacts in the
electrophysiology recordings. Recordings could be performed from
50 V
100 V 100 ms 0
the same animal across several days ($6 days per mouse; $3 days per
1 ms 103 101 101 hemisphere).
c ISI (s) A subset of our whole-cell recordings was made from place cells
7 Both directions 7 9 (overall firing rate 5 2.2 6 0.4 Hz; in-field firing rate 5 7.3 6 1.4 Hz;
Right runs
out-of-field firing rate 5 1.5 6 0.4 Hz; field size 5 43 6 13 cm; n 5 8
Firing rate (Hz)
Left runs
cells from 8 mice; Fig. 4a, b and Supplementary Figs 2a and 3).
Approximately 36% of the spontaneously active putative pyramidal
neurons had a place field along the virtual track, which is consistent
with estimates from extracellular recordings and immediate early
0 0 0
a
0 90 180
Position (cm)
d e
720
Phase (degrees)
300
65 mV 10 mV
360 200
2s
b 10 12 16
100
Firing rate (Hz)
0
ld th
ld th
fie gh
of ei
st
rs
La
Fi
Figure 3 | Extracellular recordings of CA1 place cells along the virtual linear 0 0 0
track. a, An example extracellular recording filtered between 500 Hz and
7.5 kHz. The inset shows overlaid spike waveforms from the recording. b, ISI c
60
distribution for the full duration of the recording shown in a. The time axis is 61
plotted on a log scale. c, Example firing rate maps for 3 place cells from 3 66
Vm (mV)
different mice. Top, firing rates at positions along the track are shown for 62
rightward runs (red), leftward runs (blue) and runs in either direction 68 63
(black). Bottom, the position on the track of each spike in the recording is
shown as a vertical line. A total of 23 place cells from 8 mice were recorded. 64 70 65
d, Phase precession of spike times relative to LFP theta. Top left, an example
0 90 180
extracellular recording, filtered between 2 Hz and 10 kHz, during a run Position (cm)
through the place field. Spikes and the LFP were recorded on the same d 60 e
electrode. Bottom left, the extracellular recording band-pass filtered between 12 4
6 and 10 Hz. Grey lines denote peaks (0u) in the filtered trace, black lines
Firing rate (Hz)
V (mV)
denote the times of spikes. Right, an example plot of phase (two cycles) 64
Vm (mV)
2
versus position on the virtual track for all spikes during complete runs 6
through the place field for a single cell. e, Phase values for spikes in the first 68
and last eighth of the place field. Connected points represent a single place 0
field. Horizontal lines indicate the means. n 5 10 place fields from 8 cells and 72 0
0 1
In-field
Out-of-field
3 mice (multiple place fields are owing to the directionality of firing rates). 0 1
Position (fraction Position (fraction
between the first and last eighth of the field, P , 0.01; correlation of place field) of place field)
coefficient (C) 5 20.17 6 0.09 between phase and position,
P , 0.01; n 5 10 place fields, 8 cells, 3 mice; Fig. 3d, e). These firing
Figure 4 | Ramp-like membrane potential depolarization inside place
rate and phase precession characteristics along the virtual linear track
fields. a, Example whole-cell recording during runs through the cells place
are highly similar to those measured in freely moving mice in real field. Grey boxes indicate the place field (middle example from b). b, Firing
environments2629. These data therefore demonstrate that hippocam- rates along the virtual linear track for 3 place cells recorded intracellularly
pal place-cell circuitry is operational in head-restrained mice during from 3 different animals. The grey boxes indicate the primary place field
visually guided spatial behaviours in the virtual-reality system. determined by firing rates (Methods). Bottom, vertical lines mark the
location along the track of every action potential in the recording. c, Average
Intracellular dynamics of place cells baseline membrane potential, excluding action potentials, sorted by position
We next developed methods to measure the intracellular activity of along the track for the 3 place cells from b. d, Average membrane potential
inside and outside the place field. Each pair of connected points is from a
hippocampal neurons during behaviour along the virtual linear
single cell. Horizontal lines indicate the means. n 5 8 cells from 8 mice.
track. Because the mouses head is stationary in the virtual-reality e, Average firing rates and changes in baseline membrane potential during
set-up, we were able to perform whole-cell patch-clamp recordings complete runs through the place field. To combine data from several cells,
using a patch electrode with a long taper mounted on a standard position values in the place field were normalized. Black lines indicate the
micromanipulator positioned outside the mouses field of view mean. Grey lines indicate the mean 6 s.e.m. Data are averaged over 84
(see Methods). We obtained recordings in awake mice17,18,30,31 as they complete runs through the place field (8 cells).
943
551
2009 Macmillan Publishers Limited. All rights reserved
ARTICLES NATURE | Vol 461 | 15 October 2009
gene studies in rats exploring real environments32,33 (see Methods). 6). These depolarization events occurred preferentially in place fields;
Place cells recorded intracellularly fired in theta bursts (high fre- the average membrane potential excluding action potentials was higher
quency bursts at .50 Hz occurring at theta frequencies of ,6 in place fields than outside of place fields (Vin-field 2 Vout-of-field 5
10 Hz), resulting in interspike interval (ISI) distributions with peaks 2.5 6 0.5 mV, P , 0.0001; Fig. 4c, d and Supplementary Fig. 2b).
at ,10 ms and ,130 ms, consistent with our extracellular measure- Consistently, the baseline membrane potential and firing rate were
ments (compare Supplementary Fig. 4 and Fig. 3b). The individual strongly correlated (C 5 0.55 6 0.10, P , 0.001). On complete runs
action potentials during a theta burst often occurred during the through the place field, ramps of depolarization were asymmetric, with
ascending phase of an underlying theta oscillation (Fig. 5a, d). the peak depolarization shifted towards the end of the field (position of
Spike amplitudes decreased within a burst without a change in the the peak depolarization 5 72 6 24% of the distance through the field,
peak potential of the spike, suggesting that depolarizing intracellular P , 0.05 versus 50%; slope before peak 5 3.0 mV per place field length,
oscillations contribute to intra-burst decreases in spike amplitude34. slope after peak 5 5.4 mV per field length, P , 0.001; Fig. 4e). In con-
Whole-cell recordings from place cells can therefore be obtained in trast, firing rates were symmetric within the place field (position of the
head-restrained mice behaving in virtual environments. peak firing rate 5 52 6 17% of the distance through the field, P . 0.6
From our whole-cell recordings of place-cell activity, we examined versus 50%; slope before peak 5 13.3 Hz per place field length, slope
two types of membrane potential dynamics proposed by the theoretical after peak 5 13.6 Hz per field length, P . 0.4; Fig. 4e), perhaps because
models of place-cell function: ramps of depolarization, and modu- firing rates were highest on the ascending part of a ramp depolariza-
lation of theta oscillations (Fig. 1be, hi). We first analysed changes tion13 (Supplementary Fig. 7a). An asymmetric ramp-like depolariza-
in the average baseline membrane potential during behavioural epochs tion of the baseline membrane potential is therefore a subthreshold
inside and outside the place field. As a mouse approached the recorded signature of place fields.
cells place field, the average membrane potential, excluding action We next examined the modulation of the amplitude and phase of
potentials, increased in a ramp-like manner and remained increased membrane potential oscillations occurring at theta frequencies during
until the place field was passed (Fig. 4a). The ramp of depolarization runs along the virtual track. We measured intracellular theta oscilla-
often began before action potential firing in the place field started, and tions by band-pass filtering our membrane potential recordings from
in some cases reached a steady depolarization as large as ,10 mV (peak 610 Hz, after action potentials were removed (Fig. 5a and Methods).
depolarization 5 5.7 6 2.9 mV; Fig. 4a and Supplementary Figs 5 and When the mouse entered the recorded cells place field, the amplitude
a b c
4 3
Theta power (mV2)
2.2 1.6
2
Vm
2
1.6 1
1
68 mV 15 mV 1 0 0.4
In-field
Out-of-field
2s 0 90 180
Position (cm)
4 mV
Filtered Vm d e
2s
(610 Hz) 720 400
Phase (degrees)
360 340
15 mV 64 mV
0.1 s
0 280
First eighth
Last eighth
5 mV 40 120
of field
of field
0.1 s Position (cm)
20 mV
10 mV
f 0.1 s
Spike times
Intracellular theta peaks g Spikes Spikes
Intracellular theta
LFP theta peaks peaks
720
LFP phase (degrees)
720 720
Intracellular phase
Vm
(degrees)
LFP 15 mV 8 mV
0.4 mV 0.3 mV 0 0
0.2 s 110 160 0 110 160
110 160
0.2 s
Position (cm)
Figure 5 | Membrane potential theta oscillations in place cells. a, Top, and last eighth of the place field for individual place fields. Horizontal lines
example raw and filtered (610 Hz after spikes were removed) membrane are the means. n 5 12 place fields and 8 cells from 8 mice. f, Raw (left) and
potential traces during runs through the cells place field (grey boxes). filtered (610 Hz, right) membrane potential and LFP traces in the place field
Bottom, expanded portions of the raw and filtered traces. b, Power in the from a simultaneous LFP and whole-cell recording. The times of LFP theta
theta-frequency band sorted by position along the track for the cells from peaks (grey lines), intracellular theta peaks (circles) and spikes (crosses) are
Fig. 4. Power was measured as the squared amplitude of the filtered trace. shown to illustrate the phase precession of spikes and intracellular theta
Grey boxes denote the primary place fields. c, Theta power inside and relative to LFP theta oscillations, and the absence of phase precession of
outside the place field for individual cells. Horizontal bars are means. n 5 8 spikes relative to intracellular theta oscillations. g, Phase precession of spike
cells from 8 mice. d, Spike times relative to intracellular theta. Left, example times relative to intracellular theta (left), relative to LFP theta (middle) and
raw and filtered (610 Hz) membrane potential traces in the place field. Grey phase precession of intracellular theta peak times relative to LFP theta
lines indicate peaks of theta oscillations (0u), black lines are times of spikes. (right) from a simultaneous LFP and whole-cell recording. In d and f, the top
Right, example phase (two cycles) versus position plot during place field and bottom scale bar labels denote the top and bottom traces, respectively.
traversals for a single cell. e, Intracellular phase values for spikes in the first
944
552
2009 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 461 | 15 October 2009 ARTICLES
of intracellular theta oscillations increased (Fig. 5a). Theta-band spikes recorded in slices36. Further analysis is required to assess the
power in the membrane potential trace was higher in place fields than prevalence, significance and mechanisms of these events.
outside of place fields (powerin-field 5 1.7 6 0.4 mV2, powerout-of-field 5
0.8 6 0.2 mV2, P , 0.01; Fig. 5b, c and Supplementary Figs 2c and 8). Discussion
Consistently, theta oscillation amplitude and firing rate were highly Here we have developed a visual virtual-reality system in which head-
correlated (C 5 0.61 6 0.09, P , 0.001). In contrast, the amplitude of restrained mice performed spatial behaviours along a virtual linear
membrane potential theta oscillations was similar at all spatial loca- track. Hippocampal place-cell activities were triggered during runs
tions for putative CA1 pyramidal neurons that did not have a place along the track, with similar properties to those recorded in real environ-
field (Supplementary Fig. 9) and for the LFP theta rhythm (Sup- ments. Because the mouse was head-restrained, we were able to obtain
plementary Fig. 10). whole-cell recordings lasting many minutes in awake mice using
To examine the modulation of the phase of intracellular theta, we standard patch-clamp techniques. Furthermore, this set-up can prob-
compared intracellular theta fluctuations with LFP theta oscillations. ably be combined with two-photon laser scanning microscopy, which
We began by looking for phase precession of spike times relative to has previously been performed in mice running on the spherical tread-
membrane potential theta oscillations. The intracellular phases of mill19. Virtual reality also offers the ability to design highly custom
spike times did not change during runs through the place field, environments, and to manipulate these environments rapidly through
and the phase and position of spikes were not correlated software. We therefore anticipate that our virtual-reality system will
(Dphase 5 0.6 6 10.4u between the first and last eighth of the field, make new types of experiments exploring spatial information encoding
P . 0.9; C 5 20.01 6 0.08 between phase and position, P . 0.6; possible.
Fig. 5dg). Because spike times advanced relative to LFP theta oscil- We identified three subthreshold signatures of place fields: an
lations but not intracellular theta (compare Figs 3d, e and 5d, e), it is asymmetric ramp-like depolarization of the baseline membrane
predicted that a phase shift between LFP and intracellular theta potential, an increase in the amplitude of membrane potential theta
occurs during place field traversals. We therefore performed simu- oscillations, and a phase precession of intracellular theta relative to
ltaneous whole-cell and LFP recordings to compare directly the LFP theta, such that spike times advanced relative to LFP theta but
phases of intracellular and extracellularly recorded theta. During not intracellular theta (Figs 1f, 4 and 5). These findings seem incon-
runs through the place field, the phase difference between intracel- sistent with the mechanisms underlying dual oscillator models2,4,14,
lular and LFP theta shifted such that the intracellular theta oscillation network models79 and an experience-dependent model12 of hippo-
phase precessed relative to the LFP theta rhythm (C 5 20.26 6 0.12 campal rate and temporal coding because each model predicts mem-
between LFP phase and position for the times of intracellular theta brane potential dynamics that differ significantly from our observed
peaks; n 5 2 cells from 2 mice; Fig. 5f, g and Supplementary Fig. 11a). subthreshold signatures (Figs 1b, c, eh). Our data are most consist-
Consistently, the frequency of intracellular theta oscillations in the ent with a soma-dendritic interference model that proposes interac-
place field was higher than the frequency of LFP theta fluctua- tions between a spatially independent inhibitory oscillatory input
tions (measured as a ratio of periods of intracellular theta to near the soma, and a spatially dependent (increasing in the place
periods of LFP theta; ratioin-field 5 0.97 6 0.21, P , 0.01 versus 1; field) and temporally patterned (at theta) dendritic excitatory
Dfrequency 5 0.22 Hz given a mean LFP frequency of 7.4 Hz; see input10,11,13,15. With an appropriate choice of conductances15, the
Methods and Supplementary Fig. 11b). In contrast, the frequencies soma-dendritic interference model predicts, in the place field,
of intracellular theta and LFP theta during epochs outside the place ramp-like depolarizations (Figs 1d and 4), an increase in theta ampli-
field were similar (ratioout-of-field 5 1.01 6 0.22, P . 0.2 versus 1; tude (Figs 1d and 5ac), and precession of intracellular theta relative
Supplementary Fig. 11b). Intracellular theta oscillations in place cells to extracellular theta (Figs 1i and 5f, g). We note, however, that we
were therefore not constant in amplitude or phase (relative to LFP observed an asymmetric ramp-like depolarization (Fig. 4e), which
theta) throughout runs on the linear track; rather, membrane poten- may be important for phase precession across the entire place field12,
tial theta oscillations were dynamically modulated across positions in compared with a symmetric ramp proposed in this model. We do not
virtual space. exclude the possibility that revised forms of other models could
Ramp-like depolarizations of the membrane potential and potentially explain the intracellular dynamics we observed. Other
increases in intracellular theta oscillation amplitude were present experiments will be necessary to define and quantify the cellular
simultaneously, which can be demonstrated directly by filtering the and synaptic mechanisms underlying the intracellular dynamics we
membrane potential trace from DC-10 Hz (Fig. 1f). Consistently, measured, including in the context of the entorhinal cortex2,16,37, and
intracellular theta power and the baseline membrane potential were to establish their causal relationship to rate and temporal codes in the
highly correlated (C 5 0.52 6 0.07, P , 0.001; Figs 4c and 5b and hippocampus. The virtual-reality system developed here, combined
Supplementary Fig. 2b, c). To determine whether ramps of depolariza- with electro- and opto-physiological methods, will probably facilitate
tion trigger increases in theta amplitude, we injected ramps of current this analysis.
at the soma while the animal was running along the virtual track, and
measured changes in theta power. Theta power increased weakly with METHODS SUMMARY
higher levels of depolarization; however, the increase in power was A virtual-reality system was designed using an air-supported spherical treadmill for
smaller than during runs through the place field (P , 0.01 at similar head-restrained mice19, in combination with a projection-based visual display
DV values; compare Figs 4c, 5b and Supplementary Fig. 7b). Ramp-like system20, in which a toroidal screen presented an image from a projector by an
depolarizations of the membrane potential therefore were not suf- angular amplification mirror21. Custom software to control the virtual-reality
ficient to cause the increases in theta oscillation amplitude observed system was developed on the basis of the Quake2 game engine. Rotations of the
in place fields. spherical treadmill, measured by an optical computer mouse, were used to update
Our whole-cell recordings revealed two further subthreshold the visual display. Water-scheduled C57BL/6J mice (812 weeks old) were trained
phenomena. First, in a fraction of our recordings we observed using operant conditioning to run from end-to-end of a virtual linear track
spikelets, brief small amplitude deflections of the membrane poten- (180 cm long) to obtain water rewards. For electrophysiology measurements, a
small craniotomy (,0.5 mm diameter) was made centred over dorsal hippocam-
tial (2 out of 8 place cells; amplitude 5 7.4 6 1.3 mV, full-width at
pus (2.2 mm caudal, 1.7 mm lateral to bregma). The craniotomy was sealed with
half-maximum 51.6 6 0.4 ms; 0.06 6 0.05 spikelets per second; Sup- silicone grease and then covered with silicone elastomer to allow recordings across
plementary Fig. 12a)35. Second, bursts of action potentials were occa- several days. Extracellular recordings were made using a glass electrode (filled with
sionally followed by large (,1025 mV) depolarizations lasting up to 0.5 M NaCl, ,2.5 MV pipette resistance) mounted on a micromanipulator posi-
50100 ms23 (Supplementary Fig. 12b). These events sometimes con- tioned behind the mouse. Whole-cell recordings were obtained using standard
tained broadened spikes of reduced amplitude, consistent with Ca21 blind patch methods. Patch pipettes were pulled with a long taper (,100 mm
945
553
2009 Macmillan Publishers Limited. All rights reserved
ARTICLES NATURE | Vol 461 | 15 October 2009
diameter at 1 mm from the tip) to minimize damage to the overlying cortical tissue, 21. Chahl, J. S. & Srinivasan, M. V. Reflective surfaces for panoramic imaging. Appl.
and were mounted on a micromanipulator positioned outside the field of view. Opt. 36, 82758285 (1997).
Firing rate maps were calculated for 80 spatial bins along the virtual track as the 22. Ranck, J. B. Jr. Studies on single neurons in dorsal hippocampal formation and
septum in unrestrained rats. I. Behavioral correlates and firing repertoires. Exp.
number of spikes in a bin divided by the time spent in that bin. Changes in baseline
Neurol. 41, 461531 (1973).
membrane potential in the place field were measured from membrane potential 23. Kandel, E. R. & Spencer, W. A. Electrophysiology of hippocampal neurons. II.
traces excluding spikes. Theta oscillations were analysed after band-pass filtering After-potentials and repetitive firing. J. Neurophysiol. 24, 243259 (1961).
(610 Hz) of the membrane potential recording using a linear phase finite impulse 24. Quirk, M. C. & Wilson, M. A. Interaction between spike waveform classification
response filter. and temporal sequence detection. J. Neurosci. Methods 94, 4152 (1999).
25. McNaughton, B. L., Barnes, C. A. & OKeefe, J. The contributions of position,
Full Methods and any associated references are available in the online version of direction, and velocity to single unit activity in the hippocampus of freely-moving
the paper at www.nature.com/nature. rats. Exp. Brain Res. 52, 4149 (1983).
26. Nakazawa, K. et al. Hippocampal CA3 NMDA receptors are crucial for memory
Received 8 July; accepted 15 September 2009. acquisition of one-time experience. Neuron 38, 305315 (2003).
27. Kentros, C. G., Agnihotri, N. T., Streater, S., Hawkins, R. D. & Kandel, E. R.
1. Moser, E. I., Kropff, E. & Moser, M. B. Place cells, grid cells, and the brains spatial Increased attention to spatial context increases both place field stability and
representation system. Annu. Rev. Neurosci. 31, 6989 (2008). spatial memory. Neuron 42, 283295 (2004).
2. OKeefe, J. & Burgess, N. Dual phase and rate coding in hippocampal place cells: 28. Cacucci, F., Wills, T. J., Lever, C., Giese, K. P. & OKeefe, J. Experience-dependent
theoretical significance and relationship to entorhinal grid cells. Hippocampus 15, increase in CA1 place cell spatial information, but not spatial reproducibility, is
853866 (2005). dependent on the autophosphorylation of the alpha-isoform of the calcium/
3. OKeefe, J. & Dostrovsky, J. The hippocampus as a spatial map. Preliminary calmodulin-dependent protein kinase II. J. Neurosci. 27, 78547859 (2007).
evidence from unit activity in the freely-moving rat. Brain Res. 34, 171175 (1971). 29. Sun, L. D. & Wilson, M. A. Impaired and Enhanced Spatial Representations of the
4. OKeefe, J. & Recce, M. L. Phase relationship between hippocampal place units PSD-95 Knockout Mouse. PhD thesis, Massachusetts Institute of Technology
and the EEG theta rhythm. Hippocampus 3, 317330 (1993). (2003).
5. Skaggs, W. E., McNaughton, B. L., Wilson, M. A. & Barnes, C. A. Theta phase 30. Margrie, T. W., Brecht, M. & Sakmann, B. In vivo, low-resistance, whole-cell
precession in hippocampal neuronal populations and the compression of recordings from neurons in the anaesthetized and awake mammalian brain.
temporal sequences. Hippocampus 6, 149172 (1996). Pflugers Arch. 444, 491498 (2002).
6. Buzsaki, G. Theta oscillations in the hippocampus. Neuron 33, 325340 (2002). 31. Crochet, S. & Petersen, C. C. Correlating whisker behavior with membrane
7. Tsodyks, M. V., Skaggs, W. E., Sejnowski, T. J. & McNaughton, B. L. Population potential in barrel cortex of awake mice. Nature Neurosci. 9, 608610 (2006).
dynamics and theta rhythm phase precession of hippocampal place cell firing: a 32. Wilson, M. A. & McNaughton, B. L. Dynamics of the hippocampal ensemble code
spiking neuron model. Hippocampus 6, 271280 (1996). for space. Science 261, 10551058 (1993).
8. Jensen, O. & Lisman, J. E. Hippocampal CA3 region predicts memory sequences: 33. Guzowski, J. F., McNaughton, B. L., Barnes, C. A. & Worley, P. F. Environment-
accounting for the phase precession of place cells. Learn. Mem. 3, 279287 specific expression of the immediate-early gene Arc in hippocampal neuronal
(1996). ensembles. Nature Neurosci. 2, 11201124 (1999).
9. Wallenstein, G. V. & Hasselmo, M. E. GABAergic modulation of hippocampal 34. Henze, D. A. et al. Intracellular features predicted by extracellular recordings in
population activity: sequence learning, place field development, and the phase the hippocampus in vivo. J. Neurophysiol. 84, 390400 (2000).
precession effect. J. Neurophysiol. 78, 393408 (1997). 35. Kandel, E. R. & Spencer, W. A. Electrophysiology of hippocampal neurons. IV. Fast
10. Kamondi, A., Acsady, L., Wang, X. J. & Buzsaki, G. Theta oscillations in somata and prepotentials. J. Neurophysiol. 24, 272285 (1961).
dendrites of hippocampal pyramidal cells in vivo: activity-dependent phase- 36. Wong, R. K. & Prince, D. A. Participation of calcium spikes during intrinsic burst
precession of action potentials. Hippocampus 8, 244261 (1998). firing in hippocampal neurons. Brain Res. 159, 385390 (1978).
11. Magee, J. C. Dendritic mechanisms of phase precession in hippocampal CA1 37. Hafting, T., Fyhn, M., Bonnevie, T., Moser, M. B. & Moser, E. I. Hippocampus-
pyramidal neurons. J. Neurophysiol. 86, 528532 (2001). independent phase precession in entorhinal grid cells. Nature 453, 12481252
12. Mehta, M. R., Lee, A. K. & Wilson, M. A. Role of experience and oscillations in (2008).
transforming a rate code into a temporal code. Nature 417, 741746 (2002).
Supplementary Information is linked to the online version of the paper at
13. Harris, K. D. et al. Spike train dynamics predicts theta-related phase precession in www.nature.com/nature.
hippocampal pyramidal cells. Nature 417, 738741 (2002).
14. Lengyel, M., Szatmary, Z. & Erdi, P. Dynamically detuned oscillations account for Acknowledgements We thank E. Chaffin for help with mouse behaviour,
the coupled rate and temporal code of place cell firing. Hippocampus 13, 700714 J. Carmack and id Software for providing the Quake2 code, A. Shishlov for
(2003). programming advice, G. Buzsaki, J. Magee, H. Dahmen and D. Markowitz for
15. Gasparini, S. & Magee, J. C. State-dependent dendritic computation in discussions, and C. Brody, M. Berry and E. Civillico for comments on the
hippocampal CA1 pyramidal neurons. J. Neurosci. 26, 20882100 (2006). manuscript. This work was supported by the NIH (1R01MH083686-01,
16. Maurer, A. P. & McNaughton, B. L. Network and intrinsic cellular mechanisms 5R01MH060651-09), a Helen Hay Whitney Fellowship (to C.D.H.), and a
underlying theta phase precession of hippocampal neurons. Trends Neurosci. 30, Patterson Trust Fellowship (to D.A.D.).
325333 (2007).
Author Contributions C.D.H. performed behaviour and intracellular recording
17. Lee, A. K., Manns, I. D., Sakmann, B. & Brecht, M. Whole-cell recordings in freely
experiments with technical assistance from D.A.D. C.D.H. and D.A.D. performed
moving rats. Neuron 51, 399407 (2006).
extracellular recording experiments. F.C., D.A.D. and D.W.T. designed, and C.D.H.,
18. Lee, A. K., Epsztein, J. & Brecht, M. Head-anchored whole-cell recordings in freely
F.C. and D.W.T. implemented, the virtual-reality instrumentation. F.C. performed
moving rats. Nature Protocols 4, 385392 (2009).
virtual-reality software development. C.D.H. analysed all data with strategy and
19. Dombeck, D. A., Khabbaz, A. N., Collman, F., Adelman, T. L. & Tank, D. W. Imaging methods contributions from all authors. C.D.H. and D.W.T. wrote the paper.
large-scale neural activity with cellular resolution in awake, mobile mice. Neuron
56, 4357 (2007). Author Information Reprints and permissions information is available at
20. Holscher, C., Schnee, A., Dahmen, H., Setia, L. & Mallot, H. A. Rats are able to www.nature.com/reprints. Correspondence and requests for materials should be
navigate in virtual environments. J. Exp. Biol. 208, 561569 (2005). addressed to D.W.T. (dwtank@princeton.edu).
946
554
2009 Macmillan Publishers Limited. All rights reserved
doi:10.1038/nature08499
METHODS of the Styrofoam ball to move from one end of the track to the other times the
circumference of the ball. The effective width of the track (that is, the distance the
Virtual-reality set-up. The virtual-reality system (Fig. 2a and Supplementary
mouse could actually move) was ,1 cm, rather than 9 cm, because in Quake2 the
Fig. 1) was designed around an air-supported spherical treadmill (8-inch dia-
player is surrounded by a bounding box that gives him a fixed width. The track
meter Styrofoam ball) for head-restrained mice that was previously used for in
had short proximal walls with different textures for each third of the track
vivo optical imaging at cellular resolution19. A projection-based visual display,
(060 cm: white with black dots, 61120 cm: vertical white and black stripes,
similar to the system used in a body-tethered rat virtual-reality system20, pre-
121180 cm: black with white dots). The proximal walls at the ends of the track
sented a computer rendered scene with a 270u horizontal field of view and a
were green with black dots to mark the reward zones. Tall distal walls were
vertical field from 160u to 220u relative to the mouses head. The toroidal screen
positioned at 60 cm (horizontal black and white stripes) and 120 cm (green with
displayed the two-dimensional image from a Mitsubishi HC3000 digital light
black crosses). The floor and ceiling were black throughout the track. Green was
processing projector reflected off of a 15 cm diameter round mirror and a 15 cm
selected as the only non-black or white colour because electrophysiological and
diameter angular amplification mirror (AAM)21 with an angular amplification
behavioural data indicate that mice can detect wavelengths of ,500 nm42. Mice
factor of 12. The AAM was machined out of aluminium using a CNC machine
received small water rewards (4 ml) for running between reward zones positioned
and then polished by hand. The screen was ,46 cm tall and ,80 cm in outer
at opposite ends of the track (zone 1 postion: 09 cm; zone 2 position: 171
diameter, and was constructed out of a semi-reflective nylon material (Rose
180 cm). After receiving a reward at one reward zone, the mouse then had to
Brand fabrics) supported by a frame of stainless steel rods. Water rewards were
run to the other reward zone to receive the next reward; two consecutive rewards
delivered by a solenoid valve (NResearch) attached to a water-feeding tube
could not be obtained from a single reward site. Linear track behavioural data
(Popper and Sons) positioned directly in front of the animals mouth such that
(Fig. 2ce) were from mice trained only on the linear track. Some animals used for
the mouse could lick the meniscus. Rotations of the Styrofoam ball were
electrophysiology recordings were trained in other virtual environments before
measured by an optical computer mouse (Logitech MX1000) that was posi-
training in the linear track. During electrophysiology experiments, behavioural
tioned below the field of view in front of the animal at the point where the balls
performance was in some cases degraded due to satiation from water rewards and
equator intersected the animals rostral-caudal body axis. A computer running because the visual display was turned off while patch pipettes were changed.
Labview used USB-communicated signals from the optical mouse to compute
Electrophysiology. After at least five training sessions on the virtual linear track,
low-pass filtered ball rotational velocity around both the horizontal axis (per-
a small craniotomy (,0.5 mm diameter) was made over the left hippocampus
pendicular to the body axis) and vertical axis. These velocities were then output
(2.2 mm caudal, 21.7 mm lateral to bregma). The dura was left intact for both
as analogue control voltages using a D/A converter (National Instruments) and
extracellular and intracellular recordings. Because behavioural performance was
used as input control signals to a separate computer running the virtual-reality
degraded on the day of surgery and anaesthesia, electrophysiology recordings
software.
were performed starting the next day. To preserve the craniotomy across days, it
The virtual-reality software we developed was on the basis of the open source was covered with silicone grease (Dow Corning) and then with a layer of silicone
Quake2 game engine (id software), using code ported to Visual Studio 2008 elastomer (Kwik-Sil, World Precision Instruments) until the time of recording.
(Microsoft). The rendering engine was modified, using cube map texturing in Extracellular recordings (Fig. 3 and Supplementary Fig. 10) were made using a
OpenGL graphics38, to display a fisheye transform of the perspective of the single glass electrode filled with 0.5 M NaCl (,2.5 MV pipette resistance). The
virtual player, such that the image reflected off the AAM and displayed on the electrode was mounted vertically (perpendicular to the surface of the brain) on a
screen had the geometrically correct perspective for the mouse on the ball. micromanipulator (Sutter MP285) that was positioned behind the mouse and
Software to control the water reward system, input information on ball rotation thus outside the mouses field of view. The reference electrode was positioned
velocity, and output voltages proportional to position and view angle within the outside the craniotomy in extracellular saline containing (in mM): 150 NaCl,
environment was also developed by adding A/D and D/A control to the game 2.5 KCl, 10 HEPES, pH 7.4. Signals were electronically filtered between 500 Hz
engine, using National Instruments multifunction cards. Ball velocity around the and 7.5 kHz and digitized at 20 kHz. For measurements of the LFP and phase
vertical axis was used as a control signal for changing view angle in the virtual precession, signals were electronically filtered between 2 Hz and 10 kHz; spikes
environment, whereas velocity around the horizontal axis was used as a control and the LFP were recorded on the same electrode. The position of the top of the
signal for forward and backward movement. Other open source gaming software brain was noted as a large resistance increase when the electrode made contact
(Quark, http://quark.planetquake.gamespy.com/) was used to build the virtual with the dura. As the electrode was advanced through cortex, spikes were present
linear track. To synchronize behaviour data with electrophysiology recordings, in each layer. Upon entering the external capsule, a change in the recording was
an independent computer with a Digidata 1440A interface running Clampex noted, especially from changes in sound quality using an audio monitor. The
software (Molecular Devices) digitized and stored real time information about CA1 cell body layer was ,200 mm beneath the external capsule and was char-
the mouses location in the virtual environment together with solenoid (water acterized by strongly theta-modulated spiking. Putative pyramidal neurons were
reward) control signals, ball rotational velocities, and electrophysiological data. found as units firing complex bursts separated at theta frequencies22. In all
Behavioural training. Eight-totwelve-week-old male C57BL/6J mice (Jackson animals CA1 recordings were made at a depth of ,1.1 mm; cells were found
Labs) were used for all experiments. C57BL/6J mice were selected because they reproducibly at this depth across several electrodes and days. To confirm that
have good vision compared to other inbred strains39 and are a common back- this recording position was in the CA1 cell body layer, in a separate experiment
ground strain for transgenic mice. All experimental procedures were performed we electroporated Alexa 488 dextran (5% (w/v) in extracellular saline, 4 mA
in compliance with the Guide for the Care and Use of Laboratory Animals pulses, 25 ms pulse duration, 600 pulses at 2 Hz) using a recording pipette and
(http://www.nap.edu/readingroom/books/labrats/) and were approved by found bright labelling of CA1 cell bodies in histology sections. At this depth, we
Princeton Universitys Institutional Animal Care and Use Committee. Before searched for well-isolated units of large amplitude and recorded their activity for
behavioural training, titanium head plates with a large central opening (2.5 cm ,1530 min, which was enough time for the mouse to sample thoroughly the
wide 3 0.9 cm long 3 0.08 cm thick; central opening: 0.89 cm wide 3 0.61 cm virtual linear track. Spikes were sorted offline using a threshold analysis. At most
long) were implanted on mice and affixed to the skull using dental cement one unit was isolated from a single recording. To check the quality of unit
(Metabond, Parkell). Sites for future craniotomies over the left and right dorsal isolation, we overlaid all spike waveforms to make sure that they matched by
hippocampi were marked using stereotactic coordinates (2.2 mm caudal, visual inspection (Fig. 3a), and we plotted ISI distributions to make sure no
1.7 mm lateral to bregma). After head plate implantation, mice were placed on spikes fell within the refractory period (,12 ms; Fig. 3b). Extracellular record-
a water schedule in which they received 1 ml of water per day. Body weights were ings were made from a total of 8 mice. In some cases a second craniotomy was
checked to ensure mice were ,80% of their pre-water-restriction weight40,41. made over the right hemisphere to extend the number of recording sessions from
After at least 5 days of water scheduling, behaviour training began. In each a single animal. Recordings were made for up to 4 days from the same cranio-
training session, mice were placed on the experimental apparatus with their head tomy. Place cells were found in every mouse tested.
fixed in place. The head was centred over the middle of the Styrofoam ball with Whole-cell recordings were made using the same experimental set-up as for
the headplate ,2.6 cm from the top of the ball. A lick tube to deliver water extracellular recordings. Pipettes (,57 MV) were filled with internal solution
rewards was positioned in front of the mouses mouth. The water rewards earned containing (in mM): 135 K-gluconate, 10 HEPES, 10 Na2-phosphocreatine,
during behaviour were included in the mouses daily water allotment such that a 4 KCl, 4 MgATP, 0.3 Na3GTP (pH 7.25 with KOH, 285 mOsm). We pulled pipettes
mouse received exactly 1 ml of water per day. Each training session (one per day) with a long taper (,100 mm diameter at 1 mm from the tip) to minimize damage to
lasted 45 min with the virtual-reality system turned on for all sessions. The first the overlying cortical tissue and to reduce compression of the tissue while advancing
several sessions mostly involved acclimation to the apparatus and learning to run the pipette. The pipette was mounted vertically on a standard micromanipulator
on the spherical treadmill. (Sutter MP285) positioned outside the mouses field of view; special methods to
Mice were trained to perform behaviours along a virtual linear track. The anchor the pipette17,18 were not necessary. We used standard blind patch methods to
virtual track was 180 cm long and 9 cm wide, measured as the number of rotations obtain whole-cell recordings30. In brief, we applied ,250 mbar of positive pressure
555
2009 Macmillan Publishers Limited. All rights reserved
doi:10.1038/nature08499
while moving the pipette through the cortex and reduced the pressure to ,15 mbar To quantify the average baseline membrane potential as a function of position
while searching for cells in the hippocampus. The CA1 cell body layer was located along the virtual track (Fig. 4c and Supplementary Fig. 2b), we first removed the
using a depth coordinate obtained from extracellular recordings in the same animal. contribution of action potentials by discarding all time points between 4 ms before
We attempted to form a seal when we observed large (,50%), reproducible and 7 ms after an action potentials peak. We then created a map of membrane
increases in pipette resistance. Recordings were made in current clamp mode with potential values by grouping the membrane potentials into 80 spatial bins along the
no holding current. Membrane potential values were corrected for liquid junction track and calculating an average membrane potential for each bin. The membrane
potentials. We obtained recordings lasting longer than 3 min from 46 cells from 15 potential map was smoothed using a five point Gaussian window with a standard
mice, with a peak success rate of approximately one such recording in every five deviation of one. Large, long-lasting depolarizations of the baseline membrane
attempts. Recordings were abandoned when we observed large increases in resting potential occurred infrequently outside the place field (2.3 6 1.3 events per record-
potential or large decreases in spike amplitude. We obtained whole-cell recordings ing for depolarizations of greater than 2 mV lasting longer than 0.5 s).
for up to 3 days from the same craniotomy. Whole-cell recordings from place cells To analyse ramps of depolarization during complete runs through the place
lasted on average 7.1 6 2.8 min and had series resistances of 48 6 16 MV, input field (that is, the mouse did not turn around inside the place field), we calculated
resistances of 98 6 23 MV, and resting membrane potentials of 267 6 4 mV. Firing changes in membrane potential (DV) after action potentials were removed (Fig. 4e
rates from place cells recorded intracellularly tended to be higher than those and Supplementary Figs 5 and 6b). We defined the baseline membrane potential
recorded extracellularly, both inside and outside the place field, potentially because as the mean membrane potential from 500 ms to 1 s before the mouse entered the
a fraction of spikes measured extracellularly were lost during spike sorting. All other place field and subtracted this baseline from all membrane potential values during
place-cell parameters (for example, field size, ISI distribution) were similar between the run to get DV values. The peak DV during a run through the place field
cells recorded extracellularly and intracellularly. Although we cannot exclude (Supplementary Fig. 6b) was calculated after smoothing the DV values over a
possible effects of dialysis of the cell during whole-cell measurements, such effects sliding window of length 50 ms such that the peak value was representative of the
are probably small owing to short recording durations and relatively high series ramp-like depolarization rather than brief, high frequency depolarizations. To
resistances. In experiments to assess the connection between ramps of depolariza- analyse the symmetry of firing rates and ramps of depolarization in place fields
tion and changes in theta oscillation amplitude (Supplementary Fig. 7), ramps of (Fig. 4e), we considered runs starting half the place fields width before the place
current (4 s duration, 11.5 nA at the peak) were injected at the soma after 11 s field and ending half the place fields width beyond the place field (total length was
without current injection. Only trials in which the animal was running at greater twice the length of the place field). Because place fields varied in size between cells
than 10 cm s21 were analysed. and runs through the field differed in duration, we divided each run into 60
To perform whole-cell and extracellular recordings simultaneously (Fig. 5f, g equally sized spatial bins and calculated the mean firing rates and DV for each
and Supplementary Fig. 11), we mounted two glass electrodes on separate micro- 1/60th of the run (that is, 15 spatial bins before the place field, 30 bins in the field,
manipulators, which were both positioned behind the mouse, and advanced the and 15 bins after the field). All runs through the field were plotted with increasing
electrodes independently through separate craniotomies on the same hemi- track position values so that runs in opposite directions could be combined.
sphere. Whole-cell recordings were made at the standard coordinates as To analyse subthreshold theta oscillations (Fig. 5 and Supplementary Figs 2c, 7
described earlier. The extracellular recording electrode (filled with 0.5 M and 9), spikes were removed over a window of 3 ms preceding and 5 ms after the
NaCl) was mounted at a 40u angle relative to the midline and a 45u angle with peak and were replaced using linear interpolation. The resulting trace was then
respect to the vertical. The craniotomy for the extracellular electrode was posi- band-pass filtered between 610 Hz (a peak in the ISI distribution) using a linear
tioned caudal and either lateral for the left hemisphere or medial for the right phase finite impulse response (FIR) filter with a Hamming window of width 1 s
hemisphere relative to the craniotomy for intracellular recordings. The tip of the (Matlab function fir1). To create a map of theta power along the virtual track,
extracellular electrode was positioned at a depth of ,1.2 mm and ,250 mm power was calculated as the mean of the squared amplitude for a sliding window
caudal to the centre of the craniotomy for whole-cell recordings. The extracel- of length 1 s for the entire recording, with the centre of the window providing the
lular recordings were filtered between 1 Hz and 10 kHz. position along the track. The power values were grouped into 80 spatial bins
From our 46 hippocampal whole-cell recordings, 41 cells were spontaneously along the track and smoothed using a five point Gaussian window with a stand-
active (overall firing rate .0.05 Hz). We classified 7 of the spontaneously active ard deviation of one. Power spectra for epochs in and out of the place field were
cells as interneurons on the basis of high overall firing rates (.5 Hz) and the obtained using multi-taper spectral analysis methods (Chronux toolbox, http://
absence of complex spike bursts. The remaining 34 spontaneously active cells chronux.org; Supplementary Fig. 8). Running speed, which can influence theta
were categorized as pyramidal neurons. To estimate the fraction of sponta- oscillations43, was similar inside and outside the place field (in-field
neously active pyramidal neurons that had a place field along the virtual linear speed 5 49 6 6 cm s21, out-of-field speed 5 47 6 4 cm s21, P . 0.7). LFP theta
track, we considered only those recordings during which the animal visited each oscillations were present when the animal was running (Fig. 3d and
of the 80 spatial segments at least 3 times, with visits separated by at least 2 s in Supplementary Fig. 10) and absent when the animal was resting.
time. We estimate that 36% of spontaneously active pyramidal neurons had To analyse phase precession (Figs 3d, e and 5dg), we considered complete runs
place fields in the virtual environment (8 out of 22 cells). During whole-cell through the place field (that is, the animal did not turn around in the field) that had
place-cell recordings, the place field was visited 20 6 8 times on average, with at least 5 spikes and in which the animal ran faster than 10 cm s21. Only cells with
visits separated by at least 2 s in time. more than 40 total spikes in the place field were included. Extracellular or intra-
Data analysis. To create firing rate maps (Figs 3c, 4b and Supplementary Fig. 2a), cellular (after removing spikes) voltage traces were band-pass filtered between
we divided the virtual linear track into 80 bins (2.25 cm each) and calculated the 610 Hz using an FIR filter. To assign a phase to a spike occurring at time t, we
firing rate as the total number of spikes in a bin divided by the total amount of identified, in the filtered trace, the times of the peaks immediately preceding and
time spent in a bin. The maps were smoothed using a five point Gaussian window following the spike (t1 and t2, respectively) and calculated the phase as 360(t 2 t1)/
with a standard deviation of one. Periods in which the mouse ran slower than (t2 2 t1). We circularly shifted the phase of the spikes in 1u steps from 0u to 360u,
5 cm s21, averaged over a 2-s window, were removed from the analysis. To continuing across the 360u border, and fit a linear regression line to the phase
identify place fields, we found groups of adjacent bins with firing rates greater versus position plot at each rotation4,37. We found the rotation with the best fit,
than 0.25 times the rate in the peak bin. We selected only those fields that were such that the sum of squared errors between the fit line and data was minimized,
larger than 8 bins (18 cm) in length, had mean in-field firing rates of greater than and used the correlation coefficient between phase and position at this rotation as a
1.5 Hz, and had mean in-field firing rates more than 3 times larger than the mean measure of phase precession. All phase values are from the raw data in relation to
out-of-field firing rate. To verify that place fields were a statistically significant theta phase, with 0u indicating the peak. To measure the precession of intracellular
increase in firing rate, we performed a bootstrap shuffle test. We shuffled the theta relative to extracellular theta (Fig. 5g), we first identified the times of the
times of spikes in 10 s segments, calculated new firing rate maps using the peaks of the filtered (610 Hz) membrane potential trace during runs through the
unshuffled bin dwell times, and then checked for a place field using the above- place field. For each peak of intracellular theta, we found the corresponding phase
stated criteria. This procedure was repeated 1,000 times per place-cell recording. of the simultaneously recorded extracellular theta and the position of the mouse
We called a cell a place cell only if a place field was found in the unshuffled data along the virtual track and followed the same procedures that were used for the
and in fewer than 2% of the shuffled tests (P , 0.02). Across the place cells we analysis of the phase precession of spike times. To quantify the change in times
recorded, we found place fields at all positions along the linear track. Spikes were between intracellular theta peaks and LFP theta peaks during place field traversals
separated into leftward and rightward runs based on the head direction of the (Supplementary Fig. 11a), we found the time difference between the first LFP theta
animal with regard to the virtual track (measured as the players view angle in peak and the first intracellular theta peak in the place field, the time difference
Quake2). A directionality index for firing based on firing rates in leftward and between the second LFP theta peak and the second intracellular theta peak, and so
rightward directions (FRleft, FRright) was defined as jFRleft 2 FRrightj/ on. To quantify frequency differences between intracellular and LFP theta
(FRlef t1 FRright). A directionality index of 0 indicates identical firing in both (Supplementary Fig. 11b), we calculated the ratio of the period of the first intra-
directions, whereas an index of 1 indicates firing in one direction only. cellular oscillation to the period of the first LFP oscillation, the ratio of the period of
556
2009 Macmillan Publishers Limited. All rights reserved
doi:10.1038/nature08499
the second intracellular oscillation to the period the second LFP oscillation, and so 39. Wong, A. A. & Brown, R. E. Visual detection, pattern discrimination and visual
on using peaks from the filtered (610 Hz) traces. We determined these ratios for acuity in 14 strains of mice. Genes Brain Behav. 5, 389403 (2006).
complete runs through the place field and 3 s long epochs outside the place field. 40. Rinberg, D., Koulakov, A. & Gelperin, A. Sparse odor coding in awake behaving
mice. J. Neurosci. 26, 88578865 (2006).
Data are presented as mean 6 s.d. unless noted otherwise. P values are from 41. Huber, D. et al. Sparse optical microstimulation in barrel cortex drives learned
two-tailed t-tests unless stated otherwise. Correlation coefficients (C) are from behaviour in freely moving mice. Nature 451, 6164 (2008).
Pearsons correlations. 42. Jacobs, G. H., Neitz, J. & Deegan, J. F. II. Retinal receptors in rodents maximally
sensitive to ultraviolet light. Nature 353, 655656 (1991).
38. Greene, N. Environment mapping and other applications of world projections. IEEE 43. Buzsaki, G., Leung, L. W. & Vanderwolf, C. H. Cellular bases of hippocampal EEG in
Comput. Graph. Appl. 6, 2129 (1986). the behaving rat. Brain Res. 287, 139171 (1983).
557
2009 Macmillan Publishers Limited. All rights reserved
558
J. Physiol. (1965), 178, pp. 477-504 477
With 11 text-ftgureas
Printed in Great Britain
559
478 H. B. BARLOW AND W. R. LEVICK
particular direction of motion are fed by a subset of bipolar cells that
respond to the corresponding sequence of excitation of two neighbouring
retinal regions with which they connect. Furthermore there is evidence
that this sequence-discrimination is brought about by a laterally connect-
ing inhibitory element from one of these regions, and this seems a likely
function for the horizontal cells to perform. At this stage identification of
the elements concerned is obviously tentative, but we were pleasantly
surprised to find well known histological structures already at hand to fill
the roles that the functional organization seemed to require.
All the experiments described here were performed upon on-off
directionally selective units. We have reason to believe that the mechanism
is different for the rare, on-type, directionally selective units and also for
the centrifugal and centripetal motion sensitivity of the ordinary con-
centric type of units.
METHODS
Action potentials were recorded from the unopened eyes of rabbits. As described else-
where (Barlow et al. 1964), it proved most effective to use fine tungsten electrodes, decere-
brate or lightly anaesthetized (urethane and chloralose mixture) animals, and immobilization
by continuous infusion of gallamine triethiodide (Flaxedil). Periodically the animal was
allowed to recover from paralysis by using an infusion fluid without relaxant. One could
thus ensure that the level of anaesthesia was neither too deep nor too light at the rate of
anaesthetic infusion employed.
When a good on-off, directionally selective unit was isolated the first step was always to
map out its receptive field on a plotting board 57 cm from the eye. A stationary spot was
turned on and off; in most cases this was I' diameter at an intensity of 12 cd/M2 and was
superimposed on a background of 0-6 cd/iM2. Directional selectivity was tested for and null
and preferred directions determined. Various techniques were used to provide the temporal
and spatial patterns of light stimuli. For the two-spot experiment we initially used two
glow modulator tubes controlled by pulse generators, but we later found that black and
white cards moved behind apertures in grey paper provided a more flexible means of
delivering the required stimuli. This method also has disadvantages (see p. 486) which we
finally overcame by illuminating the apertures from behind with thin Perspex light pipes lit
by low-current torch bulbs turned on and off manually. The changes of lulminance occurring
within the receptive field of the unit were monitored by a photocell whose output was
recorded with the action potentials.
RESIULTS
The results are presented in four sections. These are: (1) controls and
negative results which rule out various preliminary hypotheses on the
mechanism; (2) experiments which lead to the conclusion that the direc-
tional selectivity of the ganglion cell results from the sequence-discrimina-
ting activity of subunits-probably bipolar cells; (3) observations and
experiments which show that sequence-discrimination is achieved by an
inhibitory mechanism that prevents responses to sequences in the null
direction; (4) observations showing that inhibition also occurs with stimu-
lation of the surround of the receptive field.
560
MECHANISM OF DIRECTIONAL SELECTIVITY 479
Controls and negative results
Optical controls. It might be thought that the unequal responses to
motion in the null and preferred directions were the result of a peculiarity
of the light distribution in the retinal image caused by aberrations of the
optical system. Although none of the schemes suggested to us, and none
we could imagine, seemed at all promising, we considered this possibility,
but were forced to abandon it at an early stage. The most direct disproof is
given by the observation that two units recorded simultaneously, or
within a short period of time, in the same retinal region can have their
preferred axes opposite or at right angles to each other. An optical
explanation of the phenomenon requires that there be some asymmetry in
the light distribution to cause the asymmetry in the responses, and two
different asymmetrical light distributions cannot co-exist in the same
region.
A second disproof is provided by observing how little the phenomenon is
affected by deliberately introduced optical aberrations. Figure 1 shows that
clear-cut directional selectivity persists with spherical supplementary lenses
causing more than 10 dioptres ofrefractive error in either direction. This was
for the normal pupil diameter of our preparation-6 mm or more. Further-
more, reducing the aperture of the rabbit's optical system to 3 or 125 mm
with an artificial pupil must greatly improve retinal image quality (since
diffraction is most unlikely to be a limiting factor); yet we found that such
stopping down merely extended the range of refractive error over which
the directionally selective property was shown. It did not even improve
acuity as judged by the finest grating giving any response to movement.
The best spherical correction was determined in each preparation,
judging this opthalmoscopically and sometimes retinoscopically. In all the
preparations the cornea and media were free from clouding, except
occasionally at the very end of a 2-day experiment. Optical effects from
the shank of the electrode can be excluded, for directional selectivity is
often observed when recording from a nerve fibre. In such cases, no
portion of the electrode intercepts light reaching the retina from the
receptive field projection.
Finally, we should point out that for most of the tests used here the
human eye's performance is superior by a factor of 10. The demands made
of the rabbit's optical system are thus not at aLl severe, but of course the
blur of the retinal image should be taken into account when interpreting
quantitatively the results of our later experiments.
Latencies. The first idea about mechanism was that the latency of the
response might be shorter at successive positions along a path through the
receptive field traced in the preferred direction. It was thought that when
561
480 H. B. BARLOW AND W. R. LEVICK
the image of an object moved in this direction the excitation from succes-
sive positions would arrive synchronously at the ganglion cell, and thus
might be more effective than when movement was in the null direction and
it was dispersed in time. An alternative scheme can be devised in which
the temporally dispersed sequence is more effective because it avoids
Unit 4-40
o0 I 0
45' I
0\~~~~~~~~~
30' -//\ 4
15'_
562
MECHANISM OF DIRECTIONAL SELECTIVITY 481
Unsuccessful two-spot experiments. One sees from the records of Fig. 2
that excitation of a point in the field causes a response at on and off: yet
if a spot of light is moved through the field in the null direction no response
occurs, even though each point crossed by the spot must have received
an ' on' followed by an 'off' stimulus as the spot passed over it. The obvious
next step in the analysis seemed to be to stimulate at two points and see
how the response changed when the order and temporal interval between
the stimuli was varied. We hoped to decide whether the response to
Background 7 cd/M2 Position 45S 150A in visual field
Spot 50 cd/m2 aII- I -1I
Preferred
s
+ 0
;* ~~~~~c
Nul
X 0s 0i I I
Unit 5-26
Iw-'-- ---|i 0-25 sec
Fig. 2. Latencies of response at different positions. The receptive field map is
shown at the left; positions yielding on and off responses to a stationary spot are
marked + and -; positions yielding both are marked T if off was greater, if
on was greater; no responses were obtained on or outside the ring of O 's. Records
a, c, e, are the responses to a light turned first off, then on, at three successive posi-
tions along a line through the receptive field in the preferred-null axis, as shown to
the left (negativity is upwards). Records b, d, f, are from a photomultiplier
observing the receptive field (decreased light is downward), and the numbers show
the latency of response in msec (one impulse was ignored in a). There is no signifi-
cant trend in latency as one moves across the receptive field.
motion in the preferred direction was greater than the sum of responses to
excitation of separate points along the path, or whether the unequal
responses to motion in the two directions resulted from inhibition occurring
when the motion was in the null direction.
The question seemed clear-cut, but the results were not. In the first
place it was not nearly as easy to obtain unequal responses for the two
sequences as it was to obtain unequal responses with real moving objects.
Secondly, when we did get evidence of sequence-dependent responses, the
result seemed highly variable and we were unable to decide whether
summation or inhibition or both were occurring. This failure forced us to
31 Physiol. 178
563
482 H. B. BARLOW AND W. B. LEVICK
realize that we did not know whereabouts in the receptive field we should
put the spots, nor how far apart they should be. There was in fact a prior
question to answer before the two-spot type of experiment could be
performed and interpreted. The question, put in a form that avoids
implications as to mechanism, is this: does the ganglion cell respond
selectively to one direction of motion over all parts of its receptive field, or
is there some critical zone or line which must be crossed? The following
observations show that there is no such line and the directionally selective
property is distributed over the receptive field.
Background 7 cd/Mr2 Position 450S SOP
Spot 40 cd/M2 T Spotsge in visual field
Sposi.zO
[ 10 0MIWIi _tin;t
M&Uui.
ii02
o _,,____________
50[>
r~~ ~~~~~~~~ -e
Sequence-discrimination by subunits
Distribution of directional selectivity. Figure 3 shows the responses
obtained when a spot of light was moved across the receptive field and
back along three parallel lines. These were separated by more than the
breadth of the spot, and therefore different receptors were covered by the
geometric image of the moving spot in each case. It will be seen that the
selectivity clearly exists along all these three pathways.
Figure 4 shows typical responses obtained when the spot was moved
several times from one position to the next and back, as marked. It is
clearly not necessary for the spot to cross any definite line in order to
obtain different responses for the two directions of motion. If the experi-
ment is repeated using a black spot the same result is obtained; direction
564
c_h | m-_ ~ n
MECHANISM OP DIRECTIONAL SELECTIVITY 483
of motion, independent of contrast, can be picked out in a large number of
widely separated regions of the receptive field. However, there is an
interesting exception to the rule that aU regions of the receptive field have
the capacity to distinguish between null and preferred sequences of
excitation of the receptors they contain. There is a zone adjacent to the
edge of the field that is first crossed when motion is in the preferred
b ~b
____b fgf__gf k k
b b
c g h g m
c d c . mm
Fig. 4. Back and forth motion in different parts of the receptive field. The edge of
the receptive field is mapped at the top, and the positions a, b, c, . . . o, within it are
indicated. The spot was moved back and forth several times between a and b, then
between b and c, and so on. The records are samples of these back and forth
motions. The lower trace of each pair shows the position of the spot in the field:
downward movement of the trace corresponds to movement of the spot in the
preferred direction. Marked asymmetry of response for the opposite directions
holds in most positions in the field. Its absence in the top row of records is
expected in the inhibitory scheme (see Fig. 7 and p. 490).
565
484 H. B. BARLOW AND W. R. LEVICK
smallest distance over which responses to motion in the two directions
differ?
To answer this question we moved a strip of white card with a section
painted black behind an aperture of variable width in a sheet of grey
paper. The aperture breadth was 20 mm (subtending 20) and the width
measured in the direction in which the card moved varied from 1 mm (6')
to 11 mm (10 06'). When the card was moved there was no change until the
border of the black section entered the aperture, it then moved through a
variable distance before it disappeared behind the other edge, after which
there was again no further change and the aperture was wholly black. The
sequence described above was called 'off' stimulation since with black
trailing the receptors were successively exposed to a reduction of illumina-
tion. On the other hand, when the border of the white portion appeared in
TABLE 1. Single slit experiment
A black-white border was moved through the receptive field, but the view of the motion
was restricted by a fixed rectangular slit in a grey card placed immediately in front of the
moving border. The width of this slit, measured in the direction of motion, was varied.
Movement of the border causing the slit to change from white to black was called 'off'
stimulation, movement causing it to fill with white was called 'on' stimulation. Each of
these stimuli was applied with the edge moving in both null and preferred directions. We
graded the unit's response from 0 to 6 after listening to several repetitions of these stimuli on
the loudspeaker. This unit could distinguish the preferred from the null direction with slit
widths down to about 17' for both 'on' and 'off' stimuli.
'Off' stimulus 'On' stimulus
Slit width ,A_ A __,
Preferred Null Preferred Null
1 36' 6 1 4 0
106' 6 2 4 1
48' 5 2 4 0
34' 4 2 3 1
24' 4 2 3 1
17' 3 2 3 2
12' 3 3 3 2
8' 2 2 2 2
the aperture and moved along it they were successively exposed to 'on'
stimulation. Clearly 'off' and 'on' stimulation could be applied in any
direction and at any velocity, but we confined our attention to the pre-
ferred and null directions, and moved the cards by hand at velocities
chosen to give optimum discrimination between the two directions-in
most cases about 5/sec. The variable studied was the width of the aper-
ture, and the feature of the response attended to was the existence of a
clear-cut difference between responses to stimulation in the preferred and
null directions.
The amplitude of the response was graded subjectively from 0 to 6, and
Table 1 gives a typical result. The subjective grading was a crude but con-
566
MECHANISM OF DIRECTIONAL SELECTIVITY 485
venient way of quantification: records of typical responses are shown in
Fig. 5. Altogether 10 units have been studied in this way, and the threshold
aperture for directional discrimination varied from 6' to 24'.
The straightforward interpretation of this result is that the complete
mechanism for directional selectivity is contained within a subunit of the
receptive field extending not much more than I' in the preferred-null axis.
Since the result does not depend critically upon the position of the slit
within the receptive field, it looks, again taking the straightforward view, as
Position 450S 50P in visual field
Background
7 cd/r2 /
I1111III
I*Il 1 11
111 I I I I[110111 1I I .
17't ..
|11.111 v
Unit 4-21
0o5 sec
Fig. 5. Responses to motion of a black edge across slits of various widths
(breadth was 20 in all cases). Results are shown for four slit widths, from 10 06' to
8' measured in the preferred-null axis. The records to the left were obtained when
the black border advanced across the slit in the preferred direction, those to the
right when it moved in the null direction. The lower trace of each pair is the
response of a photomultiplier aimed at the field. Notice that the differences
between preferred and null motions are obvious in the top two records, when the
distances through which the edge could be seen moving were 10 06' and 34'; the
difference is only just detectable at 17' and has vanished at 8'. The directionally
selective mechanism appears to be contained within a retinal region subtending
I' to f, whereas the whole receptive field subtended 4j0.
if the sequence-discriminating mechanism must be reduplicated perhaps a
dozen or more times to cover the whole receptive field. Would there be any
escape from this conclusion if the image was of very poor quality so that,
even for the small slits, it was diffused over a large part of the receptive
field?
If the optics are poor, then the time course of light intensity changes at
the two edges of the receptive field will be slightly different for the two
567
486 H. B. BARLOW AND W. R. LEVICK
directions of motion, and one can set up somewhat elaborate schemes in
which these differences form the basis of directional selectivity. The
schemes have to be made even more complicated to account for the fact
that black or white edges advancing through the slit show the same
directional preference. Now it will be shown in the next section that the
interaction responsible for directional selectivity occurs better at small
separations of the stimuli than at large separations. We can think of no
way in which neighbouring retinal regions would show more interaction
than widely separated ones unless the subunit responsible for directional
selectivity is itself a small compact one. In this way the experiment about
to be described is a useful control confirming the conclusion reached from
the single slit experiments.
Discrimination of sequence. The fact that movements within a region of
the receptive field subtending less than jI' are sufficient to give directionally
selective responses suggests a possible explanation for our failure to get
clear-cut sequence-dependent results when we first attempted to excite with
a pair of static stimuli at various temporal intervals. In these experi-
ments spots closer together than about 10 had never been tried and we
therefore designed new apparatus so that two strips of light each sub-
tending 0.10 x 20 could be brought to within 001' of each other and turned
on and off in either sequence (see Methods).
At small separations the experiment gave definite evidence of sequence-
dependence, as shown in Fig. 6. The response is much greater in the
sequence corresponding to movement in the preferred direction than for
null sequences, and this is true for both 'on' and 'off' stimuli. However,
these differences become less when the separation of the slits is increased
to over 10 even though both slits remain inside the receptive field.
We have done similar two-slit experiments in which moving cards were
used to provide 'on' or 'off' excitation at the two slits. These also indicated
that sequence discrimination occurs at small separations but is reduced at
large separations as shown in Table 2. It was not easy to judge the res-
ponses accurately but the difference between null and preferred sequences
faded out for separations greater than i' and other units showed a
similar reduction for large separations. There is, however, a defect in
these experiments which was not always fully controlled. In order
to provide a vigorous stimulus with each slit it was made 0.10 wide.
This is below the threshold for directional selectivity in most prepara-
tions, hence in these cases each slit by itself was equally effective for
null and preferred sequences. However, this was not always true, and
some of our results lack the necessary controls. In addition it might
be held that there are effects of movement which are subthreshold for
each slit by itself but become suprathreshold with the pair, and that it
568
MECHANISM OF DIRECTIONAL SELECTIVITY 487
is the summation of these subliminal effects that produces the asymmetry
for the two sequences. A control observation (made on a unit in which
directional selectivity occurred across the 0-10 slit) meets this criticism and
is worth reporting because it reinforces the conclusion that sequence as
such is effective.
Unit 3-41 Preferred Null
t I 1 I l I
A B
On =
I-! Off
I!I 1 A_
17' A B B A
0-5 sec
w .
r - I I- I I I
A B
On
w
IIw I I
-%.-
Off
1006' A B B A
Fig. 6. Responses to different temporal sequences of two static stimuli. On the
left the positions of the pair of stimuli are shown within the outline of the receptive
field. Records for the small separation are shown above, those for the large
separation below. Within each half, records for on are above those for off. The
lower trace of each pair is the photomultiplier output: increasing light moves the
trace upwards, and slit A was arranged to give the bigger step in every case, even
though it was not brighter. Preferred sequences are on the left, null on the right.
Notice that the preferred sequence yields more spikes than the null at the small
spatial separation, but this difference ceases to be clearly visible when the separa-
tion of the slits is increased.
In this experiment the card behind the pair of slits had white and black
regions arranged so that when the card was moved in one direction the
sequence of lightening or darkening at the two slits corresponded to
motion in the opposite direction. Under these conditions each slit by
itself gave a greater response for movement of the card in the preferred-
direction. However, when both slits were used there was a greater response
for movement of the card in the null direction: that is, the sequence of
activation of the slits was overriding the effects of motion in the opposite
direction within each slit.
569
488 H. B. BARLOW AND W. R. LEVICK
It must be pointed out that the time interval between the two stimuli is
an important variable that we have not yet studied systematically. In
timing these stimuli manually we have varied the interval over as wide a
range as possible, but in spite of this we never found as strong an inter-
action at large as at small spatial separations of the slits. Some further
results with this type of experiment are given in Table 3 (see later) and it is
hoped that a more systematic exploration will be presented in the future.
TABLE 2. Two-slit experiment
Same unit as in Table 1. The stimulus was again a black-white border moved through the
receptive field in the null or the preferred direction, but the view of the motion was now
restricted by a pair of narrow slits in a grey card placed immediately in front of the moving
border. Each slit subtended only 6', and responses to preferred and null directions were
indistinguishable for each slit by itself. However, when the pair of slits was darkened or
lightened in sequence, the strength of the response was found to depend upon the order in
which the slits changed. The separation of the two slits, measured in the direction of motion,
was varied and the unit's response graded as in Table 1. The effect of the order of stimula-
tion was greatest at small separations, but null and preferred sequences were still dis-
tinguishable up to 24' separation.
'Off' stimulus 'On' stimulus
Slit A,AK
separation Preferred Null Preferred Null
10 36' 2 2 2 2
1006' 3 3 3 3
48' 3 3 2 2
34' 3 3 2 2
24' 4 3 3 2
17' 3 2 3 1
12' 4 2 3 1
8' 3 1 3 1
6' 3 1 3 2
We think the results already given are sufficient to establish that direc-
tional selectivity may be based upon the discrimination of the sequence of
excitation of only a pair of regions. Even though the image of a moving
object falls on a long succession of receptors in a continuous succession of
time intervals it is unnecessary to postulate the interaction of more than
two regions to account for the directionally selective property.
Responses to gratings. The results so far reported suggest that sequence-
discrimination is performed by subunits of the receptive field. Figure 1
shows that a directionally selective unit can discriminate the direction of
motion of the bars of a grating subtending 15' (period 30'). It is hard to
see how this discrimination could be performed if the bars of the grating
were small compared to the size over which each subunit integrates or
averages the light, and in fact the resolvable grating size fits, to a first
approximation, the size of subunit suggested by the preceding tests.
Another result that may also fall into line is the shape of the curves found
when determining threshold as a function of area (Fig. 5 of Barlow et at.
570
MECHANISM OF DIRECTIONAL SELECTIVITY 489
1964; see also Barlow, 1953). Complete summation (that is, threshold
oc 1/area) does not hold out to the full diameter of the receptive field in the
directionally selective units of the rabbit, and it is tempting to identify
the limit to which it does hold (approx. 20') as the integrating area of the
subunits.
One negative result in the grating tests is worth comment. Hassenstein
(1951) found paradoxical optokinetic movement responses in beetles:
when a grating of period slightly greater than the angular separation of the
axes of the ommatidia was moved in one direction, the beetles responded as
for the opposite direction of movement. Nothing of this sort was seen in
the present tests: when motion of a grating caused a response, this was
never greater in the null direction than in the preferred direction. This is
not too surprising, for the occurrence of paradoxical movement responses
in the beetle must depend upon the regular spacing of its ommatidia.
Mechanism of sequence-discrimination
The foregoing experiments show that the directional selectivity of
ganglion cells is based upon sequence-discrimination within subunits of
their receptive fields, but they tell us nothing about the mechanism where-
by a pair of stimuli causes a greater discharge in one sequence than in
reverse. Figure 7 shows two schemes in which the preferred sequence,
corresponding to motion in the preferred direction, elicits a greater
response than the null sequence. These are intended to exemplify two
broad alternatives, not to make exact specifications.
The left-hand scheme works by detecting a specific conjunction of
excitations: activity aroused by increase or decrease of illumination in
region A is delayed and arrives at the 'and' gate in the next layer
synchronously with activity aroused when the image moves on to region
B. Activity from B passes to the 'and' gate below it, and is also passed
laterally to interact with activity from C. The sequence ABC is the
preferred sequence, and the gates only respond when their respective
conjunctions 'both B and delayed A', or 'both C and delayed B' occur.
Instead of selecting the preferred stimulus by a logical conjunction, the
right hand scheme rejects the null stimulus by veto. Activity aroused at
' on) or ' off' in region B or C is again passed laterally and acts after a delay,
but in this case it inhibits the next unit. As before, CBA is the null
sequence, and when it occurs the inhibition from C prevents the response
that would have resulted from B alone, and inhibition from B likewise
vetoes A's response. On the other hand if the sequence is the preferred one,
ABC, then the inhibition from B does not arrive until the excitation from
A has already got through, and likewise C is unsuccessful in vetoing B. It
will be observed that this scheme only requires that inhibition persists
571
490 H. B. BARLOW AND W. R. LEVICK
longer than excitation; a definite delay when it is passed laterally is not
strictly necessary.
Some evidence favouring the right-hand, inhibitory, scheme has already
been given. (1) As shown in Fig. 2 a stationary spot turned on and off
elicits a response. If the excitatory conjunction scheme was modified to
account for this it would probably still predict a considerably lower
threshold for a moving than for a stationary spot. As shown in Fig. 5
of Barlow et al. (1964), the thresholds for spots of various sizes moving in
the preferred direction differ by small and inconstant amounts from those
for the same spot turned on or off. (2) The most striking feature of these
directional units is the absence of any impulses when movement is in the
null direction. This prompts one to look for a mechanism that inhibits
unwanted responses. (3) WVhen testing for directional selectivity in
Excitatory mechanism Inhibitory mechanism
A BC A BC
At At At A
572
MECHANISM OF DIRECTIONAL SELECTIVITY 491
different parts of the receptive field we found that the results obtained at
one edge were anomalous in that movements in both null and preferred
directions gave responses. Such responses are illustrated in the top row of
records in Fig. 4. On the inhibitory scheme it may be possible to resolve
this anomaly along the following lines. Responses from the last points
crossed by a spot moving in the null direction are normally prevented by
inhibition coming from the penultimate regions that have just been
crossed. If the spot is moved to and fro solely in the rim, this penultimate
region is avoided, and consequently it never inhibits the responses coming
from the rim. Measurements of this 'inhibition-free' zone at the rim
suggest that it may extend for as much as 10 inwards from the extreme
edge of the receptive field.
These observations are not decisive, but they brought the inhibitory
scheme to the front of our minds, and we now give some much stronger
evidence favouring it.
Movements in null direction evoking responses. A spot of light moved
continuously through the field in the null direction will evoke no impulses,
but if such continuous motion is interrupted while the spot is in the
receptive field, a burst of impulses occurs just when the movement starts
up again. Evidently the inhibition that prevents the response when
motion is continuous decays while the spot is stationary, so that when the
spot moves on to new receptors the activity excited escapes inhibition and
gets through to the ganglion cell. This response to intermittent motion is
illustrated in Fig. 8.
If motion in the null direction is slow enough, a discharge can also be
elicited, and this is imustrated in Fig. 9. Presumably the rate of rise and
decay of the inhibitory process, together with the distance at which it
operates, governs the range of speeds over which directional selectivity
occurs.
Responses to slits singly and in sequence. What was thought to be a
crucial test of the inhibition hypothesis was devised. Two slits were placed
close to each other and the responses to each in isolation were recorded
several times at on and off. The slits were then presented in null or
preferred sequence, and several responses again recorded. The records were
analysed by counting the impulses that occurred within i sec of stimula-
tion, and the averages of 4 to 7 responses are presented in Table 3.
First compare the figures in the last two columns, and notice that the
result confirms what has already been said. Preferred sequences are more
effective stimuli than null sequences at all separations studied, but the
difference is most pronounced at small separations and decreases at the
separations greater than 17'. Now compare the figures in the 'Null'
column with those in the 'A + B' column. In every case the 'Null' has the
573
492 H. B. BARLOW AND W. R. LEVICK
lower figure, so that inhibition certainly occurs: when the sequence is in
the null direction fewer impulses occur than when each region is excited
separately. Finally, compare the figures in the 'Preferred' column with
those in the 'A +B' column. Here there is an excess in the 'Preferred'
column at small separations, but not at large separations.
Position 50S 100A in visual field
Spot size III
/a b
Null directioni
ac
b _ b c
. \ ~ ~ ~~c
d
UniI3-26
Background 10 cd/M2 d e
Spot 60 cd/M2-
Unit 3-26
_I
-e ;c ~ ;
3- -I- 0 5 sec
Fig. 8. Escape of impulses with intermittent movement in the null direction.
Five positions are marked in relation to the outline of the receptive field shown on
the left. The lowest two pairs of records show the effect of sweeping continuously
through these positions in the null (abede) and preferred (edoba) directions. In the
upper four pairs the spot was moved discontinuously, first from a to b, then from
b to c, then from c to d, then from d to e just outside the field. The lower trace of
each pair shows the position of the spot in the field. As an example of the escape
phenomenon notice that no impulses occur when movement from c to d is part of a
continuous sweep (5th pair of records), but they do occur when this movement is
made in isolation (3rd pair). The suggested interpretation is that 'on' or 'off'
stimulation at any point inhibits 'on' or 'off' excitation of the next point in the
null direction, but this inhibition decays with time. When the spot pauses at c, off
excitation from c, and on excitation of the next point, occur after inhibition has
decayed and impulses therefore escape.
574
MECHANISM OF DIRECTIONAL SELECTIVITY 493
. . I - . I
2
I Null Degrees/sec
b 3 -s_o__-_-_i_Preferre5 5
Null Preferred
07
d Preferred
06
f Null
h
hh H-- - -----__ 0.0
Stationary
Background 10 cd/M2 0 5 sec
Spot to 60 cd/M2 Unit 3-26
Fig. 9. Paradoxical response to very slow motion in the null direction. As before
the lower trace of each pair shows the position of the spot of light. For movements
at about 50/sec a vigorous response was obtainable in the preferred direction, but
the top pair of records shows that in the null direction there is no increase over the
maintained firing rate with no stimulation (lowest pair of records). When motion
was at about 0.70/sec there was still a vigorous response in the preferred direction
(2nd pair), but in the null direction (3rd pair) there was also a distinct increase
compared with the maintained discharge (4th pair). If movement is slow enough,
the inhibition at a point in front of the advancing spot must have declined by the
time the spot reaches it to a level where extra impulses are allowed to pass.
575
494 H. B. BARLOW AND W. R. LEVICK
Individual responses are highly variable and the experimental situation
needs systematic exploration with averaging techniques. At this stage we
can say that the experiment obviously supports the idea that null
sequences are ineffective stimuli because of inhibition. It also indicates,
however, that there is some degree of facilitation for preferred sequences,
though it is fair to add that this seems a less important effect than the
inhibition.
Receptive field alone j Receptive field and surround 1 Surround alone
t
r e er re>S/// /
'I\
Preferred Preferred
62 30 l!4
0o5 sec
_l40
unit E-u
'I.
Fig. 10. Lateral inhibition and responses to movement. A black edge was moved
behind a mask of grey paper (cross-hatched) so that the advancing border crossed
a 40 hole exposing the receptive field alone with the surround masked off (left), the
surround alone with the receptive field masked off (right), or it crossed both to-
gether, with no mask (centre). The records show the responses; the lower trace of
each pair came from a photocell aimed at the receptive field. No impulses were
obtained when motion was in the null direction (lower pair of records). In the pre-
ferred direction some were obtained in each case, but the response was much
greater with the surround masked off than when it crossed surround and centre
together (62 instead of 30 impulses). Motion in the surround inhibits the response
to motion in the centre, just as light going on or off in the surround inhibits on or
off responses from the centre.
576
MECHANISM OF DIRECTIONAL SELECTIVITY 495
Inhibition from outside the receptive field
The type of inhibition postulated to account for directional selectivity,
and shown up in the experiment of Table 3, comes from within the recep-
tive field-that is, from within the region where light can evoke impulses.
There is also an inhibitory mechanism acting from outside the receptive
field-that is, from the surrounding region where light stimuli evoke little
or no response. Figure 10 shows an example of the effect of this inhibitory
mechanism on the discharge evoked by a moving object. Figure 5 of
Barlow et al. (1964) shows the effect of this inhibitory mechanism on the
threshold.
DISCUSSION
Physiological function and anatomical structure
We think that the experiments described establish without need of
further discussion these four points about directional selectivity. First, it
is not caused by optical aberrations, nor by simple differences of latency
for discharges evoked from different parts of the receptive field. Secondly,
it is not necessary to cross any critical region or line in the receptive field:
the mechanism responsible for the property resides in small subdivisions
of the field and must be extensively replicated. Thirdly, these replicated
subunits distinguish between null and preferred sequences of excitation of
a pair of regions with which they connect; thus the directional selectivity
of the ganglion cell is built up from sequence-discriminating subunits.
Fourthly, inhibition plays an important part in this discrimination by
preventing responses to sequences corresponding to motion in the null
direction.
By themselves these results probably do not justify any further con-
clusions, but the complexity of function that they have revealed is
beginning to match up to the long-known complexity of neural structure
in the retina. It is a challenging problem to fit together the jig-saw puzzle
of anatomical elements in the hope of revealing the picture of physiological
function, and a tentative solution is shown in Fig. 11. It is certainly
incomplete, for it does not specify the connexions of the concentric type of
ganglion cell, nor of those selectively responsive to fast and slow movement
(Barlow et al. 1964). Furthermore, we assume that there is a duplicate set
of bipolar and horizontal cells that are activated at 'off'. We have some
evidence, to be presented elsewhere, that 'on' and 'off' systems do not
interact with each other at this level, and therefore for simplicity we have
omitted the 'off' system. Because of the diversity of types of bipolar and
horizontal cells (on and off for at least four different directions) one can
see why a very large number of bipolar cells are required to handle the
input from a group of receptors.
577
496 H. B. BARLOW AND W. R. LEVICK
At various points there are alternatives to our scheme that are not
excluded by the evidence at present available. On the other hand the roles
of the anatomical elements and their postulated connexions are not as
arbitrarily assigned as a naive reader is liable to suppose. For discussion,
take what is perhaps the most controversial and interesting feature of the
scheme-the assignment to horizontal cells of the role of inhibitory
elements that prevent bipolar cells responding to null sequences. There are
two'main questions to be answered: why place the inhibitory element in
the inner nuclear layer? And why postulate that the horizontal cell
Null direction
* ~~~~G
- =~~~~~j
1 00je
Fig. 11. Suggested functional connexions of the retinal elements concerned with
directional selectivity. The elements are freely adapted from Cajal (1893), and are
assembled in accordance with the functional organization suggested in this paper.
The scale of the diagram is approximate and a posterior nodal distance of 1 1l5 mm
has been assumed. The pathway of excitation is from receptors (R), through bipolars
(B), to the ganglion cell (G), but activity in this direct pathway is modified by the
associational cells. The horizontal cells (H) pick up from receptors, conduct laterally
in the null direction through a teledendron (Td), and inhibit bipolars in the neigh-
bouring region. This prevents responses when an image moves in the null direction,
but has no effect when motion is in the preferred direction. Horizontal cells have
the function of the laterally conducting elements in the inhibitory scheme shown in
Fig. 7. The amacrine cells (A) are thought to pick up from bipolar endings in the
inner plexiform layer and to conduct activity throughout their axo-dendritic
ramifications; they are assumed to make synaptic connexion with the ganglion cells
and inhibit them, thus mediating lateral inhibition of the type illustrated in Fig. 5
of Barlow et al. (1964) and Fig. 10 of this paper. The off-responding mechanism is
not illustrated, but seems to require duplicate horizontal cells and bipolar cells.
Notice that the ganglion cell must connect selectively to those particular bipolars
which respond selectively to the sequences for one particular direction. Its
response is specific for this pattern of stimulation but is invariant with respect to
contrast and position in the receptive field. It may be said to achieve some degree
of 'stimulus generalization'.
578
MECHANISM OF DIRECTIONAL SELECTIVITY 497
connects from receptors to bipolar cells, rather than, for instance, from
bipolars to bipolars?
The strength of the proposed scheme arises from the fact that a function
can naturally be assigned to the neural elements that are known to exist,
without making esoteric or revolutionary assumptions about how they
work. Sequence-discrimination is assigned to bipolar cells because the
ganglion cell appears to pick up from subunits that are replicated in
different parts of the receptive field, and bipolar cells are the replicated
anatomical elements that feed ganglion cells. There is physiological
evidence of inhibitory interaction acting from one side on these subunits.
This is not like the classical lateral inhibitory interaction which counteracts
the pooled excitatory influences reaching the ganglion cell: the evidence
points to inhibition that acts locally. Excitation aroused from a particular
region of the receptive field is inhibited by preceding excitation of the
region that a light image has just crossed when motion is in the null
direction. This same inhibitory region has no influence on excitation
aroused from the neighbouring region on the opposite side, for it fails to
block the excitation when motion is in the preferred direction. The
physiological evidence thus indicates that each excitatory region has its
own private inhibitory region on one side, and one can construct a
number of schemes to account for this. Inhibition might act on the
ganglion cells, but in such a way that it only blocks one particular branch
of the dendritic tree: or it might act presynaptically on the bipolar cell
endings. Another possibility one might consider is that the inhibition is
mediated by the receptor-to-receptor connexions described by Sj6strand
(1958) in the guinea-pig. The distance over which the inhibitory effects
have to be passed may be a difficulty here, and this notion shares the
difficulty described below for other forms of inhibition which act on the
receptors. Since the horizontal cells are known to have processes conduc-
ting laterally the natural starting hypothesis is that they are the cells
carrying this inhibition from one region to another.
If this is granted there is still scope for argument as to where this
inhibition is picked up from, and where it feeds to. Might it not inhibit
receptors rather than bipolar cells? Might it not even pick up from bipolar
cells and feed back to receptors? The key observation here is that a region
which has itself been inhibited from its own private inhibitory zone on one
side can none the less inhibit activity aroused in the neighbouring zone on
the other side. Motion through the receptive field in the null direction may
elicit no impulses whatever. Consider what is happening half way through
such a traverse: one sees that excitation of the receptors at the mid-point
prevents the discharge from the next group of receptors the spot is going to
cross, even though no activity is transmitted centrally from the receptors
32 Physiol. 178
579
498 H. B. BARLOW AND W. R. LEVICK
at the mid-point. This indicates that the inhibitory connexion runs from
an early point on the path from the inhibition-arousing zone to a later
point on the path from the zone that is inhibited. Presumably then it
runs from receptors to bipolar cells, and in that case the inhibition can act
in the ordinary way by stabilizing the membrane potential of the bipolar
cells. However there is clearly a point here open to histological'investiga-
tion. Do the horizontal cells make this pattern of connexion in the rabbit?
Polyak (1941) describes the horizontal cells of the monkey as making
receptor-to-receptor connexions.
No further information has come to light on the pathway mediating
ordinary lateral inhibition of the type shown in Fig. 10. This probably acts
on the ganglion cells, and amacrine cells remain the most plausible guess.
It is clear that our allocation of functions to particular structures must
be regarded as provisional, but we were pleased to find how well the
physiological organization seems to fit in with the anatomical structure.
Other proposed anatomical correlates in other species. Maturana, Lettvin,
McCulloch & Pitts (1960) and Lettvin, Maturana, Pitts & McCulloch (1961)
have also attempted to relate structure and function in the retina, in their
case in the frog. Their discussion has something in common with ours, but
they place greater emphasis on the concept that the ganglion cell's
properties are determined by the shape and size of its dendritic tree. They
believe that the different strata of the inner plexiform layer carry informa-
tion as to different properties of the pattern of light falling on the recep-
tors; the ganglion cell is then thought to pick up the appropriate
combination of these properties by ramifying in the various layers. This
may explain how a ganglion cell is able to make connexion with a specific
subset of bipolar cells, and their notion does not contradict ours. Where we
feel that our scheme goes further is in showing how the complex task of
signalling direction of movement can be broken down into simpler tasks
that can be performed by elements making simple excitatory and inhibi-
tory connexions.
Maturana & Frenk (1963) have described directionally selective units in
the pigeon's retina. These obviously have much in common with the units
in the rabbit, for they show the same directional selectivity independent of
the path through the field and the contrast of the moving object. Further-
more, they made an interesting observation which led them to the con-
clusion that an inhibitory mechanism is involved in directional selectivity.
They turned a spot of light on and off in one place in the receptive field,
eliciting responses in the usual way. While the light was off they moved it
to another position in the field displaced in the null direction from the first
position, and turned it on and off again. No responses were obtained,
whereas, if the spot had been displaced in the preferred direction, responses
580
MECHANISM OF DIRECTIONAL SELECTIVITY 499
were obtained as usual. Clearly this is similar to the two-spot experiment
described here, but it seems from their brief description that inhibition
must persist for a long time in the pigeon. They do not attempt to make
detailed suggestions about which anatomical structures are responsible
for the specificity of the stimuli that generate responses in a particular unit,
but they give the impression that they believe it is achieved by the ganglion
cell. In our view the specificity originates with the bipolars, and the
ganglion cell generalizes for position and contrast by picking up only from
those bipolars that respond to sequences of on or off stimuli corresponding
to one particular direction of motion.
Griisser-Cornehls, Griisser & Bullock (1964) tested movement-sensitive
units in the frog with various stimuli, and came to the conclusion that
movement detection was really 'change-of-position' detection. Their
experiment suggests that they are distinguishing between continuous and
discontinuous change of position, and in that case our conclusions are not
too far apart: discontinuous change of position, as in the two-spot experi-
ment, can activate the directionally selective mechanism. However, they
were not dealing with units responding selectively to the direction of
motion, for unlike Maturana et al. (1960) they failed to find such units in
the frog, although they confirmed many of these authors' other findings.
Directional system in insects. Reichardt (1957, 1961a, b) has proposed a
mechanism capable of explaining the responses of insects to movement in
their visual field. This seems at first sight very different from the one we
have arrived at, for his scheme depends upon evaluating the cross-correla-
tion between the signal from an ommatidium and that from its neighbour
modified by passage through a low-pass filter. This is closer to the excita-
tory-conjunction scheme that we rejected than it is to the inhibitory
scheme. However, one should probably regard Reichardt's proposal as the
simplest physical system with a performance specification similar to the
beetle's eye, and one should not be too surprised if the realization of a
system in 'biological hardware' is different from what it would be in
physical hardware, even if the operation performed is very similar.
Pattern recognition, trigger features, and stimulus generalization
Maturana & Frenk (1963) suggest that an understanding of the type of
behaviour they describe in the ganglion cells of the pigeon retina clarifies
certain problems of pattern recognition. We think there are two aspects of
recent work on the visual pathway that are interesting in this respect. The
first is the specificity of the features that are effective in triggering the
activity of sensory neurones. Examples of this are provided by the 'fly
detectors' (Barlow, 1953) and 'convexity detectors' (Lettvin et al. 1959) of
the frog's retina, the Jinear elements of the cat's cortex (Hubel & Wiesel,
32.2
581
500 H. B. BARLOW AND W. B. LEVICK
1959, 1962), the 'horizontal edge detectors' of the pigeon retina (Maturana
& Frenk, 1963), and the directionally selective elements found in all these
preparations as well as in the rabbit's retina. Now in pattern recognition
by machines, Grimsdale, Sumner, Tunis & Kilburn (1959) broke the task
into two stages by first detecting the presence of certain key features of the
patterns to be discriminated and then looking for the characteristic
combinations of these features. Most of the successful systems for recogni-
zing printed or handwritten characters make use of a similar scheme
(Selfridge & Neisser, 1960; Uhr & Vossler, 1961; Frishkopf & Harmon,
1961; Kamentsky & Liu, 1963), and it is interesting to see why it is
necessary for the computer to view its text through these 'feature filters'.
It is because even the largest computor cannot recognize letter A by com-
paring the input with a complete list of all members of the class of A's.
Such an approach would require the separate representation of each of the
2n possible states of the n binary inputs and this becomes unmanageable
for values of n that are very small by biological standards. Presumably
the trigger features of the visual system likewise enable the input states
to be classified in an effective way without requiring a googolian number
of separate representations.
The second aspect we want to draw attention to is the detailed manner
in which the specific and general properties of these trigger features are
picked out. This discussion will be based upon our suggested mechanism
for directional selectivity, and we shall introduce certain simplifications
which, though not entirely justifiable, make it easier to compare the
neural process with artificial pattern recognition.
According to our analysis the operation of abstracting direction of
movement is done in two stages, each with the same two steps. The first
step in each case is the summation or pooling of selected excitatory
influences, and the second step is the inhibitory interaction of another
element that has, as it were, the power of veto. The first step loses informa-
tion, for the bipolar cell which pools inputs from a number of receptors does
not reflect in its output which particular ones were active. As pointed out by
Reichardt (1961a, b) the inhibitory step could in principle regain this lost
information, but in the case of bipolar cells it does not do this; instead it
makes the response more selective by bringing in new information.
Without this inhibitory interaction a bipolar cell would simply say, when
it became active, 'Light fell in this region'; with the inhibition it says
'Light fell in this region and was not preceded by light falling in that
region'. Compared with the receptors in the preceding layer, the bipolars
have lost some information about the exact position of the stimulus, but
they have extracted some information about the presence of a particular
sequential pattern in the stimulus.
582
MECHANISM OF DIRECTIONAL SELECTIVITY 501
The same two steps are taken in the next stage, occurring in the next
layer. Here a ganglion cell does not pool from all the bipolars in the
receptive field, but it picks up selectively from all those which respond
when a stimulus moves in a particular direction, irrespective of the location
of the bipolar cell or whether it belongs to the 'on' class or the 'off' class.
It thus discards the information as to the contrast of the stimulus object
and whereabouts in the receptive field it was; it 'generalizes' by grouping
together activity resulting from movement in a particular direction,
regardless of contrast and exact position. This is followed by inhibitory
interaction which again makes the response more specific. Light going on
or off in the surround (Fig. 5 of Barlow et al. 1964), or movement in
the surround (Fig. 10, this paper) reduces or prevents the response, so
that when activity occurs it implies that changes were not occurring
in the surrounding retina at the time they occurred within the receptive
field.
Let us now express the logical pattern of these repeated operations
symbolically. The pooling or generalizing operation is equivalent in some
ways to the formation of a logical union (inclusive 'or', symbolized by v),
and the inhibitory or veto operation is equivalent to 'and not. ..'
(symbolized by . -). If B is the class of inputs to which a bipolar cell
responds, and Ra, Rb, etc., are the inputs causing activity in the receptors
a, b, etc., then
'' B~~B= (RaVRbVRc...). (RrVRsVRt..
Likewise the class a of inputs causing activity in a ganglion cell is ex-
pressed in terms of Ba, Bb, etc., the inputs which activate the selection of
bipolars it connects with; thus
U = (BaVBbVBc...). (BrVB8VBt,..)
If we symbolize by Elk the class of inputs which is effective in exciting a
particular element after 3b synapses, and by EOA+1 the class effective for an
element after one more synapse, then E0r+1 is given by
EO+'= (E&vENvEg...). (ErvEfvE*...).
Notice that only a small proportion of the possible logical functions of the
PI can be expressed in this form, and it is therefore not at all a trivial
restriction.
We are suggesting that the classification system at one level in the
nervous system is built out of the classification at the preceding level by a
combination of pooling or union, and inhibition or veto (and not ... ). Can
we regard the proposed mechanism for directionally selective units as a
paradigm of the neural mechanisms responsible for the classification
system imposed on our sensory input? Is pooling analogous to 'stimlulus
583
502 H. B. BARLOW AND W. R. LEVICK
generalization', and is greater specificity of response always achieved by
the veto of an associational neurone, an interposed inhibitory element?
These are intriguing questions.
SUMMARY
1. The mechanism of directional selectivity has been investigated in
retinal ganglion cells of decerebrate or lightly anaesthetized rabbits.
2. The property of responding to one direction of motion (preferred) but
not to the opposite (null) direction occurs in on-off units, but the responses
to movement cannot be predicted from the map of the receptive field
obtained with static stimuli; the property cannot be explained by optical
aberrations (see Fig. 1), nor by progressive changes of latency across the
field (see Fig. 2).
3. There is no critical line or region that must be crossed to produce
unequal responses to preferred and null motion (Fig. 4): small subsections
of the receptive field possess the property (Fig. 5).
4. The response to successive stimulation of two small regions depends
upon whether the order corresponds to motion in the preferred or null
direction (Fig. 8). This effect is strong when the two regions are within
about I' of each other, but declines at greater separations.
5. This is thought to indicate that directional selectivity results from
the discrimination of sequence. Normal movement excites many points in
a long succession, but the mechanism works by discriminating the sequence
of individual pairs of regions.
6. When two stimuli are presented in the null sequence the number of
impulses elicited is much less than the sum of the numbers elicited from
each stimulus in isolation (Table 3). There is a small excess of impulses over
this sum when the stimuli are presented in the preferred sequence.
7. From this and other findings it is concluded that sequence-discrimina-
tion results primarily from an inhibitory mechanism that vetoes the
response to null sequences, rather than from the detection of the con-
junction of excitation from two regions with an appropriate delay (see
Fig. 7).
8. If the image of a moving object spreads outside the receptive field on
to its surround there are fewer impulses than when it is confined to the
receptive field alone (Fig. 10). This must be the inhibitory mechanism that
elevates the threshold for large compared with small spots, and it is
presumably different from the inhibition responsible for sequence-
discrimination.
9. The functional organization is discussed in relation to the anatomical
organization (Fig. 11). It is suggested that horizontal cells conduct
584
MECHANISM OF DIRECTIONAL SELECTIVITY 503
laterally and inhibit the bipolars on one side, thus preventing them from
responding to null sequences; the ganglion cells then pick up from the
bipolars responsive to like sequences and it is thought that the inhibition
from the surround may be mediated by amacrine cells.
10. The ability to abstract direction of motion irrespective of the
position in the receptive field and the contrast of the moving object has
elements in common with much more complex feats of pattern recognition.
The two steps-inhibition by associational neurones and selective pooling
-may also play a part in these more complex feats.
We wish to thank W. A. H. Rushton, P. A. Merton, P. E. K. Donaldson and G. West-
heimer for the loan of apparatus, and W. Hail, C. Hood, R. Rumble and P. Starling for
help in construction and photography. This work was supported in part by Grants NB 05215
and NB 03154 from the U.S. Public Health Service. The micromanipulator for the Cam-
bridge experiments was constructed in the Department of Physiology, University of Sydney
to the design of P. 0. Bishop and W. Kozak.
REFERENCES
BARLOW, H. B. (1953). Summation and inhibition in the frog's retina. J. Phy8iol. 119,
69-88.
BARLow, H. B. & HILL, R. M. (1963). Selective sensitivity to direction of motion in
ganglion cells of the rabbit's retina. Science, 139, 412-414.
BARLow, H. B., HILL, R. M. & LEVICK, W. R. (1964). Retinal ganglion cells responding
selectively to direction and speed of image motion in the rabbit. J. Phy8iol. 173, 377-407.
CAJAL, S. RAmoN y (1893). La r6tine des vert6bres. In La Cellue, 9, 119-257, ed. CARNoY,
J. B., GILSON, G. & DENYS, J. Lierre: J. van In and Co.; Louvain: A. Uystpruyst, Libraire.
FRIsHKOPF, L. S. & HARMON, L. D. (1961). Machine reading of cursive script. In Proc. 4th
Lond. Symp. on Information Theory, pp. 300-316, ed. C. CEERRY. London: Butterworth.
GRIMSDALE, R. L., SUMNER, F. H., TuNIs, C. J. & KILBURN, T. (1959). A system for the
automatic recognition of patterns. Proc. Inst. elect. Eng8, B, 106, 210-221.
GRUSSER-CORNEHLS, U., GRJssEra, 0. J. & BuLLOCE, T. H. (1964). Unit responses in the
frog's tectum to moving and non-moving stimuli. Science, 141, 820-822.
HASsENSTEIN, B. (1951). Ommatidienraster und afferente Bewegungsintegration. Z. vergl.
Physiol. 33, 301-326.
HUBEL, D. H. (1959). Single unit activity in striate cortex of unrestrained cats. J. Phy8iol.
147, 226-238.
HUBEL, D. H. & WIESEL, T. N. (1959). Receptive fields of single neurones in the cat's
striate cortex. J. Phy8iol. 148, 574-591.
HUBEL, D. H. & WIESEL, T. N. (1962). Receptive fields, binocular interaction and functional
architecture in the cat's visual cortex. J. Physiol. 160, 106-154.
KAMENTSKY, L. A. & Liu, C. N. (1963). Computer-automated design of multifont print
recognition logic. I.B.M. Journal of Research and Development, 7, 2-13.
LETTVIN, J. Y., MATURANA, H. R., MCCULLOCH, W. S. & PITTS, W. H. (1959). What the
frog's eye tells the frog's brain. Proc. Inst. Radio Engr8, N.Y., 47, 1940-1951.
LETTvIN, J. Y., MATuRANA, H. R., PITTs, W. H. & McCuLLOCH, W. S. (1961). Two remarks
on the visual system of the frog. In Sensory Communication, pp. 757-776, ed. W. ROSEN-
BLITH. M.I.T. Press; New York: John Wiley.
MATURANA, H. R., LETTVIN, J. Y., MCCULLOCH, W. S. & PITTS, W. H. (1960). Anatomy
and physiology of vision in the frog (Rana pipiens). J. gen. Physiol. 43, suppl. 2,
Mechanisms of Vision, 129-171.
MATURANA, H. R. & FRENK, S. (1963). Directional movement and horizontal edge detectors
in pigeon retina. Science, 142, 977-979.
POLYAX, S. L. (1941). The Retina. Chicago: University of Chicago Press.
585
504 H. B. BARLOW AND W. R. LEVICK
REICHARDT, W. (1957). Autokorrelationsauswertung als Funktionsprinzip des Zentral-
nervensystems. Z. Naturf. 12, 447-457.
REICHARDT, W. (1961 a). Autocorrelation, a principle for the evaluation of sensory informa-
tion by the central nervous system. In Sensory Communication, pp. 303-317, ed. W.
ROSENBLI. M.I.T. Press; New York: John Wiley.
REICHARDT, W. (1961b). Yber das optische Auflosungsvermogen der Facettenaugen von
Limulus. Kybernetik. 1, 57-69.
SELIRIDGE, 0. & NEIssER, U. (1960). Pattern recognition by machine. Sci. Amer. 203,
60-68.
S TPs.AxD, F. S. (1958). Ultrastructure of retinal rod synapses of the guinea pig eye as
revealed by three-dimensional reconstructions from serial sections. J. Ultra8tr. Re8. 2,
122-170.
THOMSON, L. C. (1953). The localisation of function in the rabbit's retina. J. Physiol. 119,
191-209.
UEm, L. & VossLER, C. (1961). A pattern-recognition program that generates, evaluates
and adjusts its own operators. Proceeding8 of the We8tern Joint Computer Conference, 19,
555-570.
586
Proc. Natl. Acad. Sci. USA
Vol. 94, pp. 1264912654, November 1997
Neurobiology
Edited by John E. Dowling, Harvard University, Cambridge, MA, and approved September 10, 1997 (received for review June 2, 1997)
ABSTRACT Deciphering the information that eyes, ears, and objects resembling them under a wide variety of illumi-
and other sensory organs transmit to the brain is important nation conditions despite large differences in the contrast of
for understanding the neural basis of behavior. Recordings their carapace (12, 13). Fig. 1b shows underwater photographs
from single sensory nerve cells have yielded useful insights, of black and gray cylindrical objects that approximate the size
but single neurons generally do not mediate behavior; net- and range of contrast of the female carapace. Even though the
works of neurons do. Monitoring the activity of all cells in a gray object is much less apparent, male crabs can detect it
neural network of a behaving animal, however, is not yet almost as well as the black one (12, 13).
possible. Taking an alternative approach, we used a realistic Such visual performance is remarkable in view of the rather
cell-based model to compute the ensemble of neural activity simple design of the animals lateral eyes. Each compound eye
generated by one sensory organ, the lateral eye of the horse- samples the underwater world with '1,000 hexagonally packed
shoe crab, Limulus polyphemus. We studied how the neural receptor units (ommatidia). Individual ommatidia collect light
network of this eye encodes natural scenes by presenting to the from a relatively large region of space ('6), giving the animal a
model movies recorded with a video camera mounted above nearly hemispheric field of view (14). A cluster of retinular cells
the eye of an animal that was exploring its underwater habitat. within each ommatidium transduces the light into electrical
Model predictions were confirmed by simultaneously record- signals that passively propagate into a single spike-firing eccentric
ing responses from single optic nerve fibers of the same cell. By using neural mechanisms generally found in more com-
animal. We report here that the eye transmits to the brain plex retinas, the eccentric cell converts the signals into a train of
robust neural images of objects having the size, contrast, action potentials that it transmits to nearby ommatidia and the
and motion of potential mates. The neural code for such brain. Detailed studies of the integrative properties of these cells
objects is not found in ambiguous messages of individual optic have yielded a quantitative description of the response of an
nerve fibers but rather in patterns of coherent activity that ommatidium in terms of its excitation by light and inhibition from
extend over small ensembles of nerve fibers and are bound neighboring ommatidia (7, 8, 15, 16, 18).
together by stimulus motion. Integrative properties of neurons Our strategy for deciphering the retinal code underlying
in the first synaptic layer of the brain appear well suited to visually guided behavior of Limulus was first to videotape the
detecting the patterns of coherent activity. Neural coding by lateral eyes view of its underwater world with a crab-mounted
this relatively simple eye helps explain how horseshoe crabs camera. Second, we recorded simultaneously from the optic
find mates and may lead to a better understanding of how nerve fiber of an ommatidium viewing the central region of the
more complex sensory organs process information. videotaped scene. Third, we computed the ensemble of optic
nerve activities in response to the video by using a model of the
Living in a world rich with information, animals are highly eye. Fourth, if the response computed for the appropriate
efficient at extracting what is essential for their survival. The first neuron in the model matched that recorded from the nerve
stage of visual processing begins in the eye where the retina fiber, we examined the array of computed responses for
transforms patterns of incident light intensity into trains of putative coding of behaviorally relevant objects.
impulses in optic nerve fibers (1). The retina can encode infor-
mation in a reliable and efficient manner (2, 3) but does not MATERIALS AND METHODS
encode all of it. Rather the retina emphasizes certain features in
the visual scene at the expense of others, presumably to help Underwater Video Recording. We recorded underwater
animals find food, locate mates, and avoid predators (46). movies with a miniature video camera (CrabCam) attached
Knowing what an animal can see is crucial for exploring the to the carapace of crabs moving in the shallow waters (,1 m
neural code the eye transmits to the brain. depth) of an estuary (Fig. 1a). At the same time, we monitored
We report here a study of neural coding in the lateral eye of the the activity of an optic nerve fiber from one of their lateral
horseshoe crab, Limulus polyphemus. We selected Limulus as a eyes. The optic axis of the CrabCam (72 3 54 field of view;
model system because its lateral eye contains the largest neural model V-1210, Marshall Electronics, Culver City, CA; model
network for which a quantitative cell-based model exists (79), SVS-1, Chinon American, Mountainside, NJ) was 4 cm above
and its visually guided behavior is well known (10, 12, 13). the eye and aligned parallel with that of the recorded neuron.
Field studies show that vision has an important role in Video signals were recorded simultaneously with spike dis-
Limulus mating behavior (10). In the spring millions of crabs charges of the nerve fiber on a VHS camcorder. Animals freely
migrate to the oceans edge from Maine to Mexico to build explored their surroundings or were moved along an under-
nests and deposit eggs (11). Male crabs use their lateral eyes water track at a speed of '15 cmys, which is the average
to find mates, whereas females use them to avoid nesting sites velocity of swimming horseshoe crabs (17). They moved past
of other crabs (10, 12). The crabs can see other horseshoe crabs stationary cylindrically shaped black and gray objects that
approximate the size and range of contrasts of adult females
The publication costs of this article were defrayed in part by page charge
This paper was submitted directly (Track II) to the Proceedings office.
payment. This article must therefore be hereby marked advertisement in
Abbreviation: ips, impulses per second.
accordance with 18 U.S.C. 1734 solely to indicate this fact. Present address: Center for Biological Rhythms, University of Vir-
1997 by The National Academy of Sciences 0027-8424y97y9412649-6$2.00y0 ginia, Charlottesville, VA 22903.
PNAS is available online at http:yywww.pnas.org. To whom reprint requests should be addressed.
12649 587
12650 Neurobiology: Passaglia et al. Proc. Natl. Acad. Sci. USA 94 (1997)
FIG. 1. (a) A horseshoe crab, Limulus polyphemus, at the waters edge mounted with a video camera, CrabCam, for recording underwater movies
and a microsuction electrode for recording responses from a single optic nerve fiber. The barrel of the electrode protrudes to the right from the
recording chamber, which is sealed with a white cap (2.5 cm diameter). Cables lead the video and optic nerve signals to recording electronics located
on an overhead skiff as the animal moves about underwater at depths of 0.5 to 1 m. (b) CrabCam images of a high-contrast black object (Left)
and a low-contrast gray object (Right). Black disks indicate the fields of view of the ommatidia whose optic nerve responses were recorded with
the microsuction electrode. (c) Light intensities (given as contrast) incident on the recorded ommatidia plotted as a function of time as the animal
moved past the black (Left) and gray (Right) objects. Contrast is the ratio of the intensity of a pixel to the average intensity of the scene. (d)
Recordings with the microsuction electrode of spike trains from single fibers in response to the 10-s sequences of light intensities shown in c. (e)
Recorded trains of spikes in d plotted as instantaneous frequencies, which are the reciprocals of the intervals between nerve impulses. ( f)
Instantaneous frequencies computed by the cell-based model in response to digitized video images of the black and gray objects. The objects passed
through the fields of view of the recorded and model neurons from 4 to 6 s after the start of the runs evoking reduced (black object) and
quasi-periodic (gray object) responses. Responses are representative of those recorded under similar conditions in other field experiments (n 5
53 trials).
588
Neurobiology: Passaglia et al. Proc. Natl. Acad. Sci. USA 94 (1997) 12651
(30 cm diameter; 15 cm high). The objects were placed 75 cm output was digitized off-line at 11 kHz, and the spike firing
away because field studies show that about half of the male times (Fig. 1d) were converted to instantaneous firing rates
crabs detect and turn toward them at this distance (13). and plotted as a function of time (Fig. 1e).
Video sequences (10-s duration) were digitized at 20 Computing Optic Nerve Responses. We computed patterns
framesys (120 3 160 pixels; 8 bits resolution) by using com- of spike activity for the ensemble of optic nerve fibers by
mercial software (Videofusion, Maumee, OH). We rescaled feeding the CrabCam movies to a realistic cell-based model of
the digitized movies by using a nonlinear mapping function the lateral eye. To determine the light intensity incident on
that inversely corrects for the gamma response characteristic each ommatidium of the model, we convolved each frame of
of the video camera and for output saturation (18). We a movie with a Gaussian point-spread function that matched
compensated for automatic gain controls of the camera, the acceptance angle of an ommatidium (6 at half-maximal
camcorder, and computer by subtracting from all frames of a sensitivity, ref. 19) by using public-domain software (NIH
movie the black level of a black and white test patch viewed by Image Version 1.6), and then we sampled the blurred frames
the CrabCam. The contrast of the black object determined by at points of intersection with the optic axes of the array of
these procedures matched well that measured from photo- ommatidia (14). The size of the model was limited to a 12 3
graphs taken underwater with high-latitude TMAX-100 film 16 array of neurons by the field of view of the CrabCam. Each
(Kodak). The turbid, shallow waters where Limulus mate frame of a movie thus yielded a 12 3 16 array of pixels
reduce the contrast of underwater objects exponentially with specifying the relative light intensities incident on each om-
distance (attenuation coefficient of '1.2 m21) such that matidium at an instant in time (see Fig. 1c), and the sequence
objects beyond 1.52 m are essentially invisible (18). of frames (Fig. 2a) comprised an optically transformed movie
Optic Nerve Recording. We recorded spike trains of single of the stimulus to the retina as the animal explored its
optic nerve fibers from behaving animals by using a water-tight underwater world.
microsuction electrode (17). Fig. 1a shows the recording The movie then was fed to the model, which computed spike
chamber and electrode attached to the carapace anterior to the trains for the ensemble of optic nerve fibers by using a set of
right lateral eye. To access the optic nerve, we trephined a hole differential equations representing the excitatory properties of
in the carapace '2 cm anterior to the eye, slid the tongue of each ommatidium and the inhibitory interactions between them.
the saline-filled recording chamber under the nerve trunk, and Details of the model are presented elsewhere (18). In brief,
fastened the chamber to the carapace. The sheath of the nerve incident light intensities were converted into voltage fluctuations
was slit open, and a single active fiber was teased free and (quantum bumps) according to an adapting bump model (20, 21).
sucked into the 200 mm electrode tip. The pixel of digitized To account for transduction noise, which contributes 610% to
movies that intercepted the optic axis of the recorded neuron the coefficient of variation of fluctuations in instantaneous firing
was identified by averaging frames coinciding with nerve rates in our experiments, quantum bumps were randomly gen-
impulses in response to a small probe light. Optic nerve signals erated by a Poisson process at a rate dependent on light intensity
were led via a cable (15 m length) to an AC amplifier (20) and summed to form the receptor potential by using a
(Electronics Shop, The Rockefeller University, New York) four-stage version of the FuortesHodgkin model (22). The
and the audio channel of the camcorder. The audio channel receptor potential then was attenuated by the passive cable
FIG. 2. (a) CrabCam images of the black and gray objects after optical sampling. The arrays of pixels show the light intensities incident on the
12 3 16 array of ommatidia viewing the videotaped scene. Rows 1 and 2 show images of the black object while the animal was moving and stationary,
respectively. Rows 3 and 4 show corresponding images of the gray object. Interval between displayed sequences is 0.25 s. (b) Computed neural images
of the optically transformed visual images in a. The arrays of pixels give the computed firing rates of optic nerve fibers mapped onto a gray scale
with black set to 0 ips and white set to twice the mean firing rate. Mean firing rates to the uniform background illumination in rows 1, 2, 3, and
4 were 12, 20, 13, and 18 ips, respectively. (c) Computed neural images of activity in the brain generated by the eyes output in b. The arrays of
pixels give the simulated responses of brain neurons having integration times of 400 ms mapped onto a gray scale with black set to 0 and white
set to twice the mean. Each image sums eight sequential neural images from the eye. Computed images are based on the temporal properties of
brain neurons and not their spatial interactions. (d) Simulated responses of brain neurons evoked by the activities of an equal number of ommatidia
located in a horizontal row of the model eye (dashed lines in c). Movies of the underwater video recordings and their computed neural images
can be viewed at Center for Vision Research at http:yywww.hscsyr.eduy;eye.
589
12652 Neurobiology: Passaglia et al. Proc. Natl. Acad. Sci. USA 94 (1997)
properties of the eccentric cell and integrated with lateral and of optic nerve responses are well represented by the model. Note
self-inhibitory potentials to form the generator potential (23). that the variability in firing rate when the black object is outside
The lateral inhibitory inputs to an eccentric cell were computed the recorded receptors field of view (coefficient of variation
by using a dynamic version of the original HartlineRatliff '0.27) is significantly greater than in the case of the gray one
formulation (7, 24) with inhibitory strength weighted as a function (coefficient of variation '0.1). This results from the field of view
of retinal distance (25, 26). The self-inhibitory input to an (black disk) of one receptor being oriented slightly lower and thus
eccentric cell was calculated by integrating a decaying exponential seeing more of the bright flickering light reflected from the sandy
function triggered by each impulse the cell fired (27). The bottom. The relative light intensities incident on these receptors
generator potential was further decremented by a putative elec- can be seen in Fig. 1c. Because transduction noise produces
trogenic pump (28, 29) and converted into a train of nerve '10% of the variation in a spike train for steady illumination,
impulses with a leaky integrate-and-fire encoder at a rate of 1 higher variability results from underwater lighting. Hence, both
impulseymV above a threshold of 1 mV (30). Computed trains of neural and environmental noise limit the detectability of these
impulses were expressed as instantaneous firing rates for each behaviorally relevant objects.
frame of the movie, converted to a linear gray scale, and mapped The good correspondence between theory and experiment
back onto two-dimensional arrays of pixels yielding a time series in Fig. 1 not only confirms the accuracy of the model but also
of neural images (31), which are snapshots of the eyes input to reveals an interesting difference in optic nerve responses to
the brain (Fig. 2b). high- and low-contrast objects. Note that as the animal moved
For all computations we set the 10 parameters of the model past the black object the firing rate of the optic nerve fiber
to those derived from the average spatiotemporal transfer initially decreased and then briefly increased (Fig. 1 e and f,
function of the eye measured in the laboratory (18, 26, 34). For Left); whereas, the firing rate to the gray object increased in a
each experiment we scaled two parameters to simulate illu- quasi-periodic fashion because of flickering light in the un-
mination levels in the field. We trimmed one parameter, the derwater scene (Fig. 1 e and f, Right). Although single optic
mean bump rate, to match the estimated transduction noise nerve fibers do not respond in the same way to the high- and
and scaled another, the sensitivity of the spike encoder, to low-contrast objects in the ocean, male horseshoe crabs detect
match the mean firing rate of the recorded neuron. We both objects about equally well.
extended these parameters to all ommatidia in the model Responses of Arrays of Optic Nerve Fibers. Clues about how
because previous studies have shown that cells in a given eye horseshoe crabs detect other horseshoe crabs and objects
have the same properties (15, 25). resembling them (Fig. 1b) can be found in the large arrays of
The accuracy of the model was checked first by comparing optic nerve responses, or neural images, computed by the
optic nerve responses computed for controlled stimuli with those model eye (Fig. 2b). Each pixel of a neural image corresponds
recorded from single optic nerve fibers in the laboratory, and to an ommatidium in the eye, and its gray level gives the optic
second by comparing responses computed for underwater movies nerve firing rate computed for that ommatidium at an instant
with those recorded in the field. Optic nerve responses to drifting in time. Inspection of the neural images in rows 1 and 3 of Fig.
bars recorded in the laboratory were indistinguishable from those 2b reveals small clusters of ommatidia responding with tran-
computed with the model, that is, responses recorded to succes- sient increases (white pixels) and decreases (black pixels) in
sive presentations of the same stimulus were as correlated with
firing rate to images of crab-size objects moving across the
those computed by the model as with each other (18). Correlation
visual field. Note that the eye encodes both moving high-
coefficients were consistently .0.95 for controlled laboratory
contrast (row 1) and low-contrast (row 3) objects with a
experiments (n 5 5) but were lower (.0.75) for field experiments
characteristic spatial pattern of optic nerve activity, even
(n 5 5) because of difficulties in determining precisely the
though the responses of single nerve fibers to the objects differ
stimulus to the recorded neuron. All field experiments yielded
(Fig. 1 d and e). Clusters of ommatidia responding together at
results similar to those reported here.
high rates (light regions) lie next to clusters responding at low
Measuring Transfer Functions. We measured the eyes
rates (dark regions). In laboratory simulations the light and
integrative properties by recording the responses of single
optic nerve fibers to sinusoidally modulated light of different dark regions merge together for smaller objects and separate
spatial and temporal frequencies presented to animals sub- for larger ones, reducing the modulation of optic nerve re-
merged in a seawater tank in the laboratory (n 5 8). Visual sponses in both cases (18). In the lower part of the visual field,
stimuli (VENUS, Neuroscientific, New York, NY) were dis- flickering light reflected off the sand modulates optic nerve
played on a monitor (10 3 13 cm, Tektronics, model 608, responses partially obscuring those to the crab-size objects.
Beaverton, OR) placed in front of a glass window on the tank,
4 cm from the eye, and centered on the optic axis of the
recorded neuron. Spatial transfer functions were measured by
drifting sinewave gratings of different spatial frequencies
across the monitor. Temporal transfer functions were mea-
sured by modulating a 6 spot of light (field of view of an
ommatidium) with the sum of five temporal frequencies
chosen to prevent phase locking of spike discharges (32).
Response gains were calculated for each of the five frequency
components by first estimating the relative modulation of the
recorded firing rate by using the method of least squares and
then normalizing the result with the contrast of the stimulus.
590
Neurobiology: Passaglia et al. Proc. Natl. Acad. Sci. USA 94 (1997) 12653
Processing Patterns of Optic Nerve Activity in the Brain. How region) for spatial frequencies from 0.008 to 0.04 cyclesydegree,
might neurons in the brain distinguish responses to moving which correspond to crab-size objects at distances of 0.25 to 1.4 m
crab-size objects from those evoked by other dynamic stimuli? A where 98% of visual detection occurs (13). With regard to the
possible answer is suggested by a recent finding that central temporal response properties of the eye, the duration of photo-
neurons integrate optic nerve signals with synaptic time constants receptor quantal responses to light (22, 33) and the dynamics of
on the order of 300500 ms (18). Integrating optic nerve re- self-inhibition (27) reduce its sensitivity to high and low temporal
sponses in Fig. 2b with a time constant of 400 ms (average of eight frequencies, respectively, resulting in a tuned temporal filter (26,
sequential neural images computed every 50 ms) yields neural 34). The eye is highly sensitive to images of crab-size objects
images of average brain activity in which responses to moving moving within the animals visual range at about the speed of a
crab-sized objects are enhanced relative to flicker in the fore- horseshoe crab (15 cmys) and up to twice these speeds for animals
ground (Fig. 2c). Profiles of simulated brain activity evoked by a passing one another in opposite directions (17). These speeds
horizontal row of optic nerve fibers reveal a 132% modulation to roughly correspond to temporal frequencies from 0.5 to 6 cyclesys
the moving black object [(18 impulsesys (ips)3 ips)y12 ips] and for which response gains exceed one (Fig. 4b, region shaded with
an 87% modulation to the moving gray one [(19 ips8 ips)y13 ips] negative slope lines), underscoring the eyes high sensitivity to
(Fig. 2d). The greater response modulation to the black object moving objects. The spatiotemporal properties of the eye can
may correspond to the field observation that animals can see it readily account for the animals ability to see high-contrast
slightly better than the gray one (13). Integrating optic nerve objects but not low-contrast ones.
responses over periods longer or shorter than 400 ms yields less Seeing Low-Contrast Objects. Natural fluctuations of under-
robust responses to the moving crab-size objects (Fig. 3). An water lighting enhance the visibility of low-contrast objects (35).
integration time of 400 ms gives maximal signal-to-noise ratio as Surface waves, acting as dynamic lenses, create beams of light that
defined by half of the peak-to-peak modulation of the response strobe the underwater scene at frequencies of '26Hz for which
to the object divided by the standard deviation of responses to the lateral eye is maximally sensitive (Fig. 4b, region shaded with
surrounding regions of the scene. Integrative mechanisms in the positive slope lines; ref. 36). Such wave-induced flicker is a
brain thus appear well suited for sensing moving crab-size objects prominent feature of the shallow waters where Limulus mate and
in the neural images it receives from the eye. persists over a range of lighting conditions (37), which is consis-
The Importance of Motion. The neural images of stationary tent with the animals ability to find mates day and night (1013).
objects dramatically illustrate the essential role of motion to Behavioral studies in mating areas show that strobic light in-
retinal coding. Whereas visual images of stationary objects creases the probability that male Limulus turn toward and
appear similar to those of moving objects (Fig. 2a), their neural approach the low-contrast gray object (38). Model computations
images do not. The objects are hardly recognizable in the show that the strobic light reflects off low-contrast objects,
computed neural images of optic nerve activity (Fig. 2b) and enhancing their neural images as the objects move across the
nearly invisible in the neural images of simulated brain activity animals visual field (Fig. 2b, compare images in rows 3 and 4). We
(Fig. 2c) when they are not moving in the animals visual field. note that strobic light also evokes coherent bursts of nerve
Indeed, response profiles in Fig. 2d reveal an amplitude of impulses from neighboring ommatidia in the absence of relative
modulation of just 27% [(20 ips14 ips)y20 ips)] for the black motion. How then does relative motion of objects enhance their
object and 22% [(20 ips16 ips)y18 ips)] for the gray one. neural image? Motion of an image across the visual field activates
Image motion thus produced over a 4-fold increase in gain for receptors along its path (Fig. 2b, row 3, white pixels) leaving
the black (132%y27%) and gray (87%y22%) objects.
Seeing High-Contrast Objects. Strongly modulated responses
to moving images of high-contrast objects (Fig. 2b, row 1) are
understandable in terms of the eyes optical and neural proper-
ties. Coarse sampling by the ommatidia (14) and lateral inhibitory
interactions among them (7) reduce the eyes sensitivity to high
and low spatial frequencies, respectively, yielding a broadly tuned
spatial filter (26). The eye is maximally sensitive (Fig. 4a, shaded
591
12654 Neurobiology: Passaglia et al. Proc. Natl. Acad. Sci. USA 94 (1997)
adapted ones in its wake (Fig. 2b, row 3, dark pixels). Without 4. Lettvin, Y. L., Maturana, H. R., McColluch, W. S. & Pitts, W. H.
motion of images across its receptor surface, the eye provides no (1959) Proc. Inst. Radio Eng. 47, 19401951.
cues for distinguishing flickering light reflected off the object 5. Levine, J. S. & MacNichol, E. F. (1979) Sens. Processes 3, 95131.
from that reflected off the sand, making these parts of the scene 6. Laughlin, S. B. (1996) Vision Res. 36, 15291541.
7. Hartline, H. K. & Ratliff, F. (1957) J. Gen. Physiol. 40, 357376.
appear continuous (Fig. 2, row 4). In short, no visual cues
8. Hartline, H. K. & Ratliff, F. (1958) J. Gen. Physiol. 41, 10491066.
distinguish object from foreground in the absence of relative 9. Barlow, R. B., Prakash, R. & Solessio, E. (1993) Amer. Zool. 33,
motion. Subsequent integration by the brain further enhances the 6678.
neural images of moving objects (Fig. 2c) because the integration 10. Barlow, R. B., Ireland, L. C. & Kass, L. (1982) Nature (London)
time of brain neurons matches the velocity range of horseshoe 296, 6566.
crabs and not that of the fast-moving flicker. Flicker-induced 11. Barlow, R. B., Powers, M. K., Howard, H. & Kass, L. (1986) Biol.
responses (26 Hz) are shorter than the integration time of 400 Bull. Woods Hole, Mass. 171, 310329.
ms whereas object motion across receptors is longer ('2 s), 12. Powers, M. K., Barlow, R. B. & Kass, L. (1991) Visual Neurosci.
yielding a substantial response envelope (Fig. 2 c and d). Image 7, 179189.
motion thus binds together the activities of neighboring clusters 13. Herzog, E. D., Powers, M. K. & Barlow, R. B. (1996) Visual
Neurosci. 13, 3141.
of ommatidia in space and time, sending robust signals to the
14. Herzog, E. D. & Barlow, R. B. (1992) Visual Neurosci. 9, 571580.
brain about potential mates. 15. Ratliff, F., ed. (1974) Studies on Excitation and Inhibition in the
Retina (Rockefeller Univ. Press, New York).
CONCLUSION 16. Knight, B. W., Toyoda, J. & Dodge, F. A. (1970) J. Gen. Physiol.
56, 421437.
Adrian and Hartline were the first to study neural coding in 17. Herzog, E. D. (1994) Dissertation (Syracuse University, New
single sensory fibers. They reported that the mean firing rate York).
of nerve impulses encoded various elementary features of 18. Passaglia, C. L. (1997) Dissertation (Syracuse University, New
stimuli to proprioreceptors (39) and visual receptors (40). York).
Later studies in the auditory systems showed that the precise 19. Barlow, R. B., Chamberlain, S. C. & Levinson, J. Z. (1980)
Science 210, 10371039.
time of occurrence of nerve impulses transmits information 20. Dodge, F. A., Knight, B. W. & Toyoda, J. I. (1968) Science 232,
about sound frequencies and their location in space (41, 42). 15431545.
Similar claims have been made recently for the visual systems 21. Wong, F., Knight, B. W. & Dodge, F. A. (1980) J. Gen. Physiol.
of salamanders and rabbits (43). Such a timing code appears 76, 517537.
not to have an important role in Limulus vision because 22. Fuortes, M. G. F. & Hodgkin, A. L. (1964) J. Physiol. 172,
neurons in the brain integrate over periods of 250500 ms 239263.
trains of optic nerve impulses with interspike intervals ranging 23. Purple, R. L. & Dodge, F. A. (1965) Cold Spring Harbor Symp.
from 25 to 200 ms (18). As a consequence of strobic under- Quant. Biol. 30, 529537.
water lighting, spatiotemporal properties of the eye, and 24. Ratliff, F., Hartline, H. K. & Miller, W. H. (1963) J. Opt. Soc. Am.
temporal properties of brain cells, objects having the size, 53, 110120.
25. Barlow, R. B. (1969) J. Gen. Physiol. 54, 383396.
contrast, and motion of horseshoe crabs are not distinguished 26. Brodie, S. E., Knight, B. W. & Ratliff, F. (1978) J. Gen. Physiol.
by unique neural codes. Low-pass integrative properties of 72, 167202.
brain neurons filter out rapid changes, yielding response 27. Stevens, C. F. (1964) Dissertation (The Rockefeller University,
envelopes that represent coherent changes in the mean firing New York).
rates among ommatidia viewing a moving object. Fig. 5 shows 28. Smith, T. G., Stell, W. K., Brown, J. E., Freeman, J. A. & Murray,
that estimates from such a rate code can account reasonably G. C. (1968) Science 162, 456458.
well for the visual performance of horseshoe crabs. 29. Fohlmeister, J. F., Poppele, R. E. & Purple, R. L. (1977) J. Gen.
The Limulus eye processes visual information with excitatory Physiol. 69, 849877.
and inhibitory mechanisms similar to those found in the first 30. Barlow, R. B. & Kaplan, E. H. (1977) J. Gen. Physiol. 69, 203220.
31. Laughlin, S. B. (1981) in Comparative Physiology and Evolution of
synaptic layer, or outer plexiform layer, of the vertebrate retina.
Vision in Invertebrates: Invertebrate Visual Centers and Behavior,
Both retinas integrate photoreceptor signals with lateral inhibi- ed. Autrum, H. (Springer, New York), pp. 133280.
tory mechanisms, but the vertebrate retina adds a second synaptic 32. Victor, J. D., Shapley, R. M. & Knight, B. W. (1977) Proc. Natl.
layer of processing, the inner plexiform layer, before transmitting Acad. Sci. USA 74, 30683072.
optic nerve signals to the brain (1). With two layers of neural 33. Kaplan, E., Barlow, R. B., Renninger, G. & Purpura, K. (1990)
processing, the frogs eye, for example, can extract specific J. Gen. Physiol. 96, 665685.
features of its visual world such as small, moving dark objects 34. Batra, R. & Barlow, R. B. (1990) J. Gen. Physiol. 95, 229244.
about the size of a fly and transmit information about them to the 35. McFarland, W. N. & Loew, E. R. (1983) Environ. Biol. Fishes 8,
brain (4, 44). The integrative mechanisms of the Limulus eye also 173183.
extract information about moving objects, encoding those resem- 36. Passaglia, C. L., Dodge, F. A. & Barlow, R. B. (1995) Biol. Bull.
Woods Hole, Mass. 189, 213215.
bling a horseshoe crab with characteristic patterns of coherent
37. Loew, E. R. & McFarland, W. N. (1990) in The Visual System of
activity among clusters of optic nerve fibers. Neurons in the first Fish, eds. Douglas, R. H. & Djamgoz, M. B. A. (Chapman and
synaptic layer of the brain integrate optic nerve signals, further Hall, London), pp. 143.
enhancing the neural images of potential mates. 38. Passaglia, C. L., McSweeney, M., Stewart, K., Kim, E., Mole, E.,
Powers, M. & Barlow, R. B. (1997) Biol. Bull. Woods Hole, Mass.,
We thank A. Meigs, R. Mitchell, A. Wixson, S. Dodge, M.E. in press.
Kelly-Manglapus, D. Porcello, L. Sher, E. Kaplan, D. Samber, and M. 39. Adrian, E. D. (1928) The Basis of Sensation: The Action of the
Powers for their help. This work was supported by the National Sense Organs (Norton, New York).
Institutes of Health (MH 49741 and EY00667) and the National 40. Hartline, H. K. & Graham, C. H. (1932) J. Cell Comp. Physiol. 1,
Science Foundation (IBN9696208). 227295.
41. Kiang, N. Y.-S., Watanabe, T., Thomas, E. C. & Clark, L. F.
1. Dowling, J. E. (1987) The Retina: An Approachable Part of the (1965) Discharge Patterns of Single Fibers in the Cats Auditory
Brain (Harvard Univ. Press, Cambridge, MA). Nerve (MIT Press, Cambridge, MA).
2. Rieke, F., Warland, D., DeRuyter van Steveninck, R. R. & 42. Carr, C. E. & Konishi, M. (1990) J. Neurosci. 10, 32273246.
Bialek, W. (1997) Spikes: Exploring the Neural Code (MIT Press, 43. Berry, M., Warland, D. & Meister, M. (1997) Proc. Natl. Acad.
Cambridge, MA). Sci. USA 94, 54115416.
3. Bialek, W. & Owen, W. G. (1990) Biophys. J. 58, 12271233. 44. Barlow, H. B. (1953) J. Physiol. 119, 6988.
592
Cell-Based Model of the Limulus Lateral Eye
Christopher L. Passaglia, Frederick A. Dodge and Robert B. Barlow
J Neurophysiol 80:1800-1815, 1998.
Updated information and services including high resolution figures, can be found at:
http://jn.physiology.org/content/80/4/1800.full.html
Additional material and information about Journal of Neurophysiology can be found at:
http://www.the-aps.org/publications/jn
Journal of Neurophysiology publishes original articles on the function of the nervous system. It is published 12 times a year (monthly)
by the American Physiological Society, 9650 Rockville Pike, Bethesda MD 20814-3991. Copyright 1998 by the American
Physiological Society. ISSN: 0022-3077, ESSN: 1522-1598. Visit our website at http://www.the-aps.org/.
593
Cell-Based Model of the Limulus Lateral Eye
Passaglia, Christopher L., Frederick A. Dodge, and Robert B. recently been used to study patterns of activity generated by
Barlow. Cell-based model of the Limulus lateral eye. J. Neurophys- ensembles of retinal neurons, but neither technique has yet
iol. 80: 18001815, 1998. We present a cell-based model of the proven practical for use with behaving animals. An alterna-
Limulus lateral eye that computes the eyes input to the brain in tive approach is to construct a realistic computational model
response to any specified scene. Based on the results of extensive
physiological studies, the model simulates the optical sampling of
of the eye from anatomic and physiological data. Because
visual space by the array of retinal receptors (ommatidia), the of their relative simplicity, the eyes of lower vertebrates and
transduction of light into receptor potentials, the integration of invertebrates appear to offer the best opportunities (Teeters
excitatory and inhibitory signals into generator potentials, and the et al. 1997; Werblin 1991). The lateral eye of the horseshoe
conversion of generator potentials into trains of optic nerve im- crab, Limulus polyphemus, is a particularly attractive model
limited the size of our model to a 16 1 16 array of ommatidia in the eye, all model ommatidia are assigned the same acceptance
the central region of the retina. Anatomic and physiological studies angle (Dr 2.35sa).
(Fig. 2A) show that interommatidial angles of adjacent columns of We specify the relative light intensities In(t) incident at time t
this array are roughly constant, but those of adjacent rows vary on all n ommatidia of the model by first convolving individual
such that the eye samples visual space more densely below the frames of the video input with the spread function A(r). We then
horizon (Barlow et al. 1984; Dodge and Kaplan 1977; Herzog and extract from the blurred images the pixel values at points of inter-
Barlow 1992; Weiner and Chamberlain 1993). We fit the diver- section {ui ,fj} with the optic axes of corresponding ommatidia
gence of optic axes with the following empiric equations that map {i,j} in the crab eye as shown in Fig. 2B. Normalizing the individual
the array of optic axes from retinal coordinates {i,j} to angular pixel values of an image by their collective mean over all images
coordinates {ui ,fj} yields the relative light intensities In(t) incident on the ommatidial
array at discrete moments in time.
ui 6i (1)
S D
1 r2
potential Sn(t) recorded from the eccentric-cell soma. This light-
A(r) 2
exp 0 2 (3) sensitive conductance acts in parallel with the passive electrical
2psa 2sa
properties of the soma represented by resistance RS and capacitance
CS . The somatic compartment is coupled to an axonal compartment
where r is the angle (in degrees) of incidence with respect to the by resistance RC . The coupling resistance represents the short
optic axis. Because no evidence exists for regional differences in length of passive membrane separating the soma from the synaptic
ates bumps of random amplitude an at random times of photon The output of the last stage gives the light-modulated excitatory
absorption tn with a rate ln(t) that is a fixed percentage of the conductance gEn(t) of the ommatidium.
incident light intensity In(t). Note that this is a nonhomogeneous LATERAL INHIBITION gLIn(t). Excitatory photoresponses drive an
process because the rate changes as In(t) changes (Shanmugan and inhibitory network of eccentric cells whose properties have been
Breipohl 1988). We select the amplitude of each bump an that well described in the classic work of Hartline, Ratliff, and col-
occurs at time t from an exponential distribution having an instanta- leagues. By measuring the interactions among individual omma-
neous mean value an(t), which constantly changes in response to tidia, Hartline and Ratliff (1957, 1958) found that the steady-state
the incident light intensity In(t). That is, an(t) shrinks over time in firing rate rV n of the nth ommatidium is equal to its uninhibited
proportion to the fraction of the available clusters being activated. response eV n minus inhibitory influences of its N 0 1 neighbors.
We model bump shrinkage as This relationship can be expressed by the following set of piece-
Shrinkage 0 F G
Pln(t)
gn(t)
where Pln(t) aP nrd(t 0 tO n)
i
(12)
wise linear equations
F N
rV n eV n 0 knmr(rV m 0 r onm)/ G /
,
nxm
H
where hats () indicate random variables. If, for example, excited
rhodopsin molecules activate one-half of the available clusters, z if z0
z/ (20)
then the bump amplitude shrinks toward one-half its previous value 0 otherwise n 1, . . , N
in the next time step. However, to maintain constant bump ampli-
tude under steady illumination, an(t) must also grow as
where superscript / is an operator that ensures nonnegative firing
Growth G[an(t)] (13) rates. The amount of inhibition exerted by neighboring ommatid-
ium m on n depends on its steady-state firing rate rV m , its threshold
We assume that the growth of bump amplitude G[an(t)] is a func-
F S D G
(18)
dt tb i2 j2
knm } KLI exp 0 2
0 2 0 exp(0i2 0 j2) (25)
si sj
gEn(t) b4n(t) (19)
where the horizontal separation between ommatidia n and m is mate lV by taking advantage of the relationship between the state
measured as i columns, and the vertical separation is measured as of bump adaptation and the characteristics of retinal noise ex-
j rows. After computing the matrix of coefficients, we rescale their pressed in Eq. 7. Specifically, we assume that the variability in the
values so that the total inhibition converging on each unit is KLI . firing rate of optic nerve impulses under steady illumination reflects
For the simulation reported here, we assigned unit amplitude and noise generated by random fluctuations in the number and ampli-
unit width to the central crater in Eq. 25 (Brodie et al. 1978b) and tude of bumps (Kaplan and Barlow 1975). We then adjust the mean
scaled the broad function equally in the vertical and horizontal bump rate lV until the coefficient of variation of firing rate computed
directions (i.e., si sj sLI 4 ommatidia). by the model matches that recorded from single nerve fibers of an
SELF INHIBITION gSIn(t). Ommatidia in the lateral eye inhibit not experimental eye in a given lighting environment.
only their neighbors but themselves as well. Every action potential To simulate phototransduction noise, we must generate two ran-
fired by an eccentric cell feeds back an inhibitory postsynaptic poten- dom numbers for each bump: one to set the interval between bumps
tial onto itself that reduces its likelihood of firing another spike in and the other to set the amplitude of a bump. In our experiments,
the immediate future (Purple and Dodge 1966; Stevens 1964). Self- mean bump rates lV typically range from 104 bumps/s in the
inhibition was not explicitly included in the original Hartline-Ratliff laboratory to 106 bumps/s outdoors. Simulation of such high
equations (1958), resulting in their overestimate of the strength of rates is a tremendous computational burden that we avoid by first
lateral inhibition KLI . We treat the self-inhibitory conductance gSIn(t) replacing the bump generator term Pln(t) in Eqs. 14 and 15 with
as a decaying exponential function having a strength KSI and time ln(t)ran(t) to compute a noise-free response. We then create bumps
constant tSI yielding the following differential equation at a lower rate and correspondingly greater amplitude without
changing the kinetics of bump adaptation and add the noise to the
dgSIn gSIn noise-free response. This allows the model to simulate the dynamic
0 / GSI fn(t) (26)
dt tSI response of the receptor potential to time-varying stimuli in an
efficient and accurate manner over a wide range of light levels.
where GSI is the strength of self inhibition in terms of an equivalent
TABLE 1. Parameters of the cell-based model of the Limulus lateral eye for simulating optic nerve responses of three experimental
eyes and the standard eye
Parameter Units Effect Eye I Eye II Eye III Standard Eye SIg SIb
Letters refer to features of the spatiotemporal transfer function in Fig. 6 most affected by the parameter; asterisks indicate parameters having nonspecific
affects on the transfer function. Sensitivity indexes for drifting gratings (SIg) were calculated as the percent change in the affected feature divided by
the percent change in the value of the modified parameter (50%). Sensitivity indexes for moving crab-size bars (SIb) were calculated by integrating the
difference between the average response computed with the standard value of a parameter and a 50% increase in its value, and then normalizing the
result by that of the parameter having the maximal effect. lU , mean bump rate; Dr, acceptance angle; tb, bump time constant; amax, maximum bump
amplitude; KLI, strength of lateral inhibition; sLI, space scale of inhibitory field; tLI, time constant of lateral inhibition; KSI, strength of self inhibition;
tSI, time constant of self inhibition; S, sensitivity of spike encoder.
FIG. 4. Stages of analysis of model computations. A: firing rate recorded from an optic nerve fiber in response to a drifting
sinusoidal grating having a spatial frequency of 0.016 cycles/deg and temporal frequency of 1 Hz. B: firing rate computed
for an optic nerve fiber in the model eye in response to the same drifting grating. C: firing rate of an optic nerve fiber
predicted by the model in response to a simulated crablike object moving across the visual field. D: firing rate of an optic
nerve fiber subsequently recorded in response to the simulated crablike object. Top and bottom traces give responses of an
optic nerve fiber to multiple presentations of the stimulus and its average response, respectively.
ground) moving repeatedly for 90 s across the monitor placed 9 spatiotemporal frequencies, whereas eye II was the most
which is equivalent to 9 ommatidia because their optic axes erties of the receptor that the spot stimulates, whereas re-
diverge by 67 (Barlow et al. 1984; Herzog and Barlow sponses to full-field stimuli include those of the inhibitory
1992). Such a configuration agrees with previous inhibitory network as well (Knight et al. 1970; Ratliff et al. 1967,
field maps (Barlow 1969; Brodie et al. 1978b). Beyond the 1969). For both small-spot and full-field conditions, the gain
peak, the gain of the eye falls sharply because the 67 of the eye increases monotonically with temporal frequency
acceptance angle of the optical apparatus (Barlow et al. to a peak in the region of 36 Hz and then falls sharply.
1980) filters fine patterns of light, limiting the spatial acuity Over a large range of temporal frequencies (0.516 Hz), the
of the animal (Fig. 6, arrow c). gain is greater than one, indicating that the eye amplifies the
The spatial band-pass tuning conferred by the optics and stimulus in its output. We typically measured peak gains of
lateral-inhibitory network diminishes with increasing tempo- up to 510, with the eye showing greater sensitivity to small
ral frequency because synaptic delays shift the phase of in- spots at low temporal frequencies and large spots at interme-
hibitory signals relative to that of excitatory signals. The diate frequencies. We conclude that the Limulus eye is mark-
shift in phase can disinhibit retinal neurons enhancing edly tuned for light oscillating at temporal frequencies in
the eyes response to temporal frequencies corresponding to the range of 46 Hz.
one-half the synaptic delay for inhibition (Ratliff et al. 1967). Dynamic interactions of phototransduction, self inhibition,
Amplification by lateral inhibition is evident in the temporal and lateral inhibition shape the temporal transfer function.
transfer functions presented next. Bump adaptation and self inhibition (KSI of 23) combine
TEMPORAL TRANSFER FUNCTION. The lateral eye is not to reduce the amplitude of the eyes response to slowly
equally sensitive to all temporal frequencies either. It is max- modulated spots of light (Fig. 6, arrow e). Lateral inhibition
imally responsive to both small-spot (s) and full-field () (KLI of 4) further decreases its response to larger areas of
flicker of intermediate temporal frequencies (Fig. 5B). Re- stimulation, i.e., low spatial frequency (Fig. 6, arrow a).
sponses to small-spot stimuli reflect only the excitatory prop- These relatively slow mechanisms become less effective at
no other mechanisms affect these features of the transfer standable because the steady-state response of the eye de-
function. The spatiotemporal response properties are, on the pends logarithmically on light intensity (Eq. 9). Insensitivity
other hand, less sensitive to comparable changes in mean to halving of self-inhibitory parameters suggests that self
bump rate and the time constant and strength of self inhibi- inhibition is not a finely tuned element of the eyes design.
tion. The eye is about threefold less sensitive to these param- Note also that changes in more than one parameter of the
eters. Insensitivity to halving of the mean bump rate is under- model often offset one another because many parameters
affect overlapping regions of the transfer function.
moving dark bar, as predicted for the corresponding omma- Indeed, not only was the overall shape of recorded responses
tidium in the model (compare with Fig. 7). The relative to moving bars predicted by the model, but also the random
modulation of firing rate recorded from the three experimen- fluctuations in recorded spike rate.
tal eyes was 31, 48, and 38% for the slowest speed, 47, 73, There are, nonetheless, detectable differences between av-
and 50% for the medium speed, and 64, 99, and 75% for erage computed and recorded responses. For example, com-
the fastest speed in good agreement with model simulations. puted transients are slightly more abrupt than those recorded
from the eye. To assess the accuracy of model predictions,
differences must be viewed in relation to the intrinsic noise
level in single optic nerve fibers, because averaging is not
an option under natural conditions. The mean correlation
coefficient between individually recorded responses of the
three eyes and their combined average was 0.95, 0.97, and
0.97 at the three stimulus velocities, whereas the correlation
coefficients between average recorded and computed re-
sponses were 0.97, 0.98, and 0.98, respectively (Table 2).
That is, average responses of the recorded neuron to dark
bars moving at different velocities are as correlated with
those computed by the model as with individually recorded
responses. Hence one cannot distinguish between responses
of the model and the eye without significant averaging to
DISCUSSION
sponses (Figs. 6 and 7), suggesting that the eyes encoded generate primarily low temporal frequencies (2 Hz), the
crablike bars in essentially the same manner. To further response of the eye was also fairly sensitive to the maxi-
ratio was over threefold less (2.7) for this snapshot. Random
fluctuations in firing rate obscure the weak pattern of neural
activity evoked by the light object. Object-independent fluc-
tuations in firing rate will be even larger in the animals
underwater environment (Passaglia et al. 1995). Yet, behav-
ioral studies show that male horseshoe crabs can see crablike
objects of similar size and contrast as those used in this
study almost equally well day and night (Herzog et al. 1996).
Passaglia et al. (1997) address how horseshoe crabs achieve
this seemingly remarkable visual performance.
In summary, retinal coding of visual behavior has been
under investigation since Hartline first isolated the spike
trains of individual nerve cells in the eye (Hartline and Gra-
ham 1932). For over 65 years, inferences have been made
about retinal processing as a whole from the vantage point
of single neurons. Multielectrode recordings from excised
retina promise to expand our knowledge about how ensem-
bles of retinal neurons transmit information (DeVries and
Baylor 1997; Meister 1996). Studies of retinal coding may
eventually progress to the point when the major portion of
Address for reprint requests: R. B. Barlow, Center for Vision Research, HARTLINE, H. K. AND GRAHAM, C. H. Nerve impulses from single receptors
Dept. of Ophthalmology, SUNY Health Science Center, Syracuse, NY in the eye. J. Cell. Comp. Physiol. 1: 227295, 1932.
13210. HARTLINE, H. K. AND RATLIFF, F. Inhibitory interactions of receptor units
in the eye of Limulus. J. Gen. Physiol. 40: 357376, 1957.
Received 24 October 1997; accepted in final form 8 June 1998. HARTLINE, H. K. AND RATLIFF, F. Spatial summation of inhibitory influ-
ences in the eye of Limulus, and the mutual interactions of receptor units.
REFERENCES J. Gen. Physiol. 41: 10491066, 1958.
ADOLPH, A. R. Spontaneous slow potential fluctuations in the Limulus pho- HARTLINE, H. K., WAGNER, H. G., AND MACNICHOL, E. F. The peripheral
toreceptor. J. Gen. Physiol. 48: 297322, 1964. origin of nervous activity in the visual system. Cold Spring Harb. Symp.
BARLOW, H. B. Summation and inhibition in the frogs retina. J. Physiol. Quant. Biol. 17: 125141, 1952.
(Lond.) 119: 6988, 1953. HERZOG, E. H. Vision in Limulus: From Optics to Neurons to Behavior
BARLOW, R. B. Inhibitory fields in the Limulus lateral eye. J. Gen. Physiol. (PhD thesis). Syracuse, NY: Syracuse Univ., 1994.
54: 383396, 1969. HERZOG, E. H. AND BARLOW, R. B. The Limulus-eye view of the world.
BARLOW, R. B., CHAMBERLAIN, S. C., AND KASS, L. Circadian rhythms in Vis. Neurosci. 9: 571580, 1992.
retinal function. In: Molecular and Cellular Basis of Visual Acuity, edited HERZOG, E. H., PASSAGLIA, C. L., DODGE, S. A., LEVINE, N. D., AND BAR-
S. R. Hilfer and J. B. Sheffield. New York: Springer-Verlag, 1984, p. LOW, R. B. Limulus vision in the ocean: comparing neural and behavioral
3153. thresholds. Biol. Bull. 185: 307308, 1993.
BARLOW, R. B., CHAMBERLAIN, S. C., AND LEHMAN, H. K. Circadian HERZOG, E. H., POWERS, M. K., AND BARLOW, R. B. Limulus vision in the
rhythms in the invertebrate retina. In: Facets of Vision, edited by D. G. ocean day and night: effects of image size and contrast. Vis. Neurosci.
Stavenga and R. Hardie, Berlin: Springer-Verlag, 1989. 13: 3141, 1996.
BARLOW, R. B., CHAMBERLAIN, S. C., AND LEVINSON, J. Z. The Limulus KAPLAN, E. AND BARLOW, R. B. Properties of visual cells in the lateral eye
brain modulates the structure and function of the lateral eye. Science of Limulus in situ. Extracellular recordings. J. Gen. Physiol. 66: 303
210: 10371039, 1980. 326, 1975.
BARLOW, R. B., IRELAND, L. C., AND KASS, L. Vision has a role in the KAPLAN, E., BARLOW, R. B., RENNINGER, G., AND PURPURA, K. Circadian
RATLIFF, F., KNIGHT, B. W., TOYODA, J.-I., AND HARTLINE, H. K. Enhance- WEINER, W. W. AND CHAMBERLAIN, S. C. The visual fields of American
ment of flicker by lateral inhibition. Science 158: 392393, 1967. horseshoe crabs: two different eye shapes in Limulus polyphemus. Vis.
RENNINGER, G. H., KASS, L., PELLETIER, J. L., AND SCHIMMEL, R. The ec- Neurosci. 11: 333346, 1993.
centric cell of the Limulus lateral eye: encoder of circadian changes in WELCH, P. D. The use of fast Fourier transform for the estimation of power
visual responses. J. Comp. Physiol. [A] 163: 259270, 1988. spectra: a method based on time averaging over short, modified periodo-
SHANMUGAN, K. S. AND BREIPOHL, A. M. Random Signals: Detection, Esti- grams. IEEE Trans. Audio Electroacoustics AU-15: 70, 1967.
mation, and Data Analysis. New York: Wiley, 1988. WERBLIN, F. S. Synaptic connections, receptive fields, and patterns of activ-
SHAPLEY, R. Fluctuations of the impulse rate in Limulus eccentric cells. J. ity in the tiger salamander retina. Invest. Ophthalmol. Vis. Sci. 32: 459
Gen. Physiol. 57: 539556, 1971a. 483, 1991.
SHAPLEY, R. Effects of lateral inhibition on fluctuations of the impulse rate. WONG, F. Nature of light-induced conductance changes in ventral photore-
J. Gen. Physiol. 57: 557575, 1971b. ceptors of Limulus. Nature 276: 7679, 1978.
SMITH, T. G., STELL, W. K., BROWN, J. E., FREEMAN, J. A., AND MURRAY, WONG, F. AND KNIGHT, B. W. Adapting-bump model for eccentric cells of.
G. C. A role for the sodium pump in photoreception in Limulus. Science Limulus. J. Gen. Physiol. 76: 539557, 1980.
162: 456458, 1968. WONG, F., KNIGHT, B. W., AND DODGE, F. A. Dispersion of latencies in
STEVENS, C. F. A Quantitative Theory of Neural Interactions: Theoretical photoreceptors of Limulus and the adapting-bump model. J. Gen. Physiol.
and Experimental Investigations (PhD thesis). New York: The Rockefel- 76: 517537, 1980.
ler Univ., 1964. WONG, R. O., CHERNJAVSKY, A., SMITH, S. J., AND SHATZ, C. J. Early
TEETERS, J., JACOBS, A., AND WERBLIN, F. How neural interactions form neural functional networks in the developing retina. Nature 374: 716 718,
responses in the salamander retina. J. Comp. Neurosci. 4: 527, 1997. 1995.
VICTOR, J. D., SHAPLEY, R. M., AND KNIGHT, B. W. Nonlinear analysis of YEANDLE, S. Studies on the Slow Potential and the Effects of Cations
cat retinal ganglion cells in the frequency domain. Proc. Natl. Acad. Sci. on the Electrical Responses of the Limulus Ommatidium (PhD thesis).
USA 74: 30683072, 1977. Baltimore, MD: The Johns Hopkins Univ., 1957.
610