1999 Nature America Inc. http://neurosci.nature.

com

contents

volume 2 no 2 february 1999

http://neurosci.nature.com

Three-dimensional reconstruction
of a Bergmann glial cell. Grosche
and colleagues show that
editorial
microdomains in these glia wrap Science and terrorism in Europe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
1999 Nature America Inc. http://neurosci.nature.com

around synaptic structures. See

page 139.

letter to the editor

Overuse of impact factors suppresses controversial ideas . . . . . . . . . . . . . . . . . . 101

news and views

Speech boundaries, syntax and the brain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Cyma Van Petten and Paul Bloom
SEE ARTICLE, PAGE 191

Coupling calcium to SNARE-mediated synaptic vesicle fusion . . . . . . . . . . . . . . . 104

Sabine Hilfiker, Paul Greengard and George J. Augustine
SEE ARTICLE, PAGE 119
Snapin may link calcium
and exocytosis.
Pages 104 and 119.
The smell of adrenaline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Stuart Firestein and Anna Menini
SEE ARTICLE, PAGE 133

Weighing the evidence: how the brain makes a decision . . . . . . . . . . . . . . . . . . . 108

Jeffrey D. Schall
SEE ARTICLE, PAGE 176

obituary
Sir Alan Lloyd Hodgkin. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
By Denis Baylor and King-Wai Yau

book review
Sir Alan Hodgkin, 19141998. The Biophysics of Computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Page 111. BY CHRISTOF KOCH
Reviewed by Eve Marder

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature America Inc., headquartered at 345 Park Avenue South, New York, NY 10010-1707. Editorial Office: 345 Park
Avenue South, New York, NY 10010. Telephone 212 726 9200, Fax (212) 696 9635. North American Advertising: Nature Neuroscience, 345 Park Avenue South, New York, NY 10010-
1707. Telephone (212) 726-9200. Fax (212) 696-9006. European Advertising: Nature Neuroscience, Porters South, Crinan Street, London N1 9SQ. Telephone (0171) 833 4000. Fax
(0171) 843 4596. New subscriptions, renewals, changes of address, back issues, and all customer service questions in North America should be addressed to Nature Neuro-
science Subscription Department, PO Box 5054, Brentwood, TN 37024-5054. Telephone (800) 524-0384, Direct Dial (615) 377 3322, Fax (615) 377 0525. Outside North America: Nature
Neuroscience, Macmillan Magazines Ltd., Houndsmill, Brunel Road, Basingstoke, RG21 6XS, U.K.. Tel: +44-(0)1256-329242. Fax: +44-(0)1256 812358. Email: subscriptions@nature.com.
Annual subscription rates: U.S./Canada: U.S. $595, Canada add 7% for GST (institutional/corporate), U.S. $195, Canada add 7% for GST (individual making personal payment BN:
14091 1595 RT); U.K./Europe:395 (institutional/corporate), 175 (individual making personal payment), 99 (student); Rest of world (excluding Japan): 450 (institutional/corpo-
rate), 195 (individual making personal payment), 110 (student); Japan: Contact Japan Publications Trading Co. Ltd., 2-1 Sarugaku-cho 1 chome, Chiyoda-ku, Tokyo 101, Japan, phone
(03) 292-3755. Back issues: U.S./Canada, $45, Canada add 7% for GST; Rest of world: surface U.S. $43, air mail U.S. $45. Reprints: Nature Neuroscience Reprints Department, 345 Park
Avenue South, New York, NY 10010-1707. Subscription information is available at the Nature Neuroscience homepage at http://neurosci.nature.com. POSTMASTER: Send address changes
to Nature Neuroscience Subscription Department, P.O. Box 5054, Brentwood, TN 37024-5054. Executive Officers of Nature America Inc: Nicholas Byam Shaw, Chairman of the Board;
Mary Waltham, President; Edward Valis, Secretary-Treasurer. Printed by Publishers Press, Shepherdsville, KY, USA. Copyright 1999 Nature America Inc.

nature neuroscience volume 2 no 2 february 1999 i

 1999 Nature America Inc. http://neurosci.nature.com

contents

scientific correspondence
Neurotrophin-3 is required for proper cerebellar development . . . . . . . . . . . . . . 115
B Bates, M Rios, A Trumpp, C Chen, G Fan, J M Bishop and R Jaenisch

articles
NT-3 and cerebellar development.
Page 115. Snapin: a SNARE-associated protein implicated in synaptic transmission . . . . . . 119
J M Ilardi, S Mochida and Z-H Sheng
SEE NEWS AND VIEWS, PAGE 104

Casein kinase-II regulates NMDA channel function in hippocampal

neurons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
D N Lieberman and I Mody

Adrenaline enhances odorant contrast by modulating signal encoding

in olfactory receptor cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
F Kawai, T Kurahashi and A Kaneko
1999 Nature America Inc. http://neurosci.nature.com

SEE NEWS AND VIEWS, PAGE 106

Microdomains for neuron-glia interaction: parallel fiber signaling to

Bergmann glial cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
J Grosche, V Matyash, T Mller, A Verkhratsky, A Reichenbach and H Kettenmann

Voltage-activated sodium channels amplify inhibition in neocortical

pyramidal neurons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
G Stuart

Retention of heroin and morphine-6-glucuronide analgesia in a new

line of mice lacking exon 1 of MOR-1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
A G P Schuller, M A King, J Zhang, E Bolan, Y-X Pan, D J Morgan, A Chang,
M E Czick, E M Unterwald, G W Pasternak and J E Pintar
Modeling the thalamic
reticular nucleus.
Page 168. SOD1 rescues cerebral endothelial dysfunction in mice overexpressing
amyloid precursor protein . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .157
C Iadecola, F Zhang, K Niwa, C Eckman, S K Turner, E Fischer, S Younkin,
D R Borchelt, K K Hsiao and G A Carlson

The construction of movement by the spinal cord . . . . . . . . . . . . . . . . . . . . . . . . 162

M C Tresch, P Saltiel and E Bizzi

Self-sustained rhythmic activity in the thalamic reticular nucleus mediated

by depolarizing GABAA receptor potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
M Bazhenov, I Timofeev, M Steriade and T J Sejnowski

Neural correlates of a decision in the dorsolateral prefrontal cortex

of the macaque . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
J-N Kim and M N Shadlen
SEE NEWS AND VIEWS, PAGE 108

Weighted combination of size and disparity: a computational model

for timing a ball catch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
S K Rushton and J P Wann
The timing of ball-catching.
Page 186.
Brain potentials indicate immediate use of prosodic cues in natural
speech processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
K Steinhauer, K Alter and A D Friederici
SEE NEWS AND VIEWS, PAGE 103

nature neuroscience volume 2 no 2 february 1999 ii

 1999 Nature America Inc. http://neurosci.nature.com
Science and terrorism in Europe
The animal rights movement, particularly in its more extreme
manifestations, has been a long-standing thorn in the side of the
biomedical research community, but the problem has taken on
1999 Nature America Inc. http://neurosci.nature.com
a new urgency in the last few weeks, with renewed death threats
being made against two prominent European neuroscientists,
Colin Blakemore of Oxford University and Wolf Singer of the
Max-Planck Institute in Frankfurt. Such threats are of course
deplorable in any context, but they are particularly disturbing
because Blakemore and Singer are among the few neuroscien-
tists who have been willing to publicly defend the use of animals
in their own research. By targeting them, the animal rights
extremists are attempting to silence public debate, at a time when
newly elected governments in both Britain and Germany appear
to be more sympathetic than their predecessors to the animal
rights movement, making it more important than ever that sci-
entists should make their voices heard.
The animal rights movement, like the anti-abortion move-
ment in the USA, comprises a broad range of opinions and tac-
tics, ranging from legitimate political activity to violent
extremism. The extreme end of the spectrum is represented pri-
marily by the Animal Liberation Front (ALF), a loosely organized
group that originated in Britain and now extends to many coun-
tries in both Europe and North America, helped in part by the
internet, which allows easy sharing of information about poten-
tial targets and tactics. The ALF home page (http://www.animal-
liberation.net), for example, sets the tone with the statement:
The earth is not dying, it is being killed. And those that are
killing it have names and addresses.
The group operates under a variety of names for
instance the Animal Rights Militia and the Justice Depart-
ment but these are generally thought to be flags of conve-
nience only. Recently, dramatic insight into the tactics of the
ALF came from a British television documentary, in which an
infiltrator posing as a potential recruit filmed several of their
leaders with a hidden camera. Among these was Robin Webb,
press officer for the British ALF and its most prominent
spokesman; although Webb has often denied ALF involvement
in acts of violence, the film showed him giving detailed
instructions on how to construct letter bombs. In the early
1990s, the ALF conducted an extended bombing campaign
against scientists in Britain, and although nobody was killed
and most of the bombs were fairly crude, some included plas-
tic explosives and were undoubtedly life-threatening. As a
result of these attacks, Blakemore, one of the most prominent
targets, is now forced to live under heavy security.
The latest crisis in Britain arose when an animal rights activist
named Barry Horne, who is serving an 18-year prison sentence
for a series of arson attacks, began a hunger strike last October.
nature neuroscience volume 2 no 2 february 1999
editorial
The ALF announced that if Horne died, ten people would be
assassinated in retribution; among the named targets were Blake-
more and Mark Matfield, the president of the Research Defence
Society, a London-based biomedical advocacy organization. The
crisis was averted when Horne called off his strike after 68 days,
and while it was not clear that he had intended to kill himself (he
had in fact been taking some food), the effect was to trigger a
wave of protests and acts of vandalism in Britain and elsewhere.
The German animal rights movement has never been as vio-
lent as its British counterpart, but it is nevertheless active and influ-
ential, a close second to the British movement, according to
Matfield. Singer, like Blakemore, has received many threats in the
past, and has been the target of a persistent campaign of vilifica-
tion and misinformation; for instance, he says that German web
sites provide translations of his publications, in which experimental
details are distorted, for instance by omitting any mention of anes-
thesia.
Singers former student Andreas Kreiter has also been targeted
by a similar campaign. Kreiter is now a faculty member at the Uni-
versity of Bremen, where he studies visual cortical physiology in
awake behaving monkeys. Soon after he had accepted the posi-
tion in Bremen, a large advertisement appeared on a billboard in
the city center, announcing that Kreiter was a monkey-torturer
and giving his home address and telephone number. This trig-
gered an intense campaign of protest, in which Kreiter received
hundreds of letters from all over Germany. Some were threatening,
and on at least one occasion there was an attempted attack on his
laboratory. He is now forced to conceal his new address, and since
the summer of 1997 he has been under daily police protection.
Kreiter attributes this campaign in part to the Deutscher Tier-
schutzbund (German Association for the Protection of Animals),
a large and influential national organization whose charismatic
leader, Wolfgang Apel, happens to live in Bremen.
Although Kreiter says that his fellow neurobiologists have been
sympathetic, he has been dismayed by the lukewarm support he
has received from the university authorities, who he feels have
failed to take a clear stand against the protesters. Many of his col-
leagues in other departments, he says, are ignorant of animal
research issues, and some are openly hostile; over a hundred fac-
ulty members at the university recently signed a memorandum
demanding that he cease his experiments. The local media have
also been unsympathetic, and on one occasion quoted without
comment a comparison to Josef Mengele, the infamous doctor
of Auschwitz.
It is tempting to see these campaigns as the work of extrem-
ists who are isolated from the rest of society, but this would be
to underestimate the extent of public sympathy for the aims,
if not the tactics, of the animal rights movement. Blakemore
99
 1999 Nature America Inc. http://neurosci.nature.com
editorial
believes that the latest crisis in Britain was due at least in part
to the actions of the Labour party, who during the 1997 elec-
tion campaign promised to adopt a more stringent policy
toward animal research. To some extent they have already done
so, and new rules requiring local as well as central approval for
animal protocols will take effect from April this year; never-
theless, the changes did not go far enough to satisfy some sup-
porters, and the result was a backlash from extremists whose
hopes had been falsely raised. Although the governments posi-
tion has been hardened to some extent by the need to resist
terrorism, the animal rights issue remains on the political agen-
da. Among those voicing their concern is Michael Dexter, the
director of the Wellcome Trust, who comments that the gov-
ernment is still under tremendous pressure from the animal
rights movement, and that scientists must stand up and be
counted. He has recently written to the Home Office minis-
ter, George Howarth, urging the government to adopt a cau-
tious approach and avoid issuing a statement that could be
1999 Nature America Inc. http://neurosci.nature.com
interpreted as a commitment to further reduce and ultimately
eliminate animal use for medical research.
In Germany, the political situation is more threatening; in
late January a bill will be introduced into the parliament that
would amend the constitution to give protection to animals.
Although the final text has not yet been agreed, the majority
of political parties support some form of amendment. The
main opposition comes from the conservative Christian
Democrats, but following their recent electoral defeat, it is
unclear whether there will be enough votes in parliament to
block the measure. According to Jan Erik Bohling of the
Gesellschaft Gesundheit und Forschung (Society for Health
and Research), a Frankfurt-based lobby group, the amendment
will be debated in the next few months and could be approved
as early as the end of April. Bohling, along with many scien-
tists, believes that this will be potentially very damaging to Ger-
man biomedical research; although freedom of research is
currently protected under the constitution, the elevation of
animal rights to constitutional level will open the door to a
protracted series of court battles, and long before research is
banned outright, it will become so difficult that researchers
will have little choice but to leave the country. Many researchers
are also concerned about what they perceive as a lack of sup-
port by the government. Reinhard Grunwald, secretary gener-
al of the Deutsche Forschungsgemeinschaft (DFG; the main
government funding agency), was recently quoted in Nature
as saying, A constitutional change would not be necessary if
researchers limited their animal experiments to those that are
really important. This has caused consternation in the research
community because it could be construed as an accusation of
illegal behavior, given that only essential experiments are per-
mitted under German law. Grunwald, however, now claims
that he was misquoted, and says that the DFG plans to make a
public statement by the end of January opposing any amend-
ment to the constitution.
The underlying problem, of course, is that scientists have been
relatively ineffectual in explaining and defending their work to
the public. Opposition to animal experimentation draws sup-
port from a widespread public distrust of science and technolo-
gy, and animal rights activists tend to be allied to a broad range of
causes, including opposition to animal farming and meat con-
sumption, genetic engineering of crops, hunting and environ-
mental destruction. Many of these causes enjoy wide support
(including that of many scientists), and as a result science finds
itself on the defensive against a broad sector of public opinion.
100
The issues have to be brought into sharper focus if the biomed-
ical community is to defend itself effectively.
But the unpalatable truth is that terrorism works the cam-
paigns of the extremists are sufficiently threatening that few sci-
entists are willing to speak out. By targeting those who do, the
ALF seeks to ensure that the arguments in favor or animal research
are not heard. If more individual scientists were to make their case
in public, the threat to any one person would be diluted; Matfield
points out that the ALF, despite its violent posturing, has never
actually killed anyone, and were they to do so, they would face an
enormous public backlash. Yet, it is understandable that few peo-
ple want to take the risk of being the first to speak out.
The degree of public ignorance, and its reflection in govern-
ment policy, testify to this reluctance. Singer, for instance, says
he is appalled by the lack of information about biomedical
research that he encounters routinely even among highly edu-
cated people. The potential consequences of this dismal situa-
tion are illustrated by the experience of Switzerland, which last
year narrowly escaped a yes vote in a national referendum that
would have banned all research on transgenic animals, and by
the current situation in Germany, where the future of animal
experiments is currently very much in the balance.
Few neuroscientists would deny that the use of animals in
research raises legitimate ethical questions; arguably, these are more
acute in the case of neuroscience than for any other discipline. The
brain cannot be studied simply by looking at its parts (for instance
in tissue culture), and certain questions about the human brain
can only be answered studying the brains of closely related species.
But these are also the species which seem likely to have the great-
est capacity for pain, fear and other forms of suffering, and for
which the ethical dilemma is therefore most pressing.
How then should neuroscientists defend their work? On the
one hand, there are practical issues; animal experiments are
tightly regulated by law, and researchers are normally very care-
ful not to cause unnecessary suffering to their experimental
subjects. Standards of laboratory animal care are undoubted-
ly better now than they were in the past, and it would be fool-
ish to deny that these improvements are in part a response to
the valid concerns of animal rights advocates. Moreover,
researchers have strong financial as well as ethical motives for
minimizing the number of experimental animals used, and the
number of animals used in research has been falling steadily
for many years. Another important point, not widely appreci-
ated, is that some of the experiments that the public finds most
troubling, those involving awake behaving monkeys, in fact
require that the animals be in good health psychologically as
well as physically; the monkeys must be calm and willing par-
ticipants if the experiment is to succeed.
But if biomedical research is to be secure in the long term,
the public must approve not only its methods but also its goals.
In cases where there is a strong focus on a specific disease, the
argument is relatively easy to make. Much current research in
neuroscience, however, is concerned with more basic questions,
whose relevance to human welfare remains to be established,
and whose likely benefits lie in the distant future. The argu-
ments for this type of research are more difficult to convey in
a short soundbite, but Singer puts it well: since we left the Gar-
den of Eden, he says, humans have been responsible for their
own fate, and with that responsibility comes the obligation to
understand, a moral obligation to be curious about ourselves
and our world. That is the message that neuroscientists must
get across, and they must not allow its communication to be
prevented by the fear of terrorism.
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com
Overuse of impact factors suppresses
controversial ideas
TO THE EDITORThe welcome criticism
of the overwhelming weight given to the
impact factor, in the December editor-
ial (Nature Neuroscience 1, 641642) has
listed some drawbacks and misuses of
this bibliometric method that raise seri-
ous concerns. Another adverse effect not
mentioned in this editorial is also worth
considering. The intense desire to get
1999 Nature America Inc. http://neurosci.nature.com
ones papers published in journals with
the highest possible impact factor pro-
duces an avalanche of papers submitted
to these journals and a progressive dis-
affection for the less-quoted journals, a
self-amplifying process that quickly
leads to a diminished reputation and
possible financial troubles for the latter.
This has the unintended effect of con-
ferring on the editors of highly ranked
journals an enormous power of dis-
nature neuroscience volume 2 no 2 february 1999
crimination, no longer regulated by gen-
uine scientific considerations. Increas-
ing submission rates require larger
panels of would-be referees, but most
editors rely on a list of people they
know, who are established experts,
hopefully honest and able to provide
their reports within reasonable time (a
very important point from the editors
point of view). In practice, papers are
examined by a rather small number of
people, a situation that tends to favor
conformity and to result in the sup-
pression of unorthodox results or ideas.
Moreover, stringent space limitations
imply high rejection rates, and this gen-
erates questionable criteria for accep-
tance; for instance, some editorial
offices are instructed to reject a manu-
script if the two referees reports dis-
letter to the editor
agree, no matter what the scientific rea-
sons for the disagreement. In other
words, there is no longer space for
debate, controversial results or diver-
gent opinions. I have experienced this
unscientific situation both as author
and as referee. Unexpected observations
and new ways of thinking vitally need
to be submitted to the scientific com-
munity for experimental confirmation
and discussions. The present concen-
tration of manuscripts in a small num-
ber of highly ranked journals is working
against these confrontations, so essen-
tial for progress in science.
J.M. Gillis
Department of Physiology,
Catholic University of Louvain
Belgium
101
 1999 Nature America Inc. http://neurosci.nature.com

news and views

Speech boundaries, syntax

and the brain
Cyma Van Petten and Paul Bloom

Speech comprehension requires rapid decoding of grammatical relationships. Electrical scalp

recordings show that the brain responds immediately to intonational cues signifying phrase
boundaries. Thus, these cues may control initial decisions about sentence structure.

Language comprehension consists alent of a comma. They describe no

of the successful transformation of Peter verspricht Anna zu ARBEITEN less than four physical differences
physical signals into abstract and und das Bro zu putzen between German sentences analo-
meaningful information. From the gous to the example above and
1999 Nature America Inc. http://neurosci.nature.com

Peter verspricht
same set of air-pressure waves, we
might understand a new idea, deter-
mine the speakers age, emotional Fig. 1. An example of the German sentences used in the experi- Jay always jogs five miles this seems
state and educational level, and add ments of Steinhauer and colleagues, illustrating how the pauses like a short distance to him;
a new word to our vocabulary. between words differ in the two conditions.
Understanding how the human
brain accomplishes this feat requires
a careful description of the stages of this
transformation, beginning with the physi-
cal signal. In speech sounds, changes in fre-
quency and amplitude occur on two time
scales. Rapid changes occurring within
20200 ms convey information about the
individual phonemes that make up words,
whereas slower prosodic modulations of fre-
quency and amplitude typically extend over
multiple words. Computer-synthesized
speech demonstrates both the adequacy and
the limitations of phonological information
alone. Although intelligible, synthetic speech
is difficult to understand because it lacks
prosody or intonation cues.
Prosody has been granted a limited role
in most theories of language comprehen-
sion. For instance, rising versus falling pitch
toward the end of a sentence distinguishes
questions from statements, and sarcastic
intonation indicates that a speakers intend-
ed meaning is the opposite of the literal one.
In these traditional examples, prosody serves
only to modify sentence meanings that have
already been computed. In this issue of
Nature Neuroscience (pages 191196), Stein-
hauer and colleagues use event-related
potential (ERP) recordings to provide per-
suasive evidence that prosody is far more
central to language comprehension. Their
data indicate that speakers use prosody to
disambiguate potentially problematic sen-
The authors are at the Department of
Psychology, University of Arizona, Tucson,
Arizona 85721, USA. email:
bloom@u.arizona.edu and
vanpettc@u.arizona.edu
ANNA zu entlasten
und das Bro zu putzen
tences, and that listeners use this informa-
tion almost immediately.
To see how this works, consider the fol-
lowing poorly written sentence: Since Jay
always jogs five miles seems like a short dis-
tance to him. In the absence of a comma
after jogs, readers find this sentence diffi-
cult to interpret. They are led down the gar-
den path of assuming that a noun phrase
(five miles) following a verb (jogs) is the
object of that verb rather than the subject of
a later verb (seems). Given this potential
ambiguity, why dont readers simply post-
pone decisions about the interpretation of
the noun phrase until they reach the end of
the sentence? This strategy avoids the gar-
den path, but at the cost of imposing a
memory burden of holding the unassigned
material in mind while continuing to read.
Working memory capacity is severely lim-
ited1, and it is needed for other aspects of
sentence comprehension, such as linking
pronouns to their antecedents2. Without
contrary evidence (such as a comma), link-
ing a noun with the earliest available verb is
an efficient processing strategy. Moreover,
the consequences of being led down the gar-
den path by a less-than-thoughtful writer
are fairly trivialat worst, one might have
to reread the sentence.
In contrast, listeners lack the luxury of
instant replay and may be severely misled
by an incorrect initial parse. So why arent
we chronically confused in our daily con-
versations? Steinhauer and colleagues
began by documenting the auditory cues a
speaker produces when saying a potential-
ly ambiguous sentencethe verbal equiv-
those with the more typical verb-
then-object structure (as in Since
Fig. 1). These differences in word
duration, pause duration, pitch
contour and loudness serve to
group the words of each sentence into dis-
tinct intonational phrases. The critical
question is whether this prosodic infor-
mation can override the usual syntactic
preference for assigning a noun to the
nearest preceding verb.
Resolution of this question requires an
immediate measure of comprehension.
Querying listeners at the end of a sentence
can reveal only their final analyses, which
may not indicate whether they avoided the
garden path, or have been there and back.
The ERP recorded from the scalp provides
a continuous measure of brain electrical
activity with millisecond temporal resolu-
tion, and so it has been widely used to
study language, for which timing is criti-
cal. Two components of the ERP elicited
by words reflect comprehension difficul-
ties. A negative voltage deflection peaking
at about 400 ms after stimulus onset
(N400) is small when a word is easily inte-
grated with prior semantic context, but
shows graded increases in amplitude as the
prior context becomes less and less useful.
A word like sugar will elicit no N400 at
the end of a sentence like I take my coffee
with cream and..., a medium-sized N400
at the end of a sentence like At the store,
I bought a pound of... and a large N400
in I finally asked my boss for a ...3. The
N400 offers a rapidly changing index of the
ease or difficulty of semantic processing,
beginning well before a spoken word is
even completely pronounced4. A second
measure of comprehension difficulty is a
positive voltage deflection following the
N400, termed the P600; large P600s are

nature neuroscience volume 2 no 2 february 1999 103


news and views
triggered by words that violate rules of
agreement, word order or other syntactic
principles. The P600 may serve as a gener-
al index of surprise at encountering a
grammatical problem, or the attempt to
revise ones initial parse of a sentence5.
Steinhauer and colleagues found that
when sentences with typical and atypical
syntactic structure were presented with nor-
mal intonation, there was no sign of an
enlarged N400 or P600 during the atypical
sentences. In other words, when subjects
heard a sentence like Since Jay always jogs
five miles seems like a short distance to him,
there was no sign of trouble at the word
seems. A critical third condition pitted the
syntactic preference against the prosodic
information. The authors spliced the first
1999 Nature America Inc. http://neurosci.nature.com
part of the sentence (Since Jay always jogs
five miles), said with the intonation sug-
gesting a verbal comma between jogs and
five, to the second part of the more com-
mon structure (this seems like a short dis-
tance to him). This created a hybrid
condition: initial prosodic cues suggested
the atypical sentence structure, while the
words specified the more typical verb-then-
object construction. In this condition, a
large N400 and P600 were elicited by the
first word that indicated the sentences true
structure, although this structure would be
entirely expected in written language. These
results show that listeners take prosodic cues
seriously and adjust their syntactic strate-
gies accordingly.
Given the importance of intonational
phrases in sentence processing, Steinhauer
and colleagues went on to ask whether
recognition of prosodic boundaries was
reflected in scalp-recorded brain activity.
The answer was yes: the authors report a dis-
tinct ERP component beginning shortly
after the conclusion of each intonational
phrase. This is an exciting finding because
discovering boundaries is a fundamental
process in multiple domains of perception
and cognition6. The contours defining a
visual object must be perceived before the
object can be recognized, and neurons early
in the visual cortical pathways respond to
even occluded contours7. Early auditory
processes partition the acoustic input into
distinct streams of sound arising from dif-
ferent locations; otherwise we might amal-
gamate the sound of a dog barking next
door with water running in the kitchen and
a radio playing in the living room into one
incomprehensible acoustic event8. The short
latency of the closure positive shift (CPS),
as the authors call this new ERP component,
suggests that detection of intonational
boundaries is an equally primary process in
speech perception. The specificity of the
104
1999 Nature America Inc. http://neurosci.nature.com
CPS to full intonational phrases remains to
be discovered. Prosodic contours are also
used to mark word boundaries within a
phrase the distinction between green
house and greenhouse. Segmentation of
the speech stream into words is a basic prob-
lem that must be solved by infants9,10, so
developmental studies of the CPS may shed
light on the perceptual strategies and brain
processes underlying language acquisition.
The precursor to a complete under-
standing of human language is a defini-
tion of the right units of analysis. Linguists
and psychologists have not always agreed
on these fundamental issues: whether
phonemic features, phonemes or syllables
are the smallest speech units11, or whether
words, phrases or sentences are the prim-
itive conceptual unit12,13. Physiological
methods for monitoring brain activity
bring a new and rich dimension of data
to bear on these old questions. Other
recent studies report brain potentials trig-
gered at syntactic boundaries (clauses) in
both spoken and written sentences14,15.
The discovery of a brain event marking
intonational units adds another tool that
can be used to understand the complex
interplay of phonological, prosodic, syn-
tactic and semantic information in deriv-
ing meaning.
Coupling calcium to SNARE-
mediated synaptic vesicle
fusion
Sabine Hilfiker, Paul Greengard and George J. Augustine
Calcium triggers vesicle fusion, but the molecular mechanism
is largely unknown. Now Ilardi and colleagues identify
Snapin, a protein that may provide the missing link.
Neurotransmitter release involves a series of
interactions between the membranes of
synaptic vesicles and the presynaptic termi-
nal, culminating in the calcium-dependent
fusion of the two membranes. However, the
molecular mechanism by which calcium
triggers fusion remains largely unknown.
Sabine Hilfiker and Paul Greengard are at
Rockefeller University, 1230 York Avenue, New
York, New York 10021, USA. George Augustine
is at the Department of Neurobiology, Duke
Medical Center, Box 3209, Durham, North
Carolina 27710, USA.
email: georgea@neuro.duke.edu
nature neuroscience volume 2 no 2 february 1999
1. Miller, G. A. Psychol. Rev. 63, 8197 (1956).
2. Carpenter, P. A., Miyake, A. & Just, M. A. in
Handbook of Psycholinguistics (ed. Gernsbacher,
M. A.) 10751122 (Academic, New York, 1994).
3. Kutas, M. & Hillyard, S. A. Nature 307, 161163
(1984).
4. Van Petten, C., Coulson, S., Rubin, S., Plante, E.
& Parks, M. J. Exp. Psychol. Learn. Mem. Cogn.
(in press).
5. Coulson, S., King, J. W. & Kutas, M. Lang. Cogn.
Processes 13, 2158 (1998).
6. Bloom, P. Curr. Directions Psychol. Sci. 5, 9094
(1996).
7. Von der Heydt, R., Peterhans, E. &
Baumgartner, G. Science 224, 12601262
(1984).
8. Bregman, A. S. Auditory Scene Analysis: The
Perceptual Organization of Sound (MIT Press,
Cambridge, MA, 1990).
9. Jusczyk, P. W. The Discovery of Spoken Language
(MIT Press, Cambridge, MA, 1997).
10. Bloom, P. in Handbook of Psycholinguistics (ed.
Gernsbacher, M. A.) 741780 (Academic, New
York, 1994).
11. Remez, R. R. in Handbook of Psycholinguistics
(ed. Gernsbacher, M. A.) 145172 (Academic,
New York, 1994).
12. Van Petten, C. Psychophysiology 32, 511525
(1995).
13. Bach, E. Informal Lectures on Formal Semantics
(SUNY, New York, 1989).
14. Muller, H. M., King, J. W. & Kutas, M. Cogn.
Brain Res. 5, 193203 (1997).
15. Kutas, M. Psychophysiology 34, 383398 (1997).
Among the many molecules implicated in
neurotransmitter release, most attention has
focused on two types of proteins, the
SNAREs and synaptotagmin. In this issue
of Nature Neuroscience, Ilardi and colleagues
(pages 119124) identify and characterize a
novel presynaptic protein, called Snapin,
which promotes the interaction of SNAREs
and synaptotagmin and may be important
for synaptic vesicle fusion.
Synaptotagmin, an integral calcium-
binding protein of synaptic vesicle mem-
branes, is thought to mediate the action of
calcium in neurotransmitter release. Evi-
dence for this notion includes the observa-
 1999 Nature America Inc. http://neurosci.nature.com

news and views

tions that transmitter release evoked by
presynaptic action potentials is greatly
decreased by presynaptic injection of synap-
totagmin fragments1 or by mutation or
2,3
deletion of the synaptotagmin I gene . The
synaptic vesicle-associated protein synap-
tobrevin (also called VAMP), a v-SNARE,
interacts with two membrane-associated
proteins (t-SNAREs), SNAP-25 and syn-
taxin, to form a stable SNARE complex.
These three SNARE proteins are essential
for neurotransmitter release46, and the
formation of a complex among these pro-
teins is the minimal molecular requirement
for membrane fusion in vitro7. Although it
is clear that both SNAREs and synaptotag-
min have important roles in neurotrans-
mitter release, the mechanism by which
1999 Nature America Inc. http://neurosci.nature.com
of docked vesicles and inhibit neurotrans-
mitter release. Snapin might also prime vesi-
cles for fusion, for example by increasing the
binding of synaptotagmin to the SNARE
complex, before a Snapin-independent reac-
tion causes vesicle fusion. Finally, vesicle
fusion per se might be mediated by a
Snapin-dependent mechanism. Earlier in-
vitro work examining the interactions of
SNARE proteins and synaptotagmin sug-
gested at least three possible models for cal-
cium-triggered vesicle fusion (Fig. 1a-c), and
the identification of Snapin now provides
an additional model for connecting SNAREs
and synaptotagmin during the fusion reac-
tion (Fig. 1d). Thus, some of the possibili-
ties for this reaction now include the
following:
2) There have been several reports that
calcium binding to synaptotagmin can dra-
matically promote association of this pro-
tein with both syntaxin8,9 and SNAP-2510
(H. Tokumaru, unpublished observations).
Based on these reports, it is widely believed
that this calcium-triggered interaction can
promote the fusion of synaptic vesicles (Fig.
1b). However, this model also has limita-
tions. First, it is not at all clear how binding
of SNAREs to synaptotagmin could account
for fusion. Second, the rate of binding
between these proteins may be too slow (H.
Tokumaru, unpublished observations) to
account for the finding that neurotransmit-
ter release occurs approximately 100 s after
calcium binding11,12.
3) There have been other reports that

synaptotagmin confers calcium sensitivity 1) The SNARE proteins and synapto- calcium binding decreases, rather than
onto the SNARE-based fusion machinery tagmin could act sequentially without bind- increases, the affinity of synaptotagmin for
is unclear. The discovery of Snapin by Ilar- ing to each other. For example, the the SNARE complex13. Such a calcium-
di and colleagues may provide the long- rod-shaped structure formed by the paired induced displacement of synaptotagmin
awaited answer to this question. parallel helices of synaptobrevin, SNAP-25 could precede the formation of the rod-
Snapin is a neuron-specific protein that and syntaxin6 could bring the vesicle and shaped complex among the SNAREs (Fig.
is found predominantly on synaptic vesi- plasma membranes in close apposition, so 1c). This helix zippering mechanism has
cles. Snapin binds to the SNARE complex, that calcium binding to synaptotagmin been postulated to underlie evoked trans-
via a direct interaction with SNAP-25, and could allow synaptotagmin to bind to mem- mitter release6 and can account for the
thereby seems to promote the interaction brane lipids and thereby cause fusion (Fig. occurrence of spontaneous release in the
of synaptotagmin with the SNAREs. The 1a). A limitation of this model is that it does absence of synaptotagmin. However, at pre-
carboxy-terminal domain of Snapin is pre- not account for observations that sponta- sent it is not clear that the SNAREs can form
dicted to form a coiled-coil structure. Pep- neous transmitter release, and thus mem- complexes rapidly enough to account for
tide fragments corresponding to this brane fusion, can occur even after the speed of synaptic vesicle fusion7,14,15.
domain reversibly inhibit this proteinpro- synaptotagmin has been genetically delet- 4) The new findings suggest that the
tein interaction, and partially block synap- ed2,3. It also neglects a variety of data in vitro coupling between synaptotagmin and the
tic transmission when introduced into indicating a direct interaction between SNARE complex may be controlled by
superior cervical ganglion neurons. Taken synaptotagmin and the SNARE complex, Snapin. Ilardi and colleagues show that this
together, these results demonstrate the phys- which is a central feature of the three other protein can promote the attachment of
iological relevance of the interaction of models we discuss. synaptotagmin to SNARE complexes in vitro
Snapin with SNAP-25 and suggest
the exciting possibility that Snapin
serves as a link between the SNAREs a b c d
and synaptotagmin during calcium-
dependent neurotransmitter release.
The precise role of Snapin in neu-
rotransmitter release awaits further
experimental efforts. It is possible
that Snapin acts as a modulator,
rather than as an essential compo-
nent, of exocytosis. This is consistent
with the authors observation that
injected Snapin peptides only par-
tially abolish neurotransmitter
release. Alternatively, Snapin might
be an essential component in the
molecular mechanism of exocytosis
by participating in vesicle docking,
priming or fusion reactions. For Fig. 1. Four possible models for calcium-triggered vesicle fusion. (a) The SNARE proteins initially bring the
two membranes into close apposition, and subsequent binding of calcium to synaptotagmin induces mem-
example, Snapin might promote the
brane fusion without a direct interaction between synaptotagmin and SNAREs. (b) The association of cal-
docking of synaptic vesicles, by link- cium-bound synaptotagmin with the SNARE complex can trigger membrane fusion. (c) A calcium-induced
ing vesicles to the plasma membrane displacement of synaptotagmin from the SNARE proteins allows formation of a SNARE complex that spans
through an interaction with SNAP- the two membranes, leading to fusion. (d) Snapin binds to the SNARE complex, by binding to SNAP-25.
25. Interfering with this interaction The subsequent formation of a calcium/synaptotagmin/SNARE/Snapin complex results in vesicle fusion.
in vivo would decrease the number Synaptotagmin may bind to the SNAREs either before or after binding calcium.

nature neuroscience volume 2 no 2 february 1999 105

 1999 Nature America Inc. http://neurosci.nature.com
news and views
and that perturbing these interactions in
vivo inhibits transmitter release. Thus, it is
possible that synaptotagmin triggers fusion
by binding to Snapin-associated SNARE
complexes, either before or after synapto-
tagmin binds calcium (Fig. 1d). This new
model is consistent with what we now know
about Snapin. However, it is still possible
that Snapin has a role in one of the many
other vesicle trafficking steps involved in
neurotransmitter release.
To clarify the precise role of Snapin in
neurotransmitter release, it would be useful
to know whether Snapin acts before or after
calcium entry into the nerve terminal. There
are several relevant biochemical questions:
Does calcium influence the ability of Snapin
to bind to SNAP-25? Is the ability of Snapin
1999 Nature America Inc. http://neurosci.nature.com
to induce binding of synaptotagmin to the
SNARE complex calcium dependent? Does
Snapin influence the binding of calcium to
synaptotagmin? Does Snapin bind directly
to synaptotagmin? Additional physiological
The smell of adrenaline
Stuart Firestein and Anna Menini
Adrenaline modulates signaling by the olfactory epithelium.
A new study suggests that this effect may result from
channel phosphorylation in olfactory receptor neurons.
The olfactory system must make sense of
thousands of chemical ligands or odorants.
This is no mean trick because the ligand
repertoire is not only enormous but also
diverse (including aromatic and aliphatic
structures with many different functional
groups) and complex (including mixtures
and elaborately structured molecules).
Increasing the difficulty, there are no appar-
ent primary odors (as for colors) or funda-
mentals (as for acoustics) or even a clear
spatial organization (as for visual or
somatosensory maps). This information
must be encoded by the primary sensory
neurons in the olfactory epithelium; each
neuron is thought to express a single recep-
tor gene from a repertoire of about 1000
receptors (in mammals), each of which is
presumably tuned to slightly different subset
of odors1,2.
Stuart Firestein is at the Department of
Biological Sciences, Columbia University,
New York, New York, USA.
email: sjf24@columbia.edu
Anna Menini is at the Instituto di Cibernetica e
Biofisica, CNR, Genova, Italy.
menini@icb.ge.cnr.it
106
and structural analyses of the consequences
of perturbing Snapin function also will help
to distinguish among the possible functions
of this protein. Of course, the same kinetic
issues raised for the other models would be
relevant for any Snapin-based fusion
hypothesis; future work will need to define
the speed of binding of synaptotagmin to
Snapin-associated SNARE complexes.
In summary, the paper by Ilardi and col-
leagues introduces Snapin, an intriguing
protein that seems to be involved in the
interaction between the SNARE complex
and synaptotagmin. This protein promises
to be an important component of the exci-
tation/secretion coupling mechanism and
may therefore have a central role in the reg-
ulation of neurotransmitter release.
1. Bommert, K. et al. Nature 363, 163165 (1993).
2 . Broadie, K., Bellen, H. J., DiAntonio, A.,
Littleton, J. T. & Schwarz, T. L. Proc. Natl. Acad.
Sci. USA 91, 1072710731 (1994).
Beyond its purely sensory role, the olfac-
tory system is closely tied to behavioral
responses, especially those with an emo-
tional component. For example, territorial-
ity, sexuality, feeding and aggression are all
strongly influenced by olfactory stimuli, and
it has been suggested that behavioral con-
text might modulate activity in the epitheli-
um via descending central inputs (as occurs
for instance in the cochlea). Whether such
inputs actually exist, however, remains
unclear. The olfactory epithelium is exten-
sively innervated by centrifugal adrenergic
fibers3, and -adrenergic receptors are
found in the olfactory epithelium, but
adrenaline was thought to affect the epithe-
lium indirectlymost likely affecting either
the gland cells that produce and secrete the
mucous that covers the epithelium or its
dense vasculature4. However, striking new
results from Kawai and colleagues (pages
133138) in this issue show that olfactory
neurons themselves are modulated by
adrenergic agonists, thus suggesting the exis-
tence of a new regulatory pathway and help-
ing to explain some previously puzzling
findings.
Transduction in the olfactory receptor
nature neuroscience volume 2 no 2 february 1999
3. Geppert, M. et al. Cell 79, 717727 (1994).
4. Sdhof, T. C. Nature 375, 645653 (1995).
5. Augustine, G. J., Burns, M. E., DeBello, W. M.,
Pettit, D. L. & Schweizer, F. E. Annu. Rev.
Pharmacol. Toxicol. 36, 659701 (1996).
6. Weis, W. I. & Scheller, R. H. Nature 395,
328329 (1998).
7. Weber, T. et al. Cell 92, 759772 (1998).
8. Li, C. et al. Nature 375, 594599 (1995).
9. Kee, Y. & Scheller, R. H. J. Neurosci. 16,
19751981 (1996).
10. Schiavo, G., Stenbeck, G., Rothman, J. E. &
Sllner, T. H. Proc. Natl. Acad. Sci. USA 94,
9971001 (1997).
11. Llins, R., Steinberg, I. Z. & Walton, K. Biophys.
J. 33, 323352 (1981).
12. Augustine, G. J., Adler, E. M. & Charlton, M. P.
Ann. NY Acad. Sci. 635, 365381 (1991).
13. Mehta, P. P., Battenberg, E. & Wilson, M. C.
Proc. Natl. Acad. Sci. USA 93, 1047110476
(1996).
14. Calakos, N., Bennett, M. K., Peterson, K. E. &
Scheller, R. H. Science 263,11461149 (1994).
15. Ungermann, C., Sato, K. & Wickner, W. Nature
396, 543548 (1998).
neuron begins in the cilia (Fig. 1). Odorant
binding to receptors in the cilia activates a
G protein, causing adenylyl cyclase to pro-
duce cAMP, which directly opens ion chan-
nels in the plasma membrane. The inward
transduction current is composed of sodi-
um and calcium ions, with the latter acti-
vating a calcium-dependent chloride
current that further depolarizes the cell. The
depolarization spreads passively from the
cilia via a single dendrite to the soma, trig-
gering action potentials that are then con-
ducted to the olfactory bulb. Until now, this
has been considered a fairly complete
description of olfactory signal transduction;
it was thought that no further processing of
the signal occurred until it reached the
olfactory bulb.
In the alternative scheme proposed by
Kawai and colleagues, adrenaline acts
downstream of the initial transduction
event, but upstream of the first synapse,
to alter the firing pattern of the receptor
neuron. The authors show that the effect
of adrenaline is two-fold: an increase in
the threshold for spike initiation and an
increase in the firing rate in response to
strong stimuli. Early experiments in the
olfactory epithelium5,6 showed that the
spike frequency is a function of odor con-
centration, and a more recent analysis has
shown that this function is quite steep
(Reisert, J. & Matthews, H. J. Physiol.
506P, 83P, 1998). Now the results of
Kawai et al. indicate that adrenaline could
produce an even steeper doseresponse
relationship, further amplifying the final
cellular response to odors. How does

Fig. 1. The major signaling pathways in
olfactory receptor neurons. In the cilia,
odorants are detected and transduced
into a generator potential through a well
defined second-messenger system.
Cyclic AMP produced in the cilia opens a
cation-permeable cyclic-nucleotide-
ODORS
gated channel, which depolarizes the
membrane from its normal resting level
(60 mV) to as much as 40 mV. An
unusual calcium-dependent chloride Cl
channel is also activated, and because the
intracellular concentration of chloride is
high, the outward flow of chloride ions
adds to the depolarization. The sensory
generator potential then spreads pas-
sively to the somatic membrane, where
it activates calcium, sodium and potas-
sium channels to produce an action
potential. A spike train of action poten-
1999 Nature America Inc. http://neurosci.nature.com
tials propagates down the axon to the
synaptic terminals in the olfactory bulb. It
is the structure of this spike train that
Kawai and colleagues showed is modu-
lated by adrenaline through the action of
the second messenger cAMP. R, odorant
receptor; G, G protein; AC, adenylyl
cyclase; A, adrenergic receptor; PKA,
protein kinase A.
adrenaline produce these effects? By com-
paring the properties of somatic voltage-
gated sodium and calcium currents with
and without adrenergic agonists, the
authors show that adrenaline increases the
amplitude of the sodium current and
decreases the amplitude of the T-type cal-
cium current. What are the molecular
mechanisms underlying these effects? The
authors find that introducing cAMP into
the olfactory receptor cell mimics the
effects of adrenaline, and that the effects
of cAMP are mediated by phosphoryla-
tion of sodium channels and T-type cal-
cium channels (Fig. 1). Note that the
cAMP generated by adrenergic activation
in the soma is unlikely to interact with the
cAMP produced by activation of the odor
receptors in the cilia, as the cilia and the
soma are almost certainly biochemically
isolated compartments.
Where does this centrifugal adrenergic
input originate? Although the data are
somewhat skimpy on this issue, most (and
perhaps all) of the fibers probably come
from the sympathetic superior cervical gan-
glia, which innervate other regions of the
face and nasopharyngeal cavities and are
associated primarily with the blood supply.
Once in the epithelium, the fibers may
branch to secretory cells, and perhaps even
neurons. Alternatively, their action on neu-
rons may be more paracrine, with no well-
defined synaptic connections. It will be
important to confirm that adrenergic
nature neuroscience volume 2 no 2 february 1999
1999 Nature America Inc. http://neurosci.nature.com
OLFACTORY
CILIUM
DENDRITE
R
-40 mV
G ATP
AC GENERATOR
cAMP
Na + POTENTIAL
++
Ca
Ca ++
SOMA
NUCLEUS ADRENALINE
A
?
G proposed that increased recep-
AC
cAMP
++
Ca
PKA ATP
Na
+
+
K
SPIKE TRAIN
AXON
ACTION
POTENTIAL
TO OLFACTORY
BULB
inputs can indeed modulate the activity of
sensory neurons in vivo. It will also be
important to determine under what cir-
cumstances this pathway is active. One pos-
sibility is that the adrenergic inputs to the
epithelium could be part of a feedback cir-
cuit, in which signals from the epithelium
are carried via the olfactory or trigeminal
nerves to interneurons that eventually acti-
vate the centrifugal adrenergic pathway.
Another possibility is that the adrenergic
neurons might be under circadian or hor-
monal control, which could elevate sensi-
tivity in the epithelium in particular
behavioral or environmental contexts.
The adrenergic pathway may not be the
only important central input to the epithe-
lium. The nervus teminalis, the so-called
zeroth cranial nerve and somewhat of an
enigma, releases gonadotropin-releasing
hormone onto the olfactory epithelium7.
There is also extensive histochemical evi-
dence for cholinergic inputs to the epitheli-
um3. These data together suggest that the
olfactory epithelium is not simply com-
posed of isolated receptors signaling from
the periphery, but is intimately tied to the
central nervous system and probably to the
endocrine system. From the work of Kawai
et al., it is clear that these external inputs
may not regulate glandular activity exclu-
sively, but may have direct consequences for
neuronal signaling.
This new finding may also throw some
light on an interesting but often ignored
news and views
report8 from 1993, which con-
cluded that animals developed
sensitivity to an odor to which
they were initially not very
responsive after being exposed
to it for several weeks. Those
changes appeared to occur at the
epithelium, rather than more
centrally, suggesting perhaps
some local feedback pathway by
which odor experience could
drive sensitivity. The results have
always been considered some-
what controversial, partly
because no clear mechanism
was shown. The investigators
tor expression could account for
the effect, but again there was
no direct evidence and no
mechanism for translating odor
exposure into modulation of
gene expression. Another possi-
bility, perhaps less likely, is that
odorants might affect the for-
mation or survival of new sen-
sory neurons, given that they are
known to be generated contin-
uously throughout life, but again there
was no evidence for how this might occur.
The work of Kawai and colleagues sug-
gests two plausible mechanisms for these
earlier results. First, the increased sensitivi-
ty might be due to the effects of adrenaline
on the firing rates of selected neurons,
working via central feedback circuits that
increase the odor sensitivity of chronically
stimulated neurons. Second, the increase in
somatic intracellular cAMP by adrenaline
provides a possible way of controlling gene
expression. Cyclic AMP acting through
CREB9 or other signal transduction path-
ways could provide a feedback pathway for
gene expression in cells that are chronically
active. Many questions remain to be
answered for either of these speculations to
be accepted, of course, but by providing a
rationale underlying the earlier observed
sensitivity increase, these findings suggest
that this issue might be worth revisiting.
With new tools available from gene target-
ing, in which all cells expressing a particular
receptor are marked by lacZ or green fluo-
rescent protein expression10, it has also
become possible to determine whether the
number of cells with a particular receptor
is increased under conditions of chronic
odor exposure. In any case, it is clear that
the olfactory system, one of evolutions old-
est inventions, still holds many secrets.
1. Buck, L. & Axel, R. Cell 65, 175187 (1991).
2. Zhao, H. et al. Science 279, 237242 (1998).
107

news and views
3. Zielinski, B. S., Getchell, M. L., Wenokur, R. L.
& Getchell, T. V. Anat. Record 225, 232245
(1989).
4. Getchell, M. L. & Getchell, T. V. J. Comp.
Physiol. A 155, 435443 (1984).
5. Getchell,T. & Shepherd, G.M. J Physiol. 282,
Weighing the evidence: how
the brain makes a decision
Jeffrey D. Schall
Kim and Shadlen investigate how neurons of the prefrontal
cortex interpret slowly accumulating signals from visual
cortex to make a perceptual decision.
1999 Nature America Inc. http://neurosci.nature.com
...in the night, imagining some fear
How easy is a bush supposd a Bear?
Shakespeare, A Midsummer Nights
Dream, Act 5
As Shakespeare observed, even perception
may require an active decision about how
to interpret the raw data. On page 176 of
this issue, Kim and Shadlen examine the
neural basis of a simple visual perceptual
decision. Their study, along with other
recent work (see refs. 13 for reviews), illus-
trates how the subjective process of decid-
ing may be soon be explained in terms of
objectively observable brain processes.
Perceptual decisions involve several steps.
First, a preliminary representation of the
sensory stimulus is converted into higher-
level explicit representations of the features
that will form the basis for the decision. Sec-
ond, one of the competing sensory repre-
sentations must be selected; in other words,
the ambiguous representation of the rele-
vant features must be translated into an
explicit representation of one of the possi-
ble alternatives. Finally, the sensory decision
must lead to an appropriate behavioral
response. A neural explanation of decision
making must describe the neural basis of
each of these steps.
Important insights into the first step
have come from the work pioneered by
William Newsome and colleagues4, who
have examined the sensory representations
on which a perceptual decision is based.
Meanwhile, other researchers have shown
how behavioral responses are generated in
the face of alternatives (reviewed in ref. 3).
Jeffrey Schall is at theVanderbilt Vision Research
Center, Department of Psychology, Wilson Hall,
Vanderbilt University,
Nashville, Tennessee 37240, USA
email: jeffrey.d.schall@vanderbilt.edu
108
1999 Nature America Inc. http://neurosci.nature.com
521-540 (1978). Science 260, 9981000 (1993).
6. Kauer, J.S. & Shepherd, G.M. Brain Res. 85, 108- 9. Silva, A.J., Kogan, J.H., Frankland, P.W. &
113 (1975). Kida, S. Annu. Rev. Neurosci. 21, 127-48
7. Wirsig-Wiechmann, C. & Jennes, L. Neurosci. (1998).
Lett. 160, 201204 (1993). 10. Mombaerts, P. Curr. Opin. Neurobiol. 6,
8. Wang, H.-W., Wysocki, C. J. & Gold, G. H. 481486 (1996).
this task can be accounted for based on the
activity of a small number (around 100) of
MT neurons5,6. They have also shown that
artificial stimulation of MT during the stim-
ulus presentation influences the monkeys
judgement in a predictable way, thereby
demonstrating directly that neurons in MT
provide the sensory representation on which
the perceptual decision is based.
How then is this sensory representation
converted into a behavioral outcome? In
other words, how and where are the signals
Kim and Shadlen have now begun to inves- from MT read? The final behavioral
tigate the intermediate stage, in which the response to the stimulus is an eye movement
sensory evidence is evaluated to arrive at a to one of two targets, so an obvious place to
decision. look is the frontal eye field and adjacent pre-
The authors took advantage of a method frontal cortex, which is known to be
developed in Newsomes laboratory, in involved in producing eye movements, and
which monkeys are trained to report the net also receives inputs from area MT. More-
direction of visual motion in an array of over, prefrontal cortex is a region in which
moving dots. They report their perception visual representations are combined with
by making an eye movement to one of two knowledge, goals and desires to generate
targets that appear on either side of the dis- action7.
play after the dots disappear.
When all the dots move in
the same direction, the task
is easy, and monkeys can
respond correctly on every
?
trial. If some of the dots
move in random directions,
the motion signal is diluted,
time
and as the fraction of ran-
domly moving dots is
increased, the task becomes
harder and the proportion of
correct responses decreases.
Eventually, when all the dots
are moving at random, there
is no net motion signal, and
so the monkeys response
reflects preference in the
absence of evidence.
This simple design allows
precise measurements of
both the stimulus and the
response. The advantage of
using motion as a cue is that Fig. 1. To earn a reward, monkeys had to decide the net direction of
the site of motion represen- motion of a stimulus composed of variable fractions of coherently
tation is well characterized; and randomly moving dots. Monkeys reported their decision by
motion cues are represented making an eye movement to either of two targets. Psychologists
seek to understand choice behavior in terms of computational or
in the extrastriate visual area
cognitive processes, symbolized by the bubble. Neurophysiologists
MT, whose neurons respond seek to understand choice behavior in terms of neural activity mon-
preferentially to moving itored in particular parts of the brain, symbolized by the raster dis-
stimuli. Newsome and col- play. Ultimately, we would like to understand the relationship
leagues have shown that between the neural activity and the behavior and associated cogni-
monkeys performance in tive processes, symbolized by the arrow.
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com
Kim and Shadlen therefore examined
activity in this area, using a modified ver-
sion of Newsomes original experimental
design. In the original experiments, the
monkey was required to report the direc-
tion of movement immediately after view-
ing the dots, but Kim and Shadlen
introduced a delay between the presentation
of the moving dots and the cue to make an
eye movement. After the moving dots dis-
appear, neurons in area MT cease to
respond, indicating that the decision must
be stored in some other area. Many studies
have shown that neurons in the prefrontal
cortex continue to fire during the delay peri-
od while the monkey remembers and plans
what he is supposed to do7. By studying
activity in the prefrontal cortex during the
1999 Nature America Inc. http://neurosci.nature.com
delay period, the authors could distinguish
the representation of the decision having
been made from the representation of the
stimulus on which it was based.
In the original studies of area MT, the
randomly moving dots were placed in the
neurons receptive field (so the neural
response was to the dots themselves). For
the present study of prefrontal neurons, the
targets for the eye movements were instead
placed in the neurons receptive field. Thus,
when a prefrontal cortex neuron fires in
response to the moving dots, it does so not
because the dots themselves are within the
receptive field (they are not), but because
the monkey has been conditioned to inter-
pret a particular pattern of moving dots as
an instruction to make an eye movement to
the receptive field of that neuron.
Kim and Shadlen show that during the
period when the moving dots are visible,
neural activity grows in prefrontal cortex to
signal one or the other choice. Important-
ly, the rate of growth and magnitude of
activity during the viewing period was pro-
portional to the quality of the evidence, that
is, to the strength of the motion signal. The
authors therefore propose that the gradual
increase in prefrontal activity reflects the
accumulation of sensory evidence. Although
this interpretation is plausible, we should
not overlook alternative explanations; for
instance, the activity may reflect the prepa-
ration of the eye movement response, or the
monkeys growing expectation that it will
earn a reward. Distinguishing these alterna-
tives will require not only further empirical
work but also operational clarification of the
concepts of preparation, expectation and so
on. What is clear from the data, however, is
that prefrontal neurons reach a state that
predicts the monkeys choice while the mov-
ing dots are visible and maintain that state
throughout the delay period when no dots
are present until the monkey is given the
nature neuroscience volume 2 no 2 february 1999
opportunity to report its choice.
In some trials, there was no net motion,
so the monkeys had to make choices in the
absence of any compelling evidence for
either alternative. In other trials, monkeys
made errors, that is, a choice that contra-
dicted the evidence. In both cases, prefrontal
neural activity during the delay period cor-
responded to the choice and not to the evi-
dence. This indicates that the prefrontal
cortex neurons are signaling something
more than just the sensory evidence during
the delay period. Based on their location and
properties, it is likely that most of the neu-
rons Kim and Shadlen recorded are not
directly involved in producing the eye move-
ment. Thus, Kim and Shadlen reason, the
neurons are encoding the decision in an
abstract sense. Similar observations have
been made in posterior parietal cortex8 and
the superior colliculus (G.D. Horwitz &
W.T. Newsome, Soc. Neurosci. Abstr. 24,
1498, 1998).
Although prefrontal cortex has been
studied before in conditions requiring an
arbitrary response based on a visual dis-
crimination7, one major advance in Kim
and Shadlens study is the use of a stimulus
with discriminability that could be manip-
ulated in a psychophysical protocol. Anoth-
er important aspect was the use of the
random dot motion stimuli, for which the
sensory representation was very well char-
acterized. This provided for what may be
the most important and innovative aspect
of the study, a quantitative model of the
decision process.
Presumably the prefrontal activity that
the authors observe must be driven by
responses in MT during the motion view-
ing period. The authors did not record from
MT in these experiments, but based on ear-
lier modeling of MT responses6, they use
signal-detection theory to suggest a model
for how the prefrontal cortex reads the evi-
dence from MT. In simple terms, their
model supposes that a neuron in prefrontal
cortex monitors the activity of two popula-
tions of MT neurons, each signaling motion
in opposite directions, say left and right.
However, the motion signals in MT are ran-
domly variable across trials. To decide which
direction the dots are moving, they propose
that the input to the prefrontal cortex neu-
ron represents a random sample of the
activity of each population. Suppose the net
motion is leftward. If the MT activity sig-
naling left exceeds the activity signaling
right, then the monkey chooses left. How-
ever, if by chance the activity signaling left
happens to be less than the activity signal-
ing right, then the monkey will make an
error. When a coherent motion stimulus
news and views
evokes a strong leftward signal, then the
total activity from a random sample of left-
signaling neurons is greater than that of a
sample of right-signaling neurons, so the
monkey makes few errors. As the stimulus
becomes less coherent, it evokes a weaker
leftward motion signal. As a result, there is
an increased probability that the activity sig-
naling left will be less than the activity sig-
naling right, and so the monkey makes
more errors. When Kim and Shadlen incor-
porate values for the magnitude and vari-
ability of the signals previously observed in
MT, they can account for the performance
of the monkeys and the responses of the
prefrontal neurons.
Through research like that of Kim and
Shadlen, neuroscience engages significant
philosophical and ethical issues9. For exam-
ple, if the neural events leading to a choice
become publicly observable, then it should
be possible to predict the choices made by
an agent if the correct brain processes are
monitored. This is not philosophical fiction.
I have had the profound experience of being
able to predict with high reliability the
choice a monkey would make during a
study of binocular rivalry10, and Shadlen
reports the same experience during this
study (personal communication). It seems
reasonable to assume that the same types of
mechanisms may underlie other types of
decisions, in our own brains as well as those
of monkeys. Ultimately, philosophical argu-
ments have been developed that reconcile a
mechanistic explanation of decision mak-
ing, one of the most sacrosanct of mental
acts, with our own sense of free will, respon-
sibility and dignity11.
1. Schall, J. D. & Bichot, N. P. Curr. Opin.
Neurobiol. 8, 211217 (1998).
2. Leon, M. I. & Shadlen, M. N. Neuron 21,
669672 (1998).
3. Schall, J. D. & Thompson, K. G. Annu. Rev.
Neurosci. 22, 241259 (1999).
4. Parker, A. J. & Newsome, W. T. Annu. Rev.
Neurosci. 21, 227277 (1998).
5. Britten, K. H., Newsome, W. T., Shadlen, M. N.,
Celebrini, S. & Movshon, J. A. Vis. Neurosci. 13,
87100 (1996).
6. Shadlen, M. N., Britten, K. H., Newsome, W. T.
& Movshon, J. A. J. Neurosci. 16, 14861510
(1996).
7. Schall, J. D. in Extrastriate Cortex of Primates,
Vol. 12 Cerebral Cortex (eds. Rockland, K,
Peters, A. & Kaas, J.) 527638 (Plenum, New
York, 1997).
8. Shadlen, M. N. & Newsome, W. T. Proc. Natl.
Acad. Sci. USA 93, 628633 (1996).
9. Editorial. Nat. Neurosci. 1, 535 (1998).
10. Logothetis, N. K. & Schall J. D. Science 245,
761763 (1989).
11. Dennett, D. C. Elbow Room: The Varieties of Free
Will Worth Wanting (MIT Press, Cambridge,
MA, 1984).
109
 1999 Nature America Inc. http://neurosci.nature.com
Sir Alan Lloyd Hodgkin 1914 1998
by Denis Baylor and King-Wai Yau
Sir Alan Hodgkin, one of the great physi-
ologists of the twentieth century, died at
Cambridge, England on December 20,
1998. He was 84.
He was born in Banbury, England and
educated at Greshams School, Holt, and
Trinity College, Cambridge. Early in his
career, his scientific genius manifested itself
in a now-classic piece of work demon-
strating that conduction of the nerve
impulse requires the spread of electrical
1999 Nature America Inc. http://neurosci.nature.com
current between active and inactive regions
of the nerve fiber. During the Second
World War, he made key contributions to
the development of radar for night fight-
ers, and although a biologist, he became
Principal Scientific Officer for radar devel-
opment by 1944. This work, as well as his
earlier life and subsequent career, is detailed
in his book Chance and Design, Reminis-
cences of Science in Peace and War.
In 19371938, he spent a year in the
United States at the Rockefeller Institute
for Medical Research and became
acquainted with K. S. Cole, who intro-
duced him to the voltage-clamp method.
In the late 1940s, Hodgkin, Andrew Huxley
and Bernard Katz used this method in a
brilliant series of experiments that eluci-
dated the ionic mechanisms underlying the
nerve impulse. These experiments, along
with an elegant and original quantitative
description of the voltage-dependent mem-
brane conductances, remain a landmark in
our understanding of how neurons work.
For this work, Hodgkin and Huxley were
awarded the Nobel Prize for Medicine or
Physiology in 1963 (shared with John
Eccles). The ion channels whose existence
Hodgkin and Huxley inferred have now
been purified, cloned and sequenced and
form a standard part of biological doctrine
taught at the high school level. Hodgkins
and Huxleys picture of how ion channels
Denis Baylor is at the Department of
Neurobiology, Stanford University School of
Medicine, 299 Campus Drive, Room D200,
Stanford, California 94305-5125, USA.
email: dbaylor@leland.stanford.edu
King-Wai Yau is at the Department of
Neuroscience and the Howard Hughes Medical
Institute, Johns Hopkins University School of
Medicine, 725 N. Wolfe Street, Baltimore,
Maryland 21205, USA.
email: kwyau@welchlink.welch.jhu.edu
nature neuroscience volume 2 no 2 february 1999
Godfrey Argent
behave has proved to be remarkably pre-
scient and robust.
Subsequently, Hodgkin turned his
attention to the sodium pump and
showed that its function depends on the
consumption of ATP. He discovered that
potassium ions move through potassium
channels in single file and characterized
the membrane conductances that under-
lie excitability in muscle fibers. He also
made the first direct measurements of the
activity of the sodiumcalcium exchanger,
which is an important pathway for calci-
um extrusion in cells.
In the early 1960s, Hodgkin collaborat-
ed with Michaelangelo Fuortes in a study
of the light response of photoreceptor cells
in the eye of the horseshoe crab Limulus.
From a kinetic analysis of the light
response, they made the then-heretical sug-
gestion that light activated ionic conduc-
tances in the membrane via a series of
chemical reactions. These types of bio-
chemical cascades continue to be studied
to the present day. The collaboration with
Fuortes sparked his interest in vision,
which he studied for the rest of his career.
Between 1970 and the late 1980s, when he
retired, his laboratory pioneered biophys-
ical investigations of signal transduction in
retinal rods and cones, again setting new
standards for experimentation and analysis.
For many scientists, particularly those
who had the privilege of knowing and
working with him, Hodgkin represented
the epitome of scientific excellence. A
renowned physiologist once remarked, I
obituary
have several heroes, but Alan is the one I
always try to imitate. His unparalleled
analytic powers grew out of formidable
mathematical skills, largely self-taught, as
well as a remarkable intuition. Whereas
some think of him as a theorist because
of the quantitative flavor of many of his
papers, he was in reality a consummate
experimentalist. He relished good mea-
surements and enjoyed complete immer-
sion in them, down to the smallest details.
When experimental results were in con-
flict with a theory, he immediately reject-
ed the theory and started over, no matter
how elegant the theory might have
seemed. In spite of his hunger to under-
stand nature, he did not try to force an
answer when there wasnt one.
For Hodgkin, science was an intensely
personal experience to be enjoyed in all its
aspects. He therefore preferred to work on
a single project at a time and to have a very
small research group. His insistence on tak-
ing on a fair share of all aspects of the work
was itself an inspiration. Discussions of the
latest results were serious and focused, and
although he was patient with coworkers
and receptive to their ideas, he was impa-
tient with error and sloppy thought. He
had an unerring sense of what would stand
the test of time and what would fall by the
wayside. His negative impression, indicat-
ed by a characteristic wrinkling of the nose,
was a bad omen indeed.
Hodgkin received many other honors,
including Knight of the British Empire,
the Order of Merit and membership in
the Royal Society as well as many foreign
academies. While serving as President of
the Royal Society and Master of Trinity
College, Cambridge, he managed to keep
his research going at full throttle. In spite
of his many accomplishments and hon-
ors, he remained a modest person with
simple tastes. He had a delightful dry
sense of humor.
His professional success and enjoyment
of life owed much to a wonderfully sup-
portive family. His devoted wife Marion
(Marni), the daughter of Nobel laureate
Peyton Rous, had prior knowledge of the
demands of the scientific way and backed
Hodgkin in all things. He is survived by
Marni and four accomplished children.
Scientists of Hodgkins talent and dis-
tinction are rare indeed.
111
 1999 Nature America Inc. http://neurosci.nature.com
Neurotrophin-3 is
required for proper
cerebellar development
Brian Bates1, Maribel Rios1,3, Andreas Trumpp2,3,
Cindy Chen1, Guoping Fan1, J. Michael Bishop2
and Rudolf Jaenisch1
1 Whitehead Institute for Biomedical Research, 9 Cambridge Center,
Cambridge, Massachusetts 02142, USA
2 G.W. Hooper Foundation, University of California, San Francisco, California
94143, USA
3 These authors contributed equally to this work
Neurotrophin-3 (NT-3) is a member of the neurotrophin fam-
1999 Nature America Inc. http://neurosci.nature.com
ily, which includes nerve growth factor (NGF), brain-derived
neurotrophic factor (BDNF) and neurotrophin-4/5 (NT-4/5).
These factors are crucial for development of the peripheral
nervous system1, but not the central nervous system (CNS),
except that NT-3 and BDNF have been implicated in the post-
natal development of the cerebellum2,3. Here we created a con-
ditional NT-3-deficient mutant, which showed abnormal
cerebellar morphology.
Postnatal cerebellar development, which involves active
proliferation and migration of neuronal precursors, is com-
plete by postnatal day 21 (P21) in the mouse 4. During this
period, the neurotrophins BDNF and NT-3 are highly
expressed in the cerebellum5,6. BDNF-deficient mutants gen-
erally die within the first two weeks of life, but cerebellar devel-
opment can be assessed during this time window 7 . NT-3
-deficient mutants, on the other hand, rarely survive past P1
(ref. 8), preventing a meaningful assessment of postnatal cere-
bellar development. To circumvent this problem, we used the
phage P1 cre recombinase-loxP system9 to create an NT-3 con-
ditional mutant that was viable, yet lacked NT-3 expression
in the CNS.
A targeting vector (Fig. 1a) designed to introduce loxP sites
around the NT-3 coding exon (exon II)10 was transfected into
embryonic stem cells. The selection cassette was removed from
homologously targeted clones by transient transfection of a cre
recombinase expression vector, leaving loxP sites surrounding
exon II (Fig. 1c; data not shown). Embryonic stem cells carry-
ing this new allele of NT-3, referred to as NT-32lox, were used to
generate a mouse line (data not shown). Mice bearing the NT-
32lox allele were intercrossed, generating viable and fertile NT-
32lox/2lox homozygotes, without behavioral abnormalities (data
not shown). To delete NT-3 in the CNS, we crossed NT-32lox/+
mice to a transgenic strain bearing the cre recombinase gene
expressed under the control of the rat nestin promoter/intron
2 enhancer (A.T., unpublished data). Nestin-cre-mediated
recombination resulted in the deletion of NT-3 exon II, leav-
ing a single loxP site remaining in the genome (Fig 1c). As exon
II contains the entire coding sequence of the protein, this
NT-31lox allele is equivalent to the NT-3-null mutation8. South-
ern blot analysis detected cre-mediated recombination in whole
embryos as early as 9.5 days post coitum (E9.5). Recombina-
tion increased during embryonic development such that it was
nearly complete in brain and spinal cord by E15.5, whereas
other tissues showed a much smaller extent of recombination
(Fig. 2a). In adults, incomplete recombination was detected in
nature neuroscience volume 2 no 2 february 1999
scientific correspondence
a number of tissues, but the NT-32lox allele was almost fully
recombined in the CNS (Fig. 2b) and in the germ line of mice
that carried the nestin-cre transgene (data not shown). There-
fore, to generate NT-3 conditional mutants, NT-32lox/+ males
bearing the nestin-cre transgene were crossed to NT-3 2lox/+
nontransgenic females. Conditional mutants generated from
this cross carried the transgene and had the genotype NT-
31lox/2lox at the NT-3 locus. Conditional mutants were produced
in the expected numbers and were fertile (data not shown),
indicating that the conditional mutation did not lead to peri-
natal lethality, unlike the NT-3-null mutation8. Northern blot-
ting revealed no NT-3 mRNA in the brain and developing
cerebellum of conditional mutants (Fig. 2c and d), demon-
strating that the NT-32lox allele was fully recombined and no
NT-3 was produced.
Because NT-3 is highly expressed during postnatal cere-
bellar development2,6 (see also Fig. 2d) and can influence the
survival and differentiation of developing cerebellar neurons2,
we examined the development of the cerebellum in the NT-3
conditional mutant. Like BDNF mutants3, NT-3 conditional
mutants had defective cerebellar foliation at P8, distinguished
by the absence of folium VII and the posterior superior fis-
sure, which separates it from folium VI (Fig. 3a and b). This
defect was not simply due to a delay in folding, as folium VII
was also absent at P28 and later (data not shown). Formation
of the characteristic layers of the cerebellum and their asso-
ciated cell types seemed to be unaffected in the conditional
a
b
c
Fig. 1. Targeting strategy and Cre mediated recombination. (a) NT-3
genomic locus (NT-3+), with the homologous targeting vector
depicted below. Filled triangles indicate loxP sites. Hatched rectangle
indicates the cytomegalovirus-promoter-driven hygror/thymidine
kinase fusion gene used for positive/negative selection. Probes used
for detection of integration and recombination indicated above.
(b) Structure of NT-3 genomic locus after homologous integration of
the targeting vector. This new allele is referred to as NT-33lox. (c) Top,
NT-3 allele resulting from cre-mediated recombination removing the
selection cassette, referred to as NT-32lox. Bottom, the result of cre-
mediated recombination of NT-32lox, referred to as NT-31lox. For (ac),
restriction sites (B, BamHI; S, Spe I; Xb, Xba I; C, Cla I; Xh, Xho I) are
shown, with the resulting fragment sizes given in kilobases (kb).
115
 1999 Nature America Inc. http://neurosci.nature.com

scientific correspondence

Fig. 2. Cre-mediated recombination and expression of NT-3 in the

a conditional mutant. (a) Southern blot of embryonically derived sam-
ples that carried the nestin-cre transgene and had the genotype
NT-32lox/+ at the NT-3 locus. NT-3+, NT-32lox and NT-31lox bands are
indicated. Note the near-complete conversion of NT-32lox into
NT-31lox by E15.5 in the brain and spinal cord. Whole emb., whole
embryo; sp.cord, spinal cord; carcass refers to whole embryo minus
head, liver, heart and limbs. (b) Southern blot showing recombina-
tion in an NT-32lox/+ adult. Rem. brain is the portion of the brain
remaining after dissection into the other brain regions shown. Sal.
gl., salivary gland; olf. bulb, olfactory bulb; cereb., cerebellum; cor-
b tex, cerebral cortex; sp. cord, spinal cord. (c) Northern blot com-
paring NT-3 mRNA in the whole brain of wild-type (w) and
conditional mutant (m) mice. 28s and 18s indicate the position of the
rRNA markers. Blot stripped and reprobed with GAPDH shown
below. (d) Northern blot comparing NT-3 mRNA in cerebella of
wild-type (w) and conditional mutant (m) mice at various postnatal
time points. Ad, adult. rRNA loading control shown below.

c
1999 Nature America Inc. http://neurosci.nature.com

d
mutant (Fig 3c and d). The external germinal layer, site of
granule cell precursor proliferation, was present at P8. Cells
in the proliferative zone of this layer incorporated 5-bromod-
eoxyuridine at P8 in wild-type mice and in conditional
mutants (Fig. 3e and f). NT-3 has been implicated in differ-
entiation of developing Purkinje neurons2. Purkinje neurons
were present in the vermis of the conditional mutant, and
although not quantified, they did not appear to be reduced in
number. The elaboration of the Purkinje dendritic tree, as
revealed by staining with an anti-calbindin D28 antibody, also
appeared qualitatively normal (Fig. 3g and h).
During cerebellar development, excess granule neurons are
produced and subsequently lost via apoptotic cell death. In
the mouse, the peak of apoptosis in the cerebellum occurs
around P8 (ref. 11). At this age, TUNEL staining showed that
conditional mutants had 75% higher density of apoptotic cells
in the granule cell layer (mutant mean standard error,
37.3 11.4 cells/mm 2 ; wild-type, 21.3 6.98 cells/mm 2 ;
p < 0.0001, unpaired students t-test) During this time, both
astrocytes and granule neurons are undergoing apoptosis in
the granule cell layer12. However, granule neurons express the
NT-3 receptor TrkC5, and their survival in vivo is enhanced
by NT-3 (ref. 2), suggesting that this increase in apoptosis is
due to granule neuron death. We observed no increase in
TUNEL-positive cells in the external germinal layer, suggest-
ing that granule neuron precursors did not require NT-3
for survival (mutant, 4.1 0.31 cells/mm; wild type,
3.8 0.26 cells/mm; p = 0.38, unpaired students t-test).
We have used the cre/loxP system to create a conditional
mutation that eliminated NT-3 expression in the CNS but did
not cause early postnatal death. The absence of NT-3 in the
developing CNS caused abnormal cerebellar development,
providing evidence that it is required for the correct forma-

a b e f

g h
c d

Fig. 3. Conditional mutants have aberrant cerebellar foliation but normal layering, external germinal layer proliferation and Purkinje arboriza-
tion. (ad) Paraffin-embedded, cresyl-violet-stained midsagittal sections through the cerebellar vermis of wild-type (a, c) and mutant mice (b,
d) at P8. Triangle in (b) marks the area of the absent folium VII in the mutant. (c, d) Higher magnification showing the characteristic layers of
the cerebellum at P8. EGL, external germinal layer; ML, molecular layer; PCL, Purkinje cell layer; GL, granule cell layer. Scale bars in (a) and (b),
500 m; in (c) and (d), 40 m. (e, f) Granule cell precursor proliferation assessed by injection of 5-bromodeoxyuridine (BrdU). Incorporation
of BrdU is detected by immunohistochemistry using an antibody to BrdU in mutant (f) and wild-type mice (e). Scale bars, 20 m.
(g, h) Purkinje cells detected with an antibody to Calbindin D28. The mutant (h) appears qualitatively similar to wild-type (g) at P8.

116 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
tion of at least some components of the CNS. Indeed, this
mutation was also associated with behavioral abnormalities,
such as increased gait width and impaired balance (data not
shown), consistent with a role for NT-3 in the development
and maintainence of the cerebellar system. The finding that
cell death was increased in the granule cell layer further sug-
gests that NT-3 may act as a survival factor for a subset of
granule neurons at P8. Because granule neurons express both
NT-3 and TrkC2,5,6, the factor may be acting in an autocrine or
paracrine fashion. NT-3 is also expressed in the adult cere-
bellum6(see also Fig. 2d), suggesting that it may be involved
in adult cerebellar function and long-term neuronal survival.
Apart from the well characterized activity of NT-3 as a neu-
ronal survival factor, it has other effects on CNS neurons,
including dendritic arborization2,13 and synaptic plasticity14.
This conditional mutant therefore will allow the exploration of
other NT-3 functions in the adult CNS, as well as the effica-
cy of NT-3 substitution therapy.
1999 Nature America Inc. http://neurosci.nature.com
ACKNOWLEDGEMENTS
The authors thank Peggy Lee and Steven OGormann for providing reagents.
nature neuroscience volume 2 no 2 february 1999
scientific correspondence
B.B. also thanks Jeanne Reis and Noah Duffey for technical support and
Scharam Akbarian, Larry Schwartz and Kuo Fen Lee for comments on the
manuscript. This research was supported by a grant from NIH/NCI.
RECEIVED 29 O CTOBER; ACCEPTED 18 DECEMBER 1998
1. Conover, J. C. & Yancopoulos, G. D. Rev. Neurosci. 8, 1327 (1997).
2. Neveu, I. & Arenas, E. J. Cell Biol. 133, 631646 (1996).
3. Schwartz, P. M., Borghesani, P. R., Levy, R. L., Pomeroy, S. L. & Segal, R.
A. Neuron 19, 269281 (1997).
4. Goldwitz, D. & Hamre, K. Trends Neurosci. 21, 375382 (1998).
5. Ernfors, P., Merlio, J. P. & Persson, H. Eur. J. Neurosci. 4, 11401158
(1992).
6. Rocamora, N., Garcia-Ladona, F. J., Palacios, J. M. & Mengod, G. Mol.
Brain Res. 17, 18 (1993).
7. Ernfors, P., Lee, K. F. & Jaenisch, R. Nature 368, 147150 (1994).
8. Ernfors, P., Lee, K. F., Kucera, J. & Jaenisch, R. Cell 77, 503512 (1994).
9. Gu, H., Marth, J. D., Orban, P. C., Mossmann, H. & Rajewsky, K. Science
265, 103106 (1994).
10. Leingartner, A. & Lindholm, D. Eur. J. Neurosci. 6, 11491159 (1994).
11. Wood, K. A., Dipasquale, B. & Youle, R. J. Neuron 11, 621632 (1993).
12. Krueger, B. K., Burne, J. F. & Raff, M. C. J. Neurosci. 15, 33663374
(1995).
13. McAllister, A. K., Katz, L. C. & Lo, D. C. Neuron 18, 767778 (1997).
14. Kang, H. & Schuman, E. M. Science 267, 16581662 (1995).
117
 1999 Nature America Inc. http://neurosci.nature.com

articles

Snapin: a SNARE-associated protein

implicated in synaptic transmission
Jeffrey M. Ilardi1, Sumiko Mochida2 and Zu-Hang Sheng1

1 Synaptic Function Unit, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Building 36, Room 5A23,
9000 Rockville Pike, Bethesda, Maryland 20892, USA
2 Department of Physiology, Tokyo Medical University, Tokyo 1608402, Japan
Correspondence should be addressed to Z.-H.S. (zsheng@codon.nih.gov)

Synaptic vesicle docking and fusion are mediated by the assembly of a stable SNARE core complex of
proteins, which include the synaptic vesicle membrane protein VAMP/synaptobrevin and the
plasmalemmal proteins syntaxin and SNAP-25. We have now identified another SNAP-25-binding pro-
1999 Nature America Inc. http://neurosci.nature.com

tein, called Snapin. Snapin was enriched in neurons and exclusively located on synaptic vesicle
membranes. It associated with the SNARE complex through direct interaction with SNAP-25. Binding
of recombinant Snapin-CT to SNAP-25 blocked the association of the SNARE complex with synaptotag-
min. Introduction of Snapin-CT and peptides containing the SNAP-25 binding sequence into presynap-
tic superior cervical ganglion neurons in culture reversibly inhibited synaptic transmission. These
results suggest that Snapin is an important component of the neurotransmitter release process
through its modulation of the sequential interactions between the SNAREs and synaptotagmin.

The molecular mechanisms that underlie calcium-dependent
exocytosis have been successfully investigated by identifying pro-
teins that are involved in the vesicle docking/fusion process at
presynaptic nerve terminals and analyzing their interactions13.
Synaptic vesicle docking and fusion at release sites require the
association of proteins on both vesicle and plasma membranes.
The SNARE core complex is a biochemical intermediate essen-
tial for vesicular transport and/or fusion processes. It consists
of the synaptic-vesicle-associated protein synaptobrevin/VAMP
and the plasma-membrane-associated proteins syntaxin and
SNAP-25 (synaptosome-associated protein-25; refs 410).
Although many classes of exocytosis-related proteins have been
identified to date, the number and identity of the essential and
regulatory proteins involved remains unclear. The identification
of neuron-specific regulators or molecular switches involved in
the assembly and disassembly of the ternary SNARE complex,
and its structural coupling to a calcium sensor during synaptic
vesicle docking and calcium-regulatory fusion processes is crit-
ical to understanding the molecular mechanisms underlying
neurotransmitter release and synaptic plasticity.
RESULTS
Cloning and subcellular localization of Snapin
To identify regulatory proteins in synaptic vesicle exocytosis that
bind to components of the SNARE complex, we used SNAP-25
(ref. 7) as a molecular bait and screened a human brain cDNA
library via yeast two-hybrid selection11. Positive prey clones were
rescued and re-transformed into fresh yeast cells with SNAP-25
bait or various control baits. The specificity of the bait and prey
interactions was then confirmed by transactivation assays of -
galactosidase induction and histidine autotrophy. We isolated
three classes of complementary DNAs, encoding syntaxin 1A5,6,
Hrs-2 (ref. 12) and a new protein called Snapin, which interact-
ed selectively with the SNAP-25 bait. The predicted open read-
ing frame for human Snapin cDNA encodes a 136-amino acid
protein with a calculated molecular mass of 15 kDa (Fig. 1a).
The start codon of the open reading frame was consistent with
Kozak consensus sequences13. EST databank searches identified
a murine Snapin homologue that shares 98% amino-acid
sequence identity with human Snapin. Analysis of the Snapin
sequence indicated that it is composed of an amino-terminal
hydrophobic segment characteristic of a transmembrane domain
and a carboxyl-terminal region with the potential (p = 0.62) to
form a coiled-coil structure, suggesting a role in mediating synap-
tic proteinprotein interactions14 (Fig. 1b).
Immunoblot analysis using a polyclonal antibody raised
against the carboxyl terminal peptide sequence from mouse
Snapin showed that a band of mass 15 kDa (as expected from
the open reading frame of Snapin) was readily detected in rat
brain and synaptosome preparations and was barely present in
other tissues (Fig. 2a). Moreover, by immunoblotting various
homogenates, we found Snapin to be widely present in all twelve
anatomically and functionally distinct areas of rat brain includ-
ing cortex, cerebellum, hippocampus and spinal cord (data not
shown). To determine where Snapin is localized within neurons,
rat cerebral synaptosomes were fractionated into cytosolic,
synaptic vesicle and synaptic plasma membrane fractions, and
then analyzed by immunoblotting. Snapin protein was exclu-
sively associated with the synaptic vesicle fraction and absent
from the cytosolic and plasma membrane fractions (Fig. 2b),
consistent with the prediction of an amino-terminal hydropho-
bic transmembrane segment from structural analysis. The vesi-
cle fraction also contained the majority of immunoreactivity
corresponding to VAMP-2, synaptophysin and synaptotagmin,
markers of synaptic vesicles. As controls, neither Na+/K+-
ATPase, a marker of the plasma membrane, nor lactate dehydro-
genase (LDH), a marker of the cytosolic fraction, were detected
in the synaptic vesicle fraction. Syntaxin-1 and SNAP-25 were
present primarily in the presynaptic plasma membrane frac-
tions and to a lesser extent in synaptic vesicle fractions.

nature neuroscience volume 2 no 2 february 1999 119

 1999 Nature America Inc. http://neurosci.nature.com
articles
Fig. 1. Molecular cloning of Snapin. a
(a) Alignment of the predicted amino-acid
sequences of human and mouse Snapins.
Residues that differ between the two iso-
forms are indicated by bold letters. The
two sequences are 98% identical at the
amino-acid level. The GenBank accession
numbers for human and mouse Snapin are
AF086837 and AF086838, respectively. b
(b) Domain structure of Snapin. A 37-
amino-acid predicted coiled-coil domain
(CC) and an amino-terminal hydrophobic
segment (HS) are indicated.
Biochemical extraction experiments showed that Snapin was
resistant to extraction from membrane by either high salt con-
centration (up to 1 M) or extreme pH (3 and 11; data not
shown), whereas it was soluble in HEPES buffer with 1% Triton
1999 Nature America Inc. http://neurosci.nature.com
X-100. These data suggest that Snapin behaves as an integral,
vesicle-associated, membrane protein.
Association of Snapin with SNARE complex
Because the yeast two-hybrid system can detect low-affinity inter-
actions that may not occur in vitro and in vivo, we tried to con-
firm the selective and direct interaction between Snapin and
SNAP-25 using in-vitro binding assays with recombinant pro-
teins. Whereas His-tagged Snapin bound strongly to GST-SNAP-
25, no binding was detectable to GST alone or to other synaptic
proteins tested including syntaxin-1A, SNAP-23 (ref. 15),
VAMP-2 (ref. 4), and the cytoplasmic domain (amino acids
80421, ref 16) of synaptotagmin 1 (Fig. 3a). SNAP-25 binding to
Snapin saturated at 100 nM, a 5-fold molar excess of the soluble
His-Snapin over the immobilized GST-SNAP-25 (Fig. 3b). His-
Snapin bound to GST-SNAP-25 at a stoichiometry of 1:1, as
observed on Coomassie-blue-stained gels (data not shown). Thus,
both in-vitro binding and yeast two-hybrid analysis indicate that
a b
Fig. 2. The tissue and subcellular distribution of Snapin.
(a) Tissue distribution of Snapin protein. Rat tissue
homogenates (50 g protein per lane) and solubilized
synaptosome preparation (10 g per lane) were
immunoblotted with the Snapin antibody. Bands were visu-
alized by ECL. (b) Distribution of Snapin in subcellular frac-
tions of rat brain synaptosomes. Equal amounts (8 g) of
synaptosome fractions enriched in presynaptic cytosol
(PC), synaptic vesicles (SV), and plasma membrane (PM)
were analyzed by SDS polyacrylamide gel electraphoresis
(PAGE) and sequentially immunoblotted with antibodies as
indicated after stripping previous antibodies in the same
blot membrane. The subcellular localization of Snapin was
determined by comparing it to markers for synaptic vesi-
cles (VAMP, synaptophysin and synaptotagmin), plasma
membrane (K/Na-ATPase, syntaxin, and SNAP-25), and
cytoplasm (lactate dehydrogenase, LDH).
120
Snapin interacts preferentially with SNAP-25 in a saturable and
stoichiometric manner, and suggest that they may interact in vivo.
Using anti-SNAP-25 affinity-column chromatography, we
examined whether Snapin is a SNAP-25- and/or SNARE-com-
plex-associated protein in the rat synaptosome preparation.
Snapin, along with other components of the SNARE complex,
was coimmunoprecipitated by the anti-SNAP-25 antibody but
not by normal control IgG, as demonstrated with both
Coomassie blue staining and sequential immunoblot analysis
on the same membrane using antibodies to synaptotagmin I,
syntaxin 1, SNAP-25, VAMP2, and Snapin (Fig. 4a). Under the
conditions used for this experiment, these proteins were copre-
cipitated at molar ratios of approximately 1:1:0.3:0.3 (syn-
taxin:SNAP-25:VAMP2:Snapin). The relative low
stoichiometry of the complex may reflect the presence of abun-
dant dimers of syntaxin-SNAP-25 in vivo that are not bound to
the vesicle-membrane-associated proteins VAMP2 and Snapin.
Furthermore, all SNARE components, including syntaxin,
SNAP-25, VAMP2 and synaptotagmin, were co-precipitated by
the anti-Snapin antibody (data not shown). Anti-VAMP2 and
anti-syntaxin antibodies also co-precipitated Snapin (data not
shown). This indicates that Snapin is a SNAP-25-binding pro-
tein that forms a newly discovered
complex by association with Syntaxin,
SNAP-25 and VAMP2.
Yeast two-hybrid selection and in-
vitro binding studies (Fig. 3) showed
that the carboxyl-terminal half of
Snapin (Snapin-CT, amino acids
79136) is sufficient for the interaction
with SNAP-25. To investigate whether
Snapin-CT could compete with the
full-length Snapin (Snapin-FL) for
binding to SNAP-25, we did in-vitro
binding competition studies with
recombinant proteins. GST-SNAP-25
was bound to glutathione-Sepharose
beads and incubated with a constant
concentration of Snapin-FL and
increasing concentrations of Snapin-
CT. The signal intensity of Snapin-FL
diminished progressively while that of
Snapin-CT increased (Fig. 4b). These
results demonstrate that both Snapin-
FL and -CT indeed compete for the
same binding region on SNAP-25.
To examine the cellular functions of
Snapin, we repeated the immunopre-
cipitation of Snapin with the SNARE
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 3. In-vitro interaction of Snapin with

SNAP-25. (a) Binding of Snapin-CT a b

Snapin bound (pixels, 100)

(79136) to SNAP-25 in vitro. GST or GST
fusion proteins (approximately 1 g each)
were immobilized on glutathione-Sepharose
and then incubated with His-tagged Snapin.
Bound protein complexes were eluted from
the matrix and separated by SDS-PAGE, and
immunoblotted with anti-His and anti-GST
antibodies. (b) Recombinant Snapin-CT
(79136) binds SNAP-25 in a saturable man-
ner. Approximately 1 g GST-SNAP-25
immobilized on glutathione-Sepharose was
incubated with increasing amounts of His-
Snapin as indicated. The bound Snapin was His-Snapin (nM)
detected by immunoblotting, and binding
intensities were quantified using NIH Image
based on standard curves.
1999 Nature America Inc. http://neurosci.nature.com

complex and synaptotagmin from brain synaptosomes (Fig. 4a)
in the presence of excess exogenous, bacterially expressed
Snapin-FL or Snapin-CT. We found that both recombinant
Snapin-CT and Snapin-FL coprecipitated with the SNARE
complex, suggesting that Snapin, syntaxin and VAMP-2 interact
simultaneously with SNAP-25 and do not compete with each
other (Fig. 4c). Interestingly, coincubation of synaptosomes
with excess Snapin-CT (containing the coiled-coil domain and
presumably competing with endogenous Snapin for SNAP-25
binding) prevented the association of synaptotagmin with the
SNARE complex even in the presence of 200 M free calcium
(Fig. 4c). Conversely, Snapin-FL significantly enhanced the
association of synaptotagmin with the SNARE complex, as
compared with the addition of a 6His-peptide control. We
observed that Snapin was less abundant in synaptosomes than
syntaxin and SNAP-25 (data not shown), and the addition of
excess recombinant Snapin-FL might increase formation of the
SNARE-Snapin complex. Thus, the opposite effects of full-
length and C-terminal recombinant Snapin on the interaction
of SNAREs with synaptotagmin suggest that Snapin is required
for high-affinity synaptotagminSNARE interaction.
The functional effect of Snapin on synaptic transmission
To assess the physiological function of Snapin, we examined its
role in synaptic transmission at the well characterized choliner-
gic synapses formed among superior cervical ganglion neurons
(SCGN) in culture 1719. We injected the carboxyl terminal
domain of mouse Snapin (Snapin-CT) into the presynaptic
SCGNs to determine whether it would disrupt the interaction of
the SNARE core complex with Snapin in vivo. This synapse is an
ideal system for these experiments for three reasons. First, pro-
teins or peptides can be introduced into the relatively large

Fig. 4. Association of Snapin with

the SNARE complex in vivo. a c
(a) Coimmunoprecipitation of
Snapin with the SNARE complex.
Immunoprecipitates with anti-
SNAP-25 antibody or control IgG
were analyzed with Coomassie
blue staining (ST, right panel) or
by sequential immunoblotting (IB,
left panel). (b) Binding competi-
tion between Snapin-FL and
Snapin-CT. Immobilized GST-
SNAP-25 (100 nM) was incubated
with equal concentrations of
recombinant Snapin-FL (250 nM)
and increasing concentrations of
Snapin-CT as indicated. Bound
b
proteins were visualized by
sequentially immunoblotting with
both anti-His and anti-GST anti-
bodies. (c) Effects of recombinant
Snapin on the association of
synaptotagmin with SNARE com-
plex. Anti-syntaxin (-stx)
antibody immunoprecipitated
synaptotagmin with SNARE pro-
teins from rat brain synaptosomes in the presence or absence of His-Snapin-CT or His-Snapin-FL as indicated. Proteins bound were analyzed by
sequential immunoblotting. Solubilized synaptosomes (SS) were loaded in the left lane as a detection control for antibodies. Note that synapto-
tagmin association with the SNARE complex is blocked by 5 M His-Snapin-CT, and significantly increased by 5 M His-Snapin-FL.

nature neuroscience volume 2 no 2 february 1999 121

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 5. Effects of Snapin-CT on synaptic

transmission of SCGNs in culture. a b

Normalized average EPSP amplitude

(a) Snapin-CT was introduced into the
presynaptic neuron by diffusion from a
suction pipette beginning when the mem- Denatured (n = 5)
brane was disrupted by applying suction at Normal (n = 5)
t = 0. The pipette concentration of Snapin-
CT was 50 M. Postsynaptic potentials
from one representative experiment
recorded 5 min before injection and 20
and 50 min after injection are illustrated.
(b) Normalized average excitatory post-
synaptic potentials (EPSPs) were plotted
from five experiments with Snapin-CT (l)
and heat-denatured Snapin-CT (L) at a
Time after injection (min)
concentration of 50 M.

(3040 m) presynaptic cell bodies by microinjection. Second, cellular fusion protein in the cell body and nerve terminal. In
1999 Nature America Inc. http://neurosci.nature.com

the injected proteins can rapidly diffuse to the nerve terminals contrast, injection of 50 M heat-denatured Snapin-CT produced
forming synapses with adjacent neurons. Third, the effects on no significant decrease in EPSP amplitude during one hour of
stimulated release of acetylcholine can be accurately monitored by recording (0.9 8.6% at 30 min after injection, n = 5; Fig. 5b),
recording the excitatory postsynaptic potentials (EPSPs) evoked indicating that the inhibitory effect of Snapin-CT on synaptic
by action potentials in the presynaptic neurons1719. After a sta- transmission is dependent on the native conformation of Snapin-
ble period of control recording for 2030 min, 50 M recombi- CT. Although the peak of EPSPs was reduced by Snapin-CT, the
nant Snapin-CT was diffused into presynaptic neurons from a time course was not significantly changed (Fig. 5a), indicating
suction pipette (at t = 0) for 23 minutes. EPSP amplitude grad- that the kinetics of exocytosis were not affected by introduction
ually decreased over a period of 20 minutes (Fig. 5a and b). The of Snapin-CT. The simplest interpretation of our results is that
maximum decrease, 29 2.3 % (n = 5, mean standard error), inhibition of synaptic transmission is due to competitive block
was observed 2040 min after commencing the injection, and of the endogenous Snapin binding to SNARE complex by the
the EPSPs recovered to the control amplitude by 6080 min. We excess injected Snapin-CT and resultant decrease of the association
assume that recovery follows proteolytic degradation of the intra- of the SNAREsynaptotagmin complex. This interpretation is con-
sistent with the results from our in-vitro
biochemical studies, which showed that
a Snapin-CT interrupted the interaction
of synaptotagmin with the SNARE
complex (Fig. 4c).
Snapin protein is composed of two
b major domains: an amino-terminal
hydrophobic segment and a carboxyl-
terminal coiled-coil domain that is
d present in different families of vesicu-
H M2 (n = 5) lar fusion proteins, including SNAP-
Normalized average EPSP amplitude

R Sc (n = 6)
g W2 (n = 6) 25, and is the binding domain of
p W4 (2.5 mM, n = 6) SNARE components. To define the
l W4 (5 mM, n = 6)
minimal sequence and critical amino-
acid residues required for Snapins
c binding to SNAP-25 and interfering
with endogenous Snapin function,
four 20-mer peptide fragments (W1,
2, 3 and 4; Fig. 6a) corresponding to
the coiled-coil domain of mouse
Fig. 6. Inhibition of binding to SNAP-25 and Snapin were synthesized. They were
neurotransmitter release by the Snapin pep- tested for their ability to affect in vitro
tides. (a) Amino-acid sequences of four Time after injection (min) binding competition (Fig. 6b and c)
mouse Snapin peptides (W1W4), two and their ability to inhibit synaptic
point-mutated peptides (M1, M2) and one transmission between SCGNs in cul-
scrambled peptide (Sc). (b) Effects of pep- ture (Fig. 6d). W4 but not W1, W2 or
tides (50 M) on the binding of Snapin (0.6 M) to immobilized GST-SNAP-25. The binding was signifi-
W3 peptides markedly competed with
cantly inhibited by W4 and M1 but not by W1, W2, W3, M2 or Sc. (c) The concentration-dependent
inhibition of Snapin-FL (250 nM) binding to SNAP-25 (100 nM) by W4 peptide. (d) Differential effects of
recombinant Snapins ability to bind
W2, W4, Sc and M2 peptides on synaptic transmission of SCGNs in culture. Normalized average EPSPs to SNAP-25 (Fig. 6b and c) and block
were plotted from six experiments with 2.5 mM W2 (g), 2.5 mM W4 (p), 5 mM W4 (l)or 2.5 mM Sc the co-immunoprecipitation of native
(R) and from 5 experiments with 2.5 mM M2 (H) peptides. Peptides were introduced into presynaptic synaptotagmin with the SNARE com-
neurons at t = 0 by diffusion from a suction pipette. plex in synaptosomes (data not

122 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
shown). The same W4 peptide also inhibited neurotransmitter
release (Fig. 6d). The maximal decrease in EPSP amplitude,
34 6.8% (n = 6), was observed 1525 min after starting injec-
tion with a pipette containing 2.5 mM W4 peptide (Fig. 6d).
Inhibition 1.3-fold greater and prolonged was observed when a 2-
fold higher pipette concentration of W4 peptide (5 mM) was used,
indicating that the effect of the W4 peptide was not near saturation
in nerve terminals. The maximum decrease in EPSP amplitude,
44 4.2% (n = 6), was observed 4050 min after injection. W2,
used as a representative of the other peptides, reduced synaptic
transmission to a lesser extent (15 1.8%, n = 6, mean of maxi-
mal decrease in EPSP amplitude in six individual experiments;
Fig. 6d). This suggests that the sequence of the W2 peptide is not
sufficient to compete with recombinant or native Snapin for bind-
ing to SNAP-25 in vitro, but may interfere with endogenous Snapin
for functional interaction with SNARE complex in nerve termi-
nals. Sc, a peptide with an amino-acid composition identical to
that of W4, but with a scrambled sequence, had no effect on the
1999 Nature America Inc. http://neurosci.nature.com
binding of Snapin to SNAP-25 (Fig. 6b), on the block of synapto-
tagmin interacting with SNARE complex (data not shown) or on
transmitter release (Fig. 6d). Moreover, replacing three central
residues of the W4 peptide with the positively charged amino acids
Lys or Arg (Leu124Lys, Asp125Lys and Val128Arg), designed to
destabilize the coiled-coil structure and interfere with
proteinprotein contact, yielded a peptide (M2) inactive in either
competition of binding or inhibition of synaptic transmission.
These results suggest that these three residues of Snapin are critical
to maintaining an active interaction with SNAP-25. The inhibitory
effects of W4 were apparent 510 min after starting injection and
reached a maximum within 1525 min at 2.5 mM or 4050 min at
5 mM pipette concentrations. The W4 peptide may act faster than
Snapin-CT not only due to the difference in the injected concen-
trations but also to the difference in their diffusion time from cell
body to active zones. These results indicate that the W4 peptide
acts in a sequence-specific manner, and that its effects reflect an
action related to Snapin.
DISCUSSION
Here we have identified a new synaptic vesicle membrane protein
called Snapin, which associates with the SNARE complex through
its direct interaction with SNAP-25. Binding of recombinant Snapin-
CT to SNAP-25 blocks the association of the SNARE complex with
synaptotagmin, one of the putative calcium sensors involved in neu-
rotransmission16,20,21. The molecular mechanisms of the interac-
tion of Snapin with SNAP-25 during the process of vesicle
docking/fusion are still unclear. However, the relatively rapid inhi-
bition of synaptic transmission resulting from injection of recom-
binant Snapin-CT and the SNAP-25-binding site peptide indicates
that the interaction between Snapin and SNAP-25 and/or the SNARE
complex is involved in a late step of exocytosis.
Only partial inhibition of synaptic transmission was observed
when 50 M Snapin-CT (29 2.3%) and 5 mM W4 peptide
(44 4.2%) were injected. Based on the color density of co-
injected Fast Green FCF dye and correction for the effect of molec-
ular mass22, the estimated concentration of Snapin-CT in the cell
body was 1.5 M (approximately 3% of the concentration in the
pipette), exceeding the concentration required for half-maximal
binding to SNAP-25 in vitro (100 nM; Fig. 3b). This partial inhibi-
tion has also been reported in other synapse systems after injection
of peptides at much higher concentrations than those required to
block the action of key synaptic proteins in vitro23,24. It is likely that
partial inhibition in these synapses may reflect barriers in neurons
that prevent diffusion of the injected molecules to all the active
nature neuroscience volume 2 no 2 february 1999
articles
zones contributing to transmitter release, rather than the inherent
inability of these molecules to fully block fast synaptic transmis-
sion. Similarly, in our SCGN experiments, the Snapin-CT protein
and W4 peptide may not reach all the active zones in the presynap-
tic terminals of cultured SCGNs that characteristically form an
extensive network of varicosities25. A 2-fold higher concentration
of W4 peptide (5 mM) produced 1.3-fold greater and prolonged
inhibition (Fig. 6d), indicating that the effect of W4 peptide was
not near saturation in nerve terminals. Alternatively, the partial
inhibition we observed probably reflects a modulatory role of
Snapin during synaptic transmission rather than its acting as an
essential factor.
Together, our results suggest that Snapin regulates a step
between vesicle docking and neurotransmitter release through
its ability to potentiate the interaction of synaptotagmin with
the SNAREs, which then leads to the final fusion step triggered
by calcium influx into nerve terminals through voltage-depen-
dent calcium channels. Alternatively, Snapin, as a v-SNARE,
could participate in vesicle docking at the release sites through
its specific interaction with the plasma-membrane-associated
protein SNAP-25 (t-SNARE). This interaction may explain the
finding that synaptic vesicles are capable of docking even when
another pair of SNAREs, synaptobrevin/VAMP (v-SNARE) and
syntaxin (t-SNARE) are deleted26 or cleaved by specific prote-
olytic neurotoxins27, 28.
Regulation of exocytosis is necessary for proper synaptic func-
tion. Our findings add another possible target of protein interac-
tions for modulation of transmitter release as well as neuronal
plasticity. Further characterization of the biochemical mecha-
nisms and modulation of the interactions between Snapin and
SNARE proteins should reveal their physiological significance dur-
ing the process of exocytosis and lead to progress in understanding
the molecular basis of learning and memory.
METHODS
Identification and cloning of Snapin. The yeast two-hybrid system was used
to clone SNAP-25-binding proteins. Full-length mouse SNAP-25b cDNA
was inserted in-frame into the pGBT9 bait vector containing the GAL4 DNA-
binding domain (Clontech, Palo Alto, California). Yeast two-hybrid screens
of a human brain cDNA library in vector pACT1 (Clontech) with the GAL4
activation domain were done and evaluated according to the protocols
described for the MATCHMAKER yeast two-hybrid system (Clontech).
Yeast cells (Y190) were sequentially cotransformed with SNAP-25 bait and
library prey vectors, and then plated on selection medium lacking Trp, Leu
and His. Two independent, but overlapping human Snapin partial clones
expressing His3 and -galactosidase activity were isolated. EST databank
searches identified both human and mouse cDNAs with a single open read-
ing frame encoding for 136 amino acid residues. The full-length mouse
Snapin cDNA was purchased from ATCC31.
Fusion proteins and in-vitro binding. Full-length syntaxin 1A, SNAP-25b,
SNAP-23, VAMP-2 and the cytoplasmic domain of synaptotagmin 1
(80421) were subcloned into GST-fusion vectors pGEX-2 or pGEX-4T
(Pharmacia, Piscataway, New Jersey). Both Snapin-FL (1136) and Snapin-
CT (79136) were subcloned into hexahistidine-tagged fusion protein vec-
tor (pET28C and A; Novagen, Madison, Wisconsin). Fusion proteins were
prepared as crude bacterial lysates by mild sonication in PBS (50 mM sodi-
um phosphate, pH 8.0, 300 mM NaCl, 1% TX-100 plus protease inhibitors).
His-Snapin proteins were purified by binding to nickel-charged nitrilotri-
acetic acid agarose columns (Qiagen, Valencia, California) and eluted with
500 mM imidazole in PBS. The eluates were concentrated with Centriprep-
10 filtration units (Amicon, Beverly, Massachusetts and dialyzed in a 10,000
molecular-weight cutoff dialysis cassette (Pierce, Rockford, Illinois) against
PBS. Approximately 1 g of GST-fusion proteins were bound to glutathione-
Sepharose beads (Pharmacia) in TBS buffer (50 mM Tris-HCl, pH 7.5, 140
mM NaCl, 0.1% TX-100 plus protease inhibitors), incubated at 4C for 1 h
123
 1999 Nature America Inc. http://neurosci.nature.com
articles
with constant agitation, and washed with TBS to remove unbound proteins.
Glutathione-Sepharose beads coupled with similar amounts of GST-fusion
proteins were added to the His-Snapin protein and incubated with gentle
shaking for 3 h at 4C. The beads were washed three times with TBS buffer,
and bound proteins were eluted in 15 mM reduced glutathione in 50 mM
Tris-HCl (pH 8). Eluates were separated from the beads by centrifugation
at 10,000 g for 1 minute and processed for 1020% SDS/tricine gradient gel
electrophoresis. Bound His-Snapin was detected by monoclonal anti-T7.Tag
antibody (Novagen) and GST-fusion proteins were detected by anti-GST
antibody (Pharmacia). HRP-conjugated secondary antibodies and ECL
chemiluminescence (Amersham, Arlington Heights, Illinois) were used to
visualize the bands.
Immunoprecipitation. Rat brain synaptosomes were prepared by differ-
ential and discontinuous Ficoll gradient centrifugation and solubilized29.
Crude synaptosomes were then separated into fractions enriched in presy-
naptic cytosol (PC), synaptic vesicles (SV) and plasma membrane (PM) as
described30. Solubilized synaptosome protein (100300 g) was incubat-
ed with 3 g of anti-SNAP-25, anti-syntaxin (10H5) monoclonal anti-
bodies3 or control mouse IgG (Zymed, South San Francisco, California) in
1999 Nature America Inc. http://neurosci.nature.com
1 ml TBS with 0.1% TX-100, 0.1% BSA and protease inhibitors, and incu-
bated on a microtube rotator at 4C for 1 h. Protein A-Sepharose CL-4B
resin (2.5 mg; Pharmacia) was added to each sample, and the incubation
continued for an additional three hours, followed by three washes with
TBS/0.1% TX-100. Subsequent interaction assays of immunoprecipitated
and immobilized protein complexes are described in the section on in-
vitro binding assays above. For multiple detection with different anti-
bodies, blots were first stripped in a solution of 62.5 mM Tris-HCl, pH
7.5, 20 mM DTT and 1% SDS for 30 min at 50C with agitation and
washed with TBS/0.1 % Tween-20 for 2 15 min.
Synaptic transmission between SCGNs. SCG cells from seven-day post-
natal rats were prepared as described1719. After 34 weeks in culture, con-
ventional intracellular recordings were made from two neighboring
neurons using microelectrodes filled with 1 M potassium acetate (4070
M). Postsynaptic responses (EPSPs) were recorded from one of the neu-
rons when action potentials were generated in the other neuron by passage
of current through an intracellular recording electrode. Neurons were
superfused with modified Krebs solution consisting of 136 mM NaCl,
5.9 mM KCl, 5.1 mM CaCl2, 11 mM glucose and 3 mM Na-HEPES (pH
7.4). Recombinant Snapin-CT or peptides were dissolved in the suction
pipette solution (150 mM KAc, 5 mM Mg2+-ATP, 10 mM HEPES, pH
7.4), and introduced into the presynaptic cell body by diffusion from a
suction glass pipette (1720 M tip resistance). Fast Green FCF (5%,
Sigma, St. Louis, Missouri) was included in the injection peptide solu-
tion to confirm their entry into the presynaptic cell body. The injection
pipette was removed 23 min after starting injection. EPSPs were record-
ed once every 20 s (0.05 Hz). Electrophysiological data was collected and
analyzed using software written by Dr. Tauc (CHRS, France). For Figs. 5b
and 6d, the peak amplitudes of EPSPs were measured and averaged. The
resultant values were smoothed by an eight-point moving average algo-
rithm and plotted against recording time with t = 0 indicating the begin-
ning of the presynaptic injection.
ACKNOWLEDGEMENTS
We thank Q. Lao for help with the yeast two-hybrid system, X. Zheng for help
with synaptosome fractionation, M Takahashi for antibodies, P.A. Roche for
SNAP-23 cDNA and S. C. Landis, H. Gainer, R. D. G. McKay, M. Mayer, H.
Arnheiter and J. Rettig for comments on the manuscript. This work was
supported by the intramural research program of NINDS, NIH (J.M.I and Z.-
H.S.), HHMI-NIH Research Scholars Program (J.M.I.), grants from The
Japanese Ministry of Education, Science and Culture (S.M.) and the Human
Frontier Science Program (S.M.).
RECEIVED 6 OCTOBER; ACCEPTED 15 DECEMBER 1998
124
1. Rothman, J. E. Mechanisms of intracellular protein transport. Nature 372,
5563 (1994).
2. Bajjalieh, S. M. & Scheller, R. H. The biochemistry of neurotransmitter
secretion. J. Biol. Chem. 270, 19711974 (1995).
3. Sdhof, T. C. The synaptic vesicle cycle: a cascade of protein-protein
interactions. Nature 375, 645653 (1995).
4. Trimble, W. S., Cowan, D. M. & Scheller, R. H. VAMP-1: a synaptic vesicle-
associated integral membrane protein. Proc. Natl. Acad. Sci. USA 85,
45384542 (1988).
5. Bennett, M. K., Calakos, N. & Scheller, R. H. Syntaxin: a synaptic protein
implicated in docking of synaptic vesicles at presynaptic active zones. Science
257, 255259 (1992).
6. Yoshida, A. et al. HPC-1 is associated with synaptotagmin and -conotoxin
receptor. J. Biol. Chem. 267, 2492524928 (1992).
7. Oyler, G. A. et al The identification of a novel synaptosomal-associated
protein, SNAP-25, differentially expressed by neuronal subpopulations. J.
Cell. Biol. 109, 30393052 (1989).
8. Sllner, T. et al. SNAP receptors implicated in vesicle targeting and fusion.
Nature 362, 318324 (1993).
9. Calakos, N., Bennett, M. K., Peterson, K. & Scheller, R. H. Protein-protein
interactions contributing to the specificity of intracellular vesicular
trafficking. Science 263, 11461149 (1994).
10. Hayashi, T. et al Synaptic vesicle membrane fusion complex: action of
clostridial neurotoxins on assembly. EMBO J. 13, 50515061 (1994).
11. Fields, F. & Song, O. A novel genetic system to detect protein-protein
interactions. Nature 340, 245246 (1989).
12. Bean, A. J., Seifert, R., Chen, Y. A., Sacks, R. & Scheller, R. H. Hrs-2 is an
ATPase implicated in calcium-regulated secretion. Nature 385, 826829
(1997).
13. Kozak, M. An analysis of 5-noncoding sequences from 699 vertebrate
messenger RNAs. Nucleic Acids Res. 15, 81258131 (1987).
14. Weimbs, T. et al. A conserved domain is present in different families of
vesicular fusion proteins: A new superfamily. Proc. Natl. Acad. Sci. USA 94,
30463051 (1997).
15. Ravichandran, V., Chawla, A. & Roche, P. A. Identification of a novel
syntaxin- and synaptobrevin/VAMP-binding protein, SNAP-23, expressed in
non-neuronal tissues. J. Biol. Chem. 271, 1330013303 (1996).
16. Matthew. W. D., Tsavaler, L. & Reichardt, L. F. Identification of a synaptic
vesicle-specific membrane protein with a wide distribution in neuronal and
neurosecretory tissue. J. Cell. Biol. 91, 257269 (1981).
17. Mochida, S. et al. Myosin II is involved in transmitter release at synapses
formed between rat sympathetic neurons in culture. Neuron 13, 11311142
(1994).
18. Mochida, S., Saisu, H., Kobayashi, H. & Abe, T. Impairment of syntaxin by
botulinum neurotoxin C1 or antibodies inhibits acetylcholine release but not
Ca2+ channel activity. Neuroscience 65, 905915 (1995).
19. Mochida, S., Sheng, Z.-H., Baker, C., Kobayashi, H. & Catterall, W. A.
Inhibition of neurotransmission by peptides containing the synaptic protein
interaction site of N-type Ca2+ channels. Neuron 17, 781788 (1996).
20. Li, C. et al Ca2+-dependent and -independent activities of neural and
nonneural synaptotagmins. Nature 375, 594599 (1995).
21. Bommert, K. et al. Inhibition of neurotransmitter release by C2-domain
peptides implicates synaptotagmin in exocytosis. Nature 363, 163165 (1993).
22. Pusch, M. & Neher, E. Rates of diffusional exchange between small cells and a
measuring patch pipette. Pflgers Arch. 411, 204211 (1988).
23. Cornille, F., Deloye, F., Fourni-Zaluski, M.-C., Roques, B. P. & Poulain, B.
Inhibition of neurotransmitter release by synthetic proline-rich peptides
shows that the N-terminal domain of vesicle-associated membrane
protein/synaptobrevin is critical for neuroexocytosis. J. Biol. Chem. 270,
1682616832 (1995).
24. Doussau, F., Clabecq, A., Henry, J.-P., Darchen, F. & Poulain, B. Calcium-
dependent regulation of rab 3 in short-term plasticity. J. Neurosci. 18,
31473157 (1998).
25. Wakshull, E., Johnson, M. I. & Burton, H., Postnatal rat sympathetic neurons
in culture. I. A comparison with embryonic neurons. J. Neurophysiol. 42,
14101425 (1979).
26. Broadie, K. et al. Syntaxin and synaptobrevin function downstream of vesicle
docking in Drosophila. Neuron 15, 663673 (1995).
27. Hunt, J. M. et al. A post-docking role for synaptobrevin in synaptic vesicle
fusion. Neuron 12, 12691279 (1994).
28. Sweeney, S. T., Broadie, K., Keane, J., Niemann, H. & OKane, C. J. Targeted
expression of tetanus toxin light chain in Drosophila specifically eliminates
synaptic transmission and causes behavioral defects. Neuron 14, 341351 (1995).
29. Sheng, Z.-H., Rettig, J., Cook, T. & Catterall, W. A. Calcium-dependent
interaction of N-type calcium channels with the synaptic core complex.
Nature 379, 451454 (1996).
30. Dunkley, P. R. et al. A rapid percoll gradient procedure for isolation of
synaptosomes directly from an S1 fraction: homogeneity and morphology of
subcellular fractions. Brain Res. 441, 5971 (1988).
31. Hillier, L. et al., Generation and analysis of 280,000 human expressed
sequence tags. Genome Res. 6, 807828 (1996).
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

Casein kinase-II regulates NMDA

channel function in hippocampal
neurons
David N. Lieberman1 and Istvan Mody2

1 Neuroscience Graduate Program, Stanford University School of Medicine, Stanford, Californina 94305, USA
2 Departments of Neurology and Physiology, UCLA School of Medicine, Los Angeles, California 900951769, USA
Correspondence should be addressed to D.N.L (liebdude@leland.stanford.edu)

Several second-messenger-regulated protein kinases have been implicated in the regulation of

1999 Nature America Inc. http://neurosci.nature.com

N-methyl-D-aspartate (NMDA) channel function. Yet the role of calcium and cyclic-nucleotide-inde-
pendent kinases, such as casein kinase II (CKII), has remained unexplored. Here we identify CKII as
an endogenous Ser/Thr protein kinase that potently regulates NMDA channel function and
mediates intracellular actions of spermine on the channel. The activity of NMDA channels in cell-
attached and inside-out recordings was enhanced by CKII or spermine and was decreased by selec-
tive inhibition of CKII. In hippocampal slices, inhibitors of CKII reduced synaptic transmission
mediated by NMDA but not AMPA receptors. The dependence of NMDA receptor channel activity
on tonically active CKII thus permits changes in intracellular spermine levels or phosphatase
activities to effectively control channel function.

The modulation of neuronal glutamate receptors by phosphory-
lation is considered to be critical for synaptic plasticity. In par-
ticular, altering NMDA-type glutamate receptor properties by
phosphorylation111 may regulate the unique function of these
receptors in developmental, physiological or pathological neu-
ronal plasticity. The numerous protein kinases that mediate
synaptic plasticity became obvious candidates for the identifica-
tion of kinases capable of increasing NMDA channel activity.
Using exogenous calcium-dependent protein kinases, initial stud-
ies on NMDA-receptor-mediated responses demonstrated
enhancing effects of such kinases on NMDA channel function.
Yet many of these studies have shown minimal effects of specific
kinase inhibitors on NMDA-receptor-mediated currents510.
Whereas these studies demonstrate the capacity of various can-
didate protein kinases to alter channel activity, they leave in ques-
tion their ability to regulate basal channel activity in situ.
Furthermore, unrecognized differences in subunit composition
and stoichiometry of the NMDA receptor channel complex in
different tissues and preparations have encouraged controversy
surrounding the various effects of protein kinases on NMDA
channel function.
More recently, the control of NMDA channel gating by
Ser/Thr kinases has been inferred from experiments in which
the cycle of phosphorylation/dephosphorylation was disrupted
by inhibiting phosphatase activity, particularly that of the
calcium/calmodulin-dependent protein phosphatase calcin-
eurin3,4. Because the protein kinase responsible for modulation
of NMDA channel gating has not been clearly characterized in
physiological studies, we explored the different effects of protein
kinases on channel gating in central neurons. Using cell-
attached recordings of single NMDA channels from acutely dis-
sociated adult hippocampal dentate granule cells to preserve
endogenous second messenger systems, we found that the con-
stitutively active protein kinase CKII continuously controls the
basal function of NMDA receptor channels. In contrast to the
phasic nature of NMDA channel regulation by second-messen-
ger-activated protein kinases, the tonic influence of CKII
establishes a continuous cycle of phosphorylation/dephos-
phorylation from which channel activity can be upregulated by
spermine through stimulation of CKII, or downregulated by
calcineurin following its activation by calcium/calmodulin.
These findings, together with our results demonstrating the
CKII insensitivity of AMPA-type glutamate receptors, reveal a
novel and selective means of regulating the gain of excitatory
synaptic transmission at hippocampal synapses by a tonically
active protein kinase.
RESULTS
We wanted to determine which kinase mediates augmentation
of NMDA channel activity following inhibition of calcineurin3
in cell-attached recordings of single NMDA channels from acute-
ly dissociated adult rat dentate gyrus granule cells. We focused
on calcium/calmodulin-dependent protein kinaseII (CaMKII)
for several reasons. First, calcineurin is frequently colocalized
with CaMKII in principal cells of the hippocampus and cortex12.
Second, exogenous CaMKII seems to regulate both AMPA and
NMDA receptors10, and third, this kinase phosphorylates Ser-
1303 of NR2B and/or a comparable site on NR2A11. The cell-
permeable specific CaMKII inhibitor KN-62 (ref. 13) and its
inactive analogue KN-04 were used to probe the involvement of
intrinsic CaMKII in altering NMDA channel function.
Exposure to KN-62 (10 M) reversibly reduced the fre-
quency of NMDA channel openings (po = 34 7% of control,
p < 0.05, paired t-test, n = 6), but did not significantly change
the open duration, the length of bursts and clusters or the
total open time in these complex openings, (Fig. 1ac,

nature neuroscience volume 2 no 2 february 1999 125

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 1. Effects of CaMKII and CKII Control Control

inhibitors on NMDA channel function in a
cell-attached patches. (a) Continuous
traces of cell-attached recordings before
(upper traces) and after (lower traces)
bath application of kinase inhibitors.
Scale bars, 2 pA/10 ms. (b) Continuous 10 M KN-62 100 M DRB
measurements of NMDA channel
N po before, during (horizontal bar)
and after addition of KN-62 (left panel)
or DRB (right panel) to the bath.
Openings were binned into 10-s inter-
vals in the KN-62 experiment and 1-
minute intervals in the DRB experiment.
b
The example shows the most extreme
reduction in po by KN-62 (po = 12% of
control). It also demonstrates that large
N popen

decreases in po are not necessarily

Npopen
accompanied by significant changes in
the duration of open times (the mean
1999 Nature America Inc. http://neurosci.nature.com

open time in the presence of KN-62 is

93% of that in control medium).
(c) Cumulative probability distributions
of all open durations before, during and
after exposure to KN-62 (left panel) and Time (min) Time (min)
DRB (right panel) for the experiments
shown in (b). DRB significantly reduced
the duration of all openings (mean open
c
time, 68% of control, p < 106,

Cumulative probability
Cumulative probability

Kolmogorov-Smirnov test), whereas

KN-62 had no significant effect on the
duration of openings.

Open time (ms) Open time (ms)

Table 1). This effect of KN-62 on NMDA channel activity did
not result from its known antagonism of voltage-dependent
calcium channels14, because the inorganic calcium channel
blocker cadmium (200 M, n = 4, data not shown) failed to
reproduce the effects of KN-62. Furthermore, it is unlikely
that KN-62 antagonized NMDA receptors, because the struc-
turally related, but CaMKII-inactive, analogue KN-04 had no
effect (n = 3, data not shown). Based on these experiments,
endogenous CaMKII tonically regulates NMDA channel
opening probability (po) by altering the frequency of open-
ings, but not their duration. Accordingly, this kinase cannot
be responsible for the characteristic prolongation of simple
and complex channel openings found after phosphatase inhi-
bition3. Thus, we looked for another kinase that could pro-
duce changes in NMDA channel openings qualitatively similar
to those found after calcineurin inhibition.
We focused our attention on CKII, an ubiquitously dis-
tributed Ser/Thr protein kinase, whose activation is indepen-
dent of intracellular calcium or cyclic nucleotides1520. CKII
is more abundant in brain than in any other tissue, and is
more concentrated in neurons than in glia17,18. Despite its
distinct nuclear, cytosolic and membrane-bound activities in
other cell types, little is known about its role in neurons.
However, a PROSITE analysis of the C-termini of the NMDA
receptor subunits NR1, NR2A and NR2B revealed that the
greatest number of consensus phosphorylation sites are those
potentially phosphorylated by CKII15.
To examine the role of CKII in modulating NMDA channel
function, we tested the effect of the cell-permeable CKII inhibitor
5,6-dichloro-1--D-ribofuranosyl benzimidazole (DRB, 50100
M)19,20 on NMDA channel activity in cell-attached recordings.
In biochemical assays using casein as a substrate and GTP as a
donor, 60 M DRB inhibits over 80% of the CKII activity mea-
sured in rat hippocampal slices, and produces a 65% reduction in
the activity of 1 pmol (140 ng) of recombinant CKII (P.T.
Tuazon, J. Roig and J.A. Traugh, personal communication). Not
only did DRB significantly reduce the po to 39 9 % of control
(p < 0.05, paired t-test, n = 6), it also significantly shortened the
mean open time, burst length, cluster duration and total open
time during these events (Fig. 1ac, Table 1). The changes in
NMDA channel gating, notably the reduction in the duration of
openings, presumably caused by inhibition of CKII by DRB,
appear conspicuously similar to those observed following cal-
cineurin activation3. Consistent with the involvement of an
active phosphatase, the reduction of NMDA channel activity fol-
lowing DRB was rapid and reversible, in contrast to the long half-
life of dephosphorylation reported for some CKII substrates in
cell-free preparations15,16. To corroborate our findings with
DRB, we then investigated the effects of other compounds, unre-
lated to DRB, that are known to affect CKII activity.

126 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

Table 1. Effects of various protein kinase inhibitors and activators on NMDA channel openings in cell-attached and inside-
out recordings.
Mean open Burst Total Cluster Total Supercluster
Cell-attached time length open/burst length open/cluster length n
KN-62 92 7 89 7 85 6 95 26 74 6 107 23 6
DRB 75 4* 60 5* 67 4* 54 6* 57 5* 55 9* 6
FK-506 + DRB 95 9 89 10 90 9 83 12 87 6 78 19 3
Chrysin 77 7* 70 6* 71 7* 52 7* 57 8* 67 13 5
Spermine 153 30* 257 56* 195 51* 238 61* 290 81* 180 49 6
Spermine + DRB 108 13 102 3 101 10 92 10 93 17 61 16 4
Phorbol ester (PDBu) 103 4 106 6 105 6 99 7 104 7 115 22 8
H-7 106 8 104 7 97 9 102 8 93 12 107 13 4
Sp-8-CPT-cAMPS 102 6 105 14 113 10 98 13 112 17 112 28 8
Rp-8-CPT-cAMPS 119 25 117 28 119 25 123 25 131 23 101 26 5

Inside-out
1999 Nature America Inc. http://neurosci.nature.com

Casein kinase II 174 17* 174 23* 183 25* 179 22* 202 32* 183 15* 9
DRB 71 14* 61 8* 62 9* 49 11 47 6* 67 25 3
ATP + Heparin 66 7* 57 4* 58 5* 60 10* 54 7* 59 10 6
BAPTA 157 26* 169 33* 188 45 147 20 229 61* 223 83 4
GTP vs ATP 136 11* 120 15 120 15 130 25 150 36 91 17 10
GTP + Heparin 70 15* 58 14* 58 16* 54 7* 47 7 57 17 3
Spermine 126 6* 147 14* 149 11* 177 39* 176 26* 167 45 4
The values represent the average standard error percent of control responses under the various experimental conditions. The bold typeface and * denote
significant (p < 0.05, two-tailed paired t-tests) differences in pair-wise comparisons of the opening characteristics in each cell. The n represents the number of
recordings with at least 5 minutes of NMDA channel activity (average > 2000 openings) during each control and experimental period. None of the treatments
had any significant effect on single-channel conductance.

Endogenous cellular regulators of CKII activity include the
polyamines spermine and spermidine, which markedly
increase the activity of this protein kinase15,16. These
polyamines also regulate NMDA channel function2124.
Whereas early studies questioned whether the site of spermine
modulation of NMDA channel activity was intracellular or
extracellular, more recent studies have shown that spermine
enhances NMDA channel activity by interacting with its N-
terminal extracellular domain2224. It is important to note,
however, that the vast majority of spermine in the brain
resides intracellularly25, where it may act as a second messen-
ger. To investigate this possibility, we examined the potential
of intracellular spermine to physiologically activate CKII and
thus enhance NMDA channel activity.
To prevent direct interaction between spermine and the
extracellular domain of NMDA channels under the patch
pipettes, we applied spermine extracellularly during cell-
attached recordings (Fig. 2). Thus, spermine would have to
enter the cytoplasm via polyamine transporters26,27, other glu-
tamate receptors28 or by other means before altering the func-
tion of the NMDA receptor channel complex. Within 510 min
following addition of spermine (200500 M) to the extracel-
lular solution, NMDA channel open times were significantly
prolonged (Fig. 2a and b, n = 6; Table 1). This experiment sug-
gests that spermine must activate an intracellular pathway that
controls NMDA channel activity.
Next, we pretreated cells with DRB to determine whether
CKII mediates the actions of spermine on the NMDA channel
(Fig. 2c). Inhibiting CKII with DRB (50100 M, n = 4,
Table 1) prevented the prolonging effect of spermine on NMDA
channel open times, suggesting that this augmentation
occurred through a CKII mediated pathway. In a separate series
of biochemical experiments, spermine was found to produce a
two-fold increase CKII activity in hippocampal slices (P.T.
Tuazon, J. Roig and J.A. Traugh, personal communication),
consistent with the enhancement of single NMDA channel
activity produced by spermine in our electrophysiological
experiments.
If CKII controls NMDA channel gating through phospho-
rylation, then inhibiting endogenous CKII should also prevent
blockers of the dephosphorylation process from prolonging
NMDA channel openings. In adult hippocampal neurons, the
protein phosphatase most effectively involved in the feedback
regulation of NMDA channel activity is the calcium-calmod-
ulin-dependent protein phosphatase calcineurin3. We applied
the calcineurin inhibitor FK-506 to cells pretreated with DRB
to examine whether CKII mediates the increase in NMDA
channel function when calcineurin-dependent dephosphory-
lation is eliminated. Indeed, pretreatment of dentate granule
cells with 60 M DRB prevented the characteristic prolonga-
tion of open times3 by the calcineurin inhibitor FK-506
(500 nM, n = 3, Table 1). Based on these findings, we conclude
that CKII is responsible for the enhancement of NMDA chan-
nel activity when the dephosphorylation process mediated by
calcineurin is inhibited.
The CKII inhibitor DRB is considered to be highly specific,
with no effect on PKC, PKA, CaMKII or the src tyrosine
kinase15,16. To ensure that the effect of DRB in granule cells was
not secondary to changing the activities of PKC5,6 or PKA7,8
(shown to alter NMDA channel function in various prepara-
tions), we used selective PKC and PKA activators and inhibitors
and examined their effects on single NMDA channel openings.
In our preparation, neither phorbol 12,13-dibutyrate (PDBu,
110 M, n = 8), an activator of endogenous PKC, nor the

nature neuroscience volume 2 no 2 february 1999 127

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 2. Extracellular spermine

increases NMDA channel activity in a Control Spermine

cell-attached patches. (a) Continuous

traces of cell-attached recordings in
the same cell before (left panel, mean
open time, 2.18 ms) and after (right
panel, mean open time, 3.94 ms) bath
application of 300 M spermine.
Scale bars, 2 pA/10 ms. (b) Open
dwell-time histograms for the record-
ings shown in (a). Each histogram
shows the frequency distributions of
log binned (9 bins per decade) shut b
and open times. The values on the
abscissae are plotted on a square- w = 116 ms w = 150 ms
root scale. The individual exponential
distributions of the dwell times are
Frequency (square-root scale)

indicated by thin lines. Each peak of

the distribution corresponds to a
time constant of the compound
1999 Nature America Inc. http://neurosci.nature.com

exponentials. The weighted mean

shut or open times (w, which also
include the dwell-times of missed
events) are indicated in each panel. Shut time (ms) Shut time (ms)
(c) Cumulative probability distribu-
tions of total open time per cluster
w = 1.4 ms w = 2.8 ms
before and after exposure of two
cells to spermine (300 m). The first
cell was recorded in the absence (left
panel) and the second cell in the pres-
ence (right panel) of the CKII
inhibitor DRB (100 m). The increase
in total open time in clusters found
with spermine alone (left panel,
p < 106, Kolmogorov-Smirnov test)
was prevented by DRB (right panel). Open time (ms) Open time (ms)
c
Cumulative probability

Cumulative probability

Total open/cluster (ms) Total open/cluster (ms)

broad spectrum kinase inhibitor H-7 (1020 M, n = 4) had
any effect on NMDA channel openings (Table 1). Furthermore,
the selective cell-permeable activator and inhibitor of PKA 8,
Sp-8-CPT-cAMPS (50100 M) and Rp-8-CPT-cAMPS
(50100 M) respectively, were also without effect on granule
cell NMDA channel openings (Table 1). Whereas these experi-
ments do not rule out a general role for PKC or PKA in regulat-
ing NMDA channel function, it is unlikely that in adult dentate
gyrus granule cells DRB could have produced its effect on
NMDA channels by inhibiting any of these protein kinases. To
corroborate our findings with DRB, we next tested the recently
described specific, but structurally unrelated, CKII inhibitor
chrysin29. Chrysin (50 M) reproduced the effects of DRB on
NMDA channel openings (Table 1), further supporting a role
for CKII in regulating the gating of NMDA channels.
In spite of their kinase specificities, DRB and chrysin may
have altered NMDA channel activity by nonspecific effects on
other CKII-mediated cellular functions, including RNA tran-
scription and translation19,20. Because CKII is also found in the
nucleus15,16,18, to exclude possible nuclear targets of CKII, we
used excised inside-out patch recordings maintained with ATP
on the cytosolic side, where application of DRB proved to be
equally effective in reducing NMDA channel openings (n = 3,
Table 1). This finding is consistent with CKII being closely asso-
ciated with the plasma membrane and/or the NMDA receptor
complex, and allowed us to undertake a more rigorous exami-
nation of CKII effects in inside-out recordings.
To further test the hypothesis that intracellular spermine can
enhance NMDA channel activity through CKII, we directly
applied spermine to the cytosolic side of inside-out membrane

128 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 3. The effects of activation or inhibi-

tion of CKII on NMDA channel activity a

Cumulative probability
recorded in inside-out patches. (a) Left

Mean open time (ms)

panel, mean open time calculated for 10-s
bins is plotted against the recording time.
The patch was excised 80 s after beginning
the recording, was subsequently trans-
ferred into a well containing GTP, and was
washed with cytosolic media before it was
exposed to ATP. Right panel, cumulative
distribution of mean open times from indi- Time (min) Mean open time (ms)
vidual experiments. Addition of 2 mM
GTP (n = 10) or 4 mM ATP (n = 18) to the b
cytosolic side of inside-out patches signifi-
cantly (p < 0.05, Students t -test)
Cumulative probability

Cumulative probability
increased the openings of NMDA chan-
nels compared to those in cell-attached or
freshly excised patches. In the absence of
exogenous ATP or GTP, channel activity
invariably runs down (data not shown).
1999 Nature America Inc. http://neurosci.nature.com

(b) Heparin (1020 g/ml) perfused onto

the cytosolic side reduces the open time
within a cluster of channel openings (mea-
sured 5 minutes after the addition of
heparin). Six minutes after its addition to Total open/cluster (ms) Total cluster length (ms)
the bath, spermine (200 M) applied to
the cytosolic face of the patch increases c
cluster duration. (c) Raw current traces
(left panel) before and after addition of
CKII (200 U/ml) to the cytosolic side of

Mean open time (ms)

the patch. Scale bars, 2 pA/10 ms. CKII
rapidly prolongs single NMDA channel
openings (right panel) when added to an
inside-out patch. Both human recombi-
nant holoenzyme and purified rabbit brain
CKII holoenzyme were used. The addition
artifact at seven minutes is blanked out.

Time (min)

patches containing the NMDA receptor. In the presence of ATP,
application of exogenous spermine (200 M) to the cytoplasmic
side, significantly prolonged NMDA channel openings (n=5,
Fig. 3b and Table 1). Thus, in addition to controlling ion flow
through inwardly rectifying potassium channels30 and certain
non-NMDA glutamate receptors,31,32 intracellular spermine
effectively regulates NMDA receptor function through activa-
tion of CKII. These findings do not contradict the known
actions of spermine on the extracellular domains of NMDA
receptors2224, but reveal an additional intracellular mecha-
nism33,34 possibly responsible for effects of extracellularly
applied spermine.
The potentiating effect of cytosolic spermine on NMDA
channel activity was mimicked by direct exposure of the
patches to purified or recombinant CKII (n = 9, Fig. 3c;
Table 1), but not by perfusion of vehicle or of heat-inactivated
enzyme (n = 4, data not shown). The rapid prolongation of
mean channel open time by addition of the holoenzyme form
of the kinase suggests a direct phosphorylation of the NMDA
receptor channel complex. Furthermore, this increase in
NMDA channel activity produced by exogenous CKII suggests
that phosphorylation by endogenous CKII is not fully satu-
rated, allowing for a CKII-dependent dynamic control of
NMDA channel function.
To better understand the role of CKII in the regulation of
NMDA channel openings, we designed experiments to address
several unique physiological and pharmacological properties of
CKII. Unlike most Ser/Thr protein kinases, CKII can use high-
energy phosphate groups provided by either ATP or GTP15,16.
We tested this characteristic of CKII in inside-out patches,
where GTP (n = 10) was used to substitute for ATP (n = 18) in
maintaining NMDA channel openings (Fig. 3a, Table 1). All
kinetic parameters but the mean open time (Fig. 3a, Table 1)
were comparable when NMDA channel activity was supported
by GTP, rather than by ATP. The 36% longer mean open time in
the presence of GTP versus ATP could be due to a more effective
utilization of GTP by CKII, or could result from the involve-
ment of some additional pathway linked to GTP hydrolysis.
If tonic CKII-dependent phosphorylation regulates NMDA
channel activity, then the effect of DRB on NMDA channel
openings should be reproduced by heparin, the classic
inhibitor of CKII in biochemical studies15,16. Indeed, in the
presence of ATP, heparin (1020 g/ml) perfused onto the
cytosolic side of inside-out patches dramatically reduced
NMDA channel open durations (n = 6, Fig. 3b; Table 1). When
the experiments were repeated in the presence of GTP instead
of ATP, heparin shortened the channel open duration to the
same extent (n = 3, Table 1). The lack of significant difference

nature neuroscience volume 2 no 2 february 1999 129

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 4. Inhibition of CKII selectively reduces the NMDA

component of synaptic transmission in the dentate gyrus. a EPSPAMPA EPSPNMDA

(a) Superimposed traces of field EPSPs recorded in the mol-

ecular layer show no effect of 100 M DRB on AMPA-
receptor-mediated EPSPs (EPSPA, recorded in 50 M
D-APV), whereas NMDA-receptor-mediated EPSPs (EPSPN,
recorded in 15 M CNQX) are significantly reduced by
DRB in a fully reversible fashion. (b) The graph shows the
time course of NMDA EPSP inhibition by 50 (n = 5) and 100
M DRB (n = 6) applied for 20 min (during the bar), without
alteration in AMPA EPSPs (50 M DRB, n = 6, and 100 M b
DRB, n = 7). In each experiment, four sweeps were averaged
at each timepoint. The graphs represent percent alterations
DRB
in the normalized mean EPSP area ( standard error) as a
function of time. (c) Chrysin (50 M; applied for 30 min

Percent of baseline
indicated by the bar), a selective CKII inhibitor structurally
unrelated to DRB, shows the same selectivity for gluta-
matergic synaptic transmission as DRB (EPSPN, n = 6, EPSPA, p EPSPN 50 M DRB
n = 5). Percentage changes ( standard error) from normal- l EPSPA 50 M DRB
ized baseline responses recorded during the control period P EPSPN 100 M DRB
1999 Nature America Inc. http://neurosci.nature.com

are plotted as in (b). L EPSPA 100 M DRB

between heparins actions on NMDA channel activ- c

Time (min)

ity in the presence of ATP or GTP excludes the possi-

bility of an effect through inhibition of a 50 M Chrysin
G-protein35. More importantly, heparin is known
Percent of baseline

for its potent antagonism of calcium release from

inositol-tris-phosphate (IP3)-sensitive cellular cal-
cium stores36, and the attachment of such cytoplas-
mic organelles to the inside of membrane patches
cannot be entirely ruled out37. Even if IP3-sensitive
calcium stores were present in our inside-out
patches, a heparin-mediated reduction of calcium
release on the cytosolic side would be expected to
increase NMDA channel open duration. Indeed,
NMDA channel openings were significantly pro-
longed by the calcium chelator BAPTA (15 mM),
added to the intracellular side of inside-out patches
(n = 4; Table 1). This finding is expected if one assumes that
activation of calcineurin is prevented following the high-affin-
ity binding of calcium to BAPTA.
Although our data clearly show that dentate granule cell
NMDA channel openings are regulated by CKII, it needs to be
demonstrated that CKII also regulates the function of synaptic
NMDA receptors. We thus examined NMDA receptor-medi-
ated synaptic transmission in hippocampal slices. In the dentate
gyrus, the CKII inhibitor DRB reversibly inhibited synaptically
elicited NMDA-receptor-mediated field responses (50 M
DRB, n = 5; 100 M DRB, n = 6). There were no concomitant
changes in AMPA-receptor-mediated EPSPs (50 M DRB,
n = 6, 100 M DRB, n = 7; Fig. 4a and b). The unaltered AMPA-
receptor-mediated synaptic events in the presence of these con-
centrations of DRB38 indicate a lack of DRB effect on glutamate
release despite possible presynaptic targets of CKII-mediated
phosphorylation26,39,40. The CKII-dependent modulation of
synaptic NMDA receptors is not restricted to the dentate gyrus
because NMDA (n = 4), but not AMPA responses (n = 3), were
significantly reduced by CKII inhibition in the CA1 region of
the hippocampus (data not shown). We further examined the
specificity of CKII inhibition for NMDA responses using
chrysin (n = 5). Like DRB, chrysin (50 M) selectively dimin-
ished NMDA (n = 6) but not AMPA (n = 5) mediated field
EPSPs (Fig. 4c). To eliminate the possibility that both CKII
L EPSPA
P EPSPN
Time (min)
antagonists exerted their effects by altering RNA transcrip-
tion19,20,29, we also tested the effect of actinomycin D
(3060 M), a known transcription inhibitor, on NMDA-
receptor-mediated synaptic responses. Actinomycin D had no
effect on NMDA-receptor-mediated synaptic transmission
(n = 5; data not shown), consistent with a specific reduction in
CKII activity by DRB and chrysin.
DISCUSSION
Casein kinase II (CKII) is a ubiquitously distributed calcium-
and cyclic-nucleotide-insensitive Ser/Thr protein kinase, highly
concentrated in the brain, where its function is virtually
unknown. Here we have demonstrated a new action of CKII in
neurons: the selective regulation of NMDA channel gating,
including the prolongation of channel openings when dephos-
phorylation by calcineurin is prevented3. According to our data,
the phosphorylation state of the NMDA receptor channel com-
plex in adult hippocampal neurons is set by a balance between
the tonic activity of the spermine-sensitive CKII and the calci-
um-mediated negative feedback provided by calcineurin. Known
biochemical properties of CKII, including its inhibition by
heparin, DRB and chrysin, potentiation by spermine and hydrol-
ysis of either GTP or ATP, were used to demonstrate the role of
this kinase in the regulation of NMDA channel gating.
Previous studies may have failed to find any effect of intracel-

130 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
lular spermine on NMDA channel function22 because the
absence of intracellular ATP or GTP in those recordings most
likely precluded the ability of spermine-activated CKII to phos-
phorylate the NMDA receptor channel complex. Judged by the
robust effect of CKII on NMDA channel function, however, the
regulation of CKII by spermine may have profound functional
consequences on NMDA-receptor-mediated events during
physiological and pathological alterations in intracellular sper-
mine levels. Following ischemic insults to the brain, polyamines
are released into the extracellular space, spermine synthesis is
stimulated, and active cellular uptake of spermine is increased in
the recovery period41,42. Thus, increased CKII activity resulting
from increased intracellular levels of spermine could contribute
to the anoxia-induced enhancement of NMDA responses.
In central neurons, CKII may modulate the activity of sev-
eral key target proteins, including calmodulin, MAP 1B, -
tubulin, tau, the regulatory subunit of type II PKA, protein
phosphatase inhibitor-2, DARP-32 (refs 1518) and CKII itself
1999 Nature America Inc. http://neurosci.nature.com
through autophosphorylation of its -subunit1518. Whereas
PKC, PKA, CaMKII and the src tyrosine kinase seem to alter
both NMDA and AMPA receptors1,2,510, CKII seems to selec-
tively control only the function of NMDA receptors. It was
beyond the scope of the present study to determine which, if
any, of the numerous consensus CKII phosphorylation sites on
NMDA receptor subunits are phosphorylated by CKII.
However, a direct phosphorylation of NMDA channels by CKII
need not be postulated. Indirect effects of CKII on NMDA
channels may include the phosphorylation of calmodulin43,
leading to a change in the direct regulatory effect exerted by the
calmodulin bound to NMDA channels44.
Spermine may be one of many endogenous regulators of
CKII activity in neurons. Other means of CKII activation may
bear on short- and long-term alterations of neuronal excitabil-
ity. For example, high-frequency synaptic activity leads to the
transient elevation of CKII activity during induction of LTP in
an NMDA-receptor-dependent manner45, and neurotrophins
may elevate CKII activity through the activation of other pro-
tein kinases46. Furthermore, the regulation of NMDA channel
function described in our study may be complemented by a
CKII-dependent regulation of biochemical events in neurons.
CKII seems to control the nuclear translocation of the tran-
scription factor NF--B via I-B phosphorylation47 and other
proteins through phosphorylation-regulated nuclear localiza-
tion sequences48. These latter findings are consistent with a role
for CKII as a second messenger relaying the consequences of
glutamate receptor activation to the nucleus49. In summary,
our results identify the NMDA receptor channel complex as a
constitutively phosphorylated target of CKII, and reveal a selec-
tive means of regulating the gain of excitatory synaptic trans-
mission in the hippocampus.
METHODS
Cell-attached recordings. Dentate granule cells from adult rats were
acutely dissociated as described50. Cell-attached recordings were done at
room temperature in a bath of 110 mM Na2SO4, 5 mM Cs2SO4, 10 mM
HEPES, 10 mM D-glucose, 1.8 mM CaCl2, 1 mM pyruvic acid and 1 M
tetrodotoxin (TTX, final pH 7.25, osmolarity 285305 mOsm). Fire-pol-
ished, Sylgard-coated borosilicate glass (Garner Glass, Claremont, Cali-
fornia) pipettes (815 M) were filled with 200500 nM L-aspartic acid
and 8 M glycine dissolved in the bath solution. Channel activity was
sampled at 20 kHz and filtered at 2 kHz for analysis and display of
records. Details of the single-channel analysis using the PAT program
(courtesy of J. Dempster) with a 50% threshold crossing algorithm for
event detection are given elsewhere50.
nature neuroscience volume 2 no 2 february 1999
articles
Inside-out patch recordings. Inside-out patches were excised in a bath
of 120 mM NaCl, 5 mM CsCl, 10 mM HEPES, 10 mM D-glucose, 1 mM
pyruvic acid and 1 M TTX with no added calcium. Fire-polished and
Sylgard-coated borosilicate glass pipettes (610 M) were filled with
the sulfate-containing extracellular solution3,50, including 1.8 mM CaCl2,
200500 nM L-aspartic acid and 8 mM glycine. The patches were moved
to a well containing 140 mM Csgluconate, 10 mM HEPES, 1 mM MgCl2,
150 nM CaCl2 and either 4 mM Mg-ATP or 2 mM GTP and were con-
tinually perfused with this solution.
Extracellular recordings. Coronal or horizontal brain slices (400 m
thick) were prepared from rats of similar age to those used in the single-
channel experiments. Slices were submerged at 35oC in a recording cham-
ber and constantly perfused with a ACSF containing 126 mM NaCl, 2.5
mM KCl, 1.25 mM NaH2PO4, 2 mM CaCl2, 0.3 mM MgCl2, 26 mM
NaHCO3 and 10 mM glucose in an atmosphere of humidified 95% O2,
5% CO2. Constant current stimuli were delivered to either the perforant
path or the Schaffer collaterals every 30 s via a bipolar stimulating elec-
trode, and extracellular field EPSPs were recorded in stratum
moleculare in the dentate and stratum radiatum of CA1. AMPA-receptor-
mediated EPSP areas were measured in the presence of 50 M D-APV,
whereas the areas of NMDA-receptor-mediated EPSPs were examined
in the presence of 15 M CNQX. Field EPSP area rather than slope was
measured to take into account the long time course of NMDA-receptor-
mediated synaptic potentials. To prevent possible effects of CKII
inhibitors on GABA-receptor-mediated inhibition, 50 M picrotoxin
was included in some experiments.
As a positive control for the phorbolester used in our cell-attached
recordings, PDBu was bath applied to hippocampal slices, producing a
clear augmentation of AMPA-receptor-mediated synaptic transmission
in extracellular recordings. The ability of Sp-8-CPT-cAMPS to regulate
neuronal function was examined in slice recordings, where this com-
pound both enhanced AMPA-receptor-mediated field EPSPs, presum-
ably through a presynaptic action, and reduced GABA-receptor-mediated
IPSCs (P. Poisbeau and I. Mody, unpublished observations).
ACKNOWLEDGEMENTS
We thank Jolinda Traugh for the gift of purified CKII, for the sharing of her
biochemical findings, and for her comments on the manuscript. We thank
Walter Pyerin for reviewing an earlier version of the manuscript, Georg Khr for
his early work with phorbol esters on NMDA channels and Brian Oyama for
technical assistance. This work was supported by a Howard Hughes predoctoral
fellowship to D.N.L and by the NINDS grant NS 27528 and the Coelho
Endowment to I.M.
RECEIVED 1 OCTOBER; ACCEPTED 16 DECEMBER 1998
1. Roche, K. W., Tingley, W. G. & Huganir, R. L. Glutamate receptor
phosphorylation and synaptic plasticity. Curr. Opin. Neurobiol. 4, 383388
(1994).
2. Moss, S. J. & Smart, T. G. Modulation of amino acid-gated ion channels by
protein phosphorylation. Int. Rev. Neurobiol. 39, 152 (1996).
3. Lieberman, D. N. & Mody, I. Regulation of NMDA channel function by
endogenous Ca2+- dependent phosphatase. Nature 369, 235239 (1994).
4. Tong, G., Shepherd, D. & Jahr, C. E. Synaptic desensitization of NMDA
receptors by calcineurin. Science 267, 15101512 (1995).
5. Chen, L. & Huang, L. Y. Protein kinase C reduces Mg2+ block of NMDA-
receptor channels as a mechanism of modulation. Nature 356, 521523
(1992).
6. Durand, G. M., Bennett, M. V. L. & Zukin, R. S. Splice variants of the N-
methyl-D-aspartate receptor NR1 identify domains involved in regulation by
polyamines and protein kinase C. Proc. Natl. Acad. Sci. USA 90, 67316735
(1993).
7. Greengard, P., Jen, J., Nairn, A. C. & Stevens, C. F. Enhancement of the
glutamate response by cAMP dependent protein kinase in hippocampal
neurons. Science 253, 11351138 (1991).
8. Raman, I. M., Tong, G. & Jahr, C. E. -adrenergic regulation of synaptic
NMDA receptors by cAMP- dependent protein kinase. Neuron 16, 415421
(1996).
9. Lu, Y. M., Roder, J. C., Davidow, J. & Salter, M. W. Src activation in the
induction of long-term potentiation in CA1 hippocampal neurons. Science
279, 13631367 (1998).
131
 1999 Nature America Inc. http://neurosci.nature.com
articles
10. Kolaj, M., Cerne, R., Cheng, G., Brickey, D. A. & Randic, M. Alpha subunit of
calcium/calmodulin-dependent protein kinase enhances excitatory amino
acid and synaptic responses of rat spinal dorsal horn neurons. J. Neurophysiol.
72, 25252531 (1994).
11. Omkumar, R. V., Kiely, M. J., Rosenstein, A. J., Min, K. T. & Kennedy, M. B.
Identification of a phosphorylation site for calcium/calmodulindependent
protein kinase II in the NR2B subunit of the N-methylDaspartate receptor.
J. Biol. Chem. 271, 3167031678 (1996).
12. Goto, S. et al. Cellular colocalization of calcium/calmodulin-dependent
protein kinase II and calcineurin in the rat cerebral cortex and hippocampus.
Neurosci. Lett. 149, 189192 (1993).
13. Ishikawa, N., Hashiba, Y. & Hidaka, H. Effect of a new Ca2+-calmodulin-
dependent protein kinase II inhibitor on GABA release in cerebrospinal fluid
of the rat. J. Pharmacol. Exp. Ther. 254, 598602 (1990).
14. Li, G., Hidaka, H. & Wollheim, C. B. Inhibition of voltage-gated Ca2+
channels and insulin secretion in HIT cells by the Ca2+/calmodulin-
dependent protein kinase II inhibitor KN-62: comparison with antagonists of
calmodulin and L- type Ca2+ channels. Mol. Pharmacol. 42, 489488 (1992).
15. Pinna, L. A. Casein kinase 2: an eminence grise in cellular regulation?
Biochim. Biophys. Acta 1054, 267284 (1990).
16. Tuazon, P. T. & Traugh, J. A. Casein kinase I and IImultipotential serine
protein kinases: structure, function, and regulation. Adv. Second Messenger
Phosphoprotein Res. 23, 123164 (1991).
17. Girault, J. A., Hemmings, H. C. J., Zorn, S. H., Gustafson, E. L. & Greengard,
1999 Nature America Inc. http://neurosci.nature.com
P. Characterization in mammalian brain of a DARPP-32 serine kinase
identical to casein kinase II. J. Neurochem. 55, 17721783 (1990).
18. Krek, W., Maridor, G. & Nigg, E. A. Casein kinase II is a predominantly
nuclear enzyme. J. Cell Biol. 116, 4355 (1992).
19. Zandomeni, R. & Weinmann, R. Inhibitory effect of 5,6-dichloro-1-beta-D-
ribofuranosylbenzimidazole on a protein kinase. J. Biol. Chem. 259,
1480414811 (1984).
20. Zandomeni, R., Zandomeni, M. C., Shugar, D. & Weinmann, R. Casein
kinase type II is involved in the inhibition by 5,6-dichloro-1-beta-D-
ribofuranosylbenzimidazole of specific RNA polymerase II transcription. J.
Biol. Chem. 261, 34143419 (1986).
21. Williams, K., Romano, C. & Molinoff, P. B. Effects of polyamines on the
binding of [3H]MK-801 to the N- methyl-D-aspartate receptor:
pharmacological evidence for the existence of a polyamine recognition site.
Mol. Pharmacol. 36, 575581 (1989).
22. Lerma, J. Spermine regulates N-methyl-D-aspartate receptor desensitization.
Neuron 8, 343352 (1992).
23. Benveniste, M. & Mayer, M. L. Multiple effects of spermine on N-methyl-D-
aspartic acid receptor responses of rat cultured hippocampal neurones. J.
Physiol. (Lond.) 464, 131163 (1993).
24. Traynelis, S. F., Hartley, M. & Heinemann, S. F. Control of proton sensitivity
of the NMDA receptor by RNA splicing and polyamines. Science 268,
873876 (1995).
25. Shaw, G. G. & Pateman, A. J. The regional distribution of the polyamines
spermidine and spermine in brain. J. Neurochem. 20, 12251230 (1973).
26. Harman, R. J. & Shaw, G. G. High-affinity uptake of spermine by slices of rat
cerebral cortex. J. Neurochem. 36, 16091615 (1981).
27. Gilad, G. M. & Gilad, V. H. Polyamine uptake, binding and release in rat
brain. Eur. J. Pharmacol. 193, 4146 (1991).
28. Bhring, R., Bowie, D., Benveniste, M. & Mayer, M. L. Permeation and block
of rat GluR6 glutamate receptor channels by internal and external
polyamines. J. Physiol. (Lond.) 502, 575589 (1997).
29. Critchfield, J. W., Coligan, J. E., Folks, T. M. & Butera, S. T. Casein kinase II is
a selective target of HIV 1 transcriptional inhibitors. Proc. Natl. Acad. Sci.
USA. 94, 61106115 (1997).
30. Lopatin, A. N., Makhina, E. N. & Nichols, C. G. Potassium channel block by
132
cytoplasmic polyamines as the mechanism of intrinsic rectification. Nature
372, 366369 (1994).
31. Bowie, D. & Mayer, M. L. Inward rectification of both AMPA and kainate
subtype glutamate receptors generated by polyamine-mediated ion channel
block. Neuron 15, 453462 (1995).
32. Kamboj, S. K., Swanson, G. T. & Cull-Candy, S. G. Intracellular spermine
confers rectification on rat calcium-permeable AMPA and kainate receptors.
J. Physiol. (Lond.) 486, 297303 (1995).
33. Herman, M. D., Reuveny, E. & Narahashi, T. The effect of polyamines on
voltage-activated calcium channels in mouse neuroblastoma cells. J. Physiol.
(Lond.) 462, 645-660 (1993).
34. Ye, C., Kanazirska, M., Quinn, S., Brown, E. M. & Vassilev, P. M. Modulation
by polycationic Ca(2+)-sensing receptor agonists of nonselective cation
channels in rat hippocampal neurons. Biochem. Biophys. Res. Commun. 224,
271280 (1996).
35. Willuweit, B. & Aktories, K. Heparin uncouples alpha 2-adrenoceptors from
the Gi-protein in membranes of human platelets. Biochem. J. 249, 857863
(1988).
36. Ehrlich, B. E., Kaftan, E., Bezprozvannaya, S. & Bezprozvanny, I. The
pharmacology of intracellular Ca2+-release channels. Trends Pharmacol. Sci.
15, 145149 (1994).
37. Sokabe, M. & Sachs, F. The structure and dynamics of patch-clamped
membranes: a study using differential interference contrast light microscopy.
J. Cell Biol. 111, 599606 (1990).
38. Nguyen, P. V., Abel, T. & Kandel, E. R. Requirement of a critical period of
transcription for induction of a late phase of LTP. Science 265, 11041107
(1994).
39. Apel, E. D., Litchfield, D. W., Clark, R. H., Krebs, E. G. & Storm, D. R.
Phosphorylation of neuromodulin (GAP-43) by casein kinase II.
Identification of phosphorylation sites and regulation by calmodulin. (Au: is
this title correct?). J. Biol. Chem. 266, 1054410551 (1991).
40. Bennett, M. K., Miller, K. G. & Scheller, R. H. Casein kinase II phosphorylates
the synaptic vesicle protein p65. J. Neurosci. 13, 17011707 (1993).
41. Gilad, G. M., Gilad, V. H. & Wyatt, R. J. Accumulation of exogenous
polyamines in gerbil brain after ischemia. Mol. Chem. Neuropathol. 18,
197210 (1993).
42. Johnson, T. D. Polyamines and cerebral ischemia. Prog. Drug Res. 50,
193258 (1998).
43. Sacks, D. B. et al. Phosphorylation by casein kinase II alters the biological
activity of calmodulin. Biochem. J. 283, 2124 (1992).
44. Ehlers, M. D., Zhang, S., Bernhardt, J. P. & Huganir, R. L. Inactivation of
NMDA receptors by direct interaction of calmodulin with the NR1 subunit
Cell 84, 745755 (1996).
45. Charriaut-Marlangue, C., Otani, S., Creuzet, C., Ben-Ari, Y. & Loeb, J. Rapid
activation of hippocampal casein kinase II during long- term potentiation.
Proc. Natl. Acad. Sci.USA. 88, 1023210236 (1991).
46. Blanquet, P. R. Neurotrophin-induced activation of casein kinase 2 in rat
hippocampal slices. Neuroscience 86, 739749 (1998).
47. Barroga, C. F., Stevenson, J. K., Schwarz, E. M. & Verma, I. M. Constitutive
phosphorylation of I kappa B alpha by casein kinase II. Proc. Natl. Acad. Sci.
USA. 92, 76377641 (1995).
48. Hbner, S., Xiao, C. Y. & Jans, D. A. The protein kinase CK2 site (Ser111/112)
enhances recognition of the simian virus 40 large T-antigen nuclear
localization sequence by importin. J. Biol. Chem. 272, 1719117195 (1997).
49. Kaltschmidt, C., Kaltschmidt, B. & Baeuerle, P. A. Stimulation of ionotropic
glutamate receptors activates transcription factor NF-kappa B in primary
neurons. Proc. Natl. Acad. Sci. USA. 92, 96189622 (1995).
50. Khr, G., De Koninck, Y. & Mody, I. Properties of NMDA receptor channels
in neurons acutely isolated from epileptic (kindled) rats. J. Neurosci. 13,
36123627 (1993).
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

Adrenaline enhances odorant

contrast by modulating signal
encoding in olfactory receptor cells
Fusao Kawai1,2, Takashi Kurahashi1,3,4 and Akimichi Kaneko1,5

1 Department of Information Physiology, National Institute for Physiological Sciences, Myodaiji, Okazaki 444-8585, Japan
2 Department of Neuroscience, University of Pennsylvania, Philadelphia, Pennsylvania 19104-6058, USA
3 Present address: Department of Biology, Osaka University, Toyonaka 560-0043, Japan
4 Present address: Precursory Research for Embryonic Science and Technology (PRESTO), Senri Life Science Center 10F,
Sinsenri, Toyonaka 565-0082, Japan
1999 Nature America Inc. http://neurosci.nature.com

5 Present address: Department of Physiology, Keio University School of Medicine, Shinjuku, Tokyo 160-8582, Japan
Correspondence should be addressed to T.K. (kurahasi@bio.sci.osaka-u.ac.jp)

Olfactory perception is influenced by hormones. Here we report that adrenaline can directly affect
the signal encoding of olfactory receptor cells. Application of adrenaline suppressed action
potentials near threshold and increased their frequency in response to strong stimuli, resulting in a
narrower dynamic range. Under voltage-clamp conditions, adrenaline enhanced sodium current and
reduced T-type calcium current. Because sodium current is the major component of spike generation
and T-type calcium current lowers the threshold in olfactory receptor cells, the effects of adrenaline
on these currents are consistent with the results obtained under current-clamp conditions. Both
effects involved a common cytoplasmic pathway, cAMP-dependent phosphorylation. We suggest
that adrenaline may enhance contrast in olfactory perception by this mechanism.

Olfactory perception, which starts in the olfactory receptor cells
(ORCs) of the olfactory epithelium17, is influenced by behav-
ioral states, presumably via hormonal regulation8. Cellular olfac-
tory sensitivities are known to be influenced by hormones. The
olfactory mucosa is innervated by adrenergic fibers911 and is
strongly immunoreactive for the -adrenergic receptor12. Sym-
pathetic nerve stimulation, which releases noradrenaline, increas-
es electrical activity in the olfactory nerve13. Adrenaline also
enhances the amplitude of the electro-olfactogram induced by
an odorant14. Because these experiments were done in olfactory
tissues containing many types of cells, however, the actual target
of adrenaline has not yet been identified.
Because olfactory cilia, the cellular site of olfactory transduc-
tion17, are exposed to the external environment, a widely accept-
ed hypothesis is that the adrenergic fibers do not affect the
transduction machinery at the receptor cell level8, but instead
regulate the secretion of mucus covering the olfactory epitheli-
um. Thick mucus would obstruct odorants from the ciliary mem-
brane of ORCs912. Here we investigated whether adrenaline
directly affects olfactory cells, focusing on electrical activity in
the cell body and axon, which are exposed to the interstitial solu-
tion. Single newt olfactory receptor cells were dissociated from
the olfactory epithelium, and their inputoutput relationship
was examined by whole-cell patch clamp. Because the site where
adrenaline could affect the olfactory cell is assumed to be the den-
drosomatic membrane, we mainly focused on cell body activity
and resulting action potential generation. Application of adren-
aline narrowed the dynamic range of spike generation, which
increased the slope of the inputoutput relationship. In general,
changing the slope of the intensityresponse curve changes the
sensory contrast. We therefore conclude that adrenaline enhances
odorant contrast in olfactory perception.
RESULTS
Adrenaline modulates spike generation
We examined the effect of adrenaline on spike generation in iso-
lated olfactory cells. The olfactory receptor potential was mim-
icked by injecting an extrinsic current from the recording pipette.
The current amplitudes were chosen to correspond to ORC
responses to a typical range of odor concentrations. There are
several advantages in using current injection over the natural
odorant stimuli. First, in isolated ORCs, odorants have secondary
effects15 of nonselectively suppressing voltage-gated ionic chan-
nels and therefore action potentials16, which do not occur with
current injection. Second, one can regulate the amount of depo-
larizing current precisely, whereas cell stimulation by odorant
molecules is difficult to control. Because the main purpose of
our study was to examine the effect of adrenaline on the spike
generation at the cell body, we mainly used ORCs that had lost
their cilia.
Bath-applied adrenaline had complex, reversible effects on
spike generation. Under weakly stimulated conditions (thresh-
old level, Fig. 1a), adrenaline suppressed spike generation. In
contrast, for strong stimuli causing repetitive spikes, adrenaline
increased spike frequency (Fig. 1b). In the presence of adrena-
line, the stimulusresponse relationship became steeper, with a
narrower dynamic range (Fig. 1c). Mean slopes between 6 and
9 pA were 0.4 spikes/secondpA in the control and 1.9 spikes/sec-
ondpA in the presence of adrenaline. Adrenaline thus amplified
the mean slope by approximately fivefold.

nature neuroscience volume 2 no 2 february 1999 133

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 1. Adrenaline narrows the

dynamic range of spike fre- a b c
quency in isolated ORCs.
(a) Responses to near-thresh-

Impulses/s
old depolarization induced by

mV
mV
injection of 5-pA current,
recorded in control Ringers
solution (thin line) or 3 minutes
after addition of 10 M adrena-
line (thick line). The recording
pipettes were filled with potas-
sium solution. (b) Responses
to a depolarization induced by injection of 7-pA current, recorded from the same cell as (a). (c) The relationship between current injection and spike
frequency in control solution (filled circles) and 10 M adrenaline (filled triangles). Each symbol represents the mean of five cells standard error.

Effects of adrenaline on INa and ICa,T and a Hill coefficient of 1.1. The activation curves of INa were
Spikes in ORCs are triggered by activation of sodium (INa)1722 fitted by a single Boltzmann function with a half-activation volt-
and T-type calcium (ICa,T) currents21,22. To understand the mech- age of 37 mV in control solution (Fig. 2e), as previously report-
1999 Nature America Inc. http://neurosci.nature.com

anism underlying modulation of spikes by adrenaline, we exam-
ined its effects on INa and ICa,T under voltage clamp. Adrenaline
(10 M) increased the peak amplitude of I Na by 17 4%
(mean standard error, n = 5) within 150200 seconds of appli-
cation in all ORCs recorded and accelerated both activation and
inactivation (Fig. 2a). The effect was reversible (Fig. 2b) and was
evident between 70 mV and 40 mV (Fig. 2c). Similar aug-
mentation of INa (18 5%, n = 4) was obtained from ORCs with
their cilia attached, suggesting that the effect is mainly due to
adrenergic receptors on the soma rather than on the cilia.
The effects of adrenaline on I Na were dose dependent
(Fig. 2d). The doseresponse curve was fitted by the Hill equa-
tion with a half-augmenting concentration (EC50) of 290 nM
ed 17 . Adrenaline increased the relative conductance (g Na )
between 70 mV and 40 mV (half-activation voltage, 41 mV).
In contrast, adrenaline did not change the inactivation curve
significantly (Fig. 2e). These results suggest that the augmenta-
tion of INa by adrenaline was caused by a shift in the activation
curve toward a more negative voltage.
Adrenaline also reversibly reduced the peak amplitude of ICa,T
in all recorded ORCs by 33 7% (n = 4), but did not change the
activation and inactivation time courses (Fig. 3a and b). A sim-
ilar reduction (31 6%, n = 4) was observed in ORCs with the
cilia attached. The reduction occurred over a wide voltage range
(Fig. 3c). The effects of adrenaline on ICa,T were dose depen-
dent, with an IC 50 of 150 nM and a Hill coefficient of 1.1

a b c
Relative amplitude
pA

Fig. 2. Adrenaline augments INa in isolated

ORCs. (a) INa induced by depolarization to 40 d e
mV (holding potential, Vh = 100 mV) in control
Relative conductance

solution (thin line) or 10 M adrenaline (thick

Relative amplitude

line) and after washout (dotted line). These

responses were recorded at 0 s (control), 180 s
(adrenaline) and 600 s (wash) of the experiment
shown in (b). The recording pipette was filled
with cesium solution. (b) Normalized ampli-
tudes of INa plotted against time. INa was
recorded every 30 s. (c) IV relationship of the
cell shown in (a). Peak INas in control solution
(filled circles) or in 10 M adrenaline (filled tri- Concentration (M) mV
angles) were plotted against test-pulse voltage.
(d) Relationship between the augmentation of
INa and adrenaline concentration (n = 5). The line represents the Hill equation obtained by least-squares nonlinear fit to the data. (e) Effects of
10 M adrenaline on the activation and inactivation curves of INa (n = 5). Vh for activation curves was 100 mV. Test voltage for inactivation curves
after a 1-s conditioning pulse was 0 mV. Lines represent a single Boltzmann function obtained by least-squares nonlinear fit to the data.

134 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
a b
Relative amplitude
pA
Fig. 3. Adrenaline inhibits ICa,T in isolated ORCs. d e
(a) ICa,T induced by depolarization to 40 mV (Vh
= 100 mV) in control solution (thin line), in 10 M
Relative amplitude
adrenaline (thick line) and after washout (dotted
line). These responses were recorded at 0 s (con-
1999 Nature America Inc. http://neurosci.nature.com
trol), 210 s (adrenaline) and 600 s (wash) of the
experiment shown in (b). (b) Time course of nor-
malized amplitudes of ICa,T, recorded every 30 s.
(c) Effects of 10 M adrenaline on the IV relation-
ship of the cell shown in (a). (d) Doseresponse
curve of ICa,T (n = 4). (e) Activation and inactiva-
tion curves of ICa,T (n = 4).
Concentration (M)
(Fig. 3d). Adrenaline did not change activation and inactivation
curves significantly (Fig. 3e), suggesting that it does not mod-
ulate the voltage dependence of ICa,T. Thus, inhibition of ICa,T
by adrenaline results from a reduction in the total T-type calci-
um channel conductance.
Effects of cAMP on INa and ICa,T
Because the -adrenergic receptor uses cyclic adenosine
monophosphate (cAMP) as a second messenger in a wide variety
of tissues2325, we next examined the effect of cytoplasmic cAMP
on INa and ICa,T. Addition of a membrane-permeable cAMP ana-
logue (1 mM 8-bromo-cAMP; note that the increase in cyto-
plasmic cAMP concentration was lower than this value) to the
superfusate increased the peak amplitude of I Na by 28 5%
(n = 6; Fig. 4a). This value was larger than that obtained with
10 M adrenaline (17 4%, Fig. 2d). This difference might be
due to the high concentration of 8-bromo-cAMP (1 mM) used.
The augmentation was observed between 70 mV and 40 mV
(Fig. 4b) and was reversible. The effect of 8-bromo-cAMP on INa
was dose dependent; the doseresponse curve was fitted by the
Hill equation with an EC50 of 75 M and a Hill coefficient of 1.5.
Cyclic AMP applied directly to the cytoplasm from the patch
pipette also increased INa (121 7%, n = 5). These results are
consistent with previous reports using other preparations from
several tissues2628. Furthermore, the effects of 8-bromo-cAMP
on the activation and inactivation curves of INa were consistent
with the effects of adrenaline. Application of 8-bromo-cAMP
strongly enhanced the activation curve between 70 mV and
40 mV (Fig. 4c).
A direct activator of adenylyl cyclase (10 M forskolin) added
to the bath also increased the amplitude of INa, by 34 6% (n = 5;
Fig. 5a), equivalent to the increase induced by cAMP. This result
suggests that cAMP was produced intracellularly by an adenylyl
cyclase in the dendrosomatic membrane of ORCs. To test whether
the enhancement of INa is mediated by a direct action of cAMP on
nature neuroscience volume 2 no 2 february 1999
articles
c
Relative conductance
mV
the sodium channel or by cAMP-dependent phosphorylation,
we used the catalytic subunit of the cAMP-dependent protein
kinase (PKA). After intracellular application of PKA by the
whole-cell pipette, INa increased by 37 8% (n = 5; Fig. 5b). In
addition, a PKA inhibitor blocked the effect of 8-bromo-cAMP
(n = 4; Fig. 5c), excluding the possibility of a direct action of
cAMP on the sodium channel. These results suggest that the
effect of cAMP on INa is mediated by PKA.
Application of 8-bromo-cAMP reduced the amplitude of ICa,T
over the entire voltage range examined (Fig. 4d and e). 8-bromo-
cAMP did not change the activation and inactivation curves of
ICa,T significantly (Fig. 4f), consistent with the effects of adrena-
line (Fig. 3e) and reports using other preparations29,30. The
doseresponse curve was fitted by the Hill equation with a half-
blocking concentration (IC50) of 11 M and a Hill coefficient of
1.8. This reduction of ICa,T is unlikely to be due to run-down,
because the 8-bromo-cAMP-induced reduction in current
(39 5%, 1 min after application) was significantly larger than
the reduction due to run-down (6 3%, at 10 min after rupture
of the patch membrane, n = 4).
When cAMP was introduced into the cell interior from the
patch pipette, ICa,T was also reduced. The amplitude of ICa,T was
decreased by 39 4% (n = 5). Extracellular forskolin also reduced
ICa,T by 37 5% (n = 5; Fig. 5d). Furthermore, application of
PKA via the patch pipette reduced I Ca,T by 34 5% (n = 4;
Fig. 5e), and the PKA inhibitor blocked the reduction caused by
8-bromo-cAMP (n = 4; Fig. 5f). These results suggest that the
effect of cAMP on ICa,T is also mediated by PKA.
DISCUSSION
In isolated ORCs under whole-cell, patch-clamp recording, we found
that adrenaline can directly affect spike encoding by modulating INa
and ICa,T via cAMP-dependent protein phosphorylation. Adrena-
line would presumably activate -adrenergic receptors on the soma,
which would shift the spike threshold to a higher voltage by sup-
135
 1999 Nature America Inc. http://neurosci.nature.com

articles

a b c

Relative conductance
pA

d e f

Relative conductance
pA
1999 Nature America Inc. http://neurosci.nature.com

Fig. 4. 8-bromo-cAMP augments INa and inhibits ICa,T. (a) INa induced by depolarization to 40 mV (Vh = 100 mV) in control solution (thin line) and
in 1 mM 8-bromo-cAMP (thick line). The recording pipette was filled with cesium solution. (b) Effects of 8-bromo-cAMP on the IV relationship of
INa. (c) Effects of 8-bromo-cAMP on the activation and inactivation curves of INa (n = 6). (d) Effects of 8-bromo-cAMP on ICa,T. (e) IV relationship
of ICa,T. (f) Effects of 8-bromo-cAMP on activation and inactivation curves of ICa,T (n = 6).

pressing ICa,T and increase the firing frequency in response to a strong
stimulus by enhancing INa. These two opposing effects on spike gen-
eration probably narrow the dynamic range of ORCs, but increase
the gain slope of signal encoding within this dynamic range. Con-
sequently, cells would be more sensitive to the difference between
the presence of an odor stimulus and its absence, because the ampli-
tude of the transduction current rises with increasing odor concen-
tration31. Under natural conditions, this is equivalent to improving
the cells ability to identify the presence of odorants.
Several findings in vivo suggest that similar effects may occur
under physiological conditions. Sympathetic nerve stimulation
increases electrical activity in the rabbit olfactory nerve13, and
adrenaline and noradrenaline enhance the amplitude of the elec-
tro-olfactogram (EOG) induced by an odorant14. Although the
EOG is thought to originate mainly from the receptor potential,
some fraction might originate from spike bursting. Here we
showed that adrenaline increases spike activity by enhancing the
amplitude of INa when strong current is injected into ORCs.
Because these previous studies used high concentrations of odor-
ant, the enhancement of EOG amplitude that they demonstrated
might be due to an increase in INa amplitude by adrenaline.
An alternative hypothesis is that adrenaline affects ORC
responses by regulating vasomotor tone and secretion from Bow-
mans glands, which modulate odorant access to and clearance
from the olfactory epithelium911,32. Adrenergic fibers are found
in human olfactory mucosa by immunohistochemistry9, and -
adrenoceptors are expressed in nasal glands as well as ORCs12.
Stimulation by the -adrenergic agonist isoproterenol causes
release of secretory granules from Bowmans glands in the dog33
and tiger salamander10. Although our data do not exclude this
possibility, adrenalin-induced modulation of activity at the cell
body of ORCs should be regarded as an important effect of adren-
aline, as a fivefold increase in the slope of intensityresponse
curve seems likely to be functionally significant.
Guanine nucleotides (GTP and GTPS) have been shown
to increase the amplitude of INa in the frog ORC34, suggesting
that activated G-proteins might affect the sodium channel
directly. Although our present data initially seem similar
because G-protein activation causes the cyclase to produce
intracellular cAMP, closer examination reveals an important
difference between our results and this previous work. Aug-
mentation of INa in our experiments was due to a shift of the
activation curve to a negative voltage, whereas in the previous
report, it was attributed to a shift of the inactivation curve to a
positive direction34. In addition, the authors did not include
Mg-ATP in their pipette, and therefore may not have been able
to see the phosphorylation process.
Odorant signal transduction is widely considered to be medi-
ated by cAMP at the cilia. One might therefore expect cross-talk
between cAMP produced at the cilia and cAMP produced by
adrenaline at the cell body. We estimate, however, that the cilia
and the cell body are independent in this regard. It is very unlike-
ly that cAMP produced by an odor stimulus in the cilia increas-
es cAMP concentration in the soma, because the total volume of
the cilia is only 0.18% of the cell body volume (five 30 m long
cylindrical cilia with 0.25 m diameter versus a spherical cell
body with 10 m diameter). Furthermore, phosphodiesterase
(PDE) activity in the cilia is so high that the cytoplasmic cAMP is
hydrolyzed quickly35. Therefore, cAMP induced by odor stimu-
lus in the cilia is unlikely to affect INa and ICa,T in the newt ORC.
This strong PDE activity would also form a boundary sep-
arating the cilia from the cell body, by breaking down cAMP
synthesized in the cell body by adrenaline. Indeed, the con-
centration of cAMP introduced from the whole-cell recording

136 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

a b c

d e f
1999 Nature America Inc. http://neurosci.nature.com

Fig. 5. Cyclic AMP augments INa and inhibits ICa,T via PKA. (a) INa induced by depolarization to 40 mV (Vh = 100 mV) in control solution (thin line) and
in 10 M forskolin (thick line). (b) INa recorded just after (thin line) and five min after (thick line) rupture of the patch membrane. The recording pipette
was filled with cesium solution containing the catalytic subunit of protein kinase A (PKA, 5 g/ml). (c) INa recorded in control solution (thin line) and in 1
mM 8-bromo-cAMP (thick line). The recording pipette contained PKA inhibitor (PKI, 400 g/ml). The superfusate was changed five min after rupture of
the patch membrane. (df) Effects of forskolin (d), PKA (e) and PKI (f) on ICa,T . Each ICa,T was recorded with the same protocols used in (ac).

pipette to the dendritic tip that is necessary to induce a
detectable response is higher than 100 M (ref. 36). This value
is much higher than that of cAMP-gated channels measured
in excised inside-out patch preparations (as low as 1 M; refs
37, 38). Moreover, olfactory cilia show adaptation to prolonged
stimulation, which is regulated by calcium feedback39,40. There-
fore, even if a small amount of cAMP produced by adrenaline
gradually diffused into cilia from the cell body, the basal trans-
duction current would not be affected, but additional odorant
stimulation would increase adenylyl cyclase activity and cause
normal transduction current.
Generally in sensory systems, changes in the dynamic range
are related to sensitivity contrast. This concept is widely used in
industrial sensors (such as photo-scanning systems). Based on
the present experimental results, therefore, we suggest that the
adrenergic system acts to enhance odorant contrast in olfac-
tory perception by modulating signal encoding at the cell body
and/or axon of ORCs.
METHODS
Preparation and recording procedures. ORCs were dissociated enzy-
matically from the olfactory epithelium of the newt, Cynops
pyrrhogaster as reported31. Isolated cells were viewed on an Olympus
upright microscope with differential interference contrast optics (40
water-immersion objective). To study I Na and I Ca,T of the somatic
membrane, we mainly selected ORCs that had lost their cilia. Mem-
brane voltages and currents were recorded in the whole-cell configu-
ration 41 using a patch-clamp amplifier (Axopatch 1-D, Axon
Instruments) linked to a computer. Recording procedures were con-
trolled by pCLAMP software (Axon Instruments). Data were low-pass
filtered (4-pole Bessel type) with a cut-off frequency of 5 kHz and then
digitized at 10 kHz by an analog-to-digital interface. All experiments
were done at room temperature (2325C).
Solutions and drugs. The recording pipette was filled with cesium solu-
tion (119 mM CsCl, 1 mM CaCl2, 5 mM EGTA, 10 mM HEPES and 1
mM Mg-ATP, pH adjusted to 7.4 with CsOH) to suppress outward cur-
rents, or pseudo-intracellular (potassium) solution (119 mM KCl, 1
mM CaCl 2 , 5 mM EGTA, 10 mM HEPES and 1 mM Mg-ATP, pH
adjusted to 7.4 with KOH) to record action potentials. The resistance of
the pipette was about 10 M. The control extracellular solution used to
record INa contained 80 mM NaCl, 3.7 mM KCl, 1 mM CoCl2, 35 mM
TEA, 2 mM HEPES and 15 mM glucose, and the solution used to
record ICa,T contained 77 mM choline-Cl, 3.7 mM KCl, 3 mM CaCl2,
0.1 mM CdCl2, 35 mM TEA, 2 mM HEPES and 15 mM glucose.
Drugs were applied either to the bath or to the cytoplasm from the
patch pipette. To examine the effect of a drug applied from the patch
pipette, we used the record obtained immediately after the establish-
ment of the whole-cell recording condition as a control. Records
obtained several minutes after rupture of the patch membrane were
compared against the control record. Forskolin was initially dissolved
in dimethylsulfoxide (DMSO) at a concentration of 10 mM and then
diluted to 1/1000 with bath solution, giving a final concentration of
10 M. 8-bromo-cAMP (1 M1 mM), forskolin and adrenaline (10
nM100 M) were bath applied from the U-tube system. The catalyt-
ic subunit of PKA (from bovine heart) was initially dissolved in 60 l
of 5 mM dithiothreithol in water and was diluted with the pipette solu-
tion. Cyclic AMP (1 mM), protein kinase inhibitor peptide (400 g/ml,
type II) and the catalytic subunit of PKA (5 g/ml) were applied from
the patch pipette. All chemicals were purchased from Sigma.
ACKNOWLEDGEMENTS
We thank Peter Sterling, Robert Smith, Noga Vardi, Michael Freed, Loren
Haarsma and Jonathan Demb for comments and support, Yoshikuni Ito for
technical support and Akiko Kawai for secretarial assistance. This work was in
part supported by grants from HFSPO and the Japanese MESSC (to TK).
ACCEPTED 19 NOVEMBER 1998

nature neuroscience volume 2 no 2 february 1999 137

 1999 Nature America Inc. http://neurosci.nature.com
articles
1. Gold, G. H. & Nakamura, T. Cyclic nucleotide-gated conductances: a new
class of ionic channels mediates visual and olfactory transduction. Trends
Pharmacol. 8, 312316 (1987).
2. Bakalyar, H. A. & Reed, R. R. The second messenger cascade in olfactory
receptor neurons. Curr. Biol. 1, 204208 (1991).
3. Breer, H. & Boekhoff, I. Second messenger signalling in olfaction. Curr. Biol.
2, 439443 (1992).
4. Ronnett, G. V. & Snyder, S. H. Molecular messengers of olfaction. Trends
Neurosci. 15, 508513 (1992).
5. Firestein, S. Electrical signals in olfactory transduction. Curr. Biol. 2, 444448
(1992).
6. Kurahashi, T. & Yau, K.-W. Tale of an unusual chloride current. Curr. Biol. 4,
256258 (1994).
7. Restrepo, D., Teeter, J. H. & Schild, D. Second messenger signaling in
olfactory transduction. J. Neurobiol. 30, 3748 (1996).
8. Takagi, S. F. in Human Olfaction (ed. Takagi, S. F.) 1481 (Univ. of Tokyo
Press, Tokyo, 1989).
9. Chen, Y., Getchell, T. V, Sparks, D. L. & Getchell, M. L. Patterns of adrenergic
and peptidergic innervation in human olfactory mucosa: age-related trends.
J. Comp. Neurol. 334, 104116 (1993).
10. Getchell, M. L. & Getchell, T. V. -adrenergic regulation of the secretory
granule content of acinar cells in olfactory glands of the salamander. J. Comp.
Neurol. 155, 435443 (1984).
11. Zielinski, B. S., Getchell, M. L., Wenokur, R. L. & Getchell, T. V.
1999 Nature America Inc. http://neurosci.nature.com
Ultrastructural localization and identification of adrenergic and cholinergic
nerve terminals in the olfactory mucosa. Anat. Rec. 225, 232245 (1989).
12. Woodhead, C. J. & Nimmo, A. J. Beta adrenoceptors in human nasal mucosa.
J. Laryngol. Otol. 105, 632634 (1991).
13. Beidler, L. M. in Sensory Communication (ed. Rosenblith, W. A.) 143157
(Wiley and MIT Press, New York, 1961).
14. Arechiga, C. & Alcocer, H. Adrenergic effects on electro-olfactogram. Exp.
Med. Surg. 27, 384394 (1969).
15. Kurahashi, T., Lowe, G. & Gold, G. H. Suppression of odorant responses by
odorants in olfactory receptor cells. Science 265, 118120 (1994).
16. Kawai, F., Kurahashi, T. & Kaneko, A. Nonselective suppression of voltage-
gated currents by odorants in the newt olfactory receptor cells. J. Gen. Physiol.
109, 265272 (1997).
17. Firestein, S. & Werblin, F. S. Gating currents in isolated olfactory receptor
neurons of the larval tiger salamander. Proc. Natl. Acad. Sci. USA 88,
62926296 (1987).
18. Schild, D. Whole-cell currents in olfactory receptor cells on Xenopus laevis.
Brain Res. 78, 223232 (1989).
19. Miyamoto, T., Restrepo, D. & Teeter, J. H. Voltage-dependent and odorant-
regulated currents in isolated olfactory receptor neurons of the channel
catfish. J. Gen. Physiol. 99, 505530 (1992).
20. Dubin, A. E. & Dionne, V. E. Action potentials and chemosensitive
conductances in the dendrites of olfactory neurons suggest new features for
odor transduction. J. Gen. Physiol. 103, 181201 (1994).
21. Kawai, F., Kurahashi, T. & Kaneko, A. T-type Ca2+ channel lowers the
138
threshold of spike generation in the newt olfactory receptor cell. J. Gen.
Physiol. 108, 525535 (1996).
22. Kawai, F., Kurahashi, T. & Kaneko, A. Quantitative analysis of Na+ and Ca2+
current contributions on spike initiation in the newt olfactory receptor cell.
Jpn. J. Physiol. 47, 367376 (1997).
23. Lamb, T. D. & Pugh, E. N. Jr. G-protein cascade, gain and kinetics. Trends
Neurosci. 15, 291298 (1992).
24. Hille, B. in Ionic Channels of Excitable Membranes (ed. Hille, B.) 170201
(Sinauer, Sunderland, 1992).
25. Hille, B. Modulation of ion-channel function of G-protein-coupled receptor.
Trends Neurosci. 17, 531535 (1994).
26. Seelig, T. L. & Kendig, J. J. Cyclic nucleotide modulation of Na+ and K+
currents in the isolated node of Ranvier. Brain Res. 245, 144147 (1982).
27. Matsuda, J. J., Lee, H. & Shibata, E. F. Enhancement of rabbit cardiac sodium
channels by -adrenergic stimulation. Circ. Res. 70, 199207 (1991).
28. Ono, K., Fozzard, H. A. & Hanck, D. A. Mechanism of cAMP-dependent
modulation of cardiac sodium channel current kinetics. Circ. Res. 72,
807815 (1992).
29. Narahashi, T., Tsunoo, A. & Yoshii, M. Characterization of two types of
calcium channels in mouse neuroblastoma cells J. Physiol. (Lond.) 383,
231249 (1987).
30. Pfeiffer-Linn, C. & Lasater, E. M. Dopamine modulates in a differential
fashion T- and L-type calcium currents in bass retinal horizontal cells. J. Gen.
Physiol. 102, 277294 (1993).
31. Kurahashi, T. Activation by odorants of cation-selective conductance in the
olfactory receptor cells isolated from the newt. J. Physiol. (Lond.) 419,
177192 (1989).
32. Getchell, T. V., Margolis, F. L. & Getchell, M. L. Perireceptor and receptor
events in vertebrate olfaction. Prog. Neurobiol. 23, 317345 (1985).
33. Hall, L. J. & Jackson, R. T. Effects of alpha and beta adrenergic agonists on
nasal blood flow. Ann. Otol. Rhinol. Laryngol. 77, 11201130 (1968).
34. Pun, R. Y. K., Kleene, S. J. & Gesteland, R.C. Guanine nucleotides modulate
steady-state inactivation of voltage-gated sodium channels in frog olfactory
receptor neurons. J. Membr. Biol. 142, 103111 (1994).
35. Lowe, G. & Gold, G. H. Nonlinear amplification by calcium-dependent
chloride channels in olfactory receptor cells. Nature 366, 283286 (1993).
36. Kurahashi, T. The response induced by intracellular cyclic AMP in isolated
olfactory receptor cells of the newt. J. Physiol. (Lond.) 430, 355371 (1990).
37. Kurahashi, T. & Kaneko, A. High density cAMP-gated channels at the ciliary
membrane in the olfactory receptor cell. Neuroreport 2, 58 (1991).
38. Kurahashi, T. & Kaneko, A. Gating properties of the cAMP-gated channel in
toad olfactory receptor cells. J. Physiol. (Lond.) 466, 287302 (1993).
39. Kurahashi, T. & Shibuya, T. Ca2+-dependent adaptive properties in the
solitary olfactory receptor cells of newt. Brain Res. 515, 261268 (1990).
40. Kurahashi, T. & Menini, A. Mechanism of odorant adaptation in the olfactory
receptor cell. Nature 385, 725729 (1997).
41. Hamill, O. P., Marty, A., Neher, E., Sakmann, B. & Sigworth, F. J. Improved
patch-clamp techniques for high resolution current recording from cells and
cell-free membrane patches. Pflugers Arch. 391, 85100 (1981).
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

Microdomains for neuronglia

interaction: parallel fiber signaling
to Bergmann glial cells
Jens Grosche1, Vitali Matyash2, Thomas Mller2, Alexej Verkhratsky2, Andreas Reichenbach1
and Helmut Kettenmann2

1 Paul Flechsig Institute for Brain Research, University of Leipzig, Jahnallee 59, D-04109 Leipzig, Germany
2 Max Delbrck Center for Molecular Medicine, Cellular Neurosciences, Robert-Rssle-Str. 10, D-13122 Berlin, Germany
The first two authors contributed equally to this work.
Correspondence should be addressed to H.K. (hk@nero.glia.mdc-berlin.de)
1999 Nature America Inc. http://neurosci.nature.com

Astrocytes are considered a reticulate network of cells, through which calcium signals can spread easily.
In Bergmann glia, astrocytic cells of the cerebellum, we identified subcellular compartments termed
glial microdomains. These elements have a complex surface consisting of thin membrane sheets,
contain few mitochondria and wrap around synapses. To test for neuronal interaction with these
structures, we electrically stimulated parallel fibers. This stimulation increased intracellular calcium con-
centration ([Ca2+]i) in small compartments within Bergmann glial cell processes similar in size to glial
microdomains. Thus, a Bergmann glial cell may consist of hundreds of independent compartments
capable of autonomous interactions with the particular group of synapses that they ensheath.

Astrocytes are described as a reticulate network of cells. A series
of cell-culture experiments established the concept that propa-
gating waves of intracellular calcium are a form of glial excitabil-
ity1,2. These waves can be initiated by a local stimulation, spread
from one cell to the next via gap junctions, and can travel
through tens of astrocytes. To analyze astrocytic calcium sig-
naling in situ, in cerebellar slices, we examined the Bergmann
glial cell, a radially aligned form of astrocyte, as a model for
local neuronalglial interactions. In cerebellar development,
the Bergmann glial cell fibers are important in guiding imma-
ture granule cells away from the external granular layer3. Dur-
ing granule cell migration, the glial cell processes are rather thin
and smooth, except for occasional bulges due to mitochondria.
This appearance changes dramatically as soon as the granule
cells have reached their target zone. Then, the stem fibers thick-
en and develop a massive outgrowth of lateral branches. These
lateral appendages determine the characteristic morphological
feature of an adult Bergmann glial cell and can be visualized,
at the lightmicroscopic level, by Golgi impregnation, immuno-
cytochemical labeling or Lucifer Yellow injection4,5. Electron-
microscopic inspection of glial elements in the molecular layer
revealed an elaborate plexus of fine processes, which varied con-
siderably in diameter (from about one micron to about twice
the membrane thickness). Unfortunately, the morphological
features of individual appendages cannot be reconstructed in
conventional electron-microscopic studies because the
appendages of these glial cells intermingle and cannot be dis-
tinguished from one another. There have been a few reports of
estimated surface-to-volume ratios of Bergmann glial process-
es, which were as high as about 13 to 15 per m (ref. 6). These
observations have created the impression that glial appendages
are irregular, random-shaped structures. Here we show instead
that these subcellular compartments appear organized and may
spatially restrict neuron-glia interactions.
RESULTS
Appendages show a complex morphology
We used serial electron microscopy and three-dimensional recon-
struction of processes from individual Bergmann glial cells to
search for basic construction principles that might have been
hidden in previous two-dimensional views. Furthermore, we
were interested in possible interactions between such glial sub-
structures and neurons. Because neither glial fibrillary acidic pro-
tein nor S-100 protein are sufficiently enriched in fine terminal
processes for their unequivocal immunocytochemical visualiza-
tion, living Bergmann glial cells were injected with Lucifer Yel-
low in a murine cerebellar slice preparation (Fig. 1a). After
fixation, the dye was photoconverted for transmission electron
microscopy. Complete series of 40 or more sections were obtained
from 5 different cells. In one case, we were able to cut a complete
series of about 450 consecutive ultrathin sections, correspond-
ing to a more than 20-m segment of one stem fiber and its
appendages. From these sections, images of the labelled glial pro-
files were transferred to a computer for three-dimensional recon-
struction (Fig. 1b and c).
Two classes of extensions from the radial fiber could be dis-
tinguished. The first type was a short lamellar or thorn-like off-
shoot that densely covered the entire fiber. The second type was
a much longer, frequently branching and highly complex out-
growth (Fig. 1b). The lateral appendages represented up to 90%
of the cells total membrane surface area. Whereas a surface-to-
volume ratio of 4.15 per m was calculated for the fiber (mean
diameter about 3 m), values of up to 25 per m were calculat-
ed for the appendages. The average surface-to-volume ratio of

nature neuroscience volume 2 no 2 february 1999 139

 1999 Nature America Inc. http://neurosci.nature.com
articles
a b c
the reconstructed part of the cell was close to 13 per m. Thus
appendages increase the cells surface-to-volume ratio more than
1999 Nature America Inc. http://neurosci.nature.com
five times as compared to the radial process.
The appendages are composed of microdomains
Although they appeared extremely irregular at first sight, sever-
al general rules of the organization of these complex outgrowths
became obvious from the reconstruction. One surprising obser-
vation was that all the many branches arose from only a few thin
stalks that left the fiber in a more or less orthogonal orientation.
a diameters of about 100 to 300 nm and could be more
Fig. 2. Fine structure of appendages and rela-
tion to synapses. (a) A small lateral appendage,
b ensheathed neuronal compartments, essential for var-
arising from the reconstructed part of the glial
fiber (blue in Fig. 1c), is shown as a slightly
turned, isolated three-dimensional reconstruc-
tion (left). Electron micrographs of four sections
contributing to the reconstruction (designated
14) are shown on the right; glial compartments
appear black after conversion of the injected
dye. The location of these sections in the recon-
struction is indicated by the labeled arrows. (1)
Region directly contacting synapses. (2) Glial
compartments without direct synaptic contacts.
(3) Bulging glial structure containing a mitochon-
drion. (4) The stalk of the appendage. (b) Another example of a synapse contacted by glial for visualization of the neuronal ultrastructure, in
compartments, appearing as black structures (white arrows); in this case, the postsynaptic several cases the postsynaptic elements of such
element can be traced back along the spine (black arrows) to the Purkinje cell dendrite synapses could be traced back to their origin from
(reddish overlay; the presynaptic terminal is labeled by a green overlay).
140
Fig. 1. Reconstruction of an appendage based on elec-
tron microscopic data. (a) Fluorescence light micro-
graph of a dye-injected Bergmann glial cell process is
shown; the red square (20 x 20 m) corresponds to the
portion that was reconstructed from consecutive ultra-
thin sections. (b) One of the lateral appendages, arising
directly from fiber (thick arrow), with all the other side
branches omitted for clarity. (c) The same appendage
as shown in (b), but with one of the appendages marked
by blue. This labeled structure is shown in isolation and
at higher magnification in Fig. 2 (left).
The outgrowths from these thin stalks were highly polymorphic.
Extremely thin, flat or thread-like profiles were connected to thick
and bulging parts in an unpredictable manner. This geometry is
in sharp contrast to that of neuronal dendrites where successive
branching occurs at more or less predictable distances, and these
branches are predictable in size7.
The geometry of the glial appendages was not entirely irregu-
lar, however. A smallest unit was recognized as consisting of a
long, thin stalk connected to a cabbage-like head (Figs. 1c and 2).
We term this structure a glial microdomain. The head shown in
Fig. 2 gave rise to a surface area of 316.9 m2 com-
pared to a volume of only 16.1 m3. The stalks had
than 7 m long. In the appendage shown in Fig. 2,
the almost cylindrical stalk had a surface area of 5.4
m2 and a volume of 0.37 m3. Thus it was mainly
the head that contributed to the enormous surface-
to-volume ratio of the appendage. Such a
microdomain may occur as a complete short out-
growth of the fiber or as a repetitive element consti-
tuting, in series, longer side branches. Thus, at some
degree of abstraction, Bergmann glial cell appendages
consist of repeating head elements connected to the
fiber, and/or to each other, by thin stalks.
The head had one or several swollen compart-
ment(s) containing mitochondria (Fig. 2); the num-
ber of mitochondria depended on the size of the
individual head. The presence of mitochondria in
each microdomain indicates that the microdomains
are metabolically independent. The metabolic sub-
strate for energy-consuming interactions with the
ious plasmalemmal transporters8,9, may be provid-
ed by the local mitochondria. An alternative energy
source might be ATP supplied from the soma or fiber;
however, this seems unlikely because the diffusion of
molecules through the long thin stalks must be sub-
ject to intolerable delay and decay.
The leaflets of the head contacted individual
synapses, or small groups of synapses (sometimes
even apparent in two-dimensional sections such as
Fig. 2a, right). Although the method was optimized
for a detection of fine glial cell processes rather than
Purkinje cell dendrites (Fig. 2b). This shows that the
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

a b

Fig. 3. Stimulation of parallel fibers triggers local calcium

signals in Bergmann glial cells. (a) Experimental protocol.
Parallel fibers (PF) were stimulated via a pipette con-
nected to a stimulator (STIM) while calcium-dependent
fluorescence responses were recorded in a Bergmann
glial cell (BG); PCL, Purkinje cell layer. (b) Confocal fluorescence intensity image of a patch-clamped Bergmann glial cell dialyzed with the calcium indi-
cator, Oregon green 488 BAPTA-1 (right). Three processes were distinguished (indicated as 13). Calcium signals in response to PF stimulation were
measured independently for each process (left). Time of PF stimulation is marked by arrows. The responding process (1) was then further subdivided
1999 Nature America Inc. http://neurosci.nature.com

into five regions of interest, in which calcium signals were measured separately (right). Time of PF stimulation is marked by an arrow and a dotted line.
The left image was obtained from a sequence of serial sections and shows the cell in depth (turned by 90 as compared to the right image). The lines
indicate the focal plane used for [Ca2+]i recordings. The process is thus within the volume from which recordings were obtained.

microdomains ensheath synapses between axon terminals of par- s. We found that not all processes from a given cell responded to the
allel fibers and dendritic spines of Purkinje cells. These sheaths stimulation, but rather observed that only small areas of the cell
were not complete, but glial leaflets were always found in the close showed an increase in fluorescence (Fig. 3b). To assess this more
vicinity of synaptic clefts (Fig. 2). quantitatively, we separately analyzed small compartments of about
20 to 50 m2 in size, and found that the signal was confined to
Local calcium signals only few of these compartments. In one example (Fig. 3b), we ana-
Next, we asked whether neuronglial signaling could be confined lyzed five compartments (~30 m2) along a Bergmann glial fiber.
to structures the size of a glial microdomain. We qualitatively mea- In two compartments, we found a fluorescence response. In three
sured intracellular calcium concentration ([Ca2+]i) in individual other compartments, no response could be detected. Because the
Bergmann glial cells in a cerebellar slice preparation by dialyzing the (projected) area covered by a microdomain is in the range of less
cell with the calcium-sensitive fluorescent dye Oregon green 488
BAPTA-1 hexapotassium salt via a patch pipette. For imaging, we
used a confocal microscope, with a lateral resolution just below 1 a b
m. Thus, we were not able to resolve the actual substructure of
the glial microdomain, but we could record the integrative response
from areas about the size of the microdomain. Slices were cut par-
allel to the orientation of the parallel fibers, which allowed us to
stimulate them by placing the electrode about 200 m away from
the Bergmann glial cell under study (Fig. 3a). Stimulation of par-
allel fibers by 10 brief pulses (200 s, 50 Hz), an experimental pro-
tocol that routinely induces synaptic responses in Purkinje
neurons10, elicited a rapid and transient increase in calcium indi-
cator fluorescence in glial processes. (We observed fluorescence
responses in 75 cells out of 126.) The signal peaked about 1 to 2 s
after stimulation and returned to the baseline level within 15 to 30
Control Control

Fig. 4. Inhibition of nerve impulse propagation and synaptic transmis-

sion blocks glial calcium responses. (a) Bergmann glial cell process fluo-
rescence intensity image (top). A region of interest is defined by the
white line. The time course of the fluorescence change (corresponding Cd2+
TTX
to intracellular [Ca2+] changes) was measured from the region of inter-
est in response to parallel-fiber stimulation (bottom). The fluorescence
signal recorded in control conditions was not observed in the presence
of tetrodotoxin (1 M TTX). The response recovered after a 10-min
washout. Time of PF stimulation is marked by an arrow. (b) Bergmann
Washout
glial cell process fluorescence intensity image (top). A region of interest
is defined by the white line. As in (a), [Ca2+]i signals in the glial cell
process were recorded in control conditions and after two minutes of
slice superfusion with 100 M cadmium (bottom). Time of PF stimula-
tion is marked by an arrow.

nature neuroscience volume 2 no 2 february 1999 141

 1999 Nature America Inc. http://neurosci.nature.com
articles
a b
Fig. 5. Spontaneous activity in Bergmann glia microdomains (a) A fluo-
rescence intensity image of the Bergmann glial cell with selected regions
of interest. (b) The normalized fluorescence intensity (F/F0) traces
demonstrate spontaneous [Ca2+]i transients in regions 1, 2 and 4;
1999 Nature America Inc. http://neurosci.nature.com
regions 3 and 5 were quiescent. Note that the responses of regions 1
and 2 were asynchronous.
than 100 m2, the extent of the fluorescence signal matches the
size of the glial microdomain. To ensure that responding and non-
responding areas were within the same focal plane, we collected
120 images above and below the analyzed area after the physio-
logical experiment (n = 5; with a z-step of 0.36 M). This allowed
us to reconstruct the position of the process in depth. Rotating the
process by 90 relative to analyzed plane (Fig. 3b) demonstrated
that the fluorescence signal was confined to the analyzed volume,
assuming 2.5 m as the depth of the focal plane for our 40 objec-
tive, suggesting that heterogeneity in calcium signaling along the
Bergmann glial process was not due to sampling artifacts.
The glial calcium response involves synaptic activation
To exclude a stimulation artifact and to verify the synaptic ori-
gin of glial calcium signals, we used pharmacological agents
known to interfere with nerve-impulse propagation and [Ca2+]i
entry into synaptic terminals. First, blockade of voltage-gated
sodium channels (and thus action potential propagation) by
tetrodotoxin (1 M) completely eliminated the local glial [Ca2+]i
changes (Fig. 4a, n = 7). This blockade was reversible; after a ten-
minute washout, stimulation again triggered a calcium signal in
the Bergmann glial microdomain. Similarly, blockade of voltage-
gated calcium channels with 100 M cadmium completely elim-
inated glial [Ca 2+ ] i elevation in response to parallel-fiber
stimulation (Fig. 4b, n = 4). Because neither we nor others have
evidence of calcium channels in Bergmann glial cells, we con-
clude that cadmium blocked neuronal calcium channels and
hence inhibited the calcium-dependent synaptic release of neu-
rotransmitter. Recovery of the response was observed only in one
cell after the five-minute washout.
Further substantiating the concept of functionally indepen-
dent subcompartments within Bergmann glial cell processes, we
observed spontaneous [Ca2+]i fluctuations within restricted areas
along the processes. In about 50% of recordings (15 of 27 cells),
[Ca2+]i increased transiently within areas less than 100 m2. We
observed regionally confined signaling in up to five compart-
ments within our visual field. Moreover, in some of these com-
partments, signaling was clearly not synchronized (Fig. 5).
Stimulation of parallel fibers could trigger [Ca2+]i signals in the
spontaneously active subcompartments.
142
DISCUSSION
The emerging picture of independent glial microdomains, capa-
ble of glianeuronal interactions in a subcellular microenviron-
ment, changes the concept of the glial cells as integrated
members of a large interacting meshwork. The common image
of astrocytes describes these cells as members of a large network,
in which calcium waves can travel with little decrement within
the syncytium11,12. Appendages of Bergmann cells are obvious-
ly not designed for the spread of calcium waves within the cell
or between adjacent glial cells. Rather, calcium signals seem to
be confined to modules that each interact with only a few
synapses. These local [Ca2+]i increases could strongly influence
the physiological properties of the glial microdomain. Our pre-
vious studies show that [Ca2+] i elevation blocks the resting
potassium conductance in Bergmann glial cells, leading to mem-
brane depolarization6, and blocks communication via gap junc-
tions either between cells or within the cell via autocellular
coupling13. This may have consequences for synaptic transmis-
sion within the extent of the microdomain. Astrocytes, including
Bergmann glia, are thought to efficiently take up transmitters,
such as glutamate14. One may speculate that the glial membrane
depolarization would slow the uptake rate, as shown for retinal
glia15, and a lack of gap junctional communication would con-
fine both the depolarization and the change in uptake to the
microdomain. This may increase transmitter availability and
synaptic efficacy for the synapses ensheathed by a glial
microdomain but not for neighboring synapses. Another pos-
sible way Bergmann glial cells could influence neuronal activity
is by releasing biologically active substances. Coactivation of
AMPA/kainate and metabotropic glutamate receptors in hip-
pocampal astrocytes triggers glutamate release by a calcium-
dependent process16. Bergmann glial cells similarly coexpress
both ionotropic (AMPA/kainate 5) and metabotropic (mGluR
1 and 5; ref. 17) glutamate receptors. Micromodular organiza-
tion of the Bergmann glial cell would confine the release of glu-
tamate to a defined and limited area. Both mechanisms could
serve to modulate only those synapses that are ensheathed by a
given microdomain, independent of the behavior of neighboring
synapses and their glial partners. Purkinje cells express two func-
tionally distinct populations of synaptic spines, and individual
spines are capable of independent activation18,19. It may well be
that the Bergmann glial microdomains are adjusted to meet this
functional diversity of Purkinje cell synapses.
METHODS
Tissue slices. Slices of 120 to 200 m were prepared from 2022 day old
mice as described 5. The slices were either cut parallel (for electron
microscopy) or perpendicular to the orientation of the Purkinje cell den-
dritic tree (for parallel-fiber stimulation). Slices were perfused with phys-
iological saline (2224C) containing 154 mM NaCl, 5.6 mM KCl, 2 mM
CaCl2, 1 mM MgCl2 1.25 mM NaH2PO4, 26 mM NaHCO3 and 10 mM
glucose, pH 7.4 when continuously gassed with 95% CO2 and 5% O2.
Calcium imaging. The patch-clamp technique was used in the whole-cell
configuration. Pipettes had resistances of 46 M. The intrapipette solu-
tion contained 150 mM KCl, 2 mM MgCl2, 10 mM HEPES, 3 mM Na2ATP
and 0.5 mM Oregon green 488 BAPTA-1 (Molecular Probes), pH 7.3. A
confocal laser microscope (Noran, Odyssey XL mounted on Zeiss Axio-
scope, with 40 water immersion objectives, either from Zeiss, numerical
aperture 0.75, or from Olympus, numerical aperture 0.8) was used to
acquire fluorescence images at 7.5 Hz. Parallel fibers were stimulated from
an isolated stimulus circuit by a glass micropipette with a resistance of
about 1 M. Calcium responses of Bergmann glial cells were recorded
while parallel fibers were stimulated by 1 and occasionally 10 square puls-
es of 200-s duration, with a frequency of 50 Hz and voltage of 50 volts.
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com
Lucifer Yellow staining. Bergmann glial cells were dialyzed via a patch
pipette containing 130 mM KCl, 0.5 mM CaCl2, 5 mM EGTA, 2 mM
MgCl2, 10 mM HEPES and 1 mg/ml Lucifer Yellow. After a 10-min dial-
ysis, the patch pipette was pulled off after damaging the seal by a large
hyperpolarizing current injection.
Tissue preparation for electron microscopy. The tissue was fixed with
2% paraformaldehyde and then incubated in 3,3 diaminobenzidinete-
trahydrochloride (DAB, Sigma); photoconversion was started by intense
blue light illumination. After further fixation in 2% paraformaldehyde
and 2% glutaraldehyde, the tissue was transferred to 1% osmium tetrox-
ide, dehydrated and embedded in EPON, cut at a constant thickness of
60 nm and examined with a Zeiss 10 electron microscope at 80 kV.
Three-dimensional reconstruction of electron-microscopic images. A series
of electron micrographs from the appendage was taken at a magnification
of 5,000. The outlines of all DAB-stained glia profiles were scanned into
special computer programs. Each single image was carefully fit to its neigh-
boring sections, to compensate for slight deformations of the sections that
unavoidably occur due to cutting and heating by the electron beam. Final-
1999 Nature America Inc. http://neurosci.nature.com
ly, a three-dimensional reconstruction of the whole appendage was made
(Image Volumes, Minnesota Datametrics Corp., St. Paul, Minnesota, USA).
ACKNOWLEDGEMENTS
The authors thank S. Lyons, A. Konnerth and B. R. Ransom for comments on the
manuscript. This work was supported by grants from Deutsche
Forschungsgemeinschaft (SFB 515; 436UKR), Volkswagen-Stiftung and
Interdisciplinary Center for Clinical Research at the University of Leipzig,
(01KS9504, Project C5).
RECEIVED 3 SEPTEMBER; ACCEPTED 23 NOVEMBER 1998
1. Finkbeiner, S. Calcium waves in astrocytesfilling in the gaps. Neuron 8,
11011108 (1992).
2. Verkhratsky, A. Orkand, R. K. & Kettenmann, H. Glial calcium: Homeostasis
and signalling function. Physiol. Rev. 78, 142 (1998).
nature neuroscience volume 2 no 2 february 1999
articles
3. Hatten, M. E. Riding the glial monorail: a common mechanism for glial-
guided neuronal migration in different regions of the developing
mammalian brain. Trends Neurosci. 13, 179184 (1990).
4. Chan-Palay, V. & Palay, S. L. High voltage electron microscopy of rapid
golgi preparations. Neurons and their processes in the cerebellar cortex of
monkey and rat. Z. Anat. Entwicklungsgesch. 137, 125153 (1972).
5. Mller, T., Mller, T., Berger, T., Schnitzer, J. & Kettenmann, H. Calcium
entry through kainate receptors and resulting potassium-channel
blockade in Bergmann glial cells. Science 256, 15631566 (1992).
6. Reichenbach, A. Siegel, D., Senitz, T. & Smith, G. Jr A comparative fractal
analysis of various mammalian astroglial cell types. Neuroimage 1, 6977
(1992).
7. Rall, W. et al. Matching dendritic neuron models to experimental data.
Physiol. Rev. 72, S159186 (1992).
8. Sweadner, K. J. in Neuroglia (eds. Kettenmann, H. & Ransom, B. R.)
259272 (Oxford Univ. Press, New York, 1995).
9. Rosenthal, M. & Sick, T. J. Glycolytic and oxidative metabolic
contributions to potassium ion transport in rat cerebral cortex. Can. J.
Physiol. Pharmacol. 70, S165169 (1992).
10. Konnerth, A., Llano, I. & Armstrong, C. M. Synaptic currents in
cerebellar Purkinje cells. Proc. Natl. Acad. Sci. USA 87, 26622665 (1990).
11. Kuffler, S. W. Nicholls, J. G. & Orkand, R. K. Physiological properties of
glial cells in the central nervous system of amphibia. J. Neurophysiol. 29,
768787 (1966).
12. Verkhratsky, A. & Kettenmann, H. Calcium signalling in glial cells. Trends
Neurosci. 19, 346352 (1996).
13. Mller, T., Mller, T., Neuhaus, J. & Kettenmann, H. Electrical coupling
among Bergmann glial cells and its modulation by glutamate receptor
activation, Glia 17, 274284 (1996).
14. Clark, B. A. & Barbour, B. Currents evoked in Bergmann glial cells by
parallel fibre stimulation in rat cerebellar slices. J. Physiol. (Lond.) 502,
335350 (1997).
15. Sarantis, M. & Attwell, D. Glutamate uptake in mammalian retinal glia is
voltage- and potassium-dependent. Brain Res. 516, 322325 (1990).
16. Bezzi, P. et al. Prostaglandins stimulate calcium-dependent glutamate
release in astrocytes. Nature 391, 281285 (1998).
17. Kirischuk, S., Kettenmann, H. & Verkhratsky, A. Metabotropic receptors
involved in calcium signalling in mouse Bergmann glial cells. J. Physiol.
(Lond.) 493, 46 (1996).
18. Eilers, J., Augustine, G. J. & Konnerth, A. Subthreshold synaptic Ca 2+
signalling in fine dendrites and spines of cerebellar Purkinje neurons.
Nature 373, 155158 (1995).
19. Denk, W., Sugimori, M. & Llins, R. Two types of calcium response
limited to single spines in cerebellar Purkinje cells. Proc. Natl. Acad. Sci.
USA 92, 82798282 (1995).
143
 1999 Nature America Inc. http://neurosci.nature.com

articles

Voltage-activated sodium channels

amplify inhibition in neocortical
pyramidal neurons
Greg Stuart

Division of Neuroscience, John Curtin School of Medical Research, Australian National University, Canberra, A.C.T. 0200, Australia
Correspondence should be addressed to G.S. (Greg.Stuart@anu.edu.au)

Inhibitory postsynaptic potentials (IPSPs) in neocortical pyramidal neurons are increased in duration
and amplitude at depolarized membrane potentials. This effect was not due to changes in the time
1999 Nature America Inc. http://neurosci.nature.com

course of the underlying synaptic current. The role of postsynaptic voltage-activated channels was
investigated by mimicking the voltage change that occurs during an IPSP with current injections.
The peak and integral of these simulated IPSPs increased during depolarization of the membrane
potential in a tetrodotoxin-sensitive manner. This amplification presumably occurs as the
hyperpolarization associated with IPSPs turns off sodium channels that are tonically active at
depolarized membrane potentials. IPSP amplification increased the ability of IPSPs to inhibit action
potential firing and promoted IPSP-induced action potential synchronization.

Synaptic integration in neurons involves the temporal and spatial
summation of synaptic events, which can lead to action poten-
tial initiation if the membrane potential in the axon becomes suf-
ficiently depolarized. The effectiveness with which a neuron sums
different synaptic inputs depends on the membrane time con-
stant, the time interval between synaptic events, their location
and how they propagate within the neuron1. Even in an electri-
cally passive neuron, synaptic integration is nonlinear because
changes in membrane potential associated with one synaptic
event affect the driving force for ion flow through other synaptic
conductances1. The ability of neurons to sum synaptic inputs is
further complicated by the large range of voltage-activated chan-
nels that they express2. Although many of these voltage-activat-
ed conductances are only active during action potential
generation, others are activated at subthreshold membrane poten-
tials. Furthermore, some voltage-activated conductances are locat-
ed in the soma and dendrites of neurons, distant from the site of
action potential initiation3.
The activation of voltage-activated conductances at mem-
brane potentials below the threshold for action potential ini-
tiation influences the ability of neurons to integrate synaptic
inputs. For example, both the peak and duration of excitato-
ry postsynaptic potentials (EPSPs) are increased at depolar-
ized, subthreshold membrane potentials414. In many cases,
amplification of these EPSPs is mediated by activation of a
non-inactivating or persistent sodium current 5,1014 . This
tetrodotoxin (TTX)-sensitive sodium current is activated at
membrane potentials approximately 10 mV below action
potential threshold, and it is thought to result from the acti-
vation of the same voltage-activated sodium channels respon-
sible for action potential generation that have switched into a
non-inactivating mode15. In many neuron types, this sodium
current is also responsible for an apparent increase in input
resistance at depolarized membrane potentials15.
A non-inactivating sodium current is likely to also influence
inhibitory postsynaptic potentials (IPSPs) at depolarized mem-
brane potentials. The present results show that non-inactivating
sodium channels can cause significant amplification of IPSPs at
depolarized, subthreshold membrane potentials. The extent of
this IPSP amplification depends on the amplitude and time course
of the inhibitory synaptic current and significantly increases the
ability of IPSPs to inhibit action potential firing. These findings
therefore indicate that voltage-activated sodium channels, which
are usually associated with excitation in the central nervous sys-
tem, can also amplify inhibition. A preliminary report of this work
has appeared (G.S. Soc. Neurosci. Abst. 23, 261.2, 1997).
RESULTS
Voltage dependence of evoked IPSPs
Somatic gramicidin perforated-patch recordings16 were used to
gain intracellular access to visually identified neocortical layer-5
pyramidal neurons without modifying the internal chloride con-
centration. Monosynaptic IPSPs evoked by extracellular stimula-
tion were recorded in the presence of the glutamate receptor
antagonists CNQX and APV at different somatic membrane
potentials (Fig. 1a). These IPSPs were completely blocked by bath
application of the GABAA receptor antagonist bicuculline (20 M;
n = 3). Their average reversal potential was 63.1 1.2 mV
(n = 10). After correction for a 3-mV junction potential between
recording and bath solutions, this gives an estimated internal chlo-
ride concentration of 11.2 mM, similar to the value reported in
cortical neurons from approximately two-week-old animals17.
IPSPs had longer durations at depolarized membrane poten-
tials than at hyperpolarized membrane potentials (Fig. 1a;
n = 10). This effect was more obvious when the amplitudes of
IPSPs at depolarized and hyperpolarized membrane potentials
were normalized (Fig. 1b). In addition, the time to peak of
IPSPs was delayed at depolarized membrane potentials com-
pared with hyperpolarized membrane potentials (Fig. 1b). This
increased IPSP duration at depolarized membrane potentials

144 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 1. Effect of membrane potential on evoked IPSPs.

(a) Gramicidin perforated-patch recording of evoked
a c
IPSPs at different membrane potentials between 71 and

Peak (mV)
56 mV. (b) IPSPs recorded at the most hyperpolarized
(71 mV) and depolarized (56 mV) membrane poten-
tials were normalized to the same peak amplitude. Note
that the IPSP at the depolarized membrane potential
peaks later and has a longer duration than the IPSP at the
hyperpolarized membrane potential. (c, d) IPSP peak
amplitude (c) and integral (d) versus membrane potential. Membrane potential (mV)
The data were fitted with a modified version of the GHK
equation between 71 and 62 mV. Note that at depolar-
ized membrane potentials, IPSP peak, and especially IPSP
integral, deviate substantially from the fitted curve. All b d
Normalized
recordings are from the same cell.

Integral (mV s)
1999 Nature America Inc. http://neurosci.nature.com

Membrane potential (mV)

was not due to a GABAB-receptor-mediated IPSP component, as
similar results were found in the presence of the specific GABAB
receptor antagonist CGP 55845A (ref. 18).
At membrane potentials more negative than the IPSP reversal
potential, the peak amplitude and integral of IPSPs were well fit
by a modified version of the Goldman-Hodgkin-Katz (GHK)
equation (Fig. 1c and d). At potentials depolarized to the IPSP
reversal potential, however, the measured values clearly deviat-
ed from the predictions of the GHK equation (Fig. 1c and d),
particularly for IPSP integral (Fig. 1d). Similar results were
observed in nine other cells.
The increased amplitude and duration of evoked IPSPs at
depolarized membrane potentials could occur because of an
increase in the underlying inhibitory postsynaptic current
(IPSC), or because the voltage response to this synaptic current
is amplified by activation or deactivation of voltage-dependent
conductances. The first possibility was addressed by voltage
clamping the somata of layer-5 pyramidal neurons
using low-access, whole-cell recording and evok-
ing monosynaptic IPSCs by extracellular synaptic a
stimulation close to the soma in the presence of
CNQX and APV. IPSCs were recorded over rough-
generated close to the soma of layer-5 pyramidal neurons, and
decayed with a double-exponential time course (Fig. 2b and
below).
Unlike IPSPs, the time course of normalized IPSCs did not
change at different membrane potentials (Fig. 2b; n = 5). Fur-
thermore, the relationships of IPSC peak and integral to mem-
brane voltage were well described by the GHK equation over the
entire voltage range examined (Fig. 2c and d; n = 5). These
results suggest that the increase in peak amplitude and duration
of IPSPs at depolarized membrane potentials cannot be
explained by changes in the time course or amplitude of the
underlying synaptic current.
IPSP amplification requires sodium channels
The second possibility, that IPSPs are modified by the activation or
deactivation of voltage-dependent conductances, was addressed
by using somatic current injections to mimic the voltage change
c
Peak (pA)

ly the range of membrane potentials used for IPSP

recordings (Fig. 2a). These IPSCs had fast rise
times (average 20-to-80% rise time, 0.48 0.12
ms; n = 5), consistent with the idea that they were

Holding potential (mV)

Fig. 2. Effect of membrane potential on evoked IPSCs.
(a) Whole-cell, voltage-clamp recordings of evoked
IPSCs at different holding potentials between 75 and
45 mV. (b) The IPSCs shown in (a) normalized to the b Normalized d
same peak amplitude. The smooth line shows the triple
Integral (pA s)

exponential (one rising, two decaying) fit to these IPSCs

(taurise = 0.5 ms; taudecay(fast) = 5.0 ms and taudecay(slow) =
37.5 ms). (c, d) Averaged IPSC peak amplitude (c) and
integral (d) versus holding potential in five cells. The
data were fitted with a modified version of the GHK
equation between 75 and 45 mV. Note that both IPSC
peak amplitude and IPSC integral are well fit over the
Holding potential (mV)
entire voltage range.

nature neuroscience volume 2 no 2 february 1999 145

 1999 Nature America Inc. http://neurosci.nature.com

articles

a b tion at depolarized membrane poten-

tials is mediated by the activation or
deactivation of postsynaptic voltage-
dependent channels.
The role of non-inactivating sodi-
um conductances was investigated by
recording simulated IPSPs at differ-
ent membrane potentials under con-
trol conditions and after bath
application of the sodium channel
blocker TTX (1 M). In TTX, the
Fig. 3. Generation of simulated IPSPs. (a) Infrared differential interference contrast image during double time course of simulated IPSPs at
somatic recording from a neocortical layer-5 pyramidal neuron. The voltage-recording and current-pass- depolarized and hyperpolarized
ing pipettes are indicated. (b) Comparison of the time course of evoked (solid line) and simulated (dot- membrane potentials were similar
ted line) IPSPs recorded at the resting membrane potential and normalized to the same peak amplitude. (Fig. 4c; n = 8); the small difference
remaining in TTX is probably due to
the hyperpolarization-activated
inward current I h (see below). TTX
1999 Nature America Inc. http://neurosci.nature.com

that occurs during evoked IPSPs. In these experiments, hyperpo-
larizing current with the same time course as the evoked IPSC
was injected via a second somatic recording pipette (Fig. 3a). The
time course of the injected current was based on triple-exponen-
tial fits to normalized and averaged evoked IPSCs (Fig. 2b). In
five cells, the average rising time constant was 0.67 0.24 ms and
the fast and slow decay time constants were 3.8 1.1 ms and
34.3 17.8 ms, with an average ratio of the amplitudes of the fast
and slow decaying time constants of 9:2. Simulated IPSPs gen-
erated by somatic current injection with this time course ade-
quately mimicked the voltage change that occurs during evoked
IPSPs (Fig. 3b).
Depolarizing the membrane potential increased both the
amplitude and the duration of simulated IPSPs (Fig. 4a; n = 8).
When these IPSPs were normalized, depolarization increased
the duration and time to peak of simulated IPSPs, as observed
for evoked IPSPs (Fig. 4b; compare with Fig. 1b). These results
show that somatic current injection with a time course similar to
that of evoked IPSCs can mimic the effects of membrane depo-
larization on evoked IPSPs, and indicate that IPSP amplifica-
completely blocked the amplification of both simulated IPSP
peak and integral at depolarized membrane potentials, reduc-
ing them to values typically observed at the resting membrane
potential (Fig. 4d and e; n = 8). Under control conditions,
depolarization increased the integral of simulated IPSPs more
than the peak, and amplification of both was maximal at the
most depolarized membrane potentials tested (Fig. 4d and e).
On average, the peak amplitude of simulated IPSPs approxi-
mately doubled (202 8%), and IPSP integral increased
approximately fourfold (443 45%) at the most depolarized,
just subthreshold, membrane potentials relative to the rest-
ing membrane potential. Thus, amplification of simulated
IPSPs at depolarized membrane potentials is substantial, and
this amplification is mediated via TTX-sensitive, voltage-acti-
vated sodium channels.
Whether evoked IPSPs are also amplified at depolarized mem-
brane potentials by voltage-activated sodium channels was inves-
tigated with the internal sodium channel blocker QX-314.
Somatic whole-cell recordings of evoked IPSPs were obtained at
different membrane potentials. As with the gramidicin perfo-

a d
Peak (mV)

Fig. 4. Effects of membrane potential and TTX on

simulated IPSPs. (a) Simulated IPSPs recorded at dif-
ferent membrane potentials between 68 and 51 mV.
control (b) Simulated IPSPs recorded at hyperpolarized (68
TTX
b Control
mV) and depolarized (51 mV) membrane potentials
were normalized to the same peak amplitude. Note
that the IPSP at the depolarized membrane potential
Membrane potential (mV) peaks later and has a longer duration than the IPSP at
e the hyperpolarized potential (compare with Fig. 1b).
(c) After bath application of TTX (1 M), simulated
IPSPs recorded at hyperpolarized (67 mV) and depo-
Integral (mV s)

larized (51 mV) membrane potentials (normalized to

the same peak amplitude) have a similar time course.
c TTX (d, e) Simulated IPSP peak amplitude (d) and integral
(e) versus membrane potential without (solid circles)
and with TTX (1 M; open circles). Note that the IPSP
peak and integral show a substantial (2 to 4-fold)
increase at membrane potentials more depolarized
than approximately 60 mV, and that IPSP amplification
is completely blocked by TTX. All recordings are from
Membrane potential (mV) the same cell.

146 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
a Control b QX-314
Normalized Normalized
1999 Nature America Inc. http://neurosci.nature.com
rated-patch recordings, depolarization increased the duration
and time to peak of evoked IPSPs (Fig. 5a). A second somatic
whole-cell recording was then made from the soma of the same
neuron using a pipette filled with the same intracellular solution
plus 1 mM QX-314, which was allowed to diffuse into the neuron
for approximately five minutes. After intracellular perfusion with
QX-314, the time courses of evoked IPSPs at depolarized and
hyperpolarized membrane potentials were similar (Fig. 5b; n = 4).
These results indicate that evoked IPSPs, like simulated IPSPs,
are also amplified at depolarized membrane potentials by volt-
age-activated sodium channels.
a Control b Control
TTX
Percent increase
Percent increase
0 percent increase in control IPSP peak (p) and integral (P)
IPSC amplitude (pA)
nature neuroscience volume 2 no 2 february 1999
articles
Fig. 5. Effect of QX-314 on evoked IPSPs. (a) Whole-cell, current-
clamp recordings of evoked IPSPs at different membrane potentials
between 75 to 55 mV (top). IPSPs recorded at the most hyperpolar-
ized (75 mV) and depolarized (55 mV) membrane potentials were
normalized to the same peak amplitude (bottom). (b) Evoked IPSPs
recorded at different membrane potentials between 75 and 55 mV
after intracellular perfusion with QX-314 (1 mM) via a second somatic
whole-cell recording pipette (top). IPSPs recorded at the most hyperpo-
larized (75 mV) and depolarized (55 mV) membrane potentials were
normalized to the same peak amplitude (bottom). Note that after intra-
cellular perfusion with QX-314, IPSPs at depolarized and hyperpolarized
membrane potentials have a similar time course. All recordings are from
the same cell.
Dependence on IPSP amplitude and time course
The dependence of IPSP amplification on the size and duration
of inhibitory events was investigated by generating simulated IPSPs
of different amplitudes or durations at depolarized membrane
potentials with or without TTX (1 M; Fig. 6). Simulated IPSPs
were used in these experiments to avoid complications arising
from changes in the driving force for IPSPs of different ampli-
tudes, and to be able to vary IPSP duration over a physiological
range. When IPSP amplitudes were varied, the kinetics of simu-
lated IPSCs were the same as those described above, whereas when
IPSP durations were varied, IPSC kinetics were varied over a range
from those of fast GABAA receptor IPSCs to slower
GABAB receptor IPSCs (see Fig. 6 legend).
Increasing the amplitude of simulated IPSPs
decreased amplification of both IPSP peak and inte-
gral (Fig. 6a; n = 4), whereas increasing the duration
of simulated IPSPs increased amplification of IPSP
peak, but reduced amplification of IPSP integral
(Fig. 6b; n = 4). This dependence of IPSP amplifica-
tion on the amplitude and duration of simulated
IPSPs was significantly reduced by ZD7288 (100 M;
TTX n = 3), which blocks Ih (ref. 19). These results indi-
Fig. 6. Effect of IPSP amplitude and duration.
(a) Simulated IPSPs with different amplitudes recorded at
depolarized membrane potentials in control (52 mV; top)
and in 1 M TTX (50 mV; middle). Simulated IPSC ampli-
tudes were 100 pA, 250 pA, 500 pA and 1 nA. The percent
increase in control IPSP peak (open circles) and integral
(solid circles) relative to that in the presence of TTX is
plotted against simulated IPSC amplitude at the bottom
(n = 4). (b) Simulated IPSPs with different durations but
similar amplitude at the resting membrane potential
recorded at depolarized membrane potentials in control
(51 mV; top) and in 1 M TTX (50 mV; middle).
Simulated IPSCs had exponentially rising and decaying time
constants of 0.5 and 4 ms, 2 and 10 ms, 5 and 20 ms, 10
and 50 ms, and 50 and 100 ms, with corresponding ampli-
tudes of 500 pA, 263 pA, 185 pA, 143 and 132 pA. The
relative to that in the presence of TTX is plotted against
IPSC half width (ms) simulated IPSC half-width at the bottom (n = 4).
147
 1999 Nature America Inc. http://neurosci.nature.com

articles

cate that as inhibition becomes more powerful during synchro-
nized activation of inhibitory input, or more prolonged during
a burst of inhibitory input or during activation of inhibitory
events of longer duration (for example, GABAB receptor IPSPs)
amplification of IPSPs by voltage-activated sodium channels is
reduced on balance because of the increased activation of Ih.
IPSP amplification and action potential firing
As amplification of IPSPs is greatest at membrane potentials close
to the action potential threshold, it might be expected to increase
the ability of IPSPs to inhibit action potential firing. Therefore,
the ability of simulated IPSPs to inhibit action potential genera-
tion was investigated under control conditions and in the pres-
ence of submaximal concentrations of TTX (5 nM). The rationale
was that these concentrations of TTX would partially block IPSP
amplification without significantly affecting action potentials,
and so reveal the extent to which amplification of IPSPs increas-
1999 Nature America Inc. http://neurosci.nature.com
es the ability of IPSPs to inhibit action potential firing. Simulat-
ed IPSPs, rather than evoked IPSPs, were used to avoid any presy-
naptic action of TTX on transmitter release. Low concentrations
of TTX were used rather than intracellular QX-314 because pre-
liminary experiments showed that QX-314 concentrations suf-
ficient to block IPSP amplification at depolarized membrane
potentials significantly reduced action potential amplitude.
Action potentials were evoked by constant depolarizing cur-
rent injection and had an average firing frequency of 13.9 0.5
Hz in control and 14.8 2.1 Hz in TTX. In control, simulated
IPSPs (approximately 2.5 mV in amplitude at the resting mem-
brane potential) inhibited action potential firing to an average
frequency of 0.5 0.5 Hz during a 35-millisecond time window
after their onset (Fig. 7a; a 96% reduction; n = 4). In low con-
centrations of TTX, however, the average action potential fre-
quency during the same time window was 7.9 1.4 Hz (Fig. 7b;
a 43% reduction; n = 4). These low concentrations of TTX
reduced amplification of simulated IPSP peak

and integral by on average 63 7% and

a Control b 5 nM TTX 62 10% respectively (Fig. 7; n = 4).
Thus, low concentrations of TTX signifi-
cantly reduced the ability of simulated IPSPs to
inhibit action potentials, which is unlikely to be
due to an effect on the action potential itself, as
these low concentrations of TTX had only a
small effect on action potential amplitude and
rate of rise. On average, action potential ampli-
tude measured from threshold was reduced by
6.4 5%, whereas action potential rate of rise
was decreased by 9.4 7%. These changes were
Simulated IPSPs not statistically significant. Furthermore, if any-
thing, low concentrations of TTX should
increase the effectiveness of IPSPs at inhibiting
action potential firing, as one would expect this
treatment to make the initiation of action
potentials less secure.
These TTX concentrations also reduced the
synchronization of action potentials following
Post-stimulus time histogram simulated IPSPs (Fig. 7). Autocorrelations of
action potential timing during a 150-millisec-
ond time window starting from the simulated
IPSP peak were used to assess IPSP-induced
Counts

action potential synchronization. These auto-

correlations (digitally filtered at 40 Hz) showed
clear peaks spaced approximately 50 ms apart
(Fig. 7a). Bath application of low concentrations
of TTX (5 nM) reduced the amplitude of the
peaks in these autocorrelations (Fig. 7b), indi-
Autocorrelation cating a reduction in IPSP-induced action
potential synchronization.

DISCUSSION
The results presented here show that IPSPs in
neocortical layer-5 pyramidal neurons are
amplified at depolarized membrane potentials.
This IPSP amplification was blocked by TTX
and intracellular QX-314, indicating that it is
Fig. 7. Effect of IPSP amplification on action potential firing. (a, b) Inhibition of action potential
mediated by voltage-activated sodium channels.
firing and subsequent IPSP-induced action potential synchronization in control (a) and in a low
concentration of TTX (5 nM; b). Top, inhibition of action potential firing by simulated IPSPs. The As this amplification is greatest at membrane
time of simulated IPSP onset is indicated by arrows. Upper middle, simulated IPSPs recorded at potentials close to action potential threshold,
resting (63 mV) and depolarized (55 mV) membrane potentials. Lower middle, post-stimulus IPSP amplification is likely to be important in
time histograms of action potential firing. Bottom, autocorrelation of spike timing during a 150- synaptic integration. Consistent with this idea,
ms time window starting after the peak of the IPSP. All recordings are from the same cell. IPSP amplification by voltage-activated sodium

148 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
channels was found to increase the ability of IPSPs to inhibit
action potential firing and to promote IPSP-induced action
potential synchronization.
Mechanisms underlying IPSP amplification
Amplification of IPSPs at depolarized membrane potentials pre-
sumably occurs because the hyperpolarization generated during
an IPSP is able to turn off a tonically active (non-inactivating)
inward sodium current, thereby generating a functional out-
ward current, which adds to the outward current occurring dur-
ing IPSPs, and so amplifies their hyperpolarization. IPSPs in
thalamocortical neurons are amplified via a similar mechanism,
but in this case by turning off a tonically active calcium current20.
The onset of IPSP amplification therefore depends on the rate
of deactivation of non-inactivating sodium channels, whereas
the duration of IPSP amplification depends on their rate of reac-
tivation. At depolarized membrane potentials, the increase in the
IPSP time to peak indicates that deactivation of non-inactivat-
1999 Nature America Inc. http://neurosci.nature.com
ing sodium channels takes several milliseconds near action poten-
tial threshold, whereas the significantly increased IPSP duration
indicates that reactivation of the same channels takes tens of mil-
liseconds. The extent and duration of IPSP amplification also
depends on the interaction of non-inactivating sodium channels
with Ih, such that larger and/or more prolonged inhibitory events
are less amplified. A role of Ih in influencing the duration of IPSPs
has been noted previously21.
The location of the sodium channels involved in IPSP ampli-
fication is unknown. Both the soma and dendrites of neocorti-
cal layer-5 pyramidal neurons contain voltage-activated sodium
channels22. The highest density of voltage-activated sodium chan-
nels in these neurons, however, is thought to be in the axon23,24.
Given that the non-inactivating sodium channels that contribute
to IPSP amplification are the same sodium channels involved in
action potential generation, switched into a different mode15,
inhibitory inputs onto the axon initial segment are likely to be
amplified locally by voltage-activated sodium channels to a
greater extent than inhibitory inputs onto the soma or dendrites.
This makes inhibitory axo-axonic synapses25 ideally placed to
inhibit action potential firing, both because of their location near
the site of action potential initiation26, and because their effect
would be significantly amplified at depolarized membrane poten-
tials by voltage-activated sodium channels. IPSPs generated by
inhibitory interneurons that target dendritic regions are also like-
ly to be amplified locally by dendritic sodium channels. Such an
effect presumably increases the ability of dendritic inhibitory
inputs to inhibit dendritic calcium electrogenesis27,28 and back-
propagation of action potentials into the dendritic tree29.
Physiological significance of IPSP amplification
The main physiological effect of amplifying IPSPs is to increase
their ability to block action potential firing. Several mechanisms
therefore act together to enhance inhibition at membrane poten-
tials near action potential threshold, as depolarization increases
both the driving force for chloride, and therefore the underlying
IPSC, and also leads to amplification of IPSPs via non-inactivat-
ing sodium channels. Strong depolarizations would also be
expected to increase the duration of GABAA-receptor-mediated
IPSCs30, although presumably this only has a significant effect
on IPSC duration at membrane potentials substantially more
depolarized than action potential threshold.
In addition to increasing the effectiveness of inhibition, ampli-
fication of IPSPs also enhanced their ability to synchronize action
potential firing. Action potential synchronization by IPSPs has
nature neuroscience volume 2 no 2 february 1999
articles
been reported in hippocampal CA1 pyramidal neurons31, where
this effect is thought to be associated with a rebound depolar-
ization after IPSPs mediated by activation of Ih. The present
results suggest that non-inactivating sodium channels are also
involved in action potential synchronization by IPSPs. Consis-
tent with this idea, action potential synchronization in other
regions of the cortex involves an interaction between both Ih and
non-inactivating sodium channels32.
As mentioned earlier, EPSPs in many neuronal types are also
amplified at depolarized membrane potentials by non-inacti-
vating sodium channels15. Although IPSP and EPSP amplifica-
tion occur through different mechanisms, with EPSP
amplification resulting from turning on non-inactivated sodi-
um channels, the effects on EPSP and IPSP time course and the
voltage-dependence of the amplification are similar12. Hence,
the same conductance amplifies both excitatory and inhibitory
synaptic events at depolarized membrane potentials in a similar
manner. Such an increase in synaptic gain at potentials near
action potential threshold is presumably important for synaptic
integration in all neurons that have a significant non-inactivating
sodium current.
METHODS
All recordings were made from the somata of visually identified layer-5
pyramidal neurons in 300-m sagittal brain slices from somatosensory
neocortex of 3 to 4 week-old rats33. During recordings, slices were con-
tinuously perfused with a solution of the following composition: 125
mM NaCl, 25 mM NaHCO 3, 25 mM glucose, 3 mM KCl, 1.25 mM
NaH2PO4, 2 mM CaCl2 and 1 mM MgCl2, bubbled with 95% O2/5%
CO2 (pH 7.4) at 35 1C.
Somatic perforated-patch and whole-cell, current-clamp recordings
were made using Axoclamp-2 amplifiers (Axon Instruments, Foster City,
California), whereas low-access, somatic, whole-cell, voltage-clamp
recordings (uncompensated series resistance < 10 M) were made using
an Axopatch 200A (Axon Instruments) with greater than 80% series
resistance compensation. Perforated-patch recordings were made using
gramicidin16. Pipette tips were filled with 150 mM KCl plus 10 mM
HEPES (pH 7.4 with KOH) and then backfilled with the same solution
containing gramicidin (9 g/ml) plus Lucifer Yellow (1 mg/ml). Perfo-
ration was detected by a slow reduction in series resistance to 3065 M.
During these experiments, fluorescence microscopy was used to deter-
mine if the recorded cell was stained with Lucifer Yellow (indicating rup-
ture of the patch membrane). If staining was detected, the data were
discarded. Whole-cell current- and voltage-clamp recordings were made
with pipettes filled with the following solution: 120 mM K-gluconate, 7
mM NaCl, 10 mM HEPES, 10 mM EGTA, 2 mM Na2-ATP and 2 mM
MgCl2, pH 7.3 with KOH. This solution has a chloride concentration
(11 mM) similar to the intracellular chloride concentration determined
from the reversal potential of IPSPs during gramicidin perforated-patch
recordings (see Results).
Monosynaptic IPSPs and IPSCs were evoked by extracellular stimu-
lation close to the soma (approximately 100 m lateral from the soma
of the recorded cell) in the presence of CNQX (10 M; Tocris, Bristol,
UK) and D-APV (50 M; Tocris), and in some cases also CGP 55845A
(1 M; gift from Novartis Pharma, Basel, Switzerland). Unless otherwise
stated, simulated IPSPs (see Results) were generated by current injection
via a second somatic recording pipette with an exponential rise and dou-
ble exponential decay (taurise = 0.67 ms; taudecay(fast) = 3.8 ms; taude-
cay(slow) = 34.3 ms; ratio of decay tau amplitudes fast:slow of 9:2). During
current clamp recordings, the membrane potential was altered by somat-
ic current pulses (500 ms duration; IPSPs evoked 200 ms after pulse
onset) or by constant current injection.
Voltage and current were filtered at 10 and 2 kHz, respectively, and
sampled at 10 to 20 kHz using a Macintosh computer together with an
analog-to-digital converter (Instrutech, Great Neck, New York). For
evoked IPSP/Cs, the stimulus artifact was either blanked or removed by
subtraction of the average response recorded at the reversal potential.
149
 1999 Nature America Inc. http://neurosci.nature.com
articles
IPSP/C peak amplitude was measured from a baseline set before the
IPSP/C, and the integral was defined as the area of an IPSP/C above or
below this baseline. IPSP/C peak and integral versus membrane or hold-
ing potential plots were fit with a modified version of the GHK equation:
Peak/integral = A V ([Cl]i - [Cl]o exp(VF/RT) )/(1 - exp(VF/RT))
where A is a constant, V is the membrane or holding potential, [Cl]i is
the intracellular chloride concentration, [Cl]o is the extracellular chlo-
ride concentration, and F, R and T have their usual meanings. All data
were analyzed with Axograph (Axon Instruments). Unless otherwise stat-
ed, no correction was made for the junction potential between pipette
and extracellular solutions. Averages at different membrane potentials
were compiled from 3 to 20 sweeps. Pooled data are expressed as mean
standard error, and statistical difference was evaluated by paired t-test at
a significance level of 0.05.
ACKNOWLEDGEMENTS
I thank J. Bekkers and S. Williams for discussions and comments on an earlier
version of the manuscript. This work was supported by an Australian Research
1999 Nature America Inc. http://neurosci.nature.com
Council QEII Research Fellowship, the Human Frontiers Science Program and
the Wellcome Trust.
RECEIVED 31 AUGUST; ACCEPTED 10 DECEMBER 1998
1. Johnston, D. & Wu, S. Foundations of Cellular Neurophysiology (MIT Press,
Cambridge, MA, 1995).
2. Hille, B. Ion Channels of Excitable Membranes (Sinauer, Sunderland, MA,
1994).
3. Johnston, D., Magee, J. C., Colbert, C. M. & Christie, R. Active properties of
neuronal dendrites. Annu. Rev. Neurosci. 19, 165186 (1996).
4. Masukawa, L. M. & Prince, D. A. Synaptic control of excitability in isolated
dendrites of hippocampal neurons. J. Neurosci. 4, 217227 (1984).
5. Stafstrom, C. E., Schwindt, P. C., Chubb, M. C. & Crill, W. E. Properties of
persistent sodium conductance and calcium conductance of layer V neurons
from cat sensorimotor cortex in vitro. J. Neurophysiol. 53, 153170 (1985).
6. Miles, R. & Wong, R. K. Excitatory synaptic interactions between CA3
neurones in the guinea-pig hippocampus. J. Physiol. (Lond.) 373, 397418
(1986).
7. Thomson, A. M., Girdlestone, D. & West, D. C. Voltage-dependent currents
prolong single-axon postsynaptic potentials in layer III pyramidal neurons in
rat neocortical slices. J. Neurophysiol. 60, 18961907 (1988).
8. Sutor, B. & Hablitz, J. J. EPSPs in rat neocortical neurons in vitro I.
Electrophysiological evidence for two distinct EPSPs. J. Neurophysiol. 61,
607620 (1989).
9. Sutor, B. & Hablitz, J. J. EPSPs in rat neocortical neurons in vitro II.
Involvement of N-methyl-D-aspartate receptors in the generation of EPSPs. J.
Neurophysiol. 61, 621634 (1989).
10. Deisz, R. A., Fortin, G. & Zieglgnsberger, W. Voltage-dependence of
excitatory postsynaptic potentials of rat neocortical pyramidal neurons. J.
Neurophysiol. 65, 371382 (1991).
11. Hirsch, J. & Gilbert, C. D. Synaptic physiology of horizontal connections in
the cat visual cortex. J. Neurophysiol. 11, 18001809 (1991).
12. Stuart, G. & Sakmann, B. Amplification of EPSPs by axosomatic sodium
channels in neocortical pyramidal neurons. Neuron 15, 10651077 (1995).
150
13. Lipowsky, R., Gillessen, T. & Alzheimer, C. Dendritic Na+ channels amplify
EPSPs in hippocampal CA1 pyramidal cells. J. Neurophysiol. 76, 21812191
(1996).
14. Urban, N. N., Henze, D. A. & Barrionuevo, G. Amplification of perforant-
path EPSPs in CA3 pyramidal cells by LVA calcium and sodium channels. J.
Neurophysiol. 80, 15581561 (1998).
15. Crill, W. E. Persistent sodium current in mammalian central neurons. Annu.
Rev. Physiol. 58, 349362 (1996).
16. Kyrozis, A. & Reichling, D. B. Perforated-patch recording with gramicidin
avoids artifactual changes in intracellular chloride concentration. J. Neurosci.
Methods 57, 2735 (1995).
17. Owens, D. F., Boyce, L. H., Davis, M. B. & Kriegstein, A. R. Excitatory GABA
responses in embryonic and neonatal cortical slices demonstrated by
gramicidin perforated-patch recordings and calcium imaging. J. Neurosci. 16,
64146423 (1996).
18. Davies, C. H., Pozza, M. F. & Collingridge, G. L. CGP 55845A: a potent
antagonist of GABAB receptors in the CA1 region of rat hippocampus.
Neuropharmacology 32, 10711073 (1993).
19. BoSmith, R. E., Briggs, I. & Sturgess, N. C. Inhibitory actions of ZENECA
ZD7288 on whole-cell hyperpolarisation activated inward current (If) in
guinea-pig dissociated sinoatrial node cells. Br. J. Pharmacol. 110, 343349
(1993).
20. Williams, S. R., Tth, T. I., Turner, J. P., Hughes, S. W. & Crunelli, V. The
window component of the low threshold Ca2+ current produces input signal
amplification and bistability in cat and rat thalamocortical neurones. J.
Physiol. (Lond.) 505, 689705 (1997).
21. van Brederode, J. F. & Spain, W. J. Differences in inhibitory synaptic input
between layer II-III and layer V neurons of the cat neocortex. J. Neurophysiol.
74, 11491166 (1995).
22. Stuart, G. J. & Sakmann, B. Active propagation of somatic action potentials
into neocortical pyramidal cell dendrites. Nature 367, 6972 (1994).
23. Mainen, Z. F., Joerges, J., Huguenard, J. R. & Sejnowski, T. J. A model of spike
initiation in neocortical pyramidal neurons. Neuron 15, 14271439 (1995).
24. Rapp, M., Yarom, M. & Segev, I. Modeling back propagating action potentials
in weakly excitable dendrites of neocortical pyramidal cells. Proc. Natl. Acad.
Sci. USA 93, 1198511990 (1996).
25. Buhl, E. H., Halasy, K. & Somogyi, P. Diverse sources of hippocampal unitary
inhibitory postsynaptic potentials and the number of synaptic release sites.
Nature 68, 823882 (1994).
26. Stuart, G., Spruston, N., Sakmann, B. & Husser, M. Action potential
initiation and backpropagation in neurons of the mammalian central
nervous system. Trends Neurosci. 20, 125131 (1997).
27. Kim, H. G., Beierlein, M. & Connors, B. W. Inhibitory control of excitable
dendrites in neocortex. J. Neurophysiol. 74, 18101814 (1995).
28. Miles, R., Tth, K., Gulyas, A. I., Hjos, N. & Freund, T. F. Differences
between somatic and dendritic inhibition in the hippocampus. Neuron 16,
815823 (1996).
29. Tsubokawa, H. & Ross, W. N. IPSPs modulate spike backpropagation and
associated [Ca2+]i changes in the dendrites of hippocampal CA1 pyramidal
neurons. J. Neurophysiol. 76, 28962906 (1996).
30. Collingridge, G. L., Gage, P. W. & Robertson, B. Inhibitory post-synaptic currents
in rat hippocampal CA1 neurones. J. Physiol. (Lond.) 356, 551564 (1984).
31. Cobb, S. R., Buhl, E. H., Halasy, K., Paulsen, O. & Somogyi, P.
Synchronisation of neuronal activity in hippocampus by individual
GABAergic interneurons. Nature 378, 7578 (1995).
32. Alonso, A. & Llins, R. R. Subthreshold Na+-dependent theta-like
rhythmicity in stellate cells of entorhinal cortex layer II. Nature 342, 175177
(1989).
33. Stuart, G. J., Dodt, H.-U. & Sakmann, B. Patch-clamp recordings from the
soma and dendrites of neurones in brain slices using infrared video
microscopy. Pflgers Arch. 423, 511518 (1993).
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

Retention of heroin and morphine-

6-glucuronide analgesia in a new
line of mice lacking exon 1 of MOR-1
Alwin G.P. Schuller1, Michael A. King2, Jiwen Zhang1, Elizabeth Bolan2, Ying-Xian Pan2,
Daniel J. Morgan1, Albert Chang2, Maureen E. Czick1, Ellen M. Unterwald3,4,5,
Gavril W. Pasternak2 and John E. Pintar1

1 Dept. of Neuroscience and Cell Biology, UMDNJ-Robert Wood Johnson Medical School, CABM, RM 326, 675 Hoes Lane,
Piscataway, New Jersey 08854, USA
2 The Cotzias Laboratory of Neuro-Oncology, Memorial Sloan-Kettering Cancer Center, 125 York Avenue, New York, New York 10021, USA
1999 Nature America Inc. http://neurosci.nature.com

3 Dept. of Psychiatry, NYU Medical Center, 550 First Avenue, New York, New York 10016, USA
4 Present address: Department of Pharmacology, Temple University School of Medicine, 3420 N. Broad Street,
Philadelphia, Pennsylvania 19140, USA
5 Present address: The Rockefeller University, 1230 York Avenue, New York, New York 10021, USA
Correspondence should be addressed to J.E.P. (pintar@mbcl.rutgers.edu)

Morphine produces analgesia by activating mu opioid receptors encoded by the MOR-1 gene.
Although morphine-6-glucuronide (M6G), heroin and 6-acetylmorphine also are considered mu opi-
oids, recent evidence suggests that they act through a distinct receptor mechanism. We examined
this question in knockout mice containing disruptions of either the first or second coding exon of
MOR-1. Mice homozygous for either MOR-1 mutation were insensitive to morphine. Heroin, 6-acetyl-
morphine and M6G still elicited analgesia in the exon-1 MOR-1 mutant, which also showed specific
M6G binding, whereas M6G and 6-acetylmorphine were inactive in the exon-2 MOR-1 mutant. These
results provide genetic evidence for a unique receptor site for M6G and heroin analgesia.

Morphine, one of the oldest and most widely used pain relievers,
elicits a variety of pharmacological responses, including analgesia,
respiratory depression and inhibition of gastrointestinal transit.
These actions are mediated by mu opioid receptors, which have been
divided pharmacologically into several subtypes1. Despite the need
for opiates in the treatment of pain, their abuse potential has limit-
ed both opiate availability and optimal use. The most widely abused
opioid is heroin, which is synthesized by acetylating morphine at
both the 3- and 6- positions. In vivo, heroin itself is very short-lived,
being rapidly converted into 6-acetylmorphine, which is thought to
mediate most of its actions2. It was initially thought that pharma-
cokinetic differences alone, rather than distinct mechanisms of
action, could account for the differences in heroin and morphine
responses26. However, recent studies indicate that heroin and 6-
acetylmorphine, as well as the major morphine metabolite mor-
phine-6-glucuronide (M6G), may act through an opioid receptor
distinct from that mediating morphine actions711. For example,
the CXBK mouse strain, developed in the early 1970s, is insensitive
to intracerebroventricular morphine injections12 consistent with
significantly lower mu1 binding levels13. In contrast, M6G, heroin
and its active metabolite 6-acetylmorphine remain potent analgesics
in CXBK mice10,14, suggesting that these agents may act on recep-
tor systems different from those activated by morphine. This possi-
bility is further supported by antisense studies that distinguish
specific G-protein subunits mediating the effects of morphine
(Gi2) from those involved with M6G (Gi1) in mice15 and rats16.
A single gene encoding the mouse mu opioid receptor
(MOR-1) has been cloned17,18, and several splice variants have
been identified19,20. Antisense probes against MOR-1 block mor-
phine analgesia in both rats7,16 and mice9,10, but detailed anti-
sense mapping of MOR-1 revealed striking differences between
morphine and M6G9,10,16. Probes targeting exon 1 all blocked
analgesia caused by morphine but not by M6G, whereas others
targeting exons 2 and 3 blocked M6G analgesia, raising the pos-
sibility that these receptors might be splice variants of a common
gene911,16. Gene knockouts offer an alternative approach to iden-
tifying the functional significance of specific genes, avoiding many
of the potential problems encountered with antisense. Multiple
disruptions of the MOR-1 gene show that the MOR-1 gene is
essential for morphine-induced analgesia as well as development
of morphine dependence2124. However, its role in mediating anal-
gesia elicited by other opioids such as heroin, 6-acetylmorphine
and M6G analgesia has not been systematically investigated.
Therefore, we determined the analgesic potential of these drugs
in two strains of MOR-1 knockout mice, in which either the first
or second coding exon of MOR-1 was disrupted.
RESULTS
Generation of MOR-1 exon-1 knockout mice
To define the role of MOR-1 in the actions of morphine, heroin and
M6G, we compared mice with disruptions in exon 2 of MOR-1,
described previously21, with a new mouse strain deficient in exon 1
of MOR-1. Exon-1-deficient mice were generated using a targeting
vector in which exon 1 was replaced by a neomycin resistance cassette

nature neuroscience volume 2 no 2 february 1999 151

 1999 Nature America Inc. http://neurosci.nature.com

articles

a H BE N EH B B HE H E E X B 1 kb N
b
MOR1 Locus 1
2.8 kb
7.9 kb

c
N EH B E H E E X
MOR1 Targeting Vector NEO HSV tk

MOR1 Targeted Allele H B E EH B E H E E X B

NEO
Int. probe 3' probe
2 kb
7.7 kb
1999 Nature America Inc. http://neurosci.nature.com

Fig.1. Creation of MOR-1 exon-1 mutant mice. (a) Schematic representation of the restriction map of the MOR-1 locus, targeting vector and the
MOR-1-targeted allele. Exon 1 is indicated by a black box. Restriction sites identified by Southern blot that are outside the MOR-1 genomic clone are
indicated in the dotted region of the upper panel. In the targeting vector, 2 kb including exon 1 of the MOR-1 gene has been replaced by a neomycin
resistance cassette. Successful targeting of the MOR-1 locus results in changes in expected BamH1 and EcoR1 fragment lengths (underlined) as
detected by respectively a 3 external and an internal probe (black bars). Positions of the upstream BamH1 (B), EcoR1 (E) and HindIII (H) are also
retained in the targeted allele. N, Not1; X, Xho1. (b) Southern-blot analysis using the 3 external probe on BamH1-digested ES DNA identifies one of
the targeted ES lines (+/). (c) Southern-blot analysis using the internal probe on EcoR1-digested tail DNA from offspring originating from MOR-1
heterozygous mating identifies homozygous (/), heterozygous (+/) and wild-type (+/+) mice.

(Fig. 1a). Introduction of the targeting vector into CCE embryonic
stem cells (ES) produced three ES lines carrying a targeted mu opi-
oid receptor allele, as determined by hybridizing BamH1-digested
ES cell DNA with an external probe located 3 of the sequences
included in the targeting construct (Fig. 1b). Targeted ES cells from
two individual clones were injected into blastocysts and both gave
rise to germline-transmitting chimeras. Heterozygous offspring were
mated to produce mice homozygous for the deleted allele (Fig. 1c).
To ensure that the region 5 of MOR-1 exon 1 was not affected by
the recombination event, tail DNA from MOR-1 wild-type and
homozygous mice was digested with BamH1, HindIII and EcoR1
and hybridized with a 3.4-kb Not-1/BamH1 fragment correspond-
ing to the 5 region of the MOR-1 gene. This strategy revealed the
alterations in EcoR1 and HindIII fragments expected from a homol-
ogous recombination event. In addition, all digests resulted in
hybridizing fragments derived from restriction sites 5 of the cloned
MOR-1 gene that were identical in wild-type and homozygous
mutant mice, indicating that this region was not altered by the
recombination event (data not shown). Offspring derived from het-
erozygous matings (n = 367) showed a Mendelian inheritance pat-
tern (24.5% wild-type, 51.8% heterozygous and 23.7% homozygous
mice), indicating the absence of embryonal or postnatal lethality of
homozygous mutant mice. Homozygous mutant mice were viable
and fertile, with no gross morphological abnormalities.
Receptor-binding assays revealed no detectable binding of the
mu receptor agonist [3H]DAMGO ([D-Ala2,MePhe4,Gly(ol)5]-
enkephalin) in the brains of the knockout mice, whereas
[3H]DAMGO labeling in heterozygous mice (mean standard error,
Bmax, 46 5.3 fmol/mg protein; Kd, 1.4 0.4, n = 4) was approxi-
mately half that found in wild-type mice (Bmax, 78.3 3.3 fmol/mg
protein; Kd, 1.3 0.3, n = 4). Binding of the tritiated delta-selective
compound [D-Pen2,D-Pen5]enkephalin ([3H]DPDPE) was similar
in all genotypes (data not shown), consistent with the lack of change
in mRNA levels for the delta DOR-1 receptor (see below). These
binding results are similar to those reported in other strains of mu
receptor knockout mice21,24. In situ hybridization with delta (DOR-
1), kappa (KOR-1) and orphan (ORL-1) receptor probes showed
no compensatory changes in mRNA levels derived from the cognate
genes in MOR-1-deficient mice. Similarly, proenkephalin and pro-
dynorphin mRNA levels did not differ significantly in mutant and
wild-type mice (data not shown). These results confirm and extend
previous observations21 in a new strain of MOR-1-deficient mice.
Opioid analgesia in exon-1 MOR-1 knockout mice
We next examined the effects of exon-1 disruption on morphine
analgesia in wild-type, heterozygous and homozygous knockout
mice using the tail-flick assay as described710,16,25. Wild-type
129SvEv/C57BL/6 mice of the F2 and F3 generations given fixed
morphine doses systemically, spinally or supraspinally (Fig. 2)
responded similarly to CD-1 mice. In contrast, morphine was vir-
tually inactive in the homozygous mutants, whereas the heterozy-
gotes had an intermediate response. Doseresponse curves
confirmed these initial observations (Table 1). The ED50 (effective
dose for 50% of the population) in the CD-1 and the wild-type mice
were indistinguishable. There was a small shift in the curve for het-
erozygotes, but morphine was inactive in the homozygous mice,
even at the highest dose tested (100 mg/kg, subcutaneous). We also
examined several additional mu analgesics. Systemic methadone
and supraspinal DAMGO were inactive in the knockout group at
doses that caused greater than 60% analgesia in wild-type mice
(Fig. 2a). The heterozygotes gave intermediate responses. The actions
of the kappa1 agonist U50,488H, the kappa3 analgesic naloxone ben-
zoylhydrazone (NalBzoH) and DPDPE were unaffected by the dis-
ruption of exon 1 of the MOR-1 gene. Thus, the elimination of exon
1 blocked the actions of three different mu analgesics, confirming
the importance of MOR-1 in traditional mu analgesia, without inter-
fering with kappa or delta analgesia. Furthermore, the intermediate
responses of the drugs in heterozygotes implies that the response to
the mu analgesics depended on gene dosage.
The exon-1-deficient mice responded quite differently to M6G

152 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

a Systemic b Supraspinal

(percent of mice)

(percent of mice)
Analgesia

Analgesia
*p < 0.001
*p < 0.005 **p < 0.002

Fig. 2. Analgesic action of opioids in wild-type, heterozygous or

homozygous mice. (a) Groups of mice (n = 8) received subcutaneous c Spinal
injections of morphine (13 mg/kg), methadone (2 mg/kg), U50,488H
(5 mg/kg) or NalBzoH (60 mg/kg) and were tested for analgesia 30

(percent of mice)
1999 Nature America Inc. http://neurosci.nature.com

min later. (b) Groups of mice (n = 9) received intracerebroventricular

Analgesia
injections of morphine (0.7 g), DAMGO (6 ng), M6G (12.5 ng) or 6-
acetylmorphine (1.2 g) and were tested for analgesia 15 min later.
(c) Groups of mice (n = 8) received intrathecal injections of mor-
phine (0.8 g), M6G (12.5 ng) or DPDPE (500 ng) and were tested for
analgesia 15 min later. Analgesia is expressed as the percentage of
mice responding. *p < 0.002

and heroin. Given systemically, M6G and heroin both retained their
analgesic activity in homozygous exon-1-deficient mice, with a slight
reduction in potency (Table 1). High doses of both drugs were anal-
gesic in all animals, confirming the lack of a ceiling effect. The anal-
gesic responses of supraspinal M6G and 6-acetylmorphine, as well as
spinal M6G, in homozygous mice did not differ significantly from
those in wild-type mice (Fig. 2b and c). The slight decrease in sen-
sitivity of systemic M6G and heroin in the exon-1 knockout animals
may reflect a small contribution of the traditional morphine recep-
tor to their overall analgesic activity.
M6G analgesia in exon-1 MOR-1 knockout mice
The pharmacological characteristics of M6G analgesia in both exon-
1-deficient and wild-type mice were indistinguishable from those
previously established in CD-1 mice. The mu-selective antagonists
-funaltrexamine (-FNA) and naloxonazine1 both effectively block
M6G analgesia in CD-1 mice26. Here both -FNA and naloxonazine
significantly antagonized M6G analgesia in all three groups of mice
(Fig. 3a). Analgesia induced by M6G was not blocked in any of the
three groups by the kappa1 antagonist nor-binaltorphimine (NorB-
NI) or by the delta antagonist naltrindole. The selective M6G antag-
onist 3-methoxynaltrexone (3-MeONtx; 0.25 mg/kg, subcutaneous),
which blocks M6G, heroin and 6-acetylmorphine analgesia at doses
that are inactive against morphine27, did not affect morphine-
induced analgesia in wild-type mice, but it effectively antagonized
M6G analgesia in all groups (Fig. 3b). Thus, the pharmacological
sensitivity of M6G analgesia in the exon-1 MOR-1 mutant mice is
identical to that found in wild-type and CD-1 mice.

a b
(percent of mice)
(percent of mice)

Analgesia
Analgesia

*p < 0.05
*p < 0.02
**p < 0.005

M6G

Fig. 3. Inhibition of M6G analgesia by selective opioid receptor antagonists. (a) Groups of mice (n = 5) received a fixed dose of intracerebroventric-
ular M6G (12.5 ng, i.c.v.) alone or with subcutaneous -FNA (40 mg/kg), naloxonazine (35 mg/kg), naltrindole (20 mg/kg) or NorBNI (10 mg/kg). -
FNA and naloxonazine were given 24 h before the agonist. The other antagonists were given 30 min before the analgesics. Analgesia is expressed as
the percentage of mice responding. -FNA and naloxonazine significantly lowered M6G analgesia, as determined by the Fischer exact test. (b) Groups
of mice received intracerebroventricular M6G (12.5 ng) alone (Control) or 15 min after subcutaneous 3-MeONTX (0.25 mg/kg). Anot her group of
wild-type mice received i.c.v. morphine (0.7 g) alone (Control) or with subcutaneous 3-MeONTX (0.25 mg/kg).

nature neuroscience volume 2 no 2 february 1999 153

 1999 Nature America Inc. http://neurosci.nature.com
articles
(percent in mice)
Analgesia
*p < 0.002
**p < 0.006
Fig. 4. Antisense effects of M6G analgesia. Groups of mice (n = 10)
received saline or the indicated oligodeoxynucleotide (5 g, i.c.v.) on
days 1, 3 and 5. On day 6, all mice were tested with M6G (12.5 ng, i.c.v.),
and analgesia was assessed. The antisense probe targeting exon 2 signif-
icantly blocked M6G analgesia.
1999 Nature America Inc. http://neurosci.nature.com
Antisense has also been used to distinguish the analgesic mech-
anisms of morphine and M6G analgesia811,16. We have demon-
strated that an antisense probe targeting exon 2 of MOR-1 blocked
M6G-induced, but not morphine-induced analgesia in CD-1
mice8,9,16. In this study, this same antisense probe blocked M6G
analgesia in wild-type and exon-1 MOR-1 mutant mice (Fig. 4). The
mismatch antisense probe was inactive, confirming the specificity
of the effect. Thus, M6G analgesia is pharmacologically indistin-
guishable in wild-type and exon-1 MOR-1 mutant mice.
The retention of M6G analgesia in exon-1-deficient mice, and
its continued sensitivity to mu receptor antagonists and MOR-1
antisense, raised the question of whether there may be an alterna-
tive mu receptor in these knockout mice. [3H]M6G labels traditional
mu receptors in brain and in CHO cells transfected with MOR-1
moderately well. However, the brain contains a high-affinity
[3H]M6G binding site of low abundance not observed in the trans-
fected CHO cells, which might correspond to the M6G receptor14.
The exon-1 MOR-1 mutant mice also had low levels of [3H]M6G
binding (Fig. 5). [3H]M6G binding was reliably detected in repeat-
ed experiments, which had to be done on fresh tissue because freez-
ing markedly diminished the binding levels. The binding was high
affinity, with a Bmax far lower than those typically observed for mu
receptors in wild-type mice. Nonlinear regression analysis of the
combined binding data yielded a Bmax value of 12.4 2.5 fmol/mg
protein with a KD value of 3.7 1.2 nM.
M6G and 6-Ac-morphine analgesia in mice lacking exon 2
The sensitivity of M6G analgesia to an exon-2 antisense probe in
both wild-type and exon-1-deficient mice, as well as in CD-1
mice8,9,16, strongly suggested that the M6G receptor contains exon 2
of MOR-1. To explore this possibility further, we next compared
M6G analgesia in mice with a disrupt-
ed exon 2 (ref. 21) to that in the exon-1-
deficient strain. M6G remained an Table 1. Analgesic activity of opioids in CD-1 and 129SvEv C57BL/6 mice.
effective analgesic in the exon-1 MOR- Drug CD-1
1 mutant mice, but it was inactive in the Morphine 5.4 (3.6, 8.2)
exon-2 MOR-1 mutant mice, even at a M6G 3.7 (2.4, 5.6)
dose over sevenfold greater than the
Heroin 1.5 (1.0, 2.0)
ED50 value in the exon-1-deficient mice
(Table 2). Similarly, 6-acetylmorphine ED50 values (mg/kg) with 95% confidence limits determined from at least three doses of each drug given sub-
was inactive in exon-2 MOR-1 mutant cutaneously in a cumulative doseresponse protocol (see Methods for doses). In the / group, morphine at
100 mg/kg did not induce analgesia in any of the mice.
mice at a dose over threefold greater
154
than its ED50 in exon-1-deficient mice (Table 2). These results
demonstrate that exon 2, but not exon 1, of the MOR-1 gene is
important for M6G and 6-acetylmorphine analgesia.
MOR-1 transcripts in exon-1-deficient mice
The above results suggested that the MOR-1 exon-2-containing tran-
script is necessary for M6G analgesia. To determine whether mRNA
transcripts containing exons 2 and 3 were present in exon-1-defi-
cient mice, we used reverse transcriptase/polymerase chain reactions
on RNA extracts from their brains (Fig. 6). Primers restricted to
exon 1 failed to reveal a product in the exon-1-deficient mice, but
not wild-type mice, confirming the deletion of this exon. However,
primers spanning exons 2 and 3 produced a transcript of the pre-
dicted size (~556 bp). The authenticity of this product was con-
firmed by sequencing in both directions. A second set of reactions on
extracts from the periaqueductal gray, again using primers spanning
exons 2 and 3 (position 621642 and 11201140), also produced a
band of the expected size with its identity also confirmed by sequenc-
ing (data not shown). Thus, expression of MOR-1 transcripts con-
taining exons 2 and 3 is retained in the exon-1-deficient mice.
DISCUSSION
Although heroin and M6G have long been considered traditional
mu analgesics that differed from morphine only in their pharma-
cokinetics26, recent studies suggest that these drugs may act through
an opioid receptor mechanism distinct from the one now known to
mediate morphine actions711,16. To genetically differentiate the
actions of morphine from those of heroin and M6G, we examined
the analgesic potential of these drugs in exon-1-deficient mice. As
expected, morphine was inactive in these mice, consistent with results
from other knockout mice containing mutations in either exon 1
(refs 22, 23), exon 2 (ref. 21) or exons 2 and 3 of MOR-1 (ref. 24).
However, M6G and heroin, as well as its active metabolite 6-acetyl-
morphine, retained their analgesic activity in the exon-1-deficient
mice. Pharmacologically, M6G analgesia in the exon-1-deficient ani-
mals was indistinguishable from that observed in wild-type mice.
Mu antagonists still reversed M6G analgesia in exon-1 MOR-1
mutant mice, whereas delta and kappa selective antagonists were
ineffective. Thus, M6G, heroin and 6-acetylmorphine analgesia can
be mediated, at least in part, through mu receptors that contain exon
2, but not exon 1, of MOR-1 and that are distinct from traditional
morphine receptors.
The blockade of M6G analgesia by an exon-2 antisense oligonu-
cleotide in the exon-1 MOR-1 mutant mice implied that the M6G
receptor is generated either from an alternative transcript from the
MOR-1 locus or from a distinct gene sharing sequence homology
with exon 2 of the MOR-1 gene. However, this latter possibility seems
quite unlikely because disruption of exon 2 of MOR-1 eliminates
M6G and 6-acetylmorphine analgesia (Table 2), as reported for a
knockout in which both exons 2 and 3 had been disrupted24. If the
M6G receptor were generated from a different locus, M6G and 6-
acetylmorphine analgesia should not have been lost in the exon-2-
+/+ +/ / ratio (/)/(+/+)
5.0 (2.6, 9.7) 6.3 (4.4, 9.2) > 100 > 19
3.8 (2.4, 6.0) 7.7 (4.5, 13.1) 8.5 (6.6, 10.8) 2.3
1.2 (0.7, 1.8) 2.7 (2.1, 3.5) 4.0 (3.4, 4.7) 2.7
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

[3H]M6G binding (cpm)

Exon 1 Exon 2 and 3

Fig. 6. RT-PCR of MOR-1 transcripts in exon-1-deficient mice. The pre-

[3H]M6G (nM)
Fig. 5. [3H]M6G binding in exon-1 MOR-1 mutant mice. Results are
1999 Nature America Inc. http://neurosci.nature.com
dicted sizes of the amplified cDNA fragments were 362 bp for exon
1and 556 bp for exons 2 and 3. Lanes 1 and 5 contain cDNA from exon-
1-deficient mice. Lanes 2 and 6 contain cDNA from wild-type mice.

Lanes 3 and 7 contain genomic DNA from wild-type mice. Lanes 4 and 8
from three independent replications. contain no added DNA.

deficient mice. Thus, the M6G receptor is probably encoded by an mice that is distinct from that responsible for morphine analgesia.
alternative transcript from the MOR-1 locus. The current data provide direct genetic evidence for a unique mech-
If the M6G receptor were an alternatively spliced variant of MOR- anism for the actions of heroin, 6-acetylmorphine and M6G, which
1, we should be able to detect the persistence of MOR-1 transcripts may prove valuable in the development of new therapeutics target-
in the mutant mice. In mRNA from the brains of exon-1-deficient ing pain and drug abuse.
mice, RT-PCR revealed no evidence of exon 1 . However, we did It is important to note that we did not target the promoter of
detect transcripts containing exons 2 and 3 with RT-PCR, using an MOR-1 in our construct and that this region remains intact, as
upstream primer in exon 2 and a downstream primer in exon 3. shown by Southern analysis. We predict that retention of an intact
Sequencing confirmed their identity. Although exon 1 in these promoter would be essential to generate the spliced RNA tran-
knockout mice was lost, alternative MOR-1 transcripts were still pre- scripts detected here and that disruption of this region could
sent. The low levels of [3H]M6G binding in the knockout mice sug- eliminate all transcripts and thus possibly yield a different anal-
gested that the abundance of these residual transcripts also might gesic phenotype. Subtle differences in targeting vectors could
be low, which is confirmed by our ability to detect them only with thus lead to different results.
RT-PCR and not with in situ hybridization or northern blots. There
is also evidence indicating that MOR-1-derived transcripts can be METHODS
translated in the MOR-1-deficient mice. For example, immunos- Gene targeting. Gene-targeting experiments were done essentially as
taining with antisera targeting a newly cloned alternative exon 4 of described30 . A MOR-1 genomic clone was isolated from a 129/ReJ genom-
MOR-1 (Y.-X.P., Catherine Abbadie, E.B., Amy Zuckerman, Jin Xu, ic library after screening with a digoxigenin-labeled mouse MOR-1 cDNA
Grace C. Rossi and G.W.P., unpublished data) demonstrates its con- probe containing exon 1 and part of exon 2, obtained from C. Evans, and
characterized by restriction mapping. A 3.4-kb Not1/BamH1 fragment, locat-
tinued expression in the exon-1-deficient animals. Although the ed 5 of exon 1 and a 6.5kb HindIII/Xho1 fragment corresponding to part
relationship of this new exon to the receptor responsible for M6G of intron 1 were subcloned into a targeting construct containing the
analgesia remains unknown, it indicates that MOR-1 transcripts are neomycin resistance cassette and the HSV-TK gene. The linearized target-
still present and can be translated in these exon-1-deficient mice. ing vector was introduced into ES cells by electroporation. DNA isolated
Persistent expression of alternative transcripts in knockout mice has from G418 and gancyclovir double-resistant ES clones was digested with
been observed in other systems. For example,
initial characterization of neuronal nitric oxide
synthase (nNOS) knockout mice revealed no Table 2. Analgesic activity of M6G and 6-acetylmorphine in MOR-1-deficient
detectable nNOS mRNA or protein28. How- mice.
ever, subsequent investigations uncovered the Drug Analgesic response
persistent expression of low levels of alterna-
Exon 1 deficient Exon 2 deficient
tively spliced transcripts in these mutant
29 +/+ / +/+ /
mice . Exon 1 encodes the first transmem-
brane spanning region of the MOR-1 recep- M6G 12.5 ng 90% 50% 80% 10% p < 0.003
tor. Because G-protein-coupled receptors all 62.5 ng N.D. 100% N.D. 0%
contain seven transmembrane regions, the 6-Acetylmorphine 1.2 g 80% 70% 75% 0% p < 0.002
M6G receptor presumably contains an alter- 3.6 g N.D. N.D. N.D. 0%
native exon 1 to maintain the structure.
Percentage of mice (n = 10, except for the homozygous exon-2-deficient mice receiving 1.2 g of 6-
Together, these observations indicate that acetylmorphine, where n = 8) that showed analgesia by the tail-flick assay after receiving the indicated
M6G analgesia in the exon-1-deficient ani- drugs by intracerebroventricular injection. The responses of exon-2-deficient mice were significantly dif-
mals was mediated, at least in part, through a ferent between the +/+ and / groups, as indicated by the Fisher exact test. Doses that were not
receptor mechanism present in wild-type examined are designated as N.D. (not determined).

nature neuroscience volume 2 no 2 february 1999 155

 1999 Nature America Inc. http://neurosci.nature.com
articles
BamH1 and analyzed by genomic Southern blotting using a 0.8-kb
Xho1/BamH1 fragment located 3 of the sequences included in the target-
ing vector. Two different lines of ES cells carrying a targeted MOR-1 allele
as determined by a 7.9-kb wild-type and a 7.7-kb targeted BamH1 fragment,
were injected into C57Bl6 blastocysts to generate chimeras. Offspring of
germline-transmitting chimeras were genotyped as described above or by
hybridizing EcoR1-digested tail DNA with a 1.2-kb EcoR1/BamH1 probe
revealing a 2.8-kb wild-type and 2-kb targeted EcoR1 fragment.
Binding assay and RT-PCR. [3H]M6G binding was measured in brain mem-
brane fractions from MOR-1-deficient mice as described14. For RT-PCR,
total RNA from the brains of exon-1-deficient mice were extracted and
reverse transcribed using random hexamers to generate a first-strand cDNA.
To detect MOR-1 exon-1 expression, two primers were designed from the
nucleotide sequence of mouse MOR-1 (GenBank submission #U26915) at
positions 195213 (sense) and 531556 (antisense). To determine expres-
sion of exons 2 and 3, we used one primer set based on sequences 624648
(sense) and 11601180 (antisense) and another based on sequences 621642
and 11201140 (not shown). Samples were heated to 94oC and cycled 35
times through 94oC for 30 s, 63oC for 30 s, 72oC for 90 s. The final incuba-
1999 Nature America Inc. http://neurosci.nature.com
tion was at 72oC for 2 min. The mixture was then run on an agarose gel with
the indicated markers.
Tail-flick assay. Analgesia was assessed in the tail-flick assay as a doubling or
greater of the baseline latencies of each mouse, which ranged from 23 s. A
maximal latency of 10 s was used to minimize tissue damage. Comparisons
were made using the Fisher exact test. Drugs were given under halothane
anesthesia by a 2-l intracerebroventricular injection.
Cumulative dose-response curves. Morphine doses tested were 3, 6, 9, 10,
13, 20, 50 or 100 mg/kg. M6G doses were 2.2, 4.4, 6.6, 8, 8.8, 11 or
13.2 mg/kg. Heroin doses were 0.8, 1.6, 2.4, 3.2, 4, 4.8, 5.6 or 6.4 mg/kg.
Cumulative doseresponse curves were determined by giving all mice the
lowest opioid dose. Mice that showed analgesia were removed from further
testing, and the remainder were given the next higher dose. Again, mice
showing analgesia were removed, and additional doses were given to the
remaining mice. ED50 values with 95% confidence limits were determined
using the Bliss program, as described710. Values for morphine in CD-1 mice
determined by cumulative doseresponse curves do not differ significantly
from those obtained with separate groups of +/+ mice from the MOR-1
knockout receiving each morphine dose.
Antisense experiments. The probe targeting exon 2 was TTGGTG-
GCAGTCTTCATTTTGG (572-593) and the mismatch was CGCCCC-
GACCTCTTCCCTT (195213). Both probes have been used previously in
CD-1 mice79.
ACKNOWLEDGEMENTS
The authors thank Rhuna Shen and Qian Ye for their contributions in mapping the
mu opioid receptor gene and constructing the targeting vector. We thank Chris
Evans, Jim Douglass and Grahaem Bell for providing probes for DOR-1,
proenkephalin/prodynorphin and KOR-1, respectively, Z.P. Chen for providing the
mouse ORL in situ probe, and Liz Robertson for providing CCE ES cells. We also
thank Brigitte Kieffer and Hans Matthes for providing the exon-2 MOR-1 mutant
mice. This work was supported by grants from the National Institute on Drug Abuse
to J.E.P. (DA-09040 and DA-08622), Y.-X.P. (DA-00296) and G.W.P. (DA-07241,
DA-02615 and a Research Scientist Award DA-00220) and a core grant from the
National Cancer Institute to Memorial Sloan-Kettering Cancer Center (CA-08748).
D.J.M. was supported by NIH training grant R25-GM-55145.
RECEIVED 2 NOVEMBER; ACCEPTED 11 DECEMBER 1998
1. Pasternak, G. W. Pharmacological mechanisms of opioid analgesics. Clin.
Neuropharmacol. 16, 118 (1993).
156
2. Way, E. L., Young, J. & Kemp, J. Metabolism of heroin and its
pharmacological implications. Bull. Narcotics 17, 2533 (1965).
3. Way, E., Kemp, J., Young, J. & Grasetti, D. R. The pharmacologic effects of
heroin in relationship to its rate of biotransformation. Pharmacol. Exp. Ther.
129, 144154 (1960).
4. Oldendorf, W. H., Hyman, S., Braun, L. & Oldendorf, S. Z. Blood-brain
barrier: penetration of morphine, codeine, heroin, and methadone after
carotid injection. Science 178, 984986 (1972).
5. Inturrisi, C. E. et al. The pharmacokinetics of heroin in patients with chronic
pain. N. Engl. J. Med. 310, 12131217 (1984).
6. Inturrisi, C. E. et al. Evidence from opiate binding studies that heroin acts
through its metabolites. Life Sci. 33 Suppl 1, 773776 (1983).
7. Rossi, G., Pan, Y.-X., Cheng, J. & Pasternak, G. W. Blockade of morphine
analgesia by an antisense oligodeoxynucleotide against the mu receptor. Life
Sci. 54, PL375379 (1994).
8. Rossi, G. C., Standifer, K. M. & Pasternak, G. W. Differential blockade of
morphine and morphine-6 beta-glucuronide analgesia by antisense
oligodeoxynucleotides directed against MOR-1 and G-protein alpha subunits
in rats. Neurosci. Lett. 198, 99102 (1995).
9. Rossi, G. C., Pan, Y.-X., Brown, G. P. & Pasternak, G. W. Antisense mapping
the MOR-1 opioid receptor: evidence for alternative splicing and a novel
morphine-6 beta-glucuronide receptor. FEBS Lett. 369, 192196 (1995).
10. Rossi, G. C., Brown, G. P., Leventhal, L., Yang, K. & Pasternak, G. W. Novel
receptor mechanisms for heroin and morphine-6 beta-glucuronide analgesia.
Neurosci. Lett. 216, 14 (1996).
11. Pasternak, G. W. & Standifer, K. M. Mapping of opioid receptors using
antisense oligodeoxynucleotides: correlating their molecular biology and
pharmacology. Trends Pharmacol. Sci. 16, 344350 (1995).
12. Pick, C. G., Nejat, R. J. & Pasternak, G. W. Independent expression of two
pharmacologically distinct supraspinal mu analgesic systems in genetically
different mouse strains. J. Pharmacol. Exp. Ther. 265, 166171 (1993).
13. Moskowitz, A. S. & Goodman, R. R. Autoradiographic distribution of mu1
and mu2 opioid binding in the mouse central nervous system. Brain Res. 360,
117129 (1985).
14. Brown, G. P. et al. 3H-morphine-6beta-glucuronide binding in brain
membranes and an MOR-1-transfected cell line. J. Pharmacol. Exp. Ther. 282,
12911297 (1997).
15. Standifer, K. M., Rossi, G. C. & Pasternak, G. W. Differential blockade of
opioid analgesia by antisense oligodeoxynucleotides directed against various
G protein alpha subunits. Mol. Pharmacol. 50, 293298 (1996).
16. Rossi, G. C. et al. Antisense mapping of MOR-1 in rats: distinguishing
between morphine and morphine-6beta-glucuronide antinociception. J.
Pharmacol. Exp. Ther. 281, 109114 (1997).
17. Reisine, T. & Bell, G. I. Molecular biology of opioid receptors. Trends
Neurosci. 16, 506510 (1993).
18. Uhl, G. R., Childers, S. & Pasternak, G. An opiate-receptor gene family
reunion. Trends Neurosci. 17, 8993 (1994).
19. Zimprich, A., Simon, T. & Hollt, V. Cloning and expression of an isoform of
the rat mu opioid receptor (rMOR1B) which differs in agonist induced
desensitization from rMOR1. FEBS Lett. 359, 142146 (1995).
20. Bare, L. A., Mansson, E. & Yang, D. Expression of two variants of the human
mu opioid receptor mRNA in SK-N-SH cells and human brain. FEBS Lett.
354, 213216 (1994).
21. Matthes, H. W. et al. Loss of morphine-induced analgesia, reward effect and
withdrawal symptoms in mice lacking the mu-opioid-receptor gene. Nature
383, 819823 (1996).
22. Sora, I. et al. Opiate receptor knockout mice define mu receptor roles in
endogenous nociceptive responses and morphine-induced analgesia. Proc.
Natl. Acad. Sci. USA 94, 15441549 (1997).
23. Tian, M. et al. Altered hematopoiesis, behavior, and sexual function in mu
opioid receptor-deficient mice. J. Exp. Med. 185, 15171522 (1997).
24. Loh, H. H. et al. Mu opioid receptor knockout in mice: effects on ligand-
induced analgesia and morphine lethality. Mol. Brain Res. 54, 321326
(1998).
25. Chen, X. H. et al. An antisense oligodeoxynucleotide to mu-opioid receptors
inhibits mu-opioid receptor agonist-induced analgesia in rats. Eur. J.
Pharmacol. 275, 105108 (1995).
26. Paul, D., Standifer, K. M., Inturrisi, C. E. & Pasternak, G. W. Pharmacological
characterization of morphine-6 beta-glucuronide, a very potent morphine
metabolite. J. Pharmacol. Exp. Ther. 251, 477483 (1989).
27. Brown, G. P. et al. 3-Methoxynaltrexone, a selective heroin/morphine-6beta-
glucuronide antagonist. FEBS Lett. 412, 3538 (1997).
28. Huang, P. L., Dawson, T. M., Bredt, D. S., Snyder, S. H. & Fishman, M. C.
Targeted disruption of the neuronal nitric oxide synthase gene. Cell 75,
12731286 (1993).
29. Brenman, J. E. et al. Interaction of nitric oxide synthase with the postsynaptic
density protein PSD-95 and alpha1-syntrophin mediated by PDZ domains.
Cell 84, 757767 (1996).
30. Joyner, A. L. Gene Targeting: A Practical Approach (Oxford Univ. Press,
Oxford, 1993).
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

SOD1 rescues cerebral endothelial

dysfunction in mice overexpressing
amyloid precursor protein
Costantino Iadecola1, Fangyi Zhang1, Kiyoshi Niwa1, Chris Eckman2, Sherry K. Turner3,
Elizabeth Fischer4, Steven Younkin2, David R. Borchelt5, Karen K. Hsiao1 and George A. Carlson3

1 Department of Neurology, University of Minnesota, 420 Delaware Street S.E., Minneapolis, Minnesota 55455, USA
2 Mayo Clinic Jacksonville, 4500 San Pablo Road, Jacksonville, Florida 32224, USA
3 McLaughlin Research Institute, 1520 23rd Street South, Great Falls, Montana 59405, USA
4 Rocky Mountain Laboratory, National Institute of Allergy and Infectious Diseases, 903 South 4th Street, Hamilton, Montana 59840, USA
1999 Nature America Inc. http://neurosci.nature.com

5 Division of Neuropathology, Johns Hopkins University School of Medicine, 720 Rutland Avenue, Baltimore, Maryland 21205, USA
Correspondence should be addressed to C.I. (iadec001@tc.umn.edu)

Peptides derived from proteolytic processing of the -amyloid precursor protein (APP), including
the amyloid- peptide, are important for the pathogenesis of Alzheimers dementia. We found that
transgenic mice overexpressing APP have a profound and selective impairment in endothelium-
dependent regulation of the neocortical microcirculation. Such endothelial dysfunction was not
found in transgenic mice expressing both APP and superoxide dismutase-1 (SOD1) or in APP
transgenics in which SOD was topically applied to the cerebral cortex. These cerebrovascular effects
of peptides derived from APP processing may contribute to the alterations in cerebral blood flow
and to neuronal dysfunction in Alzheimers dementia.

There is substantial evidence that peptides derived from proteolyt-
ic processing of the -amyloid precursor protein, including the amy-
loid- peptide (A), are involved in the pathogenesis of Alzheimers
dementia1. However, the mechanisms by which these peptides exert
their pathogenic effects remain undefined2. Generation of reactive
oxygen species by A has been proposed as a potential link between
A and neurotoxicity36. Studies in isolated blood vessels and in
endothelial cell cultures have suggested that peptides derived from
APP processing produce deleterious vascular effects that may con-
tribute to neurodegeneration in Alzheimers dementia710. Addition
of micromolar quantities of synthetic A peptide to rings of rat aorta
causes loss of vasodilation in response to acetylcholine and increased
contractility in response to vasoconstrictors, effects caused by severe
damage to the endothelial lining of the vascular rings7. These toxic
effects of A are prevented by addition of the superoxide-scavenging
enzyme SOD. However, it is not known whether nanomolar eleva-
tions in A comparable to those occurring in Alzheimers demen-
tia result in superoxide-dependent alterations of cerebrovascular
regulation in vivo. Therefore, we used transgenic mice overexpress-
ing APP to explore whether elevation of the levels of APP and A
alter cerebrovascular reactivity and, if so, whether the effect is medi-
ated by reactive oxygen species.
RESULTS
Crossing mice from a transgenic line Tg1130H expressing high
copy numbers of the human mutant APP11 with an outbred stock
overexpressing human SOD1 (ref. 12) provided littermates express-
ing APP alone (APP+/SOD1), SOD1 alone (APP/SOD1+) or both
transgenes (APP+/SOD1+), along with transgene-negative con-
trols (APP/SOD1). For cerebrovascular studies, cerebral blood
flow (CBF) was continuously monitored in the exposed parietal
cortex by a laser-Doppler probe. Resting CBF did not differ among
the groups of mice studied (p > 0.05; data not shown). Topical
neocortical application of acetylcholine, a neurotransmitter that
increases CBF by activating endothelial muscarinic receptors and
releasing endothelial nitric oxide 1315, increased CBF in
APP/SOD1 mice (Fig. 1a). This increase was substantially atten-
uated in littermates expressing APP but not SOD1 (p < 0.001).
The reduction in acetylcholine-induced vasodilation was related
to reactive oxygen species, because the altered response was not
found in mice overexpressing both APP and SOD1. Similarly, the
increase in CBF produced by bradykinin, a vasodilator that acts
through specific endothelial receptors and cyclooxygenase prod-
ucts1517, or by the calcium ionophore A23187, a compound that
produces endothelium-dependent vasodilation through receptor-
independent mechanisms14,18, was also attenuated in APP+/SOD1
mice (Fig. 1b and c). Again, this reduction in vasodilation was not
found in APP+/SOD1+ mice. In contrast to endothelium-depen-
dent cerebrovascular responses, increases in CBF produced by
vasodilators that do not act through the endothelium, such as the
nitric oxide donor S-nitroso-N-acetylpenicillamine (SNAP) or
hypercapnia19, were not reduced in APP+/SOD1 mice (Fig. 2).
These endothelial-cell-independent responses did not differ sig-
nificantly among APP+/SOD1+, APP/SOD1+ and APP/SOD1
mice. Thus, APP overexpression influences only vasodilator
responses mediated through the endothelium.
We then investigated the cerebrovascular effect of the throm-
boxane A2 analogue U46619, a potent vasoconstrictor that acts
directly on vascular smooth muscles20,21. The decrease in CBF elicit-
ed by U46619 was exacerbated in mice overexpressing APP, but not

nature neuroscience volume 2 no 2 february 1999 157

 1999 Nature America Inc. http://neurosci.nature.com

articles

a Acetylcholine b Bradykinin To examine the possibility that the nor-

mal response to endothelium-dependent
vasodilators in APP+/SOD1+ mice resulted
CBF (% increase)

CBF (% increase)
from reduced A levels, we measured
human A peptides 140 and 142 in the
brains of APP+/SOD1 and APP+/SOD1+
mice11,23. In APP+/SOD1+ mice, levels of
both forms of A were slightly, but signifi-
cantly, lower than those of APP+/SOD1
mice (Fig. 3). To determine whether the lack
of endothelial dysfunction in APP+/SOD1+
U46619 mice could be attributed to this reduction
A23187
c d in A, we studied a line of wild-type human
CBF (% reduction) APP transgenic mice (Tg6209) that express-
es less transgene-encoded protein than
CBF (% increase)

Tg1130H (ref. 11). Despite their lower A

levels (Fig. 3), Tg6209 mice showed alter-
ations in endothelium-dependent cere-
1999 Nature America Inc. http://neurosci.nature.com

brovascular responses indistinguishable

from those of Tg1130H APP+/SOD1 mice
(percent CBF increases: acetylcholine, APP
27 1, APP + 8 1; bradykinin, APP
44 4, APP+ 18 2; A23187, APP 49 2,
APP + 14 5; n = 56 per group). Thus, the
Fig. 1. Effects of endothelium-dependent vasodilators and vasocontrictors on cerebral blood flow
+
in nontransgenic mice (APP /SOD1 ) and in littermates overexpressing APP (APP /SOD1 ), SOD1 protection exerted by SOD1 overexpression
(APP/SOD1+) or both APP and SOD1 (APP+/SOD1+). (a) Acetylcholine (10 M). (b) Bradykinin in APP mice cannot be attributed to SOD1-
(50 M). (c) A23187 (3 M). (d) The vasoconstrictor U46619 (1 M). Data were collected from related reduction in A concentration.
58 mice per group. *p < 0.0010.003, analysis of variance and Tukeys test. Synthetic A has been reported to pro-
duce endothelial damage in isolated arter-
ies 79. There is no evidence for
cerebrovascular abnormalities in Tg1130H
in APP+/SOD1+ mice (Fig. 1d). SOD1 overexpression alone did not mice at the light-microscope level, with the only consistent pathol-
alter responses to vasodilators and vasoconstrictors (Figs. 1 and 2). ogy being hypertrophic corticolimbic astrocytic gliosis11. We there-
These results indicate that APP overexpression leads to attenuation fore used electron microscopy to determine whether endogenous
of endothelium-dependent cerebrovascular vasodilation, and to A produced cerebral endothelial damage in vivo. The morpholo-
enhancement of vasoconstriction by mechanisms involving reactive gy of endothelial cells was not grossly altered in APP+/SOD1 mice
oxygen species. (Fig. 4). Therefore, in contrast to previous investigations in which
Expression of SOD1 transgenes also provides protection against high concentrations of synthetic A were used in isolated vessels7,
22
APP-induced premature death . Twenty-five of thirty-one we find that the disturbance in endothelium-dependent vasodila-
APP+/SOD1 mice died within 150 days of birth, whereas 19 tion in the transgenic mice results from perturbation of endothelial
APP+/SOD1+ mice survived for at least 1 year, and 1 mouse died at cell function and not from endothelial cell destruction.
167 days. This difference in survival is statistically significant The observation that endothelium-dependent cerebrovascular
(p < 0.001; Fishers exact test). responses are intact in transgenic mice expressing both APP and

SNAP Hypercapnia
a b
CBF (% increase)
CBF (% increase)

n=8

Fig. 2. Effect of endothelium-independent vasodilators S-nitroso-N-acetylpenicillamine (SNAP) or hypercapnia (pCO2 = 5060 mmHg) in nontrans-
genic mice and in littermates overexpressing APP, SOD1 or both APP and SOD1.

158 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
Tg1130/SOD1- (n = 6)
Tg1130/SOD1+ (n = 9)
Tg6209/SOD1- (n = 8)
A (pmol/g)
A (140) A (142)
Fig. 3. Brain A levels in Tg1130H-derived APP+/SOD1 or
APP+/SOD1+, and in transgene-positive FVB/N Tg6209 mice. *p < 0.01
from Tg1130/SOD1; #p < 0.001 from both Tg1130/SOD1 and
Tg1130/SOD1+ (analysis of variance and Tukeys test).
1999 Nature America Inc. http://neurosci.nature.com
SOD1 implicates reactive oxygen species in the mechanism of the
endothelial dysfunction. To study the possible source of superoxide,
we tested whether the cerebrovascular dysfunction could be reversed
by local treatment of the cerebral cortex with SOD, an enzyme that
does not easily cross cell membranes24. In APP+ mice (Tg6209),
SOD superfusion completely reversed the attenuation of the CBF
increase evoked by acetylcholine without affecting the response to
hypercapnia (Fig. 5). In APP littermates, SOD superfusion did not
influence vasodilator responses to acetylcholine or hypercapnia
(Fig. 5). SOD superfusion did not affect resting CBF in either group
of mice (p > 0.05; data not shown).
DISCUSSION
We have demonstrated that APP overexpression in living mice leads
to a profound and selective alteration in endothelial vascular regu-
lation. The cerebral vascular reactivity to acetylcholine, bradykinin
and A23187 are all reduced in APP mice, suggesting that APP over-
expression leads to a global impairment in endothelium-dependent
responses rather than to an alteration of a specific endothelial recep-
tor or vasodilator mechanism. The impairment in endothelium-
dependent vasodilation would be expected to render cerebral blood
vessels more susceptible to vasoconstriction. Consistent with this
hypothesis, the response to the vasoconstrictor U46619 is enhanced
in mice overexpressing APP.
The fragment of APP that is involved in the mechanisms of
this endothelial dysfunction remains to be identified. The obser-
vation that the abnormality in endothelium-dependent vasodi-
lation is present both in Tg1130H and Tg6209 mice, despite a
twofold difference in A(140) levels, suggests that this APP frag-
ment may not mediate the endothelial dysfunction. However,
A(142) is not substantially reduced in Tg6209, and this APP
fragment could conceivably be involved. Transgenic mice with
deletions and mutations of candidate APP regions will have to
be studied to determine which APP domain is responsible for
vascular dysfunction.
Although insufficient to account for SOD1-mediated protec-
tion against vascular dysfunction and against premature death,
the reduction in A peptide concentration by SOD1 expression
is noteworthy. It is not clear whether SOD1 alters the processing of
APP so that less A is produced or whether A peptide is removed
more efficiently. Both the short (140) and more amyloidogenic
long (142) forms of A peptide are reduced by SOD1 overex-
pression. The ratio of the two forms is unchanged, a result more
nature neuroscience volume 2 no 2 february 1999
articles
compatible with a reduction in normal processing than with alter-
ations in peptide catabolism.
The finding that the endothelial dysfunction can be rescued by
exogenous SOD or prevented by SOD1 overexpression suggests that
reactive oxygen species are involved in mediating the effect. Super-
oxide anions could be produced either intracellularly, for example,
by activation of the receptor for advanced glycation end products5, or
extracellularly by the endothelial NADH-dependent oxidoreductase,
which is membrane bound25. The finding that both SOD1, expressed
intracellularly in endothelial cells of SOD1 transgenics26, and exoge-
nous SOD, which does not cross cell membranes24, are protective
suggests that superoxide is not generated within endothelial cells.
This hypothesis is supported by two observations in APP transgen-
ics. First, endothelial cells do not show histochemical markers or
ultrastructural evidence of free-radical-mediated damage (G. Perry
and K.K. H., unpublished observations, ref. 3 and Fig. 4). Second,
the endothelial dysfunction is not irreversible because it can be res-
cued by exogenous SOD. However, vascular production of reactive
oxygen species remains to be demonstrated in APP transgenics.
There is ample evidence that reactive oxygen species result in
cerebral endothelial dysfunction, and they are thought to be respon-
sible for loss of endothelium-dependent vasodilation in atheroscle-
rosis, diabetes and hypertension2730. However, the mechanisms of
this effect are not understood31. Considering that a common
denominator in the vasodilator effect of acetylcholine, bradykinin
and A23187 is an increase in endothelial calcium31, superoxide and
other radicals could perturb endothelial calcium homeostasis and
impair production of endothelial vasoactive factors.
Overexpression of APP in the FVB/N strain of mice leads to pre-
mature death11,22. The observation that SOD1 protects against the
lethal effects of APP overexpression could provide a link between
reactive oxygen species and early mortality22. Although SOD1
expression prevents both cerebrovascular dysfunction and premature
death, the relationship, if any, between the two phenomena remains
to be determined. Although FVB/N mice could be more sensitive
to induction of superoxide-dependent cerebrovascular dysfunction
c e
a
b d f
Fig. 4. Electron micrographs of thin sections from parietal cortex of mice.
The morphology of endothelial cells (arrowheads) in transgene-negative
control mice (a) is comparable to that of APP/SOD1+ (b), APP+/SOD1
(c and d) or APP+/SOD1+ mice (e and f). Scale bar, 1.0 m.
159
 1999 Nature America Inc. http://neurosci.nature.com
articles
a
Acetylcholine
CBF (% increase)
APP+ APP
Fig. 5. Effect of topical application of SOD to the cerebral cortex on CBF responses in Tg6209
choline. (b) Response to hypercapnia. Data were collected from 56 mice per group. * p < 0.001, t-test.
1999 Nature America Inc. http://neurosci.nature.com
than other mouse strains, it is also possible that the involvement of
reactive oxygen species is not related to their cerebrovascular effects.
The findings also raise the possibility that cerebrovascular
endothelial dysregulation may contribute to the neuronal dysfunc-
tion observed in APP+ mice11. Vascular dysregulation may impair
brain function by altering the balance between flow-dependent sub-
strate delivery and energy demands imposed by neural activity. In
addition, the endothelial dysfunction may disrupt the blood-brain
barrier and impair the delivery to the brain of critical substrates. In
support of this possibility, subtoxic concentrations of A reduce the
transfer of glucose across endothelial cells, an effect likely to be medi-
ated by reactive oxygen species10. In addition, the vascular dysreg-
ulation induced by APP overexpression makes the brain more
susceptible to the effects of cerebral ischemia32. Thus, cerebrovas-
cular dysfunction resulting from APP overexpression renders the
brain more vulnerable to injury and could amplify the pathogenic
process in Alzheimers dementia.
It is well established that Alzheimers dementia and aging are
associated with alterations in cerebral hemodynamics and cere-
brovascular regulation33,34. Resting CBF is decreased and the vas-
cular reactivity to selected vasodilator stimuli is reduced in
Alzheimers dementia33,35,36. There is evidence that endothelium-
dependent vascular relaxation is impaired in isolated cerebral ves-
sels from elderly humans and in cerebral arterioles from aged
rats37,38. Although the CBF alterations in Alzheimers dementia could
be secondary to neuronal degeneration or to morphological alter-
ations in cerebral blood vessels, these factors are less likely to be
involved in the cerebrovascular changes occurring early in the disease
when pathological changes have not yet occurred39. Our results sug-
gest that alteration of endothelial regulation of the cerebral circula-
tion could be an important factor in the mechanisms of the
cerebrovascular dysfunction in Alzheimers dementia and aging.
Furthermore, our finding that exogenous SOD rescues the endothe-
lial dysfunction in APP transgenics adds further support to the valid-
ity of antioxidant treatment in Alzheimers dementia40.
METHODS
Transgenic mice. The APP and SOD1 transgenic mice used in these studies
have been described11,12. Tg1130H mice overexpress the human APP695
isoform with the V717I familial Alzheimers dementia mutation, along with
V721A and M722V changes. Tg6209 express wild-type human APP695.
Both transgenes have a 3-myc epitope tag. A hamster prion protein gene-
derived cosmid vector41 was used to drive expression of APP in these trans-
genic lines, both of which were produced on an inbred FVB/N background.
160
Hypercapnia
b
CBF (% increase)
APP+ APP
and APP+ APP littermates. (a) Response to acetyl-
SOD1 mice (line Tg(SOD1)76) were constructed with a 12-kb genomic clone
containing the entire human gene12. Offspring of Tg(APP) Tg(SOD1)
mice were typed for the presence of the transgenes as described22. All exper-
iments were done on littermates that shared genetic background except for
the transgene(s) being overexpressed. This experimental design eliminates
the possibility that the observed effects are a consequence of differences in
genetic background rather than expression of the transgene.
Because SOD1 activity is a reflection of copper concentrations42 and
because APP is a copper-binding protein that reduces Cu2+ to Cu+ (ref.
43), we examined the effects of expression of each protein on the other.
To verify that APP overexpression did not influence SOD1 activity in dou-
ble transgenics, we measured SOD1 activity in whole blood of
APP/SOD1+ and APP+/SOD1+ mice using the Bioxytech SOD1-525 spec-
trophotometric assay44. In SOD1 mice (n = 6), SOD1 activity was 3.6
0.4 units, significantly (p < 0.05) less than the 9.2 1.5 units in
APP/SOD1+ mice (n = 5) or 10.3 0.6 units in APP+/SOD1+ mice (n =
5). APP/SOD1+ and APP+/SOD1+ were not significantly different from
one another (p > 0.05, t-test). Therefore, APP overexpression did not
interfere with the elevation in SOD1 activity in SOD1+ mice. Similarly,
western blots did not show any obvious change in APP expression as a
consequence of SOD1 overexpression.
Determination of cerebral blood flow. Techniques used for studying the
cerebral circulation in mice were similar to those described32. Mice were
anesthetized with halothane (maintenance 1%), intubated and artifi-
cially ventilated with an oxygen-nitrogen mixture. End-tidal CO2, mon-
itored by a CO2 analyzer (Capstar-100, CWI, Ardmore, Pennsylvania),
was maintained at 2.62.7% (pCO2 = 3335 mmHg)32. Arterial pO2
was 161 8, and the pH 7.41 0.03. The parietal cortex was exposed,
and the site was superfused with Ringer solution (37C; pH 7.37.4)32.
CBF was monitored in the exposed cortex with a laser-Doppler probe
positioned stereotaxically. Because CBF measurements by laser-Doppler
can vary depending on the monitoring site45, the probe was placed at
the same stereotaxic coordinates in all mice. Acetylcholine (10 M),
bradykinin (50 M), A23187 (3 M), U46619 (1 M) or SNAP (100500
M) were superfused on the cerebral cortex for 35 min. Agents were
applied at concentrations that produce 50% of maximal responses (data
not shown). Hypercapnia (pCO2 = 5060 mmHg) was produced by
introducing CO2 through the circuit of the ventilator. Arterial pressure,
monitored via a femoral arterial catheter, was 85 3 mmHg in
APP/SOD1 mice, 84 3 in APP+/SOD1 mice, 88 3 in APP+/SOD1+
mice, 86 3 in APP/SOD1+ mice, 86 3 in Tg6209 APP+ mice and 84
4 in Tg6209 APP mice. Topical superfusion with vasoactive agents
or hypercapnia did not affect arterial pressure or blood gases (data not
shown). In experiments in which effects of topical SOD were studied,
CBF responses to acetylcholine (10 M) and hypercapnia were tested
during Ringer superfusion and 30 min after superfusion with bovine
liver SOD (1000U /ml; Sigma)46.
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com
Determination of A. Techniques for A measurement have been
described in detail11,23. Brain tissue was homogenized in 1 ml of 70%
formic acid and centrifuged at >100,000 g for 1 hour. Samples were dilut-
ed and analyzed directly using either the BNT77/BA27 or BNT77/BC05
sandwich ELISA system. A values were obtained by comparing the
absorbance obtained from duplicate samples to standard curves of either
A140 (BNT77/BA27) or A142 (BNT77/BC05) with a standard curve 19. Wang, Q., Pelligrino, D. A., Koenig, H. M. & Albrecht, R. F. The role of
(Bachem, Torrance, California).
Electron microscopy. After transcardial perfusion of anesthetized mice
with 2% paraformaldehyde in phosphate-buffered saline, a sample of pari-
etal cortex was removed and fixed overnight in 2.5% glutaraldehyde and
4% paraformaldehyde in 0.1 M sodium cacodylate buffer, 0.05 M sucrose
at 4C. Samples were then postfixed, stained en bloc in 1% uranyl acetate,
dehydrated with ethanol and embedded in Spurrs resin. Thin sections
were cut with an RMC MT-7000 ultramicrotome (Research and Manu-
facturing Company, Tucson, AZ), stained with 1% uranyl acetate and
Reynolds lead citrate and observed at 80 kV on a Philips CM-10 trans-
mission electron microscope.
1999 Nature America Inc. http://neurosci.nature.com
Data analysis. Data in text and figures are expressed as mean standard
error. Two-group comparisons were analyzed by the two-tailed t-test for
independent samples. Multiple comparisons were evaluated by the analy-
sis of variance and Tukeys test. Survival data were analyzed by the Fish-
ers exact test in a 2 2 contingency table. Probability values of less than
0.05 were considered statistically significant.
ACKNOWLEDGEMENTS
This work was supported by grants from the National Institutes of Health (NS34179,
NS37853, NS35806 to C.I.; NS33249 to K.K.H; AG10681 to G.A.C.) and the
Fraternal Order of Eagles (G.A.C.). We thank G. Perry for the oxidative stress data
and Karen MacEwan for editorial assistance.
RECEIVED 24 JUNE; ACCEPTED 23 DECEMBER 1998
1. Lendon, C. L., Ashall, F. & Goate, A. M. Exploring the etiology of Alzheimer
disease using molecular genetics. JAMA 277, 825831 (1997).
2. Mattson, M. P. Cellular actions of beta-amyloid precursor protein and its
soluble and fibrillogenic derivatives. Physiol. Rev. 77, 10811132 (1997).
3. Smith, M. A. et al. Amyloid-beta deposition in Alzheimer transgenic mice is
associated with oxidative stress. J. Neurochem. 70, 22122215 (1998).
4. Hensley, K. et al. A model for beta-amyloid aggregation and neurotoxicity
based on free radical generation by the peptide: relevance to Alzheimer
disease. Proc. Natl. Acad. Sci. USA 91, 32703274 (1994).
5. Yan, S. D. et al. RAGE and amyloid-beta peptide neurotoxicity in Alzheimers
disease. Nature 382, 685691 (1996).
6. Behl, C., Davis, J. B., Lesley, R. & Schubert, D. Hydrogen peroxide mediates
amyloid beta protein toxicity. Cell 77, 817827 (1994).
7. Thomas, T., Thomas, G., McLendon, C., Sutton, T. & Mullan, M. -amyloid-
mediated vasoactivity and vascular endothelial damage. Nature 380, 168171
(1996).
8. Thomas, T., McLendon, C., Sutton, E. T. & Thomas, G. Cerebrovascular
endothelial dysfunction mediated by beta-amyloid. Neuroreport 8,
13871391 (1997).
9. Crawford, F., Suo, Z., Fang, C. & Mullan, M. Characteristics of the in vitro
vasoactivity of beta-amyloid peptides. Exp. Neurol. 150, 159168 (1998).
10. Blanc, E. M., Toborek, M., Mark, R. J., Hennig, B. & Mattson, M. P. Amyloid
beta-peptide induces cell monolayer albumin permeability, impairs glucose
transport, and induces apoptosis in vascular endothelial cells. J. Neurochem.
68, 18701881 (1997).
11. Hsiao, K. K. et al. Age-related CNS disorder and early death in transgenic
FVB/N mice overexpressing Alzheimer amyloid precursor proteins. Neuron
15, 12031218 (1995).
12. Wong, P. C. et al. An adverse property of a familial ALS-linked SOD1
mutation causes motor neuron disease characterized by vacuolar
degeneration of mitochondria. Neuron 14, 11051116 (1995).
13. Sobey, C. G. & Faraci, F. M. Effects of a novel inhibitor of guanylyl cyclase on
dilator responses of mouse cerebral arterioles. Stroke 28, 837843 (1997).
14. Rosenblum, W. I., McDonald, M. & Wormley, B. Calcium ionophore and
acetylcholine dilate arterioles on the mouse brain by different mechanisms.
Stroke 20, 13911395 (1989).
15. Rosenblum, W. I. Endothelial dependent relaxation demonstrated in vivo in
cerebral arterioles. Stroke 17, 494497 (1986).
16. Sobey, C. G., Heistad, D. D. & Faraci, F. M. Mechanisms of bradykinin-
induced cerebral vasodilatation in rats. Evidence that reactive oxygen species
nature neuroscience volume 2 no 2 february 1999
articles
activate K+ channels. Stroke 28, 22902294 (1997).
17. Mayhan, W. G. Role of activation of bradykinin B2 receptors in disruption of
the blood-brain barrier during acute hypertension. Brain Res. 738, 337341
(1996).
18. Rosenblum, W. I. & Nelson, G. H. Endothelium dependence of dilation of
pial arterioles in mouse brain by calcium ionophore. Stroke 19, 13791382
(1988).
endothelium and nitric oxide in rat arteriolar dilatory responses to CO2 in
vivo. J. Cereb. Blood Flow Metab. 14, 944951 (1994).
20. Faraci, F. M., Williams, J. K., Breese, K. R., Armstrong, M. L. & Heistad, D. D.
Atherosclerosis potentiates constrictor responses of cerebral and ocular
blood vessels to thromboxane in monkeys. Stroke 20, 242247 (1989).
21. Mayhan, W. G., Faraci, F. M. & Heistad, D. D. Responses of cerebral arterioles
to adenosine 5-diphosphate, serotonin, and the thromboxane analogue U-
46619 during chronic hypertension. Hypertension 12, 556561 (1988).
22. Carlson, G. A. et al. Genetic modification of the phenotypes produced by
amyloid precursor protein overexpression in transgenic mice. Hum. Mol.
Genet. 6, 19511959 (1997).
23. Hsiao, K. et al. Correlative memory deficits, A elevation, and amyloid
plaques in transgenic mice. Science 274, 99102 (1996).
24. Stamler, J. S. A radical vascular connection. Nature 380, 108111 (1996).
25. Mohazzab, K. M., Kaminski, P. M. & Wolin, M. S. NADH oxidoreductase is a
major source of superoxide anion in bovine coronary artery endothelium.
Am. J. Physiol. 266, H25682572 (1994).
26. Wang, P. et al. Overexpression of human copper, zinc-superoxide dismutase
(SOD1) prevents postischemic injury. Proc. Natl. Acad. Sci. USA 95,
45564560 (1998).
27. Yang, S. T., Mayhan, W. G., Faraci, F. M. & Heistad, D. D. Endothelium-
dependent responses of cerebral blood vessels during chronic hypertension.
Hypertension 17, 612618 (1991).
28. Wei, E. P., Kontos, H. A., Christman, C. W., DeWitt, D. S. & Povlishock, J. T.
Superoxide generation and reversal of acetylcholine-induced cerebral
arteriolar dilation after acute hypertension. Circ. Res. 57, 781787 (1985).
29. Mayhan, W. G. Impairment of endothelium-dependent dilatation of basilar
artery during chronic hypertension. Am. J. Physiol. 259, H14551462 (1990).
30. Sobey, C. G., Faraci, F. M., Piegors, D. J. & Heistad, D. D. Effect of short-term
regression of atherosclerosis on reactivity of carotid and retinal arteries.
Stroke 27, 927933 (1996).
31. Faraci, F. M. & Heistad, D. D. Regulation of the cerebral circulation: Role of
endothelium and potassium channels. Physiol. Rev. 78, 5397 (1998).
32. Zhang, F., Eckman, C., Younkin, S., Hsiao, K. K. & Iadecola, C. Increased
susceptibility to ischemic brain damage in transgenic mice overexpressing
the amyloid precursor protein. J. Neurosci. 17, 76557661 (1997).
33. Mentis, M. J. et al. Visual cortical dysfunction in Alzheimers disease
evaluated with a temporally graded stress test during PET. Am. J. Psychiatry
153, 3240 (1996).
34. Jagust, W. J., Eberling, J. L., Reed, B. R., Mathis, C. A. & Budinger, T. F.
Clinical studies of cerebral blood flow in Alzheimers disease. Ann. NY Acad.
Sci. 826, 254262 (1997).
35. Hock, C. et al. Decrease in parietal cerebral hemoglobin oxygenation during
performance of a verbal fluency task in patients with Alzheimers disease
monitored by means of near-infrared spectroscopy (NIRS)correlation
with simultaneous rCBF-PET measurements. Brain Res. 755, 293303
(1997).
36. Warkentin, S. & Passant, U. Functional imaging of the frontal lobes in
organic dementia. Regional cerebral blood flow findings in normals, in
patients with frontotemporal dementia and in patients with Alzheimers
disease, performing a word fluency test. Dement. Geriatr. Cogn. Disord. 8,
105109 (1997).
37. Hatake, K., Kakishita, E., Wakabayashi, I., Sakiyama, N. & Hishida, S. Effect
of aging on endothelium-dependent vascular relaxation of isolated human
basilar artery to thrombin and bradykinin. Stroke 21, 10391043 (1990).
38. Mayhan, W. G., Faraci, F. M., Baumbach, G. L. & Heistad, D. D. Effects of
aging on responses of cerebral arterioles. Am. J. Physiol. H11381143 (1990).
39. Johnson, K. A. et al. Preclinical prediction of Alzheimers disease using
SPECT. Neurology 50, 15631571 (1998).
40. Sano, M. et al. A controlled trial of selegiline, alpha-tocopherol, or both as
treatment for Alzheimers disease. The Alzheimers Disease Cooperative
Study. N. Engl. J. Med. 336, 12161222 (1997).
41. Scott, M. R., Kohler, R., Foster, D. & Prusiner, S. B. Chimeric prion protein
expression in cultured cells and transgenic mice. Protein Sci. 1, 986997 (1992).
42. Harris, E. D. Copper as a cofactor and regulator of copper, zinc superoxide
dismutase. J. Nutr. 122, 636640 (1992).
43. Multhaup, G. et al. The amyloid precursor protein of Alzheimers disease in
the reduction of copper(II) to copper(I). Science 271, 14061409 (1996).
44. Nebot, C. et al. Spectrophotometric assay of superoxide dismutase activity
based on the activated autoxidation of a tetracyclic catechol. Anal. Biochem.
214, 442451 (1993).
45. Iadecola, C. & Reis, D. J. Continuous monitoring of cerebrocortical blood
flow during stimulation of the cerebellar fastigial nucleus: A study by laser-
doppler flowmetry. J. Cereb. Blood Flow Metab. 10, 608617 (1990).
46. Zhang, F., Slungaard, A., Vercellotti, G. M. & Iadecola, C. Superoxide-
dependent cerebrovascular effects of homocysteine. Am. J. Physiol. 274,
R17041711 (1998).
161
 1999 Nature America Inc. http://neurosci.nature.com

articles

The construction of movement by

the spinal cord
Matthew C. Tresch1,2, Philippe Saltiel1 and Emilio Bizzi1

1 Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, E25526, Cambridge, Massachusetts 02139, USA
2 Present address: Section of Neurophysiology, Panum Institute, University of Copenhagen, Copenhagen, Denmark
Correspondence should be addressed to E.B. (emilio@ai.mit.edu)

We used a computational analysis to identify the basic elements with which the vertebrate spinal
cord constructs one complex behavior. This analysis extracted a small set of muscle synergies from
the range of muscle activations generated by cutaneous stimulation of the frog hindlimb. The
1999 Nature America Inc. http://neurosci.nature.com

flexible combination of these synergies was able to account for the large number of different motor
patterns produced by different animals. These results therefore demonstrate one strategy used by
the vertebrate nervous system to produce movement in a computationally simple manner.

The ease with which we move in and interact with our environ-
ment belies the complexity inherent in even the simplest of these
tasks. Movements we make effortlessly, such as reaching for an
object or walking over uneven terrain, involve the specification
and control of a vast number of variables1, ultimately involving
the activation of many thousands of motor units spanning dozens
of muscles. The question of how the nervous system overcomes
these complexities to produce movement effortlessly and effi-
ciently is central to the study of the neural control of movement.
We describe one strategy used by the vertebrate nervous sys-
tem to produce a range of behavior in a simple manner. We
show that the neural circuitry within the frog spinal cord pro-
duces motor responses to hindlimb cutaneous stimulation by
the combined recruitment of a small number of distinct muscle
groups. Such a muscle group, in which the activation level of a
set of muscles is specified together, has been termed a muscle
synergy previously2,3. Several researchers have proposed that
such spinally organized muscle groupings might underlie the
production of movement47. The present results provide direct
support for these proposals, presenting a quantitative assess-
ment of a testable form of this hypothesis. These experiments
thus provide a novel perspective on this fundamental question
of how the vertebrate nervous system produces movement in a
simple and efficient manner.
RESULTS
We examined the patterns of muscle activations evoked from
cutaneous stimulation of the hindlimb of spinalized frogs (frogs
whose spinal cord was transected). We stimulated a number of
different skin locations to evoke a range of motor responses, to
try to understand how this range was produced by the nervous
system. An example of the muscle activation patterns (Fig. 1b)
evoked from stimulating different regions (Fig. 1a) along the
rostral and caudal margins of the hindlimb is shown for one
animal. The activation level of the majority of muscles in each
frog was significantly dependent on the stimulus location on
the leg (p < 0.05, ANOVA). This dependency caused the pro-
duction of a range of different muscle activation patterns. Stim-
ulation of sites on the back of the calf (regions 14) typically
evoked responses with a strong activation of vastus externus
(VE) with a weaker activation of a few other muscles, such as
biceps femoris (BF), iliopsoas (IP) or adductor magnus (AM).
Stimulation of sites on the foot (regions 510) typically evoked
responses with a strong activation of semitendinosus (ST) and
IP and weaker activation of other muscles such as sartorius
(SA), rectus internus (RI), AM, vastus internus (VI) and BF. As
the stimulation site was moved along the front of the calf toward
the knee (regions 1114), the contribution of ST gradually
decreased, whereas the contribution of other muscles, partic-
ularly SA and AM gradually increased (n = 7).
We examined whether these responses were produced from
the combination of distinct groups of muscles. Such muscle
groups might be observed either in the variations among respons-
es evoked from different skin regions (as shown in Fig. 1b), or
in the variations among responses evoked from the same skin
location (as shown in Fig. 2a; each of these five responses were
evoked from stimulation of the front of the calf near the knee
indicated in Fig. 1a). Although the averaged and normalized
responses (Fig. 2b) were generally similar to one another, there
were clear systematic differences between them. For instance, the
fourth and fifth responses were very similar except that there was
a stronger activation of BF and IP in the fifth response. In the
first and second responses, the activation of ST was very similar,
but there was a weaker activation of SA, AM, VI, BF and IP in
the second response. These comparisons suggest the existence of
a set of at least three groups of muscles that could be controlled
independently: one with activation of ST, one with activation of
SA, AM and VI, and one with activation of BF and IP.
We investigated more systematically whether the patterns of
muscle covariations shown in Fig. 2b and the range of muscle
activations shown in Fig. 1a could both result from the combi-
nation of a small number of muscle activation patterns or mus-
cle synergies. An explicit formulation of this hypothesis has been
proposed8,9,10, and we evaluated this formulation here. Accord-
ing to this hypothesis, any given response should be describable
as the linear combination of a small number of such muscle syn-
ergies. Further, both the elements of the synergies and their
weighting within each response should be positive, because we

162 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 1. Muscle activation pat- ST SA RI

terns evoked from cutaneous a b
stimulation of the frog
hindlimb. (a) The divisions of
stimulation sites used to
examine the variation of mus-
cle activation patterns with

Normalized EMG
AM VI SM
stimulus location shown in (b).
Stimulation was applied to
many sites across the skin sur-
face, and the evoked pattern of
muscle activation was mea-
sured for each site. The VE BF IP
hindlimb configuration shown
here corresponds to the con-
figuration used in these exper-
iments. The dot within region
12 indicates the stimulus site
from which the responses
Skin location
shown in Fig. 2a were evoked.
1999 Nature America Inc. http://neurosci.nature.com

(b) The variation of muscle activation with stimulus location for one animal. Stimulation sites on the back and front of the femur did not consis-
tently evoke responses in each frog and so were excluded from analysis. The averaged, normalized activation for each recorded muscle is shown
as a function of stimulus location. For this plot, we also normalized each observed response to be of unit magnitude (taken as the vector norm).
The values here therefore reflect the relative contribution of each muscle to a given response. The numbers on the horizontal axis refer to the
stimulus regions indicated in (a). Error bars represent one standard deviation from the mean activation level at each region. The data shown here
were obtained from a total of 688 responses.

are ultimately considering muscle activations (Galagan, J. et al.
Soc. Neurosci. Abstr. 23, 512.4, 1997). This specific hypothesis can
be formalized in the following model:
N muscle activations (mean r2 = .90 .03).
mobs
j =
i=1
cwij i cij, wi 0 (1)
where mjobs is the jth observed pattern of muscle activations, cij
is the positive weighting coefficient of the ith muscle synergy for
the jth response, wi is the ith muscle synergy and N is the number
of muscle synergies. We tested how well this positive linear com-
bination of muscle synergies model explained the patterns of
muscle activations observed experimentally.
We extracted a set of muscle synergies from the observed mus-
cle activations for each animal using a computational analysis.
This analysis aimed at finding the set of muscle synergies that
could best predict the observed responses according to the model
described above. Briefly, the algorithm began with a set of arbi-
trary synergies, wi. The positive weighting coefficients of these
arbitrary synergies, cij, that best predicted each response were then
found. The muscle synergies were updated by performing a gra-
dient descent on the error between the observed response and the
best-fit, predicted response. This process was then iterated until
the algorithm converged on a particular set of muscle synergies.
In a typical animal (Fig. 2c), the algorithm extracted both a
set of synergies and the weighting coefficients of each synergy
used to reconstruct the responses evoked from cutaneous stim-
ulation. The algorithm extracted a synergy with ST, a synergy
with SA, AM and VI, a synergy with VE, and a synergy with BF
and IP. These synergies corresponded well to the patterns of
covariation observed in Fig. 2b. The derived synergies were very
robust to differences in initial conditions, with an average nor-
malized dot product of .97 .04 between the synergies found
on different iterations of the algorithm. Qualitatively, the
observed (Fig. 2b) and predicted (Fig. 2d) responses were very
similar to one another. Quantitatively, this similarity between
observed and predicted responses was found to be very large for
each animal. In each of the seven animals examined here, with
an average of 392 responses per animal, the algorithm consis-
tently explained a large amount of variance in the patterns of
The synergies shown in Fig. 2c were generally similar between
animals. The average similarity (calculated as the normalized dot
product) between synergies found in different animals was
.74 .20. The similarity of the first three synergies shown in Fig.
2c was generally higher than that of the fourth (.91 .09,
.80 .11, .80 .22, .45 .28, respectively). This similarity, how-
ever, was less than the similarity between synergies found on
repeated iterations of the algorithm for the same animal (.74 ver-
sus .97), suggesting that there were differences between the syn-
ergies found in different animals. Differences in such factors as
electrode placements, link lengths or possibly the strategies used
by different animals could have contributed to these differences.
We compared the results obtained by this algorithm to the
results obtained by the k-means algorithm11,12. K-means attempts
to explain each observed response as one of only a small num-
ber of possible responses. In contrast to our algorithm, k-means
does not allow an observed response to be a combination of more
basic patterns; in k-means, each response reflects the recruitment
of a single pattern. To explain 90% of the variance in the observed
set of responses, k-means required at least 15 different patterns.
This demonstrates that cutaneous stimulation of the hindlimb
did not simply evoke only a few distinct types of muscle patterns.
Instead, the range of responses evoked from cutaneous stimula-
tion was better explained as the combination of a small number
of muscle synergies.
We also examined how the results of the algorithm explained
the range of muscle activation patterns evoked from different
regions of the skin, such as those shown for the animal in Fig.
1b. We show for three animals how the weighting coefficient of
each synergy extracted by the algorithm varied with the stimulus
location on the hindlimb (Fig. 3). These are the weighting coef-
ficients of the synergies used to reconstruct each response
(Fig. 2c). For instance, in the animal shown in Fig. 3a, to recon-

nature neuroscience volume 2 no 2 february 1999 163

 1999 Nature America Inc. http://neurosci.nature.com

articles

a b Observed responses

Time (ms)
c d
1999 Nature America Inc. http://neurosci.nature.com

Predicted responses

Fig. 2. Example of muscle covariation patterns within evoked responses. (a) Raw EMGs for five responses evoked from stimulation of the same skin
region. The hindlimb was fixed in the configuration shown in Fig. 1a for all responses and the region of the skin indicated in Fig. 1a was stimulated using
a handheld probe (see Methods). (b) Averaged, normalized activation for the muscles recorded in each of the responses shown in (a). Note that each
muscle was normalized to the maximal value observed for that muscle across all responses evoked from any stimulation site in this animal. As a result,
the muscle balances seen in (a) are slightly different than those shown in (b). (c) Responses can be explained as a linear combination of a set of mus-
cle synergies. The synergies obtained from applying the algorithm described in the text to the entire set of responses obtained from this animal are
shown to the left. The weightings of each of these synergies used to reconstruct the responses in (b) are shown to the right. For instance, the first
response was reconstructed as .65 of the first synergy, plus 2.47 of the second synergy, 0 of the third synergy, and 1.09 of the fourth synergy.
(d) Responses resulting from the combination of muscle synergies shown in (c).

struct responses from sites of the back of the leg (regions 14),
the algorithm used a strong weighting of the synergy with VE
(the third synergy in Figs. 2c and 3a). This strong VE activation
is consistent with the responses observed from the back of the
leg (Fig. 1b). Similarly, to reconstruct responses from sites on the
foot (regions 510), the algorithm used a strong weighting of the
synergy with ST (the first synergy in Figs. 2c and 3a) along with
weaker weightings of the synergy with SA, AM and VI (the second
synergy in Figs. 2c and 3a). As stimulation sites moved along the
front of the leg (regions 1114), the weighting of the synergy with
ST gradually decreased, whereas the activation of the SA, AM
and VI synergy and of the VE synergy increased. Activation of
the fourth synergy was more variable across regions of the skin
surface. Generally similar patterns of recruitment of synergies
were observed in each animal (compare Fig. 3a, b and c).
Although Fig. 3 suggests that the use of each type of synergy
was generally similar in each animal, the exact pattern of recruit-
ment of these synergies could differ between animals. In the ani-
mal shown in Fig. 3c, for instance, there were sharp transitions
in the weighting of each synergy between different skin regions.
These sharp transitions were especially clear for the first three
synergies, with the weighting coefficients of these synergies either
being zero or large for most responses. In the animals shown in
Fig. 3a and 3b, however, there appeared to be more of a gradual
transition in the recruitment of different synergies, as indicated
in the more intermediate weighting coefficients of each synergy.
This type of gradual transition was seen in most animals, espe-
cially for the transition from the foot to the front of the leg. These
observations suggest that each animal has access to a similar set of
muscle synergies when producing responses to cutaneous stim-
ulation, but that there is a degree of flexibility in the exact pat-
tern of their use.
Finally, we examined the generality of the synergies derived
from the responses evoked from stimulation of the sites shown
in Fig. 1a. We examined how well these synergies were able to
explain the responses evoked from other regions of the skin sur-
face; if similar synergies were used to produce responses from
these other regions, we would expect the responses to be
explained well by these synergies. We evoked responses from
stimulation of sites on the dorsal surface of the foot, the dorsal
surface of the calf, the contralateral ankle, the ipsilateral and con-
tralateral back and both forelimbs. The synergies obtained from

164 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

a b c method to identify the basic elements

underlying the production of movement
in a systematic and objective manner. The
consistency of the results of this algorithm
with qualitative examination of the data
Weighting coefficient of synergy

helps to validate the approach used here.

This explicit formulation of the hypothe-
sis also allowed it to be falsified: the results
of this algorithm could easily have failed
to consistently explain the responses with
the fidelity observed here. Further, the
particular model we tested, as expressed
in Eq. 1, was relatively simple, assuming
a linear and independent combination of
muscle synergies. This simplicity makes it
all the more surprising that the model was
able to explain these responses so well.
The results and methods of our study
1999 Nature America Inc. http://neurosci.nature.com

Skin location might be used to examine how the spinal

Fig. 3. The contribution of each muscle synergy to responses evoked from different regions of the
cord is used by the rest of the nervous sys-
skin surface for three different animals (a, b, c). The order of the synergies shown here corresponds tem to produce movement. We found that
to the order of the synergies shown in Fig. 2c. The correspondence between synergies was made the set of synergies described here could
using the normalized dot product between them. The numbers on the horizontal axis refer to the explain some, but not all, of the respons-
regions shown in Fig. 1a. The vertical axis shows the average weighting coefficient for each muscle es evoked from other regions of the skin
synergy used to reconstruct the observed responses, such as those shown in Fig. 1b. surface. This result suggested that these
synergies might be utilized by different
pathways in the production of movement.
We might be able to use similar proce-
stimulation of the hindlimb were able to explain over 80% of the dures to examine whether descending systems also use these syn-
variance in the responses evoked from sites on the dorsal surface ergies to produce movement. For instance, vestibular stimulation
of the foot and calf, the ipsilateral back and the contralateral ankle in the frog evokes a low-dimensional set of hindlimb move-
(Fig. 4). Responses from the other skin regions, however, were ments13. It will be interesting to examine the relationship between
not well explained by these same synergies. The observation that the low dimensionality of that set of movements and the com-
the synergies explained some responses poorly also demonstrates bination of a small number of muscle synergies identified here.
that the synergies were not so general as to be able to explain any Further, it will be important to examine whether the model
possible motor response. These results suggest that the synergies described here is capable of explaining the complete time course
used to produce the responses from the hindlimb might also be of motor responses. In many behaviors, there is a great deal of
used in the motor responses evoked through other pathways. temporal precision in muscle activation patterns3, and it will be
interesting to examine whether this precision can be described
DISCUSSION in terms of a small number of synergies. Further experiments
These results demonstrate that the neural circuitry within the
vertebrate spinal cord seems to produce a behavior through the
combination of a small number of muscle synergies. By using a
computational analysis, we show that such a combination of mus-
cle synergies is able to explain the range of muscle activations
evoked by cutaneous stimulation of the frog hindlimb. Such a
strategy of combining a small number of basic motor elements
r2

greatly simplifies the complexity intrinsic to the production of

movement by the nervous system.
These results provide direct support for previous hypotheses
of the production of movement by the spinal cord 46 . For
instance, the nervous system was proposed to produce a range
of movement through the combination of a small number of unit
burst generators organized within the spinal cord5. Each of these
unit bursters was proposed to control the activation of a small
group of synergistic muscles. These units could then be coupled
in many different ways to produce a wide range of behavior. The Stimulation site
hypothesis considered here, that the spinal cord produces move-
Fig. 4. The ability of synergies from sites on the rostral and caudal mar-
ment through the combination of a small number of motor ele- gins of the hindlimb to explain responses evoked from other skin
ments, is similar to this unit burst generator hypothesis. regions. Synergies were derived from the regions shown in Fig. 1a and
The main contribution of this study has been to state this then used to describe the responses evoked from the other skin regions
hypothesis in an explicit form, allowing it to be examined quan- indicated on the horizontal axis. The vertical axis shows the r2 values for
titatively. The computational analysis described here provided a the fits for each of these data sets.

nature neuroscience volume 2 no 2 february 1999 165

 1999 Nature America Inc. http://neurosci.nature.com
articles
will be required to define the extent and limits of the generality of
the synergies described here to other behaviors, including those
produced by descending systems.
Sherrington first proposed that the responses to cutaneous
stimulation might also be used in other classes of movement,
such as locomotion4. Although the movements produced by the
spinal cord in response to cutaneous stimulation are often con-
sidered to be simple and stereotyped, it is well established that
these responses can be precisely tuned depending on the site of
stimulation1417. This flexibility might be used in the movements
produced by descending systems18. Our study might provide a
novel perspective from which to test this hypothesis.
METHODS
Data acquisition. All procedures were approved by the Committee on
Animal Care at M.I.T. Seven adult frogs (Rana catesbiana) were spinal-
ized at the level of the obex under tricaine anesthesia. Nine hindlimb
muscles were implanted with bipolar EMG electrodes (see ref. 9 for
1999 Nature America Inc. http://neurosci.nature.com
details): semitendinosus (ST), sartorius (SA), rectus internus (RI),
adductor magnus (AM), vastus internus (VI), semimembranosus
(SM), vastus externus (VE), biceps femoris (BF) and iliopsoas (IP). In
one animal, we recorded from rectus anterior in the place of IP.
Crosstalk from electrical pickup between adjacent muscles was elimi-
nated by suturing small pieces of insulation to the fascia in the major-
ity of animals. The effectiveness of this insulation was confirmed by
observing no EMG signals in muscles with their nerve supply cut. The
hindlimb was fixed in the horizontal plane at the level of the hip by
means of bone screws attached at the ankle and the metatarsus. This
restraint prevented all movement of the hindlimb. The foot was fixed
at a right angle relative to the calf (see Fig. 1a). The hindlimb was
mechanically stimulated by scratching restricted sites on the skin sur-
face (1 mm2 tip, .01 to .1 N force, 14 Hz, over a range of 13 mm).
Stimulation strength was regulated so as to evoke only threshold
responses. EMG signals were filtered and amplified (25 k) before being
digitally sampled (1 kHz) for offline analysis. Response onsets and off-
sets were identified offline using interactive software written in Matlab,
and the activation level of each muscle was averaged within each
response. The activation of each muscle was then normalized to the
maximal value observed for that muscle in any response. Weak
responses with a magnitude (taken as the vector norm) less than .3
were excluded from further analysis because EMG noise often obscured
the balance of muscle activations in these weaker responses.
We assessed the repeatability of the stimulation procedures used here
by examining the variability in the responses evoked from repeated stim-
ulation of a site on the skin. The standard deviation in the most activat-
ed muscle in a response was .27 across all animals. This variability was
comparable to the variability in the responses evoked from repeated elec-
trical stimulation of the same site in spinal cord (standard deviation,
0.21, data not shown). Such electrical stimulation would be expected to
activate the same neural substrate on each application of stimulation and
should therefore give a lower bound of the variability in the motor
responses evoked in this preparation. Such variability in the responses
evoked in this preparation has been observed previously19.
Algorithm to extract muscle synergies. A set of synergies, wi, consists of
coefficients chosen randomly between 0 and 1 for each muscle. We then
approximated each observed response as a positive linear combination
of these synergies, described by Eq. 1. A slightly different form of Eq. 1 in
the actual fitting procedure was used to impose the restriction of the syn-
ergies to positive numbers.
N data for the following reasons. First, we did a principal components
j m j = cijg(wi)
mobs pre
(1)
i=1
mjobs is the jth observed pattern of muscle activations, mjpre is the jth
predicted pattern of muscle activations and g(x) is a positive valued
function of x. We tried several different forms of g(x), including one in
which negative values of x were arbitrarily set to 0, the simple quadratic
g(x) = x2 and an exponential g(x) = ex. The results of the algorithm
166
using these different transformations were very similar. We only pre-
sent the results of the algorithm using the exponential transformation.
Given a certain set of synergies, the set of best-fit weighting coeffi-
cients, cij, was found using the non-negative least-squares algorithm20
supplied by Matlab. This algorithm finds the set of nonnegative coeffi-
cients that minimizes the error between the predicted and observed
responses, given a particular set of synergies. This procedure is similar
to standard regression techniques except that the weighting coefficients
are constrained to be positive.
Once the weighting coefficients were found, the error between the
predicted and observed responses was used to update the synergies. The
error was minimized using gradient descent, in which the synergies are
incrementally changed to reduce the prediction error. The change in the
synergies was calculated as Eq. 2:
wi = (mobs pre
j m j ) cijg(wi) (2)
where g is the first derivative of the transformation used in Eq. 1 and
determines the step size of the gradient descent. Because the process of min-
imizing the error using gradient descent is generally very erratic, the step
size of the descent is usually set to be low; we used a value of = .005. To fur-
ther ensure smoothness of the error minimization, we used a momentum
coefficient, in which the change in synergy at iteration t is equal to the
change in the synergy calculated by Eq. 2 plus a fraction (we used .99) of
the change in weight at iteration t 1. This momentum coefficient serves as
a type of running average of the synergy changes. These procedures are
standard in gradient descent techniques11,12. After updating the synergies,
they were normalized so that the vector norm of each synergy was equal to
one. The best-fit weighting coefficients were then again found using these
new synergies, the synergies updated, and the process repeated.
We applied this algorithm to 90% of the data for each animal, chosen
randomly from the responses evoked from the sites in Fig. 1a. After each
iteration, we assessed the ability of the synergies found by the algorithm
to predict the remaining 10% of responses. When the amount of vari-
ance explained
j
by the derived synergies in this set of test data decreased
for 20 consecutive iterations, we considered the algorithm to have con-
verged. This procedure reduces overfitting by the algorithm. We repeat-
ed the algorithm on 10 different sets of 90% of the data for each animal
with different initial synergies for each repetition, to examine effects of
different initial conditions on the results of the algorithm. Qualitative
analyses suggested that the algorithm required around 50100 respons-
es to produce consistent results. The algorithm was applied to the entire
data sets from each animal, ranging between 177 and 688 responses.
We calculated the similarity between two synergies found by the algo-
rithm as the normalized dot product between them. This value ranges
between zero and one, with one representing identical synergies. To estab-
lish correspondences between two different sets of synergies, we took all
possible dot products between the two sets of synergies. A synergy from
one set was matched to the synergy in the other set with which it had its
largest dot product. The process of correspondence was also inspected
visually in case of ambiguities.
We also applied the k-means algorithm to the set of evoked respons-
es11,12. This algorithm is similar to the algorithm described above except
that the cij are equal to zero for all except one synergy.
The algorithm described above is generally similar to factor analytic
models21, with the difference that both the factor loading and factor
scores are constrained to be positive. It is also similar to that described22
except that we are not explicitly imposing a particular prior distribution
for the weighting coefficients. We therefore implicitly impose a uniform
distribution.
We describe the results of the algorithm using four synergies to fit the
analysis on this data and found that the smallest number of components
necessary to explain at least 90% of the variance in each animal was four.
Similarly, four synergies were needed for the algorithm used here to
explain at least 90% of the variance. Adding an additional synergy to the
algorithm explained only an extra 3% of the variance. Finally, applying
the algorithm with four synergies corresponded well to a visual inspec-
tion of the data and gave consistent results between different animals.
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com
ACKNOWLEDGEMENTS
We thank Sandro Mussa-Ivaldi and Andrea dAvella for reading versions of this
manuscript and Simon Giszter, Peter Dayan, Kuno Wyler and James Galagan
for suggestions. M.C.T. was supported by a HHMI predoctoral fellowship. This
research was supported by NIH NS09343 and ONR N00014-95-I0445 to E.B.
RECEIVED 30 SEPTEMBER; ACCEPTED 30 NOVEMBER 1998
1. Bernstein, N. The Coordination and Regulation of Movements (Pergamon
Press, New York, 1967).
2. Lee, W. A. Neuromotor synergies as a basis for coordinated intentional
action. J. Motor Behav. 16, 135170 (1984).
3. Macpherson, J. M. in Motor Control: Concepts and Issues (eds. Humphrey, D.
R. & Freund, H.-J) 3347 (Wiley, Chichester, 1991).
4. Sherrington, C. S. Flexion-reflex of the limb, crossed extension-reflex and
reflex stepping and standing. J. Physiol. (Lond.) 40, 28121 (1910).
5. Grillner, S. in Handbook of Physiology, sec. 1, vol. 2 (ed. Brooks, V. B.)
11791236 (American Physiological Society, Bethesda, MD, 1981).
6. Stein, P. S. G. & Smith J. L. in Neurons, Networks, and Motor Behavior (eds.
Stein P. S. G., Grillner, S., Selverston A. I. & Stuart D. G.) 6173 (MIT Press,
1999 Nature America Inc. http://neurosci.nature.com
Cambridge, MA, 1997).
7. Jacobs, R. & Macpherson, J. M. Two functional muscle groupings during
postural equilibrium tasks in standing cats. J. Neurophysiol. 76, 24022411
(1996).
8. Bizzi, E., Mussa-Ivaldi, F A. & Giszter S. F. Computations underlying the
production of movement: a biological perspective. Science 253, 287291
(1991).
9. Giszter, S. F., Mussa-Ivaldi, F. A. & Bizzi, E. Convergent force fields organized
in the frog spinal cord. J. Neurosci. 13, 467491 (1993).
10. Mussa-Ivaldi, F. A., Giszter, S. F., & Bizzi, E. Linear combinations of
nature neuroscience volume 2 no 2 february 1999
articles
primitives in vertebrate motor control. Proc. Natl. Acad. Sci. USA 91,
75347538 (1994).
11. Bishop, C. M. Neural Networks for Pattern Recognition (Oxford Univ. Press,
Oxford, 1995).
12. Hertz, J., Krogh, A. & Palmer R. G. Introduction to the Theory of Neural
Computation (Addison-Wesley, Reading, MA 1991).
13. dAvella, A. & Bizzi, E. Low dimensionality of supraspinally induced force
fields. Proc. Natl. Acad. Sci. USA 95, 77117714 (1998).
14. Fleshman, J. W., Lev-Tov, A. & Burke, R. E. Peripheral and central control of
flexor digitorum longus and flexor hallucis longus motoneurons: the
synaptic basis of functional diversity. Exp. Brain Res. 54, 133149 (1984).
15. Pratt, C. A., Chanaud, C. M. & Loeb, G. E. Functionally complex muscles of
the cat hindlimb. IV. Intramuscular distribution of movement command
signals and cutaneous reflexes in broad, bifunctional thigh muscles. Exp.
Brain Res. 85, 281299 (1991).
16. Schouenborg, J. & Kalliomaki, J. Functional organization of the nociceptive
withdrawal reflexes I. Activation of hindlimb muscles in the rat. Exp. Brain
Res. 83, 6778 (1990).
17. Schouenborg, J. & Weng, H.-R. Sensorimotor transformations in a spinal
motor system. Exp. Brain Res. 100, 170174 (1994).
18. Schouenborg, J., Weng, H.-R. & Holmberg, H. Modular organization of
spinal nociceptive reflexes: a new hypothesis. News Physiol. Sci. 9, 261265
(1994).
19. Berkinblitt, M. B., Feldman, A. G. & Fukson, O. I. Adaptability of innate
motor patterns and motor control mechanisms. Behav. Brain Sci. 9, 585638
(1986).
20. Lawson, C. L. & Hanson, R. J. Solving Least Squares Problems. (Prentice-Hall,
Englewood Cliffs, NJ, 1974).
21. Harman, H. H. Modern Factor Analysis (Univ. of Chicago Press, Chicago,
1976).
22. Olshausen, B. A. & Field, D. J. Emergence of simple-cell receptive field
properties by learning a sparse code for natural images. Nature 381, 607609
(1996).
167
 1999 Nature America Inc. http://neurosci.nature.com

articles

Self-sustained rhythmic activity in the

thalamic reticular nucleus mediated by
depolarizing GABAA receptor potentials
M. Bazhenov1, I. Timofeev2, M. Steriade2 and T.J. Sejnowski1,3

1 Howard Hughes Medical Institute, The Salk Institute, Computational Neurobiology Laboratory, 10010 North Torrey Pines Road,
La Jolla, California 92037, USA
2 Laboratory of Neurophysiology, School of Medicine, Laval University, Quebec G1K 7P4, Canada
3 Department of Biology, University of California San Diego, La Jolla, California 92093, USA
Correspondence should be addressed to M.B. (bazhenov@salk.edu)
1999 Nature America Inc. http://neurosci.nature.com

Intracellular recordings from reticular thalamic (RE) neurons in vivo revealed inhibitory postsynaptic
potentials (IPSPs) between RE cells that reversed and became depolarizing at the hyperpolarized
membrane potentials that occur during sleep. These excitatory IPSPs can directly trigger low-thresh-
old spikes (LTSs). The oscillatory mechanisms underlying IPSP-triggered LTSs crowned by spike
bursts were investigated in models of isolated RE networks. In a one-dimensional network model,
external stimulation evoked waves of excitation propagating at a constant velocity of 25150 cells
per second. In a large-scale, two-dimensional model of the reticular nucleus, the network showed
transient or self-sustained oscillations controlled by the maximum conductance of the low-threshold
calcium current and the membrane potential. This model predicts that the isolated reticular nucleus
could initiate sequences of spindle oscillations in thalamocortical networks in vivo.

Sleep spindle oscillations consist of waxing-and-waning field poten-
tials of 7 to 14 Hz, which last 1 to 3 seconds and recur every 5 to 15
seconds. These oscillations are generated in the thalamus by inter-
action between thalamic reticular (RE) and thalamocortical (TC)
cells14. The waxing phase of spindle oscillations is associated with
recruitment of neurons from dorsal thalamic and RE nuclei5, where-
as the waning phase results from calcium-induced cAMP upregu-
lation of hyperpolarization-activated cation current, Ih, in TC cells68.
Extracellular neuron and field-potential recordings from the isolat-
ed RE nucleus in vivo show that the deafferented RE nucleus can
itself generate 714 Hz oscillations9, and so the reticular nucleus
may be involved in initiating sequences of spindle oscillations in the
intact RETC network.
The GABAergic neurons of the RE nucleus are usually thought
to interact by reciprocal inhibition. Computational models of
inhibitory RE networks produce different types of population
activity, including full and partial spatial clustering10 and tempo-
ral synchronization alternating between periods of coherent oscil-
latory activity and periods of almost chaotic behavior11. The
oscillations in these models occur because the synaptic inhibition
causes burst discharges when the reversal potential for inhibito-
ry synapses is sufficiently negative to remove inactivation of the
low-threshold calcium current.
For the reciprocal GABAA receptor synapses between RE cells,
the chloride reversal potential is about 71 mV, which is depolar-
ized compared to the reversal potential in TC cells12. Thus, at a rest-
ing membrane potential of about 78 mV (ref. 13), the
GABAA-receptor-mediated IPSPs in an RE cell reverse and could
trigger a burst of sodium spikes14. Here we used in-vivo recordings
and computational models of RE cells to investigate cellular dynam-
ics at different membrane potentials. We found, in vivo, that when RE
cells were hyperpolarized below the chloride reversal potential,
GABAA-receptor-mediated synaptic excitation led to a low-thresh-
old calcium action potential crowned by sodium spikes. In a com-
puter model of a hyperpolarized RE network, this mechanism
produced propagating patterns of spike-burst activity, which could
be transformed into self-sustained oscillations.
RESULTS
Low-threshold spike bursts in RE neurons in vivo
Intracellular recordings from RE cells (n = 97) in cats were charac-
terized by prolonged spike bursts with accelerando-decelerando pat-
terns, during either spontaneous oscillatory activity or activity
evoked by dorsal thalamic stimulation (Figs. 13). Such prolonged
spike bursts, consisting of progressively decreasing followed by
increasing interspike intervals, are in contrast to the short spike
bursts of thalamocortical neurons15. Some RE neurons were also
morphologically identified by intracellular staining with Neurobiotin
(data not shown). The resting membrane potential (Vm) of RE neu-
rons was 62.5 2.9 mV (mean standard error), which is rela-
tively depolarized compared to results reported with patch pipettes13.
After Vm was determined, most of the intracellular recordings were
done with slight (0.20.5 nA) DC hyperpolarization.
During the state of resting sleep with long-range synchroniza-
tion of electrical activity, neocortical and thalamic neurons display
three major rhythms: slow oscillations (<1 Hz), delta waves (14
Hz) and spindles (715 Hz). In animals under ketamine-xylazine
anesthesia, RE neurons are hyperpolarized during the depth-positive
cortical electroencephalogram (EEG) component of the slow oscil-
lation and depolarized during the depth-negative EEG waves16. This
cortically generated spontaneous slow oscillation was used to test
the responsiveness of RE neurons at different membrane voltages

168 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 1. Reversed IPSP in an RE neuron in vivo directly triggers a

low-threshold spike (LTS). Intact thalamocortical connections.
Left column, responses of an RE neuron to four stimuli applied to
the thalamic VL nucleus (arrowhead). At relatively depolarized Vm
(61 mV), low-intensity thalamic stimuli evoked an IPSP-rebound
sequence. When the stimulus (same parameters as above)
occurred during the hyperpolarizing phase of the slow oscillation,
the IPSP reversed and (at 78 mV) directly triggered an LTS
crowned by spike burst. The latency of LTS was in the range of
4050 ms. The early part of two responses (at 61 mV and at 78
mV) are enlarged at right. Bottom right, the amplitude of the
postsynaptic response, measured 75 ms after the stimulus, is plot-
ted against Vm. The IPSP depended linearly on Vm from depolar-
ized levels to the reversal potential (68 mV); thereafter, the
depolarizing IPSP directly activated the LTS, which added a depo-
larizing deflection to the linear function (solid and dotted lines).

because it affects both somatic and dendritic compartments

1999 Nature America Inc. http://neurosci.nature.com

of RE neurons, unlike the current pulses that are usually

applied with somatic impalement. This is particularly
important for RE neurons because their dendritic voltage is
poorly controlled from the soma17, so current pulses prob-
ably do not affect remote parts of the dendritic tree.
Electrical stimulation applied to the dorsal thalamus elic-

Amplitude of IPSPs (mV)

its an EPSP crowned by a spike burst in RE neurons, fol-
lowed by an IPSP and a period of disfacilitation: a
long-lasting absence of synaptic activity18. Here we stimu-
lated the thalamic VL nucleus and confirmed these findings.
In 10 out of 32 neurons tested, diminishing the stimulus
intensity resulted in long-latency (32.2 3.5 ms) hyperpo-
larizing IPSPs when stimuli were applied at a relatively depo-
larized Vm (Fig. 1). Because RE neurons are GABAergic and
reciprocally connected19, this primary IPSP probably orig-
inated from neighboring RE neurons that were excited by
axons of thalamocortical and/or corticothalamic cells. How-
ever, during the hyperpolarizing phase of the slow oscilla-
tion, VL stimuli with the same parameters resulted in reversed IPSPs
crowned by high-frequency spike bursts (Fig. 1).
Stimuli were delivered to VL every five seconds. Some of them
occurred at relatively depolarized Vm levels, whereas others were
delivered during the hyperpolarizing phase of the slow oscillation,
associated with different degrees of disfacilitation. Stimuli applied
during a relatively depolarized Vm elicited an inhibitory IPSP fol-
lowed by a rebound spike burst (Fig. 1). As stimuli were delivered
at Vm levels closer to the reversal potential for chloride ions, hyper-
polarizing responses disappeared and were replaced by depolariz-
ing responses. At more hyperpolarized Vm levels, these reversed
IPSPs triggered a full-blown, high-frequency spike burst.
We studied the activity patterns of RE neurons in ipsilaterally
decorticated cats. Decortication removes the inputs generated by
the cortical slow oscillation, leaving intact the thalamically generat-
ed spindle and delta oscillations20. RE neurons showed two main
forms of spontaneous oscillatory activity. The first pattern was a
spindle oscillation characterized by rhythmically recurring (712
Hz) high-frequency spike bursts on a depolarizing envelope19. The
second pattern was the occurrence of periodic spike bursts, mainly
at frequencies around 23 Hz (Fig. 2). In multi-site, multi-unit
recordings, these bursts did not occur simultaneously, but followed
each other, even when spindle oscillations occurred almost simul-
taneously. For example, in a typical recording (Fig. 2), a perispike
histogram triggered by the first action potential in the burst of one
cell shows that most spikes in a pool of neurons recorded through
another electrode occurred either before or after these bursts.
Similar burst patterns were observed in intracellularly recorded
RE neurons. At rest or under slight DC hyperpolarizing currents,
RE neurons showed periodic (0.30.6 Hz) burst activity (Fig. 3a).
Usually, this bursting activity was abolished by release of the neu-
ron from DC hyperpolarizing current or by slight DC depolariza-
tion. Under these conditions, RE neurons showed hyperpolarizing
potentials resembling IPSPs. The frequency of the hyperpolarizing
potentials was almost identical to the frequency of bursts recorded
under DC hyperpolarizing current (Fig. 3). This observation, and
the finding that the IPSPs of RE neurons can only slightly shunt the
low-threshold spikes (LTSs)21, suggest that, at the relatively hyper-
polarized Vm levels that occur during resting sleep with EEG syn-
chronization, the reversed IPSPs cause burst firing in RE neurons,
which may maintain persistent activity within the RE nucleus and
initiate spindle activity.
Computer models of activity patterns in the RE network
The ability of a reversed IPSP to trigger a low-threshold calcium
spike in RE cells was investigated in computer models of an isolat-
ed RE neuron. We first simulated an RE cell with a GABAA recep-
tor IPSP at different levels of membrane potential (Fig. 4). At the
relatively depolarized membrane potential of 65 mV, the low-
threshold calcium current in the RE cell was inactivated. Hyper-
polarization evoked by the GABA A IPSP led to the partial
deinactivation of the low-threshold calcium current and a weak
rebound LTS crowned by a single sodium spike. At a more nega-
tive level of membrane potential, the low-threshold calcium cur-

nature neuroscience volume 2 no 2 february 1999 169

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 2. Spontaneous activity of RE neurons in vivo is char-

acterized by spindle activity interspersed with low-fre-
quency ( 2 Hz) spike bursts. Ketamine-xylazine
anesthesia. Cortical activity was depressed by application
of potassium acetate. Two traces show simultaneous
recording of multi-neuron activity from RE nucleus (elec-
trodes separated by 1 mm). Spindles occurred virtually
simultaneously. Expanded fragments show spindles (top
left arrowhead) and lower-frequency spike bursts of RE
neurons during interspindle lull (bottom right arrow-
head). The histogram (10-ms bins) triggered by the first
spike in a burst in the top recording depicts the temporal
relationship between the two multi-neuron traces (filled
bars for upper trace and open bars for lower trace). The
burst firing in one neuron preferentially occurs between
n = 10

the bursts of the other neuron (peaks of filled histogram

cut off between 0 and 100 ms).
1999 Nature America Inc. http://neurosci.nature.com

ms

rent in the RE cell was partially deinactivated at rest; however, near
the chloride reversal potential, the IPSP was diminished, and, at
76 mV, the chloride IPSP in the RE neuron was reversed. Activa-
tion of the GABAA receptor synaptic current at the hyperpolarized
level led to depolarization, followed by an LTS and a burst of sodi-
um spikes in the RE cell. During burst discharge in the RE cell,
depolarization inactivated the low-threshold calcium current, so
the cell failed to produce a second burst of spikes after a 200-ms
delay. However, a third stimulus delivered after a 400-ms delay was
again followed by a burst of sodium spikes. The IPSP-evoked depo-
larization was relatively weak, and strong deinactivation of low-
threshold calcium current was required to produce an LTS followed
by a burst of sodium spikes. Furthermore, immediately after the
burst of spikes, the cell was hyperpolarized as much as 23 mV, and
additional depolarization was required to elicit an LTS. This hyper-
polarization resulted from inactivation of the low-threshold calci-
um current, which was slightly activated at membrane potentials
around 76 mV and depolarized the cell about 2 mV before the
burst. Consequently, a relatively long time delay (about 350 ms)
between shocks was required to obtain a burst of spikes after the
second stimulus. This is different from an EPSP-evoked LTS, which
can sustain burst discharges at a frequency of 10 Hz during spin-
dle oscillations5 or augmenting responses22.
Dynamical properties of the isolated RE network hyperpolar-
ized below the chloride reversal potential were investigated in a
one-dimensional network model with 100 RE cells (see Methods).
An external AMPA receptor stimulus applied to the RE cell locat-
ed at the boundary of the network triggered a localized pattern of
spike-burst activity that propagated with constant velocity through
the RE network (Fig. 5a). The mechanism of propagation depend-
ed on the level of the membrane potential in RE cells. At mem-
brane potentials of about 75 mV, the low-threshold calcium
current in RE cells was deinactivated. Bursts of spikes in presy-
naptic RE cells led to reversed GABAA receptor IPSPs, followed by
a low-threshold calcium spike and a burst of sodium spikes in
neighboring RE cells (Fig. 5b). The temporal inactivation of the
low-threshold calcium current in an RE cell after a burst discharge
prevented oscillations from developing in the cell. All RE cells were
grouped into clusters that fired simultaneously, whose size depend-
ed on the radius of synaptic interconnections (N). Only some of

a b

Fig. 3. Spontaneous activity of RE neurons. Different cells (a and b) are from two decorticated cats. At the resting membrane potential (60 mV), the
RE neurons showed periodic hyperpolarizing potentials (marked by asterisks). Under slight DC hyperpolarization (76 mV in a and 68 mV in b), the
hyperpolarizing potentials were not longer visible, but high-frequency spike bursts occurred with approximately the same period icity. Fragments
expanded at right are marked by large asterisks.

170 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 4. Response of a model RE cell stimu-

network. Only transient waves were observed in networks
lated by a 5-Hz train of GABAA receptor
IPSPs at different levels of membrane smaller than about 25 25 cells.
potential. At a resting membrane potential A similar transformation of transient oscillations into self-
of 65 mV, the low-threshold calcium cur- sustained oscillations in a two-dimensional RE network could
rent was inactivated, and the GABAA be produced by depolarizing the RE cells. In typical time
receptor IPSP was followed by a weak traces of RE cells from a network with 33 33 RE cells at dif-
rebound LTS with a single spike. For Vm = ferent levels of membrane potential produced by DC current
76 mV, the IPSP was reversed and evoked injection (Fig. 8), only transient oscillations in the form of
a burst of sodium spikes. Inactivation of cylindrical waves, similar to those in Fig. 7a, were observed
the low-threshold calcium current during
at the resting membrane potential. A weak depolarization
the first burst discharge reduced the
response to the second shock.
transformed them into an almost periodic self-sustained
oscillation (spiral wave) with a frequency around 2.5 Hz (Fig.
8a). Further depolarization increased the population fre-
quency up to about 10 Hz (Fig. 8b) and reduced the ampli-
tude of the average activity (compare the amplitude scales in
the RE cells from each cluster displayed sodium spikes, whereas Fig. 8a and b). The decrease reflected a reduction from four spikes
others showed reversed IPSPs and partial calcium spikes but no per burst discharge in the RE cells at more hyperpolarized levels of
sodium spikes (see cells 6 and 8 in Fig. 5). membrane potential into weaker and mostly single-spike discharges
1999 Nature America Inc. http://neurosci.nature.com

IPSP-based waves of excitation propagated through the network when the RE cells were only slightly hyperpolarized below the chlo-
with a constant velocity of 25150 cells per second (Fig. 6). The size ride reversal potential. Population oscillations with a frequency
of the clusters of RE cells that were synchronously activated on each around 10 Hz were produced by RE cells oscillating at around 5 Hz,
cycle of oscillation increased with N, which increased the propaga- which occurred by delayed activation of the low-threshold calcium
tion speed. Increasing the GABAA receptor conductances (gGABAA) spikes in RE cells placed just below the chloride reversal potential.
between the RE cells reduced the time delay between IPSP onset and These neurons showed long (13 s) epochs of subthreshold oscilla-
the appearance of sodium spikes, which also raised the speed when tions interrupted by the sequences of spiking-bursting activity.
gGABAA was less than 0.2 S. For gGABAA > 0.2 S, the speed of prop-
agation decreased for N < 7 or displayed slow oscillations around DISCUSSION
an average value, which depended on the radius of interconnections, Our intracellular recordings from RE neurons at resting and hyper-
for N > 7. Stronger GABAA receptor inhibition between RE cells polarized membrane potentials in vivo showed that reversed IPSPs
reduced the length of the burst discharges23, which limited an between RE cells can directly trigger an LTS. Models of thalamic net-
increase in the speed. Propagating patterns did not occur for gGABAA works examined here demonstrate that this reversal may have impor-
< 0.04 S or for gGABAA > gGABA
lim lim
A, where the limiting gGABAA
increased with increasing N (see Fig. 6). Wave activity was
also absent for all networks with N < 2.
In a one-dimensional network with flow boundary a
conditions (see Methods), the localized patterns termi-
RE cell number

nated at the boundaries. To check the effect of network

geometry, we simulated a two-dimensional network of
RE cells hyperpolarized below the chloride reversal poten-
tial. The spatiotemporal patterns in a two-dimensional
network depended on the maximal conductance for the
low-threshold calcium current in RE cells. For gCa below
Time (ms)
some critical value (gCalim
2.3 mS/cm2 for a network of 50
50 cells with N = 4), a single stimulus applied at the
boundary led to a cylindrical wave of spike-burst activity b
that traveled through the network and disappeared at the
boundary (Fig. 7a). The wavefront consisted of almost
synchronously firing RE cells. Increase of the maximal
lim
conductance for IT above gCa led to self-sustained oscil-
lations; the wave of excitation evoked by the initial stim-
ulus transformed into a rotating spiral wave (Fig. 7b). The
location of the spiral center was random and depended
on the initial stimulation as well as the variability in the
parameters of the cells in the network. The individual cells
from the network showed bursts of sodium spikes every
250300 ms corresponding to the fronts of the spiral wave.
Further increase of the maximal conductance for the low-
Fig. 5. Localized pattern of activity propagating through the model of an isolated RE net-
threshold calcium current transformed the single spiral
work. (a) RE cell #1 was stimulated at t = 0, which triggered a localized pattern that
wave into nonregular burst discharges of the RE cells. The traveled with constant velocity through the RE network. (b) Six neighbor RE cells from
critical values of the maximal conductance for IT at tran- the network. A burst of spikes in one RE cell led to a depolarizing GABA IPSP followed
A
sitions between the transient mode (single cylindrical by an LTS and a burst of sodium spikes in the neighboring RE cells. Temporal inactivation
wave), the periodic oscillations (spiral wave) and the non- of the low-threshold calcium current in RE cell after the burst discharge prevented the
regular oscillations decreased with increasing size of the oscillations from developing. Triangles mark t = 0, the time of stimulation.

nature neuroscience volume 2 no 2 february 1999 171

 1999 Nature America Inc. http://neurosci.nature.com

articles

a b

RE cell #
RE cell #
Velocity (cells/ms)

Time (ms) Time (ms)

Fig. 6. Velocity of the traveling wave as a function of GABAA receptor coupling
strength and radius of RERE connections in a one-dimensional RE network
model. (a) Velocity of propagation as a function of GABAA conductance. N is the
radii of connections. For N > 6 and gGABAA > 0.2 S, the speed of propagation
showed only a weak dependence on GABAA receptor coupling and radius of
RERE connections. The solid curves are the nonlinear fits of the calculated data
GABAA (S)
(indicated by different symbols). (b) Localized activity patterns propagating with
different speed: gGABAA = 0.1 S, N = 2 (left) and gGABAA = 0.4 S, N = 7 (right).

tant consequences for the collective dynamics of large populations of by spiral waves of electrical excitation in heart cells32. However, local
1999 Nature America Inc. http://neurosci.nature.com

RE cells. A single stimulus applied to an isolated one-dimensional electrical coupling is required to maintain spiral waves in this sys-
RE network hyperpolarized below the chloride reversal potential tem, which leads to the same mechanism of propagation as in the
triggered an isolated wave of spike-burst activity that traveled reaction-diffusion systems. In our RE network model, the mecha-
through the RE network and either terminated at the boundaries or nisms promoting pattern formation did not depend on having local
circulated indefinitely when periodic boundary conditions were interconnections between cells. The burst activity in the presynaptic
used. The mechanism of propagation was based on GABAA-recep- RE neurons initiated burst discharges in the target cells through the
tor-mediated depolarization, followed by a calcium spike in the post- GABAA-receptor-mediated depolarization when the neurons were
synaptic RE cells. Similar localized patterns have been described in below the chloride reversal potential. At the same time, these synaps-
network models of cortical excitatory cells24,25 and simplified es had an inhibitory influence among synchronously firing RE neu-
inhibitory cells26; however, in the latter study, wave propagation was rons. Such dual excitatory/inhibitory synaptic interconnections may
caused by asymmetric inhibition between neurons and
rebound bursts, but not by direct triggering of the LTS
by IPSPs. Activity patterns traveling through networks
a
of excitatory and inhibitory cells have been investigat-
ed in computational models of hippocampal slices27,28.
In these models, the propagation from one cell to
another was mediated mostly by AMPA receptors
between pyramidal neurons, which also synchronized
the burst discharges elicited by NMDA-receptor-medi-
ated synaptic currents.
A one-dimensional network of RE cells serves as a
model for an RE slice preparation cut transversely to
the thin, two-dimensional sheet of RE neurons19. We
also tested a two-dimensional RE network model
hyperpolarized below the chloride reversal potential.
In this model, the transient wave patterns were trans- b
formed into persistent oscillations in the form of the
spiral waves if the maximal conductance for the IT cur-
rent in RE cells exceeded a critical value, which depend-
ed on the size of the network and the level of membrane
potential in the RE cells. The spiral pattern that
appeared when gCa for IT reached a bifurcation is sim-
ilar to those in general reaction-diffusion systems,
which may display spiral patterns near the point of
bifurcation from spatially homogeneous solutions29. If
the size of the system is large enough, coexisting spirals
occur in the large-scale, two-dimensional, reaction-dif-
fusion model, and an increase of the control parameter
in this system leads to defect-media turbulence in the
form of spontaneously appearing and annihilating spi- Fig. 7. Localized patterns of activity in a two-dimensional network model of 50 50 RE
cells (x- and y-axes) with flow boundary conditions. (a) The maximal conductance for
ral waves30,31. This suggests that in large-scale, two- low-threshold calcium current in RE cells was g = 2.2 mS/cm2. Initial stimulation of the
T
dimensional networks of RE cells, regimes similar to RE cell #(1,1; left upper corner) at t = 0 led a cylindrical wave traveling through the RE
defect-media turbulence may be observed. One of the network with a constant velocity. (b) gT = 2.45 mS/cm2. Increasing the maximal conduc-
first known examples of spiral-wave activity in a bio- tance for low-threshold calcium current led to self-sustained patterns of activity in the
logical system is cardiac arrythmias, which are initiated form of a rotating one-arm spiral wave.

172 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
Fig. 8. The influence of the membrane poten-
a
tial on the frequency of oscillations in a two-
dimensional network model with 33 33 RE
cells. Left, the membrane potentials of two RE
cells and the average activity of 112 RE cells
from the center of the network. Right, the
Fourier spectrum of the average activity. (a) A
relatively hyperpolarized RE network showed
oscillations at around 2.5 Hz. (b) After depo-
larization (just below the chloride reversal
potential), the frequency of the population
oscillations shifted to around 9 Hz. In the
b
Fourier spectrum, the smaller peak at about
4.5 Hz corresponded to the oscillation fre-
quencies of the individual RE cells.
1999 Nature America Inc. http://neurosci.nature.com
explain the formation of localized activity patterns without highly
synchronous oscillatory states in a network with nonlocal coupling
between neurons.
The same synaptic mechanisms were responsible for traveling
patterns in one-dimensional chains and two-dimensional lattices of
RE cells. In particular, the speed of the two-dimensional waves was
close to that of traveling waves in a one-dimensional RE network.
The same conditions on cellular and network parameters generated
propagating activity patterns in all of the networks studied. In its
oscillatory mode, the average membrane potentials of RE cells in a
two-dimensional RE network varied with frequencies from 2 to 10
Hz depending on the absolute levels of the membrane potentials.
The frequency increased with depolarization until it reached around
10 Hz just below the chloride reversal potential, which is close to the
frequency of spindle oscillations recorded in the isolated RE nucle-
us in vivo9. At relatively hyperpolarized levels of membrane poten-
tial, the network oscillations were maintained through the
re-excitation of RE cells by the circulating waves of bursting neu-
rons. When cells were sufficiently depolarized, the spatiotemporal
activity in the RE network no longer resembled rotating spirals but
became more or less irregular. The remaining synchrony of network
activity was reflected in occasional 910 Hz population oscillations,
lasting 0.51 second and separated by epochs of less regular oscilla-
tions (see Fig. 8b). A similar temporal pattern of alternations was
observed in a previous network model of RE cells depolarized above
the chloride reversal potential, where coherent oscillatory activity
was accompanied by travelling waves11. However, in that model, the
oscillations were driven by GABAA-receptor-mediated hyperpolar-
ization followed by rebound bursts in RE cells.
The proposed mechanism of reciprocal excitation between
GABAergic cells is not unique to the RE nucleus of the thalamus.
In-vitro recordings from hippocampal slices have revealed that some
GABAergic synapses between interneurons and onto pyramidal neu-
rons could function as excitatory synapses3335. This effect could be
explained by the depolarizing component of GABAA receptor IPSPs
whose reversal potential depends on the chloride concentration dif-
ference across the plasma membrane36,37.
Spike-bursting activity in RE nuclei may be important in initi-
ating sequences of spindle oscillations in the intact RETC network.
Depolarization of TC cells during a sequence of spindle oscillations
terminates oscillations in these cells but may also trigger a set of pat-
terns propagating through the RE network (M.B., I.T., M.S. and
T.J.S., unpublished). Slow repolarization of TC cells during an inter-
nature neuroscience volume 2 no 2 february 1999
articles
Amplitude (mV)
50 mV
5 mV Frequency (Hz)
Amplitude (mV)
50 mV
1 mV Frequency (Hz)
spindle lull deinactivates the low-threshold calcium current, so the
local RE-evoked IPSPs could trigger a new spindle sequence. In-vivo
data presented here indicate that persistent activity in an RE nucle-
us during interspindle intervals occurs at low frequencies (0.33
Hz). This confirms the suggestion that the mechanism underlying
this activity is based on the reversal of GABAA receptor IPSPs fol-
lowed by calcium spikes in RE cells. If this hypothesis is correct, then
persistent low-frequency activity in the RE nucleus will lead to the
more or less simultaneous (within 1 to 2 cycles of spindle oscilla-
tions) activation of new spindle sequences at different thalamic foci,
whereas transient patterns will result in nonsynchronous initiation
of spatially localized spindle sequences. Future experiments will test
these predictions.
In a previous model of spindle oscillations, new spindle
sequences were triggered without persistent activity inside the RE
network by including spontaneously oscillating (initiator) TC cells38.
Our model predicts that even in the absence of such TC cells, the
RE nucleus could initiate sequences of spindle oscillations in thal-
amocortical networks in vivo.
METHODS
In-vivo recordings. In-vivo experiments were done on 46 unilaterally decor-
ticated cats and 52 cats with intact thalamocortical connections. All animals
were maintained under ketamine and xylazine (1015 mg/kg and 23 mg/kg,
i.m.) anesthesia. In addition, tissues to be incised and pressure points were
infiltrated with lidocaine. The EEG from the intact hemisphere was contin-
uously recorded, and additional doses of anesthetics were administered at
the slightest tendency toward an increase in frequency and decrease in ampli-
tude of EEG field potentials. The cats were paralyzed with gallamine tri-
ethiodide and artificially ventilated to the end-tidal CO2 of 3.53.8%. The
heartbeat was monitored and kept constant (acceptable range, 90110
beats/min). Body temperature was maintained at 3739oC. Glucose saline
(5% glucose, 10 ml i.p.) was given every 34 h during the experiments, which
lasted for 814 h. The stability of intracellular recordings was ensured by
cisternal drainage, bilateral pneumothorax, hip suspension and filling the
hole made in the skull with a solution of agar-agar (4%). All experimental
procedures were done according to national guidelines.
For microelectrode recordings from thalamic RE neurons, the surface of
the cortex that corresponds to the anterior half of the marginal and supra-
sylvian gyri was cauterized with silver nitrate. The cortex and white matter
were removed by suction until the head of the caudate nucleus was exposed.
Micropipettes were then lowered through the head of the caudate nucleus
to reach the rostrolateral sector of the RE nucleus. Intracellular recordings
were made with conventional sharp electrodes filled with a 2.5 M solution of
potassium acetate (DC resistance of 3070 M). Stable intracellular record-
ings had resting membrane potentials more negative than 55 mV and over-
173
 1999 Nature America Inc. http://neurosci.nature.com
articles
shooting action potentials. Stimulations of the ventrolateral (VL) thalamic
nucleus and motor cortical area 4 were delivered with variable durations
(0.050.2 ms) and intensities (0.050.3 mA). To suppress cortical activity in
5 experiments, the 2 M solution of potassium acetate was washed over the
cortical surface in area 4. Multi-neuron activity in the RE nucleus was record-
ed with tungsten microelectrodes (resistance 710 M). At the end of an
experiment, the animal was given a lethal dose of pentobarbital.
Computational model. We examined a single-compartment model of an
RE cell, which included voltage-dependent currents described by Hodgkin-
Huxley kinetics:
Cm dV = g (V E ) Iint Isyn (1) 15. Domich, L., Oakson, G. & Steriade, M. Thalamic burst patterns in the naturally
dt L L
where Cm = 1 /cm2 is the membrane capacitance, gL = 0.05 mS/cm2 is the
leakage conductance, EL = 77 mV is the reversal potential, Iint is a sum of
active intrinsic currents (Iintj) and Isyn is a sum of synaptic currents (Isynj).
The area of the RE cell was SRE = 1.43 x 104 cm2. The set of intrinsic cur-
rents included a fast sodium current INa, a fast potassium current IK (ref. 39),
a low-threshold calcium current IT (refs. 40, 41) and a potassium leak current
IKL = gKL(V EKL). The expressions for the transition rates of all currents
1999 Nature America Inc. http://neurosci.nature.com
are given in ref. 23.
GABAA receptor synaptic current was modeled by a first-order activation
scheme42 and was calculated according to the equation:
Isyn = gsyn [O] (VEsyn) (2)
where gsyn is the maximal conductivity, [O](t) is the fraction of open channels
and Esyn = 70 (ref. 12) is the reversal potential. The decay time constant of the
GABAA synapses between RE cells was varied from 10 ms to 100 ms43 to
ensure the robustness of the results. We found that the main difference
between the models with slow and fast synaptic decay kinetics was the dura-
tion of the burst discharges in RE cells, which was reduced slightly when slow-
er kinetics was used. The rest of the parameters were the same as in ref. 23.
We simulated two network models: (1) A one-dimensional chain of
M = 100 RE cells and (2) a two-dimensional lattice of M M (33 33
and 50 50) RE cells. In the first model, each RE cell xi (i [1,M]) was
connected with N nearest cells (xj, j [i-N,i+N], N = 1 to 9), and in
the second model each RE cell (xi,j) was connected with all cells inside a
circuit with radius N = 4 cells (xk,l, (k,l) S, S = {(k,l): k2+l2 N2 }).
Flow boundary conditions (V j = V j , j = j+1 if j [N+1,0],
j = 2Mj+1 if j [M+1,M+N] in one dimension or Vk.l = Vk,l in two 28. Traub, R. D., Miles, R. & Jefferys, J. G. R. Analysis of the propagation of
dimensions with the same rules for calculation k and l) were used to
reduce the effect of the boundaries. The maximal conductance for IKL
was initialized with some random variability (variance 10%) through
the network to insure the robustness of the results 23.
ACKNOWLEDGEMENTS
This research was supported by the Howard Hughes Medical Institute, the Sloan
Center for Theoretical Neurobiology, the Human Frontier Science Program and the
Medical Research Council of Canada.
RECEIVED 14 SEPTEMBER; ACCEPTED 16 NOVEMBER 1998
1. Steriade, M., Deschnes, M., Domich, L. & Mulle, C. Abolition of spindle
oscillations in thalamic neurons disconnected from nucleus reticularis thalami.
J. Neurophysiol. 54, 14731497 (1985).
2. Steriade, M. & Llins, R. R. The functional states of the thalamus and the
associated neuronal interplay. Physiol. Rev. 68, 649742 (1988).
3. Steriade, M., Jones, E. G. & Llins, R. R. Thalamic Oscillations and Signaling
(Wiley, New York, 1990).
4. von Krosigk, M., Bal, T. & McCormick, D. A. Cellular mechanisms of a
synchronized oscillation in the thalamus. Science 261, 361364 (1993).
5. Steriade, M., McCormick, D. A. & Sejnowski, T. J. Thalamocortical oscillations
in the sleeping and aroused brain. Science 262, 679685 (1993).
6. Bal, T. & McCormick, D. A. What stops synchronized thalamocortical
oscillations? Neuron 17, 297308 (1996).
7. Budde, T., Biella, G., Munsch, T. & Pape, H. C. Lack of regulation by
intracellular Ca2+ of the hyperpolarization-activated cation current in rat
thalamic neurones. J. Physiol. (Lond.) 503, 7985 (1997).
8. Lthi, A., Bal, B. & McCormick, D. A. Periodicity of thalamic spindle waves is
abolished by ZD7288, a blocker of Ih. J. Neurophysiol. 79, 32843289 (1998).
9. Steriade, M., Domich, L., Oakson, G. & Deschnes, M. The deafferented
reticular thalamic nucleus generates spindle rhythmicity. J. Neurophysiol. 57,
260273 (1987).
174
10. Golomb, D., Wang, X. J. & Rinzel, J. Synchronization properties of spindle
oscillations in a thalamic reticular nucleus model. J. Neurophysiol. 72,
11091126 (1994).
11. Destexhe, A., Contreras, D., Sejnowski, T. J. & Steriade, M. A model of spindle
rhythmicity in the isolated thalamic reticular nucleus. J. Neurophysiol. 72,
803818 (1994).
12. Ulrich, D. & Huguenard, J. R. Nucleus-specific chloride homeostasis in rat
thalamus. J. Neurosci. 17, 23482354 (1997).
13. Ulrich, D. & Huguenard, J. R. -Aminobutyric acid type B receptor-dependent
burst-firing in thalamic neurons: A dynamic clamp study. Proc. Natl. Acad. Sci.
USA 93, 1324513249 (1996).
14. Contreras, D., Curr Dossi, R. & Steriade, M. Electrophysiological properties
of cat reticular thalamic neurones in vivo. J. Physiol. (Lond.) 470, 273294
(1993).
sleeping cat: a comparison between cortically projecting and reticularis
neurones. J. Physiol. (Lond.) 379, 429449 (1986).
16. Contreras, D. & Steriade, M. Cellular basis of EEG slow rhythms: a study of
dynamic corticothalamic relationships. J. Neurosci. 15, 604622 (1995).
17. Destexhe, A., Contreras, D., Steriade, M., Sejnowski, T. J. & Huguenard, J. R. In
vivo, in vitro, and computational analysis of dendritic calcium currents in
thalamic reticular neurons. J. Neurosci. 16, 169185, (1996).
18. Contreras, D., Timofeev, I. & Steriade, M. Mechanisms of long-lasting
hyperpolarizations underlying slow sleep oscillations in cat corticothalamic
networks. J. Physiol. (Lond.) 494, 251264 (1996).
19. Steriade, M., Jones, E. G. & McCormick, D. A. Thalamus: Organization and
Function (Elsevier, Oxford, 1997).
20. Timofeev, I. & Steriade, M. Low-frequency rhythms in the thalamus of intact-
cortex and decorticated cats. J. Neurophysiol. 76, 41524168 (1996).
21. Ulrich, D. & Huguenard, J. R. GABAA-receptor-mediated rebound burst firing
and burst shunting in thalamus. J. Neurophysiol. 78, 17481751 (1997).
22. Timofeev, I. & Steriade, M. Cellular mechanisms underlying intrathalamic
augmenting responses of reticular and relay neurons. J. Neurophysiol. 79,
27162729 (1998).
23. Bazhenov, M., Timofeev, I., Steriade, M. & Sejnowski, T. J. Cellular and
network models for intrathalamic augmenting responses during 10 Hz
stimulation. J. Neurophysiol. 79, 27302748 (1998).
24. Golomb, D. & Amitai, Y. Propagating neuronal discharges in neocortical slices:
Computational and experimental study. J. Neurophysiol. 78, 11991211 (1997).
25. Golomb, D. Models of neuronal transient synchrony during propagation of
activity through neocortical circuitry. J. Neurophysiol. 79, 112 (1998).
26. Rinzel, J., Termal, D., Wang, X. J. & Ermentrout, B. Propagating activity
patterns in large-scale inhibitory neuronal networks. Science 279, 13511355
(1998).
27. Traub, R. D., Miles, R. & Jefferys, J. G. R. Synaptic and intrinsic conductances
shape picrotoxin-induced synchronized after-discharges in the guinea-pig
hippocampal slice. J. Physiol. (Lond.) 461, 525547 (1993).
disinhibition-induced after-discharges along the guinea-pig hippocampal slice
in vitro. J. Physiol. (Lond.) 472, 267287 (1993).
29. Kuramoto, Y. Chemical Oscillations, Waves, and Turbulence (Springer-Verlag,
Berlin, 1984).
30. Coullet, P., Gil, L. & Lega, J. Defect-mediated turbulence. Phys. Rev. Lett. 62,
16191622 (1989).
31. Bazhenov, M. & Rabinovich, M. Synchronized disorder in a 2D complex
Ginzburg-Landau equation. Physica D 73, 318334 (1994).
32. Winfree, A. When Time Breaks Down (Springer, New York, 1987).
33. Michelson, H. B. & Wong, R. K. S. Excitatory synaptic responses mediated by
GABAA receptors in the hippocampus. Science 253, 14201423 (1991).
34. Michelson, H. B. & Wong, R. K. S. Synchronization of inhibitory neurones in
the guinea-pig hippocampus in vitro. J. Physiol. (Lond.) 477, 3545 (1994).
35. Staley, K. J., Soldo, B. L. & Proctor, W. R. Ionic mechanisms of neuronal
excitation by inhibitory GABAA receptors. Science 269, 977981 (1995).
36. Nicoll, R. A., Malenka, R. C. & Kauer, J. A. Functional comparison of
neurotransmitter receptor subtypes in mammalian central nervous system.
Physiol. Rev. 70, 513565 (1990).
37. Qian, N. & Sejnowski, T. J. When is an inhibitory synapse effective. Proc. Natl.
Acad. Sci. USA 87, 81458149 (1990).
38. Destexhe, A., Bal, T., McCormick, D. A. & Sejnowski, T. J. Ionic mechanisms
underlying synchronized oscillations and propagating waves in a model of
ferret thalamic slices. J. Neurophysiol. 76, 20492070 (1996).
39. Traub, R. D. & Miles, R. Neuronal Networks of the Hippocampus (Cambridge,
Cambridge Univ. Press, 1991).
40. Huguenard, J. R. & Prince, D. A. A novel T-type current underlies prolonged
Ca2+-dependent burst firing in GABAergic neurons of rat thalamic reticular
nucleus. J. Neurosci. 12, 38043817 (1992).
41. Huguenard, J. R. & McCormick, D. A. Simulation of the currents involved in
rhythmic oscillations in thalamic relay neurons. J. Neurophysiol. 68, 13731383
(1992).
42. Destexhe, A., Mainen, Z. F. & Sejnowski, T. J. Synthesis of models for excitable
membranes, synaptic transmission and neuromodulation using a common
kinetic formalism. J. Comp. Neurosci. 1, 195230 (1994).
43. Zhang, S. J., Huguenard, J. R. & Prince, D. A. GABAA receptor-mediated Cl
currents in rat thalamus reticular and relay neurons. J. Neurophysiol. 78,
22802286 (1997).
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

Neural correlates of a decision in

the dorsolateral prefrontal cortex
of the macaque
Jong-Nam Kim and Michael N. Shadlen

Department of Physiology and Biophysics and Regional Primate Research Center, University of Washington Medical School, Box 357290,
Seattle, Washington 98195-7290, USA
Correspondence should be addressed to M.N.S. (shadlen@u.washington.edu)

To make a visual discrimination, the brain must extract relevant information from the retina,
1999 Nature America Inc. http://neurosci.nature.com

represent appropriate variables in the visual cortex and read out this representation to decide which
of two or more alternatives is more likely. We recorded from neurons in the dorsolateral prefrontal
cortex (areas 8 and 46) of the rhesus monkey while it performed a motion discrimination task. The
monkey indicated its judgment of direction by making appropriate eye movements. As the monkey
viewed the motion stimulus, the neural response predicted the monkeys subsequent gaze shift,
hence its judgment of direction. The response comprised a mixture of high-level oculomotor signals
and weaker visual sensory signals that reflected the strength and direction of motion. This combina-
tion of sensory integration and motor planning could reflect the conversion of visual motion
information into a categorical decision about direction and thus give insight into the neural compu-
tations behind a simple cognitive act.

The brain uses sensory information to form interpretations and
decisions that guide behavior. Such interpretations often out-
last the fleeting sensory impressions on which they are based,
so that sensory input can motivate subsequent behavior. To study
this process, we trained rhesus monkeys to discriminate the
direction of motion in a dynamic random dot display1. The dif-
ficulty of the task was controlled by varying the fraction of coher-
ently moving dots. At high motion coherences, the animal can
commit to an action the moment the stimulus is seen. In con-
trast, near psychophysical threshold, the monkey must base its
direction judgments on weak sensory signals perceived over hun-
dreds of milliseconds1.
The extrastriate visual cortex (areas MT and MST) contains
the neural representation of visual motion that allows the monkey
to perform this demanding task14. The direction-selective neu-
rons in these areas represent the evidence favoring one direction
or another, but this evidence must be interpreted to reach a deci-
sion. This distinction between sensory evidence and decision can
be appreciated by imposing a delay between motion viewing and
behavioral response. Direction-selective neurons stop respond-
ing when the motion stimulus is absent5, whereas neurons that
encode a decision must maintain their activity when the visual
motion is no longer present, until the animal responds.
The dorsolateral prefrontal cortex (PFC) is felt to be critical
for tasks with a delay between instruction and execution6,7. Dur-
ing the delay, many PFC neurons show sustained discharge, which
is often selective for a particular object or location811. We studied
neurons with sustained activity through the memory/delay peri-
od before an eye movement to a restricted region of the visual
field, termed the neural response field (RF). We hypothesized that
such neurons may be involved in linking the sensory evidence in
visual cortex to a behavioral plan to shift the gaze. To test this, we
arranged the motion-discrimination task so that one of two
response targets appeared in the neurons RF (Fig. 1a). The mon-
key was trained to make a delayed eye movement to one or the
other response target, depending on the direction of random dot
motion. We found that the activity of many PFC neurons reveals
the monkeys intention to make a saccade to one or the other tar-
get. The time course and intensity of neural activity seem to rep-
resent the formation of a decision about motion direction.
RESULTS
We studied 88 neurons in the frontal eye field (FEF) and the pos-
terior third of the principal sulcus (PS) region (areas 8Ar and 46)
that responded selectively when the monkey planned a saccade
to a region of space (the RF). For example, the PS neuron shown
in Fig. 2 responded during the delay preceding saccades to tar-
gets that appeared up and to the right, but not down and to the
left of the fixation point (Fig. 2a and b).
To determine this neurons behavior during a perceptual deci-
sion, we monitored its discharge while the monkey judged the
direction of random dot motion. The random dots appeared in
a five degree aperture outside the neurons RF (see Methods).
Motion direction was toward or away from the RF, and motion
strength was varied to span psychophysical threshold. After a delay,
the monkey indicated its direction judgment by making an eye
movement to one of the two targets. If the motion was up-right,
the monkey was rewarded for choosing the target inside the RF.
Conversely, if the motion was down-left, the monkey was reward-
ed for choosing the other target, outside of the RF.
This neurons response predicted the monkeys decision. The
response was larger on trials in which the monkeys eyes moved
toward the RF (Fig. 2c, e and g) and attenuated when they moved
away from the RF (Fig. 2d, f and h). The spike discharge from this

176 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 1. Behavioral tasks and neu-

ron locations. (a, b) Direction-dis- a
crimination task. The monkey
gazed at the fixation point for 350
ms. Then two targets appeared,
one of which was in the neural
response field (RF, shaded). After
200300 ms, the random dot kine-
matogram appeared between the b motion delay saccade
targets and outside the RF. The
direction of motion was toward
one of the two targets. Motion
strength was varied from trial to
trial by adjusting the percentage of
coherently moving dots. After 1 s, e
m
the random dots were turned off, Ti
leaving only the fixation point and
targets. After 0.51.5 s, the fixa-
tion point was extinguished, signal-
ing the monkey to indicate its
1999 Nature America Inc. http://neurosci.nature.com

choice by shifting its gaze to one of

the targets. The monkey was
rewarded for choosing the target
along the direction of random dot
Probability correct (mean stdev)

motion, or randomly when there c

was no net motion (0% coher- d flash delay saccade
ence). T1 and T2, saccade targets;
FP, fixation point. (c) Average psy-
chometric function for 88 experi-
ments. Error bars are standard
deviations of the proportion of
correct choices. (d) Memory-sac-
cade task used to screen neurons.
A target was flashed (100 ms) at a
random location in the visual field.
The monkey maintained fixation
through a variable delay until the Motion strength (% coherence)
fixation point was extinguished.
The monkey was then required to
shift its gaze to the remembered
location of the flashed target.
(e) Location of the recording cylin- e f
der in a schematic diagram of the
rhesus monkey brain. (f) Magnetic
resonance imaging. Fast spin-echo,
short-T1, inversion-recovery scan
through the electrode grid of mon-
key S. This is one of a series of
images obtained in the coronal
plane (slice thickness 1.5 mm). The
recording grid was filled with ster-
ile saline to reveal the angle and
location of electrode guide tubes in the coronal plane. A second series was obtained in the sagittal plane to determine the position and angle
of guide tubes in the anteriorposterior direction. The section shows the arcuate sulcus and prearcuate gyrus at the caudal end of the princi-
pal sulcus (ps, principal sulcus; as, arcuate sulcus).

neuron began to reveal the monkeys decision as early as 200300 ms
after the onset of random dot motion and remained informative
until the saccade. Moreover, the response was modulated more
strongly when the task was easier: the neuron discharged more
intensely when the monkey viewed coherent motion toward the RF
(up-right) and attenuated more profoundly to coherent motion
away from the RF. Thus, the response reflected not only the mon-
keys impending eye movement, but also the sensory input that
determined it. This mixture of visual-sensory and visuomotor
response properties is thought to occur at the nexus of sensory-to-
motor conversion1214 where the decision is computed.
Although this pattern of response was common in the FEF
and PS region, we also encountered many neurons that modu-
lated their activity only during the delay after the random dot
motion was turned off, presumably after the monkey had reached
its decision. This activity (Fig. 3a) clearly predicted the monkeys
plan to look to the target in the RF on the memory-saccade task.
However, during the motion-discrimination task, the response
did not reveal the monkeys decision until the delay (Fig. 3c
and d). During motion viewing, this neuron failed to signal the
direction of the next eye movement. Such responses may reflect
motor preparation, but they provide little insight into the deci-

nature neuroscience volume 2 no 2 february 1999 177

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 2. Response of a principalis neuron during the

motion-discrimination and memory-saccade tasks.
The response field of the neuron is shown in gray.
The arrow indicates the direction of the monkeys
saccade at the end of the trial. (a, b) Response on
memory saccades to targets in (a) and out (b) of a b
the RF. The time axis is broken to align the
response to two events: target appearance and
saccade initiation. (ch) Response during the
motion-discrimination task. Responses are aligned
to the onset of random dot motion and to the
time of the saccade. The left column (c, e, g)
depicts trials in which the monkey decided the
motion was toward the RF. The right column (d, f,
h) shows trials in which the monkey decided that
the direction was away from the RF, leading to a
saccade to the target outside the RF. Three motion
strengths are shown (sub-, near- and supra-thresh-
old). For the 0% coherence stimulus, there is no
net direction of motion. Only correct discrimina-
1999 Nature America Inc. http://neurosci.nature.com

tion trials are shown for nonzero motion

strengths. For clarity, only 10 trials are shown in
the rasters. The response preceding the onset of c d
motion was associated with fixation and the
appearance of the saccade targets.

sion-making process. Unless the monkey

makes eye movements capriciously a pos- e f
sibility precluded by the well behaved psy-
chometric function (Fig. 1c) the decision
must form during motion viewing.
For each neuron, we computed an index
of predictive activity using the responses
measured during motion viewing and dur-
ing the delay. The index measures the g h
response distribution overlap from trials in
which the monkey judged motion as
toward versus away from the RF and esti-
Spikes/s

mates the probability that an ideal observ-

er could predict the monkeys decision
based on the spike rate (Methods). For
example, an index of 0.5 indicates a chance
association (complete overlap of response
distributions associated with the two choic-
es), whereas an index of 1 indicates perfect correspondence
between predicted and observed choices (no overlap).
Most neurons (76/88, 86%) predicted the monkeys choice
reliably during motion viewing or the delay or both (Fig. 4; pre-
dictive index > 0.5 and p < 0.01 in at least one epoch). Sixty
percent (53/88) predicted the monkeys decision reliably dur-
ing motion viewing (index > 0.5 and p < 0.01) and could reflect
the association between motion processing and behavioral
response. However, about 1/4 of the neurons failed to indicate
the monkeys choices until the delay (9/26 neurons in the FEF
and 10/62 in the posterior PS region), which is too late for them
to be involved in the decision process.
Some neurons (11%, 1 FEF, 9 PS) failed to predict the mon-
keys response at any time during the discrimination task (values
near 0.5 on both axes, p > 0.01). Five neurons (1 FEF, 4 PS) reliably
predicted the monkeys choices during motion viewing but in a
pattern opposite to their response on the memory-saccade task
(predictive index < 0.5, p < 0.01). The activity of these five neurons
diminished when the monkey chose the target in the RF, but none
retained this reversed activity pattern during the delay. We tend-
ed to encounter similar patterns of predictive activity in the FEF
and PS region (Fig. 4, p > 0.08, two-dimensional Kolmogorov-
Smirnov test15), which may result from our strategy of sampling
neurons of similar type in both areas.
Time course of predictive activity
To perform above chance on this task, the monkey must decide
about the motion direction based on the evidence acquired
during motion viewing. Those neurons that predicted the mon-
keys subsequent eye movement during motion viewing may
therefore divulge properties of the decision process itself, the
linkage between sensory processing and saccade planning. A
neural correlate of the decision process is likely to reflect both
the outcome of the decision and the quality of the evidence
upon which it is founded. Moreover, neurons linking sensory
evidence to a behavioral response should undergo a temporal
transformation in their activity as the evidence produces a cat-
egorical answer. Early responses might reflect properties of the
sensory stimulus (the strength of the evidence). Responses later
in the decision process should reflect a stereotyped outcome,

178 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
Movement/memory
response field
a b
Spikes/s
1999 Nature America Inc. http://neurosci.nature.com
c d
regardless of whether the evidence was strong or weak, and
whether it was interpreted correctly or incorrectly. These prop-
erties are evident in the PFC response.
Motion strength affected the response of many PFC neurons
(Fig. 5a and b). For this analysis, we selected neurons that pre-
dicted the monkeys subsequent choice during motion viewing
(n = 53 neurons with predictive index > 0.5 and a significant
permutation test, p < 0.01). For each neuron, we computed the
average spike rate during motion viewing (from 200800 ms
after the onset of dot motion) and normalized the spike rate to
the mean. We applied this procedure separately for the two
directions of motion and restricted the analysis to correct choic-
es. For motion toward the RF, the degree of response enhance-
ment varied by 12.5% across the range of motion strengths
(~3.2 spikes/s; Fig. 5a). For motion away from the RF, the
degree of suppression varied by 22% across the range of motion
strengths (~4.3 spikes/s; Fig. 5b).
This normalization procedure removes the main determinant
of the response magnitude for these neurons: whether an eye
movement is ultimately made to the RF or away from it. The effect
of motion strength is therefore relatively subtle. The regression
analysis was statistically significant (p = 0.0032 and p < 0.00001
for Fig. 5a and b, respectively) and was unaffected by the incor-
poration of eye movement descriptors such as saccadic latency,
amplitude, duration and velocity (see Methods). This last point
implies that subtle variations in the actual saccade produced in
each trial of the experiment does not explain the variation in neur-
al response found as a function of task difficulty.
We represented the evolution of predictive activity during
motion discrimination by calculating the predictive index in 250-
ms epochs beginning 500 ms before the onset of random dot
motion and ending just after the saccade (Fig. 5c). We computed
the index separately for each of the 53 neurons that predicted the
monkeys behavior during motion viewing (as above), using only
nature neuroscience volume 2 no 2 february 1999
articles
Fig. 3. Response of a frontal eye field neu-
ron during the motion-discrimination and
memory-saccade tasks. (a, b) Response on
memory-guided saccades to targets flashed
briefly inside or outside the response field.
This neuron had a prominent visual
response at the onset of the target and a
sustained response when the monkey
planned an eye movement up-leftward.
(c, d) Response on the motion-discrimina-
tion task. The poststimulus time histogram
includes all trials in which the monkey
chose the correct direction. The response
was stronger when the monkey judged the
motion to be toward the RF, but the effect
was not apparent until the delay (lower
arrows). Responses are aligned to the onset
of random dot motion and to the saccade
initiation. The transient response at the
time of motion onset was caused by the
appearance of saccade targets.
correct choices at each of six motion strengths. On average, the
response began to predict the monkeys decision 100200 ms after
onset of the random dot motion. Later in the trial, the response
predicted the monkeys choice with greater fidelity, reaching a max-
imum during the delay, just before the eye movement. Although
each point represents the predictive activity from just 250 ms of
spike discharge, the curves look like cumulative functions.
Predictive index delay period
Predictive index motion viewing period
Fig. 4. Predictive activity for 88 neurons during motion viewing and the
delay. The predictive index approximates the accuracy with which one
could guess monkeys decision based on the spike discharge measured
during motion viewing or the delay (Methods). Values larger than 0.5
imply greater accuracy in predicting the monkeys decision from the
neural response. The histograms summarize the distribution of these
indices; shading indicates a significant departure from 0.5 (p < 0.01 by
permutation test; Methods). Blue circles, FEF; red circles, PS (area 8Ar
and Walker area 46). The open symbols indicate the neurons shown in
Figs. 2 (red) and 3 (blue).
179
 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 5. Effect of motion strength on the magnitude and time

of the prefrontal response. (a, b) Effect of stimulus strength a b
on average response during motion viewing for 53 neurons

Normalized response
with statistically significant predictive indices. (a) For deci-
sions favoring motion toward the RF, the response was larger
to stronger random dot motion. The ordinate represents the
response strength relative to the mean for all decisions
toward the RF. Filled points represent the mean standard
error of the normalized response. (b) For decisions favoring
motion away from the RF, the response was more sup-
pressed to stronger motion stimuli. Conventions are the
same as in (a) except that the response is normalized to the
mean of each neurons response associated with choices out- Motion strength (% coherence)
side the RF. (c) The predictive power of the response was
computed in 250-ms epochs whose midpoint is plotted on
the time axis. Each point represents the probability of cor- c Saccade
rectly predicting the monkeys choice from 250 ms of spike
discharge. Curves represent the averaged probabilities from
53 neurons with predictive activity during motion viewing.
The neurons predicted the monkeys choice sooner and
1999 Nature America Inc. http://neurosci.nature.com

more reliably when the motion was stronger.

Probability (mean)

The emergence of predictive activity on the most

difficult motion condition (no coherent motion) indi-
cates that the neurons primarily encode variables that
pertain to the monkeys behavioral response. On the
other hand, when the motion was stronger (that is, eas-
ier), the response predicted the monkeys choice soon-
er and better than when the motion was weaker. This
effect was subtle for individual neurons but highly reli-
able across the population (p < 0.00001, likelihood ratio
test using modified probit analysis; Eq. 3 in Methods).
It reflects a stronger and more consistent rise in the
response when motion was toward the RF and a more
profound attenuation of the discharge when motion
was toward the target outside the neurons RF (Fig. 2).
This response pattern could represent the accumulation of
motion information from the extrastriate cortex toward a plateau.
Indeed, some neurons responded to the direction of random dot
motion during passive viewing, when the monkey was not
engaged in the discrimination task (Fig. 6). The weak direction
bias observed in this control experiment favored motion toward
the RF, suggesting that instructive visual signals reach the PFC
even without a task.
Errors
Could passive visual responses explain our finding of predictive
responses during motion viewing of the discrimination task? This
seems unlikely because the neurons were predictive when the
motion strength was negligible, as in the 0% coherent motion
condition (Figs. 2c and d and 5). The neural response is domi-
nated by the direction that the monkey judges, and hence the
direction of the planned saccade. This point is further supported
by examining error trials, in which the motion direction and the
saccade choice are opposed. The response was larger when the
monkey chose the target in the RF, whether or not this was based
on a proper interpretation of the random dot motion (Fig. 7).
Thus, the response was dominated by what the monkey planned
to do, rather than by what the monkey saw.
There was, however, a subtle difference between correct and
erroneous decisions. On average, neurons modulated their
response less strongly on the error trials. During motion viewing,
the response was weaker when the monkeys choice of the target in
Time (s)
the RF was erroneous as compared to correct (p = 0.001; t-test of
mean normalized response), and the response was less attenuated
on erroneous decisions favoring motion away from the RF
(p < 0.00001). Thus, the response cannot be explained solely by
the behavioral outcome. The observation is consistent with the
idea that the neural response reflects the accumulation of senso-
ry evidence toward a decision. As shown in the next section, this
is because the strength of neural signals favoring the wrong direc-
tion (for example, the responses of rightward motion sensors
when the direction is actually leftward) are unlikely to exceed by
much the neural signals that would favor the correct direction.
Theory
Our results are consistent with the idea that neurons in the dor-
solateral PFC compare sensory signals from the extrastriate cor-
tex that favor motion toward or away from the response field.
Accumulated spike counts from pools of neurons in areas MT
and MST can account for the monkeys sensitivity and trial-to-
trial choices on the random dot motion task16. We therefore
expect that neurons involved in the decision process must accu-
mulate spikes from groups of sensory neurons and make the
appropriate comparison, and that the neural computations
underlying a decision probably involve integration in time.
The sensory signals that inform the monkeys choices on a
near-threshold discrimination (Fig. 8) are presumed to be the
accumulated spike counts from pools of noisy and weakly corre-
lated direction sensors in the extrastriate visual cortex16. These

180 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

a b

Direction away from response field

Direction toward response field
1999 Nature America Inc. http://neurosci.nature.com

Fig. 6. Response to random dot motion during passive viewing. (a) Direction-tuning function for one neuron. This neurons RF was near the upper
vertical meridian at an eccentricity of 10 degrees. The random dots were shown in a five-degree aperture centered on the fixation point, outside the
RF of the neuron. The response histograms are displayed to indicate the direction of random dot motion. Responses are aligned to motion onset. The
polar plot shows the mean response calculated during viewing. The response was largest when motion was toward the RF. (b) Summary data from 10
neurons tested during passive fixation, comparing responses to passively viewed motion toward or away from the RF. Filled symbols denote neurons
with significant predictive activity during motion viewing on discrimination trials (done in a separate block). These neurons had a stronger direction
bias (points farther from diagonal; p < 0.001, ROC area comparison and permutation test; see Methods). Error bars represent standard error.

accumulations constitute the evidence for a rightward or left-
ward choice. Imagine a PFC neuron whose RF is situated so that
a rightward decision increases its response. According to our
scheme, this neuron compares the rightward cumulant to the left-
ward cumulant. The monkey chooses right when this difference is
positive and left when it is negative. The magnitude of the differ-
ence represents the strength of the evidence favoring a decision.
Figure 8a shows idealized distributions of pooled responses
from leftward and rightward sensors to a near-threshold right-
ward motion stimulus. The size of the responses is expressed in
units of standard deviation (). Notice that the distribution of
pooled rightward signals exceeds that of the leftward signals by
1 on average (that is, the index of discriminability, d, equals 1),
consistent with a motion strength that would support 76% cor-
rect choices. When the stimulus direc-
tion is leftward, the leftward signals are
larger on average (Fig. 8b).
This analysis permits us to assess
the strength of the evidence associated
with each of four types of decisions as
Response (mean normalized)
with weaker evidence. At the near-threshold motion strength in
this example, approximately 24% of the trials would result in
errors. For erroneous rightward choices (Fig. 8f), the difference
signals on error trials tend to be small (compare Fig. 8d and f).
The expected mean is 0.84 units of signal standard deviation.
This analysis could explain the subtle difference in the neural
response that we observed on correct versus error trials (Fig. 7).
This same analysis can be extended across the range of stimu-
lus strengths (Fig. 8g). At each motion strength, there are four
possible outcomes: two directions times two choices. At all motion
strengths, whenever the monkey chooses rightward correct or
not the evidence for rightward motion (Fig. 8g) exceeds the
evidence for leftward. However, the strength of the evidence
increases with stronger motion (larger values of d), leading to
Fig. 7. Comparison of the responses on
error and correct trials. The responses from

53 neurons with predictive activity during

seen from the point of view of a
choose right neuron in the PFC (Fig.
8cf). When motion is rightward and
the correct choice follows (Fig. 8d), the
strength of the evidence favoring right-
ward is positive and large on average.
The expected mean difference between
rightward and leftward sensory signals
is 1.6 units of signal standard devia-
tion. When motion is leftward and the
correct choice follows, the evidence
favoring a rightward choice is negative
and large (Fig. 8e; expected mean,
1.6 ). The error trials are associated
Time from motion onset
motion viewing were used to construct these
curves, using all non-zero motion strengths.
Responses were normalized to the mean of
each neurons response over the 600-ms
epoch indicated by the gray bar. The black
curves depict trials in which the monkey
judged the direction as toward the RF, culmi-
nating in a saccade to the target in the RF
(T1); the gray curves indicate decisions
resulting in a saccadic eye movement to the
target outside the RF (T2). Error trials are
indicated by the dashed curves. The promi-
nent response that begins before the onset of
motion was caused by the presentation of
the saccade targets.

nature neuroscience volume 2 no 2 february 1999 181

 1999 Nature America Inc. http://neurosci.nature.com

articles

Distribution of signals from rightward and leftward sensors

Motion is rightward (d = 1) Motion is leftward (d = 1)

a b
1999 Nature America Inc. http://neurosci.nature.com

Distribution of difference signals received by a choose right neuron

c Wrong left choice d Correct right choice e Correct left choice f Wrong right choice

SRightSLeft () SRightSLeft () SRightSLeft () SRightSLeft ()

Fig. 8. Theoretical basis for variation in the strength of evidence associated with direction judg-
g
Evidence for rightward ()
ments. The graphs trace the flow of information from populations of opponent direction sensors to
a neuron in the PFC that would increase its response in association with a rightward decision. The
input to the PFC neuron represents the evidence that motion is rightward, in the form of the differ-
ence between rightward and leftward sensory signals. Four scenarios are depicted: motion is either
leftward or rightward, and the choice is either correct or erroneous. (a) Activity of the leftward and
rightward motion sensors when the true direction is rightward. The surface shows the joint proba-
bility distribution of a pair of signals from rightward and leftward sensors. The two signals should be
interpreted as the population average responses from many neurons. The graph depicts the situation
at psychophysical threshold when the rightward signal is one standard deviation larger than the left-
ward signal, on average (projected bell-shaped curves). The points below the surface are 100 ran-
Stimulus strength (d)
dom samples from the overlying distribution. Each point is labeled according to the resulting
decision (blue, rightward decision; red, leftward decision). The majority of samples fall on the side of
the main diagonal that indicates that the rightward signal exceeds the leftward signal. In this plot, the rightward choices are correct. (b) Activity of the
motion sensors when the true direction is leftward. Same conventions as in (a). The leftward sensors produce the larger signal, on average. In this plot,
the leftward choices (red) are correct. (cf) Frequency histograms of the difference in sensory signals between rightward and leftward sensors. The dif-
ferences are tabulated separately for the two directions of motion and the two decisions. The histograms are shown under the simulated points from
which they were obtained in (a) and (b). The difference, SRight SLeft, is positive when the decision is rightward (blue) and negative when the decision is
leftward (red). Notice that there are more samples corresponding to correct choices (d and e), and the difference signals tend to be larger positive and
negative values. (g) Average difference signal, SRight SLeft, accompanying correct and incorrect choices for a range of motion strengths spanning the
psychometric function. The d = 1 case corresponds to a motion strength supporting 76% correct choices (~10% coherence), as illustrated in (af). The
theoretical means for (cf) are shown. When d = 0, there is no net motion direction (0% coherence), and there is no distinction between correct and
error trials. When d = 2, there are very few errors ( 25% coherence). The difference signals associated with correct choices attain greater positive
and negative values when the motion is stronger. The opposite trend is predicted for error trials.

larger responses when the decision is rightward and greater sup-
pression when the decision is leftward. The increasing separation
of the solid curves in Fig. 8g helps explain the increase in predic-
tive index as a function of motion strength (see Fig. 5). The analy-
sis also predicts that the responses associated with errors should
diminish at higher motion strengths. We cannot evaluate this pre-
diction because errors occur rarely for strong motion.
Finally, this analysis helps to explain the relatively subtle
variation in response that we observed in the PFC as a function
of motion strength. Across the entire psychometric function,
the evidence for rightward changes by only one unit of signal
standard deviation for correct choices that lead to the same
behavioral response (Fig. 8g). The unit, , can be related to the
responses of MT neurons. At a d value of 1, the response of
pooled rightward-preferring MT neurons exceeds the response
of pooled leftward-preferring MT neurons by one standard
deviation of the pooled values. Recordings from area MT show
that at a motion strength of 10% (corresponding to d~1),
rightward- and leftward-preferring neurons differ in their
responses by ~5 spikes/s1,17. According to the analysis (Fig. 8g),

182 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
the monkeys judgments at the weakest motion strength (0%
coherence) are based on evidence of 1.13 , or ~5 spikes/s
from the average MT neuron. We do not know how to convert
this value to a spike rate in the PFC; in the example we are pur-
suing, all rightward choices are associated with a high spike rate.
However, as the motion strength increases to span the mon-
keys psychometric function, the strength of the evidence
increases (or decreases) by only another unit of or ~5 spikes/s.
This value is comparable to the change in PFC response that
we observed across the range of motion strengths used in our
experiments (see Fig. 5a, b and legend).
DISCUSSION
The dorsolateral prefrontal cortex is important in linking sen-
sation to action, especially when the linkage involves a delay6,7.
Our study provides a glimpse of this linkage in the activity of
individual PFC neurons. We used a task in which delayed sac-
cades were instructed by a direction judgment. By requiring
1999 Nature America Inc. http://neurosci.nature.com
the animal to make difficult judgments near psychophysical
threshold, we were able to characterize neural activity during
a period in which sensory processing gradually gave way to a
categorical choice, what we have termed a decision process.
Our results demonstrate a gradual unfolding of neural activ-
ity, which we interpret as a correlate of the monkeys decision.
To find a neural correlate of the monkeys decision about
the direction of motion, we searched for neurons that signaled
the monkeys commitment to a particular behavioral option.
We therefore studied neurons in the FEF and principalis region
that showed sustained activity during an oculomotor delayed-
response task. These neurons could be said to lie toward the
motor side of a sensationaction continuum 14, linking the
visual cortex with the motor system 18.
The sustained response of PFC neurons has been inter-
preted as a neural correlate of short-term memory for spatial
location19,20, but we do not know if such neural activity rep-
resents the instruction (for example, location of a spatial cue)
or the behavioral plan to move the eyes to that location. Our
results suggest that the response may conflate these represen-
tations. Most neurons modulated their response in associa-
tion with all the key ingredients of the discrimination task:
the appearance of a target within the response field, the direc-
tion and strength of random dot motion and the direction of
a planned saccade. Our analysis showed that early in the task,
during formation of the decision, the modulation in neural
response varies parametrically with the strength of visual
motion (Fig. 5). By the end of this process, many neurons in
the PFC reflect the monkeys choice. These observations sug-
gest that the PFC represents not simply the final outcome of
sensory processing but the conversion of an analog motion
representation to a binary decision variable.
This conclusion must be viewed cautiously because we do
not know the moment that the monkey decides the direction
of motion, or indeed if there is such a discrete moment. It is
possible that on different trials, the monkey reaches its deci-
sions at different times and the neural responses change short-
ly thereafter to reveal the intended action. The average
response from many trials might give the appearance of a
gradual evolution of the monkeys plan (Fig. 5c). If the deci-
sion were to occur sooner, on average, for easier discrimina-
tions, then this could explain the parametric variation in
response magnitude that we observed with different motion
strengths (Fig. 5a and b). We have looked for discrete changes
in the firing patterns of our neurons using a previously
nature neuroscience volume 2 no 2 february 1999
articles
described algorithm21, but we have failed to find evidence for
such abrupt transitions. Simultaneous recordings from two
or more neurons could, in principle, facilitate detection of
such abrupt changes in firing pattern.
Our observations support an emerging view that the dis-
tinction between sensory and motor systems may be blurred
within the association areas of the cerebral cortex18,22,23. In
many association areas, neural responses are predicted by
stimulus qualities as well as motor preparation. It is tempting
to regard the former as instructing the latter. For example,
FEF neurons represent features of a visual stimuli that instruct
an eye movement in a visual search task24. Neurons in area 46
respond at the moment an instruction is given to guide a
future eye movement11. Even neural responses in the primary
and supplementary motor cortex reflect not only the planned
behavior but the sensory cue that instructs that action12,2531.
Any brain region containing signals related to both senso-
ry processing and motor preparation may be involved in the
conversion of the former to the latter. For most tasks, howev-
er, there is insufficient time to study the process of conver-
sion: the moment the instruction is received, the animal can
prepare an action. In the threshold discrimination task that
we used, the monkey cannot process the instruction instant-
ly. The monkey, like human subjects, benefits from the tem-
poral accumulation of information 1,32 (J.D. Roitman &
M.N.S., Soc. Neurosci. Abstr. 24, 262, 1998). Our results suggest
that prefrontal neurons may do more than hold information
in short-term memory. They seem to be involved in the accu-
mulation (that is, integration) and comparison of sensory
streams toward a categorical outcome or behavioral plan.
This does not imply that the neurons we recorded are
directly responsible for deciding direction. They may reflect
the computation made at another site in the brain. In fact, the
neurons described here respond similarly to neurons in the
lateral intraparietal area (LIP) and the superior colliculus33,34
during the same motion discrimination task. This is not sur-
prising because area LIP, the FEF and the posterior principalis
region are strongly connected with each other3538. Both LIP
and the dorsolateral PFC seem to be activated during similar
delayed eye movement tasks39, and both areas project to the
superior colliculus36,4042. The relative importance of these
brain regions will need to be addressed with reversible inacti-
vation or simultaneous recording. Meanwhile, the activity in
the prefrontal and inferior parietal cortex reveals something
about the computations that may underlie the linkage between
sensory data, interpretation and behavioral planning. We pro-
pose that such neurons compute the time integral of sensory
evidence toward a plateau state.
Neurons whose dominant mode of response signals the plan
to enact a behavior must be influenced by sensory signals. The
existence of such a link comes as no surprise, but the mixture
of signals on single neurons, reflecting motor planning and
sensory strength, constrains a view of the brains logical archi-
tecture. It seems inconsistent with a central executive function
that interprets the sensory data, declares an interpretation and
recruits circuitry to enact a response. Instead, it supports a
view of brain organization that would recruit premotor cir-
cuitry in the interest of several potential actions while querying
sensory streams for evidence to select the appropriate one.
METHODS
Electrophysiology. We recorded from 88 neurons in the frontal eye
field (FEF, areas 8Ac and 45a) and posterior principalis region (areas
8Ar and Walker area 46) of two rhesus monkeys trained on a random
183
 1999 Nature America Inc. http://neurosci.nature.com
articles
dot motion discrimination task 1. Monkeys were implanted with an
eye coil, head-holding device and recording cylinder suitable for mag-
netic resonance imaging (MRI; Crist Instrument Co., Damascus, Mary-
land). The recording cylinder was placed over the arcuate sulcus and
the posterior third of the principal sulcus (PS; Fig. 1e). Sterile guide
tubes were placed through a plastic grid (Crist Instruments) to intro-
duce tungsten/glass microelectrodes to the surface of the dura mater.
The grid array was visualized in situ by MRI and registered with the
anatomy (Fig. 1f). We used the MRI to identify the FEF in the anteri-
or bank of the arcuate sulcus and to distinguish the area principalis
from more caudal portions of the prearcuate gyrus. Action potentials
were identified using a dual-voltage, time-window discriminator (Bak
Electronics, Germantown, Maryland) and stored on computer with 1-
ms precision43. All training, surgery and experimental procedures com-
plied with the National Institutes of Health Guide for the Care and
Use of Laboratory Animals and were approved by the University of
Washington Animal Care Committee.
Electrical stimulation. To aid in identifying recording sites within the
1999 Nature America Inc. http://neurosci.nature.com
frontal eye field, we attempted to elicit saccades using the electrical
microstimulation protocol described 44. We classified sites as low
threshold if we could elicit a fixed-vector saccade with a stimulating
current less than 50 A. We classified cells as in or out of the FEF
based on their location in the anterior bank of the arcuate sulcus and
their proximity to low-threshold stimulation sites. This procedure
unequivocally classified most neurons.
Behavioral tasks. Neurons were selected using a delayed-saccade task
that relies on working memory (Fig. 1d)8,45,46. We studied neurons
that responded during the delay (memory) period preceding saccades
to a restricted region of the visual field. We refer to this region as the
neural response field (RF) to remain agnostic as to function. Many
neurons also responded transiently at the onset of movement or to the
onset of saccade targets, but this was not a criterion in their selection.
Similar selection criteria were used in a related study of area LIP33.
In the motion-discrimination task (Fig. 1a and b), two response
targets appeared. One target was in the RF of the neuron; the random
dot kinematogram and the second response target appeared outside
the RF. After a brief pause (200300 ms), the random dot motion was
displayed for 1 s. The direction of motion was toward one or the other
target. Both the direction of motion and the fraction of coherently
moving dots were randomized. After a delay, the monkey indicated its
direction judgment by making an eye movement to one of the two
response targets. The monkey was trained to associate rightward
motion with the target to the right of the dots aperture, upward motion
with a target above the aperture, and so forth. The monkey received a
liquid reward for correct responses (and on half of the trials in which
the 0% coherent motion was shown). In all experiments, the monkeys
showed a smooth improvement in performance as a function of
motion strength (Fig. 1c). The average threshold motion strength sup-
porting 81% correct choices was 12.9% (range 1.5 to 34.1%; standard
deviation, 5.2%), which is typical of highly trained rhesus monkeys
in similar studies1,16.
We occasionally held neurons long enough to do a passive-fixation
control experiment (Fig. 6). No saccade targets appeared on this block
of trials, and the monkey was rewarded simply for maintaining fixa-
tion. Random dot motion (51.2% coherence) appeared in the same
location as in the discrimination block and moved toward the RF or
in one to seven directions away from the RF.
Data analysis. Responses were analyzed off-line using custom soft-
ware. To combine data from several neurons, we first normalized
each neurons response using the mean spike rate in an epoch from
200800 ms after the onset of random dot motion. This epoch cor-
responds to motion viewing, when the monkey decides the direc-
tion of motion and after the prominent transient response to the
onset of saccade targets.
The effect of motion strength on the neural response was estimat-
ed using a linear regression model, Y = 0 + 1C + where Y is the test, df = 4)49. We refer to this method as a modified probit analysis.
184
normalized spike rate and C is the rank motion strength (0 to 5). For
each trial, we also extracted five descriptors of the saccade: its laten-
cy, amplitude, accuracy, maximal speed and duration. We included
these factors along with motion strength in a multivariate regression
analysis, fitting the model,
Y = 0 + 1C + 2LAT + 3AMP + 4ACC + 5VMAX + 6DUR + (1)
The fit to equation 1 allowed us to test whether motion strength
affects the neural response in a manner that cannot be attributed to
variation in saccadic eye movements. This is a test of the null hypoth-
esis, 1 = 0, which is done by comparing the extra sum of squares
with and without this regressor and computing an F ratio47.
Analysis of predictive activity. For several analyses, we computed an
index that describes the association between neural response and the
monkeys decision. This index, the probability that the neural
response associated with one behavioral choice exceeds the neural
response associated with the other behavioral choice, approximates
the ability of the experimenter to predict the monkeys behavior from
the neural response. Denoting the responses associated with the two
choices by {r1} and {r2}, this is the joint probability of observing r1 = k
and r2 < k, over all possible values of k:
Pr(r = k)Pr(r <k)dk
Predictive index = 1 2
-
(2)
k
= Pr(r = k) [ Pr(r =)d]dk
1 2
- -
Equation 2 can be estimated by computing the area under a receiv-
er-operator-characteristic (ROC) curve obtained from the two
response distributions1,48.
The distribution of the predictive index under the null hypothesis
is typically not normal. We used a permutation test to estimate the
probability that the measured index would be observed under the
null hypothesis of a random association between neural and behav-
ioral response (that is, the true index is 0.5). Each neural response
and each behavioral response were randomly associated to construct
the two distributions, {r1} and {r2}, which therefore contain the same
number of observations as the original {r1} and {r2}. The predictive
index was computed using Equation 2, and the distribution of its
absolute difference from 0.5 was estimated using 1000 permutations
of the data. The probability of obtaining the measured index under
the null hupothesis is the fraction of this distribution exceeding the
absolute value of the measured indexs difference from 0.5.
The same procedure was used to measure of the time course of pre-
dictive activity (Fig. 5c) by computing the predictive index from spike
counts obtained from 250-ms epochs at intervals designated with
respect to the onset random dot motion or the monkeys saccadic eye
movement. The resulting sigmoid functions are well fit by the scaled
integral of a Gaussian function of time:
p(t) = a0 + a1C + (b0 + b1C)(t),
where (3)
t
[(0 + 1C]2
1
(t) = e 2[0 + 1C]2 d
2 (0 + 1C) -
Notice that (t) is a cumulative Gaussian function of time, parame-
terized by its mean and standard deviation: the mean determines the
position of the sigmoid from left to right, and the standard deviation
determines its slope. We modeled the baseline values, a; the scaling
terms, b; and the sigmoid parameters, and , as linear functions of the
motion strength, C. We fit equation 3 to the measured predictive indices
using a simplex algorithm to find the maximum likelihood solution
(assuming binomially distributed error). The null hypothesis that
motion strength does not affect the shape of these sigmoid functions
was tested by fitting the model with a1 = b1 = 1 = 1 = 0 and com-
paring likelihoods under the full and reduced models (likelihood ratio
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com
ACKNOWLEDGEMENTS
We thank Melissa Mihali for animal training and technical support. We also
thank Joshua Gold, Greg Horwitz, Mark Mazurek, Bill Newsome, Jeff Schall
and Kirk Thompson for helpful suggestions on the manuscript. This research
was supported by RR00166, EY11378 and the McKnight Foundation.
RECEIVED 24 AUGUST: ACCEPTED 23 DECEMBER 1998
1. Britten, K. H., Shadlen, M. N., Newsome, W. T. & Movshon, J. A. The
analysis of visual motion: a comparison of neuronal and psychophysical
performance. J. Neurosci. 12, 47454765 (1992).
2. Newsome, W. & Pare, E. A selective impairment of motion perception
following lesions of the middle temporal visual area (MT). J. Neurosci. 8,
22012211 (1988).
3. Salzman, C. D., Murasugi, C. M., Britten, K. H. & Newsome, W. T.
Microstimulation in visual area MT: Effects on direction discrimination
performance. J. Neurosci. 12, 23312355 (1992).
4. Celebrini, S. & Newsome, W. T. Neuronal and psychophysical sensitivity
to motion signals in extrastriate area MST of the macaque monkey. J.
Neurosci. 14, 41094124 (1994).
5. Seidemann, E., Zohary, E. & Newsome, W. T. Temporal gating of neural
1999 Nature America Inc. http://neurosci.nature.com
signals during performance of a visual discrimination task. Nature 394,
7275 (1998).
6. Jacobsen, C. F. Functions of frontal association area in primates. Arch.
Neurol. Psychiatry 33, 558569 (1935).
7. Fuster, J. The Prefrontal Cortex (Raven, New York, 1989).
8. Funahashi, S., Bruce, C. & Goldman-Rakic, P. Mnemonic coding of visual
space in the monkeys dorsolateral prefrontal cortex. J. Neurophysiol. 61,
331349 (1989).
9. Wilson, F. A., Scalaidhe, S. P. & Goldman-Rakic, P. S. Dissociation of
object and spatial processing domains in primate prefrontal cortex.
Science 260, 19551958 (1993).
10. Miller, E. K., Erickson, C. A. & Desimone, R. Neural mechanisms of visual
working memory in prefrontal cortex of the macaque. J. Neurosci. 16,
51545167 (1996).
11. Hasegawa, R., Sawaguchi, T. & Kubota, K. Monkey prefrontal neuronal
activity coding the forthcoming saccade in an oculomotor delayed
matching-to-sample task. J. Neurophysiol. 79, 322334 (1998).
12. Zhang, J., Riehle, A., Requin, J. & Kornblum, S. Dynamics of single
neuron activity in monkey primary motor cortex related to sensorimotor
transformation. J. Neurosci. 17, 22272246 (1997).
13. Riehle, A., Kornblum, S. & Requin, J. Neuronal coding of stimulus-
response association rules in the motor cortex. Neuroreport 5, 24622464
(1994).
14. Leon, M. I. & Shadlen, M. N. Exploring the neurophysiology of decisions.
Neuron 21, 669672 (1998).
15. Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. Numerical
Recipes in C. The Art of Scientific Computing (Cambridge Univ. Press,
Cambridge, 1992).
16. Shadlen, M. N., Britten, K. H., Newsome, W. T. & Movshon, J. A. A
computational analysis of the relationship between neuronal and
behavioral responses to visual motion. J. Neurosci. 16, 14861510 (1996).
17. Britten, K. H., Shadlen, M. N., Newsome, W. T. & Movshon, J. A.
Responses of neurons in macaque MT to stochastic motion signals. Vis.
Neurosci. 10, 11571169 (1993).
18. Goodale, M. A. Visuomotor control: Where does vision end and action
begin? Curr. Biol. 8, R489491 (1998).
19. Goldman-Rakic, P. in Handbook of Physiology, Sec I, The Nervous System
(ed. Plum, F.) 373417 (Am. Physiol. Soc., Bethesda, MD, 1987).
20. Funahashi, S., Bruce, C. & Goldman-Rakic, P. Dorsolateral prefrontal
lesions and oculomotor delayed-response performance: evidence for
mnemonic scotomas. J. Neurosci. 13, 14791497 (1993).
21. Commenges, D., Seal, J. & Pinatel, F. Inference about a change point in
experimental neurophysiology. Math. Biosci. 80, 81108 (1986).
22. Seal, J. & Commenges, D. A quantitative analysis of stimulus- and
movement-related responses in the posterior parietal cortex of the
monkey. Exp. Brain Res. 58, 144153 (1985).
23. Seal, J. Sensory and motor functions of the superior parietal cortex of the
monkey as revealed by single-neuron recordings. Brain Behav. Evol. 33,
113117 (1989).
24. Thompson, K. G., Bichot, N. P. & Schall, J. D. Dissociation of visual
nature neuroscience volume 2 no 2 february 1999
articles
discrimination from saccade programming in macaque frontal eye field.
J. Neurophysiol. 77, 10461050 (1997).
25. di Pellegrino, G. & Wise, S. Visuospatial versus visuomotor activity in the
premotor and prefrontal cortex of a primate. J. Neurosci. 13, 12271243
(1993).
26. Salinas, E. & Romo, R. Conversion of sensory signals into motor
commands in primary motor cortex. J. Neurosci. 18, 499512 (1998).
27. Romo, R., Ruiz, S., Crespo, P., Zainos, A. & Merchant, H. Representation
of tactile signals in primate supplementary motor area. J. Neurophysiol.
70, 26902694 (1993).
28. Shen, L. & Alexander, G. E. Neural correlates of a spatial sensory-to-
motor transformation in primary motor cortex. J. Neurophysiol. 77,
11711194 (1997).
29. Shen, L. & Alexander, G. E. Preferential representation of instructed
target location versus limb trajectory in dorsal premotor area. J.
Neurophysiol. 77, 11951212 (1997).
30. Alexander, G. E. & Crutcher, M. D. Neural representations of the target
(goal) of visually guided arm movements in three motor areas of the
monkey. J. Neurophysiol. 64, 164178 (1990).
31. Quintana, J. & Fuster, J. Mnemonic and predictive functions of cortical
neurons in a memory task. Neuroreport 3, 721724 (1992).
32. Watson, A. B. Probability summation over time. Vision Res. 19, 515522
(1979).
33. Shadlen, M. N. & Newsome, W. T. Motion perception: seeing and
deciding. Proc. Natl. Acad. Sci. USA 93, 628633 (1996).
34. Newsome, W. T. Deciding about motion: linking perception to action. J.
Comp. Physiol. A 181, 512 (1997).
35. Stanton, G. B., Bruce, C. J. & Goldberg, M. E. Topography of projections
to posterior cortical areas from the macaque frontal eye fields. J. Comp.
Neurol. 353, 291305 (1995).
36. Lynch, J. C., Graybiel, A. M. & Lobeck, L. J. The differential projection of
two cytoarchitectonic subregions of the inferior parietal lobule of
macaque upon the deep layers of the superior colliculus. J. Comp. Neurol.
235, 241254 (1985).
37. Huerta, M. F., Krubitzer, L. A. & Kaas, J. H. Frontal eye field as defined by
intracortical microstimulation in squirrel monkeys, owl monkeys, and
macaque monkeys. II. Cortical connections. J. Comp. Neurol. 265,
332361 (1987).
38. Schall, J. D., Morel, A., King, D. J. & Bullier, J. Topography of visual cortex
connections with frontal eye field in macaque: convergence and
segregation of processing streams. J. Neurosci. 15, 44644487 (1995).
39. Friedman, H. & Goldman-Rakic, P. Coactivation of prefrontal cortex and
inferior parietal cortex in working memory tasks revealed by 2DG
functional mapping in the rhesus monkey. J. Neurosci. 14, 27752788
(1994).
40. Huerta, M. F., Krubitzer, L. A. & Kaas, J. H. Frontal eye field as defined by
intracortical microstimulation in squirrel monkeys, owl monkeys, and
macaque monkeys: I. Subcortical connections. J. Comp. Neurol. 253,
415439 (1986).
41. Segraves, M. A. & Goldberg, M. E. Functional properties of corticotectal
neurons in the monkeys frontal eye field. J. Neurophysiol. 58, 13871419
(1987).
42. Stanton, G. B., Goldberg, M. E. & Bruce, C. J. Frontal eye field efferents in
the macaque monkey: I. Subcortical pathways and topography of striatal
and thalamic terminal fields. J. Comp. Neurol. 271, 473492 (1988).
43. Hays, A. V., Richmond, B. J. & Optican, L. M. A UNIX-based multiple
process system for real-time data acquisition and control. WESCON Conf.
Proc. 2, 110 (1982).
44. Bruce, C. J., Goldberg, M. E., Bushnell, M. C. & Stanton, G. B. Primate
frontal eye fields. II. Physiological and anatomical correlates of
electrically evoked eye movements. J. Neurophysiol. 54, 714734 (1985).
45. Gnadt, J. W. & Andersen, R. A. Memory related motor planning activity
in posterior parietal cortex of monkey. Exp. Brain Res. 70, 216220
(1988).
46. Hikosaka, O., Sakamoto, M. & Usui, S. Functional properties of monkey
caudate neurons. III. Activities related to expectation of target and
reward. J. Neurophysiol. 61, 814832 (1989).
47. Draper, N. & Smith, H. Applied Regresion Analysis, Second Edition (Wiley,
New York, 1966).
48. Green, D. M. & Swets, J. A. Signal Detection Theory and Psychophysics
(Wiley, New York, 1966).
49. Mood, A. M., Grabill, F. A. & Boes, D. C. Introduction to the Theory of
Statistics (McGraw-Hill, New York, 1963).
185
 1999 Nature America Inc. http://neurosci.nature.com

articles

Weighted combination of size and

disparity: a computational model
for timing a ball catch
Simon K. Rushton1 and John P. Wann2

1 Department of Psychology, University of Edinburgh, 7 George Square, Edinburgh EH8 9JZ, UK

2 Department of Psychology, University of Reading, 3 Earley Gate, Reading RG6 6AL, UK
Correspondence should be addressed to J.P.W. (J.P.Wann@reading.ac.uk)

How do we time hand closure to catch a ball? Binocular disparity and optical looming provide two
1999 Nature America Inc. http://neurosci.nature.com

sources of information about an objects motion in depth, but the relative effectiveness of the two
cues depends on ball size. Based on results from a virtual reality ball-catching task, we derive a
simple model that uses both cues. The model is sensitive to the relative effectiveness of size and dis-
parity and implicitly switches its response to the cue that specifies the earliest arrival and away from
a cue that is lost or below threshold. We demonstrate the models robustness by predicting the
response of participants to some very unusual ball trajectories in a virtual reality task.

The time to contact (TTC) of an approaching object can be approx-
imated by its instantaneous distance and velocity. The simplicity of
this mathematical solution, however, does not reflect the scale of the
problem involved in a neural or robotic implementation of TTC
estimation. The use of distance and velocity requires reconstruction
of a three-dimensional representation of the environment, which is
nontrivial and demanding. Survival in the forest, on the sports field
or in the robotics lab requires that TTC be estimated very quickly
and without undue call on cognitive resources. To operate as fast as
possible, therefore, a TTC mechanism may rely upon a simple sparse,
or even crude, abstraction of the visual environment1. In some set-
tings, specific cues may also be lost or fall below the threshold for
detection2. People seem to cope well with a range of environmental
conditions, which argues for a robust mechanism for TTC estimation
that degrades gracefully in cases of cue conflict or cue loss.
Optical size and binocular disparity are both functions of the
distance of a ball from the observer, so either could provide esti-
mates of TTC36. Many studies of interceptive timing have ignored
the role of disparity, addressing only the role of changing retinal
size (looming). Evidence for the exclusive use of retinal looming in
natural TTC judgment, however, is equivocal7, and it has been
proposed that changing disparity may be more salient than chang-
ing size for a fast-moving cricket ball2 . As a ball moves in depth,
the rate of looming (d/dt) relative to changing disparity (d/dt or
d/dt; Fig. 1) is a function of object size (R), relative to inter-
pupillary distance (I), independent of viewing distance2 :
R/I = (d/dt)/(d/dt) (1)
The salience of either input therefore depends on the size of the
object and on the individuals sensitivity to looming or disparity.
Hence a robust TTC system should take account of both chang-
ing size and changing disparity inputs and ideally be sensitive to
their relative rates of change.
An accurate geometric specification of TTC from disparity
requires the detection of the binocular angular subtense (; Fig 1).
There are precedents for considering as the optical parameter
in the specification of binocular TTC5,8,9. The finding that a sen-
sation of motion in depth requires a background reference at a
fixed optic disparity10 suggests that observers can only perceive
changing relative disparity (), but there is conflicting evidence
that a change in binocular subtense () can influence TTC judg-
ments5. Psychophysical motion-in-depth tasks10 require a declar-
ative judgment, rather than a motor response to an impending
collision5, and perceptual judgments may be processed through
a different subsystem from perception for action11. We therefore
consider the issue regarding the use of in TTC estimation to be
unresolved. We use when outlining the specification of TTC
and discuss as a potential route to estimating TTC.
Under natural circumstances, the distinction between and
may be irrelevant, as most environments provide many poten-
tial reference features, and with a distant reference . Yet given
a reference object at 10 m, then for a ball travelling at 5 m/s, the
error in using /(d/dt) as a TTC estimate is 45 ms (Appendix).
An error of this size may thwart one-handed catching12, which
demonstrates that cue conflict may be commonplace if the observ-
er uses relative disparity along with changing size. A related prob-
lem in the use of binocular disparity is that a reference point for
the detection of relative disparity may be lost during flight (for
example, when the ball is on a high parabolic arc). The issues of
cue conflict and cue loss have not been tackled in previous
accounts of TTC judgments, and so we examined how human
observers perform when there are differences in the TTC infor-
mation provided by changing size and disparity and how they
cope when cues drop out. As noted above, there may be problems
in extrapolating from perceptual judgments in psychophysical
tasks, with limited viewing periods and ocular fixation require-
ments, to active control in naturalistic settings13. Thus we stayed
within the bounds of a naturalistic task, but retained psy-
chophysical precision, by presenting observers with the task of
catching virtual tennis balls within a computer-generated envi-
ronment. By dissociating looming and disparity cues, we demon-
strate that the weighting that participants attach to either cue

186 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

Fig. 1. Retinotopic and spatiotopic information for TTC judgment. A

ball of diameter R travels at a constant velocity V toward an observer
who has an interpupilary distance I. Time to contact with the eye is
specified spatiotopically by TTC = D/V. For D >> R and D >> I, optic
size R/D and I/D, where is the binocular subtense relative to a
parallel optic axis (dotted). Their derivatives can be approximated by
(d/dt) RV/D2 and (d/dt) IV/D2. Given a reference object at a dis-
tance D + D, relative retinal disparity can be approximated by
ID/D(D + D), the derivative of which is identical to that of :
(d/dt) IV/D2. This yields four possibilities for estimating TTC without
recourse to an estimate of D or V: /(d/dt), /(d/dt), /(d/dt) or
TTC /(d/dt) if DR is large (see Appendix).
1999 Nature America Inc. http://neurosci.nature.com

depends on which specifies the shortest TTC. We present a model
that accounts for this behavior and explains why timing errors in
catching may be minor even in cases of cue conflict or cue loss.
RESULTS
In Experiment 1, we asked observers to catch virtual tennis balls
with no constraints on their eye movements. Observers were able
to catch virtual balls accurately. Their grasp response in the con-
trol condition (Condition A) was initiated 146 ms before the ball
arrived (standard error, 15 ms) and completed 22 ms after the
ball arrived (standard error, 14 ms). This accuracy is equivalent to
that required in natural catching12. We then introduced subtle
differences in the TTC information provided by changing size or
changing disparity ( 100 ms) to assess the weight that partici-
pants attached to each cue. When looming indicated a different
TTC from disparity, observers modified their grasp time, but the
degree of modification varied across conditions. The earlier, or
more immediate, estimate of arrival had the greater influence
the two modular estimates (TTCsize and TTCdisp):
TTC = 1 TTCsize + 2TTCdisp (2)
This weighting system would be robust to cue loss and could
account for the results of Experiment 1 if 1 and 2 were dynam-
ically revalued using cross ratios of the modular inputs:
1 = TTCdisp / (TTCsize + TTCdisp)
2 = TTCsize / (TTCsize + TTCdisp) (3)
This could be considered a variant of weak fusion cue com-
bination14.
The third model does not calculate TTC separately for each
input, but combines instantaneous size and disparity and uses
the summed parameter to calculate TTC. There is psychophysi-
B: TTC looming is 100 ms early
Response shift relative to control trials (ms)

C: TTC looming is 100 ms late

on grasp timing (Fig. 2). Hence looming was the dominant cue
when it specified a TTC 100 ms earlier than disparity (Condi-
tion B), whereas disparity had the most influence when looming
specified a later TTC (Condition C). The results demonstrate
that both optical size and disparity are used in catching, but the
respective influence or weighting of size and disparity is not fixed.

TTC models
To account for these findings, we needed to derive a model that
uses both size and disparity and biases its response toward the
more immediate cue. We identified three potential models that
combine optic size, disparity and their respective rates of
change to estimate TTC (see supplementary figure at http://neu-
rosci.nature.com/web_specials/supp_info/nn0299_186/). The first
model is a fixed weighting modular summation, in which two
separate estimates of TTC are derived, one from size and its rate of Initiation Completion
change, and the other from disparity and its rate of change. The
two estimates are then combined according to some weighting
factor , where = 1 would reduce the model to one that relied Fig. 2. Changes in the timing of hand grasp when catching virtual balls.
exclusively on size3 and = 0 would use only disparity. Although Looming was scaled to provide a TTC estimate 100 ms earlier (Condition
this model includes both size and disparity, its fixed weighting B) or 100 ms later (Condition C) than the binocular TTC information.
The TTC at which the grasp response was initiated and completed in
precludes it from biasing its TTC estimate toward the most imme-
Conditions B and C were compared with nested control trials (Condition
diate cue, and so it cannot account for the flexible response A), in which both TTC cues were commensurate. The unsigned change in
demonstrated in our data. It also predicts a large error in the case grasp time was significantly greater in Condition B (gain for looming, 0.78)
of sudden cue loss. than in Condition C (gain for looming, 0.41) for both initiation and com-
The second model, a variant of the first, introduces changes in pletion (t5 = 3.51, p < 0.01; t5 = 3.34, p < 0.01, one-tailed test, respec-
the weighting on the fly based on the relative immediacy of tively). Error bars are 95% confidence limits for each mean.

nature neuroscience volume 2 no 2 february 1999 187

 1999 Nature America Inc. http://neurosci.nature.com
articles
Fig. 3. Predictions from the dipole model and a modular
model for the plateau conditions of Experiment 2. Solid
lines indicate the potential inputs from changing size and
changing disparity, in which input_1 is veridical and
input_2 indicates a constant TTC of 750 ms. The modu-
Time to contact estimate (ms)
lar estimate is based on a 50:50 weighting of both inputs,
whereas the dipole ( estimate; filled circles) shows an
increasing bias toward the most immediate input. If the
desired response time was 250 ms TTC, then the dipole
would predict a response delay of 70 ms, whereas the
modular model would predict a response well after the
ball had arrived. The error in using relative disparity (;
open circles), with a reference point at 4 m, is minor if
the planned interception is at a short TTC.
1999 Nature America Inc. http://neurosci.nature.com
cal evidence2 for the early combination of size and disparity
motion signals (d/dt + d/dt), and neurophysiological evidence
for the combination of optic size and disparity ( + ) at an early
stage of visual processing15. A TTC estimate can be based on a
ratio of these combined inputs:
TTCdipole = ( + ) / (d/dt + d/dt)
We adopt the label dipole from theory on texture perception16 . A
single point viewed by two eyes specifies a binocular dipole, and two
points (such as two opposite edges of an object) viewed by one eye
specify a monocular dipole. Our model sums dipoles and does not
distinguish their origin. An alternative means of estimating the ratio
in Equation 4 is to take the change in the summed dipole length
within a logarithmic coordinate system: TTCdipole = d[ln ( + )]/dt.
Temporal error with looming TTC plateau is 750 ms
Temporal error with looming binocular TTC plateau is 750 ms
Dipole model prediction
Modular model prediction (50:50 weighting)
Error in completion of grasp (ms)
Participant
Fig. 4. Errors in grasp timing for the ball-catching task in Experiment 2. In the
plateau trials, optic size was scaled to keep TTC calculated from size constant at
750 ms, while TTC calculated from disparity continued to decrease in line with
the control, or else TTC calculated from disparity was kept constant at 750 ms,
while TTC calculated from size continued to decrease in line with the control tri-
als (see Fig. 3). Predictions from the dipole model and the fixed weighting model
were produced by using the response time for the control trials for each partici-
pant to estimate when that TTC would be achieved for each model.
188
Actual time to contact (ms)
The retinotopic quantities ( + ) rise nonlinearly as the object
approaches (d/dt, d/dt 1/D2). As a result, the cue that specifies
the earlier arrival has a disproportionate influence. For a nonde-
formable object, with a distant reference for binocular motion, Equa-
tion 4 is equivalent to the following:
(4) TTCdipole = (TTCsize + TTCdisp [I/R])/ (1 + [I/R]) (5)
Hence when the diameter of the object equals the interpupilary
distance (I = R), the dipole model bases its estimate on an equal
weighting of both inputs, but for a football, the implicit bias would
be toward an estimate based on TTCsize and for a table-tennis ball
toward TTCdisp. If one input is lost ( = 0 or = 0), then the esti-
mate relies upon the remaining single input (Eq. 4). If one rate of
change is below threshold, the model predicts a relatively small
response error (Appendix; Eq. 6). If both cues are above
threshold but indicate different TTC, then the dipole model
biases its estimate toward the more immediate cue, equiva-
lent to the explicit weighting of Equation 3. For the cue
manipulation used in Experiment 1, the dipole model pre-
dicts temporal shifts of 70 ms when size specifies an ear-
lier arrival and +34 ms when size specifies a later arrival,
which is in good agreement with the results in Fig. 2.
Experiment 2
Because the dipole model was designed to account for the
findings of Experiment 1, we wished to test its robustness
under different input conditions. We have outlined that the
model can cope with complete loss of one cue or a static state
for one input. The stalling of one TTC estimate, however,
would predict a specific response delay. In Experiment 2, we
computationally scaled either the optical size or the dispar-
ity so that they produced a TTC plateau. In one condition,
changing size presented a normal decreasing TTC estimate,
but disparity indicated a constant TTC, whereas the recip-
rocal condition produced decreasing TTC for disparity and
constant TTC for size (Fig. 3). In both cases, the perturba-
tions did not freeze disparity or size. Optic size and dispar-
ity continued to change, but their temporal derivatives were
decreased at a rate that kept TTC constant. How might an
observer cope with the strange situation in which one cue
indicates that the ball will arrive soon, and the other cue
indicates that it is moving forward but will never arrive? Par-
ticipants made only very small TTC errors (Fig. 4), which
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

Box 1. Mathematical appendix.

Using the symbols defined in Fig. 1, = optical size, = binocular D = I (d/d)/[(d/dt)] (8)
subtense and = relative disparity for an approaching ball. However, substituting Eq. 8 into Eq. 7 reduces this to /(d/dt). If
the observer tracked the ball, an independent estimate of D might
(1) Behavior of the dipole model for scale changes and cue loss be recovered from ocular vergence (I/), but it would then be more
The geometric specification of TTC in the dipole model is based expedient to use TTC = /(d/dt). Observers can estimate TTC
on Eq. 4: to within 310% (51270 ms) from changing disparity when ocu-
TTCdipole = ( + ) / [(d/dt) + (d/dt)] (4) lar vergence is fixed and size cues to D are removed6. This suggests
that TTC judgments must be based on /(d/dt) or that is recov-
This can be rewritten for the case where (d/dt) > 0 and
ered from the combination of and vergence angle. Because d/dt
(d/dt) > 0 to demonstrate that the model biases its estimate
and d/dt have geometric equivalence (Fig 1), the issue is the use
toward whichever expansion pattern is most salient:
of in place of in estimating TTC. Given ID/D(D+D),
TTCdipole = then an estimate of the actual TTC will be biased:
(d / dt) TTC (d / dt) + TTC /(d/dt) = TTC(D/(D+D) = TTC TTC(D /DR)
TTCsize + TTCdisp size (d / dt)
(d / dt) disp
= The error is proportional to the distance of the reference mark
(d / dt) (d / dt)
1+ 1+ and speed of the ball:
1999 Nature America Inc. http://neurosci.nature.com

(d / dt) (d / dt)
TTCerror = TTC D/ DR = TTC2 V/ DR (9)
TTCsize is the collision time estimate arising from changing size, In the dipole model (Eq. 4), this error is partially offset by the
and TTCdisp is the collision time estimate if only binocular infor- changing size cue:
mation is used. Substituting from Eq. 4 gives an approximation
for the ratio of object size R to interpupilary distance I: TTCdipole_error = [ TTC (I/R) ( D /DR)] /( 1 + [I/R] ) (10)
For our case of a cricket or tennis ball where R I this simplifies to
TTCdipole = (TTCsize + TTCdisp [I/R]) / (1 + [I/R]) (5)
If one input is lost, then Eq. 4 reduces to a single-input estimate TTCdipole_error = 0.5 TTC D/ DR = 0.5 TTC2 V/ DR (11)
of /(d/dt) or /(d/dt). It may be the case, however, that one Hence early in the approach, Eq. 11 specifies an error of early
of the derivatives is below threshold (e.g. for a small object when arrival, then during approach the error is reduced as a function of
d/dt is lost). Then TTC = for that input, and this will intro- TTC2. If the final acquisition of visual information for catching is
duce an error term: 300 ms before contact6,12, then for a ball travelling at 5 m/s with
DR of 10 m, the error would be 22 ms. If more distant reference
TTCdipole = /(d/dt) + /( d/dt)
points were available, the error would be negligible. The pre-
The error is then proportional to the ratio of two angular subtenses: dicted timing error for the conditions of Experiment 1 is approx-
TTCsizeTTCdisp (1 + / ) imately 18 ms early. In summary, relative disparity can provide
TTCdipole = (6)
TTCsize + TTCdisp ( / ) a useful estimate of TTC, provided the final adjustments of the
interceptive action are late in the ball trajectory. The dipole model
Given similar relative thresholds for motion detection, errors also reduces the error in using relative disparity through the
should only occur when << or << ; hence the dipole weighting of changing size.
model predicts a minor error in natural settings. In experimen-
tal conditions, size or disparity of a target may be held fixed, (3) Moving away from the eye
which then presents a TTC input of infinity. The dipole error Derivations of TTC estimates in this and previous papers36 only
would then depend on the ratio of static to moving angular sub- hold for collision with the plane of the eye. To catch the ball some
tense. TTC errors occur when TTCsize and TTCdisp are placed in distance Dhand in front of the head, the catcher must estimate the
conflict at 4 s or 8 s, reciprocally5. The dipole estimate would be offset:
a TTC of 5.3 s, which is similar to the judgments recorded. Sim-
ilar TTC errors occur when is static and only disparity changes6. TTChand = TTCeye TTCeye (Dhand/Dball)
This can be approximated by using the disparities of the hand
(2) Using relative disparity to estimate TTC and ball relative to a reference object, which converges on a veridi-
It has been proposed6 that observers could base their judgments cal estimate of TTChand as the ball approaches:
of TTC on relative disparity:
TTChand = TTCeye (1 ball/hand )
TTC = I / [D(d/dt)] (7) Balls with lateral motion may be caught eccentric to the head
There are very few cues to absolute distance (D) of an object in by combining optical expansion with the optical gap between
flight, however, and the estimation of D is likely to be difficult17. the ball and its future catching location 18. Using the dipole
We note that for small optical angles, the distance D of the object model in place of the optical expansion makes this solution more
could be estimated from retinotopic angles: robust for small objects.

means they must have switched the weighting they placed on loom-
ing or disparity from one trial to the next. Observers seemed to
rely almost exclusively on looming information when there was a
plateau in TTC from disparity. In contrast, if the next ball had a
TTC plateau on looming, they switched their TTC estimate to dis-
parity (Fig 4). Stimulus-dependent switching of this type is a basic
feature of the dipole model, and the predictions of the model are
remarkably similar to the mean tendency across participants. When
both inputs were scaled to a similar plateau, observers simply saw
a ball slowing down, and there was a long delay in their responses.

nature neuroscience volume 2 no 2 february 1999 189

 1999 Nature America Inc. http://neurosci.nature.com
articles
DISCUSSION
Catching a moving ball presents a difficult problem in visual per-
ception, in which a precise estimate of arrival (TTC) must be
gleaned from binocular and monocular looming information.
Binocular information is likely to be most effective for small balls,
and looming most effective for larger objects (Eq. 1). The dipole
TTC model implicitly compensates for changes in the relative effec-
tiveness of looming or binocular motion as a function of object
size (Eq. 5). Detection of relative disparity () for a ball in flight,
however, requires an environmental reference feature, which may
be lost for a rising ball, resulting in cue dropout. Relative dispari-
ty may kick in once again if the ball falls and a background refer-
ence becomes visible. The dipole model copes with cue dropout
by implicitly switching to the remaining cue, without any explicit
evaluation procedure, and switching back to dual weighting if both
cues become available. Finally, if the binocular reference object is
only a few meters from the observer, such as a tennis net, then the
TTC estimate from disparity will differ from the TTC estimate
1999 Nature America Inc. http://neurosci.nature.com
from looming. The dipole model can attenuate this error, reducing
the 45 ms cited for the example in the introduction to a more man-
ageable 22 ms (Appendix, Eq. 1214). Our model copes with cue
conflicts that may arise from biased estimates by implicitly weight-
ing its response toward the most immediate cue, which indicates
the shortest TTC. Cue switching and immediacy bias are both
attractive features for a robust TTC system where, in an ecologi-
cal context, responding early is often preferable to responding late.
We tested the response of human participants to small differ-
ences in the TTC inputs arising from changing disparity and chang-
ing size and found that they did indeed bias their catching response
toward the more immediate cue. We then presented the challenging
task of catching balls where one of the TTC inputs stalled, but sur-
prisingly found that participants did not make the mistakes that
would be predicted from a fixed weighting of changing size and dis-
parity. In line with the dipole model, participants seemed to switch
weighting from trial to trial toward the most immediate input. The
model is also compatible with previous findings on cue conflict5.
The dipole model allows for the influence of both optic size
and disparity on TTC judgments. Cue switching is an implicit
feature of its basic architecture, and as such the model provides a
simple appealing account of adaptive behavior, particularly with
respect to cue loss. This behavior results from the combination
of retinal cues before TTC estimation and hence avoids the com-
putational overhead of deriving modular estimates for each input,
comparing them and then weighting them appropriately (for
example, Eq. 2, 3). It is possible to design more cumbersome
modular architectures that would achieve the same ends, but par-
simony favors the simplicity and efficiency of the dipole model.
METHODS
Displays. Observers wore a head-mounted display (HMD) that placed a liq-
uid crystal screen and magnifying optics in front of each eye to project stereo-
scopic images onto an image plane. In Experiment 1, horizontal display
resolution was 220 color pixels over 60 degrees of visual angle per eye, and in
Experiment 2, horizontal resolution was 263 color pixels over 30 degrees of
visual angle per eye. The vertical field of view was 45 and 23 degrees respec-
tively. In both cases, display update was >24 frames per second.
Procedure. Participants began each trial with their hands open by their
faces, with a real tennis ball attached to their palms. A virtual stereoscop-
ic textured tennis ball approached them, and they grasped onto the real
tennis balls attached to their hands when they thought that the virtual
ball would hit them. Finger and thumb flexion was recorded at 200 Hz
during the flight of the virtual ball. In both experiments, the mean start-
ing depth D was 4 m, speed V was 2 m/s and initial ball diameter R was 7
190
cm. D, V and R were randomly varied by 10% across trials.
In Experiment 1, the HMD was fully immersive, and the ball flew
down a stereoscopic 5-m corridor toward the face of the observer. Naive
participants (n = 6) received 20 practice trials. A set of 20 control trials
(Condition A) were then interleaved with 20 trials where looming was
computationally scaled to provide a TTC estimate 100 ms earlier (Con-
dition B) or 100 ms later (Condition C) than that specified by binocu-
lar disparity. A virtual image of the moving hand was not available to
participants to avoid intertrial learning.
In Experiment 2, the HMD was see-through and allowed the image
of the ball to be overlaid onto the natural world, which contained edges
at a range of depths. The screen edges of the display were also fusible at
approximately 4 m, for someone with an interpupillary distance of 65
mm. The system was calibrated to individual participants using a static
image of the ball at three different distances. Naive participants (n = 5)
held their hands next to their cheeks, occluded from their vision, and
received 20 practice trials as in Experiment 1. A set of 20 control trials
were then interleaved with plateau trials, in which the ball moved iden-
tically to the control trials until it was 1.5 m from the participant and
then either optic size was scaled to keep TTCsize constant at 750 ms or
disparity was scaled to keep TTCdisp constant at 750 ms while the other
input continued to decrease in line with the control trials.
Note: Schematic diagrams of the models discussed may be found on the
Nature Neuroscience web site at http://neurosci.nature.com/web-
_specials/supp_ingo/nn0299_186/.
ACKNOWLEDGEMENTS
We thank Anna Plooy and Martin Smyth for their input to the second
experiment and also Julie Harris, Mike Harris and Graham Schafer for incisive
comments. This research was supported by the UK EPSRC GR/L18693 and
GR/L16125.
RECEIVED 17 AUGUST; ACCEPTED 10 DECEMBER 1998
1. Sun, H. & Frost, B. J. Computation of different optical variables of looming
objects in pigeon nucleus rotundus neurons. Nat. Neurosci. 4, 296303 (1998).
2. Regan, D. & Beverley, K. I. Binocular and monocular stimuli for motion in depth:
Changing-disparity and changing-size feed the same motion in depth
stage.Vision Res. 19, 13311342 (1979).
3. Lee, D. N. A theory of visual control of braking based on information about time-
to-collision. Perception 5, 437459 (1976).
4. Regan, D. & Hamstra, S. J. Dissociation of discrimination thresholds for time to
contact and for rate of angular expansion.Vision Res. 33, 447462 (1993).
5. Heuer, H. Estimates of time to contact based on changing size and changing
target vergence. Perception 22, 549563 (1993).
6. Gray, R. & Regan, D. Accuracy of estimating time to collision using binocular and
monocular information.Vision Res. 38, 499512 (1997).
7. Wann, J. P. Anticipating arrival: is the tau-margin a specious theory? J. Exp
Psychol. Hum. Percept. Perform. 22, 10311048 (1996).
8. Tresilian, J. R. Perceptual and motor processes in interceptive timing Hum. Mov.
Sci. 13, 335373 (1994).
9. Laurent, M., Montagne, G. & Durey, A. Binocular invariants in interception
tasks: a directed perception approach. Perception 25, 14371450 (1996).
10. Regan, D., Erkelens C. J. & Collewijn, H. Necessary conditions for the perception
of motion in depth. Invest. Ophthalmol. Vision Sci. 27, 584597 (1986).
11. Milner, A. D. & Goodale, M. A. The Visual Brain in Action (Oxford Univ. Press,
Oxford, 1996).
12. Alderson, G. J. K., Sully, D. J. & Sully, H. G. An operational analysis of a one-
handed catching task using high-speed photography. J. Motor Behav. 6, 217226
(1974).
13. Tresilian, J. R. Perceptual and cognitive processes in time-to-contact estimation:
Analysis of prediction-motion and relative judgment tasks. Percept. Psychophys.
57, 231245 (1995).
14. Landy, M. S., Maloney, L. T., Johnston, E. B. & Young, M. J. Measurement and
modeling of depth cue combination: In defense of weak fusion. Vision Res. 35,
389412 (1995).
15. Dobbins, A. C., Jeo, R. M., Fiser, J. & Allman, J. M. Distance modulation of neural
activity in visual cortex. Science 281, 552555 (1998).
16. Julesz, B. Textons, the elements of texture perception and their interactions.
Nature 290, 9197 (1981).
17. Regan, D. Visual judgements and misjudgements in cricket, and the art of flight.
Perception 21, 91115 (1992).
18. Peper, C. L., Bootsma, R. J., Mestre, D. R. & Bakker, F. C. Catching balls: How to
get the hand to the right place at the right time. J. Exp Psychol. Hum. Percept.
Perform. 20, 591612 (1994).
nature neuroscience volume 2 no 2 february 1999
 1999 Nature America Inc. http://neurosci.nature.com

articles

Brain potentials indicate immediate

use of prosodic cues in natural
speech processing
Karsten Steinhauer, Kai Alter and Angela D. Friederici

Department of Neuropsychology, Max Planck Institute of Cognitive Neuroscience, PO Box 500 355, D-04303 Leipzig, Germany
Correspondence should be addressed to K.S. (steinhau@cns.mpg.de)

Spoken language, in contrast to written text, provides prosodic information such as rhythm, pauses,
accents, amplitude and pitch variations. However, little is known about when and how these
1999 Nature America Inc. http://neurosci.nature.com

features are used by the listener to interpret the speech signal. Here we use event-related brain
potentials (ERP) to demonstrate that intonational phrasing guides the initial analysis of sentence
structure. Our finding of a positive shift in the ERP at intonational phrase boundaries suggests a spe-
cific on-line brain response to prosodic processing. Additional ERP components indicate that a false
prosodic boundary is sufficient to mislead the listeners sentence processor. Thus, the application of
ERP measures is a promising approach for revealing the time course and neural basis of prosodic
information processing.

Human everyday communication is spoken rather than written,
but the overwhelming majority of psycholinguistic research
underlying models of sentence processing is still based on read-
ing rather than speech-processing data. One reason for this asym-
metry is that written text, as compared to spoken language, can be
much more easily controlled in experimental design. Natural
speech is always contaminated with prosodic features, which
increase word-length variability and introduce pitch and ampli-
tude variations; these variables can be completely avoided with
written text. Prosodic features, however, may be important for
communication, and some studies on spoken language suggest
that prosodic information may influence sentence processing to
such a degree that preferences observed during reading may not
apply equally to speech processing13. For example, eye-tracking
measures in a reading task indicate4 that sentence 1a is much eas-
ier to understand than sentence 1b:
(1a) Since Jay always jogs a mile and a half this seems like a
short distance to him.
(1b) Since Jay always jogs a mile and a half seems like a very
short distance to him.
Initially, the reader prefers to interpret the noun phrase a mile
and a half as the grammatical object of the preceding verb jogs
rather than as the subject of the subsequent verb seems, as
required in 1b. Thus the initial analysis must be revised in 1b,
which results in a prolonged reading time, a phenomenon called
the garden-path effect.
The robustness of such effects in reading studies led to the
development of the garden-path model of sentence processing4.
According to this model, initial preferences are exclusively attrib-
uted to the inherent processing principles of an encapsulated syn-
tactic processing system (called syntactic parser), which cannot be
influenced by non-syntactic information5. Semantic interpreta-
tion (or meaning) occurs only once the syntactic parsing of
grammatical relations has been accomplished. Psycholinguists
have long debated whether or not the initial syntactic preferences
can be immediately overridden by non-syntactic influences such
as semantic or pragmatic cues6,7. When non-syntactic influences
have been found, the temporal immediacy of these effects has
been controversial5. When sentences such as 1a and 1b are pre-
sented auditorily, the initial preference in favor of 1a is consid-
erably diminished by a prosodic boundary after the first verb2.
This finding suggests that prosodic information does influence
decisions about syntactic structure at very early stages.
However, little is known about the exact relationship between
prosody and sentence processing, and in particular the time
course and neural basis of influences such as the one just
described8,9. Event-related brain potentials (ERPs) are a useful
tool for the on-line examination of both normal and impaired
language processing10,11, which may contribute to the develop-
ment of theoretical accounts of prosodic and syntactic interac-
tion. The ERP correlates of initial syntactic misanalyses for both
written and spoken presentation show that the additional costs
are reflected by a late posterior positivity between 500 and 1200
ms (termed the P600 component or syntactic positive shift)1215.
In contrast, difficulties of lexical and semantic processing (for
example, He spread the warm bread with socks) generally elic-
it an earlier centroparietal negativity between 300 and 900 ms
(the N400 component)1619 rather than a P600. These two class-
es of ERP effects indicate specific brain responses to different lin-
guistic features. Here we used ERP measures for the on-line
investigation of prosodic features.
We examined the influence of prosody on human parsing per-
formance with both behavioral and neurophysiological (ERP)
measures. The stimulus material consisted of spoken German
sentence pairs similar to the English examples in 1a and 1b. To
establish controlled experimental conditions comparable to those
in reading studies, we needed exhaustive acoustic analyses of the
speech signals. These analyses revealed systematic prosodic dif-

nature neuroscience volume 2 no 2 february 1999 191

 1999 Nature America Inc. http://neurosci.nature.com

articles

Experiment 1 Experiment 2
a b

(n = 20)
(n = 20)

Fig. 1. Prosodysyntax mismatch. ERPs of conditions B (blue) and C (red) at nine electrodes from the onset of the infinitive marker zu of the crit-
ical second verb until two seconds later, displayed separately for each experiment. Negative amplitudes are plotted upwards. (a) In Experiment 1
1999 Nature America Inc. http://neurosci.nature.com

(n = 20), the prosodysyntax mismatch condition C is characterized by an N400 component followed by a P600. (b) In Experiment 2 (n = 20), the
same biphasic N400P600 pattern as in Experiment 1 was elicited in condition C.

ferences between the two conditions, suggesting a specific prosod-
ic phrasing and accentuation pattern dependent on the syntac-
tic structure. When the sentences were presented auditorily, the
listeners event-related brain potentials reproducibly showed a
characteristic positive-going waveform at prosodic phrase bound-
aries, indicating immediate decoding of this information. Fur-
thermore, in a third condition, we introduced the initial prosodic
features of one condition into the other condition, leading to a
prosodysyntax mismatch. Both behavioral and ERP data strong-
ly suggest that the prosodic features determined the initial pars-
ing decisions. The mismatch between prosody and syntax was
reliably detected by the listeners and elicited an N400P600 pat-
tern of ERP components reflecting a prosody-induced garden-
path effect.
Our hypothesis was that these early prosodic differences of
spoken language could be sufficient to prevent the garden-path
effect in 2b. If the additional IPh boundary indeed changes the
initial interpretation of Anna, then it should even be possible
to reverse the garden-path effect. That is, if the early prosodic
cues of 2b were introduced in sentence 2a, we expected an initial
misanalysis and garden-path effect in this normally easy-to-
process structure. The noun phrase Anna would then be erro-
neously attached to the second verb, which is intransitive and
cannot take a direct object as its argument. Using a cross-splicing
technique23, we merged the acoustic signals of the first part of
2b and the second part of 2a between Anna and the infinitive
marker zu (to) of the second verb in each of the 48 sentence
pairs. This resulted in a third condition (2c) with a mismatch

RESULTS
The German sentence material consisted of 48 sentence pairs
such as 2a and 2b, where the bracketing indicates the respective
intonational phrases (IPhs)20 as described below:

(2a) [Peter verspricht Anna zu arbeiten]IPh1 [und das Bro

zu putzen]IPh2
Peter promises Anna to work and to clean the office

(2b) [Peter verspricht]IPh1 [Anna zu entlasten]IPh2 [und das

Bro zu putzen]IPh3
Peter promises to support Anna and to clean the office

As in example 1a, the noun phrase Anna is the object of the

first verb verspricht (promises) in 2a and therefore belongs to
the first IPh. In contrast to 2a but similar to 1b, in sentence 2b
Anna is the object of the second transitive verb entlasten
(support) and belongs to the second IPh. Because of the Ger-
man word order, the correct interpretation of Anna is syntac-
tically disambiguated by the second verb only (that is, arbeiten
(work) in 2a versus entlasten in 2b). Compatible with the
predictions of certain theories of syntaxprosody mapping21,22,
however, the 48 speech signals of conditions 2a and 2b differ con-
siderably even before the point of syntactic disambiguation, via
different intonational phrasing and accentuation (Methods).
Fig. 2. Closure positive shift. Grand-average ERPs of both experi-
ments (n = 40) at the PZ electrode. The waveforms of conditions A
(orange) and B (blue) are superimposed. The word onsets of the sen-
tence examples are aligned to the time axis. Both conditions evoke
closure positive shifts at their respective IPh boundaries. Only one
shift is observable in condition A, following the second verb
arbeiten, whereas two such shifts occur in condition B, before
Anna and after the second verb entlasten.

192 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com

articles

Experiment 1 Experiment 2
a b

(n = 20)
(n = 20)
(n = 20)

Fig. 3. Sentence-specific ERPs. ERPs of all three conditions at nine electrode sites from sentence onset until four seconds later, plotted separately for
1999 Nature America Inc. http://neurosci.nature.com

each experiment. (a) In Experiment 1 (n = 20), conditions B (blue) and C (red) share the same pattern of two CPSs as opposed to one CPS in condi-
tion A (orange). By two seconds, that is, after onset of the second verb, the prosodysyntax mismatch condition C diverges from B and elicits the
N400P600 pattern. (b) Experiment 2 (n = 20) generally replicates the findings of Experiment 1.

between prosodic information (Methods) and syntactic con- time windows (from 200 to 1800 ms) confirmed that both effects
straints (that is, the intransitivity of the verb arbeiten): were restricted to centroparietal electrodes. The N400 effect was
significant between 400 and 1000 ms (p < 0.01) and the P600
(2c) * [Peter verspricht]IPh1 [Anna zu arbeiten]IPh2 [und das effect between 1200 and 1800 ms (p < 0.001). For both ERP com-
Bro zu putzen]IPh3 ponents, we observed only main effects of the sentence condition
Peter promises to work Anna and to clean the office (condition B versus C) and no further interactions with the task
factor. Baseline-independent, peak-to-peak measures including
The prosodic inappropriateness of 2c becomes obvious only when condition A ruled out the possibility that the effects were simply
the intransitive verb arbeiten is encountered. At this point, the due to the different verbs (transitive verbs in B versus intransi-
sentence should initially be perceived as Peter promises to work tive verbs in C). Although condition A contained the same intran-
Anna, which is certainly ungrammatical and requires revision. sitive verbs as condition C, the amplitude difference between the
According to linguistic convention, ungrammatical sentences are N400 peak and the P600 peak in condition A was significantly
marked by an asterisk.
Forty-eight sentences of each of the three conditions were pre-
sented to 40 subjects in two ERP experiments varying in their a b
task requirements: comprehension task (Experiment 1) and
prosody judgment plus comprehension task (Experiment 2). The
prosody acceptability judgment revealed that the participants
reliably detected the prosodysyntax mismatch in condition C.
In this condition, only 6% of the trials were rated as acceptable,
as opposed to more than 80% in both conditions 2a and 2b
(p < 0.0001). The error rates in the comprehension task were not
significantly increased in the mismatch condition (C) for either
Experiment 1 or 2.
We focused on two questions: first, whether and how ERPs
reflect the influence of prosodic information on early syntac-
tic processes, and second, whether ERPs are sensitive to the
processing of prosodic features per se. We briefly discuss the
effect indicating the interplay between syntax and prosody and
then turn to a newly identified prosody-related ERP compo- B (n = 16)
nent. As Experiments 1 and 2 had very similar results, they C (n = 15)
are presented jointly.
Fig. 4. ERPs after pause removal. ERPs of conditions B (blue) and C
The syntaxprosody mismatch effect (red) in Experiment 3 (n = 16) at the CZ and PZ electrode sites.
(a) Even after removal of the pause between the first verb and the sec-
In reading studies using ERPs, violations of a verbs argument
ond noun phrase, both conditions still display the CPS at the first IPh
structure elicit an N400 component followed by a P600 compo- boundary. (b) As in the first two experiments, the intransitive verb of
nent14,24. As predicted, we found a similar biphasic N400P600 the prosodysyntax mismatch condition C elicits a biphasic
sequence during the intransitive second verb for condition C as N400P600 pattern. This comparison is more reliable than that in Fig.
compared to the grammatically correct transitive verb in condition 4a, as the average window is aligned to the critical verb and more trials
B (Fig. 1). Mean amplitude analyses across 8 consecutive 200-ms entered the averages.

nature neuroscience volume 2 no 2 february 1999 193

 1999 Nature America Inc. http://neurosci.nature.com

articles

Word and pause durations Fundamental frequency (FO)

a b

Fig. 5. Prosodic parameters. Prosodic differences between the speech signals of conditions A (orange lines) and B (blue lines). (a) Duration measures
1999 Nature America Inc. http://neurosci.nature.com

of sentence fragments and pauses (#) and respective differences between conditions B and A. Condition B shows both a lengthening of the first sen-
tence fragment Peter verspricht (p < 0.0001) and a subsequent pause insertion (#1; p < 0.0001). (b) Fundamental frequency. Whereas the main
pitch accent in condition A is on the verb arbeiten, it is aligned to the noun phrase Anna in condition B. Because of the lengthening in B, however,
both accents occur at approximately the same time.

smaller than that in C (p < 0.005), and it did not differ from B
(F < 1). Thus the increased N400 and P600 amplitudes in condi-
tion C were not due to either the verb or the prosodic pattern per
se but rather to the mismatch between the two.
An ERP component reflecting prosodic processing
The prosodic processing that induced the reversed garden-path
effect was also reflected in the ERPs. In both experiments, we
found a large positive waveform at intonational phrase bound-
aries. The grand-average ERPs across both experiments (n = 40;
Fig. 2) showed that condition A evoked a single shift at its IPh
boundary following the second verb. The ERP for condition B,
in contrast, contained two such positive shifts, corresponding to
its two IPh boundaries, the first one preceding Anna and the
second after the second verb. This pattern of one versus two pos-
itive shifts was confirmed statistically by both amplitude com-
parisons (p < 0.001) and peak localizations (p < 0.0001) and did
not differ between Experiment 1 (Fig. 3a) and Experiment 2
(Fig. 3b; F < 1). Moreover, the ERP for the mismatch condition
C, which contains two IPh boundaries, also had the two corre-
sponding shifts (Fig. 3). Additional analyses revealed that the
positive shift at IPh boundaries was not due to the word preced-
ing each IPh boundary being a verb or to so-called exogenous
ERP components (such as N100 or P200) reflecting the physical
features of the stimulus. As we assume that the shift primarily
reflects the closure of an intonational phrase, we term this ERP
component the closure positive shift (CPS).
An important question is whether the CPS actually reflects
prosodic phrasing or whether it is more directly related to the
acoustic properties marking the boundary. Because in written
sentences terminal words preceding a pause are also associated
with positive ERP components25,26, the most likely candidate for
such direct correspondence is the pause insertion after the first
verb in conditions B and C, that is, the temporary absence of any
speech input at the boundary. Therefore, we ran a third ERP
experiment in which we carefully removed the entire pause in
both conditions B and C, preserving other intonational cues.
Behavioral and ERP results for these new conditions B and C
confirmed that even without the pause the prosodic boundary
was still perceived by the listeners (n = 16) and guided their ini-
tial parsing decisions (prosodic acceptability rates, 73.8 % in B,
10.9 % in C). We still observed the CPS at the first boundary
and also the N400P600 effect in condition C (Fig. 4). This find-
ing supports the idea that the CPS reflects the processing of the
prosodic boundary rather than the perception of a pause inter-
rupting the speech input.
DISCUSSION
The present study tested whether prosodic cues in spoken lan-
guage are immediately used by the listener to solve syntactic
ambiguities that systematically result in initial misunderstand-
ings during reading, and to determine whether these prosodic
influences can be monitored on-line by ERP measures. Not only
did we demonstrate that the prosodic information was sufficient
to reverse syntactic parsing preferences, but we also identified a
specific ERP component reflecting the decoding of intonational
phrasing, the closure positive shift.
With respect to psycholinguistic modeling, the data provide
strong evidence that the syntactic parser can be directly influ-
enced by prosodic information. The presence of an early IPh
boundary preceding Anna in conditions B and C stopped fur-
ther syntactic integration into the current first verb phrase and
instead prepared an initial attachment of Anna to the second
verb. This reversed parsing preference, triggered exclusively by
prosodic information, successfully induced an initial misanaly-
sis (a garden-path effect) in the mismatch condition C and elicit-
ed the predicted N400P600 pattern of ERP components on the
incompatible intransitive verb. The cognitive processes underly-
ing both components are still subject to discussion27. Here we
follow a recent interpretation28 according to which the N400
effect presumably reflects a lexical re-access necessary to confirm
the outright violation of the intransitive verb argument struc-
ture in condition C29. The P600, on the other hand, seems to
indicate the subsequent structural revision concerning the attach-
ment site of Anna. The rapidity of this on-line revision would
also explain why we did not find any increased error rates in the
comprehension task. When the question was presented one sec-
ond later, the structure had already been repaired. As the repair

194 nature neuroscience volume 2 no 2 february 1999

 1999 Nature America Inc. http://neurosci.nature.com
process is very likely to involve subvocal corrections of the into-
national phrasing, we assume that the P600 component may
reflect the costs of both syntactic and prosodic revisions.
We also found that prosodic boundaries are associated with
a large positive-going waveform, which we labeled the closure
positive shift (CPS). This component did not depend on the pres-
ence or absence of a pause acoustically interrupting the stream
of speech input. Rather, whenever the boundary was perceived
and used to guide parsing, the CPS was found in the corre-
sponding time interval. The CPS may be associated with process-
es that serve to structure the mental representation of the speech
signal and to prepare the further analysis of subsequent input. It
seems that, at least for the sentences used in the present study,
the CPS enables monitoring of prosodically driven parsing deci-
sions long before the syntactically disambiguating element (the
second verb) is encountered.
Because of the confound between prosodic and syntactic units
in natural language, the present data leave open whether the CPS
1999 Nature America Inc. http://neurosci.nature.com
is predominantly related to prosodic structuring per se or to its
consequences on syntactic processing. Our recent ERP study
using auditorily presented artificial language (unpublished data),
however, seems to support the view of prosodic processing under-
lying the CPS. On reinspection of this data to replicate our CPS
finding, a CPS was observed even in a group of naive subjects
who had not yet acquired the syntax rules of the artificial lan-
guage. Syntactic ERP effects, in contrast, were found only in par-
ticipants who were already familiar with the syntax rules
(E. Pfeifer & A.D.R., J. Cogn. Neurosci. Suppl., p. 26, 1998).
Whereas the question of which prosodic cues are necessary or
sufficient to elicit a CPS requires future research, available data
indicate that the component is tightly linked to the cognitive
process of structuring the incoming speech signal. If this finding
holds, the component may serve as a valuable tool for systematically
exploring the relationship between prosodic parsing and syntac-
tic parsing, which has received little attention until recently3032.
Our data provide strong evidence that many garden-path
effects consistently observed during reading simply may not occur
in spoken language processing. This finding would be compatible
with so-called syntax-first models such as the garden-path model
only if prosodic information is taken to directly transmit syn-
tactic information. In any case, it underlines the necessity to take
prosodic processing into account more explicitly. On-line mea-
sures such as the CPS may help to lay the empirical foundations
for an adequate theory of natural speech processing. The present
study is only a preliminary step toward understanding the brain
functions underlying prosodic processing. However, our data
strongly suggest that ERP measures used in a controlled experi-
mental design are a promising on-line approach to shed new light
on the role of prosody with regard to both normal and impaired
speech processing33.
METHODS
Subjects. Twenty students participated in each of the first two ERP exper-
iments, and sixteen in the third experiment. All 56 subjects were right-
handed 34, and without hearing or neurological disorders. All three
experiments complied with German legal requirements.
Speech signals. Forty-eight sentence pairs such as 2a and 2b were pro-
duced by a female native speaker of standard German and recorded in a
soundproof chamber. The digitized speech signals (44.1 kHz/16 bit sam-
pling rate) of each sentence were measured with respect to word and
pause durations, fundamental frequency (pitch contour) and loudness
(amplitude squares), and the differences were statistically analyzed in
paired t-tests or with ANOVAs. There were highly significant durational
nature neuroscience volume 2 no 2 february 1999
articles
differences between conditions A and B (Fig. 5a). The additional IPh
boundary in condition B was signified prosodically by a pause insertion
before Anna (p < 0.0001), as well as by a significant lengthening of the
first constituent, Peter verspricht (p < 0.0001). Whereas a major accent
occurred on the verb arbeiten in condition A, accentuation was shift-
ed to the noun phrase Anna in condition B. These differences in accent
positions were confirmed both by a locally rising pitch contour in the
fundamental frequency (p < 0.0001; Fig. 5b) and by a corresponding
loudness maximum (p < 0.01; not shown).
Condition C was derived by cross-splicing the first part of B and the
second part of A in the silent phase of the affricate /ts/ of the infinitival
marker zu (to). This procedure plus an amplitude normalization
protected against detectability of the signal manipulation at the splicing
point. Conditions B and C of the third experiment were obtained by
removing the pause before the second noun phrase, without affecting
the signals of adjacent words.
Procedure. The 144 experimental sentences were intermixed with 144
filler sentences and presented auditorily in a pseudo-randomized order
in 8 blocks of 36 trials, distributed over 2 sessions. Block order was coun-
terbalanced across subjects. EEG was continuously recorded (250 Hz/12
bit sampling rate; Neuroscan DC amplifier) from 17 cap-mounted tin
electrodes while subjects listened to the sentences in a electromagneti-
cally shielded chamber. All electrodes were referenced against left mastoid.
Impedances were kept below 5 k. In Experiment 1, the task was to
answer yes or no to comprehension questions such as Does Anna
promise to clean the office? in 20% of the trials. In Experiments 2 and 3,
participants were asked to judge the prosodic adequacy of each sentence
immediately after presentation in addition to the comprehension task.
Data analyses. EEG epochs containing eyeblinks or movement artifacts
were rejected and did not enter the ERP averages. Averages were com-
puted both across the whole sentence (Figs. 2, 3 and 4a) and for the crit-
ical second verb (Figs. 1 and 4b) using a 200-ms prestimulus baseline.
The mean amplitudes of 8 subsequent 200-ms time windows were com-
puted from 200 ms after the onset of verb 2 until 1800 ms thereafter to
measure the N400 and P600 components in conditions B and C. Com-
parisons including condition A required additional baseline-indepen-
dent peak-to-peak measures35 to balance the diverging CPS patterns.
The closure positive shift was quantified by two different approaches.
First, we compared mean amplitudes across eight subsequent 500-ms
time windows covering the whole sentence length. Second, we evaluated
the onset and offset latencies of large positive shifts at midline electrodes
separately for each subject in each condition using peak-to-peak mea-
sures. Both behavioral and ERP data were statistically tested by ANOVAs.
ERP analyses were generally done separately for midline and lateral elec-
trodes. For the midline, a global three-way ANOVA with factors condi-
tions (3) electrode (3) and the between-subject factor task (2) was used.
For the lateral electrodes, regions of interest were defined. The resulting
ANOVA design included the factors conditions (3) hemisphere (2)
position (3) task (2). Single comparisons were only computed if the
global ANOVA revealed a significant main effect of condition with df > 1
or an interaction with the factor condition. Where appropriate, Huynh
and Feldt corrections36 and a modified Bonferroni correction proce-
dure37 were applied. For illustrative purposes only, the grand-average
ERPs were smoothed off-line using a 5-Hz lowpass filter.
ACKNOWLEDGEMENTS
We appreciate the comments of Lee Osterhout and Gerry Altman on an earlier
version of the manuscript. We also thank Katja Khn, Annett Schirmer and
Franziska Kopp for their assistance in generating the stimulus materials and in
analyzing the signals. This work was supported by the Deutsche
Forschungsgemeinschaft (FR 519/17-1).
RECEIVED 10 AUGUST; ACCEPTED 16 DECEMBER 1998
1. Marslen-Wilson, W. S. et al. Prosodic effects in minimal attachment, Q. J.
Exp. Psychol. 45A, 7387 (1992).
195
 1999 Nature America Inc. http://neurosci.nature.com
articles
2. Warren, P., Grabe, E. & Nolan, F. Prosody, phonology, and parsing in closure
ambiguities. Lang. Cogn. Processes 10, 457486 (1995).
3. Stirling, L. & Wales, R. Does prosody support or direct sentence processing?
Lang. Cogn. Processes 11, 193212 (1996).
4. Frazier, L. & Rayner, K. Making and correcting errors during sentence
comprehension: Eye movements in the analysis of structurally ambiguous
sentences. Cogn. Psychol. 14, 178210 (1982).
5. Frazier, L. in Attention and Performance XII: The Psychology of Reading (ed.
Coltheart, M.) 559589 (Erlbaum, London, 1987).
6. Taraban, R. & McClelland, J. R. Constituent attachment and thematic role
assignment in sentence processing: Influence of context based expectations. J.
Mem. Lang. 27, 597632 (1988).
7. Altman, G. T. M., Garnham, A. & Dennis, Y. Avoiding the garden-path: Eye
movements in context. J. Mem. Lang. 31, 685712 (1992).
8. Cutler, A., Dahan, D. & van Donselaar, W. Prosody in the comprehension of
spoken language: A literature review. Lang. Speech 40, 141201 (1997).
9. Blumstein, S. E. in The Cognitive Neurosciences (ed. Gazzaniga, M. S.)
915929 (MIT Press, Cambridge, MA, 1995).
10. Hagoort, P. & Kutas, M. in Handbook of Neuropsychology Vol.10 (eds. Boller, F.
& Grafman, J.) 105134 (Elsevier, Amsterdam, 1995).
11. Swaab, T. Y. The Functional Locus of Comprehension Deficits in Aphasia: An
Electrophysiological Approach (Ponsen & Looijen, Wageningen, The
Netherlands, 1996).
12. Osterhout, L. & Holcomb, P. J. Event-related brain potentials and syntactic
1999 Nature America Inc. http://neurosci.nature.com
anomaly: Evidence of anomaly detection during the perception of
continuous speech. Lang. Cogn. Processes 8, 413437 (1993).
13. Osterhout, L., Holcomb, P. J. & Swinney, D. A. Brain potentials elicited by
garden-path sentences. Evidence of the application of verb information
during parsing. J. Exp. Psychol. Learn. Mem. Cogn. 20, 786806 (1994).
14. Osterhout, L. & Holcomb, P. J. in Electrophysiology of Mind: Event-Related
Potentials and Cognition (eds. Rugg, M. D. & Coles, M. G. H.) 171215
(Oxford Univ. Press, Oxford, 1995).
15. Friederici, A. D., Steinhauer, K., Mecklinger, A. & Meyer, M. Working
memory constraints on syntactic ambiguity resolution as revealed by
electrical brain responses. Biol. Psychol. 47, 193221 (1998).
16. Kutas, M. & Hillyard, S. A. Reading senseless sentences: Brain potentials
reflect semantic incongruency. Science 207, 203205 (1980).
17. Holcomb, P. J. & Neville, H. J. Natural speech processing: An analysis using
event-related brain potentials. Psychobiology 19, 286300 (1991).
18. Kutas, M. & van Petten, C. in Handbook of Psycholinguistics (ed. Gernsbacher,
M. A.) 83143 (Academic, San Diego, 1994).
196
19. Friederici, A. D., Steinhauer, K. & Frisch, S. Lexical integration: Sequential
effects of syntactic and semantic information. Mem. Cogn. (in press).
20. Nespor, M. & Vogel, I. Prosodic Phonology (Foris, Dordrecht, 1986).
21. Selkirk, E. O. Phonology and Syntax: The Relation between Sound and
Structure (MIT Press, Cambridge, 1984).
22. Halle, M. & Vergnaud, J. R. An Essay on Stress (MIT Press, Cambridge, 1987).
23. Gernsbacher, M. A. & Jescheniak, J. D. Cataphoric devices in spoken
discourse. Cogn. Psychol. 29, 2458 (1995).
24. Rsler, F., Friederici, A. D., Ptz, P. & Hahne, A. Event-related brain
potentials while encountering semantic and syntactic constraint violations. J.
Cogn. Neurosci. 5, 345362 (1993).
25. Friedman, D., Simson, R., Ritter, W. & Papin, I. The late positive component
(P300) and information processing in sentences. Electroencephalogr. Clin.
Neurophysiol. 38, 255262 (1975).
26. Van Petten, C. & Kutas, M. Influences of semantic and syntactic context on
open- and closed-class words. Mem. Cogn. 19, 95112 (1991).
27. Osterhout, L. & Hagoort, P. A superficial resemblance doent necessarily
mean that youre part of the family: Counterarguments to Coulson, King, and
Kutas (1998) in the P600/SPS-P300 debate. Lang. Cogn. Processes (in press).
28. Friederici, A. D. The time course of syntactic activation during language
processing: A model based on neuropsychological and neurophysiological
data. Brain Lang. 50, 259281 (1995).
29. Hopf, J.-M., Bayer, J., Bader, M. & Meng, M. Event-related brain potentials
and case information in syntactic ambiguities. J. Cogn. Neurosci. 10, 264280
(1998).
30. Bader, M. in Reanalysis in Sentence Processing (eds. Ferreira, F. & Fodor, J. D.)
(Kluwer, Dordrecht, in press).
31. Beckman, M. E. The parsing of prosody. Lang. Cogn. Processes 11, 1767
(1996).
32. Fodor, J. D. Learning to parse? J. Psycholinguistic Res. 27, 285319 (1998).
33. Ross, E. D. in Behavioral Neurology and Neuropsychology (eds. Feinberg, T. E.
& Farah, M. J.) 699709 (McGraw-Hill, New York, 1997).
34. Oldfield, R. C. The assessment and analysis of handedness: The Edinburgh
inventory. Neuropsychology 9, 97113 (1975).
35. Regan, D. Human Brain Electrophysiology: Evoked Potentials and Evoked
Magnetic Fields in Science and Medicine (Elsevier, New York, 1989).
36. Huynh, H. & Feldt, L. A. Conditions under which mean square ratios in
repeated measurement designs have exact F-distributions. J. Am. Stat. Assoc.
65, 15821589 (1979).
37. Keppel, G. Design and Analysis: A Researchers Handbook (Prentice-Hall,
Englewood Cliffs, New Jersey, 1991).
nature neuroscience volume 2 no 2 february 1999