Nature Neuroscience August 1998

1998 Nature America Inc. http://neurosci.nature.
com
contents
volume 1 no 4
august 1998
http://neurosci.nature.com
1998 Nature America Inc. http://neurosci.nature.com
A cortical blood vessel surrounded by dopaminergic nerve terminals. On page 286, Krimer et al. show that dopaminergic terminals are closely apposed to the cortical microvasculature, and that they may regulate cortical blood flow. See also News and Views, page 263
editorial
Speaking of twins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259 A new voice for European neuroscience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
news and views

Collision-avoidance: natures many solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 Gilles Laurent and Fabrizio Gabbiani SEE ARTICLE, PAGE 296 Neurogenic control of the cerebral microcirculation: is dopamine minding the store? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263 Costantino Iadecola SEE ARTICLE, PAGE 286 Unexpected rewards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 Kalyani Narasimhan SEE ARTICLE, PAGE 304 The missing link: a failure of fronto-hippocampal integration in schizophrenia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266 Paul Fletcher SEE ARTICLE, PAGE 318
NPY and agouti-related protein colocalized in hypothalamus. Page 271
book review
Talking Nets: An Oral History of Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . 269 Edited by J A Anderson and E Rosenfeld REVIEWED BY P DAYAN
scientific correspondence
Coexpression of Agrp and NPY in fasting-activated hypothalamic neurons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 T M Hahn, J F Breininger, D G Baskin and M W Schwartz
Viewpoint-dependent object recognition. Page 275
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 100101707. 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 Neuroscience Subscription Department, PO Box 5054, Brentwood, TN 37024-5054. Telephone (800) 524-0328, Direct Dial (615) 377 3322, Fax (615) 377 0525. Outside North America: Nature Neuroscience, Macmillan Magazines Ltd, Brunel Road, Basingstoke, Hants RG212XS, U.K. 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); Rest of world (excluding Japan): 450 (institutional/corporate), 195 (individual making personal payment); 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 1998 Nature America Inc.
nature neuroscience volume 1 no 4 august 1998
contents
A role for NMDA-receptor channels in working memory . . . . . . . . . . . . . . . . . . . 273 J E Lisman, J-M Fellous and X-J Wang Three-dimensional object recognition is viewpoint dependent . . . . . . . . . . . . . . 275 M J Tarr, P Williams, W G Hayward and I Gauthier
articles
Growing up dopamine neurons in vitro. Page 290
Target-cell-specific facilitation and depression in neocortical circuits . . . . . . . . . 279 A Reyes, R Lujan, A Rozov, N Burnashev, P Somogyi and B Sakmann Dopaminergic regulation of cerebral cortical microcirculation . . . . . . . . . . . . . . 286 L S Krimer, E C Muly, III, G V Williams and P S Goldman-Rakic SEE NEWS AND VIEWS, PAGE 263 Transplantation of expanded mesencephalic precursors leads to recovery in parkinsonian rats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 L Studer, V Tabar and R D G McKay
Computation of different optical variables of looming objects in pigeon nucleus rotundus neurons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296 H Sun and B J Frost SEE NEWS AND VIEWS, PAGE 261 Dopamine neurons report an error in the temporal prediction of reward during learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304 J R Hollerman and W Schultz SEE NEWS AND VIEWS, PAGE 265
Deciding when to duck: how pigeons avoid looming objects. Page 296 and 261
Inter-trial neuronal activity in inferior temporal cortex: a putative vehicle to generate long-term visual associations . . . . . . . . . . . . . . . . 310 V Yakovlev, S Fusi, E Berman and E Zohary Impaired recruitment of the hippocampus during conscious recollection in schizophrenia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318 S Heckers, S L Rauch, D Goff, C R Savage, D L Schacter, A J Fischman and N M Alpert SEE NEWS AND VIEWS, PAGE 266 Genetic influence on language delay in two-year-old children . . . . . . . . . . . . . . 324 P S Dale, E Simonoff, D V M Bishop, T C Eley, B Oliver, T S Price, S Purcell, J Stevenson and R Plomin
corrections and errata

Genetic effects on language learning in two-year olds. Page 324 and 259
see page 329
classified advertising
see back pages
ii
editorial
Speaking of twins
It has been said that all parents are environmentalists until they have their second child. The insight may come even sooner to those who have twins, and studies of identical and non-identical twins have provided geneticists with some of the best data for dissecting out the relative contributions of genes and environment to the emergence of human cognitive traits. The paper on page 324 from Robert Plomin and colleagues is one such example; by analyzing a large cohort of two-year-old twins, they have shown that delayed acquisition of language has a strong genetic component, and that whereas for most children the effect of genes is small relative to that of their environment, the children who are slowest to acquire language are delayed in large part because of their genetic makeup. Twins provide an almost ideal natural experiment for studies of human heritability. Monozygotic twins are genetically identical, whereas dizygotic twins share on average 50% of their autosomal genes. Of course, parents and offspring, and nontwin siblings, are equally related, but twins provide a built-in control for many important sources of environmental variation. By comparing monozygotic and dizogotic twins with each other and with the population as a whole, it is possible to measure the extent to which a given trait is affected by genetic versus environmental influences. Plomin and colleagues took advantage of the UK Office for National Statistics to contact parents of almost every pair of twins born in England and Wales in 1994. The parents were sent a questionnaire shortly before the childrens second birthday, in which they were asked to identify their twins as identical or non-identical (this is known to be about 96% accurate), and to provide information about their vocabulary as well as various non-verbal abilities. Vocabulary was estimated by identifying which words each child knew, from a test set of 100 words that had been selected as a good predictor of total vocabulary. The response rate was an impressive 44% (parents of twins are typically very supportive of genetic studies) and after discarding from the sample those children with deafness or other complications, the authors were left with more than 3000 pairs. This may be the largest cohort of twins ever examined, and is likely to be a reasonable representation of the British population as a whole. By the age of two, children have typically passed the babbling stage and are acquiring new words rapidly, with vocabularies ranging from a few to many hundreds of words. Two-year old children typically hear thousands of words per day, mainly from their parents, and it seems plausible that differences in linguistic environment may account for at least part of the variation in vocabulary. But as Chomsky has emphasized, children show an extraordinary ability to acquire language even from an impoverished stimulus, suggesting the existence of a language instinct that is buffered against varianature neuroscience volume 1 no 4 august 1998
tions in the linguistic environment. Might this ability be affected by genetic differences between individuals? Previous studies had already suggested that there would be some genetic effect, although they were based on smaller sample sizes and had not looked as early as two years, when language is first starting to emerge. The surprising result of this new study is that the relative contributions of genes and environment differ for different parts of the population distribution. Specifically, while the heritability of vocabulary for the population as a whole is fairly low (25%) with a correspondingly large environmental contribution, a much greater heritability (73%) is seen for the bottom 5% of the distribution, that is for those children who are slowest to acquire vocabulary and who would be typically classified as having specific language impairment (SLI). These results suggests a number of conclusions. First, SLI is not simply the tail of the normal distribution, but rather a distinct genetic condition (or conditions; the data do not address the question of whether SLI is genetically or functionally heterogeneous). The genetic risk arises not from a single mutation but from the combined action of a number of alleles whose effects are individually weak and largely additive (rather than depending on specific combinations, as seems to be the case for certain other conditions such as autism). Importantly, the effects of most of these alleles appear to be specific for language; although some of the children in the bottom 5% for vocabulary also performed poorly on nonverbal tests, the majority did not, and the heritability estimate was hardly altered when the low nonverbal scorers were omitted from the analysis. So far, there is no evidence that the genetic factors predisposing children to SLI produce any other systematic effects. The results therefore set the stage for identifying what might crudely be called genes for language. How close is this prospect? The technological advances in molecular genetics over the last few years have been enormous, and continue to accelerate; on some recent estimates, the human genome may be completely sequenced within the next three years. Moreover, it is no longer necessary to perform laborious linkage studies to identify genes associated with particular quantitative traits; instead, once a trait is known to be highly heritable,
259
editorial
osci.nature.com
the entire genome can be scanned with a closely spaced array of polymorphic markers to determine which ones are associated with affected individuals. Identification of the nearby genes of interest still requires significant further searching, but this will be greatly facilitated by the emerging technologies of DNA chips and single nucleotide polymorphisms, as well as by the availability of the complete genome sequence. What can we expect to learn from the identification of such genes? A plausible guess is that at least some of them will be involved in the development of brain regions that are specifically involved in the use of language. While these genes will probably not explain most of the variation in normal human linguistic ability (since the variant alleles affect mainly the bottom 5% of two-year-olds), their identification can hardly fail to provide new insights into the neural basis of language acquisition. A particularly powerful approach may be to combine genotyping with functional brain imaging and cognitive or psychophysical studies, to determine whether specific alleles have measurable effects on brain function. Moreover, since language is such a uniquely human attribute, it will be fascinating to determine whether such genes have homologs in
other primates, and whether they can shed new light on human brain evolution. Studies of human heritability, particularly of cognitive traits, have often been met with skepticism from those who fear their implications, particularly the suggestion that traits with high heritability are in some sense genetically predetermined. But such fears are based on a fallacy; heritability in a given regime of environmental variation says nothing about the possibility that different environments might lead to different effects. Plomin himself is no genetic determinist; indeed, he emphasizes that one goal of this work is to identify children who are at risk for SLI, so that they can be given preventative treatments that may help them avoid the sequelae of their condition (for instance dyslexia). And even if some of the problems prove refractory to intervention, a genetic explanation is not obviously less palatable than an environmental one that attributes the childs difficulties to the failings of the parents. Nevertheless, as similar methods are applied to other cognitive traits, some difficult issues will inevitably arise; not least of these is the danger inherent in drawing an apparently objective and genetically-based distinction between normal and abnormal personalities.
1998 Nature America Inc. http://neurosc
A new voice for European neuroscience

Despite the scale and vigor of its research activities, Europe has until now lacked a focus for neuroscience that could compare to the American-based Society for Neuroscience (SFN) and its gargantuan annual meeting. In the past, linguistic, cultural and practical barriers have all made trans-national integration much more difficult for European researchers than for their colleagues in the United States. This situation is now changing, however, and the turning point will probably be seen as the European Neuroscience Association (ENA) forum meeting that took place last month in Berlin. Comparisons with the SFN are inevitable, and for the first time the European meeting has emerged as a serious rival, if not in scale at least in terms of quality, diversity and frenetic energy, to its American counterpart. The ENA has met biannually, but attendance has been stagnant in recent years, due in large part to a lack of coordination with national societies and with the many specialized trans-European meetings which have been competing for attendance with the ENA forum. This year, however, the ENA meeting has been totally revamped, through a combination of better organization and publicity, student discounts, and the help of outside sponsors (including Nature Neuroscience). The result has been a far stronger event, with some 2500 presentations and a total attendance exceeding 4000 (a third of them students). Participants came from all over Europe, and North America was also represented both by speakers and by poster presenters (often European) from American labs. The meeting culminated in the formation of a new body, the Federation of European Neuroscience Societies (FENS), which will replace the ENA and which seems set to emerge as the official voice of neuroscience in Europe. FENS is an umbrella organization, currently based in Holland, which includes 31 of the major European national and trans-national associations. In joining FENS, the individual associations not only contribute financial resources but also declare their intention to cooperate in forging a sense of community among European researchers. By suspending their own meetings every second year, they will avoid inter-country competition and instead promote the international meeting, starting with Brighton, UK in 2000 and rotating among member countries thereafter. In addition to running the biannual meeting, FENS will likely become the new publisher of the European Journal of Neuroscience, which it may eventually make available online to all its members. It also plans to organize summer and winter schools beginning next year, and hopes in the longer term to sponsor studentships and postdoctoral fellowships. Much will depend on its ability to attract funding support from national governments or from the EC, which in turn will depend on replacing the present cacophony with a united front in which a single organization speaks for the entire neuroscience community. Like the SFN in the United States, FENS expects eventually to play a significant role in lobbying governments on behalf of the research community, and in promoting public awareness of science. The organizers of the Berlin meeting estimate the number of neuroscientists in Europe at around 50,000, making it comparable in size to the United States. The success of FENS in representing this emerging community will depend on its ability to balance the dominant influence of the wealthy nations of Western Europe with the desire for inclusiveness and representation of less scientifically developed countries, especially in the East. It will also be important to attract international support for the biannual meeting, particularly from American researchers but also from those in other countries such as Japan. Judging from progress so far, it will probably be well worth the trip.
260
news and views
Collision-avoidance: natures many solutions

Gilles Laurent and Fabrizio Gabbiani
How does the brain sense looming danger? A new study shows that specialized visual neurons in pigeons carry out several different computations in parallel to analyze signals from approaching objects such as predators.
a na. From these variables, Anyone who was led to a an observer can derive believe that catching a some of the characterisbird simply requires tics of the objects sprinkling salt on its tail d motion, as well as predict can probably testify that the time of impending this method does not collision. The big queswork well. Whether or tion is, how does the not salt on the tail has a v brain behind the exposed paralyzing effect, live b c b c retina do it? birds rarely give you the 1/(t) One possibility is that opportunity to test the 50 500 3 1 (t) neurons (or neuronal cirmethod. In fact, most 2.5 40 400 0.8 cuits) compute (t), the animals have evolved 2 30 300 0.6 rate of angular expanmany parallel warning 1.5 (t) 20 200 0.4 sion, thus indirectly systems to escape such 1 tracking object approach. undesirable encounters (t) 10 100 0.2 0.5 (t) This solution is reasonwith predators. The smell 0 0 0 0 ably simple and could of coyote urine makes -3 -2 -1 0 -3 -2 -1 0 lead to an appropriate wild mice freeze, the Time to collision (sec) Time to collision (sec) Time to collision (s) Time to collision (s) avoidance command sound of bat calls makes when the firing rate of flying crickets dive, and Fig. 1. Elementary kinematics of object approach on a collision course these neurons crosses a the sight of an approach(a) Schematics of a typical looming experiment1. An object of fixed size d certain threshold. It also ing car makes humans approaches the eye at a constant velocity v. During approach, the angle (t) run (or brake). Among subtended by the object and its rate of increase [(t) = (t)] both grow non- has an advantage: warning signals, those linearly. (b) Time course of (t) and (t) during approach (v = 300 cm per s, because large objects coming from moving d = 30 cm). (c) Corresponding time course of the functions (t), 1(t) and start to appear big earlier than small ones, the predators present a com- (t) (with = 16, see eq. 2 in ref. 1). threshold will be crossed plex challenge to the earlier during approach brain: how can their of large objects, leaving dynamic physical characmore time for escape. The solution has teristics indicate with little ambiguity of the brain and, second, that one of its downside too, in that the rate of that they represent a looming danger? these solutions, despite its apparent expansion increases faster as collision Work by Sun and Frost1 on page 296 of complexity, is found unchanged in very approaches. If this parameter (the rate different animals (birds1 and insects2). this issue of Nature Neuroscience proof expansion) is represented by a firing poses a set of solutions expressed by the An approaching object on a collision rate, the neurons firing may saturate dynamic responses of specialized visual course projects an expanding image on close to the critical time. neurons in pigeons. What makes this the retina (Fig. 1a). If the approach To avoid this, Mother Nature found work all the more interesting to us is velocity is constant, the angle (t) suba solution: the fast expansion can be that it indicates, first, that several diftended by the object grows nonlinearly slowed down by dividing the rate of ferent computationseach with its own in a near-to-exponential fashion. This angular expansion by an exponential advantages and disadvantagesare carcan be seen in Fig. 1a, where the growth function of the objects angular size ried out in parallel in the same region of is greater over the late than over the [(t) = (t) / e .(t)] (Fig. 1c). When early half of the objects approach. Similarly, the rate at which this angle the object is far away, the growth of the Gilles Laurent and Fabrizio Gabbiani are at the expands ((t) = (t)) itself increases numerator dominates and (t) (or the Division of Biology, California Institute of nonlinearly (Fig. 1b). For a given object firing rate of the neuron representing Technology, 139-74, approach velocity, specific attributes of it) increases. When the object comes Pasadena, California 91125, USA angular expansion (size, velocity, accelnearer, the denominator gains relative e-mail: GL (laurentg@cco.caltech.edu) or FG (gabbiani@klab.caltech.edu) eration) are thus projected on the retiinfluence because of its exponentiation
Angular speed (deg/s) Angular speed (deg/sec) 1/time to collision (s-1) 1/Time to collision (sec-1)
h(t) (arbitrary units) (t) (arbitrary units)
Angular size (deg) Angular size (deg)
261
news and views
Angular size (deg) Angular size (deg)
20 15 10 5 0
(t)
d2 /v2 d3 /v3
Angular speed (deg/s) Angular speed (deg/sec)
aa1 1
a2 a2
2.5 2 1.5 1 0.5 0 0 0.05 0.1 0.15 0.2 d/v (sec) d/v (s)
Tthres (s) Tthres (sec)
b1 b
1
25 20 15 10 5
(t)
b b22
2
Tthres (s) Tthres (sec)
1.5 1 0.5 0 0.1 0.2 0.3 0.4 0.5 1/2 1/2 (d/v)1/2 (s1/2)
d1/v1
-3 -2 -1 0 Time to collision (s) Time to collision (sec)
0 0 -3 -2 -1 Time to collision (s)
Fig. 2. To classify neurons into different groups, Sun and Frost1 compared the instantaneous firing rate (PSTHs) of their recorded neurons with different functions of time such as (t), (t), and (t) (in a1, b1 and c1, respectively) for different values of d/v (d1/v1 = 0.02 s, d2/v2 = 0.1 s, d3/v3 = 0.2 s in a-c). Each function also predicts specific relations between the stimulus and the PSTH. (a) If the relationship follows (t), the time at which a given angular threshold is crossed (a1) should be linearly related to d/v (a2). (b) If the relationship follows (t), the time at which a given angular velocity threshold is crossed (b1) should be linearly related to (d/v)1/2 (b2). (c) If the relationship follows (t), the time of the peak in the PSTH (c1) should be linearly related to d/v (c2). Sun and Frost s results1 indicate the presence of two classes corresponding to (b) and (c) but none to (a). In addition, a third class has a constant firing threshold, consistent with a computation.
c1 c1
(t) (arbitrary units) (t) (arbitrary units)
c22 c
1 (t)
Tpeak (s) Tpeak (sec)
2 1.5 1 0.5 0 0 0.05 0.1 0.15 0.2 d/v (s) d/v (sec)
0.8 0.6 0.4 0.2 0 -3 0 -2 -1 Time to collision (sec) Time to collision (s)
and slows down the increase in firing rate. Another advantage is that, at a critical time preceding collision, (t) reaches a peak. This peak occurs when (and thus signals that) the object subtends a particular angle, given by 2tg -1 (2/), constant for a given neuron. There are two problems associated with , however. First, to know that this angle has been reached, downstream circuits need to detect a peak firing rate. This is not trivial, and because peak detection requires comparing successive values, this operation requires time, when time is in short supply. Second, the peak signals an angle, not an actual size; it thus confounds a large object that is farther away with a small object that is near. Less time would therefore be available for escape from a small and rapidly approaching object than from a large, slow moving one. Here also, there is a potential solution. It consists of calculating yet another variable, the time-to-contact (t) or its inverse (Fig. 1c). Tau is relatively easy to compute when (t) is small and when the approach velocity is constant: (t)/(t). This measure, introduced by Gibson3 and studied behaviorally in diving birds by Lee and Reddish 4, has the advantage that it is independent of the object size or approach velocity. Tau or 1/(t) gives a running value of the
262
time before collision. By setting an appropriate threshold, it becomes possible to trigger a motor reaction at a constant delay prior to the anticipated collision. The downsides of the computation are that it provides no information about object size or velocity and that the mathematical approximation / is not valid when is large. Remarkably, each of these three potnetial solutions is reflected in the properties of neurons in nucleus rotundus of pigeons 1 , an area homologous to the mammalian inferior caudal pulvinar5 a thalamic nucleus with visual inputs from the superior colliculus and which projects to occipital, parietal and temporal cortices. Sun and Frost1 identified three groups of neurons that respond to approaching objects on a collision course by comparing the dynamics of their firing rates with kinematic functions such as (t), (t) or (t) for different object sizes and approach velocities. (See Fig. 2 for a summary of their methods.) One group of neurons shows firing profiles that are best described by a computation. A second group shows peaked firing profiles, best fitted by a computation. This is particularly interesting to us because the algorithm was first derived2 to describe the responses of DCMD, a looming-sensitive neuron in locusts68. To find such
a remarkably similar solution in such distant species (which interestingly have similar predators) supports the idea that similar problems engender similar computational solutions. Finally, Sun and Frost describe a third group of neurons whose onset of activity during approach is independent of object size or velocity, suggesting a -style computation. The existence of this last class of neurons was previously reported by Frosts group9. The properties of these neurons were confirmed in this report and analyzed further, allowing their clear distinction from the other two neuronal and computational clusters. Why are these results important? First, they focus on the dynamics of neuronal responsesrather than mean responsesas relevant neuronal signals. An early attempt at this was made by Rind and Simmons 8 in their study of locust DCMD responses. Second, the results indicate that the brain reconstructs object approach using several parallel (and possibly serial) computations. Each one provides a different piece of information about the state of the environment, and the animal thus presumably makes an informed decision on the basis of these different inputs. But how? It will be interesting to study how downstream circuits interpret these parallel messages, so as to make the best motor
news and views
decision (for example, duck, but not too early, so as to prevent course correction by predator). Does one signal dominate the others (see ref. 9)? If so, under what circumstances? Or does some combination of these signals guide behavioral responses? It will also be fascinating to study the cellular mechanisms by which these different computations are carried out. Do the neurons for example, provide inputs to the and neurons, allowing size and velocity signals to be combined? If so, how are these opera-
tions implemented biophysically? If the neurons really fire when the variable crosses a given threshold, how is this threshold set and held constant? These are some of the many interesting questions that remain to be answered, but Barrie Frost and colleagues are getting us closer to this target.
1. Sun, H. & Frost, B. J. Nature Neuroscience 1, 296303 (1998). 2. Hatsopoulos, N., Gabbiani, F. & Laurent, G. Science 270, 10001003 (1995).
3. Gibson, J. J. The Ecological Approach to Visual Perception (Houghton Mifflin, Boston, 1979). 4. Lee, D. N. & Reddish, P. E. Nature 293, 293294 (1980). 5. Karten H. J., Cox, K. & Mpodozis, J. J. Comp. Neurol. 387, 449465 (1997). 6. Palka, J. J. Insect. Physiol. 13, 235248 (1967). 7. Schlotterer, G. R. Can. J. Zool. 55, 13721376 (1977). 8. Rind, F. C. & Simmons, P. J. J. Neurophysiol. 68, 16541666 (1992). 9. Wang, Y. & Frost, B. J. Nature 356, 236238 (1992).
Neurogenic control of the cerebral microcirculation: is dopamine minding the store?

Costantino Iadecola
Cerebral blood flow is highly regulated by neural activity. New anatomical and functional evidence suggests that dopamine neurons may play a key role in this process.
It has long been known that cerebral blood vessels receive abundant projections from central and peripheral neurons, but the functional significance of this innervation has remained elusive. A new study by Krimer and colleagues on pages 286289 of this issue of Nature Neuroscience provides evidence that central dopaminergic neurons are uniquely positioned to control the cerebral microcirculation and that they may participate in the regulation of cerebral blood flow by neural activity. The ability to map changes in cerebral blood flow produced by neural activity using functional imaging has proven to be one of the most powerful tools for localization of function in the behaving human brain1. Yet the fundamental mechanisms linking neural activity to cerebral blood flow are not fully understood. The brain, perhaps more than any other organ of the body, has the intrinsic ability to regulate its own blood flow with a high degree of
Costantino Iadecola is at the Laboratory of Cerebrovascular Biology and Stroke, Department of Neurology, University of Minnesota, Box 295 UMHC, 420 Delaware Street S.E., Minneapolis, Minnesota 55455, USA email: iadec001@tc.umn.edu
spatial and temporal precision 2 . The amount of flow that each brain region receives is directly related to the functional activity of that region. Thus, if the neural activity increases, for example in specific areas of the cerebral cortex during somatosensory or visual stimuli, cerebral blood flow to the activated regions also increases. Cerebral arteries travel on the surface of the brain (pial arteries) and then enter the brain parenchyma (penetrating arterioles), giving rise to smaller arterioles and capillaries, which supply the tissue with nutrients and remove waste. Cerebral blood vessels have the ability to contract and relax in response to a wide variety of stimuli, thereby decreasing or increasing flow to the different areas of the brain. The mechanisms of this activitydependent regulation of blood flow have been investigated for many decades. A widely accepted hypothesis, articulated by Roy and Sherrington in 1890, is that working neurons release vasoactive agents in the extracellular space, which reach blood vessels by diffusion and produce relaxation of vascular smooth muscles3. However, diffusion of vasoactive metabolites, such as nitric oxide, K+ and H+ ions and adenosine from active neu-
rons cannot account in full for the rapidity and spatial definition of the increases in cerebral blood flow produced by neural activity. It was therefore proposed that selected neurons project to cerebral blood vessels and regulate cerebral blood flow by influencing vascular diameter directly. Although it is well known that cerebral blood vessels are closely associated with neural processes, there has been a long-standing controversy as to whether this proposal is correct because clear morphological and functional evidence linking central neural processes to contractile elements of the cerebral vessel wall was lacking2. Now Krimer and colleagues provide new evidence supporting a direct role of central vascular terminals in the dynamic regulation of the cerebral microcirculation. Using immunocytochemical techniques to visualize dopaminergic terminals both at the light and electron microscopic level, they demonstrate that processes from central dopaminergic neurons terminate in close contact with penetrating arterioles and cerebral capillaries in the cerebral cortex. The density of the dopaminergic vascular innervation varies regionally, being greatest in dopaminerich regions of the frontal, sensorimotor and entorhinal cortices. The dopaminergic processes terminate either directly on the vascular basal lamina or on perivascular astrocytic end-feet. In arterioles, the terminals are located on penetrating arterioles, which are important in the distribution of cerebral blood flow. At the capillary level, the terminals are almost invariably located in close proximity to pericytes, which are contractile cells embedded in the microvascular basal lamina. Krimer and colleagues also show that microapplication of dopamine in the vicinity of cerebral microvessels by iontophoresis, produces vasoconstriction in approximately 50% of the microvessels
263
news and views
studied. These data provide a functional link between dopaminergic neurotransmission and cerebral microvascular tone and strengthen the physiological relevance of the neurovascular associations. Processes of central and peripheral neurons are known to be closely associated with cerebral blood vessels (Fig. 1). Nerve fibers originating from cranial autonomic and sensory ganglia, mainly the sphenopalatine, trigeminal and superior cervical ganglia, innervate extracranial and intracranial cerebral arteries4. They contain several neurotransmitters and neuropeptides (Fig. 1) that can either constrict or dilate arteries and arterioles. Because of the diffuse nature of the innervation and the relatively large size of the vessels innervated, extrinsic perivascular nerves are unlikely to play a role in the highly localized changes in flow initiated by functional brain activity. Rather, they modulate global increases in cerebral blood flow that occur during epileptic seizures, hypertension or following transient interruption of cerebral blood flow (post-ischemic hyperemia)2. On the other hand, terminals of selected central neurons are also in close contact with cerebral blood vessels, but unlike extrinsic nerves, they are targeted to arterioles and capillaries located deeply within the brain parenchyma (Fig. 1). These vessels are more involved in the local regulation of flow at the microvascular level. In the cerebral cortex, central perivascular terminals originate either from local interneurons or from afferent pathways arising from distant regions, such as the basal forebrain, raphe or locus coeruleus2 (Fig. 1). They terminate on the vascular basal lamina, usually with an intervening glial process5. Rarely, central terminals have been observed to contact vascular smooth muscles or other contractile cells in the wall of microvessels. The paucity of ultrastructural evidence linking terminals of central neurons to contractile cells in the vascular wall has led to skepticism about the physiological significance of the central innervation. Consequently, there has been a declining interest in studying the role of central neurogenic mechanisms in the regulation of the cerebral circulation. The discovery by Krimer and colleagues that dopaminergic perivascular terminals contact penetrating arterioles, and at the capillary level pericytes, provides morphological evidence of a direct interaction between central perivascular terminals and cerebral microvessels. This finding, in con264
cert with recent data on the cholinergic perivascular innervation6, reaffirms the concept that central neural pathways can regulate cerebrovascular tone and provides the impetus for further studies investigating the role of central pathways in the mechanisms coupling brain activity to cerebral blood flow. Dopamine profoundly influences all segments of the cerebral circulation4,7. In isolated cerebral arteries and in pial arterioles in situ, application of dopamine produces vasoconstriction8. However, in the presence of serotoninergic and adrenergic receptor blockers, this amine almost invariably produces vasodilatation, suggesting that dopamine acts as an agonist on adrenergic and serotoninergic receptors. The dopamineinduced vasodilation is mediated by activation of D1 and D2 dopaminergic receptors8. In the brain slices studied by Krimer and colleagues, microapplication of dopamine onto cerebral microvessels produces vasoconstriction. It would be of interest to determine whether such vasoconstriction is mediated by adrener-
gic or serotoninergic mechanisms and whether the constrictor effect occurs also in cerebral microvessels under conditions of normal intravascular pressure and flow. Irrespective of the pharmacological characteristics of the effect, however, the observation that dopamine constricts intraparenchymal microvessels provides functional evidence for an involvement of the dopaminergic innervation in the regulation of microvascular tone. What could be the role of this dopaminergic innervation in the neural regulation of the cerebral circulation? To increase flow to areas of increased neuronal activity, vascular adjustments involving both pial arterioles and intracerebral microvessels have to occur. Thus, dilation of intracerebral vessels located at the site of activation is associated with a corresponding dilatation of distant pial arteries whose branches supply the activated region (Fig. 1). However, dilatation of a pial artery will increase flow also in inactive regions innervated by other branches of the same pial artery. Therefore, the microvascular resistance
Cranial ganglia
Pial artery
Sup. cervical: NEPI, NPY, DA, GABA, 5-HT Trigeminal: CGRP, GAL, SP, NKA Sphenopalatine: CAT VIP, NPY, PHI, NOS CAT, VIP, Somatostatin, CCK-8, SP, M-ENK, NOS, GABA Interneurons
Arteriole
Cortical afferents CAT, 5-HT, EPI, NEPI, DA
Capillaries
Fig. 1. This cartoon illustrates the vascular innervation of the cerebral cortex. At the surface of the cerebral cortex, a pial artery is shown, giving rise to a penetrating arteriole that enters the brain parenchyma and divides to form a capillary network. Peripheral nerve fibers originating from selected cranial autonomic and sensory ganglia form a dense plexus around cerebral arteries and have terminals surrounding the smooth muscle cell coat of the arterial wall. The innervation becomes less dense as the arteries become smaller. Terminals from central neurons are in close contact with small arterioles and capillaries. An interesting feature of the central dopaminergic innervation described by Krimer and colleagues is that it terminates on penetrating arterioles or near pericytes in cerebral capillaries. Therefore, the dopaminergic innervation is strategically positioned to exert a powerful influence on the local regulation of microvascular flow. Abbreviations: 5-HT, serotonin; CAT, choline acetyl-transferase; CCK-8, cholecystokinin-8; CGRP, calcitonin gene-related peptide; DA, dopamine; EPI, epinephrine; GABA, gamma-aminobutyric acid; GAL, galanin; M-ENK, metenkephalin; NEPI, norepinephrine; NOS, nitric oxide synthase; NKA, neurokinin A; NPY, neuropeptide Y; PHI, peptide histidine isoleucin; SP, substance P; VIP, vasoactive intestinal polypeptide.
news and views
within and around the activated area has to be rearranged to restrict the increase in flow to the area of increased activity alone. Penetrating arterioles are critical for flow distribution to the cortical laminae. Pericytes are abundant at the microvascular level and are likely to be significant in the redistribution of local capillary flow. Because dopaminergic innervation is in close contact with both penetrating arterioles and pericytes, these terminals are well suited to regulate microvascular flow. Therefore, one of the functions of the dopaminergic innervation might be to focus the vascular response to the activated regions. At the same time, intrinsic dopaminergic terminals contacting cerebral endothelial cells could modulate the transfer of substrates across the bloodbrain barrier. The cerebral endothelium is the site of the blood-brain barrier and is important for controlling the homeostasis of the neuronal microenvironment. For example, during neural activity, increases in blood-brain-barrier permeability might facilitate the transfer of nutrients into the brain and the removal of metabolic waste generated by active cells. Raichle and colleagues first provided evidence that central noradrenergic terminals modulate the permeability of the blood-brain barrier 9 . Thus, dopaminergic innervation, like other catecholaminergic innervation, could modulate blood-brain barrier permeability during neural activity, thereby enhancing substrate delivery to the working brain. This possibility seems plausible, considering that dopamine activates adenyl cyclase in cerebral endothelial cells, a signal-transduction system involved in endothelial cell transport10. The origin of this dopaminergic vascular innervation has not yet been identified. However, the most likely source are dopaminergic neurons in the substantia nigra and ventral tegmental area in the mesencephalon. The axons of these neurons constitute the mesocortical dopaminergic pathway and innervate the cerebral cortex (mainly prefrontal, cingulate and entorhinal cortex) in a highly topographic fashion11. Although the function of this projection is not well understood, it is likely to be involved in cognition. Dopaminergic innervation could therefore contribute to the changes in cerebral blood flow associated with cognitive tasks involving the prefrontal cortex. This speculation is supported by studies demonstrating alterations in neocortical
activation in diseases in which there is a deficit of dopamine12, 13. The close association between dopaminergic terminals and cerebral blood vessels raises the possibility that disturbances in central dopaminergic neurotransmission could alter cerebrovascular regulation and blood-brain-barrier permeability. For example, cerebral blood flow and its regulation are abnormal in schizophrenia, depression and Parkinsonism, diseases in which there is dysfunction of dopaminergic pathways1214. In particular, in schizophrenia, alterations in blood flow in the frontal lobe often parallel the clinical manifestations of the disease15. Considering the rich perivascular dopaminergic terminals in the frontal cortex, it is likely that alterations in the dopaminergic vascular innervation contribute to the vascular disturbances observed in these conditions. In addition, it is tempting to speculate that alterations in neuronal microenvironment resulting from the dysfunction in vascular regulation and blood-brain-barrier permeability may worsen the disease process and contribute to its evolution.
1. Raichle, M. E. Proc. Natl. Acad. Sci. USA 95, 765772 (1998).
2. Iadecola, C. in Cerebrovascular Diseases (eds Ginsberg, M. D. & Bogousslavsky, J.) 319332 (Blackwell Science, Cambridge, Massachusetts, 1998). 3. Iadecola, C. Trends Neurosci. 16, 206214 (1993). 4. Edvinsson, L., MacKenzie, E. T. & McCulloch, J. Cerebral Blood Flow and Metabolism (Raven Press, New York, 1993). 5. Cohen, Z., Bonvento, G., Lacombe, P. & Hamel, E. Prog. Neurobiol. 50, 335362 (1996). 6. Vaucher, E. & Hamel, E. J. Neurosci. 15, 74277441 (1995). 7. Sun, M. H., Ishine, T. & Lee, T. J. Eur. J. Pharmacol. 334, 165171 (1997). 8. Sharkey, J. & McCulloch, J. in Neural Regulation of Brain Circulation (eds Owman, C. & Hardebo, J. E.) 111127 (Elsevier, New York, 1986). 9. Raichle, M. E., Hartman, B. K., Eichling, J. O. & Sharpe, L. G. Proc. Natl. Acad. Sci. USA 72, 37263730 (1975). 10. Bacic, F., Uematsu, S., McCarron, R. M. & Spatz, M. J. Neurochem. 57, 17741780 (1991). 11. Goldman, R. P. Adv. Pharmacol. 42, 707711 (1998). 12. Taylor, S. F. Schiz. Res. 19, 129140 (1996). 13. Drevets, W. C. Annu. Rev. Med. 49, 341361 (1998). 14. Brooks, D. J. Eur. Neurol. 2, 2632 (1997). 15. Sabri, O. et al. Lancet 349, 17351739 (1997).
Unexpected rewards
Rewarded Unrewarded
Animals (including humans) predict the outcome of their behavior, and change their behavior based on the possible rewards they receive.When the rewards are different than predicted, there are long-term changes in behavior, but when rewarded exactly as predicted, we do not change our behavior. On page 304, Jeffrey Hollerman and Wolfram Schultz examine the role of dopamine neurons in reward-based learning. Monkeys were simultaneously presented with two pictures (shown here), but rewarded with liquid only when they touched a lever below one of the pictures (marked `rewarded). Dopamine neurons in the substantia nigra and ventral tegmental area were activated during early trials when the monkeys had not yet learnt to associate the correct picture with the reward. However, as the monkeys predicted the rewards more accurately with successive trials, activation was reduced. Moreover, the dopamine neurons were activated when rewarded at unpredicted times, and depressed when not rewarded as predicted. The authors suggest that because dopamine neurons are activated more strongly by unpredicted rather than predicted rewards, they may play a role in reward-dependent learning.
Kalyani Narasimhan
265
news and views
The missing link: a failure of fronto-hippocampal integration in schizophrenia

Paul Fletcher
Schizophrenics perform poorly on memory tasks. A new study suggests that this may be due to abnormal interactions between the hippocampus and prefrontal cortex.
Schizophrenia, a chronic and debilitating syndrome that disrupts a wide range of cognitive and emotional brain functions, has been linked to structural abnormalities in prefrontal and medial temporal cortex13, but these results are inconclusive4. Even functional neuroimaging studies, which offer spatially precise ways to explore the living brain in action, have led to conflicting conclusions, for example, on the role of the prefrontal cortex in the disease5,6. In this issue of Nature Neuroscience, Heckers and colleagues (pp. 318323) take a novel and exciting approach to characterizing functional neuropathology in schizophrenia. In a verbal memory task, their patients showed abnormalities in activation of both prefrontal and medial temporal regions. Further, these abnormalities were not simple failures of activation but rather prefrontal overactivation combined with hippocampal underactivation, suggesting that one area of the brain may be compensating for abnormality in another region. Their interpretation of these results moves away from the simple notion of isolated brain regions explaining the cognitive deficits in schizophrenia to consider the brain in terms of functional systems and the schizophrenic abnormality as reflecting a disruption in integration among brain areas. Schizophrenia is a disabling psychiatric disorder characterized in the acute phase by delusions, hallucinations and disorders in the logical ordering of thoughts and, more chronically, by apathy and social withdrawal. Sufferers also experience cognitive deficits, typically in the planning and execution of more complex, nonautomatic behaviors7. For example, these patients have persistent difficulty in initiating and comPaul Fletcher is at the The Wellcome Department of Cognitive Neurology, Institute of Neurology, 12 Queen Square, London WC1N 3BG, UK e-mail: p.fletcher@fil.ion.ucl.ac.uk 266
pleting such everyday tasks as shopping trips, being easily distracted and tending to give up when confronted by any obstacles. Many researchers have suggested that the core deficit of schizophrenia is a failure to appropriately activate frontal cortex during cognitive tasks (sometimes called taskrelated hypofrontality), a notion that is supported by similarities between the symptoms of schizophrenia and the problems in initiation, planning and modification of behavior associated with frontal-lobe lesions. Schizophrenia is characterized also by deficits in memory tasks 8 that engage prefrontal and medial temporal systems9, both of which seem to be abnormal in these patients. Therefore, Heckers and colleagues used positron emission tomography (PET) scanning to evaluate schizophrenic and control subjects Principal Principal during memory-retrieval sulcus sulcus tasks. Before each memoryretrieval scan, subjects learned Cingulate and Cingulate and retrosplenial a list of words. In half the triretrosplenial cortex cortex als, the words were learned by thinking about their meanings, which makes them easier to remember. In the other half, a shallow learning task was used, in which subjects were asked to examine the letters of each word for right angles (T-junctions); this approach makes later recall less successful. The authors were thus able to scan subjects Parahippocampal Parahippocampal during high and low levels of gyrus gyrus Bob Crimi memory-retrieval performance. Their previous studies Fig. 1. Connections between the hippocampal formaon nonschizophrenic sub- tion (bottom) and the prefrontal cortex (top) in the jects1012 allow them to inter- rhesus monkey. Schizophrenia in humans has been pospret their results in light of a tulated to result from disrupted functional integration sophisticated understanding among these brain regions. This figure is based on the of the cognitive processes findings of Goldman-Rakic et al. (Neuroscience 12, 719743, 1984). involved in these tasks.
Functional neuroimaging is most frequently used to evaluate regional cerebral responses to psychological tasks. Typically, subjects are scanned while performing an experimental task and also a baseline task, which differs from the experimental task in the dimension the authors wish to study. Regions that show significantly more activity in the experimental state than in the baseline state are considered to be involved in performing the task. For example, brain areas involved in speech can be identified by subtracting brain activity during the baseline task of reading words silently from the activity in the same subject during the experimental task of reading words aloud. These task-specific activity patterns can then be compared between patients and control subjects to determine how the disease affects brain function. There is a conceptual problem, however, with applying this approach to patient populations: how are we to interpret patterns of brain activity that differ between control and patient groups when the two groups are performing experimental tasks in different ways and with differing degrees of efficiency and success?
news and views
For example, does the task-related hypofrontality discussed above represent a critical abnormality in schizophrenia? Functional imaging researchers have approached this problem in two ways. Some studies have used unmodified frontal tasks, in which the performance of schizophrenics falls below that of controls. Although these studies have commonly found hypofrontality in schizophrenia5, they have been criticized because it is unclear how to interpret a failure of activation in the face of performance failure. Does the hypofrontality reflect impaired attention? Have the patients given up trying? Perhaps there is nothing inherently physiologically abnormal about the frontal cortex in schizophrenia, but rather patients are failing to select frontally mediated cognitive strategies appropriate for the task. In short, does hypofrontality cause poor performance, does it reflect poor performance or, alternatively, are both hypofrontality and poor performance caused by some upstream phenomenon that may or may not be intimately related to the disease process? Hypofrontality has generally not been found in studies adopting the alternative approach, using modified tasks to optimize the schizophrenic subjects performance and to ensure that they continue to apply themselves to the task throughout the scanning period 6. Although this approach ensures that differences in activation cannot result from schizophrenics failure to perform the tasks, the interpretation of such differences is also beset with problems. What do functional imaging abnormalities mean when they are recorded in the context of normal task performance? The most obvious interpretation of this phenomenon is that patients and controls are using different strategies to achieve similar task performance. Therefore, interpretational difficulties remain: what is the nature of the relationship between differences in brain activity and behavior? How do these two variables relate to the schizophrenic state? The problem is simple but nonetheless profound and general in the functional neuroimaging of complex syndromes such as schizophrenia. The difficulty lies in adequately specifying the task and associated behavior and therefore in defining its relationship to brain activity. The work reported by Heckers and colleagues goes some way to circumventing these difficulties because of two crucial features. First, they have used relatively well defined tasks that they have previously explored with functional imaging in normal subjects1012. They are thus in a good
position to offer plausible cognitive interpretations of differences in activation between groups. The observation that patients show significantly greater activation of right prefrontal and bilateral parietal cortex than controls is interpreted as representing an effort to compensate for the failed recruitment of the hippocampus. Because right frontal and bilateral parietal activations are commonly observed during memory retrieval9 and have been related to the explicit effort of recall under such conditions, this conclusion is both plausible and interesting. In addition to the task-specific hippocampal underactivation, the authors observed a generally higher overall level of nonspecific hippocampal activity. They interpret these two observations (that is, high baseline hippocampal activity together with a failure to elicit task-specific activation) as suggesting an abnormal cortico-hippocampal interaction in schizophrenia. This is the second crucial feature of their observations, a finding not of isolated regional hypo- or hyperactivation but rather what seems to be an abnormality of integration between brain regions. Many studies of functional brain abnormalities in mental illness have explicitly or implicitly adhered to a functional segregationist perspective4. That is, they have explored brain function and its abnormalities in terms of regional responses to cognitive task manipulations. Friston and Frith13,14 have suggested characterizing schizophrenia in terms of abnormal interactions and influences among regions, that is, in terms of functional integration. They have proposed that a disruption of prefronto-temporal integration is a core feature of the syndrome. There is currently little direct evidence in favor of this hypothesis (though indeed, functional neuroimaging may be the technique that is best suited to evaluating its merits), but there is some evidence for such a disruption6,15. In this context it is interesting, as Friston has noted, that psychotic, schizophrenia-like symptoms are a feature of metachromatic leukodystrophy, a disease that includes demyelination of subfrontal white matter. Although Heckers and colleagues have not formally characterized their present data in terms of fronto-temporal connectivity, they discuss their findings from the perspective of functional integration and its disruption, considering the hippocampal underactivation in terms of the frontal overactivation and vice-versa. Although the observation of anatomically isolated abnormalities in brain responses has much to offer the study of schizophrenia, it is likely that complete
characterization of the syndrome will require an understanding of normal and abnormal function in terms of brain integration. A much more complete picture of the functional neuropathology of schizophrenia should emerge from observations such as those reported in the current study (an overactivation in one region associated with an underactivation in an anatomically connected region), together with attempts to explore these findings in terms of both a plausible cognitive psychological explanation and an underlying disruption of functional integration. Such efforts may lead to the use of functional imaging as a diagnostic tool. Ultimately, of course, a description of the crucial abnormalities in schizophrenia must encompass both structural and functional abnormalities. If indeed prefronto-temporal integration is disrupted in schizophrenia, this raises many questions at the structural level. Are white matter tracts measurably abnormal? Can abnormalities in the disconnected regions be reliably and consistently described in terms of cytoarchitectonic and neuroreceptor profiles? A functional understanding of which specific pathways may be abnormal in schizophrenia should greatly facilitate the identification of specific mechanistic defects in the disease.
1. Benes, F. M., McSparren, J., Bird, E. D., SanGiovanni, J. P. & Vincent, S. L. Arch. Gen. Psychiatry 48, 9961001 (1991). 2. Bogerts, B. et al. Psychiatric Res. Neuroimag. 35, 113 (1991). 3. Brown, R. et al. Arch. Gen. Psychiatry 43, 3642 (1986). 4. Chua, S. E. & McKenna, P. J. Br. J. Psychiatry 166, 563582 (1995). 5. Weinberger, D. R., Berman, K. F. & Illowsky, B. P. Arch. Gen. Psychiatry 45, 609615 (1988). 6. Frith, C. D. et al. Br. J. Psychiatry 167, 343349 (1995). 7. Frith, C.D. The Cognitive Neuropsychology of Schizophrenia (Lawrence Erlbaum, Hove, 1992) 8. Tamlyn, D. et al. Psychol. Med. 22, 101115 (1992). 9. Fletcher, P. C., Frith, C. D. & Rugg, M. D. Trends Neurosci. 20, 213218 (1997). 10. Schacter, D. L., Alpert, N. M., Savage, C. R., Rauch, S. L. & Albert, M. S. Proc. Natl. Acad. Sci. USA 93, 321325 (1996). 11. Schacter, D. L., Savage, C. R., Alpert, N. M., Rauch, S. L. & Albert, M. S Neuroreport 7, 11651169 (1996). 12. Schacter, D. L. et al. Neuron 17, 267274 (1996). 13. Friston, K. J. & Frith, C. D. Clin. Neurosci. 3, 8997 (1995). 14. Friston, K. J. Schiz Res. 30, 115125 (1998). 15. Fletcher, P. C., Frith, C. D., Grasby, P. M., Friston, K. J. & Dolan, R. J. J Neurosci. 16, 70557062 (1996).
267
scientific correspondence Coexpression of Agrp and NPY in fasting-activated hypothalamic neurons

Tina M. Hahn1, John F. Breininger1, Denis G. Baskin1,2 and Michael W. Schwartz1
1
Departments of Medicine and 2Biological Structure, University of Washington, Harborview Medical Center and VA Puget Sound Health Care System, 1660 S. Columbian Way, Seattle, Washington 98108, USA Correspondence should be addressed to M.W.S. (mschwart@u.washington.edu)
Neuropeptide Y (NPY) stimulates food intake and promotes weight gain, whereas melanocortins have the opposite effect. Yet both peptides are synthesized in the arcuate nucleus, a hypothalamic area involved in energy homeostasis. We report here that mRNA encoding NPY and the melanocortin precursor, proopiomelanocortin (POMC) are expressed in adjacent, but distinct, subpopulations of arcuate nucleus neurons. Moreover, these NPY neurons coexpress mRNA encoding Agouti-related protein (Agrp), an endogenous melanocortin receptor antagonist, and fasting increases the expression of both of these mRNA species. Our findings suggest that hypothalamic NPY/Agrp neurons constitute a unique cell type that is activated by fasting to stimulate food intake via a simultaneous increase of NPY and decrease of melanocortin. The ability to recover weight lost during periods of limited access to food is important for survival. Accordingly, a highly integrated central nervous system response to starvation and weight loss has evolved that includes changes in the activity of discrete hypothalamic pathways involved in energy homeostasis1,2. Components of this response include the activation of hypothalamic neurons that contain NPY (which stimulates food intake) coupled with reduced signaling by neurons that contain melanocortins (which inhibit food intake). Melanocortin-containing cell bodies in the forebrain have been detected only in the hypothalamic arcuate nucleus (ARC), the same nucleus that contains a population of NPY neurons that are a implicated as critical in energy home3 . Similarly, hypothalamic ostasis expression of Agrp 4,5 (also known as Agouti-related transcript, ART), a newly discovered antagonist of melanocortin receptors, has also been found only in neurons of the ARC. These anatomical considerations, as b well as the ability of NPY and Agrp to promote increased food intake and weight gain1,4, led us to propose that these two gene products are expressed by the same ARC neurons, as proposed previously5. We hypothesized further that if NPY and Agrp are coexpressed by ARC neurons, their expression should be regulated in parallel during fasting. In contrast, because c melanocortins reduce food intake6, we
b
Fig. 1. Fluorescent in situ hybridization of mRNA encoding Agrp in the ARC of mice fed ad libitum (a) or fasted (b). Adult male mice (25 g) were fasted for 48 h. Brains were removed after decapitation, frozen on dry ice and sectioned at 14 m. In situ hybridization used antisense Agrp cRNA probe. Sections were viewed with a Zeis Axioplan fluorescence microscope and quantified using MCID M2 imaging software (Imaging Research, Ontario, Canada). The expression of Agrp mRNA increases with fasting. All animals were treated in accordance with University of Washington guidelines. arc, arcuate nucleus; 3v,third ventricle; me, median eminence.
predicted that the POMC gene should be expressed in ARC neurons that do not contain Agrp mRNA. To determine if NPY mRNA is coexpressed with either Agrp or POMC mRNA in ARC neurons of fed and fasted mice and rats, we colocalized
Fig. 2. Colocalization of Agrp, NPY and POMC mRNA. (a) Fluorescent in situ hybridization of mRNA encoding NPY (green), POMC (red) and NPY/POMC colocalization (yellow/orange) in the ARC of mice fasted for 48 h. Blue staining corresponds to cell nuclei. The absence of yellow/orange indicates a lack of coexpression of NPY and POMC. (b) mRNA encoding Agrp (green) and POMC mRNA (red). The paucity of yellow/orange color demonstrates that Agrp and POMC rarely colocalize. (c) mRNA encoding Agrp (red) and NPY (green) and Agrp/NPY colocalization (yellow/orange). Agrp is highly coexpressed with NPY mRNA. In situ hybridization used antisense cRNA probes for NPY, POMC and Agrp containing either digoxigenin or biotin. Slides were simultaneously hybridized with two of the three probes. Digoxigenin-labeled riboprobes were visualized with a Cy3-coupled antibody and biotin-labeled riboprobes with a Cy2-conjugated antibody (Jackson Immuno Research). 3v, third ventricle; me, median eminence.
271
Fig. 3. Colocalization of Agrp with NPY and POMC mRNA in rat arcuate nucleus. (a) Agrp (green)/POMC (red) and (b) Agrp (red)/NPY (green) colocalization (yellow/orange) in adult male rats (325g) fasted for 48 h. The absence of yellow/orange staining in (a) contrasts with (b), indicating virtually no overlap in Agrp and POMC expression, but marked coexpression of Agrp and NPY mRNAs.
mRNAs for these peptides by fluorescent in situ hybridization. Fasting was associated with a 23% increase in the number of ARC neurons that express NPY mRNA (fed, 111 8.4; fasted, 137 4.2; n = 57 per group; p < 0.02), consistent with previous reports7. Fasting did not significantly alter the number of neurons detected as expressing POMC mRNA (fed, 112 14.9; fasted, 111 9.5; n 5) in mouse ARC, in contrast to its effect on NPY cells. Thus, although previous reports demonstrate that fasting for 48 hours reduces the overall expression of POMC mRNA by about 50% in mouse ARC8, this effect cannot be accounted for by a reduction in the number of neurons that express POMC mRNA. As observed for NPY, the number of ARC neurons that expressed Agrp mRNA increased in response to fasting, although this effect failed to reach statistical significance (fed, 133 11.8; fasted, 160 7.9; n 5). Agrp mRNA hybridization was visibly increased (compare Fig. 1a and b), and this finding was confirmed using isotopic in situ hybridization and film autoradiography as described 8 . We found a striking 18-fold increase of Agrp mRNA content in response to fasting in mouse ARC (fed, 100 40%; fasted, 1807 300%; n = 12; p < 0.0005; data expressed as percent mean value in fed animals). This effect is quantitatively greater than the two- to three-fold increase of NPY mRNA detected in the ARC after fasting 3,7. Few ARC NPY neurons coexpressed POMC mRNA (Fig. 2a). Less than 10% of ARC neurons that hybridized as positive as for NPY mRNA were also positive for POMC mRNA in mice fasted for 48 hours. A similar low level of Agrp and POMC coexpression was seen in fasted mice (Fig. 2b). In contrast, nearly all NPY neurons in the mouse ARC coexpressed Agrp mRNA (Fig. 2c). Agrp mRNA was detected in 94 1.4% of ARC NPY neurons in fed mice, and in 99 0.4% of these neurons in fasted mice. In the rat ARC, Agrp mRNA was also detected in neurons that contained NPY (Fig. 3b), but not POMC mRNA (Fig. 3a). These observations support the hypothesis that NPY neurons in the ARC express Agrp mRNA, and that the levels of both NPY and Agrp mRNA are regulated in parallel. Agrp mRNA was not detected in NPY neurons in other brain areas, such as hippocampus, thalamic reticular nucleus or cerebral cortex (data not shown). Because NPY expression in these extrahypo272
thalamic areas is not affected by fasting3, these results imply that NPY neurons in the ARC are unique both in their ability to be activated by weight loss and in their coexpression of the endogenous melanocortin receptor antagonist, Agrp. The highly integrated and redundant nature of hypothalamic systems involved in energy homeostasis facilitates stability of adipose stores by ensuring that short-term changes in energy balance elicit robust compensatory responses. Thus, fasting increases production of NPY and other hypothalamic peptides that stimulate food intake, and it also inhibits signaling by pathways that contain anorectic peptides, such as the melanocortins1,2. The current findings take this concept a step further by identifying a unique NPY cell type (NPY/Agrp) that has the potential to activate neuronal feeding pathways and simultaneously inhibit pathways that cause anorexia. Thus, these data suggest that during fasting, NPY released from NPY/Agrp neurons binds to and activates postsynaptic NPY Y1 and Y5 receptors in brain areas that regulate feeding, such as the paraventricular nucleus. The simultaneous release of Agrp antagonizes MC3 and MC4 melanocortin receptors, which leads to suppression of their anorectic effects. Leptin couples changes in energy balance and/or fat stores to the appropriate hypothalamic response. The ability of fasting to reduce leptin levels may underlie the responses of both NPY/Agrp and POMC neurons in the ARC. Leptin receptor is expressed by both POMC9 and NPY10 neurons in the ARC, and leptin-deficient ob/ob mice have increased hypothalamic NPY11 and Agrp4,5 expression, but decreased POMC8 expression. Leptin administration lowers NPY12,13, but increases POMC8 mRNA levels in the ARC of ob/ob mice and normal rats. Because hyperphagia and obesity result from increased signaling by NPY14 or from decreased signaling by melanocortins4,15, the ability of leptin to regulate these hypothalamic pathways is probably important in energy homeostasis. Our results suggest that leptin can influence both NPY and melanocortin signaling by regulating a unique subset of hypothalamic neurons containing both NPY and Agrp.
Acknowledgements
We thank S. Fisher in the Department of Metabolic Diseases at Hoffmann LaRoche, Inc for sequencing the murine Agrp clones and Daniel Porte, Jr. for his helpful comments. This work was supported by grants from the NIH (DK-12829, DK-52989 and NS-32273), the Diabetes Endocrinology Research Center and Clinical Nutrition Research Unit at the University of Washington, the VA Merit Review program, and the Metabolic Diseases Unit, Hoffmann LaRoche.
RECEIVED 15 APRIL: ACCEPTED 15 JUNE 1998

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. Schwartz, M. W. & Seeley, R. J. N. Engl. J. Med. 336, 18021811 (1997). Flier, J. S. & Maratos-Flier, E. Cell 92, 437440 (1998). White, J. D. & Kershaw, M. Mol. Cell. Neurosci. 1, 4148 (1989). Ollmann, M. M., et al. Science 278, 135138 (1997). Shutter, J. R., et al. Genes Devel. 11, 593602 (1997). Fan, W., Boston, B. A., Kesterson, R. A., Hruby, V. J., Cone, R. D. Nature 385, 165168 (1997). Brady, L. S., Smith, M. A., Gold, P. W. & Herkenham, M. Neuroendocrinology 52, 441447 (1990). Schwartz, M. W. et al. Diabetes 46, 21192123 (1997). Cheung, C. C., Clifton, D. K. & Steiner, R. A. Endocrinology 138, 44894492 (1997). Mercer, J. G. et al. J. Neuroendocrinol. 8, 733735 (1996). Wilding, J. P. H. et al. Endocrinology 132, 19391944 (1993). Schwartz, M. W. et al. Diabetes 45, 531535 (1996). Stephens, T. W. et al. Nature 377, 530532 (1995). Stanley, B. G., Kyrkouli, S. E., Lampert, S. & Leibowitz, S. F. Peptides 7, 11891192 (1986). Huszar, D. et al. Cell 88, 131141 (1997).
A role for NMDA-receptor channels in working memory

John E. Lisman1, Jean-Marc Fellous2 and Xiao-Jing Wang1
1
Brandeis University, Volen Center for Complex Systems, Waltham, Massachusetts 02254-9110, USA The Salk Institute for Biological Sciences, La Jolla, California 92037, USA Correspondence should be addressed to J.E.L (lisman@binah.cc.brandeis.edu)
The NMDA class of glutamate receptors has a critical role in the induction of long-term potentiation (LTP), a synaptic modification that may encode some forms of long-term memory. However, NMDA-receptor antagonists disrupt a variety of mental processes16 that are not dependent on long-term memory. For example, they interfere with working memory1,6, a short-lasting form of memory that is maintained by neuronal activity7 rather than by synaptic modification. This suggests that there are unknown functions of the NMDA-receptor channel. One hint is that in addition to producing the calcium entry important for LTP induction, NMDA-receptor channels produce voltage-dependent excitatory postsynaptic potentials (EPSPs)8. Here, we use a net-
work model to show that such NMDA-receptor-mediated EPSPs could be critical in maintaining working memory. These results provide a mechanistic framework useful in understanding dopamine-NMDA interactions in working memory and the disruption of working memory in schizophrenia. Working memory is stored by the maintained firing of a memory-specific subset of neurons in networks of the prefrontal cortex 7. Firing is thought to be maintained by a reverberatory process9,10, in which active neurons selectively excite each other through recurrent connections. Previous models assumed that this selectivity is due to modifications of synaptic strength during earlier learning experiences, but did not address the question of how novel items could be stored in working memory. For novel items, the storage mechanism cannot depend on pre-existing synaptic selectivity, and LTP is too slow in onset to produce it11. Here we show that the voltage dependence of NMDA-receptor-mediated EPSPs can produce the selective excitation that is needed to maintain novel items in working memory. Figure 1a shows the circuit we analyzed. The pyramidal cells are uniformly interconnected by recurrent excitatory synapses having equal synaptic strengths. When a memory item is to be stored, a subset of these cells is excited by a brief informational input from an external network. To maintain this item (the spatial pattern of the subset) in working memory, these active cells must continue to fire after the external input ceases. Moreover, excitation should not spread to inactive pyramidal cells that did not receive external input. Our main argument is this: if transmission at recurrent synapses is mediated primarily by NMDA-receptor channels, there
Fig. 1. Maintenance of worka NMDA Informational input ing memory by NMDAsynapses receptor-mediated synaptic transmission at recurrent Recurrent synapses. (a) Organization of collaterals the network used to analyze working memory. Four pyramidal cells (triangles) and one Feedback interneuron are shown. The inhibition recurrent excitatory connections are all-to-all and uniform in strength. Feedback inhibition is mediated by a group of identically connected interneurons (one shown). The spatial pattern of external informational input (*) is the pattern to be remembered. The active pyramidal cells and synapses at which depolarization has increased the NMDA conductance are also starred. (b) The steady-state currentvoltage curve for a neurons synaptic conductances. I = gGABA(V-VGABA) + gAMPA V+gNMDAV/(1+0.15e0.08V). Solid dots mark three zero crossings in the solid middle curve; two of these occur at voltages where the slope is positive and the neuron is therefore bistable (gNMDA = 2.0; gAMPA = 0; gGABA = 0.4; units mS/cm2; VGABA = -80mV). If the GABA conductance is increased (0.5), bistability disappears (upper curve). If sufficient AMPA conductance (0.05) is added, bistability also disappears (lower curve). If feedback inhibition and the NMDA conductance vary linearly with the number of active cells (n), and gGABA>>gleak, then the NMDA/GABA ratio and the bistability illustrated in the solid curve are independent of n. Because the spike-generating currents are not included, the occurrence of two stable states is a prerequisite for, but does not guarantee, bistability of the actual network. (c) Network simulation showing that pyramidal cells that receive external input continue to fire after input ceases, whereas cells that do not receive input remain silent. Dendritic membrane potential (Vd) and the NMDA current (INMDA) are shown for two pyramidal cells (one active, one inactive). Rastergram (bottom) shows that activity is limited to those pyramidal cells (p-cells) that received external input (thick vertical bar). The duration of this input and the input noise are shown in bottom trace.
b
Current (A/cm Current ( A/cm ) 2)
2
gGABA
gAMPA
-5 -90 -70 -50 -30 -10
Membrane Potential (mV) Membrane potential (mV)
c
Vdd(mV) V (mV)
-40 Active Inactive -80 0 -6 -12 100 Active Inactive
p-cell # p-cell #
IINMDA(A/cm2) NMDA (A/cm )

2
50 0
50
100 150 Time (ms) Time (ms)
200
273
a A
100 p-cell p-cell ## 50 0
Control
Control
b 60
Number of active number of active cells
c
p-cell p-cell
100 50 0 100
Reduced NMDA Reducedgg NMDA
40
Increased NMDA Increasedgg NMDA
200
400 time (ms) Time (ms)
600
800
p-cell p-cell
50 0
20 0.0
0.3
2
0.6
gGABA(mS/cm gGABA (mS/cm)2)
Fig. 2. Properties of network storage of working memory. (a) Rastergram showing that sustained activity can be turned on by depolarizing input (bottom) to an arbitrary subgroup. After 0 100 200 these cells are turned off by a brief hyperpolarizing input, a different, but overlapping group can time (ms) Time (ms) be turned on. (b) The strength of inhibition determines how many active cells can fire (from top to bottom of vertical bar) in networks that satisfy the conditions for information storage (see part (c)). (c) Rastergram shows that if the NMDA-receptor conductance is decreased (33%), there is no sustained activity in pyramidal cells (top); if it is increased (66%) activity spreads to all neurons. It is the intermediate range used in (a) that is useful for information storage because activity is maintained and restricted to the subset of cells that received external input.
will be functionally selective connections between active cells. This is because the voltage-dependence of NMDA-receptor channels makes transmission conditional; transmission requires not only the binding of glutamate, but also substantial postsynaptic depolarization. Thus, when glutamate is released by an active cell onto another active cell, the pre-existing postsynaptic depolarization allows the NMDA-receptor channels to open. The resulting inward current prevents the normal repolarization process and thereby sustains the firing of these cells. In contrast, when glutamate is released by an active cell onto an inactive cell, NMDA-receptor channels will not open significantly because the postsynaptic voltage is near resting potential, a voltage range in which NMDA-receptor channels are almost completely blocked. The small NMDA-receptor current that does occur in these cells will be counteracted by global GABAergic feedback inhibition (Fig. 1a), and inactive cells will thus remain inactive. The potential for any pyramidal cell to be either active or inactive is a form of bistability. If present, this bistability should be demonstrable in the steady-state currentvoltage curve, which is the sum of the NMDA-receptor and GABA-receptor currents as a function of voltage. Under appropriate conditions, bistability is present, because there are two voltages at which there is zero current and positive slope (Fig. 1b). This bistability disappears if the GABA/NMDA conductance ratio is made too high. To examine the memory storage capability of a network of such neurons (Fig. 1a), we simulated network dynamics using 100 twocompartment pyramidal cells and 20 single-compartment inhibitory neurons. We considered the simplest case, in which the only nonsynaptic conductances were somatic conductances that produce spikes. Maintained activity occurred in the subset of cells that were excited by a brief external input representing the memory (Fig. 1c). The NMDA-receptor current that maintains activity (middle trace) is larger in active cells than inactive cells.. The network allows flexible switching between arbitrary groups representing different memories (Fig. 2a). Selective, maintained firing of the group requires that group size (Fig. 2b) and the ratio of the NMDA to GABA conductance (Fig. 2c) be within a limited range. We conclude that NMDA-receptor-mediated, voltage-dependent trans274
mission allows a network to maintain the activity of novel memories and to do so without the need for synaptic modification. Our model provides a mechanistic framework for understanding the role of NMDA-receptor channels in working memory.
The network model
The network model consists of Np pyramidal cells and Ni interneurons. The pyramidal neuron model has two compartments, representing the passive dendrite and the active soma (plus axonal initial segment)19. The somatic (Vs) and dendritic (Vd) potentials obey the following current-balance equations: CmdVs /dt = -IL - INa - IK - g c/p(Vs - Vd) - Isyn,i + Iext; CmdVd /dt = -IL - gc/(1 - p)(Vd - Vs) - Isyn,e, with Cm = 1 F/cm2, p = 0.5 and gc = 0.5 mS/cm2. The leak current IL = gL(V - VL), while the voltage-dependent currents are described by the Hodgkin-Huxley formalism. Thus, a gating variable x satisfies a first-order kinetics, dx/dt = x (x(V)(1 - x) - x(V)x). The sodium current INa = gNam3h(V - VNa), with m = -0.1(V + 32)/(exp(-0.1(V + 32)) - 1), m = 4exp (-(V + 57)/18); h = 0.07exp (-(V + 48)/20), and h = 1/ (exp(-0.1(V + 18)) + 1). The delayed rectifier IK = gKn4(V V K ), with n = -0.01(V + 34)/(exp(-0.1(V + 34))-1), and n = 0.125exp(-(V + 44)/80). The temperature factor m = h = 2.5, n = 5. Other values are: gL = 0.1, gNa = 45, gK = 18 (in mS/cm2); VL = -80, VNa = +55, VK = -80 (in mV). The excitatory synaptic currents Isyn,e = IAMPA + INMDA are located in the dendrite, and the gating kinetics are similar as in Wang et al.19. IAMPA = (gAMPA/Np) isi(V - Vsyn), where the sum is over all pyramidal cells in the network, the conductance is normalized by Np and Vsyn = 0 mV. The synaptic gating variable s obeys ds/dt = kf F(V pre)(1 - s) - kr s, with F(Vpre) = 1/(1 + exp(-Vpre/2.0)), kf = 12 and kr = 1 (in 1/ms). INMDA = (gNMDA/Np)/(1 + 0.3{Mg}exp(-0.08V)) isi(V - Vsyn), with {Mg} = 0.5 mM. Secondorder kinetics are used to produce a slow onset and decay with different time constants: ds/dt = sx(1 - s) - ss, and dx/dt = xF(Vpre)(1 x) - xx, with x = 10, x = 0.5, s = 1, s = 0.01 (in 1/ms). Each compartment also receives Poisson-like current noise (g = .005mS/cm2/ = 500Hz)). The fast spiking interneuron model and inhibitory synaptic kinetics are taken from20. Inhibitory synapses are located in the pyramidal soma. Unless specified otherwise, the maximum conductances are gNMDA = 3 (p-to-p) and 0.3 (p-to-I); gAMPA = 0.04 (p-to-p) and 0.01 (p-to-I); gGABA = 0.15 (I-to-p) (all in mS/cm2).
1998 Nature America In
Reducing NMDA-receptor-mediated transmission at recurrent synapses should lead to a decrease in memory-associated firing (Fig. 2c). A direct test of the model would be to determine whether the firing of prefrontal units during a memory task is reduced by local application of NMDA-receptor antagonist. Unpublished results (G.V. Williams and P.S. Goldman-Rakic, personal communication) suggest that this is the case. An important requirement of the NMDA mechanism of working memory is that the contribution of the other types of ionotropic glutamate channels (e.g., AMPA channels) at recurrent synapses be low (Fig. 1b). This is because only the NMDA class of ionotropic glutamate receptors has the required voltage dependence of synaptic transmission. This requirement could be met if AMPA channels were absent or if the NMDA/AMPA ratio was made high by neuromodulation. It may therefore be significant that dopaminergic modulation through D1 receptors, which is required for working memory function12, enhances NMDA-receptor-mediated transmission while reducing the AMPA-receptor-mediated transmission13. Most of the studies of this modulation have been done on striatum, and it will be important to determine whether similar modulation occurs in the prefrontal cortex, the site of working memory. Because NMDA-receptor antagonists can produce a wide range of schizophrenic symptoms, including deficits in working memory, it has been proposed that schizophrenia is caused by the hypofunction of NMDA-receptor channels4,14, and there is some evidence that this is due to an endogenous NMDA-receptor antagonist15. NMDA-receptor-mediated transmission occurs in various brain regions, and NMDA-receptor hypofunction could thus affect several different types of information storage and processing4,8,16,17 . Our model provides a mechanistic explanation of why NMDAreceptor hypofunction in the prefrontal cortex could lead to the working memory deficits in schizophrenia.
Acknowledgements
The authors thank Patricia Goldman-Rakic, Amy Arnsten, John Krystal, Charles Yang, Matthew Shapiro, Eve Marder and Larry Abbott for comments on the manuscript and Bita Moghaddam, D. Javitt and R. Greene for helpful discussions. The authors gratefully acknowledge the support of the W.M. Keck Foundation. This work was supported by the National Science Foundation, the Office of Naval Research, the National Alliance for Research on Schizophrenia and Depression, and the Alfred P. Sloan Foundation.
RECEIVED 4 MARCH: ACCEPTED 17 JUNE 1998

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. Krystal, J. H. et al. Arch. Gen. Psychiatry 51, 199214 (1994). Caramanos, Z. & Shapiro, M. L. Behav. Neurosci. 108, 3043 (1994). Verma, A. & Moghaddam, B. J. Neurosci. 16, 373379 (1996). Javitt, D. C., Steinschneider, M., Schroeder, C. E. & Arezzo, J. C. Proc. Natl. Acad. Sci. USA 93, 1196211967 (1996). Flohr, H. Neuropsychologia 13, 11691180 (1995). Adler, C. M., Goldberg, T. E., Malhotra, A. K., Pickar, D. & Breier, A. Biol. Psychiatry 43, 811816 (1998). Funahashi, S., Bruce, C. J. & Goldman-Rakic, P. J. Neurophysiol. 61, 331349 (1989). Daw, N. W., Stein, P. S. & Fox, K. Annu Rev Neurosci 16, 207222 (1993). Amit, D. J., Brunel, N. & Tsodyks, M. V. J. Neuroscience 14, 64356445 (1994). Camperi, M. & Wang, X.-J. J. Computational Neurosci. (in press, 1998). Hanse, E. & Gustafsson, B. Neurosci. Res. 20, 1525 (1994). Williams, G. V. & Goldman-Rakic, P. S. Nature 376, 572575 (1995). Cepeda, C., Buchwald, N. A. & Levine, M. S.. Proc. Natl. Acad. Sci. USA 90, 95769580 (1993). Olney, J. W. & Farber, N. B. Arch. Gen. Psychiatry 52, 9981007 (1995). Tsai, G. et al.. Arch. Gen. Psychiatry 52, 829836 (1995). Grunze, H. C. et al. J Neurosci.16, 20342043 (1996). Jensen, O. & Lisman, J. E. Learning & Memory 3, 264278 (1996). Pinsky, P. F. & Rinzel, J. J. Comput. Neurosci. 1, 3960 (1994). Wang, X. J., Colomb, D. & Rinzel, J. Proc. Natl. Acad. Sci. USA 92, 55775581 (1995). Wang, X. J. & Buzsaki, G. J. Neurosci. 16, 64026413 (1996).
a Inc. http://neurosci.nature.com
Three-dimensional object recognition is viewpoint dependent

Michael J. Tarr1, Pepper Williams2, William G. Hayward3 and Isabel Gauthier4
1
Brown University, Department of Cognitive and Linguistic Sciences, Box 1978, Providence, Rhode Island 02912, USA University of Massachusetts, Department of Psychology, 100 Morrissey Boulevard, Boston, Massachusetts 02125-3393, USA University of Wollongong, Department of Psychology, Northfields Avenue, Wollongong, NSW 2552, Australia Yale University, Department of Psychology, PO Box 208205, New Haven, Connecticut 06520-8205, USA Correspondence should be addressed to M.J.T. (Michael_Tarr@brown.edu)
The human visual system is faced with the computationally difficult problem of achieving object constancy: identifying threedimensional (3D) objects via two-dimensional (2D) retinal images that may be altered when the same object is seen from different viewpoints1. A widely accepted class of theories holds that we first reconstruct a description of the objects 3D structure from the retinal image, then match this representation to a remembered structural description. If the same structural description is reconstructed from every possible view of an object, object constancy will be obtained. For example, in Biedermans2 oft-cited recognition-by-components (RBC) theory, structural descriptions are composed of sets of simple 3D volumes called geons (Fig. 1), along with the spatial relations in which the geons are placed. Thus a mug is represented in RBC as a noodle attached to the side of a cylinder, and a suitcase as a noodle attached to the top of a brick. The attraction of geons is that, unlike more complex objects, they possess a small set of defining properties that appear in their 2D projections when viewed from almost any position (e.g., all three views of the brick in Fig. 1 include a straight main axis, parallel edges, and a straight cross section). According to the RBC theory, a complex object can therefore be recognized from its constituent geons, which can themselves be recognized from any viewpoint.
Brick Claw Cone Cylinder Fry
Horn Lemon Noodle Soap Wedge
Fig. 1. Shaded images of the three views of the ten geons used in the experiments, along with names assigned in experiment 3. The leftmost figure in each row was arbitrarily designated the 0 view; the other two figures represent 45 and 90 rotations of the objects in the depth plane.
275
A fundamental assumption of RBC is that recognition of individual geons (and therefore objects composed of geons) should be equally accurate and fast when seen from almost any viewpoint (barring accidental viewpoints3). Do humans actually recognize geons in such a viewpoint-invariant manner? Recently published reports 3,4, including single experiments using single geons, have produced inconsistent results. Our set of experiments was designed as a definitive test of RBCs postulate that geon recognition should be viewpoint invariant, including nine experiments utilizing three different tasks (sequential matching, match-to-sample and nam1a) Sequential Matching Line Drawings, Two-Key, No Feedback n = 30, F(1, 58) = 36.0
ing), two different versions of geons (line drawings and shaded images) and several other factors that might be expected to influence recognition performance. Experiments 1ae utilized a sequential matching task, in which two images were presented back to back and participants decided whether they depicted the same or different geons. (Trials in which different geons were presented were not of theoretical interest, so only the results of same trials are discussed.) Image pairs represented geons viewed from vantage points differing by 0, 45 or 90. The null hypothesis, that geons were recognized without a cost for changes in
1c) Sequential Matching Shaded Images, Go/No-Go, No Feedback 900 n = 20, F(1, 38) = 26.0
.14 .07
900 Response Time (ms)
1b) Sequential Matching Shaded Images, Two-Key, No Feedback 900 n = 20, F(1, 38) = 33.9
.18 .12
.07
800
.05
800
.04
800
.06
.01
700
'Different' trials: 852, .07
700
'Different' trials, 823, .04 700
'Different' trials, .07
1d) Sequential Matching Shaded Images, Two-Key, Feedback n = 20, F(1, 38) = 26.4
.13 .09 .06
1e) Sequential Matching 550 Shaded Images, Go/No-Go, Feedback n = 20, F(1, 38) = 54.1
2a) Match-to-sample 550 Line Drawings, Go/No-Go, No Feedback n = 31, F(1, 60) = 27.7
450
450
.05 .02
450
.02
.10
.15
350
'Different' trials, 548, .10
350
.01
350
2b) Match-to-sample Shaded Images, Go/No-Go, No Feedback n = 34, F(1, 66) = 16.7
550
2c) Match-to-sample Shaded Images, Go/No-Go, Feedback n = 20, F(1, 38) = 37.4
1000
3) Naming Shaded Images, No Feedback n = 25, F(1, 48) = 37.9 (Block 1) Block 1
.10
450
.05 .01
450
.07
900 Block 2
350 0
'Different' trials, .04 45 Viewpoint Difference 90
.02
.03
350 0
'Different' trials, .08 800 45 Viewpoint Difference 90 0 45 Viewpoint Difference 90
Fig. 2. Results of the psychophysical experiments. (1ae) Mean response times to judge that two sequentially presented images represented the same geon; (2ac) Mean response times to judge that target geon images matched a previously presented sample geon image; (3) Mean response times to name geon images. (See Fig. 1 for names given to geons.) Error bars show within-participants 95% confidence intervals13, and error rates are given beside each data point. (Error rates were not recorded in experiment 3.) Major procedural differences between experiments are given above the graphs. Experiments 1a and 2a used line drawings scanned in from ref. 3, whereas other experiments used the shaded geons shown in Fig. 1. In experiments 1a, b and d, participants pressed one key to respond same and an alternate key to respond different, whereas experiments 1c and e and 2a, b and c used a go/no-go procedure, in which participants pressed a key to respond same or did nothing otherwise. (Participants in experiment 3 spoke geon names into a microphone.) In experiments 1d and e and 2c, participants received feedback on each trial about the accuracy and speed of their response. Numbers of participants and results of linear contrast tests are given in the upper left of each graph (all contrasts were significant at the p < .0001 level), and response times (in milliseconds) and error rates for trials in which the correct response was different are given in the bottom right. (Reponse times for different trials were not recorded in go/no-go experiments.)
276 nature neuroscience volume 1 no 4 august 1998
viewpoint, is unsupportable for any of these five experiments. (See Fig. 2 for results of all nine experiments.) The same conclusion holds for experiments 2ac, which used a match-to-sample task. In this task, trials were run in blocks, where a participant saw a target geon in the 0 view, followed by three trials each of the 0, 45 and 90 views of the target geon, interspersed with nine other geons. Again, clear viewpoint-dependent effects were found. The apparent reduction in the size of the effects here compared to experiments 1ae was at least in part a result of practice; for the first trials in each block, the difference between 0 and 90 views averaged 52 ms (averaged over all 85 participants in experiments 2ac), whereas for the third trials with each view, the difference averaged only 22 ms. This interaction of trial number and viewpoint is reminiscent of results from other experiments5, in which viewpoint effects were stronger in initial than in later blocks of trials. Indeed, such practice effects may help explain why our findings are at odds with those of experiment 4 of Biederman and Gerhardstein 3, in which the procedure was similar to our match-to-sample experiments but no viewpoint effects were found. The same phenomenon was observed for the naming task used in experiment 3, in which participants first learned labels for the 0 view of each geon, then were asked to name 0, 45 and 90 views in two subsequent blocks. The effect of viewpoint difference was highly significant in the first block but greatly reduced in the second block, as the subjects learned the new viewpoints. Although viewpoint effects in each experiment were significant, it is conceivable that these overall patterns were the result of a small subset of anomalous participants and/or stimulus items. We tested for this contingency with nonparametric sign tests, which indicated that across the nine experiments, 86% of participants and an average of 9.3 of the 10 geons were faster for 0 viewpoint changes than 90 viewpoint changes, 76% of participants and 8.2 geons were faster for 0 viewpoint changes than for 45 viewpoint changes and 70% of participants and 7.3 geons were faster for 45 viewpoint changes than for 90 viewpoint changes (all z > 4.43, all p < .001). The viewpoint-dependent effects revealed in these experiments could not have been due to certain views being inherently easier or harder to process. In the sequential matching procedure of experiments 1ae, all pairwise combinations of views were tested. That is, the 0 viewpoint difference condition includes trials testing all three of the 0, 45, and 90 views in Fig. 1, the 45 viewpoint difference condition includes trials testing 045 view combinations and 45-90 view combinations in both possible orders and the 90 viewpoint difference condition includes 090 trials and 900 trials. Thus, the decrease in performance from 0 to 45 to 90 conditions results from the changes of viewpoint in the latter two conditions, not from the particular views that were tested. Like most theories of object recognition, RBC2 attempts to explain how the visual system achieves relatively consistent identification of 3D objects even though their 2D projections vary widely when seen from different viewpoints. RBCs modus operandi is to decompose objects into collections of geons and to propose that geons are equally recognizable from almost any viewpoint. If this proposition were true, then experimen-
tal participants ability to decide that two images represent an identical geon should remain constant when the two images are taken from different viewpoints. This hypothesis was decisively rejected nine times in the experiments presented here. It is still theoretically possible that objects are recognized by first parsing them into sets of geons (or some other type of primitive). However, our experiments demonstrate that structural descriptions based on geons cannot be recovered from images in a viewpoint-invariant manner, as the recognition of geons themselves is viewpoint dependent. More generally, our findings of substantial and robust viewpoint effects for extremely simple 3D volumes argue against any scheme proposing viewpoint-invariant representations within the brain as a basis for object recognition 1. This conclusion is consistent not only with recently reported behavioral experiments using multi-part objects 4,69, but also with neurophysiological studies. For example, neurons responsive to human faces have been found in the macaque superior temporal sulcus, with the majority of these cells preferential for specific views of faces10. Neurons in inferior temporal cortex show the same viewpoint specificity for novel 3D objects that monkeys were trained to recognize 11. Taken together, these neurophysiological and behavioral findings offer persuasive evidence that object recognition is a viewpoint-dependent process. This perspective is embodied in theories assuming that collections of features, surfaces, parts or entire images of objects are encoded in a viewpoint-specific manner12 and that object recognition processes are based on the similarity between encoded and perceived images. Objects seen from viewpoints increasingly different from learned views will tend to project increasingly less-similar images, so view-based theories provide a natural account for the types of viewpoint effects found here. Note: Further details of methods may be found on the Nature Neuroscience web site at http://neurosci.nature.com/ web_specials/
Acknowledgements
This research was supported by an Air Force Office of Scientific Research Grant. We thank Jay Servidea and Jaymz Rosoff for their assistance in running the psychophysical studies.
RECEIVED 4 MAY: ACCEPTED 4 JUNE 1998

1. Marr, D. Vision: A Computational Investigation into the Human Representation and Processing of Visual Information (Freeman, San Francisco, 1982). 2. Biederman, I. Psychol. Rev. 94, 115147 (1987). 3. Biederman, I. & Gerhardstein, P. C. J. Exp. Psychol. Hum. Percept. Perform. 19, 11621182 (1993). 4. Hayward, W. G. & Tarr, M. J. J. Exp. Psychol. Hum. Percept. Perform. 23, 15111521 (1997). 5. Jolicoeur, P. Mem. Cognit. 13, 289303 (1985). 6. Blthoff, H. H. & Edelman, S. Proc. Natl. Acad. Sci. USA 89, 6064 (1992). 7. Humphrey, G. K. & Khan, S. C. Can. J. Psychol. 46, 170190 (1992). 8. Tarr, M. J. Psychonomic Bull. Rev. 2, 5582 (1995). 9. Tarr, M. J., Blthoff, H. H., Zabinski, M. & Blanz, V. Psychol. Sci. 8, 282289 (1997). 10. Perrett, D. I. et al. Proc. R. Soc. Lond. B 223, 293317 (1985). 11. Logothetis, N. K., Pauls, J. & Poggio, T. Curr. Biol. 5, 552563 (1995). 12. Poggio, T. & Edelman, S. Nature 343, 263266 (1990). 13. Loftus, G. R. & Masson, M. E. J. Psychonomic Bull. Rev. 1, 476490 (1994).
277
articles
Target-cell-specific facilitation and depression in neocortical circuits

Alex Reyes1,2, R. Lujan3,4, A. Rozov1, N. Burnashev1, P. Somogyi3 and B. Sakmann1
1 2 3 4
Abteilung Zellphysiologie, Max-Planck-Institut fr medizinische Forschung, Jahnstr. 29, D - 69120, Heidelberg, Germany Present address: Center for Neural Science, New York University, 4 Washington Place, New York, New York 10003-6621, USA Medical Research Council, Anatomical Neuropharmacology Unit, University Department of Pharmacology, Mansfield Road, Oxford, OX1 3TH, UK Present address: Instituto de Neurociencias, Universidad Miguel Hernandez, Facultad de Medicina, 03550 Alicante, Spain Correspondence should be addressed to B.S. (zpsecr@sunny.mpimf-heidelberg.mpg.de)
In neocortical circuits, repetitively active neurons evoke unitary postsynaptic potentials (PSPs) whose peak amplitudes either increase (facilitate) or decrease (depress) progressively. To examine the basis for these different synaptic responses, we made simultaneous recordings from three classes of neurons in cortical layer 2/3. We induced repetitive action potentials in pyramidal cells and recorded the evoked unitary excitatory (E)PSPs in two classes of GABAergic neurons. We observed facilitation of EPSPs in bitufted GABAergic interneurons, many of which expressed somatostatin immunoreactivity. EPSPs recorded from multipolar interneurons, however, showed depression. Some of these neurons were immunopositive for parvalbumin. Unitary inhibitory (I)PSPs evoked by repetitive stimulation of a bitufted neuron also showed a less pronounced but significant difference between the two target neurons. Facilitation and depression involve presynaptic mechanisms, and because a single neuron can express both behaviors simultaneously, we infer that local differences in the molecular structure of presynaptic nerve terminals are induced by retrograde signals from different classes of target neurons. Because bitufted and multipolar neurons both formed reciprocal inhibitory connections with pyramidal cells, the results imply that the balance of activation between two recurrent inhibitory pathways in the neocortex depends on the frequency of action potentials in pyramidal cells.
The efficacy of synaptic transmission in neuronal circuits is not constant but varies with the rate of action potentials in presynaptic neurons15. During a train of action potentials, the amplitudes of successively evoked postsynaptic potentials (PSPs) either increase (facilitate) or decrease (depress). An increase or decrease in the probability of transmitter release caused by the effects of successive action potentials in the presynaptic terminal are postulated to underlie short-term modification of PSPs613. One issue is the degree to which the identity of the presynaptic and the postsynaptic neuron determines facilitation or depression in a connection. In some synapses, the target cell is the primary determinant1425, whereas in others, it is the projecting neuron (refs 26, 27; A. Reyes & B.S., submitted). We recorded simultaneously from three neurons consisting of layer 2/3 pyramidal cells and two classes of interneurons in the neocortex of juvenile (postnatal day 14) rats and found that short-term modification of PSPs in identified interneurons is target-cell-specific but is mediated by presynaptic mechanisms. These studies represent one step in understanding organizing principles governing synaptic connections in the neocortex and may shed light on how neurons interact within a network.
Results In neocortical brain slices visualized with infrared-differential contrast optics, three classes of neurons were selected based on their morphology (Fig. 1a) and their electrical excitability (Fig. 1b and c). Both multipolar and bitufted interneurons could, upon
depolarizing current injection, discharge action potentials continuously at high rates (over 20 action potentials per second). Characteristically, however, the amplitude and frequency of action potentials decreased during a train in a class of bitufted neurons (Fig. 1b) but remained constant in multipolar cells (Fig. 1c). Pyramidal cells discharged at lower rates that decreased with time (not shown). A population of other interneurons was also identified, but they were not included in this study. Pyramidal neurons formed excitatory connections with all three cell types. Suprathreshold intracellular stimulation of presynaptic pyramidal neurons evoked unitary excitatory (E)PSPs that were glutamatergic, as they were blocked by bath application of 30 M CNQX and 50 M APV (n = 14, data not shown). Nonpyramidal neurons also formed unitary connections with the three cell types. Postsynaptic potentials evoked by bitufted and multipolar neurons were completely and reversibly blocked by 20 M bicuculline (n = 20, data not shown) and were classified as GABAergic inhibitory (I)PSPs. No evidence of nonGABAergic IPSPs was found. A further physiological criterion for classification of nonpyramidal cells was the frequency-dependent change in unitary EPSP amplitudes evoked by pyramidal cell stimulation. EPSPs facilitated in bitufted cells (Fig. 1b) but depressed in multipolar cells (Fig. 1c). Identification of three classes of neurons was subsequently confirmed by reconstruction of the dendritic arbor of biocytinlabeled neurons. Pyramidal neurons had triangular somata with a prominent apical dendrite28,29 that ended in a tuft in layer 1
279
articles
Bitufted cell, facilitating input
Multipolar cell, depressing input
Fig. 1. Selection of three classes of neurons in layer 2/3. (a) Morphological selection. Representative infrared differential interference contrast enhanced video images of a pyramidal (left), bitufted (middle) and multipolar cell (right) in a slice of the somatosensory cortex taken from a two-week-old (P14) rat. These three classes were reliably identified in the living slice by the shape of their somata and by the location and number of proximal dendrites. Calibration bar is 10 m and applies to all three images. (b, c) Functional selection. Upper pair of traces show action potential patterns of bitufted (b) and multipolar (c) neurons following injection of depolarizing current steps. The resting potentials were -68 mV and -70 mV. Lower pair of traces show the presynaptic action potentials and associated EPSPs evoked in bitufted and multipolar neurons during repetitive stimulation of the presynaptic pyramidal cell. The EPSPs evoked in the bitufted cell facilitated, whereas those evoked in the multipolar cell depressed. Amplitude calibrations refer to EPSPs. The EPSPs in this and subsequent figures are averages compiled from 50200 sweeps and were evoked by delivering a 10 Hz train of brief current pulses to the presynaptic cells.
(Fig. 2a). Bitufted cells were characterized by ovoid somata from which one to four tufts of dendrites extended only in the apical and basal directions (Fig. 2a). Multipolar cells had round somata from which multiple dendrites extended radially (Fig. 2a). Nonpyramidal cells have been grouped previously on the basis of connectivity and/or neurochemical markers30,31. To correlate these groups with the cell categories selected based on the microscope image, action-potential pattern and frequency-dependent, short-term modification of EPSPs, subsets of physiologically characterized neurons were immunolabeled for the neuropeptide
somatostatin and the calcium-binding protein parvalbumin. Ten out of thirteen tested bitufted cells, with symmetrical dendritic arbors, were immunopositive for somatostatin (Fig. 2b), identifying them as a subset of GABAergic cells32. Eight somatostatinpositive cells were tested for synaptic modification; in all cases, the evoked EPSPs facilitated. None of five somatostatin-positive bitufted cells tested were immunoreactive for parvalbumin, although nearby cells displayed parvalbumin immunoreactivity. In contrast, all of the ten physiologically tested multipolar cells were immunonegative for somatostatin, but four of the five of
Fig. 2. Anatomical and immunocytochemical b pia a identification of layer 2/3 neurons. (a) Drawings of a pyramidal (blue), a multipolar (red) and a bitufted cell (green) labeled with biocytin L1 (Methods) after simultaneous triple recording. Classification of the neurons was based on their discharge pattern and response properties of EPSPs to repetitive presynaptic action potentials. The axonal arbors are not shown for clarL2 ity. Calibration bar, 50 m. Broken line indicates approximate border between cortical layer 1 c (L1) and layer 2 (L2). The mean ( standard deviation) length, width and length per width of the somata for the cell types were, respectively, 17 3 m, 10 2 m and 1.7 0.3 m for pyramidal neurons (n = 23), 20 3 m, 8 2 m and 2.8 1 m (n = 20) for bitufted cells and 15 3 m, 11 3 m and 1.4 0.3 m for multipolar cells (n = 21) following fixation and dehydration. At the age range used, pyramidal and bitufted neurons were densely spiny, whereas multipolar cells were sparsely spiny. (b) Digital micrographs of a biocytin-labeled bitufted cell (left, visualized by AMCA-labeled streptavidin) that displayed electrophysiological characteristics typical of the population. The cell is immunopositive for somatostatin, as shown on the right by indirect FITC immunofluorescence. Calibration bar (10 m) applies to both (b) and (c). (c) Digital micrographs of a biocytin-labeled multipolar cell (left, AMCA-labeled streptavidin) that displayed electrophysiological characteristics typical of the population. The cell is immunopositive for parvalbumin, as shown on the right by indirect FITC immunofluorescence.
articles
Fig. 3. Frequency-depena b dent, short-term modification of glutamatergic excitatory postsynaptic potentials evoked in two classes of neurons. (a) Simultaneous wholecell recordings were made from a triplet PB,M PM,M PB,B PB,P (schematically shown on top) in which a pyramidal cell (P) innervated a c PB bitufted (B) and a multipolar cell (M). When the pyramidal neuron was fired at 10 Hz (upper trace), the amplitude of PM the unitary EPSPs evoked successively in the bitufted cell increased EPSP2/EPSP1 (%) (middle trace), whereas the amplitude of those evoked simultaneously in the multipolar cell decreased (lower trace). Resting potentials were -65 mV for the bitufted cell and -68 mV for the multipolar cell. (b) Pairwise comparison of short-term modification of EPSPs evoked in triple recordings. Each connected pair of symbols represents the amplitude ratio of EPSPs (amplitude of second EPSP to amplitude of first EPSP, in percent) evoked simultaneously in two target neurons during 10 Hz stimulation of the same presynaptic cell. The target neurons were either a bitufted and a multipolar cell (diamonds and circles); two multipolar cells (circles); two bitufted cells (diamonds); or a pyramidal and a bitufted cell (triangles and diamonds). (c) Distribution of amplitude ratios for EPSPs evoked in bitufted (upper histogram) and multipolar cells (lower histogram). Histograms include data from paired recordings. Symbols above histograms give the mean ( standard deviation) amplitude ratios. Means were 191 82% for bitufted cells (n = 46, diamond) and 70 13% for multipolar cells (n = 61, circle).
Observations EPSP2/EPSP1 (%)
these tested were parvalbumin immunopositive (Fig. 2c), identifying them putatively as fast-discharging basket cells. To assess how short-term modification of unitary EPSPs varied with the identity of target neurons, we recorded simultaneously from three cells where a single presynaptic neuron projected to two target neurons of different identity. In Fig. 3a, a pyramidal neuron innervated a bitufted and a multipolar neuron. Action potentials in the presynaptic pyramidal cell (at 10 Hz) evoked EPSPs that facilitated in the bitufted cell but simultaneously depressed in the multipolar cell. Facilitation and depression were maintained when the presynaptic neuron was stimulated at 20, 40 and 80 Hz (n = 6, data not shown). Figure 3b shows the amplitude ratios (amplitude of second EPSP to amplitude of first EPSP times 100, in percent) for EPSPs evoked simultaneously in bitufted and multipolar neurons. The amplitude ratios of EPSPs in bitufted cells were above 100%, whereas those of EPSPs in multipolar cells were all below 100% (significantly different, paired t-test, p < 0.05, n = 9 triple recordings). Target-cell-specific differences were also observed when one of the target neurons was pyramidal and the other bitufted (Fig. 3b). In addition, the amplitude ratios of EPSPs evoked in two multipolar target cells were both below 100%, whereas those of EPSPs evoked in two bitufted neurons were both above 100% (Fig. 3b). Figure 3c summarizes the distribution of EPSP amplitude ratios for EPSPs evoked in the two nonpyramidal target cells following stimulation of a pyramidal cell. For EPSPs evoked in bitufted cells, the mean ( standard deviation) amplitude ratio was 191 82% (n = 46 connections), whereas that for EPSPs evoked in multipolar cells was 70 13% (n = 61). The amplitude ratio of EPSPs evoked between pyramidal cells was 97 23% (n = 44, data not shown; A.Reyes and B.S., submitted). The differences in the amplitude
ratios were significant (p < 0.05; two-tailed t-test). Thus, frequency-dependent, short-term modification of unitary EPSPs evoked from axon collaterals originating from a single pyramidal cell varied with the identity of the target neuron, with the input to bitufted cells showing strong facilitation. The target-specific differences in EPSP amplitude ratios were independent of the first EPSP of a train. Mean unitary EPSPs evoked by a single stimulus (at less than 0.2 Hz stimulation frequency) in multipolar cells (1.4 1.4 mV; n = 61) were larger than those evoked in bitufted cells (0.25 0.2 mV; n = 48). In bitufted cells, a linear regression fit to a plot of EPSP amplitude ratios versus EPSP amplitudes revealed no significant correlation; the slope was 145% per mV with r2 = 0.18 (data not shown). For multipolar cells, the slope was -3% per mV (r2 = 0.08). When comparison was limited to a small range of EPSP amplitudes (0.20.4 mV), the amplitude ratios for bitufted cells (177 53%; n = 8) remained significantly (p < 0.05, two-tailed t-test) larger than those for the multipolar cells (77 12%; n = 9). Frequency-dependent facilitation of pyramid-to-bitufted cell connections implies that bitufted cells will be activated effectively by rapidly discharging pyramid cells. Bitufted cells are inhibitory and have as target neurons neighboring bitufted cells, multipolar cells and pyramidal cells. Synapses established by bitufted cells also showed short-term modification of IPSPs that depended on the identity of their targets. In Fig. 4a, a bitufted cell innervated both a bitufted and a multipolar cell. When the presynaptic bitufted cell was stimulated repetitively, the IPSPs evoked in the postsynaptic bitufted cell either exhibited small depression or facilitated weakly, while those evoked simultaneously in the multipolar cell depressed strongly. Figure 4b shows the amplitude ratios of IPSPs evoked simultaneously in a bitufted and either a
281
articles
Fig. 4. Frequency-dependent a b short-term modification of GABAergic inhibitory postsynaptic potentials in two classes of interneurons. (a) Simultaneous whole-cell recordings were made from a triplet (shown on top), in which a bitufted cell (B) innervated another bitufted (B) and a multipolar (M) cell. The bitufted cell was stimulated at 10 Hz c (upper trace). The associated IPSPs evoked in bitufted cells increased in amplitude (middle trace), whereas those evoked in the multipolar cell decreased (lower trace). (b) Pairwise comparison of short-term modification of IPSPs evoked in triplets. The connected symbols represent amplitude ratios of IPSPs evoked IPSP2/IPSP1 (%) simultaneously in a bitufted (diamonds) and either a multipolar (circles) or a pyramidal (triangles) cell following 10 Hz stimulation of a presynaptic bitufted cell. (c) Distribution of amplitude ratios of IPSPs (IPSP2 to IPSP1, in percent) evoked by bitufted cell terminals in postsynaptic bitufted (upper histogram) and multipolar cells (lower histogram). Histograms include results from dual and triple recordings. Symbols above histograms give the mean ( standard deviation) IPSP amplitude ratios (bitufted cells, diamond, 101 18%, n = 24; multipolar cells, circle, 73 12%, n = 22).
Observations IPSP2/IPSP1 (%)
multipolar or a pyramidal cell following stimulation of a presynaptic bitufted cell. Though facilitation of IPSPs, unlike the EPSPs, was not prominent in bitufted cells, the amplitude ratios of IPSPs evoked in bitufted cells were nevertheless significantly higher (p < 0.05; paired t-test; n = 9) than those evoked simultaneously in multipolar cells or pyramidal cells (Fig. 4b). The mean IPSP amplitude ratio, when the target neuron was a bitufted cell (Fig. 4c), was 101 18 % (n = 24). This value was significantly (p < 0.05, two-tailed t-test) larger than that of IPSPs evoked in multipolar (73 12%, n = 22; Fig. 4c) or in pyramidal cells (71 15 %, n = 22; Fig. 6d). To assess how transmitter release mechanisms contributed to facilitation or depression, we measured how frequently presynaptic action potentials failed to evoke an EPSP during a train of three stimuli. In a facilitating connection, the number of failures decreased progressively during the train, while the occurrence of large amplitude EPSPs increased. On average (n = 20 pairs), an increase in EPSP amplitude was accompanied by a decrease in the percentage of failures (Fig. 5a). In a depressing connection, the number of failures increased during the train while the EPSP amplitude decreased. The average (n = 20 pairs) decrease of EPSP amplitude was concomitant with an increase in the failure rate (Fig. 5c). The decrease in failure rate of the second EPSP was significant for facilitating connections (paired t-test; p < 0.001; mean difference, -14%; n = 20), as was the increase for depressing EPSPs (p < 0.001; mean difference, 15%; n = 20). In facilitating connections, the increase in the mean amplitude of the second and third EPSP in the train was independent of the occurrence of the preceding EPSPs (Fig. 5b). In addition, the failure rate of the second EPSP was unaffected by the occurrence of the first EPSP (p > 0.05; paired t-test; n = 15). Thus, facilitation depended only on the occurrence of an action potential, regardless of whether it evoked an EPSP or not. In contrast, for the depressing connection, the amplitude of an EPSP in a train was larger when the preceding EPSPs failed to occur
282
(Fig. 5d). Furthermore, the failure rate of the second EPSP increased significantly (p < 0.02; n = 9) by 9% when the first EPSP had occurred. These analyses indicate that a predominantly presynaptic mechanism underlies both facilitation and depression625. Facilitation, unlike depression, however, did not depend on release of transmitter from the presynaptic terminal, whereas depression required it. A predominantly presynaptic mechanism for the frequencydependent depression of IPSP amplitudes was further suggested by a coefficient of variation analysis of amplitude fluctuations of the first and second IPSPs in a train as determined for three connections. Plots (not shown) of the squared coefficients of variation against the mean peak amplitudes, both normalized to the respective control values, revealed that the data points were below the identity line33.
Discussion Target-cell-specific modification of PSPs has been reported in several excitatory1424 and inhibitory connections25. For excitatory projections to neocortical spiny stellate and pyramidal cells, synaptic modification depends primarily on the identity of the presynaptic neuron27,34. In the neocortical connections examined here, the postsynaptic bitufted cells determined facilitation of EPSPs and an increased paired-pulse ratio of IPSPs. Several mechanisms can account for target specificity of release properties. A target neuron could locally modify release by transmitter-like substances that are liberated rapidly from the postsynaptic cell35. Because facilitation evoked by a train of action potentials is independent of a postsynaptic response (Fig. 5a and b), it is unlikely that modification of release by bitufted cells occurs on the time scale of the train. Depression, being dependent on release (Fig. 5c and d), could be generated by a rapid postsynaptic signal. On the other hand, the main difference between facilitating and depressing terminals could be a difference in the local-release fraction of vesicles, i.e. the
articles
Fig. 5. Release mechanisms in presynaptic terminals underlie frequency-dependent short-term modification. (a) Mean ( standard deviation) percentage of failures (filled diamonds, lower graph) and mean amplitude (open diamonds, upper graph) of EPSPs evoked in bitufted cells during the first, second and third presynaptic action potentials (n = 20 connections). The number of failures decreased, whereas the EPSP amplitude increased. (b) Dependence of amplitudes of EPSPs evoked successively in a bitufted cell Stimulus no Stimulus no during suprathreshold stimulation of a presyd b naptic pyramidal cell. The time of occurrence of presynaptic action potentials is indicated schematically above voltage recordings. Numbers refer to the first, second and third action potential in a train. The average EPSPs evoked during first, second and third stimulus increased progressively (upper traces) from 0.7 mV to 1.1 mV and 1.9 mV, respectively. The middle trace shows the average EPSPs evoked by the second and third action potentials under the condition that the first action potential did not evoke an EPSP. The mean amplitudes of the EPSPs were 1.1 mV and 1.6 mV. The bottom trace shows an average of the third EPSP when the first and second action potentials failed to evoke EPSPs. The mean amplitude was 1.7 mV. The amplitude of the second EPSP was not significantly affected by the occurrence of the first EPSP (p > 0.05, paired t-test, n = 15). The amplitude of the third EPSP was similarly independent of the first and second EPSPs. (c) Plot of percentage of failures and mean amplitude of EPSPs evoked in multipolar cells during stimulation of presynaptic pyramidal neurons (n = 20). Only connections that exhibited failures to the first stimulus in the train were chosen for analysis. (d) Similar analysis for EPSP amplitudes as shown in (b) for a depressing connection between a pyramidal cell and a multipolar cell. The mean amplitudes for successive action potentials in the upper trace were 0.8, 0.5 and 0.5 mV. Those for the middle trace were 0.7 and 0.5 mV and that for the lower trace was 0.7 mV. The amplitude of the second EPSP was significantly (p < 0.05; paired t-test; n = 9) decreased by the occurrence of the first EPSP.
Failures (%) EPSP (mV)
fraction of vesicles released by a single action potential. In facilitating terminals, this fraction would be small; in depressing terminals, it would be high. This local-release fraction depends on the structure of the release sites, including the density, subtype or state of calcium channels and the concentration and binding properties of endogenous calcium buffers. Therefore, factors from the postsynaptic cell, such as neurotrophins, which act on a much longer time scale 20, could modify locally the structure of presynaptic active zones. Because pyramid-to-pyramid and pyramid-to-multipolar cell connections were different from pyramid-to-bitufted cell connections, and because both EPSPs and IPSPs show the same trend, one could argue the somatostatin-expressing bitufted neurons can influence, through a long-term mechanism, the structure of the active zone such that facilitation occurs. Somatostatin-containing bitufted neurons are unique in cortex also in that they express a very high level of the postsynaptic metabotropic glutamate receptor mGluR1a36,37 and receive input from pyramidal cell recurrent axon terminals that have a very high level of the presynaptic metabotropic glutamate receptor mGluR7, located at the vesicle fusion site36. One functional consequence of the target-cell-specific facilitation and depression described here is that cortical excitation activates different local inhibitory pathways30,31, depending on the rate of action potentials in pyramidal cells. Bitufted and multipolar cells are both connected reciprocally with pyramidal cells (Fig. 6a and b) via GABAergic inhibitory synapses. IPSPs in both connections showed the same degree of depression (Fig. 6c). Pyramidal neurons discharging at a low rate would preferentialnature neuroscience volume 1 no 4 august 1998
Failures (%)
EPSP (mV)
ly excite multipolar inhibitory neurons, which will, via the feedback circuit, inhibit pyramidal cells. At higher rates, the facilitation of bitufted neuronal inputs would increasingly ensure recruitment of this population of neurons, which also inhibit pyramidal cells. One explanation of why these cells have pronounced frequency facilitation of their inputs may lie in their position in the cortical network. Their excitatory input could be dominated primarily by the level of local pyramidal cell activity, which is fed back to the distal dendrites of the pyramidal cells as GABAergic inhibition30,31,3840. Thus, an increase in pyramidal-cell discharge frequency could shift the balance of GABAergic inhibition from a perisomatic location mediated by the parvalbumin-expressing multipolar cells to a dendritic location mediated by the somatostatin-expressing bitufted cells.
Methods Brain slice preparation and visualization of neurons both in the living slice and after labeling with biocytin are described elsewhere41,42. During recordings, slices were maintained at 340C in extracellular solution consisting of (in mM) 125 NaCl, 2.5 KCl, 25 glucose, 25 H 2CO 3, 1.25 NaH2PO4, 2 CaCl2 and 1 MgCl2 (pH:7.2). Neurons were visualized via a 40x water immersion objective and two video cameras, which were mounted on a beam splitter so that the slice could also be viewed at two different magnifications (4x difference). Whole-cell voltage recordings were performed simultaneously from two or three neurons using pipettes with d.c. resistances of 515 M when filled with (in mM) 100 K gluconate, 20 KCl, 4 ATP-Mg, 10 phosphocreatine, 0.3 GTP, and 10 HEPES (pH: 7.3; 310 mOsm). In synaptically connected neurons, suprathreshold intracellular stimulation of presynaptic cells evoked depolarizing EPSPs and IPSPs. In some experiments, intracellular solution contained 283
articles
Fig. 6. Reciprocal excitatory and a c inhibitory connections between pyramidal and nonpyramidal cells. (a) Schematic diagram of reciprocal connections between a pyramidal cell, a bitufted cell and a multipolar cell in layer 2/3. (b) Dual recording from a pyramidal and a bitufted cell that were reciprocally connected. Both cells b were stimulated alternately with trains of brief current pulses. (Action potentials are shown in first and third traces.) The EPSPs evoked in the d bitufted cell facilitated (second trace), whereas the IPSPs evoked in the pyramidal cell depressed (fourth trace). (c) Dual recording from a pyramidal and a multipolar cell that were reciprocally connected. EPSPs evoked in the multipolar cell and IPSPs evoked in the pyramidal cell both depressed. Resting membrane potentials were -63 and -68 mV, respectively. (d) Summary of freIPSP2/IPSP1 (%) quency-dependent depression of IPSPs in pyramidal cells evoked by 10 Hz stimulation of either bitufted or multipolar cells. Mean ( standard deviation) IPSP amplitude ratios were 71 15% (n = 22) and 73 17% (n = 17) for bitufted-to-pyramidal and multipolar-to-pyramidal cell connections, respectively.
Observations
4 mM KCl so that the IPSPs hyperpolarized. Both depolarizing and hyperpolarizing IPSPs were taken for analyses. Stimulus delivery, data acquisition and analyses were performed using macros in IGOR (Wavemetrics, Lake Oswego, OR). After data collection, whole-cell recording was re-established using pipettes filled with intracellular solution containing 0.5% biocytin. Morphological reconstruction of the labeled cells was subsequently performed using the Neurolucida tracing program (MicroBrightField, Colchester, VT). Presynaptic cells were stimulated with a 10-Hz train of three to five suprathreshold current pulses. Trains were delivered at intervals of longer than 5 s, so that recovery from short-term modification was complete, as evidenced by the lack of systematic changes in the amplitude of the first PSP of a train during successive trains of stimuli. Voltage traces shown are averages of 50200 sweeps. The amplitude of the first EPSP of the train was defined as the difference between the peak of the EPSP and baseline. For the second or third EPSP, the amplitude was the difference between the peak of the EPSP and the baseline measured just before the peak onset. The time-to-peak of the EPSPs were sufficiently short so that temporal summation of the EPSPs did not introduce significant errors in the measurements of the peak amplitude. IPSPs had considerably longer time-to-peaks and decayed more slowly than EPSPs. To estimate the amplitude of the second IPSP, the decay of the first IPSP was fitted with a double exponential and extrapolated to the time corresponding to the peak of the IPSP. The IPSP amplitude was defined as the difference between the IPSP peak and the value of the extrapolated fit at the time of the peak. For the neurochemical characterization of nonpyramidal cells, monoclonal antibodies to somatostatin (Code: SOMA8, recognizing somatostatin and SOM-2843, ascites fluid diluted 1:500, visualized by fluorescein (FITC)-conjugated goat anti-mouse IgG, Jackson Lab) or monoclonal antibodies to parvalbumin (Sigma, No. P-3171, diluted 1:1000, visualized as above) were used sequentially on the same cell in an indirect immunofluorescence method. Each cell was examined after each layer of immunoreaction. The two antibodies gave nonoverlapping labeling patterns, and these served as controls for the method. The biocytin-filled cells were revealed by 7-amino-4-methylcoumarin-3-acetic acid
284
(AMCA)-conjugated streptavidin (Vector Lab.). A Leica dichroic mirror system and the A4 filter block was used for recording AMCA, the L5 block for recording FITC fluorescence. Cells were recorded on a cooled CCD camera, analyzed and displayed using the Openlab software and color palette (Improvision, Coventry, UK). The immunonegativity of a cell for a marker could be due to damage caused by the recording, an undetectable low level of the molecule or the genuine absence of the marker. The consistent absence of parvalbumin from bitufted cells and somatostatin from multipolar cells strongly suggests the two cell populations are neurochemically different. Following immunocytochemical characterization, the axonal and dendritic patterns of the cells were visualized by standard avidin-biotinylated-horseradish peroxidase complex (Vector Lab) and DAB reaction.
Acknowledgments
We thank E. Neher, B. Katz and G. Borst for their comments on the manuscript and B. Katz for suggesting the term local release fraction of vesicles. We also thank J. C. Brown, at the MRC of Canada Group on Regulatory Peptides, Vancouver, for the gift of monoclonal antibodies to somatostatin, Z. Nusser and J. D. B. Roberts for assistance with digital micrography and Z. Ahmad for excellent technical assistance.
RECEIVED 24 APRIL: ACCEPTED 19 MAY 1998

1. Thomson, A. M., Deuchars, J. & West, D. C. Single axon excitatory postsynaptic potentials in neocortical interneurons exhibit pronounced paired pulse facilitation. Neuroscience 54, 347360 (1993). 2. Thomson, A. M. & Deuchars, J. Synaptic interactions in neocortical local circuits: dual intracellular recordings in vitro. Cereb. Cortex 7, 510522 (1997). 3. Markram, H. & Tsodyks, M. Redistribution of synaptic efficacy between neocortical pyramidal neurons. Nature 382, 807810 (1996). 4. Thomson, A. M. Activity-dependent properties of synaptic transmission at two classes of connections made by rat neocortical pyramid axons in vitro. J. Physiol. (Lond.) 502, 131147 (1997). 5. Buhl, E. H. et al. Effect, number and location of synapses made by single pyramidal cells onto aspiny interneurones of cat visual cortex. J. Physiol. (Lond.) 500, 689713 (1997).
articles
6. Del Castillo, J. & Katz, B. Statistical factors involved in neuromuscular facilitation and depression. J. Physiol. (Lond.) 124, 574585 (1954). 7. Katz, B. & Miledi, R. The role of calcium in neuromuscular facilitation. J. Physiol. (Lond.) 195, 481492 (1968). 8. Rahamimoff, R. A dual effect of calcium ions on neuromuscular facilitation. J. Physiol. (Lond.) 195, 471480 (1968). 9. Betz, W.J. Depression of transmitter release at the neuromuscular junction of the frog. J. Physiol. (Lond.) 206, 629644 (1970). 10. Zucker, R. S. Short-term synaptic plasticity. Annu. Rev. Neurosci. 12, 1331 (1989). 11. Winslow, J. L, Duffy, S. N. & Charlton, M. P. Homosynaptic facilitation of transmitter release in crayfish is not affected by mobile calcium chelators: Implications for the residual ionized calcium hypothesis from electrophysiological and computational analyses. J. Neurophysiol. 72, 17691793 (1994). 12. Atluri, P. P. & Regehr, W. G. Determinants of the time course of facilitation at the granule cell to Purkinje cell synapses. J. Neurosci. 16, 56615671 (1996). 13. Zucker, R. S. Exocytosis: a molecular and physiological perspective. Neuron 17, 10491055 (1996). 14. Frank, E. Matching of facilitation at the neuromuscular junction of the lobster: a possible case for influence of muscle on nerve. J. Physiol. (Lond.) 233, 635658 (1973). 15. Muller, K. J. & Nicholls, J. G. Different properties of synapses between a single sensory neuron and two different motor cells in the leech CNS. J. Physiol. (Lond.) 238, 357369 (1974). 16. Koerber, H. R. & Mendell, L. M. Modulation of synaptic transmission at Iaafferent fiber connections on motoneurons during high-frequency stimulation: Role of postsynaptic target. J. Neurophysiol. 65, 590597 (1991). 17. Davis, G. W. & Murphey, R. K. A role for postsynaptic neurons in determining presynaptic release properties in the cricket CNS: Evidence for retrograde control of facilitation. J. Neurosci. 13, 38273838 (1993). 18. Katz, P. S., Kirk, M. D. & Govind, C. K. Facilitation and depression at different branches of the same motor axon: Evidence for presynaptic differences in release. J. Neurosci. 13, 30753089 (1993). 19. Brodin, L., Shupliakov, O., Pieribone, V. A., Hellgren, J. & Hill, R. H. The reticulospinal glutamate synapse in lamprey: Plasticity and presynaptic variability. J. Neurophysiol. 72, 592604 (1994). 20. Davis, G. W. & Murphey, R. K. Long-term regulation of short-term transmitter release properties: retrograde signaling and synaptic development. Trends Neurosci. 17, 913 (1994). 21. Mennerick, S. & Zorumski, C. F. Paired-pulse modulation of fast excitatory synaptic currents in microcultures of rat hippocampal neurons. J. Physiol. (Lond.) 488, 85101 (1995). 22. Ali, A. B. & Thomson, A. M. Brief train depression and facilitation at pyramid-interneurone connections in slices of rat hippocampus; paired recordings with biocytin filling. J. Physiol. (Lond.) 501, 9P (1997). 23. Ali, A. B. & Thomson, A. M. Facilitating pyramid to horizontal oriensalveus interneurone inputs: dual intracellular recordings in slices of rat hippocampus. J. Physiol. (Lond.) 507, 185199 (1998). 24. Ali, A. B., Deuchars J., Pawelzik H. & Thomson, A. M. CA1 pyramidal to basket and bistratified cell EPSPs: dual intracellular recordings in rat hippocampal slices. J. Physiol. (Lond.) 507, 201217 (1998). 25. Atwood, H. L. & Bittner, G. D. Matching of excitatory and inhibitory inputs to crustacean muscle fibers. J. Neurophysiol. 34, 157170 (1970).
26. Bower, J. M. & Haberly, L. B. Facilitating and nonfacilitating synapses on pyramidal cells: A correlation between physiology and morphology. Proc. Natl. Acad. Sci. USA 83, 11151119 (1986). 27. Stratford, K. J., Tarczy-Hornoch, K., Martin, K. A. C., Bannister, N. J. & Jack, J. J. B. Excitatory synaptic inputs to spiny stellate cells in cat visual cortex. Nature 382, 258261 (1996). 28. Mason, A., Nicoll A. & Stratford, K. Synaptic transmission between individual pyramidal neurons of the rat visual cortex in vitro. J. Neurosci. 11, 7284 (1991). 29. Schrder, R. & Luhmann, H. J. Morphology, electrophysiology and pathophysiology of supragranular neurons in rat primary somatosensory cortex. Eur. J. Neurosci. 9, 163176 (1997). 30. Somogyi, P., Tamas, G., Lujan R. & Buhl, E. H. Salient features of synaptic organization in the cerebral cortex. Brain Res. Rev. 26, 113135 (1998). 31. Kawaguchi, Y. & Kubota, Y. GABAergic cell subtypes and their synaptic connections in rat frontal cortex. Cereb. Cortex, 7, 476486 (1997). 32. Somogyi, P. et al. Different populations of GABAergic neurons in the visual cortex and hippocampus of cat contain somatostatin- or cholecystokinin-immunoreactive material. J. Neurosci. 4, 25902603 (1984). 33. Faber, D. S. & Korn, H. Applicability of the coefficient of variation method for analyzing synaptic plasticity. Biophys. J. 60, 12881294 (1991). 34. Gil, Z., Connors I. W. & Amitai, Y. Differential regulation of neocortical synapses by neuromodulators and activity. Neuron 19, 679686 (1997). 35. Glitsch M., Llano I. & Marty, A. Glutamate as a candidate retrograde messenger at interneurone-Purkinje cell synapses of rat cerebellum. J. Physiol. (Lond.) 497, 531537 (1996). 36. Shigemoto, R. et al. Target-cell-specific concentration of a metabotropic glutamate receptor in the presynaptic active zone. Nature 381, 523525 (1996). 37. Baude, A. et al. The metabotropic glutamate receptor (mGluR1) is concentrated at perisynaptic membrane of neuronal subpopulations as detected by immunogold reaction. Neuron 11, 771787 (1993). 38. Blasco-Ibanez, J. M. & Freund, T. F. Synaptic input of horizontal interneurons in stratum oriens of the hippocampal CA1 subfield: Structural basis of feed-back activation. Eur. J. Neurosci. 7, 21702180 (1995). 39. Han, Z.-H., Buhl, E. H., Lrinczi Z. & Somogyi, P. A high degree of spatial selectivity in the axonal and dendritic domains of physiologically identified local-circuit neurons in the dentate gyrus of the rat hippocampus. Eur. J. Neurosci. 5, 395410 (1993). 40. Maccaferri, G. & McBain, C. J. Passive propagation of LTD to stratum oriens-alveus inhibitory neurons modulates the temporoammonic input to the hippocampal CA1 region. Neuron 15, 137145 (1995). 41. 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). 42. Markram, H., Lbke, J., Frotscher, M., Roth, A. & Sakmann, B. Physiology and anatomy of synaptic connections between thick tufted pyramidal neurones in the developing rat neocortex. J. Physiol. (Lond.) 500, 409440 (1997). 43. Vincent S. R., McIntosh, C. H., Buchan, A. M. & Brown, J. C. Central somatostatin systems revealed with monoclonal antibodies. J. Comp. Neurol. 238, 169186 (1985).
285
articles
Dopaminergic regulation of cerebral cortical microcirculation

Leonid S. Krimer1, E. Christopher Muly, III2, Graham V. Williams1 and Patricia S. Goldman-Rakic1
1
Section of Neurobiology and 2Department of Psychiatry, Yale University School of Medicine, New Haven, Connecticut 06510, USA Correspondence should be addressed to P.S.G.-R. (patricia.goldman-rakic@yale.edu)
Functional variations in cerebral cortical activity are accompanied by local changes in blood flow, but the mechanisms underlying this physiological coupling are not well understood. Here we report that dopamine, a neurotransmitter normally associated with neuromodulatory actions, may directly affect local cortical blood flow. Using light and electron-microscopic immunocytochemistry, we show that dopaminergic axons innervate the intraparenchymal microvessels. We also provide evidence in an in vitro slice preparation that dopamine produces vasomotor responses in the cortical vasculature. These anatomical and physiological observations reveal a previously unknown source of regulation of the microvasculature by dopamine. The findings may be relevant to the mechanisms underlying changes in blood flow observed in circulatory and neuropsychiatric disorders.
A prompt and adjustable response of the cortical microcirculation to local metabolic demands is a prerequisite for supporting cortical functions, which vary from region to region and moment to moment1. The assumption that changes in local neuronal activity are mirrored by changes in regional blood flow and energy metabolism provides the basis for brain-imaging techniques2. The mechanisms regulating the cortical microcirculation are complex and not well understood. Both direct innervation of the vasculature and locally released ions and chemicals (such as H+, K+ and Ca++ ions, adenosine and nitric oxide) have been implicated in this process36. Cortical blood vessels are known to be innervated by axons containing norepinephrine7,8, acetylcholine9,10, serotonin11 and several peptides12. Although dopaminergic fibers have been shown to contact blood vessels in the retina13 and dorsal root ganglia14, central dopamine has not been associated with the cerebral cortical microcirculation, in spite of the high density of dopaminergic afferents in many cortical areas15,16. Here we show that dopaminergic terminals are closely apposed to microvessels within the cortex, and that exogenous dopamine, when iontophoresed onto microvessels, causes them to constrict. These findings reveal a new, vasomotor role for dopamine and possibly a direct involvement of dopamine neurotransmission in the regulation of local cerebral cortical blood flow.
In this report we focus on the dopamine-rich regions of the frontal lobe. Appositions between labeled fibers and microvessels were prominent in layers I and II. Penetrating arterioles and large numbers of putative capillaries (510 m in diameter) in these layers were richly endowed with immunopositive axons stained for the dopamine transporter (Fig. 1). Immunolabeled axons generally approached the microvasculature transversely or obliquely, partially wrapped around them and gave rise to several varicosities closely apposed to blood vessel walls (Fig. 2d and e). A single giant bouton was often observed in association with a blood vessel, and this was usually the largest varicosity on that axon (Fig. 2ac). The terminals were also associated with non-endothelial components of blood vessels (Figs 2f and 3a and b).
Results
DOPAMINERGIC AXONS CONTACT MICROVESSELS

Axons immunopositive for dopamine or the dopamine transporter were readily found in close apposition to blood vessels in the frontal lobe, both in premotor cortex (area 6) and in prefrontal cortex of rhesus monkeys (areas 9 and 46; Figs 1 and 2). This association with the microvasculature was also observed in other brain areas densely populated by dopaminergic afferents, including the entorhinal cortex and the somatosensory cortex. Immunopositive axons on blood vessels were found only occasionally in the primary visual cortex of the occipital lobe, in keeping with the sparse innervation of this area by dopamine fibers16.
286
Fig. 1. Distribution of close appositions of dopamine-transporterlabeled axons on all blood vessels in a section through cortical layers I-IIIa in area 9. The vast majority of blood vessels were contacted by labeled axons at multiple discrete points or crossings (circles). Tracings were made using the Neurolucida program with an oil objective and condenser (NA, 1.4 for each) and DIC optics at a magnification of 1570x. Scale bar, 100 m.
articles
Fig. 2. High-power photographs of dopaminergic terminals associated with small cortical blood vessels. Dopaminergic axons and terminals are labeled with (ac) anti-dopamine or (df) antidopamine transporter antibodies. The terminals partially wrap around the vessel, giving rise to several small varicosities (d, e) or to one giant bouton (ac) closely apposed to the vessel. (f) A string of three terminals is apposed to a nonendothelial vascular cell, whose spindle shape is reminiscent of a smooth muscle cell. Scale bar, 10 m.
ULTRASTRUCTURE OF DOPAMINERGIC VASCULAR TERMINALS

To further characterize the association of dopaminergic axons and blood vessels, four dopaminergic terminals that were closely apposed to the blood vessels at the light-microscopic level were selected for electron microscopy (insets of Fig. 3a and c). Ultrastructural analysis confirmed that immunopositive axons were indeed apposed to the vascular basal lamina either directly (Fig. 3d) or with a thin glial process in-between (Fig. 3b). Also these contacts were invariably apposed to pericytes embedded in the basal lamina of capillaries (Fig. 3b and d). In sections immunostained with antibodies against a norepinephrine-specific markers, dopamine--hydroxylase (DH) and norepinephrine, labeled axons associated with the cortical microvasculature were much less common than in sections stained for dopamine, in accord with previous studies showing a low density of noradrenergic axons in the monkey prefrontal cortex 17 . Conversely, the extraparenchymal leptomeningeal arteries were enmeshed through their entire extent in a dense plexus of noradrenergic fibers stained for tyrosine hydroxylase (TH; Fig. 4a), and only a rare thin fiber, immunopositive for dopamine, c was occasionally observed (Fig. 4b). Most of the TH axons were extraordinarily thick and were devoid of varicosities. This dense axonal meshwork was not observed in the radial vessels descending from the horizontal pial vasculature into the cortical parenchyma. Only the largest arterioles penetrating the supragranular cortical layers still bore some thick noradrenergic axons along their walls. Dopaminergic fibers, in contrast, were never observed accompanying large blood vessels entering the cortex.
by dopamine (19.5 8.7%) to that resulting from application of the saline solution on three different vessels using an unpaired t-test. Both comparisons showed highly significant differences (p < 0.0004 and p < 0.0012, c respectively). Vasoconstriction was often observed within a few seconds of switching current polarity from retaining to ejection and was also observed in vessels that were previously unresponsive to the control solution. The microvessels f recovered completely from the constriction within two to three minutes (Fig. 5). The vasomotor effect of dopamine iontophoresis was repeatable three to five times on the same blood vessels.
Discussion The present anatomical data demonstrate that dopaminergic axons extensively contact the frontal cortical microvasculature. The dopaminergic fibers cross the microvasculature transversely, and the same axons also presumably innervate the neuropil16. These
RESPONSE OF CORTICAL MICROVESSELS TO DOPAMINE

The blood vessels selected for examination in ferret slices were 510 m in diameter. Perivascular application of dopamine, but not of the control solution caused dose-dependent vasoconstriction in 12 out of the 22 microvessels studied (Fig. 5). In 8 of the 12 vessels in which images were stored, we compared diameters of these vessels before and after dopamine application using a paired t-test. We also compared the change in average diameter produced
Fig. 3. Correlated light- and electron-microscopic analysis of dopamine terminals associated with the microvasculature. (a, b) A giant dopaminergic bouton (arrowheads), which is greater than two microns in diameter, was associated with the capillary through over fifty serial sections. It is adjacent to the basal lamina (b) of the blood vessel with only a thin glial process interposed. A pericyte (P) is associated with the vessel and one of its processes is beneath the basal lamina across from the dopamine terminal. (c, d) A string of three terminals adjacent to a vessel were also examined, and one of these is depicted in the inset photograph and micrographs (arrows). This terminal is directly apposed to the basal lamina of the capillary (between the open arrows). Once again, pericyte processes (P) are identifiable beneath the basal lamina. Scale bars, (a) and (c), 2 m; (b) and (d), 400 nm.
287
articles
Fig. 4. Catecholaminergic innervation of pial arteries. Axons immunopositive for (a) tyrosine hydroxylase (TH) or (b) dopamine that are associated with pial arteries. Note a dense, continuous meshwork of coarse TH-labeled fibers (presumably noradrenergic) surrounding an artery (a) and few, thin dopamine-labeled axons associated with another artery (b). Scale bar, 10 m.
fibers are strategically positioned to adjust both the microcirculation and neuronal activity in a discrete cortical region18. It is of interest that the dopaminergic appositions occurred not only on penetrating cortical arterioles but also on capillaries. For instance, all four microvessels that we studied with electron microscopy were capillaries. This observation also holds for other well known central vasoactive neurotransmitters19,20 and probably reflects the prevalence of the capillaries in the cerebral cortex. In our electron microscopic preparations, we always found a pericyte in close proximity to the dopaminergic terminal across the vascular basal lamina. These cells possess contractile properties21, and pericytes of central origin have been shown to modulate capillary diameters. They also respond to endothelial-derived vasoactive substances, as do vascular smooth muscle cells22. These new data on pericyte physiology suggest the possibility that dopaminergic vasomotor effects may also take place at the level of the cortical capillary. As central norepinephrine has traditionally been considered the major neurotransmitter controlling cortical blood flow7,8, we
compared its distribution on the cortical microvasculature to that of the dopaminergic innervation. Surprisingly, the innervation of the intracortical blood vessels seemed to be predominantly dopaminergic, in contrast to the extraparenchymal cerebral vasculature, which is more extensively innervated by norepinephrine. Also, in the extraparenchymal blood vessels, noradrenergic axons ran along in a dense meshwork, suggesting a global neural control of these blood vessels by the superior cervical ganglia, where these extraparenchymal axons originate8. To determine whether the dopaminergic vascular contacts were functional, we examined the vasomotor responses of small blood vessels in an in vitro cortical slice preparation using infrared DIC (differential interference contrast) video microscopy. This approach has been used successfully to assess the vasomotor effect of drugs23,24. For these experiments, we selected blood vessels with lumens of 510 m across. Considering that vessels in our in vitro preparation were possibly partially collapsed and therefore smaller than their actual diameter in vivo, the selected vessels might include arterioles as well as capillaries. We were able to demonstrate that perivascular application of dopamine results in a 19.5% vasoconstriction. According to Poiseuilles law, an equivalent change in vivo would result in a 58% reduction in blood flow. The observed vascular effects of dopamine are likely to be a direct effect of the neurotransmitter on the blood vessels rather than being secondary consequences of effects on neuronal activity. We base this conclusion on the following facts. Cortical neurons in slice preparations do not generate action potentials, either spontaneously or after application of dopamine. Dopamine alone applied iontophoretically has little or no effect on neuronal membrane potential or input conductance in cortical slice preparations25,26. These physiological results, together with the anatomical observations, strongly indicate that dopamine neurotransmission may play a role in the regulation of local cortical blood flow. Although our in vitro observations cannot be directly extrapolated to the in vivo condition, our findings are in agreement with the observation that local application of dopamine induces a 20%
Before dopamine
Initial dopamine effect
Maximal dopamine effect
Recovery
Fig. 5. DIC videomicroscopic images of cortical microvessels responding to perivascular iontophoretic application of dopamine with +10 to +60 nA current in ferret prefrontal cortical slices. The vessels are possibly small arterioles because their walls contain structures reminiscent of vascular myocytes. Application of dopamine was initiated at time 00:00. The initial contractile response was observed within 1840 s. The maximal contractile effect reached 1824%. Dopamine iontophoresis was continued for 60 s and followed by a recovery period while dopamine was retained with -4 nA negative current (fourth panel in the figure). Scale bar, 10 m.
articles
reduction in vascular diameter of pial arterioles in living cats27. These findings may be significant for the pathology of both hemodynamic and neuropsychiatric disorders, such as schizophrenia, in which changes in regional cortical blood flow are observed28,29.
Methods HISTOLOGY AND IMMUNOHISTOCHEMISTRY. Three young adult rhesus monkeys were intracardially perfused with fixative solution containing 4% paraformaldehyde, 0.2% glutaraldehyde and 15% picric acid. Brains were removed, blocked coronally and postfixed for 2 h. Prefrontal cortical areas 6, 9 and 46 were sectioned on a vibrotome into 40 m coronal sections. In addition, the medial portion of areas 6 and 9 was sectioned tangentially. Sections were pretreated in 5% goat blocking serum containing 0.4% Triton X-100 for 1 h and then incubated for 48 h with one of the following antisera: anti-TH mouse IgG (Chemicon) diluted 1:1,000, anti-dopamine transporter rat IgG (Chemicon) diluted 1:10,000 or anti-DH rabbit IgG (Eugene) diluted 1:10,000. The sections were then rinsed and incubated for 2 h with biotinylated anti-mouse, anti-rat or anti-rabbit IgG, respectively, diluted 1:200 in the blocking serum. The reaction was developed using the avidin-biotinperoxidase complex (Vector) and nickel-intensified DAB as a chromogen. To prevent excessive shrinkage of the sections in the z-axis, they were dehydrated while free-floating. This procedure helped to preserve the lumen of small blood vessels, improving their visibility. In addition, we used available sections stained with monoclonal anti-dopamine- or antinorepinephrine rabbit IgG (provided by M. Geffard) from three additional monkeys, which had been perfused with 5% glutaraldehyde fixative. Sections were analyzed under the microscope at 1570x using oil objective and condenser (NA, 1.4 for each) and DIC optics to facilitate visualization of unstained blood vessels and their related cells.

1. Cox, S. B., Woolsey, T. A. & Rovainen, C. M. Localized dynamic changes in cortical blood flow with whisker stimulation corresponds to matched vascular and neuronal architecture of rat barrels. J. Cereb. Blood Flow Metab. 13, 899913 (1993). 2. Magistretti, P. J. & Pellerin, L. The cellular bases of functional brain imaging: evidence for astrocyte-neuron metabolic coupling. Neuroscientist 3, 361365 (1997). 3. Lou, H. C., Edvinsson, L. & MacKenzie, E. T. The concept of coupling blood flow to brain function: revision required? Ann. Neurol. 22, 289297 (1987). 4. Iadecola, C. Regulation of the cerebral microcirculation during neural activity: is nitric oxide the missing link? Trends Neurosci. 16, 206214 (1993). 5. Wahl, M. & Schilling, L. Regulation of cerebral blood flow a brief review. Acta Neurochir. (Suppl.) 59, 310 (1993). 6. MacKenzie, E. T. & Scatton, B. Cerebral circulation and metabolic effects of perivascular neurotransmitters. Crit. Rev. Clin. Neurobiol. 2, 357419 (1987). 7. Hartman, B. K., Zide, D. & Udenfriend, S. The use of dopamine hydroxylase as a marker for the central noradrenergic nervous system in rat brain. Proc. Natl. Acad. Sci. USA 69, 27222726 (1972). 8. Raichle, M. E., Hartman, B. K., Eichling, J. O. & Sharpe, L. G. Central noradrenergic regulation of cerebral blood flow and vascular permeability. Proc. Natl. Acad. Sci. USA 72, 37263730 (1975). 9. Sato, A. & Sato, Y. Regulation of regional cerebral blood flow by cholinergic fibers originating in the basal forebrain. Neurosci. Res. 14, 242274 (1992). 10. Vaucher, E. & Hamel, E. Cholinergic basal forebrain neurons project to cortical microvessels in the rat: electron microscopic study with anterogradely transported Phaseolus vulgaris leucoagglutinin and choline acetyltransferase. J. Neurosci. 15, 74277441 (1995). 11. Reinhard, J. F., Liebmann, J. E., Schlosbery, A. J. & Moskowitz, M. A. Serotonin neurons project to small blood vessels in the brain. Science 206, 8587 (1979). 12. Dacey, R. G., Jr., Bassett, J. E. & Takayasu, M. Vasomotor responses of rat intracerebral arterioles to vasoactive intestinal peptide, substance P, neuropeptide Y, and bradykinin. J. Cereb. Blood Flow Metab. 8, 254261 (1988). 13. Favard, C., Simon, A., Vigny, A. & Nguyen-Legros, J. Ultrastructural evidence for a close relationship between dopamine cell processes and blood capillary walls in Macaca monkey and rat retina. Brain Res. 523, 127133 (1990). 14. Weil-Fugazza, J., Onteniente, B., Audet, G. & Philippe, E. Dopamine as trace amine in the dorsal root ganglia. Neurochem. Res. 18, 965969 (1993). 15. Goldman-Rakic, P. S., Lidow, M. S., Smiley, J. F. & Williams, M. S. The anatomy of dopamine in monkey and human prefrontal cortex. J. Neural Transm. (Suppl.) 36, 163177 (1992). 16. Williams, S. M. & Goldman-Rakic, P. S. Characterization of the dopaminergic innervation of the primate frontal cortex using a dopaminespecific antibody. Cereb. Cortex 3, 199222 (1993). 17. Morrison, J. H., Foote, S. L., OConnor, D. & Bloom, F. E. Laminar, tangential and regional organization of the noradrenergic innervation of monkey cortex: dopamine-beta-hydroxylase immunohistochemistry. Brain Res. Bull. 9, 309319 (1982). 18. Williams, G. V. & Goldman-Rakic, P. S. Modulation of memory fields by dopamine D1 receptors in prefrontal cortex. Nature 376, 572575 (1995). 19. Cohen, Z., Ehret, M., Maitre, M. & Hamel, E. Ultrastructural analysis of tryptophan hydroxylase immunoreactive nerve terminals in the rat cerebral cortex and hippocampus: their associations with local blood vessels. Neuroscience 66, 555569 (1995). 20. Cohen, Z., Molinatti, G. & Hamel, E. Astroglial and vascular interactions of noradrenaline terminals in the rat cerebral cortex. J. Cereb. Blood Flow Metab. 17, 894904 (1997). 21. Shepro, D. & Morel, N. M. L. Pericyte physiology. FASEB J. 7, 10311038 (1993). 22. Haefliger, I. O., Zschauer, A. & Anderson, D. R. Relaxation of retinal pericyte contractile tone through the nitric oxide-cyclic guanosine monophosphate pathway. Invest. Opthalmol. Vis. Sci. 35, 991997 (1994). 23. Sagher, O. et al. Live computerized videomicroscopy of cerebral microvessels in brain slices. J. Cereb. Blood Flow Metab. 13, 676682 (1993). 24. Farber, N. E. et al. Region-specific and agent-specific dilation of intracerebral microvessels by volatile anesthetics in rat brain slices. Anesthesiology 87, 11911198 (1997). 25. Cepeda, C., Radisavljevic, Z., Peacock, W., Levine, M. S. & Buchwald, N. A. Differential modulation by dopamine of responses evoked by excitatory amino acids in human cortex. Synapse 11, 330341 (1992). 26. Geijo-Barrientos, E. & Pastore, C. The effects of dopamine on the subthreshold electrophysiological responses of rat prefrontal cortex neurons in vitro. Eur. J. Neurosci. 7, 358366 (1995). 27. Edvinsson, L., McCulloch, J. & Sharkey, J. Vasomotor responses of cerebral arterioles in situ to putative dopamine receptor agonists. Br. J. Pharmacol. 85, 403410 (1985). 28. Goldberg, L. I. Dopamine receptors and hypertension. Physiologic and pharmacologic implications. Am. J. Med. 77, 3744 (1984). 29. Weinberger, D. R., Berman, K. F. & Illowsky, B. P. Physiological dysfunction of dorsolateral prefrontal cortex in schizophrenia. III. A new cohort and evidence for a monoaminergic mechanism. Arch. Gen. Psychiatry 45, 609615 (1988).
ELECTRON MICROSCOPY. Sections stained with the anti-dopamine antibodies without the addition of Triton X-100, postfixed in 1% osmium tetroxide and flat imbedded in Durcupan resin were first analyzed with the light microscope. Four immunopositive boutons, closely apposed to small blood vessels were selected, traced and photographed. They were then serially sectioned, counterstained with uranyl acetate/lead citrate and analyzed with a JEOL 1010 electron microscope. IN VITRO CORTICAL SLICE PHYSIOLOGY. Young adult ferrets were anesthetized deeply with sodium pentobarbital and decapitated. Coronal slices (400 m thick) through the medial prefrontal cortex were incubated at 32oC in oxygenated (95% O2, 5%CO2) Ringer solution containing, in mM, 125 NaCl, 2.5KCl, 1.25NaH2PO4, 26NaHCO3, 1.5 CaCl2, 1.5 MgSO4 and 10 glucose. They were submerged and studied with an Axioscop FS microscope. Twenty-two small blood vessels (~47 m across) in layers I-III of medial prefrontal cortex were selected for analysis. They were visualized with the DIC infrared video-microscopy employing a 63x water-immersed lens and an additional 4x TV tube. Glass micropipettes (tip diameter ~ 2/3 m) containing 200 mM of dopamine hydrochloride (Research Biochemicals Intl.), in dH2O with 0.1% ascorbic acid, 10 mM Hepes buffer and adjusted to pH 4.5 with 1N NaOH, or saline solution (152 mM NaCl added to the same diluent) were positioned 25 m above or to one side of the small vessels. The drug was delivered by passing 10 to 60 nA positive current or otherwise retained for control purposes inside the glass pipette by 4 to 5 nA negative current. Data were stored in real time on a VCR and later acquired on computer using Adobe Photoshop 4.0 software. Data were further analyzed with Canvas 5 software. All animals were used in this study in accordance with all appropriate regulatory guidelines.
Acknowledgements
We thank Klara Szigeti for preparing tissue for electron microscopy, and David McCormick for providing us with ferret frontal cortex for the physiological experiments. This work was supported by MH44866 and a Pfizer Postdoctoral Fellowship to ECM.
289
articles
Transplantation of expanded mesencephalic precursors leads to recovery in parkinsonian rats

Lorenz Studer1, Viviane Tabar1,2 and Ron D.G. McKay1
1 2
Laboratory of Molecular Biology, NINDS, NIH, 36 Convent Drive, Building 36, Room 5A29, Bethesda, Maryland 20892, USA University of Massachusetts, Division of Neurosurgery, Worcester, Massachusetts 01655, USA Correspondence should be addressed to R.D.G.M.(mckay@codon.nih.gov)
In vitro expansion of central nervous system (CNS) precursors might overcome the limited availability of dopaminergic neurons in transplantation for Parkinsons disease, but generating dopaminergic neurons from in vitro dividing precursors has proven difficult. Here a threedimensional cell differentiation system was used to convert precursor cells derived from E12 rat ventral mesencephalon into dopaminergic neurons. We demonstrate that CNS precursor cell populations expanded in vitro can efficiently differentiate into dopaminergic neurons, survive intrastriatal transplantation and induce functional recovery in hemiparkinsonian rats. The numerical expansion of primary CNS precursor cells is a new approach that could improve both the ethical and the technical outlook for the use of human fetal tissue in clinical transplantation.
Parkinsons disease is a neurodegenerative disorder affecting an estimated one million patients in the United States alone. Several strategies are being pursued to develop new therapies for Parkinsonian patients. These techniques range from the use of dopaminotrophic factors1 and viral vectors2 to the transplantation of primary xenogeneic tissue3. Fetal nigral transplantation is a clinically promising experimental treatment in late stage Parkinsons disease (PD). More than two hundred patients have received transplants worldwide4. Clinical improvement has been confirmed by functional studies using positron emission tomography of striatal fluorodopa uptake after transplantation4,5 and was correlated to good graft survival and innervation of the host striatum in postmortem studies of transplanted patients6. Two controlled clinical studies, sponsored by the US National Institutes of Health, are ongoing in the United States using fetal mesencephalic tissue in a larger number of patients and a large multicenter study is being planned in Europe. In spite of these promising findings, neural transplantation remains a controversial procedure. Successful numerical expansion of primary CNS precursors could alleviate some of the ethical and technical difficulties involved in the use of human fetal tissue. There has been considerable recent progress in characterizing stem cells from the central and peripheral nervous systems in vitro and in vivo711. However, controlled conversion of neuroepithelial precursors into dopaminergic neurons has not been reported. Here we demonstrate that precursor cells obtained from the rat embryonic day 12 (E12) ventral mesencephalon can be expanded in vitro and differentiated into dopaminergic neurons. Upon transplantation, differentiated precursors alleviate behavioral deficits in a rat model of Parkinsons disease.
rat embryos, mechanically dissociated, plated on polyornithine/fibronectin-coated culture dishes and grown in serumfree medium supplemented with basic fibroblast growth factor (bFGF). No cells immunoreactive for the rate-limiting enzyme in the synthesis of dopamine, tyrosine hydroxylase (TH), could be detected in the tissue at the time of dissection. In response to bFGF, cells proliferated and formed cell clusters that grew to macroscopic size after one week in vitro (Fig. 2a). Within seven days in vitro, the total cell number increased from 800,000 to 7,417,000 1,066,000 cells, (n = 12, Fig. 2b). Cultures grown without bFGF had a substantial net cell loss. Cells not responding to the mitogenic activity of bFGF, as assessed by time-lapse photomicrography, were prone to cell death (Fig. 2d and e), determined by the terminal deoxynucleotidyl transferase-mediated deoxyuridine triphosphate nick end label (TUNEL) method. Cell death within proliferating clusters was relatively rare (Fig. 2f). Immunohistochemistry showed that bFGF receptors were present in most cells (88 4%) throughout expansion. After six to eight days of bFGF treatment, all surviving cells were growing in clusters and were immunoreactive for the nestin intermediate filament protein, a marker for immature neuroepithelial precursors12,13 (Fig. 2c).
CELL DIFFERENTIATION
Removal of the mitogen initiates differentiation of nestinimmunoreactive precursor cells in vitro7. Upon removal of bFGF from the culture medium, extensive arborizations developed, interconnecting large clusters of cell bodies. Cell bodies and fiber bundles were strongly immunoreactive for -tubulin type III. After seven days of differentiation, neuronal morphologies were more mature, and 18.4 5.1% of the total cell population was immunoreactive for TH (n = 19, Fig. 3a). The TH-immunoreactive cells were also immunoreactive for dopamine (Fig. 3a, inset), dopamine transporter and -tubulin type III (data not
Results
CELL EXPANSION
An overview of the experimental procedure is shown in Fig. 1. Cells were obtained from the ventral mesencephalon of E12
290
articles
Fig. 1. Schematic illustration of the experimental procedures. Phase 1 (cell expansion), Following tissue dissection and trituration to single-cell density, precursor cells were expanded for one week in serum-free medium supplemented with bFGF and grown on precoated culture dishes. Phase 2 (cell differentiation), Conversion of dividing precursors into dopaminergic neurons was induced by withdrawal of the mitogen. Two cell differentiation systems have been developed. Cultures growing under serum-free conditions and attached to culture plate are differentiated by bFGF withdrawal only (right box). Cultures raised for transplantation purposes were differentiated as free-floating aggregates in a medium containing 10% fetal bovine serum (FBS) and maintained in a roller drum system (left box). Phase 3 (cell transplantation), Differentiated aggregates were transplanted into the ipsilateral striatum of 6-hydroxydopamine-lesioned rats. Graft function was assessed as changes in amphetamine-induced rotation behavior.
shown). In non-expanded cultures, grown without bFGF (Fig. 2b), 5.6 1.3% of cells were TH immunoreactive. Thus expansion and differentiation of precursors increased by threefold the percentage of neurons that were dopaminergic (18.4%), compared to non-expanded cultures (5.6%). This number, multiplied with a tenfold increase in total cell number during the expansion phase, leads to an estimated 30-fold increase in the in vitro yield of nigral dopaminergic neurons. One E12 embryo yielded 100,000120,000 nigral cells at the day of dissection. After these cells were expanded tenfold, 18.4% of them stained positive for TH after cell differentiation, yielding 180,000220,000 TH-immunoreactive cells. When compared to the total number of dopaminergic neurons present in the adult rat
(26,00030,000)14,15, the expansion procedure leads to an estimated sevenfold increase in TH-immunoreactive cell number. Transplantation of cells that are differentiated on culture dishes entails mechanical or enzymatic dissociation, with subsequent disruption of axodendritic trees and cell loss. Therefore we developed a reaggregation system that allows conversion of previously expanded precursors into dopaminergic neurons in free-floating spheres (Fig. 3b). These reaggregates can be directly loaded into a stereotactic needle and transplanted as a unit. Under differentiation conditions, reaggregate size increased over six to eight days, resulting in a spherical culture of 0.61.2 mm in diameter. No substantial cell proliferation occurred during the differentiation phase, as shown by immunohistochemistry for
Day 1
Day 3
Day 5
Day 7
Percentage of TUNEL+ cells
x-fold numeric expansion
Days in vitro
Days in vitro
Fig. 2. Expansion of mesencephalic precursors in vitro. (a) Time-lapse, phase-contrast photomicrography during bFGF expansion. (b) Comparison of growth rate between bFGF treated and control cultures. bFGF, control. (c) Expanded cells are immunoreactive for nestin (seven days of bFGF treatment). (d, e) TUNEL-positive nuclei (arrowheads) in clusters of dividing cells (d) and in single cells (e) after two days of bFGF treatment. (f) Single cells demonstrate a gradual increase in TUNEL labeling, reaching 96 2% TUNEL-positive cells at the end of expansion phase. Cell clusters had only a small percentage of TUNEL-positive cells at the end of the expansion phase. cells in clusters, single cells. Scale bars, 100 m (a, c); 10 m (d, e).
nature neuroscience volume 1 no 4 august 1998 291
articles
Fig. 3. Differentiated mesencephalic precursors in vitro. (a) TH-immunoreactive neurons seven days after initiation of cell differentiation. Small groups of TH-immunoreactive cells (white arrow) migrated out of the large cluster. Inset, differentiated precursors are also immunoreactive for dopamine. (b) TH-immunoreactive neurons in free-floating aggregates.(ce) Confocal images of differentiated dopaminergic neurons double labeled for TH (c, green label, and d) and BrdU (c, red label, and e). Three typical doublelabeled cells are marked with arrows in (ce). Scale bars, 100 m (a, b); 10 m (ce). (f) Representative HPLC chromatogram of the conditioned supernatant of differentiated precursors exposed to HBSS for 25 minutes (blue curve) or exposed to 56 mM KCl in HBSS for the same time period f (red curve). The retention time of dopamine was 2.7 minutes. (g) Quantification of the dopamine levels in the supernatant under basal conditions (HBSS only, blue column) or under stimulation with KCl (red column). Data are mean standard error of three independent culture series.
Retention time (min)
the proliferation marker Ki-67, which confirmed that only a small fraction (1.2 0.9%, n = 10) of the cells were mitotically active after seven days of differentiation. Supplementation of the medium with 10% fetal bovine serum (FBS) led to a dramatic increase of TH-immunoreactive neurons in differentiated reaggregates with 6178 547 TH-immunoreactive neurons per sphere, as compared to 1341 341 in cultures grown without FBS (p < 0.01). The average volume of serumtreated reaggregates, was 0.77 0.11 mm3, resulting in a mean density of 10,749 2386 TH-immunoreactive cells per cubic mm. The percentage of TH-immunoreactive cells was 14.1 4.2% in serum-treated reaggregates and 2.9 1.3% in the control group. (n = 21 for FBS-treated group; n = 11 for control group) . Treatment with GDNF (10 ng per ml), BDNF (10 ng per ml), NT4/5 (10 ng per ml) or SHH (2.5 g per ml) did not significantly increase the total number of TH-immunoreactive cells per sphere in serum-supplemented medium (data not shown). However, under serum-free conditions, GDNF did significantly improve TH-immunoreactive cell yield as compared to control cultures (GDNF 3451 621, n = 8; control group 1341 341, n = 11; p < 0.05). Because no growth factor could substantially exceed the effect of FBS, cultures produced for transplantation purposes were raised in 10% FBS without any additional growth factors. Other neuronal phenotypes present in 10% FBS differentiated reaggregates were, in descending order of frequency, GABAergic (1015%), serotonergic (35%) and cholinergic (< 0.1%) neurons. The percentage of glial cells in differentiated reaggregates was low. In serum-treated cultures, 2.8% 1.2% (n = 9) of the total cell population was immunoreactive for glial
292
fibrillary acid protein (GFAP), an astrocytic marker. The percentage of astrocytes in serum-free cultures was even lower (1.1 0.9%, n = 9). No mature galactocerebroside-immunoreactive oligodendrocytes could be detected in differentiated reaggregates under any of the conditions tested. b Cultures were exposed to 1 M BrdU throughout the expansion phase and analyzed by confocal microscopy with z-axis sectioning for TH and BrdU double labeling after differentiation. e There was no significant difference in BrdU labeling between TH-immunoreacg tive cells differentiated in reaggregate cultures and THimmunoreactive cells differentiated attached to the culture plate. An average of 41 14% (n = 12) of all dopaminergic cells had clearly incorporated BrdU Treatment (Fig. 3c). The percentage of BrdU labeling in the total cell population was 47 21% (n = 12). This is a minimum estimate, as higher BrdU concentrations (10 M) led to increased labeling rates but dramatically reduced overall cell viability (data not shown). Reverse-phase high-performance liquid chromatography (HPLC) with electrochemical detection demonstrated that one milliliter of medium collected from expanded precursors after five to seven days of differentiation contained between 80 and 250 picograms of dopamine. Similar dopamine levels were detected in the supernatant of differentiated cultures maintained in Hanks balanced buffer solution (HBSS) for 25 minutes (162 54 pg dopamine per ml HBSS, n = 3) (Fig. 3f and g). Evoked release by exposing cultures to 56 mM KCl in HBSS for 25 minutes led to a three- to fourfold increase in dopamine levels in the supernatant (572 145 pg dopamine per ml, n = 3). (Fig. 3f and g). These data demonstrate that dividing precursor populations efficiently generate dopaminergic neurons.
RESULTS IN VIVO
In vivo function of reaggregate cultures was tested by transplantation into the ipsilateral striatum of adult hemiparkinsonian rats (n = 7). All animals had been previously lesioned by unilateral injections of 6-hydroxydopamine into the medial forebrain bundle and displayed a stable rotation response to amphetamine (average 11 2 rotations per min). Gradual behavioral recovery was observed in transplanted animals (Fig. 4c). Each animal had received six to seven reaggregates, corresponding to a total number of 34 105 grafted cells. Eighty days after transplantation, rotation scores had improved substantially in five of seven animals. Rotation scores for these five animals were reduced 75% on average compared to pretransplantation scores (range 47 to
Dopamine (pg/ml)
Amplitude (mAU)
articles
ed for therapeutic grafting. Second, the culture period provides an opportunity to genetically manipulate the cultured cells to optimize neuronal survival21,22 and fiber outgrowth1 after grafting. Third, the controlled in vitro generation of dopaminergic neurons is an ideal tool to investigate the function of new genes recently implicated in the differentiation of dopaminergic neurons2330. Although non-expanded precursors transplanted into the Days post transplantation developing brain have shown an Fig. 4. In vivo results. (a) TH-immunoreactive graft in adult hemiparkinsonian rat 80 days after trans- extraordinary plasticity in neuplantation. (b) Differentiated dopaminergic neurons in graft. (c) Time course of amphetamine- ronal differentiation3134, in vitro induced rotation response. Data are given as mean change in rotation scores for each animal as expanded cells have been much compared to pretransplantation values (pretransplantation scores were 11 2 rotations per min in less successful in generating neuboth groups; n = 7, transplanted group; n = 5, control group). Significant differences between trans- rons, especially when grafted into planted and control animals could be detected at all three time points after transplantation (p < 0.05). the adult CNS35,36. We therefore Scale bars, 200 m (a); 10 m (b). differentiated precursor cells in vitro to obtain stable neuronal phenotypes prior to transplantation. Work with primary fetal mesencephalic tissue suggests that behavioral recovery is direct97%) . The remaining two animals showed only mild improvely correlated to the number of surviving dopaminergic neurons in ment (13% and 22% reduction). Control animals showed no the host striatum37. In this study, we did not use a large enough behavioral improvement. Immunohistochemical analysis of the substantia nigra, vennumber of animals to conclusively address this issue for expandtral tegmental area and striatum confirmed that all animals ed precursors. Although unlikely, non-dopaminergic effects such included in this study had a complete lesion. All animals had as the secretion of neurotrophic factors by the transplanted cells viable grafts 80 to 101 days after transplantation. The numcould have partly contributed to the behavioral recovery. Here, ber of surviving TH-immunoreactive cells per graft was the survival rate of grafted dopaminergic neurons (35%) derived 1221 431, and the average graft volume was 0.92 0.12 from expanded precursors is similar to those reported for primm3, resulting in a TH-immunoreactive cell density of 1369 mary nigral cells transplanted into the striatum of rodents or Parkinsons patients4. The successful grafting results demonstrate 389 cells per mm 3. Graft size and TH-immunoreactive cell density in this study were similar to those found after transthat long-term survival and functional integration into the adult planting primary fetal mesencephalic tissue 16. A graft with CNS can be achieved with expanded and predifferentiated precursors. The design of similar in vitro systems for other neuronal only 478 surviving TH-immunoreactive cells and a cell denphenotypes could substantially widen the use of CNS precursor sity more than threefold lower than average was found in one cells in neurodegenerative diseases. of the two animals without significant behavioral recovery. Neurotrophic factors have been characterized that protect The other animal with only mild behavioral improvement had dopaminergic neurons in animal models for Parkinsons disan average number of surviving TH-immunoreactive cells, but ease and promote the survival and fiber outgrowth of transthe graft was located ectopically. Cells in all of the grafts displanted fetal dopaminergic neurons 1,2,21,22 . The efficient played morphological features of mature dopaminergic neurons and a strong immunoreactivity for TH (Fig. 4b). delivery of such therapeutic proteins to the CNS is a major Moderate fiber outgrowth from the graft into the host brain focus in gene therapy. Genetically modified neuroepithelial was found (Fig. 4a), comparable to results from grafted pricell lines integrate into the CNS more efficiently than fibrobmary fetal mesencephalic cell cultures16. These results show lasts, and these cells can express functional gene products in vivo 38,39 . However, precursors to dopaminergic neurons, that grafted dopaminergic neurons derived from expanded precursors are functional in vivo and can alleviate behavioral manipulated to express therapeutic levels of growth factors, deficits in an animal model for Parkinsons disease. could uniquely combine the potential of genetic manipulation and neuronal function in the grafted cells. There is a considerable effort to characterize the molecular Discussion steps involved in the neuronal specification of CNS stem Cell grafting in Parkinsonian models has used primary mesencells7,40,41. We anticipate that the growing understanding of these cephalic cells4,17 or genetically modified cells that do not differentiate into neurons in the host striatum1820. Here we present mechanisms at a molecular level will lead to further improvements in the culture system reported here. However, CNS neuthe first evidence that dopaminergic neurons derived from prerons derived from stem cells have not been analyzed on a cursors expanded in vitro function in vivo. There are three main functional level in vivo. The present study demonstrates that advantages of using expanded precursors in the treatment of dopaminergic neurons can efficiently differentiate from expandParkinsons disease. First, the expansion of cell number in the ed precursors and modify behavior after grafting. The expansion proliferative phase and the enrichment in dopaminergic neurons and differentiation steps lead to an estimated thirtyfold increase in the differentiation step can reduce the amount of tissue neednature neuroscience volume 1 no 4 august 1998
Change in rotation behavior (%)
293
articles
in the number of dopaminergic neurons over primary nonexpanded control cultures and to an estimated sevenfold increase over the number of nigral TH-immunoreactive neurons present in the adult rat brain14,15. In humans, the phase of cell birth is protracted (6 to 8.5 weeks after conception) compared to the period of dopaminergic neuronal differentiation in the rat (E12 to E15), and this might offer a longer period for cell expansion in vitro. The success in expanding dopaminergic precursors in rodents suggests that a similar approach should now be attempted with primate mesencephalic material and with precursors of other neuronal subpopulations. There was no indication of uncontrolled growth in any of the animals transplanted or in over two hundred culture series tested. However, larger cohorts of animals and long-term observations beyond three months after transplantation will be needed to conclusively assess the safety of expanded precursors in neural transplantation.
1998 Nature America Inc. http://neurosci.nature.com Methods TISSUE CULTURE AND IMMUNOHISTOCHEMISTRY. Tissue was obtained from the ventral mesencephalon of E12 rat embryos (Sprague Dawley, plug day is day 0), as described42. Tissue pieces were centrifuged and mechanically triturated to quasi-single-cell suspension. 150200 x 103 cells per ml were plated at a concentration of 5 ml of cell suspension on a 10 cm dish precoated with polyornithine (15 g per ml) and fibronectin (1 g per ml). The medium, consisting of DMEM and F12 (Gibco) with N2 supplement, was changed every other day and bFGF (10 ng per ml) added daily. At the end of the expansion phase (68 days), dishes were incubated with HBSS, and cells were mechanically removed and resuspended to quasi-single-cell suspension. 200,000 cells per ml were distributed in 15 ml Falcon tubes (1 ml per tube) and placed in a roller drum (Bellco). This procedure typically resulted in the formation of one large (0.61.2 mm in diameter) sphere per tube after seven days of differentiation. The differentiation medium for reaggregate cultures consisted of Neurobasal with 2% B27, with or without FBS (10%, Gibco), GDNF 10 ng per ml, BDNF 10 ng per ml, NT4/5 10 ng per ml (Peprotech) or SHH 2.5 g per ml (kindly provided by Dr. Thomas Mller), depending on the individual experiments. BrdU (1 M, Boehringer-Mannheim) was added to the medium at days 1, 3 and 5 of the expansion phase. Cells were marked after plating with a 3 mm circle (Nikon) on the bottom of the plate7 and followed by phase-contrast microscopy and daily photomicrographs. For TUNEL labeling, cells were fixed after 7 hours, 1 day, 2 days, 4 days or 8 days in 4% paraformaldehyde and 0.15% picric acid and labeled according to the specifications of the manufacturer (Boehringer-Mannheim). All negative controls (no deoxynucleotidyl transferase in the reaction mixture) were devoid of any labeling. Reaggregates were fixed, equilibrated in 30% sucrose, cut into 20-m sections on a freezing microtome (Microm HM500) and adhered to a gelatineprecoated glass carrier. Standard immunohistochemical procedures were followed. The following antibodies were used: TH polyclonal 1:500 (Pel Freeze), dopamine polyclonal 1:500 (Chemicon), bFGF receptor monoclonal 1:100 (Sigma), -tubulin type III (TuJ1) monoclonal 1:500 (Berkley Antibody Company), GABA polyclonal 1:2000 (SIGMA), serotonin polyclonal 1:8000 (SIGMA), ChAT polyclonal 1:500 (Chemicon), GFAP polyclonal 1:100 (Chemicon), galactocerebroside monoclonal 1:50 (Boehringer-Mannheim), TH monoclonal 1:10000 (Sigma), nestin #130 polyclonal 1:500 (M. Marvin and R.D.G.M.) and Ki-67 polyclonal 1:1000 (Novocastra). Dopamine staining was carried out on cultures fixed in 5% glutaraldehyde and 1% metabisulfite in Tris buffer and then kept in metabisulfite-containing solutions prior to incubation in the secondary antibody. BrdU immunohistochemistry was performed after postfixation in 95% ethanol and 5% glacial acetic acid and incubation in primary antibodies and nucleases (Amersham Life Science). Appropriate FITC- and LRSC-labeled secondary antibodies were used for double immunohistochemistry for TH and BrdU.
immediately stabilized by adding 88 l of 85% orthophosphoric acid and 4.4 mg of metabisulfite to each sample ml. Additional cultures of the same series were placed in HBSS for 25 minutes at 37C either under basal conditions (HBSS only) or under stimulation with 56 mM KCl. The supernatant was collected and stabilized as above. Dopamine was extracted by aluminium adsorption as described43. Separation of the injected samples (20 l) was achieved by isocratic elution on a HewlettPackard Series 1050 HPLC system with a reverse-phase C18 column (3 m particle size, 80 x 4.6mm dimension, ESA, Inc.) in a commercially available MD-TM mobile phase (ESA Inc.). The flow rate was set at 1 ml per min, resulting in a working pressure of 100 bar and an elution time of 2.7 minutes for dopamine. The oxidative potential of the analytical cell (ESA Inc. Mod. 5011) was set at +325 mV. Results were validated by co-elution with dopamine standards under varying buffer conditions and detector settings. No dopamine was detected in unconditioned control medium and HBSS. SURGICAL PROCEDURES AND BEHAVIORAL TESTING. Animals were housed and treated following NIH guidelines. Adult female Sprague-Dawley rats (200250 g) were lesioned by unilateral injection of 6-hydroxydopamine bromide at two sites along the medial forebrain bundle44. Six to seven reaggregates were loaded into a blunt 18G spinal needle (Sherwood Medical) and deposited at AP +1.0 mm, ML -2.5 mm and V -4.7 mm (coordinates relative to bregma), toothbar set at -2.5. Automated assessment (Rota Count-8, Columbus Instruments) of amphetamine-induced rotational behavior (i.p injection of 5 mg d-amphetamine sulfate (SIGMA) per kg body weight) was carried out twice before transplantation (days-21 and -14) and three times after transplantation (days 30, 55 and 80). Only animals with stable pre-transplantational scores of over 7 rotations per minute (average 11 2) were included in this study. HISTOLOGICAL ASSESSMENT AND QUANTITATIVE ANALYSES. Total graft volume was estimated using Cavalieris estimator45. TH-immunoreactive cells were counted within the outlined graft area as used for volume estimation. Uniform randomly chosen sections of individual reaggregates were analyzed for the total number of TH-immunoreactive cells by means of a stereological grid (fractionator)45. Conventional and confocal images were obtained with Axiophot and Axiovert microscopes, respectively (Zeiss). Confocal image stacks were acquired by single and dual wavelength excitation at 488 and 568 nm. Identically treated but not BrdUincubated control cultures were used as a negative control for BrdU and TH double immunohistochemistry. Posttransplantational rotation scores were compared by the non-parametric Mann-Whitney U test. The effect of serum and growth factors was analyzed using ANOVA and Dunnett posthoc comparisons. Data are given as mean standard error.
Acknowledgements
We thank Nadine Kabbani and Elizabeth Rha for their help in the histological analyses and Drs C. Spenger and C. Gerfen for critically reviewing the manuscript. L.S. was supported by a grant of the Swiss foundation for biomedical grants.

1. Takayama, H. et al. Basic fibroblast growth factor increases dopaminergic graft survival and function in a rat model of Parkinsons disease. Nat. Med. 1, 5358 (1995). 2. Choi-Lundberg, D. L. et al. Dopaminergic neurons protected from degeneration by GDNF gene therapy. Science 275, 838841 (1997). 3. Deacon, T. et al. Histological evidence of fetal pig neural cell survival after transplantation into a patient with Parkinsons disease. Nat. Med. 3, 350353 (1997). 4. Olanow, C. W., Kordower, J. H. & Freeman, T. B. Fetal nigral transplantation as a therapy for Parkinsons disease. Trends Neurosci. 19, 102109 (1996). 5. Wenning, G. K. et al. Short- and long-term survival and function of unilateral intrastriatal dopaminergic grafts in Parkinsons disease. Ann. Neurol. 42, 95107 (1997). 6. Kordower, J. H. et al. Neuropathological evidence of graft survival and striatal reinnervation after the transplantation of fetal mesencephalic tissue in a patient with Parkinsons disease. N. Engl. J. Med. 332, 11181124 (1995). 7. Johe, K. K., Hazel, T. G., Mller, T., Dugich-Djordjevic, M. M. & McKay, R. D. G. Single factors direct the differentiation of stem cells from the fetal and adult central nervous system. Genes Devel. 10, 31293140 (1996).
DOPAMINE DETERMINATION BY REVERSE-PHASE HPLC. The growth medium of expanded precursors from three independent culture series was collected after 57 days of differentiation under serum-free conditions and
294
articles
8. McKay, R. D. Stem cells in the central nervous system. Science 276, 6671 (1997). 9. Shah, N. M., Marchionni, M. A., Isaacs, I., Stroobant, P. & Anderson, D. J. Glial growth factor restricts mammalian neural crest stem cells to a glial fate. Cell 77, 349360 (1994). 10. Shah, N. M., Groves, A. K. & Anderson, D. J. Alternative neural crest cell fates are instructively promoted by TGFbeta superfamily members. Cell 85, 331343 (1996). 11. Suhonen, J. O., Peterson, D. A., Ray, J. & Gage, F. H. Differentiation of adult hippocampus-derived progenitors into olfactory neurons in vivo. Nature 383, 624627 (1996). 12. Frederiksen, K. & McKay, R. D. Proliferation and differentiation of rat neuroepithelial precursor cells in vivo. J. Neurosci. 8, 11441151 (1988). 13. Lendahl, U., Zimmerman, L. B. & McKay, R. D. CNS stem cells express a new class of intermediate filament protein. Cell 60, 585595 (1990). 14. Guyenet, P. G. & Crane, J. K. Non-dopaminergic nigrostriatal pathway. Brain Res. 213, 291305 (1981). 15. Rosenthal, A. Auto transplants for Parkinsons disease? Neuron 20, 169172 (1998). 16. Spenger, C. et al. Fetal ventral mesencephalon of human and rat origin maintained in vitro and transplanted to 6-hydroxydopamine-lesioned rats gives rise to grafts rich in dopaminergic neurons. Exp. Brain Res. 112, 4757 (1996). 17. Lindvall, O. et al. Grafts of fetal dopamine neurons survive and improve motor function in Parkinsons disease. Science 247, 574577 (1990). 18. Jiao, S., Gurevich, V. & Wolff, J. A. Long-term correction of rat model of Parkinsons disease by gene therapy. Nature 362, 450453 (1993). 19. Fisher, L. J., Jinnah, H. A., Kale, L. C., Higgins, G. A. & Gage, F. H. Survival and function of intrastriatally grafted primary fibroblasts genetically modified to produce L-Dopa. Neuron 6, 371380 (1991). 20. Sabat, O. et al. Transplantation to the rat brain of human neural progenitors that were genetically modified using adenoviruses. Nat. Genet. 9, 256260 (1995). 21. Beck, K. D. et al. Mesencephalic dopaminergic neurons protected by GDNF from axotomy- induced degeneration in the adult brain. Nature 373, 339341 (1995). 22. Tomac, A. et al. Protection and repair of the nigrostriatal dopaminergic system by GDNF in vivo. Nature 373, 335339 (1995). 23. Crossley, P. H., Martinez, S. & Martin, G. R. Midbrain development induced by FGF8 in the chick embryo. Nature 380, 6668 (1996). 24. Danielian, P. S. & McMahon, A. P. Engrailed-1 as a target of the Wnt-1 signalling pathway in vertebrate midbrain development. Nature 383, 332334 (1996). 25. Hynes, M., Poulsen, K., Tessier-Lavigne, M. & Rosenthal, A. Control of neuronal diversity by the floor plate: Contact-mediated induction of midbrain dopaminergic neurons. Cell 80, 95101 (1995). 26. Hynes, M. et al. Induction of midbrain dopaminergic neurons by sonic hedgehog. Neuron 15, 3544 (1995). 27. Polymeropoulos, M. H. et al. Mutation in the alpha-synuclein gene identified in families with Parkinsons disease. Science 276, 20452047 (1997).
28. Zetterstrm, R. H. et al. Dopamine neuron agenesis in Nurr1-deficient mice. Science 276, 248250 (1997). 29. Hynes, M. et al. Control of cell pattern in the neural tube by the zinc finger transcription factor and oncogene Gli-1. Neuron 19, 1526 (1997). 30. Wang, M. Z. et al. Induction of dopaminergic neuron phenotype in the midbrain by Sonic hedgehog protein. Nat. Med. 1, 11841188 (1995). 31. Brstle, O., Maskos, U. & McKay, R. D. G. Host-guided migration allows targeted introduction of neurons into the embryonic brain. Neuron 15, 12751285 (1995). 32. Campbell, K., Olsson, M. & Bjrklund, A. Regional incorporation and sitespecific differentiation of striatal precursors transplanted to the embryonic forebrain ventricle. Neuron 15, 12591273 (1995). 33. Fishell, G. Striatal precursors adopt cortical identities in response to local cues. Development 121, 803812 (1995). 34. Vicario-Abejn, C., Cunningham, M. G. & McKay, R. D. G. Cerebellar precursors transplanted to the neonatal dentate gyrus express features characteristic of hippocampal neurons. J. Neurosci. 15, 63516363 (1995). 35. Svendsen, C. N., Clarke, D. J., Rosser, A. E. & Dunnett, S. B. Survival and differentiation of rat and human epidermal growth factor-responsive precursor cells following grafting into the lesioned adult central nervous system. Exp. Neurol. 137, 376388 (1996). 36. Svendsen, C. N. et al. Long-term survival of human central nervous system progenitor cells transplanted into a rat model of Parkinsons disease. Exp. Neurol. 148, 135146 (1997). 37. Brundin, P. et al. Survival and function of dissociated dopamine neurons grafted at different developmental stages or after being cultured in vitro. Dev. Brain Res. 39, 233243 (1988). 38. Martinez-Serrano, A., Fischer, W. & Bjrklund, A. Reversal of age-dependent cognitive impairments and cholinergic neuron atrophy by NGF-secreting neural progenitors grafted to the basal forebrain. Neuron 15, 473484 (1995). 39. Snyder, E. Y., Taylor, R. M. & Wolfe, J. H. Neural progenitor cell engraftment corrects lysosomal storage throughout the MPS VII mouse brain. Nature 374, 367370 (1995). 40. Mayer-Proschel, M., Kalyani, A. J., Mujtaba, T. & Rao, M. S. Isolation of lineage-restricted neuronal precursors from multipotent neuroepithelial stem cells. Neuron 19, 773785 (1997)30. 41. Williams, B. P. et al. A PDGF-regulated immediate early gene response initiates neuronal differentiation in ventricular zone progenitor cells. Neuron 18, 553562 (1997). 42. Studer, L. in Current Protocols in Neuroscience (eds Crawley, J. et al.) 3.3.13.3.12 (Wiley, New York, 1997). 43. Studer, L. et al. Non-invasive dopamine determination by reversed phase HPLC in the medium of free-floating roller tube cultures of rat fetal ventral mesencephalon. A tool to assess dopaminergic tissue prior to grafting. Brain Res. Bull. 41, 143150 (1996). 44. Brustie, O., Cunningham, M.G., Tabar, V. & Studer L. in Current Protocols in Neuroscience (eds Crawley, J. et al.) 3.10.13.10.28 (Wiley, New York, 1997). 45. Gundersen, H. J. G. et al. Some new, simple and efficient stereological methods and their use in pathological research and diagnosis. APMIS 96, 379394 (1988).
295
articles
Computation of different optical variables of looming objects in pigeon nucleus rotundus neurons
Hongjin Sun1,2 and Barrie J. Frost1
1 2
Visual and Auditory Neurosciences Laboratory, Department of Psychology, Queens University, Kingston, Ontario, K7L 3N6, Canada Present address: Department of Psychology, McMaster University, Hamilton, Ontario, Canada, L8S 4K1 Correspondence should be addressed to B.J.F (Frost@pavlov.psyc.queensu.ca)
Three types of looming-selective neurons have been found in the nucleus rotundus of pigeons, each computing a different optical variable related to image expansion of objects approaching on a direct collision course with the bird. None of these neurons respond to simulated approach toward stationary objects. A detailed analysis of these neurons firing pattern to approaching objects of different sizes and velocities shows that one group of neurons signals relative rate of expansion (tau), a second group signals absolute rate of expansion (rho), and a third group signals yet another optical variable (eta). The parameter is required for the computation of both and , whose respective ecological functions probably provide precise time-to-collision information and early warning detection for large approaching objects.
Gibsons ecological approach emphasized the importance of dynamic visual information for both perception and the control of ongoing behavior1,2. He and others3 have argued that when an animal interacts with its environment, several distinct types of visual motion pattern may occur, each specifying different classes of events, which in turn require different behavioral responses. For example, motion of a small part of the retinal image relative to other parts of the image that remain stationary, or move differently, usually indicates the presence and trajectory of a moving object or animal. In contrast, its path through space produces visual flow patterns across the entire retina, which can inform the moving animal of its rate of motion and heading through its environment. The relative rate and direction of motion of objects located at different distances (motion parallax) that result from self motion can help to specify the three-dimensional (3D) layout of the environment, whereas the symmetrical expansion of the retinal image of an object can indicate it is on a direct collision course (looming) toward the observer48. For many species, including humans, the image of an object looming toward them typically elicits an escape or avoidance response. This has been demonstrated experimentally in fiddler crabs, fishes, frogs, turtles, chicks, monkeys and humans916. Moreover, in these studies, the responses do not require stereo cues, as they can be elicited by monocular stimuli, and they remain essentially invariant even when the looming objects size, shape and starting distance are changed. Several other studies have shown, in a variety of species, that locomotor behavior of the moving animal itself toward stationary features in its environment is also controlled by monocular information associated with the rapid expansion of features in the retinal image 1722 . Lee 17 proposed that the expansion of the image of an object that is approaching (or the environment that the observer approaches) can trigger a behav296
ioral response, and that the precise timing of the response is controlled by the optical variable called (tau), which is equal to the inverse of the relative rate of expansion of the looming object. As the object approaches, the relative rate of expansion increases and decreases; if the movement velocity is constant, is equal to the time-to-collision, the time that will elapse before the object will collide with the observer. A number of recent studies have investigated and questioned the possibility that the brain mechanisms underlying visuomotor control tasks may involve the computation of tau2326. Figure 1a depicts a spherical object moving on a direct collision course toward an observers eye. For monocular viewing of such looming stimuli, the most salient feature is expansion of the objects image on the retina. Instantaneous image size of the object, or the angle subtended by the object on the retina, is the first-order information available to the nervous system. As the object looms toward the eye, increases approximately exponentially with time, so does the rate of change of visual angle or rate of expansion (rho). , and any other mathematical combinations of and (including ) could potentially be calculated and encoded by the brain and used to predict the time of arrival of the object. Several studies have found looming-sensitive neurons in the visual system of locusts2729, and we have previously shown that neurons in the nucleus rotundus of pigeons (which is homologous to the pulvinar in the mammalian thalamus) are selectively activated in response to looming objects at a fixed time prior to an impending collision30. Little is known, however, about which particular optical variables are being extracted and used by humans or other animals to perform these computations. Here we report the presence of several different types of looming-selective neurons in the dorsal zone of the nucleus rotundus of pigeons, each computing a different optical parameter (variable) of image expansion. One group
articles
articles
signals the absolute rate of image expansion (), another group signals the relative rate of expansion (), and the third group signals yet another optical parameter of time (), which has been proposed to model the response of locust looming-sensitive neurons29. The change in time course of these neuronal responses as we varied the time course of optical variables has helped to elucidate their function.
Looming object
retina
D(t) Moving at a constant velocity v
Results We examined the temporal response patterns of looming-sensitive neurons in the nucleus rotundus (Rt) of pigeons, using a moving soccer ball stimulus, as previously described30. From a total of 234 neurons recorded from 57 pigeons, 85 (36%) were identified as looming neurons and studied in detail. These looming neurons had large receptive fields (60120), and when monocularly presented with object motion in many different 3D directions, they revealed sharp directional tuning curves that were selective for looming objects approaching on a direct collision course toward the bird30,31. These neurons responded only during simulation of an approaching object, but not for simulation of the birds self-motion toward stationary objects32. Subsequent anatomical reconstruction of the locations of the recorded neurons showed that they were concentrated in the dorsal region of Rt. After establishing an Rt neurons looming selectivity, we examined which optical variable of the looming object these neurons encoded. The most direct way to test this is to vary the time course of optical variable changes, by presenting looming objects of different sizes or velocities. (We kept movement velocity constant during each approach.) We found that the whole population of looming neurons could be classified into three types. One class of neurons always initiates their response at the same time-to-collision, regardless of the objects size and velocity. The other two groups of neurons, in contrast, both fired earlier for larger (or slower) objects, but could be clearly distinguished from each other by different patterns of firing rate that increase immediately after the onset of the response and by the timing of the peak responses. Response patterns (peri-stimulus time histograms or PSTHs) for one neuron typical of each of the three classes is shown in Fig. 2. Figure 2a shows different responses as the size of the looming object was varied (from 10 to 50 cm), whereas Fig. 2b shows different responses for the same cells as velocity was varied (from 150 to 750 cm per s). The main effect of varying size or velocity was on the time course of the response, and there was no systematic effect on the magnitude of the peak responses. For neurons like neuron A (Fig. 2a), the response occurred at the same time prior to collision, regardless of the stimulus size, whereas for neurons like neurons B and C (Fig. 2a), the response occurs earlier for larger looming objects. What is different between neurons B and C is that neuron B maintained a similar firing pattern after the onset of the responses, whereas the firing pattern of neuron C varied systematically after the onset of firing. This is evident from the shallower slopes in the ascending phase of the responses and earlier peaks for objects of larger sizes in neuron C. Figure 1be illustrates the time course of several different optical variables that could potentially be encoded in the brain, as a spherical object moves on a direct collision course toward an observers eye. Solid and dotted lines indicate large and small objects respectively. Figure 1b depicts the time course of the (t) function. When an object moves directly towards the observer at a constant velocity V, at time t, for small values of , the (t) function can be expressed in equation 1 as follows:
b

Threshold
Threshold
1/ Time 0 Time 0
Threshold
Time
Time
Fig. 1. A schematic diagram of a spherical object of diameter d directly approaching an animals eye (a). When an object moves at constant velocity V, at time t, it is at distance D(t) away from the eye, and subtends a visual angle of (t). Time course of four optical variables, (and 1/), , , and derived from the edges of two spherical objects of different sizes, moving at a constant speed directly toward the eye are shown in (b), (c), (d) and (e) respectively (solid line corresponds to larger sphere, dashed line to smaller sphere). Horizontal lines represent hypothetical threshold values for onset of neuronal firing. Time was plotted backwards from the time of collision, t = 0.
(t)
(t) (t)
(1)
As shown in Fig. 1b, the size of the object has no effect on the value of (or 1/) at a given time point. In fact, because equals time-to-collision if movement velocity is constant, it provides accurate timing predictions with an object of unknown size and velocity17. Because neurons like neuron A always initiated firing at the same time-to-collision (Fig. 2), these neurons seem to be encoding . Figure 1c and d depicts the time course for and . These two functions increase faster for larger objects, so if a neuron initiates firing when a threshold value of or is reached, it should fire earlier for larger objects. Neurons like neuron B (Fig. 2a) all behaved in this manner. In addition to the response onset time, it is interesting to look at the temporal pattern of the response after the onset of the response, to see whether the neuronal firing patterns actually follow the temporal profile of any of these optical variables. Although constraints on maximum firing rates would prevent neurons from continuing to code or immediately before col297
articles
lision, when the value of these variables becomes very large (Fig. 1c and d), it is still possible to ask whether the firing rate might track these variables in the early stages of the response. As shown in Fig. 1c and d, if the neuronal firing rate follows either or , the rate of increase in the early phase of firing should be faster for larger objects. However, we found no systematic difference in the slopes of the PSTHs in neurons like neuron B (Fig. 2), indicating that they do not track or but instead start to fire when one or other of these variables reaches threshold. The third class of neurons, exemplified by neuron C (Fig. 2a), does not appear to track or either, because they showed shallower slopes for larger objects, which does not correspond to either function. Instead, the responses of these neurons showed a distinct peak, which occurred earlier for larger objects. A more suitable optical variable to describe this pattern would be some other mathematical function of time that becomes negatively related to when reaches large values. The function, proposed by Hatsopoulos and colleagues29 to model locusts looming-detector responses, seems to fit these requirements well. The function is represented in equation 2 as follows: (t) = C (t) e-(t) (2)
time courses (Fig. 2b) closely parallel those found for the size manipulation (Fig. 2a) for the same three neurons. Although the size and velocity manipulations produce identical retinal image expansion profiles, in the real world, and our simulated 3D space, the size and velocity of objects vary independently. We then used the data from these size and velocity manipulation experiments to examined the time course of looming responses more quantitatively. Onset of response was arbitrarily defined as the point where firing rate first reached 33% of maximal rate. We then measured the time-to-collision (Tc) for both the response onset and the response peak, that is, the time interval between these points and the moment of collision (Tconset and Tcpeak). These empirically obtained values were then compared against the quantitative predictions generated from mathematical transformations of the optical variables described above. Because equals time-to-collision regardless of object size and velocity, if a neurons onset of response is triggered by a threshold of (Th), Tconset should always be constant. If represents response latency, then Tconset can be defined (equation 3) as follows: Tconset = Th - (3)
Here, C is a proportionality constant that controls the overall If a neuron is encoding other optical variables, however, there magnitude of the response, and is a constant for a particular Fig. 2. Based on the differNeuron A Neuron B Neuron C neuron that controls the a ences in the time course of () () () inhibitory damping as the excithe neuronal responses relatatory response become very tive to the moment of collilarge. The (t) function has a dission, the looming sensitive tinct peak, and its ascending neurons in nucleus rotundus phase slope is shallower for larghave been classified into three er objects (Fig. 1e), which is the distinct classes. This figure opposite of or functions. In shows the response pattern neurons like neuron C (Fig. 2a), (post-stimulus time histhe responses showed distinct tograms) for a typical neuron peaks, and larger objects proin each of the three classes duced shallower slopes, which (neuron A, B, and C for , , matches what we expect from the and respectively) to a series theoretical function, thus sugof stimuli (a simulated moving gesting neurons in this class actusphere with a soccer-ball patally track the optical variable . tern) of varying sizes (a) or varying velocities (b), moving Object velocity was also varied b along the direct collisionwhile object size was held constant course path toward the bird. (Fig. 2b). If one only examines Responses are the sum of five retinal image size, variation of trials and are referenced to velocity (or more precisely, the time zero, which is the time inverse of velocity) actually creates when the stimulus would have the equivalent change of visual collided with the bird. The angle over time as produced by simulated path was 15 m in the object size manipulation. To length. In (a), velocity for neudescribe this more intuitively, ron B was 375 cm per s and doubling the objects velocity for neurons A and C was 500 would require the object to start cm per s. In (b), object size its movement at twice the distance was 30 cm for all three neufrom the eye (if we maintain the rons. Note that for neuron A, same movement duration). This the timing of the response would create the same visual angle remains invariant despite subchange over time as would occur stantial changes in size and Time-to-collision if an object half the size was velocity, whereas for neuron B moved at the original velocity. As and neuron C, the timing depends on object size and velocity, with larger or slower objects evoking an expected, when velocity was earlier response. Further quantitative examination suggests that neurons A, B and C encode optical manipulated, neuronal response variables , , and respectively.
298
Velocity Size
articles
a
Response onset Tc (sec)
b
Response peak Tc (sec) Square root of (object diameter/velocity)
Object diameter/velocity
Fig. 3. Quantitative examination of the timing of the responses of same three neurons shown in Fig. 2. Empirical measurements of the time intervals between response onset or response peak and time of collision ( Tconset and Tcpeak respectively) for each neuron are plotted as a function of the square root of d/V (a) and d/V (b) respectively. Best-fit linear functions were also generated for each plot. The dataset is the same as for Fig. 2; data from manipulation of object size and object velocity were pooled together for each cell. For neuron A, Tconset did not differ much over a range of values of the square root of d/V (a); for neuron B, however, the plot of Tc onset as a function of the square root of d/V demonstrates a very strong linear relationship (a). For neuron C, the plot of Tcpeak as a function of d/V also clearly demonstrates a strong linear relationship (b). This analysis further supports the idea that neurons A, B and C encode optical variables , , and respectively. Neuron A (), Neuron b (), b Neuron C ().
should be specific mathematical relationships between Tconset or Tcpeak and the object size d (and approaching velocity V). For variable , if a neuron starts to fire when a threshold of (Th ) is reached during a looming objects approach, Tconset has an approximately linear relationship with the square root of d/V, and is represented (equation 4) as: 1 d Tconset - (4) V Th For visual angle , if a neuron starts to fire when a threshold of (Th)is reached, Tconset has a linear relationship with d/V: Tconset = Th 2 tan( ) 2 1 d - V (4.1)
If a neuron tracks the function, Tcpeak would have a linear relationship with d/V: d Tcpeak = - 2 V (5)
Experimentally obtained data for response times were compared to these theoretical predictions. Empirical measurements of Tc (both Tconset and Tcpeak) from the responses of each of the three neurons shown in Fig. 2 are plotted in Fig. 3 as a function of the square root of d/V and d/V respectively. Best-fit linear functions were also generated for each plot. The Tc data for object size and velocity manipulation were pooled together for each cell. For neuron A, because there are minimal differences in Tconset values for looming objects of different sizes and velocity, the best-fit line (Tc onset versus square root of d/V) is rather flat (Fig. 3a, cf. equation 3). The empirical Tconset values closely match the theoretical prediction that the neuron would start to fire when reaches a certain threshold value. It is thus very likely that neuron A is signaling or time-to-collision. Although most of our tests were done under constant velocity condition, we also tested the responses of the neurons when the movement velocity of the looming object was not constant during
approach. We found that when an object approaches with deceleration or acceleration, the timing of the onset of the response of these neurons matched what would be predicted if the neuron responds to the same value, as revealed in the constant velocity situation. This strengthens the conclusion that these neurons are indeed computing and thus signaling accurately the time-to-collision when the movement velocity is constant. It also provides an approximate estimate of time-to-collision even when the movement velocity is not constant. It is evident that empirical Tconset values for neuron B produces a tight linear relationship to the square root of d/V (Fig. 3a, cf. equation 4). Moreover Tconset data for neuron B does not produce a strong linear relationship to d/V (plot not shown, cf. equation 4.1). This suggests that neuron B encodes and not . Moreover, according to equation 4, the exact value of the threshold for can be derived from the slope of the linear relationship. For neuron B, the regression line (R2 = 0.988) has a slope of 4.1, thus the calculated threshold was 3.4 degrees per second. Similarly Tcpeak values for neuron C are linearly related to d/V (see Fig. 3b, cf. Equation 5). This is consistent with the notion that neuron C tracks the function. From the slope of equation 5, one can obtain the exact value of , which is the constant in the visual angle component of the function shown in equation 5. Thus the complete function can be generated. The regression line for neuron C (R 2 = 0.994) has a slope of 12.6, whereas was 25.2. From Fig. 3b, one can also see that there is a weak relationship between Tc peak and d/V for neuron B (R2 = 0.195). Therefore neuron B clearly does not encode . However, there is a strong linear relationship between Tconset and square root of d/V (R2 = 0.929) for neuron C (Fig. 3a). However, this does not contradict the notion that neuron C encodes the function. This is because in the initial stages of the function, the contribution of the (t) or component is dominant, and only in the later stages does the exponential component produce a negative inhibitory contribution to the overall product of the function.
299
articles
a
% of drop-off in firing rate at the time of collision
b
R2 for reponse peak Tc Variance in response onset Tc
R2 for response onset Tc
Fig. 4. Quantitative examination of the timing of the response for the population of nucleus rotundus looming-sensitive neurons when presented with approaching objects that varied in size or velocity. In (a), the variances (standard deviation) of Tconset were plotted along the x-axis, and the average drop-off in firing rates at the time of collision, relative to the response peak (%), was plotted along y-axis. The R2 values (b) and slope values (c) of the two linear relationships (Tc onset versus square root of d/V and Tcpeak versus d/V) were plotted against each other. The data points are clustered in three separate regions in each plot, and the group membership of each neuron is consistent across (a), (b) and (c). Therefore this population of neurons can be classified into three distinct groups (, or neurons). neuron, neuron, b neuron.
Slope for response peak Tc
Slope for response onset Tc
A series of similar quantitative analyses were performed for the population of Rt looming selective neurons. From the total of 85 looming neurons found and studied in detail, 70 neurons produced complete sets of qualitative and quantitative analyses, which indicated that 37 were neurons, 20 were neurons and 13 were neurons. None of the looming-sensitive neurons were found to encode absolute visual angles (i.e., matching equation 4.1). The data for the remaining 15 neurons were discarded because of insufficient data points for a regression analysis (although the timing of their responses and the shape of their response histograms suggest that they most likely were 3 , 8 and 4 neurons) An important question is whether these response properties represent three distinct subgroups of looming-sensitive cells, or whether they form a continuum. To address this, a number of criteria were used to classify the 70 looming neurons for which complete data sets were available. The initial classification made no prior assumption of the underlying optical variable that these neurons may be encoding. After examining Tconset, we found that one group of neurons showed very little variation as object size and velocity was varied ( neurons), whereas the remaining neurons showed much larger variation ( or neurons). Those neurons with large systematic variances in Tconset could be further subdivided into two groups based on the shape of their response histograms. We found that one of these groups showed a substantial drop-off in firing rate after reaching their peak response rate, and a greatly reduced response was evident at the time of collision ( neurons). In contrast, the firing rate of the other group remained at a fairly high level for some time ( neurons). These response characteristics are displayed in Fig. 4a, where three distinct clusters are apparent.
300
In Fig. 4a, the variance (standard deviation) of the Tconset produced by variation of the size and velocity of looming objects is plotted for each cell along the x-axis, and its average drop-off (or reduction) in firing rate at the time of collision, relative to the response peak (in percentage), is plotted along the y-axis. The looming cells clearly cluster in three separate regions in the plot. One cluster ( neurons) has very small variances in Tconset compared to the other two groups, and their response drop-off is small (i.e., the responses remain at a high level even at the moment of collision). A second cluster ( neurons), like neurons, also show a small drop-off after peak firing, yet have much larger variation in Tc onset for different stimulus sizes/velocities, compared to that of neurons. The third cluster ( neurons) exhibits a significant drop-off in firing rate after peak firing compared to the other two groups. Although this initial classification reveals the existence of three distinct groups of rotundus looming neurons, Fig. 4b and c further demonstrates quantitative relationships between the timing of each cells response to object properties, by testing against certain predictions from possible underlying optical variables. The fits of each neuron with respect to the two possible linear relationships (Tconset versus square root of d/V and Tcpeak versus d/V) outlined above are plotted against each other, for both R2 values (Fig. 4b) and slope values (Fig. 4c). Again the data points of the Rt looming neuron population cluster in three separate regions (Fig. 4b and c). The cluster of data points with very low slope values (Fig. 4c) for both Tconset and Tcpeak (along both x and y-axis) functions, clearly belongs to neurons, and indicate that for these cells, Tc onset or Tcpeak show little systematic variation with changes in d/V (object size/velocity), even though the R2 values for Tconset are not always very small (Fig. 4b, x-axis).
articles
Box 1. Mathematical appendix

As shown in Fig. 1a, when a spherical object (diameter d) approaches directly towards an animals eye at constant velocity V, at time t, it is at distance D(t) away from the eye, and subtends a visual angle of (t). T0 t Tc = T0 - t Total movement duration, i.e., from the start of the movement until the moment when the object reaches the eye Time since the start of the movement Time-to-collision, i.e., time that will elapse before the object will collide with the eye If a neuron responds when a threshold of (t), i.e., Th is reached,
TcTh = To - tTh
1 d Th V
And, if we consider a response latency ()
Tconset
(1) can specify time-to-collision (Tc)

From Fig. 1a,
1 d Th V - d V
(4)
Thus, Tconset for Th has a linear relationship with with slope of
d (t) 2 tan[] = 2 D(t)

When both sides are differentiated with time (t), and V>0
1 and intercept of - Th
1 Thus, the threshold of the rate of expansion, Th = 2 slope From Equation A,
(t) 2 2 [cos ] 2
d 2 V (t) = [D(t)]2
To - t =
(t) 2 tan 2
d V
Dividing the first equation by the second,
If a neuron reponds when threshold is reached,
sin[(t)] D(t) = (t) V

For small values of (t), this becomes
TcTh = To - tTh =
Th 2 tan () 2
d V
If we then consider response latency ()
(t) D(t) (t) V

If we define (t) = Then (t)
Tconset = (t) (t) (1)
Th 2 tan( ) 2 1
d - V V
(4.1)
d Thus, Tconset for Th has a linear relationship with ,

with a slope of
D(t) = Tc V
Therefore, the optical variable (t) can specify time-to-collision (Tc) when the movement velocity is constant.
Th 2 tan( ) 2
and intercept of -
(3) The time-to-collision (Tc) when reaches peak value (t) = C (t) e-(t)
At the peak of the (t) function, (tpeak) = 0
(2) The time-to-collision (Tc) when or reaches threshold
Because (t)>0, (t)>0 and they are monotonic functions, through their inverse functions unique t values can be generated for certain (t) or (t) values. From Fig. 1a,
(2)
(t) = C [(t) e-t + e-t (-) ((t))2] (tpeak) = [(tpeak)]2 (A) (B) d V
Based on Equation B
d (t) 2 d 1 = tan( ) = 2 D(t) 2V To - t (t) = 1 d V (T -t)2 + o 4V d
(C)
2V (t) = (To - t) [(t)]2 d

Combining Equations C and D,
(D)
d 1 To - t = - 4V (t)
In our experiment,
To-tpeak = d 2 V
If a neuron tracks the function and a response latency () is considered,
d 1 1 (s) and > 5 (s/rad) 4V (t)Th d 1 (t)Th 4V d 1 To - t V (t)
Tcpeak = d - 2 V
(5)
Thus, Tcpeak for the peak of the function has a linear relationship with d with a slope of and intercept of -. 2 V
301
articles
For the function involving Tconset (along x-axes), the cluster of data points with high values of R2 (Fig. 4b) and not very small values of slope (Fig. 4c) clearly belongs to neurons. The fact that, along the y-axes (Tcpeak), they do not show high R2 values (Fig. 4b) or slopes (Fig. 4c) distinguishes them from the third cluster of cells ( neurons). This third cluster of cells shows high values of R2 (Fig. 4b) and slopes (Fig. 4c), which indicates they are most likely neurons. However, looking at the x-axes (Tconset), we find some of these neurons also have high values of R2 (Fig. 4b) and slope (Fig. 4c). This is because of the large contribution of the (t) variable in the function in the early stages, as discussed previously for neuron C (Fig. 3a). Overall, of the 20 neurons, 15 revealed a very strong linear relationship in Tconset versus square root of d/V (R2>0.9), and the 5 others had R2 values between 0.70.9. For the 13 neurons, regression analysis (Tcpeak vs. d/V) revealed that 9 neurons had R 2 >0.9 and 4 neurons had R 2 values between 0.60.9. In addition, the group membership of each neuron is completely consistent across figures 4a, b and c. Therefore, the combination of these different criteria clearly indicate the existence of three distinct neuronal groups, which appear to encode three different optic variables of looming objects.
it quick defensive actions. Indeed there have been a number of behavioral studies in animals and humans suggesting that instantaneous object size is a cue that triggers motor action in the face of impending collision33,34. Importantly, although the looming response was elicited by simulated motion of isolated objects, no response was seen when the object was combined with a background that underwent the same transformation32. The latter stimulus mimics the effect of self-motion, and so it seems likely that self-motion, and hence information about impending collision with stationary objects, is processed in separate brain structures from those that code looming objects.
Methods Methods have previously been described elsewhere 31. We recorded extracellularly from anaesthetized (ketamine/rompun) pigeons (Columba livia), which were positioned in a stereotaxic instrument. This project was reviewed by the Queens University Animal Care Committee and certified as in compliance with the guidelines of the Canadian Council of Animal Care. Visual stimuli were presented monocularly on a rear-projection tangent screen (120o x 120o), which was placed 40 centimeters in front of one of the birds eyes. The visual stimulation system consisted of a high resolution (1280 x 1024 pixels) Electrohome Trinitron color projector (ECP4000) and a Silicon Graphics IRIS 4D/310GTX computer graphics system. The visual object was designed as a paneled soccer ball with alternating black and white panels, with the area of black and white in equal proportions to control for overall luminance changes. The soccer ball could move in different directions in simulated three-dimensional space. The computer calculates the appropriate projective geometry and displays the transformed images in sequence in real time (at 60 frames per s) to yield smoothly moving and transforming animation.
Discussion It is interesting to look at the functional implications of these three types of Rt looming detectors. The three cell types found here might represent different sources of sensory information that are required to produce different behaviors, or given the nature of the mathematical functions themselves, some might also serve as a building blocks for the computation of other optical variables. Detecting time-to-collision from of an approaching object moving in depth would be a valuable strategy for survival, because it provides accurate predictions of the motion of an object of unknown size and velocity. Indeed, it was found that Rt looming neurons responses were followed by a sharp increase in EMG activity of the pigeons major flight muscle and also heart rate30, both of which are associated with natural defensive responses, such as escape and avoidance behavior. The neurons signal the rate of visual angle change, which increases nearly exponentially as a looming object approaches very close to the eye (as shown in Fig. 1c); therefore, the value of carries important information about the timing of the looming motion. By responding to certain levels of , these neurons could provide animals with gross estimates of the time-to-collision of a looming object. For animals living in a stable environment, the motion of predators may be more or less stereotypical (the size and velocity of particular predators is relatively stable), and therefore certain values of could be hardwired to provide a gross estimate of time-to-collision. In addition, given the usefulness of rate-of-expansion information, neurons most likely serve as building blocks for other neurons to calculate more complicated optical variables, such as and . As for neurons, one of the key features of the function is that Tcpeak is linearly related to the ratio of object size and velocity (see equation 5). For different-sized objects moving at the same speeds, a larger object would reach its peak firing rate earlier. This feature might have certain evolutionary benefits for prey like pigeons, as larger objects often represent a greater danger or reflect a higher probability that the object is a predator. These neurons might quickly and simply detect incoming large objects and thereby also signal urgency to elic302
After a single unit was isolated, three-dimensional directional tuning curves were obtained to determine whether a cell selectively responded to looming stimuli. Then random sequences of stimuli differing in either size or velocity were presented to determine the precise time course of responses to these parametric variations.
Acknowledgements
The authors wish to thank T. Kripalani and S. David for excellent technical assistance and D. Fleet and N. Troje for helpful discussions and comments on the manuscripts. HJS was supported by a Postgraduate Scholarship from the Natural Science and Engineering Research Council (NSERC) of Canada. This work was supported by an NSERC grant OGP0000353 and an Alexander von Humboldt Research Award to BJF.
RECEIVED 11 MARCH: ACCEPTED 21 MAY 1998

1. Gibson, J. J. The Senses Considered as Perceptual Systems (Houghton Mifflin, Boston, 1966). 2. Gibson, J. J. The Ecological Approach to Visual Perception (Houghton Mifflin, Boston, 1979). 3. Nakayama, K. & Loomis, J. M. Optical velocity patterns, velocity-sensitive neurons and space perception: a hypothesis. Perception 3, 6380 (1974). 4. Frost, B. J., Wylie, D. R. & Wang, Y.-C. in Perception and Motor Control in Birds (eds Davies, M. N. O. & Green, P. R.) 248269 (Springer-Verlag, Berlin, 1994). 5. Longuet-Higgins, H. C. & Prazdny, K. The interpretation of moving retinal image. Proc. R. Soc. Lond. B 206, 358397 (1980). 6. Wylie, D. R., Kripalani, T. & Frost, B. J. Responses of pigeon vestibulocerebellar neurons to optokinetic stimulation. I. Functional organization of neurons discriminating between translational and rotational visual flow. J. Neurophysiol. 70, 26322646 (1993). 7. Roy, J.-P. & Wurtz, R. H. The role of disparity-sensitive cortical neurons in signalling the direction of self-motion. Nature 348, 160162 (1990). 8. Tanaka, K. & Saito, H. Analysis of motion of the visual field by direction, expansion/contraction, and rotation cells clustered in the dorsal part of the medial superior temporal area of the Macaque Monkey. J. Neurophysiol. 62, 626641 (1989).
articles
9. Schiff, W. Perception of impending collision: A study of visually directed avoidant behaviour. Psychol. Monogr. 79, 126 (1965). 10. Dill, L. M. The escape response of the zebra danio (Brachydanio rerio). I. The stimulus for escape. Anim. Behav. 22, 771722 (1974). 11. Ingle, D. J. & Shook, B. L. in Brain Mechanisms of Spatial Vision (eds Ingle, D. J., Jeannerod, M. & Lee, D. N.) 229258 (Martinus Nijhoft, Dordrecht, 1985). 12. Hayes, W. N. & Saiff, E. I. Visual alarm reactions in turtles. Anim. Behav. 15, 102108 (1967). 13. Tronick, E. Approach response of domestic chicks to an optical display. J. Comp. Physiol. Psychol. 64, 529531 (1967). 14. Schiff, W., Caviness, J. A. & Gibson, J. J. Persistent fear responses in rhesus monkeys to the optical stimulus of looming. Science 136, 982983 (1962). 15. Bower, T. G. R., Broughton, J. M. & Moore, M. K. Infant responses to approaching objects: an indicator of response to distal variables. Percept. Psychophys. 9, 193196 (1970). 16. Ball, W. & Tronick, E. Infant responses to impending collision. Science 171, 818820 (1971). 17. Lee, D. N. A theory of visual control of braking based on information about time-to-collision. Perception 5, 437459 (1976). 18. Lee, D. N. The optic flow field: The foundation of vision. Philos. Trans. R. Soc. Lond. B Biol. Sci. 290, 169179 (1980). 19. Lee, D. N., & Reddish, P. E. Plummeting gannets: a paradigm of ecological optics. Nature 293, 293294 (1980). 20. Lee, D. N., & Reddish, P. E. Visual regulation of gait in long jumping. J. Exp. Psychol. Hum. Percept. Perform. 8, 448459 (1982). 21. Sun, H.-J., Carey, D. P. & Goodale, M. A. A mammalian model of opticflow utilization in the control of locomotion. Exp. Brain Res. 91, 171175 (1992). 22. Wagner, H. Flow-field variables trigger landing in flies. Nature 297, 147148 (1982).
23. Lee, D. N. & Young, D. S. in Brain Mechanisms and Spatial Vision (eds Ingle, D. J., Jeannerod, M. & Lee, D. N.) 130 (Martinus Nijhoff, Dordrecht, 1985). 24. Regan, D. & Hamstra, S. J. Dissociation of discrimination thresholds for time to contact and for rate of angular expansion. Vision Res. 33, 447462 (1993). 25. Tresilian, J. R. Four questions of time to contact: A critical examination of research on interceptive timing. Perception 22, 653680 (1993). 26. Wann, J. P. Anticipating arrival: Is the tau margin a specious theory? J. Exp. Psychol. Hum. Percept. Perform. 22, 10311048 (1996). 27. Rind, F. C. & Simmons, P. J. Orthopteran DCMD neuron: A reevaluation of responses to moving objects. I. Selective responses to approaching objects. J. Neurophysiol. 68, 16541666 (1992). 28. Simmons, P. J. & Rind, F. C. Orthopteran DCMD neuron: A reevaluation of responses to moving objects. II. Critical cues for detecting approaching objects. J. Neurophysiol. 68, 16671682 (1992). 29. Hatsopoulos, N., Gabbiani, F. & Laurent, G. Elementary computation of object approach by a wide-field visual neuron. Science 270, 10001003 (1995). 30. Wang, Y. & Frost, B. J. Time to collision is signalled by neurons in the nucleus rotundus of pigeons. Nature 356, 236238 (1992). 31. Wang, Y. C., Jiang, S. & Frost, B. J. Visual processing in pigeon nucleus rotundus: luminance, colour. motion and looming subdivisions. Visual Neurosci. 10, 2131 (1993). 32. Frost, B. J. & Sun, H.-J. in From Living Eyes to Seeing Machines (eds Srinivasan, M. V. & Venkatesh, S.) 80103 (Oxford Univ. Press, Oxford, 1997). 33. DeLucia, P. R. Pictorial and motion-based information for depth perception. J. Exp. Psychol. Hum. Percept. Perform. 17, 738748 (1991). 34. Robertson, R. M. & Johnson, A. G. Retinal image size triggers obstacle avoidance in flying locusts. Naturwissenschaften 80, 176178 (1993).
303
articles
Dopamine neurons report an error in the temporal prediction of reward during learning
Jeffrey R. Hollerman1,2 and Wolfram Schultz1
1 2
Institute of Physiology, University of Fribourg, CH-1700 Fribourg, Switzerland Present address: Department of Psychology, Allegheny College, Meadville, Pennsylvania, 16335, USA Correspondence should be addressed to W.S. (Wolfram.Schultz@unifr.ch)
Many behaviors are affected by rewards, undergoing long-term changes when rewards are different than predicted but remaining unchanged when rewards occur exactly as predicted. The discrepancy between reward occurrence and reward prediction is termed an error in reward prediction. Dopamine neurons in the substantia nigra and the ventral tegmental area are believed to be involved in reward-dependent behaviors. Consistent with this role, they are activated by rewards, and because they are activated more strongly by unpredicted than by predicted rewards they may play a role in learning. The present study investigated whether monkey dopamine neurons code an error in reward prediction during the course of learning. Dopamine neuron responses reflected the changes in reward prediction during individual learning episodes; dopamine neurons were activated by rewards during early trials, when errors were frequent and rewards unpredictable, but activation was progressively reduced as performance was consolidated and rewards became more predictable. These neurons were also activated when rewards occurred at unpredicted times and were depressed when rewards were omitted at the predicted times. Thus, dopamine neurons code errors in the prediction of both the occurrence and the time of rewards. In this respect, their responses resemble the teaching signals that have been employed in particularly efficient computational learning models.
Current theories view learning as the acquisition of specific predictions14. Humans and animals learn to predict the outcomes of their behavior, including rewards. Learning depends on the extent to which these outcomes are different than predicted, being governed by the discrepancy or error between outcome and prediction. Outcomes that affect learning in this way are termed reinforcers. Learning proceeds when outcomes occur that are not fully predicted, then slows down as outcomes become increasingly predicted and ends when outcomes are fully predicted. By contrast, behavior undergoes extinction when a predicted outcome fails to occur. (In the laboratory, predictions may fail either because the subject made an error or because the experimenter withholds the reward for correct behavior.) Recent learning algorithms employ errors in the prediction of outcome as teaching signals for changing synaptic weights in neuronal networks5. In these models, an unpredicted outcome leads to a positive signal, a predicted outcome to zero signal and the absence of a predicted outcome to a negative signal. The most efficient models capitalize on the observation that a key component of predictions concerns the exact time of reinforcement6,7. Their teaching signals use errors in the temporal prediction of reinforcement and compute the prediction error over consecutive time steps in individual trials (temporal difference algorithm8). Thus, teaching signals come to report progressively earlier reinforcement-related events and thus predict the outcome rather than simply reporting that it has occurred. They are particularly efficient for learning, as they can influence the behavioral reaction before it is executed. Reinforcement models that use predictive teaching signals can learn a wide variety of behavioral tasks, from balancing a pole on a cart wheel9
304
to playing world-class backgammon10. It is therefore of interest to determine whether real nervous systems might process rewards in a similar manner during learning. Results from lesioning and psychopharmacological experiments indicate a role of dopamine systems in behavior driven by rewards and in reward-based learning1214. We have studied the neural mechanisms underlying this role of dopamine in monkeys and have previously reported that midbrain dopamine neurons show responses to food and liquid rewards that depend on their predictability15,16. The present study investigated whether these responses could have the formal characteristics of teaching signals. We found that the magnitude of dopamine responses to a juice reward reflected the degree of reward predictability during individual learning episodes. An unexpected reward evoked a strong response in dopamine neurons. As the monkeys performance improved (i.e. as they learned to predict which response would trigger a reward), the neuronal response to the reward progressively decreased. Moreover, by varying the timing of reward, we found that dopamine neurons signal not only its occurrence but also its timing relative to expectations. Thus dopamine neurons seem to track the reward prediction error and emit a signal that has all the typical characteristics of a positive reinforcing signal for learning.
Results Dopamine neurons in pars compacta of the substantia nigra and the ventral tegmental area were studied while monkeys learned to associate visual stimuli with liquid reward. Dopamine neurons in these two different midbrain groups showed similar responsnature neuroscience volume 1 no 4 august 1998
articles
b
Familiar
Rewarded
Unrewarded
Fig. 1. The discrimination learning task. (a) Animals released a resting key when a pair of pictures appeared, touched the lever below the rewarded picture and received a drop of liquid. (b) Pictures used in the task. The same pair of two fractal pictures was used in all familiar trials (top). A new pair of two pictures was used in each block of learning trials (middle and bottom).
es and are therefore not distinguished in this report. We first tested 211 dopamine neurons during the learning of a visual-discrimination task. Animals were simultaneously presented with two pictures. If they touched a lever below one of the pictures, they received a drop of liquid, whereas the other picture was not rewarded (Fig. 1). During the initial presentations, 75% of dopamine neurons were activated when the reward occurred, comparable to other learning situations15,17,18. The same two pictures were presented repeatedly (varying randomly between left and right positions), and as the task was learned, the reward gradually ceased to activate dopamine neurons; instead, these neurons became responsive to presentation of the reward-predicting pictures, consistent with previous findings15,17. We then studied the reward responses of individual dopamine neurons during complete learning episodes. For each new
Fig. 2. Learning curves. Performance increased rapidly over successive trials with each pair of novel pictures but was stable with the familiar pictures. Percentage correct is calculated using the number of trials required to attain each two correct trials. Arrow indicates the mean number of trials required to reach the behavioral criterion for learning (second correct trial of first series of four consecutive correct trials). Data are means from 54 familiar and learning blocks in which 20 or more correct trials were performed and dopamine neurons were recorded (45 blocks) or only learning was studied (9 blocks). Each of the 54 blocks used a novel picture pair. Bars show standard errors and are not visible for familiar trial data because of their small size (01%).
episode, a novel pair of pictures was presented, whereas all other task components remained unchanged. Animals learned by trial and error to associate one of the novel pictures with reward. A learning criterion served to indicate the deviation from chance performance; this was defined as the second correct response in the first series of four consecutive correct responses. The learning criterion corresponded to the loss of chance performance and was reached on average after 12 trials, 5 of them being correctly performed (Fig. 2; see also Fig. 5a). The frequent errors during the initial learning period indicate that the pictures themselves did not contain any unintended predictive information. Dopamine neurons showed strong activation (increase in firing rate) in response to unpredicted free liquid delivered outside the task (45 of 61 neurons, 74%), consistent with previous findings15 (Fig. 3b). Strikingly, many neurons in the present study (33 of 66, 50%) were also activated by the reward during initial learning trials (Fig. 3). By contrast, reward activations were rare in trials with familiar pictures (8 of the 66 neurons, 12%) (p < 0.0001, ANOVA on magnitudes in initial ten free-liquid, learning and familiar trials; all three situations p < 0.05, Fisher test). Most neurons also showed activations following the novel pictures (not shown), resembling novelty responses and response generalization from learned, reward-predicting stimuli observed previously1722. During individual learning episodes, each responding neuron progressively lost its reward activation after the criterion was attained, and activations approached the low levels typical for familiar trials. The rate of decrease in responsiveness was related to the duration of the learning period before criterion. Some stimulus pairs were learned more quickly than others; in cases where only a few trials were required to reach criterion, reward-related activations decreased rapidly (Fig. 3a and b), whereas in cases where learning occurred more slowly (Fig. 3c), reward responses persisted even after tens of trials. In the population of 66 dopamine neurons, mean activations following reward delivery increased nearly threefold in the first two trials with novel pictures (193% above background) and declined rapidly afterwards (to 90110% above background), thus mirroring the learning curve of Fig. 2. Similarly, when results were analyzed with reference to criterion, reward activations were highest in the trials before the animal reached criterion (i.e. when the error rate was highest) and declined gradually thereafter (Figs 4 and 5). Differences were significant for learning versus familiar performance (Fig. 5b) and, in particular, for trials prior to reaching criterion versus subsequent learning blocks (Fig. 5b). This was also found when the learning criterion was redefined as the fourth correct response in the first series of four consecutive correct trials, which was attained after a mean of seven correct trials. Thus, the observed effect seems to be robust with respect to the criterion chosen. The relation to reward prediction became further evident when the predicted reward failed to occur because of behavioral errors (Fig. 6a). Activity was significantly depressed in 70% of neurons (28 of 40) that were recorded during at least six error Familiar trials during learning. Depressions Learning began at 99 29 ms after reward would have been delivered upon correct behavioral response and lasted Correct trials 401 36 ms (mean standard
305
Learning
Percent correct
articles
Fig. 3. Reward responses of b Free liquid a Learning c Learning three dopamine neurons (ac) during learning of pairs of novel pictures. Reward responses decreased after the Learning learning criterion (second of four correct responses) was reached (arrowhead). The Familiar panels (a), (b) and (c) show activity during fast, medium and slow learning episodes, in Familiar which criterion was reached after 2, 6 and 16 correct trials Reward respectively. The bottom panFamiliar els in each case show the absence of response to a predicted reward in blocks of triReward als with familiar pictures. The top panel in (b) shows the Reward response to a free reward outside the context of the task. The response lever is touched at 1.0 s before reward was delivered (-1 s), except in free liquid trials. Dots denote neuronal impulses, aligned to the electric pulse that opened the liquid reward valve (center vertical lines). Individual lines represent consecutive correct trials in sequence from top to bottom. Error trials are omitted.
error). This decrease in firing suggests that the neurons signal the (negative) discrepancy between reward occurrence and reward prediction. The depression occurred at the exact time of habitual reward, suggesting that the prediction to which the neurons respond is specific with respect to the timing of the reward. In order to investigate this, we changed the time of reward delivery, using the familiar pictures to provide stable predictions over consecutive trials. The reward was normally delivered 1.0 s after a correct lever touch and did not lead to a change in firing rate. When reward was suddenly delayed by 0.5 s, i.e. 1.5 s after the lever touch, however, dopamine neurons exhibited significant depressions at the usual time of reward (9 of 14 neurons) and in
addition were activated by reward at the new time (8 of the 14 neurons) (Fig. 6b). When reward was delivered at 0.5 s earlier than usual, i.e. 0.5 s after lever touch, dopamine neurons were activated by reward at that time, but showed no depression at the usual time of reward (6 of 8 neurons).
Fig. 4. Changes of average population response (54 neurons tested) to reward during learning. Trials with familiar pictures are also shown for comparison. Note the absence of response to reward with familiar pictures, strong activations during initial learning trials before reaching the criterion and progressive decrease after the criterion. Learning data are from episodes with at least 20 correct trials. Average population activity is shown for the first five trials (familiar pictures, top panel), for the total set of correct trials before criterion (second panel) or for sets of 5 consecutive correct trials at different stages (first to fifth, sixth to tenth, etc.) after criterion was reached (bottom four panels).
306
Correct familiar trials 14
Correct learning trials before criterion
15 after criterion
610
1115
1620
Reward
Discussion The activations of individual dopamine neurons by reward were inversely related to the progress of learning in a given learning episode, progressively decreasing as stimuli were learned and vanishing as performance was consolidated. The rapid change of responsiveness to reward observed in these discrete learning episodes indicates that dopamine neurons have a considerable degree of flexibility. Learning curves reflect the degree of reinforcer unpredictability14, and dopamine neuron activations seem to reflect this, by encoding the unpredictability of reward occurrence and tracking its decrease as learning progresses. The progressive decrease in responsiveness suggests that dopamine neurons may also code reward unpredictability in a quantitative way. Consistent with this possibility, we observed that reward activations during early trials in individual learning episodes were not as great as responses to unpredicted free liquid; this may reflect the partial (50%) reward prediction inherent in the repeated two-choice learning situation. The activations by unpredicted free liquid that was given independent of the task suggest that unpredicted rewards are reported irrespective of learning context. The results from reward omission at predicted times suggest that the predictions influencing dopamine neurons include not only the occurrence but also the time of reward. Dopamine neurons show a positive response (activation) when a reward is not predicted or when it occurs at an unpredicted time, they
imp/s
articles
Fig. 5. Comparison between progress of learning and neuronal responses to reward. (a) Behavioral performance, at chance during the initial learning period before reaching criterion and improving rapidly after criterion is reached. (b) Responses of dopamine neurons during the same sessions. Firing rate is expressed as a percentage increase over baseline activity for the same neuron (thus, 200% represents 3 x baseline firing), and is averaged for 54 neurons during the time interval 130220 ms after reward. In (a) and (b), differences were significant between blocks of five consecutive correct trials at p < 0.025, ANOVA before criterion versus subsequent blocks; * p < 0.05 familiar versus all learning blocks, post-hoc Fisher test; + p < 0.05 learning before criterion versus all subsequent blocks.
coding of temporal prediction errors may apply to both unconditioned and conditioned appetitive events. Dopamine neurons therefore seem to be sensitive and flexible detectors of errors in reward prediction, signaling not that an appetitive outcome has occurred but that this outcome is different than predicted at this exact moment. Moreover, this signal is updated very rapidly in response to b changing expectations and reward occurrence. This computation is presumably the result of synaptic inputs to dopamine neurons, possibly from the striatum23. The temporal aspects of predictions reflected in dopamine responses correspond well to the temporal components of predictions observed in animal learning experiments6,7. Dopamine responses show several essential characteristics of teaching signals of temporal-difference models of learning2326. After criterion They are increased when the primary reinforcement is unpredicted, they are also Learning increased in response to secondary reinforcers that become associated with the primary reward as a result of learning, and they are reduced when a predicted reinforcement fails to occur. The are uninfluenced when rewards occur at a predicted time, and potential influence of dopamine on synaptic plasticity2729 they display a negative response (depression) when a reward fails to occur at the predicted time, unless the reward has would allow the dopamine neurons of the substantia nigra already been delivered earlier than the predicted time. Like and ventral tegmental area to provide a teaching signal for rewards themselves, a conditioned stimulus that has been assomodifying synaptic transmission in the striatum and frontal ciated with reward will evoke a response, provided that its cortex. This may constitute a neuronal mechanism contributoccurrence is unpredicted. No response occurs, however, if ing to the well established role of dopamine in reward-driven the conditioned stimulus is predicted18. This suggests that the behavior and in appetitive learning 1214.
Percnet reward responses Percent error trials
Lever touch
Reward or no reward
Pictures on
Lever touch
Reward
Fig. 6. Responses of dopamine neurons related to errors in the temporal prediction of reward. (a) Comparison of correct and error trials, showing effect of behaviorally determined reward occurrence. Correct learning trials lead to reward and to neuronal activation (top panel). Behavioral errors lead to reward omission and induce a depression in the same dopamine neuron at the time that reward is normally given (bottom panel). Activity in error trials is aligned to lever touch, which in correct trials would be followed after 1.0 s by reward. (b) Effects of reward timing during familiar trials. Following a correct response, the reward was delivered after 1.0 s (as expected), 1.5 s (delayed) or 0.5 s (early) Activity of a dopamine neuron was depressed when reward was delayed and increased at the new time of delayed or precocious reward (familiar trials). Reward delivery is marked by a longer line; the slight jitter reflects the fact that the interval between the lever press (to which the traces are aligned) and the reward delivery was not controlled with absolute precision (timing varies 8 ms). Pictures on indicates the various times of appearance of pictures, as indicated by small vertical line in each raster. In each panel, original trial sequence is from top to bottom.
nature neuroscience volume 1 no 4 august 1998 307
articles
Only a few neuronal populations other than dopamine neurons are so far known to process reward information. These include the dorsal and ventral striatum3036, subthalamic nucleus 37 , amygdala 38 , dorsolateral prefrontal cortex 39,40 , orbitofrontal cortex41 and anterior cingulate cortex42. Some neurons in these structures are activated during the expectation of rewards and in response to rewards and reward-predicting stimuli. Also, noradrenaline neurons of locus coeruleus respond to a wide variety of attention-inducing stimuli, including rewards and punishments4345, and cholinergic neurons of nucleus basalis are activated by many rewarding and punishing events46,47. Nevertheless, despite the observation that some of these activities depend on reward unpredictability36 or behavioral errors39,42, they do not seem to code a reward prediction error in the way that dopamine neurons do; moreover, noradrenergic and cholinergic neurons differ from dopamine neurons in that rewards are not their strongest stimuli. Thus dopamine neurons display unique response characteristics, which can be conceptualized as the coding of temporal-prediction errors according to formal learning algorithms.
Methods Experiments were performed on two monkeys (Macaca fascicularis), which were moderately fluid deprived to increase their motivation to work for juice rewards. In the discrimination task, two color pictures (13 13) appeared on a computer screen to the left and right of center (Fig. 1a). The monkey released a key, touched a small lever below one picture within 1 s and, if it selected the correct picture, received 0.15 ml of apple juice 1.0 s later. No reward occurred if no response or an incorrect response occurred. Pictures varied randomly between left and right positions and extinguished upon lever touch. Liquid arrived at the mouth 55 ms after the electronic pulse activated the liquid valve. Trials lasted 46 s; inter-trial intervals were 46 s. In free-liquid trials, animals received 0.15 ml of apple juice at irregular intervals outside of any specific task.
A8, A9 and A10 were marked with small electrolytic lesions and reconstructed from 40-m thick, tyrosine hydoxylase-immunoreacted or cresyl violet-stained coronal brain sections. Experimental protocols conformed to the Swiss Animal Protection Law and were supervised by the Fribourg Cantonal Veterinary Office.
Acknowledgements
We thank Anthony Dickinson, David Gaffan and P. Read Montague for helpful discussions and advice, and B. Aebischer, J. Corpataux, A. Gaillard, A. Pisani, A. Schwarz and F. Tinguely for expert technical assistance. Supported by Swiss NSF, Roche Research Foundation and NIMH postdoctoral fellowship to J.R.H.
RECEIVED 1 APRIL: ACCEPTED 16 JUNE 1998

1. Rescorla, R. A. & Wagner, A. R. in Classical Conditioning II: Current Research and Theory (eds Black, A. H. & Prokasy, W. F.) 6499 (Appleton Century Crofts, New York, 1972). 2. Dickinson, A. Contemporary Animal Learning Theory (Cambridge Univ. Press, Cambridge, 1980). 3. Mackintosh, N. J. A theory of attention: Variations in the associability of stimulus with reinforcement. Psychol. Rev. 82, 276298 (1975). 4. Pearce, J. M. & Hall, G. A model for Pavlovian conditioning: variations in the effectiveness of conditioned but not of unconditioned stimuli. Psychol. Rev. 87, 532552 (1980). 5. Sutton, R. S. & Barto, A. G. Toward a modern theory of adaptive networks: expectation and prediction. Psychol. Rev. 88, 135170 (1981). 6. Smith, M. C. CS-US interval and US intensity in classical conditioning of the rabbits nictitating membrane response. J. Comp. Physiol. Psychol. 66, 679687 (1968). 7. Dickinson, A., Hall, G. & Mackintosh, N. J. Surprise and the attenuation of blocking. J. Exp. Psychol. Anim. Behav. Proc. 2, 313322 (1976). 8. Sutton, R. S. Learning to predict by the method of temporal difference. Machine Learning 3, 944 (1988). 9. Barto, A. G., Sutton, R. S. & Anderson, C. W. Neuronlike adaptive elements that can solve difficult learning problems. IEEE Trans Syst. Man Cybernet. SMC 13, 834846 (1983). 10. Tesauro, G. TD-Gammon, a self-teaching backgammon program, achieves master-level play. Neural Comp. 6, 215219 (1994). 12. Wise, R. A. Neuroleptics and operant behavior: The anhedonia hypothesis. Behav. Brain Sci. 5, 3987 (1982). 13. Fibiger, H. C. & Phillips, A. G. in Handbook of Physiology - The Nervous System, vol IV (ed. Bloom, F. E.) 647675 (Williams & Wilkins, Baltimore, 1986). 14. Robbins, T. W. & Everitt, B. J. Neurobehavioural mechanisms of reward and motivation. Curr. Opin. Neurobiol. 6, 228236 (1996). 15. Mirenowicz, J. & Schultz, W. Importance of unpredictability for reward responses in primate dopamine neurons. J. Neurophysiol. 72, 10241027 (1994). 16. Schultz, W., Dayan, P. & Montague, R. R. A neural substrate of prediction and reward. Science 275, 15931599 (1997). 17. Ljungberg, T., Apicella, P. & Schultz, W. Responses of monkey dopamine neurons during learning of behavioral reactions. J. Neurophysiol. 67, 145163 (1992). 18. Schultz, W., Apicella, P. & Ljungberg, T. Responses of monkey dopamine neurons to reward and conditioned stimuli during successive steps of learning a delayed response task. J. Neurosci. 13, 900913 (1993). 19. Steinfels, G. F., Heym, J., Strecker, R. E & Jacobs, B. L. Behavioral correlates of dopaminergic unit activity in freely moving cats. Brain Res. 258, 217228 (1983). 20. Schultz, W. & Romo, R. Dopamine neurons of the monkey midbrain: Contingencies of responses to stimuli eliciting immediate behavioral reactions. J. Neurophysiol. 63, 607624 (1990). 21. Mirenowicz, J. & Schultz, W. Preferential activation of midbrain dopamine neurons by appetitive rather than aversive stimuli. Nature 379, 449451 (1996). 22. Horvitz, J. C., Stewart, T. & Jacobs, B. L. Burst activity of ventral tegmental dopamine neurons is elicited by sensory stimuli in the awake cat. Brain Res. 759, 251258 (1997). 23. Houk, J. C., Adams, J. L. & Barto, A. G. in Models of Information Processing in the Basal Ganglia (eds Houk, J. C., Davis, J. L. & Beiser, D. G.) 249270 (MIT Press, Cambridge, 1995). 24. Montague, P. R., Dayan, P., Person, C. & Sejnowski, T. J. Bee foraging in uncertain environments using predictive hebbian learning. Nature 377, 725728 (1995). 25. Montague, P. R., Dayan, P. & Sejnowski, T. J. A framework for mesencephalic dopamine systems based on predictive Hebbian learning. J. Neurosci. 16, 19361947 (1996). 26. Suri, R. E. & Schultz, W. Learning of sequential movements with dopaminelike reinforcement signal in neural network model. Exp. Brain Res. (in press).
Animals first learned the discrimination task with fractal pictures as the familiar stimuli. Initially, animals reacted to a single, rewarded picture presented ipsilateral to the working hand, then contralaterally and subsequently in random alternation. Finally, the unrewarded picture was added at gradually increasing sizes until it matched the rewarded picture. After the fractal picture pair was fully learned, animals successively learned many pairs of novel pictures, similar to previous learning studies48. These pairs of pictures were presented together at equal size, varying randomly between left and right locations. The animal had to learn by trial and error which was the correct picture in each pair. Each learning picture was randomly chosen from 65536 possibilities, consisting of one of 64 yellow, red, green or blue alphanumeric symbols superimposed on one of 64 simple forms in yellow, red, green or blue (Fig. 1b). Dopamine neurons were first studied during learning of the discrimination task using fractal pictures. Subsequently, every dopamine neuron was studied in separate, randomly alternating blocks of trials with the familiar picture pair, a novel picture pair and free liquid. As described 15,17,18, activity from single dopamine neurons was recorded extracellularly during 2060 minutes in two monkeys using standard electrophysiological techniques. Dopamine neurons discharged polyphasic, initially negative or positive impulses with relatively long durations (1.85.5 ms) and low frequencies (2.08.5 impulses per s). Impulses contrasted with those of the non-dopaminergic pars reticulata neurons of substantia nigra (7090 impulses per s and < 1.1 ms duration) and neighboring fibers (< 0.4 ms duration). Neuronal activity changes were compared against a 500-ms control period before the first event in each trial by using a Wilcoxon procedure with a constant time window of 130220 ms following liquid reward, comprising 80% of onset and offset times of statistically significant increases 15,17,18 (p < 0.01). Magnitudes of activation were calculated in the constant time window for every neuron tested, independent of response significance. Recording sites of dopamine neurons randomly sampled from groups
308
articles
27. Calabresi, P., Maj, R., Pisani, A., Mercuri, N. B. & Bernardi, G. Long-term synaptic depression in the striatum: Physiological and pharmacological characterization. J. Neurosci. 12, 42244233 (1992). 28. Wickens, J. R., Begg, A. J. & Arbuthnott, G. W. Dopamine reverses the depression of rat corticostriatal synapses which normally follows highfrequency stimulation of cortex in vitro. Neuroscience 70, 15 (1996). 29. Calabresi, P. et al. Abnormal synaptic plasticity in the striatum of mice lacking dopamine D2 receptors. J. Neurosci. 17, 45364544 (1997). 30. 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). 31. Apicella, P., Ljungberg, T., Scarnati, E. & Schultz, W. Responses to reward in monkey dorsal and ventral striatum. Exp. Brain Res. 85, 491500 (1991). 32. Schultz, W., Apicella, P., Scarnati, E. & Ljungberg, T. Neuronal activity in monkey ventral striatum related to the expectation of reward. J. Neurosci. 12, 45954610 (1992). 33. Williams, G. V., Rolls, E. T., Leonard, C. M. & Stern, C. Neuronal responses in the ventral striatum of the behaving monkey. Behav. Brain Res. 55, 243252 (1993). 34. Aosaki, T. et al. Responses of tonically active neurons in the primates striatum undergo systematic changes during behavioral sensorimotor conditioning. J. Neurosci. 14, 39693984 (1994). 35. Bowman, E. M., Aigner, T. G. & Richmond, B. J. Neural signals in the monkey ventral striatum related to motivation for juice and cocaine rewards. J. Neurophysiol. 75, 10611073 (1996). 36. Apicella, P., Legallet, E. & Trouche, E. Responses of tonically discharging neurons in the monkey striatum to primary rewards delivered during different behavioral states. Exp. Brain Res. 116, 456466 (1997). 37. Matsumura, M., Kojima, J., Gardiner, T. W. & Hikosaka, O. Visual and oculomotor functions of monkey subthalamic nucleus. J. Neurophysiol. 67, 16151632 (1992).
38. Nishijo, H., Ono, T. & Nishino, H. Topographic distribution of modalityspecific amygdalar neurons in alert monkey. J. Neurosci. 8, 35563569 (1988). 39. Watanabe, M. The appropriateness of behavioral responses coded in posttrial activity of primate prefrontal units. Neurosci. Lett. 101, 113117 (1989). 40. Watanabe, M. Reward expectancy in primate prefrontal neurons. Nature 382, 629632 (1996). 41. Thorpe, S. J., Rolls, E. T. & Maddison, S. The orbitofrontal cortex: neuronal activity in the behaving monkey. Exp. Brain Res. 49, 93115 (1983). 42. Niki, H. & Watanabe, M. Prefrontal and cingulate unit activity during timing behavior in the monkey. Brain Res. 171, 213224 (1979). 43. Aston-Jones, G. & Bloom, F. E. Norepinephrine-containing locus coeruleus neurons in behaving rats exhibit pronounced responses to nonnoxious environmental stimuli. J. Neurosci. 1, 887900 (1981). 44. Sara, S. J. & Segal, M. Plasticity of sensory responses of locus coeruleus neurons in the behaving rat: implications for cognition. Prog. Brain Res. 88, 571585 (1991). 45. Aston-Jones, G., Rajkowski, J., Kubiak, P. & Alexinsky, T. Locus coeruleus neurons in monkey are selectively activated by attended cues in a vigilance task. J. Neurosci. 14, 44674480 (1994). 46. Richardson, R. T. & DeLong, M. R. Nucleus basalis of Meynert neuronal activity during a delayed response task in monkey. Brain Res. 399, 364368 (1986). 47. Wilson, F. A. W. & Rolls, E. T. Neuronal responses related to reinforcement in the primate basal forebrain. Brain Res. 509, 213231 (1990). 48. Gaffan, E. A., Gaffan, D. & Harrison, S. Disconnection of the amygdala from visual association cortex impairs visual reward association learning in monkeys. J. Neurosci. 8, 31443150 (1988).
309
articles
Inter-trial neuronal activity in inferior temporal cortex: a putative vehicle to generate long-term visual associations
Volodya Yakovlev1, Stefano Fusi2,5, Elisha Berman3 and Ehud Zohary1,4
1 2 3 4
Department of Neurobiology, Institute of Life Science, Hebrew University, Jerusalem, 91904, Israel Racah Institute of Physics, Hebrew University, Jerusalem, 91904, Israel Human Biology Research Center, Hadassah University Hospital, Jerusalem 91904, Israel Center for Neural Computation, Hebrew University, Jerusalem, 91904, Israel INFN, Universit`a di Roma, La Sapienza, Italy Correspondence should be addressed to E.Z. (udiz@lobster.huji.ac.il)
When monkeys perform a delayed match-to-sample task, some neurons in the anterior inferotemporal cortex show sustained activity following the presentation of specific visual stimuli, typically only those that are shown repeatedly. When sample stimuli are shown in a fixed temporal order, the few images that evoke delay activity in a given neuron are often neighboring stimuli in the sequence, suggesting that this delay activity may be the neural correlate of associative long-term memory. Here we report that stimulus-selective sustained activity is also evident following the presentation of the test stimulus in the same task. We use a neural network model to demonstrate that persistent stimulus-selective activity across the intertrial interval can lead to similar mnemonic representations (distributions of delay activity across the neural population) for neighboring visual stimuli. Thus, inferotemporal cortex may contain neural machinery for generating long-term stimulusstimulus associations.
Most of us can remember the next melody on an album once the current tune is over and recall the alphabet in its correct sequential order. These are classic examples of generating an association between stimuli that have been presented in a fixed temporal order. A visual example of such a phenomenon may be the way we navigate an unfamiliar environment. In such circumstances, we typically use remembered snapshots of visual scenes to verify that we are on the correct track, with one serving as a cue to lead us to the next expected landmark. Our current understanding of the neuronal basis for the formation of such long-term associative memories is only marginal. Miyashita and colleagues1 addressed this question in a recent study. Monkeys were trained on a delayed match-to-sample task, in which they remember the identity of a sample stimulus during a delay interval and indicate whether the following test stimulus is (match) or is not (non-match) identical to the sample stimulus. The novel feature in the experimental design was that the sample stimuli were presented in a fixed temporal order. Single neurons in inferior temporal (IT) cortex were recorded during the delay period between the presentation of the sample and test stimuli. Some IT neurons had increased firing rates throughout the delay interval, long after sample-stimulus presentation, as previously reported in IT and prefrontal cortex26. Although the monkey could perform the task with novel stimuli, only highly familiar stimuli evoked this delay activity. The few visual stimuli that generated delay activity in the same IT neuron were usually nearest neighbors in the fixed temporal sequence during the training period, even though the order of the sample stimuli was
310
totally irrelevant to task performance. This aspect of the neuronal response led the authors to suggest1 that the selectivity acquired by these cells represents a neuronal correlate of associative long-term memory of pictures. Based on this observation, a comprehensive theoretical framework for understanding the development of associative longterm memory was proposed7,8. According to this approach, the sustained delay activity is a feature of the pattern of connectivity between neurons, rather than of a single neuron. The persistent delay activity is maintained by recurrent synaptic feedback between interconnected neurons within a local module, built up as stimuli become familiar. The memory process is initiated by presentation of the visual stimulus, which generates a pattern of response across the neuronal population. Following removal of the visual stimulus, because of the feedback connections within the neuronal population, the dynamics of the network is such that it settles into a stable state (the attractor), in which most neurons are firing at their spontaneous level, but some distinct neurons continue firing at elevated levels even though the visual stimulus is no longer present. The stable state implies that this pattern of firing continues until a new afferent input (from a new, effective visual stimulus) changes the state of the network components. Because each visual stimulus evokes a characteristic pattern of delay activity, the delay-activity distribution is the neuronal engram of the last familiar stimulus seen. The distributed nature of the representation allows storage of a large number of patterns (stable delay-activity distributions) in the same neural module, by the same synaptic structure.
articles
Fig. 1. Experimental task, visual a stimuli and cortical area explored in this study. (a) Schematic sequence of events in the delayed match-to-sample task. A trial began with the presentation of a flickering dot (at 1 Hz) at the center of a computer screen. The monkey was required to press a lever in response to the flicker onset. Bar press led to the presentation of the sample stimulus one second later. The test stimulus was b c presented after a fixed ISI (usually five seconds). The test stimulus matched the sample stimulus in half the trials. Both stimuli were presented at the center of the screen. After a variable post-test stimulus interval (5001500 ms), the central dot stopped flickering and turned bright. This served as a go signal for the monkey to shift the bar (two-sided arrow) left if the test stimulus matched the sample stimulus, or right if the two were different, and then to release the lever to get a fruit-juice reward. Both monkeys performed over 90% of trials correctly. A set of 30 color stimuli were presented in a fixed temporal order during the training session. (b) Examples of stimuli used. Top row, fractals; bottom row, Fourier descriptors. (c) A coronal MRI image of the right hemisphere of the brain of one of the studied monkeys. ch, recording chamber; ls; lateral sulcus; sts, superior temporal sulcus; rs, rhinal sulcus. During the imaging, a tungsten electrode was placed at the center of the recording chamber (dark vertical shadow). Its depth corresponds to the approximate location of the tip of the guide tube during recording. The area between the rhinal sulcus and anterior-medial-temporal sulcus recorded during the experiments is marked by triangles.
One key property of such a network is its pattern-completion abilities. The distributed representation across a large neuronal population makes it relatively insensitive to noise. If the pattern of activity during the presentation of a modified or degraded visual stimulus resembles the pattern evoked by the original stimulus, the network will reach the same neuronal delay-activity pattern (the dynamics will flow toward the same attractor). This type of neuronal behavior, IT delay activity that is insensitive to moderate levels of noise in a visual stimulus, has been reported9. It is important to note that the stable attractors are formed during a slow learning process, which shapes the synaptic structure between the network members. Therefore, delay activity should be evident only for stimuli that have been repeatedly presented to the animal1. The memories are embedded in the synaptic structure through an unsupervised Hebbian learning rule. Thus, no special assumptions or requirements are needed to generate the appropriate synaptic structure. Last, and most important, this framework can lead to an association between stimuli repeatedly presented in temporal proximity because the delay activity can link events separated in time. Neurons that are part of an attractor of one stimulus will remain active during the delay period, until the presentation of the next stimulus. This joint activity (within a time window of tens of milliseconds) allows for Hebbian strengthing of the synapses between neurons belonging to the two populations. If the stimuli are systematically presented in a fixed temporal order, this Hebbian learning will eventually lead to similar mnemonic representations (patterns of firing rates) for the two stimuli, forming an associative memory. According to this view, sustained activity for sets of neighboring stimuli presented in a fixed temporal sequence is a single-neuron manifestation of this association, which can only form if the memory trace following one stimulus is maintained across
the intertrial interval (ITI). Theoretical considerations predict that the sustained activity following a specific stimulus will be evident during the ITI as in the interstimulus interval (ISI) because the activity evolves automatically, in a mechanical fashion, irrespective of the behavioral relevance of the stimulus. Because the sample and test stimuli are identical in half the trials in the delayed match-to-sample task, this propagation of stimulus-selective activity during the ITI could transmit information about the temporal order of the sample stimuli. Here we report that IT neurons indeed have a stimulus-selective sustained activity during the ITI. We use a simulation of a large network of (integrate and fire) neurons to illustrate the development of this sequence of events. In this simulation, the sustained activity during the ITI generates temporal correlations in the delay activity, as reported by Miyashita1.
Results Figure 1 illustrates the sequence of events and stimuli used in this study. We recorded the activity of 314 visually responsive cells in IT cortex. Twenty-three neurons (7.3%) showed stimulus-selective delay activity. Figure 2a and b shows the responses of two such neurons to three different stimuli presented as the sample stimulus or as the test stimulus. Note that the rasters and histograms on the left and right of the figure are not temporally contiguous because we were interested in the neuronal activity elicited by a specific stimulus, when presented as a test stimulus, irrespective of whether it matched or did not match the sample stimulus. Stimulus #14 elicited the most vigorous firing from the IT neuron shown in Fig. 2a, which had highly selective activity, during and after its presentation. Note that the delay activity is evident after both the test stimulus and the sample stimulus. Furthermore, the delay activity evoked by stimulus #14 as the test
311
articles
stimulus continued throughout the ITI, which lasted six to seven seconds, and was evident until the presentation of the next sample stimulus in the following trial (Fig. 2a). The few trials in which activity could be seen in the pre-sample intervals are all cases when the test stimulus in the previous trial was stimulus #14. (Individual rasters are marked by different levels of the shaded area.) Thus, the conventionally defined spontaneous activity is affected by the identity of the last stimulus seen. We therefore
analyzed the activity prior to the sample stimulus according to the identity of the previous test stimulus (Fig. 2a and b). The cell shown in Fig. 2b has a more widely distributed selective delay activity, which is maintained throughout the ITI following a specific test stimulus until the presentation of the next sample stimulus. The scatterplots in Fig. 2c and d show the delay activity in the last period of the ITI (stimulus activity before next sample) as a function of the delay activity in the first interval of the ITI (post-
STIM 8
STIM 1
Spikes per second
STIM 24 Spikes per second
STIM 29
STIM 14
Fractals
S 1s
S next
S 1s
S next
Fig. 2. Stimulus-selective sustained activity d c in the ITI. (a) and (b) show two example neurons with sustained activity throughout the ITI. Note that the different intervals within the trial are sorted and organized according to the identity of the corresponding stimulus. Consequently, the number and order of the rasters for the sample and test stimuli are not in register. The data shown during and following the test stimulus are combined for the same test stimulus across match and non-match conditions. S, sample stimulus; T, test stimulus; S next, sample stimulus in the following trial. (a) A neuron with highly selective delay activity. Post-test stimulus activity (spikes/s) Post-test stimulus activity (spikes/s) The right down arrow corresponds to a bar press in the beginning of the next trial, six to seven seconds after the termination of the trial (indicated by broken lines). Note that stimulusselective delay activity is as clear following the test as following sample stimulus #14. Sustained activity following the test stimulus was evident throughout the ITI, until the onset of the sample stimulus of the next trial. Almost all the spikes recorded in the pre-sample period were a result of sustained activity following the presentation of stimulus #14 as a test stimulus in the previous trial. (Corresponding individual rasters are marked by shaded area.) (b) An example of a neuron with more widely distributed selective delay activity. The sustained activity following a specific test stimulus is maintained throughout the ITI until the presentation of the next sample stimulus. (See stimuli #1 and #29 compared to all the fractal stimuli, which do not elicit delay activity.) (c, d) Scatterplots of the average delay activity in the last period of the ITI (one second before the sample stimulus of the next trial), as a function of the average sustained activity in the beginning of the ITI (following the test stimulus). Scatter plots (c) and (d) correspond to the data from the neurons shown in (a) and (b), respectively. Each data point is an average across different presentations of a given test stimulus. (Numerous stimuli did not elicit any activity in the cell depicted in (a), and therefore the data point at the origin of (c) represents multiple stimuli.) The + sign denotes the response to the best stimulus (#14 in c, #1 in d), and the X symbols depict the response to the ineffective stimuli (#24 and #8 in c) and less effective stimulus (#29) in (d). The diagonal lines indicate points of equal response in the two time epochs.
Pre-next-sample activity (spikes/s)
312
Pre-next-sample activity (spikes/s)
articles
Fig. 3. Sustained activity a measures across a population of IT neurons. (a) A histogram depicting the distribution of the delay selectivity index across the population of selective sustained activity neurons (n = 23). The index is the difference in activity in the end of the ITI, following the best and worst stimuli, divided by the sum of the two responses. The best and worst stimuli were Delay selectivity index chosen according to the activity they evoked during the beginning of the ITI (the post-test period). Maintained selectivity throughout the ITI corresponds to positive values. The average sustained selectivity index was 0.44 (indicated by the arrow). This corresponds to a response more than twofold stronger for the best stimulus compared to the worst stimulus in the end of the ITI. (b) Average neuronal activity for the best and worst stimuli during the ISI and the beginning (post-test) and end (before next sample) periods of the ITI. Error bars indicate standard error. Note that the neuronal activity during the ISI was based on the trials in which the best and worst stimuli appeared as sample stimuli, whereas the activity in the ITI was according to the identity of the test stimulus. (c) The difference between the activity elicited by the best and worst stimuli during the corresponding initial and last periods of the ISI and ITI, shown individually for each neuron. The vast majority of the neurons maintain their differential response during the ITI, as in the ISI.
Number of cells
Spikes per second
ISI Time interval
ITI
c
Difference in activity (spikes/s)
ISI
ITI
test stimulus activity) for the neurons shown in Fig. 2a and b, respectively. Each point is an average across all trials with the same test stimulus. To evaluate the reliability of transmission of information across the ITI in the population of delay activity neurons, we compute for each neuron a delay selectivity index: We define the best and worst stimuli as the ones that elicit the strongest and weakest activity, respectively, during the beginning of the ITI (the post-test stimulus period). The delay selectivity index is defined as (Rbest - Rworst)/(Rbest + Rworst) where R is the activity during the last period of the ITI (before the next sample). The delay selectivity index is bounded between the values of [-1,1]. A value of zero indicates that at the end of the ITI there is no difference between the responses to stimuli that elicited a very different response during the beginning of the ITI (i.e., no propagation of information). More positive values indicate the maintenance of the differential response across the ITI. Note that by definition, the best and worst stimuli will elicit a different response in the first period of the ITI, even if the neurons delay activity is not truly stimulus selective (if the difference between the best and worst stimuli is due only to random fluctuations in the response). The crucial point is whether this differential response is maintained throughout the ITI and evident in the last period of the ITI. Histograms of the delay selectivity index for the 23 neurons with selective sustained activity (Fig. 3a) show that their activity tended to continue through the ITI. The average value of the delay selectivity index for this group of neurons was 0.44 (similar to the level of selectivity, 0.52, when the same measure was
applied to the ISI). This index was significantly different from zero (one-group t-test, p < 0.0001). The average activity following the best and worst stimulus for the different time intervals (Fig. 3b) shows that the difference in response is evident in the ISI, in the classical delay period. This difference was also highly significant in the last period of the ITI, before the presentation of the next sample stimulus (paired t-test, p < 0.0001). Finally, a cell-by-cell analysis of the activity in the initial and final periods of the ISI and the ITI (Fig. 3c) demonstrates that in the vast majority of neurons, the difference in response between the best and worst stimulus was maintained across the ITI. In fact the magnitude of the differential response in the end of the ITI (5.0 spikes per s) was almost identical to that at the end of the ISI (4.6 spikes per s). The sustained activity following the best sample stimulus was disrupted if the test stimulus was different from the sample, in accordance with previous findings10. Thus, the sustained activity depended on the identity of the last stimulus seen, whether it was a sample or test stimulus. There was a strong positive correlation between the visual response to the sample stimulus and the delay activity in the following ISI, when the average activity for each stimulus was considered (average Pearson r = 0.69, n = 23). However, the correlation between the activity during the presentation of the best sample stimulus and the ISI delay activity on a trial-bytrial basis was much weaker. In fact, the visual response and the activity in the last second of the ISI were generally uncorrelated (average Pearson r = 0.10). The difference in sustained activity (between the best and worst stimulus) was greater during the post-test period (13.66 spikes per s) than during the corre313
articles
SPN 19
Sample
Test
SPN 20
Spikes per second
ISI
Post-test
Pre-next-sample
SPN 21
Serial position number
S 1s
S next
Fig. 4. Clustering of delay activity to neighboring stimuli in a fixed sequence. (a) Raster displays and spike density histograms of one neuron for three consecutive stimuli of the thirty stimuli that were presented in a fixed temporal order during the training stage. SPN denotes the serial position number of the stimulus in the training sequence. All other aspects of the data are presented in the same way as in Fig. 2b. (b) A histogram depicting the response of the same neuron during the presentation of the sample and test stimuli (top) as well as the sustained activity (bottom) during the ISI and the two ends of the ITI (post-test-stimulus period, and before next sample period) as a function of the SPN. Similar clustering of the sustained activity according to the SPN is observed in all time epochs.
sponding period in the ISI (10.34 spikes per s), but this difference was not evident at the end of the two intervals (5.0 versus 4.6 spikes per s, respectively). The response in the post-test period was also usually somewhat attenuated when the test stimulus matched the sample stimulus (13.92 versus 17.71 spikes per s for the best stimulus in the match versus non-match conditions), but again it was not significantly different at the end of the ITI (7.80 versus 8.05 spikes per s, respectively). In summary, although the delay activity immediately following a specific stimulus depended on the magnitude of the visual response, which could vary from trial to trial for different reasons, the final level of delay activity (a few seconds later) was constant. All these pieces of evidence support the suggestion that the delay activity is a result of the neural network properties, rather than a change in the state of the single neuron alone, triggered by the visual response (see also ref. 9 and below). Is there a functional role for the propagation of delay activity across the ITI? We suggest that it may allow the generation of sustained activity for neighboring stimuli that are repeatedly shown in a fixed temporal order. Indeed, the few stimuli that evoked sustained activity (example in Fig. 4) were often neighboring stimuli, as previously reported1. Clustering of delay activity according to the serial position number of the stimulus is obvious both in the ISI and in the two ends of the ITI. Such a context-dependent associative memory is formed in three stages according to the attractor model. In the first stage, the uncorrelated (context-independent) attractors build up. In the second stage, information about activity patterns is propagated from trial to trial by the sustained activity in the ITI, which leads to the buildup of correlations between stimuli from
314
consecutive trials. In the third stage, the pattern of connectivity is such that the representation of every stimulus reflects its temporal context: delay-activity distributions corresponding to neighboring stimuli in the training sequence are more correlated than the ones corresponding to distant stimuli in the training sequence. The ITI-selective activity is an essential building block for the detection and memorization of temporal correlations in the statistics of the flow of stimuli, and it is sufficient to correlate not only the nearest neighbors but also stimuli that are further apart in a temporal sequence. In the next section, following refs 7 and 11, we exemplify the mechanism underlying the formation of the temporal correlations by taking three snapshots of the behavior of the modeled network corresponding to these three stages.
MODEL NEURAL NETWORK

We present a model neural network11 to illustrate how such a context-dependent associative memory can be formed (Methods). To focus on the role of the ITI-selective activity, we expanded the analysis of the dynamics and show the typical behavior of the model neurons during all the learning stages, as they would appear in cortical recording in each interval of the trial (visual response, ISI and ITI). The network is composed of excitatory and inhibitory neurons, represented (for simplicity) by afferent currents and output rates. Each neuron in the network receives three types of input: recurrent excitatory connections from other neurons in the same module; nonselective, excitatory afferents from other areas of cortex; and local, nonselective inhibitory afferents. The statistics of the input currents determine the firing rates as in ref.
articles
External afferents
Sample Sample Test Test Next Next sample Sample
ISI
ITI
ISI
E
1 M
I
2 3
#2
Stage 1 A STAGE 1
#2
#3
SPN
100
... 1
0 100 0
2 3
100
... 1 M
0 100 0
2 3
Stage 3 A STAGE 3
100
...
0 100 0
12. The excitatory neurons in the module belong to subpopulations, each responding (for simplicity) to only one stimulus. Figure 5 shows the development of delay activity in model neurons during the training process, using stimuli that were repeatedly presented to the network in a protocol identical to the one we used in the physiology experiment above. During stimulation (when the sample or test stimuli are shown), an extra current is injected in the subpopulation of neurons responding to the stimulus presented. The elevated activity of these neurons leads to an increase in the activity of the population of inhibitory neurons, which always reflects the global activation of the excitatory population. As a result, activity is depressed in the other subpopulations that are not activated by the stimulus. In the first stage, the strength of the interclass (between subpopulations responding to different stimuli) and intraclass (within subpopulations) connections is randomly chosen, and each neuron shows a stimulus-selective visual response but no delay activity. The joint firing of two neurons activated by the same specific stimulus (for instance, neuron A and a similar neuron from the same group of neurons responding to stimulus #2) leads to the potentiation of the connection between the two. Analogously, the interclass connections tend to be depressed. With enough repetitions of the same stimulus, there are enough potentiated synapses that the network can sustain enhanced activity even after the evoking stimulus has been removed: each neuron in the subpopulation excites the others through the potentiated synapses. At the end of this stage, delay activity distributions
(attractors) are formed for each specific stimulus. This network property appears suddenly and is observed as a stimulus-specific delay activity (shown in stage 2). Because of this stimulus specificity, a positive correlation between the visual response and the following delay activity, as we report here, is expected (see also ref. 4). On the other hand, the sustained activity evoked by one specific stimulus is not affected by fluctuations in strength of the visual response from trial to trial, as we reported above. This is because the delay activity is triggered by the visual response, but it is sustained by the pattern of activity of all the neurons in the same module. The visual response determines the initial condition. All the stimuli that evoke patterns of activity in the same basin of attraction lead to the same final steady state (attractor), irrespective of the fluctuations of individual neurons activity (see also ref. 9). In stage 2, stimulus-selective delay activity exists, but the patterns of delay activity across the population of neurons are initially not overlapping (i.e., each neuron has sustained activity to only one stimulus). The delay activity of a specific subpopulation is triggered by the presentation of the corresponding stimulus and ends with the presentation of a different stimulus. This is because the global inhibition generated by a different visual stimulus is enough to suppress the activity of this subpopulation to its spontaneous activity level. When the test stimulus matches the sample, neuron A will have delay activity following test stimulus (#2), until the presentation of the next stimulus, which generates a visual response in
315
Spikes persecond Spikes per second
Stage 2 A STAGE 2
Fig. 5. Three snapshots of the behavior of the modeled network corresponding to the three stages of the development of associative memory. The left column depicts the scheme of the model network in the three stages. A network module is composed of two interconnected populations: stimulus-selective excitatory neurons and inhibitory nonselective neurons. Both populations receive external excitatory afferents from other modules in the cortex (inset in top left panel). In the excitatory population (shown in left panel, below inset), each circle denotes a subpopulation of neurons selective to a specific stimulus. Populations corresponding to stimuli that are nearest neighbors in the training sequence of length M are arranged in the picture so that they are near in the circular chain, for the convenience of presentation. Only five subpopulations are represented in the picture. Arrows denote synaptic connections. There are essentially two classes of connections: intraclass synapses, connecting neurons responding to one specific stimulus, and interclass synapses, connecting neurons responsive to different stimuli. Thicker arrows denote a higher number of potentiated connections. Two neurons (denoted by A and B) are monitored during trials of the delayed match-to-sample task in which stimuli #2 and #3 (numbered according to their serial presentation order, SPN) are repeatedly presented. (For illustration purposes only, we present ten trials in which the sample and test stimulus #2 are the same.) Their neuronal activity in the different stages is shown in the right column. (The dotted gray lines represent the average activity in each interval.) See text for an explanation of the dynamics.
articles
all neurons of subpopulation #3 (including neuron B). This joint activity (within a time window of less than 100 ms, at the end of the ITI for stimulus #2) allows for Hebbian strengthening of the synapse between the two neurons. The result of this unsupervised learning is that in the final stage (stage 3), the neurons show sustained activity also for the neighboring stimuli of their preferred stimulus, as reported above (Fig. 4; see also ref. 1). The spreading of the delay activity to the neighboring stimuli is limited to a maximal distance of a few stimuli. This is because the inhibition is faster and stronger than excitation. Thus, the dampening effect of the total inhibition becomes dominant whenever the total excitation tends to grow. Moreover, potentiation depends on the joint level of activity of the two neurons. The delay activity is generally weaker than the visual activity to a given stimulus. Therefore, the delay activity of neuron A elicited by the neighboring stimulus (#3) will usually be weaker than the activity evoked by the original stimulus (#2), and the chain reaction will be limited (see refs 7,11,13). (In our case, the parameters are such that the maximal distance is five. This limitation is not obvious in the figure because we show only the immediate neighboring stimuli in the sequence.)
Discussion The most prominent and novel finding reported here is that stimulus-selective delay activity in IT cortex persists across the ITI. We suggest that this propagation of activity across the ITI may serve to generate the synaptic structure required to form correlations between the mnemonic representations (delay-activity distributions) of successive stimuli in a sequential training protocol9. An analogous type of sustained activity that persisted across the ITI is reported in prefrontal cortex14. This activity is not related to eye movements and was seen also in a monkey that was never trained on a memory task, indicating that it evolves automatically. The sub-area within prefrontal cortex where these face-selective neurons are found receives strong input from IT. Stimulus-selective delay activity was considered to encode the memory trace during the ISI1,2,15. Sample-specific delay activity in prefrontal cortex is maintained throughout the trial, even when intervening stimuli are presented, whereas delay activity following the sample stimulus is disrupted by intervening stimuli in IT cortex10,16. These authors concluded that prefrontal cortex may subserve active working memory, whereas IT cortex may contribute to an automatic detection of stimulus repetition. Our results are in agreement with the hypothesis of a passive, automatic memory in IT. One could suggest that the sustained activity following the test stimulus was due to active working memory because the identity (match or non-match) of the test stimulus must be remembered to execute a correct response. However, the stimulus-specific sustained activity following the test stimulus was evident even after the reward, when memory of the stimulus was no longer required. The delay activity also cannot serve as the mnemonic trace of the sample stimulus throughout the trial, as it was disrupted by the presentation of a different test stimulus. Thus, the sustained activity seems to reflect the last familiar stimulus seen, irrespective of its relevance to the behavioral task. Could the sustained activity be a result of unmonitored eye movements? The sustained activity was stimulus specific and reproducible. It occurred both in the ISI, when the animal gazed at the center of screen, as was observed by the video camera, and in the ITI, when the monkey was clearly observed making eye movements. Thus, it is highly unlikely that the delay activity was caused by systematic eye movements following a specific stimu316
lus. Furthermore, eye movements do not influence the delay activity of IT neurons in the ISI4. In what circumstances would such a mechanism have behaviorally observed consequences? It was suggested11 that such activity would be highly effective in a paired association task, in which retrieval from long-term memory of the pair member associated with a given cue is required. Such associations are formed by repeated presentations of the paired associates, and monkeys with lesions of the temporal lobe (rhinal cortex) or the connections between inferotemporal and prefrontal cortex show marked impairment in this task17,18. In such a paired association task, neurons in IT cortex selectively respond to both pictures of the paired associates3. Furthermore, the neuron shown as an example in that study had a similar level of delay activity following the presentation of either of the two paired stimuli as a cue, suggesting that both stimuli now evoke the same pattern of delay activity (i.e. the same attractor). A key requirement for the buildup of an attractor network is that neurons are organized in local groups with similar stimulus specificity or that neurons with similar specificity are preferentially connected. Indeed, neighboring neurons in inferotemporal cortex tend to have similar stimulus preferences, determined by single-unit recording and optical imaging techniques19. A similar model of attractor dynamics was suggested for the generation of invariant face and object recognition in vision. In essence, it suggests that cells in IT cortex respond similarly to objects seen from different viewing angles because usually faces or objects are seen sequentially from different views in a temporal sequence as one is manipulating an object or moving in space20,21. We conclude that the stimulus-selective sustained activity in IT reflects a passive, automatic memory. The persistence of stimulus-selective activity across the ITI may serve as the link to generate associations between neighboring stimuli by modification of the synaptic structure, so that correlations between the neural representations of successive stimuli are formed. Such a scheme of association may be relevant for navigation in an unfamiliar environment. In such circumstances, we usually remember the specific route we have taken, rather than rely on a cognitive spatial map. Navigation in such circumstances relies heavily on remembered snapshots of visual scenes from specific angles, with one cue leading us to the next expected landmark. Interestingly, lesions in parietal cortex typically lead to a failure to grasp the spatial relationships between places (i.e., a failure to generate a cognitive map) with intact landmark recognition22. Temporal lobe lesions in humans, on the other hand, often result in topographical disorientation in novel environments, when landmarks along the route are used23. Furthermore, such topographical agnosia often co-occurs with prosopagnosia (inability to recognize familiar faces)24,25. This paradoxical finding is more easily understood if attractor dynamics in the temporal lobe is the common neural mechanism underlying the two mnemonic functions.
Methods BEHAVIORAL TASK AND VISUAL STIMULI. The activity of single neurons was recorded from IT cortex while monkeys performed a visual delayed match-to-sample task. The monkeys were seated in an isolated experimental chamber with a background illumination of 2 cd per m2. The only objects in front of the monkey were the computer monitor and a video camera. The background luminance of the screen was 12 cd per m2, and the colored images were high-contrast pictures. A set of 30 color stimuli were presented in a fixed temporal order during the training session. Fifteen were fractal stimuli, and the rest were Fourier descriptors.
A NIMALS AND SURGICAL PROCEDURES . Two rhesus monkeys (Macaca mulatta) weighting six to seven kg were used. A head post and a recordnature neuroscience volume 1 no 4 august 1998
articles
ing chamber were implanted above anterior-ventral IT cortex under general anesthesia with nembutal (2530 mg per kg). The monkeys were given antibiotics and analgesics postoperatively and were allowed sufficient time for recovery after surgery. All experiments, MRI tests and surgical preparations were performed in accordance with NIH and Hebrew University guidelines for use of laboratory animals for experiments. ANATOMICAL MRI. We applied magnetic resonance imaging (MRI) using a Biospec 47/40 device (Bruker) to verify the position of the recording chamber relative to the area of interest. A series of coronal T2-weighted images (1315 consecutive 2 mm slices) were recorded covering the whole area of interest in the monkey brain. A tungsten electrode (diameter, 200 m) was inserted through the chamber center above the area explored during the actual recording sessions (Fig. 1c). This area was between the rhinal sulcus and anterior-medial-temporal sulcus. The images were recorded using a spin-echo sequence with the following parameters: fieldof view of 13 x 13 cm, 256 x 256 data matrix, RARE factor of 8, TR/TE of 3000/23 ms and 8 scans yielding an effective T2-weighted contrast images corresponding to normal spin-echo taken with TE of 70 ms. The monkeys were anesthesized during the imaging session, which lasted about 15 min.
(a = 0.015). Long-term depression (LTD) occurs with a probability p- = 0.2 when one neuron is activated by the stimulus while the other is firing at its spontaneous rate.
Acknowledgements
We thank Gadi Goelman for technical support in the MRI testing, Michail Dvorkin for development of some of the data analysis tools and Nicolas Brunel for help in reproducing his simulations. Robert Shapley pointed out the possible relevance of the connection between prosopagnosia and navigation problems in patients with temporal-lobe lesions. Daniel Amit and Shaul Hochstein commented on earlier versions of the manuscript. This work was supported by grants from the Israel Academy of Science and Israel National Institute of Psychobiology (V.Y.) and a McDonnel-Pew grant for cognitive neuroscience (E.Z).

1. Miyashita, Y. Neuronal correlate of visual associative long-term memory in the primate temporal cortex. Nature 335, 817820 (1988). 2. Fuster, J. M. & Jervey, J. P. Inferotemporal neurons distinguish and retain behaviorally relevant features of visual stimuli. Science 212, 952955 (1981). 3. Sakai, K. & Miyashita, Y. Neural organization for the long-term memory of paired associates. Nature 354, 152155 (1991). 4. Nakamura, K. & Kubota, K. Mnemonic firing of neurons in the monkey temporal pole during a visual recognition memory task. J. Neurophysiol. 74, 162178 (1995). 5. 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). 6. Rao, S. C., Rainer, G. & Miller, E. K. Integration of what and where in the primate prefrontal cortex. Science 276, 821824 (1997). 7. Amit, D. J., Brunel, N. & Tsodyks, M. V. Correlations of cortical hebbian reverberations: Theory versus experiment. J. Neurosci. 14, 64356445 (1994). 8. Amit, D. J. The Hebbian paradigm reintegrated: Local reverberations as internal representations. Behav. Brain Sci. 18, 617657 (1995). 9. Amit, D. J., Fusi, S. & Yakovlev, V. Paradigmatic working memory (attractor) cell in IT cortex. Neural Comput. 9, 10711093 (1997). 10. Miller, E. K., Li, L. & Desimone, R. Activity of neurons in anterior inferior temporal cortex during a short-term memory task. J. Neurosci. 13, 14601478 (1993). 11. Brunel, N. Hebbian learning of context in recurrent neural networks. Neural Comput. 8, 16771710 (1996). 12. Amit, D. J. & Brunel, N. Global spontaneous activity and local structured (learned) delay activity in cortex. Cereb. Cortex 7, 237252 (1997). 13. Griniasti, M., Tsodyks, M. V. & Amit, D. J. Conversion of temporal correlations between stimuli to spatial correlations between attractors. Neural Comput. 5, 19 (1993). 14. Scalaidhe, S. P. O., Wilson, F. A. & Goldman-Rakic, P. S. Areal segregation of face-processing neurons in prefrontal cortex. Science 278, 11351138 (1997). 15. Fuster, J. M. Inferotemporal units in selective visual attention and short-term memory. J. Neurophysiol. 64, 681697 (1990). 16. Miller, E. K., Erickson, C. & Desimone, R. Neural mechanisms of working memory in prefrontal cortex of the macaque. J. Neurosci. 16, 51545167 (1996). 17. Murray, E. A., Gaffan, D. & Mishkin, M. Neural substrates of visual stimulusstimulus association in rhesus monkeys. J. Neurosci. 13, 45494561 (1993). 18. Gutnikov, S. A., Yuan-Ye, M. & Gaffan, D. Temporo-frontal disconnection impairs visual-visual paired association learning but not configural learning in macaca monkeys. Eur. J. Neurosci. 9, 15241529 (1997). 19. Tanaka, K. Inferotemporal cortex and object vision. Annu. Rev. Neurosci. 19, 109140 (1996). 20. Wallis, G. & Rolls, E. T. Invariant face and object recognition in the visual system. Prog. Neurobiol. 51, 167194 (1997). 21. Bartlett, M. & Sejnowski, T. Learning viewpoint invariant representations from visual experience using attractor networks. Network (in press). 22. DeRenzi, E., Faglioni, P. & Villa, P. Topographical amnesia. J. Neurol. Neurosurg. Psychiatry 40, 498505 (1977). 23. Whiteley, A. M. & Warrington, E. K. Selective impairment of topographical memory: a single case study. J. Neurol. Neurosurg. Psychiatry 41, 575578 (1978). 24. Landis, T. Cummings, J. L. Benson, D. F. & Palmer, E. P. Loss of topographic familiarity. Arch. Neurol. 43, 132136 (1986). 25. Maguire, E. A., Burke, T. Philips, J. & Staunton, H. Topographical disorientation following unilateral temporal lobe lesions in humans. Neuropsychologia 34, 9941004 (1996). 26. Petersen, C. C. H., Malenka, R. C., Nicoll, R. A. & Hopfield, J. J. All-or-none potentiation at CA3-CA1 synapses. Proc. Natl. Acad. Sci. USA 95, 47324737 (1998).
RECORDING AND DATA ANALYSIS. Single-unit activity was monitored in four hemispheres of two monkeys using standard recording techniques. Because of technical limitations (data transfer between computers, generation of new stimuli, etc.), neuronal activity was registered during the period between the beginning of the trial (presentation of flickering dot) and bar release. Therefore, the activity during the ITI was monitored in two discrete periods: the post-test stimulus activity (the firing rate between the test stimulus offset and the bar release) and the activity before the next sample, defined as the firing rate in the interval prior to the next sample stimulus, from the presentation of the flickering dot to the sample-stimulus onset. The activity during the ISI was defined as the firing rate in the interval between the sample and test stimuli. The first 200 ms following stimulus offset in the ISI and ITI were excluded to avoid the effects of a possible visual response. Neurons were considered to have a stimulus-specific delay activity if the firing rates for the various stimuli during both the ISI and post-test stimulus period were statistically different using one-way analysis of variance (ANOVA, p < 0.001). DETAILS OF THE MODEL. The parameters are as described11. The statistics of the input currents determine the firing rates as in ref. 12, where the current-to-rate transduction function was calculated for leaky integrateand-fire neurons. The integration time constant for excitatory (inhibitory) neurons is 10 ms (2 ms), and the spike emission threshold is 20 mV above the resting level. Each neuron receives 104 afferents from randomly selected excitatory neurons of the same module, 2 x 103 afferents from the population of inhibitory neurons and an external current from other unspecified areas. The mean synaptic efficacies are chosen in such a way that in the first stage, when the synaptic matrix is still not structured, the average spontaneous activity is 3.0 spikes per s for the excitatory neurons and 4.1 spikes per s for the inhibitory neurons. (The EPSPs are JE to E = 0.035 mV, JE to I = 0.054 mV, JI to E = JI to I = -0.141 mV.) The external mean excitatory current is the same as the mean recurrent excitatory current when all the neurons have spontaneous activity. During stimulation, an extra gaussian current is injected in the neurons of the subpopulation (fraction f = 0.01 of the excitatory neurons in the network) corresponding to the activated stimulus ( = 8.25 mV per ms, 2 = 0.9 mV2 per ms). Only the excitatory synapses in a module are modifiable, and each synapse has two potentiation levels26. The high level (potentiated state) corresponds to a synaptic efficacy that is 4.4 times larger than the low level (depressed state). Synaptic transitions between the two levels depend on the mean rates of the pre- and postsynaptic neurons. Long-term potentiation (LTP) corresponds to the transition between the low level and the high level and occurs with probability p+ = 0.2 if the pre- and postsynaptic neurons are simultaneously activated by the stimulus. (In other words, following each repetition, a mean fraction p+ of the depressed synapses that are connecting active neurons makes a transition to the potentiated state.) If one neuron is activated by a stimulus (high rate) and the other carries selective delay activity elicited by the previous stimulus seen, then potentiation occurs with probability p + = ap +
317
articles
Impaired recruitment of the hippocampus during conscious recollection in schizophrenia

Stephan Heckers1,2, Scott L. Rauch2,3, Donald Goff1, Cary R. Savage2, Daniel L. Schacter4, Alan J. Fischman3 and Nathaniel M. Alpert3
1
The Psychotic Disorders Unit and 2The Psychiatric Neuroimaging Research Group, Department of Psychiatry, Massachusetts General Hospital, Boston, Massachusetts 02114, USA The Positron Emission Tomography Laboratory, Division of Nuclear Medicine, Department of Radiology, Massachusetts General Hospital, Boston, Massachusetts 02114, USA Department of Psychology, Harvard University, Cambridge, Massachusetts 02138, USA Correspondence should be addressed to S.H. (heckers@psych.mgh.harvard.edu)
Poor attention and impaired memory are enduring and core features of schizophrenia. These impairments have been attributed either to global cortical dysfunction or to perturbations of specific components associated with the dorsolateral prefrontal cortex (DLPFC), hippocampus and cerebellum. Here, we used positron emission tomography (PET) to dissociate activations in DLPFC and hippocampus during verbal episodic memory retrieval. We found reduced hippocampal activation during conscious recollection of studied words, but robust activation of the DLPFC during the effort to retrieve poorly encoded material in schizophrenic patients. This finding provides the first evidence of hippocampal dysfunction during episodic memory retrieval in schizophrenia.
Schizophrenia is typically associated with the occurrence of hallucinations and delusions. However, since the original description of the disorder as dementia praecox, cognitive deficits have been recognized as core features of the disease1,2. Inattention and memory impairment are especially enduring symptoms that often do not respond to treatment and contribute to poor prognosis and disability 3. Functional neuroimaging is, arguably, the best tool to investigate the neural basis of the symptoms of schizophrenia4,5. Such studies have demonstrated three patterns of abnormal cerebral blood flow during cognitive activation. First, impairment of working memory and executive functions in schizophrenia have been associated with decreased blood flow in the dorsolateral prefrontal cortex (DLPFC)6. Second, cognitive dysmetria, the inability to receive and process information rapidly, has been associated with a dysfunction of prefrontal-thalamic-cerebellar circuitry7. Third, auditory hallucinations and the experience of psychotic symptoms have been associated with increased blood flow in medial temporal lobe, limbic and subcortical structures8,9. Recent studies have mapped the neuroanatomy of memory to a network of cortical and subcortical structures in the human brain 10,11 . This provides a foundation to test the hypothesis that schizophrenia is associated with perturbations of specific memory components. Two areas of particular interest in the study of schizophrenia, the DLPFC12 and the hippocampal formation13, are involved in the encoding, storage and retrieval of memory. Activation of the DLPFC has been associated with semantic processing during encoding and with the effort of retrieval14,15. Hippocampal activation has been associated with the detection of novelty, with the creation of
318
associations during encoding and with the experience of conscious recollection1416. We used a recently developed positron emission tomography (PET) experimental design14 to study prefrontal and hippocampal function during episodic memory retrieval in schizophrenia. Prior to scanning, subjects studied a list of written words for either (a) a shallow encoding task that required counting the number of T-junctions in the letters of each word, which usually results in poor subsequent memory for those words (low recall), or (b) a deep encoding task that required counting the number of meanings for each word, which usually produces robust subsequent memory for those words (high recall). Subjects were then scanned and tested with a stem-cued recall test, in which they were given three-letter word beginnings and were asked to retrieve studied words. This allowed us to contrast the effort of retrieval with the process of successful retrieval. Based on previous studies11,14,17,18, we predicted that activation of the prefrontal cortex would correspond with the effort of retrieval and that hippocampal and parahippocampal activation would occur during successful retrieval of memory. Consistent with our hypothesis, we found the predicted pattern of regional brain activation during memory retrieval in the control group. In contrast, the schizophrenic patients recruited the prefrontal cortex during the effort of retrieval but did not recruit the hippocampus during conscious recollection. This pattern of activation was associated with higher accuracy during low recall and lower accuracy during high recall in schizophrenics than in control subjects.
Results
BEHAVIORAL DATA
We analyzed the effects of group, condition and run on recall accuracy with a repeated-measures ANOVA using subject as
articles
Table 1. Brain regions showing significantly increased activity during low recall in control subjects and schizophrenic patients and significant differences between groups.
Low Recall Minus Baseline Region (Brodmann areas) Z score Control subjects L Prefrontal (8) R Prefrontal (8) R Precuneus (31) Schizophrenic patients R Prefrontal (11) R Prefrontal (10) R Prefrontal (9) R Parietal (7) R Thalamus Between-group comparisons Control subjects > Schizophrenic patients no regions Schizophrenic patients > Control subjects R Inferior Temporal (20) 3.50 3.46 3.18 Coordinates -28, 14, 40 42, 18, 44 20, -62, 12
patients were very close (case 6, 0.38; cases 7 and 9, 0.40) to the mean recall accuracy score of the patient group (0.35).
PET DATA
4.38 4.36 3.65 4.07 3.48 3.10
18, 42, -12 28, 56, 0 2, 54, 8 24, 32, 36 36, -64, 44 4, -18, 4
3.18
30, -36, -16
random effect. As expected, deep encoding resulted in better recall than shallow encoding in both groups, that is, they remembered many more words during the high-recall task (controls, 0.76; schizophrenics, 0.61) than during the low-recall task (controls, 0.28; schizophrenics, 0.35; main effect of condition, F(1,18) = 244.0, p < .0001). The increment in recall accuracy was significantly different between the two groups as revealed by a significant group-by-condition interaction (F(1,18) = 21.4, p < .0001). This was due to lower accuracy of the schizophrenic patients compared to controls during high recall (t-test, p = .009) and higher accuracy of the schizophrenic patients compared to controls during low recall (t-test, p = .02). It is important to note that, although the magnitude of the increase was different for the two groups (t-test, p < .001), both groups showed significantly more accuracy in high recall than in low recall (paired t-tests, control group, mean percent change, 0.48, p < .0001; schizophrenia group, 0.26, p<.0001). No other main effect or interaction was significant. To test whether the schizophrenic patients were more accurate than controls during low recall because they failed to follow the shallow, perceptual encoding strategy, we compared the accuracy of counting the correct number of T-junctions (for each set of 20 words). The correct number of T-junctions varied between 0 and 7 per word. Accuracy scores for counting T-junctions in controls (mean SD, 0.89 0.06; range, 0.820.98) and schizophrenic patients (0.78 0.18; 0.450.98) were not significantly different, but there was a trend toward a group difference (unpaired t-test, p = 0.11). To investigate this trend further, we studied the individual accuracy scores of the thirteen schizophrenic patients for counting T-junctions. The counting of T-junctions was markedly less accurate in three patients (case 6, 0.45; case 7, 0.50; case 9, 0.63) compared with the range of the other ten patients (0.780.98). The recall accuracy scores during the online test unit for these three
Regional brain activation associated with the effort to recall recently studied words was investigated with the contrast of relative regional cerebral blood flow (rCBF) during low recall minus baseline conditions (Table 1). Control subjects showed significant rCBF increases in bilateral prefrontal areas and right precuneus. Schizophrenic patients showed rCBF increases in right prefrontal areas, right parietal cortex and right thalamus. Comparing rCBF changes during the effort to recall previously studied words revealed that activation of only one region, in the right inferior temporal cortex, was greater in schizophrenic patients than in controls; no other differences were significant (Table 1). Regional brain activation during the actual recollection of recently studied words was investigated with the contrast of rCBF during high recall minus low recall conditions (Table 2). Control subjects showed significant rCBF increases in predicted regions of the right hippocampus and parahippocampal gyrus, with additional increases in retrosplenial, occipital and temporal regions. The schizophrenic patients showed no significant rCBF changes in medial temporal lobe structures during conscious recollection, but did show significant activation in right prefrontal cortex. There were several significant between-group differences during conscious recollection. Normal subjects showed significantly greater rCBF increases in the right hippocampus (Fig. 1) and the right superior temporal gyrus, whereas schizophrenic patients showed significantly greater rCBF increases in right prefrontal cortex and in bilateral inferior parietal areas (Table 2). The high-recallminus-baseline contrast revealed less robust hippocampal actiTable 2. Brain regions showing significantly increased activity during high recall in control subjects and schizophrenic patients and significant differences between groups.
High Recall Minus Low Recall Region (Brodmann areas) Z score Control subjects L Retrosplenial (29/30) R Parahippocampal (35/36) R Hippocampal L Occipital (17) R Superior Temporal (22/42) R Occipital (18) Schizophrenic patients R Prefrontal (45) Between-group comparisons Control subjects > Schizophrenic patients R Hippocampal R Superior Temporal Gyrus (22/42) Schizophrenic patients > Control subjects R Inferior Parietal (40) R Prefrontal (10) L Inferior Parietal (40) 3.66 3.12 24, -28, -4 56, -2, 8 3.36 3.27 3.21 3.16 3.13 3.12 Coordinates -8, -50, 20 26, -38, -8 24, -26, -4 -16, -90, 4 52, -4, 8 28, -84, -4
3.20
28, 30, 4
3.60 3.40 3.21
48, -62, 40 26, 48, 24 -40, -46, 44 319
articles
vation (coordinates 24, -28, -4; z = 1.81) in the control group, but between-group comparisons again revealed less hippocampal activation during conscious recollection in the schizophrenia group (coordinates 26, -28, -4; z = 2.38). We further analyzed the finding of poor hippocampal activation during conscious recollection in schizophrenia, using a randomeffects model (see Methods) to compare rCBF during the three conditions between the two groups Fig. 1. PET statistical map comparing the contrast (high recall minus low recall) between control and to retest the group-by-condi- subjects and schizophrenic patients. The PET image is co-registered with an average normal magtion interaction. The control sub- netic resonance image (MRI), transformed to Talairach space. The three slices are at the level of jects showed significantly higher the medial temporal lobe (horizontal level -4 mm, coronal level -28 mm and sagital level 24 mm in rCBF in bilateral frontal areas dur- Talairach space). Compared with the control group, the right hippocampal region (at coordinates ing all three conditions (baseline, 24, -28, -4) was significantly less activated (p < .001, z > 3.09) in the schizophrenia group. low recall and high recall). The schizophrenic patients showed significantly higher rCBF in bilateral temporal, parietal, and occipital areas during all three condiWe had found in a previous study17 that, in comparison with tions. Between-group analysis of the average of all three conyoung adults, elderly control subjects perform less accurately ditions (contrast 1,1,1,-1,-1,-1 and the reverse) revealed higher and recruit farther posterior frontal areas during low recall. rCBF in multiple frontal areas in controls and higher rCBF in The control group in this study, whose age (mean, 40.0 years) occipital, parietal and temporal areas, including the hipwas intermediate between the younger and older adults in our pocampus (at coordinates 24, -28, -4, run 1, z = 6.31, run 2, previous study, also activated farther posterior prefrontal areas. z = 5.25), in schizophrenics. The group-by-condition interacMoreover, they achieved recall accuracy scores during low recall tion (i.e., hippocampal rCBF increased significantly more in similar to those of the elderly adults studied previously. the control group during conscious recollection) was confirmed in the random-effects model, using only one run per MEMORY RETRIEVAL IN SCHIZOPHRENIC PATIENTS condition (contrast high1 - low1 (24, -28, -4) z = 2.75; conDuring the low-recall condition, the schizophrenic patients trast high2 - low2 (28, -32, -8) z = 2.85). showed increased rCBF in right parietal cortex (area 7), in Further analysis of right hippocampal rCBF during low right prefrontal areas (areas 9, 10 and 11) and in the right thalrecall and high recall focused on nine contiguous voxels cenamus. Compared to the activation of one prefrontal area (area tered in coordinates 24, -28, -4, which is the region that was 8) in the control group, the schizophrenic patients activated most significantly different between the two groups. Mean several prefrontal areas that were located farther anterior. The hippocampal rCBF was higher in the schizophrenia group more widespread activation of prefrontal areas might be relatcompared with the control group during baseline and during ed to greater effort by the patients throughout the experiment: low recall (Fig. 2a). Compared to low recall, hippocampal they achieved higher recall accuracy during low recall and rCBF increased significantly (paired t-test, p = .013) during showed more activation of area 10 in the high-recall-minushigh recall in the control subjects (Fig. 2a), whereas the schizlow-recall contrast (Table 2). ophrenic patients showed a nonsignificant decline. HipThe schizophrenic patients were less accurate when they pocampal rCBF increased during high recall in seven of eight were required to use a higher-level semantic encoding and control subjects but in only one schizophrenic patient search strategy and did not show hippocampal recruitment (Fig. 2b). The schizophrenic patient who showed hippocamduring conscious recollection. In contrast, schizophrenic pal rCBF increases was only marginally more accurate at high patients showed significant activation of regions that are assorecall (35% correct) than at low recall (28% correct). ciated with attention (area 40) and retrieval effort (area 10) during conscious recollection (Table 2). The activation of extrahippocampal areas associated with memory retrieval Discussion might represent the effort to compensate for the failed recruitMEMORY RETRIEVAL IN NORMAL SUBJECTS ment of the hippocampus, which produces variable outcomes. The activation of prefrontal areas, precuneus, medial tempoSurprisingly, the schizophrenic patients performed more ral lobe, superior temporal gyrus and visual areas that we accurately than the controls during the low-recall condition. observed during episodic memory retrieval in our control This indicates that they might have used ancillary strategies, durgroup is consistent with previous reports11,14,17,1922. As in our ing either encoding or retrieval, or both, to enhance their perprevious studies14,17, we found that replacing a perceptual with formance. As outlined above, the more widespread prefrontal a semantic encoding strategy increased recall accuracy, and cortex activation in the low-recall-minus-baseline contrast might that recollection of semantically encoded words was associated indicate greater effort during recall in the schizophrenia group, with right hippocampal activation. We have proposed14 that explaining, at least in part, their higher accuracy. prefrontal cortex activation during the low-recall condition We can only speculate about possible differences during encodreflects the effortful aspects of retrieval search and that hiping, as we do not have PET data from the study phase. The analypocampal activation reflects a conscious recollective process.
articles
Controls
Schizophrenics
Another explanation can be derived from our finding of high hippocampal rCBF in all three word-stem-completion conditions. Compared to the control group, hippocampal activity was continuously increased in schizophrenia and was not modulated by environmental contingencies. Such uncontrolled hippocampal activation might improve memory processes that do not normally recruit the hippocampus (e.g., effortful retrieval search, priming) and perturb those that depend on the hippocampus. This explanation is consistent with theoretical models of hippocampal hyperactivity causing abnormal thought processes, hallucinations and delusions in schizophrenia25,26.
rCBF
HIPPOCAMPAL DYSFUNCTION IN SCHIZOPHRENIA

Our finding of hippocampal dysfunction during episodic memory retrieval in schizophrenia complements recent evidence of abnormal hippocampal structure and function in schizophrenia 9,13,2729 . Our study demonstrates that hippocampal hyperactivity is present in schizophrenia not only at rest8,3032 but also during cognitive activation (baseline and low recall) that interferes with the normal recruitment of the hippocampus during memory retrieval. Functional hyperactivity in the setting of structural deficits of the hippocampus in schizophrenia has been interpreted as reduced efficiency of transsynaptic activity31. It is unlikely that hippocampal rCBF in our sample of schizophrenic patients was at a maximum that could not be increased, because area 40 also showed significantly higher rCBF at baseline in the schizophrenia group but still showed significantly greater rCBF increases during task performance. Increased hippocampal activity at baseline and impaired recruitment during episodic memory retrieval might represent the functional correlate of an abnormal corticohippocampal interaction in schizophrenia29,31,33,34.
rCBF
LIMITATIONS AND CONCLUSIONS

Controls Schizophrenics
Fig. 2. Hippocampal rCBF during the three test conditions. (a) Means (and standard errors) of relative rCBF during baseline, low-recall and high-recall conditions from nine contiguous voxels centered in the right hippocampus voxel (24, -28, -4) that was most significantly different between the control and schizophrenia groups. (b) Difference of rCBF values between high recall and low recall in control subjects and schizophrenic patients. Seven out of eight controls, but only one out of thirteen schizophrenic patients, showed an increase in hippocampal rCBF during high recall.
sis of the patients accuracy of counting T-junctions demonstrated that most patients performed well and followed the encoding instructions. However, schizophrenic patients might have used an additional deeper, possibly semantic, strategy to encode words from the low-recall study list. This could have occurred with the subjects awareness (explicit) or without (implicit). The observation that the patients performed poorly when instructed to employ an explicit semantic encoding strategy (high recall) makes an implicit strategy more likely. One possible mechanism is the failure to inhibit the creation of semantic associations during encoding, resulting in better performance at test. This hypothesis is consistent with previous studies showing that schizophrenic patients use automatic rather than voluntary processes to improve memory performance23 and that they have impaired recognition memory with, but not without, conscious recollection24.
The patient sample consisted of middle-aged males with predominantly hallucinatory symptoms and a disease onset before age 40. Previous studies have not found significant correlations of age, chronicity or sex with memory dysfunction in schizophrenia35. However, schizophrenia is heterogeneous, and future studies are necessary to replicate our findings in a larger sample and across the different subtypes of the disease. All patients were chronically treated with typical neuroleptic medication, and four patients were chronically treated with anticholinergic medication. Like others36, we found no correlation between correct response rates and neuroleptic dose in our sample (Spearman rank correlation, high recall rho -.18, p = 0.58; low recall rho .09, p = 0.79). There is no evidence that chronic exposure to typical neuroleptics changes blood-flow patterns in schizophrenia in the temporal lobe37 or in the DLPFC during cognitive activation 6,36 . In fact, it is likely that discontinuation of chronic neuroleptic treatment would have worsened hippocampal function and memory performance in our sample38. We have to consider a salutary effect of neuroleptic treatment on vigilance and attention, which should have improved overall task performance and is an unlikely explanation for the dissociation of prefrontal and hippocampal activation that we found. Our study provides the first evidence of impaired hippocampal function in schizophrenia during episodic memory retrieval. The pattern of increased hippocampal blood flow at baseline and abnormal recruitment during conscious recollection indicates a failure to modulate hippocampal activity
321
articles
based on contingencies of the environment. Hippocampal dysfunction might also be involved in the production of psychotic symptoms in schizophrenia8,9,31,39, contributing to poor prognosis and inadequate response to treatment.
Methods SUBJECTS. Thirteen schizophrenic patients and eight psychiatrically normal control subjects, all without any history of neurological illness, were studied using a recently developed positron emission tomography approach 14. Diagnoses were made using a structured clinical interview40. Eight patients were diagnosed as paranoid-hallucinatory and five as undifferentiated subtype of schizophrenia, according to DSM-IV criteria41.
across groups, rather than the comparison of relative rCBF changes between two conditions across groups as implemented in SPM95. Additional region-of-interest analysis used the individual rCBF values at the peak of hippocampal activation in the SPM, divided by the individual global cerebral blood flow and multiplied by 50.
Acknowledgments
The authors thank Dmitry Berdichevsky, Zakhar Levin, Avis Loring, Steve Weise, Ed Amico and Dana Ruther for technical support. This study was supported by a Dupont-Warren Fellowship (S.H.), a Young Investigator Award from the National Alliance for Research on Schizophrenia and Depression (S.L.R.) and NIMH grants R01MH57915 (D.L.S.) and MH01215 (S.L.R.).
Mean duration of illness was 18.2 6.1 years. Mean scores on the positive, negative and global scales of the positive and negative syndrome scale (PANSS) 42 were 15.2 5.6, 18.9 3.3 and 28.6 4.8, respectively. All subjects were right-handed males, and both groups were matched for age (normal subjects 40.0 6.3 years, schizophrenic patients 41.7 5.8 years). The control subjects had a higher educational status (control subjects 14.9 1.1 years, schizophrenic patients 13.1 2.3 years, p = .05, t-test), but mean parental educational status was not significantly different (control subjects 12.6 2.7 years, schizophrenic patients 11.4 2.1 years, p = .28, t-test). All patients were treated with typical neuroleptics (mean chlorpromazine equivalent dose 377 mg per day, range 50800 mg per day). Four patients were treated with benztropine (12 mg per day; they ranked 1st, 5th, 7th and 11th for mean performance scores in the schizophrenic sample). EXPERIMENTAL DESIGN. Subjects underwent six scans. During scans one and six, subjects were instructed to complete three-letter word stems presented on the computer screen into the first word that came to mind (baseline condition). Two pairs of scans (scans two and three and scans four and five) followed after two offline study sessions. During the study session, subjects were presented with a randomized list of 100 target words, composed of 20 words presented once and 20 words presented four times. The subjects were instructed to count Tjunctions in the target words presented once (perceptual encoding strategy) and to count meanings of the target words presented four times (semantic encoding strategy). We gave very specific instructions before each encoding block to count either T-junctions or the number of meanings of the word presented on the screen. All subjects successfully completed an offline trial to ensure that they were able to follow the instructions. During scanning, the subjects were asked to complete three-letter word stems of words presented either once (low-recall condition) or four times (high-recall condition). The order of the scanned recall sessions (two runs of each condition) was counterbalanced across subjects.
PET SCANNING.

1. Kraepelin, E. Dementia Praecox and Paraphrenia (Livingstone, Edinburgh, 1919). 2. Goldberg, T. E. & Gold, J. M. in Schizophrenia (eds Hirsch, S. R. & Weinberger, D. R.) 146162 (Blackwell Science Ltd, Oxford, 1995). 3. Green, M. F. What are the functional consequences of neurocognitive deficits in schizophrenia? Am. J. Psychiatry 153, 321330 (1996). 4. Andreasen, N. C. Pieces of the schizophrenia puzzle fall into place. Neuron 16, 697700 (1996). 5. Andreasen, N. C. Linking mind and brain in the study of mental illnesses: a project for a scientific psychopathology. Science 275, 15861593 (1997). 6. Berman, K. F., Zec, R. F. & Weinberger, D. R. Physiologic dysfunction of dorsolateral prefrontal cortex in schizophrenia. II. Role of neuroleptic treatment, attention, and mental effort. Arch. Gen. Psychiatry 43, 126135 (1986). 7. Andreasen, N. C. et al. Schizophrenia and cognitive dysmetria: A positronemission tomography study of dysfunctional prefrontal-thalamiccerebellar circuitry. Proc. Natl. Acad. Sci. USA 93, 99859990 (1996). 8. Liddle, P. F. et al. Patterns of cerebral blood flow in schizophrenia. Br. J. Psychiatry 160, 179186 (1992). 9. Silbersweig, D. A. et al. A functional neuroanatomy of hallucinations in schizophrenia. Nature 378, 176179 (1995). 10. Ungerleider, L. G. Functional brain imaging studies of cortical mechanisms for memory. Science 270, 769775 (1995). 11. Fletcher, P. C., Frith, C. D. & Rugg, M. D. The functional neuroanatomy of episodic memory. Trends Neurosci. 20, 213218 (1997). 12. Goldman-Rakic, P. S. & Selemon, L. D. Functional and anatomical aspects of prefrontal pathology in schizophrenia. Schizophrenia Bull. 23, 437458 (1997). 13. Arnold, S. E. The medial temporal lobe in schizophrenia. J. Neuropsychiatry Clin. Neurosci. 9, 460470 (1997). 14. Schacter, D. L., Alpert, N. M., Savage, C. R., Rauch, S. L. & Albert, M. S. Conscious recollection and the human hippocampal formation: evidence from positron emission tomography. Proc. Natl. Acad. Sci. USA 93, 321325 (1996). 15. Dolan, R. J. & Fletcher, P. C. Dissociating prefrontal and hippocampal function in episodic memory encoding. Nature 388, 582585 (1997). 16. Henke, K., Buck, A., Weber, B. & Wieser, H. G. Human hippocampus establishes associations in memory. Hippocampus 7, 249256 (1997). 17. Schacter, D. L., Savage, C. R., Alpert, N. M., Rauch, S. L. & Albert, M. S. The role of hippocampus and frontal cortex in age-related memory changes: a PET study. Neuroreport 7, 11651169 (1996). 18. Nyberg, L., McIntosh, A. R., Houle, S., Nilsson, L.-G. & Tulving, E. Activation of medial temporal structures during episodic memory retrieval. Nature 380, 715717 (1996). 19. Squire, L. R. et al. Activation of the hippocampus in normal humans: a functional anatomical study of memory. Proc. Natl. Acad. Sci. USA 89, 18371841 (1992). 20. Tulving, E., Kapur, S., Craik, F. I. M., Moscovitch, M. & Houle, S. Hemispheric encoding/retrieval asymmetry in episodic memory: positron emission tomography findings. Proc. Natl. Acad. Sci. USA 91, 20162020 (1994). 21. Buckner, R. L. et al. Functional anatomical studies of explicit and implicit memory retrieval tasks. J. Neurosci. 15, 1229 (1995). 22. Schacter, D. L. et al. Neuroanatomical correlates of veridical and illusory recognition memory: evidence from positron emission tomography. Neuron 17, 267274 (1996). 23. Spitzer, M., Braun, U., Hermle, L. & Maier, S. Associative semantic network dysfunction in thought-disordered schizophrenic patients: direct evidence from indirect semantic priming. Biol. Psychiatry 34, 864877 (1993). 24. Huron, C. et al. Impairment of recognition memory with, but not without, conscious recollection in schizophrenia. Am. J. Psychiatry 152, 17371742 (1995).
PET data were acquired with a General Electric-Scanditronix PC4096 15-slice whole body tomograph. Subjects underwent six one-minute scans and inhaled [15O]CO2 gas beginning 30 seconds after the initiation of the task. Subjects performed tasks while viewing a computer screen and responded verbally. PET images were reconstructed with a conventional convolution-backprojection algorithm, corrected for photon absorption, scatter and dead-time effects.
DATA ANALYSIS. Realignment of images and transformation into the standard stereotactic space of Talairach was performed as described 43. Images were smoothed with a two-dimensional Gaussian filter of width 15 mm FWHM. Within-group analyses and initial betweengroup analyses were performed using SPM95 (Wellcome Dept. of Cognitive Neurology, London, UK). Main effects and interactions were assessed with two contrasts (low recall minus baseline and high recall minus low-recall), using t statistics subsequently transformed into normally distributed z scores. Statistical parametric maps were thresholded at an uncorrected p <.001 (i.e., z > 3.09). Further betweengroup analyses were performed with a random-effects model in the SPM95 environment44. The data were modeled with explanatory variables for group and condition (no block effect), with one scan per condition. This allowed the comparison of rCBF in one condition
322
articles
25. Venables, P. H. Hippocampal function and schizophrenia. Experimental psychological evidence. Ann. NY Acad. Sci. 658, 111127 (1992). 26. Krieckhaus, E. E., Donahoe, J. W. & Morgan, M. A. Paranoid schizophrenia may be caused by dopamine hyperactivity of CA1 hippocampus. Biol. Psychiatry 31, 560570 (1992). 27. Dwork, A. J. Postmortem studies of the hippocampal formation in schizophrenia. Schizophrenia Bull. 23, 403421 (1997). 28. Beauregard, M. & Bachevalier, J. Neonatal insult to the hippocampal region and schizophrenia: a review and a putative animal model. Can. J. Psychiatry 41, 446456 (1996). 29. Weinberger, D. R., Berman, K. F., Suddath, R. & Torrey, E. F. Evidence of dysfunction of a prefrontal-limbic network in schizophrenia: a magnetic resonance imaging and regional cerebral blood flow study of discordant monzygotic twins. Am. J. Psychiatry 149, 890897 (1992). 30. DeLisi, L. E. et al. Increased temporal lobe glucose use in chronic schizophrenic patients. Biol. Psychiatry 25, 835851 (1989). 31. Friston, K. J., Liddle, P. F., Frith, C. D., Hirsch, S. R. & Frackowiak, R. S. The left medial temporal region and schizophrenia. A PET study. Brain 115, 367382 (1992). 32. Kawasaki, Y. et al. Regional cerebral blood flow in patients with schizophrenia: relevance to symptom structures. Psychiatry Res. 67, 4958 (1996). 33. Frith, C. D. et al. Regional brain activity in chronic schizophrenic patients during the performance of a verbal fluency task. Br. J. Psychiatry 167, 343349 (1995). 34. Goldman-Rakic, P. S. Working memory dysfunction in schizophrenia. J. Neuropsychiatry Clin. Neurosci. 6, 348357 (1994). 35. Heaton, R. et al. Neuropsychological deficits in schizophrenics. Relationship to age, chronicity, and dementia. Arch. Gen. Psychiatry 51,
469476 (1994). 36. Goldberg, T. E. & Weinberger, D. R. Effects of neuroleptic medication on the cognition of patients with schizophrenia: a review of recent studies. J. Clin. Psychiatry 57 [suppl 9], 6265 (1996). 37. Miller, D. D., Rezai, K., Alliger, R. & Andreasen, N. C. The effect of antipsychotic medication on relative cerebral blood perfusion in schizophrenia: assessment with technetium-99m hexamethylpropyleneamine oxime single photon emission computed tomography. Biol. Psychiatry 41, 550559 (1997). 38. Gilbertson, M. W. & van Kammen, D. P. Recent and remote memory dissociation: medication effects and hippocampal function in schizophrenia. Biol. Psychiatry 42, 585595 (1997). 39. Tamminga, C. A. et al. Limbic system abnormalities identified in schizophrenia using positron emission tomography with fluorodeoxyglucose and neocortical alterations with deficit syndrome. Arch. Gen. Psychiatry 49, 522530 (1992). 40. Spitzer, R. L., Williams, J. B. W., Gibbon, M. & First, M. B. Structured clinical interview for DSM-III-R. (American Psychiatric Press, Washington, DC, 1991). 41. Diagnostic and Statistical Manual of Mental Disorders, 4th edition. pp 273290 (American Psychiatric Association, Washington, DC, 1994). 42. Kay, S. R., Fiszbein, A. & Opler, L. A. The Positive and Negative Syndrome Scale (PANSS) for schizophrenia. Schizophrenia Bull. 13, 261276 (1987). 43. Alpert, N. M., Berdichevsky, D., Levin, Z., Morris, E. D. & Fishman, A. J. Improved methods for image registration. Neuroimage 3, 1018 (1996). 44. Woods, R. P. Modeling for intergroup comparisons of imaging data. Neuroimage 4, S84S94 (1996).
323
articles
Genetic influence on language delay in two-year-old children

Philip S. Dale1, Emily Simonoff2, Dorothy V. M. Bishop3, Thalia C. Eley2, Bonny Oliver2, Thomas S. Price2, Shaun Purcell2, Jim Stevenson4 and Robert Plomin2
1 2 3 4
Department of Psychology, University of Washington, Seattle, Washington 98195, USA Social, Genetic and Developmental Psychiatry Research Centre, Institute of Psychiatry, London SE5 8AF, UK MRC Cognition and Brain Sciences Unit (formerly Applied Psychology Unit), Cambridge, CB2 2EF, UK Centre for Research into Psychological Development, University of Southampton, Southampton, SO17 1BJ, UK Correspondence should be addressed to R.P. (r.plomin@iop.bpmf.ac.uk)
Previous work suggests that most clinically significant language difficulties in children do not result from acquired brain lesions or adverse environmental experiences but from genetic factors that presumably influence early brain development. We conducted the first twin study of language delay to evaluate whether genetic and environmental factors at the lower extreme of delayed language are different from those operating in the normal range. Vocabulary at age two was assessed for more than 3000 pairs of twins. Group differences heritability for the lowest 5% of subjects was estimated as 73% in model-fitting analyses, significantly greater than the individual differences heritability for the entire sample (25%). This supports the view of early language delay as a distinct disorder. Shared environment was only a quarter as important for the language-delayed sample (18%) as for the entire sample (69%).
Language development is remarkable both for striking uniformities observed across children, especially in sequence of development, and for substantial variation, especially in rate of development. The causes of eachthe role of genetic and environmental factors and their developmental interactionare the subject of lively controversy13. Here we address the question of the sources of variability. Investigators of normal language development have most often assumed an environmental explanation for individual differences, based on numerous findings of correlations between the quantity and quality of parental language and childrens rate of development46. However, such correlations are equally consistent with explanations based on child-driven influence, i.e., more rapidly developing children elicit different language than more slowly developing ones7, or with explanations based on shared genetic influences on parental and child language. A more striking form of variation is found in cases of specific language impairment (SLI), when children have difficulty with language learning but show no other cognitive deficits; SLI is estimated to affect between three and ten percent of children8. There is much greater concordance for SLI in genetically identical (monozygotic, MZ) twins than in nonidentical (dizygotic, DZ) twins911. These data, together with results from an adoption study12, point to a major genetic component in the etiology of SLI. An important question is whether there is a discontinuity between SLI and normal language development. Do children with SLI manifest a qualitatively distinct clinical syndrome, or are they simply at the extreme lower end of a continuum of language learning13? Genetic data provide a crucial test of this distinction by comparing heritability for variability of language development in the normal range with heritability at the extremes. In the present study, we used behavioral genetic data from a large, epidemiologically recruited sample of young twins to
324
address both questions: first, the magnitude of genetic effects on the rate of development across the entire population, and second, the possibility that the sources of variability at the lower extreme of the distribution are different from those operating in the normal range, which would support the view of language impairment as a distinct disorder. We used the twin method to investigate genetic and environmental influences on a key index of language development in twoyear-olds, productive vocabulary14. Two years is a particularly appropriate age to study vocabulary development because it follows a period of rapid acceleration in the use of words at about 18 months and the beginning use of word combinations at about 20 months. In addition, concern about language development is one of the most common and most serious questions about their child that parents bring to their pediatricians15. Although a substantial proportion of language delays resolve themselves in the preschool or early school years16, nearly all language impairments in school children are preceded by early language delay17. Surprisingly little is known about the genetic and environmental origins of early language delay; the twin studies cited above included only children four years and older. In addition, accurate characterization of the developmental origins of language disability may help in the selection of appropriate therapeutic targets for interventions that prevent the cascade of cognitive, social and emotional consequences of language problems6,18. Our study is unique in examining the role of environmental and genetic influences on delayed language development in an epidemiologically defined sample, rather than relying on clinical referral to identify cases. From a sample of 6,862 two-year-old twins assessed for productive vocabulary, children in the bottom 5% of the distribution were selected and compared to the rest of the distribution using vocabulary as a quantitative trait measure. This design makes it possible to apply a quantitative genetic technique
articles
that assesses genetic and environmental influences at the extreme of the distribution and compares them to the causes of variation throughout the distribution. This technique, called DF extremes analysis after its developers DeFries and Fulker19,20, addresses the fundamental issue of the etiologic links between the abnormal and normal. That is, to what extent are the causes of disorders qualitatively different from the influences on normal variation? Alternatively, are disorders merely the lower tail of the distribution for the same genetic and environmental factors that affect individual differences throughout the normal range of variation? Indeed, children with the slowest vocabulary development could reflect a mixture of children at the lower tail of the population distribution as well as children whose low performance represents a qualitatively different impairment. Because heritability is a population statistic, it cannot by itself distinguish the two groups. However, when specific genes are identified, distinct sets of genes may be associated with the two forms of variability and might be used to distinguish the two groups of low-performing children.
Results The sampling frame for the present study, called the Twins Early Development Study (TEDS), consisted of all twins born in England and Wales in 1994. Parents of 7756 pairs of twins identified from their childrens birth records were contacted by the Office for National Statistics after checking for infant mortality when the children were one year old. The project was explained briefly and parents were requested to indicate whether they would like to learn more about the project. Parents returned 5443 reply cards requesting further information. These parents were sent a booklet explaining the project in greater detail and asking for background information about the twins and the family. The background booklets were returned by 4688 families. Shortly before the childrens second birthday, these parents were sent test booklets for each twin, which included the vocabulary measure described below. Both background and test booklets were returned by 3442 families. Those who completed the two-year booklets did not differ substantially from those who did not in terms of maternal age, ethnicity, education, employment and other variables (Table 1). However, given these large sample sizes, significant differences were observed for nearly all comparisons, indicating that parents who completed both booklets were
better educated and employed and less likely to be divorced or to have been a teenager at the birth of their first child. To assess the representativeness of this sample, maternal ethnicity and education were compared to mothers from 1994 census data from the Office for National Statistics. Of our sample, 94% were described as white, as compared to 92% in the census data. At least A levels (see note 2, Table 1) were achieved by 34% of mothers in our sample, as compared to 32% for the census data. The lowest 5% vocabulary group was also comparable to others completing both booklets (Table 1). The largest difference for the lowest 5% group is that their fathers had less education. A parent-rated instrument was used to assign twin zygosity for 95% of same-sex pairs, a rate that is typical of other studies21. We excluded 70 pairs (from whom we are currently obtaining DNA) whose zygosity was uncertain based on the parent-rated instrument. Also typical of other studies, a check of our zygosity assignments for a random sample of 68 same-sex pairs of twins using DNA obtained from cheek swabs22 indicated an accuracy of 96%. We sequentially excluded a total of 333 pairs for other reasons: 41 pairs in which at least one twin had a hearing problem or specific medical syndrome such as Downs syndrome and other chromosomal anomalies, cystic fibrosis or cerebral palsy; 46 pairs who were extreme outliers for birth weight, time spent in hospital, special care after birth, gestational age or maternal alcohol consumption during pregnancy; 116 pairs for whom English is not the basic language spoken in the family; 119 pairs for whom the booklets were returned or completed six months or more after the twins second birthday and 11 pairs for whom one or other twin did not have a valid vocabulary score. As expected, in the group excluded for medical problems (n = 41) or perinatal problems (n = 46), proportionately more would have been included in the lowest 5% group (20.7% and 17.4%, respectively), and those excluded had substantially lower vocabulary scores than the rest of the sample (35.6 and 34.5, respectively, as compared to 48.1 for the rest of the sample). The final sample consisted of 3039 pairs: 1044 pairs of monozygotic (MZ) twins, 1006 pairs of samesex dizygotic (DZ) twins and 989 pairs of opposite-sex DZ twins. Productive vocabulary was assessed using an adaptation of the MacArthur Communicative Development Inventory (MCDI)14. In the MCDI, parents report on their childrens production of root words (e.g. dog, game, ear, chase, gentle, this, all and if) and
Table 1. Comparisons between families who only returned the first booklet, all families who returned both the first and second booklets and families who returned both the first and second booklet with a child in the lowest 5% vocabulary group
N Mothers age Mother white Mother has A levels2 Mother left school2 Father left school2 Mother employed Father employed Mother managerial job Father managerial job Married Divorced Teeange at birth of first child Number of older sibs First booklet only Mean 95% CI 25 89% 24% 18% 21% 40% 85% 18% 35% 88% 8% 14% 1.7 25-26 87-91 21-27 15-21 18-25 36-43 82-88 14-22 31-39 85-90 6-10 12-17 1.6-1.8 First and second booklets N Mean 95% CI 3303 3441 3318 3318 3016 3393 3180 1340 2846 3337 3337 3303 1802 27 94% 34% 9% 14% 40% 91% 19% 39% 94% 3% 8% 1.6 27-28 93-94 32-35 8-10 13-15 38-41 90-92 17-21 37-41 93-95 2-3 7-9 1.5-1.6 Lowest 5% vocabulary1 N Mean 95%CI 139 146 140 140 124 142 133 50 108 143 143 139 83 27 94% 24% 13% 43% 35% 86% 8% 36% 93% 2% 9% 1.7 26-28 89-97 17-32 8-20 34-52 27-44 79-91 2-19 27-46 88-97 0-6 5-15 1.5-1.9
775 1076 768 768 651 806 707 313 591 789 789 775 667
1based on number of individuals, not twin pairs. 2A levels in the UK involve staying on in the UK
equivalent of high school beyond the mandatory age and successfully completing college entrance examinations. Leaving school without qualifications indicates that a student did not successfully complete high school examinations.
325
articles
effects of sex and age, as is standard in twin research24 because these variables can inflate twin similarity. Total sample Lowest 5% For twin pairs in which at least one N Mean SD N Mean SD member of the pair was in the lowest 5% MZ twins 2088 46.4 25.7 126 4.7 2.5 of MCDI vocabulary, probandwise conDZ twins 3990 48.6 25.4 176 3.9 3.0 cordances (proportion of probands in Males 2953 43.6 25.2 188 4.4 2.6 concordant pairs) between impaired Females 3125 51.9 25.2 114 3.9 3.0 subjects and their twin partners (coMZ males 960 40.5 25.1 90 4.7 2.4 twins) were 81% for MZ twins (75 pairs), 42% for same-sex DZ twins (60 MZ females 1128 51.4 25.0 36 4.7 2.8 pairs) and 42% for opposite-sex DZ DZ males 1004 43.6 24.3 42 4.2 3.2 twins (79 pairs), suggesting substantial DZ females 1008 53.4 25.0 34 3.1 2.9 genetic influence and little influence of Opposite-sex DZ males 989 46.5 25.7 56 4.1 2.6 shared environment. However, twin Opposite-sex DZ females 989 50.9 25.5 44 3.8 3.2 concordances provide only a rough Zygosity x sex ANOVAs for the large total sample indicated significant zygosity (p < .001), sex (p < .0001) index of genetic and environmental and zygosity x sex interaction (p < .001). influence. Data were evaluated in more depth by DF extremes analysis 19,20 , which incorporates quantitative trait scores of the relatives (co-twins) of probands (in this case, subthe childs grammatical development. For the vocabulary meajects in the lowest 5% of MCDI vocabulary), rather than just detersure reported here, parents check the words that they have heard mining whether the condition is present or absent in the relatives their child speak, and the number of positive responses is and assessing concordance for the disorder. The essence of DF summed. The MCDI, developed during two decades, has been extremes analysis is that if the mean difference between the shown to have excellent internal consistency and test-retest reliprobands and the population is due to genetic factors, the lanability as well as concurrent validity14. Validity results include a guage scores of language-disabled subjects and their co-twins will correlation of 0.73 with a standard tester-administered measure be more similar for MZ twin pairs than for DZ twin pairs. of expressive vocabulary for unselected two-year-olds and 0.85 The vocabulary scores of the MZ and DZ probands and their for a language-impaired sample of three-year-olds. Two lists of same-sex co-twins were expressed as standardized deviation units 100 words that predict the full MCDI list of 680 words with very from the mean of the total sample. Probands were 1.7 standard devihigh accuracy (r = .98) have been identified (Fenson, L. et al. Techation units below the mean of the total sample, with similar means nical manual and users guide for the MacArthur communicative for MZ (-1.67) and DZ probands (-1.75). The main finding is that development inventories: Short form versions. unpublished manthe means for MZ and DZ co-twins are -1.51 and -0.95, respectiveuscript, 1997). One of these lists (Form A), with a few minor ly. In other words, MZ co-twins regress only 0.16 standard deviachanges to anglicise items, along with questions about other tion units to the sample mean, whereas same-sex DZ co-twins regress aspects of language not reported here, constituted the language 0.80 standard deviation units, suggesting that genetics contributes measure of the present study (MCDI:UKSF, hereafter referred to substantially to the mean difference between the probands and the as MCDI). Parent report as utilized on the MCDI has the double population. Application of DF extremes model-fitting analysis advantage of cost effectiveness and increased representativeness (Methods) to these data yields a significant estimate of 0.73 (95% of early language and is thus well suited for large-sample studies. confidence interval, CI, 0.381.00) for group-differences heritabiliThe distribution of MCDI vocabulary scores indicated a wide ty for same-sex twins only, indicating that most of the mean differdegree of variability in vocabulary at two years of age, from children ence between the probands and the population can be ascribed to with no recognizable words to children who had already reached genetic factors. This is called group-differences heritability to disthe MCDI ceiling of 100 words. The average number of words protinguish it from the usual heritability estimate, which refers to difduced from the list of 100 words was 48 25 (standard deviation) for ferences among individuals rather than to mean differences between the entire sample and 4.2 2.8 for the lowest 5% (range, 0 to 8 an extreme group and the population. The estimate of group shared words) of the sample, with 61 of these 302 children producing no environmentnongenetic factors that make twins similar, such as recognizable words. Our mean of 48 words at the average age of socioeconomic background or similar treatment by their parents 24 1/2 months is comparable to the mean for singletons at age 21 is 0.18 (00.78 CI) for same-sex twins. Although this estimate is not months (Fenson et al., unpublished manuscript), a three-and-astatistically significant because of the low power of the twin method half-month delay for twins that is consistent with other research23. to assess modest effects of shared environment25, it suggests that Sample sizes, means and standard deviations are shown in Table 2 for the entire sample and for the lowest 5%. For the total sample, about a fifth of the mean difference between the probands and the zygosity (MZ versus DZ) and sex showed slight but significant mean population may be due to environmental factors involved in groweffects favoring DZ twins and females. The difference favoring girls ing up in the same family. The remaining 9% of the difference is due is comparable to previous research using the MCDI14. As a result, to error of measurement and nonshared environmental influences. Twin analyses of individual differences in the entire sample yield proportionately fewer DZ twins and females were included in the quite different results, which suggests that the factors influencing lowest 5%. Because we included twins whose booklets were comlanguage disability differ from those influencing language ability in pleted up to six months after their second birthday, we also explored the normal range. For the entire sample, intraclass correlations were the influence of age. Reducing the sample to include only those who 0.93 for MZ twins (1044 pairs), 0.81 for same-sex DZ twins (1006 responded within three months (n = 2917) yielded highly similar pairs) and 0.74 for opposite-sex DZ twins (989 pairs). These results, twin results. Moreover, age correlated only at 0.14 with MCDI vocabsuggesting modest genetic influence and substantial influence of ulary scores. Nonetheless, for the twin analyses, we regressed out the
Table 2. Sample sizes and MCDI raw score means and standard deviations (SD) for the total sample and for the lowest 5%.
articles
Nonshared environment Shared environment Genes
in group heritability for boys and girls was caused largely by differences in DZ concordances (28% versus 58%), but these were based on only 36 male pairs and 24 female pairs. Interestingly, the much larger sample of opposite-sex DZ twins (79 pairs) yielded an intermediate concordance (42%). Although the possibility of greater group heritability for boys than girls warrants further investigation, for now we attribute the observed difference to chance.
Fig. 1. Variance due to genes, shared environment and nonshared environment for differing cut-offs for language delay and for individual differences in the entire sample (100%).
shared environment, are very similar to the results of a recent twin study of parent reports of individual differences in two-year-old language26. Model-fitting analysis (Methods) for the unselected sample of twins yields an individual differences heritability estimate of 0.25 (0.210.28 CI) for same-sex twins. In other words, about 25% of the variance in vocabulary scores throughout the entire distribution can be attributed to genetic factors. Shared environment accounts for most of the individual differences in vocabulary; the model-fitting estimate is 0.69 (0.650.72 CI) for same-sex twins. In summary, the lowest 5% probands yielded a group-differences heritability estimate of 73% and a shared environment estimate of 18%, whereas for the total sample individual-differences heritability and shared environment estimates were 25% and 69%, respectively. To our knowledge, this is the first case in which a significant difference between group-differences heritability and individual-differences heritability has been demonstrated. When progressively less severe cutoffs were used (Fig. 1), the results for the DF extremes analyses approached the results for the individual-differences analyses (shown in Fig. 1 as 100% of the sample). Heritability decreased and shared environment increased sharply even for a cutoff of 10%. These results suggest that a cutoff of 5% is needed to characterize a genetically distinct disorder of language delay. We examined the effects of birth weight, maternal education and sex as moderators of these group-differences parameter estimates. Using an extension of DF extremes analysis that assesses interactions between group-differences parameters and moderator variables27, no significant interactions were found for birth weight or maternal education for the lowest 5% group. This finding suggests that these variables do not importantly affect our estimates of group-differences heritability and shared environment, although it should be noted that outliers with respect to obstetric factors had been excluded from the sample. Sex differences were explored in both individual-differences and extreme-group analyses. For the entire sample of same-sex pairs, when the model was run separately for the two sexes, the heritabilities were similar: 0.27 (0.210.33 CI) for boys and 0.22 (0.180.27 CI) for girls. Turning to the extremes analysis, the 5% subsample yields little power to detect sex differences when this modest-sized sample is further divided by sex. Nonetheless, using the extension of the DF extremes analysis that assesses interactions, an interaction was found between sex and group heritability (b = .63, p < .05), suggesting sex differences in group heritability. DF analyses conducted separately by sex also revealed greater group differences heritability for boys (0.90, 0.401.0 CI) than for girls (0.40, 00.85 CI), although the overlapping confidence intervals for boys and girls indicate the lack of power to detect such differences. MZ concordances were similar for boys (84%) and girls (72%). The difference
Discussion Limitations of the study include attrition and differences between twins and singletons. Our sample with complete data at one and two years of age included 44% of all twins born in 1994 who survived to their first birthday. Although our sample seems representative in terms of census data for maternal ethnicity and education, it is possible that families who did not respond might yield different results. Concerning the second issue, as indicated earlier, our twins appear to lag about three months behind singletons in vocabulary development. Because such differences between twins and singletons might affect estimates of genetic and environmental influence, we are in the process of obtaining data on the younger siblings of the twins in order to provide a within-family comparison and to include twinsingleton sibling comparisons in our quantitative genetic analyses. A related issue is the possibility that twin partners influence each others language and that this might be greater for MZ twins than for DZ twins. This is unlikely given that the variances of the two types of twins are similar (Table 1), but observational studies could investigate this possibility. Another issue is whether this pattern of results simply reflects a more general cognitive delay, which includes nonverbal as well as verbal delay. Supporting this concern, 22% of the children in the language-delayed group were also in the lowest 5% of the distribution on a nonverbal measure used in TEDS. However, results were highly similar when these children with both verbal and nonverbal delays were excluded from the analysis: group heritability was 0.78 (0.391.00 CI) and group shared environment was 0.16 (00.80 CI). Moreover, for the lowest 5% on the nonverbal measure who were not in the lowest 5% of the language measure, group heritability was only 0.23 (00.53 CI). These results suggest that the high group heritability for language delay is not merely a consequence of group heritability for cognitive delay. The nonverbal measure and its relationship to vocabulary will be the topic of future TEDS analyses. In conclusion, this first study of language delay as indexed by vocabulary at two years of age yields three main findings. First, language delay at two years is highly heritable. Second, language delay is much more heritable than individual differences in language ability in the variation within the range of normal (73% versus 25%). Third, shared environment is much less important for language delay than for normal language ability (18% versus 69%). The first finding replicates and extends previous work on SLI by demonstrating a strong genetic contribution to language delay in children as young as two years. We know from longitudinal studies that many children with language delays at two years do prove to be late bloomers who subsequently catch up with their peers28. We are currently involved in further studies following up on these children to see whether the evidence for genetic contribution is strongest in those with persistent problems. The second finding supports the view of language impairment as a distinct disorder. It also leads to two predictions in relation to molecular genetic research aimed at identifying specific genes29: that genes will be found more readily for language delay than for language ability in infancy and that most genes found for language delay will not be associated with individual differences in normal language ability. The third finding concerning shared environment is surprising
327
articles
because it is reasonable to assume that shared language-learning environments are responsible for variability in language development for the low end of the distribution as well as the rest of the distribution. Although shared environment was largely responsible for individual differences in the normal range of language ability, where shared environment accounted for 69% of the variance, shared environment accounted for only 18% of the variance in children with the lowest vocabulary scores. Nonetheless, this implies that about a quarter of the difference between the low-vocabulary group and the population mean could be eliminated if the salient environmental differences could be identified and removed.
Methods TWIN METHOD. For cognitive abilities, the classical twin method yields results similar to those for the other major design, the adoption method, despite the different assumptions of the two methods30. Although the twin method has recently been robustly defended as the perfect natural experiment31, the method is not without problems. Some problems are conservative from a genetic perspective in that, if true, they make MZ twins less similar than they would otherwise have been, thus lowering estimates of heritability. For example, it has been alleged that the atypical gestation of MZ twins causes increased rates of disorder32, although other studies indicate that this is not the case33. Other problems could go in either direction, such as the fact that two-thirds of MZ twins share the same chorion (outer fetal sack), which can lead to shared infection, shared vasculature and other anomalies of sharing a crowded chorion31. Some problems might inflate heritability estimates, most notably the possibility that MZ twins share more similar postnatal environments than DZ twins, although it seems that this is not usually the cause of their greater phenotypic similarity but rather the consequence of their genetic identity.
DF EXTREMES ANALYSIS. Liability-threshold models are often used to convert
dichotomous diagnostic data to correlations on the assumption of an underlying continuous liability34,35. However, instead of assessing a dichotomy (e.g., diagnosing a disorder as present or absent) and then assuming a continuous dimension, DF extremes analysis assesses the continuous dimension directly. DF analysis was conducted following procedures described elsewhere36. Although scores for MZ and DZ probands were similar in the present study, all scores were standardized and transformed to adjust for group mean differences between MZ and DZ groups. The basic DF model is represented as the regression, C = B1P + B2R + A, in which the co-twins vocabulary score (C) is predicted from the probands vocabulary score (P) and the coefficient of relatedness (R), which is 1.0 for MZ and 0.5 for DZ pairs. Because the proband mean is transformed to a mean of 1.0 and the unselected population to a mean of 0.0, the mean of the cotwins score for MZ and DZ twins estimates their group-differences familiality (total familial similarity). The regression weight B2 estimates group-differences heritability. Group shared environment, twin resemblance not explained by genetic factors, can be estimated by subtracting group-differences heritability from MZ group-differences familiality. INDIVIDUAL-DIFFERENCES MODEL FITTING. Although an extension of DF analysis can be used to analyze individual differences, we applied the standard maximum-likelihood model-fitting analysis for the classical twin design to variance/covariance matrices for same-sex twins and for all twins including opposite-sex DZ as described by Neale and Cardon37. The ACE model (which estimates parameters for additive genetic variance, common or shared environment and environmental influences that are not shared) assumes that genetic effects are additive and that MZ and DZ twins experience equally similar environments.
Acknowledgements
We thank the parents of the twins in the Twins Early Development Study (TEDS) for making the study possible. TEDS is supported by a programme grant from the UK Medical Research Council.
1. Bates, E., Dale, P. S. & Thal, D. in Handbook of Child Language (eds Fletcher, P. & MacWhinney, B.) 96151 (Basil, Blackwell, Oxford, 1995). 2. Elman, J. L. et al. Rethinking Innateness: A Connectionist Perspective on Development (MIT Press, Cambridge, Massachusetts, 1996). 3. Pinker, S. The Language Instinct (William Morrow, New York, 1994). 4. Hampson, J. & Nelson, K. The relation of maternal language to variation in rate and style of language acquisition. J. Child Lang. 20, 313342 (1993). 5. Huttenlocher, J., Haight, W., Bryk, A., Seltzer, M. & Lyons, T. Early vocabulary growth: Relation to language input and gender. Dev. Psychol. 27, 236248 (1991). 6. Hart, B. & Risley, T. R. Meaningful Differences in the Everyday Experience of Young American Children (P. H. Brookes, Baltimore, 1995). 7. Lytton, H. Parent-Child Interaction: The Socialization Process Observed in Twin and Singleton Families (Plenum, New York, 1980). 8. Bishop, D. V. M. Uncommon Understanding: Development and Disorders of Language Comprehension in Children (Psychology Press, Hove, UK, 1997). 9. Bishop, D. V. M., North, T. & Donlan, C. Genetic basis of specific language impairment: Evidence from a twin study. Dev. Med. Child. Neurol. 37, 5671 (1995). 10. Lewis, B. A. & Thompson, L. A. A study of developmental speech and language disorders in twins. J. Speech Hear. Res. 35, 10861094 (1992). 11. Tomblin, J. B. & Buckwalter, P. R. Heritability of poor language achievement among twins. J. Speech Lang. Hear. Res. 41, 188199 (1998). 12. Felsenfeld, S. & Plomin, R. Epidemiological and offspring analyses of developmental speech disorders using data from the Colorado Adoption Project. J. Speech Lang. Hear. Res. 40, 778791 (1997). 13. Leonard, L. in Advances in Applied Psycholinguistics, Volume 1: Disorders of FirstLanguage Development (ed. Rosenberg, S.) 139 (Cambridge University Press, New York, 1987). 14. Fenson, L. et al. Variability in early communicative development. Monogr. Soc. Res. Child Dev. 59, 1173 (1994). 15. Hart, H., Bax, M. & Jenkins, S. Use of the child health clinic. Arch. Dis. Child. 56, 440445 (1981). 16. Rescorla, L., Roberts, J. & Dahlsgaard, K. Late talkers at 2: Outcome at age 3. J. Speech Lang. Hear. Res. 40, 556566 (1997). 17. Thal, D. J. & Katich, J. in Assessment of Communication and Language (eds Cole, K. N., Dale, P. S. & Thal, D. J.) 128 (Paul H. Brookes, Baltimore, 1997). 18. Beitchman, J. H. Cohen, N. J. Konstantareas, M. M. & Tannock, R. Language, Learning, and Behavior Disorders: Developmental, Biological, and Clinical Perspectives (Cambridge University Press, New York, 1996). 19. DeFries, J. C. & Fulker, D. W. Multiple regression analysis of twin data. Behav. Genetics 15, 467473 (1985). 20. DeFries, J. C. & Fulker, D. W. Multiple regression analysis of twin data: Etiology of deviant scores versus individual differences. Acta Genet. Med. Gemellol. 37, 205-216 (1988). 21. Goldsmith, H. H. A zygosity questionnaire for young twins: A research note. Behav. Genet. 21, 257-269 (1991). 22. Freeman, B., Ball, D., Powell, J., Craig, I. & Plomin, R. DNA by mail: The usefulness of cheek scrapings. Behav. Genet. 27, 251257 (1997). 23. Rutter, M. & Redshaw, J. Annotation: Growing up as a twin: Twin-singleton differences in psychological development. J. Child Psychol. Psychiatry 32, 885895 (1991). 24. McGue, M. & Bouchard, T. J. Jr Adjustment of twin data for the effects of age and sex. Behav. Genet. 14, 325343 (1984). 25. Martin, N. G., Eaves, L. J., Kearsey, M. J. & Davies, P. The power of the classical twin study. Heredity 40, 97116 (1978). 26. Reznick, J. S., Corley, R. & Robinson, J. A longitudinal twin study of intelligence in the second year. Monogr. Soc. Res. Child Dev. 249, 1160 (1997). 27. DeFries, J. C. & Gillis, J. J. in Nature, Nurture and Psychology (eds Plomin, R. & McClearn, G. E.) 121145 (American Psychological Association, Washington, DC, 1993). 28. Bishop, D. V. M. in Child and Adolescent Psychiatry (eds Rutter, M., Hersov, L. & Taylor, E.) 546568 (Blackwell Scientific, Oxford, 1994). 29. Fisher, S. E., Vargha-Khadem, F., Watkins, K. E., Monaco, A. P. & Pembrey, M. E. Localisation of a gene implicated in a severe speech and language disorder. Nat. Genet. 18, 168170 (1998). 30. Plomin, R. DeFries, J. C. McClearn, G. E. & Rutter, M. Behavioral Genetics (W.H.Freeman, New York, 1997). 31. Martin, N., Boomsma, D. & Machin, G. A twin-pronged attack on complex trait. Nat. Genet. 17, 387392 (1997). 32. Phillips, D. I. W. Twin studies in medical research: Can they tell us whether diseases are genetically determined? Lancet 341, 10081009 (1993). 33. Christensen, K., Vaupel, J. W., Holm, N. V. & Yashin, A. I. Mortality among twins after age 6: fetal origins hypothesis versus twin method. B. M. J. 310, 432436 (1995). 34. Falconer, D. S. The inheritance of liability to certain diseases estimated from the incidence among relatives. Ann. Hum. Genet. 29, 5176 (1965). 35. Smith, C. Concordance in twins: Methods and interpretation. Am. J. Hum. Genet. 26, 454466 (1974). 36. Saudino, K. J., Plomin, R., Pedersen, N. L. & McClearn, G. E. The etiology of high and low cognitive ability during the second half of the life span. Intelligence 19, 353371 (1994). 37. Neale, M. C. & Cardon, L. R. Methodology for Genetic Studies of Twins and Families (Kluwer Academic Publications, Dordrecht, 1992).
328
corrections
Neurite growth inhibitors restrict plasticity and functional recovery following corticospinal tract lesions
Michaela Thallmair, Gerlinde A.S. Metz, Werner J. ZGraggen, Olivier Raineteau, Gwendolyn L. Kartje and Martin E. Schwab Nature Neurosci. 1, 124131 (1998).
We reported in the June 1998 issue of Nature Neuroscience that the monoclonal antibody IN-1 promotes collateral sprouting in the rat spinal cord, red nucleus and pons following a lesion to the corticospinal tract. This sprouting was accompanied by functional recovery. We also published a paper in the 15 June 1998 issue of Journal of Neuroscience (Vol. 18: 47444757), which described experiments that were performed in parallel in our laboratory and reached similar conclusions; specifically, whereas the Nature Neurosci. paper concentrated on spinal effects, the J. Neurosci. paper employed identical lesions and antibody treatments, and described in detail sprouting in red nucleus and pons as well as functional recovery. Although the datasets are largely unique to each paper, the experimental design and results were very similar. Moreover, some of the data presented in the two papers are also identical (Fig. 5b of the Nature Neurosci. paper and Fig. 4a of the J. Neurosci. paper), while other data represent different time points from the same animals (Fig. 6b of the Nature Neurosci. paper and Fig. 6a of the J. Neurosci. paper). We greatly regret that neither of these papers cites the other, and that we failed to inform the editors of either Nature Neuroscience or Journal of Neuroscience of the existence of another closely related paper that was under consideration elsewhere. We also regret that an important result presented in the J. Neurosci. paper (the effect of a second lesion rostral to the first) was mistakenly described in the Nature Neurosci. paper as W.J.Z., in preparation while it was already in press at Journal of Neuroscience. We apologize to the editors, referees and readers of both Nature Neuroscience and Journal of Neuroscience for these errors, and for any confusion which we may have caused. A similar correction has been submitted to Journal of Neuroscience. Martin E. Schwab, Michaela Thallmair, Werner ZGraggen and Gerlinde Metz.
/neurosci.nature.com
errata
Visual search for motion-in-depth: stereomotion does not pop out from disparity noise
Julie M. Harris1, Suzanne P. McKee2 and Scott N. J. Watamaniuk3 Nature Neurosci. 1, 165168 (1998).
Because of an editorial error, the affiliation given for JMH was incomplete. The correct affiliations are:
1
Address until 30 June 1998: Department of Pharmacology and Centre for Neuroscience, University of Edinburgh, Crichton St., Edinburgh, EH8 9LE, UK Address after 1 July 1998: Department of Psychology, Ridley Building, Claremont Place, University of Newcastle, Newcastle upon Tyne, NE1 7RU, UK
Visual features that vary together over time group together over space
David Alais, Randolph Blake and Sang-Hun Lee Nature Neurosci. 1, 160164 (1998).
The following paragraphs were inadvertently omitted from the results section. The end of that section should read:
(page 162)...More importantly, correlated contrast modulation still produced greater coherence compared to the uncorrelated case even when the two gratings differed in contrast. Contrast modulation of a drifting grating introduces additional temporal frequencies in the stimulus, besides those associated with the smooth drift rate. Because the contrast modulations were purposefully random, the fluctuations in temporal frequency were also random. In the case of our two-component plaid, random contrast modulations effectively modulate the length (but not the direction) of the component vectors, and these random fluctuations in vector length could be perceived as irregularities in speed, or motion jerkiness. For uncorrelated contrast modulations, the changes in vector length would differ for the two components and, therefore, possibly complicate the grouping process. But why would irregular, jerky motion that is correlated among components increase perceived coherence, the result we obtained? One might argue that correlated jerkiness leads the visual system intelligently to deduce that the jerky components belong to the same object. To test this possibility, we created a plaid consisting of two components that did not modulate in contrast. In two conditions, we explicitly selected a new drift-rate every 40 ms from a set of seven ranging from 48 Hz; this caused motion to appear jerky. In one of these conditions, the random fluctuations in drift-rate were correlated, and in the other they were uncorrelated. In a third condition, the component gratings drifted smoothly at a constant rate of 6 Hz throughout the observation period. Observers again tracked periods of global coherence during 60-second observation periods. Jerky motion had no effect on perceived coherence, which was equivalent for all three conditions. Accordingly, we conclude that jerkiness is not responsible for the increased coherence produced by correlated contrast modulations.
Cocaine self-administration in dopamine-transporter knockout mice

Beatriz A. Rocha, Fabio Fumagalli, Raul R. Gainetdinov, Sara R. Jones, Robert Ator, Bruno Giros, Gary W. Miller and Marc G. Caron Nature Neurosci. 1, 132137 (1998).
On page 134, Fig. 4 was inadvertently printed in black and white instead of color. The corrected version is printed below.
Total +Alaproc Cocaine Total +Alaproc Cocaine
Wild type
DAT-/-
330
errata
Presynaptic modulation of CA3 network activity

Kevin J. Staley, Mark Longacher, Jaideep S. Bains and Audrey Yee Nature Neurosci. 1, 201209 (1998).
Because of a printing error, Figs 4c, 5c and 6 were not properly reproduced. The corrected versions are printed below.
4c
Difference from mean (%)
% difference from mean
100
6a
Interburstinterval, sec(s) interburst interval
20 15 10 5 3
interval length
b
interburst interval (s) Interburst interval, sec 3.5 500 ms 2 mV 8.5 500 25 20 15 10 5 0 5 6 7
+ +
-0
-50
250 6 K , mM K oo (mM)
++
110 8 9 10
-100 0 1 2 3 4 5 6 7 8 9 10
Seconds after burst seconds after burst
c C
interburst interval, (s) Interburst intervalsec 30 20
d D
110 120 burst length, ms
K K o mM o (mM)
EPSC frequency (Hz) EPSC frequency, Hz
4 3 2
10
Burst length burst length, ms
70 0 1 2 3 M baclofen M baclofen 4
Burst length (ms)
5c
interval length
(ms)
50 0 1 2 3 4 5 6
e
1
next burst interval Next burst interval
f
1.0
stimulus delay, (s) Stimulus delaysec
1.0
0 0 1 2 3 4 5 6 7 8 9 10 seconds after end of burst Secondsafter end of burst
Burst length burst length
0.5
0.5
0.0 0.0 0.5 burst length Burst length 1.0
0.0 0.0 0.5 1.0
Burst interval burst interval
In addition, the penultimate sentence in the legend to Fig. 4b was inadvertently omitted. The corrected legend should read: Fig. 4. .... (b) The experiment shown in (a) was repeated in a slice in which CA3 bursting was induced by tetanic stimulation of the CA3 pyramidal cell layer. The average interburst interval in this preparation was 13 seconds, which permitted longer delays between the end of a burst and glutamate application. As in (a), glutamate application at delays closer to the interburst interval triggered a burst more rapidly, but the initial response to glutamate was unchanged (inset). In this cell, bursts triggered calcium escape spikes. Glutamate was applied after every fourth burst in (b) and (c).
331
burst length, ms
Burst length (ms)
50
interval length
150
(ms)
1998 Nature America Inc. http://
Two sites of action for synapsin domain E in regulating neurotransmitter release

Sabine Hilfiker, Felix E. Schweizer, Hung-Teh Kao, Andrew J. Czernik, Paul Greengard and George J. Augustine Nature Neurosci. 1, 2935 (1998).
In preparing Fig. 2a, the same presynaptic traces were inadvertently used for all three panels. The correct version, which is virtually identical to our published Fig. 2a, is shown here. The authors regret the error.
2a
Control
s-pepE
Recovery
Also, because of an editorial error, the x-axes in Figs 2b, 2c, 2d, 3a, 3b, 3e and 5c were mislabeled as ms. The labels should read min.
errata
1

Wild type
DAT-/-
330
errata

4c
100
6a
20 15 10 5 3
interval length
b
+ +
-0
-50
250 6 K , mM K oo (mM)
++
110 8 9 10
-100 0 1 2 3 4 5 6 7 8 9 10
c C
d D
K K o mM o (mM)
4 3 2
10
Burst length (ms)
5c
interval length
(ms)
50 0 1 2 3 4 5 6
e
1
f
1.0
1.0
0.5
0.5
0.0 0.0 0.5 1.0
331
burst length, ms
Burst length (ms)
50
interval length
150
(ms)
Cortisol levels during human aging predict hippocampal atrophy and memory deficits
Sonia J. Lupien, Mony de Leon, Susan de Santi, Antonio Convit, Chaim Tarshish, N.P.V. Nair, Mira Thakur, Bruce S. McEwen, Richard L. Hauger and Michael J. Meaney Nature Neurosci. 1, 6973
One of the references in this manuscript is incorrect. Reference 18 should read: 18. Convit, A. et al Specific hippocampal volume reductions in individuals at risk for Alzheimers disease. Neurobiol. Aging 18, 131138 (1997).
329
errata
1

Wild type
DAT-/-
330
errata

4c
100
6a
20 15 10 5 3
interval length
b
+ +
-0
-50
250 6 K , mM K oo (mM)
++
110 8 9 10
-100 0 1 2 3 4 5 6 7 8 9 10
c C
d D
K K o mM o (mM)
4 3 2
10
Burst length (ms)
5c
interval length
(ms)
50 0 1 2 3 4 5 6
e
1
f
1.0
1.0
0.5
0.5
0.0 0.0 0.5 1.0
331
burst length, ms
Burst length (ms)
50
interval length
150
(ms)
corrections

2a
Control
s-pepE
Recovery
329
errata
1
errata

4c
100
6a
20 15 10 5 3
interval length
b
+ +
-0
-50
250 6 K , mM K oo (mM)
++
110 8 9 10
-100 0 1 2 3 4 5 6 7 8 9 10
c C
d D
K K o mM o (mM)
4 3 2
10
Burst length (ms)
5c
interval length
(ms)
50 0 1 2 3 4 5 6
e
1
f
1.0
1.0
0.5
0.5
0.0 0.0 0.5 1.0
331
burst length, ms
Burst length (ms)
50
interval length
150
(ms)
corrections

2a
Control
s-pepE
Recovery
329
1998

Wild type
DAT-/-
330
errata

4c
100
6a
20 15 10 5 3
interval length
b
+ +
-0
-50
250 6 K , mM K oo (mM)
++
110 8 9 10
-100 0 1 2 3 4 5 6 7 8 9 10
c C
d D
K K o mM o (mM)
4 3 2
10
Burst length (ms)
5c
interval length
(ms)
50 0 1 2 3 4 5 6
e
1
f
1.0
1.0
0.5
0.5
0.0 0.0 0.5 1.0
331
burst length, ms
Burst length (ms)
50
interval length
150
(ms)
errata

4c
100
6a
20 15 10 5 3
interval length
b
+ +
-0
-50
250 6 K , mM K oo (mM)
++
110 8 9 10
-100 0 1 2 3 4 5 6 7 8 9 10
c C
d D
K K o mM o (mM)
4 3 2
10
Burst length (ms)
5c
interval length
(ms)
50 0 1 2 3 4 5 6
e
1
f
1.0
1.0
0.5
0.5
0.0 0.0 0.5 1.0
331
burst length, ms
Burst length (ms)
50
interval length
150
(ms)

Nature Neuroscience August 1998

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Nature Neuroscience August 1998

Загружено:

Авторское право:

Доступные форматы

1998 Nature America Inc. http://neurosci.nature.

1998 Nature America Inc. http://neurosci.nature.com

news and views

NPY and agouti-related protein colocalized in hypothalamus. Page 271

nature neuroscience volume 1 no 4 august 1998

1998 Nature America Inc. http://neurosci.nature.com

1998 Nature America Inc. http://neurosci.nature.com

corrections and errata

see page 329

nature neuroscience volume 1 no 4 august 1998

1998 Nature America Inc. http://neurosci.nature.com

1998 Nature America Inc. http://neurosci.nature.com

1998 Nature America Inc. http://neurosci.nature.com

1998 Nature America Inc. http://neurosc

A new voice for European neuroscience

nature neuroscience volume 1 no 4 august 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

Collision-avoidance: natures many solutions

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 4 august 1998

h(t) (arbitrary units) (t) (arbitrary units)

Angular size (deg) Angular size (deg)

1998 Nature America Inc. http://neurosci.nature.com

news and views

Angular size (deg) Angular size (deg)

Angular speed (deg/s) Angular speed (deg/sec)

-3 -2 -1 0 Time to collision (s) Time to collision (sec)

0 0 -3 -2 -1 Time to collision (s)

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 4 august 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

1998 Nature America Inc. http://neurosci.nature.com

Neurogenic control of the cerebral microcirculation: is dopamine minding the store?

nature neuroscience volume 1 no 4 august 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

1998 Nature America Inc. http://neurosci.nature.com

Cortical afferents CAT, 5-HT, EPI, NEPI, DA

1998 Nature America Inc. http://neurosci.nature.com

news and views

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 4 august 1998

1998 Nature America Inc. http://neurosci.nature.com

news and views

The missing link: a failure of fronto-hippocampal integration in schizophrenia

1998 Nature America Inc. http://neurosci.nature.com

1998 Nature America Inc. http://neurosci.nature.com

news and views

1998 Nature America Inc. http://neurosci.nature.com

nature neuroscience volume 1 no 4 august 1998

1998 Nature America Inc. http://neurosci.nature.com

scientific correspondence Coexpression of Agrp and NPY in fasting-activated hypothalamic neurons

1998 Nature America Inc. http://neurosci.nature.com

1998 Nature America Inc. http://neurosci.nature.com

1998 Nature America Inc. http://neurosci.nature.com

RECEIVED 15 APRIL: ACCEPTED 15 JUNE 1998

nature neuroscience volume 1 no 4 august 1998

1998 Nature America Inc. http://neurosci.nature.com

A role for NMDA-receptor channels in working memory

1998 Nature America Inc. http://neurosci.nature.com

Membrane Potential (mV) Membrane potential (mV)

-40 Active Inactive -80 0 -6 -12 100 Active Inactive