Nature Neuroscience July 1998

nature neuroscience • volume 1 no 3 • july 1998 i

editorialMaking sense of channel diversity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

news and viewsWith color in mind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171Charles Heywood and Alan Cowey SEE ARTICLE, PAGE 235

Zinc, Src and NMDA receptors—a transmembrane connection . . . . . . . . . . . . . 173Philippe Ascher SEE ARTICLE, PAGE 185

Probing a complex question: when are SNARE proteins ensnared? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175Timothy A Ryan SEE ARTICLE, PAGE 192

Getting a line on pain: is it mediated by dedicated pathways? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177Edward R Perl SEE ARTICLE, PAGE 218

Cortical control of the thalamus: top-down processing and plasticity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179Josef P Rauschecker SEE ARTICLE, PAGE 226

book reviewBrain, Vision and Memory: Tales of the History of Neuroscience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181by C G GrossREVIEWED BY D PURVES

scientific correspondenceWhere is the sun? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183J Sun and P Perona

contents

http://neurosci.nature.com

volume 1 no 3 july 1998

Timm staining of the hippocam-pus, showing zinc localization.On page 185, Zheng et al.provide evidence that Src mod-ulates NMDA receptor functionby reducing its inhibition byzinc. Photo courtesy of Dr.Jacqueline McGinty, Dept. ofAnatomy, East CarolinaUniversity.

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 ParkAvenue South, New York, NY 10010. Telephone 212 726 9200, Fax (212) 696 9635. North American Advertising: Nature Neuroscience, 345 Park Avenue South, New York, NY 10010-1707. Telephone (212) 726-9200. Fax (212) 696-9006. European Advertising: Nature Neuroscience, Porters South, Crinan Street, London N1 9SQ. Telephone (0171) 833 4000. Fax(0171) 843 4596. New subscriptions, renewals, changes of address, back issues, and all customer service questions in North America should be addressed to Nature NeuroscienceSubscription Department, PO Box 5054, Brentwood, TN 37024-5054. Telephone (800) 524-0328, Direct Dial (615) 377 3322, Fax (615) 377 0525. Outside North America: NatureNeuroscience, Macmillan Magazines Ltd, Brunel Road, Basingstoke, Hants RG212XS, U.K. Annual subscription rates: U.S./Canada: U.S. $595, Canada add 7% for GST (institution-al/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 makingpersonal 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 home-page at http://neurosci.nature.com. POSTMASTER: Send address changes to Nature Neuroscience Subscription Department, P.O. Box 5054, Brentwood, TN 37024-5054. Executive Offi-cers 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.

Structure/function correlationsin neurons mediating pain.

Pages 177 and 218

Artists prefer light from the upper left.

Page 183

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

199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

http://library.neurosci.nature.com/server-java/Propub/neuro/nn0798_169.pdf








vergara

Resaltado

nature neuroscience • volume 1 no 3 • july 1998 ii

contents

articlesTyrosine kinase potentiates NMDA receptor currents by reducing tonic zinc inhibition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185F Zheng, M B Gingrich, S F Traynelis and P J Conn SEE NEWS AND VIEWS, PAGE 173

Multiple kinetic components of exocytosis distinguished by neurotoxin sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192T Xu, T Binz, H Niemann and E Neher SEE NEWS AND VIEWS, PAGE 175

Presynaptic modulation of CA3 network activity . . . . . . . . . . . . . . . . . . . . . . . . . 201K J Staley, M Longacher, J S Bains and A Yee

Input synchrony and the irregular firing of cortical neurons . . . . . . . . . . . . . . . . 210C F Stevens and A M Zador

Nociceptive and thermoreceptive lamina I neurons are anatomically distinct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218Z-S Han, E-T Zhang and A D Craig SEE NEWS AND VIEWS, PAGE 177

Cortically induced thalamic plasticity in the primate somatosensory system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226E R Ergenzinger, M M Glasier, J O Hahm and T P Pons SEE NEWS AND VIEWS, PAGE 179

Strengthening of horizontal cortical connections following skill learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230M-S Rioult-Pedotti, D Friedman, G Hess and J P Donoghue

Retinotopy and color sensitivity in human visual cortical area V8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235N Hadjikhani, A K Liu, A M Dale, P Cavanagh and R B H Tootell SEE NEWS AND VIEWS, PAGE 171

Complete sparing of high-contrast color input to motion perception in cortical color blindness . . . . . . . . . . . . . . . . . . . . . . . . . . . 242P Cavanagh, M-A Hénaff, F Michel, T Landis, T Troscianko and J Intriligator

The effects of frontal eye field and dorsomedial frontal cortex lesions on visually guided eye movements. . . . . . . . . . . . . . . . . . . 248P H Schiller and I Chou

Top-down influences on stereoscopic depth-perception . . . . . . . . . . . . . . . . . . . 254I Bülthoff, H Bülthoff and P Sinha

classified advertisingsee back pages

Mapping a new human color center.

Pages 171 and 235

*

V1V2 VP V4V

V8

* *

*

* * * * *

Motion perception in cortical color blindness.

Page 242

Do depth cues contribute to figure recognition?

Page 254

100

75

50

25

0

25

50

75

100


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om












nature neuroscience • volume 1 no 3 • july 1998 169

When the eminent British geneticist J.B.S. Haldane was asked whatGod had revealed about himself through his works, Haldane is saidto have replied “an inordinate fondness for beetles”. Were he alivetoday, Haldane might instead have cited ion channels; although theirdiversity may no longer be absorbing the creative energies of theAlmighty, they did at least attract several hundred people to a three-day meeting in New York last month. The event was organized bythe New York Academy of Sciences, and it provided an excellentoverview of recent progress in understanding the molecular basis ofionic conductances, including ionotropic receptors as well as volt-age-gated and other channels. Although the advances have beenimpressive, it was also clear that the field faces a formidable chal-lenge in making sense of what has already been discovered.

An unprecedentedly comprehensive picture of channel andreceptor diversity is now emerging from large-scale genome sequenc-ing projects, notably of the nematode worm Caenorhabditis elegans(the sequence of which is now around 80% complete). As discussedby Larry Salkoff (St Louis), the worm sequencing project has alreadyled to the identification of large numbers of new channel and recep-tor subunits. In particular, at least 80 potassium channel subunitshave been found, a remarkable number considering that the ner-vous system of C. elegans contains only 302 neurons, which havebeen classified into 118 types. About 50 of these genes belong to anew class, distinguished both by their four transmembrane domainsand by the lack of knowledge about their function (the best guess isthat they are leak channels that regulate cell excitability). By taggingthe coding sequences with green fluorescent protein, it is possibleto visualize their expression patterns; many of the subunits arerestricted to single cell types, and at least one is expressed only in asingle interneuron. If this can be extrapolated to the mammalianbrain (which is not yet clear), not only does this imply a very largenumber of channels, but some may be so restricted in expressionthat they are unlikely to be discovered except by genomic sequencing.The mammalian sequencing projects are far less advanced, the avail-able evidence suggests that any given channel family will containmany more members in mammals than in worms. For example, C.elegans has a single voltage-gated K+ channel of the Shaker class,whereas at least eight have already been identified in humans.

In addition to the large number of genes encoding channel andreceptor subunits, there are several other levels at which diversitycan arise. One is alternative splicing; some of the new K+ channelgenes discovered in C. elegans, for instance, can give rise to six orseven different isoforms. Another is RNA editing, a remarkableprocess by which single base changes (and hence changes in theencoded protein) can be introduced into an already-transcribedmRNA. Editing has been described in the mammalian brain for bothAMPA- and kainate-type glutamate receptors, where it is known toregulate ion selectivity and channel kinetics (Rolf Sprengel, Heidel-

berg; Steve Heinemann, Salk Institute). Robert Reenen (University ofConnecticut) has now found that the Drosophila Na+ channel encod-ed by the paralytic gene is also edited at several different sites, andthat like the glutamate receptors, para editing is under tight devel-opmental regulation. Why RNA editing has been exploited by thenervous system in this way, and in such diverse species, is still unclear;one possibility is that it may allow the expression of two almost iden-tical sequences without the risk of gene conversion.

The greatest source of diversity, however, arises from the fact thatmost channels and receptors are composed of multiple subunits,which can be assembled in different combinations. In many cases,a given channel can show profoundly different behavior dependingon which modulatory subunits are present. Many examples werepresented, including K+ and Ca2+ channels as well as all the majorclasses of ionotropic receptors; to cite just one, Terry Snutch (Van-couver) showed how P- and Q-type calcium currents, long thoughtto be distinct (both are voltage-gated but they differ in their ability toundergo spontaneous inactivation), can both arise from the samepore-encoding α1a subunit. Moreover, this difference can be causednot only by differential association with regulatory β subunits, butalso by alternative splicing of the α1a subunit itself. The propertiesof cloned channels must generally be studied in heterologous expres-sion systems, but just because a particular subunit combination canform in vitro does not necessarily mean that it occurs in the brain.The process of determining which subunits associate with whichothers in vivo is long and laborious, yet essential if the lessons fromrecombinant channels in vitro are to be extended to real neurons.

The number of channels and receptors that can be encoded bythe genome is thus enormous, and although this may be good newsfor pharmaceutical companies, it presents a daunting prospect forany attempt to understand the underlying principles of brain orga-nization. The challenge, of course, is to determine what this prodi-gious molecular diversity might signify in functional terms. The easyanswer is that the brain is very complex and that it needs a corre-spondingly vast number of molecules to perform its diverse func-tions; but although this may be true, it is hardly satisfying.

It is possible, of course, that not all the observed diversity has anyadaptive significance. Gould and Lewontin have warned against theuncritical acceptance of adaptive explanations in biology, and it isat least possible that some of the diversity that has arisen among dif-ferent families of channels and receptors has no purpose – genesmay duplicate and diverge in evolution simply because they can, inother words because there is no selective disadvantage to doing soand because the process is not easily reversed once it has occurred. Totake one simple scenario, imagine that a gene encoding a particularchannel undergoes duplication, and that the two sequences driftapart and acquire differences in their promoters; although they mayat first be mutually redundant, if certain populations of neurons lose

editorial

Making sense of channel diversity


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

170 nature neuroscience • volume 1 no 3 • july 1998

editorial

the ability to express one or the other gene, they both become essen-tial even if they differ very little in their functional properties.

One way to address the question of adaptive significance is toask whether individual members of gene families show conserva-tion in evolution. Broadly, the answer so far seems to be that theydo. For instance, Salkoff noted that the major classes of potassiumchannels that exist in humans all have recognizable homologs inC. elegans, suggesting that the evolution of more complex nervoussystems has not been accompanied by the appearance of new typesof channels, but rather by diversification of pre-existing types.Moreover, even the individual family members often show highconservation; the human Slo1 gene (which encodes a high-con-ductance calcium-gated K+ channel), for example, is much closerto nematode Slo1 than to human (or nematode) Slo2, and thesame principle holds for other classes of K channels. Perhaps themost striking demonstration of functional conservation was theworm homolog of one of the Long QT-type K+ channels, muta-tions of which lead to abnormal heart rhythms in humans. Theworm KQT homolog is expressed in the pharynx, which like theheart generates a rhythmic pumping action. Moreover, when thehuman mutation is introduced into the worm coding sequence,the mutant animals show a defect analogous to the human condi-tion, a ‘long pharyngeal pump syndrome’, as it were. Examples likethis offer hope that comparing model systems will reveal somegeneral principles of how different patterns of channel expressiondetermine the properties of different classes of neurons.

The current favorite method for determining the function of achannel is to knock it out genetically, but it was clear from many ofthe presentations that this approach has serious limitations. Althoughgene knockouts avoid the problems associated with lack of speci-ficity in pharmacological blocking agents, they raise interpretation-al problems of their own. Often, mutant phenotypes are eithernonexistent or too subtle to be recognized using the available tech-niques. Even in cases where a mutant phenotype is found, it is oftendifficult to rule out the possibility that the absence of the gene dur-ing development has led to compensatory changes that complicatethe interpretation of the result. The ideal gene-knockout methodwould be cell-type specific and under tight temporal control. Butdespite several apparently encouraging reports in the literature, suchtechniques are far from robust, as Peter Seeberg (Heidelberg) empha-sized. At present, the technology does not exist to inactivate specif-ic genes in specific parts of the mammalian nervous system withhigh efficiency and specificity, let alone in a rapidly inducible orreversible manner that would eliminate concerns about develop-mental compensation. A reliable method for doing this would beinvaluable, but it does not yet seem to be close to realization.

The challenge in understanding ion channel function may bereduced to two broad questions: what do specific channels contributeto the behavior of the cells in which they are expressed, and howdoes the behavior of these cells contribute to the working of the sys-tem as a whole? Although certain mutations have given interpretableand interesting phenotypes, there are major obstacles to be over-come before this can be achieved on a routine basis. For one thing,the sheer effort of descriptive analysis will be considerable. It will beessential to correlate ion channel expression with single-cell prop-erties, and this is far from trivial in vivo. In situ hybridization or anti-body staining can provide information about expression patterns,but it is also necessary to correlate this with cellular physiology. Onepowerful approach, discussed by Hannah Monyer (Heidelberg), isto record from single neurons via a whole-cell patch pipette and thento aspirate the contents of the cell into the pipette, so that mRNAexpression can be analyzed by PCR. Monyer has used this techniqueto show that principal neurons and interneurons in the primary

visual cortex show different patterns of AMPA receptor expression,and she hopes to determine how specific patterns of channel andreceptor expression can be related to cortical information processing.

Ultimately, it seems clear that to understand how channels andreceptors determine neuronal behavior, the field will have to gobeyond the level of molecular description and adopt a more quan-titative and biophysical approach. This is perhaps the greatest chal-lenge for the years ahead. To explain the electrical properties of aneuron, it is not sufficient merely to specify the types of channels itexpresses; one must also know their densities and distributions, asthey relate to the fine structure and cable properties of axons anddendrites. Such an analysis would seem essential for any seriousattempt at understanding channel function at the cellular level, butsurprisingly the question hardly came up in the meeting.

Consider, for instance, how a neuron might achieve the appro-priate number and distribution of each of the channels it express-es. To obtain the desired pattern of excitability, there mustpresumably be some form of feedback from activity to channelexpression. Yet how this might occur is almost entirely mysterious.To what extent is channel density regulated by activity, and if so bywhat feedback pathways? At what level is control exerted? It couldbe transcription, or at post-transcriptional levels such as proteinsynthesis, degradation, trafficking or association with modulatoryproteins. In muscle fibers, the distribution and turnover times fordifferent types of acetylcholine receptors are regulated with greatprecision, both during development and in response to changingpatterns of electrical activity, but whether this is also true for neu-ronal receptors and ion channels is still very unclear.

Not only the number but also the precise localization of differentmolecules must in some cases be specified. Efforts to understandthis were exemplified by the presentation from Ole Ottersen (Oslo),who has used immunogold labeling to study the fine structure ofcerebellar Purkinje cells. He has shown that different molecules aretargeted to different sites; the δ2 glutamate-like receptor, for instance,is present at the postsynaptic sites formed with parallel fibers butnot with climbing fibers. Another molecule that is precisely local-ized in these cells is the glutamate transporter EAAT4, which isknown to play a role in clearing glutamate and shortening the EPSC.How it does so is unclear; Ottersen has shown that the main site ofEAAT4 expression is at the base of dendritic spines, close to the siteof contact with glial cells and several microns away from the post-synaptic membrane where ionotropic glutamate receptors are con-centrated. How EAAT4 can affect synaptic activation given itsexclusion from the site of transduction remains to be determined,but the results highlight the importance of precise molecular descrip-tions of synaptic structure if the details of synaptic transmission areto be understood in quantitative terms.

The molecular basis of this structural specificity is even less clear,but some details are starting to emerge. Morgan Sheng (Massachu-setts General Hospital), Mary Kennedy (Caltech) Heinrich Betz(Frankfurt) and Nat Heintz (Rockefeller University) each discussedmolecular components of postsynaptic sites, and have identifiedvarious molecules that may govern how receptors and channelsbecome localized. An important goal now is to determine how thesevarious components interact, and how the appropriate density anddistribution of synaptic signaling components is achieved and main-tained. In the longer term, it will also be important to find outwhether similar mechanisms regulate channel distribution elsewhereon the membrane, and thus whether they play a more general role inregulating neuronal excitability.

New York Academy of Sciences conference: Molecular and functional diversity of ion chan-nels and receptors. New York, May 14-17, 1998. See http://www.nyas.org/brochion.html forprogram details.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


Dorothy’s whirlwind departure from amonochrome Kansas into the vividly chro-matic world of Oz, in the 1939 classic filmThe Wizard of Oz, highlights the substan-tial contribution that color makes to ourvisual world. Yet equally revealing is theease with which we view monochrome filmor television, where the absence of colordoes not compromise our enjoyment andcan pass unnoticed. This is not so in theclinical condition of cerebral achromatop-sia where patients, following characteristi-cally ventral occipitotemporal braindamage (see Fig. 1a) inhabit a drab world,devoid of color, and may be painfully awareof their complete loss of chromatic vision1.Attempts at understanding the nature ofcerebral achromatopsia and its neural basishave spawned controversy ever since LouisVerrey’s description2 of such a case in 1888(see ref. 3 for review). The ensuing debatelasted for more than a century, and not sur-prisingly the protagonists reflected oppos-ing views about whether any cognitive orperceptual function was regionally local-ized. At issue was whether achromatopsiaresults from the deletion of a corticalregion specialized for the processing ofcolor. The demonstration4, using positronemission tomography, of increased cere-bral blood flow in an area of cortex whenobservers view chromatic scenes was cer-tainly consistent with this notion, becausethe activated region, dubbed the humancolor center, is invariably damaged in casesof cortical color blindness. By then, how-ever, the cluster of 20–30 visual areas occu-pying almost half of the neocortex ofmonkeys had been identified, and thedebate turned to whether the human colorcenter was homologous to the fourth visu-al area of the monkey, cortical area V4 (seeFig. 1b). This correspondence had been

the human brain6. These areas are thepresumed homologues of those identi-fied in the monkey brain using a varietyof invasive techniques, such as cellularrecording and experimental neuroanato-my. Delineation of an area relies on thepresence of an orderly retinotopic map

With color in mindCharles Heywood and Alan Cowey

Which human brain area produces color blindness when damaged? High-resolutionfunctional neuroimaging suggests that it is area V8, not the favorite candidate V4.

proposed chiefly on the(not uncontested) view thatV4 contains a comparative-ly high proportion of cellsthat respond selectively towavelength and color5. Theresults of a functional imag-ing study by Hadjikhaniand colleagues, reported inthis issue of Nature Neuro-science (pp 235–241), sug-gest however that thehuman color center is dis-tinct from area V4. Thenewlydefined color areacontains a complete retino-topic map of the contralat-eral visual half field,responds more robustly tocolor than neighboringregions and, unlike V4, isactivated by the inductionof color aftereffects. ThefMRI signal elicited by anaftereffect thus mimics theresponse to a real coloredstimulus, providing sup-porting evidence that V8 isimplicated in processesinvolved in perceiving color.These properties, theauthors suggest, make it aready candidate for a regionresponsible for our con-scious perception of a col-ored world.

Hadjikhani and col-leagues used existing, butimproved, techniques offunctional neuroimagingto reveal, with increasedsenstivity, brain areasinvolved in the processingof color. Functional mag-netic resonance imaging(fMRI) relies on endoge-nous changes in magnetic susceptibility,which result from changes in local cere-bral blood flow and oxygenation. Suchchanges are activity dependent, and theirmeasurement in response to visually pre-sented stimuli have already establishedthe borders of a number of visual areas in

news and views

Charles Heywood is the Sir DermanChristopherson Research Fellow at theDepartment of Psychology, Science Laboratories,South Road, Durham DH1 3LE, UK([email protected])Alan Cowey is at the Department ofExperimental Psychology, South Parks Road,Oxford OX1 3UD, UK([email protected])

Fig. 1. Visual areas in the human and monkey brains. (a)The medial view of the left hemisphere of the human brain.Striate cortex, area V1, is shown in red, partly buried in thecalcarine sulcus. The region along the collateral sulcus,whose destruction leads to cerebral achromatopsia, isshown in blue. Area V8 lies in the middle of the collateralsulcus, whereas V4 lies slightly more posterior and medial.The damage indicated by blue would therefore include V4vand V8. (b) A lateral schematic view of the right hemisphereof the macaque monkey, in which the labeled sulci have beenopened up. Areas V1 (red) and V4 (purple) extend onto theventral and medial surfaces, respectively. Areas in the tem-poral lobe (TEO, orange and IT, dark green), have beenimplicated in color vision of the macaque monkey, andHadjikhani and colleagues raise the possibility that TEO maycorrespond to human V8.

Calcarinesulcus

Striatecortex

Lingualgyrus

Inferioroccipitalsulcus

Superiortemporal

sulcus

TEO ITV4

V4

Lunatesulcus

Fusiformgyrus

Collateralsulcus

B

Aa

b

Bob Crimi


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


news and views

of the visual world, which is a feature ofmany visual areas. The authors haveexploited the fMRI technique to identifysuch maps, using stimuli consisting ofslow-moving patterns of luminance mod-ulation. As the light and dark areas passacross the visual field, they elicit period-ic excitation at the associated corticallocation. Moreover, the phase of theresponse specifies the polar angle oreccentricity, for rotation or radial move-ment around the fixation point respec-tively, of the visual field regionrepresented at that location. Thus aFourier analysis on the response profileof a single voxel of the image, along witha consideration of the sign of the responseto identify mirror- versus non-mirror-image representations, will yield theretinotopy of a visual area, which can dis-played, by cortical flattening, as a two-dimensional map. Using such techniques,Hadjikhani and colleagues compared theeffect of luminance-defined visual pat-terns with that of identical patterns thatwere defined by equiluminant color vari-ation, that is, variations in color but notin luminance. In addition to findingstronger activation to color than to lumi-nance in cortical areas V1, V2, V3/VP andthe ventral subdivision of V4 (V4v), aregion in the middle of the collateral sul-cus was identified that responded prefer-entially and especially effectively to color.Its location corresponded to that previ-ously (and in the absence of identifica-tion of retinotopic boundaries,prematurely) described as ‘human V4’. Byadopting improved techniques, includinga high-field scanner, signal averaging andimproved visual displays, the authorsestablished that the true color center liesbeyond the anterior border of the previ-ously reported area V4v (i.e. outside v4altogether). Furthermore the retinotopyof the color center differs from its neigh-bors. Areas V4v, VP and inferior V2 con-tain quarter-field representations of theupper visual field and share a contiguousrepresentation of the fovea. In contrast,the color area contains a map of theupper and lower half-fields with a fovealrepresentation located at its anterior bor-der. It now seems clear that the color cen-ter is distinct from area V4, andaccordingly, the authors refer to this new,previously unreported region as area V8.

Area V8 poses as many new questionsas its identification sought to resolve.That it is distinct from area V4 is cer-tainly consistent with hitherto puzzlingdemonstrations that ablation of V4 in themacaque monkey does not result in the

red/green borders without signaling thenature of the colors of which the borderis composed, i.e. which is red and which isgreen. Form can thus be derived from theprocessing of wavelength differences andyield information about the visual scenewithout encoding its chromatic content.It should perhaps come as no surprisethat achromatopsic patients, lacking areaV8, can show a preserved capacity to usewavelength variation to detect motionand form12, presumably mediated by ear-lier and intact extrastriate areas. Detec-tion of equiluminant chromatic form caneven be achieved when it is disguised byaccompanying rapid random luminancevariation. The latter implies that patientsmust retain color-opponent processingmediated by the well known P-channel ofprimate vision, which unlike its partnerthe M-channel is blind to the introduc-tion of rapid flicker. Moreover, anotherpaper in this issue13 reports that pro-foundly achromatopsic patients lackingevidence for residual color-opponentprocesses are still able to extract motionfrom high-contrast color cues.

Achromatopsic patients can thereforeprocess wavelength differences to extractinformation about form and motion,but their brain damage nevertheless ren-ders them blind to color differences.Such brain damage is, of course, likelyto encroach on territory other than areaV8, including the adjacent V4v.Although loss of conscious representa-tion of hue characterises achromatop-sia, another very different explanationhas been offered3, namely that it is a fail-ure of color constancy—a loss of theinvariance of an object’s perceived colordespite wide variation in the wavelengthcomposition of the illuminating light.Although it has been suggested that thechromatic responses of neurons in areaV4 show color constancy14, the effectsof ablating this in the monkey have yetto convincingly demonstrate a corre-sponding deficit. Might V8 be assignedsuch a role? An explanation of cerebralachromatopsia as resulting from thedestruction of V8 with a concomitantdeficit in color constancy does not read-ily explain why two very different equi-luminant hues are indistinguishable toan achromatopsic patient, nor why theworld should be described in shades ofgray. A direct test of color constancy inone such patient does not lend unequiv-ocal support for this view15. When twopatches of different spectral composi-tion were presented against two differ-ent backgounds, one to each eye, the

severe deficits in the discrimination ofhue that is the hallmark of cerebralachromatopsia7. Conversely, impair-ments in the discrimination of visuallypresent form and pattern vision thatinvariably follow damage to V4 in themonkey are not an invariable feature ofthe vision of achromatopsic people. Thelocation of V8 within the regiondestroyed in achromatopsic patients,although the latter is always more exten-sive and includes white matter damage,is strong but not conclusive evidence thatV8 is the critical area whose removal canresult in the complete loss of the con-scious representation of color. As Had-jikhani and colleagues themselves pointout, the question naturally arises as towhere area V8 is concealed in themacaque monkey’s brain. Localization ofmacaque V8 would lead to confirmato-ry evidence that its removal results incortical color blindness. They speculatethat area TEO8,9 (Fig. 1b), a region lyinganterior to V4, may be the culprit.Indeed this region, along with moreanterior temporal lobe areas (Fig. 1b),has been implicated in color vision (Van-duffel, W. et al., Soc. Neurosci. Abstr. 23,334.7, 1997; Katsuyama, N. et al., Soc.Neurosci. Abstr. 23, 803.11, 1997). Largelesions to anterior and inferior portionsof the temporal lobe do render monkeysachromatopsic10, and a high proportionof color-selective cells, revealed by meta-bolic labeling and electrophysiologicalrecording, reside in this area11.

Although the precise role of area V8has yet to be clarified, determining whatis spared, as opposed to lost, in cerebralachromatopsia may provide some sign-posts. Color makes a ubiquitous contri-bution to vision. Color differences can,among other things, provide informationabout form, texture and motion. This ispresumably reflected in the number, andwide variation, of activated areas reportedin neuroimaging studies when subjectsperform a wide variety of color-relatedtasks. Cerebral blood flow can be modu-lated by the nature and difficulty of thebehavioral tasks (whether they entail pas-sive viewing, active discrimination ordirected attention) and the properties ofthe visual display (containing equilumi-nant color, with or without form, with orwithout associated brightness differences).The physical basis of color is the wave-length composition of light. Many corti-cal areas prior to V8 contain cells that aresensitive to, but not selective for, wave-length differences. For example, cells mayrespond vigorously to equiluminant


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

news and views


patient reported them to be indistin-guishable only when the ratios of theretinal cone response of path/back-ground were preserved. When the ratioswere different for each eye, the patchesno longer appeared identical. Theseresponses are akin to those of the nor-mal observer and are presumed to bemediated by retinal mechanisms. How-ever, the achromatopsic patient departedfrom normal performance when pre-sented with more complex scenes requir-ing multiple cone contrast comparisons.Achromatopsic patients may thereforeretain rudimentary color constancymechanisms, and it remains an openquestion as to whether V8 is essential forconsolidation of information acrosslarge regions of complex scenes.

However, what these patients do lackis the conscious representation of color. Ifit transpires that a single cortical area,area V8, is indispensable for our con-scious percept of color, it will indeed be arare triumph for the view that regionalspecialization underlies the cluster ofvisual areas that occupy a substantialproportion of the neocortex in primates.On a note of caution, however, Had-jikhani and colleagues hint that area V8responds to a wide variety of visual stim-uli. The challenge will then be to estab-lish the precise role of this, and otherareas, in the cortical processing of color.Hadjikhani and colleagues have pointedus in the correct direction.

1. Cowey, A. & Heywood, C.A. Trends Cog.Neurosci. 1, 133–139 (1997).

2. Verrey, L. Archs. Ophtalmol. (Paris) 8, 289–301(1888).

3. Zeki, S.A Vision of the Brain. (BlackwellScientific Publications: Oxford Univ. Press,1993).

4. Zeki S. et al. J.Neurosci. 11, 641–649 (1991).

5. Zeki, S. Nature 284, 412–418 (1980).

6. Sereno, M.I. et al. Science 268, 998–893(1995).

7. Heywood, C.A., Gadotti, A. & Cowey, A. J.Neurosci. 12, 4056–4065 (1992).

8. Boussaoud, D., Desimone, R. & Ungerleider,L.G. J. Comp. Neurol. 306, 554–575 (1991).

9. Zeki, S. Proc. R. Soc. Lond. B 263, 1539–1544(1996).

10. Heywood, C.A., Gaffan, D. & Cowey, A. Eur. J.Neurosci. 7, 1064–1073 (1995).

11. Komatsu, H., Ideura, Y., Kaji, S. & Yamane, S.J. Neurosci. 12, 408–424 (1992).

12. Heywood, C.A., Kentridge, R.W. & Cowey, A.Exp. Brain Res. (in press).

13. Cavanagh, P. et al. Nature Neuroscience 1,242–247 (1998).

14. Zeki, S. Neuroscience 9, 741–765 (1983).

15. Hurlbert, A.C., Bramwell, D.I., Heywood,C.A. & Cowey, A. Exp Brain Res. (in press).

but in some recombinant NMDA recep-tors (those assembled from NR1 andNR2A subunits) the IC50 of the voltage-independent inhibition is exceptionallylow (10 nM-100nM) 4-6. This is the rangeof concentration at which zinc is presentas a ‘contaminant’ in most experimentalsolutions, but also in the cerebrospinalfluid. Thus, even without adding anyadditional zinc, a large fraction of thehigh-affinity, voltage-independent,inhibitory sites on NR1-NR2A receptorsare already occupied, so that the additionof a zinc-chelating agent to the bath isusually sufficient to double the amplitudeof the baseline response4.

The interaction of tyrosine kinaseswith NMDA receptors a priori seemed tohave little relation with that of zinc. Theearly observations of Salter and col-leagues7,8 showed that tyrosine kinaseinhibitors inhibit some NMDA respons-es, which conversely can be potentiatedby intracellular injection of either a con-stitutively-active form of the intracellulartyrosine kinase Src or peptide fragmentsthat activate Src8. Similar effects wereobserved with another tyrosine kinase,fyn9. These observations were reinforcedby reports that tyrosine kinase inhibitorsinterfere with some forms of long-termpotentiation (a cellular model for learn-ing and memory), and that some tyrosinekinase deficient mice had perturbed LTPand behavioral abnormalities (for refer-ences see 10). Once again, recombinantNMDA receptors were used to reveal asubunit specificity: Src only acted onreceptors assembled from NR2A subunitsand a restricted group of splice variantsof NR19.

Zheng and colleagues on page 185 ofthis issue of Nature Neuroscience11 havelinked these two sets of apparently inde-pendent observations by experiments thatstrongly suggest that the potentiatingeffect of Src on NR1-NR2A receptorsresults from the suppression of the ambi-

Recently, scientists have become increas-ingly aware of the potent actions of zincand tyrosine kinases on NMDA receptors.NMDA receptors are a group of ionotrop-ic glutamate receptors that are permeableto both calcium and sodium, and havebeen implicated in many forms of synap-tic transmission, synaptic plasticity and incell death. Zinc had long been known tointeract with many neurotransmitterreceptors, but its inhibition of NMDAreceptors has attracted particular interestbecause of a possible functional role. Zincis stored in synaptic vesicles in a numberof glutamatergic terminals of the fore-brain, is released during synaptic activity,and its concentration in the synaptic clefthas been suggested to rise to the micro-molar range1. Early studies of the interac-tion of zinc with NMDA receptors2,3

identified two effects, a voltage-dependentinhibition which resembles that of mag-nesium, and a voltage-independent inhi-bition, which occurs at a different site. Thetwo effects were initially described for zincconcentrations in the micromolar range,and this fitted nicely with estimates of zincconcentration in the synaptic cleft fol-lowing glutamate release. Recently, afterthe cloning of NMDA receptors subunitsallowed the expression of recombinantreceptors, the two effects of zinc were fur-ther analyzed using NMDA receptor sub-types built from specific combinations ofsubunits. Using these recombinant recep-tors, additional evidence was obtained tosupport the hypothesis that the voltage-dependent effect of zinc occurs via thesame site as the voltage-dependent mag-nesium block within the channel pore, butthat zinc permeates the channel better4.The binding site involved in the voltageindependent block is less well identified,

Zinc, Src and NMDAreceptors—a transmembraneconnectionPhilippe Ascher

Zinc and tyrosine kinases produce opposite effects on theNMDA receptor; new evidence suggests that Src-inducedpotentiation is due to the relief of zinc inhibition.

Philippe Ascher is at the Laboratoire deNeurobiologie, Ecole Normale Superieure, 46rue d’Ulm, Paris 75005, Francee-mail: [email protected]


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


news and views

ent zinc inhibition. Src mimics the effectof adding an extracellular zinc chelator,and the potentiation induced by remov-ing zinc from the extracellular mediumand the potentiation induced by the addi-tion of Src occlude each other, suggestinga shared mechanism. Dose responsecurves confirm that the EC50 of the volt-age-independent, high-affinity zinc inhi-bition is increased by Src.

The picture of what exactly is happen-ing at the molecular level, however,remains fragmentary. From the work ofSalter’s group 9 we know that Src binds tothe NMDA receptor through a region thatis distinct from the catalytic site, but wedo not know whether Src directly phos-phorylates the NMDA receptor orwhether it acts via a third protein. Zhengand colleagues11 have now identified threespecific tyrosines in the NMDA receptorsubunit NR2A that appear to be involvedin NMDA receptor modulation; it istempting to speculate that they are targetsfor Src-mediated phosphorylation, butthis has not been directly demonstrated.The explanation of the high zinc senstivi-ty of NR1–NR2A remains uncertain. Thesimple explanation was that either zinc orSrc would specifically interact with NR2A.However, two of the tyrosines identifiedby Zheng and colleagues are common toNR2A and NR2B. Furthermore, recentevidence12 suggests that NR1 is involvedin the zinc modulation. Nevertheless,whatever the details of the picture, thedata strongly suggest a kind of ‘retrograde’

zinc concentration in experiments inwhich glutamate or NMDA are bathapplied, and in which the concentrationof zinc is fixed. None of the availablechelators, however, will prevent the riseof zinc concentration if zinc is releasedin the synaptic cleft in the millisecondrange, especially if the local concentra-tion in the cleft can reach the micromolarrange. This is mainly because calciumalso competes with zinc to bind to theseheavy metal chelators. The problem maybe resolved by controlling the presynap-tic zinc content, either with chelatorshaving selective access to the synapticvesicles (for example see ref. 13) or fromgenetically modified animals in which thezinc transporters and buffers (and in par-ticular the storage of zinc in synaptic vesi-cles14) will have been modified.

Zinc is also strongly suspected to playa role in various pathologic conditions,such as neuronal death following globalischemia or massive seizures 15. Zinc isknown to be cytotoxic, and there is evi-dence that this toxicity results from itsentry into the cell, even though its intra-cellular targets are not rigorously identi-fied. The cell death induced by exposureto zinc is increased by glutamate and canbe reduced by NMDA antagonists15, sug-gesting that a major route of zinc entryinto neurons could be through NMDAreceptors, which appear to be bothblocked by and permeable to zinc. TheNR1-NR2A receptors, which contain ahigh affinity, voltage independent, zincinhibitory site, could play a particular rolein the zinc toxicity because they have anadditional surprising property, namelythat the zinc inhibition remains incom-plete even when this site is saturated5–7.This means that at zinc concentrations inthe micromolar range, zinc retains accessto the channel and therefore to the cellinterior. Therefore, the NMDA receptorsmost sensitive to zinc may be, paradoxi-cally, the most dangerous route of zincentry into neurons. If we consider that Srcinduces a rightward shift of the zinc con-centration response curve, we can predictthat Src activation will potentiate theentry of zinc. Thus in pathologic diseasestates, Src and zinc may combine theireffects instead of neutralizing each other.

Although Zheng and colleagues havenot directly proven that Src can phos-phorylate NMDA receptors and have notidentified the zinc binding site, they haveraised a number of intriguing questionsabout the interaction between zinc bind-ing and tyrosine phosphorylation. Morework is needed to determine whether

signaling (see Fig. 1).Previous studies ofreceptor tyrosinekinases have popular-ized an ‘orthograde’model, where thebinding of an extra-cellular ligand to areceptor transmits asignal through themembrane to activatea tyrosine kinase onthe cytoplasmic side.The data of Zhengand colleagues sug-gest that the NMDAreceptor mediates atrans-membranemessage in the oppo-site direction; activa-tion of a tyrosinekinase on the cyto-plasmic side alters thebinding of extracellu-lar ligands. The ligandthat is most clearly

affected is zinc, which seems to be ‘dis-placed’ by the process (its apparent bind-ing affinity decreases), but the kinetics ofglutamate dissociation from the receptoris also slowed. (Glycine affinity may alsobe affected but this has not yet been stud-ied.) However, the retrograde signalinganalyzed by Zheng and colleagues doesnot exclude a mechanism working in theorthograde direction; if zinc and Src acton allosteric sites of the NMDA receptor,allosteric theory predicts that if Src alterszinc binding, zinc binding can alter Srcaction.

From a physiological point of view,the observations suggest that Src and zincplay opposite roles in synaptic transmis-sion. If there are glutamatergic synapsesin vivo where the postsynaptic NMDAreceptors have the very high zinc sensi-tivity observed in recombinant NR1-NR2A receptors, extracellular ambientzinc will set the NMDA response to abouthalf of its maximal value, which wouldplace the receptors in an optimum posi-tion to be regulated bidirectionally. Theresponse could be either increased (byactivation of a tyrosine kinase) or inhib-ited (by zinc released from glutamatergicterminals). This hypothesis remainsuntested, and in particular no study hasyet investigated the modulation ofNMDA synaptic currents by zinc. In sucha study, one experimental obstacle will bethe difficulty of chelating synapticallyreleased zinc. Zinc chelators like EDTA,TPEN or tricine allow one to control the

Fig. 1. Left: the NMDA receptor has been activated by the simul-taneous binding of glycine to the NR1 subunit and of glutamate tothe NR2A subunit, but the binding of zinc reduces the time spentin the open state. Right: the interaction of Src with the cytoplas-mic tail of the NR2 subunit induces a conformational changewhich, by preventing the binding of zinc on the extracellular side(arrow), potentiates the NMDA response.

NR1 NR2A

ZincGlyGlu

Magnesium

NR1

SrcSrcP

NR2A

Bob Crimi


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

news and views


these observations made on recombinantNMDA receptors in vitro will apply to‘real’ NMDA receptors in vivo (which mayhave different combinations of subunits).What zinc does in vivo, and why NMDAreceptors should have evolved zinc sensi-tivity at all, still remains a mystery. How-ever, Zheng and colleagues have put usone step closer to understanding the intri-cacies of this modulatory mechanism.

1. Frederickson, C.J. Intl. Rev. Neurobiol. 131,145–238 (1989).

2. Legendre, P., & Westbrook, G.L. J. Physiol.

required to tackle these issues is givenby Xu and colleagues in this issue ofNature Neuroscience 2, a study thatprobes the molecular basis of neuro-transmitter secretion using modernmicrophysiological and molecular tools.

Several different approaches tounderstanding membrane transporthave converged over the last five years,revealing a cast of molecular playersthat lie at the heart of vesicle traffickingand presynaptic neurotransmitterrelease3,4. Considerable evidence nowimplies that SNARE proteins5, a familyof compartmentally specific integralmembrane proteins with cytoplasmictails, provide a core interaction thatdetermines the specificity of membranepairing. Recently, the assembly ofappropriately paired SNAREs has beenshown to provide the minimal machin-ery required for the successful mixingof artificial membrane bilayers6. Thebest-characterized SNAREs are involvedin neuronal exocytosis. They include thesynaptic vesicle associated proteinsynaptobrevin, and the plasma mem-brane associated proteins SNAP (synap-tosome-associated protein)-25 andsyntaxin. In vitro, these three proteinsspontaneously assemble into a very sta-ble ternary complex, whose disassem-

Perhaps one of the most sensitive sub-strates for modulating information flowin the brain is the molecular machineryof the synaptic terminal. A detailed mol-ecular understanding of presynapticfunction is important because it is thetarget of many therapeutic reagents forneurological disease, as well as the siteof action of most stimulants and drugsof abuse. The molecular description ofpostsynaptic events during synaptictransmission has advanced rapidly overthe last twenty years, thanks to the fruit-ful combination of electrophysiologyand molecular biology1. Our under-standing of molecular events in thepresynaptic terminal, however, has onlyrecently begun to emerge. The charac-terization of presynaptic processes is dif-ficult because of the inherentcomplexity of the underlying organellecell biology and because, unlike with ionchannels, the study of any single mole-cule reveals little about the system as awhole. An example of the combinedfunctional and molecular approach

Probing a complex question:when are SNARE proteinsensnared?Timothy A. Ryan

A recent study uses elegant microphysiological andmolecular tools to investigate the molecular basis andkinetics of vesicle exocytosis.

Timothy Ryan is at the Department ofBiochemistry, Cornell University MedicalCollege,1300 York Avenue,New York, NY 10021, USAemail: [email protected]

(Lond.) 429, 429–449 (1990).

3. Christine, C.W. & Choi, D.W. J. Neurosci. 10,108–116 (1990).

4. Paoletti, P., Ascher, P. & Neyton, J. J.Neurosci. 17,5711–5725 (1997).

5. Williams, K., Neurosci. Lett. 215, 9–12 (1996).

6. Chen, N., Moshaver, A. & Raymond, L.A. Mol.Pharmacol. 51, 1015–1023 (1997).

7. Wang, Y.T. & Salter, M.W. Nature 369, 233–235(1994).

8. Yu, X.M., Askalan, R., Keil, G.J. & Salter, M.W.Science 275, 674–678 (1997).

9. Köhr, G. & Seeburg, P.H. J. Physiol. (Lond.) 492,445–452 (1996).

10. Lu, Y.M., Roder, J.C., Davidow, J. & Salter,M.W. Science 279, 1363–1368 (1998).

11. Zheng. F., Gingrich, M.B., Traynelis, S.F. &Conn, P.J. Nature Neurosci. 3, 185–191(1998).

12. Traynelis, S.F., Burgess, M.F., Zheng, F.,Lyuboslavsky, P. & Powers, J.L. J. Neurosci.(in press).

13. Budde, T., Minta, A., White, J.A. & Kay, A.R. Neuroscience 79, 347–358 (1997).

14. Wenzel, H.J., Cole, T.B., Born, D.E.,Schwartzkroin, P.A. & Palmiter, R.D. Proc.Natl. Acad. Sci. USA 4, 12676–12681 (1997).

15. Choi D.W. & Koh, J.Y Annu. Rev. Neurosci.21, 347–375 (1998).

bly is carried out by the ATPase NSF (N-ethyl-maleimide-sensitive fusion pro-tein) together with SNAPs (solubleNSF-attachment proteins).

The questions of where, when andhow SNARE assembly occurs withinsecretory terminals and what specificstate the stable NSF-SNAP-sensitiveternary complex corresponds to in theprogression of vesicle traffic are nowcentral to forming an accurate descrip-tion of presynaptic function. Severalrecent reports7,8 and a technical tour-de-force in the current issue of Nature Neu-roscience 2 have begun probing thesequestions. The approach of Xu and col-leagues was to apply a combination ofelegant techniques to precisely stimu-late and measure catecholamine secre-tion from neuroendocrine cells whilesimultaneously interfering with SNAREfunction. To accomplish this, they usedwhole-cell capacitance measurements tomonitor cell-surface area, carbon-fiberamperometry to detect catecholaminerelease, and photo-uncaging of chelat-ed calcium to deliver step changes inintracellular calcium, triggering secre-tion. The key to these studies is thatsynaptobrevin, syntaxin and SNAP-25are all specific substrates for digestionby various botulinum and tetanus tox-ins, and these proteins are only vulner-able to these toxins when not assembledin the tight ternary complex. Thus ameasurement of secretion after intra-cellular toxin application provides anestimate of the fraction of secretory-competent vesicles that depend onSNAREs in a toxin-sensitive, orunassembled state.

Secretory terminals are comprised ofvesicles in at least two different states ofreadiness with respect to the speed atwhich they can be caused to fuse withthe plasma membrane. In typical synap-tic terminals, there is a distinct subset


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


news and views

of vesicles, representing roughly 10% ofthe total population, that can be seen invery close proximity to the plasmamembrane, the so-called ‘morphologi-cally docked pool’. Given the very shorttime delay between the arrival of apresynaptic action potential and therelease of neurotransmitter (less thanone millisecond), it is believed thatsecretion must draw upon a pool in arelease-ready state, probably a subset ofthe morphologically docked vesicles. Inchromaffin cells, brief elevations in

that application of all but oneof the appropriate neurotoxinsleads to a complete abolition ofsecretion, which is to say thatboth the fast and slow phasesof membrane fusion areblocked. Therefore, all of theSNAREs relevant for secretionin this assay must have been ina toxin-sensitive state. Whatabout the SNAREs on dockedvesicles? The authors proposethat the docked state be areversible one, so that duringthe toxin incubation period, allof the readily releasable vesiclescycle through a state where theSNAREs are vulnerable to toxinattack (Fig. 1). This preservesthe original idea that release-

ready vesicles have their SNAREs assem-bled in a tight complex, but it predictsthat over a five-minute period (the timeto dialyze in the toxin), this state cyclescontinuously through an assembly-dis-assembly sequence. A second possibili-ty consistent with these results isdepicted in Fig. 2. Here, the dockedstate would consist of a partial assem-bly of SNAREs that is primed and readyto proceed promptly to fusion with theelevation of intracellular calcium. Onlyafter fusion would SNAREs be trappedin a tight ternary complex as was pro-posed by Hanson and colleagues9. Thismodel does not require a reversibledocking process. At present there is lit-tle in vitro biochemical evidence that apartially assembled SNARE complexsensitive to botulinum or tetanus toxinexists. However, such partial assemblycould be stabilized by one of a numberof SNARE interacting proteins such asRab, Sec1, synaptotagmin, complexinsor the recently discovered tomosyn10.The proposal that the stable ternarycomplex represents a postfusion state isalso consistent with the ability of appro-priately paired unassembled SNAREs tomediate membrane fusion6. The ener-gy released during the formation of theternary complex could potentially serveto drive the fusion event, leaving a sta-ble low-energy complex that requiresdisassembly and energy input prior tovesicle recycling.

These experiments thus constrainthe debate about the placement of thestable SNARE complex in the secreto-ry-vesicle life cycle. The use of the cel-lular milieu as the modern cuvette forstudying native molecular interactionspromises to become increasingly impor-

intracellular calcium elicit an exocytot-ic burst proceeding very rapidly forabout one second, followed by a pro-longed slower phase of secretion. Inboth types of secretory systems, it isbelieved that release-ready vesicles arethe first to go when a stimulus arrives.

Given the importance of SNAREs invesicle–plasma membrane interactions,what is the arrangement of SNAREs inthe docked or release-ready vesicles? Inits original form, the SNARE hypothe-sis proposed that vesicles would fuse

with target membranes viathe following sequence ofevents3: first, SNAREs onthe vesicle and targetmembranes assemble witheach other, forming thetight ternary complex and‘docking’ the vesicle to thetarget membrane; second,cytosolic factors, SNAPand the ATPase NSFsequentially bind. Theenergy derived from ATPhydrolysis then serves tofuse the two membranesand to disassemble theSNAREs. Thus, accordingto this scenario, one wouldpredict that docked vesi-cles would be invulnerableto toxin attack, and thatmeasurements of secretionin the presence of toxinswould reveal a smallamount of secretion corre-sponding to that derivedonly from the dockedpool.

The experiments of Xuand colleagues indicate

Fig. 1. A model of the role of SNAREsin mediating membrane fusion in neu-ronal exocytosis. Prior to docking withthe plasma membrane, SNARE proteinson the vesicle membrane are subject tocleavage by botulinum or tetanus toxin.SNARE proteins on docked vesicles arecoiled tightly in a ternary complex, pro-tecting them from toxin attack.Following the elevation of intracellularcalcium, the two membranes fuse. NSFacts to dissamble the SNAREs, freeingthem and allowing the vesicle to berecycled. In this model, the membranebinding step is reversible, allowingdocked vesicles to cycle continuouslythrough a toxin-sensitive state, withoutproceeding to fusion.

NSF-ATP? NSF-ATP

Ca++

X X X X

Syntaxin

Synaptobrevin

SNAP-25

Tetanus or Botulinum toxin

X

Syntaxin

Synaptobrevin

SNAP-25

Tetanus or Botulinum toxin

NSF-ATP

Ca++

X

Fig. 2. Alternate model for SNARE-mediated exocytosis.As in Fig. 1, the SNARES on vesicles prior to docking aresensitive to toxin attack. After vesicle docking, the SNAREsare partially assembled and still sensitive to toxins. Thisstate probably requires stabilization by other SNARE-inter-acting proteins. Vesicle fusion and recycling as in Fig. 1. Themembrane binding step need not be reversible, but a newstable but toxin-sensitive SNARE arrangement that docksvesicles is proposed.

Bob Crimi

Bob Crimi


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

news and views


thinly myelinated or unmyelinated fibersof specific thermoreceptors and variousnociceptors7, suggesting a role in sensingtissue damage and temperature. Consis-tent with this, some neurons of lamina Iare selective either for noxious stimuli orfor innocuous changes in temperature8,9.Moreover, lamina I neurons project to thethalamus and several other brain regionsvia the contralateral spinothalamic tract,which is important for the perception of

provides strong evidence favoringat least a partially dedicated system.

Pain in normal animals andhuman beings is usually caused bystimuli that are strong enough tothreaten the integrity of the tissuesinvolved. Detection of such eventsis called nociception; sense organswith the appropriate characteristicsfor nociception have been knownfor over thirty years and are highlyconserved between different mam-malian species6. However, pain canarise from stimuli or events that arenot noxious. This has been used toargue that nociception is notdirectly associated with pain3,although there is substantial evi-dence that noxious stimuli areclosely related to indications ofaversive behavior in animals or ofpain in human beings.

Han and colleagues have used acombination of electrophysiologyand histology to characterize theneurons of lamina I in the dorsalhorn of the cat spinal cord, one of the ear-liest stages in the nociception pathway.The dorsal-horn gray matter is dividedinto zones, or laminae, based on varia-tions in the size, shape and density of itsneurons. Primary sensory fibers from thesomatic and visceral tissues terminatethroughout the dorsal horn, and the ter-minals from different functional classesare segregated in different laminae. Lam-ina I receives its input largely from the

Most people accept that the sensations ofvision, hearing, taste and vibration aresubserved by dedicated neural pathways,in which the physical stimulus is trans-duced by specialized sensory cells andprocessed by certain neurons and regionsof the central nervous system that aredominated by their particular sensoryinput. Is this also true of pain? Aristotleheld that pain was an emotion ratherthan a sensation, and although this viewhad been largely abandoned by the 20th

century1, the question of whether theneural basis of pain resembles that ofother sensations remains highly contro-versial to this day2,3. One concept is thatpain is processed by dedicated painreceptors that provide input to specificcentral pain pathways. An alternateview3,4 is that pain is not signaled by aspecific ‘labeled-line’ system of neuronsbut instead by a special form of activa-tion in neurons that are also concernedwith other somatic sensations. In thisview, painful events would be distin-guished from innocuous stimuli by acombination of functional characteris-tics, including the frequency and patternof firing in neurons whose lesser or dif-ferent activities result in other sensoryexperiences. A new study in this issue(pp. 218–225) by Han, Zhang and Craig

Getting a line on pain: is itmediated by dedicatedpathways?Edward R. Perl

Are painful stimuli signaled to the thalamus by distinctpathways? A new finding of structure/function correlationsin spinal neurons suggests that the answer is yes.

Fig. 1. Diagram representing the relationshipsbetween functional category, cellular morphologyand thalamic projection of spinal dorsal horn lamina Ineurons proposed by Han, Zheng and Craig(adapted from ref. 2). MDvc, medial dorsal nucleusventral caudal portion; VMpo, ventral medialnucleus, posterior part; VPI, ventral posterior infe-rior nucleus; NS, nociceptive specific neuron; COLD,innocuous cooling neuron; HPC, heat, pinch, coolingneuron; WDR, multireceptive neuron (‘widedynamic range’).

Edward Perl is at the Department of Physiology,University of North Carolina-Chapel Hill, CB#7545, Chapel Hill, North Carolina 27599, USAemail: [email protected]

MDvc

VPI

VMpo

Centralsulcus

Midline

Thalamus

Lamina I

COLD

NS

HPC

WDR

Spinothalamictract

Dorsalanteriorinsula

Area 3a

Area 24c Cerebralcortex

Spinal cord

tant in determining the inner workingsof secretory machinery. In particular, itshould now be possible to pinpoint theroles of NSF and SNAP as well as that ofthe growing family of SNARE-interact-ing proteins in the synaptic-vesicle lifecycle, using methods like those of Xuand colleagues.

1. Coloquhoun, D. & Sakmann, B. Neuron 20,381–387 (1998).

2. Xu, T., Binz, T., Niemann, H. & Neher, E.Nature Neurosci. 1, 192–200 (1998).

3. Rothman, J.E. Nature 372, 55–63 (1994).

4. Scheller, R.H. Neuron 14, 893–897 (1995).

5. Söllner, T., Bennett, M.K., Whiteheart, S.W.,Scheller, R.H. & Rothman, J.E. Cell 75,409–418 (1993).

6. Weber, T. et al. Cell 92, 759–772 (1998).

7. Banerjee, A., Barry, V.A., DasGupta, B.R. &Martin, T.F.J. J. Biol. Chem. 271,20223–20226 (1996).

8. Schweizer, F.E. et al. Science 279, 1203–1206(1998).

9. Hanson, P.I., Roth, R., Morisaki, H., Jahn, R.& Heuser, J.E. Cell 90, 523–535 (1997).

10. Fujita,Y. et al. Neuron 20, 905–915 (1998).

Bob Crimi


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


news and views

pain from the opposite side of the body1.An important question is whether the

neurons of lamina I are specialized fordifferent types of painful stimuli, andwhether these different classes send theiroutputs to particular brain regions. Suchan arrangement would have significantimplications for understanding howafferent signals combine to give rise topain. Previous papers from Craig’s labo-ratory and others have described, in catand monkey, at least three functionalclasses of neurons in lamina I that pro-ject to the thalamus2,9. One class is noci-ceptive-specific (NS) with inputs fromone or more types of nociceptor, anoth-er responds to innocuous cooling(COLD) and a third class is multimodal,responding to heat, pinch and cooling(HPC). The authors have reported pre-viously that these different functionalclasses project to different thalamicnuclei, which in turn project to differentcortical regions (Fig. 1). The presentfindings greatly strengthen their pro-posed functional classification, by show-ing that the three classes of neuronsdefined by their physiological selectivitycorrespond to particular morphologicaltypes, distinguished by the shapes oftheir cell bodies and dendritic processes.This is consistent with a particulardefined pathway for nociception andpain and seems difficult to reconcile witha model in which pain arises from func-tional interactions between neurons thatare not themselves specified for particu-lar sensory experience.

The new findings leave unresolved therelative importance for pain of the NS andHPC neurons of lamina I and the so-called wide-dynamic-range (WDR) neu-rons described by other researchers. WDRneurons are excited only modestly byinnocuous mechanical stimuli and firemost strongly in response to noxiousmechanical or heat stimuli. They arefound mainly in laminae V and VII andproject to the thalamus via the spinothal-amic tract10. They are relatively rare inlamina I (especially in cats), and so theauthors were not able to characterize themin the present study. However, they havebeen proposed to convey pain signals tothe brain in humans and to be importantfor aversive behavior in animals6,11. WDRneurons are multipolar cells12, similar toHPC neurons, and project to at least onethalamic locus2. Is it possible that HPCand WDR neurons are of the same typeand serve similar functions?

Why has this structure/function cor-relation not been found in previous stud-

We now know from brain imaging stud-ies that a given sensory stimulus acti-vates multiple cerebral cortical zones.Somatic sensations, including pain,seem to involve an as yet unknown com-bination of ascending afferent informa-tion with an intrinsic background ofneural activity modified by past historyor disease.

It seems likely that there is merit inthe views of each of the most diametri-cally opposed camps in the controversyabout the nature of pain mechanisms.Accepting the observations of Han andcolleagues, there is a system of centralneurons, morphologically and func-tionally distinct beginning in lamina Iof the spinal cord, whose usual functionis to inform the organism about tissuedamage or the threat of such damage. Itseems to be situated so that activity insome of its constituents at several levelsof the central nervous system normallyleads to pain or equivalent reactions.However, once these messages reach thebrain, they are processed in parallel bymore than one set of central neurons.The eventual sensory experience almostsurely represents the interplay in a mix-ture of serial and parallel processingintegrated into the activity of otherneural systems related to elaboration ofconsciousness, emotion and memory.

1. Keele, K.D. Anatomies of Pain (Thomas,Springfield, Illinois, 1957).

2. Craig, A.D. Pain Forum 7, 1–14 (1998).

3. Wall, P.D. Pain 62, 389–391 (1995).

4. Melzack, R. & Wall, P.D. Science 150, 971–979(1965).

5. Belmonte, C. & Cervero, F. Neurobiology ofNociceptors (Oxford Univ. Press, New York,1996).

6. Dubner, R., Kenshalo, D.R., Maixner, W.,Bushnell, M.C. & Oliveras, J.L. J.Neurophysiology 62, 450–457 (1989).

7. Light, A.R. The Initial Processing of Pain andIts Descending Control: Spinal and TrigeminalSystems (Karger, Basel, 1993).

8. Christensen, B.N. & Perl, E.R. J. Neurophysiol.33, 293–307 (1970).

9. Dostrovsky, J.O. & Craig, A.D. J. Neurophysiol.76, 3656–3665 (1996).

10. Willis, W.D. The Pain System (Karger, Basel,1985).

11. Mayer, D.J., Price, D.D. & Becker, D.B. Pain 1,51–58 (1975).

12. Ritz, L.A. & Greenspan, J.D. J. Comp. Neurol.238, 440–452 (1985).

13. Ferrington, D.G., Sorkin, L.S. & Willis, W.D. J.Physiol. (Lond.) 388, 681–703 (1987).

14. Light, A.R., Trevino, D.L. & Perl, E.R. J. Comp.Neurol. 186, 151–171 (1979).

15. Woolf, C.J. & Fitzgerald, M. J. Comp. Neurol.221, 313–328 (1983).

ies? In most experiments, Han and col-leagues used a more selective method thanother researchers to choose their neurons,testing for antidromic activation of pro-jecting neurons by finding and stimulat-ing a restricted thalamic region thatactivates lamina I neurons. Lamina I neu-rons are difficult to record, because theyare flattened at the interface between theoutermost part of the gray matter and thelargely myelinated fibers of the spinal cordtracts. Moreover, their form is not easilyappreciated unless viewed in horizontalhistological preparations, which were notused in previous studies13–15. Anotherconsideration is that most previous stud-ies did not concentrate exclusively on lam-ina I neurons, but included observationsin the subjacent lamina II, possibly blunt-ing distinctions. A further difficulty is thevariation between individual neuronswithin the categories (e.g., fusiform,pyramidal) reported by Han and col-leagues; although this need not invalidatetheir classification, it would have madematters difficult for previous workerslacking the advantage of the horizontalplane in seeking structure/function rela-tionships within this population.

The degree of specificity implied bythe new results does not necessarily indi-cate that the system is genetically hard-wired. The shape of neurons couldequally reflect their environments,including the connections they receivefrom other neurons, rather than beingan inherent property of each neuronitself. But whichever mechanism isresponsible for the observedstructure/function relationship, the exis-tence of neurons partially dedicated toconveying information about noxious orthermal stimuli fails to explain in itselfeither the sensation of pain or its aber-rations, just as our present understand-ing of the auditory system fails toexplain hearing. Even if the functionaland structural specialization of spinalneurons projecting pain-relevant infor-mation is reminiscent of other sensorysystems, this does not indicate a tele-phonic relay arrangement of the sortimplied by the term ‘labeled line’. Mam-malian sensory systems are generallyplastic and reflect their past history. It isnot surprising that the same should betrue of a system associated with pain,and as Han and colleagues acknowledge,the specificity they describe may not beabsolute, so that transmission by the dif-ferent pathways might be modulated notonly by past experience but also byongoing activity from other brain areas.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

news and views


plasticity also is a top-down process andthat plastic changes can be propagatedin the reverse direction.

These plastic changes seem likely tobe related to the normal role of the cor-ticofugal system in the transmission ofsensory information. Tounderstand how thismight be, it is useful toreview some of thefindings on corticofugalconnections in othersystems. In one earlystudy6, the cat primaryvisual cortex wasreversibly cooled toinactivate the corticofu-gal feedback to the lat-eral geniculate nucleusof the thalamus. Thisinactivation reducedboth the light-evokedresponses and the levelof spontaneous activityin most thalamic relayneurons. Thus, corti-cofugal fibers seem todirectly but weaklyexcite relay cells, as wellas disinhibit them byreducing the strength ofintrinsic inhibitorymechanisms. A laterstudy7 provided morespecific evidence aboutthe wiring of the corti-cothalamic loop; gluta-mate, appliediontophoretically to asmall patch of the pri-mary visual cortex,excited LGN neuronsthat shared receptivefields with cortical neu-rons at the iontophore-sis site but inhibitedLGN neurons that wereout of register. More-

Sensory information is relayed to thecortex via the thalamus, yet there arenearly ten times as many fibers project-ing back from the cortex to the thala-mus as there are in the forwarddirection from thalamus to cortex1,2.The function of this massive feedbackprojection has puzzled neuroscientistsfor decades. Initial attempts to addressthis question were largely confined tothe visual system, but more recently,studies on the corticofugal modulationof subcortical responses in the auditorysystem3,4 have provided additionalinsight. The most recent contribution5,in this issue of Nature Neuroscience(p 226), extends the story to thesomatosensory system and adds a newtwist. From the combined results in dif-ferent sensory systems, we can see theemergence of a more generalized under-standing of the role of the corticothala-mic loop in sensory informationprocessing.

Ergenzinger and colleagues5 deliv-ered an NMDA receptor antagonist(APV) into the primary somatosensorycortex of monkeys, at the site of thehand representation. After severalmonths of treatment, the tactile recep-tive fields within the somatosensory partof the thalamus (the ventroposterior(VP) nucleus) showed an enormousenlargement in the hand region. Simi-lar, albeit less dramatic, enlargementswere also found after acute treatment.The results are intriguing because theychallenge the traditional view of senso-ry plasticity as a bottom-up process,whereby changes on one level are sim-ply passed on to the next higher level.By contrast, the new study shows that

Cortical control of thethalamus: top-downprocessing and plasticityJosef P. Rauschecker

Blocking cortical NMDA receptors leads to a dramatic expansionof thalamic receptive fields. The results help to illuminate therole of corticofugal connections in sensory processing.

Josef Rauschecker is at the Georgetown Institutefor Cognitive and Computational Sciences,Georgetown University, Washington, DC 20007,USAemail: [email protected]

over, simultaneous recordings of corticaland LGN activity revealed that corticalfiring was associated with an increasedprobability of firing in LGN neuronswith overlapping receptive fields but notmore distally, suggesting that corticofu-gal connections contribute to thalamicexcitation in a very specific mannerdependent on their respective locationson the retinotopic grid.

The latter result ties in well withanother study8 on the visual system, inwhich the authors recorded simultane-ously from pairs of LGN neurons thatwere stimulated by a moving bar andthen removed cortical feedback. Withcortical feedback, the LGN neuronsfired synchronously, whereas without it,this synchrony was lost. In this case, thecorticothalamic loop would be expect-ed to amplify the response to the mov-

Layer IV

Layer VI

Thalamus

Sensoryperiphery

Cortex

+

Fig. 1. Thalamocortical network in which the corticofugalprojection provides narrow, highly specific positive feedback(dark orange) to a thalamic neuron with a classical center-sur-round receptive field, as well as negative feedback (viainhibitory interneurons, shown in green) to a large number ofsurrounding thalamic neurons (pale orange), creating a wideinhibitory ‘suppression field’ (ref. 2). During normal sensoryprocessing, this circuitry helps to sharpen the contrast of thesensory input in time and space (or frequency in case of theauditory system) and suppresses irrelevant information (‘ego-centric selection’, ref. 3, 4, 9). When the cortex is inactivated,e.g. by means of NMDA receptor antagonists (ref. 5), thisleads to an unmasking of the suppressed input from recurrentcollaterals of other thalamic neurons within the suppressionfield, which causes an enlargement of thalamic receptivefields. As the cortical activity block continues for severalmonths, this effect on thalamic neurons can be surprisinglylarge, perhaps magnified by secondary plastic changes.

Bob Crimi


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


news and views

Singer’s classical paper was based solelyon data then available from the visualsystem, the concept can now be extend-ed to the auditory (and perhaps mostother) sensory systems, where corti-cofugal feedback may play a role in anal-ogous processes such as auditory streamseparation and scene analysis11.

The circuitry required for the mixedexcitatory-inhibitory feedback provid-ed by the cortico-thalamic loop isdepicted in Fig. 1. The highly repetitivestructure seemingly does not leave anyroom for selectivity, but as pointed outby Koch12, the answer may lie in thegating properties of the thalamicNMDA receptors. Thalamic relay neu-rons receive excitatory glutamatergicinput both from peripheral sensorypathways and from corticofugal con-nections. Assuming that these inputs actpartly through NMDA receptors, therelay neurons would be activated onlyif they receive sensory input and corticalfeedback at the same time. Given theresults of Sillito and colleagues8, onlythose inputs that occur within a narrowtemporal window would be amplified.Similarly, only stimuli that fall within adefined region of the sensory receptorsurface (for instance the retina orcochlea) would be amplified. Anythingthat is part of the irrelevant din wouldbe suppressed, because without corticalfeedback (and hence NMDA receptoractivation), the inputs from the periph-ery are not strong enough to overcomethe effects of thalamic inhibitoryinterneurons, which are themselves dri-ven by both sensory and corticofugalinputs (see Fig. 1). Thus, although thecorticofugal input to the thalamus helpsto amplify certain inputs (similar to a‘winner-take-all’ mechanism), it alsogenerates a large ‘suppression field’2,which exists outside the classical recep-tive field.

This circuit model for the selectivemodulation of sensory information canalso account for the results of Pons andcolleagues. Applying an NMDA recep-tor antagonist to the cortical neuronsessentially silences this part of the cor-tex (an effect that complicates the inter-pretation of the early results on theeffects of NMDA receptor antagonists invisual cortex; see ref. 13). The antago-nist has no direct effect on the thalamus(in contrast to cortical lesions, whichlead to retrograde thalamic degenera-

ing bar stimulus, because synchroniza-tion of thalamic relay neurons meansthat they will elicit temporally overlap-ping excitatory postsynaptic potentials(EPSPs) in any common cortical targetcells, thereby increasing the chance thatsuch cortical cells will fire.

The early work of Tsumoto and col-leagues in vision is complemented in anintriguing fashion by recent work ofSuga and colleagues on the corticofugalprojections of the bat’s auditory system.When cortical neurons tuned to specif-ic frequencies are inactivated by localinjections of lidocaine4, the auditoryresponses of neurons in the thalamicmedial geniculate nucleus (MGN) tunedto the same frequency are reduced,whereas responses of neurons tuned todifferent frequencies are enhanced. Thesame is true, incidentally, for neuronsin another subcortical auditory nucle-us, the inferior colliculus, suggestingthat the role of the corticofugal projec-tion may be quite general9. Moreover, ifthe cortical site is (electrically) stimu-lated rather than blocked, the effect isto increase and sharpen the responses ofa particular subclass of neurons impor-tant in the bat’s echolocation behavior,the delay-tuned neurons, in both thala-mus and colliculus3. Like frequency andretinotopic location, delay tuning ismapped in a topographic fashion corti-cally and subcortically and again, theeffect of cortical stimulation is highlyspecific for neurons with the same delaytuning on both levels.

These findings suggest a commonrole for the corticothalamic loop in sen-sory information processing in differentsystems: the corticofugal projection pro-vides positive feedback to the ‘correct’input while at the same time suppress-ing irrelevant information. It seems thatthe positive feedback is very preciselymatched between cortex and thalamus,such that a given thalamic neuronreceives excitation only from a small areaof cortex that shares the same stimulusselectivity (for instance, a particularlocation in visual space or a particularsound frequency). Based on their workin the auditory system, Suga and col-leagues have termed this principle ‘ego-centric selection’. This process leads to atemporal and spatial contrast sharpen-ing of the sensory input and, as suggest-ed by Singer10, may play a crucial role infigure-ground discrimination. Although

tion). Thus, immediately after applyingAPV to the cortex, receptive fields in thethalamus are enlarged, which can beexplained as a disinhibition of the sup-pression effects of the corticofugal fibers.Applying the APV chronically, as in themain finding of the study, renders thethalamic receptive field enlargementsenormous, suggesting that the originaldisinhibition effect leads to further plas-tic changes, perhaps in the thalamusitself. This may also yield a late explana-tion of the earlier results by Pons andcolleagues of a massive expansion of thecortical hand representation after chron-ic deafferentation14.

As Yan and Suga point out in theirlatest paper, published earlier this yearalso in Nature Neuroscience9, more sub-tle forms of plasticity could also bemediated by the corticofugal loop. Itcould be that thalamic maps are con-stantly adjusted by sensory experience,as are cortical maps. In conclusion,function and plasticity have often beenthought of as highly interwoven aspectsof cortical structure, neural activitybeing the factor that ties them together.The study by Ergenzinger et al.5 makesit clear that these dual aspects of high-er perceptual and cognitive systemsneed to be extended to the thalamus, atleast insofar as the latter is controlled byits cortical input.

1. Guillery, R. W. J. Comp. Neurol. 130,197–222 (1967).

2. Jones E. G. The Thalamus (Plenum, NewYork, 1985).

3. Yan, J. & Suga, N. Science 273, 1100–1103(1996).

4. Zhang, Y. Suga, N. & Yan, J. Nature 387,900–903 (1997).

5. Ergenzinger, E. R., Glasier, M. M., Hahm, J.O. & Pons, T. P. Nature Neurosci. 1, 226–229(1998).

6. Kalil, R. E. & Chase, R. J. Neurophysiol. 33,459–474 (1970).

7. Tsumoto, T., Creutzfeldt, O. D. & Legendy,C. R. Exp. Brain Res. 32, 345–364 (1978).

8. Sillito, A. M., Jones, H. E., Gerstein, G. L. &West, D. C. Nature 369, 479–482 (1994).

9. Yan, J. & Suga, N. Nature Neurosci. 1, 54–58(1998).

10. Singer, W. Physiol. Rev. 57, 386–420 (1977).

11. Bregman, A.S. Auditory Scene Analysis (MITPress, Cambridge, Massachusetts, 1990)

12. Koch, C. Neuroscience 23, 399–406 (1987).

13. Rauschecker, J.P. Physiol. Rev. 71, 587–615(1991).

14. Pons, T. P. et al. Science 237, 1857–1860(1991).


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


scientific correspondence

Where is the sun?Jennifer Sun1 and Pietro Perona1,2

1 California Institute of Technology 136-93, Pasadena, California 91125, USA2 Universita di Padova, Via Ognissanti 72, 35131 Padova, Italy

Correspondence should be addressed to P.P. ([email protected])

When we interpret a shaded picture as a three-dimensional (3D)scene, our visual system often needs to guess the position of the lightsource in order to resolve a convex-concave ambiguity. For morethan a century, psychologists have known that the visual systemassumes that light comes from above and have argued that thisassumption is ecologically justified because our everyday light source(the sun) is overhead. Our experiments reveal that people’s preferredlighting direction is not directly overhead, but rather shifted to theleft, and this preference is reflected in art spanning two millennia.Furthermore, we find a strong correlation between people’s hand-edness and their preferred lighting. We suggest that what counts isnot so much where the sun is, but where you like the sun to be.

The shaded shapes in Fig. 1a are typically perceived as convexbubbles surrounding concave indentations, all lit from above. Notethat this image is also consistent with a different physical scene:indentations surrounding bubbles, all lit from below. This secondperception, however, is difficult to achieve. Such an asymmetrybetween the perceptual saliency of equally valid 3D interpretationsdemonstrates that our visual system prefers the assumption thatlight is coming from above1–3. Does this preference apply uniform-ly to all lighting directions that are above the horizon? Perhaps thereis instead a preferred direction? If so, one might reason it to be direct-ly overhead. Is this intuitive guess correct?

We addressed these questions by measuring the time it takesto detect, within a group of ‘distractor’ bubbles, a single ‘target’bubble that is lit differently (Fig. 1b). Recent studies suggests thatthe light-from-above assumption is used by the visual system forinterpreting quickly and in parallel some basic aspects of 3Dscenes4–8; the target pattern may be detected quickly (pop-out)only when the distractors, but not the target, can be interpretedas convex and lit from above5–9. We simulated different directionsof lighting by varying the shading gradient of the distractor bub-bles. The target bubble was shaded to simulate illumination fromthe opposite direction (Fig. 1b).

Data from twelve naive subjects shows that the visual systemdoes not respond uniformly to all lighting directions that are abovethe horizon (Fig. 1c). There is clearly a preferred direction of light-ing where detection requires the shortest display time. Surpris-ingly, this preferred direction is not directly overhead (zerodegrees). Subject PG, for instance, performs best with a lightingdirection that is between 30 and 60 degrees left of the vertical(Fig. 1c). Furthermore, there is a consistent preference for left light-ing over right lighting. This same marked left–right asymmetry isevident in the averaged data of all twelve subjects. As the angle ofillumination increases, the preference for left lighting becomesincreasingly pronounced (Fig. 1d).

This asymmetry in our data may explain a qualitative observa-tion made by Gestalt psychologist Metzger, who noted that left-litscenes have a superior perceptual value over right-lit ones10. Heascribed this asymmetry to the convention of setting up desk lampson the left, presumably so that the writing hand does not cast ashadow on the page. Over time, he hypothesized, one learns to per-ceive left-lit scenes as being more ‘natural’. Metzger’s explanationmay be somewhat restrictive: our visual environment extendsbeyond our writing desk. We tend to position a movable light

180 150 120 90 60 30 0 30 60 90 120 150

150

200

250

300

350

400

450

500

Lighting Direction of Distractors degrees

MFV

Dur

atio

n m

sec

Performance vs Lighting Direction Subject PG

Left Lighting Right Lightinga b c

Lighting direction of distrators (degrees)

MFV

du

rati

on

(m

s)

Performance versus lighting direction(subject PG)

30, 30 60, 60 90, 900

20

40

60

80

100

120

140

Lighting Directions Compared (rightleft)

Dif

fere

nce

in M

FV

Du

rati

on

s m

sec

g g

p < 0.03

p < 0.007

p < 0.00112 Subjects

Lighting directions compared (right-left)

Perfomance differences at various lighting directions

d

Dif

fere

nce

in M

FV d

ura

tio

ns

(ms)

Fig. 1. Shaded displays that may be interpreted as 3D shapes and measurement of preferredlighting direction. (a) Rotate the page to invert the shapes. (b) Images were generated on aSilicon Graphics Indigo2. Each bubble’ spanned approximately one degree. One target patternwas present at random among 23 distractor patterns in 50% of the trials. The remaining trialscontained 24 distractors and no target. The lighting direction is determined by the shading gra-dient of the distractors. Target-present test screens are shown for 2 of the 12 lighting direc-tions used in our experiment. We denote a lighting direction by its deviation from the verticalin degrees. Positive degrees indicate lighting from the left, and negative degrees indicate lightingfrom the right. Accordingly, lighting from directly overhead is designated as 0 degrees, and light-ing from directly below as 180 degrees. (c) We used a two-alternative forced-choice stimulusonset asynchrony (SOA) design with masking. Data was collected using a staircase method thatconverged at 67% accuracy performance. The most frequently visited (MFV) duration withineach block was used to estimate 67% accuracy performance. The duration necessary for 67%accuracy for each lighting direction is shown for subject PG. (d) We computed the mean dif-ference in necessary display duration between pairs of corresponding left-right lighting direc-tions over all 12 naive subjects. The necessary duration for a left-lighting condition issubtracted from the necessary duration for the corresponding right-lighting condition.

Left lighting Right lighting

p < 0.001

p < 0.007

p < 0.08


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


source, or position ourselves in relation to a fixed light source, suchthat our hand does not cast a shadow upon the object of our manip-ulation. Right handers would then develop a preference for left-lighting, and, as suggested by van Fieandt11, left-handers may wellshow the opposite lighting preference.

To investigate this possibility, we derived the preferred light-ing direction and handedness for our subjects. We defined eachsubject’s preferred lighting direction to be the lighting directionfor which target detection performance was best. This was esti-mated by fitting a parabola to the central portion of each subject’stime versus direction curve (e.g. Fig. 1c). and calculating the direc-tion for which the parabola reached a minimum. When prefer-ence for lighting direction is considered for left- and right-handersseparately, we find that both groups have a preference for left-lighting. However, the right-handers as a group prefer a lightingdirection that is significantly more toward the left than the left-handers’ preference (Fig. 2a).

Handedness, however, is not a strictly binary trait; rather it variesin a continuum12. We used a standard ten-item questionnaire thatevaluated the relative strengths of our subjects’ handedness13. Theresulting score ranges from -10 to 10, with positive values indicat-ing a bias for the right hand, and negative values for the left. Wheneach subject’s preferred lighting direction is plotted against this hand-edness score, a strong correlation is found (Fig. 2b).

If lighting preference is indeed related to handedness, why isn’t

the effect mirror symmetric, with left-handers having a preference forright-lighting instead of left? This may be explained because left-handers live in a right-handers’ world and are often forced to func-tion in environments that are designed for right-handers. If lightingpreference is not determined by a biological trait correlated withhandedness itself, but rather by handedness-related experience, thenone would expect that right-handers would prefer left-lighting,whereas left-handers, because of their mixed experience, wouldexhibit a weaker preference for either left or right lighting.

Quantitative measurements of the perception of shape fromshading using static displays have revealed no asymmetry betweenleft- and right-lighting conditions14,15, thereby failing to confirmMetzger’s observation. One might, therefore, suspect that the pref-erence for left lighting we observe with our fast-presentation para-digm is confined to the earlier stages of visual processing, and thatthis effect may become negligible under ecological viewing condi-tions. We have reason to think otherwise. We asked two naive sub-jects, one right-hander and one left-hander, to survey 225 masterpaintings and determine the predominant angle of lighting for each.The histogram of the measurements (Fig. 3) shows that the artistsmost often chose a lighting direction that is left of the vertical. Thispreference for top-left lighting may have resulted from an accidentalartistic convention. However, this is unlikely, as preference left-light-ing is found across schools and periods: from Roman mosaics,through Renaissance, baroque, and impressionist art. It is thereforepossible that top-left lighting may actually have a higher perceptualvalue than top-right lighting in natural viewing conditions, whichinvolve frequent saccades over the entire scene. Perhaps when sub-jects are required to make their shape judgments under prolongedscrutiny of a localized portion of the test stimulus14,15, the effectdrops below a measurable level.

AcknowledgementsThe authors are grateful to Dr. Marianne Tauber for providing the references to the

Gestalt literature. Support was provided by the NSF Engineering Reasearch Center

for Neuromorphic Systems at Caltech.

1. Rittenhouse, D. Trans. Am. Phil. Soc. 2, 37–42 (1786).2. Brewster, D. Edinburgh Phil. Trans. 15, 657 (1847).3. Ramachandran, V.S. Nature 331, 163–166 (1988).4. Enns, J.T. & Rensink, R.A. Science 247, 721–723 (1990).5. Braun, J. Perception 19, A112 (1990).6. Braun, J. Spatial Vision 7, 311–322 (1993).7. Kleffner, D.A. & Ramachandran, V.S. Percept. Psychophys. 52, 18–36 (1992).8. Sun, J.Y. & Perona, P. Vision Res. 36, 2515–2529 (1996).9. Sun, J.Y. & Perona, P. Nature 379, 165–168 (1996).10. Metzger, W., Gesetze des Sehens. Frankfurt am Main (1936).11. van Fieandt, K. Psychologischen Institut Monog. (Helsinki University, 1938).12. Durost, W.N. Genet. Psychol. Monog. 16, 225–335 (1934).13. Oldfield, R.C. Neuropsychologia 9, 97–113 (1971).14. Bülthoff, H.H. & Mallot, H.A. J. Opt. Soc. Am. A 5, 1749–1758 (1988).15. Todd, J.T. et al. Invest. Ophthal. Vis. Sci. 37, 4282 (1996).

scientific correspondence

Fig. 2. The preferred light direction corre-lates with handedness. (a) Subjects werecategorized based on which hand they usedfor writing. The mean preferred lightingdirection was compared between the sixright-handed and the six left-handed sub-jects. Although both groups show a prefer-ence for left-lighting, right-handers, as agroup, prefer a larger angle left of the verti-cal (23.3 degrees) than do left-handers (7.9degrees). (b) Preferred lighting direction iscorrelated with handedness (r = 0.83,p<0.01). * lefthander, p righthander.

Fig. 3. Painters tend to light scenes from top-left. 100, 100 and 25paintings were randomly selected from catalogues of the Louvre,the Prado, and the Norton Simon Museum. Two naive subjects eval-uated these paintings for lighting direction. Using a protractor,77 ± 0.55% of the paintings were classified as being lit from the left(p<0.05). The artists most often selected lighting directions that arebetween 30 and 60 degrees to the left of the vertical.

Lefthanders Righthanders0

5

10

15

20

25

30

35

p < 0.03

p < 0.0001

10 5 0 5 10-20

-10

0

10

20

30

40

50

60

Handedness

Pref

erre

d Li

ghtin

g D

irect

ion

deg

rees

Left Preference

Right Preference

lefthanderrighthander

a Preferred angles of lighting direction

Preferred lighting direction versus handedness

HandednessLefthanders Righthanders

An

gle

left

of

the

vert

ical

(deg

rees

)

90 60 30 0 30 60 900

10

20

30

40

50

60

70

Lighting Direction of Painting degrees

g g

Left Right

Lighting direction of painting (degrees)

Evaluation of lighting direction in paintings

bp < 0.0001

p < 0.03

Pre

ferr

ed li

gh

tin

g d

irec

tio

n(d

egre

es)

- -


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Recent studies suggest that tyrosine kinases are important forsynaptic plasticity1–3. The requirement for Src activation inlong-term potentiation (LTP) induction has been attributedto its enhancement of NMDA-receptor currents4–6 because anNMDA-receptor antagonist blocks this effect3. The non-recep-tor tyrosine kinases Src and Fyn potentiate receptors composedof NR1 and NR2A subunits, but have no apparent effect onreceptors composed of NR1 and other NR2 subunits5. Fur-thermore, deletion of the C-terminal domain of NR2A elimi-nates the potentiation of NR1/NR2A-receptor currents by Src5

and impairs synaptic plasticity and contextual memory7. Thus,several lines of evidence suggest that a unique tyrosine phos-phorylation site on the C-terminal of NR2A may confer sen-sitivity to Src and be critical for the induction of LTP. However,the mechanisms by which tyrosine kinase potentiates NMDA-receptor function are unknown.

In addition, NR2B subunits are phosphorylated by one ofthe Src family of tyrosine kinases8, but no functional effect ofthis phosphorylation has been identified. This discrepancybetween biochemical and physiological data raises the possibil-ity that either tyrosine phosphorylation of NR2B affects recep-tor properties other than channel activity, or another property ofNR2A may be responsible for the apparent selective potentia-tion by tyrosine kinase of receptors containing this subunit.

NMDA receptors are allosterically modulated by a variety ofendogenous extracellular ions9–13. Zinc, one of these modulators,is accumulated in some nerve terminals in specific brain regionsand released during neuronal activity14. Zinc inhibits NMDAreceptors at two independent sites15–17. A low-affinity zinc site isprobably located inside the channel pore, and binding of zinc tothis site causes a voltage-dependent inhibition of NMDA-receptorchannels. A high-affinity site is likely located outside the channelpore and causes a voltage-independent inhibition. Receptors con-taining NR2A subunits exhibit far greater affinity for zinc at theextracellular, voltage-independent site than receptors containing

other NR2 subunits18–21. This sensitivity to zinc is high enoughto allow ambient zinc (either in vivo22 or as a contaminant ofexperimental solutions23) to tonically inhibit NR1/NR2A recep-tors. Here we show that Src reduces zinc sensitivity of recombi-nant NR1/NR2A and NR1/NR2B receptors, thereby relievingtonic inhibition by zinc and potentiating the NMDA-receptorresponse. Splice variants and mutants of NR1 subunits that havea low apparent affinity for zinc are, as predicted, potentiated to alesser degree by Src. Conversely, NR1 mutants with higher appar-ent zinc affinity are potentiated to a greater degree. Using site-directed mutagenesis, we have identified three C-terminal tyrosineresidues of NR2A that are required for Src’s modulation ofNMDA-receptor zinc sensitivity. By showing that Src potentiatesNMDA receptors by specifically reducing tonic zinc inhibition,our data link two modulatory sites of NMDA receptors that werepreviously thought to be independent.

ResultsSRC REDUCES TONIC INHIBITION OF NR1/NR2A RECEPTORS

Our hypothesis was that potentiation of NMDA receptor cur-rents by Src may be due to its relief of tonic inhibition by zincacting at the voltage-independent site, in a manner analogousto polyamine relief of tonic proton inhibition of NMDA recep-tors13. To test this hypothesis, we examined the effects ofEDTA, a chelator of transition metals such as zinc, on Src’spotentiation of NR1/NR2A receptors. The NMDA receptorcurrents were recorded with a rapid perfusion system inHEK293 cells transiently transfected with NR1-1a/NR2Areceptors. In control experiments, the amplitudes of evokedNMDA-receptor currents did not change significantly overtime (Fig. 1a and b). When purified recombinant human c-Src was added to the internal solution, the amplitude ofNMDA-receptor currents was gradually increased over a five-minute period (Fig. 1a). On average, the peak of NMDAreceptor currents evoked by 100 µM glutamate and 30 µM

Tyrosine kinase potentiates NMDAreceptor currents by reducing tonic zinc inhibition

F. Zheng, M.B. Gingrich, S.F. Traynelis and P.J. Conn

Department of Pharmacology, Emory University School of Medicine, 1510 Clifton Road, Atlanta, Georgia 30322, USA

Correspondence should be addressed to F.Z. ([email protected])

Activation of the tyrosine kinase Src potentiates NMDA-receptor currents, which is thought to benecessary for induction of hippocampal long-term potentiation. Although the carboxy(C)-terminaldomain of the NR2A subunit contains potential tyrosine phosphorylation sites, the mechanism bywhich Src modulates synaptic plasticity and NMDA receptor currents is not fully understood. Herewe present evidence from NR1 mutants and splice variants that Src potentiates NMDA-receptor cur-rents by reducing the tonic inhibition of receptors composed of NR1 and NR2A subunits by extracel-lular zinc. Using site-directed mutagenesis, we have identified three C-terminal tyrosine residues ofNR2A that are required for Src’s modulation of the zinc sensitivity of NMDA receptors. Our data linktwo modulatory sites of NMDA receptors that were previously thought to be independent.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


glycine was potentiated by 53 ± 9 % (n = 10), whereas the peakof currents evoked by 300 µM glutamate and 100 µM glycinewas potentiated by 53 ± 10% (n = 4). Our data agree with the40% potentiation of recombinant NR1-3a/NR2A receptor bySrc reported previously5. As others5 and we have used satu-rating concentrations of glutamate and glycine, it is unlikelythat the potentiation of NMDA receptors by Src is due to achange in apparent affinity for glutamate or glycine.

If our hypothesis is correct, complete removal of the tonic inhi-bition should prevent the potentiation of NR1/NR2A receptor cur-rents by Src. To remove transition-metal contamination, we addeda trace amount of EDTA (10 µM) to all recording solutions. Assum-ing 100 nM contaminant zinc and 1 mM extracellular Ca2+, the freezinc is less than 0.03 nM, whereas the free Ca2+ remains at 0.99 mM.In the presence of EDTA, Src failed to potentiate NR1/NR2A recep-tors (Fig. 1a and b). In the presence of 10 mM tricine, another metalchelator, Src also failed to potentiate NR1/NR2A receptors (n = 10,data not shown). Thus, removal of tonic inhibition of NR1/NR2Areceptors by transition metals occludes potentiation by Src.

We used the ratio of the peak NMDA-receptor current in thepresence of EDTA (IEDTA) to the peak current in the absence ofEDTA (INo EDTA) as a measure of the amount of tonic inhibitionof NR1/NR2A by transition metals. A smaller ratio indicates lesstonic inhibition. Without Src added to the patch pipettes,IEDTA/INo EDTA ranged from 1.6 to 2.5 (mean, 2.2 ± 0.1, n = 6,Fig. 1c and d). With Src added to the patch pipettes, IEDTA/INo

EDTA ranged from 1.2 to 1.7 (mean = 1.4 ± 0.1, n = 7). Thus, Srcsignificantly reduced the tonic inhibition of NR1/NR2A recep-tors by transition metals (p<0.005). Inclusion of the phosphataseinhibitor orthovanadate (500 µM) with Src results in anIEDTA/INo EDTA value of 1.18 ± 0.03 (n = 3), which is not signif-icantly different from the value with Src alone.

SRC SHIFTS IC50 FOR THE HIGH-AFFINITY, ZINC SITE

Zinc is probably the transition metal responsible for the EDTA-sensitive tonic inhibition of NR1/NR2A receptors, although

there are several other candidate ions, such as cadmium or cop-per18. To directly assess the effect of Src on zinc-induced inhi-bition of NR1/NR2A-receptor currents, we usedtricine-buffered zinc to determine the IC50 value of zinc act-ing at its high-affinity, voltage-independent site on NR1/NR2Areceptors18. In Xenopus oocytes expressing NR1-1a/NR2Areceptors, tricine-buffered zinc inhibited NMDA-receptor cur-rents with an IC50 value of 15 nM for the high-affinity zinc site(data not shown), in agreement with a previous report18. InHEK293 cells, low concentrations of zinc caused a similar con-centration-dependent inhibition (Fig. 2a). With zinc concen-trations up to 2.2 µM, the NR1-1a/NR2A-receptor current stillexhibits a linear I–V relationship (Fig. 2b), indicating that upto this concentration, zinc acts exclusively on the voltage-inde-pendent, high-affinity site (n = 3). The IC50 for the high-affin-ity zinc site of NR1-1a/NR2A in HEK cells is 86 nM (Fig. 2c),which is fourfold higher than the IC50 obtained in oocytes18.We are uncertain about the cause of this fourfold difference.Based on the fitted dose–response curves and IEDTA/INo EDTAvalues (2.86 for oocytes, n = 10; 2.13 for HEK cells, n = 6), weestimate the zinc contamination to be 275 nM for both record-ing systems, which is in line with a previously reported value(300 nM)23 and our own measurements (338 nM by VG Plas-ma Quad 3 ICP-MS mass spectrometer, 330 nM by Jarrell-Ash965 ICP plasma emission spectrometer). Addition of Srcincreased the IC50 for NR1-1a/NR2A receptors in HEK cells to392 nM, approximately fivefold (Fig. 2c). Based on this IC50value and the estimated zinc contamination level (275 nM),we predict that the potentiation caused by EDTA in the pres-ence of Src should be 42%, which is very close to the actualpotentiation observed experimentally (44 ± 9%).

SRC POTENTIATION CORRELATES WITH ZINC INHIBITION

If Src potentiates NMDA receptors by reducing tonic inhibi-tion by zinc, NMDA receptors that are less sensitive to zincshould also be potentiated to a lesser degree by Src. Recent

articles

Fig. 1. Src potentiates NR1/NR2A receptors byreducing tonic zinc inhibition. (a) Removal of toniczinc inhibition of NR1-1a/NR2A receptors occludesthe potentiation of these receptors by Src. NMDA-receptor current traces at left (labeled 1) were thefirst of a series of current traces recorded 60–90 sec-onds after obtaining the whole-cell configuration.Traces at right (labeled 30) were the last of 30 con-secutive traces recorded every ten seconds over afive-minute period. EDTA is added into extracellularsolutions to chelate zinc in some experiments,whereas recombinant human c-Src is added into theinternal solution of patch pipettes. Under controlconditions, no significant rundown of the NMDA-receptor current was observed. Inclusion of c-Src(30 units per ml) resulted in a 50% potentiation.However, this potentiation was not observed in thepresence of 10 µM EDTA. (b) Average changes ofthe NMDA-receptor current amplitudes over thesame periods. Error bars are standard error for allpanels. Number of cells is indicated in parentheses.P control (5), L Src (10), G Src and EDTA (5). (c) The effects of EDTA (10 µM) on NMDA currentsevoked by agonists (100 µM glutamate, 30 µM glycine) from representative HEK 293 cells with or without inclusion of Src in the pipette solu-tion (Vh = -50 mV). (d) The potentiation of whole-cell NR1-1a/NR2A-receptor currents by EDTA in HEK cells is reduced by inclusion of Srcin the patch pipette (*p<0.01). Circles are control cells; squares are cells perfused with Src.

a b

c d

No Src

Src

Src and EDTA

Time (min)

% o

f in

itia

l cu

rren

t

Control Src

I ED

TA/I

NO

ED

TA

Control SrcEDTA EDTA


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 2. Src shifts IC50 for thevoltage-independent zinc siteof NR1/NR2A receptors. (a)Typical current traces showingdose-dependent inhibition ofNR1-1a/NR2A receptors byzinc in HEK 293 cells (Vh, -50mV). Tricine was used to bufferzinc. (b) The I–V relationshipof the peak NMDA-receptorcurrents shows no voltage-dependent inhibition at up to 2µM buffered zinc. P 2230 nM, L 223 nM, G 0 nM. (c) Srcshifts the IC50 for the high-affinity, voltage-independentzinc site of NR1-1a/NR2Areceptors. Tricine was used tobuffer zinc. Each point on thecurve is averaged data from4–9 cells. The IC50 is 86 nMwithout Src and 392 nM withSrc. P control, L Src.


studies have identified some splice vari-ants and mutant NR1 receptors withreduced zinc sensitivity21. These recom-binant receptors were tested in HEK cellsto determine the amount of tonic inhi-bition and Src potentiation for each. TheIC50 value for NR1-1a(C744A,C798A)24,co-expressed with NR2A, is 168 nM inoocytes, about eightfold higher than thewild-type NR1-1a/NR2A receptor (Fig.3a). In HEK cells, this mutant is onlyslightly inhibited by ambient zinc, asindicated by the IEDTA/INo EDTA ratio of1.11 ± 0.03 (Fig. 3b). As predicted, NR1-1a(C744A,C798A) is not potentiated bySrc in HEK cells (Fig. 3c). NR1-1b,which has a 21-amino-acid insertion inthe N-terminal region encoded by alter-natively spliced exon 5, also showsreduced zinc sensitivity, with an IC50value of 198 nM for heteromeric recep-tors containing NR2A in oocytes (seealso refs. 18 and 21). As expected,recombinant NR1-1b/NR2A receptors,with an IEDTA/INo EDTA ratio of 1.08 ± 0.04 in HEK cells, werenot potentiated by Src (Fig. 3b and c).

With the NR1-1b and the double cysteine mutant of NR1-1a, we demonstrated that reduction of tonic inhibition indeedreduces the Src potentiation of NMDA receptors. By the samereasoning, if a mutation within NR1 exon 5 could restore zincsensitivity of the heteromeric receptors to levels comparableto wild-type receptors lacking exon 5, it should also restore thepotentiation by Src. Several point mutations within exon 5 arereported to restore zinc sensitivity to various degrees21. NR1-1bm207-211, a triple point mutant (K207G, R208G, K211G)in the exon 5 region13, exhibits the highest level of zinc sensi-tivity among all mutant receptors tested, with an IC50 value of40 nM in oocytes (Fig. 3a). Recombinant NR1-1bm207-211/NR2A receptors are under a significant amount of tonic

inhibition as indicated by an IEDTA/INo EDTA ratio of 1.77 ± 0.09(Fig. 3b). Again, consistent with our hypothesis, the peak cur-rent of NR1-1bm207-211/NR2A receptors is potentiated by Src(Fig. 3c). Overall, the amount of tonic inhibition of the NMDAreceptors (IEDTA/INo EDTA) is highly correlated (R = 0.997,p<0.005) with the amount of potentiation by Src of NMDAreceptors (Fig. 3d).

C-TERMINAL TYROSINES ARE CRITICAL FOR SRC’S ACTION

To locate the tyrosine residues critical for modulation of zincsensitivity by Src, we made a series of point mutations in the C-terminal region of NR2A (Fig. 4a). The amino-acid sequence ofthe C-terminal region contains a consensus phosphorylationsite25 for Src around tyrosine residue 1267 (Fig. 4a). Thus, apoint mutation of Y1267 to phenylalanine was constructed.

articles

a bNR1-1a/NR2A,HEK293

I (p

A)

V (mV)

I/I m

ax

[Zn2+] (nM)

c

Fig. 3. Potentiation of NMDA-receptor currents by Src corre-lates with the amount of tonic inhibition of NR1/NR2A recep-tors by zinc. (a) Dose–response curves for the high-affinity zincinhibition of heteromeric NR1/NR2A receptors in oocytes(NR1-1a, n = 14; NR1-1b, n = 12; NR1-1a(C744A,C798A), n =8; NR1-1bm207-211, n = 19). As previously reported21, a splicevariant with exon 5 (such as NR1-1b) exhibits an IC50 nearlytenfold higher than a splice variant without exon 5 (such asNR1-1a). Mutation of two cysteine residues of NR1-1a(C744A,C798A)21 also shifts the IC50 for zinc by approxi-mately tenfold. A triple mutation in exon 5 (NR1-1bm207-211:K207G, R208G, K211G)21 nullifies the effects of exon 5 on zincinhibition. L 1bm209–211, g 1b, p 1a, H 1aC744,798A (b) The ratio of the current in the presence of 10 µM EDTA tothat in the absence of EDTA is a measurement of tonic inhibi-tion of NMDA-receptor currents by trace amounts of zinc inHEK 293 cells. The amount of tonic inhibition by zinc is consis-tent with the IC50 values obtained in oocytes. (c) Effects of Srcon current amplitudes of NR1/NR2A in HEK293 cells. Notethat NR1-1a and NR1-1bm207-211 are potentiated by Src,whereas NR1-1b and NR1-1aC744,798A are not. p 1a/2A(10), P 1aC744,798A/2A (5), l 1b/2A (9), L 1bm207–211/2A (6). (d) Correlation between the tonic inhibition by zinc and potentiation by Srcin HEK cells measured six minutes after obtaining the whole-cell configuration (correlation factor, R = 0.997, p<0.005).

a b

c d

I/I m

ax

I ED

TA/I

NO

ED

TA

[Zn2+] (nM)

% o

f in

itia

l cu

rren

t

Time (min)

Src

po

ten

tiat

ion

(%)

IEDTA/INO EDTA

[Zn2+] (nM)


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


Additional tyrosine residues with high surface probability in theneighboring region were also mutated. Point mutation Y1267Fblocked the potentiation by Src (Fig. 4b and c). In addition,mutants Y1105F and Y1387F also abolished Src potentiation(Fig. 4b and c). In contrast, mutations of several other tyrosineresidues in the same region had no detectable effects on Src-induced potentiation (Fig. 4b). Consistent with our hypothesis,Src failed to change the zinc sensitivity of receptors with any oneof the three tyrosine mutations that eliminated Src potentiationof NMDA receptor currents (Fig. 4d). The zinc-inhibition curvesfor receptors with two of these mutations (Y1105F and Y1387F)were identical to the zinc-inhibition curve for the wild-typereceptor. In these cases, the only effect of the tyrosine mutationwas to prevent the Src-induced reduction in zinc sensitivity. Thethird mutation (Y1267F) reduced zinc sensitivity of the recep-tor in the absence of Src, which precluded further reduction ofzinc sensitivity by Src. The IC50 for zinc of NR1-1a/NR2A(Y1267F) in oocytes was 92 nM (n = 13, data notshown), which is approximately fivefold higher than the IC50for the wild-type NR1/NR2A. The amount of maximal inhibi-tion was not altered (data not shown). A similar shift of IC50 forzinc was also observed in HEK cells (Fig. 4d). In contrast, theEC50 values for glutamate (n = 7) and glycine (n = 7) and theIC50 value for proton inhibition (n = 8) for NR1-1a/NR2A(Y1267F) were not significantly different from thoseof the wild-type receptors (data not shown). These data suggestthat the Y1267F mutation results in a receptor that behaves likeone that is fully potentiated by Src. We have also constructeddouble tyrosine mutations NR2A(Y1105F,Y1387F) andNR2A(Y1267F,1387F). There was no detectable differencebetween NR2A(Y1105F,1387F) and the corresponding singletyrosine mutants (Y1105F and Y3187F) (Table 1).NR2A(Y1267F,Y1387F) behaved like Y1267F, with a reducedapparent zinc affinity and lack of modulation by Src (Table 1).

EDTA AND SRC INCREASE THE CURRENT DECAY TIME CONSTANT

Published evidence conflicts about the mechanism by whichSrc potentiates NMDA receptors. It has been reported that Srcincreases the open probability, open time, burst length, clus-ter length and supercluster length of NMDA channels6.Although the relaxation time constant of macroscopic currentmay be influenced by the first latency of NMDA channels26, anincrease in the duration of individual channel activation by Srcshould increase the relaxation time constant. However, nochange of the relaxation time constant is reported when themacroscopic current amplitude is potentiated5. In our whole-cell recording, the decay of NR1/NR2A current could not befitted with a single exponential component. To circumvent thepossibility that the slower, second component reflects slowremoval of the agonists due to turbulence around the wholecell, we investigated the decay time constant on NMDA-recep-

articles

Table 1. Inhibition of wild-type versus mutant NR1/NR2Areceptors by tricine-buffered zinc (223 nM) in HEK293cells

Internal Solution No Src Src

Wild-Type NR2A 0.48 ± 0.04 (5) 0.69 ± 0.05 (6)*

Y1105F 0.54 ± 0.05 (7) 0.50 ± 0.03 (5)Y1267F 0.74 ± 0.04 (4)* 0.75 ± 0.05 (4)*

Y1387F 0.49 ± 0.04 (5) 0.49 ± 0.05 (6)Y1105F,1387F 0.44 ± 0.09 (4) 0.45 ± 0.07 (5)Y1267F,1387F 0.79 ± 0.03 (3)* 0.82 ± 0.04 (4)*

Data presented in the table are I/Imax. Imax is the current recorded undernominally zinc-free conditions (in the presence of 10 mM tricine withoutadded zinc). Number of observations is indicated in parentheses. *significantdifference (p<0.05) compared to wild type by ANOVA and Newman-Keulspost-hoc test.

Fig. 4. Point mutations of tyrosineresidues in NR2A C-terminal block Src’spotentiation of NR1/NR2A currents andrelief of zinc inhibition. (a) A diagramshowing mutations of C-terminal tyrosineresidues of NR2A subunit. Note thatY1267 is part of a putative consensusphosphorylation site for Src. Other tyro-sine residues with high surface probabil-ity37 were also mutated as indicated. (b) Summary bar graph showing theeffects of nine point mutations of tyrosineresidues to phenylalanine on the Src’spotentiation of whole-cell NR1/NR2A-receptor currents measured six minutesafter obtaining the whole-cell configura-tion (*p<0.01 when compared to the wild-type NR2A by ANOVA andNewman-Keuls post hoc test). For all pan-els, number of cells is shown in parenthe-ses. (c) Point mutation of Y1105F, Y1267For Y1387F blocks Src’s potentiation ofwhole-cell NR1/NR2A-receptor currents.p wt (10), L Y1105F (5), G Y1267F (8), H Y1387F (7). (d) Zinc sensitivity of het-eromeric receptors containing three mutant NR2A subunits. Each data point is the average from 4–7 cells. Solid lines are fitted zinc curvesfor wild-type NR1/NR2A receptor without Src; dashed lines are the curves with Src. Note that Y1267F has lower affinity for zinc and thatSrc failed to relieve the zinc inhibition further. Although Y1105F and Y1387F have normal zinc affinity, Src failed to relieve zinc inhibition. pcontrol, l Src.

daY1267F

Y1105F

Y1387F

Time (min) [Zn2+](nM)

% o

f in

itia

lcu

rren

t

b

% o

f in

itia

lcu

rren

tc

Consensus sequence for Src


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


tor currents in outside-out patches exposed to brief pulses (10ms) of glutamate (with submillisecond solution-exchange ratesat the open tip27). Under such conditions, two exponentialcomponents were still needed to fit the decay of the macro-scopic NMDA receptor currents. The two averaged time con-stants were 36.6 ± 5.3 and 79.8 ± 8.9 ms without EDTA andSrc (n = 9). Inclusion of EDTA increased the peak of macro-scopic NR1/NR2A currents recorded from outside-out patch-es by 63 ± 30% (n = 9, p<0.05, paired t-test), which isconsistent with our whole-cell data as well as a previous reportthat Src increases open probability6. In addition, EDTA signif-icantly increased the slower relaxation time constant to 130 ±19 ms, whereas the faster relaxation time constant remainedunchanged with a mean of 38.9 ± 6.0 (Fig. 5). Interestingly, Srcinduced a similar selective change in the slower relaxation timeconstant. The averaged time constants in the presence of Srcwere 42.0 ± 4.1 and 117.7 ± 9.5 ms (n = 7). Because theamount of calcium influx during an EPSC is determined bythe charge integral (Q = ΣiAiτ i) of the NMDA current, theresulting increase of synaptic calcium influx could be ampli-fied by as much as 2.3-fold by Src. These observations indicatethat Src and EDTA change the NMDA-receptor channel kinet-ics in similar ways, consistent with our hypothesis that Srcpotentiates NMDA receptors by relieving tonic zinc inhibition.

SRC REDUCES ZINC SENSITIVITY OF NR1/NR2B RECEPTORS

Is the reduction of zinc sensitivity by Src unique for receptors con-taining NR2A subunits, or is it a more generalized regulatory

mechanism for other NMDA-receptor subunits? To address thisquestion, we examined the effects of Src on zinc sensitivity ofrecombinant NR1-1a/NR2B receptors. Zinc inhibits NR1/NR2Breceptors with an IC50 value of 4.6 µM under control conditionsin HEK293 cells (Fig. 6a and b). In the presence of c-Src (30 unitsper ml), the IC50 was shifted to 17.6 µM. Thus, NMDA receptorscomprised of NR2A or NR2B subunits are modulated by tyrosinekinase Src in a similar way, through a reduction of zinc sensitivity.

DiscussionTaken together, these data suggest that Src potentiates recombi-nant NR1/NR2A-receptor function by reducing tonic inhibitionof these receptors by extracellular zinc. Furthermore, Src alsoreduces zinc sensitivity of NR1/NR2B receptors. Thus, reduction ofzinc inhibition by Src could be a widely utilized mechanism forthe modulation of NMDA receptor function in the brain. Our dataare consistent with biochemical observations that both NR2A andNR2B are phosphorylated by the Src family of tyrosine kinases8,28

because they show that receptors comprised of either the NR2Aor NR2B subunit are modulated functionally by Src. The selectivepotentiation of NR1/NR2A receptors by Src in HEK293 cells isdue to the high zinc sensitivity of NR1/NR2A receptors and theambient zinc contamination in the recording solutions that caus-es them to be tonically inhibited. By contrast, receptors comprisedof NR1/NR2B are not potentiated by Src in HEK cells because theirrelatively lower sensitivity to zinc prevents tonic inhibition.

Zinc is actively accumulated in certain nerve terminals in aregion-specific manner and can be released into synaptic clefts at

articles

Fig. 6. Src reduces zinc sensitivityof NR1/NR2B receptors. (a) Theeffects of zinc (3 and 10 µM) onNR1/NR2B-receptor currentsevoked by agonists (100 µM gluta-mate, 30 µM glycine) from repre-sentative HEK 293 cells with orwithout inclusion of Src in thepatch pipette solution (Vh, -50 mV). (b) Src shifts the IC50 for the high-affinity, voltage-independent zinc site of NR1/NR2B receptors. Each point on the curve is averaged datafrom 3–5 cells. The IC50 is 4.6 without Src and 17.6 µM with Src. P control, L Src.

Fig. 5. EDTA and Src increase the time constant of theslower component of NMDA currents. (a) Typical cur-rent responses from an outside-out patch expressingNR1/NR2A receptors. A brief glutamate-concentrationjump (10 ms, to 100 µM) was applied to the patches.Saturating glycine (10–30 µM) was present in all solu-tions. Traces shown are averaged from 25 consecutiveresponses for control and 20 responses for EDTA. Thetop traces in (a) and (b) show the glutamate applicationtime. (b) The decay of NMDA-receptor-mediated cur-rents was fitted with two exponential components(smooth lines). The raw current trace is scaled to thesame size to illustrate the change of relaxation time con-stant. (c) EDTA and Src increased the time constant ofthe slower exponential component significantly (p<0.05,paired t-test for EDTA, unpaired t-test for Src), whereasthe time constant for the faster exponential componentis not changed. p control, l EDTA, g Src.

a Control Src b

I/I m

ax

[Zn2+] (µM)

a b

c

Control EDTA

τ1 (

ms)

τ2 (

ms)

Control EDTA Src Control EDTA Src

[Zn2+](µM) [Zn2+](µM)


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


concentrations up to 0.1 mM14. The reported ambient zinc con-centration in the brain varies from 150 nM to more than 2 µM22.Assuming ambient zinc levels of 1 µM, more than 70% ofNR1/NR2A receptors would be inhibited, compared to 25% ofNR1/NR2B receptors. The native NMDA receptors in the brainmost likely include both NR2A and NR2B subunits. If, as seemslikely, ambient zinc concentrations are sufficient (in excess of 1µM22) to tonically inhibit NMDA receptors containing NR2A, thensuch receptors might serve as as marginally functional receptorswhose full function can be restored through activation of tyrosinekinases such as Src29. If ambient zinc concentrations are insuffi-cient to cause significant tonic inhibition, Src could still alter theresponsiveness of NMDA receptors to synaptically released zinc.

The C-terminal region of NR2A is critical for modulation bySrc. A C-terminal deletion mutant of NR2A is not potentiatedby Src in HEK293 cells. Mutant mice expressing NR2A with theC-terminal truncation have impaired synaptic plasticity. We haveidentified three tyrosine residues in the C-terminal region ofNR2A that are critical for the modulation of NR1/NR2A recep-tors by Src. The Y1105 and Y1387 residues function as predict-ed for potential tyrosine phosphorylation sites, which is that theirmutation to phenylalanine causes no apparent changes by itselfbut blocks the reduction of zinc affinity by Src. Because boththese tyrosine residues are conserved between NR2A and NR2B,it is tempting to speculate that phosphorylation of the corre-sponding tyrosine residues in NR2B is responsible for reducingthe apparent zinc affinity of NR1/NR2B receptors. The functionof the third tyrosine residue, Y1267, is peculiar. Mutation ofY1267 to phenylalanine results in a receptor with reduced zincaffinity. This tyrosine residue and the small stretch of peptidesurrounding it are unique to NR2A. We do not know the mech-anism behind this change and can only speculate at this point.One possibility is that this residue is phosphorylated constitu-tively and may be involved in interactions with the SH2 domainof Src or other proteins. The C-terminal domains of NMDAreceptor subunits are larger in size than those of other receptorsubunits and may be involved in complex interactions with otherproteins in the postsynaptic density and cytoskeleton. One exam-ple of such an interaction is the reduction of NMDA-receptoractivity by direct binding of calmodulin to NR1 subunits, whichinactivates NMDA receptors30. Little is known about the possibleprotein–protein interactions of the C-terminal regions of NR2subunits other than their interaction with the PSD95 family ofproteins. Although the mechanism by which mutation of Y1267reduces zinc sensitivity of NR1/NR2A receptors needs to beexplored, our data do show that its lack of tonic inhibition dueto reduced zinc sensitivity occludes Src potentiation, supportingour hypothesis that Src potentiates NMDA-receptor function byreducing tonic zinc inhibition.

From a structural perspective, our results suggest that tyro-sine phosphorylation and/or Src binding on the cytosolic C-ter-minal domain of NR2A alter the apparent affinity of anextracellular zinc binding site. Although structural informationwill ultimately be required to understand the interaction betweenSrc and the zinc site, several possible explanations deserve men-tion. Conformational constraints imposed by Src binding and/ortyrosine phosphorylation (the two events may occur separately)may specifically reduce the affinity of zinc by altering the geom-etry of the coordination site. Alternatively, Src may change otheraspects of the NMDA-channel properties that perturb the bind-ing of zinc. Several modulators or mutations21 that modify zincsensitivity have previously been described. For example, expres-sion of NR1 alternative exon 5, which encodes a highly charged

surface loop, seems to reduce zinc sensitivity in a manner thatcan be mimicked by extracellular polyamines. In addition, muta-tions of acidic residues in the NR1 subunit that reduce protonsensitivity similarly reduce zinc sensitivity. One possible expla-nation for these (and by analogy Src’s) effects is that they may bedue to reduced electronegativity of the electron-donating residuesthat form the zinc coordination site, which would reduce the abil-ity of these residues to coordinate zinc. Such an effect could occurby inductive electron withdrawal through portions of thepolypeptide chain secondary to Src binding or phosphorylation.Alternatively, Src could change the channel properties in such away that occupation of the zinc site is less likely to force closure ofan open channel, requiring a greater probability of zinc occu-pancy to inhibit the channel, which might appear as a lower IC50.In either case, the effects of Src and tyrosine phosphorylationneed to be transmitted from intracellularly accessible residues toextracellular residues. Clearly additional functional as well asstructural information will be required to ultimately understandSrc’s effects on zinc modulation.

Finally, by demonstrating that potentiation of the NMDA-receptor current by Src reduces the apparent zinc affinity ofNMDA receptors, we link two modulatory sites of NMDAreceptors that have been previously thought to be indepen-dent. Our data provide further support for the convergenceof different NMDA receptor modulatory systems that hasbecome evident recently13,21,24,31.

MethodsSITE-DIRECTED MUTAGENESIS. All of the cDNAs used in this study except amutant of NR1-1b and NR2B (pCDNAI/amp, Invitrogen) were sub-cloned into pCIneo vector (Promega). Site-directed mutagenesis wasdone with Pfu DNA polymerase (Stratagene) to linearly replicate theparental strand with desired mismatch incorporated into the primer.Methylated parental DNA template was then degraded with Dpn I. Thenicked double-stranded mutant DNA was transformed into E. coli. Thenicks in the plasmid were repaired after transformation. Colonies wereselected by screening for a silent mutation that introduces a new restric-tion site. The mutations were verified by didioxy sequencing throughboth strands in the region of the mutation.

TRANSFECTION OF HEK CELLS. HEK293 cells (CRL 1573; American TypeCulture Collection, Rockville, Maryland) were maintained at 37°C and5% CO2 in DMEM supplemented with L-glutamine (200 µM), sodiumpyruvate (100 µM), penicillin/streptomycin (100 units per ml), and 10%fetal bovine serum. Low-confluence cells were transfected by calcium-phosphate precipitation32. Cells were cotransfected with a mixture con-taining NR1, NR2 and green fluorescent protein33 plasmids (1, 2, and0.3 µg per 12 mm diameter coverslip, respectively). After transfection,200 µM D-AP5 was added to the culture medium.

EXPRESSION OF NMDA RECEPTORS IN XENOPUS OOCYTES. cRNA was synthe-sized from linearized template cDNA according to manufacturer’s spec-ification (Ambion). The quality of synthesized cRNA was assessed by gelelectrophoresis and quantified by spectroscopy and gel electrophoresis.Preparation of oocytes and injection of cDNAs coding for wild-type andmutant NMDA receptors were performed as described21.

BUFFERED ZINC SOLUTIONS. The tricine-buffered zinc solutions used toobtain the zinc dose–response curves were prepared according to theempirically established binding constant18 10–5 M, by adding into 10 mMtricine (pKa, 8.15) the following concentrations of zinc (in µM): 0.26,0.78, 2.6, 7.8, 26, 77.5, and 254. The corresponding estimated concen-trations of free zinc for HEK recording at pH 7.4 were calculated withWINMAXC34 and BAD35, and were (in nM) 2.3, 6.88, 22.3, 68.8, 223,688, 2230, respectively.

VOLTAGE-CLAMP RECORDINGS FROM XENOPUS OOCYTES. Two-electrode volt-age-clamp recordings were made as described21. Briefly, oocytes wereperfused with a solution comprised of (in mM) 90 NaCl, 1 KCl, 10

articles


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


HEPES and 0.5 BaCl2 at pH7.4, and held under voltage clamp at -30 to -40 mV. Electrodes were filled with 300 mM KCl and had resistance of1–10 MΩ. Solution exchange was computer controlled through an eight-valve manifold.

WHOLE-CELL AND OUTSIDE-OUT RECORDINGS FROM HEK293 CELLS. Patch-clamp recording in the whole-cell configuration and outside-out patchrecording were done as described36 with Axopatch 200A or 200B ampli-fier. Recording electrodes (5–12 MΩ) were filled with (in mM) 140 Cs-gluconate, 5 HEPES, 4 NaCl, 2 MgCl2, 0.5 CaCl2, 1 ATP, 0.3 GTP and 5BAPTA (pH 7.4). The recording chamber was continually perfused withrecording solution composed of (in mM) 150 NaCl, 10 HEPES, 1.0 CaCl2,3 KCl and 20 mM mannitol. Glutamate (20–100 µM) and glycine (10–30µM) were applied using a multibarrel pipette driven by a piezo-electricbimorph (exchange time under 0.5 ms) as described27. In some experi-ments, recombinant human c-Src (Upstate Biotech.) was added intointernal solution to a final concentration of 5–30 units per ml.

DATA ANALYSIS AND STATISTICS. All pooled data were expressed as mean ±standard error. Statistical comparisons were done with unpaired Stu-dent’s t-test unless otherwise stated. To obtain IC50 values, pooled zincinhibition data were fitted with least-squares criterion (SigmaPlot) toequation I/Imax = (1-a)/(1+([Zn2+]/IC50)n)+a where n is the Hill slopeand a is the residual response. (n and a were not constrained.) If the fit-ting algorithm returned a value less than 0.05 for a (our estimated limitfor detection), we fixed a to 0 and refit the zinc inhibition data to equa-tion I/Imax = 1/(1+([Zn2+]/IC50)n) (as in Fig. 6).

AcknowledgementsWe thank S.F.Heinemann for NR1 and NR2B cDNAs, N. Nakanishi for NR2A

cDNA and M. Chalfie, M. Mayer and P. Seeburg for green fluorescent protein

plasmid. We also thank P. Ascher, J. Neyton and I. Mody for critically reading

the manuscript.

RECEIVED 11 MARCH: ACCEPTED 22 MAY 1998

1. O’Dell, T.J., Kandel, E.R. & Grant, S.G.N. Long-term potentiation in thehippocampus is blocked by tyrosine kinase inhibitors. Nature 353, 558–560(1991).

2. Grant, S.G.N. et al. Impaired long-term potentiation, spatial learning, andhippocampal development in fyn mutant mice. Science 258, 1903–1910 (1992).

3. Lu, Y.M., Roder, J.C., Davidow, J. & Salter, M.W. Src activation in the inductionof long-term potentiation in CA1 hippocampal neurons. Science 279,1363–1368 (1998).

4. Wang, Y.T. & Salter, M.W. Regulation of NMDA receptors by tyrosine kinasesand phosphatases. Nature 369, 233–235 (1994).

5. Köhr, G. & Seeburg, P.H. Subtype-specific regulation of recombinant NMDAreceptor-channels by protein tyrosine kinases of the Src family. J. Physiol.(Lond.) 492, 445–452 (1996).

6. Yu, X.M., Askalan, R., Keil, G.J. & Salter, M.W. NMDA channel regulation bychannel-associated protein tyrosine kinase Src. Science 275, 674–678 (1997).

7. Sprengel, R. et al. Importance of the intracellular domain of NR2 subunits forNMDA receptor function in vivo. Cell 92, 279–289 (1998).

8. Miyakawa, T. et al. Fyn-kinase as a determinant of ethanol sensitivity:Relation to NMDA-receptor function. Science 278, 698–701 (1997).

9. Nowak, L., Bregestovski, P., Ascher, P., Herbert, A. & Prochiantz, A.Magnesium gates glutamate-activated channels in mouse central neurones.Nature 307, 462–465 (1984).

10. Mayer, M.L., Westbrook, G.L. & Guthrie, P.B. Voltage-dependent block by Mg2+

of NMDA responses in spinal cord neurones. Nature 309, 261–263 (1984).11. Westbrook, G.L. & Mayer, M.L. Micromolar concentrations of Zn2+

antagonize NMDA and GABA responses of hippocampal neurons. Nature328, 640–643 (1987).

12. Peters, S., Koh, J. & Choi, D.W. Zinc selectively blocks the action of N-methyl-D-aspartate on cortical neurons. Science 236, 589–593 (1987).

13. Traynelis, S.T., Hartley, M. & Heinemann, S.F. Control of proton sensitivityof the NMDA receptor by RNA splicing and polyamines. Science 268,873–876 (1995).

14. Smart, T.G., Xie, X. & Krishek, B.J. Modulation of inhibitory and excitatoryamino acid receptor ion channels by zinc. Prog. Neurobiol. 42, 393–441(1994).

15. Mayer, M.L., Vyklicky, L. Jr & Westbrook, G.L. Modulation of excitatoryamino acid receptors by group IIB metal cations in cultured mousehippocampal neurones. J. Physiol. (Lond.) 415, 329–350 (1989).

16. Christine, C.W. & Choi, D.W. Effect of zinc on NMDA receptor-mediatedchannel currents in cortical neurons. J. Neurosci. 10, 108–116 (1990).

17. Legendre, P. & Westbrook, G.L. The inhibition of single N-methyl-D-aspartate-activated channels by zinc ions on cultured rat neurones. J. Physiol.(Lond.) 429, 429–449 (1990).

18. Paoletti, P., Ascher, P. & Neyton, J. High-affinity zinc inhibition of NMDANR1-NR2A receptors. J. Neurosci. 17, 5711–5725 (1997).

19. Williams, K. Separating dual effects of zinc at recombinant N-methyl-D-aspartate receptors. Neurosci. Lett. 215, 9–12 (1996).

20. Chen, N., Moshaver, A. & Raymond, L.A. Differential sensitivity ofrecombinant N-methyl-D-aspartate receptor subtypes to zinc inhibition.Mol. Pharmacol. 51, 1015–1023 (1997).

21. Traynelis, S.F., Burgess, M.F., Zheng, F., Lyuboslavsky, P. & Powers, J.L.Control of voltage-independent zinc inhibition of NMDA receptors by theNR1 subunit. J. Neurosci. (in press).

22. Bogden, J.D., Troiano, R.A. & Joselow, M.M. Copper, zinc, magnesium, andcalcium in plasma and cerebrospinal fluid of patients with neurologicaldiseases. Clin. Chem. 23, 485–489 (1977).

23. Li, C., Peoples, R.W. & Weight, F.F. Proton potentiation of ATP-gated ionchannel responses to ATP and Zn2+ in rat nodose ganglion neurons. J.Neurophysiol. 76, 3048–3058 (1996).

24. Sullivan, J.M. et al. Identification of two cysteine residues that are requiredfor redox modulation of the NMDA subtype of glutamate receptor. Neuron13, 929–936 (1994).

25. Kemp, B.E. & Pearson, R.B. Protein kinase recognition sequence motifs.Trends Biochem. Sci. 15, 342–346 (1990).

26. Edmonds, B. & Colquhoun, D. Rapid decay of averaged single-channelNMDA receptor activations recorded at low agonist concentration. Proc. R.Soc. Lond. B 250, 279–286 (1992).

27. Traynelis, S.T. & Wahl, P. Control of rat GluR6 glutamate receptor openprobability by protein kinase A and calcineurin. J. Physiol. 503, 513–531(1997).

28. Lau, L.-F. & Huganir, R.L. Differential tyrosine phosphorylation of N-methyl-D-aspartate receptor subunits. J. Biol. Chem. 270, 20036–20041(1995).

29. Dikic, I., Tokiwa, G., Lev, S., Courtneidge, S.A. & Schlessinger, J. A role forPyk2 and Src in linking G-protein-coupled receptors with MAP kinaseactivation. Nature 383, 547–550 (1996).

30. Ehlers, M.D., Zhang, S., Bernhardt, J.P. & Huganir, R.L. Inactivation ofNMDA receptors by direct interaction of calmodulin with the NR1 subunit.Cell 84,745–755 (1996).

31. Chen, L. & Huang, L.-Y.M. Protein kinase C reduces Mg2+ block of NMDA-receptor channels as a mechanism of modulation. Nature 356, 521–523(1992).

32. Chen, C. & Okayama, H. High-efficiency transformation of mammalian cellsby plasmid DNA. Mol. Cell. Biol. 7, 2745–2752 (1987).

33. Chalfie, M., Tu, Y., Euskirchen, G., Ward, W.W. & Prasher, D.C. Greenfluorescent protein as a marker for gene expression. Science 263, 802–805(1994).

34. Bers, D., Patton, C. & Nuccitelli, R. A practical guide to the preparation of Cabuffers. Methods Cell Biol. 40, 3–29 (1994).

35. Brooks, S.P.J. & Storey, K.B. Bound and determined: a computer program formaking buffers of defined ion concentrations. Anal. Biochem. 201, 119–126(1992).

36. Hamill, O., Marty, A., Neher, E., Sakmann, B. & Sigworth, F. Improved patch-clamp techniques for high-resolution current recording from cells and cell-free membrane patches. Pflügers Arch. 391, 85–100 (1981).

37. Emini, E.A., Hughes, J.V., Perlow, D.S. & Boger, J. Induction of hepatitis Avirus-neutralizing antibody by a virus-specific synthetic peptide. J. Virol. 55,836–839 (1985).

articles


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Functional studies on secretory cells have supplied ample evi-dence that synaptic and other secretory vesicles can exist indistinct functional states1–4. Typically, only a fraction of allvesicles of a secretory cell can be released in a certain timewindow, independent of the strength of a given stimulus.These vesicles in a specific release-ready state are termed a‘pool’. The question arises whether pools defined in this orother ways can be associated with well defined macromolecu-lar complexes of synaptic proteins, which have recently beencharacterized biochemically5–7. Some of the proteins of thesecomplexes, such as SNAP-25, synaptobrevin and syntaxin, arethe targets of very specific proteolytic neurotoxins, whichcleave the proteins at unique sites8–10. These proteins are high-ly susceptible to toxin attack in monomeric form, whereas theyare protected to various degrees in macromolecular complex-es11. This property and differences between the toxins’ actionsmay allow us to link certain functional pools to some of thebiochemically defined complexes by their toxin susceptibili-ty. Indeed, different kinetic components are impaired differ-entially by toxins12,13. Also, botulinum neurotoxin type A maybe special among the clostridial neurotoxins, as its action insome preparations can be overcome by strong stimuli14,15.

For a precise dissociation of functional steps and their cor-relation with molecular data, it would be desirable to studyexocytosis at fast resolution, as the final steps in the neurose-cretory pathway are known to be fast, and most likely neuro-toxins act at later steps. We therefore studied catecholaminesecretion from bovine chromaffin cells using capacitance mea-surement as a fast assay of exocytosis combined with flashphotolysis of caged calcium as a fast and strong stimulus. We

characterized two kinetic components that were differentiallyaffected by clostridial neurotoxins. We compared the capaci-tance signal with simultaneously measured catecholaminerelease, and thereby identified a third very prominent kinet-ic component, which apparently was not related to cate-cholamine release, and which was not affected by toxins.Together, these data suggest possible relationships betweenlate steps in the release process and their biochemically definedmolecular counterparts.

ResultsMULTIPLE COMPONENTS OF EXOCYTOSIS IN CHROMAFFIN CELLS

To study the kinetic components of secretion, we employedfast-resolution capacitance measurements to estimate the secre-tory response after spatially homogeneous elevation of inter-nal calcium concentration ([Ca2+]i) by photorelease of cagedcalcium, nitrophenyl-EGTA. Membrane capacitance (Cm) isproportional to the surface area of the cell, and it increaseswhen secretory vesicles fuse with the plasma membrane16,17.In most of our measurements, we used the following protocol(Fig. 1a). After ten minutes of whole-cell dialysis with eithertoxin-free or toxin-containing pipette solutions, we first gavelow-intensity flashes (usually two) to generate small [Ca2+]ijumps to about 20 µM. This readily elicited secretory respons-es (Fig. 1b). When there was a clear indication that secretionwas depressed, we then gave strong flashes to elevate [Ca2+]ito higher values (over 100 µM). Intervals between the flasheswere 120 seconds. In response to the first flash, there was alwaysa robust Cm increase with two clearly distinct phases, whichwe call the exocytic burst18 and the slow component. Our

Multiple kinetic components ofexocytosis distinguished by neurotoxin sensitivity

Tao Xu1, Thomas Binz2, Heiner Niemann2 and Erwin Neher1

1 Department of Membrane Biophysics, Max-Planck-Institute for Biophysical Chemistry, Am Fassberg 11, D-37077, Göttingen, Germany2 Department of Biochemistry, Medizinische Hochschule, D-30625, Hannover, Germany

Correspondence should be addressed to T.X. ([email protected])

The secretion of synaptic and other vesicles is a complex process involving multiple steps. Many mol-ecular components of the secretory apparatus have been identified, but how they relate to thedifferent stages of vesicle release is not clear. We examined this issue in adrenal chromaffin cells,where capacitance measurements and amperometry allow us to measure vesicle fusion andhormone release simultaneously. Using flash photolysis of caged intracellular calcium to induce exo-cytosis, we observed three distinct kinetic components to vesicle fusion, of which only two arerelated to catecholamine release. Intracellular dialysis with botulinum neurotoxin E, D or C1 ortetanus-toxin light chains abolishes the catecholamine-related components, but leaves the thirdcomponent untouched. Botulinum neurotoxin A, which removes nine amino acids from thecarboxy(C)-terminal end of SNAP-25, does not eliminate catecholamine release completely, butslows down both catecholamine-related components. Thus we assign a dual role to SNAP-25 andsuggest that its nine C-terminal amino acids are directly involved in coupling the calcium sensor tothe final step in exocytosis.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

nentials. One had an amplitude of 404 ± 76 fF and a time con-stant of 1.1 ± 0.2 s. The second exponential had a time con-stant of 6.4 ± 0.9 s and 361 ± 62 fF in amplitude (n = 12). Thislarge membrane increase, which constituted roughly 10% ofthe total cell membrane, was not associated with cate-cholamine release (see Figs. 1b and 6). Also, the time constantdid not change with [Ca2+]i (Fig. 2b). It rather seemed thatCm increases of this kind were elicited whenever [Ca2+]iexceeded a certain threshold.

ATP DEPENDENCE OF CALCIUM-TRIGGERED EXOCYTOSIS

To test the ATP dependence of different exocytic components,we dialyzed the cytosol of chromaffin cells with 80% calcium-loaded NP-EGTA internal solution containing nonhydrolyzable ATP analogs, either adenosine 5’-[β,γ-meth-ylene]triphosphate (AMP-PCP) or β,γ-Imidoadenosine 5’-triphosphate (AMP-PNP), via the patch pipette. While thebasal [Ca2+]i was kept at 100–300 nM, too low to support exo-cytosis, ATP was washed out of the cell and replaced by theanalogs during the first five minutes of perfusion. Subse-quently no responses could be elicited by flash photolysis, nei-ther when secretion was measured by Cm nor by amperometry(Fig. 3a; n = 28 cells in paired experiments). These data com-plement the finding that the release capability of permeabi-lized cells can be increased (‘primed’) by preincubation withATP and diminished (‘deprimed’) by preincubation in theabsence of ATP1. Our results differ from a report that MgATP-independent exocytosis remained undiminished for even sixminutes2, although that experiment was done at a lower tem-perature than ours. The authors suggested that this form of‘depriming’ may require magnesium, which was absent in theirexperiments. We therefore performed additional experiments

Fig. 1. Multiple secretorycomponents in response todifferent [Ca2+]i levels. (a) Overview of the experi-ment. Sample responsesfrom one cell are illustrated.The individual traces areintracellular free calciumconcentration ([Ca2+]i) asmeasured by furaptra, mem-brane capacitance (Cm),series conductance of theequivalent circuit (Gs),membrane conductance(Gm) and membrane current(Imon). Arrows indicatetime points, where flashesare given. They are followedby gaps in the record, duringwhich data samplingswitched to a high-resolu-tion mode. During ten min-utes of whole-cell dialysis,no loading transient wasseen in NP-EGTA-containinginternal solution. The Cmtraces remained flat until theonset of flashes (arrows).The flashes were given every 120 s. (b) Fast-resolution recordings of Cm, amperometric current and [Ca2+]i in response to individualflashes in (a) are displayed. The flashes were triggered at 200 ms in each trace.

working hypothesis is that the exocytic burst represents thosegranules that are in a release-ready state and require only ele-vation of [Ca2+]i for exocytosis. Subsequently, granules haveto undergo slower steps of recruitment to the release-readypool before they can exocytose. Therefore, further capacitanceincrease occurs at a much slower rate (see Discussion for detailsand a model). In 84 experiments, exocytic bursts were evokedat [Ca2+]i averaging 29.3 ± 18.9 µM (mean ± SD). The timecourse could be fitted by double exponentials with average timeconstants of 62.6 ± 5.9 ms (n = 44 of 84) and 295.6 ± 23.1 ms(n = 76 of 84), and average amplitudes of 368 ± 40 fF and 302± 27 fF, respectively. This is similar to previously reported val-ues19 except for somewhat slower kinetics. The slow compo-nent showed a time constant in the range of ten seconds, butthis is probably an underestimate, as capacitance at late timesmay be reduced by endocytosis and may be curtailed by a slowdecline in [Ca2+]i. The amperometric current, which remainedon a plateau level for up to ten seconds in some cases, also sug-gests that the time constant of this slow component may beover ten seconds. The increase in Cm always was accompaniedby two phases of catecholamine release as monitored by car-bon fibers (Fig. 2a). It should be noted that the faster compo-nent of the exocytic burst was also accompanied bycatecholamine release (Fig. 2a, inset).

Following flashes, [Ca2+]i dropped back to basal values attimes longer than ten seconds. Second flashes in a given exper-iment usually elicited very small Cm increases and only rarelywere accompanied by amperometric spikes, which might sug-gest the depletion of a vesicle pool during first flashes. How-ever, when [Ca2+]i was raised above 100 µM in third flashes,again an intense Cm increase was observed (Fig. 2b). This low-calcium-affinity Cm component increased with two expo-

Cm

(pF

)

10008006004002000Time (s)

400

200

0

Gs(

nS

)

10050

0

[Ca2+

] i(µM

)

1.0

0.5

0.0

Gm

(nS

)-40

-20

0

Imo

n(p

A)

A

bTime (s)

a

Imo

n(p

A)

Gm

(nS)

Gs(

nS)

Cm

(pF)

[Ca2

+ ]i(

µM)

[Ca2

+ ]i(

µM)

I AM

P(p

A)

Cm

(pF)

Time (s) Time (s) Time (s) Time (s)


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Complexes of synaptobrevin, syntaxin and SNAP-25(termed SNARE complexes) are found in synaptic vesiclemembrane in vitro24,25, and such ternary complexes arereversibly disassembled by treating synaptic vesicles with ATP-NSF and SNAPs25. It is therefore possible that, in living cells,most of the SNARE complexes are continuously formed onthe vesicle and continuously disassembled because of the activ-ity of ATP-NSF, which might be the last step of ATP hydroly-

Fig. 3. ATP dependence ofsecretion. (a) The effect of 5–6-minute dialysis of AMP-PCP onsecretion. Averaged [Ca2+]i lev-els, capacitance traces andamperometric currents inresponse to the first flashesfrom paired experiments in thepresence of MgATP as control(solid lines; n = 11) and in thepresence of 2 mM AMP-PCPsubstituting for MgATP (dashedlines; n = 17) are displayed.Flashes were triggered at 200ms in the graph. (b) The effectof 5–6 minute dialysis of AMP-PNP on secretion. Similar to (a),but 2 mM AMP-PNP was used tosubstitute for MgATP (n = 12).Control responses from pairedexperiments are from 11 cells.(c) Similar to (b) but for thesuccessive second flashes. Cellsare the same as those in (b).

using an internal solution withoutmagnesium and DM-nitrophen tochelate magnesium. The free con-centration of magnesium in thissolution was estimated to be lessthan 10 nM. Again no secretion wasobserved in the presence of 2 mMAMP-PCP (data not shown). Thuswe conclude that magnesium doesnot matter for this effect.

We also used AMP-PNP toreplace ATP. The effect of AMP-PNPwas weaker than that of AMP-PCP(Fig. 3b). In some cells, AMP-PNPonly partially blocked exocytosis,whereas in other cells AMP-PNPshowed strong inhibition of exocy-tosis. However, AMP-PNP com-pletely blocked the secretion elicitedby a second flash, suggesting thatATP is necessary for the refilling ofvesicles (Fig. 3c).

Our results confirm the impor-tance of ATP in secretion. Althoughreplacement of ATP by the nonhy-drolyzable analogs AMP-PCP andAMP-PNP completely blocked boththe exocytic burst and the slow exocy-tosis within five minutes, it is unlikelythat ATP hydrolysis is involved in theexocytic burst for the following reasons. Experiments using thecaged calcium DM-Nitrophen have shown that exocytosis canreadily be induced earlier in a whole-cell recording in the com-plete absence of free magnesium and in a nominally ATP-freesolution2,4,18. This conclusion is in line with studies on perme-abilized cells20–21 and with recent investigations showing thatthe action of N-ethylmaleimide fusion protein (NSF) is requiredbefore contact of vesicle and target membrane22,23.

Fig. 2. Kinetic analysis of multipleCm components. (a) An exampleof secretory kinetics at low[Ca2+]i (27 µM). The Cm trace isreasonably fitted by two expo-nentials (superimposed thickdashed line) with time constantsof 90 ms and 6.29 s. The corre-sponding amperometric currentalso shows a double exponentialdecay. The onset of secretionfrom 0.2 to 1 s (boxed area) isexpanded in the inset. (b) Kineticanalysis of Cm response to high[Ca2+]i. One example of Cmincrease at [Ca2+]i of 287 µM is

shown in the left panel.The Cm trace is reason-ably fitted by two expo-nentials (superimposedthick dashed line) withtime constants of 0.82 sand 4.15 s. The right paneldisplays a summary of thefaster time constants ofthe Cm increases versusthe [Ca2+]i levels from 58experiments. The dashedline is the mean value.

a

b

Time (s)

I AM

P (

pA

)C

m (

pF)

Time (s) [Ca2+]i (µM)

Cm

(p

F)

τ fas

t(s

)

a b c

[Ca2

+ ]i (

µM)

[Ca2

+ ]i (

µM)

[Ca2

+ ]i (

µM)

Cm

Cm

Cm

I Am

p (

pA

)

I Am

p (

pA

)

I Am

p (

pA

)

Time (s) Time (s) Time (s)

ControlControl

Control

AMP-PCP (2 mM)

AMP-PNP (2 mM)

AMP-PNP (2 mM)


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Fig. 4. The effect of BoNT/E and BoNT/A onsecretion. (a) BoNT/E blocks both components ofsecretion at low [Ca2+]i. Averaged [Ca2+]i levels,capacitance traces and amperometric currents inresponse to the first flashes from paired experi-ments in control (solid lines; n = 18) and in thepresence of 400 nM BoNT/E-LC (dashed lines;n = 15) are displayed. Flashes were triggered after10 min of whole cell dialysis, corresponding to 200ms in the graph. The capacitance was normalizedto its pre-flash values. (b) BoNT/A partially blockssecretion at low [Ca2+]i. Averaged [Ca2+]i levels,capacitance traces and amperometric currents inresponse to the first flashes from paired experi-ments in control (solid lines; n = 24) and in thepresence of 800 nM BoNT/A-LC (dashed lines;n = 37) are displayed. The capacitance was normal-ized to its pre-flash values. Flashes were triggeredafter 10 min of whole cell dialysis, correspondingto 200 ms in the graph. The capacitance trace givesthe impression that the slow phase of secretion isblocked completely, whereas the amperometrytrace reports some continuing release two sec-onds following the flash. This discrepancy may bedue to slow endocytosis, which conceals a slow-release process in the capacitance trace at later

time. (c) BoNT/A slows down the exocytic burst. The uppersolid trace is the averaged Cm response from 55 cells withoutBoNT/A. It can be fitted by two exponentials (superimposeddashed line) with a fast time constant of 51.6 ms and a slow oneof 195.3 ms. The lower solid trace is the averaged Cm responsefrom 40 cells poisoned with 800 nM BoNT/A-LC. The onset ofcapacitance increase only shows a slow exponential componentwith a time constant of 207.8 ms. Superimposed dashed line is asingle exponential fit. The capacitance was normalized to its pre-flash values. (d) BoNT/A reduced maximal rate of secretion.Maximal rates of secretion ∆Cm/∆t were measured from thecapacitance response. ∆t was set to half the fastest time con-stant analyzed in the exponential fit of the cell.

sis. We have shown that replacement of ATP by AMP-PCPafter a period of five minutes completely blocked both com-ponents of secretion. The in vitro results mentioned aboveoffer an explanation for these findings: when ATP is replacedby a nonhydrolyzable analogue, the continuous disassemblyof SNARE complex stops. Therefore the SNAREs on a givenvesicle tend to form ternary complexes and are no longer avail-able for complexes linking vesicle and plasma membrane.However, our data do not exclude additional requirements forATP, such as in maintaining PtdIns-4,5P2 levels26–28. In fact,our finding that the exocytic burst is lost within five minutesin the absence of MgATP implies either that one of the laststeps before fusion is affected by the prolonged absence ofMgATP, or else that transmembrane SNARE complexes canbe disintegrated without the action of NSF.

CLOSTRIDIAL NEUROTOXINS AND CATECHOLAMINE SECRETION

Clostridial neurotoxins are potent inhibitors of synaptic-vesi-cle exocytosis in nerve terminals. It is now well establishedthat the light chains (LC) of clostridial neurotoxins act as zinc-dependent metallopreoteases, which specifically cleave SNAREproteins. In particular, tetanus neurotoxin (TeNT) and botu-

linum neurotoxin (BoNT) serotypes B, D, F and G specifical-ly cleave synaptobrevin at unique peptide bonds8–10. BoNTserotypes A and E cleave SNAP-25 at two different sites locat-ed close to the carboxyl terminals29,30, whereas the targets ofBoNT serotype C are syntaxin and SNAP-2531,32. To explorethe role of SNARE proteins in the secretory pathway, we askedwhether the kinetic components described above are differ-entially influenced by different clostridial neurotoxins.

If we assume there are 22,000 vesicles per chromaffin cell33,and each vesicle contains 100 copies of synaptobrevin, the finalconcentration of synaptobrevin is calculated to be 3 µM in asingle chromaffin cell. To test the time course of the cleavageof our recombinant light chains of clostridial neurotoxins, wefirst performed in vitro cleavage assays. The results show that,in our internal solution at 32°C, 330 nM TeNT-LC are capa-ble of cleaving more than 95% of 5 µM recombinant GST-synaptobrevin2 within three minutes, and 40 nM BoNT/E-LCcan cleave more than 95% of 5 µM recombinant SNAP-25His6(data not shown). For a 10 MΩ access-resistance pipette, aprotein with molecular weight of 50 kD, which is the size ofthe light chains of clostridial neurotoxins, can diffuse into thecytosol with a time constant of five minutes34. Thus within

a b

c d

Time (s) Time (s)

Time (s)

[Ca2

+ ]i (

µM)

Cm

I Am

p (

pA

)

[Ca2

+ ]i (

µM)

Cm

I Am

p (

pA

)

Cm

(∆C

m/∆

t)m

ax (

pF/

s)

ControlControl

BoNT/E

BoNT/A

BoNT/AControl


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Fig. 5. The effect ofBoNT/C1, BoNT/D andTeNT on secretion.BoNT/C1 (a), BoNT/D(b) and TeNT (c) blockboth components of secre-tion at low [Ca2+]i. The fig-ure displays the averaged[Ca2+]i levels, capacitancetraces and amperometriccurrents in response to thefirst flashes from pairedexperiments in the controlcondition (solid lines) andin the presence (dashedlines) of 2 µM BoNT/C1-LC (n = 7; control n = 6),400 nM BoNT/D (n = 11;control n = 12) and 4 µMTeNT (n = 14; controln = 17). Flashes were trig-gered after 10 min ofwhole-cell dialysis, corre-sponding to 200 ms in thegraph. Scaling bars, 200 fF.

five minutes, the cytosol concentration of the light chains ofneurotoxins would reach 36% of the pipette concentration.Therefore, to test the role of SNAP-25 in exocytic processes,we used 400 nM BoNT/E-LC in the patch pipette and waitedfor ten minutes before elevation of [Ca2+]i to ensure completeaction of the toxin. As the action of clostridial neurotoxins arealso steeply temperature dependent35, the experiments wereconducted at 32°C.

Under our experimental protocol, BoNT/E-LC not onlyblocked the slow phase of Cm increase, but also inhibited theexocytic burst (Fig. 4a; see also Table 1). Next, we wanted toassess the effects of BoNT/A, which cleaves the Q197–R198 pep-tide bond very close to the C-terminus of SNAP-25 and there-by removes only nine amino acids, in contrast to BoNT/E,which cleaves the R180–I181 peptide bond, 17 residuesupstream. In vitro cleavage studies showed that BoNT/A hassimilar kinetics as BoNT/E (data not shown). However, even at800 nM, which is twice the concentration used for BoNT/E,BoNT/A only partially inhibited secretion (Fig. 4b). BoNT/Areduced both the exocytic burst (Table 1) and the slow phaseof secretion. The capacitance trace (Fig. 4b) seems to indicatecomplete block of the slow phase. However this result cannotbe taken at face value, because endocytosis may compensatefor ongoing exocytosis. Indeed, the amperometric trace indi-cates a small slow exocytic component after toxin treatment.By analyzing the amperometric trace, we concluded that theexocytic burst (approximately the integral response over thefirst two seconds) is reduced by a factor of 1.7, whereas theslow component (two to ten seconds) is reduced by a factorof 2.7. This strong reduction of the slow component suggeststhat BoNT/A affects a relatively early step in the secretorypathway by reducing the supply rate of release-ready granules.The complete block by BoNT/E (as compared to the partialone of BoNT/A) indicates that residues 181–197 at the C-ter-minus of SNAP-25 are essential for exocytosis, in accordancewith previous studies13,36.

Scrutinizing exocytic bursts revealed an additional change

in BoNT/A-poisoned cells, which was that exocytic bursts wereslowed down. The averaged Cm response from 50 control cellsdisplayed an exocytic burst having two exponentials with timeconstants of 51.6 ms and 195.3 ms. However, the response of46 BoNT/A-treated cells could be well fitted by a single expo-nential with a time constant of 207.8 ms, which is similar tothe second component of the control cells (Fig. 4c). As aresult, the initial maximal rate of rise of Cm was reduced 4.7-fold (see Fig. 4d). When Cm responses from BoNT/A-treat-ed cells were analyzed individually, 33 out of 38 flash responsescould be adequately fitted by a single exponential with anoverall average amplitude of 398 ± 37 fF and a time constantof 281 ± 27.8 ms. This value is not different from that of thesecond component of exocytic bursts in control cells. More-over, the time constant of this slow component was more orless constant in the [Ca2+]i range between 15 and 70 µM. Wetherefore suggest that the truncated form of SNAP-25 cannotmediate a particularly rapid interaction between calcium-sen-sor and release machinery for catecholamine release (see Dis-cussion). Alternatively, this suppression of the rapid phase ofthe exocytic burst by BoNT/A may represent the block of a

a b c

I AM

P (

pA

)[C

a2+ ]

i (µM

)C

m

I AM

P (

pA

)[C

a2+ ]

i (µM

)C

m

I AM

P (

pA

)[C

a2+ ]

i (µM

)C

m

Time (s)Time (s)Time (s)

Control

Tetx_LC (4 µM)

Control

BoNT/D (400 nM)

Control

BoNT/C1 LC(2µM)

Table 1. Amplitudes of the exocytic burst in pairedexperiments

Experiment Control (fF) Treatment (fF)

AMP-PCP/NP-EGTA 242 ± 47 (n = 9) 21 ± 3 (n = 16)AMP-PCP/DM-nitrophen 340 ± 82 (n = 4) 14 ± 7 (n = 10)AMP-PNP/DM-nitrophen 496 ± 61 (n = 11) 79 ± 27 (n = 12)BoNT/A (800 nM) 511 ± 38 (n = 50) 363 ± 27 (n = 38)BoNT/E (400 nM) 196 ± 27 (n = 13) 37 ± 9 (n = 14)BoNT/C1 (2 µM) 260 ± 37 (n = 6) 29 ± 4 (n = 6)BoNT/D (400 nM) 423 ± 86 (n = 11) 43 ± 11 (n = 13)TeNT (4 µM) 226 ± 28 (n = 13) 44 ± 8 (n = 12)

The capacitance values were measured at 800 ms after flashes. Values givenare mean ± SE.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


small, more rapid and more toxin-sensitive compo-nent of exocytosis, not necessarily related to cate-cholamine release. In conclusion, BoNT/A has a dualeffect, one at a relatively early step and another onemodifying the exocytic burst, which is considered tobe the final step in exocytosis.

BoNT/C1 cleaves both syntaxin and SNAP-25 inchromaffin cells31. This toxin cleaves SNAP-25 at apeptide bond (R198–A199) adjacent to the BoNT/Acleavage site (V.V. and H.N., unpublished results). Totest the role of syntaxin in different components ofsecretion in chromaffin cells, we used 2 µMBoNT/C1-LC in the pipette solution. If an intact syn-taxin were not essential for exocytosis, we wouldexpect the same incomplete inhibition as seen withBoNT/A. However, BoNT/C1 completely blocked thesecretion at low [Ca2+]i, suggesting that syntaxin isimportant in the final step of exocytosis (Fig. 5a).

Synaptobrevin is the substrate for TeNT andBoNT/B, D, F and G. In the presence of 400 nM BoNT/D-LC,capacitance and amperometric measurements (Fig. 5b) showthat both components of secretion are blocked at low [Ca2+]i.Similar results were also obtained for 4 µM TeNT-LC (Fig. 5c),suggesting that synaptobrevin also is critical for the final stepsof exocytosis. Comparing effective toxin concentrations showsthat BoNT/D is much more potent in blocking secretion thanTeNT. The inactive mutants of TeNT-LC (E234Q) andBoNT/C1-LC (E230A) had no effect on secretion, suggestingthat the blockage of clostridial neurotoxins that we observedhere was due to the toxins’ metalloprotease activities.

It may be considered surprising that exocytosis is completelyblocked within a time window of about ten minutes by manytoxins that do not cleave their targets as long as those proteinsare part of a SNARE complex, as many recent models envisagean intact SNARE complex as part of a release-ready fusionmachinery. If this is the case, our findings imply that therespective proteins must cycle through a nonprotected state(which is the monomeric form) within a few minutes. Other-wise, we would expect the exocytic burst to be toxin resistant.

ATP-DEPENDENT, TOXIN-INSENSITIVE CAPACITANCE COMPONENT

We have shown that there is an additional large Cm increase athigh [Ca2+]i that is not related to catecholamine release. ThisCm response is dependent on ATP. When we substituted ATPwith AMP-PCP, this component vanished completely (Fig. 6a).This result gives us some confidence that the signal is not mere-ly an artifact of Cm measurement.

We further tested this Cm increase for its sensitivity to clostridi-al neurotoxins. To do so, we first depleted the Cm responses byrepetitive stimulations at low [Ca2+]i. Then, we elevated [Ca2+]ito high levels in order to elicit this component. This way we couldinfer that this Cm increase does not represent the same pool asthat measured at low [Ca2+]i. This protocol also provided theadvantage that we could judge whether the neurotoxins were activein a given experiment. Even when the neurotoxins succeeded inblocking secretion at low [Ca2+]i, they failed to block the Cmincreases at high [Ca2+]i. Figure 6b shows that TeNT-LC did notinhibit Cm responses at high [Ca2+]i. The same results wereobtained with BoNT/E, D, A and C1. These results suggest theexistence of an additional type of vesicles devoid of catecholamines,which are fusion competent at [Ca2+]i over 100 µM. Exocytosis ofthese vesicles is not sensitive to clostridial toxins but is dependenton ATP. Above, we concluded that this intermediate componentis elicited whenever [Ca2+]i exceeded a certain threshold. This pointis strengthened by the toxin results reported here, as no indicationof this component was observed after toxin treatment when [Ca2+]iincreases were restricted to below 70 µM.

The existence of a Cm increase not related to catecholaminerelease that is triggered at [Ca2+]i above 100 µM is somewhatpuzzling. The presence of vesicles other than chromaffin gran-ules may be an explanation. One possibility might be that low[Ca2+]i mainly triggers the fusion of catecholamine-containingchromaffin granules, whereas high [Ca2+]i preferentially triggersthe exocytosis of synaptic-like microvesicles, which containacetylcholine. This postulate agrees with the general view that

articles

Fig. 6. Secretion at high [Ca2+]i is sensitive to ATP but not toclostridial neurotoxins. (a) Comparison of secretoryresponses to high [Ca2+]i after exhaustion of exocytosis atlow [Ca2+]i. Averaged [Ca2+]i levels, capacitance traces andamperometric currents from paired experiments in the pres-ence of MgATP as control (n = 7) and in the presence of2 mM AMP-PCP substituting for MgATP (n = 17) are dis-played. The flashes were usually the third ones in a given cell.(b) The effect of TeNT on secretory responses to high[Ca2+]i after exhaustion of exocytosis at low [Ca2+]i.Averaged [Ca2+]i levels, capacitance traces and amperometriccurrents from paired experiments in control (n = 7) and inthe presence of 4 µM TeNT-LC (n = 6) are displayed. Theflashes were usually the third ones in a given cell.

b

Control AMP-PCP

Time (s)Time (s)

I AM

P (

pA

)C

m[C

a2+ ]

i (µM

)

Time (s)Time (s)

I AM

P (

pA

)C

m[C

a2+ ]

i (µM

)Control Tetx LC

a


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


the [Ca2+]i threshold for triggering small clear vesicles is higherthan that for large dense-core vesicles. However, our data areinconsistent with this idea. Chromaffin cells express acetylcholinereceptor channels. Therefore one might expect to observe a tran-sient conductance following a flash if acetylcholine were released.This was not the case, however. Furthermore, the resistance ofthis non-catecholamine Cm increase to clostridial neurotoxinssuggests that the involvement of SNAREs is not necessary for thefusion of the underlying vesicles. By exclusion, then, the expla-nation for this peculiar Cm increase seems to be that of an unspe-cific membrane fusion triggered by high level of [Ca2+]i, whichthe absence of ATP somehow prevents. Alternatively, this fusionevent could represent a previously undetected process thatinvolves a toxin-resistant set of SNARE proteins, as has been sug-gested for the apical route in polarized cells37,38.

DiscussionIn previous work, at least three late steps have been distinguishedin the secretory pathway of neuroendocrine cells: morphologicaldocking, priming, and exocytosis. Various properties have beenassigned to the individual stages; thus, docking is ATP-dependent28,priming is dependent on both ATP and calcium20,39 and exocytosisshows a steep calcuim dependence18,19. Evidence for these proper-ties has been obtained from different laboratories by very differenttechniques and by experiments on diverse time scales. It is a ques-tion of intense current research how features defined in such dif-ferent ways correspond to each other, how they relate to molecularlydefined states, and which of these states are susceptible to (or elseprotected from) the proteolytic action of clostridial neurotoxins.

Electrophysiological experiments showed that sufficientlystrong stimuli elicit a so-called exocytic burst, which is inter-preted to represent complete exocytosis of a pool of release-ready

granules (the ‘readily releasable pool’, RRP). This size of this pooldepends on [Ca2+]i preceding the stimulus and on activation ofPKC4,40. Following the exocytic burst, release proceeds at a muchslower rate, which is believed to represent the combined processof recruiting vesicles to the RRP. Flash photolysis of caged calci-um together with rapid assays of exocytosis, such as membrane-capacitance measurement and amperometry, allow a detailedkinetic analysis of secretory responses. Our goal was to establishlinks between the distinct kinetic components of such [Ca2+]ijump experiments and molecular processes.

Five of the findings reported in our study bear on this issue.First, we eliminated one of the kinetic components as a candidatefor a step in the release of catecholamines. Second, we showedthat five minutes of neurotoxin action in the absence of stimula-tion are sufficient to eliminate the exocytic burst. This implies allof SNAP-25 or syntaxin or synaptobrevin must cycle through atoxin-sensitive state within that time period. Thus, any toxin-pro-tected SNARE complex as part of a mature fusion machinery musthave a lifetime shorter than five minutes. Third, cleavage of SNAP-25 by BoNT/A suppresses exocytosis only partially and manifestsitself both at a slow (seconds to minutes) and at the fastest stepby retarding the reactions in which SNAP-25 is involved. Fourth,the size of the exocytic burst is only partially affected by BoNT/A.Fifth, the time constant of the exocytic burst after BoNT/A actionis not calcium dependent above 20 µM.

In the following, we explain these findings in the frameworkof a specific model of exocytosis control although we are awarethat alternative interpretations are possible. We start with therecent proposal of a ‘productive’ reassembly of SNARE complexesbetween membranes, in which the energy released by complexformation is used to put the membranes under tension41. Weincorporate this into a model similar to those previously pro-

articles

Fig. 7. Hypothetical model for the kinetic steps leading to exocytosis. Left, Morphological docking may require cortical actin, Munc18, Doc2and similar proteins, but may not involve the interaction of v- and t-SNAREs. The 7S complex existing on the vesicle membrane prevents thev-SNAREs from interacting with plasma membrane t-SNAREs. The 20S complex is formed after the 7S complex binds to α-SNAP and NSF.It is disassembled by ATP hydrolysis, which makes v-SNAREs available to interact with plasma membrane t-SNAREs. For simplicity, 7S com-plexes are displayed as consisting only of one v-SNARE and one t-SNARE. In the productive reassembly of SNARE complexes between mem-branes, the energy released by complex formation is used to put the membranes under tension and to make the vesicle release ready.Calcium-sensor protein may interdigitate with the distal tail of the SNARE complex’s coiled coil structure. A calcium-dependent conforma-tional change might put additional torsion on the coiled-coil, forcing the two membrane anchors together and initiating fusion. The numbersrepresent functionally discrete pools of vesicles.

Ca2+

Mor

phol

ogic

alD

ocki

ng

Cortic

al

actin

α-SNAP NSFMgATP

Exocytosis10 - 100 ms

t-SNAREs

v-SNARE

NSF?

BoNT/B,D,F,GBoNT/CBoNT/A,ETeNT

Maturation 10 s

Ca2+-Sensor

ProductiveReassembly

α-SNAP

NSF

1

2 3

4 5 6


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


posed2,40, including the recent finding that v-SNARE and t-SNAREs are in parallel orientation in the complex7 and arrive atFig. 7. In this model, we allow for morphological docking, forNSF action to separate SNARE complexes42, for SNARE complexformation (productive reassembly) and for fusion. For rapidaction of calcium, the last step should be steeply dependent oncalcium and fast, such that the readily releasable pool (state 5)can be depleted rapidly within the exocytic burst. In addition, astep of SNARE complex maturation is considered in the modelto allow for the finding that this pool recovers from depletion(following strong stimulation) on the time scale of ten seconds43

and that its size depends on PKC and [Ca2+]i39,40. This step (from

state 4 to state 5) may correspond to the recruitment of othersynaptic proteins to the complex such as the calcium-dependentactivator protein for secretion (CAPS)28 and the calcium sensor(possibly synaptotagmin). We also allow morphological undock-ing on the time scale of minutes44. Reversal of the productivereassembly step may involve NSF action. Such a model fits all ofthe findings summarized above. It explains why SNARE com-plexes are not protected from clostridial toxin action becauseNSF intermittently separates SNAREs within a time span of fiveminutes, which permits proteolytic action of the toxins. Action ofmost of the toxins, with the exception of BoNT/A, would com-pletely prevent the productive reassembly and thereby preventthe formation of a release-ready pool of vesicles.

The model also provides a hint about why BoNT/A fails to elim-inate secretion completely and seems to act both at an early and at alate step. BoNT/A cleaves off SNAP-25 the nine amino acids farthestfrom the plasma membrane, at the C-terminus. Truncated SNAP-25 still is able to form SNARE complexes that are disassembled byNSF45,46, but the C-terminal end of the molecule is likely to beinvolved in initial contact during productive reassembly41. Lack ofthese nine amino acids may slow down rather than prevent this step.In our experiments, such a change would be reflected in a decreaseof the rate of secretion at high [Ca2+]i, once the readily releasablepool is consumed. In other words, it would slow down the late com-ponent of secretion, as we observed. The same lack of nine aminoacids may also explain the slowdown of the exocytic burst itself, ifthe calcium sensor for exocytosis (possibly synaptotagmin) inter-acts with the C-terminal end of the SNARE complex. Such interac-tion between the C-terminus of SNAP-25 and synaptotagmin wasactually shown to occur in vitro (R.R.L. Gerona & T.F.J. Martin, per-sonal communication). A specific proposal of how this might hap-pen is shown in Fig. 7, referring to the finding that the ternarycomplex of SNAREs has a coiled-coil structure41. If the calcium sen-sor interdigitates with the tail of this structure, a calcium-dependentconformational change might put additional torsion on the coiledcoil, forcing the two membrane anchors together and initiatingfusion. Shortening of the SNAP-25 tail would most likely reduce theefficacy of the calcium sensor in transmitting the calcium signal tothe SNARE complex. On the other hand, the affinity of the sensoritself is not expected to change (apart from possible allosteric cou-pling effects). In the model of Heinemann and colleagues19, such achange would be most conveniently represented by a reduction ofthe rate constant γ(the rate constant of exocytosis from a triply cal-cium-bound state of the sensor) to a value of about five per second.Given the intrinsic dissociation constant of the calcium sensor of 13µM19 and a calcium-binding rate to the sensor of 8⋅106 per mole persecond47, a saturation of the calcium dependence of the release-rateconstant would be expected for [Ca2+]i in the range 20 to 70 µM, asobserved after BoNT/A treatment.

Irrespective of the specific mechanism suggested here, thefinding that removal of nine amino acids from SNAP-25 slows

down the exocytic burst and does so more at high than at low[Ca2+]i is strong evidence that this portion of the molecule isinvolved in mediating the action of a calcium sensor in rapid cal-cium-regulated exocytosis.

MethodsCELL PREPARATION AND SOLUTIONS. Chromaffin cells from bovine adrenalglands were prepared and cultured as described48. Cells were used oneto four days after preparation. The external bathing solutions for exper-iments contained (in mM) 150 NaCl, 2.8 KCl, 2 CaCl2, 1 MgCl2, 10HEPES and 2 mg per ml glucose (pH 7.2, 320 mosm). For preparingpipette solutions, we generally used 2× concentrated buffers, which con-tained 250 mM Cs-glutamate, 80 mM HEPES (pH 7.2). We added dif-ferent concentrations of NP-EGTA (gift from Dr. Ellis-Davies,Philadelphia), fura-2 (Texas Fluorescence Labs, Austin, TX), furaptra(Molecular Probes, Eugene, OR), CaCl2 or ATP for different purposesas indicated in the text. The resulting mixtures were diluted with dou-ble-distilled water for the appropriate osmolarity (310 mosm). The NP-EGTA-containing internal solutions for control consisted of (in mM)84.5 Cs-glutamate, 10 NP-EGTA, 8 CaCl2, 1 MgCl2, 2 MgATP, 0.3 GTP,0.5 furaptra, 27 HEPES. For the DM-nitrophen experiment, internalsolutions contained (in mM) 110 Cs-glutamate, 5 DM-nitrophen, 4CaCl2, 2 MgATP, 0.3 GTP, 0.5 furaptra, 35 HEPES. The basal [Ca2+]i wasmeasured to be 100–300 nM by fura-2. The pipette solution was adjust-ed to pH 7.2 by either HCl or CsOH. All experiments were performedat 32-33°C.

PHOTOLYSIS OF CAGED CALCIUM AND [CA2+]I MEASUREMENT. Flashes of ultra-violet light and fluorescence excitation light were generated asdescribed48. To avoid any influence resulting from the ‘loading tran-sient’4, we used the more calcium-selective caged compound Nitro-phenyl-EGTA49. The flash photolysis efficiency was also measured asdescribed48 except for the following changes: we used 1 mM fura-2, 2mM NP-EGTA and 2 mM CaCl2 during flash experiments. Small flashintensities were applied by adding neutral-density filters. From mea-surements of [Ca2+]i before and immediately after flashes, we calculatedthe photolysis efficiency for NP-EGTA. The photolysis efficiency of a 375V discharge flash for NP-EGTA was determined to be 52%. As the [Ca2+]ishould decay significantly during 10 s of Cm measurement after flash-es48, we used the fluorescence excitation light to measure [Ca2+]i and tosimultaneously photorelease calcium after the flashes in order to keep[Ca2+]i more or less constant. [Ca2+]i was calculated from the fluores-cence ratio R according to ref. 50. The calibration constants for furap-tra measurements before and after 375 V discharge flashes were measuredas described48. For 10 mM NP-EGTA, 0.5 mM furaptra, the changes ofthe calibration constants for a ratiometric measurement at 340 and 380nm were as follows: Rmin changed from 0.238 to 0.247, Rmax from 6.554to 4.708, Keff from 2128 µM to 2053 µM.

WHOLE-CELL PATCH CLAMP AND CAPACITANCE MEASUREMENT. Convention-al whole-cell recordings were done with sylgard-coated 2-3 MΩ pipettes.Series resistance ranged from 4–12 MΩ. An EPC-9 patch-clamp ampli-fier was used together with Pulse software (HEKA Electronics, Lam-brecht, Germany). Capacitance measurements used the Lindau-Nehertechnique17 implemented as the ‘sine+dc’ mode of the software lock-inextension of pulse software, which allowed long duration Cm measure-ment in single sweeps. A 800 Hz, 50 mV peak-to-peak sinusoid voltagestimulus was applied above a DC holding potential of -70 mV. Currentswere filtered at 2 kHz and sampled at 12 kHz. The capacitance traceswere imported to IGOR Pro (WaveMetrics, Inc., Lake Oswego, OR). Theanalyses were conducted on a Macintosh computer using IGOR Pro.Unless otherwise stated, the data are given as mean ± SE.

AMPEROMETRY. Carbon-fiber electrodes were prepared from 10 µmdiameter carbon fibers (Amoco performance products, Greenville,South Carolina) and were canulated through glass capillaries. A con-stant voltage of 780 mV versus Ag/AgCl reference was applied to theelectrode. The tip of the carbon-fiber electrode was gently pressedagainst the cell surface. The amperometric current was filtered at 3kHz, sampled at 10 kHz and further digitally filtered at 1 kHz. Arti-

articles


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


facts of amperometry due to flash irradiation were subtracted usingthe averaged trace for the same fiber at the end of the experimentwhen there was no secretion.

AcknowledgmentsWe would like to thank Dr. Ellis-Davies for samples of NP-EGTA, Drs. Corey

Smith, Reinhard Jahn and Tobias Moser for feedback on the manuscript, and

Frauke Friedlein and Michael Pilot for cell preparation. This work was

supported by grants from the Deutsche Forschungsgemeinschaft (Nr. CHV-

113/65/0) and from the European Community (Nr. CHRX-CT940500 ) to E.N.

T.B. and H.N. were supported by the Fonds der chemischen Industrie and by the

Deutsche Forschungsgemeinschaft (Nr. IIB2-Bi660/1-1).

RECEIVED 29 APRIL: ACCEPTED 23 MAY 1998

1. Holz, R.W., Bittner, M.A., Peppers, S.C., Senter, R.A. & Eberhard, D.A.MgATP-independent and MgATP-dependent exocytosis. J. Biol. Chem. 264,5412–5419 (1989).

2. Parsons, T.D., Coorssen, J.R., Horstmann, H. & Almers, W. Docked granules,the exocytic burst, and the need for ATP hydrolysis in endocrine cells. Neuron15, 1085–1096 (1995).

3. Rosenmund, C. & Stevens, C.F. Definition of the readily releasable pool ofvesicles at hippocampal synapses. Neuron 16, 1197–1207 (1996).

4. Neher, E. & Zucker, R.S. Multiple calcium-dependent processes related tosecretion in bovine chromaffin cells. Neuron 10, 21–30 (1993).

5. Söllner, T. et al. SNAP receptors implicated in vesicle targeting and fusion.Nature 362, 318–324 (1993).

6. Südhof, T.C. The synaptic vesicle cycle: a cascade of protein-proteininteractions. Nature 375, 645–653 (1995).

7. Hanson, P.I. & Jahn, R. Structure and conformational changes in NSF and itsmembrane receptor complexes visualized by quick-freeze/deep-etch electronmicroscopy. Cell 90, 523–535 (1997).

8. Foran, P., Lawrence, G. & Dolly, J.O. Blockade by botulinum neurotoxin B ofcatecholamine release from adrenochromaffin cells correlates with itscleavage of synaptobrevin and a homologue present on the granules.Biochemistry 34, 5494–5503 (1995).

9. Niemann, H., Blasi, J. & Jahn, R. Clostridial neurotoxins: new tools fordissecting exocytosis. Trends Cell Biol. 4, 179–185 (1994).

10. Montecucco, C. & Schiavo, G. Mechanism of action of tetanus and botulinumneurotoxins. Mol. Microbiol. 13, 1–8 (1994).

11. Hayashi, T., Yamasaki, S., Nauenburg, S., Binz, T. & Niemann, H. Synapticvesicle membrane fusion complex: action of clostridial neurotoxins onassembly. EMBO J. 13, 5051–5061 (1994).

12. McMahon, H.T. et al. Tetanus and botulinum toxins type A and B inhibitglutamate, GABA, asparate and metenkephalin release from synaptosomes:clues to the locus of action. J. Biol. Chem. 267, 21338–21343 (1992).

13. Lawrence, G.W., Foran, P., Mohammed, N., DasGupta, B.R. & Dolly, J.O.Importance of two adjacent C-terminal sequences of SNAP-25 in exocytosisfrom intact and permeabilized chromaffin cells revealed by inhibition withBotulinum neurotoxins A and E. Biochemistry 36, 3061–3067 (1997).

14. Dreyer, F., Rosenberg, F., Becker, C., Bigalke, H. & Penner, R. Differentialeffects of various secretagogues on quantal transmitter release from mousemotor nerve terminals treated with botulinum A and tetanus toxin. Naunyn-Schmiedebergs Arch. Pharmacol. 335, 1–7 (1987).

15. Capogna, M., McKinney, R.A., O’Connor, V., Gähwiler, B.H. & Thompson,S.M. Ca2+ or Sr2+ partially rescues synaptic transmission in hippocampalcultures treated with botulinum toxin A and C, but not tetanus toxin. J.Neurosci. 17, 7190–7202 (1997).

16. Neher, E. & Marty, A. Discrete changes of cell membrane capacitanceobserved under conditions of enhanced secretion in bovine adrenalchromaffin cells. Proc. Natl Acad. Sci. USA 79, 6712–6716 (1982).

17. Gillis, K. D. in Single-Channel Recording 2nd edn. (eds. Sakmann, B. & Neher,E.) 155–198 (Plenum, NewYork, 1995).

18. Thomas, P., Wong, J.G., Lee, A.K. & Almers, W. A low affinity Ca2+ receptorcontrols the final steps in peptide secretion from pituitary melanotrophs.Neuron 11, 93–104 (1993).

19. Heinemann, C., Chow, R.H., Neher, E. & Zucker, R.S. Kinetics of thesecretory response in bovine chromaffin cells following flash photolysis ofcaged Ca2+. Biophys. J. 67, 2546–2557 (1994).

20. Bittner, M.A. & Holz, R.W. Kinetic analysis of secretion from permeabilizedadrenal chromaffin cells reveals distinct components. J. Biol. Chem. 267,16219–16225 (1992).

21. Banerjee, A., Barry, V.A., DasGupta, B.R. & Martin, T.F.J. N-Ethylmaleimide-sensitive factor acts at a prefusion ATP-dependent step in Ca2+-activatedexocytosis. J. Biol. Chem. 271, 20223–20226 (1996).

22. Nichols, B.J., Ungermann, C., Pelham, H.R.B., Wickner, W.T. & Hass, A.Homotypic vacuolar fusion mediated by t- and v-SNAREs. Nature 387,199–202 (1997).

23. Colombo, M.I., Taddese, M., Whiteheart, S.W. & Stahl, P.D. A possiblepredocking attachment site for N-ethylmaleimide-sensitive fusion protein.Insights from in vitro endosome fusion. J. Biol. Chem. 271, 18810–18816(1996).

24. Höhne-Zell, B. & Gratzl, M. Adrenal chromaffin cells contain functionallydifferent SNAP-25 monomers and SNAP-25/syntaxin heterodimers. FEBSLett. 394, 109–116 (1996).

25. Otto, H., Hanson, P.I. & Jahn, R. Assembly and disassembly of a ternarycomplex of synaptobrevin, syntaxin, and SNAP-25 in the membrane ofsynaptic vesicles. Proc. Natl Acad. Sci. USA 94, 6197–6201 (1997).

26. Hay, J.C. & Martin, T.F.J. Phosphatidylinositol transfer protein required forATP-dependent priming of Ca2+-activated secretion. Nature 366, 572–580(1993).

27. Hay, J.C. et al. ATP-dependent inositide phosphorylation required for Ca2+-activated secretion. Nature 374, 173–177 (1995).

28. Martin, T.F.J. Stages of regulated exocytosis. Trends Cell Biol. 7, 271–276(1997).

29. Binz, T. et al. Proteolysis of SNAP-25 by types E and A botulinal neurotoxins.J. Biol. Chem. 269, 1617–1620 (1994).

30. Blasi, J. et al. Botulinum neurotoxin A selectively cleaves the synaptic proteinSNAP-25. Nature 365, 160–163 (1993).

31. Foran, P., Lawrence, G., Shone, C.C., Foster, K.A. & Dolly, J.O. Botulinumneurotoxin C1 cleaves both syntaxin and SNAP-25 in intact and chromaffincells: Correlation with its blockade of catecholamine. Biochemistry 35,2630–2636 (1996).

32. Blasi, J. et al. Botulinum neurotoxin C1 blocks neurotransmitter release bymeans of cleaving HPC-1/syntaxin. EMBO J. 12, 4821–4828 (1993).

33. Plattner, H., Artalejo, A.R. & Neher, E. Ultrastructural organization of bovinechromaffin cell cortex—Analysis by cryofixation and morphometry ofaspects pertinent to exocytosis. J. Cell Biol. 139, 1709–1717 (1997).

34. Pusch, M. & Neher, E. Rates of diffusional exchange between small cells and ameasuring patch pipette. Pflügers Arch. 411, 204–211 (1988).

35. Poulain, B. et al. Differences in the multiple step process of inhibition bytetanus toxin and botulinum neurotoxins type A and B at aplysia synapses.Neuroscience 70, 567–576 (1996).

36. Bittner, M.A. & Holz, R.W. Protein kinase C and clostridial neurotoxinsaffect discrete and related steps in the secretory pathway. Cell. Mol. Neurobiol.13, 649–664 (1993).

37. Ikonen, E., Tagaya, M., Ullrich, O., Montecucco, C. & Simons, K. Differentrequirements for NSF, SNAP, and rab proteins in apical and basolateraltransport in MDCK cells. Cell 81, 571–580 (1995).

38. Weimbs, T., Low, S.H., Chapin, S.J. & Mostov, K.E. Apical targeting inpolarized epithelial cells: there’s more afloat than rafts. Trends Cell Biol. 7,393–399 (1997).

39. von Rüden, L. & Neher, E. A Ca-dependent step in the release ofcatecholamines from adrenal chromaffin cells. Science 262, 1061–1065(1993).

40. Gillis, K.D., Mößner, R. & Neher, E. Protein kinase C enhances exocytosisfrom chromaffin cells by increasing the size of the readily releasable pool ofsecretory granules. Neuron 16, 1209–1220 (1996).

41. Hanson, P.I., Heuser, J.E. & Jahn, R. Neurotransmitter release—four years ofSNARE complexes. Curr. Opin. Neurobiol. 7, 310–315 (1997).

42. Barnard, R.J.O., Morgan, A. & Burgoyne, R.D. Stimulation of NSF ATPaseactivity by alpha-SNAP is required for SNARE complex disassembly andexocytosis. J. Cell Biol. 139, 875–883 (1997).

43. Moser, T. & Neher, E. Rapid exocytosis in single chromaffin cells recordedfrom mouse adrenal slices. J. Neurosci. 17, 2314–2323 (1997).

44. Steyer, J.A., Horstmann, H. & Almers, W. Transport, docking and exocytosis ofsingle secretory granules in live chromaffin cells. Nature 388, 474–478 (1997).

45. Otto, H., Hanson, P.I., Chapman, E.R., Blasi, J. & Jahn, R. Poisoning bybotulinum neurotoxin A does not inhibit formation or disassembly of thesynaptosomal fusion complex. Biochem. Biophys. Res. Comm. 212, 945–952(1995).

46. Hayashi, T., Yamasaki, S., Nauenburg, S., Binz, T. & Niemann, H.Disassembly of the reconstituted synaptic vesicle membrane fusion complexin vitro. EMBO J. 14, 2317–2325 (1995).

47. Klingauf, J. & Neher, E. Modeling buffered Ca2+ diffusion near themembrane: implications for secretion in neuroendocrine cells. Biophys. J. 72,674–690 (1997).

48. Xu, T., Naraghi, M., Kang, H. & Neher, E. Kinetic studies of Ca2+ binding andCa2+ clearance in the cytosol of adrenal chromaffin cells. Biophys. J. 73,532–545 (1997).

49. Ellis-Davies, G.C. & Kaplan, J.H. Nitrophenyl-EGTA, a photolabile chelatorthat selectively binds Ca2+ with high affinity and releases it rapidly uponphotolysis. Proc. Natl Acad. Sci. USA 91, 187–191 (1994).

50. Grynkiewiez, G., Poenie, M. & Tsien. R.Y. A new generation of Ca2+

indicators with greatly improved fluorescence properties. J. Biol. Chem. 260,3440–3450 (1985).

articles


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


Recurrent axon collaterals of CA3 pyramidal cells make extensiveexcitatory synapses on the dendrites of neighboring pyramidalcells1, so that when one pyramidal cell fires, its neighbors are pow-erfully excited2,3. This facilitates the rapid synchronization ofaction-potential firing in CA3 neurons that underlies the normalelectroencephalographic pattern known as hippocampal sharpwave activity4,5 and periodic in vitro burst discharges6. This syn-chrony is important for activity-dependent modification of synap-tic strength4, but the positive feedback from recurrent collateralsthat underlies the synchronization should produce continuousdischarging of all CA3 neurons7, such as may occur during humantemporal lobe seizures8. Instead, in hippocampal slices from nor-mal animals, experimental manipulations that decrease inhibi-tion or increase excitation6,8 produce only brief populationdischarges that resemble sharp waves4 and the pathological elec-troencephalographic interictal spike pattern that indicates apropensity for temporal lobe seizures8. During these discharges,CA3 neurons fire a burst of action potentials during a 50 to 200millisecond depolarization that begins and ends within a few mil-liseconds of the rest of the population4,6,8, followed in vitro by arelatively silent period that lasts until the next burst6. What processterminates burst activity? Calcium influx during the burst trig-gers a potassium current that results in an afterhyperpolarizationthat can terminate bursts of action potentials initiated by a depo-larizing current injection9. Thus the afterhyperpolarization, inconjunction with inhibitory postsynaptic conductances, is a log-ical mechanism for burst termination6.

Three lines of evidence, however, raise the possibility thatinhibitory conductances are not the primary mechanism termi-nating CA3 bursts. First, the inhibitory conductance necessaryto counteract the excitatory input to a bursting CA3 pyramidalcell should produce a dramatic increase in membrane conduc-tance during the course of the burst10, but studies to date do notprovide evidence for such a conductance increase11,12. Second,CA3 bursts are only modestly prolonged by experimental manip-ulations that block the afterhyperpolarization and inhibitorypostsynaptic γ-aminobutyric acid (GABA) type A13 and B14

receptor conductances, indicating that these conductances are

not necessary for burst termination. Finally, systems characterizedby the amount of positive feedback present in CA3 becomelocked into the state favored by the feedback7,15. For example,during an action potential, depolarization of the neuronal mem-brane is accelerated by the positive feedback provided by volt-age-dependent activation of depolarizing sodium conductances.The positive feedback must be removed by inactivation of thedepolarizing sodium conductances before inhibitory potassiumconductances can repolarize the membrane16.

Analogous to the action potential, positive feedback in areaCA3 could be reduced during the burst by a use-dependent lim-itation of synaptic strength at recurrent synapses. Modeling stud-ies demonstrate that the bursts are not sustained below a criticalsynaptic strength6. Synaptic strength could be decreased by recep-tor desensitization or by inhibition of glutamate release via presy-naptic metabotropic glutamate receptors17. However, these effectsmay not produce the necessary rate of decrease in synapticstrength17–19, and use-dependent failure of synapses from CA3to CA1 pyramidal cells is independent of these two processes20.

Only a few neurotransmitter-containing vesicles are imme-diately available for release from hippocampal terminals.Anatomical studies suggest there are 2 to 36 docked glutamatevesicles per pyramidal cell synapse21,22, and electrophysiologicalstudies indicate that between 1 and 15 vesicles are available forimmediate release20,23–25. When this supply is exhausted, it isreplenished from a larger reserve pool23 with a time constant (at36° C) of two to ten seconds20,24,25. Here we demonstrate thatthe number of releasable glutamate vesicles and the probability oftheir release regulates the positive feedback mediated by recur-rent collateral synapses, and thereby the synchronous activity ofthe CA3 network.

ResultsINHIBITORY CONDUCTANCES DURING CA3 POPULATION BURSTS

We performed three experiments to assess the role of inhibitoryconductances in terminating CA3 population bursts. First, wereasoned that if CA3 population bursts are terminated by nega-tive feedback in the form of inhibitory conductances, then the

articles

Presynaptic modulation of CA3 network activity

Kevin J. Staley, Mark Longacher, Jaideep S. Bains and Audrey Yee

Departments of Neurology and Pediatrics, University of Colorado Health Sciences Center, Box B182, 4200 East 9th Avenue, Denver, Colorado 80262, USA

Correspondence should be addressed to K.J.S. ([email protected])

The simultaneous discharge of hippocampal CA3 pyramidal cells is a widely studied in vitro model ofphysiological and pathological network synchronization. This network is rapidly activated because ofextensive positive feedback mediated by recurrent axon collaterals. Here we show that population-burst duration is limited by depletion of the releasable glutamate pool at these recurrent synapses.Postsynaptic inhibitory conductances further limit burst duration but are not necessary for burst ter-mination. The interval between bursts in vitro depends on the rate of replenishment of releasableglutamate vesicles and the probability of release of those vesicles at recurrent synapses. Thereforepresynaptic factors controlling glutamate release at recurrent synapses regulate the probability andduration of synchronous discharges of the CA3 network.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


input conductance of neurons participating in the burst shouldincrease as these conductances become large enough to negateexcitatory input from recurrent collaterals and terminate theburst10. We induced CA3 population bursts by increasing theextracellular potassium concentration to 8.5 mM, and then esti-mated the membrane conductance of CA3 pyramidal cells bymeasuring the currents required to clamp the membrane at var-ious potentials during the bursts (Fig. 1). The membrane con-ductance was maximal near the beginning rather than the endof the burst, and the reversal potential of the membrane currentsdid not change substantially during the burst (n = 3), suggestingthat there was no increase in inhibitory conductances near theend of the burst. Further, blocking inhibitory GABAA conduc-tances with picrotoxin had the same effect on the reversal poten-tial of membrane currents at all time points during a burst (n =3). This indicates that GABAA currents do not increase near theend of a burst, and excludes the possibility that a progressiveincrease in GABAA inhibitory conductance was masked by a pro-portional decrease in excitatory conductances.

The second experiment was based on the effect of postsynapticinhibitory conductances on the probability of firing an actionpotential in response to excitatory postsynaptic current10. If CA3population bursts are limited by inhibitory conductances, thenthe number of action potentials triggered by depolarizing cur-rent injection should be substantially reduced at the end of a

burst compared to just prior to a burst. We induced spontaneousCA3 population bursts using 8.5 mM extracellular potassium,and recorded the number of action potentials that could be trig-gered in pyramidal cells by a 500-millisecond depolarizing currentof the same magnitude as that recorded during spontaneousbursts (1 nA; e.g. Fig. 1b). The number of action potentials elicit-ed within one second after a population burst (11.53 ± 0.3) wasnot significantly different from the number elicited one secondprior to a burst (12.44 ± 0.4; Fig. 2a and b; n = 4 cells, 35 cur-rent injections). Thus, although the pyramidal cell firing ratedrops during prolonged depolarizing currents, the time scale overwhich this occurs is poorly correlated with burst duration (Fig.2a), and bursts produce only a small decrement in the probabil-ity of action potential firing.

Finally, if CA3 population bursts are limited by negativefeedback in the form of inhibitory conductances, then phar-macological blockade of these conductances should result incontinuous activation of the CA3 network. When the after-hyperpolarization was blocked with norepinephrine26, popu-lation bursts were only moderately prolonged, and there wasno significant impact on burst frequency (n = 6; Fig. 2b and c).Burst length and frequency were also modestly affected byenhancing or blocking inhibitory postsynaptic GABAA con-ductances (n = 10; Fig. 2d). Even when the afterhyperpolariza-tion, GABAA, and GABAB conductances were simultaneously

articles

Fig. 1. Membrane currentsin CA3 pyramidal cells dur-ing population bursts. (a)Top panel, whole-cellrecording of the membranepotential of a pyramidal cellduring a CA3 populationburst induced by 6 mMextracellular potassium([K+]o) (resting membranepotential -55 mV). Middlepanel, in the same cell,membrane currentsrecorded when the mem-brane was clamped at theindicated potentials duringa population burst. Inset, ina different slice, simultane-ous recording of CA3extracellular field potentialand the membrane currentin a pyramidal cell demon-strates that the initial rip-ples in intraburst voltageclamp currents reflectspikes in the extracellularfield potential. Bottompanel, estimate of inputconductance during a pop-ulation bursts, obtained by subtracting the currents shown in the middle panel, and dividing by the difference in test potential. Thereis no evidence of an inhibitory current or conductance of sufficient amplitude to terminate the burst10. Recording performed with apotassium electrode solution. (b) Membrane currents during CA3 population bursts before and after blocking GABAA conductanceswith picrotoxin. Currents were recorded using cesium and QX314 in the electrode solution. Currents at all time points during theburst reverse at similar potentials (symbols, inset), so that there is no evidence for burst termination by a large, late inhibitory cur-rent. Bottom panel, in the same cell, the burst is prolonged after blocking GABAA receptors, and the reversal potential of the currentsat all time points during the burst is shifted positive by 10 mV. The uniform shift in reversal potential at all time points during the burstindicates that the GABAA conductance is already maximal at the earliest part of the burst.

a b

Voltage clamp

50 ms

25 mV1 nA20 nS

50 ms

1 nA

Clamp current

EC field potential


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


blocked by bath application of 40 µM norepinephrine, 100 µMpicrotoxin and 1 µM CGP55845A, prolonged CA3 dischargesnever occurred (Fig. 2e and f), indicating that inhibitory con-ductances are not necessary for burst termination.

REGULATION OF BURSTS BY SUPPLY OF RELEASABLE GLUTAMATE

We then considered the idea that burst termination is due atleast in part to a decrease in the positive feedback mediated byglutamatergic recurrent collateral synapses. In computer mod-els, population bursting is a very sharp function of the strengthof recurrent excitatory synapses6. Synaptic strength is deter-mined by several pre- and postsynaptic factors; we considered

the effect of the rate of glutamate release from the presynapticterminal (RGlu ) on synaptic strength. RGlu is the product of thenumber of releasable vesicles (NR) in the presynaptic terminalmultiplied by the probability of release of a releasable vesicle(PR) (reviewed in ref. 27). PR is considered here to be a contin-uous function of time whose value is relatively low betweenbursts and high during a burst due to the barrage of actionpotentials reaching the presynaptic terminal27. As glutamatevesicles are released at recurrent synapses during a burst, NRdecreases until exhaustion of releasable glutamate results insynaptic failure (i.e. NR → 0 causes RGlu → 0). If this occurs atenough synapses, the strength of recurrent synapses and their

articles

Fig. 2. Role of postsynaptic inhibitoryconductances in the control of CA3burst discharges. (a) Effect of the after-hyperpolarization on CA3 pyramidalcell burst firing. Similar trains of actionpotentials were elicited in a CA3 pyra-midal cell by 500 ms, 1 nA currentinjections at various times after burstsinduced by 8.5 mM [K+]o. Top tracesare membrane potential responses tocurrent injection; bottom traces areschematics of the timing of currentinjections. (b) Number of actionpotentials in each train as a function ofthe delay between the end of a popula-tion burst and the start of the currentinjection for the cell shown in (a). Theaverage interburst interval was 2.6 s.For each delay, points represent meansand bars represent standard error; n =1 to 4 for each delay. (c) Effect ofblocking IAHP (afterhyperpolarizationcurrent) on burst discharge.Intracellular and extracellular record-ings are superimposed. Top two traces,in control solution (8.5 mM [K+]o), theduration of the afterhyperpolarizationdoes not predict the timing of the nextburst. Bottom trace, blocking IAHP with40 µM norepinephrine does not alterburst timing; burst length increased by23 ± 7% (average ± standard error), andinterburst interval decreased by 1 ±12% (n = 6). Scale, 20 mV IC, 2 mV EC.(d) Effect of GABAA conductances onburst length (large symbols) and theinterval between CA3 populationbursts (small symbols) was assessed bybath application of 60 µM pentobarbital(pb) and 100 µM picrotoxin (ptx).Bursting was induced by 8.5 mM [K+]o.Heavy line is a moving average.Increasing the GABAA conductancewith 60 µM pentobarbital decreasedburst length by 12 ± 11% and decreased the interburst interval by 20 ± 14 % (n = 10). Blocking GABAA conductances with picrotoxinincreased burst length by 39 ± 19 % and increased the interburst interval by 50 ± 15 % compared to control (n = 10). (e) Bursts recordedfrom two CA3 pyramidal cells in 6 mM [K+]o (control) and after blockade of the afterhyperpolarization, GABAA, GABAA, conductances andglutamate receptor desensitization. Top panels are current-clamp recordings in a cell recorded with K+ electrode solution; bottom panels aremembrane currents in a cell voltage clamped at –55 mV with cesium and QX314 in the electrode solution. (f) Burst length (l ) and inter-burst interval (P ) for the current-clamped pyramidal cell shown in (e), during perfusion with 6 mM [K+]o and after addition of NE/PTX/CGPand cyclothiazide. Lines represent average values for each condition.

a b

c d

f

Time after spont. burst (ms)

Minutes

Burst number

Nu

mb

er o

f sp

ikes

in b

urs

tB

urs

t in

terv

al (

sec)

P

Bu

rst

len

gth

(ms)

J

Bu

rst

inte

rval

(se

c) P

e

Bu

rst

len

gth

(m

s) j

1 s


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


positive feedback would be diminished tothe point that bursts are terminated6.

The rate of increase in the ability of theCA3 network to support a burst was mea-sured by triggering a second burst at ran-dom time intervals after a spontaneousburst in 8.5 mM extracellular potassium(Fig. 3a and b). The length of the evokedburst increased in proportion to the timeelapsed since the end of the last sponta-neous burst with a time constant of 0.6 to2 seconds (n = 12). Burst length was notzero at the shortest stimulation times atleast in part because of evoked glutamaterelease at synapses that did not participatein the burst, such as at terminals on per-forant path axons, mossy fibers, and cutrecurrent collaterals. Although these slicesexhibited a wide range of spontaneousburst lengths (60–420 ms; compare Fig. 3ato Fig. 3b), the fractional increase in burstlength followed similar time courses.Because the afterhyperpolarization andpostsynaptic inhibitory conductances thatmodulate burst length (Figs 1 and 2) werenot blocked, the time constants obtainedin this experiment reflect the decay ofthose conductances as well as the re-accu-mulation of releasable glutamate. Whenthe afterhyperpolarization, GABAA, andGABAB conductances were blocked, thetime constant for the rate of increase inburst length increased to 7.1 ± 1.9 seconds(Fig. 3c; n = 7). The time course for theincrease in burst length is similar to thetime course both for successfully inducinga second population burst by triggeringaction potentials in a single CA3 pyrami-dal cell after a spontaneous burst3, and forthe replenishment of releasable glutamateat individual synapses in the hippocampalslice20. These findings support the ideathat depletion of releasable glutamate is adeterminant of burst termination, and thatthe degree of replenishment modulates theprobability of a subsequent discharge.

Alternatively, the strength of recurrentsynapses could be limited by a postsynapticmechanism such as desensitization. How-ever, when glutamate receptor desensitiza-tion was inhibited by 100 µMcyclothiazide28, there was only a modestincrease in the length of CA3 bursts record-ed during blockade of GABAA, GABAB,and after-hyperpolarization conductances (17 ± 12%; n = 4; Fig.2e and f). To directly evaluate postsynaptic glutamate receptorfunction, we measured the current induced by glutamate appli-cation to the dendrites of bursting CA3 pyramidal cells. There wasno correlation between the amplitude of the initial glutamate-induced current and the time elapsed since the end of the lastburst (basal dendrites, n = 4; distal apical dendrites, n = 22; 10cells, 9 slices, 6 animals; bursts induced by 8.5 mM external potas-sium in two cells, and by tetanization of pyramidal cell layer in

eight cells; Fig. 4a–c). However, as in the experiments utilizingelectrical stimulation (Fig. 3), the later, polysynaptic response wassubstantially increased at longer post-burst application intervals,indicating an increase in the strength of recurrent synapses.

These results suggest that during a burst, the strength of recur-rent synapses is limited by a presynaptic mechanism. Evaluation ofspontaneous excitatory postsynaptic currents recorded beforebursting began compared to events recorded between CA3 burstsinduced by 8.5 mM extracellular potassium revealed a much larg-

articles

Fig. 3. Releasable glutamate and burst timing. (a) In a CA3 network bursting spontaneouslyin 8.5 mM extracellular K+, the length of pyramidal cell bursts evoked by single electricalstimuli in s. moleculare is proportional to the time interval between stimulation and the ter-mination of the preceding spontaneous burst. Resting membrane potential is -63 mV. Leftpanel, examples of bursts evoked at the delays indicated (seconds) compared to a sponta-neous burst (Spont). Right panel, plot of evoked burst length versus post-burst stimulus delay(bars are standard error, n = 2 to 12 for each delay value). Solid line represents the leastsquares fit exponential (time constant 600 ms). (b) Extracellular recording from the sametype of experiment as shown in (a). Although the CA3 burst length in this slice is only half aslong as in (a), the rate of increase in burst length shows a similar proportionality to the inter-val since the last burst (time constant of 900 ms). Average interval between spontaneousbursts was 5.4 seconds. (c) Extracellular CA3 recording of the same experiment in a slice inwhich GABAA and GABAB postsynaptic conductances and IAHP were blocked. Evoked burstlength increased with a time constant of seven seconds following a spontaneous burst.

a

b

c

Stimulus delay (ms)

Stimulus delay (ms)

Stimulus delay (ms)

No

rmal

ized

bu

rst

len

gth

No

rmal

ized

bu

rst

len

gth

Bu

rst

len

gth

(m

s)

spont

spont

spont


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


er decrease in frequency versus amplitude (Fig. 3c; n = 4; ref. 29).This result supports the idea that the mechanism for decreasingthe strength of recurrent excitatory synapses is presynaptic18, butdoes not indicate the nature of the mechanism. To evaluate thepossibility that glutamate release is limited by negative feedback viapresynaptic metabotropic glutamate receptors17,19, we blockedthis receptor using the broad-spectrum metabotropic antagonistMCPG. Although 250 µM MCPG produced a 37 ± 11% increasein the frequency of CA3 bursts induced by 6 mM extracellularpotassium and 100 µM picrotoxin, the length of bursts was notaffected (4 ± 2 % increase, n = 8), indicating that metabotropicreceptor feedback was not the critical determinant of the strengthof recurrent synapses during a burst.

To directly assess the supply of releasable glutamate availableat the end of a burst, glutamate release was evoked by applica-tion of hyperosmotic extracellular fluid (500 mM mannitol) tothe slice30,31. Glutamate release by hyperosmotic media is notdependent on action potentials or intracellular calcium and isnot affected by known presynaptic inhibitory feedback mecha-nisms32. Bursts were induced by a tetanic stimulation of the CA3pyramidal cell layer and recorded during blockade of GABAA andGABAB postsynaptic conductances and the afterhyperpolariza-tion. Mannitol substantially increased the frequency of sponta-neous glutamate release as measured by the frequency ofspontaneous EPSCs (Fig. 5a and b). Despite blocking all inhibito-ry processes, the EPSC frequency during the two seconds pre-ceding a burst was 6.5 ± 0.8 times the frequency during the two

seconds immediately following a burst (n = 18 cells in 15 slicesfrom 13 animals). The time constants describing the rate ofchange in EPSC frequency (1.8 to 5 seconds; Fig. 5c) were simi-lar to the time constants for the change in burst length illustrat-ed in Fig. 3. The EPSC frequency was not expected to be zeroimmediately after a burst because of release of glutamate fromterminals that did not participate in the burst.

REGULATION OF CA3 BURSTING BY PROBABILITY OF RELEASE

These findings support the hypothesis that excitation at CA3recurrent collateral synapses can be terminated by depletion ofreleasable glutamate. When depletion of releasable glutamateduring a burst causes recurrent synaptic strength to fall belowthe value at which bursting is supported6, not only is the burstterminated, but in addition the next burst cannot occur until thesupply of releasable glutamate is replenished. If the probabilityof initiating the next burst is proportional to the rate of gluta-mate release between bursts29, then the next burst requires bothadequate NR so that synapses have the strength to propagate theburst, and adequate glutamate release (NR×PR) to initiate theburst. Changing the value of PR between bursts would then leadto a proportionate change in the degree to which NR needed torecover in order for glutamate release to be sufficient to triggeranother burst; the change in the recovery of NR would be reflect-ed in the interburst interval.

Thus agents that decrease PR such as adenosine and baclofen33

should increase the interval between bursts34,35, whereas increas-

articles

Fig. 4. Glutamate receptor function isnot altered following a burst. (a)Currents evoked by four consecutiveapplications of 100 µM glutamate (arrow-head) to the distal apical dendrites of apyramidal cell voltage clamped at -50 mV.Bursting induced by 8.5 mM extracellularpotassium. Intervals between the end ofthe preceding burst and glutamate appli-cation were randomly generated and arenoted next to the current recordings. Theaverage interval between bursts was 2.3seconds; thus the glutamate application2.5 seconds after the previous burst trig-gers a burst, but the initial rate of rise ofthe current is the same as for the currentelicited by glutamate application 0.2 sec-onds after the last burst. Glutamate wasapplied after every tenth spontaneousburst. (b) The experiment shown in (a)was repeated in a slice in which CA3bursting was induced by tetanic stimula-tion of the CA3 pyramidal cell layer. Theaverage interburst interval in this prepara-tion was 13 seconds, which permittedlonger delays between the end of a burstand glutamate application. As in (a), glutamate application at delays closer to the interburst interval triggered a burst more rapidly, but the ini-tial response to glutamate was unchanged (inset). Glutamate was applied after every fourth burst in (b) and (c). (c) Summary of initialresponse to glutamate application in three experiments in which the average CA3 interburst interval was 10 to 15 seconds. Glutamateresponse was calculated as the current averaged over the first 30 ms following glutamate application. Points represent the difference at eachdelay value from the mean. There is no significant change in glutamate receptor function during or after CA3 bursts. (d) Cumulative proba-bility plots of the intervals between spontaneous excitatory postsynaptic currents (EPSCs). EPSCs were recorded for two-minute intervals ina CA3 pyramidal cell voltage clamped at -50 mV under three conditions: normal extracellular media (control); five minutes after switching to8.5 mM [K+]o, representing the two-minute interval just prior to the onset of burst activity (-); and five minutes after the onset of bursting in8.5 mM [K+]o (+). Inset shows averaged EPSCs for each condition. EPSCs were recorded using a potassium electrode solution.

0.6 s

9.4 s 2.6 s8.0 s

50 ms

100 pA

glutamate

0 2 4 6 8Interevent interval, sec

0.0

1.0

Prob

abili

ty

-+

D

control

-

+control

1 pA

20 ms

glutamate

100 ms20 pA

2 s0.2 s1.5 s

2.5 s

10 ms

20 pA

0 1 2 3 4 5 6 7 8 9 10

seconds after burst

-100

-50

-0

50

100

% d

iffer

ence

from

mea

n

C

a b

c d

Seconds after burst Interevent interval (sec)

Pro

bab

ility

Dif

fere

nce

fro

m m

ean

(%

)


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


ing PR by increasing external potassium23 should decrease theinterburst interval (Fig. 6a–c). The burst length changed in pro-portion to the burst interval under these conditions, supportingthe idea that the amount of glutamate available for release isincreasing throughout the interburst interval. The effects of thesemanipulations are most consistent with a presynaptic locus ofaction, as they are the opposite of what would be expected basedon the postsynaptic effects of these manipulations (for instance,increasing potassium extracellularly would decrease inhibitiondue to the decrease the driving force for inhibitory currents36,and thus would tend to increase burst length as picrotoxin does inFig. 2d). By measuring the rate of increase in evoked burst lengthwith time elapsed from the last burst as in Fig. 3, theeffects of baclofen were shown to be mediated by adecrease in PR rather than a change in the rate ofincrease of NR (Fig. 6d). Over the wide range of burstlengths and intervals in the baclofen experiments, burstlength did not predict the time interval until the nextburst, but the length of the interval preceding a burstwas correlated with burst length (Fig. 6e). This pro-vides further evidence for the idea that burst termina-tion is the result of depletion of an excitatory factorrather than accumulation of an inhibitory factor.

DiscussionDuring synchronous CA3 population bursts in vitro,the finite supply of releasable neurotransmitter lim-its the strength of recurrent synapses. Because pop-ulation bursting is strongly dependent on the strengthof these positive-feedback synapses, the availabilityof releasable glutamate effectively controls the dura-tion of population bursts. The interval between burstsreflects the replenishment of releasable glutamate.This replenishment increases both the strength ofrecurrent synapses and the rate of glutamate release,which contributes to the initiation of the next burst29,as schematized in Fig. 7.

Although we have emphasized burst termination byexhaustion of releasable glutamate, inhibitory conduc-tances also decrease burst duration (Figs 1b, 2d–f and3c). How do these mechanisms interact to terminatebursts? The experiments in which inhibitory conduc-tances were blocked demonstrate that exhaustion ofreleasable glutamate is sufficient to terminate bursts (Fig.2c–f). Whether exhaustion of releasable glutamate is alsonecessary for burst termination can not be proved asdirectly, because it is not possible to prevent this exhaus-tion. However, three lines of evidence support thehypothesis that depletion of releasable glutamate is nec-essary for burst termination, as opposed to providing afail-safe mechanism should inhibition fail. First, inhibi-tion does not increase at the end of a burst in a mannerconsistent with a burst termination mechanism10 (Fig.1). Second, when extracellular potassium is increased,the driving forces for inhibitory chloride and potassiumcurrents are decreased36. If inhibition terminated burstsindependently of the supply of releasable glutamate, thenthe decrement in inhibition caused by increased potas-sium should increase burst length, but over a very widerange of potassium concentrations, this does not occur(Fig. 6a and b). Finally, if bursts were terminated by feed-back inhibition, then the interval following a burst wouldbe proportional to burst length, because the amount of

feedback inhibition should be proportional to the intensity of theburst, but the opposite relationship is found (Fig. 6e). However, ifthe duration of a CA3 population burst is limited by the amountof releasable glutamate at recurrent synapses, then any factor thatchanges the postsynaptic effect of glutamate should affect burstduration. Thus, although we hypothesize that inhibitory conduc-tances are neither necessary nor sufficient for burst termination,they will substantially alter the excitatory effect of released gluta-mate10. Inhibitory conductances therefore can further shorten aburst that is limited by the supply of releasable glutamate (Fig. 2c–f).

The length of CA3 population bursts described in the litera-ture varies substantially13,37–40. Much of this variation can be

articles

Fig. 5. Amount of releasable glutamate available at the end of a burst. (a)Spontaneous EPSCs recorded in a CA3 pyramidal cell before and after sponta-neous bursts (bursts are truncated). Bursting was triggered by a tetanic stimula-tion of the CA3 pyramidal layer; GABAA and GABAB postsynaptic conductancesand IAHP were blocked. Top two records are consecutive bursts that wererecorded prior to mannitol application (control); bottom ten records are con-secutive bursts recorded after application of 500 mM mannitol to the distal api-cal dendritic layer of CA3. Following a burst, the EPSC frequency is depressed.Average interburst interval was ten seconds. Holding potential was –70 mV,records were low-pass filtered at 1 kHz, and the electrode solution containedcesium and QX314. (b) Cumulative probability plots of EPSC intervals demon-strate that EPSCs are increased during application of mannitol. (c) EPSC fre-quency as a function of time elapsed since the end of the last spontaneous burst.EPSCs were binned in one-second intervals. Solid line, fit exponential function,time constant, 1.8 seconds.

0 1 2 3 4 5 6 7 8 9 10

seconds after end of burst

0

1

2

3

4

5

EPSC

freq

uenc

y, H

z

0 200 400 600 800 1000

interevent interval, ms

0.0

0.5

1.0

pro

babi

lity

control

mannitolB

man

nito

l

50 pA

500 ms

C

cont

rol

A

burst

Seconds after end of burstInterevent interval (ms)

Pro

bab

ility

Man

nit

ol

Co

ntr

ol Burst

a

b

EPSC

fre

qu

ency

(H

z)


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 6. Presynaptic modulation of CA3 population bursts.(a, b) Effect of [K+]o on CA3 burst frequency and burstlength. Burst interval and length decreased as [K+]o wasincreased, consistent with an increase in PR and a decreasein the degree to which NR had recovered at the start of theburst. Points are averaged over all bursts acquired during30-minute recording periods; bars are standard error.Experiment in (a) was performed in a transverse brain slice,and (b) in a coronal brain slice; note the difference in therange of burst lengths. (c) Dose–response relationships forthe decrease in burst frequency and increase in burst lengthwith baclofen. The length and interval of CA3 populationbursts induced by 8.5 mM [K+]o increased as baclofen wasapplied by bath at the indicated concentrations; burstingstopped in 8 µM baclofen. Picrotoxin (100 µM) was used inall baclofen and [K+]o experiments to avoid confoundingeffects due to decreased GABA release and altered Cl– gra-dients. (d) The decreased burst frequency in baclofen is notdue to a change in the rate of increase of NR. CA3 popula-tion bursts were induced by 8.5 mM [K+]o, and bursts wereevoked after every fourth spontaneous burst as describedin Fig. 2b. Under control conditions (• ) and in baclofen2 µM (o), the time constant describing the increase in burstlength at increased stimulus delays was essentially the same(solid line). Longer delays could be included in baclofenbecause the spontaneous burst interval was increased. (e,f)The duration of a burst does not predict the time intervaluntil the next burst (e) but the interval preceding a burstpredicts the burst duration (f; compare to d). Using thedata shown in averaged form in (c), normalized time inter-vals and burst lengths are plotted for each concentration ofbaclofen (control P , baclofen 1 µM p , baclofen 2 µM l ,baclofen 4 µM L ).

3 6 9K+

o , mM

5

10

15

20

inte

rbur

st in

terv

al, s

ec

250

500

burs

t le

ngth

, ms

A

interval length

500 ms2 mV

3.5

8.5

0 1 2 3 4µM baclofen

10

20

30

inte

rbur

st in

terv

al, s

ec intervallength

70

110

burs

t le

ngth

, ms

C

0 1 2 3 4 5 6

stimulus delay, sec

50

120

burs

t le

ngth

, ms

D

0.0 0.5 1.0burst length

0.0

0.5

1.0

next

bur

st in

terv

al

E

0.0 0.5 1.0

burst interval

0.0

0.5

1.0

burs

t le

ngth

5 6 7 8 9 10

K+o mM

0

5

10

15

20

25

inte

rbur

st in

terv

al, s

ec

110

150

burs

t le

ngth

, ms

intervallength

B


explained by differences in PR produced by the experimental pro-tocols used to induce bursting. The differences in PR will lead toproportional differences in NR at the start of the burst, and burstduration is limited by NR (Figs 3, 6 and 7). Some of the variationin CA3 burst length, for instance the difference noted betweenseptal and temporal slices (compare Fig. 6a and b), is not easilyexplained by differences in PR. This suggests that there may besepto-temporal differences in the number of glutamate releasesites or in the factors that modulate the effect of released gluta-mate, such as glutamate receptor desensitization or re-uptakecharacteristics41. Another potential explanation for the septo-tem-poral difference is that the interneuronal connectivity of tempo-ral transverse slices is maximized by the plane of section. Thisconnectivity makes possible the prolonged driving of CA3 by othernetworks in which epileptiform discharges are induced42–44.

Prolonged CA3 discharges that are not driven by other net-works35,45,46 consist of a primary burst followed by a series ofbrief bursts at short intervals, which are termed afterdischarges47.Although our data do not address afterdischarges explicitly, therelationship between afterdischarge length and the interval sincethe last discharge seems to follow the relationship shown in Figs3 and 7, suggesting that afterdischarge duration and interval arealso governed by the supply of releasable glutamate. If so, thenseries of afterdischarges may represent burst behavior under con-ditions in which the interburst probability of release remains veryhigh at the end of the initial burst, for instance as a consequenceof prolonged depolarizing conductances47.

In CA3 in vivo, not all pyramidal cells fire during a sharpwave4,5. This suggests that the activation of the CA3 network in

vivo is regulated not only by the supply of releasable glutamatebut also by another presynaptic mechanism, synapse-specific vari-ation in PR, which may be enhanced by neuromodulators17,33.Whether or not the supply of releasable transmitter affects othermodes of network activity, for instance those in which inhibitionalternates with excitation48,49, will depend on the rate of trans-mitter release, the number of release sites, and the rate of replen-ishment of releasable transmitter during the inhibitory periods.

Termination of network activation by exhaustion of releasableglutamate rather than postsynaptic inhibition increases the com-putational flexibility of the network, because network activity is notconstrained by a progressive dampening as inhibitory conductancesterminate the burst. The causal link between the properties of indi-vidual synapses such as PR and NR and the behavior of the networksuggests that the plasticity of synaptic properties should also bereflected in the behavior of the network. For instance, tetanization ofhippocampal afferents results in an increase in the strength of theactivated synapses31 as well as an alteration in the modes of networkactivity43. Regulation of CA3 burst duration by the supply and prob-ability of release of neurotransmitter is a novel mechanism for con-trol of a neural network that presents new avenues for understandingthe physiology of neurotransmitter release as well as new strategiesfor the treatment of epilepsy.

MethodsThe duration of reported CA3 bursts varies along the septo-temporalaxis37 and with the method used to induce bursting6,7,12–14,38–40. In orderto ascertain the range of applicability of the present findings, CA3 burstswere studied in both septal and ventral CA3, and bursts were induced

articles

Inte

rbu

rst

inte

rval

(se

c)

Bu

rst

len

gth

(ms)

Bu

rst

len

gth

(ms)

Inte

rbu

rst

inte

rval

(se

c)

bu

rst

len

gth

, ms

bu

rst

len

gth

, ms

K+o (mM) K+

o (mM)

c d

e fµM baclofen Stimulus delay (sec)

Burst length Burst interval

Bu

rst

len

gth

Nex

t b

urs

t in

terv

al

a b

Inte

rbu

rst

inte

rval

(se

c)


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


using three different manipulations: increased extracellular potassium([K+]o), GABAA receptor block, and tetanic stimulation of the CA3 pyra-midal layer43,44.

SLICE PREPARATION. Recordings from area CA3 of the adult rat hip-pocampus were made from 400-µm thick hemibrain slices at 35°C. Thebrain was cut in the coronal plane except where transverse plane is spec-ified. For both planes of section, recordings were made from slices per-pendicular to the long axis of the hippocampus. Therefore, slices fromthe septal (rostral) end of the hippocampus were used when the brainwas cut in the coronal plane, and from the temporal (caudal) end whenthe brain was sliced transversely. Extracellular solution was saturatedwith 95% O2, 5% CO2 and contained (in mM) NaCl 126, KCl 2.5,NaHCO3 26, CaCl2 2, MgCl2 2, NaH2PO4 1.25, glucose 10.

RECORDINGS. Whole-cell recordings were performed using a filling solu-tion containing (in mM) potassium methylsulfonate 123, MgCl2 2, NaCl 8,potassium ethylene glycol-bis(b-aminoethyl ether) N,N,N’,N’-tetraacteicacid (EGTA) 1, potassium adenosine 5’-triphosphate 4, and sodium guano-sine 5’-triphosphate 0.3. The whole-cell solution was buffered with 16 mMKHCO3 and saturated with 95% O2, 5% CO2. In voltage-clamp experi-ments, 2 mM QX314 (Astra Pharmaceuticals) was added to the pipettesolution to block voltage-dependent sodium conductances, and wherenoted cesium replaced potassium to improve the space clamp. Extracellu-lar recordings were performed using saline-filled patch pipettes. Record-

ings were accepted when the access resistance remained under 10 MΩ forvoltage-clamp experiments, and under 20 MΩ in experiments where nocurrent was injected. Recordings using Axoclamp 2B amplifiers were dig-itized at 2 KHz using routines written in Axobasic (Axon Instruments, Fos-ter City, California). Burst length rather than burst area was used to estimateburst intensity to facilitate comparisons between extra and intracellularrecordings (Fig. 3). Burst length was calculated as the time during which theabsolute value of the burst was above a threshold value, generally threetimes the baseline noise. Similarly, for stimuli applied after a burst, the endof a spontaneous burst was defined as 50 ms after the time at which theburst amplitude was less than a defined value, generally three times thebaseline noise. Intervals between the end of a burst and the stimulus wererandomly generated, then rounded to the nearest 200 ms. Electrical stim-uli were applied after every fourth burst to the pyramidal cell layer, andthe intensity was set to elicit one population spike in control media.

BURST INDUCTION. Bursting in CA3 was induced by increasing extracellularpotassium to between 6 and 8.5 mM as noted in the text, by blockade ofGABAA inhibition with 100 µM picrotoxin or by a single tetanic stimula-tion43 of the CA3 pyramidal cell layer. The tetanic stimulus consisted of a100 Hz, one-second train of stimuli that were of sufficient amplitude toelicit a population spike when delivered at a lower frequency. If a singletetanus did not induce bursting, it was repeated after a ten-minute inter-val43,44. When inducing bursting by tetanic stimulation, extracellular solu-tions were modified43 to Ca2+ 1.3 mM, Mg2+ 0.9 mM, and K+ 3.3 mM.

articles

Fig. 7. Schematic of proposedpresynaptic modulation of burst ter-mination and timing. (a, b) The pat-tern of bursting if the rate ofglutamate release from the presy-naptic terminal at recurrentsynapses limits synaptic strength atthese synapses. Glutamate release iscalculated as the product of PR xNR. The value of NR x PR at which apopulation burst is likely to be trig-gered29 and propagated6 is desig-nated ‘burst threshold’. Bursts stopwhen PR x NR falls below thisthreshold; bursts are possible whenPR x NR exceed threshold. For sim-plicity we assume bursts occur assoon as they are possible. In (a) and(b), NR x PR, NR, and PR are plottedfor 2 different interburst values ofPR (interburst PR = 0.25 in a and0.32 in b). Between bursts, NRincreases at a rate proportional tothe number of empty release sitesremaining, so that dNR / dT = krefill *(NRmax - NR), and NR = NRmax x (1-e (–t / τ)), where τ = 1 / krefill . NRdecreases during a burst at a rateproportional to the number of releasable vesicles25, so that during a burst dNR / dT = - kdecrease *Nr, and NR = NRmax x e (–t / τ), where τ = 1/ kdecrease. PR between bursts is assumed to have been increased by an experimental manipulation such as increased extracellular [K+]o.During a burst, Pr is assumed to increase further because of bursts of action potentials in the presynaptic cell. (c) Top panel, the relationshipsbetween interburst PR, burst length, and interburst interval. Burst interval decreases as PR increases because for a given rate of increase inNR after a burst, NR x PR reaches burst threshold more rapidly for larger values of PR. Burst length decreases as PR increases because NRdoes not recover to a very high value before PR x NR exceeds burst threshold, and so there are fewer releasable glutamate vesicles at thestart of the next burst. Bottom panel, the rate at which NR increases between bursts determines when NR x PR reaches burst threshold andtherefore sets the interburst interval. Burst length is not affected, because NR at the start of the burst is not changed. Increasing the burstthreshold increases the burst interval, because in order for NR x PR to cross burst threshold, NR must increase to closer to its maximumvalue, where the rate of increase in NR is low (a, middle panel). However, increasing the burst threshold does not change the amount bywhich NR decreases during the burst: although NR is higher at the start of the burst, the burst also terminates at a higher value of NR becauseNR x PR drops below the increased burst threshold sooner.

min maxrate of N R increase

0

2

4

6

inte

rbur

st in

terv

al, s

ec

0

100

burs

t le

ngth

, ms

length

interval

min maxP R

0

2

4

6

8

inte

rbur

st in

terv

al, s

ec

20

120

Burs

t le

ngth

, mslength

interval

CP R 0 = .32

min

max

NR

x P

R

burst

threshold

P R 0 = 0.25

2 4 6 8

seconds

0

1

PR

2 4 6 8seconds

min

max

NR

A B

max burst threshold min

b cPR0

= 0.32

NR

×P

RN

RP

R

max

aPR0

= 0.25

max

min

min

seconds seconds

min

min

max

max

rate of NR increase

burst threshold

inte

rbu

rst

inte

rval

, sec

inte

rbu

rst

inte

rval

, sec

bu

rst

len

gth

, ms

bu

rst

len

gth

, ms


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


GLUTAMATE AND MANNITOL APPLICATION. Glutamate and mannitol wereapplied via a whole-cell pipette using a 40 psi pressure pulse lasting 5 to50 ms. Intervals between the end of a burst and glutamate applicationwere randomized to a value between 200 ms and the average interburstinterval. Mannitol was applied every 4 to 30 seconds as described31.

REAGENTS. Drugs were applied by bath. Appropriate precautions weretaken to avoid oxidation of norepinephrine26. Where noted, afterhyper-polarization was blocked by addition of 40 µM norepinephrine, GABAAconductance by 100 µM picrotoxin, GABAB conductance by 1 µMCGP55845A (Ciba Geigy, Basel), and glutamate-receptor desensitizationby 100 µM cyclothiazide to the extracellular fluid. Reagents were obtainedfrom Sigma (St. Louis, Missouri). Ten micromolar acetazolamide wasapplied with pentobarbital to diminish GABAA-receptor-mediated den-dritic depolarization50.

AcknowledgementsWe thank Thomas Dunwiddie, Darrell Lewis, Charles Stevens, and Roger Traub

for comments and discussions. This work was supported by the NIH and the

Epilepsy Foundation of America.


1. Li, X.G., Somogyi, P., Ylinen, A. & Buzsaki, G. The hippocampal CA3network: an in vivo intracellular labeling study. J. Comp. Neurol. 339,181–208 (1994).

2. MacVicar, B.A. & Dudek, F.E. Local circuit interactions in rat hippocampus:interactions between pyramidal cells. Brain Res. 242, 341–344 (1982).

3. Miles, R. & Wong, R.K.S. Single neurones can initiate synchronizedpopulation discharge in the hippocampus. Nature 306, 371–373 (1983).

4. Buzsaki, G. Hippocampal sharp waves: their origin and significance. BrainRes. 398:242–252 (1986).

5. Chrobak, J.J. & Buzsaki, G. High-frequency oscillations in the outputnetworks of the hippocampal-entorhinal axis of the freely behaving rat. J.Neurosci. 16, 3056–3066 (1996).

6. Traub, R.D. & Miles R. in Neuronal networks of the hippocampus Ch. 6(Cambridge Univ. Press, 1991).

7. Traub, R.D., Knowles, W.D., Miles, R. & Wong, R.K.S. Synchronizedafterdischarges in the hippocampus: simulation studies of the cellularmechanism. Neuroscience 12, 1191–1200 (1984).

8. Jefferys, J.G.R. in The Epilepsies, 4th edn (eds Laidlaw, J., Richens, A. &Chadwick, D.W.) 241–276 (Edinburgh, Churchill Livingstone, 1993).

9. Wong, R.K.S. & Prince, D.A. Afterpotential generation in hippocampalpyramidal cells. J. Neurophysiol. 45, 86–97 (1981).

10. Bush, P.C. & Sejnowski, T.J. Effects of inhibition and dendritic saturation insimulated neocortical pyramidal cells. J. Neurophysiol. 71, 2183–2193 (1994).

11. Miles, R., Wong, R.K.S. & Traub, R.D. Synchronized afterdischarges in thehippocampus: contribution of local synaptic interactions. Neuroscience 12,1179–1189 (1984).

12. Rutecki, P.A., Lebeda, F.J. & Johnston, D. Epileptiform activity in thehippocampus produced by tetraethylammonium. J. Neurophysiol. 64,1077–1088 (1990).

13. Mueller, A.L. & Dunwiddie, T.V. Anticonvulsant and proconvulsant actionsof alpha- and beta-noradrenergic agonists on epileptiform activity in rathippocampus in vitro. Epilepsia 24, 57–64 (1983).

14. Chamberlin, N.L. & Dingledine, R. Control of epileptiform burst rate by CA3hippocampal cell afterhyperpolarizations. Brain Res. 492, 337–346 (1989).

15. Mayr, O. The origins of feedback control (MIT Press, Cambridge,Massachusetts, 1970).

16. Martina, M. & Jonas, P. Functional differences in Na+ channel gatingbetween fast-spiking interneurones and principal neurones of rathippocampus. J. Physiol. 505, 593–603 (1997).

17. Scanziani, M. et al. Use-dependent increases in glutamate concentration activatepresynaptic metabotropic glutamate receptors. Nature 385, 630–634 (1997).

18. Otis, T., Zhang, S. & Trussell, L.O. Direct measurement of AMPA receptordesensitization induced by glutamatergic synaptic transmission. J. Neurosci.16, 7496–7504 (1996).

19. von Gersdorff, H., Schneggenburger, R., Weis, S. & Neher, E. Presynapticdepression at a calyx synapse: the small contribution of metabotropicglutamate receptors. J. Neurosci. 17, 8137–8146 (1997).

20. Dobrunz, L.E. & Stevens, C.F. Heterogeneity of release probability,facilitation, and depletion at central synapses. Neuron 18, 995–1008 (1997).

21. Harris, K.M. & Sultan, P. Variation in the number, location, and size ofsynaptic vesicles provides an anatomical basis for the nonuniformprobability of release at hippocampal CA1 synapses. Neuropharmacology 34,1387–1395 (1995).

22. Schikorski, T. & Stevens, C.F. Quantitative ultrastructural analysis ofhippocampal excitatory synapses. J. Neurosci. 17, 5858–5867 (1997).

23. Liu, G. & Tsien, R.W. Properties of synaptic transmission at singlehippocampal synaptic boutons. Nature 375, 404–408 (1995).

24. Stevens, C.F. & Tsujimoto, T. Estimates for the pool size of releasable quantaat a single central synapse and for the time required to fill the pool. Proc. NatlAcad. Sci. USA 92, 846–849 (1995).

25. Murthy, V.N., Sejnowski, T.J. & Stevens, C.F. Heterogenous release propertiesof visualized individual hippocampal synapses. Neuron 18, 599–612 (1997).

26. Madison, D.V. & Nicoll, R.A. Noradrenaline blocks accommodation ofpyramidal cell discharge in the hippocampus. Nature 299, 636–638 (1986).

27. Stevens, C.F. Quantal release of neurotransmitter and long-termpotentiation. Neuron 10(suppl), 55–63 (1996).

28. Patneau D.K., Vyklicky L. Jr & Mayer M.L. Hippocampal neurons exhibitcyclothiazide-sensitive rapidly desensitizing responses to kainate. J Neurosci.13, 3496–3509 (1993).

29. Chamberlin, N.L., Traub, R.D. & Dingledine R. Role of EPSPs in initiation ofspontaneous synchronized burst firing in rat hippocampal neurons bathed inhigh potassium. J. Neurophysiol. 64, 1000–1008 (1990).

30. Fatt, P. & Katz, B. Spontaneous subthreshold activity at motor nerve endings.J. Physiol. 117, 109–128 (1952).

31. Manabe T., Renner, P. & Nicoll, R.A. Postsynaptic contribution to long-termpotentiation revealed by the analysis of miniature synaptic currents. Nature355, 50–55 (1992).

32. Rosenmund, C. & Stevens, C.F. Definition of the readily releasable pool ofvesicles at hippocampal synapses. Neuron 16, 1197–1207 (1996).

33. Capogna, M., Gähwiler, B.H. & Thompson, S.M. Presynaptic inhibition ofcalcium-dependent and independent release elicited with ionmycin,gadolinium, and α-latrotoxin in the hippocampus. J. Neurophysiol. 75,2017–2028 (1996).

34. Dunwiddie, T.V. Endogenously released adenosine regulates excitability inthe in vitro hippocampus. Epilepsia 21, 541–548 (1980).

35. Swartzwelder, H.S., Lewis, D.V., Anderson, W.W. & Wilson, W.A. Seizure-likeevents in brain slices: suppression by interictal activity. Brain Res. 410,362–366 (1987).

36. Thompson, S.M. & Gähwiler, B.H. Activity-dependent disinhibition. II.effects of extracellular potassium, furosemide, and membrane potential onECl in hippocampal CA3 neurons. J. Neurophysiol. 61, 512–523 (1989).

37. Anderson, C.M. & Jefferys, J.G.R. Ventral CA3 exhibits a lower threshold todisinhibition than dorsal hippocampus. Epilepsia 38(Supp 8), 16 (1997).

38. Korn, S.J., Giacchino, J.L., Chamberlin, N.L. & Dingledine, R. Epileptiformburst activity induced by potassium in the hippocampus and its regulation byGABA-mediated inhibition. J. Neurophysiol. 57, 325–340 (1987).

39. Scharfman, H.E. EPSPs of dentate gyrus granule cells during epileptiformbursts of dentate hilar “mossy” cells and area CA3 pyramidal cells indisinhibited rat hippocampal slices. J. Neurosci. 14, 6041–6057 (1994).

40. Whittington, M.A. & Jefferys, J.G.R. Epileptic activity outlasts disinhibitionafter intrahippocampal tetanus toxin in the rat. J. Physiol. 481, 593–604(1994).

41. Tanaka, K. et al. Epilepsy and exacerbation of brain injury in mice lacking theglutamate transporter GLT-1. Science 276, 1699–1702 (1997).

42. Barbarosie, M. & Avoli, M. CA3-driven hippocampal-entorhinal loopcontrols rather than sustains in vitro limbic seizures. J. Neurosci. 17,9308–9314 (1997).

43. Stasheff, S.F., Anderson, W.W., Clark, S. & Wilson, W.A. NMDA antagonistsdifferentiate epileptogenesis from seizure expression in an in vitro model.Science 245, 648–651 (1989).

44. Rafiq, A., DeLorenzo, R.J. & Coulter, D.A. Generation and propagation ofepileptiform discharges in a combined entorhinal cortex/hippocampal slice.J. Neurophysiol. 70, 1962–1974 (1993).

45. Traub, R.D., Borck, C., Colling, S.B. & Jefferys J.G. On the structure of ictalevents in vitro. Epilepsia 37, 879–891 (1996).

46. Taylor, G.W., Merlin, L.R. & Wong, R.K. Synchronized oscillations inhippocampal CA3 neurons induced by metabotropic glutamate receptoractivation. J. Neurosci. 15, 8039–8052 (1995).

47. Traub R.D., Miles, R. & Jefferys, J.G. Synaptic and intrinsic conductancesshape picrotoxin-induced synchronized after-discharges in the guinea-pighippocampal slice. J Physiol. (Lond.) 461, 525–547 (1993).

48. Ylinen, A. et al. Intracellular correlates of hippocampal theta rhythm inidentified pyramidal cells, granule cells, and basket cells. Hippocampus 5,78–90 (1995).

49. Kim, U., Sanchez-Vives, M.V. & McCormick, D.A. Functional dynamics ofGABAergic inhibition in the thalamus. Science 278, 130–134 (1997).

50. Staley, K.J., Soldo, B. & Proctor W. Ionic mechanisms of neuronal excitationby inhibitory GABAA receptors. Science 269, 977–981 (1995).

articles


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Cortical neurons in the waking brain fire highly irregular spiketrains that have more in common with the ticking of a Geigercounter than of a clock. What is the source of this irregular fir-ing? Softky and Koch1 noted a theoretical conundrum posed bythe irregularity of cortical neurons firing at a constant averagerate in vivo. If each cortical neuron were providing independentinput to the other similar cortical neurons it contacts, then theinput to any neuron would just be a shower of statistically inde-pendent excitatory postsynaptic potentials (EPSPs) with a con-stant mean rate. Yet such an input, when tested on a theoreticalmodel of a cortical neuron, gave rise to a firing variability muchless than that observed in vivo. What, then, is the source of theunexpectedly large variability in spike timing that characterizesin vivo firing behavior? Is the high variability due to some sub-tle interplay between the noise resulting from randomly timedsynaptic inputs and the nonlinear spike-generating mechanism,or instead due to unexpectedly large fluctuations in the synapticinput itself? Although most earlier workers have argued that thelarge variability arises from the interaction between this noisysynaptic input and the spike-generating mechanism, we will con-clude that large fluctuations in the synaptic drive, such as mightarise from the synchronous arrival of inputs from many neu-rons, are necessary to account for the high in vivo variability.

The neuronal spike generator converts input current intooutput spike trains with high fidelity2,3. Irregular firing must,then, reflect fluctuations in the currents that drive the spike gen-erator, rather than some intrinsic noise in the spike-generatingmechanism itself4. Most previous attempts to identify the sourceof irregular firing have focused on the details of the spike-gen-erating mechanism and on the importance of the inhibitorysynaptic noise5–7 (Tsodyks et al., Soc. Neurosci. Abstr., 20, 1527,1994). These workers have concluded that unstructured synap-tic noise processed by a model spike generator can produce theunexpected variability in spike timing. However, most of theseproposals have relied on theoretical models of the neuronal spikegenerator (but see ref. 8), and thus one may question the extentto which such models can be relied upon to provide a sufficientlyrealistic representation of spike generation.

We therefore set out to determine experimentally whetherthe high variability observed in vivo could arise from thesuperposition of independent excitatory and inhibitory inputs.Because reconciliation of the in vivo variability with the synap-tic drive depends critically on the details of the spike genera-tion mechanism, we have used a direct approach rather thanrely on theoretical models of the spike generator. We haveinjected ‘synthetic’ synaptic currents through a somatic elec-trode to drive neurons in neocortical slices to fire. With thismethod, we can assess the output variability in response toany input ensemble. We have complemented this approach bytesting the response to miniature excitatory postsynaptic cur-rents (EPSCs) whose rate of release was elevated at manysynapses independently through the local application ofhypertonic solution.

We find that when neocortical neurons are driven by a pop-ulation of independent inputs, the spike variability is consis-tently lower than that observed in vivo. We thus cannot confirmthe earlier conclusions that unstructured synaptic noise inter-acting with the spike-generating mechanism can account forthe unexpectedly large variability of neuronal activity in theneocortex. However, when a population of transiently syn-chronous inputs is added to the background of independentinputs, the observed firing variability is within the rangeobserved in vivo. The high variability observed in vivo is there-fore inconsistent with the activity of independent excitatoryand inhibitory inputs, but could arise from large rapid fluctu-ations in the synaptic drive, such as would result from the near-ly synchronous firing of subpopulations of afferents.

ResultsThe results are organized as follows. First we show that, inresponse to a synthetic synaptic input corresponding to anensemble of purely excitatory inputs firing at a constant rate,the coefficient of variation (CV) of the interspike interval dis-tribution is well below that observed in vivo. Next we showthat CV increases but remains below the in vivo level whenthe input consists of a steady mix of excitatory and inhibitory

Input synchrony and the irregular firing of cortical neurons

Charles F. Stevens and Anthony M. Zador

Howard Hughes Medical Institute and Sloan Center for Theoretical Neurobiology, Salk Institute for Biological Studies, La Jolla, California 92037, USA

Correspondence should be addressed to A.Z. ([email protected])

Cortical neurons in the waking brain fire highly irregular, seemingly random, spike trains in responseto constant sensory stimulation, whereas in vitro they fire regularly in response to constant currentinjection. To test whether, as has been suggested, this high in vivo variability could be due to thepostsynaptic currents generated by independent synaptic inputs, we injected synthetic synaptic cur-rent into neocortical neurons in brain slices. We report that independent inputs cannot account forthis high variability, but this variability can be explained by a simple alternative model of thesynaptic drive in which inputs arrive synchronously. Our results suggest that synchrony may beimportant in the neural code by providing a means for encoding signals with high temporal fidelityover a population of neurons.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

inputs. We then repeat the analysis of the response variabilityfor a complementary measure, the Fano factor. We then con-firm that the spiking variability in response to miniatureEPSCs evoked by hypertonic solution is still below the in vivolevels. Finally, we demonstrate that input synchrony can yieldin vivo levels of variability.

PURELY EXCITATORY INPUT

Cortical neurons in vivo are driven to fire by a shower of EPSCs.An estimate of the size of the unitary EPSCs that comprise thisshower was obtained by recording spontaneous EPSCs frompyramidal neurons in rat neocortical slices under voltage-clampconditions (Fig. 1a). As previously observed9–11, spontaneousEPSCs varied in size, with a mean amplitude of 6.4 ± 3.4 pA

(mean ± standard devia-tion) and a skewed distri-bution (Fig. 1b). Underslice recording conditions,this spontaneous releaserate was too low to causefluctuations of more thanabout a millivolt (Fig. 1c),and was never sufficient tocause spontaneous spiking.

We therefore generatedsynthetic synaptic currentsto test whether such EPSCs,arriving at a sufficientlyhigh rate, could account forthe variability observed invivo. In order to simulatethis drive in a neocorticalslice, synthetic synaptic cur-rents were generated onlineand injected through asomatic patch electrode.Chemical synaptic inputswere pharmacologicallysilenced, so that all the drivewas supplied by the elec-trode. The synthetic cur-rents were constructed tomimic the current thatwould be generated at the

soma by a population of excitatory inputs, each firing indepen-dently according to a Poisson process with a constant mean rate.These currents have the advantage that the synthetic synaptic noisecan be created with known statistical structure. The unitary eventused to construct the synthetic synaptic inputs, 30 pA, was chosento be at the extreme high end of the observed range. Because thesum of independent Poisson processes is itself a Poisson process,the input statistics were determined solely by the net rate at whichexcitatory inputs showered onto the soma. We made no explicitassumption about the number of synapses or the rate of neuro-transmitter release from each synapse, so that, for example, 100synapses each generating 20 EPSCs per second would be indis-tinguishable from 200 synapses generating 10 EPSCs per second.

Figure 2a shows the results of a typical experiment in which

Fig. 2. Fluctuating currents affect thefine structure of the spike train butnot the mean rate. (a) Sample tracesshow typical responses of a layer 2/3cortical neuron to a constant current(left), and a fluctuating current (right)consisting of the sum of independentPoisson EPSCs firing at a populationrate of 2.4 per ms. Calibration, 200ms, 10 mV, 0.3 nA. (b) The averagenumber of spikes in a one secondtrial is plotted as a function of themean input for the constant (o) and fluctuating (*) currents. Except at verylow firing rates, the mean spike rate depends only on the mean of the inputcurrent. (c) An example of a spike train recorded extracellularly from area MTof an alert macaque monkey in response to a constant-velocity visual stimulus(for details, see ref. 16).

Fig. 1. Spontaneous synaptic event recorded in alayer 2/3 cortical neuron. (a) A short recordshowing EPSCs recorded at the soma under volt-age clamp at a holding potential of -60 mV, in thepresence of tetrodotoxin (TTX). Calibration, 40ms, 30 pA. (b) The distribution of spontaneousminiature EPSCs in this neuron. Inset, a simulatedscaled mEPSC is superimposed on a typicalmEPSC from (a). (c) A short record from thesame neuron showing EPSPs. Calibration, 40 ms,0.5 mV.

0 5 10 15 20 250

0.1

0.2

0.3

0.4

0.5

Amplitude (pA)

Prob

abili

tyP

rob

abili

ty

Amplitude (pA)

a b

c

0 0.05 0.1 0.15 0.2 0.250

5

10

15

20

25

Input current (nA)

Firin

g ra

te (

Hz)

a b

Firi

ng

rat

e (H

z)

Input current (nA)

Time (ms)

c


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Fig. 3. Variability in response to mixed excitatoryand inhibitory input is less than in vivo. (a) Responseto mixed excitatory and inhibitory input. Typicalresponses of a layer 2/3 cortical neuron to a fluctuat-ing current (Ri = 0.5) consisting of the sum of a mix ofindependent Poisson EPSCs (4.4 per ms) and IPSCs(2.2 per ms). Calibration, 200 ms, 10 mV, 0.3 nA. (b)Dependence of CV, the coefficient of variation of theinterspike interval distribution, on the ratio Ri of inhi-bition to excitation. The CV increases with Ri, butremains below the in vivo level even at the largestvalue of Ri tested. The dotted line indicates the CV ofa Poisson process. The error bars indicate the standard error. (c) Dependence of the Fano factor F (thevariance divided by the mean of the spike count) on the ratio Ri. The solid line shows the actual Fano factor,whereas the dashed line shows the Fano factor predicted from F = CV2. Even for high values of Ri, the Fanofactor of the response to synthetic synaptic currents remains far below that observed in vivo. The dotted lineindicates the Fano factor of a Poisson process. The error bars indicate the standard error.

a neuron from layer 2/3 of rat neocortex was driven to firewith synthetic synaptic currents. In this example, the inputrate of 2.4 EPSCs per millisecond yielded a firing rate of 21Hz, indicating that about 115 EPSCs were required for eachaction potential. For comparison, the response to an injectionof constant current is also shown. The striking differencebetween the responses in Fig. 2a is in the fine temporal struc-ture of the spike trains. In response to a constant input, spikesarrive at regular intervals (which become longer during theone-second stimulus as a result of spike adaptation), where-as in response to the synthetic synaptic current, the interspikeintervals are much more irregular.

Although the fine temporal structure of the spike trainsgenerated by the synthetic synaptic current was very differentfrom that generated by the constant stimulus, the mean outputfiring rate in response to this ensemble depended only on themean input. The f–I curves (the firing frequency f, averagedover the one second stimulus, as a function of the input cur-rent I) for the two stimulus ensembles are fairly linear, show-ing some slight saturation at high firing frequencies, and are inclose agreement over the range of currents tested (Fig. 2b).

Is the irregularity of the spike train in Fig. 2a as high asthat seen in vivo? One approach to quantifying spike irregu-larity is based on the coefficient of variation (CV) of the dis-tribution of interspike intervals (ISIs). The CV is defined asthe standard deviation σisi divided by the mean µisi of the ISIdistribution, CV = σisi/µisi. For a Poisson process, CV is one12,

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

Ratio of inhibitory to excitatory drive

C.V

. of I

SI D

istr

ibut

ion

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1


Fano

fact

or

a b

C.V

. of

ISI d

istr

ibu

tio

n


c

Fan

o f

acto

r


but the converse is not true: one cannot conclude that aprocess is Poisson simply because CV is one.

The CV of spike trains from cortical neurons recorded invivo are generally near or above unity1,13–15. The CV dependson a number of factors, including the firing rate and the degreeof adaptation during a response. In order to provide a specificbenchmark against which to compare the present in vitroresults, we computed the CV of responses of neurons in themiddle temporal (MT) cortex of alert macaque monkeys drivenwith constant-motion stimuli16. An example of one suchresponse, emphasizing the markedly irregular activity of neo-cortical sensory neurons in vivo, is depicted in Fig. 2c. To min-imize the effect of adaptation within a trial, only responsesafter the first 500 ms were considered, and to minimize anyeffect of slow trial-to-trial drift, the CV from each trial wascomputed separately and then these CVs were averaged. Themean CV was 1.1 ± 0.16 (range, 0.91 to 1.4; n = 10) for neu-rons with a mean firing rate of µ = 17 ± 6.7 Hz (range, 11 to33 Hz). Based on these data, it is reasonable to consider a CV of0.8 as a lower limit for the in vivo range at the firing rates con-sidered here.

The spike trains from the neuron shown in Fig. 2 had a CVof 0.28. Similar results were obtained in all other neurons forwhich similar stimuli were tested (CV = 0.29 ± 0.09, µ = 21 ±2.4 spikes per s; n = 9). These values are much lower than theCV value of 1.1 observed in MT cortical neurons in vivo. Inother experiments, we examined a range of synthetic synaptic


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


currents and found that, as expected, the CV was smaller forcurrents constructed from smaller synthetic EPSCs. For thisreason, we chose to test an EPSC amplitude at the upper limit ofthe spontaneous responses recorded in slice (30 pA), becauseany smaller EPSC would have yielded a lower CV even moreinconsistent with the in vivo range. We also examined higherfiring rates and found that, as expected, the CV decreased asthe firing rate increased. The CV reported here, then, can beconsidered an upper limit on the CV likely to be generated bythe random superposition of independent inputs. Thus a pure-ly excitatory steady drive from independent inputs does notaccount for the observed irregularity of cortical spike trains.

MIXED IPSPS AND EPSPS

The synaptic drive to cortical neurons in vivo contains a sub-stantial inhibitory component mediated by GABA recep-tors17,18. It has been suggested that the irregularity observedin vivo might arise in part from added fluctuations in the drivecaused by inhibitory inputs6,7. Inhibition increases the out-put variability by increasing the variance in the input driveassociated with a given mean. To test whether the inclusion ofinhibitory inputs could account for the irregularity of spiketrains in vivo, we synthesized synaptic currents consisting ofa mixture of excitatory and inhibitory currents. Figure 3ashows the response of a neuron to such an input. In this exam-ple, the total inhibitory current Ri was half the excitatory cur-rent (Ri = 0.5); the excitatory drive (the average rate of EPSCoccurrence) was increased in order to maintain a high firingrate. As expected, the addition of this inhibitory componentincreased the irregularity, to CV of 0.44. Similar results were

obtained in all other neurons for which similar stimuli weretested (CV = 0.43 ± 0.09, µ = 20.7 ± 2.4 spikes per s; n = 9).Thus although this input does increase spike irregularity, theresultant values are still substantially below the CV value of1.1 observed in vivo.

The fraction of the total input to a cortical neuron pro-vided by inhibition has not been measured in vivo. Could ahigher inhibitory-to-excitatory ratio Ri boost the CV into thein vivo range? The dependence of the CV on Ri is shown inFig. 3b. As expected, the CV increased with increasing Ri, butremained well below unity for the highest values tested (CV= 0.72 ± 0.07, µ = 19.2 ± 1.2 spikes per s; n = 5 for Ri = 0.9).Thus at the highest levels of Ri tested, the CV approached, butremained below, the values observed in vivo. Because ratiosof inhibition to excitation are unlikely to exceed 0.9 (see ref.17), we conclude that these experiments represent an approx-imate upper limit on the variability that can arise from synap-tic noise from uncorrelated sources, and that such noise thuscannot account for the high CV observed in vivo.

TRIAL-TO-TRIAL VARIABILITY

The CV is one of several standard measures of firing variabil-ity. A high CV may reflect irregularity in the fine structure ofthe spike train, but it may arise from changes arising on a timescale longer than a typical interspike interval, such as thoseinduced by spike adaptation. Indeed, the response to the con-stant stimulus shown in Fig. 2 has a CV of 0.43, which is aboutthe same as that for the mixed excitatory/inhibitory input withRi of 0.5. The response to the purely excitatory syntheticsynaptic input also shows strong adaptation.

We therefore also considered a second measure of vari-ability, the Fano factor. The Fano factor F is defined as thevariance divided by the mean of the spike count N, F = σN

2/µN

where the spike count N is the number of spikes generated ona (typically one-second) particular trial. The Fano factor isinsensitive to slow changes such as adaptation that occur with-in a single trial, as long as they occur consistently from onetrial to the next. For any Poisson process (including bothhomogeneous and rate-modulated Poisson processes), theFano factor is exactly one; spike trains from cortical neuronsin vivo are consistently found to have a Fano factor near orabove one16,19. For example, neurons from the middle tem-poral (MT) area of alert monkeys were reported to have aFano factor of 1.3 (ref. 16).

In certain limiting cases, the Fano factor and the CV arerelated by F = CV 2. The main requirement is that the spiketrain be a stationary renewal process12, that is, a process inwhich each interspike interval (ISI) is statistically independentof every other ISI. Although there are many ways in which aspike train could deviate from a renewal process, in the pre-sent context deviations are most likely to occur if successive(or neighboring) ISIs are statistically dependent (for example,neurons show some degree of bursting or short ISIs tended tobe followed by long ISIs), or if the spike train adapts, forinstance slows down, during the one-second stimulus. Becausethe response both to the constant input and to the syntheticsynaptic drive can show considerable adaptation, at least thepossibility of adaptation during the stimulus must be consid-ered. Thus the Fano factor can provide additional and inde-pendent information about how well the responses to thesynthetic currents mimic the variability observed in vivo.

The Fano factor of the response to the constant stimulus wasonly 0.02, compared with the Fano factor (F = 0.432 = 0.18) pre-

articles

Fig. 4. Responses elicited by hypertonic solution evoked increasesin the rate of miniature EPSC release. (a) A regular train of actionpotentials driven by miniature EPSCs elicited by a one-second appli-cation of hypertonic solution from a puffer pipette positioned about40 µm from the soma. The variability of these responses (CV = 0.26,Fano factor F = 0.05) was close to that elicited by comparable syn-thetic synaptic currents. (b) The synaptic current elicited by hyper-tonic solution in the same neuron. Calibration, 300 ms, 20 mV(top), 100 pA (bottom). The bath solution contained standardRinger’s with 100 µM picrotoxin, and 30 µM cadmium added to pre-vent recurrent activity. The hypertonic solution consisted of thebath solution plus 0.5 mM sucrose.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


dicted from the renewal assumption (where F rather than F is usedto denote the Fano factor predicted from the CV). The large dis-crepancy indicates that for a constant stimulus the renewal assump-tion is not satisfied: marked adaptation within a single trial leads toa large CV that is not reflected in the trial-to-trial measure F. TheFano factor for the purely excitatory input was only slightly high-er than for the constant current (F = 0.06 ± 0.03; n = 9), and closeto that predicted from the renewal assumption (F = 0.292 = 0.08).However, for the highest inhibitory/excitatory ratios tested (Ri =0.9), the observed value (F = 0.29 ± 0.12; n = 5) was substantiallyless than the predicted value (F = 0.722 = 0.52) and well below thein vivo levels of one or above (Fig. 3c). The finding that the in vitroFano factors, even for responses driven by inputs with a stronginhibitory component, were consistently lower than those observedin vivo reinforces the previous conclusion that independent inputscannot account for the high variability observed in vivo.ASYNCHRONOUS EPSC-EVOKED ACTION POTENTIALS

The results presented so far have relied on current injectedthrough a somatic electrode as a model for the synaptic cur-rents that drive neurons in vivo. Injected current may not,however, mimic synaptic current in all respects. For example,injected current implicitly treats each synapse as a currentsource, whereas in fact a synapse is more properly consideredas a conductance in series with the driving potential at thesubsynaptic membrane. (The ‘dynamic clamp’20 can only par-tially compensate for this error, as most excitatory synapsesonto cortical neurons are onto dendrites that are electrotoni-cally remote from the soma and whose subsynaptic voltagecannot be controlled by a somatic electrode.) We thereforedeveloped a more direct approach to study the spiking vari-ability in response to synaptic stimulation.

Because conventional extracellular stimulation triggers alarge increase in the rate of transmitter release from all synap-

tic terminals simultane-ously, it is not a goodmodel for the asynchro-nous and independentbarrage of synaptic stim-ulation under considera-tion here. By contrast,local perfusion withhypertonic solution ele-vates the rate of minia-ture EPSC release frommany terminals indepen-dently and thereby gen-erates a slowly varyingaverage current. Themechanism of thissucrose-evoked trans-mitter release is indepen-dent of presynapticaction potentials and cal-cium influx11,21,22. Theminiature EPSCs re-leased under these con-ditions therefore providea much closer approxi-mation of the drive to acortical neuron expectedif all inputs were inde-pendent.

Figure 4 shows theresponse of a layer 2/3 neuron to a hypertonic solution-evokedincrease in the miniature EPSC rate. The spike train was veryregular, with CV of 0.26 and Fano factor of 0.05. These resultsare in very close agreement with values obtained for the pure-ly synthetic current (CV = 0.29 and F = 0.06; see Fig. 2). Thusthe hypertonic solution-evoked responses showed much lessvariability than observed in vivo.

We were unable to elicit spiking in the absence of blockersof (inhibitory) GABAergic inputs (n = 4), presumably becausethe hypertonic solution did not activate a sufficiently highratio of excitatory to inhibitory input. This may be due in partto the positioning of the puffer pipette near the soma, wherethe density of GABAergic terminals may be higher. We weretherefore unable to use hypertonic solution to validate thevariability measurements for the synthetic inputs that mixedexcitation and inhibition. Nevertheless, the close agreementbetween the synthetic and hypertonic solution-evoked vari-ability for the purely excitatory input suggests that currentsinjected at the soma can adequately mimic synaptic inputs.

INPUT SYNCHRONY YIELDS IN VIVO FIRING VARIABILITY

Because a steady input consisting of independent EPSCs andIPSCs fails to drive firing that is as irregular as is observed invivo, we tested the possibility that some degree of input syn-chrony might be required. We considered a simple model inwhich the input consisted primarily of occasional large brief(30–50 ms) excitatory events, randomly distributed in time(average inter-event interval, 100 ms), each consisting of thenearly synchronous (30–50 ms) arrival of on average about100 to 200 EPSCs, yielding peak currents of 1 to 2 nA (Meth-ods). Evidence from a variety of sources suggests the impor-tance of correlated firing in the neocortex9,16,23–30.

Figure 5a shows a typical experiment in which the syn-

articles

Fig. 5. Input synchrony yields in vivo variability. (a)Response to an input drive consisting primarily of occa-sional large brief (30–50 ms) excitatory events, ran-domly distributed in time, each consisting of the nearlysynchronous arrival of on average 100–200 EPSCs,yielding transient peak currents as high as 1 nA ormore. Calibration, 200 ms, 10 mV (top), 0.1 nA (bot-tom). (b) Summary of CV and Fano factor. Left, CV for

purely excitatory (E), mixed excitatory/inhibitory (E/I), and synchronous (sync), inputs. Right, same statisticsfor the Fano factor. For both graphs, the error bars indicate standard errors. The dotted lines indicate thevalue of unity expected for both CV and Fano factor for a perfect Poisson process and are near the values typ-ically observed from cortical spike trains in vivo.

E E/I sync0

0.2

0.4

0.6

0.8

1

1.2

C.V.

E E/I sync0

0.2

0.4

0.6

0.8

1

1.2

F.F.a b C.V. F.F.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


chronous input was synthesized according to this simplemodel. The irregularity of the resulting spike train was muchhigher, with both the CV and the Fano factor in the in vivorange in all neurons tested (CV = 1.3 ± 0.11; F = 0.94 ± 0.2,µ = 19.8 ± 1.0 spikes per s; n = 5). The data (Fig. 5b) demon-strate that input correlations could account for the high degreeof variability observed in vivo.

DiscussionThe observations1 on the paradox posed by the irregularity ofin vivo cortical spike trains have sparked considerable debateon the underlying mechanism5–7,32–34. Although Softky andKoch1 raised the possibility that the unexpectedly large vari-ability might arise from input synchrony, later workers haveargued that statistically stationary synaptic noise, interactingwith the spike-encoding mechanism, can account for the largevariability. All of these previous studies have, however, reliedon simulations of the spike encoder, and the results are sen-sitive to assumptions about the encoder properties.

Our results do not rely on any theoretical model of thespike encoder. Rather, we have used the actual spike encoder ofneurons in vitro to test directly whether the high variabilityobserved in vivo could arise from a constant shower of statis-tically independent synaptic inputs. We conclude that it couldnot, and propose that the high in-vivo variability arises fromcorrelations among the inputs28,29. We illustrate this by match-ing the in-vivo variability with a simple model of the inputdrive, in which occasional large brief synchronous events aresuperimposed on a tonic background.

Our conclusions rest on two main assumptions. First, wehave assumed that somatic current injection represents a rea-sonable model of synaptic drive. Although the close agree-ment between the somatic current injection and hypertonicsolution-evoked miniature EPSC results (Fig. 4) supports thisnotion, it is possible that the agreement would not be as goodwhen the inhibitory drive was very strong. Second, we haveassumed that the in-vitro biophysical properties of neocorticalneurons can be used to make inferences about their in-vivobehavior. This supposes, for example, that the spontaneousminiature EPSCs recorded in slices accurately reflect the in-vivo synaptic amplitude distribution. In addition, theinput–output transformation might be different in vivo,because of neuromodulatory influences35 for example, or theactivation of large nonlinear conductances36.

MECHANISM UNDERLYING SPIKING VARIABILITY IN VIVO

Irregular firing in vivo might in principle arise from inputvariability or from noise in the spike-generating mechanism.Because the neuronal spike generator is very reliable, inputvariability is likely to be the primary source2,3. In spinal neu-rons, ‘synaptic noise’ (fluctuations in membrane potentialarising from a barrage of EPSPs) fully accounts for outputvariability4. In the neocortex, the in-vivo synaptic driveinferred from current clamp recordings could, when injectedinto neocortical neurons in vitro, reproduce their basic firingstatistics8. These latter experiments did not interpret theinjected currents in terms of the underlying synaptic proper-ties and correlational structure of the inputs, nor did theyestablish the properties of the drive necessary or sufficient toaccount for the high variability.

In the present study, we therefore began with the assump-tion that output variability reflects fluctuations in the inputcurrent and asked what form the synaptic noise must take to

give rise to the observed spiking variability. The conundrum isthat the fluctuations in the input drive generated by inde-pendent excitatory afferents seem too small to account for thehigh variability of in-vivo cortical firing. The initial report1

emphasized the difficulty in accounting for the high variabil-ity at very high firing rates (over 100 Hz). In our experiments,we found that even at more moderate firing rates (15–25 Hz),independent EPSCs cannot account for the variabilityobserved in vivo; at higher firing rates, independent EPSCsinduced even lower output variability.

Previously proposed resolutions fall into two main classes.First, simulations suggested that inhibitory inputs increase theinput fluctuations and thereby the output variability suffi-ciently to account for the in vivo responses5–7. Our experi-mental results, however, demonstrate that the output variabilityin response even to a drive with a very strong inhibitory com-ponent remains well below the in vivo levels, indicating thatsome alternative mechanism must be responsible.

The second class of explanation invokes the details of thespike-generating mechanism. In one proposal7, the reset volt-age following an action potential was used to match the slopeor ‘gain’ of the firing-intensity curve for the first interspikeinterval in an integrate-and-fire model. In simulations, theoutput variability of responses generated by this model wasnear that found in vivo. This model failed, however, to pre-dict the responses of cortical neurons to the synthetic synap-tic currents we used (C.S. and A.Z., unpublished results). Theinitial report1 proposed a more radical modification of thespike generator. Here it was suggested that the coincidentarrival of just a few EPSPs might activate powerful nonlineardendritic conductances that would then trigger a somaticspike. Although our experiments cannot rule out this mecha-nism (and indeed, the large events shown in Fig. 5 could inprinciple arise from the activation of dendritic nonlineari-ties), we favor the notion that the dynamics of cortical net-works, rather than the properties of single neurons, underliethe requisite large fluctuations in the input drive.

The model of synaptic drive that we have used is by nomeans unique in its ability to account for the variabilityobserved in vivo; countless other input ensembles would cer-tainly have done as well. The key requirement is that the inputcurrent contain very large fluctuations, much larger thanwould be expected if all the synapses were releasing quantaindependently and at a constant rate. These fluctuations couldarise from the brief and coordinated increase in the firing rateof a large number of excitatory, or perhaps inhibitory37,synapses. Moreover, if a small number of presynaptic neuronseach made dozens of powerful synaptic contacts onto a sin-gle postsynaptic target38, then the requisite postsynaptic fluc-tuations might arise from correlated activity in just thissmaller subpopulation of neurons. Our experiments do notaddress the specific network mechanisms that might give riseto this correlated synaptic activity. Nevertheless, the presentresults indicate that, however the input current is generated,some kind of correlation among the synaptic inputs is likelyto play a critical role in generating the high degree of vari-ability observed in vivo.

SYNCHRONY AND THE NEURAL CODE

The irregularity of cortical spike trains has led to two ratherdifferent models of how cortical circuitry operates. In the con-ventional model, this irregularity represents ‘noise’ aroundthe ‘signal’, a perturbation around a mean firing rate that is

articles


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

obtained by averaging over some period (such as one second)that is much longer than a typical ISI. In this view, the precisetiming of individual spikes conveys little information, becauseit reflects only the noisy activity of other neurons attemptingto signal their own mean rate. This model follows readily fromthe idea that the drive to cortical neurons is composed of theuncorrelated activity of its synaptic inputs; it would be hardto imagine how the precise timing of spikes could representmore than noise in this model, because adding or removingjust a few EPSPs might perturb the spike timing appreciably.

In the second model, the irregular spikes reflect modula-tion on a fine time scale of the neuron’s output rate. It is nowestablished that under some conditions, the fine temporalstructure of the spike train (the precise location of each spike)can carry information16,39,40. For example, when visual stim-uli have fine temporal structure, the timing of spikes fromneurons in MT cortex can be tightly (2–5 ms) locked to thestimulus16,41. Recordings of local field potentials suggest thatthis high fidelity of spike timing is achieved by rapid comod-ulation of the rates of the input neurons16; under these con-ditions, the input correlations encode temporal edges withinthe stimulus. However, our results suggest that even when thesensory stimulus does not have fine temporal structure, spikesin cortical neurons may nevertheless arise from large eventsin the input drive that represent the correlated activity of manyneurons. Such large events are presumably more robust tonoise than the small fluctuations that are posited to drive fir-ing in the first model. Thus even when the sensory stimulusis devoid of fine temporal structure, spikes may be encodingsomething with high temporal fidelity.

MethodsSLICE PREPARATION. Brain slices were prepared from Long-Evans rats(postnatal day 14–28) that were deeply anesthetized with metofaneand then decapitated. The skull was rapidly opened and the brainplaced in ice-cold Ringer solution. The cooled brain was glued withcyanoacrylate to the stage of a vibratome and 400-µm slices were cut.Slices were transferred to a holding chamber and incubated for at leastone hour at room temperature in a solution continuously bubbledwith a mixture of 95% O2 and 5% CO2, and then placed in a record-ing chamber.

PATCH-CLAMP RECORDING. Whole-cell, patch-clamp recordings weremade from neurons in the sensory neocortex visualized through anupright microscope equipped with infrared light and differential inter-ference optics42. Recordings were performed at 33–35°C. Slices werecontinuously perfused with a Ringer’s solution containing (in mM)NaCl 120, KCl 3.5, CaCl2 2.6, MgCl2 1.3, NaH2PO4 1.25, NaHCO3 26,glucose 10, pH 7.35. Unless otherwise indicated, recordings wereobtained in the presence of the AMPA receptor antagonist 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX; 10 µM). Recording pipettes werefilled with an internal solution consisting of (in mM) K gluconate,170; HEPES, 10; NaCl, 10; MgCl2, 2; EGTA, 1.33; CaCl2, 0.133;MgATP, 3.5; GTP, 1.0, pH 7.2 and 290–300 mOsm. Resistance to bathwas 3–5 MΩ before seal formation.

In experiments in which the rate of miniature EPSCs was elevatedby locally perfusing with hypertonic solution, the bath Ringer’s con-tained 100 µM picrotoxin (to block γ-aminobutyric acid type A recep-tor responses), 30 µM Cd+2 to block synaptic responses that couldcause recurrent activity and no CNQX. To elicit hypertonic-solution-evoked miniature EPSCs, visual guidance was used to position a 2 µmpuffer pipette (containing bath Ringer’s to which 0.5 mM sucrose hadbeen added) close to dendrites about 30–50 µm from the somaticrecording electrode. A picospritzer II was then used to apply 1–2 sec-ond pulses (3–6 psi) of the pipette solution.

Recordings were obtained using an Axopatch 200A or 200B (AxonInstruments). Most recordings were obtained in the current-clampmode, without series-resistance compensation. Traces were filteredat 2 KHz and digitized at 4 KHz. Series resistance, monitored priorto each trace, was typically in the range of 8–20 MΩ, and never morethan 30 MΩ. To compensate for any slow drift in membrane potential,prior to each trace sufficient current was injected to return the mem-brane to a value determined at the start of the experiment (between -60 and -75 mV); if the actual rest potential decreased by more than5 mV from this level, the experiment was terminated. The responseto constant current injection (an ‘F–I curve’) was also periodicallymonitored to assess washout, and if the spike rate at the maximumintensity decreased by more than 30%, the experiment was termi-nated (usually 30–60 minutes).

Forty-two regular spiking neurons43 from cortical layers 2/3 wereanalyzed. In addition, results from two layer 5 neurons were consistentwith the other results and so were included in the averaged data. Cur-rent stimulation was usually applied for one second every five or tenseconds; the long interstimulus rest period helped to reduce the influ-ence of one trial on the next. For the analysis, only neurons whose aver-age firing rate was between 15 and 25 Hz were used. At higher sustainedfiring rates, recording stability tended to degrade more quickly.

Data were acquired using a National Instruments AT-MIO-16-F-5 A/D card on a 120 MHz Pentium-based computer under the Win-dow NT (Microsoft) operating system. Software written in Labview(National Instruments) with Dynamic Data Exchange links to Mat-lab (Mathworks) allowed convenient online synthesis and injectionof arbitrary synthetic current waveforms.

SYNTHETIC SYNAPTIC CURRENTS. Neurons were driven to fire throughcurrents injected through a somatic patch electrode. Several models ofcorrelated synaptic inputs were tested. In the independent EPSC model,the synaptic drive was given by Ie = P(re) * Iampa, where P(re) is asequence of independent Poisson points arriving at a rate re, Iampa =We e–t/τe is the waveform of the basic EPSC (We = 30 pA, τe = 3 ms),and * indicates convolution. The firing rate under these conditions isdetermined by a single free parameter, the rate of synaptic impulses re.

In the mixed model, an inhibitory drive given by Ii = P(ri) * Igabawas added to the excitatory drive, where as above P(ri) is a sequenceof independent Poisson points arriving at a rate ri, Igaba = Wi e–t/τe isthe waveform of the basic IPSC (Wi = 30 pA, τ i = 6 ms), and * indi-cates convolution. The total input under these conditions was given byIb = Ii + Ie. In this scenario, the firing rate depended both on ri and re.The value of ri was chosen to keep the ratio Ri of total inhibitory andexcitatory currents fixed, Ri = κ ri/ re where κ = (We τe)/(Wm τm)takes into account differences in the synaptic waveforms.

Finally, we examined the effect of synchronous excitatory inputs,Is = C(t) * Iampa, where C(t) was the stochastic function that describesthe occurrence of synchronous events. C(t) consisted of 30–50-msperiods of elevated input activity; these periods occurred in a ran-dom (Poisson) fashion at a rate of 10 Hz. During each 30–50-ms peri-od of elevated activity, about 200 EPSCs were distributed in 2–8shorter events. The synchronous input Is was added to Ie and Ii.

AcknowledgementsWe thank G. Buracas for the MT data and L. Dobrunz, E. Huang, K. Miller,

P. Latham, T. Troyer and K. Zhang for comments. This work was supported

by the Howard Hughes Medical Institute (C.F.S.), National Institutes of

Health grant NS 12961 (C.F.S.) and the Sloan Center for Theoretical

Neurobiology (A.M.Z.).


1. Softky, W. & Koch, C. The highly irregular firing of cortical cells isinconsistent with temporal integration of random epsps. J. Neurosci. 13,334–350 (1993).

2. Mainen, Z.F. & Sejnowski, T.J. Reliability of spike timing in neocorticalneurons. Science 268, 1503–1505 (1995).


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


23. Abeles, M. Corticonics—Neural Circuits of the Cerebral Cortex. (CambridgeUniv. Press, 1991).

24. Alonso, J.M., Usrey, W.M. & Reid, R.C. Precisely correlated firing in cells ofthe lateral geniculate nucleus. Nature 383, 815–819 (1996).

25. deCharms, R.C. & Merzenich, M.M. Primary cortical representation ofsounds by the coordination of action-potential timing. Nature 381, 610–613(1996).

26. Nelson, J.I., Salin, P.A., Munk, M.H., Arzi, M. & Bullier, J. Spatial andtemporal coherence in cortico-cortical connections: a cross-correlation studyin areas 17 and 18 in the cat. Visual Neurosci. 9, 21–37 (1992).

27. Singer, W. & Gray, C.M. Visual feature integration and the temporalcorrelation hypothesis. Ann. Rev. Neurosci. 18, 555–586 (1995).

28. Usher, M., Stemmler, M., Koch, C. & Olami, Z. Network amplification oflocal fluctuations causes high spike rate variability, fractal firing patterns andoscillatory local field potentials. Neural Computation 6, 795–836 (1994).

29. van Vreeswijk, C. & Sompolinsky, H. Chaos in neuronal networks withbalanced excitatory and inhibitory activity. Science 274, 1724–1726 (1996).

30. Whittington, M.A., Traub, R.D., Faulkner, H.J., Stanford, I.M. & Jefferys, J.G.Recurrent excitatory postsynaptic potentials induced by synchronized fastcortical oscillations. Proc. Natl Acad. Sci. USA 94, 12198–12203 (1997).

31. Zohary, E., Shadlen, M.N. & Newsome, W.T. Correlated neuronal dischargerate and its implications for psychophysical performance. Nature 370,140–143 (1994).

32. Shadlen, M.N. & Newsome, W.T. Noise, neural codes and corticalorganization. Curr. Opin. Neurobiol. 4, 569–579 (1994).

33. Softky, W.R. Simple codes versus efficient codes. Curr. Opin. Neurobiol. 5,239–247 (1995).

34. Reich, D.S., Victor, J.D., Knight, B.W., Ozaki, T. & Kaplan, E. Responsevariability and timing precision of neuronal spike trains in vivo. J.Neurophysiol. 77, 2836–2841 (1997).

35. Tang, A.C., Bartels, A.M. & Sejnowski, T.J. Effects of cholinergic modulationon responses of neocortical neurons to fluctuating inputs. Cereb. Cortex 7,502–509 (1997).

36. Softky, W.R. Sub-millisecond coincidence detection in active dendritic trees.Neuroscience 58, 13–41 (1994).

37. Hausser, M. & Clark, B.A. Tonic synaptic inhibition modulates neuronaloutput pattern and spatiotemporal synaptic integration. Neuron 19, 665–678(1997).

38. Markram, H. A network of tufted layer 5 pyramidal neurons. Cereb. Cortex 7,523–533 (1997).

39. Bialek, W., Rieke, F., de Ruyter van Steveninck, R.R. & Warland, D. Reading aneural code. Science 252, 1854–1857 (1991).

40. Rieke, F., Warland, D., de Ruyter van Steveninck, R.R. & Bialek, W. Spikes:Exploring the Neural Code. (MIT Press, 1997).

41. Bair, W. & Koch, C. Temporal precision of spike trains in extrastriate cortexof the behaving macaque monkey. Neural Computation 8, 1184–1202 (1996).

42. Stuart, G.J., Dodt, H.U. & Sakmann, B. Patch-clamp recordings from thesoma and dendrites of neurons in brain slices using infrared videomicroscopy. Pflug. Archiv. Eur. J. Physiol. 423, 511–518 (1993).

43. McCormick, D.A., Connors, B.W., Lighthall, J.W. & Prince, D.A.Comparative electrophysiology of pyramidal and sparsely spiny stellateneurons of the neocortex. J. Neurophysiol. 54, 782–806 (1985).

articles

3. Zador, A. The impact of synaptic unreliability on the informationtransmitted by spiking neurons. J. Neurophysiol. 19, 1230–1238 (1998).

4. Calvin, W.H. & Stevens, C.F. Synaptic noise and other sources of randomnessin motoneuron interspike intervals. J. Neurophysiol. 31, 574–587 (1968).

5. Shadlen, M.N. & Newsome, W.T. Is there a signal in the noise? Curr. Opin.Neurobiol. 5, 248–250 (1995).

6. Shadlen, M.N. & Newsome, W.T. The variable discharge of cortical neurons:Implications for connectivity, computation, and information coding. J.Neurosci. 18, 3870-3896 (1998).

7. Troyer, T.W. & Miller, K.D. Physiological gain leads to high ISI variability in asimple model of a cortical regular spiking cell. Neural Computation 9,971–983 (1997).

8. Nowak, L.G., Sanchez-Vives, M.V. & McCormick, D.A. Influence of low andhigh frequency inputs on spike timing in visual cortical neurons. Cereb.Cortex 7, 487–501 (1997).

9. Berretta, B. & Jones, S.G. A comparison of spontaneous EPSCs in layer II andlayer IV-V neurons of the rat entorhinal cortex in vitro. J. Neurophysiol. 76,1089–1100 (1996).

10. Burgard, E.C. & Hablitz, J.J. NMDA receptor-mediated components ofminiature excitatory synaptic currents in developing rat neocortex. J.Neurophysiol. 70, 1841–1852 (1993).

11. Bekkers, J.M., Richerson, G.B. & Stevens, C.F. Origin of variability in quantalsize in cultured hippocampal neurons and hippocampal slices. Proc. NatlAcad. Sci. USA 87, 5359–5362 (1990).

12. Feller, W. An Introduction to Probability Theory and its Applications, vol. 2,2nd edn. (Wiley, New York, 1971).

13. Burns, B.D. & Webb, A.C. The spontaneous activity of neurones in the cat’scerebral cortex. Proc. Royal Soc. Lond. B 194, 211–223 (1976).

14. Holt, G.R. Softky, W.R., Koch, C. & Douglas, R.J. Comparison of dischargevariability in vitro and in vivo in cat visual cortex neurons. J. Neurophysiol.75, 1806–1814 (1996).

15. Noda, H. & Adey, R. Firing variability in cat association cortex during sleepand wakefulness. Brain Res. 18, 513–526 (1970).

16. Buracas, G., Zador, A., Deweese, M. & Albright, T. Efficient discrimination oftemporal patterns by motion-sensitive neurons in primate visual cortex.Neuron 20, 959–969 (1998).

17. Berman, N.J., Douglas, R.J., Martin, K.A. & Whitteridge, D. Mechanisms ofinhibition in cat visual cortex. J. Physiol. 440, 697–722 (1991).

18. Nelson, S., Toth, L., Sheth, B. & Sur, M. Orientation selectivity of cortical neuronsduring intracellular blockade of inhibition. Science 265, 774–777 (1994).

19. Gershon, E.D., Wiener, M.C., Latham, P.E. & Richmond, B.J. Codingstrategies in monkey V1 and inferior temporal cortices. J. Neurophysiol. 79,1135–1144 (1998).

20. Sharp, A.A., O’Neil, M.B., Abbott, L.F. & Marder, E. Dynamic clamp:computer-generated conductances in real neurons. J. Neurophysiol. 3,992–995 (1993).

21. Hubbard, J.I., Jones, S.F. & Landau, E.M. An examination of the effects ofosmotic pressure changes upon transmitter release from mammalian motornerve terminals. J. Physiol. 197, 639–657 (1968).

22. Stevens, C.F. & Tsujimoto, T. Estimates for the pool size of releasable quantaat a single central synapse and for the time required to refill the pool. Proc.Natl Acad. Sci. USA 92, 846–849 (1995).


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Lamina I of the superficial dorsal horn is an integral componentof the central representation of pain and temperature sensitivi-ties in mammals1–3. It receives direct input from Aδ and C pri-mary afferent fibers, including specific nociceptors andthermoreceptors, and it is the main source of output from thesuperficial dorsal horn. In contrast to the deep dorsal horn withits modality-ambiguous ‘wide dynamic range’ (WDR) nocicep-tive neurons, which have been studied extensively, lamina I con-tains unique concentrations of modality-selective nociceptiveand thermoreceptive neurons. In primates, the axons of theseneurons project in the lateral spinothalamic tract (which is crit-ical for pain and temperature sensation) to a dedicated nocicep-tive- and thermoreceptive-specific relay nucleus in posterolateralthalamus4 and to other sites3 that together provide input to thecortical regions that are activated by painful and thermal stim-uli in humans5,6.

These characteristics support the concept that pain is sub-served by distinct sets of neurons both peripherally and cen-trally. This view is not shared by all investigators in the field ofpain research7, in part because numerous intracellular labelingstudies did not uncover distinct types of neurons in the super-ficial dorsal horn8–16. This contrasts with the visual and motorsystems, where distinct structure/function correlations havebeen identified17,18, and accumulating evidence has suggestedthat this issue deserves re-examination. Physiological studiesof lamina I spinothalamic cells have documented three majorclasses of modality-selective neurons in cat and monkey: noci-ceptive-specific (NS) cells, thermoreceptive-specific (COLD)cells, and polymodal nociceptive (HPC) cells (and also WDRcells in monkey)19–25. These classes have different axonal con-duction velocities and thalamic terminations, as shown withantidromic mapping (refs 21, 24, 26; Dostrovsky, J.O. & A.D.C.,Soc. Neurosci. Abstr., 646.9, 1993), features indicating anatom-

ical differences that could be reflected in the morphology oftheir cell bodies in lamina I. In horizontal sections, lamina Ineurons in rat, cat and monkey can be anatomically catego-rized almost comprehensively into three basic, albeit variegat-ed, cell types: fusiform, pyramidal, and multipolar27–30. Thefusiform cells appear to have unmyelinated axons, whereaspyramidal and multipolar cells have myelinated axons; thismatches the physiological difference in conduction velocitiesbetween NS cells and COLD or HPC cells. These physiologicaland morphological classifications were not distinguished inprior intracellular labeling studies of lamina I cells for method-ological reasons. We directly re-examined the characteristics oflamina I neurons, using intracellular recording and labeling,which revealed a clear correspondence between these anatom-ical and physiological classifications.

ResultsPHYSIOLOGICAL CHARACTERISTICS

We recorded intracellularly from a total of 126 cells in the super-ficial dorsal horn of 41 anesthetized cats. Stable intracellularrecordings were obtained for several minutes to over one hour.Altogether 76 cells were injected with biocytin, 53 of which wererecovered histologically, and 38 of which were lamina I neuronsthat had been characterized physiologically. The remaining recov-ered cells were uncharacterized, unresponsive and/or were locat-ed in lamina II or deeper. For some identified cells, there wasapparent injury during penetration, based on the gradual dete-rioration of the membrane potential and action-potential ampli-tude, yet the spontaneous activity pattern and the responseproperties to natural stimuli remained stable.

All 38 characterized, labeled lamina I neurons were respon-sive to natural cutaneous stimulation of the ventral hindpaw. Theresponse characteristics of these neurons were consistent with

Nociceptive and thermoreceptive lamina I neurons are anatomicallydistinct

Z.-S. Han1,2, E.-T. Zhang1 and A. D. Craig1

1 Divisions of Neurobiology and Neurosurgery, Barrow Neurological Institute, 350 West Thomas Rd., Phoenix, Arizona 85013, USA2 Present address: RS Dow Neurological Sciences Institute, 1120 NW 20th Ave., Portland, Oregon 97209, USA

Correspondence should be addressed to A.D.C. ([email protected])

Pain and temperature stimuli activate neurons of lamina I within the dorsal horn of the spinal cord,and although these neurons can be classified into three basic morphological types and three majorphysiological classes, earlier studies did not establish a structure/function correlation between theirmorphology and their physiological responses. We recorded and intracellularly labeled 38 catlamina I neurons. All 12 fusiform cells were nociceptive-specific, responsive only to pinch and/orheat. All 11 pyramidal cells were thermoreceptive-specific, responsive only to innocuous cooling. Often multipolar cells, six were polymodal, responsive to heat, pinch and cold, and four were nocicep-tive-specific. Five unclassified cells had features consistent with this pattern. These results supportthe view that central pain and temperature pathways contain anatomically discrete sets ofmodality-selective neurons.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Based on their intracellular responses to natural stimuli, 18 ofthe 38 identified lamina I neurons were categorized as NS cells,13 as COLD cells and 7 as HPC cells (examples in Fig. 1). Noneof the cells was responsive to low-threshold (weak) stimuli. Of the18 NS cells, 8 responded only to noxious pinch, and 10 respond-ed to both pinch and noxious heat (Fig. 1). The NS cells oftenshowed prolonged afterdischarge following a noxious stimulus(Fig. 1) and moderate-sized spontaneous excitatory postsynapticpotentials (Fig. 2d). The 13 COLD cells were readily excited bycooling but were not excited by pinch or heat; rather, their dis-charges were usually inhibited by heat or even by the warmth ofthe fingers when used for pinching (Fig. 1). One COLD cell (19-558) showed a paradoxical burst response to heat. Most HPC cellsresponded vigorously to noxious heat and in a graded manner to

Fig. 2. Intracellular responses of different classes of lamina I cells to prolonged current injection (0.5 nA, 500–600 ms, indicated by the bars).(a, b) The COLD cell in (a) and the HPC cell in (b) both produced repetitive discharges without significant frequency adaptation. Note thedeep but brief afterhyperpolarization following each COLD cell action potential in (a) and the more shallow but broad and variable afterhy-perpolarization following each HPC cell action potential in (b). (c, d) In contrast, both of the NS cells responded to current injection withaction potential trains that exhibited pronounced frequency adaptation, and both displayed a prolonged hyperpolarizing potential followingcurrent offset (triangular arrowheads). Both also had spontaneous EPSPs (small downward arrows). The cell in (c) had action potentials rid-ing on EPSPs and often fired in doublets (upward arrows). The cell in (d) showed variable, brief afterhyperpolarizations.

observations in prior studies20–23, and so the cells were classifiedas nociceptive-specific (NS) cells responsive only to pinch and/orheat, thermoreceptive-specific (COLD) cells responsive only toinnocuous cooling and inhibited by warming, and polymodalnociceptive (HPC) cells responsive to heat and/or pinch and alsoto cold. (WDR lamina I neurons are rare in the cat8,9,11,15,21,26.)Although the need for rapid characterization prevented us fromsystematically delimiting the receptive fields of nociceptive cells,they were generally small and confined, ranging in size from partof one toe pad to most of the ventral surface of the paw. Manyreceptive fields occupied the glabrous pads on one or two digits,sometimes including a region of hairy skin between the toes ornear the central pad. The receptive fields of COLD cells were gen-erally larger, including half or more often all of the ventral paw.

Fig. 1. Intracellular responses of NS, COLD and HPC cells. Bars indicate the application ofstimuli. The intensity of the pinch stimuli was increased during application. The NS cell (mul-tipolar cell 36-739) had a resting membrane potential of -45 mV and a spike amplitude thatvaried with firing rate. It developed a high ongoing discharge rate, which may have occluded aphasic response to the brief cold stimulus. The COLD cell (pyramidal cell 25-670) had a rest-ing membrane potential of -40 mV, a steady spike amplitude of 45 mV, and an afterhyperpolarization of 10 mV. It had irregular ongoing dis-charge at room temperature (about 23°C) prior to any stimulation. The HPC cell (multipolar cell 44-1000) had a resting membrane potentialof -55 mV and a spike amplitude of 60 mV. The small arrow indicates hyperpolarizing current pulses (0.5 nA, 50 ms, 1Hz) applied to stabilizethe recording.

NS cell COLD cell HPC cell

a

b

c

d


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


noxious mechanical stimuli; their responses to coldwere less brisk than those of COLD cells and wereoften delayed (Fig. 1). As in prior work, HPC cells hadvaried thresholds to each submodality, and with therapid characterization methods we employed, threeHPC cells were detected that apparently had highthresholds to thermal stimulation. Two of these (27-551 and 44-1000) showed only a transient responseto the brief (3 to 5 second) cold stimuli, which is char-acteristic of HPC cells that have high noxious coldthresholds and produce a tonic response to stimuli oflonger duration (30 to 60 seconds) than was practicalfor this study21. One (20-910) that responded clearlyto pinch and to cold had only a weak, poorly repro-duced response to radiant heat, similar to HPC cellsthat have high thresholds to maintained heat uponquantitative examination with a Peltier stimulator21.

We applied antidromic stimulation to the con-tralateral thalamus in 22 of the 41 experiments, in anattempt to identify spinothalamic lamina I neurons26.Antidromic responses to trains of stimuli wereobserved in two of three identified cells in whichextracellular recordings were obtained prior to pen-etration and staining (25-690 and 28-1007). Howev-er, in intracellular recordings, only one or twotime-locked action potentials were observed withantidromic stimulation in four identified lamina Icells (17-755, 19-558, 21-633, and 25-670). This isconsistent with the reported failure of antidromicsomatic invasion in intracellularly recorded lamina Ineurons following stimulation of the midbrainparabrachial region9. Thus, whereas only these 6 ofthe 24 neurons identified in 22 cats could be consid-ered putative lamina I spinothalamic cells, many oth-ers may also have been spinothalamic.

The intracellular responses to a step-depolarizingcurrent pulse (500 milliseconds) corresponded withthe classification of the identified neurons in the smallsubset that was tested. A single action potential wasnormally evoked at threshold intensities (0.2 to 0.4nA), but depolarization at suprathreshold intensities(over 0.5 nA, 500 milliseconds) produced a train ofnon-decrementing action potentials with a steadyinterspike interval in COLD cells (Fig. 2a, n = 4) and HPC cells(Fig. 2b, n = 3). In contrast, NS cells (n = 4) displayed signifi-cant frequency adaptation (Fig. 2c and d). Further, COLD cellsshowed a prominent afterhyperpolarization following each actionpotential (Figs 1 and 2a), whereas variable or no afterhyperpo-larization was generally present in NS or HPC cells (Figs 1 and2). However, NS cells displayed a prolonged hyperpolarizationfollowing intracellular current injection (Fig. 2c and d). Thesecorrelations of membrane properties with functional class resem-ble those reported for dorsal horn cells31 and are clearly sugges-tive of differences between these physiological categories involtage-gated channels. There was also a tendency for the meanhalf-amplitude spike widths of COLD cells to be greater, althoughit was not statistically significant (COLD, 0.91 ± 0.51, n = 8, ver-sus NS, 0.55 ± 0.19, n = 6, p = 0.12; COLD versus HPC, 0.59 ±0.25, n = 7, p = 0.20).

MORPHOLOGICAL CHARACTERISTICS

Of the 38 physiologically characterized lamina I cells injectedwith biocytin, 30 were stained with the ABC/DAB reaction

(Methods), examined by light microscopy and documented withphotomicrographs and camera lucida drawings. An additionaleight cells were stained with FITC-avidin (in double-labelingexperiments performed to test various immunohistochemicalmarkers) and thus were not drawn. Most of the identified neu-rons were located in the dorsal cap at the lateral apex of the dor-sal horn, where lamina I spinothalamic cells are concentrated.The somata and dendrites of all labeled cells were well stainedand generally could be followed longitudinally within lamina Ifor hundreds of microns (see Figs 6 and 7). The shape and ori-entation of the somata and proximal dendrites of most cells couldbe assessed in one section. For a few cells, labeled elements inadjacent serial sections needed to be carefully fitted with the useof tracing paper or digital overlays. Their axons, if observable,were very fine and were not included in the drawings.

We categorized the identified lamina I neurons anatomicallyas fusiform, pyramidal, multipolar and unclassified or transition-al neurons, based on the morphological criteria of somatoden-dritic shape and orientation in horizontal sections (see below) thatwe have used in prior studies of retrogradely labeled lamina I

articles

17-699 pinch + heat, U

22-964 pinch + heat, F

16-865 pinch, F

23-795 pinch, F

45-794 pinch, M

36-739 pinch, M







43-750 pinch, M

Fig. 3. Camera-lucida drawings of the soma and proximal dendrites of each iden-tified NS cell. For each cell, the response characteristics and the morphologicalclassification (F, fusiform; M, multipolar; U, unclassified or transitional) are indi-cated. The shading indicates processes that extended dorsally or ventrally out ofthe plane of the horizontal section or that were more weakly labeled. Rostral isleft, medial is up. Scale bar, 100 µm.

100 µM


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


spinothalamic cells in cat and monkey29,30. Even though there wasvariation in the fine details of the morphology of the neurons inthis sample, which is similar to the variation observed in priorwork27–30, overall there was an obvious correspondence betweenthe functional category and the somatodendritic shape of thesecells, which indicates that there are three major structural/func-tional types of lamina I neurons. Of the 18 NS cells identified, 12were fusiform neurons, whereas 4 were multipolar and 2 wereunclassified. Of the 13 COLD cells, 11 were pyramidal, whereas 2were unclassified. Of the 7 HPC cells, 6 were multipolar and 1 wasunclassified. The converse description reveals this correspondenceeven more clearly. Of the 12 fusiform neurons, all were NS cells.Of the 11 pyramidal neurons, all were COLD cells. Of the 10 mul-tipolar cells, 6 were HPC cells and 4 were NS cells. The interme-diate characteristics of the unclassified neurons were consistentwith this pattern. The drawings in Figs 3–5 show the physiologi-cal characteristics, the anatomical classification and the somato-dendritic morphology of each cell from the sample of 30 cellsstained with diaminobenzidine, organized according to the threefunctional categories identified (NS, COLD, HPC) . The pho-tomicrographs in Fig. 6 show examples of each morphological celltype. Figure 7 shows the full dendritic arborization of one repre-sentative neuron of each major structural/functional category.

Of the 18 NS neurons identified, 10 were typ-ical spindle-shaped, longitudinally oriented,bipolar fusiform lamina I neurons of varioussizes (Fig. 3). Cell 35-1169 differed slightly inthat its large soma was located at the borderbetween lamina I and lamina II, but its dendritesextended dorsally into lamina I. Two other NScells (23-795 and 17-755) belonged to the ‘T’-shaped variety of fusiform cells, using the crite-ria described previously29,30, because they had aminor dendrite protruding perpendicularly fromthe longitudinal, spindle-shaped soma (see Fig.3). One NS cell (17-699) is an example of a bipo-lar cell with an eccentric soma, a shape that wastermed unclassified in our prior studies, butwhich could simply be another subtype of thefusiform category. The other unclassified NS cellfound was an unusual bipolar cell (FITC-labeled,not shown) that has a very large, round somaand widely radiating secondary dendriticbranches. In addition, four NS cells were excep-tional in that they were multipolar neurons;three of these (45-794, 43-750 and 47-726 [notshown]) are of the ‘tube’-like variety describedearlier29,30, and one (36-739) is of the ‘radiating’subtype. Cell 36-739 was one of only two iden-tified neurons that had spiny dendrites.

Of the 13 COLD neurons identified, allshowed a general triangular shape, and 11 wereclassified as pyramidal lamina I neurons withthree major dendritic poles and three or fourdendrites. (Fig. 4 shows eight of these; three werelabeled with FITC.) As in our prior observationsof lamina I spinothalamic cells, the dendrites ofmost of these pyramidal COLD neurons issuedfew arbors and extended primarily longitudinal-ly, though a few cells had more widespread ram-ifications (e.g., 22-903 or 27-176). Cell 25-670differed in that it had only two major dendrites.In addition, two exceptional COLD cells (43-627

and 27-333) were categorized morphologically as unclassifiedor transitional neurons, because even though they had gen-erally triangular shapes, they had multiple dendritic origins.Cell 43-627 also differed because of a protruding bulge at therostral, apical (left in Fig. 4) end of the soma, and because itsbasilar dendrites on the caudal (right) end were rotated outof the horizontal plane.

Of the seven HPC cells identified, six were multipolar lam-ina I neurons that had polygonal somata with multiple den-drites that arborized both longitudinally and mediolaterally.Of these, three (21-633, 44-1000, 32-740) were of the ‘radiat-ing’ variety of multipolar cells and three (24-514, 27-551 and20-910) were of the ‘tube’-like variety29,30. Cells 21-633 and44-1000 were clearly similar to the ‘π-shaped’ subtype of mul-tipolar lamina I spinothalamic cells described in the mon-key30, but infrequently observed in the cat29. Cell 32-740 wasthe only cell that resembled somewhat the ‘quadrilateral’ sub-type that was relatively commonly observed in both cat andmonkey. In addition, one HPC cell (26-882) was a large,unclassified neuron located on the laminae I-II border thathad two major and several minor radiating, spiny dendritesextending into lamina I and the dorsal white matter and alsointo lamina II; such a cell has not been observed before.

articles

Fig. 4. Camera-lucida drawings of the soma and proximal dendrites of each identi-fied COLD cell. For each cell, the response characteristics and the morphologicalclassification (P, pyramidal; U, unclassified or transitional) are indicated. The shadingindicates processes that extended dorsally or ventrally out of the plane of the hor-izontal section or that were more weakly labeled. Rostral is left, medial is up. Scalebar, 100 µm.

23-734 cold, P

22-903 cold, P

19-558 cold, P

43-627 cold, U

27-176 cold, P

31-672 cold, P

35-1129 cold, P

45-777 cold, P

25-670 cold, P

27-333 cold, U

100 µM


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

DiscussionPrior physiological and morphological findings indicate that lam-ina I neurons belong to different classes. There are at least threedifferent physiological classes, which differ not only in theirresponse characteristics but also in the conduction velocities oftheir ascending axons, the distribution of their terminals in thethalamus, their susceptibility to descending controls, and theirresponses to pharmacological modulation20,21,23,32,33. Similarly,there are three basic morphological types of neurons that differ intheir somatodendritic shape, the myelination of their axons andtheir longitudinal distribution in the spinal cord27–30. In the pre-sent study, we re-examined the possibility of a correlationbetween cell morphology and response characteristics with intra-cellular labeling in the cat, and we have obtained strong evidence

of a homeomorphic correspondence between these respec-tive classes. This finding supports the view that there arethree basic types of lamina I neurons with correlated phys-iological and morphological characteristics: fusiform NScells, pyramidal COLD cells, and multipolar HPC cells.

This evidence provides direct support for the conceptthat pain and temperature are represented centrally by dis-tinct sets of modality-selective neurons1–3. It indicates thatthe functional and anatomical characteristics of these neu-rons are linked and developmentally specified. We infer forseveral reasons that these lamina I cell types are a generalmammalian feature. These morphological types have beenobserved in rat, cat and monkey. The three physiologicalclasses have been demonstrated in cat and in monkey. In therat, physiological data from a lamina I projection target(parabrachial nucleus34) indirectly support this idea. Fur-thermore, a thermoreceptive-(COLD)-specific subnucleuswithin trigeminal lamina I of the owl monkey has recentlybeen identified, and it contains almost entirely pyramidaltrigeminothalamic neurons, whereas fusiform and multi-polar neurons are present in the adjacent portions of laminaI, where NS and HPC cells are recorded (Blomqvist, A., E.-T.Z. & A.D.C. Soc. Neurosci. Abstr. 48.15, 1995).

The correspondence of the three morphological types withthe three major physiological classes validates the biologicalrelevance of these respective categories. Further, this evidencefor discrete categories of lamina I neurons implies the exis-tence of other pharmacological and neurochemical correlatesthat, when uncovered, will facilitate the analysis of specificpathways activated by painful and thermal stimuli. For exam-ple, many NS lamina I cells that respond to substance P in therat35 are fusiform cells that bear NK1 receptors36,37, which are

found on very few pyramidal neurons (Yu, X.H., A.D.C. et al.,unpublished observations). Similarly, the COLD-specific subnu-cleus within trigeminal lamina I of the owl monkey is devoid offibers immunoreactive for substance P or serotonin, in contrast tothe remainder of lamina I (Blomqvist, A., E.-T.Z. & A.D.C. Soc. Neu-rosci. Abstr. 48.15, 1995). In addition, these cell types may containdifferent peptides in the rat38. Conversely, physiological inferencescan now be made based on anatomical knowledge of the types oflamina I cells that terminate in a region. For example, retrogradelabeling work indicates that all three types of lamina I cells project tothe autonomic cell column in the ventrolateral medulla in cat; thissuggests that, in addition to nociceptive sensitivity39, neurons in thisregion may respond to innocuous cooling, which was recently con-firmed (Krout, K. & A.D.C., unpublished observations).

Fig. 5. Camera-lucida drawings of the soma and proximal dendrites ofeach identified HPC cell. For each cell, the response characteristics andthe morphological classification (M, multipolar; U, unclassified or transi-tional) are indicated. The shading indicates processes that extended dor-sally or ventrally out of the plane of the horizontal section or that weremore weakly labeled. Rostral is left, medial is up. Scale bar, 100 µm.

Fig. 6. Digital photomicrographs of intracellularly stained, identified lamina I neurons. F1, fusiform NS cell 28-1007, with a small arrowheadindicating a second faintly labeled cell; F2, fusiform NS cell 22-820; P1, pyramidal COLD cell 35-1129; P2, pyramidal COLD cell 23-734; M1,multipolar HPC cell 21-633; M2, multipolar NS cell 45-794; T1, ‘T-shaped’ fusiform NS cell 17-755; U1, unclassified or transitional COLD cell27-333. Scale bar, 100 µm.

21-633 heat + pinch + cold, M



20-910 pinch + cold, M



26-882 heat + pinch + cold, U

100 µM


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


It must be noted that, although the morphological charac-teristics of each of the three basic categories are generally con-sistent, there is nonetheless variation within each. This mayreflect a degree of ambiguity, or incomplete selectivity, in devel-opmental specification. This may also imply that subdivisionsof these basic categories of lamina I neurons remain to be iden-tified. We consider this latter possibility rather likely, in partbecause lamina I receives inputs from all tissues of the body(including muscles, joints, viscera, cornea and teeth26,40; forother references, see ref. 3), whereas we stimulated only skin,and in part because subtypes of each category are consistentlyrecognizable morphologically27–30. For example, in the presentsample we obtained almost no ‘quadrilateral’ multipolar cells,although these are common in the cat, which indicates that suchneurons may not receive cutaneous input.

Accordingly, the four exceptional NS cells that are multipo-lar neurons could represent cells that received additional inputfrom untested tissues or that were sensitive to untested sub-modalities that evince polymodal C fiber activity, such as chem-ical or metabolic stimuli41–43. It is also possible that thesemultipolar cells were polymodal nociceptive HPC cells with highthermal thresholds that were incompletely characterized, whichwould conform with the remaining observations. The HPC lam-ina I spinothalamic cells can have different sensitivities to thesubmodalities of heat, pinch and cold, and some respond to coldonly below 15 °C or to heat only above 50 °C when tested with aPeltier stimulator21,22 (unpublished observations). Such noxiousthermal stimuli can require a longer stimulus application than

was deemed practical during intracellular characteri-zation in the present study.

The correspondence we have identified was not dis-covered in numerous previous attempts to correlatefunction and morphology in lamina I with intracellu-lar HRP labeling8–16, probably due to methodologicalreasons. First, sagittal or transverse sections were cut inall of the previous studies. This precluded recognitionof the morphological types differentiated in the pre-sent study, which can be consistently observed only inthe horizontal (or tangential) plane in which laminaI neurons arborize. Second, innocuous and noxiouscold stimuli were not applied in most of the previousstudies. This precluded identification of the COLDand HPC functional classes. Notably, cold stimuli wereused in the earliest study8, and the single lamina I cellidentified as specifically sensitive to innocuous cool-ing was described as a pyramidal cell, although intransverse sections. The one cell that was identified asresponsive to heat, pinch and cold (explicitly ascribedto polymodal C nociceptor input, albeit perhaps sen-sitized) had several dendrites extending horizontallyinto lamina I, but its shape cannot be discerned fromthe sagittal reconstruction.

In almost all of the other previous studies, coldstimuli were not used, and lamina I neurons were char-acterized only as NS or WDR cells. Significantly, thephysiological classification of lamina I cells can be con-founded by unintentional thermal stimulation. BothCOLD cells and HPC cells with high cold sensitivitycan be misconstrued as WDR cells if the temperaturesof the probes used for characterization are uncon-trolled, because even a room-temperature probe canevoke a graded low-threshold response in such cells,whereas a neutral probe (warmed to skin temperature)

does not. The resemblance between the thermal sensitivity ofsome WDR lamina I spinothalamic cells24 and that of HPC lam-ina I spinothalamic cells22,23 supports this possibility. In addi-tion, NS lamina I cells can also be misinterpreted as WDR cells,because they can become sensitized to weak stimulation follow-ing repeated noxious or C-fiber stimuli, such as commonly usedfor searching26,44. These considerations underscore the recogni-tion of three major physiological/morphological categories oflamina I cells. Nevertheless, the possibility remains that WDRlamina I cells, which have been observed more often in monkeythan in cat, might yet be distinguishable, and this should bedirectly re-examined in light of the present observations.

Finally, the observation of distinct types of nociceptive andthermoreceptive lamina I neurons is consistent with the conceptthat these neurons supply substrates for discrete sensory chan-nels for pain and temperature sensibilities. Prior comparativepsychophysical and physiological results obtained using the ther-mal grill illusion (in which a sensation of burning, ice-like pain isevoked by interlaced cool and warm stimuli) indicated that theactivity of COLD lamina I spinothalamic neurons can be asso-ciated directly with cold sensation, whereas the activity of HPClamina I spinothalamic neurons can be associated with the burn-ing pain caused by noxious cold or unmasked by the thermalgrill6,22. Furthermore, other work indicates that these neuronscan provide the input for modality-selective somato-autonomic(homeostatic) reflexes to painful and thermal stimuli45,46. Thesefindings have led to the proposal that the fundamental role oflamina I neurons is to distribute modality-selective sensory infor-

articles

Fig. 7. Camera lucida reconstructions of the full dendritic extent of one exam-ple of each of the identified functional/morphological categories of lamina I neu-rons. Nearly all of the dendrites were contained in two or three serialhorizontal sections for each cell. Rostral is left, medial is up.

Fusiform NS28-1007

Pyramidal COLD23-734

Multipolar HPC24-514

100 µM


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

mation on the physiological status of the tissues of the entiremammalian body to both sensory and homeostatic substrates. Thepresent observations provide firm evidence supporting this con-cept and argue that modality-selective lamina I projection neu-rons may justifiably be considered to be anatomically specified.

MethodsGENERAL PREPARATION. These procedures were approved by the localinstitutional animal care and use committee. Acute recording experi-ments were performed in 41 adult cats of either sex (2.5-4.5 kg) pre-pared as described20–22. Briefly, the animals were anesthetized withpentobarbital (40 mg per kg i.p.). Cannulae were inserted in the rightcephalic vein (for infusions) and left carotid artery (for the blood pres-sure monitor). Dexamethasone (10 mg i.v.) was administered. Sup-plemental pentobarbital (approximately 4 mg per kg per h) was givenas needed to maintain areflexia and pupillary constriction and bloodpressure near 120 mmHg. A tracheostomy was performed, the animalwas artificially ventilated with 25% O2 and 75% air, and a bilateralpneumothorax was induced. End-tidal CO2 was maintained at3.5–4.5%. Body temperature was maintained at 37.5°C with a heatingpad and a feedback-controlled infrared lamp. Pancuronium was givenevery 60–90 min for paralysis; withdrawal reflexes were tested betweendoses. A dorsal laminectomy exposed the lumbosacral enlargement,and a craniectomy was performed to provide access to the right thal-amus. The animal was mounted in a stereotaxic holder, and the spinalcord was stabilized with vertebral clamps. The dura was reflected, andthe cord was covered with a pool of mineral oil or Tyrode’s solutionthat was maintained at 37–38°C with a DC heater element.

ANTIDROMIC STIMULATION. Extracellular recordings were made withtungsten-in-glass microelectrodes to identify the right somatosenso-ry thalamus in 22 of these cats. Based on these coordinates, an array ofeight concentric bipolar electrodes was implanted in order toantidromically activate lamina I spinothalamic neurons, as in priorstudies. Stimulus trains consisting of four pulses (0.1–2 mA, 2 ms) at200 or 250 Hz were delivered to individual electrodes or between pairsof electrodes. Spinal cells that responded to such trains in a one-for-one, time-locked manner at a fixed latency with a distinct thresholdwere recognized as spinothalamic neurons; we considered this suffi-cient, because lamina I cells that fulfill this criterion have in all ourprior studies demonstrated a strict antidromic collision interval whentested. Extracellular recordings were made from the lumbosacralspinal cord at sites in L7–S1 exposed by small openings in the pia, inorder to ensure that lamina I spinothalamic cells were antidromical-ly activated by the thalamic array and to localize regions containinglamina I cells with receptive fields on the ventral hindpaw for subse-quent intracellular recording.

INTRACELLULAR RECORDINGS. Recordings were made with glass micro-electrodes fabricated from 1.0 mm standard or 1.2 mm borosilicate glasspipettes and an Axoclamp 2-A (Axon Instruments) amplifier in bridgemode. The electrodes were back-filled with 2% biocytin (Sigma) in 2 Mpotassium methylsulphate; their resistance ranged from 40 to 80 MΩ.The electrodes were lowered into the superficial dorsal horn near thedorsal root entry zone in 2.0 or 2.5 µm steps. Cells were sought at a depthof 200-800 µm below the surface. Extracellular recordings were obtainedinfrequently, using antidromic stimulation as a search method. Standardmethods of current application through the microelectrode were usedto facilitate cell impalement and to clear debris from the electrode tip.Intracellular recording was signaled by a sudden large action potential,a resting membrane potential of at least -40 mV and action potentials of45 to 65 mV. Hyperpolarizing current pulses (0.5 nA, 50 ms) were oftenapplied in an effort to stabilize the recording.

RESPONSE CHARACTERIZATION. Following cell impalement, antidromicactivation was tested and natural stimuli were quickly used for unit char-acterization20–23. Innocuous stimuli (light touch, tapping, gentle pres-sure, limb movement) were generally tested first, followed by thermalstimuli (cooling with a wet ice cube or cold beaker, warming with thehand or a radiant heat lamp) and then noxious stimuli (pressure and

pinch applied with fingers or a smooth forceps, noxious heat). The cuta-neous stimuli were applied at various sites on the ventral hindpaw for3–5 s; the intensity of the radiant heat lamp was adjusted such that, heldat a fixed distance, it became painful to the investigators after about 5 s(about 45°C) and intolerable after about 8 s (about 50°C). Responseswere monitored through a speaker and an oscilloscope and were record-ed on magnetic tape.

Compromise between the competing needs for proper response char-acterization and for quick initiation of current injection for intracellularstaining meant that stimuli were repeated only three or four times todemonstrate reproducibility, that receptive fields were regionally but notprecisely delimited, and that thermal stimuli were not always maintainedsufficiently long (cold, 30–60 s; heat, 5–10 s) to generate the noxiousstimulus required by some HPC cells21–23. Cells that responded to low-threshold cutaneous stimuli, cells with high ongoing discharge, and cellsthat were silent or unresponsive were also encountered, but these werenot studied in detail in these experiments, primarily because prior stud-ies indicated that lamina I spinothalamic cells in cat do not include suchcells. These cells were occasionally stained, however, and they were sub-sequently found to lie in lamina II or deeper.

INTRACELLULAR STAINING. Following physiological characterization ofa cell, depolarizing current pulses (0.5 to 5.0 nA, 500-ms duration at1 Hz for 1 to 7 min) were applied to inject biocytin. In most experi-ments, several intracellular injections were attempted at sites separat-ed longitudinally by 0.5 to 5 mm that were carefully mapped for laterreconstruction. A period of one to eight hours was allowed followingthe injections in order to permit diffusion of the biocytin into cellprocesses. The animals were then given an overdose of barbiturate andperfused transcardially with a rinse of 0.1 M PBS (pH 7.4, room tem-perature), a fixative consisting of two liters of 4% paraformaldehydeand 0.1% glutaraldehyde in 0.1 M PB, and then one liter of 10%sucrose in 0.1 M PB. The lumbosacral spinal cord was removed andplaced in 30% sucrose in PB overnight. Serial 50-µm frozen sectionscut in the horizontal plane were rinsed, incubated in avidin-biotiny-lated horseradish peroxidase complex (ABC, 1:100, Vector) in 0.05 MTris-buffered saline (TBS) at 4 °C overnight, and reacted with 0.05%3,3’-diaminobenzidine (DAB) containing 0.01% hydrogen peroxidein TBS for 5–10 min. Labeled cells were identified in the light micro-scope as lamina I neurons if they were located in the most superficialaspect of the gray matter, dorsal to the substantia gelatinosa, which isclearly recognizable with light- and dark-field illumination. Pho-tographs were made of all cells with Kodak Technical Pan film andEktachrome film using 20x and 40x apochromatic objectives. Digitalmicroscopic images were made directly with a Leaf Microlumina scan-ner (at 3380 x 2253 pixels) and a 20x objective, and Adobe Photoshopwas used to enhance contrast and apply labels. Camera lucida draw-ings were made using 40x and 60x (oil) objectives.

In most cases, only one biocytin-labeled cell was found at theappropriate location, but for 9 of the 30 identified cells, one to fourneighboring cells were also labeled. These were usually small, light-ly labeled fusiform neurons (Fig. 6, F1, arrowhead), in contrast to thedarkly labeled cell that we identified as the characterized neuron.(Interestingly, this occurred most often, in five of these nine instances,when the identified cell was also a fusiform neuron.) In two instances,we found two adjacent, well labeled cells that had similar shapes (oncefusiform neurons, once multipolar neurons), but one was more dark-ly stained and had signs of penetration (a darkly stained, disruptedsection of membrane). In one instance, four cells with differing shapeswere labeled with similar intensity and without obvious signs of pen-etration in any particular one, and thus the physiologically charac-terized cell could not be identified and the data were excluded.

AcknowledgementsWe thank Elizabeth O’Campo, Maribeth Tatum, and Jan Carey for assistance.

This work was supported by NIH grants NS 25616 and DA 07402 and the

James R. Atkinson Pain Research Fund administered by the Barrow

Neurological Foundation.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


25. Price, D.D., Hayes, R.L., Ruda, M. & Dubner, R. Spatial and temporaltransformations of input to spinothalamic tract neurons and their relation tosomatic sensations. J. Neurophysiol. 41, 933–947 (1978).

26. Craig, A.D. Jr. & Kniffki, K.-D. Spinothalamic lumbosacral lamina I cellsresponsive to skin and muscle stimulation in the cat. J. Physiol. (Lond.) 365,197–221 (1985).

27. Gobel, S. Golgi studies of the neurons in layer I of the dorsal horn of themedulla (trigeminal nucleus caudalis). J. Comp. Neurol. 180, 375–394 (1978).

28. Lima, D. & Coimbra, A. A Golgi study of the neuronal population of themarginal zone (lamina I) of the rat spinal cord. J. Comp. Neurol. 244, 53–71(1986).

29. Zhang, E.T., Han, Z.S. & Craig, A.D. Morphological classes of spinothalamiclamina I neurons in the cat. J. Comp. Neurol. 367, 537–549 (1996).

30. Zhang, E.T. & Craig, A.D. Morphology and distribution of spinothalamiclamina I neurons in the monkey. J. Neurosci. 17, 3274–3284 (1997).

31. Lopez-Garcia, J.A. & King, A.E. Membrane properties of physiologicallyclassified rat dorsal horn neurons in vitro: Correlation with cutaneoussensory afferent input. Eur. J. Neurosci. 6, 998–1007 (1994).

32. Dostrovsky, J.O., Shah, Y. & Gray, B.G. Descending inhibitory influencesfrom periaqueductal gray, nucleus raphe magnus, and adjacent reticularformation. II. Effects on medullary dorsal horn nociceptive andnonnociceptive neurons. J. Neurophysiol. 49, 948–960 (1983).

33. Mokha, S.S., Goldsmith, G.E., Hellon, R.F. & Puri, R. Hypothalamic controlof nocireceptive and other neurones in the marginal layer of the dorsal hornof the medulla (trigeminal nucleus caudalis) in the rat. Exp. Brain Res. 65,427–436 (1987).

34. Menendez, L., Bester, H., Besson, J.M. & Bernard, J.F. Parabrachial area:Electrophysiological evidence for an involvement in cold nociception. J.Neurophysiol. 75, 2099–2116 (1996).

35. De Koninck, Y. & Henry, J.L. Substance P-mediated slow excitatorypostsynaptic potential elicited in dorsal horn neurons in vivo by noxiousstimulation. Proc. Natl Acad. Sci. USA 88, 11344–11348 (1991).

36. Brown, J.L. et al. Morphological characterization of substance P receptor-immunoreactive neurons in the rat spinal cord and trigeminal nucleuscaudalis. J. Comp. Neurol. 356, 327–344 (1995).

37. Marshall, G.E., Shehab, S.A.S., Spike, R.C. & Todd, A.J. Neurokinin-1 receptorson lumbar spinothalamic neurons in the rat. Neuroscience 72, 255–263 (1996).

38. Lima, D., Avelino, A. & Coimbra, A. Morphological characterization ofmarginal (lamina I) neurons immunoreactive for substance P, enkephalin,dynorphin and gamma-aminobutyric acid in the rat spinal cord. J. Chem.Neuroanat. 6, 43–52 (1993).

39. Sun, M.-K. & Spyer, K.M. Nociceptive inputs into rostral ventrolateralmedulla-spinal vasomotor neurones in rats. J. Physiol. (Lond.) 436, 685–700(1991).

40. Cervero, F. & Tattersall, J.E.H. Somatic and visceral inputs to the thoracicspinal cord of the cat: marginal zone (lamina I) of the dorsal horn. J. Physiol.(Lond.) 383, 383–395 (1987).

41. MacIver, M.B. & Tanelian, D.L. Activation of C fibers by metabolicperturbations associated with tourniquet ischemia. Anesthesiology 76,617–623 (1992).

42. Pickar, J.G., Hill, J.M. & Kaufman, M.P. Dynamic exercise stimulates groupIII muscle afferents. J. Neurophysiol. 71, 753–760 (1994).

43. Schmelz, M., Schmidt, R., Bickel, A., Handwerker, H.O. & Torebjörk, H.E.Specific C-receptors for itch in human skin. J. Neurosci. 17, 8003–8008(1997).

44. Woolf, C.J., Shortland, P. & Sivilotti, L.G. Sensitization of highmechanothreshold superficial dorsal horn and flexor motor neuronesfollowing chemosensitive primary afferent activation. Pain 58, 141–155(1994).

45. Craig, A.D. Propriospinal input to thoracolumbar sympathetic nuclei fromcervical and lumbar lamina I neurons in the cat and the monkey. J. Comp.Neurol. 331, 517–530 (1993).

46. Craig, A.D. Distribution of brainstem projections from spinal lamina Ineurons in the cat and the monkey. J. Comp. Neurol. 361, 225–248 (1995).

articles

1. Perl, E.R. in Handbook of Physiology, Section 1, The Nervous System, VolumeIII, Sensory Processes (ed. Darian-Smith, I.) 915–975 (American PhysiologicalSociety, Bethesda, 1984).

2. Willis, W.D. The Pain System. (Karger, Basel, 1985).3. Craig, A.D. in Somesthesis and the Neurobiology of the Somatosensory Cortex

(eds Franzen, O., Johansson, R. & Terenius, L.) 27–39 ( Birkhäuser, Basel,1996).

4. Craig, A.D., Bushnell, M.C., Zhang, E.-T. & Blomqvist, A. A thalamic nucleusspecific for pain and temperature sensation. Nature 372, 770–773 (1994).

5. Coghill, R.C. et al. Distributed processing of pain and vibration by the humanbrain. J.Neurosci. 14, 4095–4108 (1994).

6. Craig, A.D., Reiman, E.M., Evans, A. & Bushnell, M.C. Functional imaging ofan illusion of pain. Nature 384, 258–260 (1996).

7. Wall, P.D. Pain in the brain and lower parts of the anatomy. Pain 62, 389–391(1995).

8. Light, A.R., Trevino, D.L. & Perl, E.R. Morphological features of functionallydefined neurons in the marginal zone and substantia gelatinosa of the spinaldorsal horn. J. Comp. Neurol. 186, 151–172 (1979).

9. Light, A.R., Sedivec, M.J., Casale, E.J. & Jones, S.L. Physiological andmorphological characteristics of spinal neurons projecting to theparabrachial region of the cat. Somatosens. Mot. Res. 10, 309–325 (1993).

10. Bennett, G.J., Abdelmoumene, M., Hayashi, H., Hoffert, M.J. & Dubner, R.Spinal cord layer I neurons with axon collaterals that generate local arbors.Brain Res. 209, 421–426 (1981).

11. Molony, V., Steedman, W.M., Cervero, F. & Iggo, A. Intracellular marking ofidentified neurones in the superficial dorsal horn of the cat spinal cord. Q. J.Exp. Physiol. 66, 211-223 (1981).

12. Hoffert, M.J., Miletic, V., Ruda, M.A. & Dubner, R. Immunocytochemicalidentification of serotonin axonal contacts on characterized neurons inlaminae I and II of the cat dorsal horn. Brain Res. 267, 361–364 (1983).

13. Woolf, C.J. & Fitzgerald, M. The properties of neurones recorded in thesuperficial dorsal horn of the rat spinal cord. J. Comp. Neurol. 221, 313–328(1983).

14. Miletic, V., Hoffert, M.J., Ruda, M.A., Dubner, R. & Shigenaga, Y.Serotoninergic axonal contacts on identified cat spinal dorsal horn neuronsand their correlation with nucleus raphe magnus stimulation. J. Comp.Neurol. 228, 129–141 (1984).

15. Steedman, W.M., Molony, V. & Iggo, A. Nociceptive neurones in thesuperficial dorsal horn of cat lumbar spinal cord and their primary afferentinputs. Exp. Brain Res. 58, 171–182 (1985).

16. Hylden, J.L.K., Hayashi, H., Dubner, R. & Bennett, G.J. Physiology andmorphology of the lamina I spinomesencephalic projection. J. Comp. Neurol.247, 505–515 (1986).

17. Moschovakis, A.K., Burke, R. & Fyffe, R. The size and dendritic structure ofHRP-labeled gamma motoneurons in the cat spinal cord. J. Comp. Neurol.311, 531–545 (1991).

18. Tamamaki, N., Uhrich, D.J. & Sherman, S.M. Morphology of physiologicallyidentified retinal X and Y axons in the cat’s thalamus and midbrain as revealedby intraxonal injection of biocytin. J. Comp. Neurol. 354, 583–607 (1995).

19. Christensen, B.N. & Perl, E.R. Spinal neurons specifically excited by noxiousor thermal stimuli: marginal zone of the dorsal horn. J. Neurophysiol. 33,293–307 (1970).

20. Craig, A.D. & Hunsley, S.J. Morphine enhances the activity ofthermoreceptive cold-specific lamina I spinothalamic neurons in the cat.Brain Res. 558, 93–97 (1991).

21. Craig, A.D. & Serrano, L.P. Effects of systemic morphine on lamina Ispinothalamic tract neurons in the cat. Brain Res. 636, 233–244 (1994).

22. Craig, A.D. & Bushnell, M.C. The thermal grill illusion: unmasking the burnof cold pain. Science 265, 252–255 (1994).

23. Dostrovsky, J.O. & Craig, A.D. Cooling-specific spinothalamic neurons in themonkey. J. Neurophysiol. 76, 3656–3665 (1996).

24. Ferrington, D.G., Sorkin, L.S. & Willis, W.D. Responses of spinothalamictract cells in the superficial dorsal horn of the primate lumbar spinal cord. J.Physiol. (Lond.) 388, 681–703 (1987).


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Although the descending projection from the primarysomatosensory cortex to the ventroposterior nucleus (VP) of thethalamus is known to be seven to ten times greater than theascending projection from VP to cortex1, the functional role ofthis massive corticofugal projection has remained elusive2. Pre-vious studies on the contribution of neural feedback connectionsto the processing of sensory information have reported relative-ly minor and inconsistent effects on receptive field size orresponse properties of thalamic relay neurons following corticalperturbations3–7. Although the connectivity of the system sug-gests a substantial influence from cortex over the processing ofsomatosensory information at the level of the thalamus, therehas been no definitive demonstration of a major cortical influ-ence on receptive field size, response properties or somatotopicorganization in VP.

A majority of corticofugal projection cells in primarysomatosensory cortex (area 3b) express NMDA receptors8, andchronic systemic blockade of NMDA receptors produces largeunresponsive areas in somatosensory cortex9. In this study, wedetermined that chronic administration of an NMDA receptorantagonist directly into the area 3b cortical hand representa-tion results in a suppression of activity in area 3b and an enor-mous enlargement of receptive fields (RFs) in the VP handrepresentation. In addition, acute cortical suppression alsoresulted in an enlargement of RFs in VP. These findings force are-evaluation of traditional ‘bottom-up’ models of sensory pro-cessing, which view the thalamus as a simple relay nucleus tothe cortex, and they also have important implications for stud-ies of adult neuronal plasticity.

ResultsOver a period of one to five months, either saline (two controlmonkeys, Macaca mulatta) or the NMDA receptor antagonist D-2-amino-5-phosphonovaleric acid (D-APV; five monkeys, Maca-ca mulatta) was infused directly into the area 3b handrepresentation. The percentage of responsive sites and RF size forthe cortex and VP in the saline control group was completely

consistent with those reported in previous studies by us and oth-ers in normal animals10–12. In accordance with previous stud-ies9, our recordings throughout the cortical hand representationone to five months after D-APV administration revealed thatonly 10% of cortical recording sites were responsive to somaticstimulation (Table 1), compared to 100% in control animals.Analysis of the responsive 10% of recording sites in the cortex ofthe experimental animals showed that these neurons exhibitedRFs comparable in size to those found in the cortex of the controlanimals. There was no evidence of degeneration in the thalamusor cortex in any of the animals used in the present study.

These results on RF size obtained at the cortical level in thechronic D-APV treated animals contrasted sharply with thosefound at the thalamic level of these same animals, where RFswere greatly enlarged compared to those seen in controls, fre-quently encompassing more than one digit and often half ormore of the entire glabrous hand (Fig. 1c). In some instances

Cortically induced thalamic plasticity inthe primate somatosensory system

E.R. Ergenzinger1,2, M.M. Glasier1, J.O. Hahm3 and T.P. Pons1

1 Department of Neurosurgery and 2Program in Neuroscience, Wake Forest University School of Medicine, Winston-Salem, North Carolina 27127, USA3 Department of Neurosurgery, Georgetown University, Washington, D.C. 20057, USA

Correspondence should be addressed to T.P.P. ([email protected])

The influence of cortical feedback on receptive field organization in the thalamus was assessed inthe primate somatosensory system. Chronic and acute suppression of neuronal activity in primarysomatosensory cortex resulted in a striking enlargement of receptive fields in the ventroposteriorthalamus. This finding demonstrates a dramatic ‘top-down’ influence of cortex on receptive fieldsize in the somatosensory thalamus. In addition, this result has important implications for studies ofadult neuronal plasticity because it indicates that changes in ‘higher-order’ areas of the brain cantrigger extensive changes in the receptive field characteristics of neurons located earlier in theprocessing pathway.

Table 1. Responsivity of recording sites in cortex andthalamus following chronic cortical administration of D-APV or saline1

Control D-APVCortex (Area 3b)% Responsive Sites

Infusion Zone (Hand) 100%(197) 10%(754)Other 100%(47) 99%(82)Fringe Zone (Hand & Other) N.A. 78%(262)

% “Large” RF Sites 0%(197) 6%(78)

Thalamus (VP)% Responsive Sites

Hand 100%(102) 99%(388)Other 99%(116) 100%(243)

% “Large” RF Sites 2%(102) 58%(383)

1see Methods for details on data analysis.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

these RFs included the entire distal forearm and hand (Fig. 1c), atype of RF that is never seen in normal animals. Such large RFswere recorded from 58% of the recording sites within the VPhand representation of our experimental animals. The RFenlargement was specific in that these RFs were observed onlyfor recording sites within the VP hand representation and not inbody-part representations outside of the D-APV infusion zonein cortex (Fig. 1c).

These findings indicate that chronic D-APV administrationin the area 3b hand representation blocks much of the stimulus-driven activity at the cortical level and tremendously expands thesize of RFs in the portion of the VP thalamus that normally rep-resents the hand (within the ventroposterior lateral nucleus orVPL). This expansion of RF size occurred for the vast majorityof hand-responsive recording sites throughout VPL (Fig. 2). Such

an enormous expansion of RF size after delivery of pharmaco-logical agents to either the cortex or thalamus has not beenreported previously3–7.

We next determined if acute administration of D-APV wouldproduce results similar to those from chronic treatment. We firstrecorded from the cortex and thalamus of two additional mon-keys before administering D-APV. We then delivered injectionsof D-APV directly to the cortex of these same animals via aHamilton syringe and immediately assessed the results. Acutecortical recordings were similar to those found in the chronical-ly administered animals, with 93% of 83 sites unresponsive tosomatic stimulation. Recordings in the thalamus of the acutelyadministered animals showed an increase in RF size, though notas great as that observed in the chronically administered animals.Although a comparable number of sites (45% of 55 sites) showedRFs that encompassed one or more digits, compared to only 2%of 51 sites before acute D-APV administration, the magnitudeof the RF enlargement was not as great as that observed in chron-ically administered animals, in that no RFs larger than half of thehand were observed (Fig. 3).

DiscussionThe present findings demonstrate that, at least under our exper-imental conditions, top-down projections from cortex can inducelarge-scale reorganization of RFs at a relatively early processingstation in the somatosensory system. This effect seems most like-ly to be mediated directly by corticothalamic projections or indi-rectly via corticocuneate projections. Additionally, the increasein RF size for VP neurons could occur through a loss of excita-tory corticothalamic input on inhibitory interneurons within VPor on the thalamic reticular nucleus. Of course, unknown changesin additional somatosensory cortical areas such as SII, the insu-la and/or posterior parietal cortex could conceivably mediate theRF enlargement through complex corticothalamic processing,though this latter possibility seems less likely. That these resultsmight be explained by retrograde degeneration seems highlyunlikely, given that lesions of the somatosensory cortex, whichare known to produce massive retrograde degeneration, do notresult in expanded thalamic RFs13. Furthermore, the expansion ofRF’s immediately after delivery of D-APV to the cortex would

Fig. 1. Representative RFs in VP following chronic administration of D-APV to somatosensory cortex. (a) Representative RFs, indicated bythinner black lines, recorded from sites localized within the hand region of VPL from saline controls. (b) Representative electrode tracksthrough a coronal section of the thalamus in a macaque that had received D-APV to the area 3b hand representation (CM, central median; LP,lateral posterior; VPM, ventroposterior medial; VPI, ventroposterior inferior; VPL, ventroposterior lateral). Tick marks on tracks indicaterecording sites. Red, recording sites with RFs on the face. Blue, recording sites with RFs on or including the hand. Green, recording sites withRFs on the forearm. Colors are matched with RFs in (c). (c) Representative RFs recorded from sites localized within VP from a monkey withD-APV administered to area 3b. All RFs were contralateral to VP. Arrow denotes rotation of the hand and forearm to allow visualization oftheir ventral surfaces.

a b c%

rec

ord

ing

sit

es

≤ 1 Pad > 1 Pad< 1 Digit ≥ 1 Digit

Fig. 2. Distribution histogram of RF sizes for recording sites local-ized within the VPL hand representation between control and D-APV groups. RFs were categorized as to whether they were ≤ 1 pad,< 1 digit, > 1 pad, or ≥ 1 digit. RFs that were classified as > 1 padwere restricted to the pads and did not encompass any digits of thehand. RFs that were classified as ≥ 1 digit were those RFs that incor-porated at least one digit, and included RFs as large as the entirehand and forearm (see Fig. 1).

ControlD-APV

70

60

50

40

30

20

10

0


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


with an initial dose of ketaminehydrochloride (15 mg per kg) fol-lowed by intubation and admin-istration of isoflurane (0.5% to3%) to effect. Throughoutsurgery, which was performedusing aseptic precautions, the ani-mals received an intravenous dripof a solution of 5% dextrose and0.45% sodium chloride, and theirheart rate, respiratory rate, andtemperature were monitored andmaintained within normal limits.

For the chronic administrationof D-APV, we implanted a tran-scranial catheter directly into thehand representation in area 3b andattached the catheter to an osmot-ic pump containing 50 µM D-APVor physiological saline delivered ata rate of 2.5 µl per h. D-APV orsaline was administered from

between one and five months, and the osmotic pump was replaced every28 days. Ketamine was not used during osmotic pump replacements becauseof potential interactions with glutamatergic neurotransmission. Instead,valium was used to make the animal receptive to isoflurane anesthesia.

For the acute administration of D-APV, a Hamilton syringe was usedto inject D-APV directly into the hand representation in area 3b. In orderto assure that the entire hand representation was affected by the D-APVas in the chronic study, multiple injections of 2–3 µl of D-APV were madeapproximately 1 mm apart across the mediolateral extent of the handrepresentation in cortex (5–6 injections total). The injections consisted of50 µM D-APV.

ELECTROPHYSIOLOGICAL RECORDING PROCEDURES. A recording chamber andhead fixation device were attached to the skull. The chamber was madefrom dental acrylic and positioned to provide maximum access forrecording from VP and area 3b. The bone within the chamber wasremoved to expose the dura, the dura then reflected and a high resolu-tion picture of brain vasculature was then obtained with a CCD camerausing appropriate filters. The brain picture was used to help in the place-ment of electrode penetrations during recording, with the RF and respon-sivity characteristics defined for each recording point.

The mapping and recording procedures were similar to those usedpreviously25. In the chronic preparation, electrode penetrations wereplaced at 0.3–0.5 mm intervals across the mediolateral extent of VP andarea 3b and at 0.3–0.5 mm intervals across its rostrocaudal extent.Approximately 20–40 electrode penetrations were made across VP ineach hemisphere studied, and approximately 50–60 electrode penetra-tions were made through area 3b in each hemisphere studied. In the acutepreparation, 4–5 penetrations were made across VP, and 5–6 penetra-tions were made across area 3b before and after cortical D-APV injec-tions. Penetrations across VP were placed at 0.3–0.5 mm intervalsmediolaterally and rostrocaudally. Penetrations across area 3b were placedapproximately 1 mm apart mediolaterally. The location of pre- and post-injection penetrations were the same. Microelectrodes were hydraulical-ly advanced through cortex and thalamus until single- or multiple-unitresponses to mechanical stimulation of the body could be isolated. Smallmarker lesions (10 µA for 10 s) were placed in the thalamus and cortex atstrategic points to assist in locating the recordings sites post mortem.

HISTOLOGICAL ANALYSIS. On completion of all electrophysiologicalrecording experiments, animals were given a lethal dose of pentobar-bital and then perfused intracardially with 0.9% saline followed by 4%paraformaldehyde. Brains were cut in the coronal plane on a freezingmicrotome into 50 µm sections, and every fifth section was mountedand stained for Nissl substance to assess the placement of recordingtracks, marker lesions, and placement of the cannulae in cortex.Recording sites were identified by reconstructing electrode tracks andmarker lesions as described25.

articles

seem to further eliminate retrograde degeneration as a possiblemechanism explaining our results.

Although some earlier electrophysiological studies of the nor-mal thalamus in monkeys have reported ‘extra-lemniscal’ neu-rons with large RFs11,12, such RFs were localized almostexclusively in the VPLo (oralis) nucleus11, far anterior to the VPLc(caudalis)14 where our recordings were made. In addition, record-ings lateral and medial to the affected thalamic zone indicatedneurons had normal RF size, further corroborating that we wererecording within VPLc, and not in the extralemniscal pathway.Finally, histological analysis from our cases confirmed that ourrecordings were from the hand region in VPLc. Importantly, therewas no evidence of any degeneration of neurons in the thalamusor cortex. Regardless of the precise mechanism(s) responsible forthe expansion, however, the magnitude of the RF expansion iscompletely unexpected and highlights the contribution of feed-back processing loops on RF properties.

Many previous studies of adult neural plasticity after periph-eral perturbations have focused attention on the earliest point inthe ascending pathway where plastic changes could occur (spinalcord15,16, brainstem17,18, or thalamus19–21), with the interpreta-tion that plastic changes at early stations are simply relayed tocortex. The timing, nature and magnitude of our present find-ings challenge this view of the system as a simple hierarchicalpathway22,23, providing a definitive demonstration of a dramat-ic and substantial role for the cortex on neuronal processing earlyin the somatosensory pathway. This study demonstrates thatsome RF characteristics within somatosensory pathways resultfrom a series of interconnected dynamic loops, with changes atany given level capable of triggering extensive changes in the RFsof neurons at both earlier and later stations in the processingchain. Such a view of feedback connections is entirely consistentwith a recent hypothesis24 that recognizes a substantial role forcorticothalamic feedback loops on the modification of receptiveproperties but not for the major driving of thalamic neurons bythe cortex. The extent to which such feedback connections mod-ulate activity and the precise mechanisms responsible for suchmodulation should be ripe areas for future research.

MethodsSURGICAL PROCEDURES. All procedures in the present study were approvedby the Wake Forest University Animal Care and Use Committee. All mon-keys used in this study were Macaca mulatta. Monkeys were anesthetized

Fig. 3. RF size in VPL before and after acute cortical D-APV administration. (a) Distribution histogram ofRF sizes for recording sites localized within the VPL hand representation before (pre) and after (post)acute cortical D-APV administration. RFs were categorized as in Fig. 2. (b) Representative RFs recordedfrom sites localized within VPL before (pre) and after (post) acute cortical D-APV administration.

% r

eco

rdin

g s

ites

Pre Post≤ 1 Pad

PrePost

> 1 Pad< 1 Digit ≥ 1 Digit

a b80

70

60

50

40

30

20

10

0


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


8. Conti, F. & Minelli, A. in Excitatory Amino Acids and the Cerebral Cortex. (edsConti, F. & Hicks, T.P.) 81–98 (MIT Press, Cambridge, Massachusetts, 1996).

9. Garraghty, P.E. & Muja, N. NMDA receptors and plasticity in adultprimate somatosensory cortex. J. Comp. Neurol. 367, 319–326 (1996).

10. Pons, T.P., Wall, J.T., Garraghty, P.E., Cusick, C.G. & Kaas, J.H.Consistent features of the representation of the hand in area 3b ofmacaque monkeys. Somatosensory Res. 4, 309–331 (1987).

11. Poggio, G.F. & Mountcastle, V.B. The functional properties ofventrobasal thalamic neurons studied in unanesthetized monkeys . J.Neurophysiol. 26, 775–806 (1963).

12. Loe, P.R., Whitsel, B.L., Dreyer, D.A. & Metz, C.B. Body representation inventrobasal thalamus of macaque: a single-unit analysis. J. Neurophysiol.40, 1339–1355 (1977).

13. Bava, A., Fadiga, E. & Manzoni, T. Extralemniscal reactivity andcommisural linkages in the VPL nucleus of cats with chronic corticallesions. Arch. Ital. Biol. 106, 204–226 (1968).

14. Kaas, J.H. & Pons, T.P. in Comparative Primate Biology, Neurosciences.(eds Steklis, H.P. & Erwin, J.) 421–468 (Alan R. Liss, New York, 1988).

15. Dostrovsky, J.O., Millar, J. & Wall, P.D. The immediate shift of afferentdrive of dorsal column nucleus cells following deafferentation: Acomparison of acute and chronic deafferentation in gracile nucleus andspinal cord. Exp. Neurol. 52, 480–495 (1976).

16. McMahan, S. B. & Wall, P.D. Plasticity in the nucleus gracilis of the rat.Exp. Neurol. 80, 195–207 (1983).

17. Pollin, B. & Albe-Fessard, P. Organization of somatic thalamus inmonkeys with and without section of dorsal spinal track. Brain Res. 173,431–449 (1979).

18. Garraghty, P.E. & Kaas, J.H.. Functional reorganization in adult monkeythalamus after peripheral nerve injury. Neuroreport 2, 747–750 (1991).

19. Merzenich, M.M. et al. Topographic reorganization of somatosensorycortical areas 3b and 1 in adult monkeys following restricteddeafferentation. Neuroscience 8, 33–55 (1983).

20. Merzenich, M.M. et al. Somatosensory cortical map changes followingdigit amputation in adult monkeys. J. Comp. Neurol. 224, 591–605 (1984).

21. Pons, T.P. et al. Massive cortical reorganization after sensorydeafferentation in adult macaques. Science 252, 1857–1860 (1991).

22. Pons, T.P., Garraghty, P.E., Friedman, D.P. & Mishkin, M. Physiologicalevidence for serial processing in somatosensory cortex. Science 237,417–420 (1987).

23. Pons, T.P. in Somesthesis and the Neurobiology of the somatosensory cortex,(eds Franzén, O., Johansson, R. & Terenius, L.)187–195 (Birkhäuser,Basel, Switzerland, 1996).

24. Crick, F. & Koch, C. Constraints on cortical and thalamic projections:the no-strong-loops hypothesis. Nature 391, 245–250 (1998).

25. Pons, T.P., Garraghty, P.E., Cusick, C.G. & Kaas, J.H.. The somatotopicorganization of area 2 in macaque monkeys. J. Comp. Neurol. 241,445–466 (1985).

articles

DATA ANALYSIS. Percent responsive sites was determined as the number ofresponsive sites per total number of sites. Percent large RF sites are givenas number of sites with RFs encompassing one or more digits (includ-ing RFs encompassing the hand and forearm) per number of hand-responsive sites. In the chronic D-APV animals the infusion zone wasdefined as the region of cortex surrounding the cannula site in whichresponsiveness was largely suppressed (90% of recording sites) and cor-responded to a 5 mm mediolateral expanse of cortex within the handrepresentation. The fringe zone was defined as the region in whichresponsiveness was only partially suppressed (25% of sites) and extend-ed approximately 1 mm medial and lateral to the infusion zone. Corti-cal recordings indicated that the face representation was not affected bythe D-APV administration.

AcknowledgementsThis research was supported by NIH grants MH11950-01, MH53369-02 and

NS35246-01.

RECEIVED 20 MAY: ACCEPTED 26 MAY 1998

1. Liu, X.B., Honda, C.N. & Jones, E.G. Distribution of four types of synapse onphysiologically identified relay neurons in the ventral posterior thalamicnucleus of the cat. J. Comp. Neurol. 352, 69–91 (1995).

2. Jones, E.G. The Thalamus. (Plenum, New York, 1985).3. Yuan, B., Morrow, T.J. & Casey, K.L. Responsiveness of ventrobasal thalamic

neurons after suppression of S1 cortex in the anesthetized rat. J. Neurosci. 5,2971–2978 (1985).

4. Ghosh, S., Murray, G.M., Turman, A.B. & Rowe, M.J. Corticothalamicinfluences on transmission of tactile information in theventroposterolateral thalamus of the cat: effect of reversible inactivation ofsomatosensory cortical areas I and II. Exp. Brain. Res. 100, 276–286 (1994).

5. Burchfiel, J.L. & Duffy, F.H. Corticofugal influences upon cat thalamicventrobasal complex. Brain Res. 70, 395–411 (1974).

6. Shin, H.C. & Chapin, J.K. Mapping the effects of SI cortex stimulation onsomatosensory relay neurons in the rat thalamus: direct responses andafferent modulation. Somatosens. Motor Res. 7, 421–434 (1990).

7. Tsumoto, T. & Nakamura, S. Inhibitory organization of the thalamicventrobasal neurons with different peripheral representations. Exp. BrainRes. 21, 195–210 (1974).


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

The primary motor area (M1), a cortical region necessary forskilled voluntary movements, seems also to participate in learn-ing motor skills. This conclusion is based largely upon findingsthat adult M1 representations are modifiable1–4, that dendriticmorphology of M1 pyramidal neurons is altered by experience5

and that connections among M1 neurons are capable of activi-ty-dependent, long-term changes in efficacy6–11. Despite thesesuggestive findings, there is no direct evidence that learning isaccompanied by functional modifications of M1 circuits. Modi-fications outside of the cortex, which have been repeatedly doc-umented, might account for cortical motor or sensory mapchanges that can be produced by experience or nerve lesions12,13.Further, there is no evidence that morphological changes, suchas an increase in the number of dendritic branches, actually alterfunctional interactions in the cortex. Finally, there has been nodefinitive evidence that LTP-like mechanisms are engaged with-in cortex during any form of learning. Demonstration of synap-tic modification in conjunction with learning is an essential steptowards understanding how cortical circuits support motor skillsor other forms of learning.

The intrinsic horizontal pathways are a potential substratefor experience-dependent reorganization of relationships amongM1 neurons. Layer II/III pyramidal cells form a broad, intrin-sic horizontal projection system in M1, and their intracorticalpattern correlates with sites that reorganize after nervelesions14,15. Pharmacological adjustments of the excitatory-inhibitory balance within M1 restructures motor representa-tions, apparently by uncovering latent horizontal pathways16. Inaddition, horizontal connections are capable of LTP, providing apotential activity-dependent mechanism for synaptic modifica-tion3,7,9,10,17. These findings raise the possibility that changes inhorizontal connection strength may accompany motor learn-ing. Here we show that field potentials evoked by stimulation of

rat M1 horizontal connections increase after learning and prac-ticing a skilled reaching task. The amount of LTP that could beinduced by electrical stimulation was also reduced after learn-ing, implying that the observed strengthening of horizontal con-nections may involve an LTP-like mechanism. Plasticity of M1connections may therefore create cortical circuits needed toacquire or perform new motor behaviors.

ResultsRats were trained to reach through a hole in a food box with asingle forepaw in order to retrieve small food pellets using agrasping motion. Training and subsequent practice lasted three(n = 1) or five (n = 13) successive days with one training session(approximately one hour) per day. Successful performance ofthis skill occurred in the first one or two sessions, and the remain-ing sessions consisted of repeated practice and refinement of theskill. By the final two days of training, all rats achieved a perfor-mance of about 1.5 pellet retrievals per minute, with few errors inthe reach, grasp or retrieval actions. A group of comparably han-dled, age- and sex-matched cage mates (termed ‘paired controls’,n = 14) and another group of naive rats (‘unpaired controls’,n = 12) served as controls.

The strength of intrinsic horizontal synaptic connectionswithin layers II/III was evaluated ex vivo using coronal brainslices containing both hemispheres (Fig. 1a); experimenters wereblind to whether the animal was trained, and if so, which fore-limb was trained. Slices were taken 20–45 hours after the lasttraining session to rule out effects that might persist immedi-ately after practice of the task. Field-potential responses evokedin the horizontal pathway by electrical stimuli were examinedsimultaneously through stimulating and recording electrodesthat were mirror symmetrically positioned in layer II/III of theleft and right M1 (Fig. 1a). Thus, one side in each trained rat

Strengthening of horizontal corticalconnections following skill learning

Mengia-S. Rioult-Pedotti1, Daniel Friedman1, Grzegorz Hess1,2 and John P. Donoghue1

1 Department of Neuroscience, Brown University, Providence, Rhode Island 02912, USA2 Permanent address: Jagiellonian University, Institute of Zoology, 31-102 Krakow, Poland

Correspondence should be addressed to M.-S. R.-P. ([email protected])

Learning a new motor skill requires an alteration in the spatiotemporal pattern of muscle activation.Motor areas of cerebral neocortex are thought to be involved in this type of learning, possibly byfunctional reorganization of cortical connections. Here we show that skill learning is accompaniedby changes in the strength of connections within adult rat primary motor cortex (M1). Rats weretrained for three or five days in a skilled reaching task with one forelimb, after which slices of motorcortex were examined to determine the effect of training on the strength of horizontal intracorticalconnections in layer II/III. The amplitude of field potentials in the forelimb region contralateral tothe trained limb was significantly increased relative to the opposite ‘untrained’ hemisphere. Nodifferences were seen in the hindlimb region. Moreover, the amount of long-term potentiation (LTP)that could be induced in trained M1 was less than in controls, suggesting that the effect of trainingwas at least partly due to LTP-like mechanisms. These data represent the first direct evidence thatplasticity of intracortical connections is associated with learning a new motor skill.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Fig. 1. Conse-quences of motorskill learning onf i e l d - p o t e n t i a lresponses evoked inlayer II/III horizontalconnections of M1.(a) Mirror-symmet-ric placement ofstimulating (stim)and recording (rec)m i c roe l e c t rodesbilaterally in layersII/III of M1 in a coronal slice containing both hemispheres. wm, white matter. (b) Single-case examples of field potentials (averages of fivesweeps), evoked at 60% maximum stimulation intensity from a single trained (top) and a single paired-control (bottom) animal. Dark linesrepresent the trained M1 or left M1, hatched lines, the untrained M1 or right M1. (c) Group average responses for trained (top, n = 7) andcontrol (bottom, n = 20, paired and naive) rats at 60% maximal stimulation intensity, illustrating enhanced field potential in the horizontalpathway of M1 contralateral to the limb used in the reaching task. Same format as (b).

provided a within-animal control because the majority ofengaged M1 neurons are located in the hemisphere contralater-al to the limb they influence18. For trained animals, we term M1contralateral to the trained limb the ‘trained M1’, and its coun-terpart on the other side the ‘untrained M1’. In all control ani-mals, the terms ‘left M1’ and ‘right M1’ are used.

Stimulation evoked an initially negative-going field potentialof similar shape in all rats as previously described7. However,for each rat that had learned the skilled-reaching task, fieldpotentials evoked in the trained M1 were consistently larger inamplitude than in the untrained M1 (Fig. 1b and c). Ampli-tudes in the trained M1 were also larger than those observedfor the control animals. Amplitude differences between trainedand untrained M1 were not a result of stimulus intensitybecause absolute current intensities used on the two sides werenot significantly different (p = 0.23). Indeed, in 71% of thecases, the stimulation intensity was slightly larger(27.24 ± 3.38%, n = 10) on the untrained side.

The amplitude differences between trained M1 anduntrained M1 were specific to the region of the M1 forelimbrepresentation. Layer II/III field-potential measurements fromthe hindlimb region of the trained M1 and untrained M1 in anadditional group of trained rats showed no significant side-to-side amplitude differences at any stimulation intensity(p = 0.2–0.8, n = 9), whereas slices taken from the forelimbregion of the same animals showed a larger response in thetrained than untrained M1. Field potentials in the hindlimband forelimb areas were similar in shape. Peak amplitudes forthe hindlimb at 2.5 and 5 times threshold intensity were0.92 ± 0.12 mV and 1.47 ± 0.19 mV in the trained M1 and0.97 ± 0.13 mV and 1.45 ± 0.19 mV in the untrained M1.

The relationship between stimulus intensity and responseamplitude was evaluated systematically using two differentapproaches (Fig. 2a–c) to rule out the possibility that these effectswere a consequence of the particular intensities used. One series of7 trained and 20 control (8 paired, 12 unpaired) rats was testedwith stimuli that were a constant fraction of the stimulus inten-sity evoking a maximum response (absolute intensity less than orequal to 220 µA). A second series of seven trained and six pairedcontrol rats was tested with stimuli that were a constant multipleof the stimulus intensity evoking a minimal response of about 0.1mV (absolute threshold intensity 12–30 µA). In both series, theaverage responses for the trained M1 were larger than the

untrained M1 at every stimulation intensity, whereas there were nodifferences between left and right sides in the control groups(Fig. 2a and b). To compare the relative change between trainedand untrained M1, the common logarithm of the response-ampli-tude ratio between sides was calculated for each animal at eachstimulation intensity. The average log ratio obtained for trainedanimals was significantly different from zero (p < 0.05) over abroad range of stimulus intensities, reflecting larger peak ampli-tudes in the trained M1. This ratio was not different in controlanimals (p > 0.05; Fig. 2c). Figure 2d plots the distribution of theentire data set; the trained animals show a marked skew towardlarger responses on the trained side, whereas untrained controlsshow no difference between the two hemispheres.

Amplitude differences observed in the trained M1 might ariseby several mechanisms. Larger responses in the horizontal path-ways could result from newly formed synapses, from postsynap-tic excitability increase or from engaging an LTP-like mechanismthat increased the strength of existing synapses. A larger initialslope for field potentials evoked at 60% maximal stimulationintensity in the trained M1 (trained M1, 0.735 ± 0.079 mV perms; untrained M1, 0.439 ± 0.079 mV per ms; p < 0.025) suggest-ed that modifications occurred by a change in synaptic efficacy,rather than an excitability change. Because layer II/III horizontalconnections are capable of LTP, we compared the ability to poten-tiate this pathway in trained M1 and untrained M1. We postulat-ed that if learning recently engaged an LTP-like process in thetrained pathway, further electrically induced potentiation of thispathway might be occluded. LTP induction was attempted bytheta-burst stimulation simultaneously delivered during a tran-sient reduction of GABAA-receptor-dependent synaptic inhibi-tion by focal application of small amounts of bicuculline at eachof the two recording sites6,7, an established method to obtain LTPreliably in neocortex. In accord with previous results7, LTP wasreadily induced in the untrained M1 (24.1 ± 11.7% increase, n = 6)and in both hemispheres of control rats (left M1, 39.7 ± 10.5%;right M1, 36.1 ± 12.8%; p = 0.8, n = 7). The difference betweenuntrained M1 and left and right M1 of controls was not signifi-cant (p = 0.35). In contrast, the average increase in field potentialamplitude in the trained M1 was only 6.5 ± 4.2% (n = 6; Fig. 2e),which was significantly less than the untrained M1 (p<0.035,n = 6). Thus, although identical stimulation and recording pro-cedures were applied to both hemispheres, the amount of LTPthat could be induced in the trained hemisphere was reduced.

a b c

Control Controln=20

TrainedTrained

n=7 rec rec

stimstim


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

DiscussionThese results demonstrate that learning and practicing a motorskill is accompanied by an increased efficacy of horizontal con-nections in motor cortex. Although there is extensive data show-ing that motor skill learning modifies corticalrepresentations1,2,19,20 and alters dendritic morphology5 and thatcortical connections are capable of activity-dependent strengthchanges, our results provide the first direct evidence for a func-tional change of a cortical connection associated with motor skilllearning. Our results compare with changes recently observed inthe amygdala following fear conditioning, a markedly differentform of learning21,22 and might provide a basis for correlation-strength changes that have been observed during auditory con-ditioning23. Plasticity of horizontal connections could contributeto the reorganization of motor cortical representations thataccompanies motor skill learning, because information from oneregion of M1 would be spread more effectively to other regions.This hypothesis is consistent with recent findings demonstratingthat only the parts of M1 receiving strong horizontal inputs reor-ganize immediately after nerve lesions in the rat14 and that M1representations in monkeys and humans enlarge or rearrangeduring motor skill learning4,24–26. Our data cannot differentiatebetween changes that result from the early modifications in

behavior that occur when the task is first achieved and the slow-er improvements in skill that occur with subsequent practice.Both can be considered as forms of learning. Increased efficacy ofhorizontal connections would not seem to be a consequence ofmovement alone. Although the number of movements might bedifferent for the practiced limb, the actual number is small whenconsidered as a fraction of the total number of movements madeover the one to two days after training ended. The movementsmade were at a low rate (about 1.6 per minute), required littleforce and were mixed with many overlearned bilateral move-ments in consummatory actions, yet enhanced horizontal-con-nection strength persisted in M1 in the region of the forelimb.Changes in horizontal connections are also not widespread inM1 because learning did not modify the layer II/III horizontalpathway in the hindlimb area.

The marked effect of learning upon field-potential amplitudewithin the M1 forelimb region suggests that a large number ofconnections within this area have been modified in conjunctionwith skill learning. It is difficult to conceptualize how such a gen-eralized effect can provide a substrate for implementing thedetailed pattern of movements acquired. Studies of humanmotor-skill learning using transcranial magnetic stimulation sug-gest that motor-cortical maps shrink again after explicit knowl-

Fig. 2. Response differ-ences in trained andcontrol rats. (a) Field-potential amplitudes fortrained (top, n = 7) andpaired- and naive-con-trol (bottom, n = 20)animals. Stimulationintensities are plottedas a percentage of thestimulus evoking themaximal response.Filled symbols repre-sent the trained M1 orleft M1, open symbolsthe untrained M1 orright M1. Note thelarger response magni-tude in the trained M1across intensities. (b) Field-potentialamplitudes for trained(top, n = 7) and paired-control (bottom, n = 6)animals. Stimulationintensities were variedas multiples of thresh-old intensity inducing aminimal response(about 0.1 mV).Symbols as in (a). (c) Comparison of average (± standard error) log-response ratios for trained/untrained M1 (filled symbols) and left/right M1 (open symbols)for series in (a) (left) or series in (b) (right). Asterisks indicate statistical differences between control and experimental ratios. (d) Log ratio offield-potential amplitudes for trained (top) and control (bottom) groups. The histograms show the distributions of all log ratios for all rats at allstimulation intensities (entire data set). The mean value for trained animals (0.17) is significantly different from zero (p<0.001), demonstratinga shift towards the trained M1. (e) Effect of training on electrically induced LTP. Each point indicates the relative field-potential amplitudebefore and after LTP induction in both hemispheres (arrow) in trained (top, n = 6) and paired control (bottom, n = 7) animals. Averages afterLTP induction are compiled from the last 50 min of recordings because of a variable duration of the transient bicuculline effect (interruption inthe x-axis). Open symbols (+ standard error) for right M1 or untrained M1, filled symbols (- standard error) for left M1 or trained M1.

a b

c

d

e

Trained Trained

Trained

Trained

Control Control

Control

Control

Stimulation intensity

(% max) (x threshold)

Stimulation intensity

(% max) (x threshold)

log (trained M1/untrained M1)

log (left M1/right M1)

Time

Log

rat

ioFP

am

plit

ud

e (m

V)

FP a

mp

litu

de

(% b

asel

ine)


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

edge of the task is gained25 . Thus, the seemingly generalizedchanges we observe after five days of training may eventually leadto more specific circuits suitable for producing the skill. By con-trast, fMRI studies during human motor-skill learning indicatethat representations continue to enlarge with repeated prac-tice24,26. We will need to examine additional time points to deter-mine what happens to field-potential increases and decrements inLTP before we can make more definitive statements of the roleof these modification in acquiring new motor skills.

Modifications in horizontal connections seem to result fromchanges in synaptic efficacy, perhaps through LTP-like mecha-nisms, rather than other means. Learning seems to occlude LTPinduction, suggesting they share a similar mechanism. Theincrease in the initial slope of field potentials from the trainedM1 is consistent with the hypothesis that learning occurredthrough synaptic modification. Although compelling evidenceis lacking, LTP is a strong candidate mechanism for many formsof learning because it can lead to long-lasting modification ofactivated synapses27 at a number of sites, including horizontalcortical connections7,17. Other mechanisms are potentially plau-sible but are not fully consistent with our results. For instance,a different form of learning, involving classical conditioning,leads to increases in the membrane excitability of M1 neurons28,rather than synaptic modification. If increases in horizontalfield potentials were due to excitability changes in postsynap-tic neurons, no initial slope changes would be expected andtetanization would likely produce greater postsynaptic depo-larization and hence, a larger amount of LTP29. Growth of newsynaptic connections, which has been suggested to occur inadult cortex30,31, could also lead to larger responses after learn-ing. However, these new synapses would have to be unable toundergo LTP to be consistent with the finding of less LTP in thetrained M1. We therefore think it unlikely that these mecha-nisms underlie the increased field-potential amplitudes seenafter skill learning, although additional studies are required toidentify the exact mechanism involved.

We only examined the horizontal pathway after three or fivedays of practice following skill acquisition. Although synapticmodification seems to occur during this period, other mecha-nisms may operate during initial skill acquisition or during laterskill improvement. Intracellular studies and measurements of thetime course of change may help to clarify what leads to such dra-matic increases in this pathway’s efficacy. It is likely that this effectis not peculiar to M1. Plasticity of synapses formed by horizon-tally oriented axon collaterals may operate throughout many areasof the cerebral cortex to restructure various representation pat-terns. For example, within visual cortex, filling-in phenomenaand reorganization of visual receptive fields after lesions or otherperturbations appear to be mediated via horizontal connec-tions32. The common occurrence of these connections in all cor-tical areas suggests that plasticity of synapses formed by horizontalpathways may be an important contributor to learning-relatedprocesses throughout the cerebral cortex.

MethodsAnimals were cared for in accordance with National Institute of Healthguidelines for laboratory animal welfare. All experiments were approvedby the Brown University Institutional Animal Care and Use Committee.Forty adult female Sprague-Dawley rats (150–225 g) housed on a nor-mal 12:12 light-dark cycle were used for this study. Twenty-eight animalswere housed in pairs and were food restricted to maintain their bodyweight at roughly 85% of their free-feeding weight. Water was providedad libitum. One rat from each pair was placed in an operant test cage(22.8 cm cube), which contained a Plexiglas food box (3.2 x 4.5 x 5 cm)

with a 1.3 cm diameter hole through which the food was retrieved. Smallfood pellets (45 mg; Noyes Precision Food Pellets) were placed on thefloor of this food box within reaching distance. Rats learned to reach intothe food box with their preferred paw to retrieve food pellets using agrasping action. It was impossible for the rats to reach in the food boxwith both paws, although initial attempts sometimes involved trying toreach with both paws or attempting to place the snout in the food boxand reach with the tongue. All but one animal, which was excluded fromthe analysis, selected the right forelimb to perform the task. Animalsreceived one training session per day lasting for one hour. The trainingand practice period lasted five days for thirteen rats and three days forone rat. Because there was no readily apparent difference in the reach-ing behavior nor in the electrophysiological results after the three- andfive-day training, data were grouped together. The second animal from apair served as a paired control, received a comparable amount of han-dling and was given similar numbers of food pellets. The remainingtwelve animals were used as naive controls.

The experimenter was unaware of the rats’ training condition untildata analysis of the pair was completed. Coronal brain slices containingthe region of the M1 forelimb representation33, 1–2 mm anterior to thebregma, were prepared as described7 and superfused with artificial cere-brospinal fluid (ACSF) of the following composition (in mM): 126 NaCl,3 KCl, 1.25 NaH2PO4, 26 NaHCO3, 1 MgSO4, 2 CaCl2, and 10 glucose,bubbled with a 95% O2, 5% CO2 mixture at 35 ± 0.5°C. The humidifiedatmosphere over the slices was saturated with 95% O2, 5% CO2. Coro-nal slices from the hindlimb region were cut at the level of the anteriorpart of the hippocampal formation.

Field potentials were recorded using glass micropipettes placed in layerII/III, 200–350 µm below the pial surface in the region of the M1 fore-limb representation (2–2.2 mm lateral to the midline). Concentric bipo-lar stimulating electrodes were displaced horizontally by 500 µm fromeach recording electrode (Fig. 1a). Consistent mirror-symmetrical place-ment of the electrodes at identical locations in both hemispheres wasachieved with a reticle. Slices were not attached by the corpus callosum,but remained attached to each other during tissue slicing. For stimula-tion, constant current pulses (0.2 ms) were delivered at 0.033 Hz. Weused the amplitude of the field potential evoked in the layer II/III hori-zontal pathway to measure of the population excitatory synaptic responsebecause it reflects a monosynaptic current sink9 and correlates well withintracellular excitatory postsynaptic potentials evoked in this pathway7.Input–output curves for a range of stimulation intensities were con-structed for each stimulation–recording pair. In the first set of experi-ments, we averaged three to five sweeps evoked at 20, 40, 60, 80, 100 and120% of the intensity inducing a maximum response. In the second set,we averaged three responses to stimuli of 2, 2.5, 3, 3.5, 4, 4.5 and 5 timesthe intensity that evoked a 0.1 mV (threshold) response. Stimulation didnot influence the contralateral hemisphere because the slices used didnot contain the corpus callosum. Field-potential peak amplitudes werecalculated from averages of three to five waveforms, and the commonlogarithm of the left /right ratio was calculated for each animal. Becausethe log of this ratio fit a Gaussian distribution, parametric testing wasused (paired t-test). Similar input-output curves were constructed forfield potentials recorded from the hindlimb area of trained animals.

After establishing a 20 min period of stable response amplitudes usinga stimulation intensity 50-60% of maximum, LTP induction was attempt-ed with an established and reliable protocol for MI7. Prior to tetanic stim-ulation, the GABAA receptor antagonist bicuculline (3.5 mM) was appliedwithin 100 µm from the recording electrodes using a glass pipette; thetime of application on each side was separated by less than two minutes,and the same pipette was used for each side of a slice. The bicucullinepipette was retracted as soon as field-potential responses to test stimu-lation increased to about 150–200% of baseline (typically within 10–30 s).Immediately following bicuculline application, LTP was attempted bydelivering theta-burst stimulation (5 sequences of 10 bursts 10 secondsapart; 1 burst is 5 pulses at 100 Hz) at double test intensity simultane-ously through both stimulating electrodes7. The LTP effect was recordedfor at least 30 min after it reached a stable plateau.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


15. Donoghue, J.P. Limits of reorganization in cortical circuits. Cereb. Cortex 7, 97–99(1997).

16. Jacobs, K. & Donoghue, J. Reshaping the cortical map by unmasking latentintracortical connections. Science 251, 944–947 (1991).

17. Hirsch, J. & Gilbert, C. Long-term changes in synaptic strength along specificintrinsic pathways in the cat’s visual cortex. J. Physiol. (Lond) 461, 247–262(1993).

18. Price, A.W. & Fowler, S.C. Deficits in contralateral and ipsilateral forepawmotor control following unilateral motor cortical ablation in rats. Brain Res.205, 81–90 (1981).

19. Garraghty, P.E. & Kaas, J.H. Dynamic features of sensory and motor maps.Curr. Opin. Neurobiol. 2, 522–527 (1992).

20. Sanes, J.N. & Donoghue, J.P. Motor areas of the cerebral cortex. J. Clin.Neurophysiol. 11, 382–396, (1994).

21. Rogan, M.T., Staeubli, U.V. & LeDoux, J.E. Fear conditioning induces associativelong-term potentiation in the amygdala. Nature 390, 604–607 (1997).

22. McKernan, M.G. & Shinnick-Gallagher, P. Fear conditioning induces a lastingpotentiation of synaptic currents in vitro. Nature 390, 607–611 (1997).

23. Ahissar, E. et al. Dependence of cortical plasticity on correlated activity ofsingle neurons and on behavioral context. Science 257, 1412–1415 (1992).

24. Grafton, S.T. et al. Functional anatomy of human procedural learningdetermined with regional cerebral blood flow and PET. J. Neurosci. 12,2542–2548 (1992).

25. Pascual-Leone, A., Grafman, J. & Hallett, M. Modulation of cortical motoroutput maps during development of implicit and explicit knowledge. Science263, 1287–1289 (1994).

26. Karni, A. et al. Functional MRI evidence for adult motor cortex plasticityduring motor skill learning. Nature 377, 155–158 (1995).

27. Morris, R.G.M., Davis, S. & Butcher, P. in Long Term Potentiation: A Debate ofCurrent Issues (eds Baudry, M. & Davis, J.L.) 267–300 (MIT Press,Cambridge, 1991).

28. Woody, C.D., Gruen, E. & Birt, D. Changes in membrane currents duringPavlovian conditioning of single cortical neurons. Brain Res. 539, 76–84(1991).

29. Yoshimura, Y. & Tsumoto, T. Dependence of LTP induction on postsynapticdepolarization: a perforated patch-clamp study in visual cortical slices ofyoung rats. J. Neurophysiol. 71, 1638–1645 (1994).

30. Darian, S.C. & Gilbert, C.D. Axonal sprouting accompanies functionalreorganization in adult cat striate cortex. Nature 368, 737–740 (1994).

31. Kleim, J.A., Lussnig, E., Schwarz, E.R., Comery, T.A. & Greenough, W.T.Synaptogenesis and Fos expression in the motor cortex of the adult rat aftermotor skill learning. J. Neurosci. 16, 4529–4535 (1996).

32. Gilbert, C.D. Rapid dynamic changes in adult cerebral cortex. Curr. Opin.Neurobiol. 3, 100–103 (1993).

33. Donoghue, J.P. & Wise, S.P. The motor cortex of the rat: cytoarchitecture andmicrostimulation mapping. J. Comp. Neurol. 212, 76–88 (1982).

articles

AcknowledgementsWe thank Drs. Barry Connors, Mark Bear, and Marc G. Rioult for critical

comments. This work was supported by NIH grant NS22517. G. H. is an

international scholar of the Howard Hughes Medical Institute.

RECEIVED 21 JANUARY: ACCEPTED 26 MAY 1998

1. Cohen, L.G., Brasil, N.J., Pascual-Leone, L.A. & Hallett, M. Plasticity ofcortical motor output organization following deafferentation, cerebrallesions, and skill acquisition. Adv. Neurol. 63, 187–200 (1993).

2. Donoghue, J.P. Plasticity of sensorimotor representations. Curr. Opin.Neurobiol. 5, 749–754 (1995).

3. Donoghue, J.P., Hess, G. & Sanes, J.N. in Acquisition of Motor Behavior (eds.Bloedel, J., Ebner, T. & Wise, S.P.) 363–386 (MIT Press, Cambridge, 1996).

4. Nudo, R.J., Milliken, G.W., Jenkins, W.M. & Merzenich, M.M. Use-dependent alterations of movement representations in primary motor cortexof adult squirrel monkeys. J. Neurosci. 16, 785–807 (1996).

5. Greenough, W.T., Larson, J.R. & Withers, G.S. Effects of unilateral andbilateral training in a reaching task on dendritic branching of neurons in therat motor-sensory forelimb cortex. Behav. Neural Biol. 44, 301–314 (1985).

6. Hess, G. & Donoghue, J.P. Long-term potentiation of horizontal connectionsprovides a mechanism to reorganize cortical motor maps. J. Neurophysiol. 71,2543–2547 (1994).

7. Hess, G., Aizenman, C.D. & Donoghue, J.P. Conditions for the induction oflong-term potentiation in layer II/III horizontal connecitons of the rat motorcortex. J. Neurophysiol. 75, 1765–1778 (1996).

8. Hess, G. & Donoghue, J.P. Long-term depression of horizontal connectionsin rat motor cortex. Eur. J. Neurosci. 8, 658–665 (1996).

9. Aroniadou, V.A. & Keller, A. Mechanisms of LTP induction in rat motorcortex in vitro. Cereb. Cortex 5, 353–362 (1995).

10. Castro-Alamancos, M.A., Donoghue, J.P. & Connors, B.W. Different forms ofsynaptic plasticity in somatosensory and motor areas of the neocortex. J.Neurosci. 15, 5324–5333 (1995).

11. Asanuma, H. & Pavlides, C. Neurological basis of motor learning inmammals. Neuroreport 8, i–vi (1997).

12. Kaas, J.H. Plasticity of sensory and motor maps in adult mammals. Annu.Rev. Neurosci. 14, 137–167 (1991).

13. Merzenich, M.M., Recanzone, G., Jenkins, W.M., Allard, T.T. & Nudo, R.J. inNeurobiology of Neocortex (eds Rakic, P. & Singer, W.) 41–67 ( Wiley, NewYork, 1988).

14. Huntley, G.W. Correlation between patterns of horizontal connectivity andthe extent of short term representational plasticity in rat motor cortex. Cereb.Cortex 7, 143–156 (1997).


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


In Old World primates such as macaque monkeys and humans,visual information about color is processed in anatomicallysegregated columns, layers, channels or areas. It is importantto know to what extent color is processed in separate versusconvergent visual information pathways, because the addeddimension of color is so rich in visual information. For exam-ple, we can discriminate about fifteen hundred different levelsof luminance1, whereas we can make several million discrimi-nations if we also consider variations in color2. It is likely thatthis glut of color information is incorporated into the labeledlines of the neural architecture in some organized way.

In macaque monkeys, an anatomical segregation between chro-matic-opponent versus achromatic-opponent cells has been report-ed as early as the lateral geniculate nucleus. Color-specific anatomicalsegregation has also been described in primary (V1) and secondary(V2) visual cortex. In V1, prominent populations of color-selectivecells have been reported in specific layers3–5 and in the cytochrome-oxidase blobs4–6, though the latter claim has been disputed7,8. Sim-ilar (and equally controversial) claims have been made about theprominence of color-opponent cells in the ‘thin’ stripes in area V2,to which the V1 blobs project (ref. 9,10, but see 11).

However, the most prominent controversy about theanatomical segregation of color-selective neurons occurs at ahigher level, in cortical area V4. According to different reports,a high percentage of color-selective cells is either present12–15 orabsent16 in the largest and best-studied portion of that area,dorsal V4 (V4d). A high percentage of color-selective cells hasnot been reported in the smaller, ventral subdivision of V4(V4v). More recent evidence suggests that brain mechanismscritical for color selectivity are located not in macaque V4, butrather in areas anterior to it (ref. 17–19, Vanduffel et al. Soc.Neurosci. Abstr. 23, 845, 1997).

This controversy about color selectivity in V4 has now beenextended to human visual cortex. Based on human neuroimag-

ing studies, a small patch of color-selective activity near the mid-dle of the collateral sulcus has been named ‘V4’ (ref. 20–22).This choice of name presupposes that (1) an area homologous tomacaque V4 exists in humans, (2) V4 is color-selective, and (3)this region in or near the collateral sulcus is the macaque V4homolog. However, in humans, the location of this color-selec-tive region has not yet been compared with the map of retino-topic areas, to see whether color selectivity is really is in aretinotopically defined human area V4. Furthermore, the degreeof color selectivity in macaque V4 is itself controversial17–19.

This issue is not just of academic interest. In an intriguingclinical syndrome (‘achromatopsia’), human patients reportthat the visual world becomes colorless following damage to acortical region that apparently includes this color-selectivearea in the collateral sulcus23–25. This suggests that the con-scious percept of ‘color’ involves that area, although it isknown that physical wavelength-dependent differences arecoded throughout prior levels of the visual system as well. Ifwe can define better which area this is in humans, we can learnsomething about where conscious perceptions of color arise.Accurate localization in humans should also make it possibleto study the homologous area in macaques using more inci-sive (but invasive) classical neurobiological techniques.

We have attempted to clarify these issues in humans usingfunctional magnetic resonance imaging (fMRI). Technical detailswere similar to those described elsewhere26, except that here wemanipulated the color content of the visual stimuli. We also useda high-field MRI scanner and other improvements to substan-tially increase the sensitivity of the retinotopic maps (Methods).

ResultsCOLOR- VERSUS LUMINANCE-VARYING STIMULI

First, we compared the activity produced by color variations tothat produced by variations in luminance, in the same sub-

articles

Retinotopy and color sensitivity inhuman visual cortical area V8

Nouchine Hadjikhani, Arthur K. Liu, Anders M. Dale, Patrick Cavanagh and Roger B. H. Tootell

Nuclear Magnetic Resonance Center, Massachusetts General Hospital, 149 13th Street, Charlestown, Massachusetts 02129, USA

Correspondence should be addressed to N.H. ([email protected])

Prior studies suggest the presence of a color-selective area in the inferior occipital-temporal regionof human visual cortex. It has been proposed that this human area is homologous to macaque areaV4, which is arguably color selective, but this has never been tested directly. To test this model, wecompared the location of the human color-selective region to the retinotopic area boundaries in thesame subjects, using functional magnetic resonance imaging (fMRI), cortical flattening andretinotopic mapping techniques. The human color-selective region did not match the location ofarea V4 (neither its dorsal nor ventral subdivisions), as extrapolated from macaque maps. Insteadthis region coincides with a new retinotopic area that we call ‘V8’, which includes a distinctrepresentation of the fovea and both upper and lower visual fields. We also tested the response tostimuli that produce color afterimages and found that these stimuli, like real colors, caused preferen-tial activation of V8 but not V4.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


jects (Fig. 1). Our stimuli consisted of slowly moving sinu-soidal radial gratings (‘pinwheels’) of low spatial frequency,defined by either color or luminance contrast (Methods). Asshown earlier in V1 and V2 (ref. 27), we found that bothcolor- and luminance-varying stimuli produced robust acti-vation in many areas of visual cortex, when compared with auniform gray field (data not shown).

Here we focused on those locations where the color stimuliproduced more activation than the luminance stimuli. In theclassically retinotopic visual areas (V1, V2, V3/VP, V3A andV4v), we found prominent color-selective activation in the rep-resentations of the fovea (center of gaze) but not in peripheralrepresentations (Fig. 1). A foveal color bias has not been report-ed in previous imaging studies, perhaps because it is more obvi-ous in our flattened maps. However, such a foveal color bias isconsistent with the well known predominance of cone pho-toreceptors, and the corresponding absence of rods, in the fovea

of the retina. A similar foveal color bias is found in routine clin-ical perimetry and in numerous psychophysical studies.

In 25 of 26 hemispheres (13 subjects) tested, we found anadditional region that responded preferentially to color, locat-ed midway along the length of the collateral sulcus. Based onthe anatomical location and the nature of the functional com-parison used here, this collateral color-selective patch appearsequivalent to the previously reported area involved in achro-matopsia20,22, which has been proposed as the human homo-logue of macaque area V4 (refs 20–22, 28–30).

However, when we compared the location of that collateralcolor-selective patch to the retinotopic borders in the same sub-jects, we found that the color-selective patch was consistentlylocated just beyond the most anterior retinotopic area definedpreviously, area V4v. Earlier reports26,31–33 suggested thathuman V4v is a quarter-field representation of the contralater-al upper visual field. The more sensitive retinotopic mapping

articles

Fig. 1. Topography ofcolor-selective activity inhuman visual cortex. (a, b) The inferior,‘inflated’ cortex, withposterior to the left andanterior to the right, intwo subjects. (c, d) Theposterior portion ofcortex in fully flattenedformat, another view ofthe same data shown in(a) and (b), respectively.In all panels, gyri fromthe original brain areshown as light gray andsulci as dark gray. Thefundus of the collateralsulcus (cs) is indicatedby the dashed black line.The borders of previ-ously described retinotopic areas (V1, V2, V3, VP, V3A, and V4v) are indicated in white (horizontal meridians, solid lines, upper verticalmeridians, dotted lines, lower vertical meridians, dashed lines). Typically, color-varying stimuli produced relatively higher activation in thefoveal representation of V1 and often V2 and V3/VP and a distinctive patch of color-selective activation approximately midway in the collat-eral sulcus. When present, the latter patch was always located just anterior to the horizontal meridian representation marking the anteriorborder of area V4v, rather than within V4v.

Fig. 2. Retinotopic features of area V8 by fMRImapping. (a–c) Retinotopy of polar angle in the infe-rior row of cortical areas, from three flattenedhemispheres. From left to right, each panel showsthe representations of the contralateral upper quar-ter field (red through blue or vice versa; seepseudocolor logo) in inferior V1, then inferior V2,then VP, then V4v. To the right of (anterior to) V4vis the distinctive half-field representation comprisingV8 (green through blue through red, from upper tolower in this figure). (d) Retinotopic representationof eccentricity (the other dimension in polar space),from the same hemisphere shown in (c). The repre-sentations of central-through-more-peripheraleccentricities are coded in red-through-blue-through-green, respectively (see pseudocolor logo,bottom right). The representations of the center of gaze are indicated with an asterisk. Area V8 has its own representation of the fovea, quitedistinct (and 3.5 cm) from the foveal representation in adjacent area V4v.

a b

c d

a b

c d


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


methods used here confirmed that V4v represents just thatquarter field, with its foveal representation located superiorlyalongside that of adjacent areas V3/VP (Figs 2 and 3).

The improved retinotopic methods also revealed additionalretinotopic features anterior to V4v. Taken together, these fea-tures indicate the presence of an additional retinotopic map,comprising a previously undifferentiated cor-tical area that we call ‘V8’. This continues thenaming scheme begun by Zeki and colleagues,who identified areas V1 through V612–15,21,22,34.(We also identified an area ‘V7’, which is a rep-resentation of the contralateral lower visualfield anterior to human V3A.)

Area V8 has a unique polar angle retinotopyand a distinctive foveal representation. Thiscontrasts significantly with the three extrastri-ate representations posterior to V8 (V4v, VP,and the inferior wing of V2), all of which arequarter-field representations of the contralat-eral upper visual field. Although the polar-angle retinotopy in V8 includes an additionalrepresentation of this quarter field, it alsoextends further to include a lower-field repre-sentation as well (Figs 2 and 3). These threeextrastriate visual areas (V4v, VP and inferiorV2) also share a contiguous representation ofthe fovea, at the top (superior) end of this rowof areas (Fig. 2d). However, the foveal repre-sentation in V8 is not part of this contiguousfoveal band; instead it is located about 3.5 cen-timeters away along the cortical surface, at theanterior border of V8 (Fig. 2d).

In conventional Talairach coordi-nates, foveal V4v (as defined retino-topically in this study) was centered at± 32, -87, -16, whereas foveal V8 wascentered at ± 33, -65, -14. When weaveraged the Talairach coordinates ofthe color-selective area ‘V4’ describedin previous studies ( ± 26, -67, -9), wefound that it was about twice as closeto the location of our retinotopicallydefined V8, compared to our retino-topically defined V4v. This supportsall the other evidence that the color-selective activity is located in area V8,rather than in ‘V4’.

In each hemisphere, V8 comprisesa continuous representation of theentire contralateral half of the visualfield. Although the authors did notattempt polar-coordinate retinotopy orflattened mapping, a prior study alsoconcluded that this human color-selec-tive region included a representationof upper and lower visual fields22.Together with V8 in the opposinghemisphere, the entire visual field isthus represented. Such a complete rep-resentation of the visual field would beappropriate in an area processing high-er-order, large-field color information.

HUMAN VERSUS MACAQUE MAPS

Because so much of the historical controversy about corti-cal color processing arose in studies of macaque monkey, it isnatural to wonder which area in macaque corresponds toarea V8 in humans. To clarify the topographic relationshipof the human and macaque maps, we first averaged together

articles

Fig. 4. Comparison of the polar angle retinotopy in human visual cortex, relative tothat reported in macaque monkeys. In both species, visual cortex is shown in flat-tened format, with visual area boundaries and polar angle continua as in Fig. 2a–c.Area MT is shown in gray. In macaque, dorsal area V4 is also indicated (V4d). Theretinotopy of V8 is similar to that reported in area TEO, in that both areas are locatedimmediately adjacent to area V4v. However, the two areas differ in overall shape, andthe retinotopy of V8 is rotated approximately 90o relative to that reported in TEO.

Fig. 3. Detailed retinotopy of the polar angle representation, from the same hemisphere shownin Fig. 2a. This figure shows the peak fMRI response (noisy white lines) corresponding to polarangle gradients of approximately 20o, superimposed on the standard pseudocolor rendering ofareas V1, V2, VP and V4v (drawn in a). To the right of each panel is a logo indicating the specificpolar angle (white line) stimulated. The complete contralateral visual field is represented in V8,from the lower visual field (a and b), across the horizontal meridian (c–e) to the upper visualfield (f–h). Note in (e–h) that the upper visual field representation in V8 can clearly be distin-guished from, and is mirror symmetric to, that in adjacent V4v.

a

b

c

d

e

f

g

h

a b


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


six of the most robust human retinotopic maps, using digitalmorphing techniques (Fig. 4b) as described26. These aver-aged maps could then be compared to the map of macaqueretinotopy, as estimated from single-unit mapping35

(Fig. 4a). This comparison suggests that human area V8shows some retinotopic and topographic similarity tomacaque area TEO. Furthermore, TEO and/or more anteri-or areas have appeared strongly color selective in recent stud-ies of macaque visual cortex (refs 17–19, 28, Vanduffel et al.Soc Neurosci. Abstr. 23, 845, 1997).

However, the retinotopic similarity between V8 and TEO isfar from exact (Fig. 4). Furthermore, other investigators haveproposed different area boundaries in this region of macaquecortex36–38. Unfortunately, those alternative models of themacaque maps are even less similar to the empirical humanmaps in this region of cortex. Thus it is not clear which macaquearea is homologous to human area V8. However, the flat mapsin Fig. 4 do make it clear that this human collateral color area isnot topographically similar to macaque V4 (neither dorsal norventral subdivisions). Human V8 is also topographically incon-sistent with the location of subdivisions proposed in macaquedorsal V4, such as V4t (ref. 39) and V4A (e.g. ref. 14).

COLOR AFTERIMAGES

Another way to assess functional selectivity is by measuring thefMRI responses during visual aftereffects, rather than duringthe fMRI effects produced by the visual stimuli themselves. Inother dimensions such as motion40 and orientation41, such indi-rect aftereffects have ironically proven to be functionally moreselective than the effects themselves. Here we tested whetherillusory color would also activate V8, as did real color stimuli.

If one stares for a time at a saturated color, then looks awayat a uniform gray field, one sees an illusory percept of the com-plementary color. Unlike motion or orientation aftereffects,these negative color afterimages are thought to arise primarily

in the retina42,43. However, they also presumably trigger activ-ity at higher levels, as would a real stimulus that was similarlystabilized on the retina29,44. To test for the presence of fMRIresponses to these illusory colors, we produced such color after-images in the MR scanner, along with control stimuli that werevery similar but did not produce color afterimages.

Figure 5 shows the time course of fMRI activity producedby these stimuli in area V8. As expected, the colored stimuliproduced robust fMRI activity in V8, whether alternating orconstant. However, only the constant-colored stimuli pro-duced a perceptual color afterimage and a corresponding fMRIaftereffect in V8 during subsequent viewing of the uniformgray field (Fig. 5a). The alternating-colored stimuli producedneither a perceptual color afterimage nor a prolongation ofthe normal fMRI return to baseline during the subsequentviewing period (Fig. 5a). The duration of the isolated fMRIcolor aftereffect was prolonged, consistent with the prolongedduration of the illusory color percept (Fig. 5b). Overall, thisstrongly suggests that these fMRI responses were related tothe processing of the illusory colors.

One unexpected finding was that the stimuli with alter-nating colors produced slightly more activity than the stim-uli in which color remained constant (Fig. 5a). This mayreflect the fact that the hues in the constant-colored stimulibecome progressively less saturated (less densely colored) withtime because of chromatic adaptation45. Essentially one beginsto see the mixture of the color afterimage and the actual color,Because these two colors are complementary, they produce aless saturated, more ‘washed-out’ color. This decreased fMRIresponse to the decreasingly saturated colors supports theother evidence that V8 is involved in color perception.

Findings similar to our fMRI afterimages in V8 werereported from scattered voxels in single-slice imaging throughnearby posterior fusiform gyrus29, but the location of thosevoxels was not localized to any specific cortical area. However,

articles

Fig. 5. The time course of V8 activity is related to the perception of color afterimages. (a) Time course from all voxels in retinotopicallydefined V8 that responded (p<0.00001) to the colored stimuli, relative to the initial presentation of the uniform gray stimulus, averagedacross 16 MR scans, showing the response to both the actual and the illusory color stimuli. Epochs in which subjects viewed a uniform greyfield, or an illusory afterimage on a gray background, are indicated with a white background; epochs when the subjects viewed colored stim-uli (alternating- or constant-colored) are indicated with a gray background. After the color stimulus, subjects viewed a uniform gray field, oran illusory afterimage on a gray background. During a color afterimage, the fMRI response was quite prolonged, consistent with the timecourse of the percept of the illusory colors. (b) The MR afterimage is shown more directly by subtracting signal during constant color andsubsequent gray period from alternating color and subsequent grey period.

a b

Color Fixation

Time (s)Time (s)

∆ %

MIR

I sig

nal

% M

IRI i

sgn

al

Fixation

Constantcolors

FixationFixation

Alternatingcolors


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


none of these data address the possibility that similar fMRIafterimages occur nonspecifically throughout much widerareas of visual cortex. This could arise from retinal color pro-cessing that is transmitted passively to cortex, or from glob-ally increased attention when viewing the color afterimages.

To test this, we reanalyzed our data to find the areas in anactivity map that respond differentially during the presenceof the visual afterimage (Fig. 6). At lower levels of significancethan shown here, a number of additional visual areas (e.g. V1,V2) do respond more to color afterimages. However, at themore strict significance threshold used in Fig. 6, the activityin V8 was more prominent than in any other area. In partic-ular, the wider areas of foveal color selective activities includ-ing areas such as V1, V2, etc., were relatively less prominentin the afterimage test, compared to the direct comparisons ofcolor versus luminance (see Fig. 1c and Fig. 6, showing thesame hemisphere).

DiscussionThe retinotopic maps make it clear that an additional area(V8) exists beyond those areas described previously in humanvisual cortex. Area V8 is retinotopically distinct from the pre-viously described area V4v, based on at least four different cri-teria. First, V4v and V8 have separate foveal representations,approximately 3.5 cm apart along the cortical surface. Second,V4v and V8 each include separate representations of the uppervisual field, separated from each other by a representation ofthe horizontal meridian. Third, V8 differs from V4v in itsglobal functional properties, including but not limited to colorsensitivity. Fourth, the nature of the retinotopy in V8 is dif-ferent from that in V4v.

The direct comparisons between color- and luminance-vary-ing stimuli (Fig. 1) indicate that area V8 responds slightly betterthan neighboring cortical regions to colored stimuli. However,we also found that the color-varying stimuli produce preferen-

tial activation in the foveal representation of all retinotopic areas.Thus, V8 may seem especially biased for color stimuli merelybecause its foveal representation sets it topographically apartfrom the conjoined foveal color responses of its neighbors (Figs2 and 3). This is a relatively trivial explanation for the color selec-tivity reported earlier, but we cannot rule it out completely. Thisidea is further supported by the fact that area V8 responds atreasonable levels to a wider variety of visual stimuli.

However, other evidence argues that V8 is involved inwavelength-dependent processing and perhaps in the con-scious perception of color itself. The robust and selectiveresponse to illusory colors (Figs 5 and 6) strongly supportsthis idea. Also, the anatomical colocalization of V8 comparedwith the previous clinical data makes it likely that area V8 isdamaged in achromatopsic patients23–25.

What do these human data tell us about macaque visualcortex? This question is constrained by several factors. Thehuman data are based on clinical and neuroimaging data,whereas the macaque data are derived from different tech-niques (e.g., single units, lesions and DG imaging), whichcould conceivably produce different results. Also, there maybe significant biological differences between the cortical orga-nization of color sensitivity in humans compared withmacaques. As we learn more about human and macaque visu-al cortex, the number of differences between these species areincreasing correspondingly26,46–48.

Despite these caveats, the data suggest that the area ofmacaque cortex that is homologous to the human ‘achro-matopsia’ area should be located in or anterior to TEO, ratherthan in V4. This is supported by data from macaque17–19 aswell as the present data from human V8.

MethodsGENERAL PROCEDURES. Except for modifications described below, themethods in this study are similar to those described26. Informed writ-ten consent was obtained for each subject prior to the scanning ses-sion, and all procedures were approved by Massachusetts GeneralHospital Human Studies Protocol numbers 90–7227 and 96–7464.Normal human subjects, with (or corrected to) emmetropic vision,were scanned in General Electric magnetic resonance (MR) scannersretrofitted with ANMR echo-planar imaging. Most scans wereacquired in a high-field (3 T) scanner, but some early scans wereacquired in a scanner of conventional (1.5 T) field strength. Based onsignal-to-noise ratios obtained during otherwise comparable condi-tions, four functional scans at 1.5 T were found to be approximatelyequal to one functional scan at 3 T, so this was the ratio used to equatedata acquired from the two scanners. Head motion was minimizedby using bite bars with deep, individually molded dental impressions.The subject’s task in all experiments was to fixate the center of eachtype of visual stimulus throughout the period of scan acquisition.

MR images were acquired using a custom-built quadrature surfacecoil, shaped to fit the posterior portion of the head. MR slices were3-4 mm thick, with an in-plane resolution of 3.1 x 3.1 mm, orientedapproximately perpendicular to the calcarine fissure. Each scan tookeither 4 min 16 s (color-versus-luminance and color afterimage scans)or 8 min 32 s (retinotopy), using a TR of either two or four seconds,respectively. Each scan included 2,048 images, comprised of 128images per slice in 16 contiguous slices.

Improved retinotopic maps were obtained from 32 subjects (79scans polar angle, 79 scans eccentricity, 323,584 images total). Amongthem, 13 subjects were also tested for color-versus-luminance (112scans, 229,376 images total). Of these, five subjects were tested exten-sively for color afterimages (100 scans; 204,800 images total). In mostsubjects, additional scans were done to clarify the location of areaMT and other visual areas.

articles

Fig. 6. The perception of color afterimages produces relativelyhigher activation in cortical area V8, compared with other corticalareas. The activation shown here represents all regions thatresponded significantly more (p≤0.00001) during viewing of the uni-form gray stimulus following the constant color stimulus, comparedwith viewing of the same gray stimulus following the alternatingcolor stimulus.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


VISUAL STIMULI. The goal of the first color-related experiment (Fig. 1)was to map the MR activity produced by color- versus luminance-varying stimuli throughout visual cortex, using conventional psy-chophysical stimuli. Prior to scanning, the equiluminance values fordifferent color combinations (red-cyan, CIE x and y coordinates 0.645,0.345 and 0.185, 0.248 respectively, or green-purple, CIE x and y coor-dinates 0.277, 0.684 and 0.351, 0.220, respectively) were measured foreach subject, outside the scanner. Equiluminance was measured usinga motion-null test, with the same stimulus projector (NEC model MT800), lens and color software used subsequently in the MR experi-ments. In the first experiment, both color- and luminance-varyingstimuli were produced using slowly moving (0.5 Hz) sinusoidal radi-al gratings (‘pinwheels’) of low spatial frequency (3 cycles per revo-lution). The gratings varied either in achromatic luminance(maximum, 95% luminance contrast) or in equiluminant color (atmaximum available saturations of the display device within con-straints of approximately 140 cd per m2 mean luminance and approx-imately white mean chromaticity), in direct alternation, in 16-secondepochs, using 16 epochs per scan.

In a second experiment (Figs 5 and 6), color afterimages were pro-duced by showing subjects colored adaptation patterns. Subjectsadapted to two general types of stimuli; one produced a pronouncedcolor afterimage when subjects subsequently viewed a uniform grayfield, but a very similar control stimulus did not produce such a colorafterimage. There were five epochs in each scan, presented in the fol-lowing order: (1) uniform gray, (2) alternating-color (control adap-tation), (3) uniform gray, (4) constant-color (experimentaladaptation) and (5) uniform gray. All epochs were 48-s long, exceptthat the final fixation period was prolonged 16 s to reveal the finaltraces of the MR afterimage. In the analysis for Fig. 5b, the last 16 sin that final epoch was truncated to match the duration of all otherepochs. Both the constant- and the alternating-colored patterns werecomprised of complementary colors (red-cyan or green-purple), spa-tially arranged in opposed quarter-fields (i.e. two-cycle-per-revolu-tion polar square waves), akin to interleaved bow ties. All hues werepresented at equal luminance, based on motion-null tests in each sub-ject. Following adaptation to the constant colors, subjects initiallyexperienced a prominent color afterimage against the uniform graybackground, which faded over tens of seconds. The afterimage wasretinotopically similar to the adaptation stimulus, but of comple-mentary color. As controls, we presented stimuli equivalent to theconstant colors, except that the colors alternated between color andcomplementary color, reversing every 2 s. The latter condition pro-duced no perceptual color afterimages during the subsequent epoch ofuniform gray stimulation.

Stimuli for retinotopic mapping were slowly moving, phase-encod-ed thin rays or rings comprised of counterphasing black and whitechecks, scaled according to polar coordinates, similar to thosedescribed26,31,32,50. However, to produce the most informative retino-topic maps possible, several stimulus modifications and new proce-dures were implemented. First, all retinotopic measurements weremade in the 3 T scanner. This increased the MR amplitudes by a factornear four, and the physiological signal-to-noise ratio by a factor neartwo. Second, we signal-averaged the information from 4–12 scans(8,192–24,576 MR images) of polar angle or eccentricity. Data werealso combined from different slice prescriptions on the same corticalsurface, to reduce intervoxel aliasing. Third, the retinotopic stimuliwere increased in extent both foveally and peripherally, to extend from0.2° through 18–30°. This activated correspondingly more surface ineach cortical area. Fourth, the visual stimuli were presented using anew LCD projector of higher spatial resolution (800 x 600), using bet-ter optics than previously (aperture lens, bypassing shielding screen,etc.). Fifth, the retinotopic stimuli varied in color as well as luminance,to better activate any color-selective cells in the region. The sum of allthese manipulations produced very robust retinotopic maps.

DATA ANALYSIS. Data from two-condition experiments (e.g., color-versus-luminance comparisons) and phase-encoded retinotopic experiments were

initially analyzed by doing a fast Fourier transform on the MR time coursefrom each voxel. Statistical significance was calculated by converting theFourier magnitude of the response to an f-statistic. The phase of the sig-nal at the stimulus frequency was used to track stimulus location in thecase of retinotopic stimuli, and to distinguish between positive- or negative-going MR fluctuations in the case of two-condition stimulus comparisons.

Scans comparing more than two stimuli (e.g., the color afterimagedata) were analyzed by selective averaging of two conditions. This wasfollowed by statistical comparison using a t-test of the difference of thefirst seconds following onset of the next epoch (here stimulus offset).

For topographic clarity, all data were analyzed and displayed in cor-tical surface format, as described36,43,44. This made it possible toextract the MR time courses from voxels in specific cortical areas,which were defined in the same subjects. The specific areas sampledwere V1, V2, V3/VP, V3A, V4v, MT+, and V8. Area ‘MT+’ was definedon the basis of additional scans comparing moving and stationarystimuli26,40. All other areas were based on retinotopic criteria.

For ease of comparison, all hemispheres are shown in right hemi-sphere format. Above a minimum threshold, the statistical significanceof the displayed pseudocolor range has been normalized according tothe overall sensitivity of each subject, as described elsewhere.

AcknowledgmentsThis work was supported by grants from the Human Frontiers Science

Foundation and NEI EY07980 to R.B.H.T., NEI EY09258 to P.C. and Swiss

Fonds National de la Recherche Scientifique to N.H. We thank Terry

Campbell and Mary Foley for scanning and participation in these

experiments, Robert Savoy, Ken Kwong, Bruce Fischl and Kevin Hall for

advice, Tommy Vaughan for coil design and manufacture and Martin Sereno

for modifying pilot stimuli. Wim Vanduffel, Ekkehardt Kustermann and

Irene Tracy also helped in preliminary versions of this experiment.


1. Cornsweet, T.N. Visual Perception (Academic Press, New York, 1970).2. Judd, D.B. & Wyszecki, G. Color in Business, Science, & Industry 3rd edn

388 (Wiley, New York, 1975).3. Dow, B.M. Functional classes of cells and their laminar distribution in

monkey visual cortex. J. Neurophysiol. 37, 927–946 (1974).4. Livingstone, M.S. & Hubel, D.H. Anatomy and physiology of a color

system in the primate visual cortex. J. Neurosci. 4, 309–356 (1984).5. Tootell, R.B.H., Silverman, M.S., Hamilton, S.L., De Valois, R.L. &

Switkes, E. Functional anatomy of macaque striate cortex: III. Color J.Neurosci. 8, 1569–1593 (1988).

6. Ts’o, D.Y., Frostig, R.D., Lieke, E.E. & Grinvald, A. Functionalorganization of primate visual cortex revealed by high resolution opticalimaging. Science 249, 417–420 (1990).

7. Lennie, P., Krauskopf, J. & Sclar, G. Chromatic machanisms in striatecortex of macaque. J. Neurosci. 10, 649–669 (1990).

8. Leventhal, A.G., Thompson, K.G., Liu, D., Zhou, Y. & Ault, S.J.Concommitant sensitivity to orientation, direction and color of cells inlayers 2, 3 and 4 of monkey striate cortex. J. Neurosci. 15, 1808–1818 (1995).

9. Hubel, D.H. & Livingstone, M.S. Segregation of form, color and stereopsisin primate area 18. J. Neurosci. 7, 3378–3415 (1987).

10. Tootell, R.B.H. & Hamilton, S.L. Functional anatomy of the secondcortical visual area (V2) in the macaque. J. Neurosci. 9, 2620–2644 (1989).

11. Gegenfurtner, K.R., Kiper, D.C. & Fenstemaker, S.B. Processing of color,form and motion in macaque area V2. Vis. Neurosci. 13, 161–172 (1996).

12. Zeki, S.M. Colour coding in rhesus monkey prestriate cortex. Brain Res.27, 422–427 (1973).

13. Zeki, S.M. Colour coding in the superior temporal sulcus of rhesusmonkey visual cortex. Proc. R. Soc. Lond. B 197, 195–223 (1977).

14. Zeki, S. Uniformity and diversity of structure and function in rhesusmonkey prestriate visual cortex. J. Physiol. (Lond.) 277, 273–290 (1978).

15. Zeki, S. The distribution of wavelength and orientation selective cells indifferent areas of monkey visual cortex. Proc. R. Soc. Lond. B 217, 449–470(1983).

16. Schein, S.J., Marrocco, R.T. & de Monasterio, F.M. Is there a highconcentration of color-selective cells in area V4 of monkey visual cortex?J. Neurophysiol. 47, 193–213 (1982).

17. Heywood, C.A., Gadotti, A. & Cowey, A. Cortical area V4 and its role inthe perception of color. J. Neurosci. 12, 4056–4065 (1992).

articles


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


18. Heywood, C.A., Gaffan, D. & Cowey, A. Cerebral achromatopsia inmonkeys. Eur. J. Neurosci. 7, 1064–1073 (1995).

19. Cowey, A. & Heywood, C.A. There’s more to colour than meets the eye.Behav. Brain Res. 71, 89–100 (1995).

20. Lueck, C.J. et al. The colour centre in the cerebral cortex of man. Nature340, 386–389 (1989).

21. Zeki, S. et al. A direct demonstration of functional specialization inhuman visual cortex. J. Neurosci. 11, 641–649 (1991).

22. McKeefry, D.J. & Zeki, S. The position and topography of the human colourcentre as revealed by functional magnetic resonance imaging. Brain 120,2229–2242 (1997).

23. Pearlman, A.L., Birch, J. & Meadows, J.C. Cerebral color blindness: Anacquired defect in hue discrimination. Ann. Neurol. 5, 253–261 (1979).

24. Damasio, A., Yamada, T., Damasio, H., Corbett, J. & McKee, J. Centralachromatopsia: Behavioral, anatomic, and physiologic aspects. Neurology 30,1064–1071 (1980).

25. Zeki, S. A century of cerebral achromatopsia. Brain 113, 1721–1777(1990).

26. Tootell, R.B.H. et al. Functional analsis of V3A and related areas in humanvisual cortex. J. Neurosci. 17, 7076–7078 (1997).

27. Engel, S., Zhang, X. & Wandell, B. Colour tuning in human visual cortexmeasured with functional magnetic resonance imaging. Nature 388, 68–71(1997).

28. Kennard, C., Lawden, M., Morland, A.B. & Ruddock, K.H. Colouridentification and colour constancy are impaired in a patient withincomplete achromatopsia associated with prestriate cortical lesions.Proc. R. Soc. Lond. B 260, 169–175 (1995).

29. Sakai, K. et al. Functional mapping of the human colour centre with echo-planar magnetic resonance imaging. Proc. R. Soc. Lond. B 261, 89–98 (1995).

30. Kleinschmidt, A., Lee, B.B., Requardt, M. & Frahm, J. Functional mappingof color processing by magnetic resonance imaging of responses to selectiveP- and M-pathway stimulation. Exp. Brain Res. 110, 279–288 (1996).

31. DeYoe, E.A. et al. Mapping striate and extrastriate visual areas in humancerebral cortex. Proc. Natl. Acad. Sci. USA 93, 2382–2386 (1996).

32. Sereno, M.I. et al. Borders of multiple visual areas in humans revealed byfunctional magnetic resonance imaging. Science 268, 889–893 (1995).

33. Tootell, R.B.H., Dale, A.M., Sereno, M.I. & Malach, R. New images fromhuman visual cortex. Trends Neurosci. 19, 481–489 (1996).

34. Galletti, C. Fattori, P., Battaglini, P.P., Shipp, S. & Zeki, S. Functionaldemarcation of a border between areas V6 and V6A in the superiorparietal gyrus of the macaque monkey. Eur. J. Neurosci. 8, 30–52 (1996).

35. Boussaoud, D., Desimone, R. & Ungerleider, L.G. Visual topography ofarea TEO in the macaque. J. Comp. Neurol. 306, 554–575 (1991).

36. Felleman, D.J. & Van Essen, D.C. Distributed hierarchical processing inthe primate cerebral cortex. Cereb. Cortex 1, 1–47 (1991).

37. Zeki, S. Are areas TEO and PIT of monkey visual cortex wholly distinctfrom the fourth visual complex (V4 complex)? Proc. R. Soc. Lond. B 263,1539–1544 (1996).

38. Maguire, W.M. & Baizer, J.S. Visuotopic organization of the prelunategyrus in rhesus monkey. J. Neurosci. 7, 1690–1704 (1984).

39. Van Essen, D.C., Maunsell, J.H. & Bixby J.L. The middle temporal visualarea in the macaque: Myeloarchitecture, connections, functionalproperties and topographic organization. J. Comp. Neurol. 199, 293–326(1981).

40. Tootell, R.B.H. et al. Visual motion aftereffect in human cortical area MTrevealed by functional magnetic resonance imaging. Nature 375, 139–141(1995).

41. Tootell, R.B.H. et al. Functional analysis of primary visual cortex (V1) inhumans. Proc. Natl. Acad. Sci. USA 95, 811–817 (1998).

42. Craik, K.J.W. Origin of visual afterimages. Nature 145, 512 (1940).43. Brindley, G.S. Two new properties of foveal afterimages and a

photochemical hypothesis to explain them. J. Physiol. 164, 168–179(1962).

44. Schiller, P.H. & Dolan, R.P. Visual aftereffects and the consequences ofvisual system lesions on their perception in the rhesus monkey. Vis.Neurosci. 11, 643–665 (1994).

45. Jameson, D., Hurvich, L.M. & Varner, F.D. Receptoral and postreceptoralprocesses in recovery from chromatic adaptation. Proc. Natl. Acad. Sci.USA 76, 3034–3038 (1979).

46. Tootell, R.B.H. & Taylor, J.B. Anatomical evidence for MT and additionalcortical visual areas in humans. Cereb. Cortex 5, 39–55 (1995).

47. Tootell, R.B.H. et al. Functional analysis of human MT and related visualcortical areas using magnetic resonance imaging. J. Neurosci. 15,3215–3230 (1995).

48. Jacobs, G.H. & Deegan, J.F. Spectral sensitivity of macaque monkeysmeasured with ERG flicker photometry Vis. Neurosci. 14, 921–928 (1997).

49. DeYoe, E.A., Felleman, D.J., Van Essen, D.C. & McClendon, E. Multipleprocessing streams in occipitotemporal visual cortex. Nature 371,151–154 (1994).

50. Engel, S.A., Glover, G.H. & Wandell, B.A. Retinotopic organization inhuman visual cortex and the spatial precision of functional MRI. Cereb.Cortex 7, 181–192 (1997).

articles


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

In central achromatopsia, patients lose the sensation of coloras a result of damage to the cortex and often report that theworld appears solely in shades of gray, as in a black-and-whitemovie1–4. Despite this almost total loss of color sensation,there are several reports that central achromats paradoxical-ly preserve other contributions of color to visual performance.Some of these preserved functions may indicate that the dam-age to the color analysis areas was less than total. For exam-ple, tests of sensitivity to color increments and color contrastin several central achromats5–9 and in one patient with par-tial achromatopsia10 have, up to now, revealed performancequite similar to that of subjects with normal color vision. Incontrast, the achromats in our experiments all show profoundlosses on these tests, indicating more extensive damage to colorprocessing areas. On the other hand, it seems that all centralachromats, including those in our experiments, can see color-defined shapes due to the preserved visibility of bordersbetween colors they no longer distinguish8,11–14. A likely can-didate for this border response is the magnocellular, or non-opponent pathway, which responds to color transitions15,16.

Both the perception of motion for equiluminous colorstimuli11 and a color-specific motion aftereffect5 have beendemonstrated in one central achromat. However, neither studyestablished the relative strength of the preserved motionresponse. If motion responses to color stimuli were only weak-ly preserved, we could attribute this to the partial sparing ofsome color-analysis areas adjacent to the cortical damage. Any

particular spared response could mediate other responses tocolor stimuli. For example, to an observer with normal colorvision, a saturated color appears brighter than its luminancewould predict17. If an achromat lost the experience of colorbut retained this supplementary brightness response8,11, equi-luminant color bars would have a residual brightness differ-ence. In this case, the visibility of the bars and their motioncould be reduced but never eliminated. Conversely, if motionresponses to color are totally preserved in central achro-matopsia, it would argue against a residual, spared mecha-nism and suggest that color’s contribution to motion takes aroute that is anatomically remote from the damaged areas.

We now resolve this question with a population of centralachromats who show little or no preserved opponent-colorprocessing in spectral sensitivity, detection or motion tasks atthreshold. At suprathreshold levels, however, these sameachromats surprisingly demonstrate normal levels of motionresponses to moving colored gratings.

To evaluate this performance, we included two controlgroups: normal subjects and congenitally red/green-deficientobservers (where the site of loss is in the retina). These indi-viduals are commonly described as color blind; however, theysuffer only a partial loss of color experience limited to thered/green dimension, a loss much less extensive than thatexperienced by the central achromats. Because these observershave little chromatic response to red/green stimuli, however,we can use their performance to evaluate the level of residual

Complete sparing of high-contrast colorinput to motion perception in corticalcolor blindness

Patrick Cavanagh1, Marie-Anne Hénaff2, François Michel2, Theodor Landis3, Tom Troscianko4 and James Intriligator5

1 Department of Psychology, Harvard University, 33 Kirkland Street, Cambridge, Massachusetts 02138, USA2 INSERM U-280, 151 cours A.Thomas, Lyon, 69003, France3 Department of Neurology, University Hospital Geneva, rue Micheli-du Crest, CH - 1211 Genève 14, Switzerland4 Department of Experimental Psychology, University of Bristol, 8 Woodland Road, Bristol, BS8 1TN, UK5 Department of Neurology, Beth Israel Deaconess Medical Center, 330 Brookline Avenue, Boston, Massachusetts, 02215, USA

Correspondence should be addressed to P.C. ([email protected])

It is widely held that color and motion are processed by separate parallel pathways in the visualsystem, but this view is difficult to reconcile with the fact that motion can be detected inequiluminant stimuli that are defined by color alone. To examine the relationship between color andmotion, we tested three patients who had lost their color vision following cortical damage (centralachromatopsia). Despite their profound loss in the subjective experience of color and their inabilityto detect the motion of faint colors, all three subjects showed surprisingly strong responses to high-contrast, moving color stimuli — equal in all respects to the performance of subjects with normalcolor vision. The pathway from opponent-color detectors in the retina to the motion analysis areasmust therefore be independent of the damaged color centers in the occipitotemporal area. It isprobably also independent of the motion analysis area MT/V5, because the contribution of color tomotion detection in these patients is much stronger than the color response of monkey area MT.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

luminance signals in our red/green color stimuli.Our tests reveal that motion responses to high-contrast

color gratings are at normal levels in the three patients butabsent in the congenitally red/green-deficient observers. Theseresults suggest that the cortical damage that produces achro-matopsia spares a particular cortical route for high-contrastcolor information, allowing it to contribute to motion per-ception without contributing to color sensations.

ResultsCASE HISTORIES

The three patients with central achromatopsia were similar inthat they all had bilateral lesions, made several errors in read-ing the colored digits of the Ishihara text and, when asked toorder several colored disks according to their hue (Farnsworth-Munsell test), they made virtually random orderings. All threealso had visual agnosia and prosopagnosia (deficits in objectand face recognition). Two had upper-field loss (WM, JPC)and one (JPN) had only one full quadrant of vision (lowerright). Visual acuity and contrast sensitivity for achromatictests in the preserved visual fields showed that at least coarsespatial resolution was retained for all three patients2,12.

At the time of testing, JPC was 43, a former florist, and hislesions were the result of an assault that caused bilateral hem-orrhage in the occipital-temporal regions. Magnetic resonanceimaging (MRI) revealed lesions principally in both fusiformgyri but sparing the lingual gyri and the left posterior fusiformgyrus. WM was 74, a former electrician who suffered a severe

reduction of vision following a brief loss of consciousness.(WM is described in more detail in ref. 18.) MRI and com-puter tomography subsequently revealed extensive occipi-totemporal lesions bilaterally in the territory of both posteriorcerebral arteries. JPN was 41 and had suffered bilateral strokesof the occipital region, producing a large lesion of the rightoccipital pole and a lesion of the left lingual and fusiform gyri.

JPN reported no conscious sensations of color and showedchance-level color naming in a forced-choice task. JPC noticedonly reds (and some yellows) as being weakly ‘tinted’ but hedid not name them as red because the tint differed from thered he remembered. WM reported no conscious sensations ofcolor but when forced to guess, he demonstrated better-than-chance naming of red and yellow.

SPECTRAL SENSITIVITY

The increment threshold spectral sensitivity task is a measureof the degree of function of the opponent-color pathways19,20.A chromatic stimulus of varying intensity is presented brieflyon a white background, and the observer reports whether any-thing has been presented. In the normal observer, the sensi-tivity peaks at three different wavelengths, which arecharacteristic of opponent-cone interactions20. If the tests arepresented on a dark background, the luminance mechanismis more sensitive than any of the opponent mechanisms, andthe sensitivity curve has a single peak. Both of these tests wereperformed on two of the patients, JPC and WM, and a nor-mal observer, TT, one of the authors. The data for WM havebeen reported elsewhere13.

Both patients showed single-peaked functions for thresh-olds on the light and dark backgrounds. The normal observershowed the expected three-peaked function with light back-grounds and single-peaked function with dark background.These results (Fig. 1) indicate that the two patients have sig-nificant losses in the opponent-color pathways, which mediatebroad ranges of the typical three-peaked threshold functionseen in normal subjects.

This experiment was very similar to previous experi-ments7,8 on two other central achromats. Both of thosepatients showed the normal, three-peaked function. Oneexplanation of these earlier results8 was that some color-oppo-nent response was spared and contributed to a brightness per-cept for the stimuli but not a color percept. We conclude thatthe previously tested patients had some sparing of high-levelcolor centers, whereas WM and JPC have a more profoundloss with no evidence of preserved color-opponent contribu-tion to increment thresholds.

CONTRAST THRESHOLDS

These tests evaluated the color contrast that allow observersto detect a red/green sinusoidal grating or to determine itsdirection of motion. Both smoothly moving gratings and grat-ings moving in 90° steps were used. The 90° steps11,21,22 effec-tively control for motion cues from the borders, which centralachromats are known to detect well. With each step, the newpositions of the red/green transitions fall exactly halfwaybetween the previous positions, rendering the border cueambiguous.

All three central achromats showed very high discrimina-tion (Fig. 2) and detection thresholds. The thresholds for WMwere exceptionally high. These thresholds suggest that WM,JPC and JPN have severe losses in their response to color. Wedid not find any evidence of a preserved brightness response

Fig. 1. Sensitivity for increment on light and dark fields as a functionof wavelength for two central achromats, JPC and WM, and onenormal, TT. The form of the function on a dark field was similar forboth achromats and for the normal, and only one plot is shown(JPC, top right). Only the normal observer shows the typical three-peaked function on the light field. Both achromats show a single-peaked function on light and dark fields, indicating that there is littleor no contribution of opponent-color mechanisms to these detec-tion thresholds.

Log

rel

ativ

e se

nsi

tivi

ty

Log

rel

ativ

e se

nsi

tivi

ty

Log

rel

ativ

e se

nsi

tivi

ty

Log

rel

ativ

e se

nsi

tivi

ty

Wavelength (nm)

Wavelength (nm)Wavelength (nm)

JPC light JPC dark

WM light TT light normal

Wavelength (nm)


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Fig. 3. A sinusoidal, equiluminous,color grating (red/green) and a sinu-soidal luminance grating (light anddark yellow) were superimposed,moving in opposite directions. (Thestimuli are depicted here as square-wave gratings for convenience.) At aparticular contrast of the luminancegrating, the two motions null eachother, establishing the equivalentluminance contrast of the colorgrating.

to color, which was proposed as the source of the preservedchromatic sensitivity in MS and other previously tested achro-mats11.

There were no marked differences between the thresholdsfor smoothly moving or jumping stimuli, indicating that theresponse to the chromatic borders in these patients was not asignificant factor in producing these threshold responses.Despite the elevated thresholds, the red/green equiluminancevalues for the achromats were in the normal range, indicatingnormal function of at least the red- and green-sensitive coneclasses21. Of the two congenitally red/green-deficientobservers, the deutan (with losses in the green-sensitive coneclass) showed an equiluminance setting in the normal range,whereas the protan (with losses in the red-sensitive cone class)required significantly more luminance for red.

MOTION NULLING

Like a subject (MS) in a previous study11, the three patientstested in the previous experiment could report the directionof a color grating (once it was above threshold). However,unlike MS, these three patients have severe losses in chromat-ic sensitivity and require very high contrasts for accuratejudgements. A luminance mechanism, which is nominally butnever perfectly insensitive to color, may start to respond tothese very strong color stimuli. Therefore when the threepatients tested here do respond to color athigh contrasts, it may be due to a smallresponse of the luminance pathway to coloror it may be due to the response of a color-specific mechanism.

To resolve this question, we used a stim-ulus (Fig. 3) that dissociates the two com-ponents, color and luminance, atsuprathreshold levels21,23–25. This gives usa measure of the strength of the color con-tribution to motion for the patients and forthe normal observers. This test evaluatesthe low-level motion response to color as

opposed to a high-level, tracking response21,25. To make themeasurements, color and luminance gratings were superim-posed and set in motion in opposite directions. At a particu-lar balance of the contrast of the two gratings, their motionscancelled and ambiguous motion or flicker was seen. The con-trast of the luminance grating that just cancelled the motion ofthe color grating was taken as the ‘equivalent luminance con-trast’ of the color grating21. Moreover, by again using con-genitally red/green-deficient subjects as a control group, wewere able to estimate the baseline response of the luminancemechanisms to residual luminance components in the colorstimulus.

Despite catastrophic loss in color sensitivity, when testedwith these suprathreshold color gratings, the three achro-matopsic patients have equivalent luminance contrasts thatlie near or within the range of values found in normal sub-jects (Fig. 4). The responses for the 90° jump conditions canonly be mediated by color-specific motion analysis becausethe red/green borders are producing ambiguous motion sig-nals in this stimulus.

The congenitally red/green-deficient observers showedmotion strengths in the 2% to 4% range, indicating that thesum of their weak color response and luminance artifacts frommonitor and optical sources was less than 4%. The differencebetween these values and the strengths measured for the cen-tral achromats is therefore a lower bound on the actualstrength of the color-specific input to motion for thesepatients. The high levels of equivalent luminance contrastsfound for the three patients, as well as their normal equilu-minance settings, again demonstrate that they have normalred- and green-sensitive cone function21.

It is nevertheless possible that some residual cues mightdifferentiate the red and the green of the gratings for the cen-tral achromats and allow them to track individual grating bars.It is unlikely, however, that these strategies (which the patientsdid not report) could produce equivalent luminance contrastsin the normal range seen here. On the other hand, JPC doesreport some residual color sensations for red stimuli, and so allthe tests were repeated with purple/green gratings (differen-tially stimulating principally the blue-sensitive cones) for bothWM and JPC. The patients again showed the same motionstrength as normal subjects.

How can these patients have normal levels of equivalentluminance contrast but red/green-deficient levels of color-contrast threshold? How can the response to color be absentup to a significantly elevated threshold and then jump torobust, normal levels once above threshold? We tested onepatient (JPC) for blindsight26, to test the possibility that hemight be able to report the direction of motion for stimuli heclaims not to see. After instructions to guess, his reports of

Fig. 2. Threshold color contrast for the three central achromatsfor direction discrimination of the moving color gratings. The rangeof values measured for two congenitally red/green-deficientobservers is shown by a yellow band and that for two normals by agreen band. Detection thresholds are not shown but are similar tothe discrimination thresholds for all subjects, except for the nor-mals, in which detection thresholds are lower than discriminationthresholds by a factor of two to three.

Motion jump (° of phase)

Th

resh

old

co

lor

con

tras

t (%

)

Luminancegrating

Superimposedgratings

Colorgrating


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


patients appears to be the most profound yet tested. The pre-served motion responses to color were found despite thisextreme loss of color processing.

To gauge the magnitude of color’s contribution to motionin our patients, consider that the maximum red/green colorcontrast we used produced 14% and 35% of cone contrast forthe red- and green-sensitive cones, respectively. This is thesame order of magnitude as the luminance contrast requiredto null it, 12% to 23% in these patients (and the normal sub-jects). In other words, the contribution of a color stimulus tomotion, where the cone responses are out of phase, seems tobe quite similar to the contribution of a luminance stimulus,where the same cone contrasts are in phase.

This result seems at first at odds with the many reports ofdegraded motion perception for chromatic stimuli. Howev-er, recent studies have shown that thresholds for motion ofgratings defined by color are actually lower than those forluminance-defined gratings27,28. The explanation is that thethreshold for seeing a color-defined stimulus is lower still, thelowest of all stimuli29, so that color gratings can be seen beforetheir motion is noticed. The result is the phenomenon of‘stopped motion’30,31, which led to the erroneous belief thatcolor was a poor source of motion information. More recentstudies demonstrate a strong21 and independent32 contribu-tion of color to motion, and our results here show a contri-bution at a level that could not be mediated by any secondary,residual effect. The secondary, residual responses probablyaccount for the eventual detection of the color stimuli at thevery high thresholds found in the patients and the congeni-tally red/green-deficient observers. This suggest that the sec-ondary cues have about one-tenth the power of the colorsignal.

One possible site for mediating the response to color is MT,an area specialized in motion analysis, which is located farfrom the areas damaged in achromatopsia. Physiologicalrecordings in monkeys33–35 and fMRI studies in humans36

have shown responses to equiluminous chromatic gratings inarea MT. Experiments with a 90° stimulus like that used herehave shown that this response can be based on color infor-mation, not just on the color borders33. However, thisresponse is very weak, with an equivalent luminance contrastof only 2.5%, within the range that congenitally red/green-deficient observers show and therefore attributable to distor-tions in the monitor, the eye or the non-opponent pathway.This level of response is far too low to explain the performanceof normal subjects and patients whose equivalent luminancecontrasts were in the range of 12 to 23%.

If MT is not the site of the robust, low-level motionresponse to chromatic stimuli that we report here, where couldit be? The site of cortical damage in humans with achro-matopsia invariably includes the ventral surface of occipi-totemporal region, and although this deficit is often claimed toinvolve the human homologue of monkey V4, recent lesionstudies in monkeys suggest that this may not be the case37.

In humans, the most recent fMRI studies38 suggest a moreanterior site, V8, for color analysis in a location consistentwith the damage in the achromatopsic patients. The humanhomologue to V4 on the ventral surface (V4v) is probably alsodamaged in these patients but it includes only a representa-tion of the upper visual fields39. The matching area that rep-resents the lower visual field has not been identified withcertainty. If there is a dorsal equivalent to V4 with a map ofthe lower visual fields, we speculate that it might offer a pos-

articles

motion direction for red/green gratings were random for stim-uli in the 8% to 15% range of color contrast just below hisdetection threshold (but still about ten times higher than nor-mal thresholds). It seems that the loss of performance at lowcolor contrast was real and is independent of the performanceat high color contrast.

We also tested his responses in the motion-nulling test(described above) at low contrasts. The color grating had noeffect once it was below threshold; the motion of the opposingluminance grating was the only direction the patient reported.Above his color threshold, however, the contrast of the lumi-nance grating required to null the motion of the color grat-ing rose rapidly. Indeed, a normal subject (PC) showed thesame, rapid rise in equivalent luminance contrast above 20%color contrast. However, the normal observer also had a weakresidual equivalent luminance contrast (1% to 2%) for lowerchromatic contrasts, whereas the achromat showed none. Thepatient seems to share with the normal subject a motionresponse to color that rapidly gains strength at high contrast.The patient has lost a more sensitive process that operates atlow contrasts in normal subjects and apparently depends onthe functioning of higher color centers.

DiscussionAt high color contrasts, the strength of the color contributionto motion for the three central achromats was equivalent tothat for normal subjects, demonstrating that there is a directpath for color information from the retina to cortical motiondetectors that is spared in central achromatopsia. Motionresponses mediated by this pathway were fully preserved, sug-gesting that the pathway was completely independent of thelesioned color centers of the ventral surface of the occipi-totemporal region.

Previously tested central achromats have shown residualopponent-color processing as indicated by a three-peakedspectral sensitivity curve and normal detection thresh-olds7–9,11. In contrast, in our study, both patients who weretested for spectral sensitivity had only a single-peaked func-tion, and all three had significantly elevated color-detectionthresholds. The damage to the color centers in these three

Fig. 4. Equivalent luminance contrasts for color gratings moving insmooth (10.8°) and 90° jumps. Data are shown individually for thethree central achromats. The range of values measured for six con-genitally red/green-deficient observers is shown as a yellow bandand that for normals (five observers) as a green band.

Motion jump (° of phase)

Equ

ival

ent

lum

inan

ce c

on

tras

t (%

)


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


sible site for the strong contribution of color to low-levelmotion in these patients and in normals. In monkeys, V4 doeshave a fair percentage of cells that are directionally selective40.Moreover, V4 in monkeys is directly involved in maintaininga representation of motions in a delayed match-to-sampletask41. Although the site of the robust motion response tohigh-contrast color remains to be determined, our resultsdemonstrate that the path from the retina to this area doesnot pass through the damaged ventral area of these patients.This suggests that color information does not flow as a single,hierarchically organized stream through the visual system butrather projects in parallel to several different centers, only oneof which leads to the conscious experience of color.

MethodsSPECTRAL SENSITIVITY. The observers were two central achromats, WMand JPC, and one normal subject, TT, one of the authors. Observers inthis and following experiments gave informed written consent beforethe experiments, which were approved by the Ethisches Kommiteeder neurologischen Klinik des Universitätsspital Zürich, by the Comitéconsultatif de protection des personnes dans la recherche biomédi-cale du centre Léon Bérard, Lyon and by the F.A.S. Human SubjectsCommittee, Harvard University.

The uniform light surround was 20 cd per m2 and the dark sur-round was 0.3 cd per m2. The test stimuli were presented for 0.5 swith a gap of 2.5 s between stimuli. The target was a slightly blurredrectangle subtending about 2.2° by 1.6°. Fixation was not monitoredduring the experiment. The chromatic tests were provided by a mono-chromator with a bandwidth of 10 nm.

After adapting to the light or dark background for several minutes,testing began at the shortest wavelength. Increment thresholds weredetermined using a method of limits, and the procedure was repeat-ed three times, reversing the order of the wavelengths each time. Themean of the three values was used to calculate the sensitivity for eachwavelength.

CONTRAST THRESHOLDS. The observers were the three central achro-mats, two congenitally red/green-deficient (one deutan and oneprotan) and two normal subjects. The red/green-deficient subjectseach failed at least 18 of the 24 Ishihara plates, whereas neither normalsubject made any errors.

The stimulus was a rotating wheel of eight cycles of a color grat-ing, varying sinusoidally between red and green (Fig. 2). The wheelwas set in rotation either ‘smoothly’ (because of the 66.7 Hz rasterrate, the actual step was 10.8° on each frame but it appeared to movesmoothly) or in 90° jumps. The rate of rotation was fixed at 2 Hz in allcases (0.25 revolutions per second). The outer diameter of the wheelwas eight degrees of visual angle, and its inner radius was threedegrees. Its mean luminance was approximately 45 cd per m2, and thesurround was dark. The control of luminance was linearized throughgamma correction look-up tables.

The chromatic contrast of a grating was defined in terms of thepercentage of the maximum chromatic modulation obtainable withthe phosphors involved. 100% color contrast of the monitor produces,at equiluminance, 14% cone contrast for the red-sensitive cones and35% cone contrast for the green-sensitive cones.

The colors were set to equiluminance42 individually for each observ-er. Thresholds were measured at that value and often two adjacentvalues bracketing the equiluminance setting as well. In every instancetested, the highest thresholds occurred at the predetermined equilu-minance setting. Observers were instructed to fixate the central bull’s-eye, and the tests were presented on the display until a response wasgiven. Observers reported whether a stimulus was absent or present inthe detection conditions or moving clockwise versus counterclock-wise in the direction-discrimination condition. Detection thresholdswere taken as the chromatic contrast for which the subject reported‘present’ on 50% of the trials. Discrimination thresholds were taken as

the chromatic contrast for which the subject reported the directioncorrectly on 75% of the trials.

MOTION NULLING. The observers were the three central achromats, sixcongenitally red/green-deficient observers and five normal observers.The red/green-deficient subjects (two deutans, four protans) eachfailed at least 18 of the 24 Ishihara plates, whereas the normal sub-jects made no more than three errors.

The stimulus was a rotating wheel of eight cycles of a red/greensinewave as before but now (Fig. 4) superimposed additively on a sim-ilar grating with eight cycles of a light/dark yellow sinewave rotatingin the opposite direction. The color contrast was set at the maximumavailable between the red and green phosphors with allowance for thesuperimposed luminance grating. The resulting color contrast wasbetween about two (WM) and five (JPC, JPN) times threshold con-trast. The stimulus was otherwise identical in all respects to that usedin the previous experiment.

In a forced-choice procedure, while fixating the bull’s-eye, theobservers reported the direction of the global motion seen in the com-bined stimulus. The luminance contrast that produced a motion null(equal frequency of reports favoring luminance and color directions)was taken as the ‘equivalent luminance contrast’ of the color grat-ing21. We also tested red/green luminance ratios bracketing the equi-luminance setting and, in every instance, the minimum equivalentluminance contrast occurred at the predetermined setting.

AcknowledgmentsWe thank Manfred Fahle, Anne Kurtenbach and Lukas Rüttiger for the loan

of the spectral-sensitivity measuring equipment and its calibration and Seth

Hamlin for technical assistance. Supported by NEI grant EY09258 to PC,

Swiss National Science Foundation Grant NR: 2100-045699.95 to TL, and

UK Defense Research Agency grant D/ER1/9/4/2034/102/RARDE to TT.


1. Damasio, A., Yamada, T., Damasio, H., Corbett, J. & McKee, J. Centralachromatopsia: Behavioral, anatomic, and physiologic aspects. Neurology30, 1064–1071 (1980).

2. Grüsser, O.-J. & Landis, T. in Vision and Visual Dysfunction, Vol 12: VisualAgnosias. (ed. Cronly-Dillon, J.) 394–400 (MacMillan, London, 1991).

3. Pearlman, A.L., Birch, J. & Meadows, J.C. Cerebral color blindness: Anacquired defect in hue discrimination. Ann. Neurol. 5, 253–261 (1979).

4. Zeki, S.M. A century of cerebral achromatopsia. Brain 113, 1721–1777(1990).

5. Mollon, J.D., Newcombe, F., Polden, P.G. & Ratcliff, G. in Colour VisionDeficiencies (ed. Verriest, G.) 130–135 (Hilger, Bristol, 1980).

6. Wooten, B.R. Partial cerebral achromatopsia with selective hue loss. Doc.Ophthalmol. Proc. Series 33, 139–144 (1982).

7. Jaeger, W., Krastel, H. & Braun, S. Cerebral achromatopsia (symptoms,course, differential diagnosis and examination strategy). II. [Article inGerman] Klin. Monatsbl. Augenheilkd. 194, 32–36 (1989).

8. Heywood, C.A., Cowey, A. & Newcombe, F. Chromatic discrimination ina cortically colour blind observer. Eur. J. Neurosci. 3, 802–812 (1991).

9. Heywood, C.A., Nicholas, J.J. & Cowey, A. Behavioural andelectrophysiological chromatic and achromatic contrast sensitivity in anachromatopsic patient. J. Neurol. Neurosurg. Psychiatry 60, 638–643(1996).

10. Victor, J.D., Maiese, K., Shapley, R., Sidtis, J. & Gazzaniga, M.S. Acquiredcentral dyschromatopsia: Analysis of a case with preservation of colordiscrimination. Clin. Vis. Sci. 4, 183–196 (1989).

11. Heywood, C.A., Cowey, A. & Newcombe, F. On the role of parvocellular(P) and magnocellular (M) pathways in cerebral achromatopsia. Brain117, 245–254 (1994).

12. Henaff, M.A. & Michel, F. A case of central achromatopsia: Possibleimplicit treatment of hue. Invest. Ophthalmol. Vis. Sci. 34, 745 (1993).

13. Troscianko, T. et al. Human colour discrimination based on a non-parvocellular pathway. Curr. Biol. 6, 200–210 (1996).

14. Barbur, J.L., Harlow, A.J. & Plant, G.T. Insights into the different exploitsof colour in the visual cortex. Proc. R. Soc. Lond. B 258, 327–334 (1994).

15. Lee, B.B., Martin, P.R. & Valberg, A. The physiological basis ofheterochromatic flicker photometry demonstrated in the ganglion cells of themacaque retina. J. Physiol. (Lond.) 404, 323–347 (1988).

16. Schiller, P.H. & Colby, C.L. The responses of single cells in the lateralgeniculate nucleus of the rhesus monkey to color and luminance contrast.Vision Res. 23, 1631–1641 (1983).

articles


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


17. Wagner, G. & Boynton, R.M. A comparison of four methods of heterochromaticphotometry. J. Opt. Soc. Am. 62, 1508–1515 (1972).

18. Scheidler, W., Landis, T., Rentschler, I., Regard, M. & Baumgartner, G. A patternrecognition approach to visual agnosia. Clin. Vis. Sci. 7, 175–193 (1992).

19. Harwerth, R.S. & Sperling, H.G. Effects of intense visible radiation on theincrement-threshold spectral sensitivity of the rhesus monkey eye. Vision Res.15, 1193–1204 (1975).

20. King-Smith, P.E. & Carden, D. Luminance and opponent-color contributions tovisual detection and adaptation and to temporal and spatial integration. J. Opt.Soc. Am. 66, 709–717 (1976).

21. Cavanagh P. & Anstis S.M. The contribution of color to motion in normal andcolor-deficient observers. Vision Res. 31, 2109–2148 (1991).

22. Dobkins, K.R. & Albright, T.D. What happens if it changes color when it moves?:psychophysical experiments on the nature of chromatic input to motiondetectors. Vision Res. 33, 1019–1036 (1993).

23. Chichilnisky, E.J., Heeger, D. & Wandell, B.A. Functional segregation of colorand motion perception examined in motion nulling. Vision Res. 33, 2113–2125(1993).

24. Teller, D.Y. & Palmer, J. Infant color vision: motion nulls for red/green vsluminance-modulated stimuli in infants and adults. Vision Res. 36, 955–974(1996).

25. Cavanagh, P. Attention-based motion perception. Science 257, 1563–1565(1992).

26. Weiskrantz, L. Blindsight revisited. Curr. Opin. Neurobiol. 6, 215–220 (1996).27. Metha, A.B., Vingrys, A.J. & Badcock, D.R. Detection and discrimination of

moving stimuli: the effects of color, luminance, and eccentricity. J. Opt. Soc. Am.11, 1697–1709 (1994).

28. Stromeyer, C.F. III, Kronauer, R.E., Ryu, A., Chaparro, A. & Eskew, R.T. JrContributions of human long-wave and middle-wave cones to motiondetection. J. Physiol. (Lond.) 485, 221–243 (1995).

29. Chaparro, A., Stromeyer, C.F. III, Huang, E.P., Kronauer, R.E. & Eskew, R.T. JrColour is what the eye sees best. Nature 361, 348–350 (1993).

30. Ramachandran, V.S. & Gregory, R. Does colour provide an input tohuman motion perception? Nature 275, 55–56 (1978).

31. Cavanagh, P., Tyler, C.W. & Favreau, O.E. Perceived velocity of movingchromatic gratings. J. Opt. Soc. Am. 1, 893–899 (1984).

32. Cropper, S.J. & Derrington, A.M. Rapid colour-specific detection ofmotion in human vision. Nature 379, 72–74 (1996).

33. Dobkins, K.R. & Albright, T.D. What happens if it changes color when itmoves?: the nature of chromatic input to macaque visual area MT. J.Neurosci. 14, 4854–4870 (1994).

34. Saito, H., Tanaka, K., Isono, H. Yasuda, M. & Mikami, A. Directionallyselective response of cells in the middle temporal area (MT) of themacaque monkey to the movement of equiluminous opponent colorstimuli. Exp. Brain Res. 75, 1–14 (1989).

35. Gegenfurtner, K.R. et al. Chromatic properties of neurons in macaque MT.Vis. Neurosci. 11, 455–466 (1994).

36. Ffytche, D.H., Skidmore, B.D. & Zeki, S. Motion-from-hue activates areaV5 of human visual cortex. Proc. R. Soc. Lond. B 260, 353–358 (1995).

37. Heywood, C.A., Gaffan, G. & Cowey, A. Cerebral achromatopsia inmonkeys. Eur. J. Neurosci. 7, 1064–1073 (1995).

38. Hadjikhani, N.K., Liu, A.K., Cavanagh, P., Dale, A.M. & Tootell, R.B.H.Retinotopy and color sensitivity in human visual cortical area V8. NatureNeurosci., 1, 235–241 (1998).

39. Tootell, R.B., Dale, A.M., Sereno, M.I. & Malach, R. New images fromhuman visual cortex. Trends Neurosci. 19, 481–489 (1996).

40. Zeki, S.M. Uniformity and diversity of structure and function in rhesusmonkey prestriate visual cortex. J. Physiol. (Lond.) 277, 273–290 (1978).

41. Ferrera, V.P., Rudolph, K.K. & Maunsell, J.H. Responses of neurons in theparietal and temporal visual pathways during a motion task. J. Neurosci.14, 6171–6186 (1994).

42. Cavanagh, P., Anstis, S.M. & MacLeod, D.I.A. Equiluminance: Spatial andtemporal factors and the contribution of blue-sensitive cones. J. Opt. Soc.Am. 4, 1428–1438 (1987).

articles


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

In primates with a small foveal region of densely packed pho-toreceptors in their retinae that provide high acuity vision, eacheye is moved about by six extraocular muscles, innervated byneurons in the brainstem oculomotor centers. Rapid saccadic eyemovements are executed two to four times per second to bringvisual objects into central view, and slower, smooth-pursuit eyemovements are made to keep objects in view when either theobserver or the object is in motion. In the frontal lobe, two areashave been shown to be significant in visually guided eye move-ments: the frontal eye fields (FEF) and the medial eye fields(MEF), both of which make direct projections to brainstem ocu-lomotor centers1–11. The MEF, located in the dorsomedial frontalcortex, are also known as the supplementary eye fields8. Single-cell recording and microstimulation studies have shown that thesetwo areas perform different coding operations for the generationof saccadic eye movements3,7,11. Electrical stimulation of the FEFelicits saccades that have specific directions and amplitudes; pro-longed electrical stimulation yields a staircase of identical sac-cades with intervening fixations. By contrast, electricalstimulation of the MEF produces saccades that take the eyes toa particular orbital position; prolonged stimulation keeps theeyes at that position. These findings indicate that the FEF carry avector code and the MEF a place code10,11. To shed further lighton the functions of these two areas, we examined the effects ofablating them, either singly or in combination, and studied theeffects on visually guided eye movements. Most studies haveshown only mild, temporary deficits after FEF lesions12–16.Deficits in visually guided saccadic eye movements have not pre-viously been demonstrated after MEF lesions in the monkey.

Here we present results obtained on four tasks. The first taskwas saccadic eye movements to single targets. Following fixationof a small central spot, a single target appeared in one of severallocations on the monitor. Execution of an accurate saccadic eye

movement to the target was rewarded with a drop of apple juice.The second task was saccadic eye movements to paired targets.Paired targets were presented with various temporal onset asyn-chronies. Monkeys were rewarded for making a saccadic eyemovement to either target. The onset asynchrony between pairedtargets was varied to determine the temporal delay required toproduce equal probability of eye movements to each target. Thethird task was saccadic eye movements to sequential targets. Oneach trial, two targets were presented in rapid sequence with var-ious temporal durations and with various delays between them.The task was to repeat the order of the target presentations bymaking successive saccadic eye movements to the target loca-tions. The fourth task was saccadic eye movements to targets inarrays. Following fixation, eight stimuli were presented equidis-tant from the fixation spot; one of the stimuli, the target,appeared at various times prior to the other seven stimuli. Themonkey was rewarded only for eye movements directed to thetarget stimulus. In all four tasks tested, we observed prominentdeficits following FEF lesions. Considerably smaller deficits thatrecovered more rapidly were observed after MEF lesions. Theeffects of combined FEF and MEF lesions were no greater thanFEF lesions alone.

ResultsFigure 1 shows the performance of two monkeys on the single-target task. Each panel shows the distribution of rightward andleftward saccadic movement latencies made to individual visualtargets presented randomly in one of four locations. Panels (a)and (b) compare the distributions before and after a left FEFlesion. The insets display eye-movement records that show some-what less accurate saccades to the right after the left FEF lesion; asignificant change in saccadic peak velocities accompanies thisfor right saccades from a preoperative 534 to a postoperative 459

The effects of frontal eye field anddorsomedial frontal cortex lesions on visually guided eye movements

Peter H. Schiller and I-han Chou

Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA

Correspondence should be addressed to P.H.S ([email protected])

In the frontal lobe of primates, two areas play a role in visually guided eye movements: the frontaleye fields (FEF) and the medial eye fields (MEF) in dorsomedial frontal cortex. Previously, FEF lesionshave revealed only mild deficits in saccadic eye movements that recovered rapidly. Deficits in eyemovements after MEF ablation have not been shown. We report the effects of ablating these areassingly or in combination, using tests in which animals were trained to make saccadic eye movementsto paired or multiple targets presented at various temporal asynchronies. FEF lesions produced largeand long-lasting deficits on both tasks. Sequences of eye movements made to successively presentedtargets were also impaired. Much smaller deficits were observed after MEF lesions. Our findings indi-cate a major, long-lasting loss in temporal ordering and processing speed for visually guidedsaccadic eye movement generation after FEF lesions and a significant but smaller and shorter-lastingloss after MEF lesions.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

degrees per second (p < .001, t-test). In (b) an increase of 45 msis evident in the mean latency of saccadic eye movements made tothe right as compared with pre-operative performance (p < .001,t-test); the lesion also produced a broadening in the latency dis-tribution for rightward saccades. Saccadic eye movements madeto the left after the left FEF ablation yielded significantly shorterlatencies than pre-operatively by 13 ms (p < .05). Panels (d), (e),and (f) show pre- and post-operative latency distributions fromanother monkey. The right MEF lesion produced a small, butsignificant increase in latencies for leftward saccadic eye move-ments (11 ms, p < .05, t-test). Saccadic accuracy and velocity wereunaffected by the MEF lesions. The left FEF lesion made in thisanimal subsequent to the bilateral MEF lesion (Fig. 1f) produceddeficits similar to those obtained with a single left FEF lesion,yielding a difference of 54 ms in the mean latencies between leftand right saccades. Four months after the left FEF lesion in mon-key 1, the latency difference between saccades made to the leftand right dropped to 25 ms; four months after the paired FEFand MEF lesions in monkey 2, the difference dropped to 24 ms(both with p < .001, t-test). Saccadic velocity differences for left-ward and rightward saccadic eye movements were also dimin-ished but remained significant.

The paired-target task we devised permitted us to quantify thedegree to which target selection was biased by the cortical lesions.We presented paired targets with different onset asynchronies todetermine the interval required to produce equal probability oftarget selection appearing in the right and left hemifields. Resultsobtained on this task using targets with an angular separation of90 degrees relative to the fixation spot appear in Fig. 2. The pairedtargets appeared either above or below the fixation point, withone stimulus falling in the left and the other in the right visualhemifield. Paired targets were interspersed with single targets thatappeared twice as often; conditions were configured to have the

monkeys execute equal numbers of saccadic eye movements tothe left and to the right during each session, thereby forestallingthe emergence of position biases.

The inset in Fig. 2 shows preoperatively collected eye-move-ment records for paired targets that appeared either simultane-ously or with asynchronies favoring either the left or the rightstimulus in the pair. With simultaneous presentation, the prob-ability of making saccadic eye movements to the left and righttargets was equal, whereas when one target preceded the otherby 33 ms, most saccades were made to the stimulus that hadappeared first. The percent of saccades made to the left target atdifferent asynchronies are plotted in the center of the graphlabeled pre-op. Also plotted is the percent of saccadic eye move-ments made into the left visual field at various times after a leftFEF and a right MEF lesions in two different animals. Two weeksafter the left FEF lesion, the curve shifts to the far left. To achievean equal probability of left and right saccades (50% crossoverpoint), the target in the affected right hemifield had to be pre-sented 116 ms prior to the target in the left hemifield. In theweeks following the lesion, the size of the asynchrony requiredfor equivalent target choice decreased gradually, as shown by thesuccessive curves on the left side of the graph in Fig. 2.

The shift in target choice after a right MEF lesion, as shownby the curves on the right in Fig. 2, was much smaller in magni-tude. Two weeks after the lesion, an asynchrony of 31 ms wasrequired for the monkey to make saccades with equal probabili-ty to either target. Recovery after the MEF lesion was completeby the 16th week. At a comparable time after the left FEF lesion,there was still a strong bias in favor of the ipsilateral target, requir-ing a 54 ms target-onset asynchrony for equal probability per-formance. At comparable post-operative times, the asynchroniesrequired to produce equal probability choice were more thantwice as long as the latency differences between left and right-

Fig. 1. Distribution of sac-cadic latencies to single tar-gets before and after lesions.Data are shown from twomonkeys with (a) and (d) rep-resenting pre-operative data.(b) shows the distribution ofleft and right saccadic eyemovements three weeks afterleft FEF lesion; (e) shows simi-lar data two weeks after aright MEF lesion in the secondanimal. (f) shows data col-lected after a left FEF lesionhad been made in the sameanimal subsequent to bilateralMEF lesions. Eye-movementrecords before and after theleft FEF lesion appear in theinset of (a) and (b). Mean leftand right saccadic latencieswere as follows: (a) L, 125; R,121, (b) L, 112; R, 166, (d) L,130; R, 130, (e) L, 141; R, 130,(f) L, 121; R, 175. The 11-msdifference in mean left and right saccadic latencies two weeks after the right MEF lesion (e) is significant at the .05 level. Differences after FEFlesions (b and f) are significant beyond the .001 level. (c) Reconstruction of an MEF and FEF lesion from a third monkey. The major sulci arelabeled as are the locations of the FEF and MEF. Ant: anterior portion of the brain. Number of trials per histogram: (a) n = 1200, (b) n = 720,(d) n = 1120, (e) n = 2160, (f) n = 1092.

60 100 140 180 220

50

100

150

200

250

60 100 140 180 220

20

40

60

80

100

120

2 wks post MEF and LFEF2 wks post RMEF

60 100 140 180 220

20

60

60 100 140 180 220

20

40

60

80

Pre-op 3 wks post LFEF

100

140100

Nu

mb

er o

f sa

ccad

es

Latency in milliseconds

left saccades

right saccades

60 100 140 180 220

20

60

100

140

Pre-op

a b

d e f

a Pre-op 3 wks post LFEF

f2 wks post RMEF

b c

d

Nu

mb

er o

f sa

ccad

es

ePre-op

Latency in milliseconds

2 wks post MEF and LFEF


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

ward saccades made to single targets (see Fig. 1). There was nofurther improvement on the paired target task when the mon-key was tested five months after FEF lesions.

When two targets have an angular separation of 90 degrees,most of the time monkeys generate accurate saccadic eye move-ments to one target or the other, as shown in the inset in Fig. 2.However, when the target separation is decreased, animalsbegin to make averaging saccades that land the center of gaze atintermediate positions between the two targets17. Such vector-averaged saccades are presumably the product of simultane-ously arriving commands from the cortex to the superiorcolliculus or the brainstem to move the eyes to each target.Studies have shown that concurrent electrical stimulation oftwo sites in the superior colliculus of the FEF produces sac-cadic eye movements that are a vector average of saccades elicit-ed by stimulating each site alone18,19. Figure 3 shows datacollected with targets that had an angular separation of 40degrees; under such conditions vector- averaged saccadesbecome common. In the unlesioned monkey, vector-averagedsaccades occur most frequently when the targets appear simul-taneously; the frequency of such saccades falls off rapidly astarget-onset asynchrony is increased. The consequence of anFEF lesion is a dramatic shift in the asynchrony between the

targets at which vector-averaged saccades occur most com-monly; as a result of the lesion there are now no vector-aver-aged saccades when the targets are simultaneous; instead, theybecome most frequent when the target in the hemisphere con-tralateral to the lesion is presented 67–100 ms prior to the tar-get in the left hemisphere. These findings suggest that FEFlesions retard the rate at which information can be processedfor the generation of visually guided eye movements.

The dorsomedial frontal area within which the MEF resideshas been implicated in the temporal processing of events, par-ticularly in the execution of sequential motor acts20–22. We there-fore wanted to determine how MEF and FEF lesions alter theproduction of a sequence of eye movements to successively pre-sented targets. In Fig. 4, data from the sequential task are pre-sented. Four different sequences with several different durationswere presented in randomized order. Pre-operative and post-operative data collected at various times after the lesions areshown. The results demonstrate a significant but mild deficitafter MEF lesions, which is consistent with findings in humans20.However, a much larger deficit was evident after the FEF lesion.The inset shows eye-movement records collected two monthsafter a left FEF lesion using four sets of sequentially presentedtargets with a sequence duration of 117 ms. Performance on

Fig. 3. Data collected with paired targets having a 40% angular separation. (a) Eye-movement records obtained at various target asyn-chronies; text shows which target appeared first (L, left; R, right) and by how many milliseconds (ms). (b) Plot of the frequency of vector-averaged saccades as a function of target asynchrony. The data were obtained from an intact animal and from an animal with a left FEF lesion.We counted as vector averaged those saccades that fell within plus or minus nine angular degrees of the midpoint between the two targets.

Fig. 2. Pairedtargets pre-sented with var-ied asynchronies.The percent ofsaccades madeto the left targetas a function oftemporal offsetbetween thepaired targets isplotted. Data areshown at varioustimes afterrecovery from FEF and MEF lesions as well as pre-operatively (pre-op). Squares and solid lines depict data from the animal with the FEFlesion; circles and dotted lines depict data from the animal with the MEF lesion. Records of saccadic eye movements made to paired targetswith various temporal offsets in the intact animal appear as an inset. For each of the post-operative weeks plotted (wk 2–wk 16), data werecollected for 5 or 6 successive days.

Onset asynchrony in milliseconds

Perc

ent

sacc

ades

to

left

tar

gets

Left target firstRight target first

RMEF lesion

wk2

100

0

20

40

60

80

20 40 60 80 100 120 140140 120 100 80 60 40 20 0 160 180180 160220 220220200

LFEF lesion

wk2wk3

pre-op

wk16

wk16

L, 33ms 0ms R, 33msPre-op eye movements made to paired targets

Onset asynchrony in milliseconds

Per

cen

t sa

ccad

es t

o le

ft t

arg

ets

60 40 20 0 20 40 60 80 100 120 140

0

10

20

30

40

50

Onset asynchrony between paired targets

Left target first Right target first

Intact monkey LFEF lesion

sacc

ades

vec

tor

aver

aged

b

Perc

ent

a

Onset asynchrony between paired targets

Per

cen

t sa

ccad

es v

ecto

r av

erag

ed

Intact monkey

LFEF lesion

b

L, 33 ms 0 ms R, 33 ms

Pre-op eye movements made to paired targets

wk 2

wk 2wk 16

wk 3

wk 16

pre-op

L, 67 ms 0 ms R, 67 ms R, 100 ms R, 133 ms


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

sequences presented to the right was severely impaired. The mostcommon error was the execution of a single saccade, instead ofa sequence of saccades, that landed the center of gaze near eitherof the target locations or between them.

Examination of saccadic latencies made to the first target inthe sequence suggests that target reaction times make only a smallcontribution to the deficit. For example, three weeks after the leftFEF lesion, the mean latencies to the first target in the sequenceswe used were 130 ms to the left and 198 ms to the right (a laten-cy difference of 68 ms). Yet the comparable performance to theleft and right occurs with a sequence duration difference of morethan 200 ms (Fig. 4).

Full recovery on the sequential task is evident five weeksafter MEF lesions. Following FEF lesions, considerable recov-ery occurs by week 11. Combined MEF and FEF lesions pro-duced deficits and recovery times similar to those obtainedfrom the FEF lesion alone. The results from the sequential task

Fig. 4. Pre and post-operative datafor percent correct performance areshown for left and right saccadic eyemovements made to sequential tar-gets before and after a right MEFlesion (a, b) and before and after aleft FEF lesion (c, d). Two stimuliappeared in succession using four dif-ferent sets of locations and severaldifferent overall durations. The insetshows eye movements made to oneset of target sequences after a left FEFlesion, when presented for asequence duration of 117 ms, whichis completed well before the initiationof the first eye movements. The foursets of sequential targets, as indicatedin the inset, appeared at positionsA2–B1, E2–D1, A4–B5 and E4–D5.Each trial began with a central fixationspot followed by the appearance ofone sequence. The eye movementsshown in the inset demonstrate cor-rect sequences made to the intact leftside and mostly incorrect eye move-ments made for sequences presentedto the right.

suggest that for the generation of sequences of eye movements,the FEF are more important than the MEF.

The paired-target and sequential tasks do not directly test theability of animals to make a temporal discrimination. To assessdeficits in ascribing temporal order to successively appearingevents, we devised a task that required the animals to discrimi-nate stimuli on the basis of the order in which they appeared.Eight identical stimuli were presented equidistant from fixation,one of which preceded the others by various times. Only a directsaccadic eye movement made to the stimulus that appeared firstwas rewarded. This task explicitly requires that the animal dis-criminate the relative onset of the stimuli. Data collected fourmonths after a left FEF and paired MEF and left FEF lesionsappear in Fig. 5a and b. A major impairment is evident for eyemovements made into the contralateral visual field. This deficit isnot due to loss of perception of high-frequency temporal infor-mation per se, as the monkeys were unimpaired on flicker sen-

sitivity. This suggests that thedeficit lies not in perceiving tem-poral discontinuity but in a dis-ability to assess the temporal orderof presentations for the generationof saccadic eye movements.

In addition to the four testsdescribed here, the performance ofmonkeys was also assessed for eyemovements made to targets thatmoved at various velocities, onsensitivity to stimuli of varied con-trasts, on their ability to discrimi-nate targets that were of a differentsize, shape or color from other,simultaneously appearing stimuli,and their ability to discriminateflickering stimuli from steadily illu-

Perc

ent

Cor

rect

Sequence Duration in milliseconds125 150 175 200 225 250 275 300 325 350 375 400 425 450 475 500

wk2

wk3wk6

wk11

0

20

40

60

80

125 150 175 200 225 250

100

Preop

20

40

60

80

100

125 150 175 200 225 250

R

Preop

Post LFEF lesion

L

L

R

L

wk1

Post RMEF lesion

125 150 175 200 225 250 275

R

L

wk5 Eye movements, post LFEF

a

c

0

20

40

60

80

100

20

40

60

80

100

d

ba b

cPe

rcen

t cor

rect

d

Preop

Preop Post LFEF lesion

Eye movements, post LFEF

Sequence duration in milliseconds

50 100 150 200 250 300

0

20

40

60

80

100

Perc

ent

corr

ect left hemifield

right hemifield

LFEF lesion

50 100 150 200 250 300

0

20

40

60

80

100

Target onset asynchrony in milliseconds

MEF & LFEF lesionsa b

N = 3220 N = 2720

Fig. 5. Performance on the discrimination of time differences. Percent correct performance is shownseparately for targets appearing in the left and right hemispheres. (a) Data from a monkey with a leftFEF lesion. (b) Data from a monkey after paired MEF and FEF lesions. (a) n = 3220, (b) n = 2720.

a bLFEF lesion MEF and LFEF lesions

Target onset asynchrony in milliseconds

Per

cen

t co

rrec

t

wk 5

Post RMEF lesion

wk 1

wk 11

wk 6wk 3

wk 2


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

minated ones. Except for an increase in saccadic latencies, per-formance on these tasks was largely unaffected by the lesions.

DiscussionIn foveate animals, with each shift in gaze not only are stimuli incentral vision analyzed but a decision has to be made as to whereto look next. The findings we report here suggest that the FEFare central to this process of target selection. Particularly, the FEFare involved in the timing of saccadic eye-movement generationand in optimizing the speed with which eye movements can beinitiated to objects in the visual scene. In intact animals, there iscooperative interaction between the two FEF23. Following a uni-lateral FEF lesion, target selection and processing speed are biasedin favor of the intact hemisphere. Even five months after thelesion, sizeable deficits remained on our tasks that tapped thiscompetition. In contrast, the MEF lesions yielded much smallerdeficits suggesting that this area is not as centrally involved intarget selection and in the execution of sequences of eye move-ments. This is consistent with single-cell recording studies show-ing that neuronal activity in the FEF is more predictive of saccadegeneration and timing than is the MEF (ref. 24, 25, Patterson,W.F. II & Schall, J.D. Soc. Neurosci. Abstr., 185.8, 1997).

Our paired-stimulus task is similar to a task used to studypatients with brain lesions in which paired visual or tactile stimuliare presented in the right and left hemifields or parts of the body.The tendency to ignore the stimulus that appears in space con-tralateral to the lesion has been termed the ‘extinction phenome-non.’26 Adding precisely defined temporal asynchronies to thepresentation of paired targets, as we did in this study, makes it pos-sible to obtain exact measurements of the temporal factors involved.As a result, it is possible to ascribe an exact value in time for themagnitude of the deficit incurred and its recovery. Our results raisethe possibility that the extinction phenomenon seen after frontaland parietal lesions is due not to a loss of attention per se, but to aloss in the speed of processing in the impaired hemisphere.

What brain structures might be responsible for bringing aboutthe gradual improvement in performance found on the tasks wehad used? It has been proposed that two major cortical streamscontribute to the control of visually guided saccadic eye move-ments, the anterior and the posterior27. Because even afterremoval of both major areas of the anterior stream, the FEF andthe MEF, there is notable improvement in performance over time,it is probable that the posterior stream is involved in the recovery.This stream originates in the occipital and parietal lobes andreaches the brainstem oculomotor areas predominantly throughthe superior colliculus28–30.

MethodsFour monkeys were trained on a variety of visually guided eye-movementtasks that allowed us to assess the relative effects of frontal eye field, dor-somedial frontal cortex and combined lesions on the execution of sac-cadic eye movements to single and paired target stimuli. Followingtraining, in the first animal, a unilateral lesion was made of the left FEF. Inthe second animal, successive lesions of the left dorsomedial frontal cor-tex and of the right FEF were made several months apart. In the third ani-mal, three lesions were made, also several months apart; initially the leftdorsomedial frontal cortical area containing the left MEF was ablated,then the right MEF area, and finally the left FEF. The monkeys were test-ed extensively after each lesion for several months before the second andthird lesions were made. The fourth monkey served as a control animal.

All but one of the lesions were made by aspiration under aseptic con-ditions in anesthetized animals with the aid of a surgical microscope.The FEF was removed by aspirating the anterior bank of the arcuate sul-

cus 6 mm medial and 6 mm lateral from the posterior tip of the princi-pal sulcus as well as the gray matter between the arcuate and the posteriortip of the principalis. The fundus and the posterior bank of the arcuateinvolved in smooth pursuit eye movements were spared31–33. Anteriorlythese lesions encroached upon area 46. In the second animal, the leftdorsomedial cortex area was destroyed after we recorded and stimulatedthe area extensively to establish the exact location of the MEF; this par-ticular lesion was produced by repeated, closely spaced lidocaine injec-tions until neural activity was permanently shut down, resulting in alesion verified histologically and shown in Fig.1c. Subsequently in thisanimal, the right FEF was removed by aspiration and was also verifiedhistologically. In the third animal, the same region of dorsomedial frontalcortex was removed as in the second animal. The lesions extended 8 mmlaterally from the midline and 6 mm anterior and posterior from a pointthat was in the same coronal plane as the posterior tip of the principalsulcus. Detailed photographs were taken during surgery before and aftereach aspiration. Electrophysiological mapping was carried out only inthe second animal. All protocols were approved by the MIT Animal CareCommittee and followed NIH guidelines.

The animals were tested using a color monitor placed at a distance of57 cm. The head was restrained during testing, and eye movements weremeasured using the scleral search coil method. Monkeys readily per-formed 800–3000 trials per day. Background luminance for the single,paired and multiple target tasks was 2.26 cd per m2. The targets weresmall, 0.34 degree of visual angle squares with a luminance of 90.27 cd perm2. Saccadic latencies were computed from the time of target onset tothe time of departure of the eye-movement trace from an electronic win-dow set around the fixation spot. The targets most commonly appearedat an eccentricity of 12 degrees from fixation.

The sequential task was made distinct from the other tasks in fourways. First, twenty-five equally spaced small outline squares of low con-trast were present on the screen throughout. Second, the targets appearedinside the selected squares and were bright red in color. Third, the tar-gets flashed on briefly in rapid succession, with a gap between them.Fourth, the animal was rewarded only when two correct successive sac-cadic eye movements were made to the successively appearing targetpositions. The target durations and the delays introduced between thefirst and second target were arrived at experimentally and were chosento optimize the ability of monkeys to perform the sequential task suc-cessfully. Several different sets or paired target locations were studied,which were run in blocks always using four pairs arranged in a mirror-image fashion to allow us to compare performance in the left and righthemispheres. The first target duration ranged between 33 and 300 ms,the interval between 50 and 117 ms and the second target between 33and 83 ms using 16.7 ms steps (the frame rate for the monitor). Withineach block the sequence durations as well as the mirror-imaged pair loca-tions appeared in randomized order. (See inset in Fig. 4 showing eyemovements made to one set of four pairs used in a block.)

For the temporal-discrimination task, eight stimuli appeared, one ofwhich preceded the others by various times ranging from 16 ms to 300ms. The eight stimuli were identical and the same size and luminance asthe targets presented in the single- and paired-target case. All the stimuliwere equally spaced around a circle with a radius of 12 degrees centeredon the fixation spot.

AcknowledgementsWe thank J. Colby and W. Slocum for their technical assistance.


1. Bizzi, E. Discharge of frontal eye field neurons during eye movements inunanesthetized monkeys. Science 157, 1588–1590 (1967).

2. Bruce, C.J. & Golberg, M.E. Primate frontal eye fields: I. Single neuronsdischarging before saccades. J. Neurophysiol. 53, 603–635 (1985).

3. Chen, L.L. & Wise, S.P. Supplementary eye field contrasted with frontal eyefield during acquisition of conditional oculomotor associations. J.Neurophysiol. 73, 1122–1134 (1995).

4. Mann, S.E., Thau, R. and Schiller, P.H. Conditional task-related responses inmonkey dorsomedial frontal cortex. Exp. Brain Res. 69, 460–468 (1988).


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


20. Gaymard, B., Pierrot-Deseilligny, C. & Rivaud, S. Impairment of sequencesof memory-guided saccades after supplementary motor area lesions. AnnalsNeurol. 28, 622–626 (1990).

21. Mushiake, H., Inase, M. & Tanji, J. Neuronal activity in the primate premotor,supplementary and precentral motor cortex during visually guided andinternally determined sequential movements. J. Neurophysiol. 66, 705–718(1991).

22. Tanji, J. & Shima K. Role for supplementary motor area cells in planningseveral movements ahead. Nature 371, 413–416 (1994).

23. Schlag, J., Dassonville, P. & Schlag-Rey, M. Interaction of the two frontal eyefields before saccade onset. J. Neurophysiol. 79, 64–72 (1998).

24. Hanes, D.P. & Schall, J.D. Neural control of voluntary movement initiation.Science 274, 427–430 (1996).

25. Hanes, D.P., Patterson, W. F. & Schall, J.D. The role of frontal eye fields incountermanding saccades: visual, movement, and fixation activity. J.Neurophysiol. 79, 817–834 (1998).

26. Bisiach, E. & Vallar, G. in Handbook of Neuropsychology Vol. 1 (eds Boller, E. &Grafman, J.) 195–222 (Elsevier, Amsterdam, 1988).

27. Schiller, P.H. in Cognitive Neuroscience of Attention. (ed Richards, J.) 3–50(Lawrence Erlbaum, London, 1998).

28. Keating, E.G., Gooley, S.G., Pratt, S. & Kelsey, J. Removing the superiorcolliculus silences eye movements normally evoked from stimulation of theparietal and occipital eye fields. Brain Res. 269, 145–148 (1983).

29. Schiller, P.H. The effect of superior colliculus ablation on saccades elicited bycortical stimulation. Brain Res. 122, 154–156 (1977).

30. Tehovnik, E.J., Lee, K-M. & Schiller, P.H. Stimulation-evoked saccadesfrom the dorsomedial frontal cortex of the rhesus monkey followinglesions of the frontal eye fields and superior colliculus. Exp. Brain Res. 98,179–190 (1994).

31. Gottlieb, J.P., Bruce, C.J. & MacAvoy, M.G. Smooth eye movements elicitedby microstimulation in the primate frontal eye field. J. Neurophysiol. 69,786–799 (1993).

32. Keating, E.G. Lesions of the frontal eye field impair pursuit eye movementsbut preserve the predictions driving them. Behav. Brain Res. 53, 91–104(1993).

33. Lynch, J.C. Saccade initiation and latency deficits after combined lesions offrontal and posterior eye fields in monkeys. J. Neurophysiol. 68, 1913–1916(1992).

articles

5. Robinson, D.A. & Fuchs, A.F. Eye movements evoked by stimulation offrontal eye fields. J. Neurophysiol. 32, 637–648 (1969).

6. Russo, G.S. & Bruce, C.J. Neurons in the supplementary eye field of therhesus monkeys code visual targets and saccadic eye movements in anoculocentric coordinate system. J. Neurophysiol. 76, 825–848 (1996).

7. Schall, J.D. Neuronal activity related to visually-guided saccades in the frontaleye fields of rhesus monkeys: comparison with supplementary eye fields. J.Neurophysiol. 66, 530–579 (1991).

8. Schlag, J. & Schlag-Rey, M. Evidence for a supplementary eye field. J.Neurophysiol. 57, 179–200 (1987).

9. Shook, B.L., Schlag-Rey, M. & Schlag J. Primate supplementary eye field: I.Comparative aspects of mesencephalic and pontine connections. J. Comp.Neurol. 301, 618–642 (1990).

10. Tehovnik, E.J. The dorsomedial frontal cortex: eye and forelimb fields. Behav.Brain Res. 67, 147–163 (1995).

11. Tehovnik, E.J. & Lee, K.M. The dorsomedial frontal cortex of the rhesusmonkey. Topographic representation of saccades evokes by electricalstimulation. Exp. Brain Res. 96, 430–442 (1993).

12. Deng, S-Y., Goldberg, M.E., Segraves, M.A., Ungerleider, L.G. & Mishkin, M.in Adaptive Processes in the Visual & Oculomotor Systems (eds Keller, E. & Zee,D. S.) 201–208 (Pergamon, Oxford, 1986).

13 Dias, E.C., Kiesau, M. & Segraves, M.A. Acute activation and inactivation ofmacaque frontal eye field with GABA-related drugs. J. Neurophysiol. 74,2744–2748 (1995).

14. Latto, R.A. & Cowey, A. Visual field defects after frontal eye field lesions inmonkeys. Brain Res. 30, 1–24 (1971).

15. Schiller, P.H., Sandell, J.H. & Maunsell, J.H.R. The effect of frontal eye fieldand superior colliculus lesions on saccadic latencies in the rhesus monkey. J.Neurophysiol. 57, 1033–1049 (1987).

16. Sommer, M.A. & Tehovnik, E.J. Reversible inactivation of macaque frontaleye field. Exp. Brain Res. 116, 229–249 (1997).

17. Ottes, F.P., Van Gisbergen, J.A.M. & Eggermont, J.J. Metrics of saccade responsesto visual double stimuli, two different modes. Vision Res. 24, 1169–1197 (1984).

18. Robinson, D.A. Eye movements evoked by collicular stimulation in the alertmonkey. Vision Res. 12, 1795–1808 (1972).

19. Schiller, P.H., True, S.D. & Conway, J.L. Paired stimulation of the frontal eyefields and the superior colliculus of the rhesus monkey. Brain Res. 179,162–164 (1979).


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Our visual system can often recognize objects not only on the basisof their static appearance but also by observing how they move.In some cases, impoverished image sequences can be recognizedfrom their pattern of motion despite the fact that no single framehas enough figural information to support recognition (Fig. 1).Image sequences such as those devised by Johansson1, which con-vey vivid impressions of humans engaged in various dynamic activ-ities, are elegant and powerful demonstrations of this fact.

Since the first reports of this work, the issue of how motionsequences are interpreted by the primate brain has been exten-sively studied. Previous work2 has shown that observers canidentify human individuals from their gait alone. Coding the-ory principles have also been applied to model gait percep-tion3. A model for the interpretation of these sequences hasbeen developed4 by adapting structure-from-motion ideas5.Similar approaches have also been developed by otherresearchers 6,7. Neurons that respond selectively to biologicalmotion sequences have been described in the visually respon-sive areas of the primate temporal cortex8, a region that isknown to be involved in object recognition.

Most of the studies that have attempted to explain humanperception of biological motion sequences have done so largely interms of bottom-up mechanisms, in which the object geometryis extracted from low-level features without recourse to higher-level internal representations of objects. In this report, however,we examine the possibility that internal object representationsmay also play a top-down role in this process. Our stimuli werestereo views of walking human figures that were defined by asmall number of dots; we have used depth-distorted versions ofthese figures to study the interactions between depth cues andrecognition. Our results provide two pieces of evidence in thisregard. First, they suggest that the anomalous stereo-depth cuesdo not significantly influence the recognizability of the stimuli.

Second, they show that top-down recognition-based influencescan strongly alter depth perception, such that expectations abouta familiar object’s 3D structure override the true stereoscopicinformation. Consistent with this hypothesis, we have found thatreducing the recognizability of objects reduces the magnitude ofthe top-down influence on depth perception.

ResultsAs stimuli, we used variants of previously described biologi-cal motion sequences1 (see Fig. 1). They were 3D stereo ani-mations showing twelve points on a male human (three pointspositioned at the joints of each limb) as he walked on a tread-mill at a normal pace. We compared the normal version of thisstimulus with a ‘depth-scrambled’ version in which the depthpositions of the joints were randomly altered in the z-axis. Thisrendered their 3D trajectories arbitrary within the volumedefined by the original structure, while leaving their 2D pro-jections on the retina unchanged (discounting the minor posi-tional shifts induced by the need to incorporate depth-disparityinformation). The degree of arbitrary depth scrambling of thescrambled walker was a continuously variable parameter. Wedefined the extent of added depth noise as a function of thedepth-extent (say, D) of the original undistorted walker. Thus,a noise level of 0 % corresponded to an undistorted sequence,whereas a noise level of 50 % implied a sequence wherein theindividual body points could assume any depth value (withuniform probability) within a bound of D/2 about their orig-inal depth position. Additionally, we generated ‘random’ ver-sions of both the normal and the depth-scrambled walkers byscrambling the x and y positions of their constituent points.Unlike the purely depth-scrambled version, this created dis-tortions of the 2D retinal projection. In all sequences, theadded offsets were kept constant from frame to frame.

Top-down influences on stereoscopicdepth-perception

Isabelle Bülthoff1, Heinrich Bülthoff1 and Pawan Sinha2

1 Max-Planck-Institut für biologische Kybernetik, 72076 Tübingen, Germany2 Department of Psychology, University of Wisconsin, Madison, Wisconsin 53706, USA

Correspondence should be addressed to I.B. ([email protected])

The interaction between depth perception and object recognition has important implications for thenature of mental object representations and models of hierarchical organization of visualprocessing. It is often believed that the computation of depth influences subsequent high-levelobject recognition processes, and that depth processing is an early vision task that is largely immuneto ‘top-down’ object-specific influences, such as object recognition. Here we present experimentalevidence that challenges both these assumptions in the specific context of stereoscopic depth-perception. We have found that observers’ recognition of familiar dynamic three-dimensional (3D)objects is unaffected even when the objects’ depth structure is scrambled, as long as their two-dimensional (2D) projections are unchanged. Furthermore, the observers seem perceptuallyunaware of the depth anomalies introduced by scrambling. We attribute the latter result to a top-down recognition-based influence whereby expectations about a familiar object’s 3D structure over-ride the true stereoscopic information.


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

In our first experiment (recognizability experiment), weasked whether randomizing the depth structure of the movingfigure while preserving its 2D traces would adversely affect itsrecognizability as a human. Subjects viewed in stereo the dis-torted walker sequences interspersed with random and human(unscrambled) sequences. The viewing position was variedbetween 0° (the figure is seen walking in place facing right; depthdistortion does not affect the positions of the points in the imageplane) and 90° (the walker is walking toward the observer; thesame distortion now results in displaced points in the new imageplane, see Fig. 2). The subjects rated all sequences for their struc-tural goodness as a human on a scale from 1 (completely ran-dom) to 5 (completely human). We expected that if stereo-depthinformation were critical for the recognition processes, subjectswould perceive our depth-scrambled sequences as randomobjects from all viewpoints because the 3D structure of thedepth-distorted walker would be completely different from ahuman figure. The ratings assigned to these sequences would,therefore, be uniformly low for all viewing positions. If, howev-er, recognition required merely 2D congruence, then the rating

would be expected to increase in going from a viewing directionperpendicular to the distorted depth axis (90°) to one parallelto it (0°), as the monocular view comes to resemble more close-ly a 2D human figure.

The key result (Fig. 3) is that irrespective of the depth struc-ture of the sequences, viewpoints preserving the ‘normal’ 2D pro-jections yielded biological motion percepts (high ratings). Otherviewpoints for the scrambled sequence yielded percepts of ran-domly moving dots (low ratings). The data strongly indicate thatthe recognition process used by the subjects in this task is heav-ily biased towards 2D traces. Stereo-depth information does notseem to contribute significantly to the recognition processes.

The most surprising result of this experiment is that depth-scrambled motion sequences that had ‘normal’ 2D traces wererated as highly as unscrambled sequences. There are at least twopossible explanations for this. Either subjects might be percep-tually aware of the depth scrambling but decide nevertheless tobase their ratings on the similarity of the 2D projection, or theymight be perceptually unaware of the depth scrambling, possi-bly due to a top-down object-specific influence that activelyimposes the expected structure on the input and thus suppress-es the perception of 3D anomalies.

To distinguish between these possibilities, we designed a sec-ond experiment (depth-plane experiment) to test for the exis-tence of any recognition-dependent influences that might serve tosuppress information about depth-anomalies being provided bylow-level stereo processes. To assess observers’ ability to perceivethe true depth structure of these sequences, we designed a simpletask that required them to report whether three indicated pointsin the structure were in the same fronto-parallel depth plane.Stereo viewing was used throughout the experiment. Each exper-imental trial commenced with a presentation of either a depth-scrambled walker or a random pattern for the duration of onewalk-cycle, this being sufficient time to allow the moving figureto be recognized. As the presentation continued into the nextcycle, three of the points were highlighted by thin red outlines.After two-thirds of the duration of a walk-cycle, the screen thenturned blank. Subjects had been instructed to report whether thethree red dots lay in the same fronto-parallel depth plane. In 50 %of the sequences, the three dots were in the same plane and in50% they were not. In the depth-distorted figures, we could varythe depth of the highlighted dots independent of their positionson the limbs (same versus different limb); this allowed us to askwhether the perceived depth was influenced by the expectationthat points on the same limb would be at the same depth.

Fig. 1. Example of stimulus. (a) Stereogramof a single frame of a motion sequence of ahuman walker represented by dots only. Thestructure of a human walker is only recog-nizable if the dots begin to move accordinglyto their biological motion pattern for asshort as 100 to 200 ms1. (b) We illustratethe structure of the human figure by addingconnecting lines to the dots.

a

Z

(a)

90°0°

X

Y

Fig. 2. In the recognizability experiment, subjects viewed depth-scrambled and normal sequences in stereo from different viewing positionsindicated by the arrows along the equator at waist level in (a) and (b). (a) Undistorted walker. (b) Depth-scrambling a biological motionsequence involves adding depth noise to the positions of the joints while leaving their 2D positions (in the xy-plane) largely unchanged. Fromother viewing positions (e.g., in the yz projection shown on the right), the original 2D pattern of a human figure is severely distorted. In all fig-ures, connecting lines serve to enhance recognizability of the human figure and are not shown in the experiments.

a Y

Z

(b)

90°0°

X

b

b


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

In Fig. 4a and b, thefalse-alarm rate (a responseof ‘in same plane’ when thedots are in fact not in thesame plane) is shown as afunction of depth-noiselevel and depth-disparityrespectively. The curvesshow data for three stimu-lus types: (1) humansequences with dots on thesame limb, (2) humansequences with dots on dif-ferent limbs, and (3) ran-dom sequences. If theperceived depth is affectedby prior expectations based on object recognition, the dots onthe same limb would be most likely to be perceived as copla-nar, and those on different limbs would be least likely. The falsealarm rate is highest for the ‘same limb’ condition. In Fig. 4c,the hit-rate (correct responses of ‘in same plane’) is shown asa function of depth noise. It is uniformly high for the ‘samelimb’ condition (average value, 94%, SE 1%), lower for the ran-dom sequences (average value, 72%, SE 1%) and lowest for the‘different limbs’ condition. Thus, the perceived depth is influ-enced by prior expectations about the depth structure of theimage, which are in turn determined by its recognizability.

DiscussionOur results have important implication for the nature of the men-tal representations for dynamic objects. The limited influence ofdepth information on object recognition observed in the first

experiment suggests that the recognition process is based large-ly on matching the stimulus to an internal representation of theobject’s 2D trace-structure rather than its 3D geometry. Consis-tent with this idea, the monkey infero-temporal cortex, which isknown to be involved in object recognition, has recently beenreported to contain ‘view-tuned’ neurons, which respond to 3Dobjects only when they are seen from a certain viewpoint9. Analternative possibility, which we must consider, is that recogni-tion might involve structure-from- motion processes. This ideais based on the fact that a vivid perception of 3D structure canarise from the 2D projection of a rotating object, in the absenceof stereoscopic depth cues. Structure-from-motion perceptioncan occur independent of object recognition, because even unfa-miliar rotating objects give rise to a 3D percept. Our subjectsmight therefore have derived a 3D structure from the moving2D projection of the walking figure and matched this to an inter-

nal 3D representation. Webelieve, however, that this isunlikely, because it has been10

demonstrated that recogni-tion based on 2D cues pro-ceeds unhindered even whenthe 3D structure suggested bystructure-from-motionprocesses is inconsistent withthe object identity.

We interpret the results ofthe depth-plane experimentas pointing to the existence ofa top-down influence capableof modulating the informa-tion provided by the earlydepth-perception processesbased on binocular disparities.There are other related exam-ples, e.g. the hollow maskeffect11 and the cyclopeanNecker-cube12 (in which anambiguous 2D image givesrise to two alternating per-cepts with different depthstructures in conflict with thedisparity given by the stere-ogram), which argue in favorof the influence of high-levelcues on depth perception.This influence can, in turn, be

Fig. 4. Results of the depth-plane experiment aver-aged across 11 subjects. The false alarm rate (aresponse of ‘in same plane’ when the dots are physi-cally not in the same plane) is plotted against themaximum random depth-distortion allowed in thesequence in (a) and against the maximum disparity inpixels (each pixel subtends 0.015 degrees of visualangle) between the three highlighted dots in (b). Forcomparison the hit rate (a correct response whenthe three dots are in the same plane) is shown in (c).Three conditions are plotted in each graph.__ Human figure with the 3 marked dots on thesame limb; ._._._. Human figure with the 3 markeddots on different limbs; …….. Random figure. Theviewing position for all trials was 0° (the ‘walker’ is seen walking to the right); this viewing position insured goodrecognizability of the moving figure. The specific noise levels we used for the distorted walker in this experimentwere 0, 25, 50, 100, 150, and 200 %, which is around the noise level used in the first experiment.

Fig. 3. Results of the recognizability experimentaveraged across 22 subjects. The following stimulussequences were presented: human walker (no dis-tortion), depth-distorted walker (a constant z-dis-tortion was applied from frame to frame, the amountof depth noise added was 100%) and random pattern(constant xz-distortion). The subjects rated them ona scale from 1 (very random) to 5 (very human).Abscissa: viewing position in degrees, at 0° thewalker is seen walking to the right with its depth axisparallel to the viewing axis; at 90° the walker is seenwalking towards the observer with its depth axisperpendicular to the viewing axis. P z-distortion, Rxz-distortion, H no distortion. n = 22.

1

2

3

4

5

0° 9 ° 18 ° 27 ° 36 ° 45 ° 54 ° 63 ° 72 ° 81° 90 °

z-distxz-distno-dist

Stru

ctur

al g

oodn

ess

as a

hum

an

Viewing position (deg)

n=22

z-distortionxz-distortionno distortion

Stru

ctu

ral g

oo

dn

ess

as a

hu

man

Viewing position (deg)

0% 50% 100% 150% 200%

Fals

e al

arm

rat

e

Noise level

100%

80%

60%

40%

20%

0%

n=11

0%

20%

40%

60%

80%

100%

5 10 15 20 25 30

Fals

e al

arm

rat

e

Disparity (pixels)

n=11

b

0%

20%

40%

60%

80%

100%

0% 50% 100% 150% 200%

Noise Level

n=11

c

Hit

rat

e

Noise level

Hit

rat

e

c

Disparity (pixels)

Fals

e al

arm

rat

e

Fals

e al

arm

rat

e

Noise level

a

n = 11

n = 11

n = 11

b


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

modulated by factors that change the recognizability of a stimu-lus. This hypothesis would explain why the human sequences giverise to false depth perception so much more frequently than dothe random sequences. Other important factors that need to beconsidered in interpreting these results include grouping inducedby common motion or proximity. That is, dots may be more like-ly to be perceived as coplanar if they move in synchrony or if theyare close together. We believe, however, that in our experimentsthe effect of such factors would be relatively limited, because thesequences in the different conditions had very similar mid-levelattributes such as motion and density distributions. Specificallythe differences in performance between the walker and the ran-dom conditions suggest the importance of object-specific, recog-nition-based influences over general configurational ones. Themotion trajectories of the individual dots were the same in bothconditions, and only their 2D offsets were randomized. Thus, twodots of the walker that moved in phase continued to do so in therandom stimulus, thereby largely maintaining the mid-level group-ing cues and the articulation geometry in the two conditions. Yet,the suppression of binocular disparity perception occurs only whenthe walker is recognizable.

The idea that top-down influences can affect perception iscertainly not a new one. Several well known visual illusions,such as the Dalmatian dog picture13 or the mother-in-law/daughter-in-law figure14 demonstrate the significance oftop-down expectations in interpreting ambiguous stimuli.Recent computational models of the neocortex have argued thatfeedback cortico-cortical projections might allow top-downinfluences to propagate from the higher cortical areas that areinvolved in object recognition back to the earlier areas that sup-port lower-level processes15,16. Our study now provides evi-dence that even the very low-level process of stereo-depthperception, which was previously considered to be a purely bot-tom-up process17,18, is in fact susceptible to top-down influ-ences. Additionally, our experimental results provide indirectevidence that dynamic three-dimensional objects might be rec-ognized by the visual system based on their 2D traces ratherthan on their 3D structural descriptions.

MethodsThe biological motion sequences used in our experiments were based ondata collected at the Gait Analysis Laboratory of the Spaulding Rehabil-itation Hospital in Boston, Massachusetts. The data comprised the 3Dpositions of twelve points on a male human as he walked in place. Thepoint positions were updated 39 times over the course of one complete

walk cycle. The stimuli were generated by displaying each point as a brightdot (0.03 degree of visual angle) on a gray background (mean luminance:20 cd per m2). The experiments were conducted on a Silicon GraphicsIndigo 2 workstation. All sequences were presented in stereo using a pairof StereoGraphics Crystal-Eyes (TM) LCD shutter glasses synchronizedwith the display. Two views were generated for each frame to allow stereo-scopic vision. All subjects were tested to ensure that they had functioningstereoscopic ability, two subjects were rejected and all subjects were naiveas to the purpose of the experiments.

AcknowledgmentsWe wish to thank N. Logothetis, B. Tjan and D. Kersten for insightful comments

on earlier versions of the manuscript and P. Lipson for providing the biological

motion data set.

RECEIVED 29 JANUARY: ACCEPTED 22 MAY 1998

1. Johansson, G. Visual perception of biological motion and a model of itsanalysis. Percept. Psychophys. 14, 201–211 (1973).

2. Cutting, J. E. & Kozlowski, L. T. Recognition of friends by their walk. Bull.Psychonom. Soc. 9, 353–356 (1977).

3. Cutting, J. E. Coding theory adapted to gait perception. J. Exp. Psychol. Hum.Percept. Perform. 7, 71–87 (1981).

4. Hoffman D. D. & Flinchbaugh B. E. The interpretation of biological motion.Biol. Cybern. 42, 195–204 (1982).

5. Ullman S. The Interpretation of Visual Motion (MIT Press, Cambridge, 1979).6. Shibata T., Sugihara, K., & Sugie, N. Recovering three-dimensional structure

and motion of jointed objects from orthographically projected optical flow.Trans. IECE 68-D, 1689–1696 (1985).

7. Webb J. & Aggarwal, J. Structure from motion of rigid and jointed objects.Artif. Intell. 19, 107–131 (1982).

8. Oram, M. W. & Perrett, D. I. Responses of anterior superior temporalpolysensory (STPa) neurons to ‘biological motion’ stimuli. J. Cog. Neurosci. 6,99–116 (1994).

9. Logothetis, N. K., Pauls, J. & Poggio, T. Shape recognition in the inferiortemporal cortex of monkeys. Curr. Biol. 5, 552–563 (1995).

10. Sinha, P. & Poggio, T. Role of learning in three-dimensional form perception.Nature 384, 460–463 (1996).

11. Gregory, R. L. in Illusion in Nature and Art (eds Gregory, R. L. & Gombrich, E.H.) 49–96 (Duckworth, London, 1973).

12. Julesz, B. Foundations of Cyclopean Perception (Univ. of Chicago, Chicago,1971).

13. Goldstein, B. Sensation and Perception (Brooks/Cole, Pacific Grove, 1996).14. Boring, E. G. A new ambiguous figure. Am. J. Psychol. 42, 444–445 (1930).15. Ullman, S. Sequence seeking and counter streams: a computational model for

bidirectional information flow in the visual cortex. Cereb. Cortex 5, 1–11(1995).

16. Mumford, D. On the computational architecture of the neocortex. II. Therole of cortico-cortical loops. Biol. Cybern. 66, 241–252 (1992).

17. Julesz, B. Early vision and focal attention. Rev. Mod. Phys. 63, 735–772(1991).

18. Nakayama, K., Shimojo, S. & Silverman, G. H. Stereoscopic depth: Its relationto image segmentation, grouping, and the recognition of occluded objects.Perception 18, 55–68 (1989).


199

8 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

Documents

Nature Neuroscience July 1998