Nature Neuroscience September 2001

editorialStretching the definition of espionage...............................................................851

letters to the editorSynaptic scaling in vitro and in vivo.................................................................. .853

news and viewsFigure and ground in the brain.........................................................................857Nava Rubin SEE ARTICLE, PAGE 937

Protofibrils, the unifying toxic molecule of neuro-degenerative disorders?.........859Christian Haass and Harald Steiner SEE ARTICLE, PAGE 887

Neuronal migration and the evolution of the human brain................................860Yi Rao and Jane Y. Wu SEE ARTICLE, PAGE 931

Virtual neurology..............................................................................................862Robert Rafal SEE ARTICLE, PAGE 953

From neuron to BOLD: new connections...........................................................864Peter A. Bandettini and Leslie G. Ungerleider

book reviewThe many faces of memory...............................................................................867From Conditioning to Conscious Recollection: Memory Systems of the Brainby Howard Eichenbaum and Neal J. CohenREVIEWED BY LARRY R. SQUIRE

brief communicationsLactate enhances the acid-sensing Na+ channel on ischemia-sensing neurons.........869D C Immke and E W McCleskey

Opposing actions of protein kinase A and C mediate Hebbian synaptic plasticity.....871M X Li, M Jia, H Jiang, V Dunlap and P G Nelson

contents

http://neurosci.nature.com

volume 4 no 9 september 2001

The assignment of figure andground for a given visual displaycan dramatically alter the shapethat human observers perceive.For example, depending on thefigure/ground assignment in theimage shown here, one seeseither a head-on view of awoman, or two profiles. Baylisand Driver report a neural corre-late of this effect: neurons in infe-rior temporal cortex of behavingmonkeys respond to componentsof visual shape only when theyare perceived as figure ratherthan ground. Shape analysis thusseems to be intertwined with fig-ure/ground segregation. Coverimage copyright 2001, created byRoger Shepard, reprinted withpermission from his book MindSights, W H Freeman, 1990. See pages 857 and 937

nature neuroscience • volume 4 no 9 • september 2001 i

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 Neuro-science Subscription Department, PO Box 5054, Brentwood, TN 37024-5054. Telephone (800) 524-0384, Direct Dial (615) 377 3322, Fax (615) 377 0525. Outside North America:Nature Neuroscience, Macmillan Magazines Ltd., Houndsmill, Brunel Road, Basingstoke, RG21 6XS, U.K.. Tel: +44-(0)1256-329242. Fax: +44-(0)1256 812358. Email: [email protected]. Annual subscription rates: U.S./Canada: U.S. $650, Canada add 7% for GST (institutional/corporate), U.S. $199, Canada add 7% for GST (individualmaking personal payment BN: 14091 1595 RT); U.K./Europe:£435 (institutional/corporate), £185 (individual making personal payment), £99 (student); Rest of world (excludingJapan): £480 (institutional/corporate), £195 (individual making personal payment), £110 (student); Japan: Contact Japan Publications Trading Co. Ltd., 2-1 Sarugaku-cho 1 chome,Chiyoda-ku, Tokyo 101, Japan, phone (03) 292-3755. Back issues: U.S./Canada, $45, Canada add 7% for GST; Rest of world: surface U.S. $43, air mail U.S. $45. Reprints: NatureNeuroscience Reprints Department, 345 Park Avenue South, New York, NY 10010-1707. Subscription information is available at the Nature Neuroscience homepage at http://neu-rosci.nature.com. POSTMASTER: Send address changes to Nature Neuroscience Subscription Department, P.O. Box 5054, Brentwood, TN 37024-5054. Application to mail periodicalspostage rate is pending at New York, NY. Executive Officers of Nature America Inc: Annette Thomas, President; Edward Valis, Secretary-Treasurer. Printed by Publishers Press, Shepherdsville,KY, USA. Copyright ©2001 Nature America Inc.

Human thalamic neurons fromtelencephalic precursors

Pages 860 and 931.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

© 2001 Nature Publishing Group http://neurosci.nature.com

contents

nature neuroscience • volume 4 no 9 • september 2001 ii

New transporter blockers from cone snail venom.

Page 902.

Subunit specificity of AMPAreceptor surface delivery.

Page 917.

Selective destruction of neuronsthat drive breathing rhythms.

Page 927.

Reinforcing and locomotor stimulant effects of cocaine are absentin mGluR5 null mutant mice.............................................................................873C Chiamulera, M P Epping-Jordan, A Zocchi1, C Marcon, C Cottiny, S Tacconi, M Corsi, F Orzi and F Conquet

Synchrony does not promote grouping in temporally structured displays..........875H Farid and E H Adelson

reviewThe fundamental plan of the retina...................................................................877R H Masland

articlesThe ‘Arctic’ APP mutation (E693G) causes Alzheimer’s disease by enhanced Aβ protofibril formation ...................................................................887C Nilsberth, A Westlind-Danielsson, C B Eckman, M M Condron, K Axelman, C Forsell, C Stenh, J Luthman, D B Teplow, S G Younkin, J Näslund and L Lannfelt SEE NEWS AND VIEWS, PAGE 859

Allosteric interaction between the amino terminal domain and the ligand binding domain of NR2A..................................................................894F Zheng, K Erreger, C-M Low, T Banke, C J Lee, P J Conn and S F Traynelis

Two new classes of conopeptides inhibit the α1-adrenoceptor and noradrenaline transporter...........................................................................902I A Sharpe, J Gehrmann, M L Loughnan, L Thomas, D A Adams, A Atkins, E Palant, D J Craik, D J Adams, P F Alewood and R J Lewis

GABAA receptor cell surface number and subunit stability are regulated by the ubiquitin-like protein Plic-1.....................................................908F K Bedford, J T Kittler, E Muller, P Thomas, J M Uren, D Merlo, W Wisden, A Triller, T G Smart and S J Moss

Subunit-specific temporal and spatial patterns of AMPA receptor exocytosis in hippocampal neurons...................................................................917M Passafaro, V Piëch and M Sheng

Normal breathing requires preBötzinger complex neurokinin-1 receptor-expressing neurons.............................................................................927P A Gray, W A Janczewski, N Mellen, D R McCrimmon and J L Feldman

Telencephalic origin of human thalamic GABAergic neurons..............................931K Letinic and P Rakic SEE NEWS AND VIEWS, PAGE 860

Shape-coding in IT cells generalizes over contrast and mirror reversal, but not figure-ground reversal.............................................................937G C Baylis and J Driver SEE NEWS AND VIEWS, PAGE 857

The role of withdrawal in heroin addiction: enhances reward or promotes avoidance?.......................................................................................943D M Hutcheson, B J Everitt, T W Robbins and A Dickinson

Prefontal cortex in long-term memory: an “interference” approach using magnetic stimulation ..............................................................................948S Rossi, S F Cappa, C Babiloni, P Pasqualetti, C Miniussi, F Carducci, F Babiloni and P M Rossini

Enhanced visual spatial attention ipsilateral to rTMS-induced ‘virtual lesions’ of human parietal cortex...........................................................953C C Hilgetag, H Théoret and A Pascual-Leone SEE NEWS AND VIEWS, PAGE 862

errata.........................................................................................................959

classified advertising .......................................................see back pages

Prefrontal cortex TMS interfereswith long-term memory.

Page 948.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


nature neuroscience • volume 4 no 9 • september 2001 851

In May, the US Department of Justice indicted two Japaneseresearchers, alleging that they engaged in economic espionage onbehalf of Japan. A central claim is that one of the researchers,Takashi Okamoto, stole reagents from the Cleveland Clinic Foun-dation (CCF) in Ohio, and transferred them to the RIKEN BrainScience Institute near Tokyo. Because RIKEN is funded by theJapanese government, the Justice Department considers that theftof trade secrets on its behalf constitutes foreign economic espi-onage. This is an extraordinary allegation against a major researchinstitute, but the available information suggests it is unfounded.

Until July 1999, Okamoto was a lab head at the Lerner ResearchInstitute of the CCF. According to the indictment, Okamoto con-spired with another Japanese scientist, Hiroaki Serizawa of KansasUniversity Medical Center, to steal DNA samples and cell linesfrom the CCF and transfer them to RIKEN via Serizawa’s lab.Other materials from Okamoto’s lab were said to be destroyed,and the researchers sought to conceal the damage by substitutingvials filled with tap water. Colleagues became suspicious whenexperiments started to fail, and a few weeks later Okamoto abrupt-ly resigned his position at CCF and returned to Japan to take up aposition at the Brain Science Institute. CCF reported the incidentto the local police, who turned the case over to the FBI.

Okamoto had told RIKEN that he was in a dispute with hisformer employer, but they considered it to be a personal matterand did not pursue it. Neither the FBI nor CCF informed themthat a criminal investigation was underway, according to RIKENsources, so the May indictment, almost two years after Okamotoleft CCF, took them by surprise. When it was announced, Okamoto disappeared on administrative leave, and had no fur-ther contact with RIKEN until his resignation on 31 July. Okamoto’s motives remain mysterious. His current location isunknown, and it seems unlikely that Japan will allow him to beextradited to face trial in the US, given that it does not recognizeeconomic espionage as a crime. Serizawa, however, was arrestedin Kansas and released on bail; his trial will begin in November.

Although the indictment does not explicitly allege wrongdo-ing by RIKEN, the implication seems unmistakable from the claimthat Okamoto’s and Serizawa’s actions “would and did confer abenefit on RIKEN” by providing them with the stolen reagents. Anaffidavit from the investigating FBI agent is more specific, claim-ing that RIKEN employees discussed with Okamoto how the stolenreagents could be transported and stored there.

RIKEN, which has conducted its own investigations sinceOkamoto’s disappearance, insists that it has done nothing wrong.It acknowledges that stolen reagents did reach RIKEN, but claimsthat they disappeared soon after Okamoto’s arrival and werenever used. It also acknowledges that there were e-mail discus-sions about storing boxes of reagents from CCF, but claims that

these were normal for any researcher relocating to a new insti-tute. Finally, it rejects any suggestion that Okamoto was hiredbecause of his access to reagents developed at the CCF.

RIKEN may have been sloppy in its hiring process, but that isnot the same as conducting foreign espionage. The charge seemsinherently implausible, and unless the Justice Department or theCCF can provide evidence to the contrary, RIKEN’s denials shouldbe taken at face value. Whether the Serizawa trial will provideRIKEN with an opportunity to clear its reputation, however,remains to be seen. The prosecution has asked the presiding judgeto conduct the trial in closed session, citing concerns about theeffect on CCF of revealing its trade secrets. CCF supports the pros-ecution’s request, according to public relations director MarkCohen, who declined to discuss the case further with Nature Neu-roscience. However, the prosecution has a vested interest in demon-strating the commercial sensitivity of the information, so thisargument should be treated with skepticism, particularly as thedirector of the Lerner Research Institute has been quoted as sayingthat Okamoto’s research there produced no patentable discoveries.

Most intellectual property disputes are resolved by either nego-tiation or civil litigation. A criminal charge of economic espionageis far more serious, and this case has caused considerable offensein Japan, where it has been widely reported. Finding treatments forAlzheimer’s disease is a high priority for drug companies in bothcountries, and as one prominent Japanese newspaper put it, “manyfeel the United States’ intention is to launch a pre-emptive strikeagainst an emerging rival.” This may be overstated, but the ques-tion remains why the Justice Department chose to bring espionagecharges, rather than more mundane charges of theft or vandalism.

It is hard to escape the suspicion that the charges are politi-cally motivated. The Economic Espionage Act was passed by theUS Congress in 1996, reflecting widespread concerns about thethreat to the United States from theft of its trade secrets. How-ever, no espionage convictions have yet been obtained, andindeed this case is the first indictment under section 1831 of theact, which refers to espionage by foreign governments. Economicespionage is a serious threat, and no doubt the Justice Depart-ment is under pressure to obtain convictions under the new act.It is difficult to believe, however, that the actions attributed toOkamoto and Serizawa, deplorable though they may be, repre-sent the type of serious crime that the act was meant to deter.

Meanwhile, the reputation of a major international researchinstitute has been tarnished, probably unfairly, by the JusticeDepartment’s interpretation of an individual researcher’s con-duct. It would be too much to expect an apology from the Jus-tice Department if the charges prove to be unsubstantiated, butin that event, one hopes at least that the CCF will help RIKENto restore its good name.

editorial

Stretching the definition of espionage

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



TO THE EDITOR—The recent study byMagee and Cook1 in CA1 pyramidalneurons in vitro (see also ref. 2) raises afundamental issue. Is the dependence ofsomatic EPSPs on the location of thedendritic synapses, which is expectedfrom dendritic filtering, a ‘bug’ thatshould be rectified (for example, bymechanisms that eliminate voltageattenuation in the dendritic tree), or isthis dependence a ‘feature’ thatenhances the computational capabilityof the neuron? Magee and Cook’s directdendritic measurements show that thesynaptic conductance change, gsyn,becomes larger as one moves along theapical dendrite, away from the soma.This progressive increase in gsyn coun-terbalances the voltage attenuationimposed by dendritic cable properties,and consequently, the amplitude of uni-tary somatic EPSPs is insensitive to itsdendritic origin (‘location-independent’somatic EPSPs). If the location depen-dence of soma EPSPs is indeedremoved, then “…all synapses will havethe same ability to initiate action poten-tials and to induce long-term synapticplasticity regardless of their location inthe dendritic arborization”1 and, func-tionally, the neuron could be treated asa ‘point neuron’.

But is it valid to assume that if, invitro, the size of individual somaticEPSPs is independent of the dendriticinput location, this would also remaintrue when many synapses bombard thedendritic tree, as is the case in vivo? Weshow that in the latter case, the location-independence found in the quiescent invitro condition is lost, and distal synaps-es become weaker at the soma than doproximal synapses (Fig. 1; see web sup-plement, http://www.nature.com/neuro/web_specials/, for detailed figure legend).This is the result of a several-foldincrease in dendritic membrane con-ductance, Gm, due to the activity of manysynapses in vivo3–6. In other words, pre-cisely the same mechanism of synapticconductance change that is used for scal-ing up distal synapses destroys the ‘loca-tion independence’ (it is ‘self defeating’)when the network is active.

The general argument is that if, insome reference cases, the scaling ofsynaptic conductance gives rise to loca-tion-independent EPSP amplitude at the

Synaptic scaling in vitro and in vivo

soma, any change in Gm will instanta-neously eliminate this property. In par-ticular, if Gm increases, the somaticEPSPs from distal sites will decrease rel-atively more than do somatic EPSPs thatoriginate proximally. This can be demon-strated using the simplest case of an infi-nitely long passive uniform cylinder withlinear steady-state current inputs (Iin).Voltage attenuation with distance in thiscase is exponential,V(x)∝ e-x/λ, where λ = √(d/(4RiGm) is the space constant, dis the cylinder diameter and Ri is the spe-cific axial resistance. In order to gener-ate the same V at some point (forexample, at x = 0) for all input locations,Iin(x) must increase as ex/λ to compen-sate for the exponential attenuation of V.If Gm is increased uniformly by some fac-tor, then λ is reduced by the square rootof this factor, and a steeper profile ofIin(x) is now required for preserving loca-tion-independent V at x = 0. The scalingthat was sufficient to preserve location-independence prior to the increase in Gmis now insufficient, particularly for largex values (such as distal synapses). Forexample, if the distal and proximal inputsites are 1λ apart and the distal input isscaled such that V(x=0) is identical fromthe two sites, then increasing Gm by a fac-tor of 4 results in a distal input that isonly 37% of the proximal input at x = 0.

The effect of network activity that islikely to be found in vivo on the degreeof location-independence of somaticEPSP amplitude is simulated using amodel of a CA1 neuron (Fig. 1a). In the‘in vitro’ case, a progressive increase ingsyn with distance (Fig. 1b) removes thelocation dependence and produces uni-tary somatic EPSPs with a 0.2 mV peakfor all input locations (example in Fig.1a, ‘in vitro’ case). This location inde-pendence is abolished due to networkactivity (Fig. 1c). First we show that auniform increase in Gm over the den-drites, resulting in a 4-fold reduction insoma input resistance, Rin (similar to theexperimental findings)4–6, significantlyweakens (by a factor of 5 at x = 600 µm)distal synapses (that are ‘location-inde-pendent’ in vitro) as compared to proxi-mal synapses (black line in Fig. 1c). Theother three curves in Fig. 1c incorporatethe synaptic scaling (shown in Fig. 1b)that preserves the in vitro location-inde-pendence into the in vivo simulations.

letters to the editor

Note that in these three cases, the reduc-tion in Rin underestimates the actualreduction found in vivo (25% reductionfor blue and red cases, 50% for the greencase). Because distal synapses inducelarger local conductance changes com-pared to proximal synapses, the distaldendritic membrane becomes moreshunted (and more depolarized, an effectthat was not simulated here) when manysimilar excitatory synapses bombard thedendritic tree. This ‘self-defeating’ mech-anism (Fig. 1c) dramatically weakens dis-tal synapses, and this effect is robustunder a wide range of model parameters.

Is it possible to circumvent the mutu-al synaptic shunt and still use the synap-

Fig. 1. Network activity eliminates the loca-tion independence of somatic EPSP ampli-tude found in vitro. See web supplement,http://www.nature.com/neuro/web_specials/,for detailed figure legend.

a

b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


854 nature neuroscience • volume 4 no 9 • september 2001

tic scaling mechanism to generate ‘loca-tion independent EPSPs’ in the in vivocondition? One possibility is that synaps-es in vivo are not activated randomly andasynchronously over the dendritic sur-face, as assumed in Fig. 1. Rather, groupsof equally distant synapses may fire insynchrony among themselves and out ofsynchrony with other groups of synapses(Fig. 2a). Such temporally coherent andspatially stratified synaptic activationcould potentially reduce the mutualshunting effect described above, andmight partially restore the location inde-pendence of the somatic EPSP amplitudefound in CA1 neurons in vitro. Figure 2billustrates the results for different groupsizes (10, 60 and 100 synapses pergroup). Surprisingly, the location inde-pendence found in vitro was partiallyrestored only for the small group size

erties, and it is, therefore, highly unlike-ly that this mechanism will retain loca-tion independence of somatic EPSPamplitude in the dynamic transitions(and fluctuations) that neuronal net-works undergo in vivo (for example, inCA16). Whether the location dependenceof somatic EPSPs is a ‘bug’8 or a ‘fea-ture’9 will be resolved if we continue tolisten to what neurons (as well as synaps-es) tell us, while keeping in mind that inmany instances, what they say in vitro isnot necessarily what they say in vivo.

Michael London and Idan SegevDepartment of Neurobiology and Interdisci-plinary Center for Neural Computation, TheHebrew University of Jerusalem, Jerusalem,91904, Israel.e-mail: [email protected]

Note: The source code for the simulations in this

work and detailed figure legends are available at

http://www.nature.com/neuro/web_specials/

REPLY—We welcome the intent of the let-ter by London and Segev to reaffirm thewell-known fact that network activitycan change the cable properties of neu-rons11 (also see ref. 13 for further refer-ences). We show the same effect inFigure 4d (ref. 1), and one of us hasrecently published papers specifically onthis issue12, 13. We do not agree, howev-er, with the supposition that computersimulations using unrealistic model neu-rons can tell us much of substance aboutsynaptic integration under in vivo con-ditions. If computer models are to pro-vide enhanced understanding ofneuronal integration, they need toreflect as closely as possible the condi-tions they wish to simulate.

London and Segev’s passive dendritemodel contains too many assumptionsand omissions to justify their conclu-sions. The most glaring omission is a lackof voltage-dependent conductances in thedendrites8; active properties can com-pletely alter synaptic integration and theoverall electrical behavior of dendrites.For example, in a model containing Na+

and Ca2+ channels, increased synapticactivity might generate locally initiatedspikes rather than the saturation to 0 mVshown by London and Segev. Models thatincorporate K+ channels show that thesechannels can regulate the amplitudes ofEPSPs, the threshold for local spikes, andthe shapes, amplitudes and frequency ofback-propagating spikes. Also, H- chan-nels will reduce the location dependence

(Fig. 2b, green line). For larger groups(blue and red lines), the average com-posite somatic EPSP from distal locationswas attenuated relatively more than itwas in the reference asynchronous case(black line). This is due to significantvoltage saturation when a large numberof up-scaled synapses are co-activateddistally. Moreover, composite EPSPsfrom distal synapses are further attenu-ated because they are likely to encountersubstantial shunt resulting from the syn-chronous activity of other more proxi-mal groups of synapses.

Using a steeper synaptic scaling (Fig. 2c, red line), it is still possible toobtain location-independent somaticEPSP for a given in vivo condition (Fig. 2c, horizontal red dots). However, assoon as the network statistics change (forexample, the average background firingrate increases) the location independenceis instantaneously lost (Fig. 2c, green dots).In addition, achieving location indepen-dence for a given in vivo condition is crit-ically dependent on the target somaticEPSP value and on the dendritic mor-phology, and in many cases (for example,for red dots beyond 550 µm) this is impos-sible to obtain with reasonable values forthe synaptic conductance change.

Other membrane mechanisms, suchas voltage-dependent amplification7

could still render distal and proximalsynapses equally effective at the soma,even in the presence of network activi-ty. Still other voltage-dependent mech-anisms such as Ih and IA currents, as wellas synaptic inhibition, are expected toeffectively increase Gm, thus intensifyingthe location dependence of somaticEPSP amplitude. It remains to be shownexperimentally whether, indeed, den-dritic attenuation is actually removed inthe in vivo condition. Such experimentsare currently feasible, including intra-cellular recordings from pairs of synap-tically connected neurons in vivo, as wellas the use of two-photon microscope formeasuring unitary somatic EPSPs fol-lowing the activation of a single den-dritic synapse.

The main purpose of this letter is toemphasize that the behavior of unitarysynaptic EPSPs found in vitro is boundto be markedly different when the neu-ron is embedded in an active network.Experiments confirm that network activ-ity changes the cable properties of thepostsynaptic neuron dramatically. Themechanism of synaptic scaling that pre-serves ‘location independence’ in vitro ishighly sensitive to dendritic cable prop-


Fig. 2. Attempting to restore location-indepen-dence of somatic EPSP amplitude in vivo usingthe mechanism of synaptic scaling. See web sup-plement, http://www.nature.com/neuro/web_specials/ for detailed figure legend.

a

b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



of EPSP decay and temporal summation,and thereby drastically alter the way inwhich dendrites respond to the patternsof inputs used by London and Segev.

Other problems in the model by Lon-don and Segev include the use of uniformsynapse types and synaptic densities (forexample, proximally located inhibitionproduces much of the somatic shuntingseen in vivo)14, the use of very slow kinet-ics for AMPA conductances (more realis-tic kinetics would increase the requiredsynchrony that was used by London andSegev)15, and the complete omission ofvoltage-dependent NMDA conductances(NMDARs reduce the impact of the typeof network activity used by London andSegev)13. We would also like to remindreaders that our data and conclusionscovered input from only a single synap-tic pathway, which is located in the regionof the dendrites that is the least sensitiveto London and Segev’s simulated changesin input patterns. (Schaffer collaterals arewithin ∼ 300 µm of the soma.) Whetherother more distal pathways might use thesame normalizing mechanism or be nor-malized to the same level is simplyunknown (although the large size, com-plicated geometry and sparse density oftuft spines suggest that they may indeedhave a larger conductance)16.

In short, we find the modeling of Lon-don and Segev to be accurate and infor-mative only within the confines ofexamining the impact of synaptic conduc-tances on passive cables. Given what wenow know about dendritic physiology, webelieve that their simulations do not pre-sent a realistic picture of neurons in vivo.

1. Magee, J. C. & Cook, E. P. Nat. Neurosci. 3,895–903 (2000).

2. Iansek, R. & Redman, S. J. J. Physiol. (Lond.)234, 665–688 (1973).

3. Rapp, M., Yarom, Y. & Segev, I. NeuralComput. 4, 518–533 (1992).

4. Borg-Graham, L. J., Monier, C. & Fregnac, Y.Nature 393, 369–373 (1998).

5. Pare, D., Shink, E., Gaudreau, H., Destexhe, A.& Lang, E. J. J. Neurophysiol. 79, 1450–1460(1998).

6. Kamondi, A., Acsady, L., Wang, X. J. &Buzsaki, G. Hippocampus 8, 244–261 (1998).

7. Reyes, A. Annu. Rev. Neurosci. 24, 653–675(2001).

8. Graham, B. in ICANN (Proceedings of NinthInternational Conference on Artificial NeuralNetworks) 1006–1011 (Institute of Electricaland Electronics Engineers ConferencePublication No. 470, 1999).

9. Poirazi, P. & Mel, B. W. Neuron 29, 779–796(2001).

10. Hines, M. L. & Carnevale, N. T. NeuralComput. 9, 1179–1209 (1997).

11. Bernander, O., Douglas, R. J., Martin, K. A. &Koch, C. Proc. Natl. Acad. Sci. USA 88,11569–11573 (1991).

12. Cook, E. P. & Johnston, D. J. Neurophysiol. 81,535–543 (1999).

13. Cook, E. P. & Johnston, D. J. Neurophysiol. 78,2116–2128 (1997).

14. Papp, E., Leinekugel, X., Henze, D. A., Lee, J.& Buzsaki, G. Neuroscience 102, 715–721(2001).

15. Forti, L., Bossi, M., Bergamaschi, A., Villa,A. & Malgaroli, A. Nature 388, 874–878(1997).

16. Megias, M., Emri, Z., Freund, T. F. & Gulyas,A. I. Neuroscience 102, 527–540 (2001).

17. Magee, J. C. Nat. Rev. Neurosci. 1, 181–190(2000).

18. Ulrich, D. & Stricker, C. J. Neurophysiol. 84,1445–1452 (2000).

It is clear that neuronal dendrites are farmore than passives cables and, as a result,support a wider range of functionality thandepicted by the model of London andSegev. Furthermore, CA1 pyramidal neu-rons should indeed discriminate amongdifferent spatio-temporal patterns ofsynaptic input, but not in the way suggest-ed by the passive cable model of Londonand Segev17. We would expect CA1 den-drites to be capable of linearly summatinglower levels of synaptic activity withoutrespect to location (at least for Schaffer col-laterals)18. However, we do not believe thatthe most important result of increasedsynaptic activity is a change in the cableproperties of the dendrites. Instead, we pre-dict that high levels of synaptic input willmove dendrites into a completely differentintegration mode, one that is more non-linear and that perhaps includes local spikeinitiation9 (see ref. 17 for further refer-ences). Such a wide range of processing ismade available by the wonderfully com-plex, nonlinear properties of dendrites.

In closing, it is true that dendritic cableproperties are a foundation upon whichdendritic function is constructed. How-ever, when one views a remarkable struc-ture, it is always most enlightening to lookat more than just its foundation.

Jeffrey C. Magee1 and Erik P. Cook2

1 Louisiana State University Medical Center,2020 Gravier St., New Orleans, Louisiana70112, USA.2 Howard Hughes Medical Institute, BaylorCollege of Medicine, One Baylor Place,Houston, Texas 77050, USA.e-mail: [email protected]


We welcome short letters on matters arising from previous papers in Nature Neuro-science or on other topics of widespread interest to the neuroscience community.Because space in this section of the journal is limited, priority is given to short (fewerthan 500 words), well-written letters addressing the most topical issues. Typically, newdata are not presented in this section, although they may occasionally be allowed at thediscretion of the editors. Letters concerning material previously published in Nature Neuroscience are usually sent to the authors of the original piece for their commentsand/or a formal reply. References, if absolutely necessary, should be restricted to ten orfewer. Letters may be edited for brevity and clarity.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Study the shapes in Fig. 1a for a moment.Now look at Fig. 1b: which shape was pre-sent in 1a—the top or the bottom one?Many observers answer “top”, not noticingthat Fig. 1a also contains a region that cor-responds to the bottom shape. The reasonobservers often do not notice this initially isthat this region is attributed to the back-ground. The observation that the shape ofbackground regions is often not registeredperceptually dates back to the Gestalt psy-chologist Edgar Rubin1,2, who wrote, “whatis perceived as figure and what is perceivedas ground do not have shape in the sameway. In a certain sense, the ground has noshape.” The different perceptual quality offigure and ground has profound ecologi-cal justification. Imagine rearranging thesurfaces in Fig. 1; their own shapes wouldremain intact, but the shape of the back-ground regions would change entirely.Whereas the shape of a figure is intrinsicto it, the shape of the ground is an acci-dental outcome of a specific arrangementof figural objects. It therefore makes sensethat shape analysis (such as for recogniz-ing objects) should be performed only forfigural regions. To encode or analyze theshape of background regions would be awaste of resources.

Papers from Baylis and Driver in thisissue of Nature Neuroscience3 and Kourtziand Kanwisher in this week’s issue of Sci-ence4 present evidence for a neural corre-late of the differential treatment of figureand ground by the brain. Baylis and Dri-ver3 studied the inferotemporal (IT) cor-tex of awake behaving monkeys, whichcontains cells whose responses are oftenselective to specific, complex shapes5. Theyasked, when a cell is selective to a certainshape, does it matter whether this shape ispresented as figure or as ground? Theiranswer is a resounding “yes.” The authorsdevised stimuli that allowed them toreverse whether a particular shape wasseen as figure or ground, while making

Figure and ground in the brainNava Rubin

Segregation of an image into figure and ground is an important step in visual processing. Twonew papers show that responses in human and monkey brain areas known to be involved inshape perception depend critically on whether a region is perceived as figure or ground.

minimal changes to the image. They foundthat IT cells that produced a vigorousresponse to a particular shape when it wasfigure could have a very weak responsewhen a region of nearly identical shapewas present in the background. Over thelarge population of IT cells from whichthey recorded, there was no correlationbetween a cell’s response to the figure ver-sus ground conditions for each shape. Thismeans, for example, that an IT cell thatresponds strongly to the bottom shape inFig. 1b typically will not respond to Fig. 1a, where a region of the same shape ispresent as part of the background. Theeffect of figure/ground assignment wasobserved even in the earliest portions ofthe cells’ responses, suggesting an intimatecoupling between shape selectivity and fig-ure/ground segregation. These new phys-iological findings are in excellentagreement with the perceptual effectsdescribed by Edgar Rubin and extended inseveral behavioral studies by Baylis andDriver (see references in ref. 2). Becauseinferotemporal cortex has been implicat-ed in a variety of shape perception andobject recognition processes5, it is impor-tant to know that monkey IT cells areinfluenced by figure/ground segregationin a manner similar to human perception.

As Baylis and Driver point out, theirresults indicate that the selectivity of ITresponses is not determined simply by thecontours in a display. This may seem atodds with the strong emphasis vision

news and views

research has put on edges (sharp gradientsin luminance, color or texture) as a majorsource of visual information. The ‘edge doc-trine’ has deep roots, ranging from the pio-neering physiology experiments of Hubeland Wiesel6 to perception studies with linedrawings and decades of research in com-puter vision7. Indeed, edges provide impor-tant information about the shape of thesurfaces they bound. But detecting edges isonly a first step toward interpreting images.An edge is only informative about the shapeof the ‘front’ surface (Fig. 1). The region onthe other side of the edge typically contin-ues behind the front surface, and its ownshape will have nothing to do with theedge’s shape. This is true not only for theground (the distant-most, shapeless region,which nothing can come behind), but alsowhen the back region is a surface with awell-defined shape of its own. For example,the red–blue border in Fig. 1a defines theshape of the blue, but not the red shape.The classic problem of figure/ground seg-regation is therefore a special case of themore general problem of determiningwhich of the two sides of an edge is in front.Edgar Rubin used the term “belongingtogether” (Zusammmengehörigkeit) torefer to the inseparability of a contour andthe region it bounds. In recent years, theproblem has been stated as determining thedirection of ‘border-ownership’: the side infront is said to ‘own’ the edge8. Baylis andDriver’s results indicate that IT cells haveinformation not just about edges, but also

Nava Rubin is in the Center for Neural Science,New York University, New York, New York10003, USA. e-mail: [email protected]

Fig. 1. The assignment of image regions as ‘figure’ or ‘background’ has a dramatic effect on theirperceived shape. The top shape in (b) is easily found in (a), but the region that corresponds to thebottom shape in (b) is often unnoticed, because it is part of the background (top right corner) in(a). Two papers appearing this week report a neural correlate for this perceptual phenomenon.

a b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



about their border-ownership polarity. Thisallows the neurons to avoid responding tomeaningless background regions, andinstead reliably encode the shapes ofobjects.

Kourtzi and Kanwisher4 address a sim-ilar question in humans, using functionalmagnetic resonance imaging (fMRI).Brain imaging studies suggest that the lat-eral occipital complex (LOC), a region inhuman occipitotemporal cortex, is impor-tant for shape perception and object recog-nition9–11. Kourtzi and Kanwisher studiedwhether shape selectivity in this region isalso sensitive to figure and ground. LikeBaylis and Driver, they manipulatedwhether a region appeared as figure orground by making small changes to theimage. Stereo was used to reverse the rela-tive depth of two bordering regions, andthus the polarity of border-ownership, byswapping the images to the two eyes8.

Kourtzi and Kanwisher used an adap-tation technique to investigate the selectiv-ity of LOC neurons to these stimuli.Single-unit and fMRI responses typicallyadapt to repeated presentations of similarstimuli, becoming smaller with repeti-tion11,12. Selective adaptation can be takento indicate neuronal selectivity, and is thusa useful tool for fMRI research. Previouswork has shown that LOC adapts to repeat-ed presentations of the same shape but notto different shapes, suggesting the presenceof neurons tuned to shape. LOC alsoadapts to translated and scaled versions ofan object, but not objects shown from dif-ferent vantage points, indicating that shapeencoding in LOC is invariant to size andlocation, but not to changes in viewpoint11.Kourtzi and Kanwisher asked whether thisadaptation also depended on border-own-ership. In their critical comparison, a par-ticular contour was followed by the samecontour with border-ownership reversed.They found no detectable LOC adaptationwhen the repeated stimulus had reversedborder-ownership—as if a wholly newstimulus were presented. The magnitudeof the effect is striking: after reversing bor-der-ownership, the response amplitude wasjust as high as when presenting an entire-ly new contour, fully reflecting the dramaticchange in the perceived identity and shapeof the figure. Thus, LOC cells seem tobehave much like the IT cells studied byBaylis and Driver: a ‘shape’ is not merelyan arrangement of contours; it does notcount as a shape unless it is also figural.

Given that monkey IT and human LOCare the brain regions most often implicat-ed in shape and object perception, the find-ings of these two papers3,4 suggest that

fields in higher visual cortical areas couldfacilitate the computations considerably.Moreover, perceptual results showing thatobject familiarity can have marked effectson figure/ground resolution15 also impli-cate higher cortical regions in the process.

Taken together, computational con-siderations and experimental results suggest that the resolution of border-ownership, and the resulting assignmentof image regions as figure or ground,involve computations in early as well ashigh-level visual areas. Asking how thesecomputations are done by the brain maytherefore be more appropriate than ask-ing where they are done. Traces of activ-ity before border-ownership resolutionmight be found not in a specific area, butin the early activity of the network. Slow-ing the process down, for example byusing stimuli where border-ownershipresolution is more difficult, may help infuture studies. The striking efficiencywith which the visual system assigns bor-der-ownership to edges makes this com-putational feat hard to unravel, but alsopoints to its central importance.

1. Rubin, E. Visuell wahrgenommene Figuren(Gyldendals, Copenhagen, 1921).

2. Rubin, E. in Visual Perception EssentialReadings (ed. Yantis, S.) 225–230 (PsychologyPress, Philadelphia, 2001).

3. Baylis, G. & Driver, J. Nat. Neurosci. 4,937–942 (2001).

4. Kourtzi, Z. & Kanwisher, N. Representation ofperceived object shape by the human lateraloccipital cortex. Science (in press).

5. Logothetis, N. K. & Sheinberg, D. L. Annu.Rev. Neurosci. 19, 577–621 (1996).

6. Hubel, D. H. & Wiesel, T. N. J. Physiol. (Lond.)195, 215–243 (1968).

7. Marr, D. Vision: A Computational Investigationinto the Human Representation and Processingof Visual Information (W.H. Freeman, SanFrancisco, 1982).

8. Nakayama, K., He, Z. J. & Shimojo, S. inVisual Cognition (eds. Kosslyn, S. M. &Osherson, D. N.) 1–70 (MIT Press,Cambridge, Massachusetts, 1995).

9. Malach, R. et al. Proc. Natl. Acad. Sci. USA 92,8135–8139 (1995).

10. Mendola, J. D., Dale, A. M., Fischl, B., Liu, A. K. & Tootell, R. B. J. Neurosci. 19,8560–8572 (1999).

11. Grill-Spector, K. & Malach, R. Acta Psychol.(Amst.) 107, 293–321 (2001).

12. Miller, E. K., Li, L. & Desimone, R. Science254, 1377–1379 (1991).

13. Zhou, H., Friedman, H. S. & von der Heydt,R. J. Neurosci. 20, 6594–6611 (2000).

14. Pao, H., Geiger, D. & Rubin, N. Measuringconvexity for Figure/Ground separation. Proc.7th IEEE Intl. Conf. Comp. Vision, 948–955(1999).

15. Peterson, M. A. & Gibson, B. S. Psychol. Sci. 5,253–259 (1994).

there may be no place in the brain whereshape is represented independently of bor-der-ownership. This is important for the-ories of shape perception and objectrecognition. At the same time, these papersleave open the question of where and howborder-ownership is computed in thebrain. One possibility is that border-own-ership is resolved at earlier stages of corti-cal processing, in areas that provide inputto human LOC and monkey IT. Anotherrecent study provides some support forthis idea, but at the same time poses newquestions. Zhou et al.13 reported that cellssensitive to border-ownership can befound in early monkey visual cortex—V1,V2 and V4. They oriented a large squareso that one of its edges fell over the ‘mini-mal receptive field’ (MRF) of a cell. Theycentered one edge and then the oppositeedge on the cell’s MRF, thus stimulatingthe cell with a contour in both cases, butwith inverted border-ownership polarity.They found that over 50% of the cells inV2 and V4, and 18% of V1 cells, showedmarked differences in their response to thetwo conditions. These cells thereforeappear to signal only edges with particu-lar border ownership assignments—notthe mere presence of any edge.

Can we therefore conclude that bor-der-ownership is computed early in visu-al cortex, and that the reported behaviorof monkey IT cells3 and human LOC4

reflect input from early visual areas? Sev-eral considerations suggest that the pic-ture may be more complex than that.Resolving border-ownership is a nontriv-ial process not yet well understood com-putationally. A central difficulty is theneed to combine local and global infor-mation. In the experiments of Zhou et al.,for instance, the size of the square wasmuch larger than the size of the cell’s MRF(as mapped by small bar stimuli). Theresponses of the cells were therefore affect-ed by manipulations that took place inimage regions represented by widely dis-tant cells in those early cortical areas.Although it is possible to compute border-ownership with a network of small-recep-tive-field units that mediate informationvia a cascade of lateral connections14, thisis a time-consuming computation forlarge figures. A key finding of Zhou et al.was that the effects of border-ownershipcould be observed very early, within 10–25milliseconds from response onset (simi-lar to what Baylis and Driver found). Theauthors note that V2 mechanisms wouldbe at their limit to resolve border-owner-ship so quickly for large figures, and thatfeedback from cells with larger receptive

news and views©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co

m© 2001 Nature Publishing Group http://neurosci.nature.com

Significant progress has recently been madein identifying the molecular mechanismscausing common neurodegenerative disor-ders such as Alzheimer’s disease (AD) andParkinson’s disease. Rare genetic cases, typ-ically with a very early onset, have providedmechanistic insights that have also provenrelevant to the much more common spo-radic cases, which occur late in life. Thefamilial cases have been helpful for identi-fying not only the genes involved but alsothe pathogenic mechanisms. This has ledto the unifying hypothesis that familial AD(and probably sporadic AD as well) iscaused by increased production of a toxicpeptide called amyloid β-peptide (Aβ)1.However, Nilsberth et al., in this issue2, nowreport the surprising discovery that an earlyonset form of familial AD is associated withdecreased Aβ production. What looks like aparadoxical finding turns out to support aunifying pathogenic mechanism associat-ed with all AD-causing mutations.

Aβ belongs to the family of amyloido-genic, β-sheeted molecules, which rapidlyaggregate and deposit as highly insolublelesions. Aβ is a 40–42 amino acid peptide,which we all continuously produce through-out life1. Proteases termed ‘secretases’ gen-erate Aβ by proteolytic cleavage of a largetransmembrane protein, the β-amyloid pre-cursor protein (βAPP; Fig. 1). First, cleav-age by β-secretase generates the N-terminusof the Aβ domain, leaving behind a C-ter-minal membrane-associated stub. This stubis then cleaved by the second secretase, theγ-secretase, yielding the full length Aβ pep-tide. The γ-secretase apparently cleavesβAPP within the membrane, a surprisingand unanticipated process. Although the

from the site after amino acid 40 to the siteafter amino acid 42 without affecting totalAβ generation1. A selective increase ofAβ42 production was also found for allpathogenic presenilin mutations analyzed1.Thus, familial AD mutations at the β- andγ-secretase sites and in the presenilins allhave a common phenotype, the increasedproduction of the longer Aβ variant. Fur-thermore, the additional two amino acidsat the C-terminus increase Aβ aggregationand amyloid plaque formation1. This ledto the hypothesis that all familial AD muta-tions affect the kinetics of Aβ deposition.

However, additional genetic mutationshad been identified that did not fit neatlyinto this model. In addition to the muta-tions located at the N- and C-terminus ofthe Aβ domain, some mutations were alsofound to occur within the Aβ domain closeto the cleavage site of the third secretase,called α-secretase (Fig. 1). Nilsberth et al.2

now describe one such mutant (E693G),which has been named the Arctic mutationbecause it was found in a family fromnorthern Sweden. Several other mutationshad been previously described near this site:the Dutch (E693Q)5 and Italian (E693K)6

mutations at codon 693, and the Flemish(A692G)7 mutation one codon upstream.

The pathogenetic mechanisms associ-ated with these mutations within the Aβdomain are not entirely understood. Clin-ically, affected members of these kindredsdisplay prominent cerebral amyloidangiopathy with or without other typicalAD-like signs. Although the Flemishmutation leads to alternative processing ofAβ8, the Dutch mutation may have moresubtle effects on processing and aggrega-tion9. Therefore, these mutations do not

γ-secretase has not been strictly identified,presenilin proteins seem to be at least essen-tial cofactors, and may indeed be the γ-sec-retase3. Presenilins were originally identifiedby genetic linkage, and mutations in the pre-senilin genes seem to be involved in mostfamilial AD cases1. In addition to mutationsin presenilins, a variety of much less abun-dant mutations have also been identifiedwithin the βAPP gene itself. Most of thesemutations cluster at the sites of secretase-mediated cleavage (Fig. 1). One mutation,the Swedish mutation, has been identifiedexactly at the site of β-secretase cleavage, andseveral other mutations have been found ator near the γ-secretase cleavage site.

How do these mutations accelerate ADpathology? For most identified mutations,pathogenicity is related either to increasedproduction of Aβ in general or increasedproduction of a longer variant of the Aβpeptide, Aβ42. The Swedish mutationincreases Aβ production by providing abetter substrate for β-secretase, thusincreasing cleavage4. In contrast, mutationsclose to the γ-secretase site bias cleavage

news and views

Protofibrils, the unifying toxicmolecule of neurodegenerativedisorders?Christian Haass and Harald Steiner

A newly identified Alzheimer’s mutation leads to thesuggestion that protofibril intermediates in amyloid plaqueformation may be a crucial factor in pathogenicity.

The authors are in the Laboratory forAlzheimer’s Disease Research, Department ofBiochemistry, Adolf-Butenandt-Institute;Schillerst Ludwig-Maximilians-University-Munich44, 80336 Munich, Germany. e-mail: [email protected];[email protected]

Membrane

N

NL (Swedish) N (Iowa)G (Flemish)

Q (Dutch)K (Italian)G (Arctic)

C

Aβ

KMDAEFRHDSGYEVHHQKLVFFAEDVGSNKGAIIGLMVGGVVIATVIVITLVML

β α γ

I (Austrian)P (Australian)

M (French)

V (Florida)

I (London)L F G

Fig. 1. The structure of βAPP and the location of familial AD-associated mutations. The Aβ domainand its flanking amino acids are enlarged. Mutations associated with Alzheimer’s disease (http://mol-gen-www.uia.ac.be/ADMutations/) are indicated, along with the cleavage sites of the three secretases.


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



really fit into the model of familial AD, andone even tends to forget about them bystating that “all familial AD associatedmutations increase Aβ42 generation”.

The new Arctic mutation is a strikingexample of this paradox. Surprisingly, plas-ma samples from carriers of the Artic muta-tion displayed reduced levels of Aβ40 andAβ42. The same was found in conditionedmedia derived from cells transfected withthe corresponding βAPP cDNA2. The find-ing was further substantiated by demon-strating that cells transfected with the Italianand Dutch mutation also showed reducedsecretion of Aβ40 and Aβ42 (ref. 2).

Therefore, all three mutations atcodon 693 result in a pathogenic pheno-type opposite to what would have beenpredicted. At first glance, these findingsare incompatible with models of ADpathogenesis in which increased Aβ42production or ratio is the commondenominator. However, AD pathology isultimately linked with the assembly andextracellular accumulation of fibrillar Aβ.

One of the intermediates in the path-way of Aβ fibril formation, the structuralcomponent of all amyloid plaques, is theprotofibril. Protofibrils, originally identi-fied by Teplow10 and Lansbury11, areshort, flexible assemblies ∼ 5 nm in diame-ter and rarely exceeding 200 nm in length.They are not only important intermedi-ates in amyloid fiber formation, but alsocause selective neuronal cell death12,13.

Nilsberth et al.2 investigated how theArctic amino acid substitution affectedprotofibril formation in vitro. Whereas noapparent difference was found between the

Protofibrils may also represent an idealtarget for anti-amyloidogenic drugs. How-ever, one still needs to prove that protofib-ril formation is critically required fordisease-specific pathology in the humanbrain—a very difficult task which proba-bly needs several additional years of inten-sive research. In the end, drugs againstprotofibrils may provide an importanttherapeutic alternative to the secretaseinhibitors and vaccination approaches thatare already under investigation.

1. Steiner, H. & Haass, C. Nat. Rev. Mol. CellBiol. 1, 217–224 (2000).

2. Nilsberth, C. et al. Nat. Neurosci. 4, 887–893(2001).

3. Wolfe, M. S. & Haass, C. J. Biol. Chem. 276,5413–5416 (2001).

4. Vassar, R. & Citron, M. Neuron 27, 419–422(2000).

5. Levy, E. et al. Science 248, 1124–1126 (1990).

6. Tagliavini, F. Alz. Report 2, 28 (1999).

7. Hendriks, L. et al. Nat. Genet. 1, 218–221(1992).

8. Haass, C., Hung, A. Y., Selkoe, D. J. & Teplow,D. B. J. Biol. Chem. 269, 17741–17748 (1994).

9. Watson, D. J., Selkoe, D. J. & Teplow, D. B.Biochem. J. 340, 703–709 (1999).

10. Walsh, D. M., Lomakin, A., Benedek, G. B.,Condron, M. M. & Teplow, D. B. J. Biol.Chem. 272, 22364–22372 (1997).

11. Harper, J. D., Wong, S. S., Lieber, C. M. &Lansbury, P. T. Chem. Biol. 4, 119–125 (1997).

12. Walsh, D. M. et al. J. Biol. Chem. 274,25945–25952 (1999).

13. Hartley, D. M. et al. J. Neurosci. 19, 8876–8884(1999).

14. Conway, K. A. et al. Proc. Natl. Acad. Sci. USA97, 571–576 (2000).

overall fibrilization rates of wild-type andArctic Aβ, the mutant peptide producedprotofibrils at a much higher rate and inlarger quantities. The Arctic mutation thusincreases the quantity of an Aβ assemblythat not only has potent intrinsic neuro-toxic activity, but also converts into fibrils,neurotoxic moieties in their own right.

Taking this into consideration, the Articmutation can indeed fit into the unifyinghypothesis, as one can state that all famil-ial AD mutations facilitate Aβ assembly (beit into protofibrils, fibrils or other toxicmoieties). One could speculate that selec-tive increases in the levels of protofibrilsmay be a common cause for the early onsetof AD pathology in all the familial cases.This would be in line with the recent find-ing that very similar protofibrils areformed by α-synuclein, the protein foundwithin the major lesions of Parkinson’s dis-ease14. Moreover, rare mutations causingearly-onset Parkinson’s diseases also accel-erate protofibril formation14, although ithas not yet been shown that they can causeselective neuronal cell death.

Unifying assembly mechanisms may bea common phenomenon of neurodegener-ative disorders associated with the deposi-tion of amyloidogenic peptides, a hypothesisstrongly supported by the findings of Nils-berth et al.2. A popular theory within thefield has been that amyloid plaques are thetoxic unit directly associated with neurode-generation. However, it became clear thatthe density of amyloid plaques does notnecessarily correlate with the dementia andneuronal cell loss. The level of protofibrilsmay finally fulfill this critical correlation.

news and views

Neuronal migration and theevolution of the human brainYi Rao and Jane Y. Wu

A new study demonstrating a pathway for neuronal migrationin humans, but not in monkeys, suggests that migration has akey role in the evolution of the brain, as well as its development.

Neurons are natural migrants; most, if notall, of the neurons in the mammalian ner-vous system migrate from their places ofbirth to their locations of function. In thebrain, neurons usually originate in the ven-tricular zone, where their precursor cellsproliferate. They can then migrate radial-

ly to other layers in the brain, or tangen-tially (in a direction parallel to the surfaceof the brain) to other regions of thebrain1–3. Radial migration is dependenton radially aligned glial fibers, whereastangential migration is independent ofglial cells and perhaps relies on contactswith other neurons. Although radialmigration was the focus of research in the1970s and 1980s, tangential migration wassuggested in the 1960s and, through workin the 1990s, has now been established as amajor mode of neuronal migration. Neu-ronal migration is a crucial step in neuraldevelopment, as defects in neuronalmigration cause multiple human diseases.

Similar to other fields of experimen-tal biology, our knowledge of neuronalmigration is based primarily on experi-ments with brains of laboratory animals.

The authors are in the Departments of Anatomyand Neurobiology, Pediatrics, Molecular Biologyand Pharmacology, Washington UniversitySchool of Medicine, Box 8108, 660 S. EuclidAvenue, St. Louis, Missouri 63110, USA. e-mail: [email protected];[email protected]

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 1. Guidance activities involved in theGE–DT pathway. Schematic coronal sectionsof mouse and human brains at the level of thethalamus. +, attractive activity for cells fromthe GE; –, repulsive activity. GE, ganglioniceminence; DT, dorsal thalamus; ST, subthala-mus; CP, choroid plexus.


Because developmental mechanisms arewidely conserved across species rangingfrom flies and worms to humans, therehas not been much effort devoted toexperimental studies of neuronal migra-tion in the human brain. However, a seriesof studies carried out by Pasko Rakic andcolleagues, culminating in the paper pub-lished in this issue, show that certainimportant questions about the humanbrain can only be addressed by studyinglive tissue from the human brain (in thiscase, human brain slices from abortedfetuses)4. This study extends earlier sug-gestions, based on histological analysis offixed human brains, that neuronal pre-cursor cells migrate from a structure inthe telencephalon, the ganglionic emi-nence (GE), to the thalamus in the dien-cephalon5. By comparing neuronalmigration in humans with those in miceand monkeys, Rakic and colleaguesdemonstrate that the human brain maypossess migratory pathways that do notexist in other mammals, or perhaps evenin other primates.

Human thalamic nuclei connected tothe frontal cortex are larger than those inother primates6. For example, the pulv-inar nucleus in the dorsal thalamus (DT)is larger in primates than in other mam-mals and, among primates, is larger inhumans than in chimpanzees andmacaque monkeys. In previous work,Rakic and Sidman asked whether the larg-er pulvinar nucleus results from increasedcell proliferation in the ventricular zoneof the diencephalon. They found thatthere are two phases of pulvinar develop-ment in humans5. Whereas the earlyphase correlates with cell proliferation inthe diencephalon, the late phase does not;cell proliferation in the diencephalon wasnot detected from the eighteenth to thethirty-fourth week of gestation, which is

major inhibitory neurotransmitter in thebrain. Taken together with earlier studiesof GABAergic neuronal migration from theGE to the neocortex9,10, the new results inhuman tissue indicate that the GE con-tributes to GABAergic neurons in multipleregions of the brain.

Using similar techniques, Letinic andRakic did not detect cell migration fromthe GE to the DT in the monkey or themouse brains4. Because earlier work onhuman brains5 was done at times andunder conditions different from the workon the mouse or the monkey brains7, thepresent study provides the strongest evi-dence that the GE to DT migratory path-way is apparent only in the human brain.

Previous studies in rodents showedthat regions surrounding the migratingneurons in the GE can influence migra-tion10. To address the question of whatcontributes to the difference in neuronalmigration between human and mousebrains, Letinic and Rakic isolated explantsof human and mouse GE and co-culturedthem with either the DT or the subthala-mus (ST), which is part of the path fromthe GE to the DT4. Human GE cells wereattracted by human DT, whereas mouseGE cells were neither attracted norrepelled by the mouse DT (Fig. 1). ST wasrepulsive in the mouse explants, but nei-ther repulsive nor attractive in the humanexplants. The repulsive and attractiveactivities are contact-independent, indi-cating that they are diffusible guidancecues. These results suggest that guidancecues in the DT and the ST could explainthe presence of the GE to DT pathway inthe human, and its absence in the mouse.

The Letinic and Rakic paper thusprovides not only direct evidence for anew migratory pathway in the humanbrain, but also suggests possible cellularmechanisms that may underlie the dif-ferential migration of GE cells inhumans and other species. It also raises anumber of questions. For example, is thespecies difference in GE to DT migrationdue solely to changes in the positioningof the guidance cues, or to changes incellular responsiveness in the GE cells?It will be interesting to see results fromcross-species co-cultures of the GE withthe DT and the ST, which may provide

the major period of human pulvinargrowth. This suggests that cells con-tributing to the late phase of pulvinargrowth are not likely to be derived fromthe ventricular zone of the diencephalon5.During the late phase, the ganglionic emi-nence (GE), contains proliferative cells,and streams of cells extend from the GEto the thalamus. Cells in these streams arebipolar in the tangential direction, whichsuggests that they are migrating. Rakicand Sidman thus proposed that cells fromthe GE migrate through these streams tothe thalamus in the human brain5. Thepositioning of the streams, their transientnature, and the direction of the leadingand trailing processes of cells in thestreams are consistent with the possibilitythat these streams were migratory path-ways. However, there was no direct evi-dence that cells actually migrate from theGE to the thalamus.

In similar studies, Ogren and Rakicfound in macaque monkeys that only theearly phase of pulvinar developmentoccurs, and that the pulvinar nucleus doesnot receive contributions of neurons fromthe telencephalon7. These findings led tothe suggestion that neuronal migrationfrom the GE to the pulvinar nucleusmight be unique to humans7.

In the work reported in this issue,Letinic and Rakic report the first direct evi-dence that neurons indeed migrate fromthe GE to the DT in human brain slices4.They placed the lipophilic dye DiI in theGE and found labeled cells in the DT,including the pulvinar and mediodorsalnuclei4. These neurons seem to be migrat-ing in a fashion similar to other types oftangential migration described in the olfac-tory system8, as they seem to be indepen-dent of glial fibers, but instead rely oncontacts with other neurons. Furthermore,the migrating neurons contain GABA, the

news and views

CP

ST

DT(+)

GEDT

ST(-)

CP

GE

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



a more definite answer to the questionof whether changes in GE responsivenessare involved in the species differences inGE to DT migration. It will also be inter-esting to characterize the molecularidentities of the repulsive and attractiveguidance cues in the ST and the DT. Twosecreted proteins, Slit and netrin, canrepel rodent GE neurons11,12. Their pat-terns of expression remain to be charac-terized in monkey and human brains, aswell as in relevant regions of the mousethalamus. Because a guidance cue canact as both a repellent and an attrac-tant13, it is also possible that the samecues may function differently in the GEto DT pathway of different species.

One of the most interesting sugges-tions from the work of Rakic and col-leagues is that new neuronal migrationpathways may be involved in brain evo-lution. During the evolution of the mam-malian brain, regions connected to eachother anatomically and functionally arethought to co-evolve14, but mechanismsfor co-evolution are not known. Resultsfrom Rakic and colleagues suggest a

1. Rakic, P. Experientia 46, 882–891 (1990).

2. Hatten, M. E. & Heintz, N. in FundamentalNeuroscience (eds. Zigmond, M. J., Squire, L.R. & Landis, S. C.) 451-480 (Academic, NewYork, 1998).

3. O’Rourke, N. A., Chenn, A. & McConnell, S. K. Development 124, 997–1005 (1997).

4. Letinic, K. & Rakic, P. Nat. Neurosci. 4,931–936 (2001).

5. Rakic, P. & Sidman, R. L. Z. Anat. Entwickl.Gesch. 129, 53–82 (1969).

6. Armstrong, E. A. Am. J. Phys. Anthropol. 55,369–383 (1980).

7. Ogren, M. P. & Rakic, P. Anat. Embryol. 162,1–20 (1981).

8. Lois, C. & Alvarez-Buylla, A. Science 264,1145–1148 (1994).

9. Anderson, S. A., Eisenstat, D. D., Shi, L. &Rubenstein, J. L. R. Science 278, 474–476 (1997).

10. Zhu, Y., Li, H. S., Zhou, L., Wu, J. Y. & Rao, Y.Neuron 23, 473–485 (1999).

11. Wu, W. et al. Nature 400, 331–336 (1999).

12. Hamasaki, T., Goto, S., Nishikawa, S. & Ushio,Y. J. Neurosci. 21, 4272–4280 (2001).

13. Song, H.-J., Ming, G.-L. & Poo, M.-M. Nature388, 275–279 (1997).

14. Barton, A. R. & Harvey, P. H. Nature 405,1055–1058 (2000).

15. Letinic, K. & Kostovic, I. J. Comp. Neurol. 384,373–395 (1997).

novel and specific mechanism for co-evolution of brain structures. Thus, theGE to DT pathway may enable the co-evolution of the frontal cortex and thethalamic nuclei that are connected to it.There are perhaps multiple migrationpathways from the GE to thalamicregions15, and it will be interesting toknow whether all of those pathways cor-relate with the evolution of the neocortexand the thalamus.

Evidence obtained so far indicatesthat the evolution of a new migratorypathway could, in principle, contributeto the presence of more neurons in thehuman thalamus. The significance ofthese pathways in vivo could be tested ifthese pathways could be experimentallymanipulated in slices of mouse, monkeyand human brains after the identifica-tion of distinct guidance cues. Perhapsstudies of human patients with geneticdefects disrupting a specific migratorypathway(s) may help answer the ques-tion of whether a migratory pathwayleads to the evolution of a larger thala-mus in humans.

news and views

Virtual neurologyRobert Rafal

Using transcranial magnetic stimulation to induce a 'virtual lesion' in the parietal lobe, a new study reveals the mechanisms of hemispatial neglect, a neurological disorder of attention.

Patients with unilateral brain lesions, espe-cially those involving the temporoparietalcortex, are often inattentive to objects andevents contralateral to the lesion. This syn-drome, known as hemispatial neglect,seems to involve a deficit in the orientingof attention rather than perceptual pro-cessing, as failure to detect contralesionalstimuli is more likely when an ipsilesion-al object is competing for attention. Thiscan be demonstrated by testing for a signcalled visual extinction: the patient maybe able to detect and report an object inthe contralesional field when it is pre-sented alone, but fails to do so when there

is a competing item in the ipsilesional fieldthat must also be ‘reported’; that is, theobject is extinguished from awareness bythe competing stimulus.

One classic theory of neglect andextinction posits mutual inhibitionbetween the hemispheres such that whensystems for orienting attention in onehemisphere are damaged, homologousregions in the opposite hemisphere aredisinhibited1. The presentation of a com-peting stimulus, which activates the dis-inhibited intact hemisphere, then furtherinhibits the lesioned hemisphere, caus-ing extinction. A key feature of the hemi-spheric rivalry account is that it predictsbetter-than-normal performance in thefield ipsilateral to the brain lesion.

In this issue, Hilgetag et al.2 adaptthe technique of repetitive transcranialmagnetic stimulation (TMS) to tem-

porarily inactivate parietal cortex innormal volunteers and produce a modelof hemispatial neglect (Fig. 1), allowingthem to test the hemispheric rivalryaccount of visual attention. Subjectswere stimulated for 10 minutes with 1 Hz TMS at a point that was subse-quently demonstrated, using structuralMRI, to overlie the intraparietal sulcus.After TMS was terminated, the authorsmeasured subjects’ ability to detect visu-al stimuli presented in the field con-tralateral to TMS, in the ipsilateral fieldor simultaneously in both fields. Com-pared to before TMS, detection of con-tralateral stimuli presented alone wasreduced, contralateral detection was fur-ther reduced by a competing ipsilateralstimulus (visual extinction), and detec-tion of ipsilateral stimuli presentedalone was facilitated—consistent withdisinhibition of the unstimulated hemi-sphere as predicted by the hemisphericrivalry hypothesis.

Several previous observations arealso consistent with the hemisphericrivalry account. In neurological neglectpatients, not only is detection of con-tralesional stimuli impaired, but detec-tion of ipsilesional stimuli isenhanced3,4. The rivalry account alsopredicts that a second lesion in the

Robert Rafal is in the School of Psychology,University of Wales, Bangor, The BrigantiaBuilding, Penrallt Road, Bangor, GwyneddLL57 2AS, Wales, UK.e-mail: [email protected]

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



opposite hemisphere might result inrecovery from neglect; and indeed thishas been reported to occur both with anatural lesion5 and when patients are‘treated’ by TMS inactivation of thehemisphere opposite to the originallesion6. Moreover, a previous studyusing single-pulse TMS showed thatinactivation of the right parietal loberesults in disinhibition of the contralat-eral hemisphere as measured by adecrease in the threshold to detect tac-tile stimuli in the right hand7. Hilgetaget al.2 used repetitive TMS not only todemonstrate contralateral disinhibition,a signature of hemispheric rivalry, butto do so in a protocol that models thekey features of hemispatial neglect.Their findings highlight a cardinal les-son of neuropsychology: that lesion-induced deficits may not be understoodsimply in terms of the absence of a puta-tive function that is normally mediatedby the lesioned tissue. Rather, the patho-logical behavior reflects the normaldynamic interactions of the region withother interconnected structures8,9.

The new application of TMS extendsits emerging role as an important toolin cognitive neuroscience. There are,broadly, two converging approaches forinvestigating brain–behavior relation-ships in humans. Activation techniques,which include event-related electricalpotentials, functional magnetic reso-

provide excellent spatial resolution andtemporal resolution.

Unlike previous TMS techniques thatused single pulses or brief trains ofrepetitive TMS at higher frequencies(typically 10–25 Hz) to disrupt corticalfunction for, at most, a few hundredmilliseconds, the continuous 1 Hz stim-ulation used by Hilgetag et al.2 producesa reduction in neuronal excitability thatpersists for several minutes after termi-nation of stimulation. This decrease inexcitability has been measured, after 1 Hz TMS over motor cortex, as areduction in the amplitude of theMEP12,13. Recently, the technique hasbeen shown to decrease the excitabilityof visual cortex. In that study,phosphenes (perceived flashes of light)were evoked by TMS, and the thresholdfor eliciting them (that is, the stimula-tor output required to evoke them)increased after 1 Hz TMS14. Moreover,this decrease in visual cortex excitabili-ty is associated with a decrement invisual imagery15. The study reported byHilgetag et al.2 is the first, however, touse the technique to induce a ‘virtuallesion’ persisting after the stimulationthat produces a faithful model of a clas-sical neurological syndrome. Confirma-tion of a hemispheric rivalry account forvisual orienting makes an exciting debutfor the method.

The conjoined use of TMS coregis-tered with structural MRI in this studyhighlights the importance of converg-ing techniques in cognitive neuro-science. It permitted the investigators toidentify the precise cortical site stimu-lated. The study did not, however, spec-ify the extent of the neural tissue thatwas inactivated. Although the effect ofindividual TMS pulses may, as noted, befocal, the spatial extent of the ‘virtuallesion’ produced by the cumulativeeffect of repetitive 1 Hz stimulation isnot known. A special opportunityafforded by the 1 Hz method, becauseits effects persists after the terminationof stimulation, is that it will be possibleto study the extent of cortical inactiva-tion after stimulation by measuringblood flow with fMRI in individual sub-jects. Furthermore, this mapping of the‘virtual lesion’ can be coregistered withfMRI activations in the same subjectsperforming the same task. The allianceof this new TMS technique with struc-tural and functional neuroimagingextends the promise of virtual neurolo-gy as an exciting new tool for exploringhuman psychobiology.

nance imaging (fMRI),positron emission tomogra-phy (PET) and optical imag-ing, all record the activationof neural tissue in relationto mental events. These arecorrelational measures thatcan determine what brainareas are active during men-tal events, but not whether agiven region is necessary forthem to occur. For this, con-verging evidence must besought by studying thebehavioral consequences ofbrain inactivation. Untilrecently, the only opportu-nity to do so in humans was in neurological patients.Although structural MRIhas now made possible pre-cise lesion localization and,thus, very high spatial reso-lution in neuropsychologi-cal studies10, it is obviouslynot generally possible to testpatients before and after thelesion. Furthermore, natur-

al lesions not only inactivate corticalneurons, but also can produce deficitsby damage to the white matter tractsthat connect cortical areas to oneanother. In addition, the reorganizationthat occurs during recovery, thoughoffering precious insights into the braindynamics relevant to rehabilitation,nevertheless requires caution in draw-ing inferences about the normal func-tion of the damaged tissue.

TMS obviates some of the con-founds inherent in neuropsychologicalinvestigation. Developed just a fewdecades ago to study motor systemphysiology in humans, the techniquehas been adapted by cognitive neuro-scientists in the past few years to pro-duce non-invasive, focal and transientinactivation—‘virtual lesions’11—ofsmall regions of human cortex. Using afigure-of-eight coil that generates atightly focused magnetic field at theintersection of the coils, a TMS pulseinduces an electrical current that tran-siently stimulates the underlying cortexand disrupts its normal function. Whenapplied over the motor cortex at inten-sities just above the threshold for elic-iting a motor evoked potential (MEP)recorded with electromyography, indi-vidual movements of different fingerscan be obtained at sites millimetersfrom one another. Single-pulse TMSthus can be time-locked to stimuli to

news and views

Fig. 1. Schematic depiction of hemispatial neglect as induced byrepetitive TMS. In neurological neglect, patients suffer fromimpaired attentional resources on one side of space. Hilgetag etal.2 produce a similar condition by repeatedly stimulating theparietal lobe in normal subjects.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Only ten years ago, the first papersappeared on functional magnetic reso-nance imaging (fMRI) using blood oxy-genation level dependent (BOLD)contrast. The explosive world-widegrowth of fMRI as a tool for non-inva-sively visualizing dynamic, localizedprocesses in the human brain that fol-lowed reflects the promise that the tech-nique will open up new avenues ofresearch in cognitive neuroscience. Thephysical basis of BOLD contrast is oxy-genation-dependent magnetic suscepti-bility of hemoglobin. Deoxyhemoglobinis paramagnetic, causing slightly attenu-ated signal intensity in MRI image voxelscontaining deoxygenated blood. Duringbrain activation, localized increases inblood flow increase blood oxygenationand consequently reduce deoxyhemoglo-bin, causing the MRI signal to increase.It is therefore assumed that these local-ized increases in BOLD contrast reflectincreases in neuronal activity.

Two primary questions remain aboutthe interpretation of fMRI signals: thequantitative relationship between neur-al activity and BOLD contrast, and thebiological mechanism underlying thisrelationship. The MRI signal in activat-ed regions begins to increase approxi-

ronal activity simultaneously recordedwithin monkey primary visual cortex.Stimulus-driven unit activity and thelocal field potential (LFP) were record-ed through a microelectrode while theanesthetized monkey was visually stim-ulated with a rotating checkerboard pat-tern. Unit activity represents the firing,or action potentials, of single or multi-ple neurons recorded near the electrodetip (within about 100 µm for single unitsand 200 µm for multi-units). In contrastto this fast ‘spiking’ activity, which rep-resents the transmitted output of one ora few neurons, the LFP is a relatively slowoscillatory electrical wave, resembling anEEG recorded from the scalp. However,the spatial resolution of the EEG is verycoarse, whereas the LFP reflects aggre-gate activity from a population of neu-rons located within a few millimeters ofthe electrode tip. This activity is thoughtto be a weighted sum of the membranepotentials generated from the popula-tion, with neurons closer to the electrodetip making the greatest contribution.Although changes in membrane poten-tials (both excitatory and inhibitory)mainly reflect synaptic activity localizedto dendrites and soma, action potentialswithin the neuronal population may alsocontribute to the LFP.

Logothetis et al. distinguished unitactivity from the LFP by filtering thebroad band of activity into high-fre-quency and low-frequency components:300–1500 Hz for units and 40–130 Hz(the so-called gamma range) for theLFP. They then correlated the timecourses of activity and BOLD responses.(Because single-unit activity and multi-unit activity were highly correlated, themain comparisons were correlationsbetween BOLD and multi-unit activityversus BOLD and LFP.) Both LFP andmulti-unit activity correlated with theBOLD response, but the better predic-tor was the LFP. This was probablybecause about one quarter of the multi-unit responses were transient, return-

mately 2 seconds after neural activitybegins, and plateaus in the ‘on’ stateafter about 7 to 10 seconds, remainingelevated while the activity continues.When activity ends, the signal returnsto baseline after about 8 to 11 seconds.Transient signal changes are alsodescribed, including a ‘pre-undershoot’(reduced BOLD signal within the firsttwo seconds of activity) and a morecommonly observed ‘post-undershoot’(reduced signal for 10 to 40 secondsafter activity ends). Despite extensivemodeling, the biological basis and het-erogeneity of BOLD signal dynamicsand magnitude remain unclear, pri-marily because they reflect the interplayof many uncharacterized variables,including neural activity, metabolism,blood volume, blood flow and subse-quent oxygenation changes. The mostsignificant steps toward understandingthe relationship between neural activityand BOLD contrast have come fromdirect, simultaneous and spatially reg-istered measurement of these variables.Several papers have shown an essential-ly linear relationship between non-simultaneous measures of neuronalactivity and hemodynamic changes inmonkeys1, and simultaneous measuresin rats2–4, but the recent paper inNature by Logothetis et al.5 should beconsidered a landmark because it is themost comprehensive, detailed anddefinitive set of comparisons yet made.

Logothetis et al. examined how wellthe BOLD signal correlated with neu-

news and views

From neuron to BOLD: newconnectionsPeter A. Bandettini and Leslie G. Ungerleider

A recent paper in Nature directly compares fMRI with simulta-neously recorded neural activity in the monkey, yielding newinsights into the interpretation of BOLD contrast.

The authors are in the Laboratory of Brain andCognition, National Institute of Mental Health,National Institutes of Health, Bethesda,Maryland 20892, USA. e-mail: [email protected];[email protected]

1. Kinsbourne, M. in Advances in Neurology(eds. Weinstein, E. A. & Friedland, R. P.)41–49 (Raven, New York, l977).

2. Hilgetag, C. C., Theoret, H. & Pascual-Leone,A. Nat. Neurosci. 4, 953–957 (2001).

3. Smania, N. et al. Brain 121, 1759–1570(1998).

4. Ladavas, E. Brain 113, 1527–1538 (1990).

5. Vuilleumier, P., Hester, D., Assal, G. & Regli, F.Neurology 19, 184–189 (1996).

6. Oliveri, M. et al. Brain 122, 1731–1739 (1999).

7. Seyal, M., Ro, T. & Rafal, R. Ann. Neurol. 38,264–267 (1995).

8. Sprague, J. M. Science 153, 1544–1547 (1966).

9. Henik, A., Rafal, R. & Rhodes, D. J. Cogn.Neurosci. 6, 400–411 (1994).

10. Sapir, A., Soroker, N., Berger, A. & Henik, A.Nat. Neurosci. 2, 1053–1054 (1999).

11. Pascual-Leone, A., Walsh, V. & Rothwell, J.

Curr. Opin. Neurobiol. 10, 232–237 (2000).

12. Chen, R. et al. Neurology 48, 1398–1403(1997).

13. Muellbacher, W., Ziemann, U., Boroojerdi, B.& Hallett, M. Clin. Neurophysiol. 111,1002–1007 (2000).

14. Boroojerdi, B., Prager, A., Muellbacher, W. &Cohen, L. G. Neurology 54, 1529–1531 (2000).

15. Kosslyn, S. M. et al. Science 284, 167–170(1999).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


ing to baseline levels within a few sec-onds of visual stimulation. By contrast,LFP responses, like BOLD, remainedelevated for the duration of visual stim-ulation, and hence were more consis-tently coupled with the BOLD response(Fig. 1). Further, a more formal analy-sis of the estimated BOLD response, aspredicted by both LFP and multi-unitactivity, indicated that LFP gave a sig-nificantly better estimate. This is prob-ably because synaptic activity consumesmore energy, an important determinantof the BOLD response magnitude, thandoes the transmission of action poten-tials. So, the good news is that BOLD isrelated to neuronal activity. The badnews is that apparently onecannot assume a close rela-tionship between BOLD andthe kind of signals (single-and multi-unit activity) typ-ically recorded throughmicroelectrodes in physio-logical studies.

The LFP time courses inthe Logothetis et al. paperalso clarify a previouslyunresolved observationregarding the dynamic ‘non-linearity’ of BOLD contrast.Boynton et al.6 first demon-strated that with brief (< 3second) stimuli, the magni-tude of the BOLD signal waslarger than expected from alinear system, assuming thatthe neuronal input was con-stant for all stimulus dura-tions. Since then, severalpapers have demonstratedthis same effect (mostrecently, ref. 7; Fig. 2). Acentral question in the inter-pretation of dynamic BOLDcontrast has been what caus-


tionship between neural activity andmeasured BOLD signal. The hemody-namically derived high neural activityduring the first three seconds is strik-ingly similar to the LFP recordings ofLogothetis et al.5 The degree of resem-blance between these direct physiologicalrecordings and Fig. 2 provides strongsupport for the idea that these hemody-namic ‘nonlinearities’ are primarily dueto transiently high neuronal activity dur-ing the first three seconds of stimulation.The BOLD signal for brief responsesthus seems to be more faithful to theunderlying neural activity than mighthave been expected.

Other discrepancies remain, however.Several fMRI studies have reportedBOLD increases that are larger thanwould have been expected from single-unit recordings in awake, behaving mon-keys. Two examples, in which theimaging and physiological tasks are verysimilar, come immediately to mind. Instudies of spatial attention, covertlydirecting attention to a particular loca-tion and waiting for a target to appearthere increases the BOLD response inhuman extrastriate visual areas V2 andV4 about 35–50% as much as target pre-sentation9, an order of magnitude largerthan the increase in single-unit activitymeasured in homologous visual areasduring a similar expectation period10.

es these ‘nonlinearities’; are there non-linearities in hemodynamics or neuronalinput? Mathematical models of hemo-dynamic changes show that such non-linear BOLD signal behavior is possible8

without invoking nonlinearities causedby transiently high neuronal activity inthe first three seconds. On the otherhand, Boynton et al. suggested that thesource of the observed BOLD nonlin-earities could be transiently high neu-ronal activity, but evidence for thisexplanation was lacking. The neuronalactivity necessary to create the hemody-namic changes observed with increasingstimulus duration has been estimated(Fig. 2, bottom), assuming a linear rela-

news and views

Fig. 1. Simultaneous neural and hemodynamic recordings from a cortical site showing transientneural response to a pulse stimulus of 24 seconds. Both single- and multi-unit responses adapt acouple of seconds after stimulus onset, with LFP remaining the only signal correlated with theBOLD response. SDF, spike-density function (see text); ePts, electrode ROI—positive time series.Reprinted by permission from Nature 412, 150–157, copyright 2001 Macmillan Magazines Ltd.

Fig. 2. The BOLD response has larger amplitudes than expected from a linear response to a constant input acrossstimulus durations. Top, spatial variation of the nonlinearity. The amplitude of the fMRI response relative to a linearresponse with constant neuronal input is shown for one voxel. Bottom, estimated neuronal activity (red), derivedcompletely from the measured BOLD responses, with the assumption of a direct linear relationship between neu-ronal activity and BOLD contrast for different stimulus durations, shows a striking resemblance to the time courseof LFP in Logothetis et al.5 Adapted from ref. 7.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Similarly, in binocular rivalry experi-ments, in which disparate visual stimuliare presented simultaneously to the twoeyes and the subject experiences an alter-nating percept of one and then the other,BOLD signal modulation in humanvisual cortex (V1, V2 and V4) evokedduring the rivalry condition is about halfas strong as the response to physicallyalternating the two visual stimuli11, ortwice as large as the modulation mea-sured in single units in homologousmonkey visual areas12. The most obvi-ous explanation for these mismatchesbetween BOLD and single-unit activitywould be that BOLD is driven by theLFP rather than unit activity and, insuch cases, the LFP reflects mainly sub-threshold processes that fail to drive theunits.

There is, however, an alternativeexplanation, which relates to the exper-iments by Logothetis et al. on stimuluscontrast. The LFP, multi-unit activityand BOLD were all found to decrease asa function of reduced stimulus contrast,although at differing rates. Importantly,at low levels of contrast (12.5%), theBOLD response is about 50% of its max-imum amplitude, whereas the LFP andmulti-unit activity both drop to about10–15% of their maximum amplitude(Fig. 3). These findings suggest that atlow levels of neuronal activity, the LFPis not necessarily a better predictor ofBOLD than multi-unit activity, andBOLD signal changes will be overesti-mated relative to both the LFP andmulti-unit activity. This is exactly the sit-uation in the spatial attention studiesdescribed above, where there was a verylarge BOLD response but the neuronalactivity measured during the expectationperiod consisted of small, though sig-nificant, increases in baseline, sponta-neous unit firing. Although it seems thatat higher levels of neuronal activity, mis-

intrinsically on local neurons ratherthan on distant neurons in other corti-cal regions13. Thus, both the LFP andmulti-unit activity likely reflect mainlythe activity in local cortical circuits.Finally, it may be valuable to considerthat the better coupling of LFP, com-pared to multi-unit activity, with theBOLD signal reflects a difference notbetween LFP and multi-unit activityper se but between gamma frequencyoscillations (the frequency band exam-ined in the LFP) and other frequencycomponents. If, as suggested by others,the gamma frequency band holds spe-cial importance in both perceptualbinding and attention14,15, then onemight predict a good correlationbetween the gamma component ofmulti-unit activity and BOLD. Indeed,it has been proposed that rhythmic fir-ing of neurons in this frequency band ismetabolically expensive. These com-ments notwithstanding, Logothetis etal. have made a significant step in elu-cidating the neural origins of the BOLDsignal. Studies to come will build on theseminal findings reported in this paper.

1. Rees, G., Friston, K. & Koch, C. Nat.Neurosci. 3, 716–723 (2000).

2. Matheiesen, C., Caesar, K., Akgoren, N. &Lauritzen, M. J. Physiol. (Lond.) 512,555–566 (1998).

3. Ogawa, S. et al. Proc. Natl. Acad. Sci. USA 97,11026–11031 (2000).

4. Brinker, G. et al. Magn. Reson. Med. 41,469–473 (1999).

5. Logothetis, N., Pauls, J., Augath, M.,Trinath, T. & Oeltermann, A. Nature 412,150–157 (2001).

6. Boynton, G. M., Engel, S. A., Glover, G. H. &Heeger, D. J. J. Neurosci. 16, 4207–4221(1996).

7. Birn, R. M., Saad, Z. S. & Bandettini, P.Neuroimage (in press).

8. Buxton, R. B., Wong, E. C. & Frank, L. R.Magn. Reson. Med. 39, 855–864 (1998).

9. Kastner, S., Pinsk, M. A., De Weerd, P.,Desimone, R. & Ungerleider, L. G. Neuron22, 751–761 (1999).

10. Luck, S. J., Chelazzi, L., Hillyard, S. A. &Desimone, R. J. Neurophysiol. 77, 24–42(1997).

11. Polonsky, A., Blake, R., Braun, J. & Heeger,D. J. Nat. Neurosci. 3, 1153–1159 (2000).

12. Leopold, D. A. & Logothetis, N. K. Nature379, 549–553 (1996).

13. Braitenberg, V. & Schuz, A. Anatomy of theCortex: Statistics and Geometry (Springer,Berlin, 1991).

14. Gray, C. M., Konig, P., Engel, A. K. & Singer,W. Nature 338, 334–337 (1989).

15. Fries, P., Reynolds, J. H., Rorie, A. &Desimone, R. Science 291, 1560–1563(2001).

matches between BOLD and unit activ-ity become less of a problem, they maystill come into play and explain, at leastin part, the greater-than-expected BOLDresponse in the binocular rivalry exper-iment as well. Thus, until we are able tocharacterize the precise relationshipbetween BOLD and neuronal activity atall levels of activation, or better yet,understand the neural vascular couplingmechanism governing this relationship,interpretation of BOLD signal in thiscontext remains limited.

Overall, the work of Logothetis et al.5 is outstanding not only for thedepth and sophistication with which itaddresses some pressing questionsabout BOLD contrast, but also for theclarity with which it brings unansweredquestions to bear. It may be prudent,however, to make a few cautionaryremarks in closing. First, the resultswere obtained in anesthetized monkeys,and therefore it will be important todetermine how well the results hold upin the awake, behaving animal. Second,although the LFP was a significantlybetter predictor of the BOLD responsethan multi-unit activity, the differencein predictability was not large. On aver-age, the LFP accounted for only 7.6%more of the variance in the BOLDresponse compared to multi-unit activ-ity. Third, it may be an overstatementto conclude, as the authors did, thatbecause the LFP correlated more close-ly than multi-unit activity with theBOLD response, BOLD signals “reflectthe input and intracortical processingof a given area rather than its spikingoutput”. As mentioned earlier, althoughthe LFP reflects mainly summated den-dritic and somatic currents arising fromsynaptic activity, action potentials canalso contribute to the LFP, dependingon their phase and decay time and thedistance of the spiking neurons fromthe electrode tip. Therefore, the LFPreflects more than just input and intra-cortical processes. Conversely, eventhough multi-unit activity is, by defin-ition, spike activity, it may not, in thestrictest sense, reflect the output of anarea. Indeed, it has been estimated thatabout 80% of cortical axons terminate

news and views

12.5 25 50 1000

0.2

0.4

0.6

0.8

1.0

1.2

Michelson contrast

Nor

mal

ized

res

pons

e

Contrast response

BOLD

LFPMUA

Fig. 3. Normalized response amplitude ofLFP, multi-unit activity and BOLD against con-trast. Data from five sessions with a pulseduration of 12.5 seconds. Reprinted by per-mission from Nature 412, 150–157, copyright2001 Macmillan Magazines Ltd.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


brief communications


COS7 cells transiently transfected with ASIC cDNA were used).Neuron somas were patch-clamped and used only if a change inextracellular pH from 7.4 to 7.0 evoked a large, amiloride-sen-sitive Na+ current; response to this small pH change indicatesthe presence of ASIC3 (ref. 6), the most acid-sensitive of the 5types of ASICs expressed in rats7. We used either unlabeled sen-sory neurons or neurons for which the site of innervation wasidentified by transport of dye placed in the pericardial sack5 orskeletal muscle8 (Fig. 1a); results were identical regardless of thesite of sensory innervation.

When 15 mM lactate was present, decreasing extracellular pHto 7.0 depolarized the neurons about 70% more than pH 7.0alone, often triggering or increasing action potential activity (Fig. 1b). Likewise, currents evoked by a step to pH 7.0 increaseddramatically in the presence of lactate (Fig. 1c). Muscle ischemiacauses extracellular lactate to rise to about 15 mM from restinglevels below 1 mM9; thus, both the pH and lactate concentrationused are physiologically relevant. This response likely explainswhy, in whole animal studies, topically administered lactic acidcauses greater firing of sensory axons than do other acid stimuliat the same pH2,10,11.

The response increased with lactate concentration and showedno saturation up to 30 mM (Fig. 1d). ASIC3 as well as ASIC1achannels transfected into COS7 cells exhibited the lactate mod-ulation (Fig. 1e), indicating that the modulation is not uniqueto ASIC3. Lactate applications at pH values that did not activatethe acid-sensing channel (8.0 or 7.4) caused no depolarization,no current and no change in input resistance (n = 3, data notshown); thus, lactate acts by modulating but not activating theASIC.

What is the mechanism of lactate modulation? Lactate doesnot use a separate receptor and signaling cascade; it works onASICs in excised membrane patches, and it completes itsaction too fast (within 20 ms; see Supplementary Fig. 1 on thesupplementary information page of Nature Neuroscienceonline). Lactate may act using its ability to chelate extracellu-lar divalent ions. If so, it should be mimicked by other mono-carboxylic acids and by direct alteration of divalent ions.Lactate, pyruvate and formate each enhanced ASIC currentand their order of potency—lactate (62% enhancement) ismore potent than pyruvate (35%), which is more potent thanformate (20%)—matches their order of divalent ion bindingaffinities12 (Supplementary Fig. 2).

Fig. 1. Lactate enhances ASICs. (a) Brightfield and fluorescent photos ofdissociated DRG sensory neurons. Fluorescence is from retrogradelytransported dye (DiI) injected into the sensory target (which, in differentpreparations, was either skeletal8 or cardiac5 muscle). Representative volt-age (b) and current (c) recordings from a neuron briefly exposed to pH 7.0(resting pH 8.0) in the presence or absence of 15 mM lactate. The channelis an ASIC because it passed Na+ better than K+ and was blocked by 10 µMamiloride (data not shown). (d) Average (± s.e.m.) increase of pH 7.0-evoked current by the indicated lactate concentrations. (e) Currents fromCOS-7 cells transfected with ASIC3 or ASIC1a evoked by indicated pHwith and without 15 mM lactate. External solution consisted of 130 mMNaCl, 5 mM KCl, 10 mM glucose, 1 mM MgCl2, 2 mM CaCl2 and 20 mMHEPES, pH 8.0 with NMG+. Test solutions were identical to this except forpH (7.0–6.0), pH buffer (MES for lower pH, MOPS near pH 7) or divalention concentrations. Lactic acid (Sigma, St. Louis, Missouri) was addedbefore pH adjustment. Internal (patch pipette) solution consisted of 100 mM KCl, 10 mM EGTA, 40 mM HEPES, 5 mM MgCl2, 2 mM NaATPand 0.3 mM Na3GTP, pH 7.4 with KOH. Scale bars (a) 50 µm; (b) 250 ms,20 mV; (c) 500 ms, 1 nA; (d) 500 ms, 500 pA; (e) 500 ms, 250 pA (top),500 pA (bottom).

Lactate enhances theacid-sensing Na+ channelon ischemia-sensingneuronsD. C. Immke and E. W. McCleskey

Vollum Institute, Oregon Health Sciences University, Portland, Oregon 97201-3098, USA

Correspondence should be addressed to D.I. ([email protected]) or E.M. ([email protected])

Lactic acid produced by anaerobic metabolism during cardiacischemia is among several compounds suggested to trigger angi-nal chest pain1,2; however, the pH reached when a coronary arteryis occluded (pH 7.0 to 6.7)3,4 can also occur during systemic aci-dosis, which causes no chest pain. Here we show that lactate, act-ing through extracellular divalent ions, dramatically increasesactivity of an acid-sensing ion channel (ASIC) that is highlyexpressed on sensory neurons that innervate the heart5,6. Theeffect should confer upon neurons that express ASICs an extrasensitivity to the lactic acidosis of local ischemia compared toacidity caused by systemic pathology.

Rat sensory neurons were used except in Fig. 1e (in which

a

b c

d e

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om




To explicitly examine the role of divalent ions, wecalculated their free concentrations in various lactatesolutions from published equilibrium constants12.Figure 2a shows currents from a representative cellexposed to pH 7.0 stimuli with and without 15 mMlactate, and with the indicated concentrations of Ca2+

and Mg2+ ions added to the media. The first twotraces show the effect of lactate in our standardrecording medium (2 mM Ca2+, 1 mM Mg2+). Withthese as the total divalent ion concentrations in solu-tion, the concentrations of Ca2+ and Mg2+ unbound to 15 mM lactate are 1.71 mM and 0.88 mM, respectively. The thirdtrace shows that a pH 7.0 stimulus with these lower divalent ionconcentrations perfectly mimics the action of 15 mM lactate. Toobtain 2 mM and 1 mM free Ca2+ and Mg2+ when 15 mM lac-tate is present, there must be 2.35 mM and 1.12 mM total Ca2+

and Mg2+ added to solution; these concentrations concealed thelactate effect (trace 4). Analogous experiments were done with 1 mM and 30 mM lactate; in each case, there was perfect agree-ment between lactate and its divalent ion mimic (Supplemen-tary Fig. 3). We conclude that lactate acts by decreasing divalentions in the extracellular media.

The large effect of small changes in divalent ion concentra-tion was unexpected, because there is little block of ASIC3 chan-nels by Ca2+ between 1 and 2 mM6. However, previouslypublished data was obtained with pH 6.0 stimuli, not the pH 7.0used here. Indeed, stimulating pH explains the inconsistency; thelactate/divalent effect is lost at lower pH (Fig. 2b). For a currentto increase only at high pH, divalent ions likely increase the sen-sitivity of the channel to protons. Consistent with this expecta-tion, the ASIC opened at much lower [H+] in the absence ofdivalent ions (Fig. 2c). Separately decreasing either Ca2+ or Mg2+

was sufficient to increase current (Supplementary Fig. 4). Becauseof the very steep activation curve of ASIC3 (Hill coefficient, 4)6,it takes only a 0.05 pH unit shift to mimic the action of 15 mMlactate. This mechanism—a shift in gating—distinguishes lac-tate modulation of ASICs from modulation by FMRFamide,which acts by slowing desensitization13.

Lactate modulation is an example of heterosensitization, aphenomenon in pain transduction in which a stimulus (lactate)increases nociceptor sensitivity without directly activating thetransducer (ASICs on ischemia-sensing neurons)14. In this case,

Fig. 2. Lactate modifies the medium, not the channel. (a) Currents from a representative cell activated by pH 7.0with and without 15 mM lactate and with the total divalention concentrations indicated. 15 mM lactate added to 2 mMCa2+ and 1 mM Mg2+ reduces the free concentrations to 1.71and 0.88 mM; 2.35 mM Ca2+ and 1.12 mM Mg2+ decrease to2 and 1 mM (equilibrium constants, KCa, 85 mM; KMg, 117 mM; pKa, 3.66)12. Currents vary according to free diva-lent ion concentrations rather than lactate itself. (b) Modulation is lost at more acidic pH. Percent (± s.e.m., n = 4–15 cells) increase in current evoked by the indicatedpH with 15 mM lactate, the 15 mM lactate mimic (1.71 Ca2+/0.88 Mg2+), or with 15 mM lactate on COS-7 cellsexpressing ASIC3. (c) Activation range depends on divalentions. Fractional current (normalized to that at pH 5; n ≥ 3)versus activating pH with and without added divalent ions.Inset, representative currents without added divalents. Scalebars, (a) 1 s, 400 pA; (c) 500 ms, 500 pA.

heterosensitization should enhance the specificity of the sensory response, making neurons that express ASICs more sen-sitive to acidity caused by local muscle ischemia, which produceslactate, than to acidity caused by systemic problems such as kid-ney dysfunction or diabetic ketoacidosis. The results are relevantto cardiac pain, and perhaps also to pain caused by other vaso-occlusive disorders—notably, sickle cell anemia—and to musclepain that follows severe exercise or grand mal seizures9.

Note: Supplementary figures are available on the Nature Neuroscience web site

(http://neuroscience.nature.com).

ACKNOWLEDGEMENTS

We thank R. Waldmann and M. Lazdunski for providing clones, M. Bobo for

tissue culture, and S. Sutherland, J. Adelman, S. Cook and C. Benson for

comments. The NIH supported this work.

RECEIVED 14 MAY; ACCEPTED 5 JULY 2001

1. Armour, J. A. Cardiovasc. Res. 41, 41–54 (1999).2. Pan, H. L., Longhurst, J. C., Eisenach, J. C. & Chen, S. R. J. Physiol. (Lond.)

518, 857–866 (1999).3. Cobbe, S. M. & Poole-Wilson, P. A. J. Mol. Cell Cardiol. 12, 745–760 (1980).4. Yan, G. X. & Kleber, A. G. Circ. Res. 71, 460–470 (1992).5. Benson, C. J., Eckert, S. P. & McCleskey, E. W. Circ. Res. 84, 921–928 (1999).6. Sutherland, S. P., Benson, C. J., Adelman, J. P. & McCleskey, E. W. Proc. Natl.

Acad. Sci. USA 98, 711–716 (2000).7. Waldmann, R. & Lazdunski, M. Curr. Opin. Neurobiol. 8, 418–424 (1998).8. Honig, M. G. & Hume, R. I. J. Cell Biol. 103, 171–187 (1986).9. Cohen, R. D. & Woods, H. F. Diabetes 32, 181–191 (1983).10. Stahl, G. L. & Longhurst, J. C. Am. J. Physiol. 262, H748–H753 (1992).11. Hong, J. L., Kwong, K. & Lee, L. Y. J. Physiol. (Lond.) 500, 319–329 (1997).12. Martell, A. & Smith, R. Critical Stability Constants (Plenum, New York, 1977).13. Askwith, C. C. et al. Neuron 26, 133–141 (2000).14. Woolf, C. J. & Salter, M. W. Science 288, 1765–1769 (2000).

a

b c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om




Opposing actions ofprotein kinase A and Cmediate Hebbian synapticplasticityMin-Xu Li1, Min Jia1, Hao Jiang2, Veronica Dunlap1 andPhillip G. Nelson1

1 Laboratory of Developmental Neurobiology, National Institute of Child Healthand Human Development, Building 49, Room 5A38, National Institutes of Health, Bethesda, Maryland 20892, USA

2 William T. Gossett Neurology Laboratories, Henry Ford Health Sciences Center, One Ford Place, 4D, Detroit, Michigan 48202, USA

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

A compartmental nerve–muscle tissue culture system expressesHebbian1,2 activity-dependent synapse modulation. Proteinkinase C (PKC) mediates a heterosynaptic loss of efficacy, andwe now show that protein kinase A (PKA) is involved in homosy-naptic stabilization. Both work through postsynaptic changes inthe acetylcholine receptor (AChR) as measured electrophysio-logically and by imaging techniques.

Previous studies3,4 indicated that activation of PKC is neces-sary and sufficient for the stimulation-induced heterosynapticloss of synaptic efficacy in our system, and is involved in neuro-muscular synapse elimination in vivo5. Loss of efficacy in vitrowas related to a decrease in postsynaptic responsiveness (areduced spontaneous mini-EPP amplitude) and reduction inacetylcholine receptors (AChR) at the synapse. Here we exam-ine neuromuscular synapses in a chamber system with two com-peting cholinergic neuronal populations converging on a

common target of skeletal muscle cells. We demonstrate bothpositive and negative effects of activation upon synaptic strength,examine some of the cell biological mechanisms and moleculartargets involved in early stages of this process, and provide a gen-eral model for the Hebb synapse. (For detailed methods, see thesupplementary information page of Nature Neuroscience online.)

In addition to electrical stimulation of synapses, we used directactivation of PKC by phorbol 12-myristate 13-acetate (PMA) as aconvenient and reproducible surrogate for such stimulation4 to produce synapse downregulation. Treatment with the inactive phor-bol ester 4 alpha-phorbol 12,13 didecanoate did not result in anydecrement in synaptic efficacy (data not shown).

If, as our previous results indicate, PKC acts to reduce synapticefficacy by downregulating receptor aggregates, what preserves theintegrity of active synapses on these cells? We wished to identifyand isolate a ‘protective’ effect of stimulation6 to study its mecha-nism. Treatment with 60 nM of PMA for 2 h resulted in reductionin AChR aggregate labeling to about 60%. Over that period, com-plete loss of individual aggregates did not occur, but rather a reduc-tion occurred in the number of receptors in each initially identifiedaggregate. Electrical stimulation of the neural input to the musclefibers during the PMA treatment completely prevented the decreasein labeling produced by PMA (Fig. 1a and b). The result with intra-cellularly recorded end plate potentials (EPPs) was similar (Fig. 1c).The average EPPs from a series of singly innervated cells werereduced by PMA, but stimulation blocked the PMA effect.

What might be responsible for this activity-dependent sparing?The AChR may be affected by the action of PKA7, suggesting thatthe kinase might be involved in Hebbian behavior. We testedwhether an inhibitor of PKA, H-89, would prevent the synapsepreservation produced by activation of the synapse. In the presenceof H-89, stimulation of singly innervated myotubes resulted in asignificant decrement in synaptic efficacy (Fig. 1d). Furthermore,measurement of spontaneous mini EPPs showed that this was atleast in part a postsynaptic effect, as mini EPP amplitude was sig-nificantly decreased in the H-89-treated, stimulated preparations

Fig. 1. Neuronal electrical stimulation prevents the downregulation of synapse efficacy by PKC activation; this effect involves PKA. (a) AChR images of sin-gle AChR aggregates labeled with rhodaminated α-bungarotoxin (Rh-α-BTX) before and after treatment from control preparations, from preparationstreated with 60 nM PMA or with PMA plus electrical stimulation. PMA alone reduced AChR fluorescence intensity. PMA plus electrical stimulation did notreduce AChR intensity. Treatment and stimulation were for 2 h. Scale bar, 10 µm. Preparations were kept in a low concentration of Rh-α-BTX (25 nM)throughout the experiment, so the intensity of labeling was the net result of both insertion and loss of receptor. (b) Results of experiments using quantita-tive fluorescence imaging technique as in (a) show that after treatment with 60 nM PMA for 2 h, AChR fluorescence intensity was reduced (*p < 0.0001, n = 35), but electrical stimulation (0.3 Hz) blocked the PMA induced reduction of fluorescence intensity of AChR aggregates (**p > 0.2, n = 23). Control, n = 27. (c) Intracellular EPP recordings showed that without electrical stimulation, PMA reduced EPP amplitude (n = 5, *p < 0.023). In preparations withneuronal electrical stimulation (0.3 Hz), PMA (100 nM, 2 h) did not induce significant EPP reduction (n = 7, **p > 0.37). (d) Left bars, control group with sin-gle inputs stimulated for 2 h (n = 5). There was no significant difference before and after stimulation (*p > 0.4). Right bars, EPPs recorded in 10 experimentsbefore and after the treatment of 1 µM H-89 in the center chamber, plus electrical stimulation for 2 h (n = 10, **p < 0.025). , before treatment; , aftertreatment. (e) H-89 reduced miniature end-plate potentials (MEPPs). MEPPs (recorded in 0.1 µM TTX) in control group before () and after 2 h electricalstimulation (; n = 5, *p > 0.2), and in the group treated with H-89 in the central chamber with electrical stimulation before () and after treatment (; n = 4, **p < 0.05). (f) Intracellular recording of EPPs were taken after PKI injection (5 Hz, 30 ms, approximately; 1 nA for 30 min) but before and aftera 20 min period without (left) or with (right) 5 Hz neural stimulation. For both data sets, n = 4; p < 0.005.

Contrl 0 h

0

20

40

60

80

100

120

140

AC

hR in

tens

ity (

%)

Control PMA PMA+stim 2 h/0 h 2 h/0 h 2 h/0 h

*

**

EP

P a

mpl

itude

(m

V)

0

No elec stim Elec stim

ME

PP

(m

V)

0

0.2

0.4

0.6

0.8

Stim Stim+H-89 in center

* * **

**

EP

P a

mpl

itude

(m

V)

0

2

4

6

8 * ***

**

Stim Stim +H-89 center

Before After PMA PMA

2

4

6*

**

****

*

Before After PMA PMA

EP

P s

lope

(V

/s)

% o

f ini

tial v

alue

PKI inject PKI inject+ stim

40

80

120

160

0

Contrl 2 h PMA 2 h PMA + stim 2 h

BeforePMA

Before PMA+ stim

a b c

de f

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om




Fig. 2. PKA activation prevents stimulation-related synapse downregulation. (a) End-platepotentials (EPPs) recorded from a doubly inner-vated muscle. Unilateral electrical stimulationfor 1 h induced EPP reduction primarily of theunstimulated side. The preparation was incu-bated with 1 µM µ-conotoxin to block muscleaction potentials during the recordings. Scalebar, 2 mV, 10 ms. (b) Single AChR aggregateslabeled with Rh-αBTX from stimulated orunstimulated synapses of doubly innervatedmyotubes, before and after a 2-h stimulationperiod. Scale bar, 10 µm. (c) Quantitative resultsfrom several experiments such as in (b). Nostim, n = 27, p > 0.7 (compared with 0 h); stimside, n = 37, p > 0.95; unstim side, n = 9, p < 0.0001. (d) cAMP prevents the het-erosynaptic loss of synapse strength producedby unilateral stimulation of bilaterally innervatedmyotubes. EPPs were measured before and after2 h stimulation with and without cAMP (2 mM)in the center chamber. For the unstimulatedinputs, db cAMP produces a significant block ofEPP decrement (p < 0.004). , stimulation only,no cAMP (n = 10); , stimulation with cAMP inthe center chamber. The cAMP was then washed out and the EPPs were re-measured (n = 5). (e) Treatment with 2 mM db cAMP completely prevents thedecrement in synaptic AChR produced by PMA (treatments were for 2 h). PMA-treated, n = 29; PMA plus cAMP (2 mM), n = 29. *p < 0.0032. (f) Calcitoningene related peptide (CGRP) increases cellular cAMP level and leads to the activation of PKA in muscle. In the present experiments, 1 µM CGRP reducedthe functional synapse loss produced by unilateral electrical stimulation. Treatment and stimulation were for 20 h. Control, n = 14; CGRP, n = 15, p < 0.02.

(Fig. 1e). H-89 treatment without stimulation produced no synap-tic efficacy decrease (data not shown). Experiments with anotherblocker of PKA, Rp-cAMPS, confirmed the H-89 experiments (datanot shown). Confirming the postsynaptic action of the PKA, wefound that a PKA blocker, PKI, injected postsynaptically, induced astimulation-dependent loss of synaptic efficacy (Fig. 1f).

As PKA is activated by an increase in cAMP, and cAMP isinduced by calcitonin gene-related peptide (CGRP)7, we used boththese agents to further test the hypothesis that PKA mediatessynapse retention. The Hebbian decrement of synaptic efficacy, pre-dominantly of non-stimulated inputs to doubly innervated, uni-laterally stimulated myotubes (Fig. 2a), is reflected in changes inthe AChR (Fig. 2b and c). The inputs were identified by stimulatingeither the labeled or the unlabeled side of preparations with GFPneurons in only one side compartment. There was no significantloss of AChR from stimulated endings, whereas significant lossoccurred from the unstimulated synapses of doubly innervatedmyotubes stimulated through the other synapse. The reduction ofEPPs of unstimulated inputs was completely prevented by treat-ment with 2 mM dibutyryl cAMP (Fig. 2d). As with electrical stim-ulation, cAMP treatment of the synapses in the center chamberprevented the loss of synapse-associated AChR induced by PMA(Fig. 2e). To rule out the possibility that this was a presynapticallymediated effect, we used aneural myotube cultures. In such cul-tures lacking nerve cells, PMA treatment produced a 29 ± 4% (n =22) loss of AChR from aggregates, and PKA activation blocked thisloss (PMA + db cAMP gave 13 ± 5% loss, n = 18; p < 0.025). Treat-ment with CGRP reduced stimulation-induced synapse reduction(Fig. 2f; see ref. 3 for assay methodology).

Wortmannin (10 nM), an inhibitor of IP3 kinase, did not causeloss of stimulated inputs; nor did it block the loss of unstimulatedinputs in doubly innervated myotubes (data not shown).

We attribute the decrement in the efficacy of the unstimulatedinput in doubly innervated, unilaterally stimulated myotubes to ageneral action of PKC, whereas we attribute the retention of the

stimulated inputs to a local action of PKA. These differential actionsof the kinases are due to differences in their spatial localization8 andthe demonstrably different phosphorylations they produce on theAChR9, among other molecules, with concomitant differentialeffects on AChR stability in the muscle membrane. Thus, the kinas-es may work both directly and in opposition on the AChR.

Similar mechanisms may be involved in the process of cen-tral synapse elimination during brain development. Glutamateand GABA receptors are substrates for a number of kinases10,and elimination of the polyinnervation of Purkinje cells byclimbing fibers in the cerebellum is reduced in mutant micelacking PKCγ11. PKA and PKC actions produce differentialphosphorylation of the NMDA receptor with concomitanteffects on receptor stability12.

We propose that for a major stage of the activity-dependentsynapse-elimination process, both at the neuromuscular junctionand at central synapses, the conjoint but opposed action of thekinases is critical in conferring stimulus specificity.

Note: Supplementary methods are available on the Nature Neuroscience web site

(http://neuroscience.nature.com/web_specials).

RECEIVED 13 JUNE; ACCEPTED 17 JULY 2001

1. Hebb, D. O. The Organization of Behavior (Wiley, New York, 1949).2. Stent, G. Proc. Nat. Acad. Sci. USA 70, 997–1001 (1973).3. Jia, M., Li, M. X., Dunlap, V. & Nelson, P. G. J. Neurobiol. 38, 369–381 (1999).4. Lanuza, M. A. et al. J. Neurosci. Res. 61, 616–625 (2000).5. Lanuza, M. A. et al. J. Neurosci. Res. 63, 330–340 (2001).6. Akaaboune, M., Culican, S. M., Turney, S. G. & Lichtman, J. W. Science 286,

503–507 (1999).7. Lu, B., Fu, W. M., Greengard, P. & Poo, M.-M. Nature 363, 76–79 (1993).8. Coghlan, V. M., Bergeson, S. E., Langeberg, L., Nilaver, G. & Scott, J. D. Mol. Cell

Biochem. 127–128, 309–319 (1993).9. Nimnual, A. S. et al. Biochemistry 37, 14823–14832 (1998).10. Yee, G. H. & Huganir, R. L. J. Biol. Chem. 262, 16748–16753 (1987).11. Kano, M. et al. Cell 83, 1223–1231 (1995).12. Tingley, W. G. et al. J. Biol. Chem. 272, 5157–5166 (1997).

1 1

1

Unstim sideStim side

0

120100

80

60

40

20

Perc

ent c

hang

e of

ACh

R int

ensit

y

2 h

/ 0

h

No stim Stimside

Unstimside A

Ch

R f

luo

resc

en

cein

ten

sity

2 h

/0 h

(%

)

100

E P

P a

mp

litu

de

aft

er

stim

vers

us

be

fore

stim

(%

) Unstimulated side ofdouble innervated inputs

control

020

40

60

80

100

120

140

160

180

PMA60 nM

PMAcAMP

center

150

50

0

*

*control

NocAMP

cAMP Functional synapse

loss (

%)

Elec stim Elec stim+CGRP0

10

20

30

40

50

Stim side Unstim side

Beforestim

Beforestim

Afterstim

Afterstim

a

b

cd e f

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om




Fig. 1. Locomotor response tococaine in mGluR5 WT (n = 14) andnull mutant (n = 16) mice. Valuesrepresent mean percent activitycounts ± s.e.m. *p < 0.05 versussaline; **p < 0.01 versus saline(Dunnett’s test after two-wayrepeated-measures analysis of vari-ance; ANOVA). For detailed meth-ods, see the supplementaryinformation page of NatureNeuroscience online.

Reinforcing andlocomotor stimulanteffects of cocaine areabsent in mGluR5 nullmutant miceChristian Chiamulera1*, Mark P. Epping-Jordan2*,Alessandro Zocchi1,3, Clara Marcon1, Cécilia Cottiny2,Stefano Tacconi1, Mauro Corsi1, Francesco Orzi3,4 andFrançois Conquet2

1 Department of Biology, GlaxoSmithKline Laboratories, via Fleming 4, 37100 Verona, Italy

2 GlaxoSmithKline R&D, Institut de Biologie Cellulaire et de Morphologie, rue du Bugnon 9, 1005 Lausanne, Switzerland

3 Neuromed Institute, via Atinense 18, 86077 Pozzilli, Italy 4 Department of Neurological Sciences, University of Rome La Sapienza,

Piazzale Aldo Moro 5, 00185 Rome, Italy * The first two authors contributed equally to this work.

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

Both ionotropic and metabotropic glutamate receptors (mGluRs)are involved in the behavioral effects of pyschostimulants1–3; how-ever, the specific contributions of individual mGluR subtypesremain unknown. Here we show that mice lacking the mGluR5gene do not self-administer cocaine, and show no increased loco-motor activity following cocaine treatment, despite showingcocaine-induced increases in nucleus accumbens (NAcc)dopamine (DA) levels similar to wild-type (WT) mice. Theseresults demonstrate a significant contribution of mGlu5 recep-tors to the behavioral effects of cocaine, and suggest that theymay be involved in cocaine addiction.

Both acute and repeated cocaine administration increaseglutamate concentrations in the NAcc4,5, a brain region asso-ciated with the reinforcing and locomotor effects of cocaine6,7.Systemic and brain injections of non-selective mGluR agonistsand antagonists mediate baseline and psychostimulant-inducedlocomotor activity1–3. mGluR5 is highly expressed in theNAcc8, and repeated systemic cocaine treatment increasesmGluR5 mRNA levels in the NAcc shell and dorsolateral stria-

tum9; however, the functional role of mGluR5 in cocaine effectsremains unknown.

To investigate this role, we first examined the locomotorresponses to cocaine in F5 generation mGluR5 WT and nullmutant mice. Baseline locomotor activity did not differbetween mutant and WT mice (mutant, 2710 ± 642; WT, 2393± 479, mean horizontal activity counts/45-min session ±s.e.m.). Cocaine induced a significant, dose-dependent increasein locomotor activity in WT mice, but did not alter locomo-tor activity in mutant mice at any time point or dose tested(Fig. 1). Although mice received repeated cocaine exposurethat may have induced some behavioral sensitization in WTmice, locomotor activity was not increased in mutant mice.Our results indicate that mGluR5 is essential for cocaine-induced hyperactivity.

To investigate whether the reinforcing properties of cocainewere affected by the mGluR5 mutation, we examined intravenouscocaine self-administration (SA) in WT and mutant mice. Acqui-sition of a discriminated two-lever food-reinforced task did notdiffer between WT and mutant mice (Fig. 2a). When intravenouscocaine was substituted for food, WT mice acquired stable cocaineSA across a typical dose range10, but mutant mice did not self-administer cocaine at any dose tested (Fig. 2b). Active leverresponding in mutant mice extinguished within three to five ses-sions at all cocaine doses, and no mutant mouse acquired stableSA, suggesting that the reinforcing properties of cocaine are absentin mice lacking mGluR5. Data from the food training suggest thatthe failure to acquire cocaine SA was not due to an inability to learnthe lever-press task, and that the reinforcing properties of food areunchanged in mutant mice.

The selective mGluR5 antagonist 2-methyl-6-(phenylethynyl)-pyridine (MPEP)11 dose-dependently decreased cocaine SA (Fig. 2c) in C57Bl/6J mice. The effect of MPEP was specific tococaine reinforcement, as MPEP had no effect on the rate of food-reinforced lever pressing (Fig. 2d) under the same schedule ofreinforcement as during cocaine SA. These results suggest that


Fig. 2. Food and cocaine reinforcement. (a) Acquisition to criterion offood-reinforced lever pressing did not differ in mGluR5 WT (n = 5) andnull mutant (n = 6) mice. Values represent mean ± s.e.m. (Student’s t-test). (b) Cocaine SA in mice shown in (a). Values represent mean num-ber of injections for 2 sessions at each dose ± s.e.m. *p < 0.05 versussaline within genotype; ***p < 0.05 WT at 0.4 versus WT at 3.2 mg/kg/injection (Bonferroni-corrected Student’s t-tests after two-wayrepeated-measures ANOVA). (c) MPEP dose-dependently decreasedcocaine SA in C57Bl/6J mice (n = 5). Values represent mean percent ofbaseline number of injections per 1 h session at 0.8 mg/kg/injection, *p < 0.05 versus saline; ***p < 0.05 versus 3 mg/kg MPEP (means compar-isons after one-way ANOVA). (d) MPEP (30 mg/kg i.v.) had no effect onthe number of food-reinforced lever presses per minute in C57Bl/6J mice(n = 5) (repeated-measures t-test). For detailed methods, see the supple-mentary information page of Nature Neuroscience online.

a b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om




the absence of cocaine SA in mutant mice was due to the loss ofmGlu5 receptors and not to developmental alterations resultingfrom the genetic mutation.

Numerous studies have reported that self-administered cocaineincreases DA levels in the ventral striatum6, and psychostimulant-induced locomotor activity and increased mesoaccumbens DA lev-els in mice are closely correlated12. Because mutant mice showedneither a cocaine-induced locomotor response nor cocaine SA, weexamined the effect of cocaine on NAcc DA levels using microdial-ysis in conscious, freely moving WT and mutant mice12. Wild-typeand mutant mice had similar basal levels of extracellular DA.Cocaine-induced (10 mg/kg, i.p.) increases in extracellular DA levels did not differ between WT and mutant mice (Fig. 3b). These results suggest that the absence of mGluR5 affectsneither baseline nor cocaine-induced increases in NAcc DA levels.

To ensure that changes in responses to cocaine in mutant micewere not due to mGluR5 mutation-induced alterations indopaminergic elements, we investigated the brain distributionand expression of DA receptors and the DA transporter (DAT)in mutant and WT mice. No differences were found between WTand mutant mice in binding of the selective radiolabeled com-pounds 3H-SCH23390 to D1-like and 3H-YMO91512 to D2-likeDA receptors and 3-WIN35,428 to the DAT, or in expression ofD1 or D2 DA receptor mRNA examined by in situ hybridization(data not shown). These findings indicated that the expressionand distribution of DA receptors and of the DAT are not alteredin the mutant mice.

Several neurotransmitters and peptides contribute to cocainedependence6,7. Although evidence supports a primary involve-ment for DA, the precise roles of specific DA receptor subtypes

remain unclear. D1- or D2-like DA receptor antagonists reducethe reinforcing effects of cocaine6,7; however, D1 receptor mutantmice acquire a cocaine-conditioned place preference13 and D2receptor mutant mice self-administer cocaine (S.B. Caine et al.,Soc. Neurosci. Abstr. 26, 681.8, 2000). Likewise, the exact mecha-nisms of the mGluR5 contribution to cocaine dependence arenot known. It is possible that glutamate acts in synergy withmesolimbic DA afferents into the NAcc to mediate the effects ofcocaine. Acute cocaine increases extracellular NAcc dopamine14

and glutamate4 levels, and these effects are enhanced after repeat-ed cocaine administration5,14. Excitatory amino acid (most like-ly glutamatergic) and mesolimbic dopaminergic terminals formsynapses on single NAcc neurons2,3. Nucleus accumbens outputneurons express both DA2,6,7 and mGlu58 receptors. As has beensuggested for locomotor activity2, mGluR subtypes expressed byNAcc projection neurons may interact with dopaminergic inputsthrough their respective intracellular signaling pathways to influ-ence the reinforcing effects of cocaine. Regardless of the specificmechanisms involved, the present results suggest that mGluR5is essential in cocaine SA and locomotor effects.

Note: Supplementary methods are available on the Nature Neuroscience web site


ACKNOWLEDGEMENTSWe thank A. Morrison and C. Corti for their contribution in the establishment

and detection of the mutation, F. Fornai and E. Grouzman for HPLC analyses,

and M. Geyer, D. Lavery and E. Ratti for reviewing the manuscript. This work

was supported by GlaxoSmithKline R&D.


1. Kim, J. H. & Vezina, P. J. Pharmacol. Exp. Ther. 284, 317–322 (1998).2. Vezina, P. & Kim, J. H. Neurosci. Behav. Rev. 23, 577–589 (1999).3. Swanson, C. J. & Kalivas, P. W. J. Pharmacol. Exp. Ther. 292, 406–414

(2000).4. Smith, J. A. et al. Brain Res. 683, 264–269 (1995).5. Pierce, R. C., Bell, K., Duffy, P. & Kalivas, P. W. J. Neurosci. 16, 1550–1560

(1996).6. Koob, G. F., Sanna, P. P. & Bloom, F. E. Neuron 21, 467–476 (1998).7. White, F. J. & Kalivas, P. W. Drug Alcohol Depend. 51, 141–153 (1998).8. Tallaksen-Greene, S. J., Kaatz, K. W., Romano, C. & Albin, R. L. Brain Res.

780, 210–217 (1998).9. Ghasemzadeh, M. B., Nelson, L. C., Lu, X. Y. & Kalivas, P. W. J. Neurochem.

72, 157–165 (1999).10. Caine, S. B., Negus, S. S. & Mello, N. K. Psychopharmacology 147, 22–24

(1999).11. Gasparini, F. et al. Neuropharmacology 38, 1493–1503 (1999).12. Zocchi, A. et al. Neuroscience 82, 521–528 (1998).13. Miner, L. L. et al. Neuroreport 6, 2314–2316 (1995).14. Parson, L. H. & Justice, J. B. J. Neurochem. 61, 1611–1619 (1993).

Fig. 3. Extracellular NAcc DA levels in mGluR5 WT and null mutantmice. (a) Dialysis probe location. Solid and dashed boxes indicate theminimum and maximum extent of probe placements. ac, anterior com-missure; CPu, caudate putamen; LV, lateral ventricle; NAccC, nucleusaccumbens core; NAccSh, nucleus accumbens shell. (b) DA level analysis.10 µl samples were collected every 20 min from WT and mutant mice(saline, n = 3/genotype; cocaine, n = 3/genotype) and were analyzed byHPLC. Values represent mean pg/sample DA ± s.e.m. *p < 0.05; **p < 0.01 cocaine versus saline at the same time point within genotype(Bonferroni-corrected Student’s t-test after separate two-way repeated-measures ANOVA). Extracellular DA levels were not significantly differ-ent between cocaine-treated WT and mutant groups (repeated-measuresANOVA, no significant main effects or interaction). For detailed methods,see the supplementary information page of Nature Neuroscience online.

a

b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om




Synchrony does notpromote grouping intemporally structureddisplaysHany Farid1 and Edward H. Adelson2

1 Department of Computer Science and The Center for Cognitive Neuroscience, Dartmouth College, Hanover, New Hampshire 03755, USA

2 Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA

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

It has been proposed that the human visual system can use tempo-ral synchrony to bind image regions into unified objects1,2,3, as pro-posed in some neural models4. Here we present experimental resultsfrom a new dynamic stimulus suggesting that previous evidencefor this hypothesis can be explained with the well-established mech-anisms of early visual processing, thus obviating the need to positnew synchrony-sensitive grouping mechanisms (see also ref. 5 for acritique of the binding by neural synchrony hypothesis).

In a previous study, the hypothesis for binding by temporalsynchrony3 was demonstrated with a dynamic texture displaycomposed of randomly oriented Gabor elements (Fig. 1a). Oneach frame, the phase of each Gabor shifted forward or back-ward according to a random process. By using one randomprocess for all the Gabors in a central rectangular region and adifferent process for all the Gabors in the surrounding region,the authors created a form cue that they claimed was definedsolely by temporal synchrony. Subjects were readily able to dis-tinguish the shape of the central region. This led the authors toconclude that the visual system must precisely register and cor-relate changes in motion across a spatially distributed area.

We have argued, however, that these results can be explainedwith the well-established filtering mechanisms of early visual pro-cessing6. Due to the stochastic nature of the reversal sequences,there were moments when the central Gabors rapidly alternatedbetween forward and backward shifts (thus ‘jittering’ in place),whereas the surrounding Gabors had a run of all forward or allbackward shifts, or vice versa. We showed that a temporal band-pass filter (with a temporal integration window on the order of 10frames7,8) can convert these relatively large-scale temporal changedifferences into a contrast cue (Fig. 1b). This led us to concludethat this cue, not a finer temporal synchrony cue, is responsiblefor the perception of form in these displays.

Here we report on a new dynamic textured stimulus thatdissociates the potential grouping cues of temporal synchronyand integrated contrast. Our basic stimulus (Fig. 1c) consistedof an array of small windows, each containing dots driftingwith a constant speed and direction. Across windows, the speedwas constant, but the direction was randomized. On eachframe, the dots moved randomly forward or backward alongtheir specified direction (Fig. 2a). As with the original Gaborstimulus, a form cue defined by temporal synchrony was intro-duced; the motion reversals of all the dots in a central regionwere synchronized to one random process, whereas the rever-sals in the surround were synchronized to another process. Thecentral region was a horizontally or vertically oriented rectan-

gle, and subjects were asked to determine its orientation.As with the original Gabor stimulus, the dot stimulus con-

tained a temporal contrast cue. The cue emerged when, forexample, the central dots repeatedly alternated between for-ward and backward shifts while the surround dots had a runof shifts in one direction. Just as before, the rapid reversalscaused all the dots in one region to repeatedly fall back ontothemselves (Fig. 2a), thus temporally integrating to low con-trast while the surround integrated to high contrast. This tem-poral contrast cue could be eliminated by simply changing thereversal angle, so that reversing dots no longer fell back ontothemselves. In the first condition, for example, the reversalangle was 140°, so that repeated reversals yielded a zigzag pat-tern (Fig. 2b). In the second condition, the reversal angle wasthe same, but the sign of each reversal was randomized, yield-ing a random walk pattern (Fig. 2c). In both of these condi-tions, the integration of a region rapidly reversing directionswas now largely the same as an area repeatedly shifting alongthe same direction (Fig. 1d). (We verified this by passing thevarious stimuli through a temporal band-pass filter.) At thesame time, we preserved the temporal synchrony cue that pur-portedly gives rise to the perception of form. (QuickTimemovies of these stimuli are available on the supplementaryinformation page of Nature Neuroscience online.)

We asked subjects to judge the orientation of a horizontallyor vertically oriented rectangle defined by synchronous motionreversals. When the motion reversal was 180°, performance wasnearly perfect, but when the reversal angle was decreased, sub-jects’ performance fell to near chance levels (Fig. 3). Perfor-mance in the 140° zigzag condition was slightly above chancebecause of a slight contrast cue remaining due to the spatialextent of the dots. If, as it has been argued, the perception ofform in these displays is a result of grouping mechanisms andprocesses based on temporal synchrony, then performance

Stimulus Filter output

Fig. 1. One frame of the Gabor and dot stimulus, and sample output oftemporal band-pass filtering. (a) One input frame of a Gabor stimulus.(b) Sample output of temporal band-pass filtering revealing a contrastcue. (c) One input frame of our dot stimulus. (For clarity, the regionbetween windows is shown in gray; in the actual stimulus, this regionwas black.) (d) A representative output of temporal band-pass filtering,revealing the lack of a contrast cue (see 120-degree zigzag condition,Fig. 2). The band-pass impulse response is h(t) = (kt/τ)n e–kt/τ (1/n! –(kt/τ)2/(n + 2)!), with τ = 0.01, k = 2 and n = 4, and t in (0, 10) frames.The particular value choices for these parameters are not crucial torevealing the contrast cue.

a b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om




should have been unaffected by these relatively minor changesin the angle of reversal. Instead, the perception of form wasgreatly diminished along with the temporal contrast cue. This

result, in combination with objections9 (W.H.A. Beaudot, VisionSci. 1, 147, 2001) to the conclusions of grouping by temporalsynchrony based upon periodic motion stimuli1,2 (as opposed tothe stochastic stimuli discussed here), provides strong evidencethat synchrony is not responsible for the perception of form inthese or earlier displays.

Note: Supplementary movies of stimuli are available on the Nature

Neuroscience web site (http://neuroscience.nature.com/web_specials).

ACKNOWLEDGEMENTS

H.F. is supported by a National Science Foundation Career Award (IIS-99-

83806) and a departmental National Science Foundation Infrastructure

grant (EIA-98-02068). E.H.A. is supported by a National Institute of Health

grant (EY12690-02).


1. Fahle, M. Proc. R. Soc. Lond. B Biol. Sci. 254, 199–203 (1993).2. Usher, M. & Donnelly, N. Nature 394, 179–181 (1998).3. Lee, S. & Blake, R. Science 8, 1165–1168 (1999).4. Singer, W. & Gray, C. Annu. Rev. Neurosci. 18, 555–586 (1995).5. Shadlen, M. N. & Movshon, J. A. Neuron 24, 67–77 (1999).6. Adelson, E. H. & Farid, H. Science 286, 2231 (1999).7. Watson, A. B. Handbook of Perception and Human Performance (Wiley,

New York, 1986). 8. Geisler, W. S. & Albrecht, D. G. Vis. Neurosci. 14, 897–919 (1997).9. Kandil, F. I. & Fahle, M. Eur. J. Neurosci. 13, 2004–2008 (2001).10. Brainard, D. H. Spat. Vis. 10, 443–446 (1997).

Fig. 3. Experimental results. Subjects’ ability to judge the aspect ratioof a rectangular figure. The results are the average of three subjects(two naive, one practiced); each bar corresponds to the averageacross 50 trials per subject, and the error bars indicate one standarderror. In the zigzag (z) and random-walk (r) conditions (Fig. 2c) thereversal sequence was designed to eliminate the temporal contrastcue present in the straight condition (s). Subjects’ judgment was at ornear chance (50%) when this cue was absent. The angle of motionreversal is noted below each condition. The motion reversals weresynchronized in all conditions.

s z z z r r r

50

60

70

80

90

100

Per

cent

cor

rect

180 140 120 90 140 120 90

Fig. 2. Temporal properties of the dot stimulus. (a) Schematic of a por-tion of our dynamic dot stimulus. The dots in each window moved in ran-dom directions, and on each frame, the dots either continued along theirspecified direction or reversed their direction. The reversals of all thedots within the dashed rectangle were synchronized. Shown is the motionsequence of three windows; two are in synchrony and the third is not.The vertical dashed lines mark reversal points. (b) In the zigzag condition,the synchronous motion reversals were preserved, but the reversal direc-tion was slightly altered. (c) All three motion advance/reversal conditions.The angle of motion reversal is noted below each schematic. Although allconditions preserved synchronous motion reversals, only the ‘straight’condition contained a strong temporal contrast cue. In the random-walkcondition, the sign of each reversal was randomized from frame to frameand across windows, but within a window, all the dots reversed along thesame direction. The dot stimulus consisted of 15 × 15 windows each ofsize 16 × 16 pixels. On a black background, white dots generated as two-dimensional Gaussians with a standard deviation of 2 pixels afforded sub-pixel motion. The initial placement of the dots within each window wasdetermined by randomly jittering (by ± 2 pixels) a lattice of dots separatedby 8 pixels. On each frame, the dots in each window moved randomly (by2 pixels) forward or backward along their specified direction. The direc-tion of motion across windows was randomized. The stimuli were dis-played for 0.75 s at 60 frames/s on a standard Apple monitor. The stimuliwere generated in Matlab (Natick, Massachusetts) and displayed using thePsychophysics toolbox10. A form cue was introduced by synchronizing themotion reversals of a horizontally or vertically oriented rectangle span-ning 7 × 5 windows. The motion reversals in the surround were synchro-nized to a separate process.

a

b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


A simple concept of the retina’s function—lateral inhibitionby horizontal and amacrine cells, a direct pathway mediatedby bipolar cells—is part of the everyday canon of neurobiolo-gy. In reality, the retina is a more complex and more subtlestructure than the textbooks imply. This is of course true alsofor other structures of the central nervous system—such as thehippocampus or cortex—where a similar mismatch existsbetween a simple iconic physiology and the facts of the bio-logical structure. Here I make an initial attempt to come togrips with the real retina, to encompass the system’s actual cel-lular complexity.

Neuroanatomical studies have reached a milestone. The iden-tification and classification of retinal neurons (Fig. 1), begunmore than 100 years ago by Santiago Ramon y Cajal, is nearingcompletion—the first time that this has been accomplished forany significantly complex structure of the mammalian CNS. Thisstatement is possible because much of the recent work on reti-nal cell populations has been quantitative. Staining cells as wholepopulations permits comparison of their numerical frequency.More importantly, when the number of cells of a general class(such as amacrine cells) is known, one can then determine whenthe identified types add up to the class total1–4. Much detailremains to be learned, and a few additional cell types are sure tobe discovered. However, we now know at least that no large cellpopulations remain unidentified, that there are no major pieces‘missing’ within the retina’s machinery5.

Unexpectedly, for most mammals, the numbers of bipolarand amacrine cells are distributed fairly evenly among the dif-ferent types. This differs from initial impressions, which weremuch influenced by early studies in primates. The primate foveais anomalous in being dominated numerically by a single type ofretinal ganglion cell, with an associated, specialized type of bipo-lar cell (see below). In other mammalian retinas, and away fromthe fovea in primates, individual bipolar, amacrine and ganglioncell types are numerically distributed in a more level way.Although variations certainly exist (generally, there are fewerwide-field than narrow-field neurons), there are no dominanttypes. In other words, the retina is not composed of a few majorplayers surrounded by a diverse cast of minor ones. Instead, itconsists of many parallel, anatomically equipotent microcircuits.

How can this awesome list of cell types be sorted? What uni-fying principles might allow us to conceive of the retina more

The fundamental plan of the retina

Richard H. Masland

Howard Hughes Medical Institute, Wellman 429, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts 02114, USA

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

The retina, like many other central nervous system structures, contains a huge diversity of neuronaltypes. Mammalian retinas contain approximately 55 distinct cell types, each with a differentfunction. The census of cell types is nearing completion, as the development of quantitativemethods makes it possible to be reasonably confident that few additional types exist. Althoughmuch remains to be learned, the fundamental structural principles are now becoming clear. Theygive a bottom-up view of the strategies used in the retina’s processing of visual information and sug-gest new questions for physiological experiments and modeling.

simply? From the work of many laboratories6–11, the fundamen-tal backbone of the retina’s structural organization has come intoview. It reinforces certain principles learned from physiologicalexperiments, and suggests new questions for further ones. HereI review the retina’s structure and point out some unresolvedfunctional issues that it suggests.

Parallel pathways from cones to ganglion cellsA typical mammalian retina contains 9–11 different types ofcone-driven bipolar cells. These represent an assortment of path-ways from cones to the inner retina, each carrying a different typeof information. This diversity was initially shown in the cells’structures and the distinct proteins that each expresses. Electro-physiological experiments are now beginning to reveal its func-tional consequences.

In most mammalian species, rods outnumber cones byapproximately 20-fold, and rods were once considered the pri-mordial photoreceptors. However, molecular cloning of thevisual pigments (opsins) that render these cells light-sensitiveled to the conclusion that cone pigments evolved long beforerhodopsin, the rod pigment12–14. The early photoreceptor thusseems to have been some type of cone (Fig. 2a). In retrospect,this makes sense; in building a cell to detect light, one wouldsurely design it for times when copious light is available. (Instarlight, a human rod photoreceptor has been calculated toreceive only one photon every 10 minutes8,15.) Cones are asso-ciated with a complex network of postsynaptic cells, whereasthe circuitry strictly associated with rods is minimal; eventhough rods outnumber cones, most mammalian retinas have8 to 10 cone-driven neurons for every cell associated primari-ly with the rod pathway.

The existence of multiple subclasses of cone-driven bipo-lar cells (‘cone bipolars’) was initially predicted on structuraland molecular grounds11,16,17. First, bipolar cells branch at dif-ferent levels of the inner plexiform layer18, which containprocesses of different types of amacrine and ganglion cells.Some ganglion cell types have dendrites confined mainly tolevel 1 of the inner plexiform layer, others to level 2, and so on.The inner plexiform layer, named as though it formed a sin-gle, tangled ‘plexus,’ is in fact an ordered stack of synapticplanes, more like a club sandwich than a plate of spaghetti.Specific bipolar cells make their synapses within specific planes,

review


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



and this confines their possiblesynaptic partners to cells withprocesses that occupy thosesame levels. Second, differenttypes of bipolar cells have dif-ferent numbers and distribu-tions of synapses, without agradation of intermediate formsbetween the types. The conclu-sion reflects more than neu-roanatomical anecdote; a formalcluster analysis showed thatcone bipolars segregate into dis-crete groups based on synapsenumber and distribution16,19.Third, individual bipolar celltypes have characteristic sets ofneurotransmitter receptors andcalcium-binding proteins20–22.These molecular distinctionsreflect different modes of intra-cellular signaling and differenttypes of excitatory and inhibito-ry inputs from other retinalneurons, either at their inputsfrom cones or from amacrinecells that synapse on their axonterminals. At the cone synapses,different glutamate receptors arepresent. At their axon terminals,different bipolar cells canreceive inhibitory glycinergic orGABAergic input via one of twodifferent kinds of GABA recep-tors. The different receptors andtheir channels have differing affinities and rates of activationand inactivation, which give the cells different postsynapticresponsiveness22–25.

How are these differences manifested physiologically? First,the output of the cone photoreceptors is separated into ON andOFF signals (Fig. 2b). All cone synapses release glutamate, butbipolar cell types respond to glutamate differently. Some bipo-lar cells have ionotropic glutamate receptors: glutamate opens acation channel, and the cell depolarizes. Other bipolar cells havea sign-inverting synapse mediated by metabotropic glutamatereceptors, mainly mGluR6; these bipolar cells hyperpolarize inresponse to glutamate26,27. As it happens, photoreceptor cellswork ‘backward’ (they hyperpolarize when excited by light,causing their synapses to release less glutamate), but the ensu-ing series of sign-reversals is not important for present pur-poses. When the retina is stimulated by light, one type of bipolarcell hyperpolarizes, and the other type depolarizes. OFF andON bipolar cells occur in approximately equal numbers. Thedistinction, created at the first retinal synapse, is propagatedthroughout the visual system.

The classes of ON and OFF bipolars are each further subdivid-ed; there are three to five distinct types of ON and three to five typesof OFF bipolars (Figs. 2c and 3). The purpose of the subdivisionis, at least in part, to provide separate channels for high-frequency(transient) and low-frequency (sustained) information. Thus, thereare separate ON-transient, ON-sustained, OFF-transient and OFF-sustained bipolar cells28–30. An elegant series of experiments showsthat the distinction is caused by different glutamate receptors on

the respective OFF bipolar cells; they recover from desensitizationquickly in the transient cells and more slowly in the sustained cells31.

An often-cited reason for splitting the output of the cones intoseparate temporal channels is to expand the overall bandwidth ofthe system. However, this would imply that the frequency band-width present at the output of a cone is too broad for transmis-sion through the cone-to-bipolar synapse, which is uncertain giventhe many modes of synaptic transmission available. An alterna-tive is that fractionating the temporal domain facilitates the cre-ation of temporally distinct types of ganglion cells (Fig. 4).

An important point here is that there are no dedicatedcones—cones that provide input, say, only to ON bipolars oronly to OFF bipolars (as shown for simplicity in Fig. 2).Instead, the output of each cone is tapped by several bipolarcell types to provide many parallel channels, each communi-cating a different version of the cone’s output to the inner reti-na (Figs. 3, 4 and 6).

The foundations of color visionThe bipolar cells discussed so far are not chromatically selective,and this would prevent the retina from discriminating amongwavelengths. A single type of cone, no matter how narrow itsspectral tuning, cannot create color vision. A cone’s synaptic out-put is a single signal, which can vary only in magnitude. For thatreason, a cone’s signal to the brain is inevitably ambiguous; thereare many combinations of wavelength and intensity that willevoke the same output from the cone. To specify the wavelengthof a stimulus, the outputs of at least two cones must be compared.

review

Fig. 1. The major cell types of a typical mammalian retina. From the top row to the bottom, photoreceptors,horizontal cells, bipolar cells, amacrine cells and ganglion cells. Amacrine cells, the most diverse class, havebeen studied most systematically in the rabbit3,4, and the illustration is based primarily on work in the rabbit.Most of the cells are also seen in a variety of mammalian species. The bipolar cells are from work in the rat39;similar ones have been observed in the rabbit, cat16 and monkey17. For steric reasons, only a subset of thewide-field amacrine cells is shown.©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


This combination of a short-wavelength cone and one ormore long-wavelength cones is avirtually universal feature ofmammalian retinas14. At onetime, many mammals werethought to lack color vision, and

indeed an animal with only these two visual pigments is adichromat—in everyday language, red–green ‘color blind.’ Butthe phrase is misleading; the distance between the peak sensi-tivities of the short and long opsins spans the wavelengthsreflected by important objects in the natural world, and an ani-mal with only those opsins has a strong form of color vision.If any doubt exists on this point, one should remember thatroughly 5% of humans inherit this form of dichromacy, butmany learn of it only during adulthood, when first confront-ed by tests designed to reveal variations in color vision.

The pathway from rods to ganglion cellsMost amacrine cells and all ganglion cells receive their main bipo-lar cell synapses from cone bipolars, but retinas work in starlightas well as daylight, and this range is created by a division of laborbetween cones (for bright light) and rods (for dim light). Signalsoriginating in rod photoreceptors reach the retinal ganglion cellsvia an indirect route using as its final path the axon terminals ofthe cone bipolar cells34–37.

That a single set of ganglion cells is used for both starlightand sunlight represents an obvious efficiency, long known fromelectrophysiological findings. However, it was not obvious a pri-

Fig. 3. The connections with cones and axonalstratification of different types of bipolar cells.Five different types of bipolar cells are illus-trated. Two of them are diffuse (chromaticallynonselective) ON bipolar cells terminating inthe inner half of the inner plexiform layer. Twoare diffuse OFF bipolar cells terminating in theouter half. Each samples indiscriminately fromthe spectral classes of cones. The blue conebipolar, however, contacts only blue cones andthus is spectrally tuned to short wavelengths.Within the ON or OFF sublayer, axons of thebipolar cells terminate at different levels, indi-cating that they contact different sets of postsy-naptic partners. After refs. 9 and 17.

Fig. 2. The bipolar cell pathways ofmammalian retinas, assembled fromindividual components. This diagramis intended to emphasize the overallorganization of the parallel channels,and much detail is omitted. Many pri-mate retinas have midget bipolar andganglion cells, but only a few have aseparate red and green channels.Rods are not as clumped as would besuggested here. For visual clarity,cones are shown contacting only asingle bipolar cell each; in fact, allcones contact several bipolar cells, asshown in Figs. 3, 4 and 6. For thedetailed synaptology of the rod path-way, see refs. 36, 37, 125.

Early in evolution, two cone opsins diverged, one with max-imal absorption at long wavelengths and one with maximalabsorption at short wavelengths12–14. Because an individual conecontains only a single spectral type of opsin, this creates two typesof cones, one reporting on long wavelengths and one on short;by comparing their outputs, the retina can create a single signalthat reflects the spectral composition of the stimulus.

The short-wavelength-sensitive cone, familiarly termed the‘blue cone,’ occupies a distinct and simple position in the arrayof retinal circuitry: blue cones synapse on their own specializedtype of bipolar cell, which in turn synapses on a dedicated class ofretinal ganglion cells32,33. Blue cones generally make up less than15% of all cones. The retina thus contains many long-wavelengthcones, which communicate to ganglion cells via a variety of bipo-lar cells, a single type of blue cone, and a single type of blue cone-driven bipolar cell (Figs. 2d and 3).

The synaptic connections of the inner retina are arrangedso that the outputs of some ganglion cells compare the respons-es of the blue cones with those of the long-wavelength cones.For example, the ganglion cell may be excited by short-wave-length stimuli and inhibited by long wavelengths. This repre-sents an economy; a single signal tells the brain where alongthe spectrum from blue to yellow thestimulus lies.

review


a b c d

e f

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


The retina of a macaque monkey con-tains approximately 1,500,000 retinal gan-glion cells; a cat, 160,000; a rabbit, 380,000(refs. 1, 9, 42) . Around 70% of the gan-glion cells of the monkey’s retina are midgetcells. They have a simple center–surroundorganization with linear spatial summationin the receptive field center. Associated with

midget ganglion cells is a special midget bipolar cell. In thefovea, an individual ganglion cell receives direct input fromonly a single cone. The fundamental advantage offered by amidget system is a high sampling density, which enables greatspatial resolution. In the central fovea, the spatial resolutionof the entire system—photoreceptors, bipolars and ganglioncells—is limited only by the cone packing density43.

In humans and some species of monkey, gene duplicationfollowed by mutations affecting a few amino acids caused thelong-wavelength opsin present in all mammals to evolve intotwo closely related opsins with slightly different absorptionmaxima44,45. Such retinas thus contain the widely conservedblue cone (with its specialized bipolar and ganglion cells), along-wavelength ‘green’ cone and a slightly different long-wave-length ‘red’ cone. This does not change the fundamental orga-nization of color vision; it simply creates better colordiscrimination between long wavelengths.

How the output of red and green cones is transmitted to thecentral visual system is a matter of controversy. The majorityopinion is that it is transmitted via the midget system10,11,46,47.Midget bipolar and ganglion cells automatically have the spec-tral sensitivity of the single cone from which they receive input, sothat the existence of the midget system perforce creates separatechannels for the two longer wavelengths. A minority view holdsthat there is an as-yet-undiscovered ganglion cell, analogous inits circuitry to the blue/yellow ganglion cell, that compares redand green wavelengths48.

Two types of horizontal cellsAll rods and cones receive feedback from horizontal cells, butthese cells are a numerically small proportion of the retina’sinterneurons, generally less than 5% of cells of the innernuclear layer2,38,40. In most mammals, there are two morpho-logically distinct types of horizontal cells49–52. (Mice and ratshave only one.) In monkeys, these have different numbers ofsynapses with different types of cones. The reason for this bias-ing is not yet certain; it may involve chromatic opponency inthe red–green system. Traditionally, horizontal cells are said toenhance contrast between adjacent light and dark regions. Exci-tation of a central cone causes feedback inhibition of both theexcited cone and a ring of neighboring ones. Because eachcone—both the central one and its neighbors—transmits a sig-

Fig. 4. How transient (high-pass) and sustained(low-pass) bipolar cells decompose the output ofa cone. The resulting high- and low-frequencychannels can contact narrowly stratified ganglioncells (a), in which case the two frequency bandsare transmitted via separate, parallel channels tothe brain. Bottom, a more broadly stratified gan-glion cell (such as a beta cell) receives input fromboth types of bipolar cells123. Such a ganglion cell(b) has a broadband response. Many such combi-nations are possible, as are many permutations ofinput from amacrine cells.


ori that rod-driven information would reach the ganglion cellsby an indirect path. Furthermore, rod photoreceptors far out-number cones in most mammalian retinas; it was a surprise tolearn, when quantitative methods became available, that conebipolars outnumber rod bipolars in all but a few mammalianretinas2,38. The reason is that more rods converge onto a singlerod bipolar than cones onto cone bipolars; the rod system tradesacuity for sensitivity, and the circuitry associated with rods issimpler than that of cones (Fig. 2e).

Because rods evolved after cones, the likely scenario is thatthe rod circuitry was grafted onto the cone pathways. Only onekind of rod photoreceptor exists, and rods drive only a single typeof bipolar cell. It synapses on a specialized amacrine cell, termedAII, which then transmits the output of rod bipolar cells to gan-glion cells. This occurs largely via synapses (chemical or gap junc-tional) by AII onto axon terminals of cone bipolar cells, whichthen excite the ganglion cells.

It may seem strange that rod bipolar cells would not simplydrive retinal ganglion cells directly, but seems less strange whenone appreciates the complexity of the pre-existing inner retinalcircuitry of the cone pathways. By synapsing on the axon of thecone bipolar cell, the rod pathway gains access to the elaboratecircuitry of the cone pathway, including its associated amacrinecircuitry. For example, the directionally selective type of ganglioncell retains its function in very dim light, even though it receivesno direct synapses from the rod bipolar cells. The rod system pig-gybacks on the cone circuitry rather than re-inventing it.

Added complexities in the primate retinaAt one time, primate retinas were thought to be somehow sim-pler than those of lower mammals, because recordings fromthe central retina of monkeys show mainly a simple type ofcenter–surround ganglion cell physiology; complex propertieslike direction selectivity are statistically rare. However, the rel-ative conservation of bipolar and amacrine cell types in mon-keys and other mammals is now well documented7,17,22,38–41.Furthermore, such a conclusion would imply, remarkably, thatretinal circuitry evolved over millennia was discarded. Instead,to the already existing retina were added three specializations:an additional chromatic class of cone, a rod-free fovea, and ahuge number of small bipolar and ganglion cells, the so-calledmidget system (Fig. 2f).

review

a b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


nal to the inner retina, the upshot is that a small stimulusexcites those ganglion cells that lie directly under the stimu-lus, but inhibits neighboring ganglion cells. This is the classical‘center–surround’ organization, in which a ganglion cell isexcited or inhibited by stimuli falling in its receptive field cen-ter, whereas stimulation of the surrounding region has anopposite effect.

An alternate formulation of the same facts is that horizontalcells adjust the system’s response to the overall level of illumina-tion—they measure illumination across a broad region and sub-tract it from the signal that is transmitted to the inner retinaabout a local image8. In effect, this reduces redundancy in thesignal transmitted to the inner retina. The mean luminance acrossa large region of retina is shared by many cones and contains lit-tle information. When a local stimulus occurs, it exceeds or fallsbelow the mean; the occurrence of that local event is the mainsignal transmitted to the inner retina.

Rods receive a separate type of horizontal cell feedback; this isaccomplished by a specialization of one of the horizontal cells(the b/H2 type) that contacts cones. An axonal process of thishorizontal cell contacts the rods, but does it far enough awayfrom the horizontal cell’s soma that the axonal arbor is electro-tonically isolated53. The rod feedback system is thus isolated fromthe cone feedback system, sensibly because the ranges of bright-ness covered by rods and cones are so enormously different. Thismay be another consequence of the late evolution of rods. Itallows the rods to have an independent horizontal cell feedback,driven by rods and feeding back to rods, without the creation ofa third type of horizontal cell.

Twenty-nine types of amacrine cellsAll retinal ganglion cells receive input from cone bipolar cells,but direct synapses from bipolar cells are a minority of allsynapses on the ganglion cells; most are from amacrinecells54–56. The exact fraction varies among different functionaltypes of ganglion cells, ranging from roughly 70% for alphacells (large, movement-sensitive ganglion cells found in mostmammals) to 50% for the midget ganglion cells located in themonkey central fovea. Amacrine cells also make inhibitorysynapses on the axon terminals of bipolar cells, thus control-ling their output to ganglion cells. In contrast to horizontal cells,which have a single broad role, amacrine cells have dedicatedfunctions—they carry out narrow tasks concerned with shapingand control of ganglion cell responses.

Traditional presentations of the retina underweight theimportance of amacrine cells, which are sometimes illustrat-ed in a 1:1 ratio with horizontal cells57,58. They in fact out-number horizontal cells by amounts that range from 4:1 to 10:1(depending on the species) and can outnumber ganglion cellsby 15 to 1 (refs. 2, 38, 40) . How can this complexity be under-stood? A first impulse is to deny that it exists—perhaps the tax-onomy has been made artificially complex, or cells that lookdifferent actually have identical functions? It turns out thatneither of these is tenable. The different amacrine cells havedistinct pre- and postsynaptic partners, contain a variety ofneurotransmitters, survey narrow areas of the visual scene orbroad ones, branch within one level of the inner synaptic layeror communicate among many3,4. Both their molecules andtheir form point to diverse functions.

Amacrine cells seem to account for correlated firing amongganglion cells. Shared input from a common amacrine cell willtend to make ganglion cells fire together; the cross-correlation isbroad if mediated by chemical synapses and narrower if mediat-

ed by gap junctions, known to couple amacrine and ganglioncells59. Correlated firing between ganglion cells has been pro-posed to represent a form of multiplexing, which could expandthe information-carrying capacity of the optic nerve60,61.

Those amacrine cells with functions that are more preciselyunderstood do remarkably specific jobs. The dopaminergicamacrine cells globally adjust the retina’s responsiveness underbright or dim light62–64. They are numerically sparse (9000 cellsin a rabbit retina that has 4,500,000 amacrine and 380,000 gan-glion cells)65 and have wide-spreading arbors located in innerplexiform layer 1. Dopamine affects many elements of the reti-na’s circuitry; it alters the gap-junctional conductance betweenhorizontal cells and between amacrine cells66,67, potentiates theresponses of ionotropic glutamate receptors on bipolar cells,and ultimately affects the center–surround balance of ganglioncells68,69. Remarkably, retinal dopamine can even cause pigmentmigration in cells of the retinal pigment epithelium, a neigh-boring non-neural tissue70. In the latter case (and very likelysome of the former as well), this is mediated non-synaptically,via a diffuse, paracrine release of the neurotransmitter. Elegantexperiments using transgenically labeled amacrine cells in cul-ture show that the extrasynaptic release is controlled by spon-taneous action potentials in the absence of synaptic input andmodulated by inputs, presumably also paracrine ones, fromother retinal neurons71,72.

In contrast, the starburst amacrine cells seem to be narrowlyassociated with a particular computational circuit. They arborizein thin (2–4 µm) strata within the inner plexiform layer, wherethey make excitatory cholinergic synapses on certain retinal gan-glion cells, notably those particularly sensitive to moving stim-uli. By feedforward excitation and/or inhibition (these neuronsrelease both acetylcholine and GABA73), they are important fordirection selectivity74–76.

Ten to fifteen types of retinal ganglion cellsIt became possible to record from retinal ganglion cells beforemodern anatomical techniques were invented, and early ideas ofthis population were much influenced by electrophysiologicalresults, with their inherent sampling biases. These described twotypes of concentrically organized receptive fields, one with asmall, linearly summing receptive field center (X cell) and anoth-er with a large, non-linear responsive area (Y cell). Systematicanatomical studies now make it apparent that many other typesof ganglion cells exist. These are easily distinguished by theirbranching level, their dendritic arbor width (that is, the area ofthe visual field that they sample), and in many cases, their direct-ly recorded physiology77–80 (Figs. 1 and 5).

In all cases studied thus far, cells distinguished by structur-al criteria have turned out to have distinct physiologies. In thecat, the correspondence between X-cells and β, and Y-cells andα was established long ago, as was the analogous match betweenP and M, midget and parasol cells in the monkey17. Other celltypes were studied early in the rabbit, using direct recordingfrom the retina (where the problem of electrode selectivity islessened)81–85. A bistratified neuron is the famous ON–OFFdirection-selective cell. A similar but monostratified medium-field neuron is the ON-type direction-selective cell, which pro-jects to the accessory optic system and provides an error signalfor eye velocity in optokinetic nystagmus. An extremely small,monostratified neuron is the local edge detector described inclassic electrophysiological studies.

In the monkey, a small bistratified neuron is the blue ONcell, and a larger, sparser neuron is a blue OFF cell. In both the

review


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 5. The types of ganglion cellsidentified thus far in the retina ofthe cat. Ongoing work in the rab-bit and monkey confirms thisdiversity, and many of the cellsobserved are probably homologsof those seen in the cat. Courtesyof D. Berson77–80.


cat and monkey, a very large, very rare neuron has tonicresponses to light and projects to a pretectal nucleus; it seemsto control pupillary size. A similarly rare neuron projects to thecat suprachiasmatic nucleus, presumably to entrain circadianrhythms. Remarkably, this cell seems to be directly photosen-sitive (D.M. Berson, F.A. Dunn & M. Takao, Invest. Ophthal-mol. Vis. Sci. 42, S113, 2001)86.

The primate fovea, with its huge number of midget cells,seems to have been superimposed upon existing ganglion cellpopulations that were little changed during the primate’s evo-lution from earlier mammals. Some of these cells seem to cor-respond to neurons present in lower mammals and carry out‘vegetative’ functions, such as the control of pupil size and opto-kinetic responses. Evidence for autonomous subcortical path-ways that mediate these functions in the monkey is that bothsurvive combined lesions of the visual cortex and superior col-liculus87. It takes only a few neurons to measure the ambientlevel of illumination, which controls the pupillary aperture.There is no particular need for this number to increase as thetotal number of ganglion cells increases, and they end up as asmall fraction of the total cells. A monkey retina that has1,050,000 midget ganglion cells could comfortably ‘contain’ theganglion cell population of an entire cat or rabbit retina withinits remaining 450,000 cells11.

For this purely statistical reason, non-midget, non-parasolcells in the monkey have largely been ignored. However, mod-ern methods, notably, visually guided microinjection88,89, arenow providing an increasingly clear anatomical view of the otherganglion cells of the monkey90–93. There is some reason to sus-pect that the geniculostriate system receives non-midget, non-parasol types of information, and learning more about these cells’physiology seems important (see below).

Visual function: new certainties and new questionsA reward of structural studies is the level of certainty that theirhard-won conclusions provide. The demonstration that X and

Y cells are anatomically distinctentities helped still an acrimo-nious taxonomic controversyamong electrophysiologists.Psychophysicists had long sus-pected that vision along theblue–yellow axis is differentfrom vision along the red andgreen axis, which is given aconcrete basis in the sparsenessof blue cones and their bipolarcells. An exact synapticwiring33,47,91,94 now underpinsthe receptive field of the blue-ON ganglion cell, accuratelypredicted 35 years ago95.

A different kind of contri-bution comes from the quan-

titative nature of such studies. Human visual acuity, forexample, is now known to precisely match the packing densityof the foveal cones43,96. This contribution is sometimes takenfor granted, but should not be; our concept of central visualprocessing would be different if primate M cells were not 8%of all ganglion cells, as shown anatomically, but 30–50%, aswould be concluded from their encounter frequency in elec-trophysiological experiments. As modeling of higher visualprocesses becomes more precise, knowledge of such physicalparameters becomes increasingly useful.

Structural results also raise new questions; the cell popula-tions of the retina hint at unsuspected subtleties in the retina’sinput–output relationships, some of which must have conse-quences for vision. For example, what are the remaining phys-iological types of retinal ganglion cells, and how do theycontribute to behavior? The question here is the physiologicalresponse properties of the non-concentric (X and Y, M and P)types of cells and their function in the central structures towhich they project. For subcortically projecting cells, thoseroles may be very sophisticated. The ON directionally selec-tive cell of the rabbit, for example, projects to the accessoryoptic system and drives optokinetic responses85,97; the baroquemorphologies of non-midget, non-parasol cells that projectsubcortically in the monkey suggest equally subtle physiolo-gies. These questions should be answerable by in vitro record-ing followed by microinjection89,92.

We need to complete our understanding of the synaptic basisof color vision. Here our colleagues who study higher visual cen-ters are struggling; the cortical coding of color has been a tan-gled subject98–100. If the red–green axis is coded in the retina bya distinct, dedicated set of retinal ganglion cells, then one mightexpect a single cortical mechanism to code for color along boththe red–green and blue–yellow axes. If red and green are trans-mitted separately, via the late-evolving midget system, highercenters may have anatomically and/or computationally inde-pendent ways of handling the two axes.

review©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


In the lateral geniculate body of the monkey, several spe-cialized types of cells project to the K (koniocellular) layers ofthe lateral geniculate body101,102. There are hints of other typesof cells mixed among the cells of the magnocellular and par-vocellular layers, and history teaches that it is possible to misseven a sizable class of cells when using metal microelec-trodes103. Even though the remaining cells may be few in num-ber, they are not necessarily unimportant for vision. Theblue-ON ganglion cells make up less than 6% of all ganglioncells in the monkey but are a fundamental basis of primatecolor vision. Similarly, parasol cells make up 8% of all ganglioncells, yet are thought to be the source of a major stream of cor-tical information flow. Newly expanded techniques for record-ing from ganglion cells backfilled from specific central targets(D.M. Dacey et al., Invest. Ophthalmol. Vis. Sci. 42, 114, 2001)should soon provide a more complete description of the infor-mation that enters the geniculostriate system.

Microstructure within the receptive field centerA surprise when the complete array of amacrine cells was revealedwas the plethora of narrow-field amacrines, which make upalmost 50% of amacrine cells in the rabbit, rat and monkey andthus represent 20–30% of neurons in the inner nuclearlayer3,4,104,105. How do they affect ganglion cell physiology?

In addition to amacrine AII (a link between the rod system andthe ganglion cells), there are, in the mid-periphery of the rabbitretina, 11 types of amacrine cells with dendritic arbors less than100 µm in diameter. In the same region, the diameters of retinalganglion cell arbors range from 200 to 1000 µm. This means thatmany narrow-field amacrine cells exist within the dendritic field,and thus the receptive field center, of most ganglion cells.

If nothing else, the finding invalidates the textbook gener-alization that the function of amacrine cells is to carry infor-mation laterally across the retina; these cells are scarcely morelaterally conducting than are the bipolar cells. It also suggeststhat more information processing occurs within the center ofthe ganglion cell’s receptive field than is usually credited.Indeed, many narrow-field amacrine cells of each of severaltypes tile the retina within each ganglion cell’s receptive field.They must affect the transfer of information through the reti-na, with a spatial resolution similar to that of the bipolar cells,but the nature of the transformation remains to be learned.

A likely possibility is that some of the narrow-fieldamacrines are involved in contrast gain control106, which maycause, among other things, a ‘predictive’ response of ganglioncells to moving stimuli107. However, it is not at all apparentwhy a conceptually simple function such as a negative, con-trast-driven feedback would require 11 different kinds ofamacrine cells. Other narrow-field amacrine cells carry outtemporal sharpening; amacrine AII generates regenerative cur-rents, which give the leading edge of its response to light a fastrise time108,109. Many narrow-field cells communicate amongseveral layers of the inner plexiform layer and thus carry out‘vertical inhibition’110, named by analogy to the familiar lat-eral inhibition mediated by horizontal cells.

Too many wide-field amacrinesWhy there are so many wide-field amacrine cells? The rabbit hasfive kinds of medium-field amacrine cells (dendritic arbors ~175µm) and at least ten wide-field types3,4. The latter can have den-drites that run for millimeters across the retinal surface111,112,suggesting that long-range lateral integration, spreading far acrossthe retina, may be more important than has been recognized113.Some of the cells have sparse, relatively simple arbors. Othershave garden-variety dendritic arbors but also have axon-likeprocesses that can span 5 to 10 mm across the retina’s surface.Recording from two types in mammals reveals that they havereceptive fields coterminous with their dendritic arbors and thatthey generate action potentials, which should conduct activityfar from the main dendritic arbor114,115.

Hints that activity spreads over long trans-retinal distanceswere evident long ago from the ‘periphery effect,’ a simpledemonstration that stimulation outside the classical receptivefield can change retinal sensitivity within the receptive field. Thereis also a recent report of oscillatory 40-Hz activity correlated forup to 10 mm across the cat’s retina116,117. However, the exactfunction of these lateral effects is not known, nor is the need formultiple types of wide-field amacrine cells explained. Perhapslateral conduction is required in viewing natural scenes, whichcontain wider ranges of contrast and more complex trans-reti-nal motion than the usual laboratory stimuli.

Contrast gain control is a critical ‘normalization’ functionat the front end of the visual system, and there is direct evidencefor both narrow and wide forms of it. Recently, two studies eval-

review


Fig. 6. The fundamental signal-carrying pathways of a generic mammalian retina, reduced to a conceptual minimum. Each type of bipolar cell (black)transmits a different type of information to the inner retina. The information that it transmits is determined by the bandwidth of the cones that it con-tacts, the number and type of those cones, the transfer function of the cone to bipolar synapse, and its interplay with amacrine cells. This is a minimalrepresentation of the amacrine cells, which also include wide-field cells and which have synaptic contacts among each other. The different types ofbipolar cells are contacted by distinct types of amacrine cells, in a variety of synaptic arrangements. These converge upon the retinal ganglion cells, inwhich specific combinations of bipolar and amacrine inputs create many functional types of ganglion cells.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



uated temporal contrast adaptation using reverse correlationand flickering checkerboards. They produced evidence for botha mechanism that works on a large spatial scale118 and one thatis extremely local—operating on a scale, in the rabbit, ofapproximately 100 µm, a fraction of the size of the receptivefield center for many ganglion cells119. There is some evidencethat the rate of adaptation is different for different-sized stim-uli. This suggests the existence of multiple, independent formsof contrast adaptation. One form of temporal contrast adapta-tion seems to operate entirely within the bipolar cells them-selves, because it persists in the presence of pharmacologicalagents that should block amacrine cell function. For larger stim-uli, the array of amacrine cells may contain several mechanismsby which the responsiveness of the retina is tuned to the char-acteristics of the visual environment.

What are the fundamental channels of vision?A final question concerns events at the heart of the retina’s design.What are the separate filters represented by the different types ofbipolar cells, and how are they reflected in the information trans-mitted centrally?

The diffuse bipolar cells represent as-yet-undeciphered par-allel channels by which the retina parses the visual input (Fig.6). In some cases, the operation performed by bipolar cells isobvious. The blue bipolar cell acts as a spectral filter tuned towavelengths peaking at about 420 nm, and to moderate spatialfrequencies. The red and green midget bipolars of primates aretuned to their particular wavelengths and to higher spatial fre-quencies. Roughly half the diffuse bipolar cells carry out a signinversion creating the ON and OFF classes of response. Withineach broad class (ON or OFF) of diffuse bipolars, though, thereare at least four specific subtypes of bipolar cells of uncertaintuning. We learn their approximate spatial tuning from theirdendritic spread, but we have only hints from their neurotrans-mitter receptors and channels about their dynamic properties.

From early studies in cold-blooded vertebrates28,29,120, andmore recent studies in mammals, bipolar cells were found to comein sustained (low-pass) and transient (high-pass) varieties. Resultsfrom salamander retina30,121 point to even greater diversity, andthis is also clear in the existing recordings from cone bipolar cellsof mammals25,31,122. Although these experiments are technicallydifficult, a critically important challenge to physiologists is to pre-cisely characterize the behavior of each channel.

Another challenge is to learn how the bipolar channels arerecombined at the level of the ganglion cell (Figs. 4 and 6). Here,modeling techniques may be useful. The central problem is tounderstand, especially in the temporal domain, how the finalresponse of a ganglion cells is created from one or several bipolarcell inputs123. It is unlikely that anyone will soon record simul-taneously from one ganglion cell and two bipolar cells; modelsor simulations may clarify our thinking in this realm.

A higher-order question is how the parallel channels createdby bipolar cells are reflected in the central visual system. The firstlimiting event for scotopic vision is the capture of photons by thecone mosaic. Even though cones’ output is much transformedlater—within the retina and higher in the visual system—vision’soverall sensitivity, chromatic selectivity and resolution dependexactly on the number and spacing of the different types ofcones43,124. The second limiting event in vision is the transmis-sion of signals from the cones to the inner retina by the bipolarcells. Bipolar cells are the mandatory link between cones (or rods)and the rest of the visual system—all visual information mustflow through them. Even though these signals, too, are later

shaped and recombined, it is inescapable that the separate chan-nels inherent in bipolar cell diversity represent fundamentals ofvision, basic building blocks from which all further codings areconstructed. In principle, we should eventually be able to decon-volve the outputs of individual bipolar channels from signalsencountered even deep within the central visual system.

ACKNOWLEDGEMENTSR. Rockhill made the illustrations. B. Boycott and P. Sterling made comments on

the manuscript. The author is a Senior Investigator of Research to Prevent

Blindness.

RECEIVED 15 JUNE 2000; ACCEPTED 19 JULY 2001

1. Vaney, D. I. A quantitative comparison between the ganglion cell populationsand axonal outflows of the visual streak and periphery of the rabbit retina. J. Comp. Neurol. 189, 215–233 (1980).

2. Martin, P. R. & Grünert, U. Spatial density and immunoreactivity of bipolarcells in the macaque monkey retina. J. Comp. Neurol. 323, 269–287 (1992).

3. MacNeil, M. A. & Masland, R. H. Extreme diversity among amacrine cells:implications for function. Neuron 20, 971–982 (1998).

4. MacNeil, M. A., Heussy, J. K., Dacheux, R., Raviola, E. & Masland, R. H. Theshapes and numbers of amacrine cells: matching of photofilled with Golgi-stained cells in the rabbit retina and comparison with other mammalianspecies. J. Comp. Neurol. 413, 305–326 (1999).

5. Masland, R. H. Neuronal diversity in the retina. Curr. Opin. Neurobiol. 11431–436 (2001).

6. Vaney, D. I. in Progress in Retinal Research (eds. Osborne, N. & Chader, G.)49–100 (Pergamon, New York, 1990).

7. Wässle, H. & Boycott, B. B. Functional architecture of the mammalian retina.Physiol. Rev. 71, 447–480 (1991).

8. Sterling, P. in The Synaptic Organization of the Brain Vol. 4 (ed. Shepherd, G. M.), 205–253 (Oxford Univ. Press, New York, 1998).

9. Rodieck, R.W. The First Steps in Seeing (Sinauer Associates, Sunderland,Massachusetts, 1998).

10. Nathans, J. The evolution and physiology of human color vision: insights frommolecular genetic studies of visual pigments. Neuron 24, 299–312 (1999).

11. Boycott, B. B. & Wässle, H. Parallel processing in the mammalian retina: TheProctor Lecture. Invest. Ophthalmol. Vis. Sci. 40, 1313–1327 (1999).

12. Okano, T., Kojima, D., Fukada, Y. & Shichida, Y. Primary structures ofchicken cone visual pigments: vertebrate rhodopsins have evolved out of conevisual pigments. Proc. Natl. Acad. Sci. USA 89, 5932–5936 (1992).

13. Johnson, R. L. et al. Cloning and expression of goldfish opsin sequences.Biochemistry 32, 208–214 (1993).

14. Jacobs, G. H. The distribution and nature of colour vision among themammals. Biol. Rev. 68, 413–471 (1993).

15. Westheimer, G. The maxwellian view. Vision Res. 6, 669–682 (1966).16. Cohen, E. & Sterling, P. Demonstration of cell types among cone bipolar

neurons of the cat retina. Phil. Trans. R. Soc. Lond. B Biol. Sci. 330, 305–321(1990).

17. Boycott, B. B. & Wässle, H. Morphological classification of bipolar cells of theprimate retina. Eur. J. Neurosci. 3, 1069–1088 (1991).

18. Cajal, S. R. The Structure of the Retina (Thomas, Springfield, 1972).19. McGuire, B. A., Stevens, J. K. & Sterling, P. Microcircuitry of bipolar cells in

rat retina. J. Neurosci. 12, 2920–2938 (1984).20. Grünert, U., Martin, P. R. & Wässle, H. Immunocytochemical analysis of

bipolar cells in the macaque monkey retina. J. Comp. Neurol. 348, 607–627(1994).

21. Hartveit, E. et al. Localization and developmental expression of the NMDAreceptor subunit NR2A in the mammalian retina. J. Comp. Neurol. 348,570–582 (1994).

22. Haverkamp, S. & Wässle, H. Immunocytochemical analysis of the mouseretina. J. Comp. Neurol. 424, 1–23 (2000).

23. Greferath, U. et al. GABAA receptor subunits have differential distributions inthe rat retina: in situ hybridization and immunohistochemistry. J. Comp.Neurol. 353, 553–571 (1995).

24. Euler, T. & Wässle, H. Different contributions of GABAA and GABACreceptors to rod and cone bipolar cells in a rat retinal slice preparation. J. Neurophysiol. 79, 1384–1395 (1998).

25. Euler, T. & Masland, R. H. Light-evoked responses of bipolar cells in amammalian retina. J. Neurophysiol. 83, 1817–1829 (2000).

26. Nawy, S. & Jahr, C. E. Suppression by glutamate of cGMP-activatedconductance in retinal bipolar cells. Nature 346, 269–271 (1990).

27. Nawy, S. & Jahr, C. E. cGMP-gated conductance in retinal bipolar cells issuppressed by the photoreceptor transmitter. Neuron 7, 677–683 (1991).

28. Kaneko, A. Physiological and morphological identification of horizontal,bipolar and amacrine cells in goldfish retina. J. Physiol. (Lond.) 207, 623–633(1970).

29. Werblin, F. S. & Dowling, J. E. Organization of the retina of the mudpuppy,

review©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


necturus maculosus II. Intracellular recording. J. Neurophysiol. 32, 339–355(1969).

30. Awatramani, G. B. & Slaughter, M. M. Origin of transient and sustainedresponses in ganglion cells of the retina. J. Neurosci. 20, 7087–7095 (2000).

31. DeVries, S. H. Bipolar cells use kainate and AMPA receptors to filter visualinformation into separate channels. Neuron 28, 847–856 (2000).

32. Kouyama, N. & Marshak, D. W. Bipolar cells specific for blue cones in themacaque retina. J. Neurosci. 12, 1233–1252 (1992).

33. Dacey, D. M. & Lee, B. B. The ‘blue-on’ opponent pathway in primate retinaoriginates from a distinct bistratified ganglion cell type. Nature 367, 731–735(1994).

34. Famiglietti, E. V. Jr. & Kolb, H. A bistratified amacrine cell and synaptic circuitryin the inner plexiform layer of the retina. Brain Res. 84, 293–300 (1975).

35. Smith, R. G., Freed, M. A. & Sterling, P. Microcircuitry of the dark-adaptedcat retina: functional architecture of the rod–cone network. J. Neurosci. 6,3505–3517 (1986).

36. Strettoi, E., Dacheux, R. F. & Raviola, E. Synaptic connections of rod bipolarcells in the inner plexiform layer of the rabbit retina. J. Comp. Neurol. 295,449–466 (1990).

37. Strettoi, E., Raviola, E. & Dacheux, R. F. Synaptic connections of the narrow-field, bistratified rod amacrine cell (AII) in the rabbit retina. J. Comp. Neurol.325, 152–168 (1992).

38. Strettoi, E. & Masland, R. H. The organization of the inner nuclear layer ofthe rabbit retina. J. Neurosci. 15, 875–888 (1995).

39. Euler, T. & Wässle, H. Immunocytochemical identification of cone bipolarcells in the rat retina. J. Comp. Neurol. 361, 461–478 (1995).

40. Jeon, C.-J., Strettoi, E. & Masland, R. H. The major cell populations of themouse retina. J. Neurosci. 18, 8936–8946 (1998).

41. Masland, R. H. Processing and encoding of visual information in the retina.Curr. Opin. Neurobiol. 6, 467–474 (1996).

42. Williams, R. W., Cavada, C. & Reinoso-Suarez, F. Rapid evolution of the visualsystem: a cellular assay of the retina and dorsal lateral geniculate nucleus of theSpanish wildcat and the domestic cat. J. Neurosci. 13, 208–228 (1993).

43. Williams, D. R. Seeing through the photoreceptor mosaic. Trends Neurosci. 9193–198 (1986).

44. Nathans, J., Piantanida, T. P., Eddy, R. L., Shows, T. B. & Hogness, D. S.Molecular genetics of inherited variation in human color vision. Science 232,203–210 (1986).

45. Nathans, J., Thomas, D. & Hogness, D. S. Molecular genetics of human colorvision: the genes encoding blue, green, and red pigments. Science 232,193–202 (1986).

46. Mollon, J. D. in Evolution of the Eye and Visual System (eds. Cronly-Dillon, J. R. & Gregory, R. L.) 306–319 (CRC Press, Boca Raton, Florida, 1991).

47. Dacey, D. M. Parallel pathways for spectral coding in primate retina. Annu.Rev. Neurosci. 23, 743–775 (2000).

48. Calkins, D. J. & Sterling, P. Evidence that circuits for spatial and color visionsegregate at the first retinal synapse. Neuron 24, 313–321 (1999).

49. Boycott, B. B., Hopkins, J. M. & Sperling, H. G. Cone connections of thehorizontal cells of the rhesus monkey’s retina. Proc. R. Soc. Lond. B Biol. Sci.229, 345–379 (1987).

50. Hack, I. & Peichl, L. Horizontal cells of the rabbit retina are non-selectivelyconnected to the cones. Eur. J. Neurosci. 11, 2261–2274 (1999).

51. Peichl, L., Sandmann, D. & Boycott, B. B. in Development and Organization ofthe Retina (eds. Chalupa, L. M. & Finlay, B. L.) 147–172 (Plenum, New York,1998).

52. Dacey, D. M., Lee, B. B., Stafford, D. K., Pokorny, J. & Smith, V. C. Horizontalcells of the primate retina: cone specificity without spectral opponency.Science 271, 656–658 (1996).

53. Nelson, R., Lützow, A. V., Kolb, H. & Gouras, P. Horizontal cells in cat retinawith independent dendritic systems. Science 189, 137–139 (1975).

54. Freed, M. A. & Sterling, P. The ON-alpha ganglion cell of the cat retina and itspresynaptic cell types. J. Neurosci. 8, 2303–2320 (1988).

55. Calkins, D. J., Schein, S. J., Tsukamoto, Y. & Sterling, P. M and L cones inmacaque fovea connect to midget ganglion cells by different numbers ofexcitatory synapses. Nature 371, 70–72 (1994).

56. Jacoby, R., Stafford, D., Kouyama, N. & Marshak, D. Synaptic inputs to ONparasol ganglion cells in the primate retina. J. Neurosci. 16, 8041–8056 (1996).

57. Hubel, D. H. Eye, Brain and Vision (Freeman, New York, 1988).58. Nolte, J. The Human Brain (Mosby-Year Book, St. Louis, Missouri, 1988).59. Vaney, D. I. Patterns of neuronal coupling in the retina. Prog. Ret. Eye Res. 13,

301–355 (1994).60. Meister, M., Lagnado, L. & Baylor, D. A. Concerted signaling by retinal

ganglion cells. Science 270, 1207–1210 (1995).61. Meister, M. Multineuronal codes in retinal signaling. Proc. Natl. Acad. Sci.

USA 93, 609–614 (1996).62. Ehinger, B. Functional role of dopamine in the retina. Prog. Ret. Res. 2,

213–232 (1983).63. Dowling, J. E. Dopamine: a retinal neuromodulator? Trends Neurosci. 9,

236–240 (1986).64. Witkovsky, P. & Dearry, A. Functional roles of dopamine in the vertebrate

retina. Prog. Ret. Res. 11, 247–292 (1991).65. Tauchi, M., Madigan, N. M. & Masland, R. H. Shapes and distributions of the

catecholamine-accumulating neurons in the rabbit retina. J. Comp. Neurol.293, 178–189 (1990).

66. Hampson, E. C. G. M., Vaney, D. I. & Weiler, R. Dopaminergic modulation of

gap junction permeability between amacrine cells in mammalian retina. J. Neurosci. 12, 4911–4922 (1992).

67. Piccolino, M., Witkovsky, P. & Trimarchi, C. Dopaminergic mechanismsunderlying the reduction of electrical coupling between horizontal cells of theturtle retina induced by D-amphetamine, bicuculline, and veratridine. J. Neurosci. 7, 2273–2284 (1987).

68. Jensen, R. J. & Daw, N. W. Effects of dopamine antagonists on receptive fieldsof brisk cells and directionally selective cells in the rabbit retina. J. Neurosci. 4,2972–2985 (1984).

69. Jensen, R. J. & Daw, N. W. Effects of dopamine and its agonists andantagonists on the receptive field properties of ganglion cells in the rabbitretina. Neuroscience 17, 837–855 (1986).

70. Bruenner, U. & Burnside, B. Pigment granule migration in isolated cells ofthe teleost retinal pigment epithelium. Invest. Ophthalmol. Vis. Sci. 27,1634–1643 (1986).

71. Gustincich, S., Feigenspan, A., Wu, D. K., Koopman, L. J. & Raviola, E.Control of dopamine release in the retina: a transgenic approach to neuralnetworks. Neuron 18, 723–736 (1997).

72. Puopolo, M., Hochstetler, S. E., Gustincich, S., Wightman, R. M. & Raviola,E. Extrasynaptic release of dopamine in a retinal neuron: activity dependenceand transmitter modulation. Neuron 30, 211–225 (2001).

73. O’Malley, D. M., Sandell, J. H. & Masland, R. H. Co-release of acetylcholineand GABA by the starburst amacrine cells. J. Neurosci. 12, 1394–1408 (1992).

74. Masland, R. H. & Tauchi, M. The cholinergic amacrine cell. Trends Neurosci.9, 218–223 (1986).

75. He, S.-G. & Masland, R. H. Retinal direction selectivity after targeted laserablation of starburst amacrine cells. Nature 389, 378–382 (1997).

76. Yoshida, K. et al. A key role of starburst amacrine cells in originating retinaldirectional selectivity and optokinetic eye movement. Neuron 30, 771–780(2001).

77. Berson, D. M., Isayama, T. & Pu, M. The eta ganglion cell type of the catretina. J. Comp. Neurol. 408, 204–219 (1999).

78. Berson, D. M., Pu, M. & Famiglietti, E. V. The zeta cell: a new ganglion celltype in cat retina. J. Comp. Neurol. 399, 269–288 (1998).

79. Isayama, T., Berson, D. M. & Pu, M. Theta ganglion cell type of the cat retina.J. Comp. Neurol. 417, 32–48 (2000).

80. Pu, M., Berson, D. M. & Pan, T. Structure and function of retinal ganglioncells innervating the cat’s geniculate wing: an in vitro study. J. Neurosci. 14,4338–4358 (1994).

81. Barlow, H. B. & Levick, W. R. The mechanism of directionally selective unitsin rabbit’s retina. J. Physiol. (Lond.) 178, 477–504 (1965).

82. Levick, W. R. Receptive fields and trigger features of ganglion cells in thevisual streak of the rabbit’s retina. J. Physiol. (Lond.) 188, 285–307 (1967).

83. Oyster, C. W., Takahashi, E. & Collewijn, H. Direction selective retinalganglion cells and control of optokinetic nystagmus in the rabbit. Vision Res.12, 183–193 (1972).

84. Simpson, J. I. The accessory optic system. Annu. Rev. Neurosci. 7, 13–41(1984).

85. Oyster, C. W., Simpson, J. I., Takahashi, E. S. & Soodak, R. E. Retinal ganglioncells projecting to the rabbit accessory optic system. J. Comp. Neurol. 190,49–61 (1980).

86. Provencio, I. et al. A novel human opsin in the inner retina. J. Neurosci. 20,600–605 (2000).

87. Pasik, P. & Pasik, T. in The Oculomotor System (ed. Bender, M. B.) 40–80(Harper & Row, New York, 1964).

88. Tauchi, M. & Masland, R. H. Local order among the dendrites of an amacrinecell population. J. Neurosci. 5, 2494–2501 (1985).

89. Yang, G. & Masland, R. H. Direct visualization of the dendritic and receptivefields of directionally selective retinal ganglion cells. Science 258, 1949–1952(1992).

90. Rodieck, R. W. & Watanabe, M. Survey of the morphology of macaque retinalganglion cells that project to the pretectum, superior colliculus and theparvicellular laminae of the lateral geniculate nucleus. J. Comp. Neurol. 338,289–303 (1993).

91. Dacey, D. M. Morphology of a small-field bistratified ganglion cell type in themacaque and human retina. Vis. Neurosci. 10, 1081–1098 (1993).

92. Dacey, D. M. Physiology, morphology and spatial densities of identifiedganglion cell types in the primate retina. Ciba Found. Symp. 184, 12–34(1994).

93. Peterson, B. B. & Dacey, D. M. Morphology of wide-field, monostratifiedganglion cells of the human retina. Vis. Neurosci. 16, 107–120 (1999).

94. Calkins, D. J., Tsukamoto, Y. & Sterling, P. Microcircuitry and mosaic of ablue-yellow ganglion cell in the primate retina. J. Neurosci. 18, 3373–3385(1998).

95. Wiesel, T. N. & Hubel, D. H. Spatial and chromatic interactions in the lateralgeniculate body of the rhesus monkey. J. Neurophysiol. 29, 1115–1156 (1966).

96. Curcio, C. A., Sloan, K. R., Kalina, R. E. & Hendrickson, A. E. Humanphotoreceptor topography. J. Comp. Neurol. 292, 497–523 (1990).

97. Buhl, E. H. & Peichl, L. Morphology of rabbit retinal ganglion cells projectingto the medial terminal nucleus of the accessory optic system. J. Comp. Neurol.253, 163–174 (1986).

98. Ts’o, D. Y. & Gilbert, C. D. The organization of chromatic and spatialinteractions in the primate striate cortex. J. Neurosci. 8, 1712–1727 (1988).

99. Hubel, D. & Livingstone, M. Color puzzles. Cold Spring Harb. Symp. Quant.Biol. 55, 643–649 (1990).

review


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



100. Conway, B. R. Spatial structure of cone inputs to color cells in alertmacaque primary visual cortex (V1). J. Neurosci. 21, 2768–2783 (2001).

101. Hendry, S. H. & Yoshioka, T. A neurochemically distinct third channel inthe macaque dorsal lateral geniculate nucleus. Science 264, 575–577(1994).

102. Hendry, S. H. C. & Reid, R. C. The koniocellular pathway in primate vision.Annu. Rev. Neurosci. 23, 127–153 (2000).

103. Saul, A. B. & Humphrey, A. L. Spatial and temporal response properties oflagged and nonlagged cells in cat lateral geniculate nucleus. J. Neurophysiol.64, 206–224 (1990).

104. Menger, N., Pow, D. V. & Wässle, H. Glycinergic amacrine cells of the ratretina. J. Comp. Neurol. 401, 34–46 (1998).

105. Kalloniatis, M., Marc, R. E. & Murry, R. F. Amino acid signatures in theprimate retina. J. Neurosci. 16, 6807–6829 (1996).

106. Shapley, R. & Enroth-Cugell, C. Visual adaptation and retinal gain controls.Prog. Ret. Res. 3, 263–346 (1984).

107. Berry, M. J., Brivanlou, I. H., Jordan, T. A. & Meister, M. Anticipation ofmoving stimuli by the retina. Nature 398, 334–338 (1999).

108. Nelson, R. AII amacrine cells quicken time course of rod signals in the catretina. J. Neurophysiol. 47, 928–947 (1982).

109. Boos, R., Schneider, H. & Wässle, H. Voltage- and transmitter-gatedcurrents of AII-amacrine cells in a slice preparation of the rat retina. J. Neurosci. 13, 2874–2888 (1993).

110. Roska, B. & Werblin, F. Vertical interactions across ten parallel, stackedrepresentations in the mammalian retina. Nature 410, 583–587 (2001).

111. Famiglietti, E. V. Polyaxonal amacrine cells of rabbit retina: size anddistribution of PA1 cells. J. Comp. Neurol. 316, 406–421 (1992).

112. Dacey, D. M. Axon-bearing amacrine cells of the macaque monkey retina. J. Comp. Neurol. 284, 275–293 (1989).

113. Cook, P. B. & Werblin, F. S. Action potentials are propagated by wide-fieldamacrine cells in the tiger salamander retina. J. Neurosci. 14, 3852–3861(1994).

114. Stafford, D. K. & Dacey, D. M. Physiology of the A1 amacrine: a spiking,axon-bearing interneuron of the macaque monkey retina. Vis. Neurosci. 14,507–522 (1997).

115. Taylor, W. R. Response properties of long-range axon-bearing amacrinecells in the dark-adapted rabbit retina. Vis. Neurosci. 13, 599–604 (1996).

116. McIlwain, J. B. Receptive fields of optic tract axons and lateral geniculatecells: peripheral extent and barbituate sensitivity. J. Neurophysiol. 27,1154–1173 (1964).

117. Neuenschwander, S. & Singer, W. Long-range synchronization ofoscillatory light responses in the cat retina and lateral geniculate nucleus.Nature 379, 728–733 (1996).

118. Smirnakis, S. M., Berry, M. J., Warland, D. K., Bialek, W. & Meister, M.Adaptation of retinal processing to image contrast and spatial scale. Nature386, 69–73 (1997).

119. Brown, S. P. & Masland, R. H. Spatial scale and cellular substrate of contrastadaptation in retinal ganglion cell. Nat. Neurosci. 4, 44–51 (2001).

120. Dowling, J. E. The Retina: An Approachable Part of the Brain (Belknap,Cambridge, 1987).

121. Wu, S., Gao, F. & Maple, B. R. Functional architecture of synapses in theinner retina: segregation of visual signals by stratification of bipolar cellaxon terminals. J. Neurosci. 20, 4462–4470 (2000).

122. McBillem, G. S. & Dacheux, R. F. Responses to glutamate ofmorphologically defined rabbit cone bipolar cells. Invest. Ophthalmol. Vis.Sci. (2001).

123. Freed, M. A. Parallel cone bipolar pathways to a ganglion cell use differentrates and amplitudes of quantal excitation. J. Neurosci. 20, 3956–3963(2000).

124. Packer, O. S., Williams, D. R. & Bensinger, D. G. Photopigmenttransmittance imaging of the primate photoreceptor mosaic. J. Neurosci.16, 2251–2260 (1996).

125. Vaney, D. I. Neuronal coupling in rod-signal pathways of the retina. Invest.Ophthalmol. Vis. Sci. 38, 267–273 (1997).

review©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


articles


Alzheimer’s disease (AD) is characterized neuropathologically byprogressive deposition of the 40–42 residue amyloid β-protein(Aβ) in brain parenchyma and cerebral blood vessels. Several path-ogenic mutations have been identified in the amyloid precursorprotein (APP) gene, all located close to the major APP process-ing sites (for reviews, see refs. 1, 2). These processing sites arelocated either adjacent to the Aβ domain in APP (the β- and γ-secretase sites) or within the Aβ sequence itself (the α-secretasesite). Increased production of the more amyloidogenic Aβ42 isseen with mutations positioned in the vicinity of the γ-secretasecleavage site. The only known AD mutation close to the β-secre-tase site, the Swedish mutation3, results in elevation of both Aβ42and Aβ40 in plasma and fibroblasts from mutation carriers4–6.

Pathogenic mutations within the Aβ sequence (Fig. 1a) gen-erally result in a phenotype different from AD, with massiveamyloid accumulation in cerebral blood vessel walls in addi-tion to parenchymal amyloid plaques. The Dutch mutationcarriers (E693Q) are clinically characterized by the occurrenceof intracerebral hemorrhages7. Carriers of the Flemish muta-tion (A692G), another intra-Aβ mutation, frequently sufferfrom intracerebral hemorrhage, and individuals who survivedevelop a progressive dementia with features of AD8. Differ-ent pathogenic mechanisms have been proposed for the Dutchand Flemish mutations. It has been observed that the Flemishmutation leads to increased Aβ levels, whereas a reduced ratioof Aβ42/40 was seen in media from cells transfected with theDutch mutation9. In addition, the Flemish mutation resulted inan increased Aβ/p3 ratio, indicating effects on α-secretase

The ‘Arctic’ APP mutation (E693G)causes Alzheimer’s disease byenhanced Aβ protofibril formation

Camilla Nilsberth1, Anita Westlind-Danielsson1,2, Christopher B. Eckman3, Margaret M. Condron4, Karin Axelman1, Charlotte Forsell1, Charlotte Stenh1, Johan Luthman2,David B. Teplow4, Steven G. Younkin3, Jan Näslund1 and Lars Lannfelt1

1 Karolinska Institutet, Department of Neurotec, Geriatric Medicine, Novum KFC, S-141 86 Huddinge, Sweden2 Bioscience, Discovery Research Area CNS & Pain Control, AstraZeneca, S-151 85 Södertälje, Sweden3 Mayo Clinic Jacksonville, 4500 San Pablo Road, Jacksonville, Florida 32224, USA 4 Center for Neurologic Diseases, Brigham & Women’s Hospital, 77 Avenue Louis Pasteur (HIM756), Boston, Massachusetts 02115, USA

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

Several pathogenic Alzheimer’s disease (AD) mutations have been described, all of which causeincreased amyloid β-protein (Aβ) levels. Here we present studies of a pathogenic amyloid precursorprotein (APP) mutation, located within the Aβ sequence at codon 693 (E693G), that causes AD in aSwedish family. Carriers of this ‘Arctic’ mutation showed decreased Aβ42 and Aβ40 levels in plasma.Additionally, low levels of Aβ42 were detected in conditioned media from cells transfected withAPPE693G. Fibrillization studies demonstrated no difference in fibrillization rate, but Aβ with theArctic mutation formed protofibrils at a much higher rate and in larger quantities than wild-type(wt) Aβ. The finding of increased protofibril formation and decreased Aβ plasma levels in the ArcticAD may reflect an alternative pathogenic mechanism for AD involving rapid Aβ protofibril formationleading to accelerated buildup of insoluble Aβ intra- and/or extracellularly.

cleavage (Fig. 1a) that were not observed for the Dutch muta-tion10,11. Furthermore, radiosequence analysis of the Aβ andp3 proteins derived from APP with the Flemish and Dutchmutations suggests that these mutations affect both α- and β-secretase processing10,11. Investigations of synthetic Aβ pep-tides have indicated that the Dutch mutation acceleratesprotofibril formation compared to the wild-type peptide. Incontrast, the Flemish Aβ peptides display increased solubilityand decreased fibrillogenesis rates compared to wild-type pep-tides12,13. Thus, distinct Aβ fibrillization kinetics and effectson APP metabolism by these mutations may underlie the dif-ferences in their clinical features. A third pathogenic intra-Aβmutation was discovered in an Italian family (E693K), whichhad clinical manifestations similar to Dutch patients14. Final-ly, a mutation at codon 694 (D694N) in a family from Iowacauses progressive dementia and severe cerebral amyloidangiopathy15. There are currently no reports on effects by eitherof these mutations on cellular Aβ formation.

The Aβ protein forms amyloid fibrils that accumulate intosenile plaques in AD brains. The Aβ fibrillization process is a com-plex multi-step reaction, proceeding through a nucleation andextension phase16,17. A number of soluble fibril intermediates havebeen identified, including protofibrils and Aβ-derived diffusibleligands (ADDLs)12,18–23. Both species are neurotoxic20,21,23. Littleis known about the involvement of protofibrils or ADDLs in thepathogenesis of AD. However, findings indicate that protofibrilsgenerated by another protein, α-synuclein, are involved in earlyonset Parkinson’s disease, and pathogenic missense mutations in α-

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


synuclein accelerate protofibril formation24. Thus, protofibrils mayhave general importance as triggers of neurodegeneration.

We identified a pathogenic mutation located at codon 693within the Aβ region of APP, at which glutamic acid is substitut-ed for glycine (E693G). Affected subjects have clinical featuresof early-onset AD. We have named the mutation the ‘Arctic’mutation, because the family in which it was detected is fromnorthern Sweden. Plasma levels of both Aβ42 and Aβ40 werelower in mutation carriers compared to healthy family members,and the Aβ42 concentration was reduced in media from cellstransfected with APPE693G. Furthermore, in vitro studies showedthat the Aβ peptide with the Arctic mutation (Aβ1–40Arc) had ahigh propensity to form protofibrils. Taken together, these datasuggest that in carriers of the Arctic mutation, an alternate path-ogenic mechanism for AD operates, in which increased Aβprotofibril formation may be a primary event.

RESULTSClinical description and genetic analysisThe family carrying the Arctic mutation spans over 4 generations(Fig. 1b) and clinical information was available on 11 affected casesin 3 generations. An autosomal dominant pattern of inheritancewas seen in the family with a mean age of onset at 57 ± 2.9 years(range 54–61 years). Clinical examination, neuropsychological test-ing, brain imaging (computed tomography or magnetic resonanceimaging) and EEG were used in evaluating patients according toDSM-IV criteria. Clinical history was typical for AD, with a slow,insidious progression and decline in memory for recent events asthe first presenting symptom. Signs of strokes or vascular lesionswere not found on brain imaging in seven investigated patients.

The family carrying the Arctic mutation was screened formutations in exons 16 and 17 of the APP gene by single strandconformation polymorphism analysis (SSCP). An abnormalmobility pattern was observed in exon 17. Sequencing revealed

an A→G nucleotide substitu-tion, representing a glutamicacid to glycine substitution atAPP codon 693 (E693G), corre-sponding to position 22 in theAβ sequence. The mutation wasfully penetrant; no escapeeswere found. Two-point linkageanalysis was performed betweenthe mutation and affection sta-tus in the family, with an age-dependent penetrance, giving alod score of 3.66 at recombina-tion fraction 0.00. The E693Gmutation was previously report-ed in one patient in the US witha Swedish origin, in whom neu-ropathological examinationrevealed neuritic plaques andneurofibrillary tangles, con-firming the diagnosis of AD25.

Investigations have revealed that this subject is likely to be a mem-ber of the Arctic kindred (T.D. Bird, personal communication).

Decreased Aβ plasma levels in Arctic mutation carriersPathogenic APP mutations affect APP processing, as reflected in anincrease of either total Aβ or Aβ42 in the plasma of affected fam-ily members6. We investigated whether the Arctic mutation alsoaffected Aβ plasma levels. Plasma from 9 mutation carriers, ofwhom 4 were symptomatic, and from 11 non-carriers in the fam-ily was analyzed by well-characterized sandwich enzyme-linkedimmunosorbent assay (ELISA) systems detecting Aβ42(BAN50/BC05) and Aβ40 (BAN50/BA27)26. To prove that theArctic mutation did not change any of the antibody recognitionsites, Aβ1-40wt and Aβ1-40Arc peptides were tested and foundto be recognized equally well. However, the antibodies BNT77and 4G8, both raised against epitopes in the middle of the Aβ pro-tein, had reduced affinity for the Arctic peptide; thus, they werenot used. The Aβ42 plasma concentration was 11.7 ± 3.9 fmol/mlin mutation carriers, 27% less than the concentration in non-car-riers, 16.0 ± 5.6 fmol/ml (p = 0.04, Fig. 2). The Aβ40 plasma con-centration was 105 ± 22 fmol/ml, 26% less than the concentrationin non-carriers, 141 ± 34 fmol/ml (p = 0.01, Fig. 2). The Aβ42/40ratio was calculated for each individual, but no significant differ-ence between mutation carriers and controls was found (p = 0.13).

Aβ42 is reduced in media from Arctic cellsThe effect of the Arctic mutation on Aβ formation was furtherinvestigated in vitro in transiently transfected HEK293 cells. APPwtwas compared to the Arctic (APPE693G), Dutch (APPE693Q), Ital-ian (APPE693K) and Flemish (APPA692G) mutations (Fig. 1a). Con-structs containing the Swedish double mutation (APPSwe) andone APP mutation at codon 717 (APPV717F), both with well-stud-ied APP processing characteristics1, were used as positive con-trols. Media were conditioned and analyzed for Aβ levels by the


articles

Fig. 1. The intra-Aβ mutations. (a) The APP molecule with localizationof the Aβ and p3 proteins, containing pathogenic intra-Aβ mutations. (b)Pedigree showing the segregation of AD and of the E693G mutation inthe Arctic family, compatible with an autosomal dominant pattern ofinheritance. The pedigree has been disguised to protect the confidential-ity of family members. Solid symbols, affected; open symbols, unaffected;slashed symbols, deceased; +/–, mutation carrier; –/–, non-carrier.

a

b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


same Aβ42- and Aβ40-specific sandwich ELISA systems as usedfor human plasma (Table 1). Care was taken to obtain similar APPexpression between different mutations and experiments, whichresulted in almost identical APP levels (data not shown). In accor-dance with previous reports27, Aβ42 and Aβ40 protein concen-trations in media were increased 4–5 times by the Swedishmutation. Increased Aβ42 levels were seen in media by the APP717mutation, leading to an approximately three-fold increase in theAβ42/40 ratio, in agreement with data previously reported26. Allmutations located at codon 693 (Arctic, Dutch and Italian) showedthe same effect on Aβ levels: the Aβ42 concentration was signifi-cantly lower, whereas Aβ40 levels were similar to that in APPwt-transfected cells. Consequently, the Aβ42/40 ratio showed a22–33% reduction (Table 1). This finding was in contrast to theAPP692 Flemish mutation, which caused increased levels of bothAβ42 and Aβ40 in conditioned media, producing a slight increasein Aβ42/40 ratio, in accordance with previous reports26.

Radiosequence analysisRadiosequence analysis was done to exclude the possibility thatthe Arctic mutation caused an increase in the amount of N-ter-minally truncated forms of Aβ, which might not be detected bythe BAN50 ELISA. To increase Aβ levels, hybrid APP mutants wereused containing the Swedish mutation alone or the Swedish muta-tion together with the Arctic mutation. The 4-kD Aβ proteins pro-duced by each sample were almost identical, yielding peaks ofradioactivity (cycles 4, 19, 20) corresponding to Aβ beginning atAsp 1 (Fig. 3a and c). Both p3 bands produced profilesof radioactivity consistent with APP cleavage eventsoccurring near the α-secretase site (Lys16–Leu17). Forexample, most of the protein produced from APP-Swestarts at Val 18 but peptides starting at Leu 17 and Phe 19 could also be observed (Fig. 3b). In the p3 sam-ple from APP-Swe/Arc, peptides started primarily atVal 18 and Phe 19. In addition, a smaller peak wasobserved at cycles 9 and 10, corresponding to Aβ begin-ning at Glu 11 (Fig. 3d). All of these species areobserved commonly in samples from HEK cells10,11.

Aβ fibril formationWe investigated the effect of the Arctic amino acidsubstitution (Glu22→Gly) on amyloid fibrillizationkinetics. Size exclusion chromatography (SEC)analysis of freshly dissolved Aβ1-40wt produced a

single peak at a retention time of approximately 20 minutes(Fig. 4a). This peak represented the monomeric/dimeric formof Aβ1-40wt12. With increasing incubation time, a second dis-tinct peak appeared in the gel-excluded fraction with a reten-tion time of about 12 minutes. This earlier peak containedprotofibrils (Fig. 4b and c), as verified by ultracentrifugationand by negative stain and transmission electron microscopy(TEM) of Aβ1-40wt (data not shown), in line with previousfindings12. Similar retention times were obtained for the Aβ1-40Arc peptide (Fig. 4d–f). However, Aβ1-40Arc generatedprotofibrils much faster and in larger quantities than Aβ1-40wt(Fig. 4). The monomeric/dimeric Aβ1-40Arc peak declined inparallel with the growth of the protofibrillar peak (Fig. 4d–f).The maximum concentration (111 µM) of Aβ1-40Arc protofib-rils was observed at 6.5 hours. Kinetic studies up to 48 hoursshowed that Aβ1-40wt generated a small quantity of protofib-rils with a maximum concentration at 25 hours (Fig. 5a). Incontrast, a rapid formation of protofibrils was seen with theAβ1-40Arc within the first five hours of incubation with asimultaneous rapid decline in the concentration of themonomeric/dimeric peptide (Fig. 5b). Despite the dramatickinetic difference between Aβ1-40wt and Aβ1-40Arc, there wasno detectable difference in fibrillization rate. Because carriersof the Arctic mutation are heterozygotes, they produce bothAβwt and AβArc. Assuming equimolar in vivo production, thekinetics of protofibril formation was studied in a 1:1 mixture ofAβ1-40wt and Aβ1-40Arc. This mixture of peptides showedkinetics that were intermediate to the single peptide curves(Fig. 5c).

The fibrillar and protofibrillar morphology of Aβ1-40Arc insedimented samples from kinetic studies was confirmed by nega-tive stain and TEM. Many Aβ1-40Arc fibrils displayed larger diam-eters (10–18 nm) relative to the Aβ1-40wt samples, due tointertwining of several thinner fibrils (Fig. 6b). Protofibrils couldalso be discerned in the sedimented samples (Fig. 6). The presenceof protofibrils was confirmed in parallel from the same prepara-tion using SEC. Many longer and straighter protofibrils were foundin the Aβ1-40Arc preparation, relative to the wild-type sample,although curved variants were also found (Fig. 6b and c). Similarto the Aβ1-40wt, mesh-works of curved Aβ fibrils could be seen

articles


Fig. 2. Plasma Aβ levels in the family carrying the Arctic mutation.Plasma from 9 mutation carriers and 11 non-carriers in the familywere analyzed by end-specific ELISAs. A significant decrease (*) ofAβ42 (p = 0.04) and Aβ40 (p = 0.01) was observed in carriers of theArctic mutation.

Table 1. Aβ42/40 ratio and Aβ42 and Aβ40 levels in conditioned mediafrom transiently transfected HEK293 cells.

Aβ42/40 ratio Aβ42 ± s.d. Aβ40 ± s.d.APP constructs (%) ± s.d. (fmol/ml) (fmol/ml)

APPwt 9.6 ± 0.7 13.8 ± 1.0 144 ± 6Arctic (E693G) 7.5 ± 0.5* 11.2 ± 0.6 149 ± 3Dutch (E693Q) 6.6 ± 0.6* 9.6 ± 0.7 147 ± 12Italian (E693K) 6.4 ± 0.6* 8.0 ± 0.7 126 ± 17Flemish (A692G) 11.7 ± 1.6* 27.0 ± 2.0 232 ± 25Swedish 8.8 ± 0.9 59.1 ± 4.9 675 ± 59V717F 27.4 ± 1.7* 24.4 ± 2.4 89 ± 11Mock (vector only) 7.2 ± 2.4 2.1 ± 1.0 28 ± 5

*p = 0.004 in comparison to APPwt.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


in the Aβ1-40Arc preparations. These are likely to be assembliesformed through intermolecular association of many protofibrils.

DISCUSSIONWe identified the first pathogenic mutation, designated the Arc-tic mutation, located within the Aβ protein domain of the APPgene, which produces a classic AD phenotype. Carriers of thismutation develop progressive dementia with clinical features typ-ical of AD, but without the severe cerebral amyloid angiopathythat characterizes other mutations in the Aβ region of APP7,8,14,15.Aβ42 is elevated in plasma from subjects with APP717 mutationsas well as from those with presenilin (PS) 1 and PS 2 mutations.In carriers of the Swedish mutation (APP KM670/671NL), bothAβ42 and Aβ40 were elevated in plasma6,28. On the basis of thesefindings, it has been suggested that overproduction of Aβ42 is acentral event in the pathogenesis of AD1. However, inconclusiveresults on Aβ plasma levels in sporadic AD have been report-ed6,29–31. In the present study, Aβ42 and Aβ40 plasma levels wereshown to be significantly decreased in carriers of the Arctic muta-tion. Low levels of Aβ were also observed in the youngest muta-tion carriers investigated, 20–30 years before the expected onsetof the disease, suggesting a long period of biochemical abnor-mality before clinical onset. The other four intra-Aβ mutationsidentified (Dutch, Flemish, Iowa and Italian) give rise to a clini-cal phenotype different from AD. To our knowledge, there is noinformation on Aβ plasma levels available for these families.

In vitro studies of the effects of APP and PS mutations on APPmetabolism correlate well with in vivo findings (for review, seeref. 32). Thus, the metabolic effects of the Arctic mutation werecompared to those of the other intra-Aβ mutations in transient-ly transfected cells by ELISA measurements. Strikingly, all three

Fig. 4. Size exclusion chromatograms of Aβ. SEC profiles illustratingtime-dependent growth of protofibril peak (left peak) and concomitantdecline of monomeric/dimeric peak (right peak) for Aβ1-40wt (a–c)versus Aβ1-40Arc (d–f) at 5 (a, d), 45 (b, e) and 125 (c, f) minutes incu-bation. Initial peptide concentrations were 143 µM (Aβ1-40wt) and 138µM (Aβ1-40Arc). Each SEC analysis was run for 35 min (x-axis).Chromatograms are from one experiment representative of four.

Fig. 3. Radiosequencing of Aβ 3H-phenylalanine-labeledAβ and p3. We sequenced 4-kD bands from APP contain-ing the Swedish mutation (a) or the Swedish/Arctic muta-tion (c) and corresponding 3-kD bands with the Swedish(b) or the Swedish/Arctic mutation (d). The peaks atcycle numbers 4, 19 and 20 in (a) and (c) correspond tothe positions of phenylalanine residues in Aβ beginning atAsp 1. Differences between the Swedish andSwedish/Arctic samples in the relative heights of thepeaks at cycles 4 and 19/20 are related to sequencingchemistry. The differences do not bear on the conclu-sions reached in the experiment. The major peaks in (b)and (d) also originate from the two phenylalanines atpositions 19 and 20 in Aβ. In these cases, however, theproteins start primarily at Val 18, Phe 19 or Phe 20.


mutations at codon 693 (Arctic, Dutch and Italian)led to decreased Aβ42 concentrations in condi-tioned media, whereas increased levels of both Aβ42and Aβ40 in media were found for the FlemishAPP692 mutation. These results were verified by asecond independent ELISA using monoclonal anti-body 6E10 as capture antibody and end-specificpolyclonal antibodies for detection (unpublished

observations). This is consistent with previously reported in vitroeffects of the Flemish and Dutch mutations9. Increased Aβ levelsin conditioned media were reported for the Flemish mutation andreduced Aβ42/40 ratios were observed in cells transfected withthe Dutch mutation. The reduced levels of Aβ could occur as aresult of the BAN50 ELISA used, because this method does notcapture truncated forms of Aβ. However, both Aβ40 and Aβ42levels would be reduced in equal amounts if this were the case,whereas our results showed that only Aβ42 was reduced in themedia. In addition, radiosequence analysis revealed that most ofthe Arctic Aβ is not N-terminally truncated. When the p3 bandwas sequenced, the Swedish/Arctic mutation produced increasedamounts of protein beginning at Glu 11 compared to the amountproduced by the Swedish mutation alone. Even though this pro-tein does not contribute greatly to the total amount of protein

articles

a b

c d

a d

b e

c f

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 6. Electron micrographs of Aβ protofibrils. High-magnification transmission electron micrographs,104,000× (insets, 630,000×), of sedimented negativelystained Aβ1-40wt (a) and Aβ1-40Arc (b, c) samples.Aβ1-40wt protofibrils are seen with their typical curvedappearance and with lengths of ∼ 30-60 nm (a). Bothcurved (arrows; b, c) and straight Aβ1-40Arc protofibrilscan be discerned (c) in this preparation. The Aβ1-40Arcprotofibrils appear longer than the Aβ1-40wt protofibrils,and a larger percent of straight, short protofibrils werenoted. The diameters of the two protofibril types weresimilar (∼ 4–6 nm; insets). Aβ1-40Arc fibrils were alsoobserved that exhibited large diameters (∼ 10–18 nm)resulting from intertwining of two or more thinner fibrils(b). Scale bars, 100 nm, 20 nm (insets).

produced, it could be important as a nidus for Aβ oligomeriza-tion. Proteins beginning at Glu 11 are more prone to aggregateand may be important in the initial events of AD pathogenesis13,33.

It is intriguing that mutations located at the same Glu693codon of APP lead to such different phenotypes, whereas in vitrostudies indicate a similar effect on Aβ concentration in condi-tioned media. It is evident that Aβ deposition is a central eventin the pathological cascade, but why does the E693G Arctic muta-tion lead to AD, whereas the E693Q, E693K, A692G and D694Nmutations lead to a different clinical phenotype with predomi-nately vascular symptoms? The answer may lie within the Aβsequence. The KLVFF motif at position 16–20 in the Aβ protein iscentral in the fibrillization process34,35. Mutations at position 21–23 in Aβ are located close to the KLVFF region and couldtherefore affect the conformation of the peptide and its fibrilliza-tion process. Indeed, in vitro studies on the Dutch peptide havedemonstrated that it polymerizes into protofibrils and then intofibrils significantly faster than the wild-type peptide12. However,at present, the understanding is limited concerning the differencesin clinical expression by the pathogenic intra-Aβ mutations.

The Glu to Gly single amino acid substitution at position 22in the Aβ1-40Arc molecule caused a dramatic increase in rateand capacity to form protofibrils compared to the Aβ1-40wt pep-tide, even when protofibril formation was measured at equimo-lar amounts of wild-type:Arc. We propose that when Aβ42Arcand Aβ40Arc are formed in situ in the brain, they are more proneto be retained by cellular systems due to the accelerated drive toform protofibrils. Protofibril formation will significantly enhancethe bulk and insolubility of Aβ. This, in turn, could effectivelyaccelerate disease initiation and progression. Recent cell biolog-

ical studies suggest that Aβ is generated intracellularly (for review,see ref. 36). In addition, it has been reported that human neu-rons preferably accumulate Aβ42 over Aβ4037, consistent withstudies showing that C-terminally extended proteins have a muchhigher tendency to form amyloid fibrils38. This offers an expla-nation to our finding of decreased Aβ42 concentration in mediafrom cells transfected with mutations at APP codon 693. Theeffect of the Arctic mutation on Aβ42 fibrillization was not stud-

ied here because of technical problems associatedwith Aβ42 proteins, although it may be visualizedindirectly by intracellular quantification of bothAβ40 and Aβ42 (ongoing studies). However, pre-vious studies have shown that Aβ42 also formsprotofibrils12; thus, we anticipate that Aβ42Arc

articles


Fig. 5. Kinetics of Aβ. Kinetics of Aβ1-40wt (88 µM) (a) and Aβ1-40Arc (92 µM) (b) monitoring decline in monomeric/dimeric peak area() and concomitant increase and subsequent decline of protofibrillarpeak area (). Each data point is the mean ± s.d. of three injections inone experiment representative of three. Protofibril formation moni-tored for 7 h using 50-µM peptide concentrations. Aβ1-40wt () andAβ1-40Arc () were compared to a 1:1 mixture of Aβ1-40wt:Aβ1-40Arc () (c). Maximum protofibrillar concentrations were 1.3 µM(Aβ1-40wt), 14 µM (Aβ1-40wt + Arc) and 21 µM (Aβ1-40Arc). Eachcurve is the mean of three separate experiments (Aβ1-40Arc, oneexperiment) ± s.d. (with 2–3 injections/point).

a

b

c

a

b c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



will be more prone to form protofibrils than both the Aβ42wtand Aβ40Arc.

The Arctic form of AD may be an example of a rare diseasevariant that helps explain molecular mechanisms that may under-lie more common forms of the disorder. Clinical, genetic and bio-chemical data support the hypothesis that the Arctic mutationproduces AD through an alternative pathogenic mechanism. Wepropose that this mechanism involves rapid Aβ protofibril for-mation leading to accelerated buildup of insoluble Aβ intra-and/or extracellularly. (The vast majority of sporadic AD casesdo not show increased Aβ plasma levels6,30.) Thus, factors pro-moting protofibril formation should be considered in the patho-genesis of sporadic AD. Increased protofibril formation may alsobe operating in these more common forms of the disease. Indeed,our findings open new avenues for possible therapeutic interven-tion using drugs targeted at preventing protofibril formation24,39.

METHODSPCR amplification and sequencing. Venous blood was drawn into tubescontaining EDTA, and DNA was prepared according to standard proce-dures. SSCP was done as described40. To sequence exon 17 of the APPgene, a 319-bp fragment was amplified with the following primers: 5´-CCTCATCCAAATGTCCCCGTCATT-3´ and 5´-GCCTAATTCTCT-CATAGTCTTAATTCCCAC-3´. Direct sequencing was done in both 3´and 5´ directions using the same primers. The Arctic mutation was seenin one family and not in 56 controls or 254 cases with dementia.

Linkage analysis. Two-point lod score was calculated using Mlink fromthe Linkage software package (version 5.1) at each of the following recom-bination fractions 0.00, 0.10, 0.20, 0.30 and 0.40 (q males = q females). Asingle-locus model with an autosomal dominant inheritance wasassumed, which was compatible with the inheritance as it appeared inthe pedigree. A cumulative age-dependent penetrance was assigned fromthe known ages of onset in the family. Individuals were put into differentliability classes depending on the age at onset (affected) or age at lastexamination (unaffected). The disease gene frequency and the markerallele frequency were estimated to be 0.001 and the phenocopy rate wasset to 0.0001.

Aβ42 and Aβ40 measured by ELISA. The concentration of Aβ in plasmaand in the conditioned media was measured by the end-specific ELISAs,as described previously26. Aβ was captured with BAN50. Aβ40 and Aβ42were subsequently detected with BA27 and BC05, respectively. Plasmawas spiked with synthetic peptides, revealing that both Aβ1-40Arc andAβ1-40wt peptides were recovered by ELISA to the same extent. The dataobtained was analyzed by non-parametric Mann–Whitney analysis.

In vitro mutagenesis of APP cDNA and transfection of HEK293 cells.The mutations were introduced to APP695 cDNA in pcDNA3 usingQuikChange Site-Directed Mutagenesis Kit according to the manufac-turer’s instructions (Stratagene, La Jolla, California). The mutated con-structs were verified by sequencing. For the ELISA measurements, HEK293cells were transfected with the different constructs using FuGENE 6 Trans-fection Reagent (Roche Diagnostics, Basel, Switzerland) according to themanufacturer’s instructions. Twenty-four hours after transfection, thecells were conditioned for 48 h in OptiMEM containing 5% newborn calfserum. The APP expression in the cells were investigated by western blotusing monoclonal antibody 22C11 (Roche Diagnostics).

Radiosequence analysis. HEK293 cells were transiently transfected withAPP containing the Swedish mutation alone or APP with both theSwedish and the Arctic mutation. Twenty-four hours after transfection,the cells were metabolically labeled for 48 h with 360 µCi/ml L-[2, 3, 4, 5,6-3H]phenylalanine (Amersham Pharmacia Biotech, Buckinghamshire,UK). The media were collected and the Aβ protein was isolated byimmunoprecipitation using a polyclonal Aβ40-end specific antibody,KI2Ger, followed by SDS-PAGE and electroblotting. The bands corre-

sponding to Aβ and p3 were excised from the membrane and radiose-quenced as described10.

Synthetic peptides. Aβ1-40wt was purchased from Bachem, Bübendorf,Switzerland or Biosource International/QCB (Camarillo, California) andAβ1-40Arc from Biosource International/QCB. The peptides were tri-fluoroacetic salts; they were stored at –20°C. All other chemicals were ofthe highest purity available.

Aβ fibrillization studies. Samples of each peptide were incubated with-out agitation at 30°C in 50 mM Na2HPO4·NaH2PO4 (pH 7.4) contain-ing 0.1 M NaCl, for various time points. Initial peptide concentrationswere 88–143 µM, similar for both peptides in each experiment. Aftercentrifugation (17,900 g for 5 min at 16°C) Aβ1-40 species, sampledfrom the supernatant, were separated using SEC. Chromatographicseparation and analysis were done with a Merck Hitachi D-7000LaChrom HPLC system, having a diode array detector model L-7455, aL-7200 model autosampler and a model L-7100 pump, coupled to aSuperdex 75 PC3.2/30 column (Amersham Pharmacia Biotech). Sam-ples were eluted at a flow rate of 0.08 ml/min (ambient temperature)using 50 mM Na2HPO4 NaH2PO4 (pH 7.4) and 0.15 M NaCl. Chro-matograms were obtained by measuring ultraviolet absorbance at 214nm. Peak areas for monomeric/dimeric and protofibrillar Αβ were inte-grated using Merck-Hitachi Model D-7000 Chromatography Data Sta-tion Software (Amersham Pharmacia Biotech). The mean of triplicateintegrated peak values from the SEC measurements were used to gen-erate each data point. In addition, a standard curve was produced bycorrelating integrated peak areas with peptide concentrations as deter-mined by quantitative amino acid analysis. The concentrations of total(at t = 0 h) and soluble peptides remaining in solution after centrifu-gation were calculated from the standard curve.

Transmission electron microscopy. Aβ peptide samples were preparedand incubated as indicated for the kinetic studies, using higher peptideconcentrations (617 µM). After eight days, aggregated Aβ species weresedimented using the same centrifugation parameters as described above.Buffer was removed and pelleted material was suspended in 50 µl waterusing gentle sonication (2 × 6 s). We applied 8-µl samples to carbon-sta-bilized Formvar film grids (Ted Pella, Redding, California). Samples werenegatively stained with 8 µl uranyl acetate (1%; E. Merck, Darmstadt,Germany). Four grids were prepared for each sample and examined usinga Philips CM10 TEM. We also examined samples from pellets sedimentedduring the kinetic experiments.

ACKNOWLEDGEMENTSWe thank G. Arnerup, R. Kaiser, L. Lilius, S. Petrén and D. Yager for scientific

support, and L. Tjernberg and colleagues in our department for comments on

the manuscript. The following foundations are acknowledged: the Swedish

Society for Medical Research, Trygg-Hansa, Stiftelsen för Gamla tjänarinnor,

Åke Wibergs stiftelse, Erik Rönnbergs Stiftelse, Stiftelsen Clas Groschinskys

minnesfond, Artur Eriksson, Gun & Bertil Stohnes stiftelse, Loo and Hans

Ostermans stiftelse för geriatrisk forskning, Ulf Widengrens Minnesfond,

Swedish Society for Medicine, the Alzheimer Foundation, the Swedish Medical

Research Council (project 10819) and the US National Institutes of Health

(grants NS38328 and AG14366 to D.B.T.).


1. Hardy, J. Amyloid, the presenilins and Alzheimer’s disease. Trends Neurosci.20, 154–159 (1997).

2. Selkoe, D. J. Translating cell biology into therapeutic advances in Alzheimer’sdisease. Nature 399, A23–A31 (1999).

3. Mullan, M. et al. A pathogenic mutation for probable Alzheimer’s disease inthe APP gene at the N-terminus of β-amyloid. Nature Genet. 1, 345–347(1992).

4. Citron, M. et al. Excessive production of amyloid β-protein by peripheralcells of symptomatic and presymptomatic patients carrying the Swedishfamilial Alzheimer’s disease mutation. Proc. Natl. Acad. Sci. USA 91,11993–11997 (1994).

articles©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


5. Johnston, J. A. et al. Increased β-amyloid release and levels of amyloidprecursor protein (APP) in fibroblast cell lines from family members with theSwedish Alzheimer’s disease APP670/671 mutation. FEBS Lett. 354, 274–278(1994).

6. Scheuner, D. et al. Secreted amyloid β-protein similar to that in the senileplaques of Alzheimer’s disease is increased in vivo by the presenilin 1 and 2and APP mutations linked to familial Alzheimer’s disease. Nat. Med. 2,864–869 (1996).

7. Levy, E. et al. Mutation of the Alzheimer’s disease amyloid gene in hereditarycerebral hemorrhage, Dutch type. Science 248, 1124–1126 (1990).

8. Hendriks, L. et al. Presenile dementia and cerebral haemorrhage linked to amutation at codon 692 of the β-amyloid precursor protein gene. Nat. Genet.1, 218–221 (1992).

9. De Jonghe, C. et al. Flemish and Dutch mutations in amyloid β precursorprotein have different effects on amyloid β secretion. Neurobiol. Dis. 5,281–286 (1998).

10. Haass, C., Hung, A. Y., Selkoe, D. J. & Teplow, D. B. Mutations associated witha locus for familial Alzheimer’s disease result in alternative processing ofamyloid β-protein precursor. J. Biol. Chem. 269, 17741–17748 (1994).

11. Watson, D. J., Selkoe, D. J. & Teplow, D. B. Effects of the amyloid precursorprotein Glu693→ Gln ‘Dutch’ mutation on the production and stability ofamyloid β-protein. Biochem. J. 340, 703–709 (1999).

12. Walsh, D. M., Lomakin, A., Benedek, G. B., Condron, M. M. & Teplow, D. B.Amyloid β-protein fibrillogenesis—detection of protofibrillar intermediate.J. Biol. Chem. 272, 22364–22372 (1997).

13. Walsh, D. M., Hartley, D. M., Condron, M. M., Selkoe, D. J. & Teplow, D. B. Invitro studies of amyloid β-protein fibril assembly and toxicity provide clues tothe aetiology of Flemish variant (Ala692→Gly) Alzheimer’s disease. Biochem J.355, 869–877 (2001).

14. Tagliavini, F. et al. A new βPP mutation related to hereditary cerebralhaemorrhage. Alz. Report 2, S28 (1999).

15. Grabowski, T. J., Cho, H. S., Vonsattel, J. P. G., Rebeck, G. W. & Greenberg, S.M. Novel amyloid precursor protein mutation in an Iowa family with dementiaand severe cerebral amyloid angiopathy. Ann. Neurol. 49, 697–705 (2001).

16. Hasegawa, K., Yamaguchi, I., Omata, S., Gejyo, F. & Naiki, H. Interactionbetween Aβ(1-42) and Aβ(1-40) in Alzheimer’s β-amyloid fibril formation invitro. Biochemistry 38, 15514–15521 (1999).

17. Teplow, D. B. Structural and kinetic features of amyloid β-proteinfibrillogenesis. Amyloid 5, 121–142 (1998).

18. Harper, J. D., Lieber, C. M. & Lansbury P. T. Jr. Atomic force microscopicimaging of seeded fibril formation and fibril branching by the Alzheimer’sdisease amyloid-β protein. Chem. Biol. 4, 951–959 (1997).

19. Harper, J. D., Wong, S. S., Lieber, C. M. & Lansbury, P. T. Jr. Observation ofmetastable Aβ amyloid protofibrils by atomic force microscopy. Chem. Biol.4, 119–125 (1997).

20. Lambert, M. P. et al. Diffusible, nonfibrillar ligands derived from Aβ1–42 arepotent central nervous system neurotoxins. Proc. Natl. Acad. Sci. USA 95,6448–6453 (1998).

21. Walsh, D. M. et al. Amyloid β-protein fibrillogenesis—structure andbiological activity of protofibrillar intermediates. J. Biol. Chem. 36,25945–25952 (1999).

22. Harper, J. D., Wong, S. S., Lieber, C. M. & Lansbury, P. T. Assembly of Aβ

amyloid protofibrils: an in vitro model for a possible early event inAlzheimer’s disease. Biochemistry 38, 8972–8980 (1999).

23. Hartley, D. M. et al. Protofibrillar intermediates of amyloid β-protein induceacute eletrophysiological changes and progressive neurotoxicity in corticalneurons. J. Neurosci. 19, 8876–8884 (1999).

24. Conway, K. A. et al. Acceleration of oligomerization, not fibrillization, is ashared property of both α-synuclein mutations linked to early-onsetParkinson’s disease: implications for pathogenesis and therapy. Proc. Natl.Acad. Sci. USA 97, 571–576 (2000).

25. Kamino, K. et al. Linkage and mutational analysis of familial Alzheimerdisease kindreds for the APP gene region. Am. J. Hum. Genet. 51, 998–1014(1992).

26. Suzuki, N. et al. An increased percentage of long amyloid β protein secretedby familial amyloid β protein precursor (βAPP717) mutants. Science 264,1336–1340 (1994).

27. Citron, M. et al. Mutation of the β-amyloid precursor protein in familialAlzheimer’s disease increases β-protein production. Nature 360, 672–674(1992).

28. Citron, M. et al. Mutant presenilins of Alzheimer’s disease increaseproduction of 42-residue amyloid β-protein in both transfected cells andtransgenic mice. Nat. Med. 3, 67–72 (1997).

29. Tamaoka, A. et al. Amyloid β protein in plasma from patients with sporadicAlzheimer’s disease. J. Neurol. Sci. 15, 65–68 (1996).

30. Iwatsubo, T. Amyloid β protein in plasma as a diagnostic marker forAlzheimer’s disease. Neurobiol. Aging 19, 161–163 (1998).

31. Mayeux, R. et al. Plasma amyloid β-peptide 1-42 and incipient Alzheimer’sdisease. Ann. Neurol. 46, 412–416 (1999).

32. Selkoe, D. J. The cell biology of β-amyloid precursor protein and presenilin inAlzheimer’s disease. Trends Cell Biol. 8, 447–453 (1998).

33. He, W. & Barrow, C.J. The Aβ 3-pyroglutamyl and 11-pyroglutamyl peptidesfound in senile plaque have greater β-sheet forming and aggregationpropensities in vitro than full-length Aβ. Biochemistry 38, 10871–10877 (1999).

34. Tjernberg, L. O. et al. Arrest of β-amyloid fibril formation by a pentapeptideligand. J. Biol. Chem. 271, 8545–8548 (1996).

35. Soto, C., Kindy, M. S., Baumann, M. & Frangione, B. Inhibition ofAlzheimer’s amyloidosis by peptides that prevent β-sheet conformation.Biochem. Biophys. Res. Commun. 226, 672–680 (1996).

36. Wilson, C. A., Doms, R. W. & Lee, V. M. Intracellular APP processing and Aβproduction in Alzheimer disease. J. Neuropath. Experiment. Neurol. 58,787–794 (1999).

37. Gouras, G. K. et al. Intraneuronal Aβ42 accumulation in human brain. Am. J.Pathol. 156, 15–20 (2000).

38. Harper, J. D. & Lansbury, P. T. Jr. Models of amyloid seeding in Alzheimer’sdisease and scrapie: mechanistic truth and physiological consequences of thetime-dependent solubility of amyloid proteins. Annu. Rev. Biochem. 66,385–407 (1997).

39. Klein, W. L., Krafft, G. A. & Finch, C. E. Targeting small Aβ oligomers: thesolution to an Alzheimer’s disease conundrum? Trends Neurosci. 24, 219–224(2001).

40. Forsell, L. & Lannfelt, L. Amyloid precursor protein mutation at codon 713(Ala→Val) does not cause schizophrenia: non-pathogenic variant found atcodon 705 (silent). Neurosci. Lett. 184, 90–93 (1995).

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

Glutamate is the most widely used neurotransmitter in the ver-tebrate central nervous system. The NMDA receptor, a subtypeof glutamate receptor exhibiting high calcium permeability, is crit-ical in synaptic plasticity, and is implicated in a variety of neu-ropathological processes, including seizures and excitotoxicityassociated with stroke1. Alteration of the time course of macro-scopic NMDA currents has important functional implicationsbecause it changes the amount of calcium influx. One mechanismby which the time course of macroscopic currents can be modu-lated is desensitization2, defined as a reduction of macroscopicNMDA currents in the continuous presence of glutamate. Sever-al forms of desensitization have been reported (for review, see refs.1, 2). The term ‘glycine-dependent desensitization’ is used todescribe the decrement of NMDA receptor currents that occurswhen glycine concentrations are not over-saturating; this is dueto a negative allosteric interaction between glutamate binding andglycine binding3,4. The term ‘glycine-independent desensitization’describes desensitization of NMDA receptors that cannot be pre-vented by a high concentration of glycine5–8. The glycine-inde-pendent desensitization described in some earlier works mayinvolve calcium-dependent inactivation. An increase in intracel-lular calcium, resulting from either activation of NMDA recep-tors or from activation of other calcium-permeable channels, canreduce NMDA receptor currents9–13. More recently, the term‘glycine-independent desensitization’ has been used to describeall forms of calcium-independent and glycine-independent desen-sitization, which are particularly prominent for recombinantNR1/NR2A receptors. Recent progress in molecular biology hasbegun to make it possible to differentiate various forms of desen-sitization on a structural basis. Molecular determinants for the

Allosteric interaction between theamino terminal domain and theligand binding domain of NR2A

F. Zheng1, K. Erreger1, C.-M. Low1,2, T. Banke1, C. J. Lee1, P. J. Conn1,3 and S. F. Traynelis1

1 Department of Pharmacology, Emory University School of Medicine, 1510 Clifton Road, Atlanta, Georgia 30322, USA2 Present address: National Neuroscience Institute, 11 Jalan Tan Tock Seng, S308433, Singapore3 Present address: Merck Research Laboratories, 770 Sumneytown Pike, PO Box 4, WP46-300, West Point, Pennsylvania 19486-0004, USA

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

Fast desensitization is an important regulatory mechanism of neuronal NMDA receptor function.Only recombinant NMDA receptors composed of NR1/NR2A exhibit a fast component of desensi-tization similar to neuronal NMDA receptors. Here we report that the fast desensitization ofNR1/NR2A receptors is caused by ambient zinc, and that a positive allosteric interaction occursbetween the extracellular zinc-binding site located in the amino terminal domain and theglutamate-binding domain of NR2A. The relaxation of macroscopic currents reflects a shift to anew equilibrium due to increased zinc affinity after binding of glutamate. We also show a similarinteraction between the ifenprodil binding site and the glutamate binding site of NR1/NR2Breceptors. These data raise the possibility that there is an allosteric interaction between theamino terminal domain and the ligand-binding domain of other glutamate receptors. Ourfindings may provide insight into how zinc and other extracellular modulators regulate NMDAreceptor function.

glycine-independent desensitization have been mapped to twodistinct extracellular domains of NR2 subunits14,15. A four-amino-acid domain just upstream of the M1 region (‘pre-M1 domain’)has been reported to influence the slower component of glycine-independent desensitization (τ = 2 s). A leucine/isoleucine/valine-binding protein (LIVBP)-homologous amino terminal domain(‘LIVBP domain’) is required for the fast component of glycine-independent desensitization14,15.

Despite this progress, the kinetics for desensitization ofNMDA receptors are not well understood. Several groups havedescribed two components of glycine-independent desensiti-zation of NR1/NR2A receptors15,16, whereas others observe asingle component14. The onset of calcium-dependent inacti-vation is slow in some studies17,18, but faster in others19,20. Theunderlying basis of these discrepancies seems to be the vari-able degree of a fast component with a time constant of200–300 ms. In the present study, we present several lines ofevidence to suggest that this fast component of desensitizationis caused by ambient zinc. The time course of this fast compo-nent reflects binding of ambient zinc to the extracellular zincsite in the amino terminal domain (ATD) after binding of glu-tamate to the agonist site in the S1/S2 domain, as binding ofglutamate increases zinc affinity through an allosteric interac-tion of the two sites. We show a similar interaction betweenthe ifenprodil and glutamate binding site for receptors com-prising NR1/NR2B subunits (see also refs. 21, 22). We proposethat such an allosteric interaction might exist for other mem-bers of the glutamate receptor family and may be involved inthe regulation of glutamate receptor function by endogenousmodulators and compounds used therapeutically.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


RESULTSEDTA alters desensitization of NR1/NR2A receptorsWe investigated the desensitization of recombinant NR1/NR2Areceptors under conditions that favor the glycine-independentdesensitization. We use the term ‘desensitization’ to describe thereduction in the amplitude of macroscopic currents during a pro-longed application of glutamate. In the internal solution forwhole-cell patch-clamp recording, we used a strong calcium buffer(5 mM BAPTA) that prevents calcium-dependent inactivation17,18.Glycine (at least 20× median effective concentration (EC50) forNR1/NR2A), a co-agonist for NMDA receptors, was added to allsolutions to prevent glycine-dependent desensitization. The degreeand time course of desensitization of the inward and outward cur-rents were indistinguishable (Fig. 1a). The desensitization ofNMDA receptor current could not be fitted with a single expo-nential component (Fig. 1b, top); it was best fitted with two expo-nential components (Fig. 1b, bottom; τ1 = 0.27 ± 0.04 s, τ2 = 1.86 ± 0.26 s). On average, χ2 was also reduced by 2.5 ± 0.6-fold by introducing the second exponential component (n = 9).Our data confirmed the presence of two distinct kinetic compo-nents of desensitization of NR1/NR2A receptors, as reported pre-viously15,19,20. We also confirmed the previous finding14,20 thatthis fast component is not consistently present when the C-ter-minal of NR1 is truncated (n = 7, data not shown).

NR1/NR2A receptors are inhibited in a voltage-independentmanner by zinc that acts at a high-affinity extracellular site1.The median inhibitory concentration (IC50) of zinc for thishigh-affinity site23–25 of NR1/NR2A is 10–80 nM; thus,NR1/NR2A receptors are tonically inhibited by ambient zinc(~300 nM) in the recording solution. Removal of ambient zinc

by 10 µM EDTA not only enhanced the peak currents, but alsoreduced the desensitization of NR1/NR2A (Fig. 2a and b). Theratio of the steady-state current (Iss) over the peak current (Ipk)was increased from 0.47 ± 0.06 to 0.75 ± 0.05 by addition ofEDTA (Fig. 2c, n = 9), indicating a reduction in desensitization.In the presence of EDTA, the desensitization of NR1/NR2A wasbest fitted with a single exponential decay (τ = 1.62 ± 0.20 s,Fig. 2d). Introducing a second exponential component intocurve fitting for desensitization in the presence of EDTA result-ed in no significant reduction of the Z value from the runs testor the χ2 value. These data suggest that the fast component (τ = 0.27 s) of desensitization of NR1/NR2A receptors is selec-tively abolished by the metal chelator EDTA. Furthermore, thesedata also suggest that zinc does not accelerate the slow compo-nent of desensitization (Fig. 2d).

EDTA is also a chelator for calcium, and could potentiallyreduce the decay of NR1/NR2A currents by disrupting calcium-dependent inactivation. However, we used a low con-centration of EDTA (10 µM), which reduced the free extracel-lular calcium concentration by less than 1%, but reduced thefree extracellular zinc concentration to the subpicomolar range(Winmax26). Furthermore, we recorded under conditions thatprevented the calcium-dependent inactivation. Followingremoval of extracellular ambient zinc with 10 µM EDTA, wefound a similar degree of desensitization in calcium-contain-ing and nominally calcium-free external solution (Iss/Ipk = 0.78 ± 0.07, nominally 0 Ca2+; Iss/Ipk = 0.76 ± 0.05,

Fig. 1. Desensitization of recombinant NR1/NR2A receptors hastwo kinetic components. (a) Typical current traces recorded fromHEK 293 cells expressing NR1/NR2A receptors. A rapid perfusionsystem applied 100 µM glutamate for 5 s. Glycine (30–60 µM) waspresent all the time. Ambient zinc in the recording solution was esti-mated to be 300 nM24. Similar desensitization was observed for boththe inward (Vh = –50 mV) and outward current (Vh = +50 mV). (b) Fitted curves are plotted as smooth lines; data points are shownas dots. The desensitization was poorly fitted with a single exponen-tial component (top, τ = 963 ms; runs test, Z = 12.78). It was fittedwith two exponential components (inset, bottom panel, τ1 = 205 ms,τ2 = 2233 ms; runs test, Z = 1.499).

Fig. 2. EDTA abolishes the fast component of NR1/NR2A receptordesensitization. (a) Typical current traces in the presence and absenceof EDTA (10 µM) were recorded from the same HEK 293 cell express-ing NR1/NR2A (Vh = –50 mV; 100 µM glutamate, 3 s). (b) Fitted curveswere plotted as smooth lines superimposed over the actual data (plot-ted as dots). Without EDTA, the desensitization was fitted with twoexponential functions (τ1 = 160 ms, τ2 = 1.85 s). In the presence ofEDTA, the desensitization was fitted with a single exponential function(τ = 1.80 s). (c) The degree of desensitization was shown as the ratio ofthe steady state current measured at the end of the glutamate applica-tion (3–5 s) over the peak current (*p < 0.01, n = 9; **p < 0.05, n = 9).(d) Time constants were obtained through curve fitting of the outwardcurrents recorded at +40 or +50 mV (n = 9). In the absence of EDTA,the desensitization time course was fitted by two exponential compo-nents. In the presence of EDTA, the desensitization was fitted satisfac-torily with a single exponential component. The time constant fordesensitization in EDTA was not significantly different from the timeconstant for the slow component of desensitization in the presence ofambient zinc (p > 0.1).

a b

a b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

Fig. 3. Tricine-buffered zinc causes fast desensitizationof recombinant and neuronal NR1/NR2A receptors. (a) NMDA receptor currents were recorded from thesame HEK293 cell transiently transfected withNR1/NR2A at both positive and negative holding poten-tials (+50, –50 mV; 100 µM glutamate, 3 s). Free zincconcentrations are shown. (b) Current in tricine-buffered zinc was best fit by two exponential compo-nents (0.20 s and 2.8 s), whereas current in the absenceof zinc was fit by one component (1.6 s). (c) NMDA cur-rents in the same cultured cerebellar granule neuronmaintained in 25 mM K+ (Vh = +50 mV, pH 7.3; 1 mMNMDA, 5 s). Glycine (50 µM) was present at all times.(d) Normalized ifenprodil and ifenprodil/Zn2+ tracesfrom (c). Tricine-buffered zinc caused a current relax-ation (τfast = 221 ± 33 ms from 2 component exponen-tial fits, n = 6) similar to the fast desensitization ofrecombinant NR1/NR2A receptors. (e) The mean Iss/Ipkfor ifenprodil/Zn2+ was significantly different from bothifenprodil and washout (ANOVA with Tukey post hoctest, p < 0.01, n = 6). There was no statistically significantdifference among control, ifenprodil and washout condi-tions (p > 0.05).

1.8 mM extracellular Ca2+; n = 3, Vh = –50 mV). Furthermore,with 1.8 mM extracellular calcium, the Iss/Ipk for inward cur-rents was similar to the Iss/Ipk for outward currents (0.70 ± 0.06,Vh = –50 mV; 0.75 ± 0.06, Vh = 50 mV; p > 0.1, n = 7). Thesedata suggest that the reduction of the degree of desensitizationby EDTA that we observed is not likely due to alteration of cal-cium-dependent inactivation of these receptors. Rather, theselective removal of the fast component of glycine-independentdesensitization by EDTA suggests that this fast component maybe caused by ambient zinc.

Zinc causes fast desensitization of NMDA receptorsTo confirm that extracellular zinc can cause the fast desensiti-zation of NR1/NR2A receptors, we used tricine buffer to obtainthe desired free zinc concentration as described previously23,24.The desensitization of NR1/NR2A receptors in the absence offree zinc was again best fitted with a single exponential decay(Fig. 3a and b). In the presence of tricine-buffered zinc, an addi-tional fast component of desensitization appeared (Fig. 3aand b; see also Fig. 6c), and the overall degree of desensitiza-tion was increased (n = 3–5 for each free zinc concentrationtested; Fig. 5b). Thus, the EDTA-sensitive fast component ofdesensitization of recombinant NR1/NR2A receptors can berestored by elevating the extracellular free zinc concentration.

Fast desensitization of NMDA receptors in neurons has beenwell documented1,2 and has typically been described as glycine-independent desensitization or calcium-dependent inactivation.To determine whether zinc could also cause fast desensitizationof neuronal NMDA receptors, we recorded from cultured cere-bellar granule neurons, using perforated patch recording to elim-inate dialysis as a confounding variable. Previous analysis ofmRNAs shows that the predominant NR2 subunit expressed inthese neurons after a week in high potassium media is NR2A27.We confirmed that NR2A is the predominant subunit, based onpharmacological properties of NMDA currents. NMDA currentsshowed little sensitivity to 3 µM ifenprodil (peak current, 85.6 ± 6.8% of control, n = 6; Fig. 3c). This concentration wassufficient to block 75 ± 6% of the NMDA current in granuleneurons that were maintained in media not supplemented withhigh potassium (n = 3), and that were presumably expressingpredominantly NR2B-containing receptors (data not shown).

Granule neurons maintained in high potassium also showedhigh sensitivity to zinc. The peak current was reduced by 48.6± 5.1% in the presence of 100 nM free zinc (tricine-buffered;Fig. 3c, n = 6), suggesting the presence of high-affinity voltage-independent zinc inhibition23,24. At the same concentration, zincalso caused a fast desensitization of the outward NMDA cur-rents in these granule neurons (Fig. 3c–e) that was similar to thefast desensitization of recombinant NR1/NR2A receptors. Fur-thermore, the enhanced desensitization was reversible uponwashout of zinc (Fig. 3c and e). Taken together, our data sug-gest that zinc causes fast desensitization of neuronal NMDAreceptors when NR2A is the predominant NR2 subunit.

Zinc acts at the extracellular site in the ATDPrevious studies have identified the amino terminal domain ofNR2A as the location for the high-affinity zinc site that isresponsible for the voltage-independent inhibition by zinc28–31.If the fast component of desensitization is caused by zinc act-ing at this site, it should be disrupted by mutations that disruptthe voltage-independent zinc inhibition of NR1/NR2A recep-tors. We selected two histidine mutations (H44G and H128A)that greatly reduce the zinc affinity28–30, and we tested the effectsof these two mutations on the desensitization of NR1/NR2Areceptors. Both histidine point mutations significantly alteredthe desensitization of NR1/NR2A receptors in the presence ofambient zinc (Fig. 4a). The degree of desensitization ofNR1/NR2A(H44G) receptors or NR1/NR2A(H128A) receptorswas significantly less than the desensitization of wild-typeNR1/NR2A receptors in the presence of ambient zinc, but wasidentical to the desensitization of wild-type receptors in thepresence of 10 µM EDTA (Fig. 4b). The onset of desensitiza-tion of mutant receptors was best fitted with a single exponen-tial component, with a time constant comparable to thatobserved in the presence of EDTA for the wild-type NR1/NR2Areceptors (Fig. 4c). The elimination of the fast component ofdesensitization by mutations that disrupt high-affinity zinc inhi-bition supports our hypothesis that binding of zinc to the high-affinity site in the amino terminal domain causes fastdesensitization. Henceforth, we refer to the fast desensitizationcaused by zinc binding to the high-affinity site in the aminoterminal domain as ‘zinc-induced desensitization.’

a b

c d e

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


ing sites could be underestimated. Given these limitations, westill observed a 2.5-fold difference in the zinc IC50 for the peakand steady-state currents. The proposed allosteric model impliesreciprocity in that there should be a similar magnitude differ-ence in the affinity of glutamate for the zinc-bound versus thezinc-unbound receptor. Consistent with previous reports28–30,the glutamate EC50 was 4.4 µM in the presence of EDTA, and2.35 µM in the presence of 1 µM zinc as measured in oocytes (n = 15, data not shown). The IC50 for the peak currents is nota direct measurement for the zinc affinity of glutamate-unboundreceptors, as the peak is determined by the net balance of twoprocesses: the rate of fully liganded receptors going into the openstate, and the rate of the fully liganded receptors going into thezinc-bound, high-affinity state. Therefore, the IC50 for peak cur-rents is partially influenced by the higher-affinity state, and wouldbe an overestimation for the true zinc affinity of glutamate-unbound receptors. The true degree of the allosteric interaction

Fig. 5. The degree and the time course of the zinc-induced fast desensiti-zation of NR1/NR2A receptors depend on extracellular free zinc concen-tration. (a) A rapid perfusion system applied a 400-ms pulse of 100 µMglutamate. Tricine buffer was used to achieve the desired free zinc con-

centrations. The maximal currents for the peak and steadystate were defined as the peak and steady-state currents innominally zinc-free solutions (10 mM tricine without addedzinc). For the peak currents, the zinc IC50 was 97 nM. Forthe steady-state currents, the zinc IC50 was 37 nM. The Hillslopes were 0.92 and 1.09 for the peak and steady-statecurrents, respectively (n = 4–7). (b) The predicteddose–response curve for the degree of zinc-dependentdesensitization (smooth curve) was in agreement with datameasured using a 400–500 ms glutamate pulse (circles, n = 5–9 for each free zinc concentration). (c) Normalizedsample current traces recorded in the same cell in the pres-ence of free extracellular zinc were shown on an expandedtime scale. Scale bar, 100 ms; 15, 12 and 9 pA for currentsrecorded in the presence of 23, 223 and 2230 nM zinc,respectively). Tricine buffer was used to achieve desiredfree zinc concentrations. (d) The time constant of the fastdesensitization depended on the free zinc concentration. Ashort glutamate pulse was used to minimize slow glycine-independent desensitization. The onset of desensitizationwas then fitted with a single exponential component (n = 3–12 for each concentration of free zinc tested; r, cor-relation efficiency).

The zinc and glutamate sites are allosterically coupledHow could zinc act at an extracellular site to cause desensitiza-tion of NR1/NR2A receptors? One possible mechanism is that thefast desensitization of NR1/NR2A receptors is caused by anallosteric interaction between the zinc and glutamate binding sites.Our hypothesis is that zinc binds to NR1/NR2A receptors with alow affinity in the absence of glutamate, and binds to the receptorswith a high affinity in the presence of glutamate. By definition,allosteric interaction32 dictates that glutamate binds NR1/NR2Areceptors with higher affinity in the presence of zinc. Because thezinc-bound/glutamate-bound state is more stable than either thezinc-bound state or the glutamate-bound state, more receptorswill gradually enter this state after a concentration jump into glu-tamate as the system relaxes to a new equilibrium. This re-equili-bration of zinc will result in the time-dependent reduction ofwhole-cell NR1/NR2A receptor current.

We compared the zinc sensitivity for the peak and steady-statecurrents of NR1/NR2A receptors (Fig. 5a). If the zinc and gluta-mate binding sites are allosterically coupled, the steady-state cur-rent will be more sensitive to zinc than the peak current. Tominimize the amount of the slow zinc-independent desensitiza-tion, we applied glutamate for only 400–500 ms. We chose thisshort protocol because separation of the two components (thefast zinc-induced component and the slow zinc-independentcomponent) by curve fitting would fail at very low and very highconcentrations of free zinc. With this short protocol, whole-cellcurrents do not reach steady state at lower free zinc concentra-tions. Subsequently, the zinc IC50 value determined for steadystate currents may be overestimated, and thus, the degree of pos-sible allosteric interaction between the zinc and glutamate bind-

Fig. 4. Histidine point mutations in the amino terminal domain of NR2Aabolish the fast desensitization of NR1/NR2A receptors. (a) Sample tracesof wild-type and mutant NR1/NR2A receptor currents recorded fromHEK293 cells in the presence of ambient zinc (Vh = +50 mV; 100 µM glutamate, 5 s). The fast component of desensitization was absentfor NR1/NR2A(H44G) and NR1/NR2A(H128A). Horizontal scale bar, 1 s;vertical scale bar, 500, 160 and 800 pA for wild type, H44G and H128A,respectively. The Iss/Ipk (b) and time constants (c) for wild-type (n = 7),H44G (n = 5) and H128A (n = 5) receptors were determined for the out-ward currents (Vh = +50 mV; **p < 0.05, ANOVA with Tukey post hoc test).

a

b c

a b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

esis implies that the amount of residual current during full occu-pation of the zinc binding site is pivotal in determining the degreeof zinc-induced desensitization (Eq. 5), the degree of zinc-induced desensitization should be pH-dependent in the samemanner. At alkaline pH values, the degree of zinc-induced desen-sitization would be greatly diminished, because the lack of freeprotons prevents zinc/glutamate-bound receptors from beinginhibited. At lower pH values, the increased free proton concen-tration leads to reduced residual currents, and therefore greaterdegree of zinc-induced desensitization.

Indeed, the degree of zinc-induced desensitization showedthe predicted pH-dependency. The zinc dose–response curve forsteady-state currents in HEK293 cells (Fig. 6a) showed that theresidual current in the presence of saturating zinc concentrationwas smaller at more acidic pH values. The degree of desensitiza-tion predicted using these residual current values (Fig. 6b and c)was in agreement with the observed values.

Ifenprodil produces desensitization of NR1/NR2B Our data suggest that the amino terminal domain and the glu-tamate-binding domain of NR2A interact, causing the fast desen-sitization of NR1/NR2A receptor currents in the presence ofsubmaximal levels of zinc. Because homologous amino terminaldomains are present in all glutamate receptor subunits, it is pos-sible that the allosteric interaction between the amino terminaldomain and glutamate-binding domain is a general form of reg-ulation of glutamate receptor function. Data from chimeric recep-tors and point mutations suggest that the ifenprodil binding sitemight be located in the amino terminal domain of NR2B33,34.Furthermore, zinc inhibition of NR2A-containing receptors andifenprodil inhibition of NR2B-containing receptors share thesame mechanism, enhancement of tonic proton inhibition atphysiological pH22,28,29. By drawing analogy to the effects of zincon NR2A, one would predict that submaximal ifenprodil shouldproduce desensitization of macroscopic NR1/NR2B receptor cur-rents. Thus, we examined the effects of submaximal concentra-tions of ifenprodil on the relaxation of whole-cell currents ofNR1/NR2B receptors.

Under control conditions, there isonly minimal desensitization forNR1/NR2B receptors with an averagedIss/Ipk of 0.90 ± 0.02 (n = 6, Fig.7a andb). Ifenprodil (270 nM) induced a time-dependent relaxation in current respons-es. As a result of this time-dependentrelaxation, the Iss/Ipk was significantlyreduced (Fig. 7b). Previous studies haveshown that a point mutation in theamino terminal domain of NR2B,E201R, reduces ifenprodil binding33,34.The desensitization caused by ifenprodilwas abolished by the same point muta-tion (Fig.7c and d). These data are con-

is likely to be greater than the 2.5-fold difference indicated by thedifferences in zinc IC50 for the peak and steady-state currents.

If the underlying mechanism for the zinc-induced desensiti-zation is an allosteric interaction that occurs when glutamatebinds to the S1/S2 domain of NR2A and increases affinity forzinc, then one would predict that the degree of zinc-induceddesensitization would be less at very low and very high concen-trations of free zinc. At very low concentrations, there is notenough free zinc to substantially alter the occupancy of the highaffinity zinc site in the amino terminal domain. At very high con-centrations, zinc would saturate all binding sites even before theapplication of glutamate, reducing the impact of enhancementof zinc affinity by glutamate. Thus, the predicted dose–responsecurve for the degree of zinc-induced desensitization would bebell-shaped (Fig. 5b, Eq. 5). This predicted curve is in generalagreement with the degree of desensitization measured at vari-ous concentrations of free zinc (Fig. 5b).

If the onset of the fast desensitization of NR1/NR2A receptorsreflects zinc binding, then the time constant for this componentshould be dependent on the free zinc concentration, and from theconcentration dependence of the time constant, we can calculatethe dissociation constant (Kd) for zinc. Indeed, the time constantshowed strong dependence on the free zinc concentration (p < 0.001; Fig. 5c and d). Based on the linear regression of datapresented in Fig. 5d (Eq. 6), the kon and koff rates for zinc are 3.22 × 107/M/s and 1.35/s, respectively (see Methods). From thesemicroscopic rate constants, we calculate that the Kd for zinc is 42 nM, which is in agreement with the zinc IC50 determined forthe steady-state currents (37 nM; Fig. 5a).

Zinc-induced desensitization is pH-dependentZinc, acting at the high-affinity site in the amino terminal domainof NR2A, inhibits NR1/NR2A receptors by enhancing tonic pro-ton inhibition28,29. Such a mechanism provides accurate predic-tion for the pH-dependency of the residual currents observedwhen the high-affinity zinc site is saturated29. The residual cur-rents are smaller at lower pH values, at which more free protonsare available to inhibit the zinc-bound receptors. As our hypoth-

Fig. 6. pH influences zinc-induced fast desensitization of NR1/NR2A receptors. (a) The zincdose–response relationship was determined for steady-state current at the end of a 5-s applica-tion of glutamate, normalized to the steady-state current in the absence of zinc (10 mM tricineand 1 µM EDTA) for each cell. For pH 7.8, IC50, 41 nM; Hill slope, 1.2; residual current, 0.59. ForpH 6.8, IC50, 30nM; Hill slope, 1.3; residual current, 0.17 (n = 7–13 for each condition). (b) Thepredicted dose–response curves for the degree of zinc-induced desensitization (smooth curves)at pH 7.8 and 6.8 were in agreement with observed data (circles and squares) measured using a5-s application of glutamate (n = 7–13 for each free zinc concentration). (c) Normalized tracesfrom one cell, demonstrating the pH-dependency of zinc-induced desensitization.

a

b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


sistent with the presence of an allosteric interaction betweenthe amino terminal domain and the glutamate-binding domainof NR2B (see also refs. 21, 22) that is analogous to the allostericinteraction between the amino terminal domain and the glu-tamate-binding domain of NR2A.

DISCUSSIONRecombinant NR1/NR2A receptors have been used extensively asa model system to investigate the underlying mechanism ofdesensitization of NMDA receptors, because NR1/NR2A recep-tor and NMDA receptor forms of desensitization are thought toclosely resemble one another. In particular, only receptors com-prising NR1/NR2A subunits exhibit a fast onset of glycine-inde-pendent desensitization similar to that of neuronal NMDAreceptors. In this study, we demonstrated by four lines of evi-dence that this fast component of the glycine-independentdesensitization of NR1/NR2A is likely caused by extracellularambient zinc, and the time course of the relaxation of macro-scopic currents reflects zinc binding to the extracellular high-affinity site. First, removal of ambient zinc by the metal chela-tor EDTA selectively abolishes this fast component without anyeffect on the slow component of glycine-independent desensi-tization. Second, tricine-buffered zinc reproduces the fastglycine-independent desensitization of NR1/NR2A. Third, muta-tions that disrupt the voltage-independent zinc inhibition alsoabolish the fast desensitization. Fourth, the Kd for zinc deter-mined from kinetic analysis of the concentration dependence ofthe onset time constants of the fast desensitization is identicalto the zinc IC50 for the amplitudes of steady-state currents. Thelast point is perhaps the strongest evidence in support of ourhypothesis. Thus, the fast component of the glycine-indepen-dent desensitization of NR1/NR2A receptors (τ = 0.3 s) is mech-

anistically distinct from the slow component (τ = 1.8 s). Itrequires not only the presence of the agonist, but also the pres-ence of an allosteric modulator, that is, zinc. Our results furtherextend the model for desensitization of neurotransmitter recep-tors. In addition to the fast glycine-dependent desensitizationof NMDA receptors that is caused by an allosteric interactionbetween the agonist (glutamate) and co-agonist (glycine) bind-ing sites3,4, our data suggest that desensitization could also becaused by an allosteric interaction between the binding sites forthe agonist and an allosteric modulator.

This zinc-induced desensitization has caused some confu-sion in the literature. Because the onset kinetics depend on thefree zinc concentration, variations in the amount of ambientzinc would make this fast component variable and at times dif-ficult to separate from the slow component of glycine-indepen-dent desensitization. For example, a single component with atime constant of 700 ms has been observed under nominally cal-cium-free conditions16. This time constant is significantly fasterthan the slower component of glycine-independent desensitiza-tion (t = 1.8 s), suggesting that they may have observed a mixtureof the zinc-dependent component and the slow glycine-inde-pendent component. The prominent fast desensitizationdescribed as ‘calcium-dependent inactivation’ in the presenceof 5 mM intracellular EGTA has been reported to disappear inthe presence of extracellular EDTA (1 mM)20, which is used toremove extracellular free calcium, but also removes ambientzinc. C-terminal truncation of NR1 may alter zinc or pH-sen-sitivity of NR1/NR2A receptors (F.Z., unpublished data). Sub-tle changes in zinc or pH sensitivity could alter the degree andkinetics of zinc-induced desensitization, making it more diffi-cult to detect through curve fitting14.

The zinc-induced desensitization of NR1/NR2A receptorscould be described as a defined sequence of events involvingtwo specific structural domains of NR2A, the ligand bindingdomain and the amino terminal domain. Previous studies have

Fig. 8. A model for an allosteric interaction between the amino termi-nal domain and S1/S2 domain of NR2. We hypothesize that the desensi-tization caused by zinc results from the following events. (1) Glutamatebinds to the S1-S2 domain of NR2A. (2) Glutamate binding leads toallosteric changes in the amino terminal domain that alter zinc affinity.(3) As the system relaxes into a new equilibrium, the occupancy of thezinc binding site increases in a time-dependent manner. (4) Zinc bindingto the amino terminal domain of NR2A causes conformational changesof the receptor that enhance binding of protons to the pH-sensitive gat-ing elements28,29, shifting more receptors into closed states35.

Fig. 7. Ifenprodil produces desensitization of NR1/NR2B receptors. (a) Typical current traces in the presence and absence of ifenprodil froma HEK 293 cell expressing NR1/NR2B. A rapid perfusion system applieda 10-s pulse of 100 µM glutamate. Glycine (20 µM) and ifenprodil (270nM) were added to both the glutamate and wash solution (Vh = –50mV). (b) The degree of desensitization was determined by the ratio ofthe steady state current measured at the end of a long glutamate pulse(10 s) to the peak current. Ifenprodil caused desensitization ofNR1/NR2B receptors (*p < 0.01, n = 6). (c) Sample traces of mutantNR1/NR2B(E201R). Recording conditions were identical to that for thewild-type NR1/NR2B receptors. Ifenprodil (270 nM) had no effect ondesensitization. (d) Desensitization of NR1/NR2B(E201R) receptorswas identical in the presence and absence of ifenprodil (p > 0.01, n = 4).

a b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

proposed two functional roles for the amino terminal domain ofNR2A: involvement in the fast component of desensitization ofNR1/NR2A receptors14,15 and location of the high-affinityextracellular zinc site28–31. Our data demonstrate that involve-ment of the amino terminal domain in the fast desensitizationis the functional consequence of the zinc-binding site in theamino terminal of NR2A. Although the amino terminal domainand the glutamate-binding domain of NR2A (that is, S1/S2domain) are modular, they interact to produce use-dependentregulation of NMDA receptor function (Fig. 8). Specifically, wepropose that the zinc-induced desensitization is due to the fol-lowing molecular events. First, glutamate binds to the S1/S2domain of NR2A. Second, glutamate binding leads to allostericchanges of the amino terminal domain of NR2A, increasing theaffinity of the zinc binding site. Third, as the system relaxes intoa new equilibrium, the occupancy of the zinc binding siteincreases in a time-dependent manner. Fourth, zinc binding tothe amino terminal domain of NR2A causes conformationalchanges of the receptor that enhance binding of protons to thepH sensitive gating elements28,29, shifting more receptors intoclosed states35. In this model, the zinc-induced desensitizationresults from re-equilibration of zinc binding to the NMDAreceptors and subsequent enhancement of tonic proton inhibi-tion. The time course of zinc-induced desensitization repre-sents the rate-limiting step that is likely zinc binding, asglutamate binding36 and protonation35 both occur more rapid-ly than the 270-ms time constant measured for the fast com-ponent of desensitization. A similar sequence of events mayoccur for ifenprodil-induced desensitization of NR2B.

It is possible that our finding of an allosteric interactionbetween the amino terminal domain and the glutamate-bindingdomain of NR2A and NR2B is subunit-specific. However, it maywell extend to other members of glutamate receptor family. Theamino terminal domain of glutamate receptor subunits beyondNR2A and NR2B may contain a binding site for other extracel-lular regulators, and such regulatory sites could be allostericallycoupled to the agonist-binding site in the S1/S2 domain. At pre-sent time, this hypothesis could not be tested for different sub-types of glutamate receptors because little is known aboutwhether ligands exist for the amino terminal domain of variousglutamate receptor subunits. However, the multiplicity of com-mon mechanistic features in regulation of receptor function bythe amino terminal domain of NR2A and NR2B make it unlike-ly that the allosteric interaction between the amino terminaldomain and the agonist-binding domain reported here is inci-dental. Rather, it seems to be a fundamental part of the regula-tion of NMDA receptor function.

METHODSTransfection of HEK cells. HEK 293 cells (CRL 1573; ATCC, Rockville,Maryland) were maintained at 37° and 5% CO2, as described previous-ly24. Low-confluency cells were transfected by the calcium phosphate pre-cipitation method37, with a mixture containing NR1-1a, NR2A (or NR2B)and GFP38 plasmids (1, 2 and 0.3 µg per 12-mm diameter coverslip, respec-tively). After transfection, NMDA antagonists (100–200 µM AP5, 2 mMMg2+, 5–10 mM kynurenic acid) were added to the culture medium.

Buffered zinc solutions. The tricine-buffered zinc solutions used to obtainthe zinc dose–response curves were prepared according to the empiri-cally established binding constant 10–5 M as described previously23,24.

Whole-cell patch-clamp recording from HEK 293 cells. Patch-clamprecording in the whole-cell configuration39 was made with an Axopatch200B amplifier (Axon Instruments, Union City, California) or a PC501Aamplifier (Warner, Hamden, Connecticut). Recording electrodes

(5–12 MΩ) were filled with 140 mM Cs-gluconate, 5 mM HEPES, 4 mMNaCl, 2 mM MgCl2, 0.5 mM CaCl2, 1 mM ATP, 0.3 mM GTP and 5 mMBAPTA (pH 7.4, 23°C). The recording chamber was continually perfusedwith recording solution composed of 150 mM NaCl, 10 mM HEPES, 1mM CaCl2, 3 mM KCl and 10–20 mM mannitol (pH 7.4 unless other-wise noted). Glutamate (100 µM) was applied using a multibarrel pipettedriven by a piezo-electric bimorph40 or a nanostepper (SF77B, Warner)with an exchange time of 0.5 ms and 2–8 ms, respectively. Glycine (30–60 µM) was present all the time. Data used for analysis were collect-ed within 5–15 min after initial break-in to minimize time-dependentchange of the glycine-independent desensitization. In some experiments,series resistance was corrected off-line41. Correction of series resistancedid not alter the zinc-induced current relaxation.

Whole-cell perforated-patch recording from cerebellar granule cells.Cerebella from 4- to 7-day postnatal Sprague–Dawley rats were isolated,passed through a 210 µM nylon mesh and plated onto glass coverslipscoated with 5 µg/mL poly-D-lysine (approved by Institutional AnimalCare and Use Committee of Emory University). Cultures were maintainedfor 6–8 days at 37°C and 5% CO2 in DMEM supplemented with L-gluta-mine (0.2 mM), pyruvate (0.1 mM), penicillin/streptomycin (100 units/mL), 10% fetal bovine serum and 25 mM KCl. Recording elec-trodes (5–9 MΩ) were filled with the same solution used for HEK293 cellrecording with 25 µg/mL gramicidin (Sigma, St. Louis, Missouri). It took20–30 min to achieve acceptable perforation with series resistances rang-ing from 15 to 40 MΩ. In the continued presence of glycine (50 µM),NMDA (1 mM) was applied by local perfusion through a capillary tube(1.1 mm inner diameter) positioned near the cell. The solution flow wasdriven by gravity (flow rate, 1–5 ml/min) and controlled by solenoid valves(Lee, Westbrook, Connecticut). Series resistance correction was not appliedgiven the low amplitude of typical currents (∼ 100 pA).

Curve fitting. The time course of desensitization was fitted with one ortwo exponential components with NPM (S.F. Traynelis, Emory Univ.)using the following equation.

A(t) = A0 + ∑n An exp (–t/τn) (1)

Here, A0 is the offset; An and τn are the amplitudes and time constantsfor each exponential components.

The zinc IC50 was determined by fitting the dose–response curve tothe following equation.

I/Imax = (1 – a)/(1 + ([Zn2+]/IC50)n) + a (2)

Here, a and n are the residual and the Hill slope, respectively.For a given free zinc concentration, the following equations apply.

Iss = Imax((1 – a)/(1 + ([Zn2+]/IC50,ss)n

ss) + a) (3)

Ipeak= Imax((1 – a)/(1 + ([Zn2+]/IC50, peak)npeak) + a) (4)

Therefore, the degree of zinc-dependent desensitization, 1 – Iss/Ipk, can becalculated by the following equation.

1 – Iss/Ipk = 1 – ((1 – a)/(1 + ([Zn2+]/IC50,ss)n

ss) + a)/((1 – a)/(1 + ([Zn2+]/IC50,peak)n

peak) + a) (5)

To predict the desensitization for Fig. 6b, an additional offset term wasadded to Eq. 5 to account for the slow desensitization observed in theabsence of zinc. This empirically determined value was 0.20 ± 0.02 (n = 30) for pH 6.8 and 0.14 ± 0.02 (n = 36) for pH 7.8.

Assuming that zinc binds with the extracellular domain of NMDAreceptor at a single site, the onset of zinc binding could be described bythe following equation.

1/τon = kon[Zn2+] + koff (6)

Here, kon and koff are the association and dissociation rate constants,respectively. These two rate constants could be estimated based on lin-ear regression of Fig. 5d.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


Statistics. All pooled data are expressed as mean ± s.e.m. Unpaired Stu-dent’s t-test was used unless stated otherwise. Quality of fits with one ortwo exponential components was assessed using χ2 and runs test42. Thecritical Z value was 1.96 for the random distribution (that is, symmetri-cal distribution above and below the fitted curve; n > 100, α = 0.05). If theZ from runs test was less than the critical Z value, the fit was consideredas good. If the Z was greater than the critical Z value, the fit was rejected.

ACKNOWLEDGEMENTSWe thank M.L. Mayer, J. Neyton and P. Paoletti for reading the manuscript.

S. F. Heinemann provided NR1 and NR2B. S. Nakanishi provided NR2A. D.

Lynch and E. Aizenman provided NR2B(E201R). This work is supported by

grants from NINDS (NS 39418 to F.Z., NS 36654 to S.F.T. and NS 31373 and

NS 34876 to P.J. Conn), HHMI (K.E.), the Benzon Society (T.B.) and a Young

Investigator Award from NARSAD to F.Z.


1. Dingledine, R., Borges, K., Bowie, D. & Traynelis, S. F. The glutamate receptorion channels. Pharmacol. Rev. 51, 7–61(1999).

2. McBain, C. J. & Mayer, M. L. N-methyl-D-aspartic acid receptor structureand function. Physiol. Rev. 74, 723–760 (1994).

3. Mayer, M. L., Vyklicky, L. Jr. & Clements, J. Regulation of NMDA receptordesensitization in mouse hippocampal neurons by glycine. Nature 338,425–427 (1989).

4. Benveniste, M., Clements, J., Vyklicky, L. & Mayer, M. L. Kinetic analysis ofthe modulation of N-methyl-D-aspartic acid receptors by glycine in mousecultured hippocampal neurones. J. Physiol. (Lond.) 428, 333–357 (1990).

5. Sather, W., Johnson, J. W., Henderson, G. & Ascher, P. Glycine-insensitivedesensitization of NMDA responses in cultured mouse embryonic neurons.Neuron 4, 725–731 (1990).

6. Sather, W., Dieudonne, S., Macdonald, J. F. & Ascher, P. Activation anddesensitization of N-methyl-D-aspartate receptors in nucleated outside-outpatches from mouse neurones. J. Physiol. (Lond.) 450, 643–672 (1992).

7. Tong, G. & Jahr, C. E. Regulation of glycine-insensitive desensitization of theNMDA receptor in outside-out patches. J. Neurophysiol. 72, 754–761 (1994).

8. Tong, G., Shepherd, D. & Jahr, C. E. Synaptic desensitization of NMDAreceptors by calcineurin. Science 267, 1510–1512 (1995).

9. Clark, G. D., Cliford, D. B. & Zorumski, C. F. The effect of agonistconcentration, membrane voltage and calcium on N-methyl-D-aspartatereceptor desensitization. Neuroscience 39, 787–797 (1990).

10. Vyklicky, L. Calcium-mediated modulation of N-methyl-D-aspartate(NMDA) responses in cultured rat hippocampal neurones. J. Physiol. (Lond.)470, 575–600 (1993).

11. Legendre, P., Rosenmund, C. & Westbrook, G. L. Inactivation of NMDAchannels in cultured hippocampal neurons by intracellular calcium. J. Neurosci. 13, 674–684 (1993).

12. Rosenmund, C. & Westbrook, G. L. Calcium-induced actin depolymerizationreduces NMDA channel activity. Neuron 10, 805–814 (1993).

13. Rosenmund, C. & Westbrook, G. L. Rundown of N-methyl-D-aspartatechannels during whole-cell recording in rat hippocampal neurons: role ofCa2+ and ATP. J. Physiol. (Lond.) 470, 705–729 (1993).

14. Krupp, J. J., Vissel, B., Heinemann, S. F. & Westbrook, G.L. N-terminaldomains in the NR2 subunit control desensitization of NMDA receptors.Neuron 20, 317–327 (1998).

15. Villarroel, A., Regalado, M. P. & Lerma, J. Glycine-independent NMDAreceptor desensitization: localization of structural determinants. Neuron 20,329–339 (1998).

16. 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).

17. Krupp, J. J., Vissel, B., Heinemann, S. F. & Westbrook, G. L. Calcium-dependent inactivation of recombinant N-methyl-D-aspartate receptors is

NR2 subunit specific. Mol. Pharmacol. 50, 1680–1688 (1996).18. Krupp, J. J., Vissel, B., Thomas, C. G., Heinemann, S. F. & Westbrook, G. L.

Interactions of calmodulin and alpha-actinin with the NR1 subunit modulateCa2+-dependent inactivation of NMDA receptors. J. Neurosci. 19, 1165–1178(1999).

19. Medina, I. et al. Calcium-dependent inactivation of heteromeric NMDAreceptor-channels expressed in human embryonic kidney cells. J. Physiol.(Lond.) 482, 567–573 (1995).

20. Zhang, S., Ehlers, M. D., Bernhardt, J. P., Su, C. T. & Huganir, R. L.Calmodulin mediates calcium-dependent inactivation of N-methyl-D-aspartate receptors. Neuron 21, 443–453 (1998).

21. Kew, J. N. & Kemp, J. A. An allosteric interaction between the NMDAreceptor polyamine and ifenprodil sites in rat cultured cortical neurones. J. Physiol. (Lond.) 512, 17–28 (1998).

22. Mott, D. D. et al. Phenylethanolamines inhibit NMDA receptors byenhancing proton inhibition. Nat. Neurosci. 1, 659–667 (1998).

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

24. Zheng, F., Gingrich, M. B., Traynelis, S. F. & Conn, P. J. Tyrosine kinasepotentiates NMDA receptor currents by reducing tonic zinc inhibition. Nat.Neurosci. 1, 185–191 (1998).

25. 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. 18, 6163–6175 (1998).

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

27. Vallano, M. L., Lambolez, B, Audinat, E. & Rossier, J. Neuronal activitydifferentially regulates NMDA receptor subunit expression in cerebellargranule cells. J. Neurosci. 16, 631–639 (1996).

28. Choi, Y .P. & Lipton, S. A. Identification and mechanism of action of twohistidine residues underlying high-affinity Zn2+ inhibition of the NMDAreceptor. Neuron 23, 171–180 (1999).

29. Low, C.-M., Zheng, F., Lyuboslavsky, P. & Traynelis, S. F. Moleculardeterminants of coordinated proton and zinc inhibition of NR1/NR2Areceptors. Proc. Natl. Acad. Sci. USA 97, 11062–11067 (2000).

30. Fayyazuddin, A., Villarroel, A., Le Goff, A., Lerma, J. & Neyton, J. Fourresidues of the extracellular N-terminal domain of the NR2A subunit controlhigh-affinity Zn2+ binding to NMDA receptors. Neuron 25, 683–694 (2000).

31. Paoletti, P. et al. Molecular organization of a zinc binding n-terminalmodulatory domain in a NMDA receptor subunit. Neuron 28, 911–925 (2000).

32. Monod, J., Wyman, J. & Changeux, J.-P. On the nature of allosterictransitions: a plausible model. J. Mol. Biol. 12, 88–118 (1965).

33. Gallagher, M. J., Huang, H., Grant, E. R. & Lynch, D. R. The NR2B-specificinteractions of polyamines and protons with the N-methyl-D-aspartatereceptor. J. Biol. Chem. 272, 24971–24979 (1997).

34. Brimecombe, J. C., Gallagher, M. J., Lynch, D. R. & Aizenman, E. An NR2Bpoint mutation affecting haloperidol and CP101,606 sensitivity of singlerecombinant N-methyl-D-aspartate receptors. J. Pharmacol. Exp. Ther. 286,627–634 (1998).

35. Traynelis, S. F. & Cull-Candy, S. Pharmacological properties and H+

sensitivity of excitatory amino acid receptor channels in rat cerebellar granuleneurons. J. Physiol. (Lond.) 433, 727–763 (1991).

36. Lester, R. A. & Jahr, C. E. NMDA channel behavior depends on agonistaffinity. J. Neurosci. 12, 635–643 (1992).

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

38. 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).

39. 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).

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

41. Traynelis, S. F. Software based correction of single compartment seriesresistance errors. J. Neurosci. Meth. 86, 25–34 (1998).

42. Zar, J. H. in Biostatistical Analysis (ed. Zar, J. H.) 584–585 (Prentice Hall,Upper Saddle River, New Jersey, 1999).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

Marine snails of the genus Conus (‘cone snails’) are predatorygastropod mollusks found on or near coral reefs in tropical watersthroughout the world. They have evolved an elaborate strategyfor the rapid immobilization of fish, worms or other mollusksthat involves the injection of a complex mixture of bioactive pep-tides into their prey. These conopeptides are typically 10–30 amino acid residues long and contain multiple intramol-ecular disulfide bonds. The venom of any single Conus speciesmay contain more than 100 different peptides1. Conopeptidesinterfere with neurotransmission by targeting a variety of pro-teins expressed on the cell surface. Presently, distinct classes ofconopeptides are recognized to act at voltage-sensitive Ca2+ chan-nels2, Na+ channels3–5, K+ channels6,7, nicotinic ACh recep-tors8–10, 5-HT3 receptors11, NMDA receptors12, vasopressinreceptors13 and neurotensin receptors14. The potential ofconopeptides as tools in neuroscience or as therapeutic agentsremains largely unexplored, given that the conopeptides charac-terized to date are estimated to represent only 0.1% of theconopeptides present in the ∼ 500 species of Conus15.

Here we describe the discovery and characterization of twonew classes of conopeptides: the ρ-conopeptide class defined byconopeptide TIA from Conus tulipa, and the χ-conopeptide classdefined by conopeptides MrIA and MrIB from C. marmoreus(Table 1, Fig. 1). Both classes interfere with the action of nora-drenaline (NA), a key neurotransmitter in the central andperipheral nervous systems. 1H NMR studies reveal marked dif-ferences in the three-dimensional structures of these two-looppeptides. To achieve their different pharmacologies, the ρ-conopeptides use a new combination of amino acids arrangedon an α-conopeptide fold (1–3, 2–4 disulfide connectivity),whereas the χ-conopeptides use a different sequence andconopeptide disulfide framework (1–4, 2–3).

Two new classes of conopeptidesinhibit the α1-adrenoceptor andnoradrenaline transporter

Iain A. Sharpe1,2, John Gehrmann1, Marion L. Loughnan1, Linda Thomas1, Denise A. Adams1, Ann Atkins1, Elka Palant3, David J. Craik1, David J. Adams2, Paul F. Alewood1 and Richard J. Lewis1,3

1 Institute for Molecular Bioscience, University of Queensland, Brisbane 4072, Australia2 School of Biomedical Sciences, Department of Physiology and Pharmacology, University of Queensland, Brisbane 4072, Australia3 Xenome Ltd., 50 Meiers Road, Indooroopilly, Brisbane 4068, Australia

Correspondence should be addressed to R.J.L. ([email protected])

Cone snails use venom containing a cocktail of peptides (‘conopeptides’) to capture their prey.Many of these peptides also target mammalian receptors, often with exquisite selectivity. Herewe report the discovery of two new classes of conopeptides. One class targets α1-adrenoceptors(ρ-TIA from the fish-hunting Conus tulipa), and the second class targets the neuronalnoradrenaline transporter (χ-MrIA and χ-MrIB from the mollusk-hunting C. marmoreus). ρ-TIAand χ-MrIA selectively modulate these important membrane-bound proteins. Both peptides actas reversible non-competitive inhibitors and provide alternative avenues for the identification ofinhibitor drugs.

RESULTSPeptide discoveryWhile searching for α-conopeptides (competitive antagonists ofthe nicotinic ACh receptor), we isolated and sequenced thepolypeptides TIA ([M + 2H]2+, observed m/z = 1195.57, expect-ed m/z = 1195.57; calculated and expected [M + H]+

m/z = 2390.15), MrIA ([M + H]+, observed m/z = 1408.53,expected m/z = 1408.53), and MrIB ([M + H]+ observed m/z = 1393.52, expected m/z = 1393.55) from the crude venom ofC. tulipa and C. marmoreus collected from the Great Barrier Reef,Australia (Table 1, Fig. 1). ([M + H]+ and [M + 2H]2+ refer tothe ionization states of the molecules, and m/z is the mass/chargeratio determined by mass spectrometry.) Surprisingly, thesepolypeptides had little sequence homology to previously identi-fied conopeptides, despite the cysteine residues being positionedin the linear sequence in a manner reminiscent of α-conopep-tides. To allow investigations into their mechanisms of action andthree-dimensional structures, chemically identical synthesizedforms of these peptides were assembled.

PharmacologyThe likely modes of action of TIA and MrIA were identified fromstudies of their effects on the biphasic contractile response of theelectrically stimulated rat prostatic vas deferens. Both peptidesaffected the second phase of contraction, which peaks approxi-mately 600 ms after stimulation due to the action of neurallyreleased NA on postjunctional α1-adrenoceptors16. This NA-mediated response was preceded by a faster contraction, whichpeaked approximately 200 ms after stimulation due to the actionof the sympathetic co-transmitter ATP at postjunctionalP2X-purinoceptors17. TIA and MrIA had opposite effects on thestrength of the NA-mediated component of the vas deferens con-

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


traction. TIA (concentration producing 50% inhibition (IC50),500 nM) caused a concentration-dependent inhibition of the sec-ond phase of the contraction, without affecting the ATP-medi-ated response (Fig. 2a). This inhibition was slowly reversed uponwashout with conopeptide-free solution. In the epididymal seg-ment of rat vas deferens, TIA (3 µM and 10 µM) increased theconcentration of exogenously applied NA required to producethe half-maximal effect (EC50) by 5.5- and 17.8-fold, anddepressed the level of the maximum response to 82% and 42%of the control response, respectively. Because the effects of TIAwere not surmountable, the peptide acted as a non-competitiveantagonist of NA-induced contractions in the vas deferens. Theconcomitant decrease in NA potency is explained by the under-lying hyperbolic relationship between α1-adrenoceptor activa-tion and the contractile response in the rat vas deferens18. Atrelatively low concentrations of a non-competitive α1-adreno-ceptor antagonist, part of the receptor pool can be removed with-out affecting the level of the maximum response. This gives riseto the phenomenon of ‘spare receptors.’ The EC50 for NA isincreased in this situation because the fractional receptor occu-pancy required to generate the half-maximal response is increasedwith the reduction in the size of the available receptor pool. Inaddition, the half-maximal response does not occur at 50% recep-tor occupancy because of the underlying non-linear relationshipbetween receptor activation and response. As the receptor poolis further diminished by increasing concentrations of a non-com-petitive antagonist, the maximum response to NA is depressed.This reduction in the level of the maximum response is accom-panied by a rightward shift of the concentration–response curve(that is, increased EC50). Again, this is due to the increase in thefractional receptor occupancy required by NA to produce thehalf-maximal response19.

To define the molecular pharmacology of TIA, we investigat-ed its action on the cloned hamsterα1B-adrenoceptor expressed in COS-1cells. TIA dose-dependently displacedbinding of the α1-adrenoceptor antago-nist 125I-HEAT to α1B-adrenoceptors(IC50, 124 nM) (Fig. 2b), confirmingthat TIA is an inhibitor of theα1-adrenoceptor. Thus, TIA belongs toa previously unknown class of conopep-tides we name ρ-conopeptides. Thisinhibition was non-competitive, as sat-uration binding studies revealed thatρ-TIA (1 µM) reduced maximum125I-HEAT binding to α1B-adrenocep-tors by 85% without affecting the affin-ity of the receptor for the radioligand(Kd) (Fig. 2c). Thus, TIA acts non-com-

petitively to inhibit the bindingof agonists (NA) and antagonists(HEAT) that act directly at theNA binding site20.

In contrast to ρ-TIA, MrIA,its amidated derivativeMrIA-NH2, and MrIB-NH2 causeda concentration-dependent in-crease in the second phase of thecontraction of the rat prostaticvas deferens in response to elec-trical field stimulation (Fig. 3a).Again, the ATP-mediated re-

sponse was unaffected, and the effect of the peptides was reversedupon washout. MrIA-NH2 (EC50, 430 nM), being 1.5- and 2-foldmore potent than MrIA and MrIB-NH2, respectively, was used toestablish the mechanism for this potentiation. Log concentra-tion–response curves to exogenously applied NA were dose-depen-dently shifted leftward in a parallel manner 3.3- and 7.4-fold byMrIA-NH2 (1–3 µM), indicating that the sensitivity of the tissueto NA was enhanced by the peptide. However, MrIA-NH2 (3 µM)had no effect on the responsiveness of the tissue to methoxam-ine. Because methoxamine is an α1-adrenoceptor agonist, but nota substrate for the noradrenaline transporter21 (NET), we hypoth-esized that MrIA-NH2 acts to inhibit the NET, the principal routeof elimination of NA from the synapse.

To define the molecular pharmacology of MrIA-NH2, we inves-tigated its action on the cloned rat and human NET expressedin COS-1 cells. The rate of cellular accumulation of 3H-NA viathe rat NET was reduced by MrIA-NH2 (IC50 615 nM) in a dose-dependent manner (Fig. 3b). MrIA-NH2 also inhibited transportof 3H-NA by the human NET (IC50, 1.26 µM). Thus, MrIA-NH2is an inhibitor of the NET, making it the first of a previouslyunknown class of conopeptides we name χ-conopeptides. Thisaction was found to be non-competitive in nature, as χ-MrIA-NH2(1 µM) reduced the maximum rate of 3H-NA uptake by thehuman NET by 42% without affecting the affinity of the trans-porter for substrate (Km) (Fig. 3c). Of the family of monoamineneurotransmitter transporters, χ-MrIA-NH2 is selective for theNET, because at 100 µM it did not inhibit the closely relatedhuman dopamine and serotonin transporters (data not shown).

A range of tissue assays were used to determine the selectivityof the ρ- and χ-conopeptides. Responses of the guinea pig ileumto nicotine and the mouse phrenic nerve–hemidiaphragm to elec-trical stimulation were unaffected by ρ-TIA (10 µM) andχ-MrIA-NH2 (3 µM), indicating that unlike the α-conopeptides,

0 10 20 30 40 50 60

Rel

ativ

e in

tens

ity

Elution time (min)

1.00

0.75

0.50

0.00

0.25

TIA

0 10 20 30 40 50 60

Elution time (min)

1.00

0.75

0.50

0.00

0.25Rel

ativ

e in

tens

ity

MrIA

Fig. 1. Liquid chromatography/mass spectroscopy (LC/MS) analysis of crude venom from C. tulipa andC. marmoreus. (a) TIA eluted as a minor component at 24.5 min (Inset C. tulipa shell). (b) MrIA elutedat 21.8 min, and MrIB eluted as a minor component at 24.2 min (Inset C. marmoreus shell). Over 90%of the conopeptides present in the crude venom of these species have not been characterized.

a b

Table 1. Sequence and pharmacological diversity of 2-loop conopeptides.

Class Name Sequence Connectivity Physiological target

ρ TIA FNWRCCLIPACRRNHKKFC* 1–3, 2–4 α1-adrenoceptor

χ MrIA NGVCCGYKLCHOC 1–4, 2–3 noradrenaline transporterMrIB VGVCCGYKLCHOC 1–4, 2–3

α GI ECCNPACGRHYSC* 1–3, 2–4 nicotinic acetylcholinePnIB GCCSLPPCALSNPDYC* 1–3, 2–4 receptors

Cysteines involved in disulfide bonds are labeled 1–4 with respect to the linear sequences. *Amidated C-termini. O,4-hydroxyproline. GI was isolated from Conus geographus8, and PnIB was isolated from C. pennaceus38.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

Fig. 2. Mode of action of ρ-TIA. (a) Inhibition of electrically evokedcontractions of the prostatic portion of the rat vas deferens by ρ-TIA.(b) Displacement of 125I-HEAT binding to expressed α1B-adrenoceptorby ρ-TIA. (c) Saturation curves for 125I-HEAT binding to the α1B-adrenoceptor expressed in COS-1 cells in the absence () andpresence of either 30 nM () or 1 µM () ρ-TIA. Symbols representthe mean ± s.e.m. of n = 3 separate experiments performed in dupli-cate. Some error bars are obscured by the symbols.

these peptides do not target neuronal or muscle subtypes of thenicotinic ACh receptor. ρ-TIA and χ-MrIA-NH2 had no effect onthe first component of vas deferens contraction. Thus, both arewithout effects on the voltage-sensitive ion channels involved inaction potential propagation and neurotransmitter exocytosis,which are targeted by certain other conopeptide classes2,3,7. Localanesthetic-type activity, such as blockade of neuronal Na+ chan-nels, has been previously reported for prazosin and otherα1-adrenoceptor antagonists22,23, and for cocaine24. Also unaffect-ed by ρ-TIA (10 µM) were presynaptic α2-adrenoceptors, of whichthe activation by NA inhibits evoked contractions in the prazosin-treated rat vas deferens. χ-MrIA-NH2 did not affect contractions tomethoxamine, indicating that the peptide does not act at α1-adreno-ceptors, distinguishing it from some other NET inhibitors, such asthe antidepressants desipramine and amitriptyline25.

Three-dimensional structuresTo begin to understand the structural features underlying thebiological activity of ρ- and χ-conopeptides, we used 1H NMRtechniques to determine the structures of ρ-TIA and a repre-sentative χ-conopeptide, χ-MrIB-NH2. ρ-TIA gave NMR spec-tra with good NH chemical shift dispersion, indicative of awell-structured molecule. A total of 213 distance restraintsderived from 17 intra-residue, 93 sequential, 67 medium and 36long-range NOEs (nuclear Overhauser effects), 7 dihedral and19 αH chemical shift restraints were obtained for ρ-TIA from500 and 750 MHz data. In contrast, χ-MrIB-NH2 showed peak‘brothering,’ indicating the presence of a minor conformation(<10% relative population). However, changing the solvent con-ditions to 30% acetonitrile/H2O or 30% TFE/H2O emphasizedthe major conformation. Comparable NMR data were alsoobtained for χ-MrIA and χ-MrIA-NH2, indicating similar struc-tures among the χ-conopeptides. Like MrIB-NH2, these peptidesshowed evidence for minor conformations, most likely associ-ated with cis–trans isomerization of the His10–Hyp11 peptidebond. This proposal is supported by a doubling of peaks forresidues flanking this bond (His11, Cys13) and those physicallyclose to it (Gly2, Val3) due to nearby disulfide linkages. Themajor conformer clearly has a trans His10–Hyp11 peptide bondbased on a strong sequential His10αH→Hyp11δH NOE in allof the χ-conopeptides, but the low intensity and limited numberof peaks observed for the minor conformers prevented theirdefinitive characterization. The possibility that configurationalisomerization of the disulfide bonds (ProR/ProS forms) alsocontributes to the minor forms cannot be excluded. However,whatever the cause of the minor forms, they are very similar inoverall structure to the major conformer, as the NH chemicalshifts of the brothered peaks are all within 0.05 p.p.m. of themajor conformer. A total of 82 distance restraints derived from14 intra-residue, 42 sequential, 3 medium, 23 long range NOEsand 4 dihedral restraints were obtained for the major confor-mation of χ-MrIB-NH2 from 500 MHz data.

For both ρ-TIA and χ-MrIB-NH2, 50 structures were energy-minimized, and the 20 lowest-energy structures were chosen to

represent the final structures (Fig. 4). ρ-TIA consisted of a stretchof 310 helix between Arg4 to Leu8, a helical turn from Pro9 toArg12, followed by four nested β-turns between Arg12 and Cys19,which almost comprised another turn of helix. These compriseda type I β-turn between residues 12–15, and three type IV β-turnsbetween residues 13–16, 14–17 and 15–18. ρ-TIA had a back-bone pairwise root-mean-square deviation (RMSD) of 0.73 Å(1.94 Å for all heavy atoms), and no NOE violations greater than0.2 Å (4 violations greater than 0.1 Å).

In contrast to ρ-TIA, χ-MrIB-NH2 contained no helical ele-ments, and instead formed a small β-hairpin structure. Specifi-cally, χ-MrIB-NH2 contained a flexible loop between Tyr7 andLeu9, which was presented on the β-hairpin structure, with main-chain H-bonds between Cys4 NH and His11 C=O (all structures),and Gly6 NH and Leu9 C=O (13/20 structures). χ-MrIB-NH2 hada backbone pairwise RMSD of 1.23 Å (2.90 Å for all heavy atoms)and no NOE violations greater than 0.2 Å (3 violations greaterthan 0.1 Å). Although the RMSD was higher than is typically seenfor larger proteins, it is not unusual for small peptides to exhibithigh RMSD values. NOEs are weaker than for larger proteins dueto correlation time effects, and small peptides lack a core regionthat typically gives many NOE contacts. Both factors can lead tolower precision (RMSD), but the structures reported here are suf-ficient to define the global fold. In addition, χ-MrIB-NH2 was lessdefined between residues Tyr7–Leu9, possibly reflecting flexibili-ty in this region. With these residues excluded from the calcula-tion, the backbone RMSD was 0.74 Å.

DISCUSSIONThe ρ- and χ-conopeptides are the first examples of peptides thatact selectively at α1-adrenoceptors and the NET. Both classes actallosterically through sites distinct from those targeted by nora-

00.00

0.25

0.50

0.75

1.00

1.25

Time (ms)

∆ te

nsio

n (g

)

250 500 750 1000 1250

control

100 nM ρTIA

300 nM ρTIA

1 µM ρTIA

3 µM ρTIA

log [ρ-TIA] (M)

125 I

-HE

AT

bin

ding

(%

)

– 9 –8 –7 –60

25

50

75

100

0 200 400 600

0.0

1.5

3.0

4.5

6.0

125 I

-HE

AT

bin

ding

(pm

ol/m

g pr

otei

n)

125I-HEAT (pM)

a

b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


drenaline. They are structurally distinct from the various otherclasses of small molecule inhibitors of the α1-adrenoceptor andNET that act competitively at the site of NA interaction, manyof which are important therapeutic agents. Both ρ-TIA andχ-MrIA-NH2 lack the common and often therapeutically limit-ing pharmacology of α1-adrenoceptor antagonists (α2-adreno-ceptor and Na+-channel inhibition) and NET inhibitors(α1-adrenoceptor and muscarinic ACh receptor antagonism),and thus may be useful clinically. Given the physical character-istics of conopeptides, we expect that these peptides will not crossthe cell membrane, and that their site of action is on the extra-cellular surface of these proteins. Previously, a peptide withsequence identical to χ-MrIA was found to have antinociceptiveactivity when administered intrathecally to mice26. This is notunexpected given the role of NA in antinociception27, and indi-cates that χ-MrIA may have potential as an analgesic.

ρ-TIA, with a 4/7 two-loop framework, adopts the canonicalα-conopeptide fold with a 1–3, 2–4 disulfide bond connectivity,as exemplified by PnIB (Fig. 4). The ρ-TIA structure is also relat-ed to the published X-ray structures for α-conopeptides [Tyr15]-PnIA, [Tyr15]-PnIB and [Tyr15]-EpI, with a backboneRMSD of approximately 1.0 Å over the two loops28–30. Thusρ-conopeptides achieve their pharmacology by displaying dif-ferent side-chain elements on a peptide backbone that can alsobe used to target nicotinic ACh receptors. In contrast,χ-MrIB-NH2 has a distinctly different structure. The presence of

a Cys5 αH to Cys13 NH NOE, strong Cys4 βH to Cys13 αH andβH NOEs, and Cys5 αH to Cys10 αH NOE is not compatiblewith the canonical 1–3, 2–4 disulfide bond connectivity, but isinstead indicative of a 1−4, 2−3 disulfide bond connectivity. Thisconnectivity was also shown chemically by selective reduc-tion/alkylation studies (data not shown), and by the selective syn-thesis of the three possible structural isomers26,31. Thus,χ-conopeptides achieve their pharmacology through a conopep-tide structure that uses different side-chain elements and a pre-viously unidentified conopeptide backbone. Although we assumethat the observed activity for the χ-conopeptides arises from themajor (>90%) conformation described here, minor forms, mostlikely associated with cis–trans isomerisation of Hyp11, do occurin solution, and could potentially contribute to the biologicalactivity. In any case, these minor forms are very similar to themajor conformation and exhibit the same connectivity of disul-fide bonds. Given that this is a previously unknown disulfidearrangement for native conopeptides, it is not surprising thatχ-MrIB-NH2 superimposed poorly with α-conopeptides such asGI (Fig. 4). For example, the heavy atom RMSD for the cysteineresidues of χ-MrIB-NH2 and GI was only 4.02 Å. These high-res-olution three-dimensional structures of ρ- and χ-conopeptidesprovide templates to guide the design of peptidomimeticinhibitors of α1-adrenoceptors and the NET.

These two conopeptide classes open up alternative avenuesfor modulating α1-adrenoceptors and the NET. Because thesepeptides are non-competitive inhibitors, they seem to act alloster-ically at modulatory sites distinct from the site of NA interaction,presumably at the extracellular loops of these membrane-boundproteins. These modulatory sites on the α1-adrenoceptor and theNET may also be sites where endogenous mammalian peptideinhibitors act. An example of such an endogenous modulatorypeptide is 5-HT-moduline, a non-competitive inhibitor of the

00.0

0.5

1.0

1.5

Time (ms)

∆ te

nsio

n (g

)

250 500 750 1000 1250

3 µM

1 µM

300 nM

100 nM

30 nM

control

χ-MrIA-NH 2

χ-MrIA-NH 2

χ-MrIA-NH 2

χ-MrIA-NH 2

χ-MrIA-NH 2

–8 –7 –6 –5 –4–9

0

25

50

75

100

log [χ-MrIA-NH2] (M)

3 H-N

A tr

ansp

ort (

%)

0 10 20 300.0

1.0

2.0

3.0

4.0

3 H-N

A tr

ansp

ort

(fm

ol/m

in/1

06 ce

lls)

3H-noradrenaline (µM)

a

b

c

Fig. 3. Mode of action of χ-MrIA-NH2. (a) Enhancement of electricallyevoked contractions of the prostatic portion of the rat vas deferens byχ-MrIA-NH2. (b) Inhibition of 3H-NA uptake into COS-1 cells viaexpressed rNET by χ-MrIA-NH2. (c) Kinetics of 3H-NA transport byhNET expressed in COS-1 cells in the absence () and presence of 1 µM χ-MrIA-NH2 (). Symbols represent the mean ± s.e.m. of n = 3separate experiments performed in duplicate. Some error bars areobscured by the symbols.

Fig. 4. NMR-derived structures of ρ-TIA and χ-MrIB-NH2. The NMRstructure of α-conopeptides PnIB and GI are shown for comparison.Based on loop size, ρ-TIA (4/7) and χ-MrIB-NH2 (4/2) are closest instructure to α-conopeptides PnIB (4/7) and GI (3/5), respectively.Disulfide bonds are shown in red.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

5-HT1B/D G-protein-coupled receptor32. It remains to be seen ifnon-competitive modulatory sites are a common feature acrossG-protein-coupled receptors and transporters.

The involvement of ρ-TIA and χ-MrIA in prey capture is lessclear, as neither conopeptide alone was toxic to Gambusia affinisat 25 nmol/g. The peptides have not been tested for biologicalactivity in mollusks. It is possible that these peptides act (per-haps even at a different physiological target in the prey species) ina ‘cabal’ with other bioactive peptides in the venom to produce asynergistic effect that aids prey capture33. The two-loopCCXnCXnC peptides, in which X defines residues comprising theintercysteine loops, seem to represent a ‘privileged’ structuralclass that can access diverse pharmacology, reminiscent of thefour-loop CXnCCXnCXnCXnC peptides, which target voltage-sensitive Ca2+ (ω-conopeptides)2, Na+ (µO-conopeptides)5 andK+ (κ-conopeptides)7 ion channels. However, the two-loop struc-tural class act at membrane proteins targeted by the neurotrans-mitters NA (ρ- and χ-conopeptides) and ACh (α-conopeptides),rather than at voltage-sensitive ion channels.

METHODSPeptide isolation. The peptides were purified to homogeneity, and theirprimary structures were determined by Edman sequencing, followingprocedures previously described34. ρ-TIA, χ-MrIA and χ-MrIB wereidentified in crude venom from C. tulipa and C. marmoreus(0.6 mg/mL, 15 and 5 µL injected, respectively) by reverse-phase HPLCusing a C18 column (3.5 µm, 2.1 × 50 mm Zorbax 300 SB, Palo Alto, Cal-ifornia) eluted with a 1% linear gradient from 0 to 60% B (A = 0.1%formic acid, B = 90% acetonitrile/0.1% formic acid) using a HP1100pump. The eluant was continuously monitored with a QSTAR pulsar(PE-Sciex, Foster City, California) mass spectrometer. Mass spectrome-try and sequence data indicated that the C-terminus of ρ-TIA is ami-dated, whereas χ-MrIA and χ-MrIB are the free-acid forms.

Peptide synthesis. ρ-TIA (amidated), χ-MrIA (amidated and free-acid)and χ-MrIB (amidated) were synthesized using Fmoc and Bmoc chem-istry, and oxidized in NaHCO3, pH 8.034. The final overall yield was 65%for TIA and 30% for MrIA, with purity greater than 98% by HPLC. Oncoinjection onto a Waters C18 symmetry column (Milford, Massachus-sets), TIA (amidated) coeluted with native TIA, and HPLC purified MrIA(free acid), but not the amidated form of MrIA, coeluted with native MrIA.

Rat vas deferens. Segments of vasa deferentia from male Wistar rats(250–350 g) were mounted under a tension of 0.5 g in 5 mL organ bathscontaining 119 mM NaCl, 4.7 mM KCl, 1.17 mM MgSO4, 1.18 mMKH2PO4, 25.0 mM NaHCO3, 5.5 mM glucose, 2.5 mM CaCl2 and 0.026 mM EDTA (pH 7.4, 37°C, bubbled with 5% CO2:95% O2). Bisect-ed prostatic segments were field-stimulated with single electrical pulsesof 55-V amplitude and 1-ms duration generated by a Grass S44 stimu-lator (Quincy, Massachussets) at 3-min intervals. Isometric contractionswere recorded digitally on a Power Macintosh computer with aMacLab/8s data acquisition system (ADInstruments, Sydney, Australia).The resulting contractions could be abolished by tetrodotoxin (0.1 µM),indicating a neural origin. Drugs were added cumulatively to the organ.All animal experiments were conducted with the approval of the Uni-versity of Queensland Animal Experimentation Ethics Committee.

To examine if ρ-TIA affected presynaptic α2-adrenoceptors, electri-cal pulses as described above were delivered at 20-s intervals to the pro-static portion of vas deferens treated with prazosin (0.5 µM). Cumulativeconcentration–response curves for the inhibitory effect of NA were estab-lished before and 20 min after addition of ρ-TIA (10 µM). To determinewhether ρ-TIA affected postjunctional contractile responses to NA, weused unstimulated bisected epididymal segments. Cumulative concen-tration–response curves to NA were established in the absence or presenceof ρ-TIA (3 µM or 10 µM applied 20 min before NA). The bisected epi-didymal segments were also used to establish concentration–responsecurves to NA and methoxamine in the absence and presence of

χ-MrIA-NH2. χ-MrIA-NH2 (1 µM or 3 µM) was applied 20 min beforecumulative additions of NA or methoxamine, which contract the smoothmuscle via postjunctional α1-adrenoceptors.

Mouse phrenic nerve–hemidiaphragm. Left and right hemidiaphragmwere dissected from male Quackenbush mice (20–30 g), mounted undera tension of 1.0 g in 5 mL organ baths containing 135.0 mM NaCl, 5.0 mM KCl, 2.0 mM CaCl2, 1.0 mM MgCl2, 11.0 mM glucose, 15.0 mMNaHCO3 and 1.0 mM KH2PO4 (pH 7.4, 37°C, bubbled with 5%CO2:95% O2). Field stimulation of the phrenic nerve with 3-V pulses of0.2-ms duration delivered at 20-s intervals evoked contractions recordedas described above. We examined the effect of ρ-TIA (10 µM) andχ-MrIA-NH2 (3 µM) on the contractions.

Guinea pig ileum. Segments of ilea (∼ 1.5 cm) from male guinea pigs(280–430 g) were mounted under a resting tension of 1.0 g in 5 mL organbaths containing 136.9 mM NaCl, 2.68 mM KCl, 1.84 mM CaCl2, 1.03mM MgCl2, 5.55 mM glucose, 11.9 mM NaHCO3, 0.45 mM KH2PO4 and10 µm indomethacin (pH 7.4, 37°C, bubbled with 5% CO2:95% O2). Non-cumulative doses of nicotine (4 µM) were added until reproducible con-tractile responses were obtained. Tissue was then exposed to the ρ-TIA (10µM) or χ-MrIA-NH2 (3 µM) for 25 min and nicotine was again applied.

125I-HEAT binding. Membranes from COS-1 cells (ATCC, Manassas, Vir-ginia) grown in Dulbecco’s modified Eagle’s medium (DMEM) contain-ing 5% fetal bovine serum and transiently transfected with plasmid DNAencoding the hamster α1B-adrenoceptor were prepared as described pre-viously20. For radioligand binding studies, we used duplicate tubes con-taining 80 pM 125I-HEAT, COS-1 membranes and various concentrationsof ρ-TIA, in HEM buffer (20 mM HEPES, 1.5 mM EGTA, 12.5 mMMgCl2, pH 7.5). For saturation binding studies, 125I-HEAT was added to575 nM in the presence and absence of TIA. Total reaction volume was250 µL. Nonspecific binding (5–10% of total) was determined by theinclusion of phentolamine (10–4 M). After 80 min of incubation at roomtemperature, the reactions were stopped by the addition of ice-cold HEMbuffer and were transferred onto Whatman GF/C glass filters (Kent, UK).

Uptake of 3H-NA. COS-1 cells grown in 24-well plates containing DMEMand 10% fetal bovine serum were transiently transfected with plasmidDNA encoding the rat35 or human NET36 using Lipofectamine 2000reagent (Gibco, Carlsbad, California). Assays were done at room tem-perature, 24 h after transfection in transport buffer containing 125 mMNaCl, 4.8 mM KCl, 1.2 mM MgSO4, 1.2 mM KH2PO4, 1.3 mM CaCl2,25 mM HEPES, 5.55 mM D-(+)-glucose, 1.02 mM ascorbic acid, 10 µMU-0521 and 100 µM pargyline, pH 7.4. The cells were exposed to vari-ous concentrations of χ-MrIA-NH2 or desipramine (10−5 M; for the deter-mination of nonspecific accumulation) for 15 min before 3H-NA (100 nM, supplemented with unlabeled NA as required) was added foreither 8 (rat NET) or 20 min (human NET). 3H-NA was added at con-centrations up to 30 µM to determine the kinetics of NA transport byhNET expressed in COS-1 cells. The solution was then rapidly removed,and the cells were washed with ice-cold PBS. The cells were lysed with0.1% Triton-X-100 in 10 mM Tris·HCl (pH 7.5) for 90 min, and the celllysate was taken for scintillation counting.

Drugs. The following drugs were used: desipramine hydrochloride,indomethacin, methoxamine hydrochloride, nicotine hydrogen tartrate(–)-noradrenaline bitartrate (NA), pargyline, tetrodotoxin (Sigma, St. Louis,Missouri); phentolamine mesylate (Research Biochemicals International,Natick, Massachussets); U-0521 (Biomol, Plymouth Meeting, Pennsylva-nia); 125I-HEAT, 3H-NA (NEN Life Science, Boston, Massachussets).

Statistics and data analysis. Data are expressed as means ± s.e.m. ofresults obtained from 3–6 separate experiments. Student’s two-tailed t-testor analysis of variance was used for statistical evaluation; values of p lessthan 0.05 were considered significant. Sigmoidal concentration–responsecurves for the calculation of EC50 or IC50 values were fit by non-linearregression using Prism software (GraphPad, San Diego, California).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


NMR studies. NMR experiments were typically done on 2−5 mM peptidein either D2O or 90% H2O/10%D2O at pH 3.5 and 278−285 K. Addi-tional data on χ-MrIA-NH2 were obtained in 30% d3-acetonitrile/10%D2O/60% H2O and 30% TFE/10% D2O/60% H2O. Spectra were acquiredon ARX 500 MHz or AVANCE 750 MHz spectrometer (Bruker, Billerica,Massachussets) equipped with a shielded gradient unit, and referencedto a DSS internal standard. Spectra were processed using UXNMR orXWIN-NMR (Bruker), and analyzed in Felix (Hare Research, Bothell,Massachussets). NOE crosspeaks were assigned as strong (1.8−2.7 Å),medium (1.8−3.5 Å) or weak (1.8−5.0 Å). NOE peaks observed at longmixing times (≥400 ms) but not at shorter mixing times were classed asvery weak (1.8−6 Å). Pseudo-atom corrections of 1.5 Å for methyl, 1.0Å for methylene, and 2.0 Å for phenyl and tyrosine ring protons wereadded. Structures were calculated from distance and angle restraints usingthe torsion-angle dynamics/simulated annealing protocol as previouslydescribed37. After the initial structures showed discrepancies due to theunusual disulfide connectivity in χ-MrIB-NH2, the simulated annealingprotocol was modified to first calculate the structures without disulfidebonds. With this protocol, 41/50 structures had the preferred disulfideconnections 4−13 and 5−10. Incorporating this disulfide connectivityinto the simulated annealing protocol improved structural convergence.

The coordinates for ρ-TIA and χ-MrIB-NH2 have been deposited withthe Research Collabatory for Structural Biology (RCSB) with accessioncodes RCSB013213 (PDB ID 1IEN) and RCSB013214 (PDB 1IEO).

ACKNOWLEDGEMENTSWe thank R. Graham (Sydney) and D. Kaye (Melbourne) for assistance and

advice and for providing the α1B-adrenoceptor (R.G.) and human NET (D.K.)

clones. L. Bryan-Lluka (Brisbane) provided the rat NET clone. A. Jones, N.

Daly, K. Nielsen and T. Bond (I.M.B.) provided assistance. This work was

supported by grants from AusIndustry and the National Health and Medical

Research Council, Australia.


1. Olivera, B. M. et al. Diversity of Conus neuropeptides. Science 249, 257–263(1990).

2. Olivera, B. M., McIntosh, J. M., Cruz, L. J., Luque, F. A. & Gray, W. R.Purification and sequence of a presynaptic peptide toxin from Conusgeographus venom. Biochemistry 23, 5087–5090 (1984).

3. Cruz, L. J. et al. Conus geographus toxins that discriminate between neuronaland muscle sodium channels. J. Biol. Chem. 260, 9280–9288 (1985).

4. Fainzilber, M., Kofman, O., Zlotkin, E. & Gordon, D. A new neurotoxinreceptor site on sodium channels is identified by a conotoxin that affectssodium channel inactivation in molluscs and acts as an antagonist in ratbrain. J. Biol. Chem. 269, 2574–2580 (1994).

5. McIntosh, J. M. et al. A new family of conotoxins that blocks voltage-gatedsodium channels. J. Biol. Chem. 270, 16796–16802 (1995).

6. Craig, A. G. et al. An O-glycosylated neuroexcitatory Conus peptide.Biochemistry 37, 16019–16025 (1998).

7. Shon, K. J. et al. κ-Conotoxin PVIIA is a peptide inhibiting the shaker K+

channel. J. Biol. Chem. 273, 33–38 (1998).8. Gray, W. R., Luque, A., Olivera, B. M., Barrett, J. & Cruz, L. J. Peptide toxins

from Conus geographus venom. J. Biol. Chem. 256, 4734–4740 (1981).9. Hopkins, C. et al. A new family of Conus peptides targeted to the nicotinic

acetylcholine receptor. J. Biol. Chem. 270, 22361–22367 (1995).10. Shon, K. J. et al. A noncompetitive peptide inhibitor of the nicotinic

acetylcholine receptor from Conus purpurascens venom. Biochemistry 36,9581–9587 (1997).

11. England, L. J. et al. Inactivation of a serotonin-gated ion channel by apolypeptide toxin from marine snails. Science 281, 575–578 (1998).

12. Mena, E. E. et al. Conantokin-G: a novel peptide antagonist to the N-methyl-

D-aspartic acid (NMDA) receptor. Neurosci. Lett. 118, 241–244 (1990).13. Cruz, L. J. et al. Invertebrate vasopressin/oxytocin homologs.

Characterization of peptides from Conus geographus and Conus striatusvenoms. J. Biol. Chem. 262, 15821–15824 (1987).

14. Craig, A. G. et al. Contulakin-G, an O-glycosylated invertebrate neurotensin.J. Biol. Chem. 274, 13752–13759 (1999).

15. Adams, D. J., Alewood, P. F., Craik, D. J., Drinkwater, R. D. & Lewis, R. J.Conotoxins and their potential pharmaceutical applications. Drug Dev. Res.46, 219–234 (1999).

16. McGrath, J. C. Adrenergic and ‘non-adrenergic’ components in thecontractile response of the vas deferens to a single indirect stimulus. J. Physiol. (Lond.) 283, 23–39 (1978).

17. Sneddon, P. & Westfall, D. P. Pharmacological evidence that adenosinetriphosphate and noradrenaline are co-transmitters in the guinea-pig vasdeferens. J. Physiol. (Lond.) 347, 561–580 (1984).

18. Minneman, K. P., Fox, A. W. & Abel, P. W. Occupancy of α1-adrenergicreceptors and contraction of rat vas deferens. Mol. Pharmacol. 23, 359–368(1983).

19. Kenakin, T. P. Pharmacologic Analysis of Drug–Receptor Interaction(Lippincott-Raven, Philadelphia, Pennsylvania, 1997).

20. Hwa, J., Graham, R. M. & Perez, D. M. Identification of critical determinantsof α1-adrenergic receptor subtype selective agonist binding. J. Biol. Chem.270, 23189–23195 (1995).

21. Trendelenburg, U., Maxwell, R. A. & Pluchino, S. Methoxamine as a tool toassess the importance of intraneuronal uptake of l-norepinephrine in thecat’s nictitating membrane. J. Pharmacol. Exp. Ther. 172, 91–99 (1970).

22. Northover, B. J. A comparison of the electrophysiological actions ofphentolamine with those of some other antiarrhythmic drugs on tissuesisolated from the rat heart. Br. J. Pharmacol. 80, 85–93 (1983).

23. Pérez, O., Valenzuela, C., Delpón, E. & Tamargo, J. Class I and IIIantiarrhythmic actions of prazosin in guinea-pig papillary muscles. Br. J. Pharmacol. 111, 717–722 (1994).

24. Przywara, D. A. & Dambach, G. E. Direct actions of cocaine on cardiaccellular electrical activity. Circ. Res. 65, 185–192 (1989).

25. Cusack, B., Nelson, A. & Richelson, E. Binding of antidepressants to humanbrain receptors: focus on newer generation compounds. Psychopharmacology(Berl.) 114, 559–565 (1994).

26. McIntosh, J. M. et al. Isolation and characterization of a novel Conus peptidewith apparent antinociceptive activity. J. Biol. Chem. 275, 32391–32397(2000).

27. Fürst, S. Transmitters involved in antinociception in the spinal cord. BrainRes. Bull. 48, 129–141 (1999).

28. Hu, S. H. et al. The 1.1 Å resolution crystal structure of [Tyr15]-EpI, a novelα-conotoxin from Conus episcopatus, solved by direct methods. Biochemistry37, 11425–11433 (1998).

29. Hu, S. H. et al. The 1.1 Å crystal structure of the neuronal acetylcholinereceptor antagonist, α-conotoxin PnIA from Conus pennaceus. Structure 4,417–423 (1996).

30. Hu, S. H., Gehrmann, J., Alewood, P. F., Craik, D. J. & Martin, J. L. Crystalstructure at 1.1 Å resolution of α-conotoxin PnIB: comparison with α-conotoxins PnIA and GI. Biochemistry 36, 11323–11330 (1997).

31. Balaji, R. A. et al. λ-Conotoxins, a new family of conotoxins with uniquedisulfide pattern and protein folding. Isolation and characterization from thevenom of Conus marmoreus. J. Biol. Chem. 275, 39516–39522 (2000).

32. Massot, O. et al. 5-hydroxytryptamine-moduline, a new endogenous cerebralpeptide, controls the serotonergic activity via its specific interaction with 5-hydroxytryptamine1B/1D receptors. Mol. Pharmacol. 50, 752–762 (1996).

33. Olivera, B. M. & Cruz, L. J. Conotoxins, in retrospect. Toxicon 39, 7–14(2001).

34. Loughnan, M. et al. α-Conotoxin EpI, a novel sulfated peptide from Conusepiscopatus that selectively targets neuronal nicotinic acetylcholine receptors.J. Biol. Chem. 273, 15667–15674 (1998).

35. Brüss, M., Pörzgen, P., Bryan-Lluka, L. J. & Bönisch, H. The ratnorepinephrine transporter: molecular cloning from PC12 cells andfunctional expression. Brain Res. Mol. 52, 257–262 (1997).

36. Percy, E. et al. Catechol-O-methyltransferase activity in CHO cells expressingnorepinephrine transporter. Br. J. Pharmacol. 128, 774–780 (1999).

37. Nielsen, K. J. et al. Effects of chirality at Tyr13 on the structure–activityrelationships of ω-conotoxins from Conus magus. Biochemistry 38,6741–6751 (1999).

38. Fainzilber, M. et al. New mollusc-specific α-conotoxins block Aplysianeuronal acetylcholine receptors. Biochemistry 33, 9523–9529 (1994).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

GABAA receptors are the major sites of fast synaptic inhibitionin the brain. These receptors are pentameric hetero-oligomersthat can be assembled from 6 subunit classes with multiple mem-bers: α1–6, β1–3, γ1–3, δ, ε and θ1–4. Each of these subunitsshares a common transmembrane topology, incorporating anextracellular N-terminal domain and 4 transmembrane domainswith a major intracellular domain between transmembranedomains 3 and 4 (ref. 5). It is likely that, in vivo, most GABAAreceptor subtypes are constructed from α, β and γ subunits1–4.

For efficient inhibitory synaptic transmission, it is critical thatGABAA receptors are correctly targeted to and clustered at inhibito-ry synapses. The clustering of GABAA receptors at synaptic sites isat least partially dependent upon gephyrin, a microtubule bind-ing protein that is also essential in glycine receptor clustering6–8.Another microtubule protein, GABA receptor-associated protein(GABARAP), which binds to the γ2 subunit, has been suggestedto have a role in the membrane trafficking of GABAA receptors9,10.

GABAA receptor number at synaptic sites can be modified bygrowth factor receptor activation, kindling and neuronal activi-ty11–14. Furthermore, the endocytosis of synaptic GABAA receptorshas been demonstrated both in hippocampal and cortical neurons,and receptors are found localized both to endosomes and clathrin-coated pits14–19. Together, these results demonstrate that synapticGABAA receptors cycle between the cell surface and intracellularendocytic compartments. Whereas GABAA receptor internaliza-tion can be largely attributed to dynamin-dependent endocyto-sis14,17, the neuronal mechanisms that facilitate the insertion ofnewly synthesized or recycled receptors into neuronal membranesto maintain a stable cell surface receptor number remain unknown.

GABAA receptor cell surface numberand subunit stability are regulatedby the ubiquitin-like protein Plic-1

Fiona K. Bedford1, Josef T. Kittler1, Emilie Muller2, Philip Thomas3, Julia M. Uren1, DanielaMerlo4, William Wisden4, Antoine Triller2, Trevor G. Smart3 and Stephen J. Moss1

1 Medical Research Council Laboratory of Molecular Cell Biology and Department of Pharmacology, University College London, Gower Street, London WC1E 6BT,UK

2 Laboratoire de Biologie Cellulaire de la Synapse Inserm U497, Department of Biology, ENS 46 Rue d’Ulm Paris 75005 Paris, France3 Department of Pharmacology, The School of Pharmacy, 29-39 Brunswick Square, London WC1E 1AX, UK4 Medical Research Council Laboratory of Molecular Biology, Cambridge, CB2 2QH, UK

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

Controlling the number of functional γ-aminobutyric acid A (GABAA) receptors in neuronalmembranes is a crucial factor for the efficacy of inhibitory neurotransmission. Here we describe thedirect interaction of GABAA receptors with the ubiquitin-like protein Plic-1. Furthermore, Plic-1 isenriched at inhibitory synapses and is associated with subsynaptic membranes. Functionally, Plic-1facilitates GABAA receptor cell surface expression without affecting the rate of receptorinternalization. Plic-1 also enhances the stability of intracellular GABAA receptor subunits, increasingthe number of receptors available for insertion into the plasma membrane. Our study identifies apreviously unknown role for Plic-1, a modulation of GABAA receptor cell surface number, which sug-gests that Plic-1 facilitates accumulation of these receptors in dendritic membranes.

Here we demonstrated that the ubiquitin-related protein Plic-1(refs. 20, 21) can facilitate the membrane insertion of GABAA recep-tors, an affect mediated via increasing the stability of intracellularreceptor pools. This ability of Plic-1 to control GABAA receptor cellsurface stability suggests that this protein may be important in theaccumulation of GABAA receptors at inhibitory synapses.

RESULTSPlic-1 binds to the GABAA receptorTo search for proteins involved in the membrane trafficking ofGABAA receptors, we used yeast two-hybrid screens22 (Y2H) witha bait encoding the predicted major intracellular domain betweentransmembrane domains 3 and 4 of the GABAA receptor α1 sub-unit1–5. Screening a rat hippocampal library23, we isolated sever-al partially overlapping clones that encoded a 67-kD protein thatshowed very high identity (99%) with the mouse and human pro-tein Plic-1 (refs. 20, 21). Plic-1 homologs have also been identi-fied previously in Xenopus, and Plic-1 is homologous to the yeastprotein Dsk 2 (ref. 24). Plic-1 is a ubiquitin-like protein that has anamino terminus that is 33% identical to ubiquitin and a C-ter-minal ubiquitin-associated (UBA) domain25,26 (Fig. 1a), with thecentral portions predicted to be rich in α helices. To examine thebinding specificity of Plic-1 for various GABAA-receptor subunits,the interaction of the Y2H Plic-1 clone 8A with the intracellulardomains of other receptor subunits was tested in yeast. Clone 8Abound to the GABAA receptor α1–3, α6 and β1–3 (refs. 1–4, 27)subunit intracellular domains but not to the correspondingregions of the GABAA receptor γ2L or δ subunits27 or to the intra-cellular domain of the neuronal nicotinic acetylcholine receptor α7

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


lysates from A293 cells transiently expressing the GABAA recep-tor α1 subunit modified at its N-terminus with the Myc 9E10epitope (9E10α1)15–18. GST-clone 8A (Fig. 2a, construct 3) andfull-length GST-Plic-1 (Fig. 2a, construct 1) but not GST alonewere able to bind to the GABAA

9E10α1 subunit (Fig. 2a) as deter-mined by immunoblotting. Further deletion constructs of Plic-1were analyzed, demonstrating that the UBA domain of Plic-1 wassufficient to mediate the interaction of Plic-1 with the 9E10α1subunit (Fig. 2a, constructs 2, 4 and 5).

To further examine the association between GABAA recep-tors and Plic-1, we produced antisera (anti-Plic-1) against thisprotein using GST-clone 8A as an antigen. This affinity-purifiedantisera detected a band of 67 kD in extracts from total brain, aswell as from isolated cortex and cerebellum (Fig. 2b), similar insize to that predicted by cDNA cloning (Fig. 1a). Detection ofthis band could be completely abolished with an excess of immu-

subunit (Fig. 1b), which is a member of the same receptor super-family as GABAA receptors5. As all GABAA receptor subtypes inthe CNS contain α and β subunits, these results suggest that Plic-1 interacts with all GABAA receptor subtypes. To identify theamino acids on the α1 subunit that were responsible for bindingPlic-1, we performed Y2H assays using fragments of the α1 sub-unit intracellular domain. This analysis revealed that amino acids346–355 (NYFTKRGYAW) within the α1 subunit were sufficientto mediate interaction with Plic-1 (Fig. 1c). The first 5 residuesof this sequence are conserved in the GABAA receptor α1–3, α6and β1–3 subunits, but are not conserved in the γ2 and δ receptorsubunits1,2, which could explain the observed specificity of Plic-1for GABAA receptor α and β subunits.

To confirm our observations from yeast, we examined theinteraction of Plic-1 with GABAA receptor subunits using affin-ity purification assays (pull-downs) and western blotting. Clone8A and other fragments of Plic-1 were expressed as glutathione-S-transferase (GST) fusion proteins and immobilized on glu-tathione-agarose28. These reagents were then used to probe cell

Fig. 2. Interaction of Plic-1 and GABAA receptors in neurons and recombinant systems. (a) The UBA domain of Plic-1 specifies interaction with the 9E10α1subunit. Full-length Plic-1 (1) and deletion constructs (2–5) of Plic-1 GST fusion proteins used to map the binding domain on Plic-1 for GABAA receptors.Positive interactions were determined by affinity purification of 9E10-tagged 9E10α1 subunits and western blotting of purified complexes with anti-9E10.The lysate lane (Lys) represents 10% of the homogenate used. +++, strong interaction; –, no interaction. (b) Adult rat tissue extracts (50 µg), from totalbrain (B), cortex (Co), cerebellum (Ce) and A293 cell extract (293; 250 µg) were western blotted with anti-Plic-1 (5 µg/ml).

(c) Complexes of Plic-1 and GABAAreceptors are found in neurons.Solubilized brain homogenates were sub-jected to immunoprecipitation with anti-Plic-1 with (+) or without (–) a 50-foldmolar excess of immunizing antigen (Ag)or control IgG. Immunoprecipitates wereimmunoblotted using anti-Plic-1 or anti-GABAA β2/3 (bd17, 5 µg/ml) as indi-cated. The lysate lane represents 10% ofthe input used for the immunoprecipita-tions. (d) Co-immunoprecipitation of Plic-1 with GABAA subunits in A293 cells.A293 cells were either untransfected (lane1) or transiently transfected with Plic-1alone (lane 2) or co-transfected with the9E10α1 subunit (lanes 3 and 4), 9E10β3 sub-unit (lane 5) or 9E10γ2S subunit (lane 6).The cells were [35S]-methionine labeledand extracts were immunoprecipitated(IP) with anti-Plic-1 antibodies. Lane 4(+Ag) indicates that the anti-Plic-1 anti-bodies were incubated with a 50-foldexcess of the immunizing antigen.

Plic-1

UBL UBAα-HR1 α-HR2

α1 α2 α3 α6 β1 β2 β3 γ2L

8A +++ +++ +++ +++ +++ +++

nAchRα7δ

+++ –– –

TM4α1

β-Gal Activity

323–454 +++

323–385 +++323–364 +++

323–355 +++346–385 +++

346–364 +++

386–454 –

TM3

Fig. 1. Identification of an interaction between GABAA receptors andPlic-1. (a) Plic-1. The full-length Plic-1 cDNA encodes a predicted pro-tein of 69.84 kD. The encoded protein shows two conserved domains,an amino terminal ubiquitin-like domain (UBL, residues 26–103) and aC-terminal ubiquitin-associated domain (UBA, residues 534–582), inaddition to two potential central domains rich in α-helices (α-HR1,residues 160–317 and α-HR2, residues 365–476). (b) Plic-1 interactswith many GABAA receptor subunits. Interactions between clone 8Aand various GABAA receptor intracellular domains was assayed by Y2Hand quantified using a liquid β-galactosidase assay. (c) Identification ofthe Plic-1 binding site on the α1 subunit. Deletion constructs of theGABAA receptor α1 subunit intracellular domain were used as Y2Hbaits to map the minimal Plic-1 binding domain. Interactions betweenthe α1 subunit intracellular domain deletion constructs and clone 8Awere assayed by Y2H and quantified as described in (b). +++, stronginteraction; –, no interaction.

1 2 3 4 5Lys Gst

α19E10

+++

—

+++

+++

+++

UBA

UBAα-HR-2

UBA

UBL α-HR-1 α-HR-2 UBA

Construct

1

2

3

4

5

Bound

UBL α-HR-1

Plic-1 –

GABAA β2/3–

Anti-Plic-1

IgG

Lys +

Ag–

66

97.4

B Co Ce 293

Plic-1

IP: anti- Plic-1

Plic-1

+Ag

66

451 2 3 4 5 6

a

c

ab

c d

b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

nizing antigen (data not shown). Expression of the Plic-1 pro-tein could also be detected in A293 cells, but at lower levels thanin brain (Fig. 2b). Previous studies have revealed that Plic-1mRNA has a broad pattern of tissue expression20,21. WhetherPlic-1 and GABAA receptor complexes occur in neurons was nexttested via immunoprecipitation. Solubilized brain extracts weresubjected to immunoprecipitation with anti-Plic-1 antibodiesand immunoblotted with either anti-Plic-1 or bd17, a mono-clonal antibody that recognizes GABAA receptor β2/3 subunits.GABAA β2/3 subunits assemble with α and γ2 subunits in neu-rons to form benzodiazepine-sensitive GABAA receptors29, thepredominant GABAA receptor subtype in the CNS. Co-immuno-precipitated GABAA β2/β3 subunits were detected by westernblotting with bd17, as bands of 55–58 kD from immunoprecip-itations with anti-Plic-1, but not with control IgG, or when thePlic-1 antibody had been blocked with antigen (Fig. 2c).Immunoprecipitation of Plic-1 was also confirmed by westernblotting precipitated material with anti-Plic-1 antibodies (Fig. 2c). To analyze the specificity of the Plic-1 interaction tofull-length GABAA receptor subunits, we expressed recombinantreceptor subunits and Plic-1 in A293 cells. In this system, theGABAA receptor 9E10α1 (bands 52 and 48 kD)15,16 and β3 sub-unit (major band, 55 kD)15,16 co-immunoprecipitated with anti-Plic-1 but not with control IgG (Fig. 2d) or anti-Plic-1 blockedwith the immunizing antigen. The γ2 subunit did not co-immunoprecipitate with Plic-1, when co-expressed in A293 cells,confirming our observations in yeast (Fig. 1b).

Subcellular distribution of Plic-1 and GABAA receptorsWe next examined the subcellular localization of Plic-1 expressionusing immunofluorescence. In cultured hippocampal neurons,Plic-1 exhibited an intracellular clustered distribution with

Fig. 3. Colocalization of Plic-1 with GABAA receptors. Hippocampalneurons, 21 days old, were immuno-stained for Plic-1 (anti-Plic-1, 5µg/ml; a) and GABAA receptors (bd17, 10 µg/ml; b); merged image (c).Scale bar, 20 µm. A293 cells transiently expressing 9E10α1β3γ2S GABAAreceptors and Plic-1 were stained for Plic-1 (anti-Plic-1, 5 µg/ml; d) andthe GABAA receptor β3 subunit (anti-9E10, 5 µg/ml; e); merged image(f). Arrows (a–c) identify colocalized clusters of Plic-1 and GABAAreceptors on dendritic shafts of hippocampal pyramidal neurons.Arrowheads (f) identify colocalized pools of Plic-1 and GABAA receptorsin transiently transfected A293 cells. Scale bar, 10 µm. The data are single0.5-µM optical sections; representative cells are shown in each case.

immunoreactivity in neuronal processes (Fig. 3aand c). Immunoreactivity, for the GABAA receptorβ2/3 subunits, exhibits a membrane and intracel-lular clustered staining pattern in hippocampal neu-rons (Fig. 3b and c). A significant subset of GABAAreceptor clusters were within or just below the cellsurface that contained Plic-1 immunoreactivity(Fig. 3c). In A293 cells, low levels of endogenousPlic-1 prevented reliable detection of this proteinby immunofluorescence. However, cells transient-ly co-expressing Plic-1 with α1β3γ230 GABAAreceptors revealed that the distribution of these pro-teins often co-localized in intracellular structuresbeneath the plasma membrane (Fig. 3d–f).

We examined the relationship of Plic-1 tosynapses at the EM level in the ventral spinal cordwhere neuronal activity is controlled by glycineand GABA-mediated inhibition32,33. Using animmunoperoxidase method (data not shown),Plic-1 immunoreactivity (IR) was detected in

intracellular compartments and consistently found at synapticsites. With pre-embedding immunolabeling and gold-toned sil-ver-intensified nanogold particles, the staining was found scat-tered in the dendroplasm (Fig. 4a) and could also be detectedin the vicinity of postsynaptic membranes (Fig. 4a–c), the latterbeing in front of presumed inhibitory synaptic boutons con-taining pleiomorphic vesicles. The particle density over the den-drites was 3.8/µm2 ± 0.2 (n = 83) and was significantly higher (p < 0.0001) than that found over presynaptic boutons (0.9 ± 0.1, n = 44). To estimate the particle density associatedwith the plasma membrane, we have computed particles at lessthan 30 nm from the plasma membrane, thus defining a stripeof that width. The particle density was then 14.8/µm2 ± 1.1 (n = 83). Comparison with the density over the dendroplasmindicate that the association with the plasma membrane did notresult from a random process (p < 0.0001). These Plic-1 asso-ciated particles were frequently adjacent to minute cisternae.Such subsynaptic cisternae contain many components of theprotein synthetic machinery33 and may be part of subsynaptictubulovesicular endosomes. The inhibitory nature of the presy-naptic boutons was confirmed as they were GAD-IR (Fig. 4d) orwere apposed to gephyrin (Fig. 4e), which is involved in the sta-bilization of glycine and GABA6–8. In some cases, there was evi-dence of the presence of the GABAA receptor β2–3 subunit inassociation with the Plic-1-IR (Fig. 4f) and with the bd17 anti-body (see Methods). Plic-1-associated gold particles were alsofound at coated pits, at the border of the postsynaptic differen-tiation, with both pre- (Fig. 4g) and post- (Fig. 4h) embeddingimmunogold methods. In addition to its association with den-dritic and postsynaptic cisternae (Fig. 4a–c), Plic-1-IR was alsodetected at the edge of the Golgi apparatus (Fig. 4i) and in asso-ciation with other intracellular membranes. These associationsdo not result from a random distribution of particles, becausethe intracellular membrane-associated particle density(18.3/µm2 ± 1.3, n = 83) was higher than the density at the plas-ma membrane (p < 0.05) and the density over the whole den-droplasm (p < 0.0001).

The localization of Plic-1 to intracellular compartments in addi-tion to GABAergic synapses suggests that Plic-1 is not an anchor-ing protein for GABAA receptors at the cell surface. Furthermore,the distribution of Plic-1 within clathrin-coated pits, the Golgiapparatus and intracellular membranes suggests that this proteinmay be involved in the membrane trafficking of GABAA receptors.

a b c

d e f

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


Fig. 5. Functional regulation of GABAA receptors by the Plic-1-blockingpeptide pepPBα. (a) PepPBα but not a scrambled control inhibits thebinding of Plic-1 to the GABAA receptor 9E10α1 subunit. Extracts of A293cells expressing 9E10α1 were exposed to the UBA domain of Plic-1expressed as a GST fusion protein (lanes3–5) or GST alone (lane 2), in the pres-ence of 1 µM pepPBα (lane 5) or scram-bled control (lane 4); lane 1 represents10% of the input used. Bound 9E10α1 wasdetected by western blotting with anti-9E10. (b, c) Blocking the interaction ofPlic-1 modifies GABA-activated mem-brane currents. GABA- (10 µM) acti-vated membrane currents recorded fromA293 cells expressing α1β3γ2S GABAAreceptors at a holding potential of–40mV at selected times (t) after forma-tion of the whole-cell recording mode(defined as t = 0). The patch pipette con-tained either pepPBα (b) or a scrambledpeptide (c) at 200 µg/ml. (d) Time depen-dence of the effect of pepPBα and scram-bled peptides on 10 µM GABA-activatedcurrents. Data were collated from cellsby normalizing successive GABA currentamplitudes to the initial response toGABA (defined as 1) in each A293 cellwithin 3 min of whole-cell formationeither in the absence (control) or pres-ence of pepPBα or scrambled peptides.All points are mean ± s.e.m. from n = 4control and n = 5–7 peptide-treated cells.

Plic-1 modulates GABAA receptor activityTo analyze the functional consequences of the interactionbetween Plic-1 and GABAA receptors, we first focused on recom-binant receptors expressed in A293 cells. In contrast to neuronalreceptors, the assembly, membrane insertion and endocytosisof GABAA receptors are well documented in this system15–18,31.We examined the effects of disrupting the binding of Plic-1 toGABAA receptor function by measuring GABA-activated mem-brane currents in A293 cells expressing recombinant GABAAreceptors via whole-cell patch clamp recording. For these exper-iments, we used a peptide corresponding to the Plic-1 bindingsite on the α1 subunit (pepPBα). To test if this reagent coulddisrupt the interaction between the GABAA receptor α1 subunit

and Plic-1, pull-down assays were used. GST-clone 8A or GSTwas exposed to extracts of A293 cells expressing the GABAAreceptor 9E10α1 subunit, in the absence and presence of pepP-Bα. GST-clone 8A, but not GST alone, was able to bind to the9E10α1 subunit, and this interaction could be specifically blockedby pepPBα but not by a scrambled control peptide (Fig. 5a).Transfected A293 cells expressing α1β3γ2S constructs were inter-nally perfused with pepPBα, which resulted in a time-depen-dent reduction in whole-cell GABA-activated current amplitudes(Fig. 5b and d) soon after formation of the whole-cell record-ing conformation. The depression in response was up to 40%after a 30-minute recording period. In contrast, the scrambledpeptide had no effect; recordings made with patch pipettesincluding this peptide were similar to recordings performedusing a control patch pipette electrolyte (Fig. 5c and d). Equi-librium GABA concentration–response curves showed that therewas little change in GABA potency (Fig. 6a), suggesting thatpepPBα had no overt effects on GABAA channel function. How-ever, cells treated with pepPBα still displayed smaller currentamplitudes (Fig. 5b and d). PepPBα also had no effect on sin-gle-channel conductance or the mean open time (data notshown). These data are therefore consistent with an involvementof Plic-1 in controlling receptor cell surface stability and number,rather than modulating ion channel function.

Fig. 4. Ultrastructural localization of Plic-1 in neurons. (a–c) Gold-toned silver intensified nanogold particles (arrows) are detected in thepostsynaptic cytoplasm, frequently associated with citernae (arrow-heads). Gold particles associated to cisternae are frequently seen closeto the postsynaptic membrane (b, c). Presynaptic boutons containpleiomorphic populations of vesicles. (d) Plic-1-IR in front of a GAD-IRbutton. (e) Simultaneous detection of gephyrin (*) and Plic-1-associatednanogold particles. (f) Presence of Plic-1-associated particle close toGABAA receptor immunoreactive (bd17, crossed arrow) membrane.(g, h) Association of Plic-1-associated particles with coated invagina-tions (crossed arrows). (i) Detection of Plic-1-IR at the membrane ofthe Golgi apparatus. All images except (h) are from pre-embeddingimmunocytochemistry; (h) is from post-embedding immunolabeling.Scale bars, 0.25 µm.

α19E10

1 2 3 4 5

10 µM GABA + internal scrambled peptide

5 s

200 pA

t = 1 min t = 3 min t = 8 min t = 15 min

t = 1 min t = 3 min t = 8 min t = 15 min

10 µM GABA + internal pepPBα peptide

5 s

200 pA

GA

BA

-act

iva

ted

cu

rren

t

Recording time (min)

(n

orm

alis

ed)

0 5 10 15 20 25 300.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

controlscrambledpepPBα

a b c

d e f

g h i

a b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

We also analyzed the effects of pepPBα on GABA-mediatedcurrents in CA1 neurons of hippocampal brain slices using intra-cellular dialysis to introduce either pepPBα or scrambled con-trol peptides (Fig. 6b). Neurons exposed to scrambled peptideor just patch electrode electrolyte showed stable response cur-rents, which were well maintained with time. However, in neu-rons treated with pepPBα, a statistically significant decrease of20% in current amplitude occurred after 10 or 30 minutes ofrecording. Our electrophysiological results from both neuronaland recombinant preparations strongly suggest that Plic-1 is adeterminant of GABAA receptor surface stability.

Plic-1 modulates GABAA receptor cell surface expressionTo further address the mechanism of Plic-1 action, we examinedthe levels of α1β3γ2S GABAA receptors expressed on the cell sur-face of A293 cells using a whole-cell ELISA (enzyme-linkedimmunosorbent assay)34. For these experiments, we used pep-tides modified with a sequence encoding the Antennapedia (Antp)internalization sequence35. In A293 cells exposed to the Antp-PBα peptide, a significant reduction in cell surface receptor lev-els was detected after two to four hours of treatment, which wasnot seen with the scrambled peptide (Fig. 7a). The differencebetween the onset of this reduction compared to the electro-physiological observations is presumably due to the faster deliv-ery of the peptide via the patch pipette and the higherintracellular concentrations of peptide in the electrophysiologi-cal experiments. These data are consistent with Plic-1 actingeither to prevent GABAA receptor removal (for example, by endo-cytosis) or to promote its insertion in the plasma membrane.Plic-1 is not likely to be involved in controlling receptor anchor-ing at the cell surface, as this protein does not seem to be entire-ly associated with cell surface receptor pools (Figs. 3 and 4). Todirectly test the role of Plic-1 in controlling receptor endocyto-sis and membrane anchoring, the internalization of cell-surfaceα1β3 γ2S GABAA receptors in A293 cells was measured in thepresence of Antp-PBα or control peptide using a modified whole-cell ELISA assay34. Cells expressing GABAA receptors compris-ing 9E10α1, 9E10β3 and 9E10γ2S subunits were pre-labeled with9E10 antibody. Cells were then allowed to endocytose surface-labeled receptors in the presence or absence of Antp-PBα or con-

trol peptide. This approach demonstrated that GABAA receptorsinternalized at similar rates under all conditions, suggesting thatPlic-1 plays a minimal role in mediating receptor endocytosis ormembrane anchoring (Fig. 7b).

To determine if Plic-1 is involved in controlling cell-surfacelevels of native neuronal GABAA receptors, a similar approachwas adopted using cultured hippocampal neurons. In this sys-tem, we previously established that synaptic GABAA receptorsare undergoing constitutive endocytosis, with receptors cyclingbetween the cell surface and intracellular endocytotic struc-tures16,17. Antp-PBα peptide produced a significant reduction inthe cell-surface levels of GABAA receptors as monitored using anantibody directed against an extracellular epitope in the β2/3subunits29 (Fig. 7c). The percentage loss of cell-surface GABAAreceptors was smaller in the neurons compared to A293 cells.This is probably due to the slower rate of constitutive endocyto-sis of GABAA receptors in neurons compared to A293 cells6. Ourcombined observations using both neuronal and recombinantpreparations suggest that Plic-1 enhanced the cell-surface levelsof GABAA receptors independent of both anchoring or effects onthe endocytosis of receptors.

Plic-1 enhances the stability of GABAA receptor subunitUbiquitin-related proteins have been suggested to have a numberof functions, including linking binding partners to integrins, mod-ulating target protein degradation and regulating the protea-some25,26. To test the role of Plic-1 activity in controlling GABAAreceptor stability, the receptor 9E10β3 subunit, which is capable offunctional homomeric cell surface expression30,36,37, was co-expressed with Plic-1 in A293 cells. Subunit stability was then ana-lyzed after a brief labeling period with [35S]-methionine followedby a 4-hour chase period. This approach revealed that co-expres-sion of the 9E10β3 subunit with Plic-1 significantly enhanced thestability of the 9E10β3 subunit (Fig. 8a). This effect was quantifiedusing a phosphoimager (Fig. 8b). In cells that expressed the 9E10β3subunit alone, much of 9E10β3 subunit was degraded with only57 ± 4.5% remaining after the chase period (Fig. 8b; n = 3), con-sistent with a short half-life for the 9E10β3 subunit, as demon-strated previously36,37. Co-expression of Plic-1 with the 9E10β3subunit produced an increase in amount of the 9E10β3 subunit

Fig. 6. Further characterization of theeffects of pepPBα on GABAA receptorfunction. (a) Equilibrium GABA concen-tration–response curves recorded fromα1β3γ2S GABAA receptor-expressingcells internally perfused with controlpipette electrolyte or electrolyte sup-plemented with 100 µg/ml pepPBα pep-tide. The curves are nonlinearleast-square fits to the data using the Hillequation generating EC50 values of 10.36± 0.82 µM (control) and 15.83 ± 0.66µM (pepPBα), and Hill coefficients of1.04 ± 0.08 (control) and 1.3 ± 0.06(pepPBα). Data are from n = 4 cells andnormalized to the responses induced by500 µM GABA in each cell (defined as 1). (b) Whole-cell GABA-activated currents recorded from hippocampal slices following brief exposure to 10 µM GABA using a control pipette solution or a solution supplemented with either 200 µg/ml scrambled version of pepPBα or 200 µg/ml ofpepPBα. Two recording time (t) points after establishing the whole-cell recording mode are represented. Current amplitudes have been normalizedto the control currents recorded at t = 10 min. All values are mean ± s.e.m. from n = 8–10 neurons. A one-way ANOVA with Tukey–Kramer after thetest indicated that at t = 30 min, the GABA-activated currents exposed to pepPBα are significantly reduced compared to control currents and cur-rents recorded from cells treated with the scrambled peptide (P < 0.05); however, GABA-activated currents recorded from controls and scrambledpeptide treated cells are not significantly different.

control+pepPBα

10 min 30 min

GA

BA

-act

iva

ted

cu

rren

t

(

no

rma

lized

)

Whole-cell recording time

0.0

0.2

0.4

0.6

0.8

1.0

1.2

Control+200 µg/ml PepPBα+200 µg/ml Scram.

GA

BA

-act

iva

ted

cu

rren

t

(

no

rma

lized

)

a b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


Fig. 7. Plic-1 modulates cell surface expression ofGABAA receptors. (a) Cell-surface levels ofGABAA receptors in A293 cells are modified byAntp-PBα . A293 cells expressing 9E10-taggedα1β3γ2S GABAA receptors were incubated withAntennapedia modified Plic-1 binding site peptides(Antp-PBα , 5 µM), an Antennapedia scrambledversion of this peptide (Antp-sc, 5 µM) or vehicleonly, at 37°C for 2 or 4 h. Cell surface receptorlevels were quantified by non-permeabilizedwhole-cell ELISA assays n > 6 in each case. Resultsare normalized and presented as means ± s.d. aspercent of control (cont). Asterisk, significantreduction of 34 ± 4% (p < 0.05, Student’s t-test, n = 8) in surface GABAA receptor expression. (b) Internalization of GABAA receptors in A293cells is not affected by Antp-PBα . Expressing cellswere labeled with saturating concentrations of9E10 antisera and then incubated at 37°C for var-ious time periods in the presence and absence of5 µM Antp-PBα or Antp-Sc peptides. Cell surfacelevels of antibody were then measured by whole-cell ELISA as described in (a). (c) Antp-PBα pep-tides reduce cell-surface levels of GABAAreceptors in hippocampal neurons. Cultured neu-rons were incubated for 4 h with 5 µM Antp-PBα ,Antp-sc or vehicle only (cont) and were then assayed by non-permeabilized cellular ELISA using bd17 and treated as in (a). Antp-PBα but not con-trol peptide produced a statistically significant reduction in GABAA receptor cell surface expression as indicated by the asterisk of 19.5 ± 2% (p < 0.05, Student’s t-test, n = 6). (d) Cell-surface levels of GABAA receptors in A293 cells are modulated by Plic-1. A293 cells transiently express-ing 9E10α1β3γ2S GABAA receptors alone, co-expressing Plic-1 or co-expressing a control vector were assayed for surface receptor levels by non-permeant whole-cell ELISA as described in (a). Results are normalized and presented as means ± s.e.m. as percentage of control. Co-expressionof Plic-1 produced a statistically significant increase in GABAA receptor cell surface expression, by 268 ± 3.2% (p < 0.0001, Student’s t-test, n = 3),as indicated by an asterisk.

remaining after the chase period to 87 ± 9%, which is significantlydifferent from cells expressing 9E10β3 alone (Fig. 8b; n = 3, P >0.001, Student’s t-test). The stabilization of GABAA receptor sub-units by Plic-1 is consistent with a proposed role for some ubiq-uitin-like proteins as negative regulators of proteasomeactivity25,26. To confirm that GABAA receptors are substrates ofthe proteasome, cells expressing either 9E10α1 or 9E10β3 subunitswere treated with lactacystin for 6 hours. Lactacystin, a specificinhibitor of proteasome activity, produced a large increase in thesteady state levels of both subunits (Fig. 8c). To determine if lac-tacystin modulates GABAA receptor subunit stability, pulse chaseanalysis was used in the presence and absence of lactacystin. Thisrevealed that inhibiting the activity of the proteasome produced adramatic increase in the stability of the 9E10β3 subunit (Fig. 8d).

To determine whether the stabilization of GABAA receptor sub-units by Plic-1 produces a resulting increase in cell surface expres-sion of receptors, we overexpressed Plic-1 and measured surfacereceptor levels by whole-cell ELISA. In A293 cells expressing9E10α1β3γ2 GABAA receptors, overexpression of Plic-1 produced astatistically significant increase in total surface receptors by 268 ±3.25%, p < 0.0001, compared to cells expressing receptor subunitsalone or in cells expressing a control 117 ± 2.7% (Fig. 7d). There-fore, our combined observations strongly suggest that Plic-1 facil-itates the insertion of GABAA receptors in the plasma membranelargely by its ability to stabilize intracellular receptor pools.

DISCUSSIONGABAA receptors are critical mediators of synaptic inhibition inthe brain1,2. At synapses, GABAA receptors constitutively under-go endocytosis via clathrin-coated pits in a dynamin-dependentprocess16,17. Therefore, to maintain a stable cell-surface receptornumber, continual membrane insertion of newly synthesized or

recycled receptors is required. However, how neurons facilitatethe insertion of GABAA receptors into synaptic membranesremains to be determined. This issue is not only of importancefor inhibitory synaptic transmission, as AMPA-type glutamatereceptors, the major sites of excitatory synaptic transmission inthe brain, also cycle between the plasma membrane and intra-cellular compartments34,38.

To address the mechanisms underlying GABAA receptormembrane trafficking, we searched for previously unidentifiedbinding partners of GABAA receptor subunits. Here we iden-tified an interaction between GABAA receptors and the ubiq-uitin-like protein Plic-1 (refs. 20, 21). This interaction wasmediated via the UBA domain25,26 of Plic-1, and the intracel-lular domain of GABAA receptor α and β subunits. The broadsubunit binding specificity of Plic-1 suggested that most pop-ulations of GABAA receptors are probably capable of interact-ing with this protein. Blocking the interaction of Plic-1 andGABAA receptors resulted in a loss of cell-surface receptornumber with no effects on receptor internalization. Consistentwith this, overexpression of Plic-1 resulted in an increase intotal cell surface receptor number. This suggested that Plic-1could facilitate the membrane insertion of GABAA receptors.Plic-1 increased GABAA receptor subunit half-life, which isconsistent with a previously suggested role for Plic-1 or thecorresponding human homolog hPlic-1 as negative modula-tors of proteasome activity21,39. Importantly, constitutive degra-dation of GABAA receptor subunits by the proteasome was alsoidentified in our study.

The ability of Plic-1 to increase the half-life of intracellularGABAA receptor pools would account for Plic-1’s ability to facil-itate the insertion of these receptors into synaptic membranes.Plic-1 may also be involved in controlling receptor biosynthesis by

110

90

100

80

70

0Cont 5 µM

Antp-PBα5 µM

Antp-sc

110

90

100

80

70

0Cont 2 h 4 h

Per

cen

t co

ntr

ol,

ab

sorb

ance

(65

5 n

M)

2 h 4 h

5 µM Antp-PBα

5 µMAntp-sc

50

100

200

300

150

250

*

α1β3γ2S α1β3γ2S+ Plic-1

α1β3γ2S+ vector

*

*

0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Time (min)12090300 60

0

Ab

sorb

ance

Control

Antp-scAntp-PBα

Per

cen

t co

ntr

ol,

abso

rban

ce

Per

cen

t co

ntr

ol,

ab

sorb

ance

(65

5 n

M)

(65

5 n

M)

(65

5 n

M)

a b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

stabilizing unassembled subunits in the ER, which may increasethe production of heteromeric receptors. Oligomerization of indi-vidual GABAA receptor subunits into heteromers occurs within 5minutes after translation, but is inefficient; less than 25% of trans-lated subunits are assembled into heteromeric receptors40. Fur-thermore, transport of these assembled receptors to the cellsurface is slow, taking between four to six hours. Because dis-ruption of the interaction of Plic-1 with GABAA receptors pro-duced a rapid decrease in cell surface receptor expression, evidentafter 15 minutes, the primary affect of Plic-1 on GABAA recep-tor cell-surface expression is unlikely to be at the level of receptorassembly in the ER.

The precise mechanism underlying Plic-1 mediated GABAAreceptor stabilization remains to be established, but it is likelythat Plic-1 inhibits GABAA receptor poly-ubiquination, whichwould reduce receptor targeting to the proteasome as suggestedfor other ubiquitin-like proteins25,26. This would then increasethe size of the intracellular pool of GABAA receptors in neuronsavailable for membrane insertion. Whether Plic-1 acts to stabilizenewly synthesized or previously internalized receptors remainsto be determined. Also, the precise mechanisms or signals thatmark receptors for degradation remain to be identified. Howev-er, given the presence of Plic-1 at GABAergic synapses in clathrin-coated vesicles and on intracellular membranes, Plic-1 is thereforesuitably located to stabilize both recycled and de novo synthe-sized GABAA receptors.

The stability of GABAA receptors at synaptic sites depends inpart on a specific interaction with the subsynaptic cytoskeleton com-pared to extrasynaptic receptors6,7,14. However, whether local mem-brane insertion of receptors is also required for synaptic stabilizationof GABAA receptors remains to be determined. Given that Plic-1 ispresent at inhibitory synapses, it will be interesting to examine if thisprotein is specifically involved in facilitating the membrane inser-tion of GABAA receptor at these membrane specializations.

Fig. 8. GABAA receptor subunits arestabilized by Plic-1. (a) Pulse chase of9E10β3 subunits expressed in A293 cellswith Plic-1. A293 cells transiently trans-fected with 9E10β3 subunits alone orwith Plic-1 using equimolar ratios of the9E10β3 cDNA. + denotes transfection ofa cDNA; –, absence of cDNA.Expressing cells were labeled with [35S]-methionine for 1 h and lysed immedi-ately or chased in cold media for 4 h.Extracts were immunoprecipitated withanti-9E10 antibody and analyzed bySDS-PAGE. (b) Levels of 9E10β3 proteinremaining in the presence and absenceof Plic-1, after a 4-h chase period (a),were quantified using a Biorad phos-phoimager (Hemel, Hempstead, UK).After subtraction of background levels,incorporated radioactivity was normal-ized to control, 9E10β3 (t = 0). Resultsare presented as means ± s.d. and per-cent of control, n = 3. (c) GABAAreceptors are substrates of the protea-some. A293 cells transfected with9E10α1 or 9E10β3 were treated with 10 µM lactacystin for 6 h. Lysates wererun on SDS-PAGE and immunoblotted with anti-9E10 antibody. (d) Pulse chase of 9E10β3 subunits exposed to the proteasomal inhibitor, lacta-cystin. A293 cells transiently transfected with 9E10β3 subunits were [35S]-methionine labeled for 1 h and lysed immediately or chased in coldmedia for 30 or 120 min in the presence (+ Lact) or absence of 10 µM lactacystin. The levels of remaining 9E10β3 subunits were quantitated onthe Biorad phosphoimager and are presented as percent control at t = 0.

Finally, given that GABAA receptors cycle between synapticsites and intracellular endocytic structures14–18, the capacity ofneurons to modulate the removal and/or insertion of GABAAreceptors in synaptic membranes may have profound effects onthe efficacy of synaptic transmission. Therefore, the ability ofPlic-1 to modulate the membrane insertion of GABAA receptorssuggests a involvement for this protein in the accumulation ofthese receptors at inhibitory synapses.

METHODSYeast two-hybrid screens. Baits encoding the intracellular domain includ-ing TM3 and 4 of the α1 subunit were amplified by PCR and cloned intothe yeast expression vector pPC97. This bait was used to screen a rat hip-pocampal library of 6.5 × 106 independent clones constructed in pPC86, ofwhich 1.25 × 107 were screened23. The plasmids were transformed intoyeast strain Y190 and transformants were selected on –Leu/His/Trp mediacontaining 25 mM 3AT and assayed for β-galactosidase activity22. Positiveclones were co-transformed with either the bait vector or the pPC86 vec-tor backbone into yeast to confirm interactions. Baits encoding the GABAAreceptor α1, α2, α3, α6, β1–3, γ2 and δ1–4,16 subunits including TMs 3and 4 in pPC97 were screened against clone 8A. To map the region of theα1 subunit responsible for binding Plic-1, deletions of the α1 bait wereconstructed in pPC97 and screened against clone 8A as described above.

Antibodies. Rabbit polyclonal antibodies were raised against GST-clone 8A.The resulting antisera was then affinity purified with GST-clone 8A immo-bilized on Affigel 10 (Biorad, Hemel, Hempstead, UK). A monoclonal anti-body directed against the GABAA receptor β2/3 subunits (bd17) waspurchased from Chemicon (Luton, UK). Secondary antibodies were fromMolecular Probes (Eugene, Oregon) and Jackson Immunochemicals (Har-row, UK) and were used at 1:500. The 9E10 antibody was isolated fromhybridoma cell lines via chromatography on protein A–Sepharose16.

Pull down assays and immunoprecipitation. Extracts of A293 cellsexpressing the GABAA receptor α1 subunit modified with the 9E10 epi-tope16 were made in 25 mM HEPES, pH 7.6, 100 mM KCl, 5 mM EGTA,

66

45

–

+Lact 6 h

–+ +

α19E10

β39E10

Plic-1

+

–

+ + +

++–

β3

0 4 0 4 h

66

45

0 30 120 30 120 min

+ Lact

100 70 54 73 68 %

66

45

100

80

β3 p

rote

in r

emai

nin

g (

% c

on

tro

l)

0

20

40

60

Plic-1

+ +

+–

β39E10

β39E10

β39E10

a b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


5 mM EDTA, 1% Triton-X-100 and a cocktail of protease inhibitors16.Extracts (250 mg protein) were then incubated with GST Plic-1 fusionproteins (10 mg) bound to agarose beads at 4°C for 2 h. Bound proteinswere detected via western blotting with anti-9E10 antibody and visualizedwith ECL16. For immunoprecipitation experiments, extracts from trans-fected A293 cells or brain were prepared in 25 mM Tris-Cl pH 7.6, 150mM NaCl, 1 mM EDTA, 2% Triton-X-100 and a cocktail of proteaseinhibitors. Antibodies used for the immunoprecipitation of GABAA/Plic-1 complexes were cross-linked to protein A–Sepharose16; precipitatedcomplexes were subsequently detected by western blotting and visual-ized using ECL after SDS-PAGE. In some experiments, transfected A293cells were pre-labeled with [35S]-methionine (Amersham, Amersham,UK) and either lysed immediately or chased for defined time periodswith media containing an excess of unlabeled methionine. GABAA recep-tor subunits were then isolated from cell extracts via immunoprecipita-tion and resolved by SDS-PAGE; incorporated radioactivity was thenquantified using a Biorad phosphoimager (Hemel, Hempstead, UK).

Cell culture and immunohistochemistry. Cultures of hippocampal neu-rons were made from embryonic day 18 embryos as described previ-ously16. A293 cells were maintained in DMEM with 10% fetal calf serumand transfected via electroporation16. For immunolocalization studies,neurons (21 days in vitro) or A293 cells were fixed in 4% paraformalde-hyde and processed for immunohistochemistry using antisera againstGABAA receptor β2/3 subunits, the 9E10 epitope or anti-Plic-1, as pre-viously described16. Cells were then immunolabeled with the appropri-ate fluorescently labeled secondary antibodies and images were collectedvia confocal microscopy. For the non-permeant cellular ELISA assays34,neurons or A293 cells were incubated at 37°C with peptides corre-sponding to the Plic-1 binding site on the α1 subunit (RRWRRWWR-RWWRRWRRNYFTKRGYAW), which had been modified at theN-terminus with the Antennapedia internalization sequence35 (includ-ing a biotin residue at the N-terminus) or a scrambled control (RRWR-RWWRRWWRRWRRAGKFNWYRTY) at 5 µM for 2 h or 4 h. Cellswere then fixed and stained under non-permeabilizing conditions withbd17 or 9E10 monoclonal antibodies and goat anti-mouse IgG, washed6 times and incubated with 500 µl of 3,3´,5,5´-tetramethylbenzidine(TMB) for exactly 30 min. The supernatant was transferred to a cuvetteand absorbance was determined at 655 nm34. For the internalizationassays, antibodies were pre-bound at 4°C under saturating conditionsthat do not induce internalization34. The cells were then incubated forvarying time periods at 37°C and the remaining cell surface antibodywas measured as above.

Electrophysiology. GABA-activated currents were recorded from cul-tured hippocampal slices containing native GABAA receptors30 and alsofrom single A293 cells transfected with cDNAs encoding for α1β3 γ2SGABAA receptor subunits and the reporter plasmid for the S65T mutantgreen fluorescent protein. Membrane currents were measured usingwhole-cell patch clamp in conjunction with a List EPC7 patch amplifier(Los Angeles, USA) and filtered at 10 kHz. Patch electrodes (1–8 MΩ)contained 140 mM KCl, 2 mM MgCl2, 1 mM CaCl2, 10 mM HEPES, 11mM EGTA and 2 mM adenosine triphosphate, pH 7.2. The A293 cellswere continuously perfused with a Krebs solution containing 140 mMNaCl, 4.7 mM KCl, 1.2 mM MgCl2, 2.5 mM CaCl2, 10 mM HEPES, 11mM glucose, pH 7.4. GABA was rapidly applied to A293 cells using amodified U-tube as previously reported30

. The Plic-1 peptide (NYFTKR-GYAW) and the scrambled control peptide (AGKFNWYRTY) were dis-solved in pipette electrolyte at 200 µg/ml concentration. The scrambledcontrol contained the same amino acids in a random order. To avoidinterference with forming cell-attached patches, the tip of the electrodewas filled with electrolyte devoid of peptide. GABA concentrationresponse curve data were fitted with the Hill equation.

I/Imax = 1/[1 + (EC50/[A])nH] (1)

Here, I and Imax represent the GABA-activated current at a given con-centration and the maximum current induced by a saturating con-centration of GABA, respectively. EC50 defines the concentration of

GABA, [A], that induces a 50% of the maximum response, and nH isthe Hill coefficient.

Electron microscopy. Adult Sprague–Dawley rats were anesthetized withpentobarbital (60 mg/kg) and intracardially perfused with 4%paraformaldehyde (PFA) and 0.1% glutaraldehyde in PBS, in keepingwith animal use policy at the Ecole Normale Superiéure (ENS). After dis-section, the cervical spinal cord was kept overnight at 4°C in 4% PFA.Vibratome sections (100 µm) were cryoprotected, permeabilized byfreeze-thawing, extensively rinsed in PBS and immersed for 20 min in50 mM ammonium chloride and for 30 min in PBS with 0.1% gelatin(PBSg). For the detection of Plic-1, vibratome sections were incubated12 hours (4°C) in the anti-Plic-1 rabbit polyclonal antibody (1/50 inPBSg) and the antibody binding sites were detected using a goat anti-rabbit nanogold-coupled antibody (1% in PBSg, Nanoprobe, StonyBrook, New York). Gold particles were intensified with HQ silver kit(Nanoprobe) and subsequently gold-toned33. For detection of gephyrin,we have used a monoclonal antibody (Mab 7a, 1:200), the presence ofwhich was determined using immunoperoxidase methods41. After dehy-dration and osmification, the sections were flat-embedded. Observationsof ultrathin sections (pale yellow) were contrasted with uranyl acetateand Reynolds lead citrate (Poole, UK). For post-embedding immuno-gold labeling, we have used rapid freezing followed by freeze substitu-tion and finally embedding in a lowicryl HM20 resine using an AFSReichert Leica apparatus (Munich, Germany). Ultrathin sections werestained with the anti-Plic-1 antibody (1:50, 12 h), and the binding siteswere revealed using a goat anti-rabbit coupled to 10 nm gold41. Imageswere collected with a JEOL 100CXII electron microscope (Paris, France).For quantification of electromicrographs, with reference to the associa-tion of Plic-1 with membrane structures, immunoreactivity within 30nm was computed as an indication of membrane association. Thirtynanometers corresponds to the expected distance for an antigenic deter-minant at about 10 nm from the plasma membrane and being labeledwith two immunoglobulins and a pre-embedding method42,43.

ACKNOWLEDGEMENTSThis work was supported by the MRC, the Wellcome Trust and the Institut de

Recherche sur la Moelle Epinière.


1. Macdonald, R. L. & Olsen, R. W. GABAA receptor channels. Annu. Rev.Neurosci. 17, 569–602 (1994).

2. Rabow, L. E., Russek, S. J. & Farb, D. H. From ion currents to genomicanalysis: recent advances in GABAA receptor research. Synapse 21, 189–274(1995).

3. Davies, P. A., Hanna, M. C., Hales, T. G. & Kirkness, E. F. Insensitivity toanaesthetic agents conferred by a class of GABAA receptor subunit. Nature385, 820–823 (1997).

4. Bonnert, T. P. et al. Theta, a novel γ-aminobutyric acid type A receptorsubunit. Proc. Natl. Acad. Sci. USA 96, 9891–9896 (1999).

5. Unwin, N. Neurotransmitter action: opening of ligand-gated ion channels.Cell Suppl. 72, 31–41 (1993).

6. Essrich, C., Lorez, M., Benson, J. A., Fritschy, J. M. & Luscher, B. Postsynapticclustering of major GABAA receptor subtypes requires the γ2 subunit andgephyrin. Nat. Neurosci. 1, 563–571 (1998).

7. Kneussel, M. et al. Loss of postsynaptic GABAA receptor clustering ingephyrin-deficient mice. J. Neurosci. 19, 9289–9297 (1999).

8. Feng, G. et al. Dual requirement for gephyrin in glycine receptor clusteringand molybdoenzyme activity. Science 282, 1321–1324 (1998).

9. Kittler, J. T. et al. The subcellular distribution of GABARAP and its ability tointeract with NSF suggest a role for this protein in the intracellular transportof GABAA receptors. Mol. Cell Neurosci. 18, 13–25 (2001).

10. Wang, H. B., Bedford, F. K., Brandon, N. J., Moss, S. J. & Olsen, R. W.GABAA-receptor-associated protein links GABAA receptors and thecytoskeleton. Nature 397, 69–72 (1999).

11. Wan Q, et al. Recruitment of functional GABAA receptors to postsynapticdomains by insulin. Nature 388, 686–689 (1997).

12. Nusser, Z., Hajos, N., Somogyi, P. & Mody, I. Increased number of synapticGABAA receptors underlies potentiation at hippocampal inhibitory. Nature395, 172–177 (1998).

13. Nusser, Z., Cull-Candy, S. & Farrant, M. Differences in synaptic GABAAreceptor number underlie variation in GABA mini amplitude. Neuron 19,697–709 (1997).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

14. Moss, S. J. & Smart, T. G. Constructing inhibitory synapses. Nat. Rev.Neurosci. 2, 241–250 (2001).

15. Connolly, C. N. et al. Subcellular localization and endocytosis of homomericγ2 subunit splice variants of γ-aminobutyric acid type A receptors. Mol. CellNeurosci. 13, 259–271 (1999).

16. Connolly, C. N. et al. Cell surface stability of γ-aminobutyric acid type Areceptors. Dependence on protein kinase C activity and subunit composition.J. Biol. Chem. 274, 36565–36572 (1999).

17. Kittler, J. T. et al. Constitutive endocytosis of GABAA receptors by anassociation with the adaptin AP2 complex modulates inhibitory synapticcurrents in hippocampal neurons. J. Neurosci. 20, 7972–7977 (2000).

18. Kittler, J. T. et al. Analysis of GABAA receptor assembly in mammalian celllines and hippocampal neurons using γ2 subunit green fluorescent proteinchimeras. Mol. Cell Neurosci. 16, 440–452 (2000).

19. Marsh, M. & McMahon, H. T. The structural era of endocytosis. Science 285,215–220 (1999).

20. Wu, A. L., Wang, J., Zheleznyak, A. & Brown, E. J. Ubiquitin-related proteinsregulate interaction of vimentin intermediate filaments with the plasmamembrane. Mol. Cell 4, 619–625 (1999).

21. Kleijnen, M. F. et al. The hPLIC proteins may provide a link between theubiquitination machinery and the proteasome. Mol. Cell 6, 409–401 (2000).

22. Field., S. & Song, O. A novel genetic system to detect protein–proteininteractions. Nature 340, 245–246 (1989).

23. Dong, H. et al. GRIP: a synaptic PDZ domain-containing protein thatinteracts with AMPA receptors. Nature 386, 279–284 (1997).

24. Funakoshi, M., Geley, S., Hunt, T., Nishimoto, T. & Kobayashi, H.Identification of XDRP1; a Xenopus protein related to yeast Dsk2p binds tothe N-terminus of cyclin A and inhibits its degradation. EMBO J. 18,5009–5018 (1999).

25. Jentsch, S. & Pyrowolakis, G. Ubiquitin and its kin: how close are the familyties? Trends Cell Biol. 10, 335–341 (2000).

26. Hofmann, K. & Bucher, P. The UBA domain: a sequence motif present inmultiple enzyme classes of the ubiquitination pathway. Trends Biochem. Sci.21, 172–173 (1996).

27. Hanley, J. G., Koulen, P., Bedford, F., Gordon-Weeks, P. R. & Moss, S. J. Theprotein MAP-1B links GABAC receptors to the cytoskeleton at retinalsynapses. Nature 397, 66–90 (1999).

28. Smith, D. B. & Johnson, K. S. Single step purification of polypeptides expressedin E. Coli as fusions with glutathione S-transferase. Gene 67, 31–40 (1988).

29. Benke, D., Fritschy, J. M., Trzeciak, A., Bannwarth, W. & Mohler, H.Distribution, prevalence, and drug binding profile of gamma-aminobutyric

acid type A receptor subtypes differing in the beta-subunit variant. J. Biol.Chem. 269, 27100–27107 (1994).

30. Wooltorton, J. R., Moss, S. J. & Smart, T. G. Pharmacological andphysiological characterization of murine homomeric β3 GABAA receptors.Eur. J. Neurosci. 9, 2225–2235 (1997).

31. Connolly, C. N., McDonald, B. M., Krishek, B. J., Smart, T. G. & Moss, S. J.Assembly and cell surface expression of heteromeric and homomeric GABAAreceptors. J. Biol. Chem. 271, 89–97 (1996).

32. Todd, A. J., Watt, C., Spike, R. C. & Sieghart, W. Co-localization of GABA,glycine, and their receptors at synapses in the rat spinal cord. J. Neurosci. 16,974–982 (1996).

33. Gardiol, A., Racca, C. & Triller, A. Dendritic and postsynaptic proteinsynthetic machinery. J. Neurosci. 19, 168–179 (1999).

34. Noel, J. et al. Surface expression of AMPA receptors in hippocampal neuronsis regulated by a NSF-dependent mechanism. Neuron 23, 365–376 (1999).

35. Williams, E. J. et al. Selective inhibition of growth factor-stimulatedmitogenesis by a cell-permeable Grb2-binding peptide. J. Biol. Chem. 272,22349–22354 (1997).

36. Taylor, P. M. et al. Identification of amino acid residues within GABAAreceptor β subunits that mediate both homomeric and heteromeric receptorexpression. J. Neurosci. 19, 6360–6371 (1999).

37. Taylor, P. M. et al. Identification of residues within GABAA receptor alphasubunits that mediate specific assembly with receptor beta subunits. J. Neurosci. 20, 1297–1306 (2000).

38. Garner, C. C., Nash, J. & Huganir, R. L. PDZ domains in synapse assemblyand signaling. Trends Cell Biol. 10, 274–280 (2000).

39. Mah, A. L., Perry, G., Smith, M. A. & Monteiro, M. J. Identification ofubiquilin, a novel presenilin interactor that increases presenilin proteinaccumulation. J. Cell Biol. 151, 847–862 (2000).

40. Gorrie, G. H. et al. Assembly of GABAA receptors composed of α1 and β2subunits in both cultured neurones and fibroblasts. J. Neurosci. 17,6587–6588 (1997).

41. Colin, I., Rostaing, P., Augustin, A. & Triller, A. Localization of componentsof glycinergic synapses during rat spinal cord development. J. Comp. Neurol.398, 359–372 (1998).

42. Triller, A., Cluzeaud, F., Pfeiffer, F., Betz, H. & Korn, H. Distribution ofglycine receptors at central synapses: an immunoelectron microscopy study.Cell Biol. 101, 683–688 (1985).

43. Triller, A., Cluzeaud, F., Pfeiffer, F. & Korn, H. in Molecular Aspects ofNeurobiology (eds. Levi Montalcini, R. et al.) 101–105 (Springer, Berlin,Heidelberg, 1986).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


AMPA-type glutamate receptors mediate most of the fast excitato-ry synaptic transmission in mammalian brain1 and represent animportant target for postsynaptic mechanisms that control synap-tic strength. Accumulating evidence indicates that AMPA receptorsare not statically localized at synapses, but rather move in and out ofthe postsynaptic membrane on a rapid time scale. The regulatedchange in the postsynaptic trafficking of AMPA receptors seems tobe a major mechanism contributing to synaptic plasticity2–5.

Long-term depression (LTD) in hippocampus and cerebellumis associated with AMPA receptor internalization6–9. Moreover, avariety of factors that induce synaptic depression, such as NMDAreceptor activation or insulin, also promote rapid AMPA recep-tor endocytosis in cultured hippocampal neurons8,10–14. Recentstudies have provided the first quantitative measurements ofAMPA receptor internalization and offered initial insights intotheir cell biological mechanisms12–16. Long term potentiation(LTP) in the hippocampus, on the other hand, is correlated witha recruitment of AMPA receptors to the postsynaptic mem-brane6,17–20. Compared with AMPA receptor endocytosis, how-ever, little is known about the cell biological pathways or molecularmechanisms of AMPA receptor exocytosis (insertion into the plas-ma membrane). Nor is it clear whether the delivery of AMPAreceptors to synapses (as inferred from electrophysiological assays)results from surface insertion of AMPA receptors from intracel-lular pools or from lateral translocation of AMPA receptors fromnon-synaptic to synaptic sites on the plasma membrane.

The evidence for postsynaptic AMPA receptor exocytosis hasbeen largely indirect, based on pharmacological and electro-physiological approaches21. The regulated delivery of AMPAreceptors to the synapse has been inferred by experiments inwhich virally transfected AMPA receptor subunits were ‘tagged’by distinctive electrophysiological properties3,17,19,20. Synaptic

Subunit-specific temporal andspatial patterns of AMPA receptorexocytosis in hippocampal neurons

Maria Passafaro1, Valentin Piëch1 and Morgan Sheng1,2

1 Department of Neurobiology and Howard Hughes Medical Institute, Massachusetts General Hospital and Harvard Medical School, 50 Blossom Street, Boston, Massachusetts 02114, USA

2 Present address: Center for Learning and Memory, Howard Hughes Medical Institute, Massachusetts Institute of Technology (E18-215), 77 Massachusetts Avenue, Cambridge, Massachusetts 02139, USA

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

Using a thrombin cleavage assay in cultured hippocampal neurons, we studied the kinetics,regulation and site of AMPA receptor surface delivery. Surface insertion of the GluR1 subunit occursslowly in basal conditions and is stimulated by NMDA receptor activation and insulin, whereas GluR2exocytosis is constitutively rapid. Although both subunits ultimately concentrate in synapses, GluR1and GluR2 show different spatial patterns of surface accumulation, consistent with GluR1 beinginserted initially at extrasynaptic sites and GluR2 being inserted more directly at synapses. Thespatiotemporal pattern of surface accumulation is determined by the cytoplasmic tails of GluRsubunits, and in heteromeric receptors, GluR1 acts dominantly over GluR2. We propose that GluR1controls the exocytosis and GluR2/3, the recycling and endocytosis of AMPA receptors.

delivery of AMPA receptors also depends on stargazin, a proteinrequired for AMPA receptor expression on the cell surface22,23.However, these electrophysiological studies did not directly mea-sure the kinetics of surface insertion or synaptic accumulationof AMPA receptors. With much higher temporal resolution, themovement of GFP-tagged AMPA receptor subunit GluR1 to den-dritic spines was directly visualized18; however, the GFP imag-ing approach is not ideal for distinguishing surface versusintracellular or synaptic versus non-synaptic delivery.

In this study, we exploited a thrombin surface cleavage assayto visualize AMPA receptors newly inserted into the cell surface.This method, which has been used previously for G-protein-cou-pled receptors24,25 and glycine receptors (M. Rosenberg et al.,Soc. Neurosci. Abstr. 25, 180.7, 1999) allowed us to quantify thetime course of AMPA receptor exocytosis and to approach thequestion of where on the neuronal surface the AMPA receptorexocytosis occurs.

AMPA receptors are heteromeric complexes composed of up tofour distinct subunits, of which GluR1, GluR2 and GluR3 pre-dominate in mature hippocampal neurons26. GluR1 and GluR2/3differ in their cytoplasmic tail domains, the C-termini of whichinteract with distinct PDZ scaffold proteins: GluR1 with SAP97(ref. 27), and GluR2/3 with GRIP/ABP and PICK-1 (refs. 28–31).These PDZ-based interactions have been implicated in the synap-tic targeting and/or surface stability of AMPA recep-tors8,15–17,20,28,32–35. The relative roles of GluR1 and GluR2/3 inthe regulated exocytosis of AMPA receptors are poorly under-stood. Here we show that the surface delivery of GluR1 and GluR2is differentially controlled in terms of rate, regulation by NMDAreceptors, and spatial pattern of surface accumulation. Our resultsimply distinctive roles for GluR1 and GluR2 subunits in control-ling AMPA receptor surface expression and synaptic plasticity.

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 1. Specificity of thrombin cleavage of surface HA/T-GluR. (a) Thrombin treatment eliminates surface staining of HA/T-GluR1 in heterologous cells. HEK293 cells 48 h after transfectionwith HA/T-GluR1 were treated with thrombin or with DMEM(control), and processed for HA immunostaining as in the neu-ronal thrombin cleavage assay protocol. Surface HA staining(green in merge), but not intracellular HA staining (red inmerge), was eliminated by thrombin treatment. (b) Cells trans-fected with HA/T-GluR1 were treated with thrombin (+) or con-trol (–), then lysed and immunoblotted sequentially with HA andGluR1 C-terminus-directed antibodies. (c) Thrombin treatmentdoes not affect total levels of surface GluR1 and GluR2 on neu-rons. Hippocampal neurons transfected with HA/T-GluR1 orHA/T-GluR2 were treated with thrombin or DMEM-only con-trol, stained for total surface GluR1 and GluR2 using antibodiesdirected against the N-terminal extracellular regions of thesesubunits (top), and double-stained for intracellular HA (bottom).


RESULTSTime course of GluR1 and GluR2 exocytosisTo allow visualization of AMPA receptors newly inserted into theplasma membrane, we introduced into GluR1 and GluR2 an HAepitope tag near the N-terminus of the protein, followed imme-diately by a specific cleavage site for the extracellular proteasethrombin (HA/T-GluR1 and HA/T-GluR2). After transfectionand expression in HEK293 cells or cultured hippocampal neu-rons, the N-terminal HA tag allowed labeling of surface HA/T-GluR1 or -GluR2 by HA antibody applied in non-permeabilizingconditions (Figs. 1 and 2). Thrombin treatment of live trans-fected HEK cells eliminated surface HA staining of HA/T-GluR1without affecting intracellular HA immunoreactivity (Fig. 1a).On western blots, HA/T-GluR1-transfected HEK cells afterthrombin treatment showed ∼ 50% reduction in HA immunore-activity but no change in immunoreactivity detected with an anti-body directed against the C-terminus of GluR1 (Fig. 1b). Theseresults are consistent with the thrombin-mediated cleavage of ashort N-terminal peptide containing the HA epitope from thesurface population of HA/T-GluR1.

In transfected neurons, steady-state surface staining of HA/T-GluR1 and -GluR2 was abundant on dendrites and enriched ondendritic spines (see below). Treatment of live neurons withthrombin effectively cleaved off the N-terminal HA tag from sur-face HA/T-GluR1 or HA/T-GluR2, rendering these ‘clipped’ sub-units invisible to HA staining (Fig. 2a and b). Thrombintreatment had no effect on intracellular staining of HA/T-GluR1

or HA/T-GluR2 by HA antibodies (Fig. 1c, Fig. 2a and b).Following washout of thrombin and return of cultures to37°C, the surface insertion of AMPA receptors from intra-cellular compartments could be visualized by the reap-pearance of surface HA immunoreactivity and could bequantified over time (Fig. 2). Other than releasing the N-terminal tag, the thrombin treatment did not change thetotal amount of HA/T-GluR and endogenous AMPAreceptors in the plasma membrane (Fig. 1c). Using anti-bodies directed against the N-terminal extracellular regionof GluR1 and GluR2, the total surface fluorescence inten-sity for GluR1 was 154.4 ± 12.4 arbitrary units before, and146.9 ± 10.6 after, thrombin treatment, and for GluR2,162 ± 12.3 before and 156 ± 9.8 after thrombin. There-fore, the rate of reappearance of surface HA immunore-activity should reflect the steady-state rate of insertion ofAMPA receptors into the plasmalemma.

We first compared the time course of exocytosis of GluR1versus GluR2. HA/T-GluR2 reappeared on the cell surface

rapidly after washout of thrombin; intensity of surface HAimmunofluorescence recovered to 15 ± 0.3% of steady-state level at5 minutes, 28 ± 4.6% at 10 minutes, 38 ± 2.7% at 30 minutes, and63 ± 4.5% at 60 minutes (n = 7 for each time point, Fig. 2b andc). HA/T-GluR1 showed a much slower time course of surfaceinsertion. After thrombin treatment, surface HA staining recov-ered to only 3 ± 0.4% at 5 minutes, 5.6 ± 2.5% at 10 minutes, 12 ±2.1% at 30 minutes and 22 ± 3.8% at 60 minutes (n = 7, Fig. 2aand c). Based on a single exponential best fit to the data, the timeconstant (τ) of surface insertion was ∼ 27 minutes for HA/T-GluR1and ∼ 10 minutes for HA/T-GluR2. At steady state, the intracellu-lar and surface HA staining intensity for HA/T-GluR1 and -GluR2were not significantly different (Fig. 2a, b and d). Thus, the dis-tinct kinetics of surface insertion of HA/T-GluR1 and -GluR2 areunlikely to be due to different levels of expression of these con-structs. We observed no obvious differences between cell body anddendrites with respect to the kinetics of surface appearance ofHA/T-GluR1 or -GluR2.

Recycling of thrombin-cleaved (‘HA-invisible’) receptors backto the surface, and internalization of HA/T-GluR after surfaceinsertion, would both tend to reduce the rate of surface accu-mulation of intact HA/T-GluR and contribute to causing the sur-face HA signal to plateau. However, the early time points (up to∼ 10 min) show an approximately linear increase in HA staining(Fig. 2c), and should have been relatively unaffected by recycling.Using a combined antibody-feeding and acid-strip assay, we mea-sured the resurfacing of HA/T-GluR subunits following a 10-

articles

a

b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 2. Time course of surface delivery of HA/T-GluR1 and -GluR2 in cultured hippocampal neurons. (a, b) Surface andintracellular HA-immunostaining of HA/T-GluR1 (a) and HA/T-GluR2 (b) are shown at steady-state (before thrombin treat-ment) and at 0, 10 and 30 min after recovery from thrombintreatment of live neurons (see Methods). (c) Time course ofHA/T-GluR1 and -GluR2 surface accumulation. Points representmean ± s.e.m. of quantitative surface immunofluorescence inten-sity measurements for at least five neurons per condition. Theplotted lines represent the best fit to a single exponential (deter-mined by the least-squares method). (d) Steady-state levels ofintracellular and surface HA/T-GluR1 and -GluR2 in transfectedneurons, as measured by quantitative HA immunofluorescence.Histograms show mean ± s.e.m. Scale bar, 20 µm.

minute internalization period in basal conditions (Fig. 3). Less than 10% of surface HA/T-GluR1 or -GluR2was endocytosed during the internalization period inthese conditions. The fraction of internalized HA/T-GluR1 that resurfaced in the ensuing period was unde-tectable at 10 minutes and was 1.7 ± 1.0% at 30 minutes.The fraction of internalized HA/T-GluR2 that resurfacedwas 5.7 ± 2.2% at 10 minutes and 8.5 ± 2.6% at 30 min-utes. Thus, the rate of surface appearance of HA/T-GluRmeasured during the first 10 minutes of the thrombincleavage assay is not significantly affected by recycling.The more rapid recycling of GluR2 relative to GluR1 isconsistent with the faster exocytosis of GluR2 measuredin the thrombin cleavage assay.

To investigate whether the cytoplasmic C-terminal tailof GluR1 and GluR2 might be responsible for the differentkinetics of surface insertion, we constructed chimeras HA/T-GluR1/R2C and HA/T-GluR2/R1C, in which the cytoplas-mic tails of these subunits are ‘swapped.’ In the thrombincleavage assay, HA/T-GluR2/R1C showed a time course ofexocytosis similar to that of wild-type GluR1 (Fig. 4a andc). On the other hand, the surface delivery of the chimera HA/T-GluR1/R2C showed a time course similar to that of GluR2. We con-clude that the differential rate of exocytosis of GluR1 (slow) and

GluR2 (fast) is determined by the cytoplasmic tails of these sub-units.

We cotransfected neurons with N-terminally myc-taggedGluR1 (myc-GluR1) and HA/T-GluR2 to test the effect of het-eromultimerization on receptor exocytosis. In cells coexpressingmyc-GluR1, the time course of exocytosis of HA/T-GluR2 resem-bled that of HA/T-GluR1 expressed alone (Fig. 4b and c). Asexpected, thrombin had no effect on surface myc-GluR1 staining,confirming the specificity of thrombin cleavage (Fig. 4b). Assum-ing that the coexpressed subunits assemble into heteromers, theseresults imply that GluR1 acts ‘dominantly’ over GluR2 to restrainGluR2 exocytosis in GluR1/GluR2 heteromeric receptors.

articles


Fig. 3. Time course of HA/T-GluR1 and HA/T-GluR2 recycling in hip-pocampal neurons. (a, b) Live neurons transfected with HA/T-GluR1(a) or HA/T-GluR2 (b) were surface-labeled with mouse anti-HA anti-bodies and internalization was allowed to proceed for a further 10 minat 37°C in the CO2 incubator. After this internalization period,remaining surface antibodies were ‘acid-stripped’ (see Methods), andcells returned to 37°C CO2 incubator to allow resurfacing of theinternalized receptor/primary HA antibody complex. Immediatelyafter acid wash (Acid Wash), and at 10 and 30 min of recovery afteracid wash, neurons were fixed and stained with anti-mouse FITC sec-ondary antibody in non-permeabilizing conditions to label resurfacedreceptors (top), and then double-stained with Cy3-conjugated sec-ondary antibody in permeabilizing conditions to label remaining inter-nalized receptors (bottom). The left column shows surface andinternalized staining before acid wash.

a

b

c

a

b

d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


exocytosis, accelerating the reappearance of surface HA/T-GluR1at 10 minutes (Fig. 5c). In contrast, the exocytosis of HA/T-GluR2was unaltered by insulin. Wortmannin (200 nM) also blocked theeffect of insulin on GluR1 exocytosis, but by itself had no effecton basal GluR1 and GluR2 exocytosis (Fig. 5c). Thus, GluR1 andGluR2 are differentially regulated; GluR1 surface delivery is accel-erated by NMDA receptor activation and by insulin, whereasGluR2 exocytosis shows no response to these stimuli. After treat-ment with NMDA, glycine or insulin, there was also a small butsignificant reduction in intracellular HA immunofluorescence,consistent with the export of a fraction of HA/T-GluR from intra-cellular compartments to the surface (data not shown).

Brefeldin A (BFA) inactivates various Arf GTPases and dis-rupts ER-Golgi transport and other membrane trafficking stepsin eukaryotic cells37,38. BFA (10 or 20 µg/ml, applied for 10 minafter thrombin cleavage) inhibited the basal insertion of GluR2and GluR1 (Fig. 5c), and the glycine-stimulated insertion ofGluR1 (data not shown). These data indicate that a BFA-sensi-tive step is involved in the rapid process of surface insertion ofboth GluR1 and GluR2 in hippocampal neurons.

Sequence determinants of GluR exocytosisBecause the cytoplasmic C-terminal tails of GluR1 and GluR2determine the kinetics of surface insertion, we focused on specificsequences within these regions. Deletion of the last 4 residues ofGluR1 (HA/T-GluR1∆4), which should prevent interactionbetween GluR1 and SAP97 (ref. 27), resulted in a drastic loss of

surface expression at steady-state (surface HA stain-ing intensity 8 ± 0.2 arbitrary units compared with115 ± 6.6 for wild-type HA/T-GluR1; Fig. 6a and c).The intracellular levels of GluR1∆4 were only slightlylower than wild type (Fig. 6c). These data suggest thatinteractions mediated by the last four amino acids ofGluR1 are critical for surface delivery or surface sta-bility of GluR1. Unfortunately, the surface expressionof HA/T-GluR1∆4 was too low to measure the rate ofinsertion of this construct following thrombin cleav-age. Coexpression of HA/T-GluR1∆4 with myc-GluR2in neurons reduced the surface level of myc-GluR2(58 ± 10.4 arbitrary units) compared to cotransfec-tion with wild-type HA/T-GluR1 (99 ± 10.4).

We also introduced a more subtle mutation in theC-terminus of GluR1, changing the third-from-lastresidue (threonine-887) to alanine (-ATGL to -AAGL). At steady-state, the surface level of the HA/T-GluR1(T887A) mutant was moderately reduced (Fig.6b and c). In the thrombin cleavage assay, the rate ofsurface accumulation of HA/T-GluR1(T887A) wasimpaired (Fig. 6b and d). The τ of recovery of sur-face staining was slowed to 42 minutes forGluR1(T887A) compared with 27 minutes for wild-type GluR1. The prolongation of τ suggests that thereduced steady-state surface expression of T887A isdue largely to impaired exocytosis of this mutant sub-

Differential regulation of GluR1 and GluR2 exocytosisElectrophysiological studies suggest that activity can stimulatethe delivery of GluR1 and GluR4 to synapses17,19. We used thethrombin cleavage assay to test directly whether activity promotesGluR1 or GluR2 exocytosis, measuring the recovery of surfaceHA immunoreactivity at an early time point (10 min) followingthrombin treatment (Fig. 5). Drugs were added 1 hour beforethrombin cleavage and maintained for the 10 minutes afterthrombin washout. Activating NMDA receptors with 300 nMglycine36 or 20 µM NMDA caused an increase in the surfaceinsertion of HA/T-GluR1 (Fig. 5a and c). In contrast, glycine orNMDA had no effect on GluR2 surface insertion (Fig. 5b and c).The glycine stimulation of HA/T-GluR1 exocytosis was blockedby the NMDA receptor antagonist APV (100 µM), and by wort-mannin (200 nM), an inhibitor of phosphatidylinositol-3-kinase(PI3K; Fig. 5c). The time course of surface reappearance of HA/T-GluR1 and HA/T-GluR2 in basal conditions was unaffected by100 µM APV. The recovery of surface HA/T-GluR1 and HA/T-GluR2 after thrombin treatment was, respectively, 4.1 ± 2.0%and 26 ± 7.2% of steady-state at 10 minutes, and 13.0 ± 1.8%and 36 ± 4.9% at 30 minutes. Even in the constant presence ofAPV (100 µM) starting immediately after transfection, there wasno significant difference in the steady-state surface levels of HA/T-GluR1 measured at 3 days after transfection (107 ± 11.8 arbitraryunits compared to 115 ± 7.1 in the absence of APV).

Insulin stimulates AMPA receptor endocytosis in hippocampalneurons8,12,13. Insulin (300 nM) also enhanced AMPA receptor


articles

Fig. 4. Time course of exocytosis of GluR chimeras and heteromeric GluR1/GluR2 receptors. (a) Surface HAstaining of HA/T-GluR2/R1C (top) and -GluR1/R2C (bottom) at steady state, and at 0, 10 and 30 min follow-ing thrombin treatment. Intracellular HA staining of HA/T-GluR2/R1C and -GluR1/R2C did not change follow-ing thrombin treatment (data not shown). (b) Surface HA-staining of HA/T-GluR2 in neurons cotransfectedwith myc-GluR1, at steady state, and at 0, 10 and 30 min following thrombin (top). Bottom, coexpression ofmyc-GluR1 in the same neurons by surface staining with myc antibodies. (Myc immunoreactivity is unaffectedby thrombin.) Scale bar, 20 µm. (c) Time course of surface accumulation of chimeras HA/T-GluR2/R1C andHA/T-GluR1/R2C, and of HA/T-GluR2 in cells cotransfected with myc-GluR1.

a

b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 6. Effect of C-terminal mutations on the surface level and exo-cytosis of GluR1. (a) Steady-state intracellular and surface HA stain-ing of neurons expressing HA/T-GluR1∆4. (b) Surface staining ofHA/T-GluR1(T887A) at steady state, 10 min after thrombin treat-ment in the absence of drugs (control), and 10 min after thrombintreatment in the presence of glycine (300 nM). (c) Quantitation ofthe steady-state intracellular and surface levels of HA/T-GluR1∆4and -GluR1(T887A). (d) Time course of surface accumulation ofHA/T-GluR1(T887A) compared to wild-type HA/T-GluR1. Plottedlines represent the best fit to a single exponential curve determinedby the least-squares method. Scale bar, 15 µm.

unit. (A primary reduction in surface stability of GluR1(T887A)should decrease the value of τ.) Our data, however, do not excludesome involvement of the C-terminus in the stabilization of GluR1in the plasmalemma. In addition to a reduced exocytosis rate inbasal conditions, the surface insertion of HA/T-GluR1(T887A)was also not stimulated by glycine (Fig. 6b).

Deletion of the last 4 amino acids of GluR2, which shouldabolish interaction with PDZ proteins GRIP/ABP and PICK-1,had only a modest negative effect on steady-state surface level ofHA/T-GluR2∆4 (Fig. 7a and b). The intracellular levels ofHA/T-GluR2∆4 and wild type were not significantly different.In the thrombin cleavage assay, the time course of surface accu-mulation of HA/T-GluR2∆4 (τ = 18 min) was slower than wildtype (Fig. 7c). Truncation of the last 15 amino acids of GluR2(HA/T-GluR2∆15) reduced the steady-state surface expressionfurther (Fig. 7a and b), though the effect was still not as dra-matic as with the 4-residue C-terminal truncation of GluR1(Fig. 6). Correlated with the reduced steady-state surface expres-sion, the time course of surface accumulation of GluR2∆15 wasslower than either wild-type GluR2 or GluR2∆4 (τ = 29 min;Fig. 7a and c). The slower time constants of recovery of surfacestaining imply a reduced rate of exocytosis of HA/T-GluR2∆4and HA/T-GluR2∆15; however, a reduced stability on the cellsurface of these mutants cannot be excluded.

Site of surface accumulation of GluR1 and GluR2The thrombin cleavage assay allowed us to approach an outstand-ing question in AMPA receptor trafficking. At which subcellularsites are AMPA receptors initially inserted on the neuronal sur-face? Using triple-label immunofluorescence, we first examined

the steady-state distribution of surface HA/T-GluR1 and HA/T-GluR2 in hippocampal neurons (Fig. 8). Cell-surface HA/T-GluR1 and -GluR2 were concentrated like the endogenousGluR proteins in numerous clusters that colocalized with thesynaptic marker Shank, particularly in dendritic spines (Fig. 8). The intensity of staining at ‘synaptic’ sites (defined bycolocalization with Shank) was 67 ± 10.6 arbitrary units (aver-age intensity per area) for HA/T-GluR1 and 69 ± 11.1 forHA/T-GluR2 (Fig. 8a and d). In addition, however, there wasdiffuse staining of these AMPA receptor subunits on the neu-ron surface that did not overlap with Shank, which we definedas ‘non-synaptic’ (33 ± 9.6 for HA/T-GluR1 and 31 ± 7.7 forHA/T-GluR2, Fig. 8a and d). Endogenous GluR1 and GluR2also showed non-synaptic surface staining in addition to synap-tic clusters (Fig. 8b). The degree of synaptic enrichment of

articles


Fig. 5. Regulation of exocytosis of HA/T-GluR1 and -GluR2 by NMDAreceptor activation and insulin. (a, b) Cultured hippocampal neuronstransfected with HA/T-GluR1 (a) or HA/T-GluR2 (b) were treated withglycine (300 nM), NMDA (20 µM), wortmannin (Wtn, 200 nM), insulin(300 nM), Brefeldin A (BFA; 10 or 20 µg/ml) and APV (100 µM), individ-ually or in combination as indicated, during the thrombin surface cleav-age assay (see text for details). For each condition, the surface HAstaining at 10 min of thrombin recovery is shown. Control panels showthe surface staining of HA/T-GluR1 and -GluR2 at 10 min followingthrombin in the absence of drugs. (c) Quantitation of results in (a) and(b). Histograms show mean ± s.e.m. (n = 6). Scale bar, 20 µm.

a

b

c

a

b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 7. Effect of C-terminal mutations on surfacelevels and exocytosis of GluR2. (a) Surface HA-immunostaining of HA/T-GluR2∆4 and -GluR2∆15at steady state, and at 10 and 30 min after recoveryfrom thrombin treatment. Surface staining at 0 minafter thrombin was undetectable (data not shown).Scale bar, 20 µm. (b) Quantitation of steady-stateintracellular and surface levels of HA/T-GluR2, -GluR2∆4 and -GluR2∆15 (mean ± s.e.m.). (c) Timecourse of HA/T-GluR2, -GluR2∆4 and -GluR2∆15surface accumulation.


HA/T-GluR1 and HA/T-GluR2 was less pronounced than that ofendogenous GluR1 and GluR2 (Fig. 8d). The more diffuse patternof surface distribution of HA/T-GluR1 and -GluR2 could be due tooverexpression of these transfected subunits and/or their unnat-ural homomeric configuration. Surface HA/T-GluR2∆4 also con-centrated in puncta that colocalized to a considerable extent withShank, indicating some synaptic localization of this C-terminalmutant (Fig. 8c). However, the degree of synaptic enrichment ofGluR2∆4 was less than that of wild type (Fig. 8d).

To probe where GluR1 and GluR2 are first inserted into theplasma membrane, we examined the surface distribution of newlyinserted HA/T-GluR1 and HA/T-GluR2 at early time points afterrecovery from thrombin cleavage and compared it with the synap-tic marker Shank. As shown above, thrombin treatment resultedin a complete loss of surface HA/T-GluR1 and -GluR2 staining(Fig. 9, 0 min). No loss of Shank clusters was observed in the samecells. Three minutes after recovery from thrombin, weak stainingfor HA/T-GluR1 was observed on the neuronal surface in amicropunctate non-synaptic pattern that showed no detectableoverlap with Shank (mean staining intensity per area, 0.2 ± 0.1arbitrary units at synaptic sites, defined by overlap with Shank;8.4 ± 2.6 at non-synaptic sites, defined by non-overlap with Shank;Fig. 9a and b). At 5 minutes, total surface intensity increased andHA/T-GluR1 began to accumulate in synaptic regions (15 ± 3.7units at synaptic sites, 9 ± 3.6 in extrasynaptic areas; Fig. 9a and b).Over the next 10 minutes, the intensity of surface HA/T-GluR1signal increased roughly in parallel in synaptic and extrasynapticareas. Because non-synaptic surface area is much greater thansynaptic area, these results indicate that most HA/T-GluR1 accu-mulates during the first few minutes after insertion in non-synap-tic areas of the neuronal surface. After stimulation of HA/T-GluR1

exocytosis by NMDA, the newly inserted HA/T-GluR1 also accumulated mainly in non-synap-tic regions of the neuronal surface during thefirst 5 minutes (Fig. 9d).

HA/T-GluR2 behaved quite differently thanGluR1 with respect to the spatial pattern of sur-face accumulation (Fig. 9a and b). As early as 3minutes after recovery from thrombin, thenewly inserted GluR2 exhibited a sharply punc-tate pattern that colocalized with Shank (38 ±6.0 units were ‘synaptic;’ 6.1 ± 3.9 were ‘non-synaptic’). At 5 minutes, the surface intensityincreased to 50 ± 3.7 units for synaptic areas and7 ± 3.8 for extra-synaptic areas. Thus, the meanstaining intensity of HA/T-GluR2 at 3–5 min-utes was ∼ 7-fold higher in synaptic than in non-synaptic areas, and the synaptic intensity at 5minutes had already reached ∼ 70% of the valueat steady state, before thrombin treatment. From5 to 15 minutes, the level of surface HA/T-GluR2 at synapses reached a plateau, whereas

extra-synaptic staining of HA/T-GluR2 continued to rise morerapidly (tripling to 21 ± 4.9 arbitrary units at 15 min; Fig. 9a andb). Thus, HA/T-GluR1 and -GluR2 showed distinct spatiotem-poral patterns of surface accumulation. GluR2 accumulated moreimmediately and more selectively in synapses; the synaptic/non-synaptic intensity ratio at 3 minutes and 5 minutes (∼ 7) was high-er than at steady-state (2.2). GluR1 appeared first in non-synapticregions before accumulating at synaptic sites; the synaptic/non-synaptic ratio at 3 minutes (<0.1) and at 5 minutes (1.7) was lowerthan at steady-state (2.0).

Upon coexpression with myc-GluR1, however, HA/T-GluR2appeared on the surface in a pattern similar to HA/T-GluR1. Inthe first few minutes after thrombin treatment, newly emergedHA/T-GluR2 in presumptive heteromeric configuration withmyc-GluR1 showed a relatively diffuse non-synaptic distributionthat contrasted with the punctate synaptic pattern seen whenHA/T-GluR2 was expressed alone (Fig. 9, compare a and c). At3 minutes, staining intensity was 2.1 ± 0.4 arbitrary units insynaptic areas, and 10 ± 2.9 in extrasynaptic areas. The intensityof synaptic staining of heteromeric HA/T-GluR2 increased withtime; at 5 minutes after thrombin treatment, HA/T-GluR2 stain-ing intensity was 18 ± 3.7 (synaptic) and 9.0 ± 1.9 (non-synaptic),and at 10 minutes, staining intensity was 30 ± 4.9 (synaptic), and18 ± 4.9 (non-synaptic).

The chimera HA/T-GluR2/R1C also behaved similarly toGluR1 with respect to the spatial pattern of surface accumula-tion during the first 15 minutes following thrombin cleavage(data not shown). Surface intensity of HA/T-GluR2/R1C was 9± 3.1 (synaptic) and 6.5 ± 3.5 (non-synaptic) at 5 minutes, andincreased to 21 ± 4 (synaptic) and 14 ± 2.8 (non-synaptic) at 15minutes. Thus determination by the cytoplasmic tails of GluR

articles

a

b c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


subunits, and the dominance of GluR1 over GluR2 in heteromericreceptors, apply to both the temporal and the spatial pattern ofsurface delivery of AMPA receptors.

DISCUSSIONSubunit-specific regulation of exocytosisUsing the thrombin surface cleavage assay, we were able to direct-ly measure the rate of surface insertion of specific subunits of theAMPA receptor in cultured neurons. Although we found no evi-dence for an effect of thrombin on surface AMPA receptor levels,endocytosis of native receptors (M.P., unpublished observations),or on exocytosis of HA/T-GluR via stimulation of NMDA recep-tors, we cannot exclude that thrombin might have some unknownspecific action on AMPA receptor trafficking. Nevertheless, giventhat thrombin treatment was identical for all assays, a conclusion ofthis study is that GluR1 and GluR2 behave differently with respectto the rate and regulation of surface delivery. GluR2 exocytosis israpid and ‘constitutive’ (no stimulation by NMDA, glycine orinsulin), whereas GluR1 exocytosis is slow but ‘inducible’ (a trick-le in basal conditions that is accelerated by NMDA receptor andinsulin stimulation). These cell biological conclusions correlate inessence with the subunit-specific rules for synaptic delivery ofAMPA receptors inferred from electrophysiological tagging stud-ies17,20. Activation of NMDA receptors (particularly of synapticNMDA receptors by glycine) also triggers the surface insertion ofendogenous AMPA receptors, correlating with synaptic potentia-tion36. Our finding that NMDA receptor stimulation boosts theexocytosis of GluR1 (but not GluR2) suggests that GluR1 is thesubunit providing the ‘driving force’ for activity-dependent synap-tic delivery of AMPA receptors. This idea is consistent with mousegenetic data indicating that GluR1 but not GluR2 is important forLTP39,40. Therefore, the molecular control of GluR1 exocytosis mayhold the key to postsynaptic mechanisms underlying LTP.

PDZ proteins are widely implicated in targeting, anchoringand/or trafficking of their membrane protein binding partners41.

A C-terminal point mutation (T887A) in GluR1 that blocks bind-ing of the PDZ protein SAP97 slows the basal rate and abolishesthe NMDA receptor-dependent stimulation of GluR1 surface inser-tion. The loss of NMDA receptor-stimulated exocytosis couldexplain why the GluR1(T887A) mutant fails to be functionallyrecruited to synapses by activity17. A difference in our study, how-ever, is that NMDA receptor activity is not essential for the basalexocytosis and the steady-state surface expression of wild-typeGluR1 in cultured neurons. This is at apparent odds with previ-ous results that showed, in slice infection experiments, that GFP-GluR1 failed to localize in spines or to incorporate functionallyinto synapses in basal conditions17,20. Possible explanations includedifferences in the experimental preparations, the mode of trans-fection and the duration of expression of the exogenous GluRs,and the size of the N-terminal tag incorporated into GluR1.

The site of AMPA receptor exocytosisBecause thrombin cleavage effectively removes surface stainingof HA/T-tagged AMPA receptors, it allowed us to visualize thesites on the cell surface where newly inserted GluR subunits firstaccumulate. At the earliest time points we examined (3 to 5 minafter thrombin), considerable time had elapsed for receptor redis-tribution on the neuronal surface. Nevertheless, we observed astriking difference in the spatiotemporal patterns of GluR1 ver-sus GluR2 surface accumulation, raising the possibility that thesesubunits are delivered to different sites on the cell surface. Theinitial appearance of GluR1 in non-synaptic regions precedingits accumulation in synaptic areas is consistent with the surfacedelivery of GluR1 to non-synaptic loci followed by its lateraltranslocation into synapses. In contrast, surface GluR2 accumu-lated relatively immediately at synaptic sites (as defined by lightmicroscopic colocalization with Shank), following which synap-tic levels essentially reached a plateau, but non-synaptic levelscontinued to rise. This spatiotemporal pattern might be explainedby an initial insertion of GluR2 in (or in the vicinity of) synaps-

articles


Fig. 8. Synaptic concentration of HA/T-GluR1 and HA/T-GluR2 in hippocampal neurons. (a) Immunostaining of neurons for HA/T-GluR1 (left) orHA/T-GluR2 (right) 4–6 days after transfection. Cell-surface HA/T-GluR was labeled on live cells at 4ºC with mouse anti-HA antibody (green). Aftersurface labeling, the neurons were fixed, permeabilized and stained for intracellular HA/T-GluR with rabbit anti-HA antibody (blue, data not shown),whereas synaptic sites were labeled with guinea pig antibodies against the PSD protein Shank (red). Staining for intracellular AMPA receptors wasobserved in both a diffuse and punctate pattern throughout the dendrites (data not shown). Top, merge of the green, red and blue channels; individ-ual channels are shown in gray-scale. (b) Immunostaining of untransfected neurons for endogenous surface GluR1 or GluR2 (green), and Shank (red);bottom, merge. (c) Immunostaining of transfected neurons for surface HA/T-GluR2∆4 (green), and Shank (red). (d) Quantitation of mean immuno-fluorescence intensity in synaptic versus non-synaptic regions for HA/T-GluR1, HA/T-GluR2 and HA/T-GluR2∆4 compared with endogenous GluR1and GluR2. Synaptic area is defined by colocalization with Shank. Histograms show mean ± s.e.m. Scale bar, 5 µm.

a

b

c

d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 9. Differential spatiotemporal pat-tern of surface accumulation of HA/T-GluR1 and -GluR2. (a) Distribution ofHA/T-GluR1 (top) and HA/T-GluR2(bottom) at 0, 3, 5, 10 and 15 min fol-lowing thrombin treatment. SurfaceHA-staining, green; intracellular HA-staining, blue; Shank synaptic marker,red. The merge is shown in color; theindividual channels, in grayscale. In the3-min and 5-min color-merged imagefor (a), the green channel (surfaceHA/T-GluR1 staining) has beenenhanced to show more clearly theextrasynaptic location of newlyinserted HA/T-GluR1. The grayscaleimage for surface HA/T-GluR1 at the 3min and 5 min points was notenhanced. (b) Time course of surfacestaining intensity of HA/T-GluR1 andHA/T-GluR2 in synaptic regions(defined by colocalization with Shank)and non-synaptic regions (defined bylack of colocalization with Shank), dur-ing recovery from thrombin treatment.(c) Distribution of surface HA/T-GluR2(green) in cells cotransfected with myc-GluR1 (blue) at 0, 3 and 5 min followingthrombin treatment. Red, Shank synap-tic marker. (d) Distribution of surfaceHA/T-GluR1 at 0 and 5 min afterthrombin treatment in cells stimulatedwith NMDA (20 µM). Scale bar, 5 µm.


es, followed by lateral translocation to extra-synaptic locations.Alternatively, GluR2 is also exocytosed at non-synaptic sites buttranslocates to synapses more rapidly than GluR1. Our experi-ments look at patterns of surface accumulation rather than direct-ly at exocytosis. Technical advances that offer real-time imagingof AMPA receptor insertion and movement will be necessary todistinguish the above possibilities.

Our data indicate that an initial extrasynaptic accumulationof receptors also applies for GluR1/GluR2 heteromers and forNMDA receptor-stimulated GluR1 exocytosis. The idea thatAMPA receptors are first exocytosed at non-synaptic regions ofthe plasmalemma is in keeping with a study of stargazin-deficientneurons that implied a two-step mechanism for synaptic deliveryof AMPA receptors22,23. The lateral translocation of AMPA recep-tors from non-synaptic to synaptic regions may be an importantsite for regulation of synaptic plasticity. Indeed, an additional con-trol mechanism at this step could explain the apparent discrep-ancy between our data showing only modest effects of C-terminalmutations on the surface delivery of GluR1(T887A) and GluR2∆4,and the previous findings showing that such mutants fail to incor-porate functionally into synapses17,20.

A model for postsynaptic AMPA receptor traffickingIt is well documented that insulin stimulates AMPA receptorendocytosis in hippocampal neurons8,12,13,42. Here we show thatinsulin stimulates the exocytosis of AMPA receptors, specificallythe GluR1 but not the GluR2 subunit. This is exactly opposite to

insulin-stimulated endocytosis, inwhich the GluR2 but not GluR1subunit is responsive to insulin8,12.Based on the differential cell bio-logical behaviors of GluR1 and

GluR2 subunits, we propose the following model in which GluR1determines the rate and site of exocytosis of AMPA receptors,whereas GluR2 controls recycling and endocytosis (Supplemen-tary Fig. 1, available on the Nature Neuroscience web site). Insteady-state basal conditions, surface AMPA receptors(GluR1/GluR2 or GluR2/GluR3 heteromers) are concentrated inthe postsynaptic membrane. GluR2/GluR3 receptors in an intra-cellular compartment undergo constant recycling with surfacereceptors, thereby accounting for the high constitutive rate ofexocytosis of GluR2 in our experiments. This exchange occurswith surface AMPA receptors at the synapse, thereby insertingGluR2/GluR3 heteromers ‘directly’ into the postsynaptic mem-brane (or as close as would afford colocalization with Shank atthe light microscopic level). Meanwhile, intracellularGluR1/GluR2 heteromeric receptors lie dormant, as GluR1 actsdominantly to restrain constitutive exocytosis. Upon NMDAreceptor activation, however, GluR1 exocytosis is triggered, lead-ing to surface accumulation of GluR1/GluR2 receptors. The exo-cytosis of GluR1/GluR2 receptors requires the interaction ofGluR1 C-terminal tail with PDZ proteins (such as SAP97) andperhaps other cytoplasmic proteins. We propose that the activi-ty-stimulated insertion of GluR1/GluR2 occurs at extra-synap-tic sites and is followed by translocation into synapses; this secondstep may also be regulated by activity.

In synaptic depression, the prevailing evidence suggests thatGluR2- or GluR3-mediated endocytotic mechanisms are activat-ed, perhaps involving phosphorylation of the GluR2/3 C-terminus

articles

a

b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


and consequent release from GRIP/ABP anchors15,16,35. We sug-gest that the released GluR1/GluR2 and GluR2/GluR3 receptors‘drift’ outside of the synapse before being captured by the endo-cytotic machinery. First, this would be consistent with the kinet-ics of surface HA/T-GluR2 appearance at non-synaptic sites, whichlags behind its accumulation in synapses. Second, this would facil-itate the endocytosis of AMPA receptors without co-internalizationof NMDA receptors and other synaptic membrane proteins10,12,14.

METHODSDNA constructions. HA epitope tag (YPYDVPDYA) and thrombin cleav-age site sequence (LVPRGS) were inserted three amino acids from the C-terminal of the signal peptide of GluR subunits. HA/T cassette wasgenerated by PCR using two overlapping oligos. The 3´ oligo containedthe HA epitope and thrombin cleavage site. The 5´ oligo contained the N-terminus of the respective GluR subunits. This cassette was used togeth-er with a 3´ oligo containing the C-terminus of respective GluR subunitsto amplify by PCR the entire GluR1or GluR2 coding region. The chimericGluR1/R2C and GluR2/R1C were made by fusing the cytoplasmic tail ofGluR2 (820–862) to amino acid 812 of GluR1 and the cytoplasmic tail ofGluR1 (812–889) to amino acid 820 of GluR2. Point mutant T887A wasmade using the QuikChange system (Stratagene, La Jolla, California). Thedeletion constructs were made by PCR amplification using appropriateoligos and HA/T-GluR cDNA as template. Altered regions of mutant con-structs were checked by sequencing. The HA/T-GluR constructs wereexpressed using mammalian expression vector GW1 (CMV promoter) orpβactin-16-pl (β-actin promoter).

Cell cultures and transfection. Primary hippocampal cultures were pre-pared from embryonic day 18–19 rat brains as described previously43

and grown in Neurobasal medium supplemented with B27 (Gibco BRL,Rockville, Maryland). Neurons were plated on coverslips coated withpoly-D-lysine (30 µg/ml) and laminin (2 µg/ml) at a density of75,000/well. At 14 days in vitro (d.i.v.), cultures were transfected usingthe calcium phosphate method. DNA (5–8 µg) was mixed with 250 mMCaCl2 and added to the same volume of 2× HEPES-buffered saline (274mM NaCl, 10 mM KCl, 1.4 mM Na2HPO4, 15 mM D-glucose, 42 mMHEPES, pH 7.05). The DNA mix was incubated for 15 min in the darkand then added to the neurons at 37ºC in 5% CO2 until a fine precipi-tate formed. Transfected neurons were washed with DMEM and culturedin the original medium at 37°C at 5% CO2 for 4 to 6 days. HEK 293 cellswere transfected using Lipofectamine (Gibco-BRL).

Thrombin surface cleavage assay. Live HEK cells or transfected neurons(18–20 d.i.v.) were treated for 5 min at room temperature with thrombin(Pharmacia, Piscataway, New Jersey; 1 unit/ml in DMEM). After thor-ough washing with DMEM, neurons were returned to 37°C for varioustimes to allow for surface insertion of new receptors. Neurons werelabeled with mouse anti-HA antibody (1 µg/ml; Boehringer Mannheim,Indianapolis) for 1 h at 4°C to visualize the surface HA/T-GluR recep-tors, then washed extensively with cold DMEM and fixed for 8 min in4% paraformaldehyde/4% sucrose. Neurons were then incubated withrabbit anti-HA (Santa Cruz Biotechnology, Santa Cruz, California; 1µg/ml) and Shank guinea pig antibody 1123 (1 µg/ml, donated by E.Kim, Kaist, Taejon, Korea) in GDB (30 mM phosphate buffer pH 7.4containing 0.2% gelatin, 0.5% Triton-X-100 and 0.8 M NaCl) for 1 h atroom temperature to label intracellular receptors and Shank. After 3washes with 20 mM phosphate buffer containing 0.5 M NaCl, neuronswere incubated with FITC-, Cy3- and Cy5-conjugated secondary anti-bodies (Jackson ImmunoResearch, West Grove, Pennsylvania) diluted inGDB for 1 h at room temperature. Images were acquired with a BioradMRC1024 confocal microscope.

Image analysis and quantification. Labeled transfected neurons were cho-sen randomly for quantification from 2–5 platings of neurons. The num-ber of neurons used for quantification for each experimentalcondition/time point ranged from five to seven. Confocal images wereobtained using a Zeiss 63× (NA 1.4) objective with sequential acquisitionsettings at the maximal resolution of the confocal (1280 × 1024 pixels).

Each image was a z-series of 7–12 images each averaged 2–3 times with aKalman filter and taken at 0.75 µm depth intervals. The resultant stackwas ‘flattened’ into a single image using a maximum projection. The con-focal microscope settings were kept the same for all scans. All morpho-metric measurements were done using MetaMorph image analysissoftware (Universal Imaging Corporation, West Chester, Pennsylvania).Single neurons were selected and carefully manually traced for maximumaccuracy. The average intensity of surface fluorescence staining was mea-sured in the traced regions and the background staining (determined overneuron-free areas of the culture) was subtracted. Intensity measurementsare expressed in arbitrary units of fluorescence per square area. To measurethe distribution of surface insertion, ‘synaptic’ staining was determinedby measuring the average surface fluorescence intensity in regions thatoverlapped with Shank immunofluorescence; ‘non-synaptic’ stainingintensity was calculated by subtracting the synaptic intensity value fromthe total (synaptic and non-synaptic) value. All values in figures and textrefer to mean ± s.e.m.

Receptor recycling assay. Four days after transfection with HA/T-GluR1or HA/T-GluR2, live neurons were surface-labeled at 37°C for 10 minwith mouse anti-HA antibodies, washed and returned to 37°C CO2 incu-bator for a further 10 min to allow for internalization. After this inter-nalization period, the remaining surface antibodies were ‘stripped’ by anacid wash11 (0.5 M NaCl/0.2 M acetic acid) on ice for 4 min. Cells werethen replaced in DMEM and returned to 37°C CO2 incubator for 0, 10 or30 min (to allow resurfacing of the internalized receptor/antibody com-plex). After this incubation, neurons were fixed and stained with FITC-conjugated anti-mouse antibody for 30 min at RT in non-permeabilizingconditions, washed and then stained with another anti-mouse antibody(Cy3-conjugated) for 30 min at RT in GDB (permeabilizing conditions).Confocal images were acquired using the same threshold setting.

Endogenous AMPA receptor staining. Live hippocampal neurons (18d.i.v.) were labeled for 10 min at 37°C with an antibody (10 µg/ml) direct-ed against the extracellular region of either AMPA receptor subunit GluR1(Oncogene Research Products, Cambridge, Massachusetts) or GluR2(Chemicon International, Temecula, California). After washing, neuronswere fixed with paraformaldehyde/4% sucrose and incubated with Shankguinea pig antibody in GDB for 1 h at room temperature. Cells werewashed in 20 mM phosphate buffer containing 0.5 M NaCl and incu-bated with FITC-, Cy3- and Cy5-conjugated secondary antibodies (Jack-son ImmunoResearch Labs).

Note: Supplementary figure of a model for postsynaptic AMPA receptor

trafficking is available on the Nature Neuroscience web site (http://neuroscience.

nature.com/web_specials).

ACKNOWLEDGEMENTSWe thank C. Sala, S. Hyoung Lee, M. Wyszynski and Y. Tian Wang for

discussions. This work was supported by NIH grant NS 35050 (M.S.). M.S. is

Associate Investigator of the Howard Hughes Medical Institute.


1. Dingledine, R., Borges, K., Bowie, D. & Traynelis, S. F. The glutamate receptorion channels. Pharmacol. Rev. 51, 7–61 (1999).

2. Turrigiano, G. G. AMPA receptors unbound: membrane cycling and synapticplasticity. Neuron 26, 5–8 (2000).

3. Malinow, R., Mainen, Z. F. & Hayashi, Y. LTP mechanisms: from silence tofour-lane traffic. Curr. Opin. Neurobiol. 10, 352–357 (2000).

4. Scannevin, R. H. & Huganir, R. L. Postsynaptic organization and regulationof excitatory synapses. Nat. Rev. Neurosci. 1, 133–141 (2000).

5. Lüscher, C., Nicoll, R. A., Malenka, R. C. & Muller, D. Synaptic plasticity anddynamic modulation of the postsynaptic membrane. Nat. Neurosci. 3,545–550 (2000).

6. Lüscher, C. et al. Role of AMPA receptor cycling in synaptic transmission andplasticity. Neuron 24, 649–658 (1999).

7. Wang, Y. T. & Linden, D. J. Expression of cerebellar long-term depressionrequires postsynaptic clathrin-mediated endocytosis. Neuron 25, 635–647(2000).

8. Man, Y. H. et al. Regulation of AMPA receptor-mediated synaptic transmissionby clathrin-dependent receptor internalization. Neuron 25, 649–662 (2000).

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



9. Carroll, R. C., Lissin, D. V., von Zastrow, M., Nicoll, R. A. & Malenka, R. C.Rapid redistribution of glutamate receptors contributes to long-termdepression in hippocampal cultures. Nat. Neurosci. 2, 454–460 (1999).

10. Lissin, D. V. et al. Activity differentially regulates the surface expression ofsynaptic AMPA and NMDA glutamate receptors. Proc. Natl. Acad. Sci. USA95, 7097–7102 (1998).

11. Carroll, R. C. et al. Dynamin-dependent endocytosis of ionotropic glutamatereceptors. Proc. Natl. Acad. Sci. USA 96, 14112–14117 (1999).

12. Lin, J. W. et al. Distinct molecular mechanisms and divergent endocytoticpathways of AMPA receptor internalization. Nat. Neurosci. 3, 1282–1290(2000).

13. Beattie, E. C. et al. Regulation of AMPA receptor endocytosis by a signalingmechanism shared with LTD. Nat. Neurosci. 3, 1291–1300 (2000).

14. Ehlers, M. D. Reinsertion or degradation of AMPA receptors determined byactivity-dependent endocytic sorting. Neuron 28, 511–525 (2000).

15. Xia, J., Chung, H. J., Wihler, C., Huganir, R. L. & Linden, D. J. Cerebellarlong-term depression requires PKC-regulated interactions between GluR2/3and PDZ domain-containing proteins. Neuron 28, 499–510 (2000).

16. Chung, H. J., Xia, J., Scannevin, R. H., Zhang, X. & Huganir, R. L.Phosphorylation of the AMPA receptor subunit GluR2 differentially regulatesits interaction with PDZ domain-containing proteins. J. Neurosci. 20,7258–7267 (2000).

17. Hayashi, Y. et al. Driving AMPA receptors into synapses by LTP and CaMKII:requirement for GluR1 and PDZ domain interaction. Science 287, 2262–2267(2000).

18. Shi, S. H. et al. Rapid spine delivery and redistribution of AMPA receptorsafter synaptic NMDA receptor activation. Science 284, 1811–1816 (1999).

19. Zhu, J. J., Esteban, J. A., Hayashi, Y. & Malinow, R. Postnatal synapticpotentiation: delivery of GluR4-containing AMPA receptors by spontaneousactivity. Nat. Neurosci. 3, 1098–1106 (2000).

20. Shi, S., Hayashi, Y., Esteban, J. A. & Malinow, R. Subunit-specific rulesgoverning AMPA receptor trafficking to synapses in hippocampal pyramidalneurons. Cell 105, 331–343 (2001).

21. Lledo, P.-M., Zhang, X., Südhof, T. C., Malenka, R. C. & Nicoll, R. A.Postsynaptic membrane fusion and long-term potentiation. Science 279,399–403 (1998).

22. Chen, L. et al. Stargazin regulates synaptic targeting of AMPA receptors bytwo distinct mechanisms. Nature 408, 936–943 (2000).

23. Nakagawa, T. & Sheng, M. Neurobiology. A stargazer foretells the way to thesynapse. Science 290, 2270–2271 (2000).

24. Hein, L., Ishii, K., Coughlin, S. R. & Kobilka, B. K. Intracellular targeting andtrafficking of thrombin receptors. A novel mechanism for resensitization of aG protein-coupled receptor. J. Biol. Chem. 269, 27719–27726 (1994).

25. Daunt, D. A. et al. Subtype-specific intracellular trafficking of α2-adrenergicreceptors. Mol. Pharmacol. 51, 711–720 (1997).

26. Wenthold, R. J., Petralia, R. S., Blahos, J. I. & Niedzielski, A. S. Evidence for

multiple AMPA receptor complexes in hippocampal CA1/CA2 neurons. J. Neurosci. 16, 1982–1989 (1996).

27. Leonard, A. S., Davare, M. A., Horne, M. C., Garner, C. C. & Hell, J. W. SAP97is associated with the alpha-amino-3-hydroxy-5-methylisoxazole-4-propionic acid receptor GluR1 subunit. J. Biol. Chem. 273, 19518–19524(1998).

28. Dong, H. et al. GRIP: a synaptic PDZ domain-containing protein thatinteracts with AMPA receptors. Nature 386, 279–284 (1997).

29. Srivastava, S. et al. Novel anchorage of GluR2/3 to the postsynaptic density bythe AMPA receptor-binding protein ABP. Neuron 21, 581–591 (1998).

30. Wyszynski, M. et al. Association of AMPA receptors with a subset ofglutamate receptor-interacting protein in vivo. J. Neurosci. 19, 6528–6537(1999).

31. Sheng, M. & Pak, D. T. S. Ligand-gated ion channel interactions withcytoskeletal and signaling proteins. Annu. Rev. Physiol. 62, 755–778 (2000).

32. Li, P. et al. AMPA receptor-PDZ interactions in facilitation of spinal sensorysynapses. Nat. Neurosci. 2, 972–977 (1999).

33. Osten, P. et al. Mutagenesis reveals a role for ABP/GRIP binding to GluR2 insynaptic surface accumulation of the AMPA receptor. Neuron 27, 313–325(2000).

34. Daw, M. I. et al. PDZ proteins interacting with C-terminal GluR2/3 areinvolved in a PKC-dependent regulation of AMPA receptors at hippocampalsynapses. Neuron 28, 873–886 (2000).

35. Matsuda, S., Launey, T., Mikawa, S. & Hirai, H. Disruption of AMPA receptorGluR2 clusters following long-term depression induction in cerebellarPurkinje neurons. EMBO J. 19, 2765–2774 (2000).

36. Lu, W. et al. Activation of synaptic NMDA receptors induces membraneinsertion of new AMPA receptors and LTP in cultured hippocampal neurons.Neuron 29, 243–254 (2001).

37. Klausner, R. D., Donaldson, J. G. & Lippincott-Schwartz, J. Brefeldin A:insights into the control of membrane traffic and organelle structure. J. CellBiol. 116, 1071–1080 (1992).

38. Chardin, P. & McCormick, F. Brefeldin A: the advantage of beinguncompetitive. Cell 97, 153–155 (1999).

39. Zamanillo, D. et al. Importance of AMPA receptors for hippocampal synapticplasticity but not for spatial learning. Science 284, 1805–1811 (1999).

40. Jia, Z. et al. Enhanced LTP in mice deficient in the AMPA receptor GluR2.Neuron 17, 945–956 (1996).

41. Sheng, M. & Sala, C. Pdz domains and the organization of supramolecularcomplexes. Annu. Rev. Neurosci. 24, 1–29 (2001).

42. Zhou, Q., Xiao, M. & Nicoll, R. A. Contribution of cytoskeleton to theinternalization of AMPA receptors. Proc. Natl. Acad. Sci. USA 98, 1261–1266(2001).

43. Brewer, G. J., Torricelli, J. R., Evege, E. K. & Price, P. J. Optimized survival ofhippocampal neurons in B27-supplemented Neurobasal, a new serum-freemedium combination. J. Neurosci. Res. 35, 567–576 (1993).

articles©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


Breathing is an exceptionally reliable and continuous mammalianbehavior that regulates blood gases and pH at rest and in responseto diverse challenges such as exercise, sleep or altitude. Therhythm underlying breathing is postulated to depend criticallyon neurons in the preBötC1,2. Two recent developments allowedus to determine in awake adult rats whether this critical regulatorybehavior would be affected by lesions in the preBötC. First, theextent of the preBötC can be anatomically defined by the sub-population of propriobulbar respiratory neurons within the ven-trolateral respiratory column expressing NK1R3–5. Second, NK1Rneurons can be specifically lesioned by substance P conjugatedto saporin (SP-SAP)6, an effect that takes several days, allowingfor complete recovery from surgery6.

RESULTSSP-SAP was effective in eliminating preBötC NK1R neurons inadult rats. Injections of 0.1–0.2 pmol of SP-SAP or 0.3 pmoleach of unconjugated saporin and SP were made in the pre-BötC. Unilateral SP-SAP injection transiently produced sighs(large inspiratory efforts followed by prolonged expiration),consistent with the effects of injection of SP alone into the pre-BötC in vitro7 and in vivo (P.A.G., D.R.M. and J.L.F., unpub-lished observations). All rats returned to normal behavior uponrecovery from surgery. Two to eighteen days after injection, ratswere perfused and their medullas were stained for NK1Rimmunoreactivity (n = 20). The lesion extent was estimated bycounting NK1R immunopositive neuronal soma in the rostralmedulla inside a circle of 600-µm diameter approximating thepreBötC (ventrolateral to the nucleus ambiguus) and in a rec-tangle (1600 × 1070 µm) outside this circle (Fig. 1, inset).

Normal breathing requirespreBötzinger complex neurokinin-1receptor-expressing neurons

Paul A. Gray1,2, Wiktor A. Janczewski1, Nicholas Mellen1, Donald R. McCrimmon3 andJack L. Feldman1

1 Department of Neurobiology, University of California Los Angeles, Box 951763, Los Angeles, California 90095-1763, USA2 Interdepartmental Ph.D. Program in Neuroscience, University of California Los Angeles, Box 951761, Los Angeles, California 90095-1761, USA3 Department of Physiology and Northwestern Institute for Neuroscience, Northwestern University Medical School, 303 East Chicago Avenue,

Chicago, Illinois 60611-3008, USA

The first two authors contributed equally to this work

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

The normal breathing rhythm in mammals is hypothesized to be generated by neurokinin-1 receptor(NK1R)-expressing neurons in the preBötzinger complex (preBötC), a medullary region proposed tocontain the kernel of the circuits generating respiration. If this hypothesis is correct, then completedestruction of preBötC NK1R neurons should severely perturb and perhaps even fatally arrest breath-ing. Here we show that specific and near complete bilateral (but not unilateral) destruction ofpreBötC NK1R neurons results in both an ataxic breathing pattern with markedly altered blood gasesand pH, and pathological responses to challenges such as hyperoxia, hypoxia and anesthesia. Thus,these ∼ 600 neurons seem necessary for the generation of normal breathing in rats.

Counts were estimated from four transverse sections beginningwithin 100 µm of the rostral border of the lateral reticular nucle-us and spanning the preBötC3,4 containing approximately12–15% of the total preBötC volume. Uninjected controls (n = 4) had 35 ± 5.1 (mean ± s.e.m., per side) NK1R neuronswithin the preBötC (∼ 600 preBötC NK1R neurons total, whichwe estimate represent less than 10% of all preBötC neurons)and 82 ± 9 NK1R neurons outside the preBötC. Most of the lat-ter were immediately ventral to the preBötC. Unilaterally SP-SAP injected rats (n = 4) had 0 ± 0 NK1R neurons inside and29 ± 9 NK1R neurons outside the preBötC on the injected sidecompared to the uninjected side, which had 47 ± 6 inside and 62 ± 6 outside. In contrast, tyrosine hydroxylase (TH)-immunoreactive neurons, which do not express NK1Rbut which have soma and dendrites within the target site3,4,were largely unaffected (Fig. 1c; unilateral injections, injectedside, 111 ± 15; control side; 121 ± 15, n = 4), although someTH neurons at the injection site showed damage. NK1R stain-ing in dorsal brainstem regions such as the solitary tract andmotor nucleus of the vagus appeared normal with no signs oftoxin-mediated degradation or receptor internalization (n = 20)8, although collateral damage in other regions was pos-sible. In some rats, nucleus ambiguus motoneurons adjacent tothe preBötC were mildly damaged (Fig. 1b); however, motoneu-ronal damage could not account for the changes in breathingfrequency or pattern we observed. Bilateral injection of uncon-jugated saporin and SP had little effect on preBötC NK1R neu-rons (inside, 30 ± 4; outside, 70 ± 14; per side, n = 2).

Bilateral but not unilateral SP-SAP lesions profoundly affect-ed breathing. We measured respiratory period and inspiratory

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 2. PreBötC– rats breathe with an ataxic pattern. Spontaneousbreathing from control (top of each pair) and preBötC– (bottom of eachpair) rats. Plethysmographic traces breathing room air (top), 5% CO2/95% O2 (middle) and 100% O2 (bottom). Upward deflectionrepresents inspiration. Control rats breathing room air spontaneouslygenerate an occasional sigh (arrow) followed by an extended expiratorypause; this pattern seems absent in preBötC– rats. (Movies of apreBötC– rat at 2 and 8 days after injection are available on the supple-mentary information page of Nature Neuroscience online.)


amplitude from unrestrained, awake adult rats. Bilaterally SP-SAP injected rats with near-complete bilateral destructionof preBötC NK1R neurons (see below) exhibited a transforma-tion in breathing pattern 4–5 days after injection, going froma eupneic to a severely ataxic pattern in room air (preBötC–; n = 10; Fig. 2). The changes in breathing pattern were stronglycorrelated to the loss of preBötC NK1R neurons. PreBötC– ratsthat were perfused 10–18 days after injection (sufficient timefor clearance of dead NK1R neurons) had markedly fewer NK1Rneurons within (4 ± 1; range, 0–7; p < 0.001, total, both sides)and somewhat fewer NK1R neurons outside (70 ± 10; range,66–96; p < 0.05, total) the preBötC, per side (n = 6). Bilateral-ly injected rats that maintained normal breathing patterns andthat were perfused 10–12 days after injection had 30 ± 9 NK1Rneurons inside (range, 15–52; p < 0.05) and 105 ± 13 NK1Rneurons outside (range, 81–131; p > 0.05, n = 4) per side, witha maximum estimated lesion sparing more than 20% of preBötC NK1R neurons.

The respiratory period of uninjected (control) rats was 0.53 ± 0.001 s (n = 10) and normalized peak inspiratory ampli-tude (Ipeak) was 1.00 ± 0.006 (n = 3), with occasional sniffing(period, 0.18 ± 0.001 s) and sighing (Figs. 2 and 3). Ataxia inpreBötC– rats was characterized by shortened respiratory peri-ods (0.29 ± 0.01 s, n = 8, p < 0.0001) and an irregular sequenceof inspiratory efforts of near normal amplitude interspersedwith (prolonged) periods of apnea or very low amplitude inspi-ration (Ipeak = 0.85 ± 0.004; n = 3, p < 0.001, Figs. 2 and 3).One additional rat (>90% bilateral destruction of preBötCNK1R neurons perfused at day 18) with an ataxic breathingpattern with noticeable apneas did not show periods of lowamplitude inspiration. Arterial blood gases and pH values ofpreBötC– rats four days after injection, when their breathingpatterns still seemed normal, suggested a slight hyperventila-tion giving rise to reduced CO2 with consequent alkalosis andelevated O2 (pH 7.44 ± 0.01; PCO2, 31.5 ± 1.0 mm Hg;

PO2, 97.25 ± 2.7 mm Hg; n = 4) compared to control rats(pH 7.40 ± 0.01; PCO2, 36.5 ± 1.9 mm Hg; PO2, 92.5 ± 5.5 mmHg; n = 2; T = 37.4 ± 0.1°C; n = 17). Five days after injection,however, preBötC– rats had a profile typical of respiratorydepression giving rise to CO2 retention and consequent aci-dosis (pH 7.26 ± 0.07; PCO2, 56.4 ± 3.7 mm Hg; PO2,76.5 ± 5.5 mm Hg; n = 2; T = 36.8 ± 0.1°C; n = 14).

PreBötC– rats had pathological responses to changes ininspired gases. In response to 100% O2, control rats had anincreased period (0.6 ± 0.002 s, n = 9, p < 0.001) and anincreased Ipeak (1.3 ± 0.008, n = 3, p < 0.0001; Figs. 2 and 3).PreBötC– rats responded with a marked (further) depression ofbreathing, with a small decrease (∼ 3%) in period (0.28 ± 0.002 s, n = 4, p < 0.05) and a decrease in Ipeak(0.7 ± 0.006, n = 3, p < 0.0001) due primarily to the loss of nor-mal amplitude events (Fig. 3). Two rats developed fatal apneas.The depressive effects of 100% O2 in ataxic animals are in con-trast to its normalizing effects in humans with Cheyne–Stokesbreathing, which has a vaguely similar respiratory phenotype;this suggests a different etiology for preBötC– rats. In controlrats, 5% CO2 and 95% O2 caused a significantly decreased(∼ 25%) period (0.4 ± 0.001 s, p < 0.0001) and increased Ipeak(1.8 ± 0.009, n = 3, p < 0.0001) compared to response whenbreathing room air (Figs. 2 and 3). PreBötC– rats, in contrast,showed a smaller (∼ 14%) decrease in respiratory period frombreathing room air (0.25 ± 0.001 s, n = 5, p < 0.0001) with anincrease in Ipeak (1.1 ± 0.007, p < 0.0001) due primarily to theappearance of a number of higher amplitude respiratory eventsamong continued low-amplitude inspiration (Fig. 3). In responseto severe hypoxia (4.4% O2/95.6% N2), control and injected ratsup to four days after injection showed shortened (∼ 34%) periods(0.35 ± 0.001 s, n = 7, p < 0.0001) and increased amplitude; theserats tolerated this challenge for at least 15 minutes (n = 16; datanot shown). PreBötC– rats (n = 8) had a brief increase in fre-

articles

NK1R

ControlscNA scNA

IO IOV

SP-SAP

5 SPscNA

XII

preBötC IOscNA

IO IO

L

ChAT TH

Fig. 1. Local injection of SP-SAP selectively ablates preBötC NK1Rneurons. Inset, transverse section of adult rat brainstem showing rela-tive location of preBötC (circle) ventrolateral to scNA. Rectangle(excluding circle) indicates area outside preBötC analyzed for neuroncounts (see text). Comparison of NK1R (a), choline acetyltransferase(ChAT) (b) and tyrosine hydroxylase (TH) (c) immunohistochemistry inunilaterally injected rat. scNA, semicompact division of nucleusambiguus; IO, inferior olive; 5SP, trigeminal nucleus; XII, hypoglossalnucleus. Scale bars, 200 µm.

a

b c

control Room air

5% CO2

100% O2

preBötC-

1.0 Ipeak5 s

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 4. Anesthetized (fluorothane, 1–2%) preBötC– rats do not havespontaneous respiratory activity. However, lung deflation inducesdiaphragmatic activity (Dia. EMG, top). Arrows point to short-lasting(<100 ms) lung deflations that induce bursts of diaphragmatic activityleading to airflow.

quency and amplitude, followed within 2–7 minutes by apneas of increasing and even-tually fatal duration. Only one preBötC– ratrecovered spontaneous breathing when quicklyreturned to room air. In addition, breathing inpreBötC– rats was unusually sensitive to depres-sion by anesthesia, similar to humans with Jou-bert’s Syndrome9. Thus, in contrast to controlrats, no spontaneous respiratory drive could berecorded in preBötC– rats at modest levels ofinhaled fluorothane (1–2%; n = 6, Fig. 4).Moreover, premotor circuits driving respirato-ry motoneurons appeared intact, as lung defla-tion in anesthetized preBötC– rats induced arobust reflex diaphragmatic output (Fig. 4).

DISCUSSIONAlthough the importance of the preBötC ingenerating a respiratory-related rhythm in in vitro preparations from neonatal rodents iswell established1,2,10, the role of the preBötCin generating the normal rhythm of breathingin intact mammals is controversial1,11. The res-piratory pattern of in vitro preparations issomewhat similar in pattern, frequency andchemosensitivity to gasping in anesthetizedadult rats7,11 and it has been suggested thathypotheses derived from these preparationshave little relevance to normal breathing inintact animals1,11. Thus, we tested the hypoth-esis that the preBötC is critical for the genera-tion of normal breathing in awake, behavinganimals. We found that bilateral injections ofSP-SAP that eliminated greater than 80% ofpreBötC NK1R neurons resulted in a highlyirregular breathing pattern in conscious rats.These rats were unable to maintain appropriate blood gasesand pH, and had pathological responses to hypoxia or hyper-oxia. However, these lesions did not result in the rapid onsetof fatal apnea when these preBötC– rats were conscious. Whentaken together with previous results1–3,10,12,13, our interpreta-tion is that normal breathing in mammals requires an intactpreBötC, with NK1R neurons having a necessary role. In theirabsence, an ataxic rhythm sufficient to maintain life can stillbe generated in conscious rats by undetermined structures,which may include non-NK1R expressing neurons in the pre-BötC or ventrolateral respiratory column14, other respiratoryregions13, vestigial brainstem rhythmogenic networks15,16 oreven structures normally underlying voluntary or emotionalcontrol of respiratory muscles. The extent to which brainregions in addition to the preBötC are also necessary for thenormal pattern of breathing is yet to be determined. PreBötC

Fig 3. PreBötC rats (3–5 days after injection) and control rats breathe differently. Each of 9graphs represents grouped data (n = 3, 10 consecutive minutes) of instantaneous respiratoryperiod (0–0.75 s) versus Ipeak (0.0–2.0) for 4800–7000 breaths, plotted as two-dimensionaldensity for control (top), preBötC– (middle) and combined control (red)/preBötC– (green;bottom). Individual breaths (and period) were converted into a two-dimensional histogram(bin width, 0.02 s and 0.02) and plotted as density maps. Top and middle, colors represent therelative number of breaths in a given bin as percentages of breaths in the maximum bin, asindicated in legend at right; densities less than 10% of maximum are black. Bottom, flattenedcomparison between control (red) and preBötC– (green) rats of all bins with densities ≥25%of maximum. Arrow points to sniffing behavior. Data was obtained from rats breathing roomair (left), 5% CO2/95% O2 (middle) and 100% O2 (right).

NK1R neurons may contain the kernel for the genetically deter-mined behavior of breathing in mammals, and delineating theirproperties may provide ways to understand and treat life-threatening and poorly understood central failures of breathingsuch as central sleep apnea17,18, congenital central hypoventi-lation syndrome (Ondine’s curse)19,20, Joubert’s Syndrome21

and Sudden Infant Death Syndrome22.

METHODSInjection of SP-SAP. We used 54 adult Sprague–Dawley male rats. Allprocedures were approved by the UCLA Animal Care Committee(ARC# 94-159-22). Rats were anesthetized with 100 mg/kg of ketamine(Ketaject, Phoenix Scientific, St. Joseph, Missouri) and 10 mg/kg ofXylazine (AnaSed, Lloyd Laboratories, Shenandoah, Iowa) adminis-tered intraperitoneally. Rats were intubated and respiratory flow wasrecorded (MLT1L flow head from ADInstruments, Grand Junction,Colorado and DRAL501 transducer, Honeywell Data Instruments,Acton, Massachusetts). When necessary, rats were mechanically venti-lated. Rats breathed 1–2 volume percent fluorothane (IsoFlo, AbbottLaboratories, North Chicago, Illinois) in 1:1 air:oxygen mixture. Ratswere positioned in a stereotaxic frame with bregma 5 mm below lamb-da. Microinjections (100–150 nl) of SP-SAP (Advanced Targeting Sys-tems, San Diego, California) were made from a glass capillary tube with

articles


10%

90%

0.75

2.0

Period (s)

Room air 5% CO2 100% O2C

on

trol

PreB

ötC

-

I pea

k

00

Dia.EMG

+

-

Pressure

2 V

10 cmH2O

5 s

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



a 40-µm tip diameter. Coordinates were 0.7 mm rostral, 1.95 mm lateraland 2.6 mm ventral to the obex. When the inferior cerebellar vein wasin the way of injection, the micropipette was moved rostrally (50–150 µm) to avoid puncture. A second injection was made just cau-dal to the vein. Fluorescent microspheres (Molecular Probes, Eugene,Oregon) were added to the solution to allow identification of injectionsites. For arterial blood sampling, the femoral artery was cannulatedand an artery access port was implanted subcutaneously. Blood oxy-gen and carbon dioxide levels were assessed with a blood gas analyzer.

Histology. Rats (n = 39, 24 brainstems were of sufficient quality todo cell counts) were transcardially perfused with 4% paraformalde-hyde in PBS, cryoprotected in 25% sucrose PBS, embedded in OCT,sectioned at 40 µm on a cryostat and processed free floating. Sectionswere incubated in primary antibody diluted in sera at 4°C overnight,placed in biotin conjugated species-specific secondary antibody (Vec-tor Laboratories, Burlingame, California), stained using the ABCmethod (Vector Laboratories) and mounted on gelatin subbed slides.The following antibodies were used: rabbit anti-NK1R (1:20,000,Chemicon, Temucula, California), goat anti-ChAT (1:400, Chemi-con) and mouse anti-tyrosine hydroxylase (1:1000, BoehringerMannheim, Indianapolis, Indiana). Brightfield images were digitallyacquired into Photoshop (Adobe Systems, San Jose, California). Allimages were filtered and adjusted for contrast and light levels in Pho-toshop for clarity. NK1R somata were visually identified by the pres-ence of at least 2 labeled processes and their location plotted viacamera lucida and counted within a 1600 × 1040 µm region boundedlaterally by the 5SP and dorsally by the nucleus ambiguus. Neuronswithin the preBötC were counted by placing a 600-µm circle just ven-tral to and centered upon the lateral edge of the NA. Cells in sectionsseparated by 200 µm were counted.

Phethysmography. All respiratory recordings of unrestrained rats wereobtained using a seven-liter whole-body plethysmograph (Buxco,Sharon, Connecticut) and a pressure transducer (DRAL501, Honey-well Data Instruments). The data was recorded using Axoscope soft-ware (Axon Instruments, Union City, California). Respiratory periodswere determined using peak detection software written in LabView soft-ware (National Instruments, Austin, Texas) or by hand measurementof plethysmographic traces. Peak inspiratory amplitudes were calcu-lated from pressure transducer output (in millivolts) and normalized.For control rats breathing room air and 100% oxygen, mean respira-tory period was calculated after removal of periods less than 0.3 s toexclude sniffing. For calculation of mean inspiratory amplitude, val-ues greater than 10 ml were excluded as motion artifacts. All numbersare expressed as mean ± s.e.m. Mean values were calculated from 15-min (period, room air) or 10-min (all else, 5 min after gas exchange)recordings of continuous respiratory behavior. Control means werecompared against control room air, and ataxic means were comparedagainst control means for the same gas mixture for statistical analysis(Student’s t-test for independent variables) in Origin (Microcal,Northampton, Massachusetts).

Apneic rats. Rats that developed irreversible apnea were anesthetized andmechanically ventilated at a rate adequate to keep end-tidal CO2 at the5% level. Bipolar wire electrodes were implanted into the costal part ofdiaphragm (n = 6) and abdominal muscle (n = 2). Needle electrodeswere used to monitor ECG. Signals were amplified filtered and integrat-ed. For lung deflation, a constant negative pressure source was used.Intratracheal pressure was monitored.

Note : Supplementary movies of a preBötC– rat at two and eight days after

injection are available on the Nature Neuroscience web site


ACKNOWLEDGEMENTSWe thank M. Sofroniew, N. Brecha, T. Otis, R. Fregosi and L. Kruger for

assistance and G. Li for histological work. A. Monnier participated in early

experiments. Funding was provided by a Ford Foundation Pre-Doctoral

Fellowship for Minorities and the Porter Physiology Development Program of

the American Physiological Society to P.A.G., and by the National Institutes of

Health (HL40959). W.A.J. is on leave of absence from the Medical Research

Centre, Polish Academy of Sciences, Warsaw, Poland.


1. Rekling, J. C. & Feldman, J. L. PreBötzinger Complex and pacemakerneurons: hypothesized site and kernel for respiratory rhythm generation.Annu. Rev. Physiol. 60, 385–405 (1998).

2. Smith, J. C., Ellenberger, H. H., Ballanyi, K., Richter, D. W. & Feldman, J. L.Pre-Bötzinger Complex: a brainstem region that may generate respiratoryrhythm in mammals. Science 254, 726–729 (1991).

3. Gray, P. A., Rekling, J. C., Bocchiaro, C. M. & Feldman, J. L. Modulation ofrespiratory frequency by peptidergic input to rhythmogenic neurons in thepreBötzinger Complex. Science 286, 1566–1568 (1999).

4. Wang, H., Stornetta, R. L., Rosin, D. L. & Guyenet, P. G. Neurokinin-1receptor-immunoreactive neurons of the ventral respiratory group in the rat.J. Comp. Neurol. 434, 128–146 (2001).

5. Guyenet, P. G. & Wang, H. Pre-Bötzinger neurons with preinspiratorydischarges “in vivo” express NK1 receptors in the rat. J. Neurophysiol. 86,438–446 (2001).

6. Mantyh, P. W. et al. Inhibition of hyperalgesia by ablation of lamina I spinalneurons expressing the substance P receptor. Science 278, 275–279 (1997).

7. Lieske, S. P., Thoby-Brisson, M., Telgkamp, P. & Ramirez, J. M.Reconfiguration of the neural network controlling multiple breathingpatterns: eupnea, sighs and gasps. Nat. Neurosci. 3, 600–607 (2000).

8. Mantyh, P. W. et al. Receptor endocytosis and dendrite reshaping in spinalneurons after somatosensory stimulation. Science 268, 1629–1632 (1995).

9. Habre, W., Sims, C. & D’Souza, M. Anaesthetic management of children withJoubert syndrome. Paediatr. Anaesth. 7, 251–253 (1997).

10. Ballayni, K., Onimaru, H. & Homma, I. Respiratory network function in theisolated brainstem-spinal cord of newborn rats. Prog. Neurobiol. 59, 583–684(1999).

11. Remmers, J. E. Central neural control of breathing. in Lung Biology in Healthand Disease: Control of Breathing in Health and Disease (eds. Altose, M. &Kawakami, Y.) 1–41 (M. Dekker, New York, 1999).

12. McCrimmon, D. R., Monnier, A., Hayashi, F. & Zuperku, E. J. Patternformation and rhythm generation in the ventral respiratory group. Clin. Exp.Pharmacol. Physiol. 27, 126–131 (2000).

13. Dobbins, E. G. & Feldman, J. L. Brainstem network controlling descendingdrive to phrenic motoneurons in rat. J. Comp. Neurol. 347, 64–86 (1994).

14. Speck, D. F. & Feldman, J. L. Effects of microstimulation and microlesion inthe ventral and dorsal respiratory groups in medulla of cat. J. Neurosci. 2,744–757 (1982).

15. Champagnat, J. & Fortin, G. Primordial respiratory-like rhythm generationin the vertebrate embryo. Trends Neurosci. 20, 119–124 (1997).

16. Gdovin, M. J., Torgerson, C. S. & Remmers, J. E. The fictively breathingtadpole brainstem preparation as a model for the development of respiratorypattern generation and central chemoreception. Comp. Biochem. Physiol. A.Mol. Integr. Physiol. 124, 275–286 (1999).

17. Guilleminault, C. & Robinson, A. Central sleep apnea. Neurol. Clin. 14,611–628 (1996).

18. Thalhofer, S. & Dorow, P. Central sleep apnea. Respiration 64, 2–9 (1997).19. Gozal, D. Congenital central hypoventilation syndrome: an update. Pediatr.

Pulmonol. 26, 273–282 (1998).20. Nattie, E., Bartlett Jr., D. & Rozycki, A. Central alveolar hypoventilation in a

child: an evaluation using a whole body plethysmograph. Amer. Rev. Resp.Dis. 112, 259–265 (1975).

21. Maria, B. L., Boltshauser, E., Palmer, S. C. & Tran, T. X. Clinical features andrevised diagnostic criteria in Joubert syndrome. J. Child Neurol. 14, 583–590(1999).

22. Guilleminault, C. & Robinson, A. Developmental aspects of sleep andbreathing. Curr. Opin. Pulm. Med. 2, 492–499 (1996).

articles©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


articles


During human brain evolution, considerable enlargement ofthe association areas of the cortex is accompanied by a com-parable enlargement of corresponding association thalamicnuclei in the diencephalon, but developmental mechanismscoordinating these expansions remain unknown1. Telencephalicand diencephalic neurons in mammalian species originate earlyfrom distinct proliferative zones2–5. However, histological stud-ies of the embryonic human brain have shown a putativemigratory stream of cells extending from the telencephalic gan-glionic eminence (GE) to the dorsal thalamic association nucleiafter the end of neurogenesis in the diencephalic proliferativezone5–7. This suggests that some thalamic nuclei import a sec-ond, late wave of their neurons from the telencephalon6. Thismigratory-like stream was not evident in the developingmacaque monkey brain or in any other mammalian speciesthat has been examined so far2–4,8–12, suggesting that it could bea special feature of human or hominoid thalamic development.Such a transdivisional crossing of neurons is unusual, and theonly other example of neurons crossing brain boundaries mightbe cells migrating from the dorsal rhombic lip to the precere-bellar pontine nuclei3,13–14.

Here, using organotypic slice cultures of human forebrain,we provide direct evidence for the migration from the GE tothe dorsal thalamus (DT). Furthermore, because the pheno-type, function and mode of migration of these cells isunknown, we studied their nature, guidance mechanisms andmigratory substrate. Finally, we examined whether modifica-tions in the guidance cues for neuronal migration in the courseof evolution might be responsible for the presence of thismigratory pathway in the human but its absence in other mam-malian species.

RESULTSWe confirmed the existence of a distinct migratory-like path-way extending from the GE to the DT in all specimens between

Telencephalic origin of humanthalamic GABAergic neurons

Kresimir Letinic and Pasko Rakic

Yale University School of Medicine, Section of Neurobiology, New Haven, Connecticut 06510, USA

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

In non-primate mammalian species, telencephalic and diencephalic neurons originate from theirrespective local proliferative zones. Using vital dye labeling in organotypic slice cultures, we showthat in human brain, a contingent of neurons from the ganglionic eminence of the telencephalonmigrate to the dorsal thalamic association nuclei of the diencephalon. These neurons rely onhomotypic–neurophilic guidance during their migration, are GABAergic, and express Dlx1/2homeodomain-containing proteins. Similar experiments in a non-human primate and in rodentembryos did not reveal a similar migratory pathway. Migration assays demonstrated that the humandorsal thalamus attracts telencephalic cells, an effect not observed in the mouse, in which such migra-tion is inhibited by chemorepulsive cues. These data suggest that modifications in the pattern ofmigratory guidance cues in the forebrain may underlie the appearance of a new migratory pathwayand thus contribute to the evolutionary expansion of the thalamic association nuclei in the human.

15 and 26 weeks of gestation, which is long after neuronal pro-liferation in the human diencephalon has been completed. Thispathway is located just under the surface of the DT, extendsmedially from the GE and, in Nissl-stained sections, is clearlyvisible as a thin cellular layer (Fig. 1).

To directly demonstrate the migration of cells from the GEto the human DT, we prepared living slices from 15- to 25-week-old fetal forebrains. We also examined slices preparedfrom macaque monkey brains at comparable fetal ages (embry-onic day 65–85; E65–85) to explore whether previous histo-logical and autoradiographic studies might have overlooked asmall number of cells migrating from the GE to the thalamus inthis species. Crystals of 1,1´-dihexadecyl-3,3,3´-tetramethyl-indocarbocyanine perchlorate (DiI) were placed into the GE,and the slices were cultured for 36 to 48 hours. In slices ofhuman but not monkey brain, labeled unipolar and bipolarcells migrated from the GE to the DT at the level of themediodorsal nucleus and pulvinar (Fig. 2). As in the monkey,we observed no migration from the GE to the thalamus in slicesof embryonic and perinatal mouse brains cut at different planes(data not shown). Whereas the lateral and medial subdivisionsof the human GE are fused at the fetal ages examined, theplacement of DiI into various parts of the GE showed that themedial portion is the major source of cells destined for the thal-amus. After 36 to 48 hours in vitro, cells migrated up to 1.5mm from the site of crystal deposit at an average rate of 28 µmper hour, comparable to the rate of tangential cell migrationto the cortex15 and the olfactory bulb in rodents16. Althoughthe migration from the GE to the thalamus apparently existsonly in the human, GE cells migrated freely to the cerebral cor-tex in both the human and macaque monkey brain (data notshown), indicating the existence of a similar migration pro-gram for other classes of GE neurons.

A combination of immunocytochemistry and electronmicroscopy revealed that the GE–DT migratory stream is com-

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

explants, similar to the organization observed in vivo by elec-tron microscopy (Fig. 3h–j).

Dlx1/2 homeodomain-containing proteins are specificallyexpressed by progenitors and postmitotic neurons in the GE, butnot in the DT17. In agreement with their origin in the GE, neu-rons in the migratory stream extending from the GE to the DTexpress Dlx1/2 proteins (Fig. 4a, b, d and f). The GE is the majorsource of GABAergic neurons for various forebrain structures,such as the basal ganglia and the cortex; the GE neurons migrat-ing to the DT in the human brain express GABA (Fig. 4c, e and g).

Because neuronal migration from the GE to the DT was notobserved in other species, we investigated whether there are dif-ferences in the long-range guidance cues between the human andthe non-human forebrain. We co-cultured (in collagen gel orMatrigel) mouse or human GE explants with explants or slicesisolated from different forebrain structures (both types of gelsgave similar results; Fig. 5). When E14–18 mouse GE explantswere cultured at a distance from the DT region of diencephalicslices, cell migration from explants was symmetrical (37/37explants), indicating that this structure had no repellent effect onmigrating GE cells (Fig. 5c). However, when mouse GE explantswere cultured at a distance from the subthalamic (ST) region,migrating GE cells were repelled, resulting in more cells in thequadrant distal to the subthalamus than in the quadrant proxi-mal to it (44/50 explants; Fig. 5d). In contrast, when the humanGE explants were co-cultured with the DT explants, most cellsmigrating from the explants were directed toward the dorsal thal-amus (51/57 explants; Fig. 5e). However, a symmetrical migra-tion of cells was observed around human GE explants whencultured with control tissue, in this case, the cortex (23/23explants; data not shown). This finding indicated a chemoattrac-tant effect of the human DT on migrating GE cells. The ST region

posed of closely associated, immature bipolar neurons,immunoreactive for TuJ1 (an early neuronal marker) and PSA-NCAM (the polysialated form of the neural cell adhesionmolecule), that are connected by a zonula-adherens type ofmembrane specialization (Fig. 3a–c). The close relationshipamong the migrating neurons as well as the absence of radialglial or axonal tracts indicates their reliance onhomotypic–neurophilic interactions during migration. In addi-tion, we found that distinct tubular formations, composed ofastroglial cell processes immunoreactive for glial fibrilar acidicprotein (GFAP), enveloped and isolated streams of migratingneurons from the surrounding parenchyma. The extracellularmatrix protein tenascin, expressed by glial cells, is densely dis-tributed along the migration pathway, suggesting an involve-ment of these non-neuronal cells in the guidance of migratingGE neurons (Fig. 3f). To directly support the hypothesis thatmigrating neurons rely onhomotypic–neurophilic guid-ance, we cultured explants of theDT surface region in Matrigel.We observed chains of migratingneurons coming out of the

Fig. 2. Cell migration from the GE tothe DT in slice cultures of human andmonkey brains. (a) A coronal slice pre-pared from a 20-week-old fetal brain,showing the migration of DiI labeledcells (arrows) to the pulvinar of theDT. Inset illustrates the orientation ofslices shown in (a) through (c) (d, dor-sal; v, ventral; m, medial; l, lateral). Lightgray, GE; dark gray, DT; asterisk, site ofDiI crystal. (b) Another coronal sliceshowing the migration of cells (arrows)from a DiI crystal placed in the GE tothe mediodorsal nucleus. (c) A coronalslice prepared from an E85 fetal mon-key forebrain, with a DiI crystal placedin the GE. No cell migration to the DToccurred (asterisk).

Fig. 1. Localization of the GE–DT migratory stream in the fetal humanforebrain. (a) Nissl-stained coronal section through a 20-week-oldhuman forebrain shows the localization of major telencephalic struc-tures. C, cortex; BG, basal ganglia. (b) Diagram of section in (a) illustrat-ing the GE–DT migratory stream (arrows). (c) Low magnification imageof migratory stream in a Nissl-stained section (arrows).

a b

c

a

b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


GABAergic neurons from the telencephalon to the diencephalonin the human forebrain, which has not been observed in othermammalian species2–4,8–12. Here we further confirm its absenceduring development in the mouse and macaque monkey brain.It should be noted that GABAergic cells are lacking in dorsalthalamic nuclei in rodents, except in lateral geniculate nucle-us20. However, in the primate brain, GABAergic cells formapproximately 30% of the neurons in every thalamic nucleus21.Therefore, the population of telencephalon-derived dorsal thal-amic neurons in the human brain may serve to specificallyincrease the population of GABAergic neurons in the large asso-ciation nuclei such as the mediodorsal and pulvinar nuclei.Whether the telencephalon also contributes some projectionneurons to these thalamic nuclei remains to be determined.

The homotypic–neurophilic guidance of neurons from theGE to the human DT is reminiscent of the chain migration ofneurons from the subventricular zone to the olfactory bulb inadult rodents22. Thus, the expansion of the DT in the humanbrain may be achieved by a developmental mechanism alsooperating in the olfactory bulb of non-primate mammals.Because this mode of migration critically depends on the pres-ence of PSA-NCAM23, any disturbance of this molecule couldaffect the migration from the GE to the human thalamus. Bothdecreased levels of PSA-NCAM expression24 and a reductionin cell number in some association thalamic nuclei25 have beenreported in the brains of schizophrenia patients.

Our data indicate that in the mouse, repulsive activities ofthe CP and the ST region together inhibit the migration of GEcells to the DT. Furthermore, they also indicate that in thehuman brain, the chemoattractant influence of the DT and theabsence of repellent influence of the CP on the migrating GEcells are evolutionary adaptations that allow the migration ofthe GE neurons into the DT. These adaptations could involvespecific molecules such as Slit, which regulates neuronal migra-tion in the mammalian brain26,27.

in the human did not have a repulsive activity on the migratingGE neurons (data not shown). The choroid plexus (CP) in themouse repels migrating cortical neurons18, so we reasoned that ifCP had a similar effect on the GE cells, it could prevent theirmigration to the diencephalon, particularly because it is attachedto the DT surface, where cells could enter the area. Co-culturingexperiments indicated that mouse CP indeed repelled GE cells(47/50 explants; Fig. 5f), whereas human CP did not have anyeffect (59/62 explants; Fig. 5g). In control experiments in human,we confirmed that GE cells respond to the repulsive activity ofanother forebrain structure, the septum (18/20 explants; Fig. 5h),which is also repellent for these cells in the rodent brain19.

To determine whether mouse DT exhibits a short-range,contact-dependent inhibitory signal that would prevent migra-tion of GE cells, we directly apposed slices of the GE to the DTor ST region of the diencephalic slices and labeled migratingGE cells using DiI crystals. Migrating GE cells crossed the inter-face and dispersed into the mouse DT, as they did in the humanassays, but did not migrate into the ST (Fig. 6b–e). Further-more, when small GE explants, labeled with PKH-26, weretransplanted over the DT region of the diencephalic slices, neurons migrated extensively onto the slice, but remained confined to the GE explants when placed over the ST region(Fig. 6f–h). In addition, more than 80% of dissociated mouseGE cells extended a migratory process when transplanted overthe DT (Fig. 6i), indicating that this structure allows migra-tion of most GE cells.

DISCUSSIONThe present study provides direct evidence for migration of

Fig. 3. Neurophilic–homotypicmechanism of guidance of neuronsmigrating from the GE to the DT. (a) Cross-section through migratingcells, illustrating their tight apposi-tion; zonula-adherens-like contactsoccur between membranes ofneighboring cells (inset). (b, c)Electron micrographs showing Tuj1(b) and PSA-NCAM immunoreactiv-ity (c) in migrating neurons. (d) Anelectron micrograph illustratingastroglial cells that have cellularprocesses that ensheathe migratingneurons (cross-section). Neuronalcell somas and processes weretraced in green; glia were traced inred. (e) GFAP immunostaining ofglial tubes (arrow). Asterisks markspaces occupied by migrating neurons (cross-section). (f) Laserconfocal image showing tenascinimmunoreactivity (red, arrows)closely surrounding the migratoryneuronal stream that is stained forTuj1 (green, asterisks). (g) Summarydiagrams of a cross section (top), ora tangential section (bottom)through the migratory stream. Gliaare depicted in red; neurons, ingreen. (h–j) Explants prepared fromthe surface region of the human DT give rise to migrating chains of neu-rons in Matrigel. High-power DIC image (h), Tuj1 immunostaining (i)and electronmicrograph (j) of neuronal chains.

a

b

c

d

e f

h i j

g

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

We suggest that the telencephalic neurons contribute to theexpansion of the association thalamic nuclei that are anatomical-ly related to the association cortex involved in higher cognitivefunctions such as symbolic reasoning or language28. During phy-logenesis, brain structures with majoranatomical and functional links haveapparently evolved together, but how

Fig. 5. Regulation of cell migration acrossthe boundary between the GE and DT bylong-range attractive and repulsive guid-ance cues. (a) Mouse diagram. Red andgreen color labels indicate the lower edgeof the ST and the lateral edge of the DTto which mouse GE explants wereapposed. CP, orange. (b) Human diagram.GE explants were apposed to explantsdissected from DT, ST or CP (labeledgreen, red and orange, respectively). (c) Symmetric distribution of migratingGE cells when co-cultured at a distancefrom the mouse DT. (d) Asymmetric dis-tribution of migrating GE cells because ofa repulsive effect of mouse ST. (e) Asymmetric distribution of migratingGE cells with most cells migrating towardthe explant of the human DT. (f) Repulsive effect of the mouse CP onmigrating GE cells. (g) Symmetric distrib-ution of migrating GE cells when co-cul-tured with the human CP. (h) Repellenteffect of the septum (S) on migrating GEcells in the human.

the size of participating structures is coordinated is not under-stood. Our results indicate that, to accommodate the input fromthe expanding neocortex, an additional complement of neurons isimported to the thalamic association nuclei during a protractedperiod of fetal development (from 15 to approximately 34 weeks),when the GE of the telecephalon is the only proliferative centerstill active in this region.

METHODSImmunohistochemistry and electron microscopy. Post mortem humanbrain tissue (20 specimens ranging from 15 to 26 weeks gestation) wasobtained from the Albert Einstein College of Medicine, New York Statelicensed Human Fetal Repository, and was approved by the institutionalboard. Research protocols used were approved by the Human Investiga-tion Committee at Yale University School of Medicine. The tissue wasfixed in 4% paraformaldehyde and 0.1% glutaraldehyde in PBS and eithercryoprotected and cut on a cryostat at 10–30 µm for light microscopy,or cut on a vibratome at 50 µm for electron microscopic examination.The sections were blocked and incubated 24 h at 4°C with primary anti-bodies and then overnight at 4°C with biotin-, CY2- or CY3-conjugat-ed secondary antibodies (Jackson, West Grove, Pennsylvania). Primaryantibodies and dilutions were as follows: anti-meningococcus B mono-clonal IgM, recognizing PSA-N-CAM, 1:2000; TUJ1 monoclonal, rec-ognizing neuron-specific class III beta tubulin (Becton–Dickinson,Franklin Lakes, New Jersey), 1:500; anti-GFAP polyclonal (Sigma, St.Louis, Missouri), 1:200; anti-Tenascin monoclonal (Sigma), 1:4000; anti-DLX 1 and anti-DLX2 polyclonals (donated by J.L.R. Rubenstein, UCSF),1:20 and 1:200, respectively; and anti-GABA polyclonal (Sigma), 1:10,000.Some tissue blocks were fixed in 2% paraformaldehyde and 2.5% glu-taraldehyde, osmicated in 1% osmium tetroxide in a phosphate buffer,and embedded in Durcupan, Fluka Chemie (Buch, Switzerland). Select-ed blocks were cut serially into ultrathin sections on Leica/Reichart-YungUltramicrotome (Bannockburn, Illinois), stained with lead citrate and

a

b

c d

e f

g h

a

b c

d e

f g

Fig. 4. Neurons migrating from the GE to the DT are GABAergic. (a) Dlx2 immunoreac-tivity in cells of the GE–DT migratory stream (arrows). (b, d and f) Double staining for TuJ1(b) and Dlx2 (d) of the migratory stream; (f) is a double exposure. (c, e and g) Doublestaining for TuJ1 (c) and GABA (e) of the migratory stream; (g) is a double exposure.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


uranyl acetate, and examined with a JEOL transmission electron micro-scope (Peabody, Massachusetts). Vibratome sections (50 µm) wereprocessed for immunoperoxidase histochemistry, osmicated, embeddedflat in Durcupan, and processed as described above.

Organotypic slice cultures and explant cultures. Human, monkey andmouse fetal brains were put in HEPES-buffered MEM (Gibco, Life Tech-nology, Gaithersburg, Maryland) with L-glutamine (Gibco) on ice(timed-pregnant CD1 mice were obtained from Charles River (Wilm-ington, Massachusetts); macaque monkey embryos at E65 and E85 wereobtained by Cesarean sections according to the procedure approved bythe Institutional Animal Care and Use Committee at Yale UniversitySchool of Medicine). We used an embryonic organotypic slice proce-dure, which reproduces both the timing and mode of cell proliferationand migration in the cerebral tissue for up to at least 72 h29. In short,dissected blocks of cerebral hemispheres were cut using a McIlwain tis-sue chopper (Mickle Lab Engineering, Surrey, UK) into 300–350-µm-thick coronal or horizontal slices and transferred with a paintbrush todishes containing HEPES-buffered MEM, a slices procedure that pre-serves cell production and migration for up to 72 hours29. Slices wereseparated using dissecting needles or forceps and transferred onto theinside of Millicell tissue culture plate inserts (Millipore, Bedford, Mass-achusetts), placed into 35-mm petri dishes and incubated in Neurobasalmedium with L-glutamine, B27, N2 and antibiotics (Gibco) at 37°C with5% CO2 Gibco, Invitrogen (San Diego, California). DiI crystals (Mole-cular Probes, Eugene, Oregon) were placed using insect pins under adissection microscope. Small explants of human thalamus were isolatedand embedded in Matrigel, Becton–Dickinson (Franklin Lake, New Jer-sey). After incubation for 24 to 48 h, slices and explants were fixed in4% paraformaldehyde and slices were mounted on glass slides.

Neuronal migration assays. Brain slices 300–400 µm thick were cuton a tissue chopper, and desired tissues were isolated from the rest ofthe brain under a dissection microscope. In apposition experiments,isolated thalamic and GE slices were directly apposed to one anotherand cultured on Millicell (Millipore) inserts in 35-mm petri dishesafter, in each experiment, a crystal of DiI was put in the GE. In trans-plantation experiments, isolated GE tissue was further trimmed intosmaller explants (100–200 µm), labeled vitally with PKH-25 (Sigma)and placed over the thalamic slices. In a similar way, PKH-26-labeled,

Fig. 6. Mouse DT is permissive for the migration of neurons from the GE. (a) The left portion of the schematic diagram of E14 mouse forebrain illus-trates apposed edges (blue, red and green) of the GE and ST/DT regions of diencephalic slices. Migrating cells were labeled with DiI crystals placed inthe GE. Right side of diagram, PKH-26-labeled GE mini-explants (blue), transplanted over the ST or DT. Dashed lines in (b) to (d) represent GE–thal-amic border. Migrating cells freely cross from the GE into the DT in both human (b) and mouse (c). (d) Cell migration is restricted to the GE anddoes not cross into the mouse ST. (e) Higher magnification image of (d) showing labeled GE cells near the border with the ST, aligned parallel to theborder or directing their migratory processes away from it. In (f) to (h), GE explants are delineated by dashed lines. (f) GE transplant immediatelyafter the transplantation. (g) Cells disperse from a mouse GE explant onto the DT. (h) Cells do not disperse from mouse GE explants into the ST. (i) Dissociated mouse GE cells transplanted onto the DT extend a migratory process.

dissociated GE cells were placed over the thalamic slices. In explantco-culture experiments, small tissue explants (200–400 µm) or wholediencephalic slices were embedded in Matrigel (Collaborative Bio-medical Products, BD Micosciences, Bedford, Massachusetts) or incollagen gel (Vitrogen, Cohesion, Palo Alto, California), in 35-mmglass-bottom microwell dishes (MatTek Corporation, Eugene, Oregon). After congealing, the gel with explants was overlaid with aNeurobasal medium (containing N2, B27, L-glutamine and antibiotics)and incubated for 24 h. In all experiments, the tissue was fixed in 4%paraformaldehyde and analyzed by laser confocal microscopy.

ACKNOWLEDGEMENTSWe thank J.L.R. Rubenstein and S. Anderson for providing Dlx1/2 antibodies.

We thank all colleagues in P. Rakic’s laboratory for their advice and comments

on the manuscript.


1. Barton, A. R. & Harvey, P. H. Mosaic evolution of brain structure inmammals. Nature 405, 1055–1058 (2000).

2. Rakic, P. Principles of neuronal cell migration. Experientia 46, 882–891(1990).

3. Hatten, M. E. Central nervous system neuronal migration. Annu. Rev.Neurosci. 22, 511–539 (1999).

4. Wilson, S. W. & Rubenstein, J. L. R. Induction and dorsoventral patterning ofthe telencephalon. Neuron 28, 641–651 (2000).

5. Sidman, R. L. & Rakic, P. Neuronal migration, with special reference todeveloping human brain: a review. Brain Res. 62, 1–35 (1973).

6. Rakic, P. & Sidman, R. L. Telencephalic origin of pulvinar neurons in the fetalhuman brain. Z. Anat. Entwicklungsgesch. l29, 53–82 (1969).

7. Letinic, K. & Kostovic, I. Transient fetal structure, the gangliothalamic body,connects telencephalic germinal zone with all thalamic regions in thedeveloping human brain. J. Comp. Neurol. 384, 373–395 (1997).

8. Anderson, S. A., Eisenstat, D. D., Shi, L. & Rubenstein, J. L. R. Interneuronmigration from basal forebrain to neocortex: dependence on Dlx genes.Science 278, 474–476 (1997).

9. Anderson, S., Mione, M., Yun, K. & Rubenstein, J. L. R. Differential origins ofneocortical projection and local circuit neurons: role of Dlx Genes inneocortical interneuronogenesis. Cereb. Cortex 9, 646–654 (1999).

10. Lavdas, A. A., Grigoriou, M., Pachnis, V. & Parnavelas, J. G. The medialganglionic eminence gives rise to a population of early neurons in thedeveloping cerebral cortex. J. Neurosci. 19, 7881–7888 (1999).

j

a

a

b c d e

f g h i

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

11. Parnavelas, J. G. Origin and migration of cortical neurons: new vistas. TrendsNeurosci. 23, 126–131 (2000).

12. Wichterle, H., Garcia-Verdugo, J. M., Herrera, D. G. & Alvarez-Buylla, A.Young neurons from medial ganglionic eminence disperse in adult andembryonic brain. Nat. Neurosci. 2, 461–466 (1999).

13. Rodriguez, C. I. & Demecki, S. M. Origin of the precerebellar system. Neuron27, 475–486 (2000).

14. Rakic, P. Illegal immigrations. Neuron 27, 409–410 (2000).15. O’Rourke, N. A., Chenn, A. & McConnell, S. K. Postmitotic neurons migrate

tangentially in the cortical ventricular zone. Development 124, 997–1005(1997).

16. Lois, C. & Alvarez-Buylla, A. Long-distance neuronal migration in the adultmammalian brain. Science 264, 1145–1148 (1994).

17. Porteus, M. H. et al. DLX-2, MASH-1, and MAP-2 expression andbromodeoxyuridine incorporation define molecularly distinct cell populationsin the embryonic mouse forebrain. J. Neurosci. 14, 6370–6383 (1994).

18. Hu, H. Chemorepulsion of neuronal migration by Slit2 in developingmammalian forebrain. Neuron 23, 703–711 (1999).

19. Hu, H. Y. & Rutishauser U. A septum-derived chemorepulsive factor formigrating olfactory interneuron precursors. Neuron 16, 933–940 (1996).

20. Harris, R. M. & Hendrickson A. E. Local circuit neurons in the ratventrobasal complex—a GABA immunocytochemical study. Neuroscience 21,229–236 (1978).

21. Montero, V. M. & Zempel, J. The proportion and size of GABA-

immunoreactive neurons in the magnocellular and parvocellular layers ofthe lateral geniculate nucleus of the rhesus monkey. Exp. Brain Res. 62,215–223 (1986).

22. Lois, C., Garcia-Verdugo, J. M. & Alvarez-Buylla, A. Chain migration ofneuronal precursors. Science 271, 978–981 (1996).

23. Hu, H., Tomasiewicz, H., Magnuson, T. & Rutishauser, U. The role ofpolysialic acid in migration of olfactory bulb interneuron precursors in thesubventricular zone. Neuron 16, 735–743 (1996).

24. Barbeau, D., Liang, J. L., Robitalille, Y., Quirion, R. & Srivastava, L. K.Decreased expression of the embryonic form of the neural cell adhesionmolecule in schizophrenic brains. Proc. Natl. Acad. Sci. USA 92, 2785–2789(1995).

25. Jones, E. G. Cortical development and thalamic pathology in schizophrenia.Schizophr. Bull. 23, 483–501 (1997).

26. Zhu, Y., Li, H.-S., Zhou, L., Wu, J. Y. & Rao, Y. Cellular and molecularguidance of GABAergic neuronal migration from an extracortical origin tothe neocortex. Neuron 23, 473–485 (1999).

27. Wu, W. et al. Directional guidance of neuronal migration in the olfactorysystem by the protein Slit. Nature 400, 331–336 (1999).

28. Goldman-Rakic, P. S. Topography of cognition: parallel distributed networksin primate association cortex. Annu. Rev. Neurosci. 11, 137–156 (1988).

29. Haydar, T. F. Bambrick, L. L., Kruger, B. K. & Rakic, P. Organotypic slicecultures for analysis of proliferation, cell death, and migration in theembryonic neocortex. Brain Res. Brain Res. Protoc. 4, 425–437 (1999).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Inferotemporal (IT) cortex is involved in visual shape repre-sentation and visual object recognition, based on evidence fromsingle-cell recording1–9, functional imaging10,11 and lesionstudies9,12. In comparison with earlier visual areas, cells in IThave larger receptive fields and show more abstract preferencesfor complex shape properties2,4,6,9,13, but exactly how thisregion represents shape remains controversial3–5,13. Here weexamined shape representation within IT in relation to fig-ure–ground reversal, as well as other stimulus manipulationsthat served as control comparisons.

The figure–ground assignment of a given visual display candramatically alter the shape that human observers perceive (exam-ples, top of Fig. 1). Adjacent figure and ground regions definedby a common contour are perceived as very different. Humanobservers typically recognize the figure later (for example, the facein the top row of Fig. 1), but not the ground (white shape in thatrow), even for judgments based on exactly the same shared con-tour14–18. Moreover, they rate a mirror image of the figure as moresimilar19 to the original figure–ground display than an image ofthe ground in isolation. This arises even though the ground probeshares exactly the same curved contour as in the originally exposeddisplay, whereas the mirror image of the figure has a mirror-reversed contour. These phenomena also arise for shapes madeby unfamiliar contours15–20 (see below), not just for profiles ofmeaningful shapes. Such effects reveal the influence of one-sidededge assignment on visual shape perception in humans15–20.

Here we examined how the shape preferences of IT cells inthe primate brain may relate to these psychological phenomena.Specifically, we tested how the preferences of individual IT cellsfor stimuli drawn from a population of pseudorandom two-dimensional shapes would generalize across three different trans-formations: figure–ground reversal, reversal of contrast-polarityand mirror-image reflection about the vertical (Fig. 1a–h). Allshapes were polygons with straight edges at the top, bottom andalong one side, and with a pseudo-randomly curved contour on

Shape-coding in IT cells generalizesover contrast and mirror reversal,but not figure-ground reversal

Gordon C. Baylis1,2 and Jon Driver3

1 University of Plymouth, Plymouth Institute of Neuroscience, 12 Kirkby Place, Plymouth, PL4 8AA, UK2 Department of Psychology, University of South Carolina, Columbia, South Carolina 29208, USA3 Institute of Cognitive Neuroscience, University College London, 17 Queen Square, London WC1N 3AR, UK

Correspondence should be addressed to G.C.B. ([email protected]) or J.D. ([email protected])

We assessed how the visual shape preferences of neurons in the inferior temporal cortex of awake,behaving monkeys generalized across three different stimulus transformations. Stimulus-preferences of particular cells among different polygon displays were correlated across reversedcontrast polarity or mirror reversal, but not across figure–ground reversal. This corresponds withpsychological findings on human shape judgments. Our results imply that neurons in inferior tempo-ral cortex respond to components of visual shape derived only after figure–ground assignment ofcontours, not to the contours themselves.

the other side15–20. The curved contours of possible polygons dif-fered in their identity and location (left or right side of polygon).

It is possible that any selectivity in the responses of IT cells tothese stimuli is determined just by these physical differencesamong the displays. Alternatively, IT responses might show pat-terns that are more like shape judgments in human observers,where figure and ground regions are perceived to have very dif-ferent shapes despite their common defining contour, with themirror image of any figure being perceived as more similar tothat figure than its ground (as confirmed for the present stimulialso; see below). For the displays used here, exactly the samecurved contour was present across a reversal of figure–groundassignment (Fig. 1, compare a to e, c to g, b to f, and d to h), yetthis contour produces shapes that look very different to humanobservers when figure and ground are reversed14–20.

The curved contour was necessarily on opposite sides of thefigure region versus the adjacent ground region within any display(stimuli a–h, Fig. 1). Our further manipulation of mirror-imag-ing (see also ref. 6) controlled for this, as the curved contour ofany mirror image of an original figure is on the same side as thecurved contour of the original ground (stimuli a–h, Fig. 1). Thefigure and ground region of each individual display also differedin contrast polarity (one white, the other black). Our orthogo-nal manipulation of reversing contrast-polarity (see also ref. 8)controlled for this, as a contrast-reversal of an original figure hasthe same polarity as the original ground (stimuli a–h, Fig. 1).

We recorded activity from IT cells in monkeys to determinetheir firing rates for the different stimuli, and to determine howthese rates correlated across the three different stimulus trans-forms. We also required human observers to rate the similarityof the displays across the same transforms.

RESULTSWe recorded from 88 cells in areas TEa, TEm and TE3 (ref. 20)of 2 awake monkeys while they viewed displays drawn from 32

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


32 stimuli (Fig. 3; the same nomenclature is used for these stim-uli as in Fig. 1). The pattern of firing rates was similar across thecontrast-reversal and mirror-image transforms, but not acrossthe figure–ground reversal. (Compare the four appropriate pair-ings of graphs, each with four bars, across each of these trans-forms, Fig. 3.) Peri-stimulus time histograms for this cell inresponse to the 16 stimuli generated from types 1 and 2 showthat responses were similar across mirror-image and especiallycontrast-reversal transforms, but that they differed markedlyacross the figure–ground transform (Fig. 4). For instance, type2 received a much more vigorous response than type 1 in versionb, but the opposite ordering applied for version f (thefigure–ground transform of version b).

In histograms of the distributions of correlations for all cellsin the population across the three transformations, most cellsshowed significant positive correlations in shape preference acrossreversed contrast polarity in the display (mean correlation coef-

Fig. 1. Example stimuli. Top, classic figure–ground display, togetherwith its components. Humans rate a mirror image of the figure as moresimilar to the original figure–ground display than the original ground inisolation. Stimuli a–f, Visual displays for the single-cell recording exper-iment, showing how 8 different displays were generated from one par-ticular curved contour (type 1 is shown). Bottom right, 3 additionaltypes of curved contours (2–4); each of these analogously generated a8 different displays (2a–h, 3a–h, 4a–h). Three aspects of the displayswere manipulated orthogonally, in a 2 × 2 × 2 fashion illustrated by thelayout of shapes 1a–h, which shows one 2 × 2 table of possible displaysin the ‘front’ plane (b, d, f, h), with another 2 × 2 table of possible dis-plays in the ‘back’ plane (a, c, e, g). All displays comprised either a whitefilled polygon on a black background, or vice versa. The differencebetween examples in the front plane and the back plane in the illustra-tion depicts this contrast-polarity transform for otherwise equivalentdisplays. Each display also appeared in mirror image form. The differ-ence between examples in adjacent columns for each of the 2 × 2tables in the schematic illustrates this mirror-reversal transform.Finally, a given curved contour could have the figural region (as definedby surroundedness14,17) on its left or right, leading to thefigure–ground transform between examples in adjacent rows for each2 × 2 table in the schematic. Only the figure–ground transform leavesthe curved contour entirely unchanged.


possibilities (stimuli a–h, Fig. 1, equivalent transforms wereimplemented for types 2, 3 and 4, thus yielding 8 × 4 = 32 stim-uli in total). Eighty-nine percent of cells (78/88) showed signifi-cant differences in mean evoked firing rate in the interval 100 to600 ms after stimulus onset, as a function of which of the 32 pos-sible stimuli were shown (at p < 0.01 or better). We then exam-ined how the shape preferences revealed by these differentialevoked firing rates correlated across the three transformationswe had applied to the stimuli. Most cells showed significant andsubstantial correlations in stimulus preferences across reversalsof contrast polarity and across mirror imaging, but not acrossfigure–ground reversal.

We first show the correlations across these three transformsfor one illustrative neuron (Fig. 2). All 32 stimuli contributed toeach of these correlations, but stimuli were paired differently foreach correlation (see Methods). We plot the mean firing rates ofthis cell (in the 100–600 ms interval after stimulus onset) for all

articles

0

1 0

2 0

3 0

4 0

0 1 0 2 0 3 0 4 00

1 0

2 0

3 0

4 0

0 1 0 2 0 3 0 4 00

1 0

2 0

3 0

4 0

0 1 0 2 0 3 0 4 0

Contrast reversal Mirror reversal Figure-ground reversal

Response (spikes/s)

Res

pons

e (s

pike

s/s)

Res

pons

e (s

pike

s/s)

Res

pons

e (s

pike

s/s)

Response (spikes/s) Response (spikes/s)

Fig. 2. Correlation plots for a single cell. Plots show mean firing rates (in the period 100–600 ms after the stimulus) for an illustrative neuron, for par-ticular stimuli along the x-axis, and for transformed versions of the same stimuli along the y-axis. (a) Contrast reversal transform. (b) Mirror reversal.(c) Figure–ground reversal. The total set of 32 stimuli all contribute to each plot. For each plot, this set was divided into 2 subsets of 16, with eachmember of one subset providing a transformed version for one member of the other subset. (a, b) Stimuli that induced a particular firing rate led toa similar rate when transformed (correlations of 0.92 and 0.68 respectively, for this particular cell). No such relationship is apparent in (c) across thefigure–ground transform (R = 0.0).

a b c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


For figure–ground reversals, the correlation coefficient averagedvery close to zero throughout the trial.

Finally, we examined how the average firing rates changedduring the trial for preferred versus non-preferred stimuli, andhow this generalized across the three transformations. To do this,we first determined for each cell which of the 32 stimuli producedthe maximal mean firing rate in a 100-600 ms time bin after stim-ulus onset; this defined the preferred stimulus for that cell. Wealso identified the non-preferred stimulus for each cell, produc-ing the lowest mean firing rate in the same time window. Weshow the mean firing rates across all cells for their preferred ver-sus non-preferred stimuli, at different times after stimulus onset(Fig. 6a). Firing rates for contrast-polarity and mirror-imagereversals of these stimuli show how the preference was largelymaintained across both these transformations (Fig. 6b and c). Incontrast, the preference disappeared across the figure–groundtransformation, consistent with our other findings (Fig. 6d). Weconfirmed the site of cellular recordings by histology (Fig. 7).

In a matching task (see Methods) on the same shapes as used inour physiological work, 12 human observers selected the untrans-formed figure on 87.5% of trials, the contrast-reversed version on68.3% trials and the mirror-reversed version on 54.2% of trials.The latter two transforms were each selected significantly moreoften (p < 0.01) than the figure–ground reversal (only 19.7%).Moreover, the figure–ground reversal was not selected any moreoften than a shape with an entirely different contour (20.3%), andmost selections for either of these two types arose when they werethe only two alternatives presented (see Methods). These data con-

Fig. 4. Peri-stimulus time histograms of firing, using 20 ms bins, for theillustrative cell. Response to the 8 variants of type 1 and type 2 shapes.(Detailed responses to 16 stimuli are shown here, rather than all 32; type3 and type 4 data (Fig. 3) are omitted for brevity.) 2 × 2 × 2 layout andnomenclature for the stimuli are as in Figs. 2 and 3. The y-axis brace rep-resents a firing rate of 100 spikes/s; the bar on the x-axis represents thefirst 500 ms of stimulus presentation time.

Fig. 3. Firing rates in response to the 32 stimuli for a singlecell. Histograms show mean firing rate in the 100–600 msperiod after stimulus onset for the illustrative cell from Fig.2, now shown for each individual stimulus. The layout in theillustration has the same 2 × 2 × 2 arrangement as in theschematic in the center of Fig. 1, and uses the same nomen-clature for the 32 different stimuli (variants a–h on contourtypes 1–4). Hence, comparing laterally adjacent pairs of his-tograms addresses the mirror image transform of the stim-uli; comparing histograms between the apparent ‘front’ and‘back’ plane addresses contrast reversal; vertically adjacenthistograms represent a figure–ground reversal. The patternseen within each of the paired histograms stays similaracross both the contrast and mirror-image transforms, butnot across the figure–ground reversals, hence the correla-tions in Fig. 2 for the same cell.

ficient, R = 0.59), and likewise across mirror-imagereflection of the presented shape (mean R = 0.46; Fig. 5). In contrast, correlation coefficients for fig-ure–ground reversal were typically low, and centeredaround zero (mean R = 0.04). Chi-square tests, com-paring the number of cells showing significant corre-lations across the different transforms, found manymore such correlations for contrast versusfigure–ground reversal (χ2

1 = 80.4, p < 0.0001), andfor mirror-image versus figure–ground reversal (χ2

1 = 44.96, p < 0.0001). In addition, correlationswere somewhat more pronounced for contrast than mirror imagereversal (χ2

1 = 9.44, p < 0.05), in accord with the human simi-larity ratings reported below.

The poor generalization across figure–ground reversal wasfound equivalently for cells that showed a significant correlationacross both contrast and mirror-image reversal (black, bottomhistogram of Fig. 5) and those that did not (white, Fig. 5); thesedistributions for figure–ground reversal did not differ. We alsoassessed how the correlation coefficients developed as a functionof time after stimulus onset. For contrast-polarity and mirror-image transformations, the average coefficients climbed rapidly,reaching asymptote at around 200–300 ms post stimulus onset.

articles


spik

es /

s

0

1 0

2 0

3 0

4 0

1a 2a 3a 4a0

1 0

2 0

3 0

4 0

1c 2c 3c 4c

0

1 0

2 0

3 0

4 0

1e 2e 3e 4e0

1 0

2 0

3 0

4 0

1g 2g 3g 4g

0

1 0

2 0

3 0

4 0

1d 2d 3d 4d0

1 0

2 0

3 0

4 0

1b 2b 3b 4b

0

1 0

2 0

3 0

4 0

1 f 2 f 3 f 4 f0

1 0

2 0

3 0

4 0

1h 2h 3h 4h

The pattern made by all fourbars in each graph remains thesame across this transform.

The patternmade by allfour bars ineach graphchangesacross thistransform.

The pattern made by all four bars in eachgraph remains similar across this transform.

Mirror reversal

Figure-groundreversal

Contrast reversal

spik

es /

ssp

ikes

/ s

spik

es /

ssp

ikes

/ s

spik

es /

s

spik

es /

ssp

ikes

/ s

Contrastreversal

Mirror reversal

Fig

ure–

grou

nd r

ever

sal

1a

2a

1c

2c

1e

2e

1g

2g

1b

2b

1d

2d

1f

2f

1h

2h

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 6. Population response to preferred versus non-preferred stimuli across the three stimulus transforms.Mean firing rate across the population of neurons, forsuccessive 20-ms time bins, with standard errors. (a) Responses to the preferred (blue) and non-preferred(red) stimulus selected for each neuron. (b) Responsesto the contrast-reversed versions of each neuron’s pre-ferred versus non-preferred stimuli, showing that thepreference is maintained. (c) Responses to the mirrorimaged versions of each neuron’s preferred versus non-preferred stimuli, again showing that the preference isstill maintained. (d) Responses to the figure–groundreversed versions of each neuron’s preferred versusnon-preferred stimuli, showing that the preference isnow abolished.


firm that human perception of shape generalizes well across con-trast reversal, and fairly well across mirror reversal. In contrast,figure–ground reversal alters shape perception as much as the gen-eration of a new shape from an entirely different contour. Ourfindings for the shape preferences of IT cells in the primate brainclosely parallel these aspects of human perception.

DISCUSSIONThe shape preferences of IT cells generalized wellacross contrast reversals of the stimuli and acrossmirror imaging of the stimuli, but not across fig-ure–ground reversals. The two-dimensional poly-gons we used varied in their particular curvedcontour. The other lines (three straight edges) wereheld constant in the stimulus sets assessed for eachtransformation, as we tested for generalization acrosscontrast reversal, mirror imaging or figure–groundreversal. These three transformations have very dif-

ferent influences on the critical curved contour. Contrast reversalchanges the polarity of this critical contour. Mirror reversal reflectsthis contour about the vertical. Only figure–ground reversal leavesthe critical curved contour itself unchanged. (Its relative positionwith respect to the body of the shape changes, but this is appliedequally to the mirror-reversal transform.) Thus, if the selectiveresponses of IT cells had been caused primarily by just the curvedcontour that distinguished the various displays physically, then weshould have found maximum generalization across figure–groundreversal, as only this keeps the curved contour constant. Howev-er, the opposite result was found, with generalization absent onlyfor the figure–ground transform.

This demonstrates that the selectivity of IT responses is notdetermined simply by the distinctive contours in a display, con-trary to simple edge-based models of shape recognition discussedelsewhere5,22. Instead, coding in IT follows similar principles tothat observed for human shape judgments. Human observersrate the mirror image of an original figure as more similar to theoriginal display than the original ground18,19, as confirmed herefor the displays used in our physiological work. This arises eventhough the ground shares the same informative contour as theoriginal figure, and hence has the ‘profile’ of the original figureembedded in it as background. We found here that IT cells like-wise generalized more strongly across mirror imaging than acrossfigure–ground reversal. Our findings for mirror imaging are con-

articles

Contrast reversal

Mirror reversal

Figure-ground reversal

0

5

1 0

1 5

0 0.2 0.4 0.6 0.8–0.2–0.4 1.0

0

5

1 0

1 5

0 0.2 0.4 0.6 0.8–0.2–0.4 1.0

0

5

1 0

1 5

0 0.2 0.4 0.6 0.8–0.2–0.4 1.0

Num

ber

of c

ells

Spearman correlation coefficient



Num

ber

of c

ells

Num

ber

of c

ells

Fig. 5. Correlations of ranked stimulus preferences for each of the threetransforms in the cell population. Histograms show the population distri-butions of Spearman rank-order correlations in firing rate (for 100–600 ms following stimulus onset) between transformed versions ofthe stimuli. Each bar indicates the number of cells from the populationshowing a particular size of correlation. Most cells show reliable positivecorrelations (with 15 degrees of freedom) across the contrast-reversal transform and mirror-reversal transform. Correlations for thefigure–ground transform are much lower overall, averaging near zero.For figure–ground reversal plot, cells that showed significant correlationsfor both contrast and mirror-image reversal are represented in black;those that did not, in white.

0

1 0

2 0

3 0

–200 0 200 400 600

Res

pons

e (s

pike

s /

seco

nd)

Time relative to stimulus onset (ms)

0

1 0

2 0

3 0

–200 0 200 400 600

Res

pons

e (s

pike

s /

seco

nd)


0

1 0

2 0

3 0

–200 0 200 400 600


Res

pons

e (s

pike

s /

seco

nd)

0

1 0

2 0

3 0

–200 0 200 400 600


Res

pons

e (s

pike

s /

seco

nd)

Untransformed Contrast reversal

Mirror reversal Figure-ground reversal

a b

c d

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


contour. The present study finds no support for the latter viewat the level of IT responses, as the cells did not respond differen-tially to the presence of their preferred ‘profile’ in the backgroundto the current stimulus (for example, Fig. 6d). Instead, our resultsshow that shape description in IT cortex is entirely constrained byone-sided assignment of contours to figural objects.

METHODSAnimals and surgery. The experiment was conducted with two malemacaque monkeys (Macaca fascicularis, 4.8 and 5.8 kg). With asepticsurgery, we placed a recording chamber and inserted a scleral coil in theleft eye. All procedures were approved by the Institutional Animal Careand Use Committee.

Recording techniques. The activity of single neurons was recorded withepoxy-insulated tungsten microelectrodes (FHC, Brunswick, Maine) asthe monkey sat in a primate chair, using standard techniques for single-cell recording7. Action potentials of single cells were amplified using BAKneurophysiological hardware, passed through a dual-window discrimi-nator, with output TTL pulses timed to a resolution of 0.1 ms by thecomputer controlling the experiment. Maintenance of fixation was con-firmed using the scleral search-coil technique27, measuring eye positionwith an accuracy of 30´ every 16 ms. Data were rejected from trials dur-ing which the monkey was not fixating appropriately when the stimulusappeared, or during which eye movements of more than 2° occurred inthe first 600 ms following stimulus onset.

X-radiographs were used to locate the position of the microelectrodeon each recording track relative to bony landmarks. The position ofcells was reconstructed from the X-ray coordinates taken, together withserial 50-µm histological sections showing the micro-lesions made atthe end of some of the microelectrode tracks. Recording sites were alllocated within the lower bank of the superior temporal sulcus and inthe adjacent dorsal part of the inferior temporal gyrus. All recordingsites were localized within cytoarchitectonic areas TEa, TEm and TE3,as described previously21 (Fig. 7).

Stimulus presentation and task. The 32 visual stimuli (Fig. 1) were storeddigitally on a computer disk, and displayed on a Sony video monitorusing a Data Translation video framestore (512 × 480 pixels; Marlboro,Massachusetts). Maximum and minimum luminances on the screen were5.2 and 0.22 footlamberts, respectively. The exposed shape was eitherwhite on a black background, or vice versa (Fig. 1). Each shape averaged2.8° in width and 3.5° in height, with the center of the curved contourlocated centrally at fixation.

Before a trial, the whole screen was gray. The screen then went black orwhite for 1 s, so that the subsequent figural stimulus could later appearagainst this background with the opposite polarity, which produces entire-ly unambiguous figural assignment in human observers. This preliminarychange to the luminance of the whole screen was unrelated to our com-parisons (see also the baseline data, before onset of the experimental stim-ulus, in Fig. 6). After 1 s, a central fixation dot of opposite polarity to therest of the screen appeared for 500 ms. The fixation dot was followed bythe experimental shape for 1 s, then the screen returned to gray for 3.5 s,before the start of the next trial. The monkeys performed a simple visualtask during testing (adapted from ref. 29) to ensure that they fixated thestimuli. If the shape shown centrally was any one of the 32 in the experi-mental set (a black or white shape, Fig. 1), then the monkeys could obtaina fruit juice reward during its exposure, provided they were fixating with-in 1° of the central location. If the central shape was a red square (11% oftrials, excluded from analyses), then the monkey had to withhold licking toavoid ingesting aversive hypertonic saline. A 0.5-s signal buzzer preceded thepresentation of the stimulus. (This sounded concurrently with the centralfixation point.) Thus, if the monkey fixated correctly before the stimulusappeared, he had sufficient time to discriminate black or white experi-mental shapes from the red square, and then obtain fruit juice while it wasstill available (during the central stimulus).

Before the experiment, the monkeys had been trained on a simplevisual discrimination task. They viewed the monitor with a central fix-ation point, and could lick to obtain fruit juice when a white or black

sistent with other single-cell evidence6; the contrast-reversal find-ings also agree with previous studies8,23. The additional com-parison with figure–ground reversal here reveals that the selectiveresponses of IT neurons correspond with psychological obser-vations15–20 on how one-sided assignment of edges to figuresconstrains human perception of contoured shapes, and accordwith human similarity ratings.

A longstanding question in vision research14,15,18–20 is why fig-ures and their abutting grounds are perceived as so different inshape, despite the common contour. One computational proposalis that the visual system may decompose shapes into convexparts16,19,24. A convexity in the outline of a figure (for example, the‘nose’ in the face profile at the top left of Fig. 1) will produce a cor-responding concavity in the abutting ground region of the image,and vice-versa19. This will lead to different convex parts on eitherside of a given contour. Our finding that IT neurons were drivenby the figural shapes resulting from one-sided edge assignment, notby contours per se, seems consistent with shape representation with-in IT in terms of the layout of such component parts5,19,24.

We re-analyzed the data in terms of another transform amongthe stimuli, to assess this hypothesis further. We correlated stim-ulus preferences across a transform that can be illustrated with ref-erence to Fig. 1, comparing the lower left stimulus in the frontpanel (stimulus f, Fig. 1) to the top right stimulus in the back panel(stimulus c, Fig. 1), and so on. We thus compared pairs of stimulithat had the same contrast polarity and faced in the same direc-tion, but had just the curved edge itself (not the shape as a whole)reflected. Thus, after figural assignment, the two members of thepair should have different convex parts. (Indeed, the change inconvex parts is the same as the change for a figure–ground reversal,except that the parts now ‘point’ in the same direction.) We foundthat the (null) correlations across this transform were equivalent tothose for our standard figure–ground reversal transform, averaging0.09 versus 0.04, respectively, with no difference between the pop-ulation distributions of these correlations.

Taken together, our results accord with theories of objectrecognition that propose15–20,25,26 that one-sided edge assign-ment precedes shape description in the visual system, withdecomposition into component parts proceeding only for thefigural side of any contour. A rival account27 proposes insteadthat part decomposition initially arises for both sides of every

articles


Fig. 7. Regions of IT cortex in which the cells were recorded, drawn onsections from the brain of monkey A. Top, locations of these coronalsections are shown on a schematic monkey brain.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



circle was presented centrally for 1 s immediately after the buzzer, buthad to withhold a lick when a red square was presented instead. Havingmastered this with greater than 95% accuracy, they were trained tomaintain central fixation. After several weeks of this training (withoutany exposure to the experimental stimuli like those in Fig. 1), the exper-imental trials were run in blocks of 36, each comprising the 32 experi-mental stimuli, plus 4 trials with red squares, all in random order. Foreach cell studied, three to seven blocks of trials were run, each with dif-ferent random orders of stimuli.

Analyses. The two monkeys gave the same pattern of results, and soare considered jointly here. For each trial, the number of action poten-tials occurring in a 500-ms period starting 100 ms after stimulus onsetwas initially considered. This period was chosen because most of theneurons studied typically showed vigorous responses to visual stim-uli with latencies just above 100 ms, and the monkeys consistently heldcentral fixation for the first 600 ms of stimulation. To test whether aneuron was showing selectivity among the set of 32 shapes, analysisof variance was performed on the response rates to the different stim-uli. Only those cells (78/88) that showed a significant effect of stimu-lus (at p < 0.001 or better) were included in the further analyses (31 from one monkey, 47 from the other), as only these could addressour experimental questions.

For each of our three orthogonal transformation of the stimuli (con-trast polarity, mirror-imaging and figure–ground reversal) the set of32 stimuli can be divided into two subsets of 16, one subset providingtransformed versions of each member of the other subset. To calcu-late the influence of one specific transformation (such as contrastreversal) on the stimulus selectivity of a single cell, we correlated thefiring rates to the 16 stimuli in one subset against those for the corre-sponding members of the other subset (Fig. 2; data from one illustra-tive cell), using Spearman’s rank-order correlation. This was initiallydone for firing rate across the 100–600-ms time bin following stimu-lus onset, for every cell (Fig. 5).

To see how the correlations (and thus the generalization of stimulusselectivity across a particular transform) developed over time, we next cal-culated the correlation coefficients for time bins of increasing extent. Thesewere calculated for spikes in response to each stimulus in the first 20 ms,then the first 40 ms, and so on, up to 500 ms after the stimulus period.Average correlations climbed rapidly to form an asymptote around 200–300 ms after stimulus onset for contrast and mirror-image transforms,but remained near zero throughout the trial for figure–ground reversal.

As another way to study the effects of stimulus transforms on thestimulus selectivity of the cell population, we examined how the dif-ference in responses to the preferred and the non-preferred stimulusdeveloped over time. This was done for each cell by calculating theresponse to its optimal stimulus in successive 20-ms time bins. A sim-ilar time course of firing rate was then calculated for the ‘non-pre-ferred’ stimulus for each cell. These values were then averaged acrossthe population of 78 cells to produce the diagram shown in Fig. 6a.Analogous procedures were used to plot the responses to contrast-reversed versions of the same two stimuli (Fig. 6b), mirror-reversedversions (Fig. 6c) or figure–ground reversed versions (Fig. 6d).

For the human shape-judgment task, observers were presented with asample shape for 400 ms, and two test stimuli were then added to the dis-play at bottom left and bottom right. They were asked to judge which ofthese two test stimuli was more similar in shape to the sample. The teststimuli were two different shapes drawn with equal probability withoutreplacement from the following set: the original shape, a contrast reversal,a mirror reversal, a figure–ground reversal and a shape with a differentcontour. Observers indicated by pressing a left or right key which testshape was more like the sample shape. The 12 repetitions of each of 20possible permutations of test pairings were averaged together to generateoverall preferences for each type of transform when presented at test.

ACKNOWLEDGEMENTSG.C.B. was supported by grants from the National Institutes of Health (R29

NS27296) and the National Science Foundation (SBR 96-16555). J.D. was

supported by the Biotechnology and Biological Sciences Research Council (UK).

RECEIVED 27 APRIL; ACCEPTED 30 JULY 2001

1. Baylis, G. C., Rolls, E. T. & Leonard, C. M. Functional subdivisions of thetemporal lobe neocortex. J. Neurosci. 7, 330–342 (1987).

2. Desimone, R., Schein, S. J., Moran, J. & Ungerleider, L. G. Contour, color andshape analysis beyond the striate cortex. Vision Res. 24, 441–452 (1985)

3. DiCarlo, J. J. & Maunsell, J. H. R. Form representation in monkeyinferotemporal cortex is virtually unaltered by free viewing. Nat. Neurosci. 3,814–821 (2000)

4. Logothetis, N. K., Pauls, J. & Poggio, T. Shape representation in the inferiortemporal cortex of monkeys. Curr. Biol. 5, 552–563 (1995).

5. Riesenhuber, M. & Poggio, T. Nat. Neurosci. 3, 1199–1204 (2000).6. Rollenhagen, J. E. & Olson, C. R. Mirror-image confusion in single neurons

of the macaque inferotemporal cortex. Science 287, 1506–1508 (2000).7. Rolls, E. T., Judge, S. J. & Sanghera, M. K. Activity of neurons in the

inferotemporal cortex of the alert monkey. Brain Res. 130, 229–238 (1977).8. Sary G., Vogel, R. & Orban, G. Cue-invariant shape selectivity of macaque

inferior temporal neurons. Science 260, 995997 (1993).9. Tanaka, K. Inferotemporal cortex and object vision. Annu. Rev. Neurosci. 19,

109–139 (1996).10. Malach, R. et al. Object-related activity revealed by functional magnetic

resonance imaging in human occipital cortex. Proc. Natl. Acad. Sci. USA 92,8135–8139 (1995).

11. Farah, M. J. & Aguirre, G. K. Imaging visual recognition: PET and fMRIstudies of functional anatomy of human visual recognition. Trends Cogn. Sci.3, 179–185 (1999).

12. Farah, M. J. Visual Agnosia (MIT Press, Cambridge, Massachusetts, 1990).13. Plaut, D. C. & Farah, M. J. Visual object representation: Interpreting

neurophysiological data within a computational framework. J. Cogn.Neurosci. 2, 320–343 (1990)

14. Rubin, E. Visuell Wahrgenommee Figuren (Gyldendalske Boghandel,Copenhagen, Germany, 1915).

15. Baylis, G. C. & Driver, J. One-sided edge-assignment in vision: 1. Figure-ground segmentation and attention to objects. Curr. Dir. Psychol. Sci. 4,140–146 (1995).

16. Driver, J. & Baylis, G. C. One-sided edge-assignment in vision: 2. Partdecomposition, shape description, and attention to objects. Curr. Dir.Psychol. Sci. 4, 201–206 (1995)

17. Driver, J. & Baylis, G. C. Edge-assignment and figure-ground segmentation inshort-term visual matching. Cognit. Psychol. 31, 248–306 (1996)

18. Baylis, G. C. & Cale, E. The figure has a shape, but the ground does not:Evidence from covert testing of shape recognition. J. Exp. Psychol. Hum.Percept. Perform. 27, 633–643 (2001).

19. Hoffman, D. D. & Richards, W. A. Parts of recognition. Cognition 18, 65–96(1984).

20. Baylis, G. C. & Driver, J. Obligatory edge-assignment in vision: the role offigure and part segmentation in symmetry detection. J. Exp. Psychol. Hum.Percept. Perform. 6, 1323–1342 (1995).

21. Selzer, B. & Pandya, D. N. Afferent cortical connections and architectonics ofthe superior temporal sulcus and surrounding cortex in the rhesus monkey.Brain Res. 149, 1–24 (1978)

22. Pinker, S. Visual cognition: an introduction. Cognition 18, 1–64 (1984).23. Rolls, E. T. & Baylis, G. C. Size and contrast have only small effects on the

responses to faces of neurons in the cortex of the superior temporal sulcus ofthe monkey. Exp. Brain. Res. 65, 38–48 (1986).

24. Biederman, I. Recognition-by-components: a theory of human imageunderstanding. Psychol. Rev. 94, 115–147 (1987).

25. 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).

26. Palmer, S. & Rock, I. Rethinking perceptual organisation: the role of uniformconnectedness. Psychon. Bull. Rev. 1, 29–55 (1994).

27. Peterson, M. A. Object recognition processes can and do operate beforefigure-ground organisation. Curr. Dir. Psychol. Sci. 3, 105–111 (1994).

28. Robinson D. A. A method of measuring eye-movements using a scleral searchcoil in a magnetic field. IEEE Trans. Biomed. Eng. 101, 131–145 (1963).

29. Baylis, G. C., Rolls, E. T. & Leonard, C. M. Selectivity between faces in theresponses of a population of neurons in the cortex of the superior temporalsulcus of the monkey. Brain Res. 342, 91–102 (1985).

articles©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


articles


Although the intrinsic rewarding properties of addictive drugssuch as heroin are important in the acquisition of drug self-administration, compulsive drug-seeking and drug-taking byaddicts is not readily explained in terms of simple reward orpositive reinforcement processes alone. Abstinence from thedrug in dependent subjects induces aversive withdrawal symp-toms that are thought to contribute to the compulsive natureof drug self-administration in addiction; an addict may self-administer heroin to escape from ‘abstinence agony’1–4. Indeed,heroin-dependent animals not only avoid places previouslyassociated with opiate withdrawal5 but also actively choose toavoid the precipitation of withdrawal6.

An alternative to this avoidance theory regards the withdrawalstate as analogous to natural motivational states such as hunger7.Just as non-food-deprived animals work for and consume foodswith an intrinsic incentive value, so will non-drug-dependentanimals self-administer addictive drugs. To render food-seekingcompulsive, however, it is necessary to make an animal hungryby depriving it of food. Contemporary theories of motivation8

assume that hunger enhances food-seeking and food-taking byincreasing the incentive value of the food to such an extent that itbecomes the dominant behavior in the animal’s current reper-toire. In other words, hunger makes food-seeking compulsive.The incentive-motivational account of addiction assumes thatthe withdrawal state induced by depriving an addict of heroinenhances the incentive value of the drug to such an extent thatheroin-seeking becomes the dominant behavior.

Both the avoidance and the incentive-motivational theoriesassume that drug self-administration is initially acquiredbecause the intrinsic, incentive value of a drug renders it reward-ing. Moreover, both theories assume that the consequent expo-sure to the drug produces dependence, so that abstinenceinduces a state of withdrawal. The theories differ in definingthe function of the withdrawal state in producing the compul-sive drug-seeking and drug-taking characteristic of addiction.According to the avoidance theory, drug-seeking and drug-tak-

The role of withdrawal in heroinaddiction: enhances reward orpromotes avoidance?

D. M. Hutcheson, B. J. Everitt, T. W. Robbins and A. Dickinson

Department of Experimental Psychology, University of Cambridge, Downing Street, Cambridge CB23EB, UK

Correspondence should be addressed to D.M.H. ([email protected])

The compulsive nature of heroin abuse has been attributed to the fact that drug self-administrationenables an addict to escape from and avoid the severe withdrawal symptoms resulting from opiatedependence. However, studies of incentive learning under natural motivational states suggest analternative hypothesis, that withdrawal from heroin functions as a motivational state that enhancesthe incentive value of the drug, thereby enabling it to function as a much more effective reward forself-administration. In support of this hypothesis, we show here that previous experience withheroin in withdrawal is necessary for subsequent heroin-seeking behavior to be enhanced whendependent rats once again experience withdrawal.

ing are maintained simply through the negative contingencybetween self-administration and the aversive state of with-drawal. In contrast, the incentive-motivational theory assumesthat the withdrawal state enhances the incentive value of thedrug to such an extent that drug-seeking and drug-taking dom-inate the behavioral repertoire.

One prediction of the incentive-motivational (but not theavoidance) theory is that the withdrawal state should not direct-ly increase heroin-seeking behavior. Rather, animals should haveto learn about the enhanced incentive value of heroin duringwithdrawal, through previous experience with the drug specifi-cally in that state. Hungry rats do not automatically increase theirfood-seeking behavior when shifted from a non-food-deprivedstate to a hunger state8,9. For food deprivation to enhance theperformance of a food-seeking response, rats have to have eatenthe food previously when hungry, consistent with the predictionthat animals must learn about the enhanced incentive value ofthe food in the deprived state8. We therefore investigated whetherthe control over heroin-seeking by withdrawal also depends onthis process of incentive learning.

RESULTSHeroin-seeking behaviorRats were trained to self-administer heroin by first respondingon a drug-seeking lever for a variable period to gain access to adrug-taking lever. A single press on the drug-taking lever thendelivered a heroin infusion. This was followed by a time-out peri-od, during which both levers were retracted, and after which anew cycle of seeking and taking was commenced by presentationagain of the drug-seeking lever. Previous studies of cocaine-selfadministration using this ‘seeking–taking’ schedule10 havedemonstrated a positive relationship between the drug-seekingand the infusion dose when the time-out period is long enoughfor the short-term satiety effects of the preceding infusion toabate. To determine a time-out value under which drug-seekingreflected the incentive value of heroin, we compared responding

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

for a high (0.12 mg/kg infusion) and a low (0.03 mg/kg infusion)dose both in the absence of a time-out period and with a 24-minute time-out period.

As there were no significant differences between latencies ofthe first drug-seeking response during the first and secondassessments of performance with the 0-minute time-out period(F1,12 = 1.09, p = 0.32), the mean of these values was comparedto performance under the 24-minute time-out period. Therewas a significant dose × time-out period interaction (F1,12 = 55.40, p = 0.001), and simple main effects analysesshowed that the latency was shorter (F1,12 = 44.50, p = 0.001)with the 0.12 mg/kg dose than with the 0.03 mg/kg dose underthe 24-minute time-out period. In contrast, under the 0-minutetime-out period, increasing the dose lengthened the latency(F1,12 = 51.06, p = 0.001). From these results, we concluded thatperformance on the ‘seeking’ lever reflected the incentive valueof heroin when assessed with a 24-min time-out period, andtherefore we used this period to assess the effects of withdraw-al on drug-seeking behavior (Fig. 1).

Incentive learning during opiate withdrawal A separate group of rats were trained to self-administer hero-in using the seeking–taking schedule (Fig. 2, Stage 1) beforeestablishing opiate dependence through regular, experimenter-administered, repetitive systemic (intraperitoneal; i.p.) injec-tions of morphine for the rest of the study (Fig. 2, Stage 2). Ananalysis of performance on the final training session after theinduction of dependence revealed no significant main effectsor interactions (all p-values > 1) between each of the subse-quent group assignments; the mean latency of the first seekingresponse was 114.7 ± 24.3 s (s.e.m.).

All rats then received the incentive learning stage, whichallowed the heroin groups but not the control (saline) groups toexperience the drug in the withdrawal state (Fig. 2, Stage 3).Throughout the incentive learning stage, only the drug-takinglever was available; the seeking lever was permanently retracted.During the first session of the incentive learning stage, the twoheroin groups received the same treatment: they self-adminis-tered heroin during spontaneous withdrawal, which allowed them

to experience the high incentive value of drug-taking in the with-drawal state. The control saline animals either self-administeredsaline or remained in the home cage in the first session while inwithdrawal. In the second session, all rats were in a normal, opi-ate-maintained state. One heroin group was allowed to self-administer saline in the experimental chamber; the other heroingroup remained in the home cage. The control group was allowedto self-administer heroin. This second session matched the expo-sure of the control and heroin groups to heroin-taking and thewithdrawal state, without allowing the control animals to expe-rience heroin in withdrawal.

Finally, in the critical test stage (Fig. 2, Stage 4), we mea-sured the performance of the drug-seeking response by theheroin groups when they were in a withdrawal state (elicitedby the omission of the two preceding scheduled morphineinjections). Responding on the drug-seeking lever did not giveaccess to the drug-taking lever during this test. If we hadallowed the rats to receive heroin infusions, they would havehad an opportunity during the test to learn about the relativeincentive value of the drug in the withdrawal and maintainedstates. By omitting drug infusions during the test, we couldassess in a controlled way the influence of the incentive learn-ing treatments on heroin-seeking.

The incentive learning and test stage were repeated twice, toenhance experience with heroin in withdrawal. When in with-drawal, rats with previous experience of heroin in withdrawal ini-tiated drug-seeking with a shorter latency (Fig. 3a), and showedmore completed cycles of drug-seeking compared to either salinecontrols or control groups without experience (Fig. 3b).

This effect of incentive learning depended upon the rats beingin withdrawal during the test. When tested in the maintainedstate, latency to initiate drug-seeking (Fig. 3a) and the numberof drug cycles achieved (Fig. 3b) were similar for the saline con-trol group and the heroin-experienced group. However, perfor-mance of the control rats was enhanced when tested in themaintained versus the withdrawal state, showing that withdraw-al from this level of dependence has a general suppressive effect onbehavior.

An analysis of the latencies (Fig. 3a) revealed a significantgroup × test-state (maintained versus withdrawal state) interac-tion (F1,41 = 6.05, p = 0.018). Analysis of the main effect of expe-rience (ANOVA with Newman–Keuls) showed no significantdifference between groups that had previously experienced hero-in in withdrawal (F2,18 = 1.30, n.s.). When tested in the with-drawal state, the group that had not previously experiencedheroin in withdrawal, compared to the same group tested on themaintained state, showed a significant increase in the latency toinitiate drug-seeking (F2,20 = 8.50, p ≤ 0.01). Analysis of simplemain effects showed that for groups tested in withdrawal, theheroin-experienced groups had lower average latencies in com-parison to the groups that either received saline (F1,9 = 9.58, p = 0.013) or remained in the home cage (F1,17 = 4.96, p = 0.04).

An analysis of the mean cycles (Fig. 3b) achieved yielded agroup × test-state interaction (F1,41 = 4.72, p = 0.036). Analysisof factor experience (ANOVA with Newman–Keuls) showed nosignificant difference between groups that had previously expe-rienced heroin in withdrawal (F2,18 = 3.06, p = 0.08). Groups thatdid not experience heroin in withdrawal and were tested in thewithdrawal state achieved fewer seeking cycles than the grouptested in the maintained state (F2,21 = 11.33, p ≤ 0.001). Analy-sis of simple main effects showed that for groups tested in with-drawal, respective heroin-experienced groups attained moreseeking cycles than the group that received saline (F1,9 = 12.36,

Fig. 1. The influence of drug dose and time out on heroin-seeking activ-ity. Rats initiated drug-seeking more rapidly for higher doses of heroinwhen they experienced a time-out period after each drug infusion. Allrats received the 10 infusions available in each session of this phase. Thedata represent values from the last two days of responding from eachfour-day period at the same time-out period. Asterisks represent indi-vidual group comparisons with one way ANOVA at each time-out point.*p < 0.05, ***p < 0.001.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


state9. Similarly, in the present study, ratsthat have previously taken heroin whenin opiate withdrawal subsequentlyshowed greater drug-seeking behaviorwhen again in withdrawal as comparedto the parallel no-experience control.Moreover, as in the case of hunger, theeffect of the previous incentive learning

experience was only observed in the presence of the relevantmotivational state, which in this case was withdrawal. Althoughthe opiate withdrawal state can be aversive, these findings nev-ertheless support an incentive-motivational hypothesis8 of opiateaddiction11,12 whereby withdrawal increases the incentive value ofheroin, an increase that is manifest as increased heroin-seekingbehavior during withdrawal.

It is unclear how the incentive learning effect can be explainedsimply in terms of the avoidance of or escape from withdrawal. Asself-administration of heroin in the withdrawal state during theincentive learning phase occurred in the absence of the opportunityto perform the seeking response, at no stage did the rats experi-ence an association of heroin self-administration and consequen-

Fig. 2. Incentive learning study. Stage 1, train-ing on the seeking–taking schedule. Stage 2,induction of opiate dependence. Stage 3,incentive learning, during which the heroingroups took heroin in the withdrawal stateand saline (or nothing) in the maintainedstate, and the saline group took saline (ornothing) during the withdrawal state andheroin in the maintained state. Stage 4, per-formance on the seeking lever was tested ineither the withdrawal state or the maintainedstate. Stages 3 and 4 were then repeated.

p = 0.01) and the group that remained in the home cage (F1,17 = 15.29, p = 0.001).

Measurement of physical signs of opiate withdrawalWe observed clear signs associated with the opiate withdrawalsyndrome in rats in withdrawal from morphine compared withthose who continued to receive their scheduled morphine injec-tions. Rats in abstinence had significantly more ‘wet dog’ shakesand tremors (mean shakes, 8.3 ± 1.4; mean tremors, 2.2 ± 0.7)than those rats in which dependence was maintained (meanshakes, 0.2 ± 0.2; mean tremors, 0; Z > 2.29 and p < 0.05 foreach). Paw tremor and ptosis were observed in some but notall subjects in the abstinence group (mean paw tremor, 1.2 ± 0.48; mean ptosis, 0.7 ± 0.3) and not at all in dependentsubjects. There were no significant differences between with-drawal and maintained subjects in the appearance of the othermeasured signs. The rats lost significant body weight duringthe 30-minute withdrawal observation period (Z = 3.11, p < 0.01) with larger decreases in body weight in the abstinent(mean, –3.7 ± 0.3 g) versus maintained group (mean, –2.3 ± 0.4 g). Rats maintained on morphine showed no evi-dence of being in a withdrawal state at any time.

DISCUSSIONThese results favor the incentive-motivational hypothesis of hero-in-seeking behavior under opiate withdrawal. Previous studieshave shown that rats having experienced a food reward whenhungry show greater food-seeking when once again deprived,versus rats with no experience of food reward in the hungry

Fig. 3. Effect of previous heroin experience on heroin-seeking activity.Performance on the seeking lever was enhanced in the heroin groupsrelative to the saline and ‘none’ groups during the first half of the testsession when the rats were tested in the withdrawal state, but not in themaintained state. Asterisks represent individual group comparisons withone way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001. (a) Mean latenciesof the first seeking response (± s.e.m.) were longer for the heroin non-experienced, saline and ‘none’ groups tested in the withdrawal state,compared to other conditions. (b) Mean cycles of the seeking achieved(± s.e.m.) during drug-seeking were lower for the groups that did notexperience heroin in withdrawal than for the heroin experiencedgroups when tested in the withdrawal state, but not when tested in themaintained state.

a

b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

tial escape from withdrawal when responding on this seeking lever.The elevated responding at test by the heroin group while in with-drawal, relative to the saline group, cannot be explained in terms ofprevious experience of such an avoidance or escape contingency.

However, from an incentive-motivational perspective, it is sur-prising that the rats tested in withdrawal did not perform the seek-ing response more rapidly than those in the maintained state,especially when they had previously taken the drug in withdraw-al. Indeed, the latencies for heroin-seeking by heroin-experiencedrats in withdrawal were similar to but no larger than those of hero-in-experienced rats in the maintained state. In contrast, hungryrats with the equivalent previous experience with food seek foodmore readily than undeprived rats. We suggest that the best expla-nation for this difference is the general suppression of seekingbehavior by the debilitating nature of the withdrawal state. Indeed,it is well established that opiate withdrawal suppresses appetitivebehavior13. Therefore, although the heroin group assigned a high-er incentive value to the drug in the withdrawn versus the main-tained state, this difference was masked by a general behavioralsuppression brought about by disruptive withdrawal symptomsand possibly a general aversive state produced by withdrawal1,18.

The incentive learning treatment experienced by the heroingroup is an example of a reinforcer revaluation procedure9

, anddemonstrates that heroin-seeking must, at least in part, be medi-ated by an associative structure that encodes the contingencybetween the seeking response and drug-taking. As the test wasconducted in the absence of the opportunity to perform the tak-ing response for heroin, previous experience with heroin-takingunder withdrawal could only influence the seeking response dur-ing testing if the animals had encoded the seeking–taking con-tingency. In this sense, heroin-seeking, like cocaine-seeking, is agoal-directed action rather than an inflexible stimulus–responsehabit. We have previously shown that following the acquisitionof cocaine-seeking behavior under an identical seeking–takingschedule, devaluing the taking response by extinction produces areduction in cocaine-seeking behavior during testing14.

The withdrawal state in opiate-dependent animals functionsas a primary motivational state that enhances the incentive valueof heroin so that drug-seeking becomes the compulsive form ofbehavior characteristic of opiate addiction15. Thus, reward orpositive reinforcement processes provide an integratedaccount11,12,16 of drug self-administration both in the non-addicted and in the dependent states, once it is recognized thatwithdrawal functions as a motivational state to enhance theincentive value of the drug. Although there are other integrativereward theories of addiction, such as the incentive salience-sen-sitization theory17, the incentive-motivational hypothesis isunique in predicting an involvement of incentive learning, andcontrasts with other accounts based on simple negative rein-forcement3 or opponent process theory1,18,19. The neural mech-anisms mediating incentive learning about drug rewards remainto be determined, although it is known that cortical processesare involved in the control of the incentive value of foodrewards20,21. Our hope, however, is that the incentive-motiva-tional theory, while acknowledging the importance of with-drawal11, also specifies precisely how it influences the nature ofthe transition from heroin use to compulsive heroin addiction.

METHODSAdult male Lister Hooded rats weighing 280–300 g at the time of surgery(Charles River, UK) were implanted with indwelling intravenous cathetersas previously described22. Rats were trained to self-administer heroin inchambers fitted with two retractable levers (4 cm wide, 10 cm apart and5 cm from the grid floor). During each infusion, 0.2 ml of solution was

administered over 7.3 s. All procedures were conducted in accordancewith licenses issued under the United Kingdom 1986, Animals (Scien-tific Procedures) Act.

Heroin-seeking behavior. Fifteen rats were trained to press the drug-tak-ing lever for a 0.06-mg/kg infusion of heroin in daily, 2-h pre-trainingsessions, which started with the insertion of the drug-taking lever andended with its retraction. Each press on this lever delivered a heroin infu-sion, during which the lever was retracted, the stimulus light above thelever was illuminated, and the house light was switched off. After acquir-ing this drug-taking response (7–10 sessions) on the fixed-ratio (FR) 1schedule, the rats were trained on the seeking–taking chain schedule10

in daily 3-h sessions. Each cycle of the chain schedule started with theinsertion of the drug-seeking lever with the taking lever retracted, andthe first press on the seeking lever initiated a random interval (RI) sched-ule. The first press after the RI had elapsed terminated the first link ofthe chain with the retraction of the seeking lever and presentation of thedrug-taking lever to initiate the second link. One press of the drug-takinglever then resulted in drug infusion accompanied by the same stimulusevents as during pretraining. A time-out period followed the infusion,after which the drug-seeking lever was reinserted, thereby starting thenext cycle of the chain. The RI parameter was progressively increased to30 s.

In summary, the rats were trained under a heterogeneous multiple(chain (tandem FR 1 RI) FR 1) time-out schedule for heroin with theassignment of the seeking and taking responses to the left and rightlevers counterbalanced across rats. The rats received a series of test ses-sions with different doses of heroin and different time-out periodsbefore the assessment of dose and time-out effects reported here. Forhalf the rats, the dose was 0.03 mg/kg/infusion; for the remaining rats,0.12 mg/kg/infusion. The time-out periods were then varied in thesequence of 0, 24 and 0 min with 4 consecutive daily training sessionsat each time-out period. Each session terminated at 10 infusions or 4 h, whichever occurred first.

Incentive learning during spontaneous opiate withdrawal. The proce-dure was divided into a number of stages (Fig. 2). In Stage 1 (Fig. 2), 45naive rats were initially trained to self-administer heroin under the mul-tiple (chain (tandem FR 1 RI 30 s) FR 1) time-out 0-s schedule using theprocedure used in the first experiment. This stage concluded followingsix two-hour sessions of training under the chain schedule.

After acquisition of self-administration, the rats were given twice-daily intraperitoneal injections of morphine hydrochloride (McFar-land-Smith, Edinburgh, UK) dissolved in sterile saline (0.9%), atincreasing doses from 8–30 mg/kg (0.1 mg per 100 g body weight, onaverage) over 3 days. The rats were subsequently maintained on 30mg/kg morphine by morning and evening injections given at least 12hours apart (Fig. 2, Stage 2). This morphine maintenance schedule,which has previously been shown to result in physical dependence23,was maintained throughout the rest of the experiment except duringwithdrawal. The morning morphine injection was always given pre-cisely four hours before all self-administration sessions. After 2 dayson the maximum daily dose of morphine, self-administration train-ing under the chained schedule was resumed with a 24-min time-outperiod. The heroin dose was increased from 0.06 to 0.12 mg/kg/infu-sion, and training continued for 6–8 sessions.

In Stage 3 (Fig. 2), spontaneous withdrawal was induced by replacingboth the scheduled evening dose of morphine after the last training ses-sion and the morphine injection on the following morning with salineinjections. The rats were then placed in the operant chambers with accessto the drug-taking lever only and allowed to self-administer under an FR 1 schedule with a 24-min time-out period during a 4-h session. Therats in the heroin group (n = 13) self-administered 0.12 mg/kg heroinper infusion, whereas the remaining rats in the saline group (n = 13) self-administered saline. The rats in both these groups received the maxi-mum 10 infusions. The maintenance schedule was then reinstated,commencing with an injection of morphine in the evening. After a daywithout training, the rats were given a second self-administration ses-sion with the drug-taking lever alone, having received their scheduled

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


morphine injections before the session. In this session, the heroin groupreceived on average 7.9 ± 0.3 saline infusions, whereas the saline groupreceived 8.8 ± 0.4 heroin infusions. This incentive learning procedureensured that the heroin group received the heroin in the withdrawal stateand saline in the maintenance state, whereas the saline group receivedthe opposite state–infusion assignment.

Additional groups were used to assess any influence of self-adminis-tration of saline. These two groups were trained in a fashion identical tothe others except that instead of receiving the saline self-administrationsessions, they remained in the home cage. Therefore, in the second hero-in group, the animals (n = 8) self-administered 9.8 ± 0.2 heroin infusionsin withdrawal, on average, whereas the ‘none’ group (heroin non-expe-rienced) remained in their home cages (n = 11). When in the dependentstate, the heroin group that previously self-administered heroin in with-drawal remained in the home cage, whereas each rat in the ‘none’ controlgroup self-administered the maximum 10 heroin infusions.

On the following day (Fig. 2, Stage 4), six animals from the first hero-in group and five from the saline group received saline injections (insteadof morphine) at the scheduled time in the evening and on the followingmorning to induce spontaneous withdrawal before the drug-seeking test.The remaining rats remained on the maintenance schedule of morphineinjections. The second heroin and the ‘none’ groups, which remained inthe home cage instead of self-administering saline, were all tested in with-drawal. All rats were then placed in the testing chambers for the test ses-sion during which they received a maximum of 10 trials with thedrug-seeking lever alone during a 4-h session. The first press after theseeking lever was inserted initiated a fixed interval 30-s schedule, and thefirst lever press after the schedule had timed out resulted in retraction ofthe seeking lever without an infusion. After a 24-min time-out period,the next trial started with the reinsertion of the seeking lever into thechamber. Stages 3 and 4 were then repeated using the same procedure.

Measurement of physical signs of spontaneous opiate withdrawal. Themaintenance schedule of morphine injections was then reinstated afterthe second test (for two days), and the rats that had been tested in a main-tained state were assigned to two matched groups (half saline-experi-enced and half heroin-experienced in each group). The first group (n = 6) were observed for physical withdrawal signs during spontaneouswithdrawal (28 hours after the last morphine injection) and comparedwith a second group (n = 6) in which dependence was maintained (4 hafter the last scheduled morphine injection). Rats were placed in a clearplastic circular observation area (24 cm wide by 50 cm high), allowed tohabituate for 15 min, and then observed for the following 30 min. Anobserver, who was blind to the group assignment of the rat, determinedthe number of wet dog shakes, the incidence of teeth chattering and mas-tication, and the number of 5-min periods during which ptosis, pilo-erection or diarrhea occurred. The body weight of the rats was alsorecorded before and after the 30-min observation period.

ACKNOWLEDGEMENTSThis work was supported by a Medical Research Council UK Programme Grant

(9537855) and was done within the MRC Cooperative for Brain, Behaviour and

Neuropsychiatry. D.M.H. holds a Medical Research Council UK Training

Fellowship.


1. Koob, G. F., Stinus, L., Le Moal, M. & Bloom, F. E. Opponent process theoryof motivation: neurobiological evidence from studies of opiate dependence.Neurosci. Biobehav. Revs. 13, 135–140 (1989).

2. Koob, G. F. Drug addiction: the yin and yang of hedonic homeostasis. Neuron16, 893–896 (1996).

3. Wikler, A. Dynamics of drug dependence. Implications of a conditioningtheory for research and treatment. Arch. Gen. Psychiatry 28, 611–616 (1973).

4. Khantzian, E. J. The self-medication hypothesis of addictive disorders: focuson heroin and cocaine dependence. Am. J. Psychiatry 142, 1259–1264 (1985).

5. Mucha, R. F., Gritti, M. D. & Kim, C. Aversive properties of opiate withdrawalstudied in rats. NIDA Res. Monogr. 75, 567–570 (1986).

6. Goldberg, S. R., Hoffmeister, F., Schlichting, U. & Wuttke, W. Aversiveproperties of nalorphine and naloxone in morphine-dependent rhesusmonkeys. J. Pharmacol. Exp. Ther. 179, 268–276 (1971).

7. Nader, K., Bechara, A. & van der Kooy, D. Neurobiological constraints onbehavioral models of motivation. Annu. Rev. Psychol. 48, 85–114 (1997).

8. Dickinson, A. & Balleine, B. Motivational control of goal-directed action.Anim. Learn. Behav. 22, 1–18 (1994).

9. Balleine, B. Instrumental performance following a shift in primarymotivation depends on incentive learning. J. Exp. Psychol. (Anim. Behav.) 18,236–250 (1992).

10. Olmstead, M. C., Parkinson, J. P., Miles, F. J., Everitt, B. J. & Dickinson, A.Cocaine-seeking by rats: regulation, reinforcement and activation.Psychopharmacol. (Berl.) 152, 123–131 (2000).

11. Stewart, J., de Wit, H. & Eikelboom, R. Role of unconditioned andconditioned drug effects in the self-administration of opiates and stimulants.Psychol. Rev. 91, 251–268 (1984).

12. Wise, R. A. & Bozarth, M. A. A psychomotor stimulant theory of addiction.Psychol. Rev. 94, 469–492 (1987).

13. Hand, T. H., Koob, G. F., Stinus, L. & Le Moal, M. Aversive properties ofopiate receptor blockade: evidence for exclusively central mediation in naiveand morphine-dependent rats. Brain Res. 474, 364–368 (1988).

14. Olmstead, M. C., Lafond, M. V., Everitt, B. J. & Dickinson, A. Cocaine-seeking by rats is a goal-directed action. Behav. Neurosci. 115, 394–402(2001).

15. Ahmed, S. H., Walker, J. R. & Koob, G. F. Persistent increase in the motivationto take heroin in rats with a history of drug escalation.Neuropsychopharmacology 22, 413–421 (2000).

16. Bozarth, M. A. & Wise, R. A. Anatomically distinct opiate receptor fieldsmediate reward and physical dependence. Science 224, 516–517 (1984).

17. Robinson, T. E. & Berridge, K. C. The neural basis of drug craving: anincentive-sensitization theory of addiction. Brain Res. Rev. 18, 247–291(1993).

18. Solomon, R. L. & Corbit, J. D. An opponent-process theory of motivation. I.Temporal dynamics of affect. Psychol. Rev. 81, 119–145 (1974).

19. Solomon, R. L. The opponent process theory of acquired motivation. Am.Psychol. 35, 691–712 (1980).

20. Balleine, B. W. & Dickinson, A. Goal-directed instrumental action:contingency and incentive learning and their cortical substrates.Neuropharmacology 37, 407–419 (1998).

21. Balleine, B. W. & Dickinson, A. The effect of lesions of the insular cortex oninstrumental conditioning, evidence for a role in incentive memory. J. Neurosci. 20, 8954–8964 (2000).

22. Arroyo, M., Markou, A., Robbins, T. W. & Everitt, B. J. Acquisition,maintenance and reinstatement of intravenous cocaine self-administrationunder a second-order schedule of reinforcement in rats: effects ofconditioned cues and continuous access to cocaine. Psychopharmacol. (Berl.)140, 331–344 (1998).

23. Maldonado, R. et al. Inhibition of morphine withdrawal by the association ofRB 101, an inhibitor of enkephalin catabolism, and the CCKB antagonist PD-134, 308. Br. J. Pharmacol. 114, 1031–1039 (1995).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

In vivo investigation of the functional correlates of learning andmemory in humans is currently possible with neuroimaging tech-niques measuring regional cerebral blood flow and metabolism,such as positron emission tomography (PET) and functional mag-netic resonance imaging (fMRI). The results of neuroimaginginvestigations are largely convergent with the clinical findings inamnesic patients1, which suggest a pivotal role of medial tempo-ral lobe structures—in particular, the hippocampal formation—in long-term episodic memory2. More controversial is theneuropsychological evidence for the involvement of the frontallobe in human episodic memory. Lesions of the frontal lobes arenot usually associated with clinically evident amnesia. However, aconsistent activation of the prefrontal cortex has been found notonly during working memory tasks3, but also during long-termepisodic learning4,5. In addition, deficits in source memory6 ormemory for temporal order (recency)7 have been reported fol-lowing frontal lobe lesions in man, and meta-analytical evidenceexists for impairment in free recall tasks after frontal damage8.

Imaging studies of episodic memory, mostly for verbal stim-uli, suggest a hemispheric encoding–retrieval asymmetry; the leftprefrontal cortex is crucial in encoding, and the right prefrontalcortex in retrieval. The hemispheric encoding–retrieval asym-metry (HERA) model9, developed from these observations, isnow the focus of a number of imaging studies that have tried tocharacterize other factors affecting both the hemispheric asym-metry and the functional neuroanatomical subdivisions of frontalactivation. In agreement with clinical neuropsychological evi-dence10, prevalent right-sided or bilateral activations have been

Prefontal cortex in long-termmemory: an “interference” approachusing magnetic stimulation

Simone Rossi1,2, Stefano F. Cappa3, Claudio Babiloni1,4, Patrizio Pasqualetti5, Carlo Miniussi1,Filippo Carducci1,4, Fabio Babiloni4,5 and Paolo M. Rossini1,6

1 IRCCS S Giovanni di Dio, Via Pilastroni 4, I-25125, Brescia, Italy 2 Dipartimento di Neuroscienze, Sezione Neurologia, U.O. di Neurofisiopatologia, Policlinico Le Scotte, Viale Bracci, I-53100 Siena, Italy 3 Centro di Neuroscienze Cognitive, Università Salute-Vita S. Raffaele, DIBIT Via Olgettina 58, 20132 Milano, Italy 4 Dipartimento di Fisiologia Umana e Farmacologia, Università La Sapienza, P. le A. Moro 5, I-00185 Roma, Italy 5 AFaR (Associazione Fatebenefratelli per la Ricerca)—Dipartimento di Neuroscienze, S. Giovanni Calibita, Fatebenefratelli Isola Tiberina,

I-00186 Roma, Italy 6 Neurologia, Università Campus Biomedico, Roma, Italy

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

Neuroimaging has consistently shown engagement of the prefrontal cortex during episodic memorytasks, but the functional relevance of this metabolic/hemodynamic activation in memory processingis still to be determined. We used repetitive transcranial magnetic stimulation (rTMS) to transientlyinterfere with either left or right prefrontal brain activity during the encoding or retrieval of picturesshowing complex scenes. We found that the right dorsolateral prefrontal cortex (DLPFC) was crucialfor the retrieval of the encoded pictorial information, whereas the left DLPFC was involved in encod-ing operations. This ‘interference’ approach allowed us to establish whether a cortical area activatedby a memory task actually contributes to behavioral performance.

observed during the encoding of non-verbal items such as unfa-miliar faces11 or complex scenes12. However, the the left prefrontalcortex is also activated in response to non-verbal stimuli such asunfamiliar faces or complex figures13. Prevalent right prefrontalactivation has been associated with successful retrieval14,15,retrieval effort16 or monitoring of the retrieved information17;left prefrontal activation has also been observed in studies deal-ing with recognition18 and source memory19.

Thus, both the material and the type of memory process mayaffect the lateralization of frontal activation during memory tasks.In addition, distinct portions within the frontal lobe (such asventrolateral, dorsolateral and anterior prefrontal cortex) maybe engaged in different aspects of memory performance5. VLPFChas been associated with the maintenance of information;DLPFC, with manipulation and/or monitoring; and AFC, withselection of processes and/or subgoals.

To better clarify the role of the DLPFC(s) in encoding andretrieval memory process, it might be helpful to use methodolo-gies which do not depend solely on the measurement of the meta-bolic or hemodynamic response to cognitive challange, but whichdirectly and transiently block the functional participation of a “can-didate” brain region. This can be achieved with transcranial mag-netic stimulation, a widely used technique for motor controlresearch and clinical testing20,21. Repetitive TMS (rTMS) dischargestrains of magnetic impulses repetitively in a few hundred mil-liseconds, reaching cortical regions of interest. This method allowssafe interference with the complex neural networks underlyingsomatosensory perception22, motor-related cerebral activity23 and

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

articles


higher cognitive functions21,24, and provides direct insights intothe involvement of stimulated areas by means of measurable behav-ioral performance.

Here we used focal rTMS to transiently disrupt the function ofthe left or right DLPFC, in order to clarify the roles and func-tional prevalence of these regions in the mechanisms of encod-ing and retrieval of complex images (Fig. 1). Six conditions werestudied, R-Enc (right rTMS in encoding, no stimulation inretrieval), L-Enc (left rTMS in encoding, no stimulation inretrieval), sham (left rTMS in encoding and right in retrieval),R-Ret (no stimulation in encoding and right rTMS in retrieval),L-Ret (no stimulation in encoding and left rTMS in retrieval)and baseline (reference condition, no stimulation in encoding orin retrieval). Thus, the effects of right and left prefrontal stimu-lation applied during encoding and retrieval were compared withbaseline and sham rTMS conditions. This interference approachmay help to better clarify the functional significance of the frontallobe in long-term memory.

RESULTS

The two measures ‘C’ (criterion) and ‘d´’ (discrimination),derived from signal detection theory, were computed on thebehavioral data (Table 1). C was inversely related to the propor-tion of false positives (when, during retrieval, subjects erroneouslyanswered that a distractor had been seen in the encoding phase).The index d´ indicated the ability of subjects to distinguishbetween ‘already seen’ and ‘never seen’ pictures. Both indices sig-nificantly varied across experimental conditions (C, F5,60 = 5.224,p < 0.001; d´, F5,60 = 7.921, p < 0.001). However, after Tukey’scorrection, only in the R-Ret condition did C decrease (versusbaseline, p = 0.012; versus R-Enc, p = 0.006; versus sham,

p = 0.003), suggesting that subjects tended to be lessspecific, with more intrusions of unseen picturesduring right DLPFC rTMS. None of the other pair-wise comparisons were significant. On the otherhand, both L-Enc (versus baseline, p = 0.003; versussham, p = 0.002) and R-Ret (versus baseline, p < 0.001; sham, p < 0.001) lowered d´. R-Enc andL-Ret did not produce any significant differencecompared to either baseline or sham (p > 0.40).These results confirm that the left DLPFC duringencoding and right DLPFC during retrieval areinvolved in the modulation of memory tracestrength. However, whereas a specific interferenceof right DLPFC stimulation was further highlight-ed by R-Ret versus L-Ret contrast (p = 0.047 afterTukey’s correction), the L-Enc versus R-Enc con-trast was not significant (p = 0.175).

Focusing the analysis of variance (ANOVA) onthe HERA model (with ‘right versus left hemisphere’ and ‘stim-ulus side in encoding versus retrieval’ as within-subjects factors),the main effects of hemisphere and stimulus were not statistical-ly significant (F1,12 = 0.119, p = 0.736 and F1,12 = 0.435, p = 0.552,respectively), but a significant interaction effect occurred (F1,12 = 35.08, p < 0.001; Fig. 2a).

Reaction times for each response were consistently faster inencoding than in retrieval. Logistic regressions in search of corre-lations between reaction time and error rate were not statisticallysignificant. However, taking reaction time as dependent variable

Table 1. Percentages of hits, false alarms and measures of signaldetection (C and d´) in the different experimental conditions.

Conditions Hits False alarms Criterion C d´(mean % ± s.d.) (mean % ± s.d.) (mean ± s.d.) (mean ± s.d.)

R-Enc 65 ± 18 20 ± 13 1.02 ± 0.67 1.44 ± 0.62L-Enc 54 ± 28 26 ± 14 0.68 ± 0.42 0.79 ± 0.83Sham 74 ± 19 21 ± 17 1.05 ± 0.82 1.87 ± 0.93R-Ret 58 ± 29 41 ± 23 0.28 ± 0.65 0.58 ± 0.58L-Ret 77 ± 15 33 ± 13 0.47 ± 0.39 1.37 ± 0.69Baseline 76 ± 14 22 ± 15 0.97 ± 0.72 1.86 ± 1.01

L-Enc, left DLPFC rTMS in encoding; no stimulation in retrieval. R-Enc, right DLPFC rTMS inencoding; no stimulation in retrieval. Sham, sham rTMS (left DLPFC in encoding and rightDLPFC in retrieval). R-Ret, right DLPFC rTMS in retrieval; no stimulation in encoding. L-Ret,left DLPFC rTMS in retrieval; no stimulation in encoding. Baseline, no stimulation. Each valuerefers to pooled subjects (13 for each condition).

Fig. 1. Sites of TMS and experimental timing. Top, position of scalpelectrode site F3 (10–20 International EEG system) on a scalp model,and position of the perpendicular projection of F4 on a cortical model(F4p). Models were obtained by averaging the magnetic resonanceimages of 152 subjects (SPM96). Talairach coordinates for F4p are (42,32, 31), which correspond to superior frontal gyrus/Brodmann area 9.Analogous results can be obtained with the projection of F3 on the scalpmodel. This allows a rough localization of the cortical site stimulated byrTMS with the focal coil. Bottom, typical time course of the experimen-tal condition with respect to visual and rTMS stimulations; EMG of theECD and FDI muscles and of the motor response. rTMS stimulationapplied to the scalp electrode site F3 did not elicit EMG activity.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce


articles

with condition as an independent variable, two patterns distin-guishing the effects of rTMS on performance from that on reactiontime emerged. First, reaction times in retrieval were significantlyassociated with condition (F5,60.76 = 2.884, p = 0.021). Second, theTukey’s procedure identified two statistically different (p < 0.001)subsets that were internally homogeneous: blocks in which real orsham TMS were applied during retrieval (sham, R-Ret and L-Ret; p = 0.188) and blocks without stimulation (p = 0.857); (Fig. 2b).

DISCUSSIONrTMS transiently and safely interferes with the function of cor-tical networks involved in cognitive processes; this offers advan-tages for the investigation of the neurophysiologicalmechanisms underlying cognitive task performance. PET andfMR, indeed, are able to detect regional ‘activations’ with excel-lent spatial resolution, but their relatively low time discrimi-nation does not allow for tracing the hierarchical organizationin a distributed network. Moreover, it is difficult to determineunequivocally if the detected metabolic changes result in a netfacilitatory or inhibitory effect on behavior. In contrast, if thetransient interference induced by rTMS results in the worsen-ing of behavioral performance, this may provide strong evi-dence for the active involvement of the stimulated brain areain the process under study25 and for its place in a functionalhierarchy. Nevertheless, the intimate mechanisms of rTMSinterference—and its selectivity within functional subregions—still need to be fully determined. The present findings providedirect evidence for a functional role of the prefrontal cortex inlong-term episodic memory processes. The hemispheric asym-metry effects observed in this study seem to extend the HERAmodel of verbal episodic memory organization in the brain26 tothe visuospatial domain.

The high error rate after right stimulation during retrieval(Table 1) suggests that the rTMS-induced disrupting effect isdirect, as it takes place immediately after the stimulation period,while the retrieval effort is active. In other words, the interfer-ence of rTMS persisted for at least 1.5 seconds after the end ofstimulation (Fig. 1), and was associated with an increased num-ber of false positive responses. Patients with frontal lesions tendto produce more ‘false alarms’ in recognition memory tasks27,28.Taken together, these findings indicate a selective specialization ofthe right DLPFC in the monitoring phase of retrieval17 duringyes/no recognition tasks of complex visuospatial stimuli.

Left rTMS applied during the encoding process significantlyreduced the probability of successful retrieval of the encodedinformation (Table 1), providing direct confirmation of previ-ous neuroimaging evidence suggesting that the left DLPFC is cru-cial in encoding mechanisms29. This finding is striking,considering that the encoded information (complex scenes) hasshown prevalence in the right hemisphere12. The effect mightresult from less efficient (‘shallow’) encoding and/or from a fasterdecay of the information, due to concomitant rTMS. However,as the L-Enc versus R-Enc contrast was not significant (despitethe finding that R-Enc, unlike L-Enc, did not differ between shamand baseline), the present findings suggest that a bilateral PFCengagement, with left functional prevalence, is associated withencoding of pictorial material memory traces.

The regions affected by rTMS in the present study are proba-bly the same as those engaged in working memory tasks5,30,31.Many neuroimaging investigations during working memory taskshave suggested that the DLPFC is crucial in the short-term reten-tion of information. In particular, Brodmann area 46 seems tobe associated with the selection of response, whereas areas 9 and8 seem crucial for the maintenance of the representations32. Theseregionally specific nodes within the working memory distributedneuronal network are capacity-constrained in the physiologicaldomain31; rTMS might transiently disable the processing con-tribution of DLPFC and adjacent structures to the circuitry ofworking memory, inducing a dramatic decrement of its capacityin the active manipulation of information5. Indeed, pictures werealways available during the task (Fig. 1), so that working memo-ry processes might occur only during self-monitoring of theresponses in the previous trials or together with an influence onexecutive frontal functions (management of instructions, visuo-motor transformation and response selection). The location ofthe activations associated with material specificity, however, seemsto be in the VLPFC5, which was unlikely to be directly affectedby rTMS of middle/superior frontal regions. This part of the PFCmay be less sensitive than the VLPFC to the nature of informationcontent, and seems to show a left-sided functional prevalence.

The low error rate in retrieval during left rTMS ruled out thepossibility that any interference of rTMS with frontal eye fields(affecting the accuracy of saccades33 needed to scan the picture)contributed to recognition errors.

Fig 2. Effects of rTMS on retrieval and reaction time. (a) Hemisphericinteraction of rTMS effects on d´. (b) Mean values of reaction times inthe different conditions of retrieval. The presence of rTMS, either activeor sham, shortens reaction times, irrespective of the stimulated site.See text for statistical evaluations.

a

b

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

articles


Reaction times were consistently faster during encoding thanduring retrieval, reflecting the expected different cognitivedemand. Concomitant rTMS, either sham or active, ipsi- or con-tralateral to the moving hand, significantly shortened reactiontimes (Fig. 2b). This suggests no direct motor-related effects ofthis type of rTMS, but rather, a nonspecific arousal effect thatdid not influence the cognitive process, but intersensory facili-tation mechanisms34 due to the noise of the coil discharging can-not be ruled out.

However, the physical effect of low-intensity TMS is to inducecurrents in the brain that flow almost parallel to the cortical surface.These currents result in an immediate trans-synaptic activation ofa discrete brain volume underneath the coil35 followed by activa-tion of other regions functionally connected with the stimulatedone23,36,37,38. The observed interference effect of the rTMS mighttake place on the whole distributed neural cortical network that isinvolved in a particular task. In this framework, it is difficult toascertain whether some remote effects of DLPFC stimulation mightextend to more ventral regions through extant functional connec-tions, whereas the low intensity and the selectivity of stimulationmake improbable a direct spread of the magnetic stimulus toVLPFC and AFC. This might explain the relatively low specificityof the effects induced by rTMS of DLPFC(s). Indeed, the same siteof brain stimulation may lead to interference with other aspects ofmemory function, including working memory38,39, procedurallearning40 and semantic memory41.

Possible trans-synaptic effects of rTMS on memory process-es, including an ‘at distance’ (diaschisis) interference on hip-pocampal function, could not be addressed in the presentprotocol, but may be amenable to investigation using combinedPET–TMS protocols.

METHODSSubjects. Thirteen healthy volunteers (9 female), 22 to 41 years old (aver-age, 30.1), naive to the pictures presented, gave their written informedconsent for the study, after the approval of the protocol by the local EthicsCommittee. All were right handed (mean dexterity, 89%) according tothe Edinburgh handedness inventory42. Their medical history and exam-ination were normal, and they had never taken neuroactive drugs.

Subjects sat on reclining chairs with their heads stabilized by restraints,in front of a 17-in monitor. Their right index finger rested between 2buttons spaced 6 cm apart.

Experimental conditions. Six blocks of encoding were followed by sixblocks of retrieval, and the order of presentation was pseudorandomizedand counterbalanced among subjects. For each block of the encoding phase,16 complex colored magazine pictures (8 interiors and 8 external land-scapes) were randomly presented on the monitor for 2 s each, with 2 inter-trial intervals (18 or 25 s long, 8 and 5 subjects, respectively). Images werepreceded by a visual warning stimulus (a red spot lasting 1 s). Subjectswere instructed to press with their right index finger one of the two but-tons (left, to indicate internal; right, external) as quickly as possible afterthe presentation of a green circle in the middle of each picture (the ‘go’signal), which appeared 1 s after the picture presentation (Fig. 1).

One hour after encoding, the corresponding retrieval blocks occurredin which 16 pictures of interiors were again randomly presented. Eightof these pictures had previously been seen (tests), and eight were novel(distractors). The timing of warning and go signals, picture presenta-tion and intervals were the same as in the encoding blocks. Subjects wereagain asked, in a yes–no recognition task, to discriminate between thepictures by pressing one of two buttons (left, test; right, distractor)immediately after the go stimulus. The correct choice in encoding and inretrieval (test) was always the left button. The encoding and the retrievalphases lasted 28 to 38 min each, depending on the interstimulus inter-val used. Each picture and each response produced an appropriate trig-ger signal. A 10-min training session, performed with a different set ofpictures, allowed the subjects to practice with the task and with either

sham or active rTMS (both left and right DLPFCs) before the actualexperimental session.

The six encoding/retrieval blocks (R-Enc, L-Enc, sham, R-Ret, L-Retand baseline) were labeled according to the type (active or sham) andthe side (left/right) of the rTMS applied on the DLPFCs.

Recording and stimulating procedures. Triggers and electromyographic(EMG) signals were recorded continuously and off-line analyzed. EMGsignals were recorded with surface electrodes glued on the skin in a shortbipolar montage, with the active electrode placed on the motor pointsof the right and left first dorsal interosseous muscles (FDI) and on theright extensor communis digitorum (ECD).

Before applying rTMS, individual resting excitability thresholds formotor cortex stimulation were determined for both hemispheres by mea-suring the amplitude of motor twitches evoked by single TMS stimuli inthe contralateral FDI muscle. Threshold was defined as the minimalintensity of the stimulator output (Mag-Stim Super Rapid, Carmarth-henshire, Wales, UK) capable of evoking a motor evoked potential (MEP)greater than 50 µV with 50% probability (see International StandardGuidelines43). The stimulating figure-eight coil was tangential to the areaof scalp surface corresponding to the primary motor cortex (C3 or C4positions of the 10–20 EEG international system), with its handle point-ing backward and angled about 45° from the midline. Excitability thresh-old measurements were taken after the presentation of the warningstimulus, as during the experimental setup. Once individual thresholdswere determined (mean, 62.6 ± 9.2%, without interhemispheric differ-ences), the intensity of stimulation was reduced by 10%. Thus, left andright DLPFC were stimulated, when required, with a subthreshold inten-sity for eventual motor cortex activation that would have overtly inter-fered with motor performance (mean intensity of stimulation used, 55.7± 9.1%). Then, left and right DLPFCs were stimulated by lining up the tipof the middle bar of the coil on F3 and F4, respectively37,38, correspond-ing to the Brodmann area 9 (Fig. 1). A mechanical arm fixed the coil inthat position (marked on the scalp) and its correct position was checkedby an experimenter repeatedly throughout the session. Trains of 10%subthreshold rTMS (500 ms, 20 Hz) were delivered, when required bythe experimental design, at the same time as picture presentation (Fig. 1). The same intensity and timing of rTMS was used for sham stim-ulation. In this case, the coil was still centered on F3 and F4, but it washeld perpendicularly to the scalp surface, so that scalp contact and dis-charging noise were similar to the active stimulation, but the inducedmagnetic field did not activate cortical neurons39.

Data analysis. For each subject’s answer, the trial-to-trial performance(wrong/right choice) and reaction time (from the go signal to the firstEMG burst, either in the right FDI or ECD muscles) were considered.The cutoff to define effective the earliest EMG activation was a burstgreater than 50 µV in one of the two muscles, taking into account possi-ble different response strategies of subjects.

Behavioral data were initially composed in a spreadsheet with 1248rows (13 subjects × 16 answers × 6 blocks) and then grouped in two cat-egories. The first category grouped variables derived from the experi-mental design, including subjects, sequence and type of blocks, order ofthe stimuli within each block, and type of picture (interiors or externals inencoding, and internal test or distractor in retrieval). The second catego-ry included variables related to the response to each stimulus during eitherencoding or retrieval (right or wrong), with corresponding reaction times.

To take into account all possible sources of variations of the first group(plus the reaction time as continuous covariate) on the two dependentdichotomous variables (failure during encoding and retrieval), two logis-tic regressions were applied. The ‘forward likelihood-ratio’ method waschosen as a screening procedure, by individuating which variables couldplay a role and, therefore, to limit the factors to be included in theANOVA models. Thereafter, ANOVA for repeated measures (with exper-imental blocks as within-subjects factor) was applied to two psychome-tric measures (C and d´) commonly used to describe the ability to rejectdistractors during retrieval and to discriminate between the two items(tests and distractors). These measures can be obtained by applying asimple algorithm derived by the signal detection theory44. C can be inter-preted as an index of ‘specificity’ (the ‘willingness’ of a subject to endorseitems as old); d´ can be considered as the ‘true memory strength’ (the

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce


articles

ability of subjects to distinguish between already seen or novel pictures).After verifying the differences between blocks with active rTMS and

control conditions (baseline and sham), a specific two-way ANOVA forrepeated measures (‘hemisphere’ and ‘side of stimulus in encoding andretrieval’ as within-subjects factors) was applied to better address theHERA model. Throughout ANOVA for repeated measures, Mauchly’stest did not allow rejection of the sphericity assumption. Thus, no attemptto correct the degrees of freedom (Greenhouse–Geisser procedure) wasmade. Tukey’s method was used for post hoc comparisons.

ACKNOWLEDGEMENTSWe thank A. Polese, K. Sosta and I. Benaglio for help with experiments. The

work was partly supported by Ministero della Sanità (Progetto Finalizzato 1999)

and by Fondazione Telethon Onlus (E.C0985).

RECEIVED 20 APRIL; ACCEPTED 27 JULY 2001

1. Squire, L. R. Memory and the hippocampus: a synthesis from findings withrats, monkeys, and humans. Psychol. Rev. 99, 195–231 (1992).

2. Schacter, D. L. & Wagner, A. D. Medial temporal lobe activations in fMRI andPET studies of episodic encoding and retrieval. Hippocampus 9, 7–24 (1999).

3. Haxby, J. V., Petit, L., Ungerleider, L. G. & Courtney, S. M. Distinguishing thefunctional roles of multiple regions in distributed neural systems for visualworking memory. Neuroimage 11, 380–391 (2000).

4. Cabeza, R. & Nyberg, L. Neural bases of learning and memory: functionalneuroimaging evidence. Curr. Opin. Neurol. 13, 415–421 (2000).

5. Fletcher, P. C. & Henson, R. N. A. Frontal lobes and human memory. Insightsfrom functional neuroimaging. Brain 124, 849–881 (2001).

6. Janowsky, J. S., Shimamura, A. P., Kritchevsky, M. & Squire, L.R. Cognitiveimpairment following frontal lobe damage and its relevance to humanamnesia. Behav. Neurosci. 103, 548–560 (1989).

7. Milner, B., Corsi, P. & Leonard, G. Frontal-lobe contribution to recencyjudgements. Neuropsychologia 29, 601–618 (1991).

8. Wheeler, M. A., Stuss, D. T. & Tulving, E. Frontal lobe damage producesepisodic memory impairment. J. Int. Neuropsychol. Soc. 1, 525–536 (1995).

9. Tulving, E., Kapur, S., Craik, F. I., Moscovitch, M. & Houle, S. Hemisphericencoding/retrieval asymmetry in episodic memory: positron emissiontomography findings. Proc. Natl. Acad. Sci. USA 91, 2016–2020 (1994).

10. Milner, B. in The Frontal Agranular Cortex and Behavior (eds. Warren, J. M. &Akert, K.) (New York, McGraw-Hill, 1964).

11. McDermott, K. B., Buckner, R. L., Petersen, S. E., Kelley, W. M. & Sanders, A. L.Set- and code-specific activation in frontal cortex: an fMRI study of encodingand retrieval of faces and words. J. Cogn. Neurosci. 11, 631–640 (1999).

12. Kirchhoff, B. A., Wagner, A. D., Maril, A. & Stern, C. E. Prefrontal-temporalcircuitry for episodic encoding and subsequent memory. J. Neurosci. 20,6173–6180 (2000).

13. Fletcher, P. C., Frith, C. D. & Rugg, M. D. The functional neuroanatomy ofepisodic memory. Trends Neurosci. 20, 213–218 (1997).

14. Nyberg, L. et al. Functional brain maps of retrieval mode and recovery ofepisodic information. Neuroreport 29, 249–252 (1995).

15. Buckner, R. L., Koutstaal, W., Schacter, D. L., Wagner, A. D. & Rosen, B. R.Functional–anatomic study of episodic retrieval using fMRI. I. Retrievaleffort versus retrieval success. Neuroimage 7, 151–162 (1998).

16. Schacter, D. L. et al. Neuroanatomical correlates of veridical and illusoryrecognition memory: evidence from positron emission tomography. Neuron17, 267–274 (1996).

17. Shallice, T. et al. Brain regions associated with acquisition and retrieval ofverbal episodic memory. Nature 368, 633–635 (1994).

18. Schacter, D. L., Buckner, R. L., Koutstaal, W., Dale, A. M. & Rosen, B. R. Lateonset of anterior prefrontal activity during true and false recognition: anevent-related fMRI study. Neuroimage 6, 259–269 (1997).

19. Rugg, M. D., Fletcher, P. C., Chua, P. M. & Dolan, R. J. The role of theprefrontal cortex in recognition memory and memory for source.

Neuroimage 10, 520–529 (1999).20. Rossini, P. M. & Rossi, S. Clinical application of motor evoked potentials.

Electroencephalogr. Clin. Neurophysiol. 106,180–194 (1998).21. Hallett, M. Trascranial magnetic stimulation and the human brain. Nature

406, 147–150 (2000).22. Oliveri, M. et al. Time-dependent activation of parieto-frontal networks

for directing attention to tactile space: a study with paired TMS pulses inright brain damaged patients with extinction. Brain 123, 1939–1947(2000).

23. Rossi, S. et al. Effects of repetitive transcranial magnetic stimulation onmovement-related cortical activity in humans. Cereb. Cortex 10, 802–808(2000).

24. Pascual-Leone, A., Walsh, V. & Rothwell, J. Transcranial magnetic stimulationin cognitive neuroscience—virtual lesion, chronometry, and functionalconnectivity. Curr. Opin. Neurobiol. 10, 232–237 (2000).

25. Walsh, V. & Cowey, A. Transcranial magnetic stimulation and cognitiveneuroscience. Nat. Rev. Neurosci. 1, 73–79 (2000).

26. Lepage, M., Ghaffar, O., Nyberg, L. & Tulving, E. Prefrontal cortex andepisodic memory retrieval mode. Proc. Natl. Acad. Sci. USA 97, 506–511(2000).

27. Curran, T., Schacter, D. L., Norman, K. A. & Galluccio, L. False recognitionafter a right frontal lobe infarction: memory for general and specificinformation. Neuropsychologia 35, 1035–1049 (1997).

28. Swick, D. & Knight, R. T. Contributions of prefrontal cortex to recognitionmemory: electrophysiological and behavioral evidence. Neuropsychology 13,155–170 (1999).

29. Desgranges, B., Baron, J. C. & Eustache, F. The functional neuroanatomy ofepisodic memory: the role of the frontal lobes, the hippocampal formation,and other areas. Neuroimage 8, 198–213 (1998).

30. Wagner, A. D. Working memory contributions to human learning andremembering. Neuron 22, 19–22 (1999).

31. Callicott, J. H. et al. Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI. Cereb. Cortex 9, 20–26(1999).

32. Rowe, J. B., Toni, I., Josephs, O., Frackowiak, R. S. & Passingham, R. E. Theprefrontal cortex: response selection or maintenance within workingmemory? Science 288, 1656–1660 (2000).

33. Brandt, S. A., Ploner, C. J., Meyer, B. U., Leistner, S. & Villringer, A. Effects ofrepetitive transcranial magnetic stimulation over dorsolateral prefrontal andposterior parietal cortex on memory-guided saccades. Exp. Brain Res. 118,197–204 (1998).

34. Nickerson, R. S. Intersensory facilitation of reaction time: energy summationor preparation enhancement? Psychol. Rev. 80, 489–509 (1973).

35. Amassian, V. E., Qirck, G. J. & Stewart, M. A comparison of corticospinalactivation by magnetic coil and electrical stimulation of monkey motorcortex. Electroencephalogr. Clin. Neurophysiol. 77, 390–401 (1990).

36. Illmoniemi, R. J. et al. Neuronal responses to magnetic stimulation revealcortical reactivity and connectivity. Neuroreport 8, 3537–3540 (1997).

37. Paus, T. et al. Transcranial magnetic stimulation during Positron EmissionTomography: a new method for studying connectivity of the human cerebralcortex. J. Neurosci. 17, 3178–3184 (1997).

38. Mottaghy, F. M. et al. Modulation of the neural circuitry subserving workingmemory in healthy human subjects by repetitive transcranial magneticstimulation. Neurosci. Lett. 280, 167–170 (2000).

39. Jahanshahi, M. et al. The effects of transcranial magnetic stimulation over thedorsolateral prefrontal cortex on suppression of habitual counting duringrandom number generation. Brain 121, 1533–1544 (1998).

40. Pascual-Leone, A., Wassermann, E. M., Grafman, J. & Hallett, M. The role ofthe prefrontal cortex in implicit procedural learning. Exp. Brain Res. 107,479–485 (1966).

41. Flitman, S. S. et al. Linguistic processing during repetitive transcranialmagnetic stimulation. Neurology 50, 175–181 (1998).

42. Oldfield, R. C. The assessment and analysis of handedness: the Edinburghinventory. Neuropsychologia 9, 97–113 (1971).

43. Rossini, P. M. et al. Non-invasive electrical and magnetic stimulation of thebrain, spinal cord and roots: basic principles and procedures for routineclinical application. Electroencephalogr. Clin. Neurophysiol. 91, 79–92 (1994).

44. Kamman, R. in Workbook for General Psychology 228–236 (Prentice–Hall, 1970).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

Prefontal cortex in long-term memory: an “interference” approach using magnetic stimulationSimone Rossi, Stefano F. Cappa, Claudio Babiloni, Patrizio Pasqualetti, Carlo Miniussi, Filippo Carducci, Fabio Babiloniand Paolo M. RossiniNat. Neurosci. 4, 948–952 (2001)

The title of this article contained a typographical error. It should have read as follows:

Prefrontal cortex in long-term memory: an “interference” approach using magneticstimulation

Ion channel properties underlying axonal action potential initiation in pyramidalneuronsCosta M. Colbert and Enhui PanNat. Neurosci. 5, 533–538 (2002)

A printer’s error introduced an extraneous diagonal line into Fig. 2b on page 534. The correct figure is reproduced below.

corrigenda

Neurotrophins use the Erk5 pathway to mediate a retrograde survival responseFiona L. Watson, Heather M. Heerssen, Anita Bhattacharyya, Laura Klesse, Michael Z. Lin and Rosalind A. SegalNat. Neurosci. 4, 981–988 (2001)

In Fig. 5e on page 986, the pluses and minuses for lines “PD to DA” and “PD to CB” were incorrect. The conclusions stated in the textand the experimental description in the figure legend were correct. The corrected figure is reproduced below.

A-kinase anchoring proteins in amygdala are involved in auditory fear memoryMarta A.P. Moita, Raphael Lamprecht, Karim Nader and Joseph E. LeDouxNat. Neurosci. 5, 837–838 (2002)

The authors wish to correct their supplementary methods online, which gave the wrong sources for three antibodies. The mouseanti-RIIα and anti-RIIβ antibodies were obtained from Transduction Laboratories (San Diego, California), and the rabbit anti-AKAP150 antibody was obtained from Santa Cruz Biotechnology (Santa Cruz, California).

Fig. 5. Activation of Erk5 promotes survival. (e) Neurons in compart-mented cultures were treated with PD98059 (PD) to distal axons or cellbodies, as indicated. Distal axons were stimulated with neurotrophinsand cell body lysates were immunoblotted for P-CREB. PD treatment ofdistal axons alone does not prevent CREB phosphorylation. When PD isapplied to the cell bodies, CREB phosphorylation is inhibited.

nature neuroscience • volume 5 no 10 • october 2002 1017

errata

–80 –60 –40 –20 200Command potential (mV)

Nor

mal

ized

cond

ucta

nce

0.2

0.4

0.6

0.8

1.0b

AxonSoma

Fig. 2. Na+ channel properties differ betweenthe soma and axon. (b) Voltage dependence ofactivation in somatic and axonal patches. Eachcurve is the best-fit Boltzmann for an individualpatch. Axonal Na+ channels (solid lines) wereactivated by less depolarization than somaticchannels (dotted lines).

e

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

Visual spatial attention is an essential neural function basedon wide-ranging brain networks including cortical and sub-cortical stages and prominently involving the right parietal cor-tex1–4 (for review, see ref. 5). Much of the experimentalevidence about the extent and function of this network stemsfrom behavioral studies in patients with localized brain damagewho have deficits in focusing spatial attention, mostly towardthe contralesional space. Depending on whether the deficitsappear for unilateral stimuli presented in contralesional spaceor predominantly for bilateral stimuli, such syndromes havebeen classified as neglect or extinction, respectively3.

Experimental studies and models of spatial attention haveinvestigated the representation of external attentional space acrossthe two cortical hemispheres and have suggested the involvementof reciprocal inter-hemispheric inhibition6–12. Although thesemodels differ in their exact structural basis, their functional prin-ciples and the topography of attentional representation, a sim-ple prediction resulting from the shared idea of inter-hemisphericcompetition is the disinhibition of structures in the unimpairedhemisphere caused by lesions of structures in the impaired hemi-sphere. This functional release might also result in a measurablebehavioral enhancement. However, any such effect is difficult totest reliably in patients with permanent brain lesions; locationsof lesions are not systematic, their influence is frequently wide-spread, and patients’ attentional baseline before the onset of theimpairment is not normally known13. Experimental ‘virtuallesion’ approaches such as transcranial magnetic stimulation(TMS)14,15 offer a practical alternative.

Enhanced visual spatial attentionipsilateral to rTMS-induced 'virtuallesions' of human parietal cortex

Claus C. Hilgetag1,3, Hugo Théoret2 and Alvaro Pascual-Leone2

1 Boston University School of Medicine, Department of Anatomy and Neurobiology, 700 Albany Street W746,Boston, Massachusetts 02118, USA

2 Beth Israel Deaconess Medical Center and Harvard Medical School, Department of Neurology, Laboratory for Magnetic Brain Stimulation, 330 Brookline Avenue, Boston, Massachusetts 02215, USA

3 Present address: International University Bremen, Campus Ring 1, D-28759 Bremen, Germany

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

The breakdown of attentional mechanisms after brain damage can have drastic behavioralconsequences, as in patients suffering from spatial neglect. While much research has concentratedon impaired attention to targets contralateral to sites of brain damage, here we report theipsilateral enhancement of visual attention after repetitive transcranial magnetic stimulation (rTMS)of parietal cortex at parameters known to reduce cortical excitability. Normal healthy subjectsreceived rTMS (1 Hz, 10 mins) over right or left parietal cortex. Subsequently, detection of visualstimuli contralateral to the stimulated hemisphere was consistently impaired when stimuli were alsopresent in the opposite hemifield, mirroring the extinction phenomenon commonly observed inneglect patients. Additionally, subjects’ attention to ipsilateral targets improved significantly overnormal levels. These results underline the potential of focal brain dysfunction to produce behavioralimprovement and give experimental support to models of interhemispheric competition in thedistributed brain network for spatial attention.

Here we studied the potential enhancement in visual spatialattention resulting from cortical impairment. In a simple behav-ioral protocol, subjects had to detect small rectangular stimulibriefly presented on a computer monitor either unilaterally inthe left or right visual periphery, or bilaterally in both. The sub-jects’ performance in stimulus detection was assumed to reflectthe efficiency of their attentional mechanisms. Peripheral stim-uli, well outside the foveal focus, were chosen because a theo-retical model for attention10 indicated that these should beparticularly effective. Stimulus sizes were adapted to the sub-jects’ visual acuity and attentional capability, to be at peri-threshold stimulus strength (with a resulting median baselinelevel of correct stimulus detection around 20% for small stim-uli and 70% for slightly larger ones; Fig. 1). We used an repet-itive transcranial magnetic stimulation (rTMS) protocol that,when applied to the motor cortex, induces a transient reduc-tion of cortical excitability in most subjects16,17. The applica-tion of the same protocol to other cortical areas also results inbehavioral effects consistent with transient suppression of cor-tical excitability18–20. We tested spatial detection performanceat baseline and after rTMS during the window of cortical dys-function. The observations of the experimental conditions werecontrasted with control conditions.

RESULTSParietal rTMS produced a significant ipsilateral enhancement forthe correct detection of unilateral stimuli compared to the pre-TMS baseline (evaluated by one-tailed z-test of all within-sub-

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


ject differences, p < 0.01; Fig. 2a). Conversely, there was a non-sig-nificant trend toward decremented performance for unilateralcontralateral stimuli. The TMS also caused a significant reduc-tion in correct responses to bilateral stimuli (p < 10–7). Detailedinvestigation of the results showed that although trends were mir-ror-symmetric for rTMS of left and right parietal cortex, theenhancement produced by right-hemispheric rTMS was signifi-cantly larger than that after stimulation of the left hemisphere (p < 0.05), and only the right-hemispheric parietal stimulationproduced a significant ipsilateral detection enhancement (p < 10–3; Fig. 2b and c). However, both left- and right-hemi-spheric rTMS led to significantly reduced correct detection ofbilateral stimuli (p < 10–6 and p < 0.005 respectively). All trendshad the same direction for the two different stimulus sizes used

Fig. 1. Methods. (a)Stimulus detection per-formance versus stimu-lus size. Representativecurves showing thedetection performanceof one subject (H.T.)for the three differentstimulus conditions, left(L) or right (R) unilat-eral and bilateral (B), aswell as for catch trials(N) during initial baselinetesting. Such titrationcurves were obtainedfor all subjects, to selectsuitable peri-thresholdstimuli for the experimental trials. In this particular case, stimulus sizes 2 and 3 were chosen for subsequent testing. (b) Experimental paradigm, seeMethods. MT, motor threshold. (c) Location of parietal rTMS sites (subject C.C.H.). The horizontal cross section at the bottom shows left and right(P3 and P4) stimulation sites; the sagittal section in the top panel shows the location of right-hemispheric P4. Magnetic stimulation sites are indicatedby red arrows. IPS, intraparietal sulcus; AG, angular gyrus; SMG, supramarginal gyrus; CS, central sulcus.


(not shown separately); however, the effects were more pro-nounced for the smaller stimuli. No significant increases ordecreases in stimulus detection occurred in the control experi-ments, during either sham trials or rTMS of right-hemisphericM1. (Subjects’ performance in all stimulus conditions decreasedafter sham stimulation, although not significantly (z-test, p > 0.09). Detection of left unilateral stimuli very slightlyimproved after rTMS of right M1, whereas performance in thetwo other stimulus conditions was somewhat diminished. How-ever, none of these trends were significant (p > 0.1).)

Responses after parietal rTMS tended to deviate toward theside of the magnetic stimulation, producing a significant increasein the detection of right unilateral stimuli after right rTMS (one-tailed pairwise t-test of subjects’ responses for both stimulus sizes,

articles

Fig. 2. Changes in correct stimulus detection after parietal rTMS. The diagrams are based on changes in the number of correctly detected stimuli (rela-tive to the total number of presented stimuli) averaged for both stimulus sizes and all subjects. (a) The pooled data show a significant increase in perfor-mance ipsilateral to the parietal rTMS location (increase in relative percentage points, 7.3 ± 2.6% s.e.m.), and a trend toward decreased contralateralperformance (reduction by 2.5%). In addition, detection of bilateral stimuli decreased significantly (–11.7 ± 2.0%). These trends are also apparent afterseparating data for left parietal TMS (b) and right parietal rTMS (c). Significant trends (as determined by z-tests, see Results) are marked by asterisks.

a b c

a b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


p < 0.01; Fig. 3b), and a significant increase in the number ofincorrect unilateral responses on the ipsi-rTMS side for bilateralstimuli (p < 0.05 for both left and right TMS, Fig. 3a and b). Inaddition, there was a significant decrease in the detection of bilat-eral stimuli for either rTMS side (p < 0.005 for left TMS, p < 0.01for right TMS), and a strong leftward deviation of responses forbilateral stimuli already at baseline.

We confirmed the parietal location of stimulated scalp sites withthe help of MRI in representative subjects (for example, in Fig. 1c)and found them in broad agreement with earlier investigations4,11.

DISCUSSIONOur results are consistent with the idea that a temporary pari-etal impairment induced by rTMS disinhibits the contralateralcortex and thus leads to improved attentional performance. Itseems unlikely that the effects were produced by a global facili-tatory impact of the rTMS, as there was a trend toward dimin-ished detection of stimuli in the visual hemispace represented inthe stimulated parietal cortex. Similarly, it seems unlikely thatthe effects were mainly due to the transcallosal spread of TMS-induced cortical activation. TMS does induce distant, includingtranscallosal, effects21,22; however, if local TMS effects werematched by transcallosal ones (such as bilateral inhibition), nobehavioral consequences on attention would be expected. Tran-scallosal effects need to have a differential, specific impact in thetwo hemispheres in order to produce the observed behavioraleffects. The inhibition induced at the site of stimulation wouldhave to be matched by increased excitability in the contralateralhemisphere, consistent with the proposed competitive mecha-nism. An intriguing possibility would be that no net local effectsare achieved by rTMS while contralateral excitability is inducedthrough transcallosal influences. However, motor cortex effects of1 Hz rTMS do not support this hypothesis16,23.

One may further exclude a performance increase simply dueto training and growing familiarity with the protocol, as the sub-jects’ performance did not improve after sham stimulation. Afacilitatory ipsilateral effect solely resulting from the auditoryand mechanical stimulation provided by the rTMS can also beruled out. First, no effect has been shown yet whereby this wouldresult in a prolonged facilitation, lasting for several minutes afterthe mechanical and auditory stimulation have ended. Second,the sham trials and our control stimulation of right cortical area

M1, which also provided auditory and auditory-mechanical stim-ulation, did not result in a performance increase. Finally, theenhanced stimulus detection is unlikely to be caused by a modi-fied, more liberal approach by the subjects to reporting detect-ed stimuli during the trials after rTMS. Such a shift in theresponse approach would also result in an increased error rate inthe catch trials (when no stimulus was presented); however, theerror rate was the same in baseline and test trials. It is also notclear how this shift could explain the differential performanceenhancement and decrements in the different unilateral and bilat-eral stimulus conditions.

We used a recently developed rTMS protocol, which from pre-vious studies16,17,24,25 seems to produce a decrease of corticalexcitability in a large proportion of subjects, continuing evenafter the acute rTMS has ceased. The duration of the suppress-ing effects of rTMS trains on cortical excitability depends on theduration of the rTMS train itself but can last for up to 15 min-utes16,25. However, these are excitability studies (based on motor-evoked potentials and phosphenes), and there are as yet very fewstudies detailing the behavioral effects of this stimulation proto-col18,20. Previous results from our laboratory show that a 10-minute 1-Hz rTMS train to the dorsolateral prefrontal cortexaffects spatial memory for approximately 5 minutes19. It seemslikely from our present results that this rTMS protocol also had aprolonged impact on neural function in the parietal cortex.

In agreement with earlier studies26, we found an rTMS-induced performance deficit for correctly reporting bilateral visu-al stimuli coupled with an increase in incorrect responses foripsilateral unilateral targets; that is, subjects frequently onlyreported the ipsilateral stimulus of a bilateral stimulus pair. Theseeffects correspond to the clinical picture of spatial extinction. Wealso saw a trend toward diminished detection of contralesionalunilateral stimuli, although we did not find a significant neglecteffect after parietal TMS inhibition as did an earlier study27 forright-hemispheric acute rTMS. This divergence may be due tothe different attentional protocol, the different parietal stimula-tion site or the different TMS protocol.

A more general question is posed by the finding of a largeripsilateral enhancement than contralateral decrement in stimulusdetection. Although this result is puzzling at first, it is in line withseveral earlier proposals for shifts and modifications of the atten-tional topography after unilateral cortical impairment12,13,28,29.Because of gradients in the representation of attentional space inthe brain, the exact shape of the intact and lesion-modified dis-

articles


Fig. 3. Response vectors for baseline performance and after parietalTMS. Shown in black (before TMS) and red (after TMS) are relativecomposite vectors for all responses, both correct and incorrect, to uni-lateral left, unilateral right and bilateral stimuli. Detection vectors arefound closest to the coordinate axis that represents the respective stim-ulus condition, along the axis to the left for left unilateral stimuli, to theright for right unilateral stimuli, and toward the top for bilateral stimuli.Length of the coordinate axes indicates 100% detection. The extent ofthe vectors along the B-axis is determined by the relative number ofbilateral responses and along the R- or L-axis by the relative number ofcorrect unilateral minus the relative number of (incorrect) contralateralunilateral responses. (Typically, the number of incorrect contralateralresponses was very small.) The vectors are averaged across all subjectsand both stimulus sizes. Response vectors for bilateral stimuli had astrong leftward deviation even at baseline. Asterisks indicate significantincreases in responses compared to the pre-TMS baseline for unilateralstimuli; crosses indicate significant reductions in bilateral stimulusresponses. (a) Responses after left parietal rTMS; (b) responses afterright parietal rTMS.

a

b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



tribution of attention affect how large the relative impairment-induced changes in attentional performance may be at a giveneccentricity of visual space (Fig. 4). Our results, however, argueagainst the hypothesis made previously29 that pathophysiologicattentional effects are merely due to the residual gradient of atten-tional representation in the intact hemisphere. This would notexplain the enhancement seen in our results. On the other hand,an apparent enhancement in attention could also result from aphenomenological translation or rotation of attentional spacetoward the ipsilesional side13,28. Although our experimentaldesign could not test this idea, an apparently rotated attentionalgradient as suggested previously28 could also be explained on thebasis of competitive principles in the interaction of the hemi-spheres12. In any case, the detailed study of intact and impairedattentional topography within the different theoretical frame-works will be a worthwhile challenge for the future.

Although the observed trends were mirror-symmetric forrTMS of right and left parietal cortex, indicating that right- andleft-hemispheric structures make similar, mirror-symmetric con-tributions, the enhancement after right parietal rTMS was morepronounced. This reinforces the idea of a frequently observedright-hemispheric dominance for spatial attentional representationin the human brain30. Lateralization was also seen in the leftwarddeviation of responses in the baseline condition (correct averagebaseline detection 71% for left unilateral stimuli and only 52%for unilateral right ones, as well as the large number of wrong uni-lateral left responses for bilateral stimuli), which corresponds tothe effect of pseudoneglect31. The significant deterioration of per-formance for bilateral stimuli after rTMS, on the other hand,might indicate the higher attentional load of this condition and,potentially, the competitive interaction between stimuli32.

Our results are thus in line with studies finding effects thatcould result from mechanisms of inter-hemispheric competi-tion4,9,11,33. Such studies have further shown that, in addition toparietal cortex, other cortical or subcortical components in thenetwork for spatial attention may also interact competitively11.Competition between different brain structures might, thus, be ageneral principle of brain function34 and may explain the ‘para-doxical’ behavioral enhancement or recovery observed after var-ious brain lesions10,35. Future studies will need to systematicallyexplore the specific contributions of the different neural com-ponents, as well as their functional interactions, in the widelydistributed network for spatial attention.

METHODSSubjects. Experiments were done in accordance with NIH guidelines forhuman studies and were approved by the Beth Israel Deaconess MedicalCenter Investigational Review Board. Subjects were seven healthy males(27–37 years old) with normal or corrected to normal vision. All subjectswere right-handed according to the Edinburgh Inventory of Manual Pref-erence36. An eighth subject was studied, but his results were excluded from

further analysis because of left-handedness. Six of the subjects were test-ed after left parietal (P3) rTMS as well as (on a different day) after rightparietal (P4) rTMS. One subject was only tested after right parietal (P4)rTMS. The order of testing left and right regions was random.

Stimulus presentation. Stimuli consisted of small black rectangles (longeraxis horizontal) or squares sized 2 × 2, 2 × 3, 3 × 3, 3 × 4 or 4 × 5 pixels(approximate pixel size 0.25 mm × 0.25 mm) presented against whitebackground. Two suitable adjacent peri-threshold stimulus sizes wereselected for each subject after an initial block of trials involving all fivestimulus sizes (Fig. 1a). This procedure was implemented to avoid floor orceiling effects. Averaged for left, right and bilateral stimuli, the subjectscorrectly detected 10–33% (median, 21%) of the smaller stimuli and43–80% (median, 69%) of the larger stimuli. Stimuli were presented 177.5 mm left or right of a fixation cross (at approximately 24° eccen-tricity) on a 19-inch Viewsonic G790 CRT display with an effective 18-inch diagonal, at 71 Hz and 1024 × 768 pixel resolution, with a view-ing distance of 400 mm. Stable viewing was supported through a chin-rest, and subjects were instructed to keep fixation on the center of thescreen throughout the experiment. Eye movements were monitored withthe help of a video camera mounted behind the CRT display. The strong-ly peripheral location of the stimuli made target saccades unlikely in anycase, and none of the trials had to be eliminated because of eye move-ments. Stimuli were presented and responses recorded using SuperlabPro2.0 (Cedrus, San Pedro, California). Subjects used their dominant righthand to report the detection of stimuli through a RB-410 response box(Cedrus), pressing the index finger for detection of left unilateral stim-uli, ring finger for right unilateral stimuli, and middle finger for bilateralstimuli. No button was pressed when no stimulus was detected.

The same block design was used before and after the train of TMS(Fig. 1b). A fixation cross appeared for 1000 ms. After its disappearance,the stimuli (left, right, bilateral or none) were presented for 40 ms. Sub-sequently the subject had a 1250-ms window in which to respond (Fig. 1b) before a new trial was started (complete trial length, 2290 ms).In every block, each stimulus condition (left, right or bilateral), for twoadjacent sizes of stimuli adapted to the individual subject, was repeated20 times in random order (total 120 trials). The trials were randomlyinterspersed with 10 catch trials (no stimulus presentation), so that acomplete block lasted just under 5 min. Correct non-responses to catchtrials were very high for all subjects during both baseline (98 ± 1%; s.e.m.)and testing after rTMS (97 ± 1%).

TMS protocol. Locations for TM stimulation were determined within the10–20 EEG (electroencephalogram) coordinate system and confirmed byMRI using vitamin E capsules as markers of stimulated skull positions(Fig. 1c). Test locations were P3 and P4; control stimulations were doneover right M1, determined topographically and verified as the punctummaximum for TMS-induced excitation of the contralateral first dorsalinterosseus. Stimulation was produced through a MagStim SuperRapidstimulator (MagStim, Whitland, UK) using a 70-mm figure-eight coil.Additionally, sham trials were conducted in which the coil was held closeto the right side of the subject’s skull, but angled away so that no currentwas induced in the brain. The output strength of the TMS was set to 90%

articles

Fig. 4. Schematic representation of a modified attentional gradient fol-lowing parietal impairment (shown here for right impairment; trans-formed representation in red, baseline in black), resulting in potentiallyunequal performance changes at mirror-symmetric stimulus eccentrici-ties (–x and x). Although the diagram depicts a Gaussian-like distribu-tion of attention, unequal performance changes might also be found forvarious other attentional distributions. This representation assumes atopographic representation of attentional space with center-magnifica-tion for both intact and unilaterally impaired attention28, and alsoaccounts for the amplified and narrowed attentional gradient due toimpairment and contralateral disinhibition.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


of the subject’s motor threshold, defined as the minimal intensity of stim-ulation capable of inducing motor evoked potentials greater than 50 µVpeak-to-peak amplitude in at least 6 of 10 trials. At threshold, SuperRapidoutput ranged between approximately 48% to 68%. In experimental andcontrol conditions, the TM stimulation consisted of a train of TMSimpulses applied at 1 Hz for 600 s. Previous studies have suggested thatthis protocol produces temporarily reduced excitability of stimulated cor-tical sites outlasting the period of actual rTMS16,17,25, and is accompaniedby significant behavioral changes18. The time window of reduced excitabil-ity has been estimated to last between 5 and more than 15 min. To max-imize the potential effect, we designed the testing phase of the experimentto be completed within a 5-min window after the end of rTMS.

To assess the potential behavioral impact of rTMS, the relative changesin correctly detected stimuli before and after rTMS were computed (Fig. 2) as the percentage of the following:

(1)

We also derived response vectors for the different stimulus conditions(Fig. 3). In addition to showing changes in correct responses as in Fig. 2, these vectors also present incorrect responses, and so allow anassessment of the performance relative to the total number and kind ofstimuli presented in the different conditions, before and after rTMS.

ACKNOWLEDGEMENTSThe work was supported in part by grants from the Wellcome Trust (C.C.H.),

Canadian Institutes of Health Research (H.T.) and the US National Institute of

Mental Health (RO1MH60734, RO1MH57980), and National Eye Institute

(RO1EY12091) (A.P-L.).

RECEIVED 22 MAY; ACCEPTED 1 AUGUST 2001

1. Mesulam, M. M. A cortical network for directed attention and unilateralneglect. Ann. Neurol. 10, 309–325 (1981).

2. Nobre, A. C. et al. Functional localization of the system for visuospatialattention using positron emission tomography. Brain 120, 515–533 (1997).

3. Vallar, G. Spatial hemineglect in humans. Trends. Cogn. Sci. 2, 87–97 (1998).4. Rushworth, M. F. S., Ellison, A. & Walsh, V. Complementary localization and

lateralization of orienting and motor attention. Nat. Neurosci. 4, 656–661(2001).

5. Toga, A. W., Frackowiak, R. S. J. & Mazziotta, J. C. (eds). Special issue (issue 1,part 2): Action and Visio-Spatial Attention: Neurobiological Bases andDisorders. Neuroimage 14, 1-146 (2001).

6. Kinsbourne, M. Hemi-neglect and hemisphere rivalry. Adv. Neurol. 18, 41–49(1977).

7. Kinsbourne, M. in Neurophysiological and Neuropsychological Aspects ofSpatial Neglect (ed. Jeannerod, M.) 69–86 (North-Holland, Amsterdam,1987).

8. Kinsbourne, M. in Unilateral Neglect: Clinical and Experimental Studies (eds.Robertson, I. & Marshall, J.) 63–86 (Lawrence Erlbaum, Hove, UK, 1993).

9. Seyal, M., Ro, T. & Rafal, R. Increased sensitivity to ipsilateral cutaneousstimuli following transcranial magnetic stimulation of the parietal lobe. Ann.Neurol. 38, 264–267 (1995).

10. Hilgetag, C. C., Kötter, R. & Young, M. P. Inter-hemispheric competition ofsub-cortical structures is a crucial mechanism in paradoxical lesion effectsand spatial neglect. Prog. Brain Res. 121, 121–141 (1999).

11. Oliveri, M. et al. Left frontal transcranial magnetic stimulation reduces

detected – detectedpre-rTMS post-rTMS

number of presented stimuli

contralesional extinction in patients with unilateral right brain damage.Brain 122, 1731–1739 (1999).

12. Hilgetag, C. C. Spatial neglect and paradoxical lesion effects in the cat—amodel based on midbrain connectivity. Neurocomputing 32–33, 793–799(2000).

13. Smania, N. et al. The spatial distribution of visual attention in hemineglectand extinction patients. Brain 121, 1759–1770 (1998).

14. Pascual-Leone, A., Bartres-Faz, D. & Keenan, J. P. Transcranial magneticstimulation: studying the brain-behaviour relationship by induction of‘virtual lesions.’ Phil. Trans. R. Soc. Lond. B Biol. Sci. 354, 1229–1238 (1999).

15. Walsh, V. & Cowey, A. Transcranial magnetic stimulation and cognitiveneuroscience. Nat. Rev. Neurosci. 1, 73–79 (2000).

16. Chen, R. et al. Depression of motor cortex excitability by low-frequencytranscranial magnetic stimulation. Neurology 48, 1398–1403 (1997).

17. Maeda, F., Keenan, J. P., Tormos, J. M., Topka, H. & Pascual-Leone, A.Interindividual variability of the modulatory effects of repetitive transcranialmagnetic stimulation on cortical excitability. Exp. Brain Res. 133, 425–430(2000).

18. Kosslyn, S. M. et al. The role of area 17 in visual imagery: convergent evidencefrom PET and rTMS. Science 284, 167–170 (1999).

19. Mottaghy, F., Gangitano, M., Sparing, R., Krause, B. & Pascual-Leone, A.Segregation of areas related to visual working memory in the prefrontalcortex revealed by rTMS. Cereb. Cortex (in press).

20. Theoret, H., Haque, J. & Pascual-Leone, A. Increased variability of pacedfinger tapping accuracy following repetitive magnetic stimulation of thecerebellum in humans. Neurosci. Lett. (in press).

21. Ferbert, A. et al. Interhemispheric inhibition of the human motor cortex. J. Physiol. (Lond.) 453, 525–546 (1992).

22. Di Lazzaro, V. et al. Direct demonstration of interhemispheric inhibition ofthe human motor cortex produced by transcranial magnetic stimulation.Exp. Brain Res. 124, 520–524 (1999).

23. Wassermann, E. M., Wedegaertner, F. R., Ziemann, U., George, M. S. & Chen,R. Crossed reduction of human motor cortex excitability by 1-Hztranscranial magnetic stimulation. Neurosci. Lett. 250, 141–144 (1998).

24. Maeda, F., Keenan, J. P., Tormos, J. M., Topka, H. & Pascual-Leone, A.Modulation of corticospinal excitability by repetitive transcranial magneticstimulation. Clin. Neurophysiol. 111, 800–805 (2000).

25. Boroojerdi, B., Prager, A., Muellbacher, W. & Cohen, L. G. Reduction ofhuman visual cortex excitability using 1-Hz transcranial magneticstimulation. Neurology 54, 1529–1531 (2000).

26. Pascual-Leone, A. et al. Induction of visual extinction by rapid-ratetranscranial magnetic stimulation of parietal lobe. Neurology 44, 494–498(1994).

27. Fierro, B. et al. Contralateral neglect induced by right posterior parietal rTMSin healthy subjects. Neuroreport 11, 1519–1521 (2000).

28. Karnath, H. O., Niemeier, M. & Dichgans, J. Space exploration in neglect.Brain 121, 2357–2367 (1998).

29. Pouget, A. & Driver, J. Relating unilateral neglect to the neural coding ofspace. Curr. Opin. Neurobiol. 10, 242–249 (2000).

30. Weintraub, S. & Mesulam, M. M. Right cerebral dominance in spatialattention. Further evidence based on ipsilateral neglect. Arch. Neurol. 44,621–625 (1987).

31. Jewell, G. & McCourt, M. E. Pseudoneglect: a review and meta-analysis ofperformance factors in line bisection tasks. Neuropsychologia 38, 93–110(2000).

32. Fink, G. R., Driver, J., Rorden, C., Baldeweg, T. & Dolan, R. J. Neuralconsequences of competing stimuli in both visual hemifields: a physiologicalbasis for visual extinction. Ann. Neurol. 47, 440–446 (2000).

33. Vuilleumier, P., Hester, D., Assal, G. & Regli, F. Unilateral spatial neglectrecovery after sequential strokes. Neurology 46, 184–189 (1996).

34. Walsh, V., Ellison, A., Battelli, L. & Cowey, A. Task-specific impairments andenhancements induced by magnetic stimulation of human visual area V5.Proc. R. Soc. Lond. B Biol. Sci. 265, 537–543 (1998).

35. Kapur, N. Paradoxical functional facilitation in brain-behaviour research. Acritical review. Brain 119, 1775–1790 (1996).

36. Oldfield, R. C. The assessment and analysis of handedness: the Edinburghinventory. Neuropsychologia 9, 97–113 (1971).

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Because of an advertising production error, the 'Recruitment & Events' pages in the July issues of Nature Neuroscience and NatureReviews Neuroscience implied incorrectly that the Burnham Institute in La Jolla, California, and the Center for Neuroscience atUniversity of California-Davis are recruiting new directors. In fact, neither position is open. The new Director for the Center forNeuroscience and Aging at The Burnham Institute is Stuart A. Lipton, and the Scientific Director of the Institute is John Reed. TheDirector of the Center for Neuroscience at UC Davis is Edward Jones. We apologize to all concerned for any confusion or embar-rassment that this error may have caused.

Fig. 2. TMS sites. The left SMG and right ANG sites are detailed in the left and rightcolumns, respectively. The top row shows frameless stereotaxic recording of the TMS coilposition14 during an experimental session for a single subject. The MRI brain sections arein an approximately parasagittal orientation. The red and green markers indicate the TMScoil position (outside the parasagittal section) for left SMG (left) and right ANG (right). Ineach case, a yellow line extends from the center of the coil, along the direction of maxi-mum field intensity, into the underlying cortex. The middle row shows vitamin E capsulesplaced over the left SMG and right ANG sites (insets, left and right) in an MRI, taken afteran experiment, and registered into standard space33,34. A red + sign is placed at the centerof the underlying brain area. The main pictures show the same brain areas in sagittal sec-tions. The main axis of the coil was in the sagittal plane at each of the positions, which wereapproximately 2 cm lateral of one another. The electrical field intensity (shown by thecurved yellow lines) induced by the ANG stimulation would be reduced by a third in thesagittal plane of the SMG50. Similarly, the field induced by SMG stimulation would bereduced by a third in the sagittal plane of ANG. Bottom row, mean MRI scan of nine sub-jects, after registration into standard space. The sagittal sections are taken at the meanmedio-lateral position of the SMG (x = –52) and ANG (x = 41) sites. Yellow circles indicatethe rostro-caudal and dorso-ventral positions of the centers of the stimulated brain areasin individual subjects superimposed on these sagittal planes. Although the anatomical land-marks are less clear in the averaged MRI scan, it is clear that all the SMG sites clusterbetween the anterior intraparietal sulcus and the post-central sulcus, and are therefore inthe supramarginal gyrus. The ANG sites cluster in the angular gyrus, around the posterior,superior temporal sulcus as it runs through the angular gyrus.


errata

Complementary localization and lateralization of orienting and motor attentionMatthew F. S. Rushworth, Amanda Ellison and Vincent WalshNat. Neurosci. 4, 656–661 (2001)

The lettering in Figure 2 reproduced poorly. The correct figure appears below. We regret the error.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Because of an advertising production error, the 'Recruitment & Events' pages in the July issues of Nature Neuroscience and NatureReviews Neuroscience implied incorrectly that the Burnham Institute in La Jolla, California, and the Center for Neuroscience atUniversity of California-Davis are recruiting new directors. In fact, neither position is open. The new Director for the Center forNeuroscience and Aging at The Burnham Institute is Stuart A. Lipton, and the Scientific Director of the Institute is John Reed. TheDirector of the Center for Neuroscience at UC Davis is Edward Jones. We apologize to all concerned for any confusion or embar-rassment that this error may have caused.

Fig. 2. TMS sites. The left SMG and right ANG sites are detailed in the left and rightcolumns, respectively. The top row shows frameless stereotaxic recording of the TMS coilposition14 during an experimental session for a single subject. The MRI brain sections arein an approximately parasagittal orientation. The red and green markers indicate the TMScoil position (outside the parasagittal section) for left SMG (left) and right ANG (right). Ineach case, a yellow line extends from the center of the coil, along the direction of maxi-mum field intensity, into the underlying cortex. The middle row shows vitamin E capsulesplaced over the left SMG and right ANG sites (insets, left and right) in an MRI, taken afteran experiment, and registered into standard space33,34. A red + sign is placed at the centerof the underlying brain area. The main pictures show the same brain areas in sagittal sec-tions. The main axis of the coil was in the sagittal plane at each of the positions, which wereapproximately 2 cm lateral of one another. The electrical field intensity (shown by thecurved yellow lines) induced by the ANG stimulation would be reduced by a third in thesagittal plane of the SMG50. Similarly, the field induced by SMG stimulation would bereduced by a third in the sagittal plane of ANG. Bottom row, mean MRI scan of nine sub-jects, after registration into standard space. The sagittal sections are taken at the meanmedio-lateral position of the SMG (x = –52) and ANG (x = 41) sites. Yellow circles indicatethe rostro-caudal and dorso-ventral positions of the centers of the stimulated brain areasin individual subjects superimposed on these sagittal planes. Although the anatomical land-marks are less clear in the averaged MRI scan, it is clear that all the SMG sites clusterbetween the anterior intraparietal sulcus and the post-central sulcus, and are therefore inthe supramarginal gyrus. The ANG sites cluster in the angular gyrus, around the posterior,superior temporal sulcus as it runs through the angular gyrus.


errata

Complementary localization and lateralization of orienting and motor attentionMatthew F. S. Rushworth, Amanda Ellison and Vincent WalshNat. Neurosci. 4, 656–661 (2001)

The lettering in Figure 2 reproduced poorly. The correct figure appears below. We regret the error.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Documents

Nature Neuroscience September 2001