Nature Neuroscience May 2001

editorialListening to postdocs........................................................................................451

letters to the editorDo ‘smart’ mice feel more pain, or are they just better learners?.........................453

news and viewsFaces and places: of central (and peripheral) interest.........................................455Nancy KanwisherSEE ARTICLE, PAGE 533

Beam me up, Scottie! TREK channels swing both ways.......................................457James Maylie and John P. AdelmanSEE ARTICLE, PAGE 486

Brain glucosensing and the KATP channel...........................................................459Barry E. Levin, Ambrose A. Dunn-Meynell and Vanessa H. RouthSEE ARTICLE, PAGE 507

Representing retinal image speed in visual cortex...............................................461Eero P. Simoncelli and David J. HeegerSEE ARTICLE, PAGE 526

book reviewParallel tales of a changing brain........................................................................463The Dying of Enoch Wallace: Life, Death and the Changing Brainby Ira B. BlackREVIEWED BY BRADLEY T. HYMAN

contents

http://neurosci.nature.com

volume 4 no 5 may 2001

In early visual cortical areas,inputs from adjacent retinal loca-tions are known to project toneighboring regions of cortex.Rafael Malach and colleaguesnow find such retinotopic maps inobject-related occipito-temporalcortex, suggesting that mostobject representations are alsoorganized with respect to retinalposition. See pages 455 and 533.

nature neuroscience • volume 4 no 5 • may 2001 i

Purkinjecell Climbing

fiber

InferiorolivePrecerebellar

nuclei

Premotorareas

Deepcerebellar

nuclei

Mossyfibers

Parallel fibers

Granulecell

Inhibitoryinterneuron

Climbingfiber

collateral

Plasticity in the cerebellum.Page 467.

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 paid at New York, NY. Executive Officers of Nature America Inc: Annette Thomas, President; Edward Valis, Secretary & Treasurer. Printed by Publishers Press, Shep-herdsville, KY, USA. Copyright ©2001 Nature America Inc.

A candidate sweet taste receptor.

Page 492.

©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

brief communicationsRegeneration of CNS axons back to their target following treatment of adult rat brain with chondroitinase ABC............................................................465L D F Moon, R A Asher, K E Rhodes and J W Fawcett

reviewBeyond parallel fiber LTD: the diversity of synaptic and non-synaptic plasticity in the cerebellum................................................................................467C Hansel, D J Linden and E D’Angelo

articlesProtein mobility and GABA-induced conformational changes in GABAAreceptor pore-lining M2 segment......................................................................477J Horenstein, D A Wagner, C Czajkowski and M H Akabas

KCNK2: reversible conversion of a hippocampal potassium leak into a voltage-dependent channel...............................................................................486D Bockenhauer, N Zilberberg and S A N GoldsteinSEE NEWS AND VIEWS, PAGE 457

A candidate taste receptor gene near a sweet taste locus....................................492J P Montmayeur, S D Liberles, H Matsunami and L B Buck

The scaffold protein, Homer1b/c, regulates axon pathfinding in the central nervous system in vivo............................................................................499L Foa, I Rajan, K Haas, G Y Wu, P Brakeman, P Worley and H Cline

ATP-sensitive K+ channels in the hypothalamus are essential for the maintenance of glucose homeostasis.................................................................507T Miki, B Liss, K Minami, T Shiuchi, A Saraya, Y Kashima, M Horiuchi, F Ashcroft, Y Minokoshi, J Roeper and S SeinoSEE NEWS AND VIEWS, PAGE 459

Joint-encoding of motion and depth by visual cortical neurons: neural basis of the Pulfrich effect........................................................................513A Anzai, I Ohzawa and R D Freeman

Learning to see: experience and attention in primary visual cortex.....................519R E Crist, W Li and C D Gilbert

Speed skills: measuring the visual speed analyzing properties of primate MT neurons..........................................................................................526J A Perrone and A ThieleSEE NEWS AND VIEWS, PAGE 461

Center–periphery organization of human object areas.......................................533I Levy, U Hasson, G Avidan, T Hendler and R MalachSEE NEWS AND VIEWS, PAGE 455

Musical syntax is processed in Broca’s area: an MEG study..................................540B Maess, S Koelsch, T C Gunter and A D Friederici

Effect of subjective perspective taking during simulation of action: a PET investigation of agency.............................................................................546P Ruby and J Decety

classified advertising see back pages

nature neuroscience • volume 4 no 5 • may 2001 ii

contents

Homer and axon pathfinding errors.

Page 499.

Distinguishing self-produced actions.

Page 546.

Speed tuning in MT neurons.Pages 461 and 526.

©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 5 • may 2001 451

Many groups have expressed concern about the low pay andpoor job prospects for young biomedical scientists in the US; atlast, there are signs that the message is being heard. TheNational Academy of Sciences (NAS), whose mandate is toadvise the US Congress on matters of science policy, last yearpublished a report, Addressing the nation’s changing need for bio-medical and behavioral scientists, which made a number of rec-ommendations. The NIH has now responded to this report,and there is both good news and bad news in their reply. Thegood news is a promise of substantially larger stipends for bothgraduate students and postdocs. The bad news is that the NIHposition does not address some of the deeper structural prob-lems that underlie the current concerns.

The present funding situation for students and postdocsreflects a fundamental tension between two aims. In one view,the purpose of these programs is to train young scientists to pre-pare them for independent research careers. The other view isthat they are hired hands, paid to research a specific field becauseit is in the national interest. This dichotomy is embodied in twodifferent funding mechanisms, with different philosophies.

The first is the National Research Service Award (NRSA) pro-gram, whose explicit purpose is to train young scientists to becomeindependent investigators. NRSAs are awarded to individuals,either directly or via institutional training grants; they are com-petitive (applicants are evaluated on their track record, the strengthof the proposal, and the overall training environment) and pres-tigious. As the NAS report describes, the number of NRSAs hasbeen deliberately limited over the years, to ensure that the supplyof newly trained scientists is matched to the likely demand.

Although the NRSA program has remained almost constant insize since its inception in 1974 (it currently makes around 15,000awards per year), the last quarter-century has seen a large increasein the number of students and postdocs funded by a secondmechanism, namely support from research grants. The fundingfor these positions has a fundamentally different rationale; aresearch grant is awarded for a specific research project, andapplications are evaluated on the project’s perceived importanceand likelihood of success, not on its value in training the peoplewho will be hired to work on it. Unlike the NRSA program, thereis no attempt to control the numbers of young scientists who arefunded through research grants, and whereas NRSAs are openonly to US citizens and permanent residents, there are no suchrestrictions on grant-based funding.

The NIH has now announced a plan to make substantialincreases in the NRSA stipends. The values of these stipends werenever high, and over the years the postdoctoral stipend in par-ticular has been steadily eroded by inflation (recent increasesnotwithstanding). The current rates for graduate students andfirst-year postdocs—$16,500 and $28,260 respectively—are fru-gal to say the least, and in the context of expensive cities such as

San Francisco or Boston, they are little short of dismal. But theNIH plans to increase these by 10–12% per year until they reachnew targets of $25,000 and $45,000, after which their value will bemaintained through annual cost-of-living adjustments.

This is welcome news, not only for NRSA recipients but alsofor young scientists generally. The NRSA scale has historicallybeen a benchmark for setting grant-based salaries as well as pri-vate fellowships; although universities are not obliged to followthe NRSA scale, they are likely to do so to avoid disparities, andthe NIH will encourage grant applicants to take account of thenew figures when setting their own budgetary requests.

The NIH should be applauded for taking a leadership role inboosting the income of young scientists. The applause will betempered, however, for at least two reasons. First, the NRSA pro-gram will remain closed to foreigners. The NAS argued that thereis no clear rationale for this policy; foreign scientists frequentlystay in US after they have completed their studies, and even ifthey return to their home countries, they are still likely to con-tribute to the growth of biomedical knowledge and global health.The NIH pays lip-service to these arguments, but seems unwill-ing to act on them. Admittedly, it may be difficult to persuadeCongress that they should be paying to train foreigners, but itcan be argued that the current system institutionalizes the ideaof foreign scientists as hired labor rather than future colleagues.Certainly, this perception is reinforced by the current salaries,which may be attractive by (say) Chinese standards but are unac-ceptably low to many Americans.

Second, neither the NAS nor the NIH seem inclined to con-front the hard question of how many new researchers the systemshould produce each year. The NAS report acknowledges that thesupply currently exceeds the demand; the US awards about 5400biomedical PhDs per year, whereas even by the year 2005, the NASestimates that demand for trained researchers will not rise muchabove 3000. Despite these estimates, however, the NAS recom-mended only stabilization—not reduction—in the number ofnew PhDs, on the grounds that any decrease might disrupt theresearch enterprise. Yet the NIH appears to reject even this mod-est recommendation, on three grounds: they argue that the NASprojections are unreliable, that the specific NAS proposal for con-trolling numbers (to shift resources away from research grantsand back into the NRSA program) is inappropriate, and thatresponsibility for graduate student and postdoctoral enrollmentslies not with the NIH but with universities.

Clearly it would be in nobody’s interests to see the researchenterprise collapse for lack of manpower. But nor should itbecome chronically dependent on young people being willing toinvest a decade or more, working long hours for low pay, in pur-suit of jobs that may never materialize. The fact that the NIH iswilling to raise stipends is certainly encouraging, but it is stillonly the first step toward real reform.

editorial

Listening to postdocs

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



letters to the editor

TO THE EDITOR—Whereas Wei and col-leagues in a recent Nature Neurosciencearticle1 claim that overexpression of theNMDA receptor subunit NR2B in theforebrain leads to an enhancement ofinflammatory pain, the data presenteddo not necessarily warrant such a claim;instead, the data may suggest that thesetransgenic mice are better at remember-ing painful events, and that they learnhow to detect certain cues for avoidingpotential new injury.

First, these authors did not find anyalteration in pain sensitivity and noci-ceptive threshold in transgenic mice inthree types of measurements (tail-flickertest, hot-plate test and cold-plate test).These findings are consistent with datafrom our laboratory2 that these miceexhibited indistinguishable responses tomild foot shock, therefore suggesting thatNR2B mice have a normal pain sensitiv-ity and threshold.

Second, the authors examined for-malin-injection-induced responses inmice. Injection of 5% formalin producedindistinguishable responses (licking) inboth phase I (initial 10 minutes) andphase II (10–55 minutes), but with morebehavioral response at phase III (55–120minutes). However, judging from theirFig. 6a, it seems that the NR2B trans-genic and wild-type mice exhibited asimilar degree of residual lickingresponses in the second half of phase III(between 90–120 minutes). This suggeststhat the transgenic mice did not feel dif-ferently than the wild type mice at thislate stage, thus providing unfavorableevidence to the authors’ own conclusionthat raises doubts about the conclusionthat the transgenic mice suffered moreinflammatory chronic pain.

Finally, experiments using mechani-cal non-noxious stimuli may represent agood indication that these transgenicmice are indeed better learners. Typicallynon-painful, mechanical poking of vonFrey fiber to the dorsum of a hind pawelicited no response in animals. However,at one or three days after complete Fre-und’s adjuvant (CFA) injection to thehind paw, wild-type animals respondedto the stimulation of the injected (ipsi-lateral) hind paw, but not much to thecontralateral (un-injected) hind paw.With the same protocol, NR2B mice

Moreover, our recent study has demon-strated that the activation of NMDAreceptors in CA1 neurons is necessary forconverting short-term memories intolong-term memories3.

Therefore, it seems that the behav-ioral responses reported by Wei and col-leagues may be due to the enhancedlearning and memory capacity in NR2Bmice. Furthermore, their results nicelyextended our original finding that thesemice are better in learning and memoryregardless of whether experiences arepleasant or unpleasant. In fact, remem-bering bad experiences should be evolu-tionarily meaningful so that animals canavoid harmful situations or predators,thus increasing their survival.

Yaping Tang, Eiji Shimizu and Joe Z. TsienDepartment of Molecular Biology, PrincetonUniversity, Washington Road, Princeton,New Jersey 08544-1014, USAe-mail: [email protected]

REPLY—Something more than clevernesscaused forebrain-targeted NR2B overex-pressing mice to respond so robustly toinflammatory stimuli1. Tang et al. rightlyindicate that transgenic and wild-typemice were no different in tests of acutepain. However, they fail to recognize thesignificance of a selective enhancement ofinflammatory pain in the transgenics.Acute pain relies on intact reflex circuit-ry, whereas inflammatory pain-inducedbehavioral responses involve NMDAreceptor-mediated synaptic plasticity incentral modulatory pathways (for review,see ref. 4). In the latter category, formalinand CFA tests are widely used (for exam-ple, see refs. 5–7).

Tang and colleagues’ technical con-cerns about the formalin test are notvalid. The fact that phase III responsesrose and fell in magnitude during theperiod of observation (a well-establishedobservation) is irrelevant to the findingthat those responses were greatlyenhanced in transgenic mice. Further-more, the demonstration that formalin’seffects were dose dependent hardly dis-credits our study, particularly becauseformalin-induced central synaptic plas-ticity depends on the activity level ofinput fibers8,9. The differences between

showed 30% more paw withdrawalresponses than wild-type animals inresponse to stimulation of the injectedhind paw. The authors conclude that thisis evidence that the mice developedenhanced pain sensitivity.

However, such a conclusion is imme-diately called into question by one of theauthors’ additional experiments: NR2Bmice also exhibited the same 30% increasein enhanced paw withdrawal responses tothe poking of the contralateral hind paw(un-injected side)! Such a non-specificwithdrawal behavior is analogous to awell-known human behavior in which achild who cried after receiving a flu shota year ago will likely cry once he walksinto the doctor’s office and sees thesyringe and needles. This type of behav-ioral response is not because the child hasdeveloped increased pain sensitivity orfeels more pain, but simply reflects thefact that environmental cues re-activateunpleasant memories about the previouspainful event; crying (or running out ofthe clinic) is a defensive behavioralresponse as a result of memory retrieval.Therefore, the paw withdrawal in responseto the mechanical stimulation of the unin-jured site strongly suggests that theseNR2B mice remembered the bad experi-ence better. Moreover, they are better ableto generalize and recognize these cues inorder to avoid potential new assaults. Tolink the non-specific withdrawal respons-es to the enhancement of pain perception,the authors need to use in vivo recordingtechniques to show that non-painful pok-ing to the uninjected paw is now capableof eliciting enhanced neural activities inCNS pain perception centers.

A possible effect of learning and mem-ory on paw withdrawal behavior is furthersupported by the authors’ own observa-tion that nociceptive stimuli producedstronger neuronal activation in hip-pocampal CA1 and CA3 regions in thetransgenic animals (Fig. 5 of ref. 1). It iswell known that the hippocampus is notinvolved in pain perception, but is a keystructure in the formation of memoriesabout people, places and events. Theenhanced activation (400% increase) ofhippocampal CA1 and CA3 cells is con-sistent with the idea that these transgenicmice produced stronger memories aboutthe occurrence of the injury or insults.

Do ‘smart’ mice feel more pain, or are they just betterlearners?

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


454 nature neuroscience • volume 4 no5 • may 2001

letters to the editor

transgenic and wild-type mice in the for-malin test were significant and mean-ingful, and we challenge Tang et al. toconnect learning with increased lickingand biting of an injured hind paw.

It is also wrong to assert that trans-genic mice simply learned to avoidmechanical stimulation after CFA injec-tion. Our experiment has nothing to dowith why a child may be scared of a doc-tor’s office. If a child repeatedly visits aclinic and gets a painful injection eachtime, then the child may learn to associatethe clinic with pain, by classical condi-tioning. By contrast, how could the micehave known that the von Frey filament,which had never caused pain before,would suddenly become painful the firsttime it was applied after CFA injection?Mice had no chance to learn this. A betteranalogy is to the way an inflamed patchof skin may feel unusually tender to thelight touch of clothing: putting on clothesis never associated with pain until sud-denly, after induction of inflammation,the experience may become painful, not

a sensory ‘memory.’ NMDA receptorslikely contribute to synaptic plasticity inparts of the forebrain dedicated to painprocessing, much as they do in the hip-pocampus. We believe that this com-monality explains the alteration in bothbehavioral learning and sensitivity toinflammatory pain in NR2B-overex-pressing mice. Future studies will beneeded to determine the molecularmechanisms by which the forebrain con-tributes to inflammatory pain.

Geoffrey A. Kerchner, Feng Wei,Guo-Du Wang, Susan J. Kim, Hai-Ming Xu, Daphné A. Robinson,Ping Li, Zhou-Feng Chen and Min ZhuoDepartments of Anesthesiology, Anatomy &Neurobiology, Psychiatry, and MolecularBiology and Pharmacology, WashingtonUniversity Pain Center, WashingtonUniversity School of Medicine, St. Louis,Missouri, 63110, USAe-mail: [email protected]

1. Wei, F. et al. Nat. Neurosci. 4, 164–169 (2001).

2. Tang, Y. P. et al. Nature 401, 63–69 (1999).

3. Shimizu, E. et al. Science 290, 1170–1174(2000).

4. Wall, P. D. & Melzeck, R. The Textbook of Pain3rd edn. (Churchill Livingstone, New York,1994).

5. Ji, R. R., Baba, H., Brenner, G. J. & Woolf, C. J.Nat. Neurosci. 2, 1114–1119 (1999).

6. Ren, K., Hylden J. L., Williams, G. M., Ruda,M. A. & Dubner, R. Pain 50, 331–344 (1992).

7. Calejesan, A. A., Ch’ang, M. H.-C. & Zhuo, M.Brain Res. 798, 46–54 (1998).

8. Taylor, B. K., Peterson, M. A. & Basbaum, A. I.J. Neurosci. 15, 7575–7584 (1995).

9. Kim, S. J., Calejesan, A. A., Li, P., Wei, F. &Zhuo, M. Brain Res. 829, 185–189 (1999).

10. Zhuo, M. & Gebhart, G. F. J. Neurophysiol. 67,1599–1614 (1992).

11. Zhuo, M. & Gebhart, G. F. J. Neurophysiol. 78,746–758 (1997).

12. Urban, M. O. & Gebhart, G. F. Proc. Natl.Acad. Sci. USA 96, 7687–7692 (1999).

13. Calejesan, A. A., Kim, S. J. & Zhuo, M. Eur. J.Pain 4, 83–96 (2000).

14. Wei, F., Xu, Z. C., Qu, Z., Milbrandt, J. &Zhuo, M. J. Cell Biol. 149, 1325–1333 (2000).

because learning has takenplace, but because changehas occurred in the sensoryexperience. We found thatthis effect was enhanced inNR2B-overexpressing mice.Indeed, nociceptive respons-es to non-noxious stimuli(allodynia) occurred ininjected and uninjectedhind paws (notably, for bothwild-type and transgenicmice); this reflects the factthat severe inflammation, asinduced by CFA, can inducebilateral changes in painsensitivity due to activationof higher-order brain cen-ters and bilaterally project-ing descending facilitatorypathways10–13.

Can pain trigger mem-ory formation? Tang et al.point out that hippocam-pal neurons were activated

by pain, which is not surprising, becausemore than a third of CA1 neuronsrespond directly to peripheral nocicep-tive stimulation14. Mice undoubtedlyremember painful situations, but thesememories did not influence theirresponses to tests of pain sensitivity.Indeed, when placed repetitively on ahot plate, both wild-type and NR2Btransgenic mice showed consistentresponse latencies over time (Fig. 1),suggesting that this test was insensitiveto any learning that may have occurred.Like the hot plate, behavioral responsesto CFA and formalin injection measuresensory experiences and not memory, ashighlighted by the fact that theseresponses require ongoing peripheralsensory input.

There was something special aboutinflammatory pain. We believe thatsynaptic plasticity in the forebrain influ-ences the sensory experience of mice afterformalin or CFA injection. This plasticityis not ‘learning’ in the strict sense putforth by Tang et al., but it might be called

Fig. 1. Mice showed stable behavioral responses in a hot plate.Wild-type and NR2B transgenic mice were subjected multipletimes, at 10-min intervals, to a hot plate test. No significantchange in response latency was observed in either group overtime. Points represent means ± standard error of the mean.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



news and views

Primates are highly visual animals, devot-ing as much as half their cerebral cortex tosolving visual problems such as objectrecognition and visually guided action. Inthe macaque visual cortex, researchershave discovered and characterized rough-ly 30 different areas, each with a distinctfunctional and/or anatomical profile1. Yetuntil the mid-1990s, little was knownabout the functional organization ofhuman visual cortex. This landscapehas changed dramatically with theinvention of functional magnetic res-onance imaging (fMRI), a powerfultool for the noninvasive mapping ofthe normal human brain.

Early fMRI investigations inhumans located the first few areas ofthe visual cortical pathway, corre-sponding to those previously chartedin the macaque2,3. The inferredhomologies between human andmacaque visual areas were basedlargely on a long-known property ofvisual cortex called ‘retinotopy,’ inwhich nearby regions in the visualimage are represented by nearbyregions in the brain, producing a‘map’ of the visual field in each cor-tical area. Retinotopic cortex seemsto favor polar coordinates, witheccentricity (distance from the centerof gaze) mapped onto one corticalaxis, and polar angle mapped onto aroughly perpendicular axis.

More recent investigations have iden-tified several higher-level regions of cor-tex not characterized previously in themacaque. These areas are defined by a dis-tinctive functional profile, such as a selec-tive response to images of faces4,5 orplaces6,7. A new study by Levy and col-leagues8 in this issue shows that two of

was increased with eccentricity, enablingsubjects to identify the objects despite thelower acuity of the peripheral field.

As in earlier reports4,5,9, Levy and col-leagues8 found stronger responses to facesthan houses in two regions: area LO (lat-eral occipital cortex) and the fusiform facearea (FFA)5. They also found a third pre-viously described region6,7 that respond-ed more to houses than to faces. This

region included the parahippocam-pal place area (PPA), whichresponds to a wide variety of imagesdepicting places, including a maxi-mal response to indoor and outdoorscenes, and a weaker but still strongresponse to pictures of houses6.

The new finding8 is that the face-selective regions responded morestrongly to central than to peripher-al objects, whereas the house-selec-tive regions had the oppositepreference. Thus, category selectivityand eccentricity bias co-exist in thesame region of cortex, with faceselectivity associated with a center-field bias and place selectivity with aperipheral-field bias.

Levy and colleagues also showedthat face and place selectivity is notsimply an artifact of the use of smallface stimuli and larger house stim-uli: face-selective regions respondedmore strongly to large face images(that extend into the visual periph-

ery) than to small house images (that donot), and house-selective regions preferthe reverse. This result further demon-strates that when category selectivity andvisual field biases are pitted against eachother, category selectivity wins.

One puzzle raised by these findings iswhy the center-field bias for face-selectivecortex is apparently not mirrored by asimilar center-field bias in face discrimi-nation performance (beyond thatobserved for simpler visual stimuli)10.

More generally, why should categoryselectivity and eccentricity bias be associ-

these dimensions along which visual cor-tex is organized—category selectivity andeccentricity bias (a preferential responseto either central or peripheral stimuli)—not only co-exist in the same corticalregion, but have a systematic relationshipto each other. Their findings may haveimplications for our understanding notonly of the functions of these corticalregions, but also of their origins.

In the new study, Levy and colleaguesfirst identified regions of cortex thatresponded more strongly when the sub-jects viewed faces than when they viewedhouses, and other regions with the oppo-site property. Second, to identify corticalregions that respond differentially to stim-uli presented at the center of gaze versusthe periphery, the authors scanned sub-jects while they viewed a set of tiny cen-trally located images of common objects,or a ring of the same object stimuli aroundthe center of gaze presented at either amedium or large eccentricity. Object size

Faces and places: of central (andperipheral) interestNancy Kanwisher

Early visual cortical areas are organized in retinotopic coordinates. Levy and colleagues nowreport that category-selective regions show a similar organization, with face-selective areasresponding more to central stimuli and place-selective areas responding to peripheral stimuli.

Nancy Kanwisher is in the Department of Brainand Cognitive Science, MIT, NE 20-454, 3 Cambridge Center, Cambridge, Massachusetts02139, USA.e-mail: [email protected]

Fig. 1. Eye movement scanpath (in red) of a person viewing apainting by Rein; note the tendency to fixate on the faces.Adapted from Yarbus (Eye Movements and Vision, Plenum, NewYork, 1967).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


456 nature neuroscience • volume 4 no 5 • may 2001

news and views

ated in visual cortex in the first place? Anatural interpretation of these resultsdraws on a currently popular view that thecortex is highly plastic, and that its func-tional organization is shaped substantiallyby experience. If the stimulus selectivitiesand receptive field locations of many visu-al neurons are learned rather than speci-fied innately, the association reported byLevy and colleagues8 might follow direct-ly from the statistics of everyday percep-tual experience. The authors suggest thatwe generally fix our eyes on faces directly(Fig. 1), whereas in place perception moreof the relevant visual information comesfrom the periphery. Thus this associationmight be a direct mirror of experience.

A strong version of this experientialhypothesis would predict that if a personwere raised seeing (or attending to) facesonly in the periphery and places only atthe center of gaze, his or her cortex mightshow the reverse of the pattern reportedby Levy and colleagues8. Such a resultmight enable us to further determinewhich dimension—category selectivity oreccentricity bias—is more fundamental indetermining the organization of high-levelvisual cortex. Would the hypothesized faceand periphery-selective region of cortexbe located in the usual site of the face area,or in the usual site of the (normally place-selective) periphery-biased cortex, or insome new location? Although this partic-ular experiment could never be done, itmay be possible to test for sensitivity toexperience in adulthood by finding peo-ple with unusual visual experience. Forexample, what happens to the locus offace-selective cortex when central visionis impaired by macular degeneration?

If the experiential hypothesis sounds far-fetched, consider the following strands ofevidence. First, after monkeys receive exten-sive training recognizing originally novelstimuli, some neurons in their visual cor-tex respond selectively to these learnedstimuli10,11. Crucially, recent findings sug-gest that such neurons may respond selec-tively not only to the trained stimuli, butalso to the retinal positions where the stim-uli were presented during training, and per-haps even to the exact conjunction of aparticular stimulus and the location whereit was presented (J.J. DiCarlo and J.H.R. Maunsell, Soc. Neurosci. Abstr., 26,498.2, 2000). Second, an elegant psy-chophysical study12 found that after peopleare trained to discriminate certain novelvisual stimuli presented in fixed retinal loca-tions, performance drops when the samestimuli are subsequently presented at newuntrained locations. Why would the visual

ple, eccentricity biases might be geneti-cally specified, with selectivity for faces orplaces then developing near the regionsthat provide their relevant inputs. (Alongthese lines, it has been argued14 that cor-tical regions responsive to tools developnear the visual motion area MT.) Alter-natively, cortical neurons originallyresponsive to one stimulus property maysubsequently develop an additional selec-tivity for a second stimulus property thatis experientially associated with the first15.

In sum, Levy and colleagues’ new dataare suggestive, but they remain consistentwith a wide range of accounts of the ori-gins of the functional organization of high-level visual cortex. The good news is thatclever experimentation should enable us tonarrow the range of viable hypotheses.Indeed, at the recent meeting of the Cog-nitive Neuroscience Society (New York,2001), Malach reported that letter strings(like faces) also activate center-biasedregions of visual cortex. Because peoplehave only been reading for a few thousandyears—probably not enough time for evo-lutionary change—this pattern is unlikelyto be genetically specified, and is more like-ly to result from each individual’s experi-ence fixating words. Whether center-biasedface selectivity and periphery-biased placeselectivity arise in the same experientialfashion remains to be determined.

1. Felleman, D. J. & Van Essen, D. C. Cereb.Cortex 1, 1–47 (1991).

2. Tootell, R. B., Dale, A. M., Sereno, M. I. &Malach, R. Trends Neurosci. 19, 481–489(1996).

3. Engel, S. A. et al. Nature 369, 525 (1994).

4. McCarthy, G., Puce, A., Gore J. & Allison, T. J. Cogn. Neurosci. 9, 605–610 (1996).

5. Kanwisher, N., McDermott, J. & Chun, M. J. Neurosci. 17, 4302–4311 (1997).

6. Epstein, R. & Kanwisher, N. Nature 392,598–601(1998).

7. Aguirre, G. K., Zarahn, E. & D’Esposito, M.Neuron 21, 373–383 (1998).

8. Levy, I., Hasson, U., Avidan, G., Hendler, T. &Malach, R. Nat. Neurosci. 4, 533–539 (2001).

9. Gauthier, I., Skularski, P., Gore, J. C. &Anderson, A. W. Nat. Neurosci. 3, 191–197(2000).

10. Rovamo, J., Makela, P., Nasanen, R. &Whitaker, D. Invest. Ophthalmol. Vis. Sci. 38,1029–1039 (1997).

11. Logothetis, N. K. & Pauls, J. Cereb. Cortex 3,270–288 (1995).

12. Nazir, T. A. & O’Regan, J. K. Spat. Vis. 5,81–100 (1990).

13. Ullman, S. & Soloviev, S. Neural Networks 12,1021–1036 (1999).

14. Martin, A., Wiggs, C. L., Ungerleider, L. G. &Haxby, J. V. Nature 379, 649–652 (1996).

15. Sakai, K. & Miyashita, Y. Nature 354, 152–155(1991).

system use such inflexible and literal repre-sentations, rather than extracting represen-tations that are invariant to positionalchanges? Computational models of objectrecognition suggest that, counterintuitively,the most effective way to achieve position-al invariance may be to store representa-tions of fragments of familiar objects ateach spatial position13.

Despite this evidence in favor of theexperiential hypothesis, the ability of visu-al cortex to learn from experience is highlyconstrained. For one thing, neuropsycho-logical patients who selectively lose facerecognition abilities as a result of focal braindamage are rarely able to relearn this abili-ty, suggesting that the remaining visual cor-tex (which is adequate for visual recognitionof nonface objects) cannot be trained onface recognition in adulthood. Indeed, astudy reported at the Cognitive Neuro-science Society meeting last year (R. Le Grand, C. Mondloch, D. Maurer &H.P. Brent, San Francisco, April, 2000) sug-gested that the development of normal facerecognition depends on visual experienceextremely early in life. These researchersstudied people born with dense bilateralcataracts that precluded pattern vision untilsurgical correction between two and sixmonths of age. Surprisingly, these peoplenever develop normal configural process-ing of faces. That is, they are impaired atdiscriminating between faces that differ inthe relative positions of facial features, butunimpaired at discriminating faces on thebasis of individual face parts or on eithertask when the face stimuli are presentedupside-down. Thus, pattern vision in thefirst few months of life is necessary for thedevelopment of configural face processingas an adult; years of subsequent visual expe-rience with faces is not sufficient to bringabout normal face processing.

Although I have so far emphasized thecritical role of experience in the develop-ment of visual cortex, Levy and colleagues’findings8 are also consistent with hypothe-ses that place a much greater emphasis oninnate constraints. After all, it has been truethroughout primate evolution that face per-ception is done best with high-resolutionfoveal information, and place perception innatural environments usually requires awide field of view. Thus natural selectionmay have led to a genetic blueprint not onlyfor specific cortical machinery for face andplace perception, but also for the newlyreported visual field biases in these areas.

Levy and colleagues’ results are alsoconsistent with intermediate hypothesesthat incorporate critical instructive rolesfor both genes and experience. For exam-

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


news and views


The resting membrane potential sets thepoint of reference for the dynamics of theaction potential, but for many years thecurrents underlying the resting potentialin neurons were overlooked. It is now rec-ognized that many neurons contain ‘back-ground’, or ‘leak’, potassium channels thatcontrol the resting membrane potentialand are closed by neurotransmitters. Inthis issue, Steve Goldstein and colleagues1

provide insight into the versatility of back-ground channels, and offer a new mech-anism for broadening the diversity ofpotassium channel function. They showthat phosphorylation of native or clonedKCNK2 (TREK1) ‘background’ potassi-um channels by protein kinase A (PKA)reversibly converts them into voltage-dependent channels.

Background K+ channels are open atrest and voltage independent. Their inhi-bition induces a profound depolarizationand an increased probability of actionpotential discharge. Background K+ cur-rents in different cell types are closed by awide range of neurotransmitters2,3. Despitea growing appreciation of their impor-tance, the mechanisms through which neu-rotransmitters inhibit background K+

channels are not well understood. Perhapsthe best-understood example of a back-ground K+ current comes from a behav-ioral model in Aplysia, where sensitizationof the gill-withdrawal reflex with a noxiousstimulus to the tail results from presynap-tic facilitation between the sensory and themotor neurons of the reflex pathway. Sero-tonin released from interneurons activatesPKA, which inhibits a background, K+

inhibition12. Although TREK1 channelshave not been unambiguously identifiedin intact brain preparations, Goldsteinand colleagues1 examined native channelsin an immortalized hippocampal cell lineand found TREK1-like channels (andTREK1 mRNA) only after differentiationinto the neuronal phenotype. These chan-nels had the same biophysical attributesas cloned TREK1 channels expressed inXenopus oocytes. PKA and alkaline phos-phatase modulated native or cloned chan-nels examined in inside-out patchessimilarly. Dephosphorylated channelsshowed a high, voltage-independent Po(open probability), being open approxi-mately 80% of the time at either –100 mVor 100 mV, whereas treatment with PKAand ATP reduced the Po dramatically.Importantly, the Po of phosphorylatedchannels showed a clear voltage depen-dence that was insensitive to the potassi-um equilibrium potential (EK), being∼ 100-fold more likely to open at 100 mV than at –100 mV. TREK1 con-tains a single consensus site for PKAphosphorylation, within the intracellularC-terminal domain following the fourthtransmembrane segment, which has beenshown to underlie PKA-mediatedresponses12,13. Replacement of serine 348with alanine resulted in ‘background’channels that were unaffected by PKA,whereas a negatively charged amino acid,aspartate, produced voltage-dependentchannels that were unaffected by phos-phatase treatment1.

To those who have spent the pastdecade examining the biophysical andmolecular gymnastics underlying the gat-ing mechanisms of various classes ofpotassium channels, the results presentedby Goldstein and colleagues1 seem hereti-cal, a kind of cross-kingdom fertilization.An accepted tenet might be paraphrased,“Once a voltage-(in)dependent channel,always a voltage-(in)dependent channel.”The cumulative results from many labo-ratories create an exquisitely detailed mol-ecular dynamic description of voltage-dependent gating. All voltage-dependentchannels have among their six transmem-brane domains a specialized S4 segmentdecorated with positively charged aminoacids that respond to voltage by reorient-ing their side chains and mobilizing thetransmembrane helix to rotate up andaway from the tightly packed closed state.This controlled disruption induces a waveof coordinated secondary rearrangementsthat spread throughout the molecule,resulting in pore opening. Inward rectifierchannels with two transmembrane

conductance, the S-current, in presynap-tic terminals. This prolongs action poten-tial duration, allowing more calcium toflow into the terminal, and increasing neu-rotransmitter release4–6.

During the past decade, dozen ofpotassium channel subunits have beencloned and categorized based on theirstructural, biophysical and pharmacolog-ical characteristics. This has engendereda new generation of experiments aimed atmatching cloned channels with nativeconductances. Although this effort hasbeen exciting and largely successful, back-ground channels have escaped molecularidentification until recently. In the pastseveral years, our understanding of thediversity of structural motifs underlyingK+ channels has expanded with thecloning of member of the two-pore-domain family. In these channels, eachsubunit polypeptide is a tandem arrange-ment of two inward rectifier-like motifs,each with two segments bracketing apore-forming domain. In contrast to theother families, in which four subunitscoassemble into functional channels, thetwo-pore-domain subunits form four-foldsymmetrical channels from the assemblyof two subunit polypeptides, a dimer ofdimers. To date, twelve mammalian two-pore-domain subunits have been cloned.Although they share remarkably low pri-mary sequence homology, they demon-strate voltage independence andnonrectifying conduction in symmetricalK+ solutions7,8. Recent reports have con-vincingly demonstrated that at least oneof the cloned two-pore-domain K+ chan-nels, TASK1, represents well-characterizednative background K+ conductances inthree different central neurons9–11.

The report highlighted here featuresanother two-pore K+ channel, TREK1,which shares the hallmark characteristicsof the S-current, including PKA-mediated

Beam me up, Scottie!TREK channels swingboth waysJames Maylie and John P. Adelman

TREK1 was known as a voltage-independent ‘background’potassium channel, but a new study suggests that proteinkinase A can reversibly convert it to a voltage-dependent state.

John Adelman is in the Vollum Institute, andJames Maylie is in the Obstetrics andGynecology Dept.,Oregon Health SciencesUniversity, 3181 S.W. Sam Jackson Park Rd.,Portland, Oregon 97201-3098, USA.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



news and views

domains for each subunit, bracketing apore domain, could not possibly be volt-age dependent, lacking an S4 helix alto-gether. Lean and mean structuralmachines, these channels are geared forresponding to a completely different set ofmetabolic stimuli. Some (IRK-like chan-nels) are ‘open holes’ blocked by intracel-lular cations at positive potentials. Others(GIRK channels) respond to the touch ofG-proteins on their intracellular tails. KATPchannels are gated by the relatively simplemetabolite, ATP, which works in a com-plex coordinated fashion with a large βsubunit, the sulfonylurea receptor. At firstblush, the two-pore channels, includingTREK1, appear as simple tandems of twoinward rectifier motifs, and would not beexpected to have voltage dependenceunder any circumstances.

Therefore, one of the most intriguingquestions that arise from the Goldsteinpaper is the mechanism underlying volt-age dependence in the phosphorylatedTREK1 channels. One possibility is thatthe phosphate somehow blocks the chan-nel pore directly. However, initial resultsindicate that this intuitively unlikelyprocess, a negative charge being accom-modated in a pore that is cation-selective,is not responsible. A pore-occlusionmechanism from a C-terminal block inwhich voltage has a stronger role than ionoccupancy cannot yet be entirely ruledout, however. As Goldstein and colleaguespoint out, the gating mechanism conver-sion observed for TREK1 channels maybe analogous to the ‘willing’ (easilyopened) and ‘reluctant’ (harder to open)states visited by voltage-dependent calci-um channels. If this is the case, then chan-nel biophysicists may rest easier becauseit would mean that TREK1 channels areintrinsically voltage dependent, and in theunphosphorylated state have a very neg-ative V1/2. Alternatively, the presence ofthe negative charge may introduce a con-formational alteration in the subunit pro-tein that closes the pore and exposescharged residues that reside in the mem-

augmented. Simultaneously, phosphory-lated TREK1 channels emerge that oper-ate at depolarized voltages only andfacilitate repolarization. The conversion ofone population of channels between thetwo gating modes will alter the opposingforces of depolarization and repolariza-tion, permitting a delicate, highly regulat-ed balance that is emphasized by the useof a single population of channels, per-mitting a direct intrinsic proportionalityto be imposed. The extent to which phos-phorylated TREK1 channels contribute tothese processes is difficult to assess, giventhe relatively low open probability at depo-larized potentials. Transgenic mice har-boring a single substitution of alanine forserine at the PKA target site in the mouseTREK1 gene would provide profoundinsight into this important question.Although the rapid conversion of openrectifier channels to voltage-dependentchannels may be unsettling for biophysi-cists, the plasticity of TREK1 channelstructure and function endows the neuronwith a previously unappreciated mecha-nism for regulating excitability.

1. Bockenhauer, D., Zilberberg, N. & Goldstein,S. A. N. Nat. Neurosci. 4, 486–491 (2001).

2. Weight, F. F. & Votava, J. Science 170, 755–758(1970).

3. Katayama, Y. & North, R. A. Nature 274,387–388 (1978).

4. Castellucci, V. & Kandel, E. R. Science 194,1176–1178 (1976).

5. Klein, M. & Kandel, E. R. Proc. Natl. Acad. Sci.USA 77, 6912–6916 (1980).

6. Siegelbaum, S. A., Camardo, J. S. & Kandel, E. R. Nature 299, 413–417 (1982).

7. Lesage, F. & Lazdunski, M. Am. J. Physiol.Renal Physiol. 279, F793–801 (2000).

8. North, R. A. Trends Neurosci. 23, 234–235(2000).

9. Millar, J. A. et al. Proc. Natl. Acad. Sci. USA 97,3614–3618 (2000).

10. Talley, E. M., Lei, Q., Sirois, J. E. & Bayliss, D. A. Neuron 25, 399–410 (2000).

11. Buckler, K. J., Williams, B. A. & Honore, E. J. Physiol. (Lond.) 525, 135–142 (2000).

12. Patel, A. J. et al. EMBO J. 17, 4283–4290(1998).

13. Maingret, F. et al. EMBO J. 19, 2483–2491(2000).

14. Backx, P. H. & Marban, E. Circ. Res. 72,890–900 (1993).

brane electric field to changes in mem-brane potential. At depolarized potentials,these side chains would move in a man-ner that permits pore opening. Detailedstructure–function experiments will like-ly resolve this mystery.

The effects of blocking TASK1 back-ground channels are clearly illustrated incerebellar granule neurons, where block-ade, presumably by phosphorylation, caus-es depolarization9. Similar, thoughquantitatively distinct, effects are likely inmany neuronal populations where othermembers of this family, such as TREK1, areexpressed. The results from Goldstein andcolleagues predict that in central neurons,as the membrane depolarizes, phosphory-lated TREK1 channels will contribute tothe voltage-dependent K+ channel orches-tra that coordinates repolarization, enhanc-ing recovery and re-firing (Fig. 1). TREK1is also expressed in heart, where the actionpotential contains a pronounced plateauphase following the peak. Here theunphosphorylated background channelsmay also supply a background K+ currentduring the action potential plateau14.Phosphorylation will enhance depolariza-tion from rest and also reduce the outwardplateau currents, by 86% for TREK1 chan-nels based on the relative Po values, in thedirection of prolonging the action poten-tial. Therefore, the functional consequenceof phosphorylation should be both a depo-larization of the resting membrane poten-tial, and a tendency to prolong the actionpotential and actually slow recovery.

The conversion of the background cur-rent into a voltage-dependent current sug-gests that phophorylation does not simplysilence the channels that hold the mem-brane potential near EK. It adds a new levelof fine-tuning and gives them new roles;phosphorylation of TREK1 channels addsa new component to the dynamic of theaction potential without the need for adedicated family of genes, or energyexpended to express them. Moreover, theresponse to PKA activation in vivo is prob-ably not like that seen when PKA is patch-applied, but more like a rheostat, tuning agraded response across the population ofTREK1 channels. As channels becomephosphorylated, the rise to threshold is

Fig. 1. At rest (left), unphosphorylated TREK1 channels help set the resting membrane potential.Neurotransmitter binding to receptors on the postsynaptic membrane activates second messengers,including PKA, which phosphorylates TREK1 channels, silencing them (second from left). As theaction potential depolarizes the membrane, phosphorylated TREK1 channels open and contribute torepolarization (third from left). As the membrane potential drops, voltage-dependent TREK1 activityis reduced (far right). Phosphatase activity may then recycle TREK1 channels back to the voltage-independent state, reintroducing them as background channels.

PKA

K+K+

K+

K+K+

K+K+

K+ Phosphate (–)

K+K+

K+

K+K+

K+

K+K+K+

K+K+K+ K+

K+K+

K+ K+

Bob

Cri

mi

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


used neurophysiology in slices of theVMH to show that Kir6.2–/–mice have nodemonstrable GR neurons. They exam-ined the VMH because it contains GR neurons and because it has beenimplicated in the control of both foodintake7 and the counter-regulatoryresponse to intracellular glucose defi-ciency (glucoprivation)8. Administrationof 2-deoxyglucose, a competitiveinhibitor of intracellular glucose metab-olism, produced glucoprivic feeding inwild-type (Kir6.2+/+) mice, but thisresponse was greatly attenuated inKir6.2–/– mice. On the other hand, theresponses to both leptin and neuropep-tide Y (NPY) were normal in the Kir6.2–/–

mice. Leptin, a hormone released by adi-pose tissue, inhibits food intake by act-ing, in part, on orexigenic (food intakestimulating) NPY neurons in the VMH.Indeed, leptin also activates the KATPchannel in VMH9. The demonstrationthat Kir6.2–/– mice had defective gluco-privic feeding but retained normalresponses to both leptin and NPY sug-gests that there is a clear dissociationbetween mechanisms underlying thephysiological control of intake and theemergency response to severe cellular glu-coprivation. It also suggests that the KATPchannel, and the action of leptin on thischannel, are not required for the controlof normal feeding. Further studies of theregulation of ingestive behavior and body

the KATP channel opens (activates), thecell membrane becomes hyperpolarized,and neuronal firing ceases (Fig. 1).

Miki and colleagues3 now show thatmice with deletion of the Kir6.2 pore-forming subunit of the KATP channel(Kir6.2–/–) both lack functional GR neu-rons in their ventromedial hypothalamus(VMH) and show severely impaired neu-rohumoral counter-regulatory responses(to mobilize glucose) and feedingresponses to glucoprivic stimuli. Thissuggests that there might be a causal linkbetween the KATP channel on GR neuronsand these physiological responses.

Populations of GR neurons are locat-ed in the hypothalamus (ventromedial,arcuate and paraventricular nuclei), mid-brain (substantia nigra), pons (locuscoeruleus) and medulla (nucleus of thesolitary tract)4–6. Miki and colleagues3

Fig. 1. Proposed role of the ATP-sensitive K+ (KATP) channel on glucose-responsive (GR) neurons.GR neurons contain KATP channels composed of the Kir6.2 pore-forming unit and a sulfonylureareceptor (SUR). Glucose is transported into the cell body and terminal by a glucose transporter,possibly Glut2. The rate of glycolysis is regulated by glucokinase (GK). Glycolytic and oxidative(mitochondrial) metabolism of glucose raises the ATP/ADP ratio, causing ATP to bind to the KATPchannel complex. This inactivates(closes) the channel, producingaccumulation of intracellular K+,membrane depolarization, influxof Ca2+ through a voltage-gatedCa2+ channel and increased neu-ronal firing. Nerve terminal KATPchannels can also lead to trans-mitter (glutamate and γ-aminobu-tyric acid, GABA) release. Duringstates of energy deficiency suchas hypoglycemia or the adminis-tration of 2-deoxyglucose, theATP/ADP ratio and ATP bindingto the channel would be reduced,placing the channel predomi-nately in the open state. Intracel-lular K+ would exit through theopen channel along a concentra-tion gradient, leading to cellmembrane hyperpolarization andcessation of firing. This wouldlower the metabolic demands onthe neuron.

news and views


The brain depends on a constant supplyof glucose to fuel its energetic demands,which may explain the evolution of sev-eral mechanisms for detecting and regu-lating levels of plasma glucose. Becausefood intake is the major source of glu-cose, Mayer proposed his ‘glucostatichypothesis,’ that food intake is increasedin response to low levels of plasma glu-cose1. The existence of specialized brainglucose-responsive (GR) neurons hasbeen known for many years2, but nodirect, causal link between a glucose-mediated change in their firing rate and aspecific physiological response such asfeeding has ever been established. In thisissue, Miki and colleagues3 show that anATP-sensitive K+ (KATP) channel is animportant link between GR neurons and‘glucoprivic feeding,’ that is, food intakethat occurs when glucose availability isseverely limited.

Most neurons use glucose to fueltheir activity. GR neurons also use glu-cose to regulate their firing rate by alter-ing the activity of the KATP channel. Thischannel is an octomeric protein com-posed of four pore-forming units for K+

(Kir6.1 or Kir6.2) and four receptors(SUR1, SUR2A or SUR2B) for sulfony-lureas, drugs used to increase insulinsecretion in type II diabetics. Uptake andintracellular metabolism of glucose leadsto an increase in the ATP/ADP ratio,which promotes ATP binding to thechannel complex. This inactivates (clos-es) the channel, leading to intracellularK+ accumulation followed by membranedepolarization, calcium influx andincreased cell firing. When glucose sup-ply is limited, the ATP/ADP ratio falls,

Brain glucosensing andthe KATP channelBarry E. Levin, Ambrose A. Dunn-Meynell and Vanessa H. Routh

An ATP-sensitive K+ channel in glucose-responsive neurons isshown to be required for the emergency response to severeglucose deprivation, but not necessarily for normal feeding.

The authors are in the Neurology Service(127C), VA Medical Center, E. Orange, NewJersey 07018, USA, and the Departments ofNeurosciences and Pharmacology andPhysiology, New Jersey Medical School(UMDNJ), Newark, New Jersey 07103, USA.e-mail: [email protected]

Glut2?

Glucose

Glucose

ATPADP

GK

Ca2

Sur

K+

Kir6.2

Sulfonylurea

+

Am

y C

ente

r

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



news and views

weight in Kir6.2–/– mice are required toshed light on this important issue.

It is curious that mammals haveevolved mechanisms for dealing withsevere hypoglycemia, a condition thatalmost never occurs in nature. Evolu-tionary origins aside, the counter-regu-latory response has gained considerableclinical significance since the advent ofinsulin therapy for diabetes mellitus.Episodes of inadvertent insulin-inducedhypoglycemia have become common-place as clinicians attempt to maintainever-tightening control of blood glucoselevels. Further, repeated bouts of hypo-glycemia lead to reduced awareness ofhypoglycemia and attenuation of thecounter-regulatory response10,11. Thisneurohumoral response is designed tomobilize every last ounce of the body’sglucose stores to prevent brain damage.Sympatho-adrenal activation releasesnorepinephrine and epinephrine fromsympathetic nerves and the adrenalmedulla. Corticosterone (in rodents) isreleased from the adrenal cortex, growthhormone from the pituitary andglucagon from the α-cells of the pan-creas. In Kir6.2–/– mice, the brain-medi-ated glucagon component of thecounter-regulatory response to hypo-glycemia is attenuated, and the ability tocorrect hypoglycemia is impaired. It isunclear how the remainder of thecounter-regulatory response is affectedby deletion of Kir6.2. No corticosteroneor growth hormone levels were mea-sured, and the epinephrine responseseemed to be intact. Because the path-ways mediating these other responsesand those regulating glucagon release arevery different, this is an important issue.Methodological problems make it diffi-cult to interpret the epinephrine responseto hypoglycemia. Both Kir6.2+/+ andKir6.2–/– mice had enormous elevationsof basal epinephrine levels (almost 50times those found in undisturbed ani-mals), which might have obscured anydeficit of hypoglycemia-induced epi-nephrine release in the Kir6.2–/– mice.Thus, additional studies in Kir6.2–/– miceare required to gain a full understandingof the role of the KATP channel in thiscomplex neurohumoral reaction.

The most important and unambigu-ous finding of this study is that VMHGR neurons require the Kir6.2 subunitof the KATP channel to sense glucose. Asthe authors point out, this channel is

there are other glucosensing mechanismsin the body that are likely to be involvedin the counter-regulatory response, butmay be independent of the KATP chan-nel. These include glucose-sensitive neu-rons, which increase their firing rate asglucose levels fall, and glucosensing ele-ments in the portal vein15. That bothglucoprivic feeding and the counter-reg-ulatory response are not totally abol-ished in Kir6.2–/– mice suggests thatthese other glucosensing mechanisms donot require the KATP channel.

In summary, the current studies inthe Kir6.2–/– mouse demonstrate that anintact KATP channel is critical to the func-tioning of GR neurons in the VMH. Inaddition, a functional KATP channelseems to be required to mount a normalcounter-regulatory response and toincrease food intake when intracellularglucose metabolism falls to pathologi-cally low levels. However, further studieswill be required before we can accept thecontention that VMH GR neurons arethe specific effectors of these responsesor that they are truly involved in the reg-ulation of blood glucose levels and foodintake under physiological conditions.

1. Mayer, J. N. Engl. J. Med. 249, 13–16 (1953).

2. Oomura, Y., Ono, T., Ooyama, H. & Wayner,M. J. Nature 222, 282–284 (1969).

3. Miki, T. et al. Nat. Neurosci. 4, 507–512(2001).

4. Mizuno, Y. & Oomura, Y. Brain Res. 307,109–116 (1984).

5. Levin, B. E. Brain Res. 874, 158–164 (2000).

6. Liss, B., Bruns, R. & Roeper, J. EMBO J. 18,833–846 (1999).

7. Anand, B. K. & Brobeck, J. R. Yale J. Biol. Med.24, 123–146 (1951).

8. Borg, W. P., Sherwin, R. S., During, M. J., Borg, M. A. & Shulman, G. I. Diabetes44, 180–184 (1995).

9. Spanswick, D., Smith, M. A., Groppi, V. E.,Logan, S. D. & Ashford, M. L. Nature 390,521–525 (1997).

10. Dagogo-Jack, S. E., Craft, S. & Cryer, P. E. J. Clin. Invest. 91, 819–828 (1993).

11. Tkacs, N. C., Dunn-Meynell, A. A. & Levin, B. E. Diabetes 49, 820–826 (2000).

12. Dunn-Meynell, A. A., Rawson, N. E. & Levin,B. E. Brain Res. 814, 41–54 (1998).

13. Lynch, R. M., Tompkins, L. S., Brooks, H. L.,Dunn-Meynell, A. A. & Levin, B. E. Diabetes49, 693–700 (2000).

14. Ritter, S., Dinh, T. T. & Zhang, Y. Brain Res.856, 37–47 (2000).

15. Adachi, A., Shimizu, N., Oomura, Y. &Kobashi, M. Neurosci. Lett. 46, 215–218(1984).

necessary but not sufficient to define aGR neuron. Only a small percentage ofneurons within a few select brain areasare GR neurons. However, both GR andnon-GR VMH neurons, indeed neuronsthroughout the brain, contain Kir6.2 incombination with SUR1 (ref.12). It islikely that glucokinase, a high Km hex-okinase not found in most neurons, isthe critical piece of cellular machineryrequired to provide GR neurons withtheir glucosensing capacity (Fig. 1)13. Inthe many other non-GR neurons, theKATP channel may have a protective rolethat comes into play when energy sub-strates (glucose and/or oxygen) becomeseverely limited. Reduced substrate forATP production would lower theATP/ADP ratio, open the KATP channel,hyperpolarize the cell membrane andreduce cell firing and metabolic demand.If the KATP channel does have such aneuroprotective role, Kir6.2–/– miceshould be at increased risk for hypo-glycemia-induced cell damage, particu-larly in the hypothalamic arcuatenucleus, where hypoglycemia producesapoptotic neuronal damage in intact ani-mals11. The Kir6.2–/– mouse is an idealmodel to test this postulated neuropro-tective role for the KATP channel.

The authors are on less solid footingwhen they propose a direct causal linkbetween the absence of VMH GR neu-rons and the impairments of glucopriv-ic feeding and counter-regulatoryresponse in Kir6.2–/– mice. First, becauseof the widespread distribution of GRneurons in the brain, there is no way todirectly link defects in VMH GR neu-rons with attenuated behavioral andneurohumoral responses in Kir6.2–/–

mice. Second, although systemic orintraventricular infusions of 2-deoxyglu-cose induce feeding, direct injections ofother glucoprivic agents into a variety ofbrain areas suggest that glucoprivic feed-ing is mediated by hindbrain rather thanVMH sites14. Third, the authors did notexamine the function of VMH neuronsin Kir6.2–/– mice under the type of low-glucose conditions required to elicit glu-coprivic feeding or a counter-regulatoryresponse. Indeed, they defined GR neu-rons by their response to changes inextracellular glucose levels from a phys-iological level (2.5 mM) to one so high(25 mM) that is unlikely ever to beencountered in the brain, even duringsevere diabetic hyperglycemia. Finally,

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


news and views


When we move, or when objects in theworld move, the visual images projectedonto our retinae change accordingly.Over 50 years ago, the psychologist J.J. Gibson noted that important envi-ronmental information is embedded inlocal retinal image velocity (that is, bothspeed and direction), and thus beganthe investigation of the mechanisms bywhich such velocities might be esti-mated. Most physiological models ofvisual motion posit that neurons in areaMT (a small extrastriate region of visualcortex) of the primate brain are velocityselective, responding most strongly to avisual stimulus moving in a preferreddirection and with a preferred speed,and disregarding other stimulus attrib-utes such as color and pattern. Manystudies have confirmed the encoding ofdirection by these neurons. In this issue,Perrone and Thiele1 present experimen-tal data from neurons in area MT thatconfirm a prediction of these theoriesregarding the representation of speed.

Some neurons in the primary visualcortex (area V1) of primates are motionsensitive. Each of these neurons respondsvigorously to a visual pattern with a pre-ferred orientation moving in a preferreddirection, but less so or not at all for pat-terns with the wrong orientation or inthe opposite direction. These V1 neu-rons do not seem to provide the charac-teristics one expects in a velocityrepresentation, however, because they

J.A. Movshon, Invest. Opthal. Vis. Sci.Suppl. 24, 106, 1983) for the invariance ofMT speed preferences with respect tostimulus pattern.

Perrone and Thiele tested whetherMT speed preferences are invariant tochanges in stimulus pattern, by measur-ing responses to moving sinusoidal grat-ing stimuli (adopting the experimentalprotocol of Newsome, Gizzi andMovshon, 1983). A moving sinusoidalgrating may be characterized by its ori-entation, spatial frequency (in units ofcycles per degree of visual angle) andtemporal frequency (in units of Hertz orcycles per second). The grating speed isdetermined by its temporal frequencydivided by its spatial frequency8. Thisrelationship leads to a simple predictionregarding the responses of speed-tunedneurons to drifting gratings9 (Fig. 1). Aset of speed tuning curves for a hypo-thetical MT neuron (Fig. 1a) showsresponses to gratings of different spatialfrequencies. Changing the spatial fre-quency leads to a rescaling of the curve,but its shape and position are unaffect-ed. That is, the neuron shows stableselectivity for stimulus speed, regardlessof spatial frequency. Replotting the samedata as a set of temporal frequency tun-ing curves (Fig. 1b) shows that the tem-poral frequency tuning is stronglyaffected by changes in spatial frequency.

confound changes in stimulus pattern(such as orientation) with changes instimulus velocity2,3.

As an alternative, it has been proposedthat MT neurons provide an unambigu-ous representation of velocity. These neu-rons receive their primary input fromdirection-selective cells in V1 (ref. 4). Thevast majority of MT neurons, like their V1afferents, are direction selective5, but asubstantial number of these neurons,unlike their V1 afferents, maintain theirdirection preference while largely ignor-ing changes in the stimulus pattern3. MTcells are also tuned for speed5–7. Untilnow, however, there was only unpublishedevidence (W.T. Newsome, M.S. Gizzi and

Representing retinalimage speed in visualcortexEero P. Simoncelli and David J. Heeger

Speed preferences in MT neurons are found to be unaffectedby changes in stimulus pattern, supporting the hypothesisthat these neurons represent retinal image velocities.

Eero Simoncelli is in the Howard HughesMedical Institute, Center for Neural Science,and Courant Institute of MathematicalSciences, New York University, New York, NewYork 10003-1056, USA. David Heeger is in theDepartment of Psychology, Stanford University,Stanford, California 94305-2130, USA. e-mail: [email protected] [email protected]

1 2 4 8 16

Res

pons

e

1 2 4 8 16 32

Speed (deg/s)

1 2 4 8 16

Res

pons

e

1 2 4 8 16 32

Temporal frequency (Hz)

SF=1SF=2SF=4

Fig. 1. Speed tuning versus temporal-frequency tuning. (a) Speed tuning curves of a hypotheticalMT neuron, measured with drifting sinusoidal gratings. (b) Temporal frequency tuning curves ofthe same MT neuron (replotted from a, using the relationship that grating speed is temporal fre-quency divided by spatial frequency). (c) Speed tuning curves of a hypothetical V1 neuron. (d) Temporal frequency tuning curves of the same V1 neuron.

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



news and views

This behavior may be contrasted withthat of a typical V1 neuron that is tunedfor temporal frequency10,11, but not forspeed (Fig. 1c and d).

Perrone and Thiele recorded MTresponses to an array of spatial and tem-poral frequencies, and plotted theresponses of each neuron as a surfacelike those in Fig. 2. The surface repre-sents the full spatiotemporal frequencyselectivity (dubbed the ‘spectral recep-tive field’ by the authors). They fit theresulting spectral receptive field of eachneuron with an elliptical Gaussian func-tion, and used the tilt of the best-fitGaussian to classify the neuron as beingselective for temporal frequency or forspeed. The spectral receptive fields of V1neurons are parallel to the spatial andtemporal frequency axes10,11 (Fig. 2a),but Perrone and Thiele report that forroughly 60% of their MT neurons, thespectral receptive fields were tilted alonglines emanating from the origin (Fig. 2b). Thus, for these neurons, pre-ferred speed was largely independent ofspatial frequency. In addition, theydemonstrated that the tilt of the best-fitGaussian accurately predicted the pre-ferred speed of each neuron, as mea-sured directly with a moving bar.

Computational theories explain theemergence of velocity selectivity in MTneurons through a suitable combinationof V1 afferents2,3,6,9,12,13. A simplifiedversion of this construction is shown inFig. 2, in which the spectral receptivefield of a hypothetical MT neuron (Fig. 2b) was computed directly by sum-ming those of three V1 neurons (Fig. 2a). The resulting MT neuron istuned for a much broader range of spa-tial frequencies than its V1 afferents, andhas a spectral receptive field that is tiltedalong a line emanating from the origin.

The experimental results of Perroneand Thiele provide a much-needed addi-tion to the body of evidence supporting

quent authors. Thus, even though theresponse properties of the neurons inthese two cortical areas are very different,they seem to be based on a commoncanonical computation. The underlyingcortical circuitry and biophysical mecha-nisms responsible for orientation selec-tivity in V1 are still hotly debated, butgiven the commonality of the computa-tional principles that seem to apply inboth V1 and MT, we may optimisticallyhope that the circuitry and biophysicalmechanisms in these two (and perhapsother) cortical areas may also follow acanonical form.

1. Perrone, J. A. & Thiele, A. Nat. Neurosci. 5,526–532 (2001).

2. Adelson, E. H. & Movshon, J. A. Nature 300,523–525 (1982).

3. Movshon, J. A., Adelson, E. H., Gizzi, M. S. &Newsome, W. T. in Experimental Brain ResearchSupplementum II: Pattern RecognitionMechanisms (eds. Chagas, C., Gattas, R. &Gross, C.) 117–151 (Springer, New York, 1986).

4. Movshon, J. A. & Newsome, W. T. J. Neurosci.16, 7733–7741 (1996).

5. Maunsell, J. H. & Van Essen, D. C. J. Neurophysiol. 49, 1127–1147 (1983).

6. Albright, T. D. J. Neurophysiol. 52, 1106–1130(1984).

7. Rodman, H. R. & Albright, T. D. Vision Res.27, 2035–2048 (1987).

8. Watson, A. B. & Ahumada, A. J. in Motion:Perception and Representation (ed. Tsotsos, J. K.) 1–10 (Association for computingmachinery, New York, 1983).

9. Grzywacz, N. M. & Yuille, A. L. Proc. R. Soc.Lond. B Biol. Sci. 239, 129–161 (1990).

10. Tolhurst, D. J. & Movshon, J. A. Nature 257,674–675 (1975).

11. Hamilton, D. B., Albrecht, D. G. & Geisler, W. S. Vision Res. 29, 1285–1308 (1989).

12. Heeger, D. J. J. Opt. Soc. Am. A 4, 1455–1471(1987).

13. Simoncelli, E. P. & Heeger, D. J. Vision Res. 38,743–761 (1998).

14. Levitt, J. B., Kiper, D. C. & Movshon, J. A. J. Neurophysiol. 71, 2517–2542 (1994).

15. Okamoto, H. et al. Vision Res. 39, 3465–3479(1999).

the hypothesis that MT neurons computeand represent local retinal image veloci-ties. Although many previous authorshad measured speed tuning in these neu-rons, those experiments did not rule outthe possibility that the neurons wereactually selective for temporal frequency(as are V1 neurons) instead of speed. Thedata reported by Perrone and Thiele, onthe other hand, demonstrate that at leastsome MT neurons are selective for speedper se. A relatively minor criticism oftheir work is that the measured spectralreceptive fields are not very well charac-terized by elliptical Gaussian functions.Instead of using Gaussian fits, the analy-sis could have been strengthened byusing previously published non-para-metric methods for evaluating whetherthe spectral receptive fields were parallelwith the axes or tilted14, and by relyingmore heavily on existing computationalmodels of MT responses (cited above). Itis unlikely that these alternate analyseswould have changed the main conclusionthat some MT neurons are speed-tuned,but it might, for example, have resultedin a larger proportion of neurons beingclassified as speed-tuned.

A more subtle but fundamental issueis that although a tilted spectral recep-tive field is sufficient to produce speedselectivity, it is not absolutely necessary.For example, if an MT neuron has a rel-atively narrow spatial frequency band-width, the tilt of its spectral receptivefield would be difficult to measure.There is an alternative test that can pro-vide more definitive evidence for veloc-ity selectivity (E.P. Simoncelli, W.D. Bair, J.R. Cavanaugh & J.A. Movshon, Invest.Opthalm. Vis. Sci. Suppl. 37, 1996)15.

The construction of velocity selectivi-ty in MT from V1 afferents (Fig. 2) isanalogous to the construction of orienta-tion selectivity in V1 from LGN afferents,as originally proposed by Hubel andWiesel and elaborated by many subse-

Fig. 2. Construction of a speed-tuned MT neu-ron from V1 afferents13. (a) Spectral receptivefields for three hypothetical V1 neurons. (b)Spectral receptive field of a hypothetical speed-tuned MT neuron, constructed by summing thethree V1 afferents (compare with Fig. 6 of Perrone and Thiele1).

0 1 2

2

4

8

Tem

pora

l fre

quen

cy (

Hz)

0 1 2Spatial frequency (cyc/deg)

0 1 2 0 1 2

2

4

8

Tem

pora

l fre

quen

cy (

Hz)

Spatial frequency (cyc/deg)

6 6

a b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Following CNS injury in the adult mammal, axon regenerationfails in scar regions containing a number of different chondroitinsulfate-bearing proteoglycans (CSPGs)1. Degradation of chon-droitin sulfate using chondroitinase ABC reduces growth inhi-bition associated with many CSPGs2–13. Here we demonstratethat it is possible to enhance CNS axon regeneration in the adultrat nigrostriatal tract following chondroitinase ABC degradationof chondroitin sulfate.

In a preliminary experiment to determine the time course ofCSPG deposition following nigrostriatal tract axotomy, six anes-thetized adult rats were given unilateral nigrostriatal axotomylesions14 using a Scouten wire knife, and killed after 1, 2 or 4 weeks. Immunolabeling using various antibodies against chon-droitin sulfate and against CSPG core proteinsshowed that in this model of injury, as in oth-ers3–5, peak levels of many CSPGs were detect-ed at injury sites between one and two weekspost-axotomy (data not shown).

To degrade chondroitin sulfate throughoutthis period, we repeatedly delivered chondroiti-nase ABC to the site of axotomy. Twenty-threeanesthetized adult rats were given unilateral

nigrostriatal axotomy lesions, as above. Cannulae were securedtranscranially to allow sham infusion or repeated perilesional infu-sions of 3 µl saline containing 600 ng of either high-purity, pro-tease-free chondroitinase ABC (Seikagaku, Japan), or penicillinase(control bacterial protein, Roche, UK), on days 0, 3, 7 and 10 post-axotomy (Fig. 1a).

To assess regeneration of dopaminergic nigrostriatal axons, seriesof parasagittal sections of brains from animals killed 11, 18 or 100days post-axotomy were immunolabeled using antibodies againsttyrosine hydroxylase (TH). At all times examined following axo-tomy and treatment with chondroitinase ABC, dopaminergic nigralaxons had grown through the injury site (Fig. 1d) and along thecourse of the original nigrostriatal tract (Fig. 1e) back to their orig-inal principal target (the ipsilateral striatum), where they branched(Fig. 1f). Axons did not always grow directly toward the target butoften grew ectopically, branching extremely frequently. In contrast,at no time did dopaminergic nigral axons regenerate beyond thesite of axotomy following axotomy with either sham infusion ortreatment with control protein (Fig. 1g–i). The few TH-immunore-active processes observed in control animals beyond the site of axo-tomy and within occasional striatal foci (Fig. 1b and c) were notthought to be degenerating or regenerating nigral axons, becausethe number of these axons did not change with time; further, theseaxons are seen following either complete axotomy14 or chemicalablation of the substantia nigra15. Estimates for the number ofregenerating dopaminergic nigral axons in axotomized animalstreated with chondroitinase ABC have been adjusted accordingly.

The number of TH-immunoreactive processes observedgrowing 1 mm beyond the site of axotomy toward the originalprincipal target differed greatly between treatment groups(Kruskal Wallis, H2 = 16.9, p < 0.001; Dunn’s post hoc tests, p < 0.05; Fig. 1b). Following treatment with chondroitinase ABC,the mean number of dopaminergic nigral axons growing this farwas estimated as 1914 on day 11, 2162 on day 18 and 2020 on

Fig. 1. Treatment with chondroitinase ABCenhanced dopaminergic nigrostriatal axon regener-ation in vivo. All animals were treated in accordancewith the UK Animals (Scientific Procedures) Act1986. Axotomized, sham-infused animals examined11, 18 or 100 days post-axotomy did not differ sig-nificantly one from another and were combined.Error bars, standard error; n, number of animalsper group. (a) Parasagittal schematic showing thesite of axotomy and transcranial infusion. Threesmall diamonds indicate regions shown at highermagnification in (d–f) and (g–i). Scale bar, 2 mm. (b) The estimated mean number of tyrosinehydroxylase (TH)-immunoreactive axons counted1 mm anterior to the site of the axotomy was signif-icantly greater in axotomized animals treated with chondroitinase ABC than in those treated with control protein. (c) The estimated mean numberof TH-immunoreactive processes entering the ipsilateral striatum was significantly greater in animals treated with chondroitinase ABC than thosetreated with control protein. (d–i) TH immunolabeling examined 11 days post-axotomy indicated that following treatment with chondroitinase ABC(d–f), but not following treatment with control protein (g–i), dopaminergic nigral neurons regrew through the site of axotomy (d, g) and along thecourse of the original nigrostriatal tract (e, h) and into the striatum (f, i), where they arborized. Scale bars, 50 µm.

brief communications


Regeneration of CNSaxons back to their targetfollowing treatment ofadult rat brain withchondroitinase ABCLawrence D. F. Moon1,2, Richard A. Asher1, Kate E. Rhodes1 and James W. Fawcett1

1 Physiological Laboratory, University of Cambridge, Downing Site, TennisCourt Road, Cambridge CB2 3EG, UK

2 Present address: The Miami Project to Cure Paralysis, PO Box 16960, MailLocator R-48, Miami, Florida 33101, USA

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

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


day 100 (Fig. 1b). However, because these numbers includenumerous axon collaterals, they may overestimate the numberof primary regenerating axons. In intact brains, approximately50,000 TH-immunoreactive axons were counted at this level.

The number of TH-immunoreactive processes entering theoriginal target, 4 mm beyond the site of axotomy, also differedgreatly between treatment groups (Kruskal Wallis, H2 = 15.3, p < 0.001; Dunn’s post hoc tests, p < 0.05; Fig. 1c). Followingtreatment with chondroitinase ABC, the mean number ofdopaminergic processes entering the original target was esti-mated as 1786 on day 11, 1665 on day 18 and 1045 on day 100(Fig. 1c). In intact brains, approximately 70,400 TH-immunore-active processes were counted at this level.

Finally, the mean number of TH-immunoreactive cell bod-ies in the substantia nigra either ipsilateral or contralateral to thelesion did not differ between groups (data not shown). Takentogether, these results indicate that treatment with chondroiti-nase ABC promotes long-distance regeneration of cut dopamin-ergic nigral axons.

To demonstrate that chondroitinase ABC degraded chon-droitin sulfate in the region of axon regeneration, series of sec-tions were immunolabeled using three different monoclonalantibodies. Antibody 2B6 (Seikagaku, Japan) recognizes an epi-tope created following chondroitinase ABC degradation of chon-droitin-4 sulfate. (2B6 does not recognize intact chondroitinsulfate.) At all times examined, 2B6 immunoreactivity was absentin controls (Fig. 2a), whereas there was intense 2B6 immunola-beling within 4 mm of the infusion site in axotomized animalstreated with chondroitinase ABC (Fig. 2b). Thus, chondroitinaseABC (but not control protein) effectively degraded chondroitinsulfate from the entire region between the site of axotomy andthe proximal part of the original target. Further, the same stain-ing patterns were observed following immunolabeling using the3B3 antibody (Seikagaku, Japan), which recognizes an epitopecreated following chondroitinase ABC degradation of chon-droitin-6 sulfate. (3B3 does not recognize intact chondroitin sul-fate.) Finally, immunolabeling using the CS-56 antibody (Sigma,UK), which recognizes an epitope on some intact chondroitinsulfate glycosaminoglycan chains, showed intense reactivity sur-rounding the site of axotomy in axotomized animals treated withcontrol protein (Fig. 2c), whereas in axotomized animals treat-

ed with chondroitinase ABC, CS-56 immunoreactivity was muchreduced (Fig. 2d). Differences between groups in labeling (2B6,3B3 and CS-56) were less pronounced at 18 and 100 days post-axotomy, indicating that CSPGs depleted of chondroitin sulfateare removed slowly from injury sites. These results clearly indicatethat treatment with chondroitinase ABC (but not control pro-tein) effectively degraded chondroitin sulfate in vivo from the siteof axotomy and the surrounding tissue.

Finally, on an animal-by-animal basis, following axotomy andtreatment with chondroitinase ABC (examined 11, 18 or 100 dayspost-axotomy), the extent of chondroitin sulfate degradation(quantified by optical densitometry of 2B6 immunoreactivity)correlated significantly and positively with the number of TH-immunoreactive processes entering the original target (Pearsonproduct moment correlation, r = 0.70, p = 0.011). This indicatesthat with an increase in chondroitin sulfate degradation, thenumber of dopaminergic nigral axons able to grow long distancesincreased as well.

Our results show that degradation of chondroitin sulfate usingchondroitinase ABC can render the environment of the damagedCNS more permissive to axon regeneration. This result confirmsthat CSPGs are a significant source of inhibition after CNS injuryand suggests that the effects of chondroitinase ABC should beevaluated in other models of CNS injury, with the long-termintention of treating human spinal cord injuries.

ACKNOWLEDGEMENTS

This work was funded by grants from the Medical Research Council, Action

Research, the International Spinal Research Trust and the Wellcome Trust.

Various antibodies were donated by J. Levine and A. Oohira, or obtained from the

Developmental Studies Hybridoma Bank maintained by the University of Iowa.

RECEIVED 18 DECEMBER 2000; ACCEPTED 21 MARCH 2001

1. Davies, S. J., Goucher, D. R., Doller, C. & Silver, J. J. Neurosci. 19, 5810–5822(1999).

2. Grumet, M., Flaccus, A. & Margolis, R. U. J. Cell Biol. 120, 815–824 (1993).3. Asher, R. A. et al. J. Neurosci. 20, 2427–2438 (2000).4. Levine, J. M. J. Neurosci. 14, 4716–4730 (1994).5. McKeon, R. J., Jurynec, M. J. & Buck, C. R. J. Neurosci. 19, 10778–10788

(1999).6. Yamada, H. et al. J. Neurosci. 17, 7784–7795 (1997).7. Dou, D.-L. & Levine, J. M. J. Neurosci. 14, 7616–7628 (1994).8. McKeon, R. J., Hoke, A. & Silver, J. Exp. Neurol. 136, 32–43 (1995).9. Snow, D., Lemmon, V., Carrino, D. A., Caplan, A. I. & Silver, J. Exp. Neurol.

109, 111–130 (1990).10. Smith-Thomas, L. C. et al. J. Cell Sci. 108, 1307–1315 (1995).11. Brittis, P. A., Canning, D. R. & Silver, J. Science 255, 733–736 (1992).12. Chung, K. Y. et al. Development 127, 2673–2683 (2000).13. Yick, L. W. et al. Neuroreport 11, 1063–1067 (2000).14. Brecknell, J. E., Dunnett, S. B. & Fawcett, J. W. Neuroscience 64, 219–227 (1995).15. Brecknell, J. E. et al. Neuroscience 71, 913–925 (1996).

brief communications


Fig. 2. Treatment with chondroitinase ABC degraded chondroitin sul-fate in vivo, examined 11 days post-axotomy. (a, b) 2B6 immunolabelingindicated that chondroitin-4 sulfate was degraded between the site ofaxotomy and the original target following treatment with chondroiti-nase ABC (b) but not control protein (a). Fields of view correspond tolarge rectangle shown in Fig. 1a. The region lacking 2B6 immunoreactiv-ity in (b) is due to cavitation. Scale bars, 2 mm. (c, d) CS-56 immunola-beling indicated that chondroitin sulfate was degraded at the site ofaxotomy following treatment with chondroitinase ABC (d), but notcontrol protein (c). Right half of image corresponds to lesion core, lefthalf corresponds to gliotic surround. Scale bars, 100 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


review


Cerebellar circuits and motor learningWhy bother to study the cellular substrates of memory in an atyp-ical brain structure like the cerebellum? The simple answer is thatit is the location in the mammalian brain that holds the greatestpromise for uncovering the holy grail: an understanding of(admittedly simple) forms of learning and memory that will flowcontinuously from molecules and cells through synapses and cir-cuits to behavior without any grossly embarrassing gaps in themiddle. The factors that allow for this include an unusually well-defined circuit diagram, the fortuitous discovery of some naturalcell-type specific gene promoters that will facilitate the produc-tion of transgenic mice, behavioral outputs such as eye and eyelidmovements that are simple and easily quantified, and, mostimportantly, a clear understanding of the pathways that conveybehaviorally relevant signals (such as the conditioned stimulus(CS) and unconditioned stimulus (US) in eyelid conditioning).

The cerebellar circuitry is essentially composed of a relay sta-tion in the deep cerebellar nuclei (DCN) and a cortical ‘side-loop’(Fig. 1a). Cerebellar output to premotor centers originates in theDCN and is in turn driven by direct excitatory input from themossy fibers. In addition, it is modulated by the inhibitory inputfrom the Purkinje cell axons, which convey computations andinteractions in the Purkinje cell. These computations will be per-formed upon a matrix of subtle and informationally rich excita-tory parallel fiber input (∼ 200,000 axons), massive andsynchronous excitation produced by the one climbing fiber axoninnervating each mature Purkinje cell, and input from inhibitoryinterneurons. This unusual anatomical configuration inspired amodel of motor learning by Marr1, who proposed that the paral-lel fiber–Purkinje cell synapses could provide contextual infor-

mation, that climbing fiber–Purkinje cell synapses could signalan ‘error’ in motor performance that required alteration of sub-sequent behavior, and that the conjunction of these two signalscould strengthen the parallel fiber–Purkinje cell synapse to cre-ate a memory trace for motor learning. This model was modifiedby Albus2, who realized that a decrease in synaptic strength wouldbe more appropriate given the sign-reversing function of the Purk-inje cell inhibitory output. Albus also noted that this model isanalogous to classical conditioning with the parallel fibers con-veying a CS, the climbing fiber a US and a depression of the par-allel fiber–Purkinje cell synapse giving rise to a conditionedresponse (CR) via disinhibition of the DCN. To place this modelin a behavioral context, let us consider associative eyelid condi-tioning (Fig. 1b). Before training, an airpuff to the eye (US) givesrise to an immediate reflexive blink (the unconditioned response,UR). During training, a neutral stimulus such as a tone (CS) ispaired with the airpuff stimulation so that the tone onset precedesthe airpuff and the two stimuli coterminate. As the association isacquired, a carefully timed eyelid response is performed (the CR),the onset of which shortly precedes the airpuff. This learning canalso be actively reversed. In well-trained animals that reliably per-form CRs, this response can undergo rapid ‘extinction’ if tone(CS) stimuli are repeatedly presented without airpuffs.

There is extensive evidence to support the involvement ofcerebellar cortical and deep nuclear circuits in associative eyelidconditioning (for review, see refs. 3 and 4). First, extracellularrecording showed that in well-trained animals, populations ofcells in the DCN begin to fire during the CS–US interval (Fig. 1b). This firing is predictive of and correlated with the per-formance of the CR, suggesting that the CR behavior is expressed

Beyond parallel fiber LTD: thediversity of synaptic and non-synaptic plasticity in the cerebellum

Christian Hansel1, David J. Linden2 and Egidio D’Angelo3,4

1 Department of Anatomy, Institute of Neuroscience, Erasmus University Rotterdam, PO Box 1738, 3000 DR Rotterdam, The Netherlands2 Department of Neuroscience, Johns Hopkins University School of Medicine, 725 N. Wolfe Street, Baltimore, Maryland 21205, USA3 Department of Cellular/Molecular Physiology and Pharmacology, University of Pavia and INFM (Pavia Unit), Via Forlanini 6, Pavia, I 27100, Italy4 Department of Evolutionary and Functional Biology, University of Parma, Parco Area delle Scienze 11, Parma, I 43100, Italy

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

In recent years, it has become clear that motor learning, as revealed by associative eyelidconditioning and adaptation of the vestibulo-ocular reflex, contributes to the well-establishedcerebellar functions of sensorimotor integration and control. Long-term depression of the parallelfiber–Purkinje cell synapse (which is often called ‘cerebellar LTD’) is a cellular phenomenon that hasbeen suggested to underlie these forms of learning. However, it is clear that parallel fiber LTD, byitself, cannot account for all the properties of cerebellar motor learning. Here we review recentelectrophysiological experiments that have described a rich variety of use-dependent plasticity incerebellum, including long-term potentiation (LTP) and LTD of excitatory and inhibitory synapses,and persistent modulation of intrinsic neuronal excitability. Finally, using associative eyelid condition-ing as an example, we propose some ideas about how these cellular phenomena might function andinteract to endow the cerebellar circuit with particular computational and mnemonic properties.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 1. Cerebellar circuitry. (a) The neurons of the deep cerebellar nuclei(DCN) receive their main excitatory input from glutamatergic mossyfibers (MFs), which are the axons of a large number of pre-cerebellarnuclei. The main outflow of information from this structure is carried byexcitatory axons that originate from the large neurons of the DCN andproject to premotor areas and thalamic structures. The sole output ofthe cortical side-loop is the inhibitory, GABAergic projection fromPurkinje cells to the neurons of the DCN. Purkinje cells receive twomajor excitatory inputs, which are organized in very different ways. Eachmature Purkinje cell is innervated by a single glutamatergic climbing fiber,which is the axon of cells in the inferior olive. This is potentially the mostpowerful synaptic contact in the brain, as a each Purkinje cell receives∼ 1,400 synapses from a single climbing fiber axon. Climbing fibers alsoprovide a weak collateral innervation of the DCN consisting of a fewsynapses in the distal dendrites, the function of which is poorly under-stood. In contrast, each Purkinje cell receives ∼ 200,000 synapses fromparallel fibers, which are the axons of granule cells. Parallel fibers alsodrive GABAergic inhibitory interneurons (stellate and basket cells),which project to the axo-somatic regions of Purkinje cells. Closing theloop, granule cells receive excitatory synapses from branches of the samemossy fibers that innervate the DCN directly. Excitatory presynaptic ter-minals are drawn as triangles, and inhibitory presynaptic terminals aredrawn as bars. (b) Associative eyelid conditioning. In each panel, theupper trace represents the eyelid response and the lower histogram rep-resents extracellularly recorded spiking in the DCN. No response is seento the CS and a reflexive UR is seen to the US (CS and US onset are indi-cated by vertical lines. With repeated pairing of the CS and US, a blinkresponse develops that is timed to precede the US onset. This CR is cor-related with an increase in the firing rate of DCN cells. This panel is mod-ified from ref. 75. Copyright 1984 by the Society for Neuroscience.

in the firing rate and pattern of DCN neurons. Furthermore,stimulation of the DCN can elicit eyelid responses in naive ani-mals. Second, during training, stimulation of mossy and climb-ing fibers can substitute for the CS and US, respectively. Finally,reversible inactivation of the DCN and overlying cerebellar cor-tex prevented the acquisition of the eyelid CR, but not the per-formance of the UR. In contrast, inactivation of the superiorcerebellar peduncle or red nucleus, sites through which excita-tory DCN output is conveyed, prevented the expression of theCR during training, but not its acquisition, as evidenced by thefact that the CR was present after inactivation. These studies sug-gest that the cerebellum and its associated projections are essen-tial for acquisition and expression of the eyelid CR. Morespecifically, the memory trace seems to be localized upstream ofthe red nucleus but downstream of the mossy and climbing fibers,that is, in the cerebellar cortex and/or the DCN. Similar evidenceimplicates the cerebellar cortex and vestibular nuclei in anotherform of motor learning, adaptation of the vestibulo-ocular reflex5

(reviewed in ref. 6). Although it is likely that there are other formsof motor learning that require plasticity in the cerebellar circuit(multijoint limb load adjustment is a good candidate), it shouldbe cautioned that there are also forms of motor learning (suchas improvement in the performance of rotating rod and thin rodbalancing tasks) that can persist even when some forms of cere-bellar plasticity are blocked5.

Parallel fiber LTDCellular mechanisms that might underlie acquisition of associa-tive eyelid conditioning would be expected to be located at pointsin the cerebellar circuit where CS and US signals converge andwhere their repeated co-occurrence results in an increase in DCNfiring. As suggested by Marr–Albus models, parallel fiber LTD isone such cellular mechanism. In parallel fiber LTD, first report-ed by Ito and collaborators, a persistent attenuation of the par-

allel fiber–Purkinje cell synapse is produced when parallel fiberand climbing fiber inputs to a Purkinje cell are stimulated togeth-er at low frequency7. The requirements for parallel fiber LTDinduction, like those for eyelid conditioning, are typically asso-ciative—parallel fiber synaptic strength is not attenuated byclimbing fiber stimulation alone, parallel fiber stimulation alone,or stimulation of both in an unpaired manner. (See refs. 8 and 9for examples of LTD induced with very strong PF stimulationalone.) Repeated induction of parallel fiber LTD results in satu-ration of the response; a maximum depression of ∼ 50% is typical.Parallel fiber LTD is generally understood to have three initialrequirements for induction. The climbing fiber contributes toLTD induction via Ca2+ influx through voltage-gated channels,occurring during the complex spike. The parallel fibers releaseglutamate, which acts upon both mGluR1 metabotropic receptorsand AMPA receptors. After these initial signals, there is a require-ment for transient protein kinase C (PKC) activation. Theinvolvement of a nitric oxide/cGMP/protein kinase G/phos-phatase inhibition pathway has also been suggested. (For review,see ref. 10.) Parallel fiber LTD is expressed postsynaptically, as areduction in the number of functional AMPA receptors producedby clathrin-mediated endocytosis11,12.

Parallel fiber LTPIn an animal that has been trained to achieve asymptotic perfor-mance of eyelid CRs, one way in which repeated application ofthe CS alone could be made to result in CR extinction would be toproduce homosynaptic LTP of parallel fiber–Purkinje cell synaps-es. Indeed, this form of LTP may be observed after low-frequency(2–8 Hz) parallel fiber stimulation in the absence of climbing fiberactivation13–18. In contrast to LTD, LTP at the parallel fiber synapseseems to be triggered by a presynaptic Ca2+ influx. Its induction isblocked by removal of external Ca2+, but not by antagonists ofglutamate receptors or by loading of the Purkinje cell with the

review


Purkinjecell

Climbingfiber

Granulecell


nuclei

Premotorareas

Deepcerebellar

nuclei

Mossyfibers

Parallel fibers


Climbingfiber

collateral

a

b

Bob

Crim

i

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Ca2+ chelator BAPTA19–22. Several lines of evidencesuggest that parallel fiber LTP is mediated by the presy-naptic activation of Ca2+-sensitive adenylyl cyclase,leading to a rise in the concentration of cAMP and theconsequent activation of protein kinase A (PKA). AnLTP-like effect may be produced by bath applicationof the adenylyl cyclase activator forskolin or mem-brane-permeable cAMP analogs19,23–25. In addition,cerebellar LTP induced by granule cell stimulation (butnot an exogenous cAMP analog) is attenuated in cellcultures from a type I Ca2+-sensitive adenylyl cyclasenull mutant mouse25. Induction of LTP produced bytetanic stimulation is blocked when PKA inhibitors areapplied in the bath19,24 or when presynaptic granulecells are transfected with an expression vector encoding a peptideinhibitor of PKA22.

The expression of parallel fiber LTP also seems to be presy-naptic. It is associated with a decrease in the rate of synaptic fail-ures16,20,21,25 and the extent of paired-pulse facilitation19,21.Further evidence comes from the observation that LTP could bedetected in granule cell–glial cell pairs in culture when postsy-naptically recording evoked AMPA/kainate receptor-mediatedcurrents20 or glutamate transport currents21. The observationthat such ‘reporter currents’ measured in glial cells can alsoundergo LTP indicates a presynaptic locus of expression. Thus,whereas LTD at the parallel fiber–Purkinje cell synapse is post-synaptically induced and expressed, LTP is presynapticallyinduced and expressed. In such a scheme, cerebellar LTP and LTDwould not truly reverse each other, but rather would be additive,independent phenomena, a suggestion that has recently beenconfirmed experimentally10. This is in striking contrast to resultsobtained at excitatory synapses received by pyramidal cells in thehippocampal CA1 subfield, in which both LTP and LTD can bepostsynaptically expressed and the synapses are endowed withreversible bidirectional modification (for review, see ref. 10).

Climbing fiber LTDStimulation of the one climbing fiber axon that innervates amature Purkinje cell results in the activation of ∼ 1,400 synapseswith a high release probability26, thereby reliably producing amassive excitation of the Purkinje cell known as the complexspike. The classic Marr–Albus–Ito models of cerebellar functionare generally silent regarding changes in the strength of this

climbing fiber–Purkinje cell synapse. However, climbing fibersshow a form of developmental plasticity during the first threeweeks of postnatal life. After an initial innervation of each Purk-inje cell by several climbing fibers, an activity-dependent elimi-nation of surplus climbing fibers takes place until the 1:1connection ratio that is typical for the adult cerebellar cortex isachieved27. Moreover, even in adult rats, climbing fibers showsome degree of morphological plasticity, depending on the avail-ability of target Purkinje cells and the overall electrical activityof the cerebellar cortex28. Furthermore, the molecular machineryrequired for LTD at the parallel fiber synapse (mGluR1, P/Q-type Ca2+ channels, GluR2-containing AMPA receptors andPKC) is also present at the climbing fiber input.

These observations inspired a study using whole-cell patch-clamp recordings in a slice preparation in which climbing fiberLTD was observed and characterized for the first time29. LTD ofclimbing fiber-evoked EPSCs was observed after 5 Hz × 30 stetanization (Fig. 2). This climbing fiber LTD did not spread toneighboring parallel fiber synapses, and, like parallel fiber LTD,it could be saturated when tetanization was repeated. LTD wasnot associated with a change in the paired-pulse ratio. This latterresult is preliminary evidence for a postsynaptic locus of expres-sion (assuming that a presynaptic modification would changethe release probability). Similar to parallel fiber LTD, climbingfiber LTD could be blocked by loading the Purkinje cells with theCa2+ chelator BAPTA, or by bath application of a group I mGluRantagonist or a PKC inhibitor.

What might be the consequences of climbing fiber LTD forcerebellar function? Perhaps, because of a high safety factor for

review

Fig. 2. Climbing fiber long-term depression. (a) 5 Hz × 30 sclimbing fiber stimulation resulted in a depression of climbingfiber-evoked EPSCs (n = 15). Under control conditions, nochange in EPSC amplitude was observed (n = 5). Each teststimulus consisted of a pulse pair applied with an interval of50 ms. Traces are averages of two to five sweeps. In the time-course graph, each data point represents the average of threesuccessive responses evoked at 0.05 Hz. Only the amplitudeof the first EPSC is shown. (b) Climbing fiber tetanization didnot change the amplitudes of parallel fiber EPSCs as mea-sured in a subset of the tetanized cells (n = 9) shown in (a). (c) Climbing fiber stimulation produced an alteration inthe complex spike waveform (current-clamp mode).Representative single traces are shown before and 25 minafter tetanization. The time-course graphs show the changesin the amplitudes of the first two spike components undertetanization (n = 5) and control conditions (n = 4). Each datapoint represents the average of five successive responsesevoked at 0.05 Hz. This figure is modified from ref. 29. Usedwith permission from the copyright holder, Cell Press.

a b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


evoking complex spikes, the ∼ 20% attenuation of EPSC ampli-tude produced by climbing fiber LTD is of no relevance for cere-bellar function and is merely an ‘evolutionary byproduct’ of thebiochemical mechanisms used for parallel fiber LTD. However,when recordings were performed in current-clamp mode, 5 Hz × 30 s stimulation resulted in a selective modification ofthe complex spike. Complex spikes are composed of an initial,somatically mediated Na+ spike, followed by a series of smallerspikes that are produced by dendritic Ca2+ influx30,31 and arecently described ‘resurgent Na+ current’32. After 5 Hz tetaniza-tion, the second complex spike component was selectivelyreduced, suggesting that climbing fiber LTD can indeed influ-ence the integrative response of the Purkinje cell. Complex spikesare associated with a large dendritic Ca2+ transient33,34, which islikely to be attenuated as a consequence of climbing fiber LTD.This Ca2+ transient can have several important effects. It canbriefly modulate dendritic excitability through effects on ionchannels such as Ca2+-dependent K+ channels as well as otherelectrogenic proteins such as the Na+/Ca2+ exchanger. In addi-tion, climbing fiber-evoked Ca2+ transients are critical for induc-tion of parallel fiber LTD14,34, short-term potentiation ofmGluR-mediated responses at the parallel fiber synapses35, andboth LTP36 and transient depression37 of GABAergic interneu-ron–Purkinje cell synapses.

LTP of GABAergic interneuron–Purkinje cell synapsesLTP of both spontaneous and evoked GABAA receptor-mediat-ed IPSCs may be induced by repetitive climbing fiber activation36.This inhibitory LTP (called ‘rebound potentiation’) requires apostsynaptic Ca2+ transient, as it can be blocked by loading Purkinje cells with BAPTA and can be induced after application ofdepolarizing voltage steps in place of climbing fiber stimulation.The Ca2+ transient that triggers rebound potentiation seems torequire Ca2+ mobilization from internal stores38,39. Inhibitors ofeither calcium/calmodulin-dependent kinase II (CaMKII)40 orPKA41 have been reported to block this form of LTP. The role ofinhibitory synaptic activity in this process is somewhat contro-versial. It has been suggested that simultaneous activity ofinhibitory synapses is needed for the induction of rebound poten-tiation42. This finding is in contrast to the recent demonstrationthat simultaneous inhibitory activity can suppress rebound poten-tiation when inhibitory interneurons are activated within ∼ 1 s ofclimbing fibers. This effect was described to be mediated througha cascade involving postsynaptic GABAB receptors and inhibitionof adenylyl cyclase and cAMP production41. If the latter findingprevails, it would comprise an unusual computation: a heterosy-naptic potentiation that is suppressed by homosynaptic synchrony.A protocol for reversing this form of LTP has yet to be described.

Alterations of interneuron–Purkinje cell synaptic strength arelikely to have a major influence on Purkinje cell throughput.Indeed, it has been demonstrated that single action potentialsevoked in inhibitory interneurons can generate delays in Purkinjecell action potential firing and that tonic inhibitory input candrastically modulate the spike firing pattern in Purkinje cells43.

LTP of mossy fiber–granule cell synapsesMossy fibers constitute a large glutamatergic system of the brainand have granule cells as their primary target (there are as manyas 1011 granule cells and 4 times as many mossy fiber–granulecell synapses), providing a large potential substrate for informa-tion storage. Activation of mossy fibers using either a theta burstpattern (100 Hz for 100 ms repeated 8 times at 250 ms intervals),or a single tetanus (100 Hz for 1 s) that mimics endogenous activ-

ity44,45, paired with postsynaptic depolarization to –40 mV, resultsin LTP of mossy fiber EPSCs46,47 (Fig. 3d and e), whereas pro-tracted low-frequency stimulation (2 Hz for 5 min) causes LTD(Maffei, Rossi and D’Angelo, unpublished data). LTP is similarto that commonly recorded in hippocampal area CA1 in that itrequires postsynaptic depolarization, NMDA receptor activationand consequent Ca2+ influx. As such, it is quite sensitive to thelevel of inhibitory input47, which, in this case, is provided byGolgi cells48. Mossy fiber LTP can also be blocked by an mGluRantagonist or a PKC inhibitor46,49. During mossy fiber LTPexpression, amplitudes of both the AMPA and the NMDA recep-tor-mediated synaptic currents are increased. Mossy fiber LTPexpression is also associated with a change in the kinetics of theNMDA-EPSC such that deactivation is protracted. This couldconstitute an extension of the window for EPSP temporal sum-mation and, thereby, coincidence detection. The role of mossyfiber–granule cell LTP/LTD is not completely understood. Where-as LTP would promote population coding, LTD would promotesparse coding in the cerebellum granular layer1,2. A regulation ofsynaptic strength at this synapse, together with changes in intrin-sic excitability and in the strength of Golgi cell–granule cell con-nections, may have a critical influence on the selection of mossyfiber patterns to be relayed to the cerebellar cortex50.

LTP and LTD of synapses received by DCN cellsAlthough LTP and LTD of synapses that are received by granule orPurkinje cells are one possible way to achieve acquisition andextinction of eyelid conditioning, plasticity of the synapses receivedby the DCN cells could also modulate DCN firing and hence CRexpression. Unfortunately, only a single preliminary report of use-dependent plasticity of mossy fiber–DCN synapses has appeared inthe literature. Racine and collaborators51 used field potentialrecording in both acute and chronic preparations of intact rat cere-bellum to show that a component of the field potential was poten-tiated by repeated bursts of 300 Hz stimuli to the inferior peduncle.This potentiation ranged from 118–150% of baseline, was seen in5 of 6 animals and lasted up to 8 days in chronic recordings.

Recent work has indicated that the GABAergic Purkinjecell–DCN synapses also exhibit LTP and LTD52. Activation ofthese synapses with a burst-and-pause stimulus results in a trainof summating hyperpolarizing IPSPs followed by a prominentrebound depolarization and associated spike burst. This rebounddepolarization provides a mechanism by which inhibitory inputscan drive postsynaptic excitation. In these cells, LTP is elicited bytrains of IPSPs, which reliably evoke a rebound depolarizationin the DCN neurons. LTD is induced if the same protocol isapplied while the amount of postsynaptic excitation is reduced(by postsynaptic hyperpolarization or an internal Na+ channelblocker), and no change is produced if a Ca2+ chelator is appliedpostsynaptically. The polarity of the change in synaptic strengthis correlated with the amount of rebound depolarization-evokedspike firing and the amplitude of the resulting postsynaptic Ca2+

transient. These results speak to two important points that haveimplications beyond cerebellar function. First, they suggest asolution for a long-standing problem in cellular synaptic plas-ticity: how do inhibitory synapses produce postsynaptic Ca2+ sig-nals that are required for use-dependent plasticity when they donot directly gate a Ca2+ conductance? Second, they show that animportant computational principle that governs the inductionof use-dependent change in excitatory synaptic efficacy, namelythat LTP and LTD are triggered by stimuli that result in high andmoderate Ca2+ signal amplitudes, respectively53–55, can also applyto inhibitory synapses. It remains to be determined whether the

review


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



LTP and/or LTD thresholds can also be changed as a function ofrecent synaptic activity (so-called ‘metaplasticity,’ another pre-diction of the excitatory synapse models cited above).

It is likely that any stimulus that appropriately elevates subsy-naptic Ca2+ in DCN neurons will result in LTP/LTD of Purkinjecell–DCN synapses. Indeed, LTP may be produced either by repeat-ed postsynaptic hyperpolarizing pulses, which evoke rebounddepolarization52 or by depolarizing pulses56. The expression mech-anism remains unclear, as it could be detected using exogenouspulses of a GABAA receptor agonist, suggesting a postsynaptic locusof expression, yet it was associated with an increase in the frequencybut not the amplitude of mIPSCs, suggesting a presynaptic locus ofexpression56. LTD of inhibitory input to the DCN may also be pro-duced using postsynaptic depolarizing pulses delivered at a lowerfrequency57. Both LTP and LTD of inhibitory input may be inducedheterosynaptically, by repeated stimulation of glutamatergic mossyfiber synapses at 100 or 10 Hz, respectively56,57. Both LTP and LTDof these inhibitory synapses induced in this manner require NMDAreceptor activation and elevation of postsynaptic Ca2+. At present,it is unclear whether induction of LTP/LTD in an NMDA recep-tor-independent manner52 and an NMDA receptor-dependentmanner56,57 engage the same signaling cascades downstream ofpostsynaptic Ca2+ elevation.

Non-synaptic plasticity in cerebellar neuronsMemory storage in the brain is generally assumed to be mediatedby long-term modifications in the strength of synaptic transmis-sion. Whereas LTP and LTD are computationally appealing (inpart because they allow for the synapse-specific modi-fication of a large array of inputs), and it is likely thatthey are central to memory storage, there are also rea-sons to believe that they are not the whole story. Oneidea, which has received relatively little attention, is that

information storage in the mammalian brain may also involveactivity-dependent changes in the intrinsic excitability of neu-rons. This type of plasticity could be a local phenomenon that isrestricted to one synapse, but it could also have the property ofaffecting postsynaptic signals evoked by a large number of (orpossibly even all) synapses impinging on a particular neuron,depending upon the identity and location of the intrinsic (volt-age-sensitive) conductances altered, thereby creating a more glob-al change in signal integration. Until recently, there have only beenscattered results in support of a role for intrinsic, non-synapticplasticity in memory storage. For example, there is evidence forpersistent changes in intrinsic neuronal excitability induced bybehavioral training. Studies using Hermissenda have shown thatassociative pairing of light and rotation produces a long-termincrease in the intrinsic excitability of type B photoreceptors, dueto a reduction of K+ currents58. A similar effect has been seen fol-lowing various forms of associative conditioning in neocorticalneurons59, CA1 pyramidal neurons of the hippocampus60 andcerebellar Purkinje cells61. In addition, slowly developing changesin the intrinsic excitability of both invertebrate and mammalianneurons have been reported in cell culture systems when activi-ty was blocked with tetrodotoxin62,63. However, rapid, synapti-cally driven changes in intrinsic excitability of mammalianneurons had not been convincingly demonstrated.

Aizenman and Linden64 made current-clamp recordings fromthe DCN of juvenile rat cerebellar slices in which a short depo-larizing test pulse that evoked a small number of action poten-tials was applied. After a baseline period, a synaptic tetanus was

review

Fig. 3. Rapid, synaptically driven increases in intrinsic excitabil-ity of DCN and granule cells. (a) Tetanization of excitatorysynapses produced a sustained increase in intrinsic excitabilityof DCN neurons. Averaged time course of the number ofspikes evoked during 200 ms, depolarizing current test pulses.Control experiments (open circles) show that a stable baselinecan be recorded for at least 25 min. Application of a synapticconditioning tetanus (at arrowheads, filled circles) resulted in aslowly developing increase in the number of spikes evoked bythe test stimulus. Error bars, s.e.m. Representative traces,evoked by a +0.5 nA test pulse, from tetanized (b) and control(c) experiments at the time points indicated by asterisks.Traces evoked by sub-threshold test pulses demonstrate thatRinput remains stable when an increase in spiking is observedwith larger depolarizing test pulses. (d) Tetanization of mossyfibers (arrowheads) produced a sustained increase in synapticresponses and intrinsic excitability of granule cells when synap-tic inhibition was turned off (bicuculline) but not when it wasleft active (bicuculline-free). The nature of synaptic and non-synaptic changes is illustrated in (e). EPSCs (comprising bothAMPA and NMDA components) increased following tetaniza-tion (arrowhead). Membrane excitability (the opposite of cur-rent threshold in current-clamp recordings) also increased. Thecombination of these changes caused a large increase in synap-tic excitation, evaluated as the probability of action potentialgeneration by EPSPs. Dotted lines in the top and middle panelare interpolation of control experiments, showing stable base-lines throughout LTP recordings. (a–c) From ref. 64. (d, e)From refs. 46 and 47. Copyright 1999 and 2000 by theAmerican Physiological Society and the Society forNeuroscience, respectively.

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


delivered to the DCN neuron using a stimulating electrode placedin adjacent white matter, to activate glutamatergic mossy fibers.(GABAA receptors were blocked with picrotoxin.) This synapticconditioning tetanus resulted in a slowly developing but dramaticincrease in the number of spikes evoked by the test pulse, whichstabilized over a period of 15–20 min and which was associatedwith a decrease in the spike threshold (Fig. 3a–c). Subsequentexperiments showed that induction of this persistent increase inintrinsic excitability required Ca2+ influx mediated by eitherNMDA receptors (when triggered by a synaptic conditioningtetanus) or voltage-sensitive Ca2+ channels (when triggered byrepeated, direct somatic current injection). Thus, whereas thesestudies showed this persistent increase in intrinsic excitability tobe driven by bursts delivered to the mossy fibers, it is likely thatany stimulus that results in the appropriate postsynaptic Ca2+

transient will produce this effect. For example, a burst-and-pausestimulus delivered to the GABAergic Purkinje cell–DCN synaps-es results in rebound depolarization and an associated Ca2+ tran-sient52,65, which would be an excellent candidate for such a signal.

When theta burst stimulation (bursts of 10 pulses with aninterpulse interval of 10 ms, repeated with an interburst intervalof 200 ms, designed to mimic firing during theta oscillations)plus postsynaptic depolarization was applied to mossyfiber–granule cell synapses, a persistent increase in the intrinsicexcitability of the postsynaptic granule cell was observed47

(Fig. 3d and e). As previously described, this form of stimulationalso gives rise to LTP of the mossy fiber EPSCs. Like mossyfiber–granule cell LTP, the increased excitability was dependentupon postsynaptic depolarization, NMDA receptor activationand a postsynaptic Ca2+ transient. However, the increase inintrinsic excitability required less membrane depolarization (≥ 60 mV) than the mossy fiber LTP (≥ 40 mV). Thus, weak gran-ule cell depolarization shall only induce the increase in mem-brane excitability, whereas strong granule cell depolarization shallincrease membrane excitability and synaptic conductance togeth-er. Once again, the influence of membrane depolarization sug-gests that Golgi cell inhibition is critical for fine tuning plasticityat the mossy fiber–granule cell relay. Potentiation of intrinsicexcitability may be relevant in conditions of low synaptic excita-tion, restoring granule cell readiness (compensatory effect66) andregulating the number of granule cells effectively activated bydischarging mossy fibers. At present, it has yet to be determinedif the use-dependent increases in intrinsic excitability in granuleor DCN cells may be actively reversed.

Use-dependent alterations in intrinsic excitability, as shownin cerebellar granule and DCN cells, have the potential to broad-ly affect neuronal throughput. Indeed, the probability of spike fir-ing as evoked by every excitatory synapse on a postsynaptic neuron

may be altered. This is based on the assumption that either thepostsynaptic neuron is extremely electrotonically compact (as isthe case for the cerebellar granule cell67) or, in noncompact neu-rons, the conductances that are persistently altered by depolariz-ing current injection are physically interposed between excitatorysynapses and the spike initiation zone, as would be expected ifthey were localized to the soma or axon hillock. The spatial local-ization of these conductances has yet to be determined and may beof great computational importance. For example, if a group ofsynapses impinging upon a DCN neuron were activated to evokea local Ca2+ transient, this might produce a local change of volt-age-gated ion channels and thereby modulate throughput, notfrom individual synapses, or from a whole neuron, but ratherfrom a dendritic module consisting of the originally activatedsynapses plus all those more distal on the same dendritic branch.

Emergent properties of a plastic cerebellar circuitLet us return to the case of associative eyelid conditioning. Can thecerebellar circuit, with the complement of use-dependent mecha-nisms described herein (Fig. 4), be reasonably expected to underliethis simple form of learning and memory? To account for acquisi-tion of the CR, we need a location or locations in the circuit whereCS and US information converge and where an associative com-putation is performed that ultimately results in an increase in fir-ing in the DCN. This increase in firing must be appropriately timedto shortly precede the US onset. Furthermore, to account for extinc-tion of the CR, we need a location or locations in the circuit whereCS information is received and where its repeated presentationalone will ultimately result in a decrease in DCN firing during theCS–US interval. In addition to acquisition and extinction, it wouldbe useful if the cerebellar circuit could account for some of the morecomplex phenomena related to associative eyelid conditioning. Forexample, if an animal is repeatedly presented with a tone (CS)–air-puff (US) pairing until it reliably acquires tone-evoked eyelidresponses (CRs), but is then presented with repeated tones alone,it will undergo ‘extinction’ so that it no longer performs eyelidresponses to the tone. However, if tone–airpuff pairing is thenresumed, the rate of acquiring the tone-evoked eyelid response willbe greater in this second training session. This is referred to as ‘sav-ings.’ A related phenomenon is called ‘CS generalization.’ This refersto the finding that an animal that has previously undergone acqui-sition and complete extinction to a CS–US pair (tone–airpuff pairsfollowed by repeated tones presented alone), in a second trainingsession, will acquire eyelid response CRs faster to pairing of theairpuff US with a different CS, such as a light. Finally, ‘US pre-treatment’ refers to the observation that repeated presentations of aUS before CS–US pairing will result in slower CR acquisition, as ifthe US were somehow less effective.

Clearly, one place where the streams of CS and US informa-tion converge is the Purkinje cell, in which parallel fiber LTD

review


Purkinjecell Climbing

fiber


nuclei

Premotorareas

Deepcerebellar

nuclei

Mossyfibers

Parallel fibers

Granulecell


Climbingfiber

collateral

Fig. 4. A summary of synaptic and non-synaptic plasticity in the cerebel-lar circuit. In this drawing, the occurrence of long-term use-dependentplasticity has been coded with color: red indicating potentiation and blueindicating depression. Red action potentials in neuronal somata indicatepersistent increases in intrinsic excitability, whereas bars of color atsynapses indicate conventional synaptic LTP or LTD. The location of thecolor bar in the presynaptic or postsynaptic membrane represents thebest current understanding of the locus of expression of these forms ofLTD. In some locations, such as the PF–PC synapse, the data in supportof these expression sites are good, whereas in others, such as thePC–DCN synapse, much remains to be determined. It is likely that addi-tional forms of long-term synaptic and non-synaptic plasticity will have tobe added to this diagram in the near future.

Bob

Crim

i

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



could drive CR acquisition through DCN disinhibition as envi-sioned in the models of Marr, Albus, Ito and Thompson. Howcan this mechanism produce CRs that are appropriately timed(so that the CR begins shortly before the US onset)? If oneassumes that a large population of granule cells codes for the CSin a continuum of different ways, with some firing at the begin-ning of the CS, others at the end of the CS, and others in the mid-dle, then those PF synapses from granule cells that fire in closeproximity to US onset will be depressed, whereas those that firefar from US onset will be potentiated by parallel fiber LTP. Whentransduced by the Purkinje cell–DCN inhibitory synapse, thiswill result in lower DCN firing rates in the period immediatelyfollowing CS onset and higher firing rates shortly before USonset, thereby producing a well-timed CR68,69.

However, the Purkinje cell is not the only site that receivesconvergent CS and US information. The DCN cells receive bothmossy fiber and climbing fiber collateral synapses. Unfortunate-ly, the latter have not been well characterized physiologically. Ifclimbing fiber/mossy fiber coactivation resulted in either LTP ofmossy fiber DCN synapses51 or a persistent increase in DCNintrinsic excitability64, this could potentially underlie at least aportion of CR acquisition (see below). The main route for climb-ing fibers to excite the DCN may be disynaptic. When climbingfibers fire, there is a transient increase in the firing rate of Purkinje cells, followed by a pause. When this pattern of activityis applied to Purkinje cell–DCN synapses, a strong rebound depo-larization and consequent Ca2+ transient is produced52,65.

We must also consider that the memory trace for CR acquisi-tion may be stored in multiple cerebellar locations and that plas-ticity at one site may induce plasticity at another site as asecondary consequence. For example, in a model proposed byMauk and collaborators, the memory trace of the eyelid CR issequentially stored, first as a depression of the parallel fiber–Purkinje cell synapse in the cerebellar cortex. This would result inan attenuation of Purkinje cell firing and hence Purkinjecell–DCN synaptic input, thereby disinhibiting the DCN targets.This disinhibition, when coupled with activation of the mossyfiber–DCN synapse, could then potentiate the latter or increasethe DCN intrinsic excitability, resulting in a distributed memorytrace. In this scheme, aspects of the CR (mostly amplitude) arestored at the mossy fiber–DCN synapse, whereas other aspectsof the CR (mostly timing information) are retained in the par-allel fiber Purkinje cell synapse68,70,71. A computational analysishas suggested that this model could constitute a memory tracethat is unusually resistant to degradation by ongoing ‘back-ground’ activity in the cerebellar circuit72. Furthermore, granulecell populations active at specific times may be consolidated byLTP or LTD at the mossy fiber–granule cell synapse, or by changesin granule cell intrinsic excitability.

There are several ways in which repeated CS informationcould ultimately give rise to decreased DCN firing and conse-quent extinction of CRs. Both the mossy fiber–granule cellsynapse and the parallel fiber–Purkinje cell synapse undergo LTPwith repeated activation. In addition, granule cell intrinsicexcitability is increased following repeated mossy fiber stimula-tion. If the acquisition of the CR involves a particular modifica-tion at a specific synaptic or cellular location, should extinctionproduce a complete and exact reversal of these effects? WhereasLTD and LTP of the parallel fiber synapse are obvious candidatesto underlie CR acquisition and extinction, respectively, they donot truly reverse each other, as the former is expressed postsy-naptically, whereas the latter is expressed presynaptically.Although a parallel fiber synapse may undergo LTD, and then

LTP, to return it to the same aggregate synaptic strength as mea-sured with single pulses, this synapse will not then function inthe same way as it did in the naive state. For example, if it nowhas a higher probability of release, it will be more subject tofatigue during prolonged firing. This situation may become evenmore complex if a portion of the memory trace for CR acquisi-tion (such as the mossy fiber synaptic strength or intrinsicexcitability in the DCN) is not altered at all following extinction.

Lest we become too discouraged here, it should be noted thatthe ‘reversibility problem’ may have a silver lining. Clearly, toaccount for savings and CS generalization, the state of the cere-bellar circuit after CR acquisition followed by extinction cannotbe identical to that before training. Residual plasticity in one ormore cerebellar locations is likely to underlie these phenomena.Recently, local reversible inactivation during training, combinedwith a computer simulation, has suggested that the deep nucleimight be one site of this residual plasticity71. Whereas this resid-ual plasticity could take the form of either LTP of the mossyfiber–DCN synapse or an increase in intrinsic excitability, thelatter may provide a better explanation for generalization betweenCSs that are coded by distinct populations of mossy fibers.

In some forms of associative classical conditioning, repeatedUS presentation before CS–US pairing slows the rate of CR acqui-sition. Presumably, it is less behaviorally relevant to learn a CS–USassociation if the US alone has been previously applied repeated-ly. Repeated US presentation will activate climbing fibers and mayresult in LTD of climbing fiber–Purkinje cell synapses. This couldthen slow down acquisition of CRs in two ways, both related toreduction of the climbing fiber-evoked Ca2+ transient. First, asmaller Ca2+ transient could make it harder to induce parallelfiber-Purkinje cell LTD from subsequent parallel fiber/climbingfiber coactivation. Second, a smaller Ca2+ transient could reducethe amplitude of the Ca2+-dependent K+ conductance, which isa good candidate for causing the climbing fiber-evoked pause inongoing Purkinje cell spiking. This could then have a secondaryeffect of delimiting rebound excitation in the DCN and therebyattenuating the potential contribution of climbing fiber activa-tion to synaptic and non-synaptic plasticity in this structure.

To this point, we have made a set of proposals about howsome cellular phenomena in the cerebellum might constitute thememory trace for a simple form of motor learning and how theirinteraction might account for some known properties of thislearning such as extinction, savings, CS generalization and soforth. Although this might seem to be sufficient, it is likely thatthis circuit must have the capacity to simultaneously encode morethan this one simple form of motor learning. Thus, in addition tothe cellular alterations that initially lay down the memory trace,there must also be a set of homeostatic phenomena that regulatethe overall activity of the circuit so that the efficacy of synapses orthe intrinsic properties of neurons are not driven either to theirabsolute minimum or maximum63,73,74. The latter is particular-ly dangerous in the case of excitatory synapses, because not onlyis the subsequent information storage capacity of the networkcompromised, but there is also the risk of damage by Ca2+-medi-ated excitotoxic processes.

There are several phenomena in the cerebellar circuit thatmay serve homeostatic or antiexcitotoxic functions. For exam-ple, the climbing fiber input evokes massive excitation of thePurkinje cell, which results in large dendritic Ca2+ transients33,34.At the same time, bursts of climbing fiber activity may providea means for general activity reduction in the cerebellar cortex inthat it can induce LTP of inhibitory interneuron–Purkinje cellsynapses and LTD of coactive excitatory parallel fiber synapses, as

review©

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

well as homosynaptic climbing fiber LTD. Similarly, LTP and LTDof Purkinje cell–DCN synapses may serve as a homeostatic mech-anism to regulate the overall excitability and dynamic range ofthe DCN cell. For example, the increase in intrinsic excitabilitythat can be produced by strong postsynaptic spiking64 could con-stitute a positive feedback loop that would drive spiking to dan-gerously high levels. However, large amounts of postsynapticspiking will also potentiate Purkinje cell-evoked IPSPs, makingthe cell less likely to fire more action potentials, whereas LTDtends to be induced when the postsynaptic activity is limited52,increasing the probability of postsynaptic spiking.

Some future directionsAlthough the holy grail of relating cellular phenomena to behav-ior through the intermediate levels of cerebellar synapses and cir-cuits is not yet at hand, the good news is that it seems possible.Perhaps the greatest gap in our present attempts to build theseconnections is our level of understanding of the ways in whichbehaviorally relevant signals are coded. For example, does repeat-ed presentation of a tone CS result in a pattern of mossy fiber acti-vation that will support LTP of mossy fiber-granule cell or parallelfiber–Purkinje cell synapses? Does repeated presentation of a toneCS when combined with an airpuff US at intervals that supportassociative conditioning result in a pattern of activation that willproduce parallel fiber LTD? Does repeated presentation of an air-puff US result in a mode of climbing fiber activation that will pro-duce climbing fiber LTD? As these data become available, cellularelectrophysiologists must then make their best efforts to simulatethese in vivo training signals in reduced preparations such as brainslices and to provide a greater data set for the parameters of theseplastic phenomena, particularly in regard to duration.

Another important line of work will involve incorporating awider range of cerebellar plasticity into computer simulations. Eventhough the cerebellar network is ‘simple’ by the standards of theCNS, the proliferation of cellular plastic phenomena has nowreached the point where it is not possible to make straightforwardpredictions about the behavior of the circuit during and after train-ing. Computer models of multiple plastic phenomena in the cere-bellar circuit have already given rise to some very unexpectedbehavioral predictions that have been confirmed experimental-ly69. Moreover, many of the critical experiments to address thebehavioral/cellular link will continue to involve the analysis ofbehavior together with pharmacological and genetic manipula-tions. Further molecular understanding of the second messengersand substrates involved in these various forms of plasticity will notonly constrain certain aspects of these models but will allow forthe development of more sophisticated transgenic and pharma-cological strategies. Whereas these general approaches are beingused in the analysis of memory storage in many brain regions, webelieve that the cerebellar circuit offers a unique opportunity todevelop a complete and specific theory of the engram.

ACKNOWLEDGEMENTSThis work was supported by NWO-ALW 810.37.003 (C.H.), USPHS MH01590,

MH51106, MH61974 and the Develbiss Fund (D.J.L.) and EC grants PL97 0182

and PL97 6060, and INFM PRA-Cady (E.D.). We thank C. De Zeeuw,

M. Mauk, M. Schmolesky and J. Weber for comments.

RECEIVED 16 JANUARY; ACCEPTED 27 MARCH 2001

1. Marr, D. Theory of cerebellar cortex. J. Physiol. (Lond.) 202, 437–455 (1969).2. Albus, J. S. A theory of cerebellar function. Math. Biosci. 10, 25–61 (1971).3. Kim, J. J. & Thompson, R. F. Cerebellar circuits and synaptic mechanisms

involved in classical eyeblink conditioning. Trends Neurosci. 20, 177–181(1997).

4. Mauk, M. D. Roles of cerebellar cortex and nuclei in motor learning:contradictions or clues? Neuron 18, 343–346 (1997).

5. De Zeeuw, C. I. et al. Expression of a protein kinase C inhibitor in Purkinjecells blocks cerebellar long-term depression and adaptation of the vestibulo-ocular reflex. Neuron 20, 495–508 (1998).

6. Raymond, J. L. & Lisberger, S. G. Hypotheses about the neural trigger forplasticity in the circuit for the vestibulo-ocular reflex. Prog. Brain Res. 124,235–246 (2000).

7. Ito, M., Sakurai, M. & Tongroach, P. Climbing fiber induced depression ofboth mossy fiber responsiveness and glutamate sensitivity of cerebellarPurkinje cells. J. Physiol. (Lond.) 324, 113–134 (1982).

8. Hartell, N.A. Strong activation of parallel fibers produces localized calciumtransients and a form of LTD that spreads to distant synapses. Neuron 16,601–610 (1996).

9. Eilers, J., Takechi, H., Finch, E. A., Augustine, G. J. & Konnerth, A. Localdendritic Ca2+ signaling induces cerebellar long-term depression. Learn.Mem. 4, 159–168 (1997).

10. Bear, M. F. & Linden, D. J. in The Synapse (W. M. Cowan & K. Davies, eds.)455–517 (Johns Hopkins Univ. Press, 2000).

11. Wang, Y.-T. & Linden, D. J. Expression of cerebellar long-term depressionrequires postsynaptic clathrin-mediated endocytosis. Neuron 25, 635–647(2000).

12. 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).

13. Sakurai, M. Synaptic modification of parallel fibre-Purkinje cell transmissionin in vitro guinea-pig cerebellar slices. J. Physiol. (Lond.) 394, 463–480(1987).

14. Sakurai, M. Calcium is an intracellular mediator of the climbing fiber ininduction of cerebellar long-term depression. Proc. Natl. Acad. Sci. USA 87,3383–3385 (1990).

15. Hirano, T. Depression and potentiation of the synaptic transmission betweena granule cell and a Purkinje cell in rat cerebellar culture. Neurosci. Lett. 119,141–144 (1990).

16. Hirano, T. Differential pre- and postsynaptic mechanisms for synapticpotentiation and depression between a granule cell and a Purkinje cell in ratcerebellar culture. Synapse 7, 321–323 (1991).

17. Crepel, F. & Jaillard, D. Pairing of pre- and postsynaptic activities in cerebellarPurkinje cells induces long-term changes in synaptic efficacy in vitro. J. Physiol. (Lond.) 432, 123–141 (1991).

18. Shibuki, K. & Okada, D. Cerebellar long-term potentiation under suppressedpostsynaptic Ca activity. Neuroreport 3, 231–234 (1992).

19. Salin, P. A., Malenka, R. C. & Nicoll, R. A. cAMP mediates a presynaptic formof LTP at cerebellar parallel fiber synapses. Neuron 16, 797–803 (1996).

20. Linden, D. J. Long-term potentiation of glial synaptic currents in cerebellarculture. Neuron 18, 983–994 (1997).

21. Linden, D. J. Synaptically-evoked glutamate transport currents may be usedto detect the expression of long-term potentiation in cerebellar culture. J. Neurophysiol. 79, 3151–3156 (1998).

22. Linden, D. J. & Ahn, S. Activation of presynaptic cAMP-dependent proteinkinase is required for induction of cerebellar long-term potentiation. J. Neurosci. 19, 10221–10227 (1999).

23. Chavis, P., Mollard, P., Bockaert, J. & Manzoni, O. Visualization of cyclicAMP-regulated presynaptic activity at cerebellar granule cells. Neuron 20,773–781 (1998).

24. Kimura, S., Uchiyama, S., Takahashi, H. E. & Shibuki, K. cAMP-dependentlong-term potentiation of nitric oxide release from cerebellar parallel fibersin rats. J. Neurosci. 18, 8551–8558 (1998).

25. Storm, D. R., Hansel, C., Hacker, B., Parent, A. & Linden, D. J. Impairedcerebellar long-term potentiation in type I adenylyl cyclase mutant mice.Neuron 20, 1199–1210 (1998).

26. Dittman, J. S. & Regehr, W. G. Calcium dependence and recovery kinetics ofpresynaptic depression at the climbing fiber to Purkinje cell synapse. J. Neurosci. 18, 6147–6162 (1998).

27. Crepel, F., Mariani, J. & Delhaye-Bouchaud, N. Evidence for a multipleinnervation of Purkinje cells by climbing fibers in the immature ratcerebellum. J. Neurobiol. 7, 567–578 (1976).

28. Strata, P. & Rossi, F. Plasticity of the olivocerebellar pathway. Trends Neurosci.21, 407–413 (1998).

29. Hansel, C. & Linden, D. J. Long-term depression of the cerebellar climbingfiber-Purkinje neuron synapse. Neuron 26, 473–482 (2000).

30. Llinas, R. & Sugimori, M. Electrophysiological properties of in vitro Purkinjecell somata in mammalian cerebellar slices. J. Physiol. (Lond.) 305, 171–195(1980).

31. Llinas, R. & Sugimori, M. Electrophysiological properties of in vitro Purkinjecell dendrites in mammalian cerebellar slices. J. Physiol. (Lond.) 305, 197–213(1980).

32. Raman, I. M. & Bean, B. P. Resurgent sodium current and action potentialformation in dissociated cerebellar Purkinje neurons. J. Neurosci. 17,4517–4526 (1997).

33. Ross, W. N. & Werman, R. Mapping calcium transients in the dendrites ofPurkinje cells from the guinea-pig cerebellum in vitro. J. Physiol. (Lond.) 389,319–336 (1987).

review


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



34. Konnerth, A., Dreessen, J. & Augustine, G. J. Brief dendritic calcium signalsinitiate long-lasting synaptic depression in cerebellar Purkinje cells. Proc.Natl. Acad. Sci. USA 89, 7051–7055 (1992).

35. Batchelor, A. M. & Garthwaite, J. Frequency detection and temporallydispersed synaptic signal association through a metabotropic glutamatereceptor pathway. Nature 385, 74–77 (1997).

36. Kano, M., Rexhausen, U., Dreessen, J. & Konnerth, A. Synaptic excitationproduces a long-lasting rebound potentiation of inhibitory synaptic signalsin cerebellar Purkinje cells. Nature 356, 601–604 (1992).

37. Llano, I., Leresche, N. & Marty, A. Calcium entry increases the sensitivity ofcerebellar Purkinje cells to applied GABA and decreases inhibitory synapticcurrents. Neuron 6, 565–574 (1991).

38. Hashimoto, T., Ishii, T. & Ohmori, H. Release of Ca is the crucial step for thepotentiation of IPSCs in the cultured cerebellar Purkinje cells of the rat. J. Physiol. (Lond.) 497, 611–627 (1996).

39. Khodakhah, K. & Armstrong, C.M. Induction of long-term depression andrebound potentiation by inositol trisphosphate in cerebellar Purkinjeneurons. Proc. Natl. Acad. Sci. USA 94, 14009–14014 (1997).

40. Kano, M., Kano, M., Fukunaga, K. & Konnerth, A. Ca-induced reboundpotentiation of γ-aminobutyric acid-mediated currents requires activation ofCa/calmodulin-dependent kinase II. Proc. Natl. Acad. Sci. USA 93,13351–13356 (1996).

41. Kawaguchi, S. & Hirano, T. Suppression of inhibitory synaptic potentiationby presynaptic activity through postsynaptic GABAB receptors in a Purkinjeneuron. Neuron 27, 339–347 (2000).

42. Kano, M. Long-lasting potentiation of GABAergic inhibitory synaptictransmission in cerebellar Purkinje cells: its properties and possiblemechanisms. Behav. Brain Sci. 19, 354–361 (1996).

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

44. Kase, M., Miller, D. C. & Noda, H. Discharges of Purkinje cells and mossyfibres in the cerebellar vermis of the monkey during saccadic eye movements.J. Physiol. (Lond.) 300, 539–555 (2000).

45. Pellerin, P.-P. & Lamarre, Y. Local field potential oscillations in primatecerebellar cortex during voluntary movement. J. Neurophysiol. 78, 3502–3507(1997).

46. D’Angelo, E., Rossi, P., Armano, S. & Taglietti, V. Evidence for NMDA andmGlu receptor-dependent long-term potentiation of mossy fibre–granulecell transmission in rat cerebellum. J. Neurophysiol. 81, 277–287 (1999).

47. Armano, S., Rossi, P., Taglietti, V. & D’Angelo, E. Long-term potentiation ofintrinsic excitability at the mossy fiber-granule cell synapse of rat cerebellum.J. Neurosci. 20, 5208–5216 (2000).

48. Maex, R. & De Schutter, E. Synchronization of Golgi and granule cell firing ina detailed network model of the cerebellar granule cell layer. J. Neurophysiol.80, 2521–2537 (1998).

49. Rossi, P., D’Angelo, E. & Taglietti, V. Differential long-lasting potentiation ofthe NMDA and non-NMDA synaptic currents induced by metabotropic andNMDA receptor coactivation in cerebellar granule cells. Eur. J. Neurosci. 8,1182–1189 (1996).

50. Schweighofer, N., Doya, K. & Lay, F. Unsupervised learning of granule cellsparse codes enhances cerebellar adaptive control. Neuroscience 103, 35–50(2001).

51. Racine, R. J., Wilson, D. A., Gingell, R. & Sutherland, D. Long-termpotentiation in the interpositus and vestibular nuclei in the rat. Exp. BrainRes. 63, 158–162 (1986).

52. Aizenman, C., Manis, P. B. & Linden, D. J. Polarity of long-term synaptic gainchange is related to postsynaptic spike firing at a cerebellar inhibitorysynapse. Neuron 21, 827–835 (1998).

53. Bienenstock, E. L., Cooper, L. N. & Munro, P. W. Theory for the development

of neuron selectivity: orientation specificity and binocular interaction invisual cortex. J. Neurosci. 2, 32–48 (1982).

54. Artola, A. & Singer, W. Long-term depression of excitatory synaptictransmission and its relationship to long-term potentiation. Trends Neurosci.16, 480–487 (1993).

55. Hansel, C., Artola, A. & Singer, W. Relation between dendritic Ca levels andthe polarity of synaptic long-term modifications in rat visual cortex neurons.Eur. J. Neurosci. 9, 2309–2322 (1997).

56. Ouardouz, M. & Sastry, B. R. Mechanisms underlying LTP of inhibitorysynaptic transmission in the deep cerebellar nuclei. J. Neurophysiol.84,1414–1421 (2000).

57. Morishita, W. & Sastry, B. R. Postsynaptic mechanisms underlying long-termdepression of GABAergic transmission in neurons of the deep cerebellarnuclei. J. Neurophysiol. 76, 59–68 (1996).

58. Alkon, D. L. Changes of membrane currents during learning. J. Exp. Biol.112, 95–112 (1984).

59. Brons, J. F. & Woody, C. D. Long-term changes in excitability of corticalneurons after Pavlovian conditioning and extinction. J. Neurophysiol. 44,605–615 (1980).

60. Disterhoft, J. F., Coulter, D. A. & Alkon, D. L. Conditioning-specificmembrane changes of rabbit hippocampal neurons measured in vitro. Proc.Natl. Acad. Sci. USA. 83, 2733–2737 (1986).

61. Schreurs, B. G., Gusev, P. A., Tomsic, D., Alkon, D. L. & Shi, T. Intracellularcorrelates of acquisition and long-term memory of classical conditioning inPurkinje cell dendrites in slices of rabbit cerebellar lobule HVI. J. Neurosci.18, 5498–5507 (1998).

62. Desai, N. S., Rutherford, L. C. & Turrigiano, G. G. Plasticity in the intrinsicexcitability of cortical pyramidal neurons. Nat. Neurosci. 2, 515–520 (1999).

63. Turrigiano, G. G. Homeostatic plasticity in neuronal networks: the morethings change, the more they stay the same. Trends Neurosci. 22, 221–227(1999).

64. Aizenman, C. D. & Linden, D. J. Rapid, synaptically driven increases in theintrinsic excitability of cerebellar deep nuclear neurons. Nat. Neurosci. 3,109–111 (2000).

65. Aizenman, C. D. & Linden, D. J. Regulation of the rebound depolarizationand spontaneous firing patterns of deep nuclear neurons in slices of ratcerebellum. J. Neurophysiol. 82, 1697–1717 (1999).

66. Fregnac, Y. Homeostasis or synaptic plasticity? Nature 391, 845–846 (1998).67. D’Angelo, E. et al. Theta-frequency bursting and resonance in cerebellar

granule cells: experimental evidence and modeling of a slow K+-dependentmechanism. J. Neurosci. 21, 759–770 (2001).

68. Medina, J. F. & Mauk, M. D. Computer simulation of cerebellar informationprocessing. Nat. Neurosci. 3, 1205–1211 (2000).

69. Ohyama, T. & Mauk, M. D. Latent acquisition of timed responses incerebellar cortex. J. Neurosci. 21, 682–690 (2001).

70. Raymond, J. L., Lisberger, S. G. & Mauk, M. D. The cerebellum: a neuronallearning machine? Science 272, 1126–1131 (1996).

71. Medina, J. F., Garcia, K. S. & Mauk, M. D. A mechanism for savings in thecerebellum. J. Neurosci. (in press).

72. Medina, J. F. & Mauk, M. D. Simulations of cerebellar motor learning:computational analysis of plasticity at the mossy fiber to deep nucleussynapse. J. Neurosci. 19, 7140–7151 (1999).

73. Abbott, L. F. & Nelson, S. B. Synaptic plasticity: taming the beast. Nat.Neurosci. 3, 1178–1183 (2000).

74. Stemmler, M. & Koch, C. How voltage-dependent conductances can adapt tomaximize the information encoded by neuronal firing rate. Nat. Neurosci. 2,521–527 (1999).

75. McCormick, D. A. & Thompson, R. F. Neuronal responses of the rabbitcerebellum during acquisition and performance of a classically conditionednictitating membrane-eyelid response. J. Neurosci. 4, 2811–2822 (1984).

review©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


articles


Ligand-gated ion channels are allosteric proteins that coupleneurotransmitter binding to the opening and subsequentdesensitization of an ion channel that is part of the receptorprotein. A complete understanding of how ligand binding con-trols channel gating requires not only identification of bind-ing site and channel-lining residues, but also knowledge of theconformational movements that occur within these domainsduring channel activation.

GABAA receptors are members of the ligand-gated ion chan-nel gene superfamily that includes receptors for glycine, acetyl-choline, serotonin and, in invertebrates, glutamate1–5. Drugs usedclinically for general anesthesia, and those used to treat anxietydisorders and epilepsy, target GABAA receptors1,3,6, which areformed by the pseudosymmetrical assembly of five homologoussubunits around a central transmembrane channel. Co-expres-sion of α and β subunits mainly forms receptors with a subunitstoichiometry of two α and three β7,8; however, a small subpop-ulation with other stoichiometries such as three α and two β mayexist. In vivo, the most common stoichiometry of subunits isprobably two α , two β and one γ7,9,10. Each subunit has an N-terminal extracellular domain of about 200 amino acids, whichincludes the agonist binding sites, and a similar sized C-termi-nal domain with four membrane-spanning segments (M1, M2,M3, M4). The ion channel is primarily lined by five M2 segments,one from each subunit. Ten of 26 residues in and flanking the α1M2 segment are on the water-accessible protein surface, andprobably line the channel (Fig. 1a)11. On an α-helical wheel, thechannel-lining residues lie on one face, implying that the M2 seg-ment secondary structure is α helical. To facilitate comparisonswith other receptor subunits in the gene superfamily, aligned

positions in the M2 segment are referred to by an indexing num-ber system in which the 1´ position refers to α1Thr256 andβ1Val251 at the cytoplasmic end of M2, and the 20´ position refersto α1Asn275 and β1Glu270, the extracellular ring of charge12,near the extracellular end (Fig. 1a)13. In the homologous acetyl-choline receptor, at 9-Å resolution, the M2 segments bend at thecenter, which is the narrowest point of the channel in the closedstate. The segment ends angle away from the axis. At both ends,the center-to-center distance between adjacent M2 segments isabout 20 Å (ref. 14). The M2 segments change conformationduring gating11,15,16. Little is known about the thermal mobili-ty of the M2 segments in the closed and open states, or how thechannel lining residues of different subunits move relative to oneanother during gating transitions.

Disulfide trapping has been used to study protein mobilityand proximity relationships between residues in both water-sol-uble and integral-membrane proteins17–19. In proteins of knownstructure, the average α-carbon separation of disulfide-bondedcysteine (Cys) residues is about 5.6 Å (refs. 17, 20). The disulfidebond formation rate depends on the collision frequency of thesulfhydryls, the energy of the collision and the presence of anoxidizing environment17. Collision frequency depends on theaverage separation distance of the sulfhydryls, their relative ori-entation in the protein and the protein’s mobility and/or flexi-bility, especially near Cys residues.

We used disulfide bond trapping to probe the proximity andmobility of M2 segment channel-lining residues located on dif-ferent GABAA receptor subunits. We assayed the ability to formintersubunit disulfide bonds between engineered Cys residues ataligned M2 segment positions in the rat α1 and β1 subunits by

Protein mobility and GABA-inducedconformational changes in GABAAreceptor pore-lining M2 segment

Jeffrey Horenstein1,2, David A. Wagner3, Cynthia Czajkowski3 and Myles H. Akabas1

1 Department of Physiology and Biophysics and Department of Neuroscience, Albert Einstein College of Medicine, 1300 Morris Park Avenue, Bronx, New York 10461, USA

2 Department of Physiology, Columbia University, 630 W 168th Street, New York, New York 10032, USA 3 Department of Physiology, University of Wisconsin-Madison, 1300 University Avenue, Madison, Wisconsin 53706, USA

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

Protein movements underlying ligand-gated ion channel activation are poorly understood. Here weused disulfide bond trapping to examine the proximity and mobility of cysteines substituted foraligned GABAA receptor α1 and β1 M2 segment channel-lining residues in resting and activatedreceptors. With or without GABA, disulfide bonds formed at α1N275C/β1E270C (20´) andα1S272C/β1H267C (17´), near the extracellular end, suggesting that this end is more mobile and/orflexible than the rest of the segment. Near the middle of M2, at α1T261C/β1T256C (6´), a disulfidebond formed only in the presence of GABA and locked the channels open. Channel activation mustinvolve an asymmetric rotation of two adjacent subunits toward each other. This would movealigned engineered cysteines on different subunits into proximity and allow disulfide bondformation without blocking conduction. Asymmetric rotation of M2 segments is probably a commongating mechanism in other ligand-gated ion channels.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 1. Effects of oxidation and reduction on wild-type α1β1 GABAAreceptors (a) Aligned channel-lining residues in the α1 and β1 M2 mem-brane-spanning segments. Index numbers are shown in the center, tofacilitate comparisons with other ligand gated ion channels. Pairs ofaligned Cys-substitution mutants were expressed and tested for thepresence of spontaneous or inducible disulfide bonds. Shaded squares,channel-lining residues11; black squares, levels at which disulfide bondsformed; shaded circles, non-channel-lining positions. Residues morecytoplasmic than α1Val257 that were screened by the substituted-cys-teine-accessibility method but were not accessible to sulfhydryl reagentsapplied extracellularly are not shown11. Top, extracellular. (b) DTT (10 mM) applied without GABA has no lasting effect on cur-rents induced by 0.5 µM GABA test pulses in an oocyte expressing wild-type α1β1 receptors. Current trace not shown during DTT application.(c) A 3-min application of 100:400 µM Cu:phen has no effect on cur-rents induced by 50 µM GABA in an oocyte expressing wild-type recep-tors. Current trace not shown during Cu:phen application. (d) A 2-minapplication of 100:400 µM Cu:phen plus 20 µM GABA has no effect onwild type. GABA concentration, 20 µM. Bars above records indicateperiod of application of the reagent at left. Holding potential, –50 mV. InFigs. 1 to 4, the current traces are separated by a 3–5 min wash periodwith buffer. The currents during these washes are not shown.

examining the effects of oxidation and reduction on the func-tion of wild-type and mutant receptors expressed in Xenopusoocytes. The presence of spontaneously formed disulfide bondswas tested by examining the ability of the reducing agent dithio-threitol (DTT) to alter GABA-activated currents. The ability toinduce disulfide bond formation was assayed by studying theeffect of the oxidizing reagent copper-o-phenanthroline(Cu:phen) on channel function. Cu:phen promotes oxidationby catalyzing the formation of hydroxyl radicals and superox-ide from ambient molecular oxygen17. To identify gating-induced structural rearrangements within the channel, Cu:phenwas applied in the presence or absence of GABA. The presenceof disulfide-linked dimers was also assayed biochemically, toconfirm that the functional effects were due to intersubunitdisulfide bond formation. At two positions near the extracel-lular end of the M2 segment, 17´ and 20´, disulfide bondsformed with or without GABA. At the 6´ position, near the mid-dle of the channel, a disulfide bond formed only in the pres-ence of GABA.

RESULTSCharacterization of wild-type and mutant receptorsReduction of wild-type α1β1 GABAA receptors by a 3-minuteapplication of 10 mM DTT potentiated the GABA-inducedcurrent (IGABA) 3 minutes after DTT washout by 33 ± 3% (n = 5; Fig. 1b); however, this potentiation washed out within5 to 10 minutes. Others have interpreted similar effects of DTTas redox effects21,22. However, they may be pharmacologicaleffects of DTT similar to the potentiation induced by otheralcohols and general anesthetics1,3,6.

Oxidation by application of 100:400 µM Cu:phen increasedsubsequent IGABA by 12 ± 4% (n = 4) when applied for 3 min-utes without GABA (Fig. 1c) and by 5 ± 8% (n = 3) when appliedfor 2 minutes with saturating GABA (Fig. 1d). Effects of this mag-nitude (± 10–15%) reflected the variability of sequential GABAresponses over the approximately 30-minute duration of theseexperiments, and were not an effect of the applied reagents11.

All the Cys-substitution mutants tested expressed IGABA,except α1R19Ćβ1R19Ć and α1R0Ćβ1R0Ć. The GABA EC50was 3.4 µM for wild type, and, for the mutants, ranged between0.78 µM for α1S17Ćβ1 and 10.7 µM for α1β1E20Ć (Table 1).

GABA test pulses at an EC50 concentration were applied beforeand after DTT or Cu:phen applications. To maximally activatethe receptors during application with DTT or Cu:phen, a GABAconcentration at least five times EC50 was used.

Disulfide bond forms in α1 N275Cβ1E270, 20’ positionGABA (20 µM) elicited little current from oocytes expressingα1N20Ćβ1E20Ć (–110 ± 17 nA, n = 21). After a 3-minuteapplication of 10 mM DTT, peak IGABA increased by 1330 ± 286% (n = 21; Fig. 2a; Table 2). A subsequent 1-minuteapplication of 100:400 µM Cu:phen nearly completely reversedthe effect of DTT, reducing IGABA to its initial level. Applica-tion of 10 mM DTT restored the subsequent currents (Fig. 2a).We inferred that the receptor contained a spontaneouslyformed disulfide bond(s) that could be broken and re-formedby sequential reduction and oxidation. Spontaneous disulfidebond formation indicated that mildly oxidizing conditions dueto ambient oxygen in our buffer were sufficient to promotedisulfide bond formation. This is not fundamentally differentfrom disulfide bond formation in the more oxidizing environ-ment created by Cu:phen. It is a measure of the propensity toform disulfide bonds.

articles


a b

c d

Table 1. GABA EC50, Hill coefficient and maximal currentsfor the engineered Cys mutants.

Mutant EC50 (mM) nH Imax (nA) number*

α1β1 (WT) 3.4 ± 0.7 0.97 ± 0.12 –1378 ± 61 3

α1N20Ćβ1E20Ć nd –110 ± 17 21

α1N20Ćβ1 0.97 ± 0.22 0.82 ± 0.09 –713 ± 79 3

α1β1E20Ć 10.7 ± 1.9 1.17 ± 0.02 –677 ± 152 4

α1S17Ćβ1H17Ć 2.1 ± 0.9 1.39 ± 0.16 –768 ± 87 3

α1S17Ćβ1 0.78 ± 0.16 0.73 ± 0.07 –623 ± 75 3

α1β1H17Ć 4.5 ± 0.8 1.48 ± 0.03 –1358 ± 78 7

α1T6Ćβ1T6Ć 1.9 ± 0.8 1.51 ± 0.10 –1307 ± 103 3

α1T6Ćβ1 1.2 ± 0.5 1.47 ± 0.60 –930 ± 199 2

α1β1T6Ć 1.0 ± 0.1 1.21 ± 0.08 –1008 ± 133 3

*Number of oocytes; nd, not determined, current before DTT applicationwas too small

Fig. 2. Disulfide bonds form in α1N20Ćβ1E20Ćreceptors. (a) Currents from an α1N20Ćβ1E20Ć-expressing oocyte. Initial application of 20 µMGABA elicits very small currents. Application of 10 mM DTT (3 min) increases the current elicitedby subsequent GABA application. Application of100:400 µM Cu:phen causes subsequent GABA-induced currents to return to initial levels. TheCu:phen effect is reversed by subsequent applica-tion of 10 mM DTT. Bars above records indicateperiod of application of the reagent at left. Currentsduring application of DTT and Cu:phen are notshown. Holding potential, –50 mV. (b) Applicationof 0.3 mM NEM blocks spontaneous and Cu:phen-induced disulfide bond formation. Currents duringapplication of DTT, NEM and Cu:phen are notshown. (c) Determination of the rate of re-oxida-tion in the absence of GABA. Before the start of thetrace, 10 mM DTT was applied for 3 min to reducethe receptors. Brief (10 s) test pulses of 20 µMGABA were applied every 4 min. The peak currentswere fit with a single exponential decay function todetermine the time constant for re-oxidation. (d) Determination of re-oxidation rate in the pres-ence of GABA. Before the start of the trace, 10 mMDTT was applied to the oocyte for 3 min to reducethe receptors. GABA (20 µM) was applied, and thecurrent decay was recorded and fit with a singleexponential decay function to determine the timeconstant for re-oxidation in the presence of GABA.(e) β1-mycE20Ć forms DTT-reducible dimers in thepresence of both α1-FLAG and α1-FLAGN20Ć subunits on blot probed with anti-myc (anti-β) antibody. (f) α1-FLAGN20Ć forms DTT-reducible dimersstrongly in the presence of β1-mycE20Ć and weakly in the presence of wild-type β1 on blot probed with anti-FLAG (anti-α) antibody. (e, f) Western blotsof HEK 293 cells expressing combinations of FLAG and myc-tagged wild-type, α1-FLAGN20Ć and β1-mycE20Ć subunits. *Subunit containing an engineeredCys residue. Each combination was either left untreated or reduced with 50 mM DTT. All samples were alkylated with NEM and subjected to SDS-PAGE/western blot analysis. Molecular weight of monomers (indicated by arrow): α1-FLAG, 48, 50, 52 kD and β1-myc, 56, 50 kD. MW of crosslinked subunits,∼ 115–130 kD. MW of non-specific myc band in non-transfected cells (indicated by star), ∼ 62 kD.


IGABA spontaneously decayed back to original levels severalminutes after reduction with DTT. To test whether this decay wasdue to spontaneous re-formation of the disulfide, DTT-reducedreceptors were treated with 0.3 mM N-ethylmaleimide (NEM; 2 min). This should have irreversibly alkylated free sulfhydryls,thereby preventing disulfide formation. NEM application causedan immediate reduction in the subsequent GABA currents, butthe current level was stable, and subsequent application of100:400 µM Cu:phen had no effect (Fig. 2b).

To determine the state dependence of disulfide bond forma-tion, we measured the spontaneous reoxidation rates with orwithout GABA. To determine the closed state rate, we reducedα1N20Ćβ1E20Ć receptors with 10 mM DTT (3 min), and, for30 minutes after reduction, measured the peak current elicitedby brief (20 s) GABA test pulses applied every 3 to 5 minutes (Fig. 2c). The peak currents were fit with a single exponentialdecay function. In the closed state, τ for peak current decay was8.9 ± 1.2 minutes (n = 6). In the presence of GABA, channelsmake transitions between the open, closed and desensitized states;however, on the time scale of these experiments, the channelswere mostly desensitized. To determine the disulfide bond for-mation rate in the presence of GABA, receptors were reducedwith 10 mM DTT (3 min). GABA (20 µM) was applied for sev-eral minutes (Fig. 2d). The GABA current decay was fit with asingle exponential function (τ = 2.3 ± 0.2 min; n = 5). This rateoverestimates the actual disulfide bond formation rate becausedesensitization contributed an uncertain percentage of currentloss. Estimation of the desensitization rate could only be done

by co-applying GABA and DTT to prevent re-oxidation. How-ever, DTT potentiates GABA currents, possibly by a pharmaco-logical mechanism, and might alter desensitization rates1,3,6.Thus, we could not measure the rates with DTT and subtractthem from the rate of current decline without DTT.

Similar results were obtained in α1β1E20Ć receptors, inwhich the engineered Cys was only in β. DTT application (10 mM, 3 min) increased subsequent IGABA by 246 ± 49% (n = 8). This was smaller than the 1330% increase observed withα1N20Ćβ1E20Ć following DTT application. The difference inincrease may relate to a difference in the ability to hold the chan-nel closed with an α–β disulfide bond (or possibly two bonds)for α1N20Ćβ1E20Ć, as opposed to a single β–β bond forα1β1E20Ć. As with α1N20Ćβ1E20Ć, for α1β1E20Ć receptors,the DTT effect on subsequent GABA currents reversed sponta-neously in several minutes. Cu:phen application returned thecurrents to near their original levels (data not shown). BeforeDTT application, Cu:phen with or without GABA had little effecton currents: 3 ± 2% inhibition in the closed state (n = 4), and 2 ± 5% inhibition in the open/desensitized state (n = 4). Thus,as with the 20´ double Cys mutant, a disulfide bond was formedspontaneously and could be broken and re-formed by sequen-tial reduction and oxidation. The Cu:phen effects at this level orat the other reactive positions were not reversed by EGTA appli-cation (data not shown). This implies that Cu:phen effects werenot due to free Cu2+ binding between engineered Cys residues.There was no evidence of disulfide bond formation in α1N20Ćβ1receptors in which the engineered Cys was only in α. The effects

articles

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


of DTT and Cu:phen on α1N20Ćβ1 were similar to those seenwith wild type (Table 2).

Biochemical experiments confirmed that disulfide-linked sub-unit–subunit dimers formed under conditions similar to thoseused in the electrophysiological experiments. For these experi-ments, mutant and wild-type α1-FLAG subunits (tagged with theFLAG epitope) and β1-myc subunits (tagged with the myc 9E10epitope) were expressed in HEK 293 cells, and inter-subunitcrosslinking was identified by observing subunit dimers on west-ern blots. In unreduced wild-type receptors, no β1-myc dimersand only trace amounts of α1-FLAG dimers were detected (Fig. 2eand f). When β1-mycE20Ć was co-expressed with either α1-FLAGN20Ć or wild-type α1-FLAG, the β1-mycE20Ć subunit wasincorporated into a dimer band in the unreduced lane (Fig. 2e).When α1-FLAGN20Ć and β1-mycE20Ć were co-expressed, the α1-FLAGN20Ć subunit was strongly incorporated into a dimerband in the unreduced lane. In contrast, when α1-FLAGN20Ć wasco-expressed with wild-type β1-myc, only trace amounts of α1-FLAG dimer were detected, which was similar to wild type (Fig. 2f). In all cases, DTT reduction eliminated the dimer bands,consistent with a disulfide linkage. The dimer band migratedmore slowly than expected given the monomer mobility. Disul-

fide-linked acetylcholine receptor subunitdimers also migrate anomalously slowly onSDS-PAGE gels23. Thus, the biochemical andelectrophysiological results demonstrated disul-fide formation between 20´ engineered Cys.

α1S272Cβ1H267C, the 17’ positionThe 17´ position, α1Ser272 and β1His267, is thechannel-lining position one helical turn belowthe 20´ level11. In about half of theα1S17Ćβ1H17Ć-expressing oocytes, follow-ing transfer from the cysteine-containing cul-ture media to the cysteine-free electro-physiology buffer, the peak current induced byrepeated brief applications of GABA decayedover a 10-minute period. After the currentdecay stabilized, application of 10 mM DTTrestored the GABA-induced currents to theirinitial level. Thus, disulfide bonds formed spon-taneously at the 17´ position. In the other halfof the oocytes, the currents elicited by repeat-ed applications of GABA showed little or nodecrease. In this group, a 3-minute applicationof 10 mM DTT caused a 41 ± 12% increase incurrent (n = 7, Table 2), which was compara-ble to the DTT effect on wild type. We cannotexplain the difference between the two popula-tions of α1S272Cβ1H267C-expressing oocytes.In the oocytes, however, disulfide bonds wereprobably not present initially at the 17´ level.

Application of 100:400 µM Cu:phen toα1S17Ćβ1H17Ć-expressing oocytes in theabsence of GABA decreased the subsequentIGABA by 55 ± 2% (n = 11; Fig. 3, Table 2). Inthe presence of GABA, it inhibited IGABA by 64 ± 10% (n = 7). Additional applications ofCu:phen caused no further inhibition. A 3-minute application of 10 mM DTT reversedthis inhibition. Thus, under oxidizing condi-tions, a disulfide bond(s) formed inα1S17Ćβ1H17Ć receptors.

Similar effects occurred in α1β1H17Ć receptors. DTT (10 mM) potentiated IGABA by 12 ± 17% (n = 7), comparable towild type. Application of 100:400 µM Cu:phen for up to 5 min-utes in the closed state inhibited subsequent GABA currents by32 ± 7% (n = 19; Fig. 3a). Application of 100:400 µM Cu:phenplus GABA inhibited subsequent GABA currents by 26 ± 11%(n = 3). DTT reversed this inhibition.

There was no evidence of disulfide bond formation inα1S17Ćβ1 receptors in which the engineered Cys was only in α.The effects of DTT and Cu:phen on the IGABA were comparable tothose seen with wild type (Table 2).

Western blots of α1-FLAGS17Ćβ1-mycH17Ć receptors showedspontaneously formed disulfide-linked dimers recognized by bothanti-FLAG and anti-myc antibodies, indicating the presence of bothα and β subunits in the dimers (Fig. 3b and c). DTT reductioneliminated the dimers. Following oxidation with Cu:phen, therewas no apparent increase in dimer formation (Fig. 3b and c). In HEK cells, unlike in oocytes, these disulfide bondsformed spontaneously because the HEK culture media containedcystine, which creates an oxidizing extracellular environment.

Western blots of α1β1H17Ć were similar to α1S17Ćβ1H17Ć;however, the DTT-sensitive dimers were only recognized by the

articles


Table 2. Effect of reduction and oxidation on all of Cys mutants tested.

DTT Cu:phen Cu:phen+GABA

Mutant (% effect) s.e.m n (% effect) s.e.m.n (% effect) s.e.m. n

Wild-type 33 3 5 12 4 4 5 8 3αA-1ĆβA-1Ć 7 3 3 2 8 3 –16 13 4αR0ĆβR0Ć neαV2ĆβA2Ć 93 29 3 8 4 9 7 6 9αT6ĆβT6Ć 13 23 2 6 6 5 # 13 4αT7ĆβT7Ć 29 1 3 –2 2 3 –3 7 5αL9ĆβL9Ć 77$ 36 6 –7$ 6 5 ndαT10ĆβT10Ć 53 29 2 –9 4 4 –10 4 4αT13ĆβT13Ć 38 19 2 8 6 5 –10 7 7

αI16ĆβT16Ć 56 21 2 –8 5 8 –4 5 15αS17ĆβH17Ć & & &αA18ĆβL18Ć 75 17 4 4 12 9 –21 1 4αR19ĆβR19Ć neαN20ĆβE20Ć 1330 286 21 –81@ 3 6 –85@ 2 5

αN20Ćβ 16 4 4 9 7 4 –10 1 3αS17Ćβ 10 11 11 –5 5 3 –13 6 7αT6Ćβ 71 54 5 nd% –14 10 2αβE20Ć 246 49 8 –3 2 4 –2 5 4αβH17Ć 12 17 7 –32 7 19 –26 11 3αβT6Ć –9 11 3 –5 4 4 # 6 5

ne, no expression; nd, not determined. #In the presence of GABA, Cu:phen locked these channelsopen. See text for details. $Effects on leak current only. &Oocytes expressing the 17´ mutant fellinto two populations. In one population, there was spontaneous ‘rundown’ of GABA-induced cur-rents when the cells were transferred from OR3 culture media to the electrophysiology perfusionbuffer, which lacked cysteine. DTT reversed the ‘rundown’ in these cells, potentiating GABA-induced currents by 173 ± 41% (n = 6). Cu:phen, in turn, reversed the effects of DTT. In the otherpopulation of oocytes, there was little or no ‘rundown.’ In this population, DTT potentiated GABA-induced currents by 41 ± 12% (n = 7), which is comparable to the effect of DTT on wild type. In thissecond population of oocytes, Cu:phen inhibited currents by 55 ± 2% (n = 11) when applied in theresting state and by 64 ± 10% (n = 7) when applied in the presence of GABA. @The percent poten-tiation of the initial currents by DTT is larger than the percent inhibition by Cu:phen becauseCu:phen application does not inhibit the current to the level of the initial currents before DTT (Fig. 2a). %Because there was no effect of Cu:phen on the 6´ double mutant in the absence ofGABA, the effect of Cu:phen on α1T6Ćβ1 was not tested.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


nel. In some cells, the persistent current gradually returned tobaseline over a period of more than 20 minutes. On this timescale, current relaxation was most likely due to receptor turnoverrather than channel closure.

Cu:phen application to α1T6Ćβ1T6Ć or α1β1T6Ć recep-tors produced no changes in either the leak current in theabsence of GABA or in the subsequent GABA-induced cur-rents (Fig. 4b). Thus, disulfide bond formation between theengineered Cys in the β1T6Ć subunits was state dependent.Only in the open state were the Cys in sufficiently close prox-imity or sufficiently mobile to allow for disulfide bond for-mation.

Biochemical experiments were consistent with the electro-physiological results. β1T6Ć subunit dimers were onlyobserved following the co-application of Cu:phen and GABA(Fig. 4c). As with the electrophysiology experiments, onceformed, these dimers were resistant to DTT reduction. Therewere no spontaneous β1T6Ć dimers and none following appli-cation of Cu:phen in the absence of GABA (Fig. 4c). In Fig. 4d, there is a faint dimer of α1T6Ć observed followingthe co-application of Cu:phen and GABA, which is resistantto DTT reduction. The α1 dimer was seen in only two of fourexperiments and was always much fainter than the β1 dimer,which was robustly detected in all four experiments. A smallpercentage of receptors with a three α , two β stoichiometrymight account for the presence of this faint α dimer band.

No disulfides between other aligned M2 CysWith the Cys mutants at the other channel-lining positions(including α1A254Cβ1A249C, –1´; α1V257Cβ1A252C, 2´;α1T262Cβ1T257C, 7´; α1L264Cβ1L259C, 9´; α1T265Cβ1T260C,10´; α1T268Cβ1T263C, 13´; and α1I271Cβ1T266C, 16´; Fig. 1a),application of DTT or Cu:phen in the presence or absence ofGABA had no irreversible effects, similar to responses with wildtype (Table 2). For the double Cys mutants at the 2´, 9´ and 18´positions, the effect of DTT application was greater than the effecton wild type (Table 2); however, as with wild type (Fig. 1b), thiseffect was transient. Furthermore, application of Cu:phen imme-diately after DTT application did not reverse the transient DTT-induced potentiation any faster than washout with buffer. Weinfer that no disulfide bonds formed between these engineeredCys, and that the DTT-induced potentiation was probably a phar-macological effect. The α1L9Ćβ1L9Ć receptors were constitu-itively open, which caused a large baseline leak current withminimal response to GABA. Similar effects of 9´ mutations have

anti-myc antibody, indicating the exclusive formation of β–βdimers. A trace amount of DTT-sensitive dimer was present onthe western blots of α1S17Ćβ1, indicating that trace amounts ofα–α dimers may have been formed. In conclusion, at the 17´ level,the substituted cysteines of at least one α subunit and one β sub-unit, and at least one pair of β subunits, come into sufficientlyclose proximity to form disulfide bonds.

Cu:phen+ GABA locks α1T261Cβ1T256C open, 6’ levelApplication of GABA to α1T6Ćβ1T6Ć- or α1β1T6Ć-express-ing oocytes induced normal-looking currents that returned tobaseline following GABA washout (Fig. 4a, left two traces). Incontrast, following a 2-minute application of Cu:phen plus sat-urating GABA, the current remained elevated and stable forover 10 minutes (Fig. 4a, middle trace and the starting currentlevel of the final two traces). The persistent elevated currentmagnitude was similar to the current magnitude induced bysaturating GABA concentrations before application of GABAplus Cu:phen (data not shown). (In Fig. 4a, the initial two andfinal two GABA-induced test currents were elicited by EC50GABA concentrations, whereas saturating GABA was coappliedwith the Cu:phen.) We infer that disulfide bond formationbetween 6´ engineered β Cys residues locked the channels in anopen, conducting state. For α1T6Ćβ1 receptors, which onlycontain a Cys in α, application of Cu:phen or DTT in the pres-ence of GABA had no effect on IGABA (data not shown).

To probe the structure of the locked-open state, we testedthe effect of two GABAA receptor inhibitors that act by bindingin the channel, picrotoxin and penicillin. The persistent ele-vated current following washout of Cu:phen plus GABA wasinsensitive to either 100 µM picrotoxin or 10 mM penicillin(data not shown). Disulfide bonds, however, could interferewith the binding of both drugs. The picrotoxin binding site isat a more cytoplasmic position in the channel, α 1Val2´ (ref. 24) and mutations at the 6´ position altered picrotoxinbinding25. Based on the voltage dependence of block, penicillinalso binds deep within the channel26.

The disulfide bond formed in the locked-open channels wasresistant to reduction. When applied at a concentration of 10 mM, DTT, Tris(carboxyethyl)phosphine (TCEP) and 2-mercaptoethanol had little or no effect on the persistent ele-vated current following application of Cu:phen plus GABA (datanot shown). It is unclear whether the disulfide was buriedbetween two subunits and inaccessible to these reagents, orwhether these reagents were unable to reach this level in the chan-


articles

Fig. 3. Disulfide bonds form in α1S17Ćβ1H17Ć and in α1β1H17Ćreceptors. (a) GABA-induced currents from α1β1H17Ć-expressingoocyte. Initial application of 2 µM GABA elicits robust currents. Followingapplication of 100:400 µM Cu:phen (1 min), the GABA-induced currentmagnitude is ∼ 50% smaller. This is reversed by 10 mM DTT application. (b, c) Western blots of HEK cells expressing combinations of FLAG- andmyc-tagged wild type, α1-FLAGS17Ć and β1-mycH17Ć subunits. *Subunitcontaining an engineered Cys residue. Each subunit combination was leftuntreated, reduced with 50 mM DTT or oxidized with 200:400 µMCu:phen. All were treated with NEM before SDS-PAGE/western blotanalysis. (b) β1-mycH17Ć forms DTT-reducible dimers in the presence ofboth wild-type α1-FLAG and mutant α1-FLAGS17Ć subunits on blot probedwith anti-myc (anti-β) antibody. Dimer formation is not increased by oxi-dation. (c) α1-FLAGS17Ć forms DTT-reducible dimers strongly in the pres-ence of β1-mycH17Ć and weakly in the presence of wild-type β1-myc. Dimerformation is increased by oxidation on blot probed with anti-FLAG (anti-α)antibody. Arrows, GABA subunit monomers. Star, non-specific band innon-transfected cells recognized by the anti-myc antibody.

a

b c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


been reported9. The effect of DTT or Cu:phen application on the9´ leak current was comparable to the effects on the GABA-induced currents in wild type.

Disulfide bond formation at the non-channel-lining positionsbetween the 17´ and 20´ levels was also tested. The mutantα1R19Ćβ1R19Ć did not express GABA-induced currents, andapplication of DTT or Cu:phen did not facilitate GABA-inducedcurrents. The mutant α1A18Ćβ1L18Ć was functional, but appli-cation of DTT or Cu:phen in the presence or absence of GABAhad no irreversible effects (Table 2). For two of the mutants,α1A18Ćβ1L18Ć and α1I16Ć β1T16Ć, during the Cu:phen plusGABA application, currents were significantly lower than withGABA alone. This inhibition, however, washed out with removalof the Cu:phen and was most likely due to a transient interactionof free Cu2+ with the engineered Cys residues in the vicinity ofthe Zn2+ binding site residue, β1His17´ (refs. 27, 28).

No disulfides between engineered and endogenous CysTo rule out the possibility that a disulfide bond formed betweenan engineered Cys and an endogenous Cys residue, the 6´, 17ánd 20´ experiments were repeated in a Cys-minus background.The Cys-minus background was formed by mutating the endoge-nous Cys in membrane-spanning segments, α1C234S in M1 andα1C293S, β1C288A in M3. The Cys-minus α1β1 receptors had aGABA EC50 of 2.7 µM (n = 3), similar to wild type. Using theelectrophysiological assay, DTT and Cu:phen had similar effectson α1N20Ćβ1E20Ć, α1S17Ćβ1H17Ć and α1T6Ćβ1T6Ć inthe Cys-minus background (data not shown) and in the wild-type background. Therefore, the observed effects were not dueto disulfide bond formation between the endogenous M1 andM3 Cys and an engineered Cys.

DISCUSSIONThe ability to form a disulfide bond between engineered Cysresidues depends on structural factors including the average dis-tance separating the Cys residues and the relative side chain ori-entations, and on dynamic factors such as the thermal mobility

and flexibility of the protein regions containing the residues17.Using electrophysiological and biochemical assays, we showedthat disulfide bonds formed between aligned, engineered, chan-nel-lining Cys residues on different subunits at three levels in theM2 segment, two near the extracellular end (the 17´ and 20´ posi-tions), and one below the middle of the segment (6´ position;Fig. 1a). Disulfide bond formation at the 6´ position required anoxidizing environment and the presence of GABA. Because the6´ disulfide bond locked the channels into an open, conductingstate, we infer that the disulfide bond formed in the open stateand not in a non-conducting, desensitized state. The 6´ disulfidebond probably linked adjacent subunits, because the channellumen would be occluded by a disulfide bond between non-adja-cent subunits. We knew that the 6´ disulfide bond formed main-ly between β subunits because the electrophysiological resultswere identical whether the engineered Cys was in both the α andβ subunits or only in β, and there was no effect if the Cys wasonly in α. Adjacent β subunits imply a subunit stoichiometry oftwo α and three β (Fig. 5, left), a stoichiometry consistent withprevious results7,8. Furthermore, β subunits, unlike α subunits,form homomeric channels and thus have interfaces that can co-assemble28,29. As the two GABA binding sites are located at β–αsubunit interfaces30,33, two β and two α subunits are involved informing the GABA binding sites. The third β subunit (β*, Fig. 5) in αβ receptors, which is presumably replaced by the sin-gle γ subunit in αβγ receptors7,9,10, is thus the only subunit notinvolved in forming the agonist binding site. Thus, we infer thatat the 6´ level, the disulfide bond forms between the third β sub-unit and an adjacent β subunit, which is part of an α–β pair thatforms a GABA binding site (Fig. 5).

Based on the fact that the 6´ disulfide bond forms betweenadjacent β subunits in the open but not in the closed state, these6´ engineered Cys residues move into closer proximity and/or aremore mobile/flexible in the open rather than the closed state.What conformational change would bring engineered Cys intocloser proximity? If in the closed state the subunits are roughlyfive-fold symmetric relative to the channel axis, as is suggested

articles


Fig. 4. In α1T6Ćβ1T6Ć, co-application of Cu:phen andGABA induces formation of a disulfide bond that locks thechannels in an open state. (a) Co-application of Cu:phenand GABA induces persistently elevated currents in anα1T6Ćβ1T6Ć-expressing oocyte. Two responses to 2 µMGABA (EC50) elicit reversible currents (left two traces).Center trace, current elicited by 10 µM GABA (near-satu-rating GABA) pre-applied for 20 s to open the channelsand followed immediately by co-application of 100:400 µMCu:phen and 10 µM GABA. Following washout of Cu:phenand GABA, the current remained elevated for longer than10 min. Initial levels of the two subsequent EC50 GABAtest pulses (right two traces) start at a level similar to theend of the center trace. Traces are separated by 5-minwashes. (b) Application of Cu:phen in the absence ofGABA does not induce persistently activated currents. (c, d) Western blots of HEK cells expressing combinationsof FLAG and myc-tagged wild type, α1-FLAGT6Ć and β1-

mycT6Ć subunits. Each combination was either leftuntreated, or treated with 50:100 µM Cu:phen, 100 µMGABA, or 50:100 µM Cu:phen and 100 µM GABA, as indi-cated below the lanes. Some samples were subsequentlytreated with DTT, as indicated below the lanes. All sampleswere alkylated with NEM, purified by immunoprecipitationwith anti-FLAG agarose and subjected to SDS-PAGE/western blot analysis. (c) Blot probed with anti-myc (anti-β) antibody shows that β1-mycT6Ćforms DTT-reducible dimers only when simultaneously treated with both GABA and Cu:phen. (d) Blot probed with anti-FLAG (anti-α) shows thatα1-FLAGT6Ć forms dimers very weakly and only when simultaneously treated with both GABA and Cu:phen.

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



by the electron density maps of the homologous acetylcholinereceptor14, the engineered cysteines would be at roughly fivefoldsymmetrical positions around the channel axis (Fig. 5, left). Tobring the Cys in adjacent subunits into closer proximity, the M2segments must rotate asymmetrically relative to each other. Eitherone M2 segment remains stationary and the other turns towardit, or both rotate toward each other (Fig. 5, right). If both M2segments rotate toward each other, they may rotate by differentamounts. We cannot distinguish between these possibilities. TheM2 segment channel-lining face forms an arc of about 120°(ref. 11); therefore, the sum of the rotational movement of thetwo subunits must be approximately 120°. We propose that GABAbinding at the two α–β interfaces induces a rotation at the 6´ levelof the M2 segments of these subunits relative to the M2 segmentof the fifth subunit (Fig. 5). This asymmetry of motion may inpart explain why mutations at aligned M2 positions in differentsubunits have different effects.

In the muscle acetylcholine receptor, the agonist binding sitesare located at the αγ and αδ interfaces2. This implies that theacetylcholine receptor β subunit has a function similar to thethird β subunit in αβ GABAA receptors. The hypothesis that gat-ing from the closed to the open state involves an asymmetric rota-tion at the 6´ level of two adjacent subunits differs from apreviously proposed mechanism16, involving a symmetrical rota-tion of all five M2 segments, based on a comparison of fivefoldrotationally averaged cryo-electron microscopic images of theacetylcholine receptor in the closed and open states16. In theseimages, however, an asymmetric movement of one subunit rela-tive to the other four subunits would be lost.

Other groups have inferred that the closed channel gate islocated at the 9´ position14, between the 9´ and 13´ positions15,and at positions more cytoplasmic than the 2´ position34. Weobserved no functional evidence for disulfide bond formation atany of these positions in the closed state. However, expression ofα1L9Ćβ1L9Ć resulted in a large leak conductance, and muta-tions at this level increased the channel’s spontaneous open prob-ability9. Application of 100 µM bicuculline did not close thechannel (data not shown); thus, for α1L9Ćβ1L9Ć, during theCu:phen application in the absence of GABA, the channels wereundergoing transitions between the closed and open states. Nev-ertheless, no evidence of disulfide bond formation was observed.Thus, the residues proposed to form the closed channel gate donot seem to form disulfide bonds. Perhaps the residues formingthe gate are rigidly held in position, and are thus prevented fromcolliding with sufficient frequency or proximity necessary fordisulfide bond formation. Alternatively, the gate may not beformed by the close apposition of aligned residues pointingtoward each other near the central channel axis.

At the 17´ and 20´ positions, disulfide bonds formed bothin the presence and in the absence of GABA. In the presenceof GABA, the rate of disulfide bond formation was, at most,

fourfold faster. The faster rate may represent moremobility/flexibility or closer average separation distance in thepresence of GABA. During prolonged application of GABA,however, the open probability is low, and the channels spendmost of the time in desensitized states. We cannot distinguishwhether disulfide bond formation at the 17´ and 20´ positionsin the presence of GABA occurs in the open or desensitizedstates, or both. In contrast, at the 6´ position, the rate in theclosed state was immeasurably slow, so we could assess the dif-ference in closed and open state rates.

At both the 17´ and 20´ positions, disulfide bond formationinhibited channel opening; however, at the 17´ position,GABA-induced currents were inhibited only by about 50%. Thisimplies that the receptor can open, albeit less efficiently, with adisulfide bond linking at least two M2 segments. This suggestseither that the closed to open transition at the 17´ level does notrequire a rotational or translational movement of the two linkedsubunits relative to each other, or that the other subunits cancompensate. Alternatively, the remaining current might comefrom a subset of receptors unable to form disulfide bonds; how-ever, there is no obvious reason for this. As with the 6´ position,the disulfide bond must form between adjacent subunits becausea bond between non-adjacent subunits would occlude the chan-nel, a conclusion consistent with the western blot observations.

The 20´ position aligns with the acetylcholine receptor extra-cellular ring of charge12. In αβ GABAA receptors, the β subunit17´ histidines form a bidentate, high-affinity Zn2+ binding sitepresumably between the two adjacent β subunits8,27. The maxi-mum separation between the α carbons of two histidines form-ing such a site is 13 Å (ref. 35). This provides a distance constrainton the 17´ α carbons on adjacent subunits consistent with thelow-resolution acetylcholine receptor structure14.

In proteins of known crystal structure, the average separationbetween disulfide-bonded Cys α carbons is ∼ 5.6 Å (refs. 17, 20).Thus, to form a disulfide bond, Cys residues must approach tothis distance; however, their average separation may be greater,and the closer apposition may represent a transient conforma-tion. In the aspartate chemotaxis receptor, engineered Cys wereused to deduce a relationship between average separation dis-tance in the crystal structure, disulfide bond formation rates andsulfhydryl collision frequency17. For the closest pair of sulfhydrylsrequiring a 4.6 Å translocation to form a disulfide bond, the t1/2was 3.3 s, with an inferred collision frequency of ∼ 104 s–1; for themost distant pair studied, requiring a 15.2 Å translocation, the

articles

Fig. 5. Asymmetric rotation of adjacent β subunits upon channel opening.Left, top view cross section of the channel in the closed state. Circles, fiveM2 segments; SH, position of the 6´ engineered Cys residues; gray trian-gles, closed channel gate. (This should not be taken to imply that the gateis at the 6´ level.) White circles, M2 segments from the subunits involvedin forming the two GABA binding sites. Shaded circle, third β subunit thatis not involved in forming the GABA binding sites. Right, open state fol-lowing treatment with Cu:phen in the presence of GABA. The four sub-units forming the GABA binding sites have rotated in a counterclockwisedirection, whereas the shaded subunit has rotated in a clockwise direc-tion. Arrows indicate direction of rotation relative to the initial startingposition. (The handedness of rotation, clockwise versus counterclock-wise, was chosen arbitrarily, our data do not provide information on thedirection of rotation.) Such a movement, at least of the two adjacent βsubunits, would be necessary to bring sulfhydryls at aligned positions onadjacent subunits into close proximity to allow disulfide bond formation.As discussed in the text, the two adjacent β subunits may not rotate to asimilar extent. Note the disulfide bond indicated between the sulfhydrylson the adjacent β subunits after Cu:phen.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


t1/2 was 3630 s, and the inferred collision frequency was ∼ 10 s–1

(ref. 17). In our experiments, disulfide bond formation in thepresence of Cu:phen was complete within 120 s. This places alower limit on the reaction rate. Extrapolating from the previousresults implies a minimum collision rate of ∼ 103 s–1. This is like-ly to be a lower limit on the collision rate because the previousrates were measured in a more oxidizing environment, that is, inthe presence of 10-fold higher Cu:phen.

At the 17´ and 20´ positions, disulfide bonds formed in theabsence of Cu:phen, that is, under milder oxidizing conditions.This implies a greater collision frequency of the engineered Cysresidues, presumably due to greater thermal protein motion atthe extracellular end of the M2 segment than in the rest of M2,which may be more rigid.

The inability to form disulfide bonds between aligned engi-neered Cys at many channel-lining positions is probably due tolower mobility/flexibility or relative Cys orientation in theseregions; the average separation distance of the Cys residues atthese positions is predicted to be smaller than at the 17´ and 20´positions in the 9 Å acetylcholine channel structure14. Thus, wepropose that the extracellular end of M2 between the 17´ and 20´positions is more mobile and/or flexible than the more cyto-plasmic M2 regions. Based on effects of amide-to-ester backbonemutations with unnatural amino acids on the acetylcholine recep-tor, it was suggested that the M2 segment extracellular end under-goes greater conformational changes during gating, whereas themore cytoplasmic M2 region undergoes a rigid body motion thatdoes not alter the pattern of backbone hydrogen bonds36. Thisis consistent with our finding that at the 6´ position a disulfidebond only formed in the presence of GABA, presumably due toan asymmetric rotation of neighboring subunits. Given the like-ly structural similarity of ligand-gated ion channel superfamilymembers, our findings may provide insights into the M2 seg-ment mobility and the conformational changes occurring dur-ing gating in other members of the superfamily.

METHODSMutagenesis and oocyte expression. Mutagenesis of rat α1 and β1 sub-units, in vitro transcription of mRNA, and preparation and injection ofXenopus oocytes were done as described previously8. The oocytes werekept in OR3 media, which contains free cysteine. This cysteine mayhave prevented disulfide bond formation until it was oxidized or untilthe oocytes were transferred to the cysteine-free electrophysiologyrecording buffer.

Electrophysiology. Currents were recorded under two-electrode voltageclamp from oocytes continuously perfused at 5 ml/min with buffer (115 mM NaCl, 2.5 mM KCl, 1.8 mM MgCl2, 10 mM HEPES pH 7.5 withNaOH) using equipment and procedures described previously8. Perfusionchamber volume was about 200 µl. Currents are reported as mean ± s.e.m.Cu:phen solutions were prepared by diluting stock solutions of CuSO4 and1,10-phenanthroline (Sigma, St. Louis, Missouri) in buffer. Concentrationsof Cu:phen solutions are given as the [Cu2+]:[phenanthroline] in µM. WhenGABA was coapplied with DTT or Cu:phen, a near-saturating GABA con-centration was used to ensure that all of the channels were in the activatedstate during the period of reagent application. The percent effect was ((IGABA after/IGABA before) – 1) × 100. IGABA after is the peak current of theGABA test pulses after the application of reagent, and IGABA before is the peakcurrent of the initial GABA applications. Data are given as mean ± s.e.m.

GABA dose–response relationships were determined and fitted withthe Hill equation as previously described8. Generally, an EC50 GABA con-centration was used for test pulses before and after applications of DTT orCu:phen, although, occasionally, near-saturating GABA concentrationswere used with similar results (see Table 1 for mutant EC50 values). EC50GABA test pulses are more sensitive than near-saturating GABA test puls-es to effects of oxidation and reduction on channel gating kinetics37.

Biochemical detection of cross-linked subunits. Cell culture and tran-sient transfection of HEK-293 cells were done as described previously33.HEK cells were maintained in minimal essential media with Earle’s salts,which contains cystine.

The α1-FLAG subunit contained the FLAG epitope sequence (DYKD-DDDK) inserted between the sixth and seventh amino acid of the maturesubunit, and the β1-myc subunit had the myc 9E10 epitope (EQK-LISEEDL) inserted between the fifth and sixth amino acid of the maturesubunit. These epitopes do not affect function29.

Detection of crosslinked subunits with mutations at the 20´ and 17´positions used the following protocol. Adherent cells in dishes, exceptthose to be treated with Cu:phen, were incubated with 10 mM NEM inPBS (137 mM NaCl, 2.7 mM KCl, 1.3 mM MgCl2, 1.5 mM KH2PO4, 14 mM K2HPO4, pH 7.1) for 10 min at room temperature (RT). Disheswere washed twice with PBS. Appropriate dishes were incubated with200:400 µM Cu:phen for 10 min at RT. Dishes to be treated with DTTwere washed twice with PBS and then treated with 50 mM DTT for 10 min at RT. All dishes were washed twice with PBS + 10 mM NEM.Cells were solubilized in 1 ml of ice cold lysis buffer (1% Triton X-100,Sigma; 50 mM Tris-HCl pH 7.5; 150 mM NaCl; 5 mM EDTA; 0.5 mgml–1 Pefabloc, Boehringer, Indianapolis; 1 µg/ml leupeptin, Sigma; 1µg/ml pepstatin, Sigma), and incubated 15 min on ice. Lysates were trans-ferred to microfuge tubes, incubated on a rotating wheel for 2 h at 4°C,and spun at 25,000 × g for 15 min, at 4°C. Supernatant was transferred tofresh tubes and, with 25 mM DTT added to appropriate samples, wasincubated for 15 min at RT. All samples were heated to 100°C for 3 min,and were spun at 16,000 × g for 5 min at 4°C. The supernatant was storedfor western blot analysis.

To detect GABA-dependent crosslinking of subunits with 6´ engineeredCys residues, contaminating proteins that interact with the myc antibodywere removed using the following affinity purification step. Cells werewashed twice with PBS, incubated in either PBS, 50:100 µM Cu:phen inPBS, 100 µM GABA in PBS, or 50:100 µM Cu:phen plus 100 µM GABAin PBS, 15 min, RT. Cells were washed once with PBS, incubated in PBSand 4 mM NEM for 15 min at RT and solubilized in 1 ml ice cold lysisbuffer for 15 min at 4°C. Lysates were transferred to microfuge tubes,incubated on a rotating wheel for 2 h at 4°C and spun at 25,000 g for 15 min at 4°C. To pre-clear samples, 200 µl of equilibrated Sepharose(Sigma, see below) was added and samples were incubated on a rotatingwheel for 1 h at 4°C. Beads were pelleted at 16,000 × g for 15 min at 4°C.Samples were immunoprecipitated by transferring supernatant to cleantubes, adding 50 µl equilibrated anti-FLAG-agarose beads (Sigma) and100 µl equilibrated Sepharose, and incubating on a rotating wheel for 2h at 4°C. Beads were pelleted at 16,000 × g for 15 min at 4°C, washed 4times with 1 ml wash buffer (0.1% Triton-X-100, 150 mM NaCl, 5 mMEDTA, and 50 mM Tris-Cl, pH 7.5) and once with 1 ml 25 mM Tris-Cl.The protein was eluted by addition of 100 µl of 400 µg/ml FLAG pep-tide in wash buffer and was incubated on a rotating wheel for 1 h at 4°C.Beads were pelleted, and the eluate was stored for western blot analysis(see below). FLAG-agarose and Sepharose beads were equilibrated bywashing 5 times in 1 ml wash buffer.

Immunoblotting. Laemmli sample buffer (LSB, 5×, 3% SDS, 0.6 Msucrose, 0.325 M Tris-HCl pH 6.8, and, in some instances, 10 mM DTT)was added to the protein samples that were subjected to SDS-polyacry-lamide gel electrophoresis (10% acrylamide) using a Mini-Protean IIapparatus (Bio-Rad, Hercules, California) at 200 V for ∼ 1 h. Sampleswere transferred to nitrocellulose membrane (0.45 µm) using a MiniTrans-Blot apparatus (Bio-Rad) at 100 V for 1 h. The nitrocellulose mem-brane was washed 3 times with 20 mM Tris-HCl, 500 mM NaCl, pH 7.5(TBS), blocked with 0.5% non-fat powdered milk in TBS for 1 h at RT,and washed 3 times with 0.5% Tween-20 in TBS (TTBS). Blots were incu-bated in primary antibody diluted in TTBS + 0.5% powdered milk (5 µg/ml M2 anti-FLAG, Kodak, Rochester, New York, or anti-myc 9E101:10000 dilution, gift from J. Hell) for 2 h, and washed 4 times in TTBS.Blots were incubated in secondary antibody (peroxidase-conjugated goatanti-mouse IgG, Pierce, Rockford, Illinois) for 45 min, washed 6 timeswith TTBS and developed with Super Signal ECL substrate (Pierce).Sometimes different amounts of dimer relative to monomer wereobserved (for example, compare the 20´ position to the 6´ position). It

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



is difficult to interpret these differences quantitatively because the blotsshow total receptor subunits (unassembled, partially assembled and fullyassembled). The monomer staining observed can be from a variety ofsubunit pools, whereas the cross-linked dimers are most likely frommature assembled receptors. Because the ratio of the total number ofassembled receptors to the total pool of receptor subunits can varybetween transfections and between the different Cys mutants, these exper-iments should only be interpreted qualitatively.

ACKNOWLEDGEMENTSThis work was supported in part by grants from the National Institutes of Health

NS30808 (M.H.A.), GM61925 (M.H.A.) and NS34727 (C.C.), and a

Burroughs Wellcome Fund New Investigator Award (C.C.). We thank A.

Finkelstein for comments on this manuscript.


1. Macdonald, R. L. & Olsen, R. W. GABAA receptor channels. Annu. Rev.Neurosci. 17, 569–602 (1994).

2. Karlin, A. & Akabas, M. H. Toward a structural basis for the function ofnicotinic acetylcholine receptors and their cousins. Neuron 15, 1231–1244(1995).

3. 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).

4. Gage, P. W. Signal transmission in ligand-gated receptors. Immunol. Cell Biol.76, 436–440 (1998).

5. Sieghart, W. et al. Structure and subunit composition of GABA(A) receptors.Neurochem. Int. 34, 379–385 (1999).

6. Krasowski, M. D. & Harrison, N. L. General anaesthetic actions on ligand-gated ion channels. Cell Mol. Life Sci. 55, 1278–1303 (1999).

7. Tretter, V., Ehya, N., Fuchs, K. & Sieghart, W. Stoichiometry and assembly of arecombinant GABAA receptor subtype. J. Neurosci. 17, 2728–2737 (1997).

8. Horenstein, J. & Akabas, M. H. Location of a high affinity Zn2+ binding site inthe channel of α1β1 gamma-aminobutyric acid A receptors. Mol. Pharmacol.53, 870–877 (1998).

9. Chang, Y., Wang, R., Barot, S. & Weiss, D. S. Stoichiometry of a recombinantGABAA receptor. J. Neurosci. 16, 5415–5424 (1996).

10. Farrar, S. J., Whiting, P. J., Bonnert, T. P. & McKernan, R. M. Stoichiometry ofa ligand-gated ion channel determined by fluorescence energy transfer.J. Biol. Chem. 274, 10100–10104 (1999).

11. Xu, M. & Akabas, M. H. Identification of channel-lining residues in the M2membrane-spanning segment of the GABAA receptor alpha1 subunit. J. Gen.Physiol. 107, 195–205 (1996).

12. Imoto, K. et al. Rings of negatively charged amino acids determine theacetylcholine receptor channel conductance. Nature 335, 645–648 (1988).

13. Charnet, P. et al. An open-channel blocker interacts with adjacent turns ofalpha-helices in the nicotinic acetylcholine receptor. Neuron 2, 87–95 (1990).

14. Unwin, N. Nicotinic acetylcholine receptor at 9 Å resolution. J. Mol. Biol. 229,1101–1124 (1993).

15. White, B. H. & Cohen, J. B. Agonist-induced changes in the structure of theacetylcholine receptor M2 regions revealed by photoincorporation of anuncharged nicotinic noncompetitive antagonist. J. Biol. Chem. 267,15770–15783 (1992).

16. Unwin, N. Acetylcholine receptor channel imaged in the open state. Nature373, 37–43 (1995).

17. Careaga, C. L. & Falke, J. J. Thermal motions of surface alpha-helices in the

D-galactose chemosensory receptor. Detection by disulfide trapping. J. Mol.Biol. 226, 1219–1235 (1992).

18. Yu, H., Kono, M., McKee, T. D. & Oprian, D. D. A general method formapping tertiary contacts between amino acid residues in membrane-embedded proteins. Biochemistry 34, 14963–14969 (1995).

19. Wu, J. & Kaback, H. R. A general method for determining helix packing inmembrane proteins in situ: helices I and II are close to helix VII in the lactosepermease of Escherichia coli. Proc. Natl. Acad. Sci. USA 93, 14498–14502(1996).

20. Krovetz, H. S., VanDongen, H. M. & VanDongen, A. M. Atomic distanceestimates from disulfides and high-affinity metal-binding sites in a K+

channel pore. Biophys. J. 72, 117–126 (1997).21. Amato, A., Connolly, C. N., Moss, S. J. & Smart, T. G. Modulation of

neuronal and recombinant GABAA receptors by redox reagents. J. Physiol.(Lond.) 517, 35–50 (1999).

22. Pan, Z., Zhang, X. & Lipton, S. A. Redox modulation of recombinant humanGABA(A) receptors. Neuroscience 98, 333–338 (2000).

23. Hamilton, S. L., McLaughlin, M. & Karlin, A. Formation of disulfide-linkedoligomers of acetylcholine receptor in membrane from Torpedo electrictissue. Biochemistry 18, 155–163 (1979).

24. Xu, M., Covey, D. F. & Akabas, M. H. Interaction of picrotoxin with GABAAreceptor channel-lining residues probed in cysteine mutants. Biophys. J. 69,1858–1867 (1995).

25. Gurley, D., Amin, J., Ross, P. C., Weiss, D. S. & White, G. Point mutations inthe M2 region of the alpha, beta, or gamma subunit of the GABAA channelthat abolish block by picrotoxin. Receptors Channels 3, 13–20 (1995).

26. Twyman, R. E., Green, R. M. & MacDonald, R. L. Kinetics of open channelblock by penicillin of single GABAA receptor channels from mouse spinalcord neurones in culture. J. Physiol. (Lond.) 445, 97–127 (1992).

27. Wooltorton, J. R., McDonald, B. J., Moss, S. J. & Smart, T. G. Identification ofa Zn2+ binding site on the murine GABAA receptor complex: dependence onthe second transmembrane domain of beta subunits. J. Physiol. (Lond.) 505,633–640 (1997).

28. Sigel, E., Baur, R., Trube, G., Mohler, H. & Malherbe, P. The effect of subunitcomposition of rat brain GABAA receptors on channel function. Neuron 5,703–711 (1990).

29. Connolly, C. N., Krishek, B. J., McDonald, B. J., Smart, T. G. & Moss, S. J.Assembly and cell surface expression of heteromeric and homomericgamma- aminobutyric acid type A receptors. J. Biol. Chem. 271, 89–96(1996).

30. Sigel, E., Baur, R., Kellenberger, S. & Malherbe, P. Point mutations affectingantagonist affinity and agonist dependent gating of GABAA receptorchannels. EMBO J. 11, 2017–2023 (1992).

31. Amin, J. & Weiss, D. S. GABAA receptor needs two homologous domains ofthe beta-subunit for activation by GABA but not by pentobarbital. Nature366, 565–569 (1993).

32. Smith, G. B. & Olsen, R. W. Identification of a [3H]muscimol photoaffinitysubstrate in the bovine gamma-aminobutyric acidA receptor alpha subunit.J. Biol. Chem. 269, 20380–20387 (1994).

33. Boileau, A. J., Evers, A. R., Davis, A. F. & Czajkowski, C. Mapping the agonistbinding site of the GABAA receptor: evidence for a beta-strand. J. Neurosci.19, 4847–4854 (1999).

34. Wilson, G. G. & Karlin, A. The location of the gate in the acetylcholinereceptor channel. Neuron 20, 1269–1281 (1998).

35. Higaki, J. H., Fletterick, R. J. & Craik, C. S. Engineered metalloregulation inenzymes. Trends Biochem. Sci. 17, 100–104 (1992).

36. England, P. M., Zhang, Y., Dougherty, D. A. & Lester, H. A. Backbonemutations in transmembrane domains of a ligand-gated ion channel:implications for the mechanism of gating. Cell 96, 89–98 (1999).

37. Zhang, H. & Karlin, A. Contribution of the beta subunit M2 segment to theion-conducting pathway of the acetylcholine receptor. Biochemistry 37,7952–7964 (1998).

articles©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


articles


Potassium leak currents have been recognized as essential to neu-romuscular function for over fifty years1–5. Also called resting(or background) conductances, their existence as unique molec-ular transport entities was questioned even as they were attrib-uted key roles in sympathetic ganglia, invertebrate axons,vertebrate myelinated axons and cardiac myocytes6–11. Operat-ing under control of agents as disparate as serotonin, cyclicnucleotides, molecular oxygen and γ-aminobutyric acid, potas-sium leak establishes and maintains a hyperpolarized restingmembrane potential that suppresses excitatory responses, where-as inhibition of the leak produces a permissive state11–14. Nativeleaks have defied coherent description, perhaps because they seemto be invariant, and are easily camouflaged by time-dependentprocesses. However, cloning of TOK1 from Saccharomyces cere-visiae15, KCNK0 from Drosophila melanogaster16 and, to date,twelve mammalian KCNK genes for 2 P domain potassium-selec-tive leak channels, has affirmed that these currents are carried bydedicated pathways amenable to detailed study17.

The human KCNK2 gene encodes a 426-residue KCNK2 chan-nel subunit with two pore-forming P domains and four trans-membrane segments (a 2P/4TM topology), and gains highexpression in central nervous system tissues, especially the hip-pocampus18. Referred to as TPKC1 (ref. 18) or TREK1 (ref. 19),and now KCNK2 for clarity, the channel has been implicated inneural responses to temperature, arachadonic acid, mechanicalstretch and volatile anesthetics19–21. Here we sought to under-stand the discrepant attributes accorded KCNK2 channels18–23.We report that the single canonical site in KCNK2 for proteinkinase A (PKA) phosphorylation, previously shown to mediatetemperature sensitivity20, controls not only the amount of chan-nel activity, but its fundamental response to voltage.

RESULTSNative KCNK2 channels: phenotypic interconversionTo allow direct comparison of native and cloned channels, ratKCNK2 was cloned and found to share 95% identity with the

human gene at the amino acid level, including a PKA consensussite with serine at position 348 (see Methods). Rat hippocampalcells carrying a temperature-sensitive simian large tumor anti-gen gene24 expressed KCNK2 messenger RNA only when drivento differentiate (Fig. 1a). Thus, reverse transcription generatedno signal from undifferentiated cells but a band of the expectedsize for KCNK2 with RNA isolated from cells induced to matureto primary neurons. Accordingly, KCNK2-like channels were notobserved in small on-cell patches with undifferentiated cells (n = 46), but were apparent with mature cells in 5 of 20 patcheshaving only one to three channels. (Seventeen other patches hadtoo many channels to evaluate, and 111 were quiet; to reduceactivity of other native channels, long pre-pulse depolarizationswere used to silence channels subject to inactivation, and pipetteand bath solutions contained 1 mM magnesium and no sodiumor calcium.) When native KCNK2-like channels were excisedfrom cells in inside-out mode and treated with the catalytic sub-unit of PKA (20 U/ml) and ATP (1 mM), they showed a decreasein activity at –60 mV that was reversed by exposure to alkalinephosphatase (AP, 20 U/ml; Fig. 1b).

Evaluation of single native channels at a variety of potentialsrevealed that enzyme-induced changes in activity at –60 mV result-ed from a unique regulatory alteration in channel phenotype. AfterAP treatment, the channels opened frequently, in a voltage-insen-sitive fashion: at 100 mV, open probability (Po100) was 0.80 ± 0.08(n = 3), which was steadily maintained from –100 to 100 mV(Po–100/Po100 = 0.93 ± 0.10, n = 3; Fig. 1c, left). Conversely, treat-ment with PKA and ATP yielded channels with decreased activity(Po100 = 0.11 ± 0.09, n = 3) that were ∼ 100-fold more active atdepolarized potentials (Po–100/Po100 = 0.01 ± 0.00, n = 3; Fig. 1c,right). These voltage-dependent changes in activity produced sig-nificant outward currents and limited inward currents despitesymmetric potassium levels across the membrane. Three singlenative channels were thus reversibly interconverted by AP or PKAand ATP between two phenotypes: a voltage-insensitive open poreand a voltage-dependent outward rectifier.

KCNK2: reversible conversion of ahippocampal potassium leak into avoltage-dependent channel

Detlef Bockenhauer, Noam Zilberberg and S. A. N. Goldstein

Departments of Pediatrics and Cellular and Molecular Physiology, Boyer Center for Molecular Medicine, Yale University School of Medicine, 295 Congress Avenue, New Haven, Connecticut 06536, USA

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

Potassium leak channels are essential to neurophysiological function. Leaks suppress excitabilitythrough maintenance of resting membrane potential below the threshold for action potential firing.Conversely, voltage-dependent potassium channels permit excitation because they do not interferewith rise to threshold, and they actively promote recovery and rapid re-firing. Previously attributed todistinct transport pathways, we demonstrate here that phosphorylation of single, native hippocampaland cloned KCNK2 potassium channels produces reversible interconversion between leak and voltage-dependent phenotypes. The findings reveal a pathway for dynamic regulation of excitability.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

Native and cloned KCNK2: biophysical equivalenceSeveral lines of evidence argued that the native channel understudy was KCNK2. First, single native and cloned KCNK2 chan-nels had the same primary slope conductance (85 ± 2 and 84 ± 1 pS, respectively, in symmetrical 100 mM KCl from –20and 100 mV; Fig. 2a). Second, isotonic replacement of rubidiumfor potassium as charge carrier yielded a new primary slope con-ductance in both native and cloned KCNK2 channels (64 ± 1.4and 64 ± 3.3 pS, respectively) and similarly increased Po (Fig. 2b). Third, magnesium at physiological concentration (1 mM) similarly inhibited single native and cloned KCNK2channels to produce two open states at hyperpolarized poten-tials, consistent with unblocked (O) and blocked current levels(O*; Fig. 2c); thus, the amplitude of O* was –3.4 ± 0.9 and –3.3 ± 0.9 pA at –60 mV and 2 kHz for native and cloned chan-nels, respectively (Figs. 1b, c and 2c), and half-maximal at –81 ± 4 mV (data not shown). Fourth, the effects of PKA andATP or AP on native and cloned KCNK2 channel gating para-meters were similar (Fig. 2d). Thus, dwell time in the principleopen state (O*) at –60 mV was the same for the two channelsand unchanged by application of PKA and ATP or AP (1.90 ±0.05 and 2.00 ± 0.05 ms for native and cloned channels with AP,respectively, n = 3). Furthermore, both channels demonstratedtwo closed states at –60 mV. The briefer closure was unaffectedby phosphorylation (0.50 ± 0.10 and 0.50 ± 0.05 ms for nativeand cloned channels with AP, respectively, n = 3) and was appar-ent in single channel records as ‘flickery’ transitions within openbursts. The longer closure had a mean dwell time of ∼ 5 ms in thepresence of AP (5.0 ± 0.4 and 6.1 ± 1.0 ms for native and clonedchannels, respectively, n = 3) and increased in frequency ∼ 10-fold (from ∼ 0.02 to 0.20 of closures) and duration ∼ 200-fold(1200 ± 900 and 1700 ± 2000 ms for native and cloned channels,n = 3, respectively) after PKA and ATP treatment. This was rem-iniscent of kinase regulation of KCNK0, in which the frequencyand duration of visits to a long closed state were altered; although,in that case, phosphorylation increased activity in a voltage-inde-pendent fashion25.

A direct effect via KCNK2 serine 348To investigate the mechanistic basis for these phenotypic inter-conversions, cloned human KCNK2 channels were studied fur-ther. Single wild-type channels showed expected changes in

channel activity with enzyme exposure at –60 mV (Fig. 3a). Thus,PKA and ATP decreased KCNK2 activity (Po100 = 0.18 ± 0.15, n = 2) and yielded channels ∼ 100-fold more active at depolar-ized voltages (Po–100/Po100 = 0.03 ± 0.04, n = 2), whereas APincreased channel activity (Po = 0.68 ± 0.03, n = 2) in a uniformfashion across a wide voltage range (Po–100/Po100 = 0.90 ± 0.10, n = 2). To determine if the enzymes were acting directly onKCNK2, serine 348 (in the PKA consensus site) was first changedto alanine (S348A) to mimic the non-phosphorylated state. Sin-gle S348A channels (Fig. 3b, left) were similar to AP-treated nativeor wild-type cloned channels, demonstrating a high, uniform Poacross the voltage range (Po100 = 0.53 ± 0.08, Po–100/Po100 = 1.0 ± 0.2, n = 6). Next, serine 348 was mutated to aspar-tate (S348D) to coarsely resemble KCNK2 in the phosphorylatedstate. Single S348D channels (Fig. 3b, right) recapitulated the effectof PKA and ATP on native and cloned wild-type channels, yield-ing voltage-dependent outward rectification (Po100 = 0.21 ± 0.05,Po–100/Po100 = 0.02 ± 0.02, n = 6). The function of S348A orS348D channels was not altered by application of either PKA andATP or AP, consistent with regulation of wild-type channels byphosphorylation of serine 348 (n = 2; Fig. 3c).

The idea that changes in KCNK2 phenotype were mediateddirectly by phosphorylation of serine 348 (rather than by involve-ment of other proteins26) was supported by study of channelsaltered to cysteine (S348C) and exposed to 2-sulfonatoethylmethanethiosulfonate (MTSES), a reagent that covalently modi-fies cysteine to leave a negatively charged sulfonatoethyl moiety.In unmodified form, S348C channels showed a voltage-insensi-tive Po like S348A channels or wild-type channels treated with AP(data not shown). Whereas wild-type channels were insensitive(data not shown), S348C channels treated with 10 mM internalMTSES decreased in activity at –60 mV (Fig. 3d), similarly to wild-type channels treated with PKA and ATP (Fig. 3a); the change wasstable, despite reagent wash-out, until chemical reduction (Fig. 3d), consistent with covalent modification at position 348.

DISCUSSIONPotassium channels control neuronal excitability through influenceover the duration, frequency and amplitude of action potentials.Potassium channels that are active at rest inhibit depolarizationtoward firing threshold, and thus suppress excitation. Conversely,potassium channels activated at depolarized potentials do not inter-

Fig. 1. Native KCNK2 leak pores are reversibly transformedinto voltage-dependent channels by PKA-mediated phosphory-lation. Recordings were taken in symmetric 100 mM KCl and 1 mM MgCl2 solution. C, closed state (solid line); O* and O,open states (dashed lines). In the presence of magnesium, chan-nels show two open states at negative potentials. (a) Reversetranscription with KCNK2-specific primers demonstrates thepresence of KCNK2 messenger RNA in a rat hippocampal cellline in the mature (lane 2) but not the undifferentiated state(lane 3). Markers (lane 1), 0.5, 1, 1.5 and 2 kb; arrow, size ofexpected product (1427 bp). (b) A single native channel ininside-out patch mode shows decreased activity when treatedwith PKA and ATP and increased activity with AP at –60 mV.Trace duration, 7 min. Sampled at 1 kHz, filtered at 200 Hz.Open levels O* and O, –3.0 and –6.7 pA, respectively. (c) A sin-gle native channel at various voltages (as indicated) shows a highuniform Po in the presence of AP (left panel) and voltage-dependent outward rectification after treatment with PKA andATP (right panel). Scale bars, 5 pA and 100 ms; sampled at 20 kHz, filtered at 2 kHz. At 100 mV open level 8.0 pA and –100 mV, O* and O are –3.2 and –8.1 pA, respectively.

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


fere with rise to threshold, but do facilitate recovery and repetitivefiring. Here we identified a pathway for regulation of neuronalexcitability. Native hippocampal and cloned KCNK2 channels trans-form in a regulated and reversible fashion from potassium-selec-tive leak conductances (open across the physiological voltage range)to strictly voltage-dependent channels.

Comparison to voltage effects on other channelsInwardly rectifying (Kir) channels contribute to potassium leak,stabilizing cells near the equilibrium reversal potential of potas-sium (EK)27–30. Formed of subunits with one P domain and twotransmembrane segments (1P/2TM), Kir channels pass smalloutward currents because of pore blockade by internal magne-sium and polyamines; at potentials negative to EK, large inwardcurrents are passed upon relief from blockade. An analogousprocess, that is, pore occlusion by an external cation, cannotexplain outward currents in KCNK2 channels, first, because theactivation midpoint (V1/2) for S348D channels was insensitiveto changes in EK. Thus, in symmetric 100-mM potassium, V1/2was 10 ± 7 mV (predicted EK ∼ 0 mV), whereas after isotonicreplacement of 95 mM bath potassium for sodium, V1/2 was 8 ± 10 mV (predicted EK ∼ –85 mV; Fig. 4a). Moreover, whereasexternal magnesium did produce inhibition at negative voltages(Fig. 2c), both S348A and S348D channels were influenced equal-ly (Fig. 4b). Finally, the V1/2 of S348D channels was also unal-tered in the absence of added divalent cations (with EGTA), orwhen ultra-pure potassium was used to reduce heavy metal con-tamination (data not shown).

TOK1 has a predicted 2P/8TM topology, and thus shares withKCNK2 a two-P-domain subunit structure; however, TOK1 func-tions distinctly15. Reminiscent of a Kir channel (but with recip-rocal rectification properties), this target for viral K1 Killer toxin31

is a non-voltage-gated outward rectifier that passes significantoutward potassium currents at potentials positive to EK by amechanism that seems to involve both conformational changes ofthe protein and ion occupancy of the pore15,32–35.

Classical voltage-gated potassium channel (Kv) subunits, inan EK-insensitive fashion, are quiet at rest and show increasedPo with membrane depolarization; this is similar to KCNK2 whenit functions in a voltage-dependent manner (Fig. 4a). Kv channelsare formed by subunits with a 1P/6TM topology, and carry a spe-cial fourth span (S4) with positively charged residues at everythird or fourth position that acts as a voltage sensor and moves inresponse to changes in membrane potential. KCNK2 has nocharges in its predicted transmembrane segments to suggest ituses a similar type of voltage sensor. The voltage dependence ofS348D KCNK2 activation (estimated roughly from a fit to thePo–voltage relationship, Fig. 4a) suggests a net movement of ∼ 2.5elemental charges across the field, compared to ∼ 4–6 elementalcharges (determined by the same method) for the classical volt-age-gated potassium channel Shaker.

Phenotypic conversion of KCNK2 recalls the highly regu-lated transition of voltage-gated calcium channels between‘willing’ and ‘reluctant’ states, which differ significantly in theirvoltage dependence for activation. (These channels are formedby large proteins resembling four 1P/6TM units linked in tan-dem, and bear four S4 voltage sensors38,39.) In this analogy,dephosphorylated KCNK2 is still voltage dependent, but hasa hyperpolarized V1/2 producing a sustained high Po in thephysiological potential range.

Other pathways also alter KCNK2 phenotypeWhereas phosphorylation of KCNK2 on serine 348 produces avoltage-dependent phenotype, it seems unlikely that the phosphate

a

b

c

d

Fig. 2. Native and cloned KCNK2 channels show the same biophysicalproperties. Patches were excised in inside-out mode from differenti-ated hippocampal cells or oocytes expressing cloned KCNK2, andwere studied with symmetric 100 mM KCl unless otherwise indicated.Traces of 1 s are shown; sampled at 20 kHz, filtered at 2 kHz. Channelstate designations as in Fig. 1. (a) Native and cloned channels have thesame major unitary conductance and appearance. Shown at 100 mV;principle conductance values are in text. (b) Native and cloned chan-nels have the same current level and appearance with rubidium ascharge carrier. In this case, bath solution was replaced isotonically by100 mM RbCl. Shown at 100 mV; dashed line open level, 5.4 pA. (c) With 1 mM MgCl2 added, native and cloned channels show twoopen state levels at negative voltages. Shown at –60 mV, lines for O*and O are indicated at –3.0 and –6.7 pA, respectively. (d) Dwell-timehistograms of native and cloned KCNK2 channels at –60 mV reveal oneopen and two closed states. One open state is apparent in the presenceof AP (dashed line) or PKA and ATP (solid line) with mean duration 2.3 ± 0.1 ms (native) and 2.1 ± 0.2 ms (cloned) with PKA and ATP (n = 3); values for AP are in the text. Histograms suggest two closedstates for both native and cloned channels in the presence of AP(dashed line) or PKA and ATP (solid line); the briefer closure was unaf-fected by phosphorylation, and was 0.7 ± 0.3 ms (native) and 0.6 ± 0.3 ms(cloned) with PKA and ATP (n = 3); values for AP are in the text. Thelonger closure had a mean dwell time of ∼ 5 ms with AP and increasedin frequency and duration with PKA and ATP treatment (values in text).Closed and open-time duration were determined using half-amplitudethreshold detection46. Dwell-time distributions were plotted on a log-arithmic time axis with a square-root vertical axis to best discern eventpopulations; histograms were fitted with TacFit software (Bruxton,Seattle, Washington) using a sums of exponential probability densityfunction and maximum likelihood method.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

is a structural component of the channel gate (although it couldbe a source of gating charge). Two other manipulations that placeda negative charge on the residue yielded similar changes, that is,mutation to aspartate (Fig. 3b) and mutation to cysteine followedby MTSES modification (Fig. 3d). Moreover, KCNK2 channelsremain subject to phenotypic regulation after ablation of the PKAconsensus site at serine 348. In two of nine excised single-channelpatches, S348A KCNK2 revealed an intermediate phenotype atbaseline (Po100 = 0.4 ± 0.2, Po–100/Po100 = 0.3 ± 0.1); in these cases,internal application of 1 mM ATP without PKA increased Po andabolished voltage dependence (Po = 0.7 ± 0.2, Po–100/Po100 = 0.9 ± 0.1) in an AP-reversible fashion. This suggests thatan endogenous kinase can act on the mutant channel (or an acces-sory protein) to alter KCNK2 phenotype. Furthermore, 10 µMarachidonic acid, a regulator of whole-cell KCNK2 currents21,modified whole-cell S348D currents to resemble those of S348A:magnitude was increased and rectification was altered from out-ward toward open (data not shown). These findings (albeit withmutant channels) suggest other pathways may regulate KCNK2 ina manner reciprocal to that of PKA. Thus, the phenotypic changein KCNK2 described here follows channel phosphorylation, butits mechanistic basis remains to be discerned.

A broader role for an emerging superfamilyFifty years after the description of potassium-selective leak chan-nels, their genes have been identified17. KCNK genes are numer-ous, have wide tissue distribution, and encode 2P/4TM subunitsthat form KCNK channels. Studies of cloned single KCNK0 chan-nels have confirmed that leak accrues from channels open at rest40

and have demonstrated that leak channels are not always open,but rather, are subject to strict regulation like their native coun-terparts25. Thus, the excitatory influence of neurotransmitter-medi-ated inhibition of KCNK3 in cortical and thalamic neurons seems

key to ‘state-switching’ and, hence, transitions between sleep andwakefulness41–43. In contrast, the volatile anesthetic halothane acti-vates KCNK3 in hypoglossal and locus coeruleus neurons hyper-polarizing the cells and decreasing spike activity and this is thoughtto mediate immobilization, as well as analgesic and hypnoticactions of the drug44. Thus, regulation of KCNK channels has beenassociated with more or less leak; this report demostrates that therepertoire of these channels is more diverse.

KCNK channels that show function (KCNK0, 2, 3, 4, 5, 6, 9, 10,13 and TWK18) all demonstrate attributes expected for potassiumleak conductances: currents are present at rest and seem to riseinstantly with voltage steps. Activity is voltage independent andpermeation behavior follows expectations for open(Goldman–Hodgkin–Katz) rectification: both inward and outwardsingle channel currents are large with symmetric potassium acrossthe membrane. So KCNK0 (refs. 16, 40) and KCNK2 in the absenceof magnesium (Fig. 4) show linear unitary current–voltage rela-tionships, whereas KCNK2 in the presence of magnesium passessomewhat larger outward currents (Fig. 4b), and KCNK9 slightlylarger inward currents45. Nonetheless, because current flows moreeasily from a side of high permeant ion concentration, and potas-sium inside an animal cell is higher than outside, all KCNK channelspass larger outward unitary currents under physiological condi-tions. These open channel attributes are not altered when KCNK2channels transform from voltage independent to voltage gated.

Discordant descriptions of whole-cell KCNK2 currents can nowbe understood. Macroscopic currents like those we observed whenserine 348 was altered to alanine (Fig. 4c) have been reported undersome baseline conditions22,23 and after exposure to chloroform19

or lysophospholipids21; these demonstrate operation of KCNK2 asan open rectifier that is weakly blocked by external magnesium in avoltage-dependent fashion. Other treatments19–21 yielded currentslike those we observed after mutation of serine 348 to aspartate,and KCNK2 was voltage gated (Fig. 4c). These disparate pheno-types can be difficult to distinguish under physiological conditionsbecause both produce large, positive outward currents (Fig. 4c).However, at voltages negative to 0 mV, the two behaviors are read-ily distinguished and are expected to have very different effects onexcitability. Operating as an open rectifier, KCNK2 passes currentsboth above and below EK (∼ –85 mV), stabilizing the cell at rest (Fig. 4d). When voltage gated, KCNK2 only passes significant cur-

Fig. 3. KCNK2 channels have a voltage-dependent phenotype whenresidue 348 is negatively charged via phosphorylation, mutation or mod-ification. Cloned KCNK2 channels were studied in inside-out mode with100 mM symmetric KCl solution without MgCl2. Channel state designa-tions as in Fig. 1. Open levels have a step size of 6 pA at –60 mV; O* wasnot observed in the absence of magnesium. On, open level for that num-ber of channels. (a) Wild-type cloned KCNK2 channels show decreasedactivity when treated with PKA and ATP (solid bar) and increased activ-ity when treated with with AP (dashed bar). Shown at –60 mV; traceduration, 7 min; sampled at 1 kHz, filtered at 200 Hz. (b) Mutation ofKCNK2 serine 348 to alanine (S348A) or aspartate (S348D) recapitu-lates the effects of enzyme treatments. A single S348A KCNK2 channel(left) shows a uniform Po from –100 to 100 mV, whereas a single S348DKCNK2 channel (right) shows voltage dependence. Trace duration, 2 s;sampled at 20 kHz, filtered at 2 kHz. (c) Single KCNK2 channelsmutated to alanine or aspartate at residue 348 are insensitive to PKAand ATP (solid bar) or AP (dashed bar); shown at –60 mV. Trace dura-tion, 5 min; sampled at 1 kHz, filtered at 200 Hz. (d) Application of 10 mM MTSES (solid bar) to four S348C KCNK2 channels held at –60 mV reduces channel activity. Whereas washout did not reverse theeffect, application of 5 mM dithiothreitol (DTT, dashed bar) restoredfunction. Trace duration, 5 min; sampled at 1 kHz, filtered at 200 Hz.

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


KCNK2 cRNA. Whole-cell currents were measured 1–5 days after cRNAinjection by two-electrode voltage clamp using an oocyte clamp (Warn-er, Hamden, Connecticut) and pCLAMP8 software (Axon Instruments,Foster City, California) with bath solution containing 100 mM KCl, 1 mM MgCl2, 0.3 mM CaCl2, 5 mM HEPES, pH 7.5. Data were sampledat 5 kHz and filtered at 1 kHz. Electrodes were fabricated from borosil-icate tubes (Garner Glass, Claremont, California), contained 3 M KCland had resistances of 0.2–1 MΩ. Recordings were made under constantperfusion at room temperature.

For patch clamp studies, data were sampled with an Axon 200B ampli-fier using pCLAMP8 software (Axon Instruments, Foster City, California)and stored on videotape. Data were digitized at 10 or 20 kHz and filteredat 2 or 4 kHz with a Gaussian software filter. Data were analyzed withTAC software (Bruxton, Seattle, Washington). Amplitude levels wereobtained by all-points histograms. Bath and pipette solution contained100 mM KCl, 1 mM EGTA, 5 mM HEPES, pH 7.5 unless otherwise stat-ed. Bath and pipette solutions for recordings with rat hippocampal cellsincluded 1 mM MgCl2. Neither AP nor PKA and ATP revealed silentchannels in patches without channel activity from either native or exper-imental cells (n = 3). To identify native KCNK2 channels, recordingswere taken in 5-s steps and 20-mV increments from 100 to –100 mV,with an interpulse interval of 6 s and a holding voltage of 0 mV. Ultra-pure KCl was purchased from Alfa Aesar, Ward Hill, Massachusetts. ATP(sodium salt) was used at 1 mM and dithiothreitol (DTT), at 5 mM; bothwere purchased from Sigma (St. Louis, Missouri). PKA catalytic subunitwas purchased from Calbiochem (La Jolla, California). Alkaline phos-phatase was purchased from New England Biolabs (Beverley, Massa-chusetts). Stock enzyme solutions were stored at –20°C (AP) or –80°C(PKA) at 500× and diluted to a final level of 20 U/ml with bath solutionbefore use. We stored 2-sulfonatoethyl methanethiosulfonate (MTSES;Toronto Research Chemicals, Downsview, Ontario) as a solid at 4°C anddissolved it in bath solution before use.

ACKNOWLEDGEMENTSThis work was supported by grants to S.A.N.G. from the National Institutes of

Health. D.B. is supported by the Child Health Research Center (Yale University)

rents above threshold (Fig. 4d), because changes in open probabil-ity are half maximal above 0 mV (Fig. 4a), rise to threshold is unim-peded, and recovery and repetitive firing are facilitated.

METHODSMolecular biology. To clone rat KCNK2, primers were designed byhomology to human KCNK2 (AF004711) and a rat EST and used with arat brain cDNA library (Clontech, Palo Alto, California) and RACEmethodology; the primers for 3´ extension were 5´-GGCGGCCCCT-GACTTGCTGGATCC-3´ and 5´-GCCTTGCTGGGAATTCCCCTCTTTGGTTTTCTACTGG-3´; reverse complement primers were used for the5´ reactions. Two clones were sequenced on both strands and deposited(AF325671). Reverse transcription and amplification were done with 5´-GGCGGCCCCTGACTTGCTGGATCC-3´ and 5´-GACAGCTCAGGAGCCTCCTCATGAGTGCGG-3´ using an RT-PCR ONE STEP kit (Qia-gen, Valencia, California) after RNA was isolated from single 10-cm con-fluent culture dishes using an RNeasy Mini kit (Qiagen). Human KCNK2was cloned18, mutated and transcribed with T7 RNA polymerase(Ambion, Austin, Texas), as described25.

The line, provided by S. Rivkees (Yale University) and previouslydescribed24, was produced from embryonic rat hippocampal cells byretroviral transduction with temperature-sensitive simian virus 40large tumor antigen. Cells were passed at 33°C in DMEM with 10%FBS, 50 mg/l penicillin/streptomycin and 200 mg/l of geneticine intissue culture dishes coated with poly-lysine. Differentiation wasinduced by increasing culture temperature to 38°C, reducing FBS to1%, and adding 0.1 mg/ml sodium pyruvate, 2 mM glutamine, 10 ng/ml human basic fibroblast growth factor, and N2 supplement(GIBCO BRL, Rockwell, Maryland; provides 5 mg/l insulin, 100 mg/l transferrin, 20 nM progesterone, 0.1 mM putrescine and 30 nM sodium selenite). These conditions were applied for 2 to 4 daysbefore reverse transcription or biophysical study.

Electrophysiology. Oocytes were isolated from Xenopus laevis frogs(Nasco, Atkinson, Wisconsin), defolliculated by collagenase treatmentand injected the same day with 46 nl of cRNA containing 0.25 to 2 ng of

Fig. 4. Voltage-dependent activation of KCNK2 isnot sensitive to potassium reversal potential or dueto magnesium blockade. KCNK2 channels studied ininside-out patches or by two-electrode voltageclamp. (a) Normalized open probability–voltage rela-tionships for cloned KCNK2 channels. S348A chan-nels (solid squares) have a uniform Po, whereasS348D channels (circles) are voltage dependent (n =6–7). V1/2 for S348D channels was determined by fit-ting to a Boltzmann function, A1–A2/(1 + exp[V1/2 –V]/VS)–1 + A2, and was almost the same in symmetric100 mM KCl (solid circles) and asymmetric condi-tions (open circles, 5 mM KCl in bath, 100 mM KClin pipette, n = 6). This plot excludes the two S348Asingle channel patches that were of mixed phenotypebefore application of ATP. No solutions includedmagnesium. (b) Magnesium similarly inhibits inwardcurrent through S348A (squares) and S348D (cir-cles) single channels. The principle single channelcurrent level was measured in the absence of magne-sium (0 Mg, solid symbols) and with 1 mM MgCl2(1 Mg, gray symbols; the latter state is O*. The volt-age dependence of block was modeled by a simplifi-cation of the approach of Woodhull, as describedpreviously18. The apparent electrical distance traversed by magnesium, δ, was –0.31 ± 0.04 and –0.29 ± 0.03 for S348A and S348D channels, respec-tively. Half-block was seen with 1 mM MgCl2 at –87 ± 4 and –85 ± 4 mV for S348A and S348D channels, respectively. (c) KCNK2 whole-cell currentsin physiological and high external potassium. Sample current families from oocytes expressing S348A or S348D channels bathed with 5 or 100 mMpotassium solution. Scale bars, 2 µA and 100 ms; protocol indicates 200-ms pulses from –60 mV to voltages of –120 to 60 mV in steps of 10 mV. (d) Current–voltage relationships for groups of 9 cells expressing S348A channels (open squares) or S348D channels (open circles) in 5 mM potas-sium solution. Inset, current–voltage relationships for cells in 5 mM potassium, and for the same cells expressing S348A channels (filled squares) orS348D channels (filled circles) in 100 mM potassium solution. Error bars, s.e.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

and an award from the National Institute of Diabetes and Digestive and Kidney

Diseases. We thank F. Sesti and R. Goldstein for discussions and advice during

the course of this work.

RECEIVED 16 FEBRUARY; ACCEPTED 21 FEBRUARY 2001

1. Goldman, D. E. Potential, impedance, and rectification in membranes.J. Gen. Physiol. 27, 37–60 (1943).

2. Hodgkin, A. L. & Katz, B. The effect of sodium ions on the electrical activityof the giant axon of the squid. J. Physiol. (Lond.) 108, 37–77 (1949).

3. Hodgkin, A. L. & Huxely, A. F. A quantitative description of membranecurrent and its application to conduction and excitation in nerve. J. Physiol.(Lond.) 117, 500–544 (1952).

4. Hille, B. Ionic selectivity of Na and K channels of nerve membranes.Membranes 3, 255–323 (1975).

5. Adams, D. J., Smith, S. J. & Thompson, S. H. Ionic currents in molluscansoma. Annu. Rev. Neurosci. 3, 141–167 (1980).

6. Schmidt, H. & Stampfli, R. The effect of tetraethylammonium chloride onsingle Ranvier’s nodes. Pflugers Arch. Gesamte Physiol. Menschen Tiere 287,311–325 (1966).

7. Hille, B. Potassium channels in myelinated nerve. Selective permeability tosmall cations. J. Gen. Physiol. 61, 669–686 (1973).

8. Baker, M., Bostock, H., Grafe, P. & Martius, P. Function and distribution ofthree types of rectifying channel in rat spinal root myelinated axons.J. Physiol. (Lond.) 383, 45–67 (1987).

9. Chang, D. C. Is the K+ permeability of the resting membrane controlled bythe excitable K+ channel? Biophys. J. 50, 1095–1100 (1986).

10. Jones, S. W. On the resting potential of isolated frog sympathetic neurons.Neuron 3, 153–161 (1989).

11. Backx, P. H. & Marban, E. Background potassium current active during theplateau of the action potential in guinea pig ventricular myocytes. Circ. Res.72, 890–900 (1993).

12. Siegelbaum, S. A., Camardo, J. S. & Kandel, E. R. Serotonin & cyclic AMP closesingle K+ channels in Aplysia sensory neurones. Nature 299, 413–417 (1982).

13. Shen, K. Z., North, R. A. & Surprenant, A. Potassium channels opened bynoradrenaline and other transmitters in excised membrane patches ofguinea-pig submucosal neurones. J. Physiol. (Lond.) 445, 581–599 (1992).

14. Buckler, K. J. A novel oxygen-sensitive potassium current in rat carotid bodytype I cells. J. Physiol.(Lond.) 498, 649–662 (1997).

15. Ketchum, K. A., Joiner, W. J., Sellers, A. J., Kaczmarek, L. K. & Goldstein, S. A. N. A new family of outwardly-rectifying potassium channel proteinswith two pore domains in tandem. Nature 376, 690–695 (1995).

16. Goldstein, S. A. N., Price, L. A., Rosenthal, D. N. & Pausch, M. H. ORK1, apotassium-selective leak channel with two pore domains cloned fromDrosophila melanogaster by expression in Saccharomyces cerevisiae. Proc.Natl. Acad. Sci. USA 93, 13256–13261 (1996).

17. Goldstein, S. A. N., Bockenhauer, D., O’Kelly, I. & Zilberberg, N. Potassiumleak channels with 2 P domains. Nat. Reviews Neurosci. 2, 175–184 (2001).

18. Goldstein, S. A. N., Wang, K. W., Ilan, N. & Pausch, M. Sequence andfunction of the two P domain potassium channels: implications of anemerging superfamily. J. Molec. Med. 76, 13–20 (1998).

19. Patel, A. J. et al. A mammalian two pore domain mechano-gated S-like K+

channel. EMBO J. 17, 4283–4290 (1998).20. Maingret, F. et al. TREK-1 is a heat-activated background K+ channel.

EMBO J. 19, 2483–2491 (2000).21. Maingret, F., Patel, A. J., Lesage, F., Lazdunski, M. & Honore, E.

Lysophospholipids open the two-pore domain mechano-gated K+ channelsTREK-1 and TRAAK. J. Biol. Chem. 275, 10128–10133 (2000).

22. Fink, M. et al. Cloning, functional expression and brain localization of anovel unconventional outward rectifier K+ channel. EMBO J. 15, 6854–6862

(1996).23. Meadows, H. J. et al. Cloning, localisation and functional expression of the

human orthologue of the TREK-1 potassium channel. Eur. J. Physiol. 439,714–722 (2000).

24. Eves, E. M. et al. Immortal rat hippocampal cell lines exhibit neuronal andglial lineages and neurotrophin gene expression. Proc. Natl. Acad. Sci. USA89, 4373–4377 (1992).

25. Zilberberg, N., Ilan, N., Gonzalez-Colaso, R. & Goldstein, S. A. N. Openingand closing of KCNK0 potassium leak channels is tightly-regulated. J. Gen.Physiol. 116, 721–734 (2000).

26. Zhou, Y. et al. A dynamically regulated 14-3-3. Slob, and slowpoke potassiumchannel complex in Drosophila presynaptic nerve terminals. Neuron 22,809–818 (1999).

27. Vandenberg, C. A. Inward rectification of a potassium channel in cardiacventricular cells depends on internal magnesium ions. Proc. Natl. Acad. Sci.USA 84, 2560–2564 (1987).

28. Kubo, Y., Reuveny, E., Slesinger, P. A., Jan, Y. N. & Jan, L. Y. Primary structureand functional expression of a rat G-protein-coupled muscarinic potassiumchannel. Nature 364, 802–806 (1993).

29. Lopatin, A. N., Makhina, E. N. & Nichols, C. G. Potassium channel block bycytoplasmic polyamines as the mechanism of intrinsic rectification. Nature372, 366–369 (1994).

30. Fakler, B. et al. Strong voltage-dependent inward-rectification of inward-rectifier K+ channels is caused by intracellular spermine. Cell 80, 149–154(1995).

31. Ahmed, A. et al. A molecular target for viral killer toxin: TOK1 potassiumchannels. Cell 99, 283–291 (1999).

32. Vergani, P., Miosga, T., Jarvis, S. M. & Blatt, M. R. Extracellular K+ and Ba2+

mediate voltage-dependent inactivation of the outward-rectifying K+

channel encoded by the yeast gene TOK1. FEBS Lett. 405, 337–344 (1997).33. Vergani, P., Hamilton, D., Jarvis, S. & Blatt, M. R. Mutations in the pore

regions of the yeast K+ channel YKC1 affect gating by extracellular K+.EMBO J. 17, 7190–7198 (1998).

34. Loukin, S. H. et al. Random mutagenesis reveals a region important forgating of the yeast K+ channel Ykc1. EMBO J. 16, 4817–4825 (1997).

35. Loukin, S. H. & Saimi, Y. K+-dependent composite gating of the yeast K+

channel, Tok1. Biophys. J. 77, 3060–3070 (1999).36. Sigworth, F. J. Voltage gating of ion channels. Q. Rev. Biophys. 27, 1–40

(1994).37. Yellen, G. The moving parts of voltage-gated ion channels. Q. Rev. Biophys.

31, 239–295 (1998).38. Bean, B. P. Neurotransmitter inhibition of neuronal calcium currents by

changes in channel voltage dependence. Nature 340, 153–156 (1989).39. Colecraft, H. M., Patil, P. G. & Yue, D. T. Differential occurrence of reluctant

openings in G-protein-inhibited N- and P/Q-type calcium channels. J. Gen.Physiol. 115, 175–192 (2000).

40. Ilan, N. & Goldstein, S. A. N. KCNK0: single, cloned potassium leak channelsare multi-ion pores. Biophys. J. 80, 241–254 (2001).

41. McCormick, D. A. & Bal, T. Sleep and arousal: thalamocortical mechanisms.Annu. Rev. Neurosci. 20, 185–215 (1997).

42. Millar, J. A. et al. A functional role for the two-pore domain potassiumchannel TASK-1 in cerebellar granule neurons. Proc. Natl. Acad. Sci. USA 97,3614–3618 (2000).

43. Talley, E. M., Lei, Q. B., Sirois, J. E. & Bayliss, D. A. TASK-1, a two-poredomain K+ channel, is modulated by multiple neurotransmitters inmotoneurons. Neuron 25, 399–410 (2000).

44. Sirois, J. E., Lei, Q., Talley, E. M., Lynch, C. III & Bayliss, D. A. The TASK-1two-pore domain K+ channel is a molecular substrate for neuronal effects ofinhalation anesthetics. J. Neurosci. 20, 6347–6354 (2000).

45. Kim, Y., Bang, H. & Kim, D. TASK-3, a new member of the tandem pore K+

channel family. J. Biol. Chem. 275, 9340–9347 (2000).46. Colquhoun, D. & Sigworth, F. J. in Single-Channel Recording (eds. B.

Sakmann & E. Neher) 483–588 (Plenum, New York, 1995).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


The initial step in taste perception is the detection of tastants bytaste cells, which are clustered in taste buds on the tongue andother parts of the mouth1. On the tongue, taste buds are groupedin circumvallate, foliate and fungiform taste papillae, which havedistinct locations. Taste cells are depolarized by tastants, leadingto the transmission of signals to the brain via gustatory nervefibers that contact taste cells within the taste buds2.

Mammals can distinguish five taste qualities: sweet, sour, bit-ter, salty and umami (glutamate taste). Biochemical and electro-physiological studies indicate that the detection of salty and sourtastants is mediated by ion channels3. In contrast, sweet, bitterand umami taste transduction are likely to involve G protein-coupled receptors (GPCRs)3. A role for the taste-specific G pro-tein, gustducin, in bitter and sweet taste is further suggested bythe finding that mice lacking gustducin have deficits in the detec-tion of both bitter and sweet tastants4.

Consistent with these observations, GPCRs that function astaste receptors, or that are likely to do so, have now been identi-fied for both glutamate and bitter tastants5. A splice variant ofthe brain metabotropic glutamate receptor 4 (taste-mGluR4)6 isexpressed in taste cells, and, appropriate to the chemical diver-sity of bitter compounds, there is a family of approximately 25candidate bitter receptors7,8 (T2Rs), several of which have beenshown to recognize bitter compounds9. Although a sweet tastereceptor has been identified in Drosophila10, a mammalian sweetreceptor has not yet been found. The functions of two otherGPCRs expressed in mammalian taste cells, T1R1 and T1R2, arenot yet known, but their subcellular location suggests that theyfunction as taste receptors11.

Several genetic loci that have been identified in mouse orhuman control sensitivity to bitter or sweet compounds12. One ofthese, the Sac locus, governs the sensitivity of mice to certainsweet tastants, including sucrose and the artificial sweetener, sac-charin13–18. C57BL/6J (B6) mice (‘tasters’) prefer water that con-tains 1.6 mM saccharin (or 50 mM sucrose) to water that doesnot14. In contrast, DBA/2J (D2) mice (‘nontasters’) exhibit thispreference only at higher concentrations of the sweet tastants

(8 mM saccharin)13–19 and even then, the preference is less pro-nounced. At higher concentrations, saccharin and sucrose alsoelicit greater responses in gustatory fibers of B6 than D2 (or othernontaster) mice17,20.

Previous studies showed that some T2r genes are located ator near genetic loci that control sensitivity to bitter tastes7,8.To explore whether the mouse Sac locus might similarly con-tain a gene encoding a sweet taste receptor, we searched thesyntenic region of the human genome for genes encodingGPCRs. Using this approach, we identified T1r3, a gene encod-ing a GPCR that is expressed in a subset of taste cells in mouse.Chromosomal mapping studies using both a radiation hybridpanel and recombinant inbred strains showed that T1r3 isclosely linked to the Sac locus on mouse chromosome 4.Sequence analyses of T1r3 in Sac taster versus nontaster strainsfurther revealed allelic differences among mouse strains thatcould result in differences in Sac phenotype. Surprisingly, insitu hybridization studies showed that T1R3 is expressed in thesame taste cells as T1R2, a related receptor, raising the possi-bility that the two receptors function as heterodimers or thatthese cells recognize more than one ligand.

RESULTSIdentification of a candidate taste receptor, T1R3To investigate whether Sac phenotypes might result from poly-morphisms in a sweet receptor gene, we first asked whether thereis a gene that encodes a GPCR in the vicinity of the Sac locus.The Sac locus maps near the distal end of mouse chromosome 4at about 83 cM from the centromere13–15,17,18. Using The Jack-son Laboratory Mouse Informatics database, we determinedthat the corresponding region in human is at chromosome 1p36.

To examine whether there is a gene encoding a GPCR at 1p36,we first used the NCBI human genome sequence database tosearch for genes in this region that encode receptors related toother GPCRs8. We found none in the finished sequence data-base. However, in the draft sequence database (htgs), we foundtwo BAC clones (AC026283 and AL139287) that overlapped with

A candidate taste receptor genenear a sweet taste locus

Jean-Pierre Montmayeur, Stephen D. Liberles, Hiroaki Matsunami and Linda B. Buck

Howard Hughes Medical Institute, Department of Neurobiology, Harvard Medical School, Boston, Massachusetts 02115, USA

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

The mechanisms underlying sweet taste in mammals have been elusive. Although numerous studieshave implicated G proteins in sweet taste detection, the expected G protein-coupled receptors havenot been found. Here we describe a candidate taste receptor gene, T1r3, that is located at or near themouse Sac locus, a genetic locus that controls the detection of certain sweet tastants. T1R3 differs inamino acid sequence in mouse strains with different Sac phenotypes (‘tasters’ versus ‘nontasters’). Inaddition, a perfect correlation exists between two different T1r3 alleles and Sac phenotypes inrecombinant inbred mouse strains. The T1r3 gene is expressed in a subset of taste cells incircumvallate, foliate and fungiform taste papillae. In circumvallate and foliate papillae, most T1r3-expressing cells also express a gene encoding a related receptor, T1R2, raising the possibility that thesecells recognize more than one ligand, or that the two receptors function as heterodimers.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

contigs (NT_004384/NT_002209) assigned to chromosome 1p36,and that contained a gene encoding a potential taste receptor.This receptor, ‘hT1R3,’ is related to the putative taste receptorsT1R1 and T1R2 (Fig. 1). Although we also identified anothergene at 1p36 that encodes a novel GPCR (as well as one encodinga serotonin receptor), further studies indicated that it is notexpressed in taste tissue (data not shown).

We next isolated the mouse ortholog of hT1r3. Using degen-erate primers, we first amplified a fragment of the mT1r3 genefrom mouse genomic DNA. Consistent with the expression ofT1r3 in taste tissue, using this fragment as probe, we were ableto isolate two T1r3 cDNA clones from a cDNA library preparedfrom mouse circumvallate and foliate taste papillae. In situhybridization experiments verified that the T1r3 gene is expressedin mouse taste cells (see below).

In Fig. 1, the deduced sequences of mT1R3 and hT1R3 arealigned with mT1R1 and mT1R2 sequences11 that we obtainedfrom cDNA clones. mT1R3 is related to mT1R1 (32% aminoacid identity) and mT1R2 (30%) as well as the calcium sens-ing receptor21 (29%) and the V2R family of candidatepheromone receptors22–24 (28% maximum identity). Aminoacid identity between mT1R3 and hT1R3 (72%) is similar to

that between mT1R1 and hT1R1 (74%). Allof the receptors shown have an extremelylong N terminal extracellular domain(NTD), which, by analogy with the struc-turally related mGluRs, may be involved inligand binding25,26.

In previous studies, we obtained evidencethat alternative RNA splicing generatesmT1R2 variants that differ in the NTD (J-P. M & L.B.B., unpublished data). Wefound that all three T1Rs appear to be encod-ed by 6 exons, 5 of which encode the NTD.However, RT-PCR amplification of mT1r3and mT1r1 cDNAs encoding NTD segmentsdid not yield PCR products of different sizes,suggesting that variants of T1R3 and T1R1similar to those seen for T1R2 are not pro-duced (data not shown). As reported for ratT1r1 and T1r2 (ref. 11), probes prepared fromsegments encoding membrane-spanningparts of mT1r1, mT1r2 and mT1r3 hybridized

to single bands in Southern blots of restriction-enzyme-digest-ed mouse genomic DNA (data not shown), suggesting theabsence of closely related genes.

The mT1r3 gene maps close to the Sac locusThe presence of hT1r3 at human chromosome 1p36 suggestedthat the mT1r3 gene would be in the corresponding region ofmouse chromosome 4, near the Sac locus. To investigate thisissue, we used the T31 radiation hybrid panel27 to map the chro-mosomal location of mT1r3 (Fig. 2). We also mapped the loca-tions of the mT1r1 and mT1r2 genes with this panel. To refinethe order of markers in the distal region of chromosome 4beyond what was currently available, we also mapped the loca-tions of other markers in this region (Dvl28; Gnb1, ref. 29;D4Smh6b30; D18346, ref. 31 and V2r2, ref. 23). Several othermarkers previously mapped in this region using the T31 panelare also shown in Fig. 2.

These experiments showed that the mT1R3 gene is indeedlocated near the end of chromosome 4, close to the Sac locus(Fig. 2). Previous mapping studies indicate that the Sac locus islocated on chromosome 4 in a region distal to D4Smh6b15.More recent studies position Sac in an approximately 2.6-cM

Fig 1. T1R3 differs in mice with different Sac phe-notypes. The deduced protein sequences ofhT1R3 and the mT1R1, mT1R2 and mT1R3 ofC57BL/6J (B6) mice are aligned with those of themT1R3s of DBA/2J (D2) and SWR/J (SW) mice.Predicted signal sequences are boxed, amino acidspresent in all the proteins are shaded in gray, andthe seven potential transmembrane domains char-acteristic of GPCRs are indicated by horizontalbars and Roman numerals. Amino acid differencesin mT1R3 in mice that differ in Sac phenotype areshaded in black. B6 mice are tasters, D2 mice arenontasters, and SW mice have an intermediate Sacphenotype. B6 and D2 mT1R3s differ at 6 posi-tions (residues 55, 60, 61, 371, 706 and 855).SWR/J (SW) mice have the same amino acid as B6mice at T1R3 positions 55 and 60, but the same asD2 mice at the other positions except at residues261 and 692, where they differ from both B6 andD2 mice.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


segment, and in a location distal to D4Mit256 and proximal toD18346 (ref. 31). Similarly to Sac, T1r3 mapped in a locationdistal to both D4Smh6b and D4Mit256 and proximal toD18346. As in previous studies, mT1r1 and mT1r2 alsomapped to the distal region of chromosome 4 (ref. 11; N. Ryba,personal communication). However, whereas Sac and mT1r3are both distal to D4Smh6b15, mT1r1 and T1r2 are both prox-imal to this marker (Fig. 2). This is consistent with a reportthat mT1r1 is in a location about 5 cM proximal to Sac31.

Polymorphisms in the coding region of the T1r3 geneTo investigate whether differences in mT1R3 could be respon-sible for Sac phenotypes, we compared the sequences of mT1r3cDNAs amplified from B6 (taster) and D2 (nontaster) mice(Fig. 1). These experiments revealed six amino acid changes inT1R3 in D2 compared to B6 mice. Four changes (T55A, I60T,P61L, R371Q) are located in the NTD, one (I706T) is at theend of transmembrane domain 4, and one (G855E) is in the C terminal cytoplasmic domain. The first three amino acid dif-ferences are located near residues that are altered in severaldefective forms of the calcium sensing receptor32. One or moredifferences in T1R3 in this region in B6 versus D2 mice maysimilarly affect the function of T1R3.

We also examined whether there are differences in mT1R1,mT1R2 or Gnb1 in B6 versus D2 mice by comparing cDNAsamplified from taste tissue of the two strains. Gnb1 (whichmaps near the Sac locus; Fig. 2) encodes the G protein beta-1subunit, which is expressed in taste cells, and is therefore a pos-sible candidate for a Sac locus gene29. We found that the cod-ing regions of both the T1r1 and Gnb1 genes are identical inB6 and D2 mice. We identified several nucleotide differencesin B6 versus D2 T1r2 genes, two resulting in changes in theNTD (Fig. 1), but the chromosomal location of the T1r2 geneexcluded a role for these changes in Sac phenotype. Weobtained further evidence against a role for Gnb1 in Sac phe-

notype by in situ hybridization experiments, which failed toreveal any difference in the level or patterning of expression ofGnb1 mRNAs in B6 versus D2 mice (data not shown).

T1r3 alleles and Sac phenotypes are correlatedPrevious studies determined the Sac phenotypes of recombinantinbred (RI) strains derived from crosses between B6 and D2 mice(BXD/Ty strains)14. To determine whether individual RI strainshave the B6 or D2 T1r3 allele, we used DNA from each RI strainin PCR reactions with primers that would amplify only the B6or D2 allele (Fig. 3). The 3´ end of the 3´ primers matchedsequence-encoding amino acids 60 and 61 in one or the otherallele. Taking advantage of a DraIII site in only the D2 allele atthis site (because of the amino acid 60 codon), we also amplifieda larger T1r3 DNA segment containing this region from each RIstrain DNA, and then digested the segment with DraIII.

These experiments revealed a perfect correlation betweenSac phenotypes and T1r3 alleles (Fig. 3). All of the taster RIstrains gave a PCR product with only the B6-specific primer,and all of the nontaster strains yielded a product with only theD2-specific primer. Moreover, DraIII cleaved all of the T1r3gene segments amplified from nontaster DNAs, but none ofthose obtained from taster DNAs. In contrast, similar studies ofT1r2 alleles showed a correlation with the Sac phenotype inonly 16 of 21 RI strains (Fig. 3). Additional experiments usingprimers matching only one allele of V2r2 showed a correlationin only 18 of 21 strains (Fig. 3), whereas allelic primers forD18346 (ref. 31) gave results identical to the T1r3 primers (datanot shown), consistent with its location proximal to V2r2.These results indicated that T1r3 is located 0–5.3 cM (p = 0.05)from the Sac locus between D4Smh6b andV2r215.

These results are consistent with those obtained using theradiation hybrid mapping panel, which placed T1r3 in the same∼ 2.6-cM segment of chromosome 4 as the Sac locus betweenD4Mit256 and D18346 (ref. 31). In summary, the order of mark-ers determined from the RI strains was D4Smh6b–Sac/T1r3/Dvl/D18346–V2r2; in the higher resolution radiation hybrid map,the order was D4Mit256–D4Smh6b–T1r3/Dvl–D18346–V2r2.Studies of the Sac locus using F2 hybrids and partially congenicmouse strains indicate an order of D4Mit256–Sac–D18346(ref. 31). Thus, both T1r3 and Sac are flanked proximally byD4Mit256 and D4Smh6b and distally by D18346 and V2r2(D4Mit256– D4Smh6b–Sac/T1r3–D18346–V2r2).

Using the same methods, we examined the DNAs of severalother inbred mouse strains for which the Sac phenotypes areknown14,19,33,34 (Fig. 3). DNAs from 129/SvJ, Balb/cBy andC3H/HeJ mice, all of which are nontasters14,16, gave T1r3 PCRproducts only with the D2-specific primer, whereas DNA from

Fig. 2. The mT1r3 gene is located near the Sac locus on mouse chro-mosome 4. Right, results of chromosome mapping studies conductedwith the T31 radiation hybrid mapping panel, with relative distancesshown in centirays (cR; see scale bar). The distal region of mouse chro-mosome 4 displayed is indicated on the chromosome 4 diagram at left.The locations of T1r1, T1r2, T1r3, Gnb1, Dvl, and V2r2, D18346 and theD4Smh6b marker were mapped in the present studies; selected othermarkers that were mapped previously in other studies are shown forreference. Data for the anchor loci was obtained from The JacksonLaboratory Mouse Radiation Hybrid Database (http://www.jax.org/resources/documents/cmdata/rhmap/4data.html). All marker namesshould be italic but are in plain text for legibility. The Sac locus has beenmapped distal to D4Smh6b. T1r3 also mapped distal to D4Smh6b,whereas Gnb1 mapped to the same location as D4Smh6b, and T1r1 andT1r2 both mapped proximal to it.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

C57BL/6By, a taster strain closely related to B6, gave a productwith only the B6 primer. In addition, T1r3 segments amplifiedonly from the first 3 strains were cleaved with DraIII, providingfurther evidence that the T1r3 of nontaster strains have the sameamino acid as D2 mice at position 60.

SWR/J mice have an intermediate Sac phenotype and cannotbe classified as either tasters or nontasters14. Using DNA fromSWR/J mice, neither T1r3 allele-specific primer yielded a PCRproduct, even though digestion with DraIII indicated that thisstrain has the same amino acid as B6 at position 60. Sequenceanalysis of the SWR/J T1r3 gene showed that amino acid 60 isthe same as in B6 mice, but amino acid 61 is the same as in D2mice (Fig. 1). In addition, this strain has the same amino acid asB6 at position 55, but the same amino acids as D2 at positions371, 706 and 855. It also differs from both B6 and D2 at posi-tions 261 (R261C) and 692 (L692S). These findings are consis-tent with a role for T1R3 in Sac phenotypes.

The expression of T1r3 in taste cellsTo examine the expression of the T1r3 gene in taste tissue, wehybridized mT1r3 cRNA probes (or mT1r1 or mT1r2 probes)to sections through circumvallate, foliate and fungiform tastepapillae of both B6 and D2 mice. These studies showedthat the mT1r3 gene is expressed in all three types of tastepapillae (Fig. 4) as well as in taste buds on the palate (data

not shown). The T1r3 probe hybridized to approx-imately 24% of cells (55 of 229) in circumvallatepapillae, 15% of cells (16/108) in fungiform and14% of cells (33/239) in foliate papillae. In con-trast, as previously reported in rat11, the mT1r2gene appeared to be expressed mainly in circum-vallate and foliate taste papillae, whereas themT1r1 gene was expressed strongly only in fungi-form papillae (Fig. 4). None of the probeshybridized to sections of the olfactory epitheliumor vomeronasal organ (data not shown).

We next used two-color in situ hybridization toask whether T1R3 is expressed in the same cells asthe taste-specific G protein, gustducin. In sectionsof circumvallate and foliate papillae, mT1r3 andgustducin probes hybridized largely to separatepopulations of taste cells (Fig. 5). Only about 10%of T1r3+ cells were also gustducin+ and only about10% of gustducin+ cells were T1r3+ in circumval-late papillae. In foliate papillae, ∼ 19% of T1r3+ cellswere also gustducin+ and vice versa. Previous stud-ies in rat indicate a similarly small overlap betweencells expressing gustducin versus T1r1 or T1r2 (ref.11). These results suggested that most taste cellsthat express T1R3 do not express gustducin. How-ever, a low level of gustducin expression in the

remaining T1r3+ cells could not be excluded.We then compared the expression of T1r3 in individual cells

with that of T1r1 and T1r2. Surprisingly, 98% of circumvallatetaste cells that hybridized to the T1r3 probe also hybridized tothe T1r2 probe. In addition, all T1r2+ cells were also T1r3+. Inpreliminary experiments, analysis of a few fungiform taste cellsexpressing the T1r3 gene similarly suggested that a proportionof those cells may coexpress the T1r1 gene, but further experi-ments are needed to clarify this issue. The finding that T1r3 andT1r2 genes are expressed in the same cells is in striking contrastto previous observations that T1r1 and T1r2 genes are predom-inantly expressed in different taste papillae, and that only a smallpercentage of T1r2+ foliate taste cells coexpress T1r1 (ref. 11).

DISCUSSIONThe results presented here identify T1R3 as a candidate tastereceptor in mouse and human, possibly a receptor for sweet tas-tants. First, consistent with previous studies implicating G pro-teins in sweet taste transduction1,3, T1R3 is a member of theGPCR superfamily that is expressed in a subset of taste cells inthe mouth. Second, the mT1r3 gene is located at or near the Saclocus, a mouse genetic locus that controls sensitivity to certain

Fig. 4. The pattern of expression of T1r3 differs from that ofT1r1 and T1r2. Sections of circumvallate (CV), foliate (Fol), andfungiform (Fun) taste papillae of C57BL/6J (B6) or DBA/2J (D2)mice were hybridized with digoxigenin-labeled mT1r1, mT1r2,mT1r3 or gustducin cRNA probes. The T1r3 probe labeledtaste cells in taste buds of all three types of papillae, whereasthe T1r1 probe predominantly labeled taste cells in fungiformpapillae and the T1r2 probe labeled cells in circumvallate andfoliate, but rarely in fungiform papillae. No significant differenceswere seen in the extent or patterning of labeling in B6 versusD2 mice. Scale bar, 50 µm.

Fig. 3. Polymorphisms in the T1r3 gene are correlated with Sac phenotypes. PCR withallele-specific primers and restriction enzyme digestion was used to examine mT1r3 andmT1r2 alleles in a series of recombinant inbred (RI) strains (1–32) and other inbred(B6–SW) mouse strains with known Sac phenotypes (+, taster; –, nontaster; +/–, interme-diate taster) as previously determined by Lush14. Reaction products were electrophoresedon agarose gels with 100-bp ladder size markers (M). The primers and enzyme used and thegene analyzed in each row are indicated at left. A perfect correlation is seen between B6and D2 mT1r3 alleles and Sac phenotypes in the RI strains, whereas the correlation is onlypartial for mT1r2 and V2r2 alleles. A perfect correlation was also seen between amplifica-tion with mT1r3 allele-specific primers and restriction enzyme digestion and the taster andnontaster Sac phenotypes of the other inbred strains, but an intermediate taster strainyielded mixed results.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


sweet tastants13–15,17,18. And finally, there are amino acid differ-ences in T1R3 in Sac taster, nontaster and intermediate tastermice that could affect the function of T1R3 and thereby accountfor differences in Sac phenotype.

Null mutant mice lacking gustducin show deficiencies in thedetection of both sweet and bitter tastants4. This suggests thatgustducin may be expressed by many taste cells that recognizesweet tastants. In contrast, we detected gustducin in only approx-imately 10% of T1R3-expressing taste cells in the circumvallatepapillae and approximately 20% of those in the foliate papillae.There are several possible explanations for this apparent dis-crepancy. One is that T1r3 is not a sweet receptor gene and thatSac phenotypes are determined by another gene, for example, atranscription factor gene that controls the expression of a sweetreceptor. Indeed, the 2.6-cM chromosomal segment to which Sacis mapped undoubtedly contains numerous genes. Another pos-sible explanation is that many T1R3-expressing cells express alow level of gustducin that was not detectable by the in situhybridization methods used in our studies. Yet another possibil-ity is that some cells that detect sweet tastants do express gust-ducin, whereas others do not. Consistent with this idea, the abilityto detect sweet tastants is reduced, but not eradicated in gust-ducin null mutant mice4. In addition, there is biochemical evi-dence for the coupling of bitter taste receptors, but not sweet tastereceptors, to gustducin3,9,35,36.

Our studies also argue against an involvement of T1R1, T1R2or Gnb1 in Sac phenotypes. We found that T1R1 and Gnb1 bothhave identical sequences in the Sac taster and nontaster mousestrains B6 and D2. In addition, in situ hybridization indicatedthat the levels and patterning of expression of T1r1 and Gnb1genes are the same in taster and nontaster mice. Moreover, con-sistent with one previous report31, we found that the T1r1 genemaps at a distance from the Sac locus on mouse chromosome4, proximal to a marker (D4Smh6b) that is, in turn, proximal tothe Sac locus and the mT1r3 gene. Although we identified dif-ferences in T1R2 in taster and nontaster mice, we found that theT1r2 gene is located even further from the Sac locus than theT1r1 gene, excluding T1R2 from a role in Sac phenotype.

Surprisingly, we found that the T1r3 gene is coexpressed withthe T1r2 gene in many taste cells. Our studies indicated that near-ly all circumvallate and foliate taste cells that express T1R3 orT1R2 express both receptors. In fungiform papillae, preliminaryexperiments suggested that T1R3 might similarly be coexpressedwith T1R1, at least in some cells. This raises the possibility that,like the structurally-related GPCR, GABAB

37, T1R3 is expressedas a heterodimer with T1R2 or T1R1. Another possibility is thatthe T1Rs function separately, perhaps each recognizing one ormore different ligands. Finally, the receptors might exist as bothhomodimers and heterodimers. This scheme could conceivablyexplain how sucrose and saccharin can stimulate different signaltransduction cascades in a single taste cell38.

As in previous studies with rat T1R1 and T1R2 (ref. 11, 39),we were unable to obtain functional expression of any of the

mouse T1Rs in heterologous cells. Although functional expres-sion has been obtained by attaching a short N terminal segmentof rhodopsin to the N termini of some T2Rs9, this approach wasunsuccessful with the T1Rs, including when T1R3 was coex-pressed with T1R1 or T1R2, and when receptors or pairs of recep-tors were coexpressed with various G proteins. The cellsexpressing these receptors did not respond to saccharin, othersweeteners, or other tastants that were tested, including severalbitter tastants and glutamate. Similar difficulties have beenencountered in expressing most other types of chemosensoryreceptors, including V1Rs, V2Rs and most odorant receptors. Asfor those receptors, the functions of T1R3 and the other two T1Rsremain to be defined.

How many sweet receptors are there? Sweet taste chemicalsinclude sugars, certain amino acids and proteins, and several arti-ficial sweeteners with differing chemical structures, which sug-gests that there might be multiple different sweet receptors40.Although some modeling studies predict the existence of a singlesweet receptor41,42, the existence of at least several different sweetreceptors is supported by a number of human taste perceptionstudies, as well as electrophysiological studies in rodents43–45. Inhumans, cross-adaptation studies indicate that perceptual adap-tation to one sweet tastant does not necessarily result in an inabil-ity to detect another sweet tastant46,47. In mice, single taste cellscan be depolarized by multiple different sugars, but adaptation ofa cell to one of those sugars does not result in adaptation to allof the others48. Although these studies suggest that there is morethan one sweet receptor, the actual number of receptors thatdetect sweet compounds remains to be determined.

If T1R3 is indeed a sweet receptor, are there additional recep-tors that recognize sweet tastants? One possibility is that allthree T1Rs are sweet receptors. In this scenario, the differentT1Rs could recognize different or partially overlapping sets ofsweet tastants. Alternatively, different homodimers or het-erodimers of the T1Rs could have different specificities. Theremight also be additional members of the T1R family that areinvolved in sweet taste, but that are not yet identified. Anotherpossibility is that some members of the T2R family recognizesweet rather than bitter tastants. Future studies should ulti-mately reveal whether one or both of these receptor families isindeed involved in sweet taste or whether sweet taste involvesyet another receptor family.

Fig. 5. T1r3 is coexpressed with T1r2, but not gustducin. Sections of B6 cir-cumvallate papillae were hybridized simultaneously with a fluorescein-labeledT1r3 probe and a digoxigenin-labeled T1r3, T1r2 or gustducin (gust) probe.The digoxigenin-labeled probe was revealed with Alexa 488 (green), and thefluorescein-labeled probe, with HNPP/fast red (red). Each row shows redand green signals obtained from the same field and an overlay in which doublylabeled cells are yellow. The results indicate that most taste cells that expressT1R3 or T1R2 express both receptors, whereas T1R3 and gustducin are pri-marily expressed in different cells. Scale bar, 50 µm.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

METHODSCloning of T1Rs. The sequence of hT1R3 was deduced from BAC clonesAC026283 and AL139287 (NCBI). Pairs of degenerate primers (SW1/SW14or SW7/SW14) matching hT1R3 were used to amplify a T1r3 gene segmentfrom mouse genomic DNA. PCR products of the expected size were cloned(TA Topo-4 vector; Invitrogen, Carlsbad, California), sequenced, and usedto screen (55°C) a mouse (C57BL/6J) circumvallate and foliate taste papil-lae cDNA library prepared in λΖAPII (Stratagene, La Jolla, California)49.One 2.8-kb cDNA clone encoded most of mT1R3. cDNAs encoding theremaining coding region were obtained from the cDNA library by PCR.(SW1, 5´CA(C/T)(C/T)CITG(C/T)TG(C/T)T(A/T)(C/T)GAITG-(C/T)(A/G)TIGA(C/T)TG-3´; SW7, 5´-CCI(A/T)CICCIGCIAGITG-(T/C)TTIGCI-CA(A/G)CA-3´; SW14, 5´-A(A/G)IA(A/G)IGCICCC-AT(C/T)TGIACIGCIGG-3´).

We obtained mT1r1 and mT1r2 cDNAs in previous studies in whichwe used PCR with degenerate primers matching various known GPCRsto search for candidate taste receptors (unpublished data). A 220-bp frag-ment obtained with one primer pair (LB2/LB4) was cloned andsequenced, revealing a segment of the mT1r2 coding region. Two ESTswith related, but distinct sequences (W18663, mouse; AA853697, human)that encoded part of T1R1 were identified in the dbest database. Usingthe T1r2 cDNA and T1r1 ESTs (Genome Systems, Palo Alto, California)as probes, clones encoding mT1R1 and mT1R2 were isolated from thetaste tissue cDNA library. The sequences of these cDNAs are identical tothose reported11,31. (LB2, 5´-AA(C/T)CTICCIGAIAA(C/T)TA(C/T)-(A/T)A(C/T)GA(A/G)G(G/C)IAA-3´; LB4, 5´- GGICGI(G/C)(A/T)-IAGIATIA(C/T)(A/G)TA(A/G)CA(C/T)TTIGG-3´.)

Chromosome mapping. DNA segments were amplified from the RH T31mapping panel (Research Genetics, Huntsville, Alabama), using condi-tions specified by the manufacturer, and primers matching the genesencoding mT1R1 (20R9/43R9; 186-bp fragment), mT1R2 (31R7/33R7;776-bp fragment), mT1R3 (15TR3/11TR3; 357-bp fragment), Dvl (3Dvl-1/4Dvl-1; 191-bp fragment), V2r2 (7VR2/8VR2; 302-bp fragment), themicrosatellite marker D4Smh6b, the STS D18346 (1D18346/5D18346;170-bp fragment) and Gnb1 (7TB1/8TB1; 371-bp fragment). Resultswere interpreted using The Jackson Laboratory Mouse Radiation Hybriddatabase (http://www.jax.org/resources/documents/cmdata/rhmap/rhsubmit.html). Advice on data interpretation was provided by L. Rowe(The Jackson Laboratory, Bar Harbor, Maine). Primer sequences usedin the mapping are as follows: 20R9, 5´-CCACTCTGAGTGGCGGC-3´;43R9, 5´-GACCCGCCACCTTCAGCC-3´; 31R7, 5´-AGCTGCCATG-GATAGACG-3´; 33R7, 5´-ACATATAGCGCCATCACC-3´; 15TR3, 5´-TTCCATCACTACAGATGAC-3´; 11TR3, 5´-GCTCAGGCAGCCTGTGAG-3´; 3Dvl-1, 5´-GGCCGCATTGAGCCGGGC-3´; 4Dvl-1, 5´-CCCTGTCTGGGACACGAT-3´; 7VR2, 5´-ACTTACAAACTGGAC-AGC-3´; 8VR2, 5´-CTTTGTCATAGAACATTG-3´; 7TB1, 5´-GATGATGGCATGGCTGTG-3´; 8TB1, 5´-ATTGGACCAACCCAAAAC-3´; 1D18346,5´-TGTCCGCAGTGTGGAAACTA-3´; 5D18346, 5´-AAGGGAT-GTCCAGGGTAGAG-3´.

Analyses of inbred mouse strains. Fifty nanograms of genomic DNAfrom BXD/Ty RI or other inbred strains (The Jackson Laboratory, BarHarbor, Maine) were used in PCR reactions: 94°/1 min, 54°/1 min,72°/2 min + 6 s extension for 38 cycles. For T1R3, a 5´ primer match-ing both B6 and D2 alleles (38TR3) was used in combination with a3´ primer specific for B6 (BXD-B6) or D2 (BXD-DBA) allele. Alter-natively, primers matching both alleles (38TR3/36TR3) were used andthe resulting 265-bp product was digested with DraIII. For T1r2, a 5´primer matching both B6 and D2 alleles (5R7) was used in combina-tion with a 3´ primer specific for the B6 allele (BXD.B6TR2). Alterna-tively, primers matching both alleles (5R7/53R7) were used and the529 bp product was digested with SphI, which cleaves only the D2allele, because of a difference in the codon for amino acid 352. ForV2r2 (‘VR2’23), a 5´ primer matching both B6 and D2 alleles (5VR2)was used with a B6-allele specific primer (6VR2). Primer sequencesused for allele typing are as follows: 38TR3, 5´-AGGGGACTACAT-ACTGGG-3´; BXD-B6, 5´-GAGAACCTGTTGCACGGGA-3´; BXD-DBA, 5´GAGAACCTGTTACACAGGG-3´; 36TR3, 5´-TTGATC-TCCTCCACAGCC-3´; 5R7, 5´-CCTGTACCAGAACCCAAC-3´;BXD.B6Tr2, 5´-GGATACTCTGGCTTGTCG-3´; 53R7, 5´-GATAGAC-

GATTTGCTTCG-3´; 5VR2, 5´-AGATGTCCAGATAATAAA-3´; 6VR2,5´-AGGTGACCTGGTCGGGAT-3´.

To compare mT1R sequences in different strains, cDNA was prepared(First Strand cDNA Synthesis Kit, Life Technologies, Rockville, Mary-land) from RNA isolated from circumvallate, foliate and fungiform tastepapillae of adult mice49, and used in PCR reactions to specific with spe-cific primers. The RT-PCR products49 were isolated from agarose gelsand then directly sequenced using specific primers50.

In situ hybridization. Sixteen-micrometer sections of adult mouse tastepapillae were hybridized (58°C) to digoxigenin-labeled cRNA probesprepared from cloned segments of cDNAs encoding mT1R1 (nt1881–2564 plus nt 29–1778), mT1R2 (nt 152–1754), mT1R3 (nt1–1700), or gustducin (fragment encoding amino acids 22–328). Exper-iments were done as described previously23, except that, following fixa-tion, sections were treated with 0.2 M HCl (8 min, RT) and thenproteinase K (10 µg/ml; Roche, Indianapolis, Indiana; 10 min, RT). Sec-tions were counterstained with Hoechst 33258 (Sigma, St. Louis, Mis-souri) to visualize taste cell nuclei and determine the percentage of tastecells labeled with individual probes23.

Two-color in situ hybridization experiments used the same conditions,with the following modifications. Twelve-micrometer sections of tastepapillae were hybridized to both digoxigenin- and fluorescein-labeled(Roche) cRNA probes23. Following hybridization, sections were incu-bated with tyramide blocking reagent (NEN, Boston, Massachusetts),incubated with peroxidase-anti-digoxigenin and alkaline phosphatase-anti-fluorescein antibodies (Roche; 1 h, RT), washed, incubated withtyramide-biotin (NEN, Boston, Massachusetts; 10 min), washed, incu-bated with streptavidin-Alexa 488 (Molecular Probes, Eugene, Oregon)in the dark (30 min), washed, treated with HNPP/ Fast Red alkaline phos-phatase substrate (Roche), and then counterstained with Hoechst 33258(ref. 23; Sigma). Reagents were used essentially as recommended by themanufacturers. Sections mounted in gelvatol were examined using con-focal microscopy (Biorad, Herts, England).

ACKNOWLEDGEMENTSWe thank L. Rowe at The Jackson Laboratory for suggestions and contributions

to the chromosome mapping studies. We also thank C. Gao for technical

assistance, and members of the Buck lab, in particular, C. Neophytou, for help,

comments and discussions throughout this project. This work was supported by

the Howard Hughes Medical Institute, grants from the National Institutes of

Health (L.B.B.), and fellowship support from the Alice and Joseph Brook Fund

(J.-P.M.), the Naito Foundation (H.M.), and the Japan Society for the

Promotion of Science (H.M.). GenBank accession number for mT1R1, mT1R2

and mT1R3 are AF337039-41

RECEIVED 19 MARCH; ACCEPTED 28 MARCH 2001

1. Lindemann, B. Taste reception. Physiol. Rev. 76, 718–766 (1996).2. Herness, M. S. & Gilbertson, T. A. Cellular mechanisms of taste transduction.

Annu. Rev. Physiol. 61, 873–900 (1999).3. Gilbertson, T., Damak, S. & Margolskee, R. The molecular physiology of taste

transduction. Curr. Opin. Neurobiol. 10, 519–527 (2000).4. Wong, G. T., Gannon, K. S. & Margolskee, R. F. Transduction of bitter and

sweet taste by gustducin. Nature 381, 796–800 (1996).5. Kinnamon, S. A plethora of taste receptors. Neuron 25, 507–510 (2000).6. Chaudhari, N., Landin, A. & Roper, S. A metabotropic glutamate receptor

variant functions as a taste receptor. Nat. Neurosci. 3, 113–119 (2000).7. Adler, E. et al. A novel family of mammalian taste receptors. Cell 100, 693–702

(2000).8. Matsunami, H., Montmayeur, J.-P. & Buck, L. A family of candidate taste

receptors in human and mouse. Nature 404, 601–604 (2000).9. Chandrashekar, J. et al. T2Rs function as bitter taste receptors. Cell 100,

703–711 (2000).10. Ishimoto, H., Matsumoto, A. & Tanimura, T. Molecular identification of a

taste receptor gene for trehalose in Drosophila. Science 289, 116–119 (2000).11. Hoon, M. A. et al. Putative mammalian taste receptors: a class of taste-specific

GPCRs with distinct topographic selectivity. Cell 96, 541–551 (1999).12. Reed, D. Gene mapping for taste related phenotypes in human and mice.

Appetite 35, 189–190 (2000).13. Fuller, J. Single-locus control of saccharin preference in mice. J. Hered. 65,

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


33–36 (1974).14. Lush, I. The genetics of tasting in mice: VI. Saccharin, acesulfame, dulcin and

sucrose. Genet. Res. 53, 95–99 (1989).15. Lush, I., Hornigold, N., King, P. & Stoye, J. The genetics of tasting in mice.

VII. Glycine revisited, and the chromosomal location of Sac and Soa. Genet.Res. 66, 167–174 (1995).

16. Capeless, C. & Whitney, G. The genetic basis of preference for sweetsubstances among inbred strains of mice: Preference ratio phenotypes andthe alleles of the Sac and dpa loci. Chem. Senses 20, 291–298 (1995).

17. Bachmanov, A. et al. Sucrose consumption in mice: major influence of twogenetic loci affecting peripheral sensory responses. Mamm. Genome 8,545–548 (1997).

18. Blizard, D., Kotlus, B. & Frank, M. Quantitative trait loci associated withshort-term intake of sucrose saccharin and quinine solutions in laboratorymice. Chem. Senses 24, 373–385 (1999).

19. Belknap, J. et al. Single locus control of saccharin intake in BXD/Tyrecombinant inbred (RI) mice: some methodological implications for RIstrain analysis. Behav. Genet. 22, 81–100 (1992).

20. Frank, M. & Blizard, D. Chorda tympani responses in two inbred strains ofmice with different taste preferences. Physiol. Behav. 67, 287–297 (1999).

21. Brown, E. et al. Cloning and characterization of an extracellular Ca2+-sensingreceptor from bovine parathyroid. Nature 366, 575–580 (1993).

22. Herrada, G. & Dulac, C. A novel family of putative pheromone receptors inmammals with a topographically organized and sexually dimorphicdistribution. Cell 90, 763–773 (1997).

23. Matsunami, H. & Buck, L. A multigene family encoding a diverse array ofputative pheromone receptors in mammals. Cell 90, 775–784 (1997).

24. Ryba, N. J. & Tirindelli, R. A new multigene family of putative pheromonereceptors. Neuron 19, 371–379 (1997).

25. Takahashi, K., Tsuchida, K., Tanabe, Y., Masu, M. & Nakanishi, S. Role of thelarge extracellular domain of metabotropic glutamate receptors in agonistselectivity determination. J. Biol. Chem. 268, 19341–19345 (1993).

26. Kunishima, N. et al. Structural basis of glutamate recognition by a dimericmetabotropic glutamate receptor. Nature 407, 971–977 (2000).

27. McCarthy, L. et al. A first-generation whole genome-radiation hybrid mapspanning the mouse genome. Genome Res. 7, 1153–1161 (1997).

28. Eicher, E. & Shown, E. Molecular markers that define the distal end of mouseautosomes 4, 13, and 19 and the sex chromosomes. Mamm. Genome 4,226–229 (1993).

29. Danciger, M., Farber, D., Peyser, M. & Kozak, C. The gene for the beta-subunit of retinal transducin (Gnb-1) maps to distal mouse chromosome 4,and related sequences map to mouse chromosomes 5 and 8. Genomics 6,428–435 (1990).

30. Greenfield, A., Brown, S., Friedman, J. & Bahary, N. Mapping of cloneD4Smh6b to the distal end of mouse chromosome 4. Mouse Genome 90,94–95 (1992).

31. Li, X. et al. High-resolution genetic mapping of the saccharin preferencelocus (Sac) and the putative sweet taste receptor (T1R1) gene (Gpr70) to

mouse distal Chromosome 4. Mamm. Genome 12, 13–16 (2001).32. Bai, M. Structure and function of the extracellular calcium-sensing receptor.

Intl. J. Mol. Med. 4, 115–125 (1999).33. Pelz, W., Whitney, G. & Smith, J. Genetic influences on saccharin preference

of mice. Physiol. Behav. 10, 263–265 (1973).34. Ramirez, I. & Fuller, J. Genetic influence on water and sweetened water

consumption in mice. Physiol. Behav. 16, 163–168 (1976).35. Ruiz-Avila, L. et al. Coupling of bitter receptor to phosphodiesterase through

transducin in taste receptor cells. Nature 376, 80–85 (1995).36. Ming, D., Ruiz-Avila, L. & Margolskee, R. Characterization and

solubilization of bitter-responsive receptors that couple to gustducin. Proc.Natl. Acad. Sci. USA 95, 8933–8938 (1998).

37. Marshall, F., Jones, K., Kaupmann, K. & Bettler, B. GABAB receptors—thefirst 7TM heterodimers. Trends Pharmacol. Sci. 20, 396–399 (1999).

38. Bernhardt, S., Naim, M., Zehavi, U. & Lindemann, B. Changes in IP3 andcytosolic Ca2+ in response to sugars and non-sugar sweeteners intransduction of sweet taste in the rat. J. Physiol. (Lond.) 490, 325–336 (1996).

39. Gimelbrant, A., Haley, S. & McClintock, T. Olfactory receptor traffickinginvolves conserved regulatory steps. J. Biol. Chem. 276, 7285–7290 (2001).

40. Faurion, A., Saito, S. & Mac Leod, P. Sweet taste involves several distinctreceptor mechanisms. Chem. Senses 5, 107–121 (1980).

41. Glaser, D., Tinti, J.-M. & Nofre, C. Evolution of the sweetness receptor inprimates. I. Why does alitame taste sweet in all prosimians and simians, andaspartame only in old world simians? Chem. Senses 20, 573–584 (1995).

42. Nofre, C., Tinti, J.-M. & Glaser, D. Evolution of the sweetness receptor inprimates. II. Gustatory responses of non-human primates to ninecompounds known to be sweet in man. Chem. Senses 21, 747–762 (1996).

43. Froloff, N., Faurion, A. & Mac Leod, P. Multiple human taste receptor sites: amolecular modeling approach. Chem. Senses 21, 425–445 (1996).

44. Nimomiya, Y., Inoue, M., Imoto, T. & Nakashima, K. Lack of gurmarinsensitivity of sweet taste receptors innervated by the glossopharyngeal nervein C57BL mice. Am. J. Physiol. 272, R1002–R1006 (1997).

45. Nimomiya, Y., Imoto, T. & Sugimura, T. Sweet taste response of mousechorda tympani neurons: Existence of gurmarin-sensitive and -insensitivereceptor components. J. Neurophysiol. 81, 3087–3091 (1999).

46. Schiffman, S., Cahn, H. & Lindley, M. Multiple receptor sites mediatesweetness: evidence from cross adaptation. Pharmacol. Biochem. Behav. 15,377–388 (1981).

47. Froloff, N., Lloret, E., Martinez, J.-M. & Faurion, A. Cross-adaptation andmolecular modeling study of receptor mechanisms common to four tastestimuli in humans. Chem. Senses 23, 197–206 (1998).

48. Tonosaki, K. & Funakoshi, M. Cross-adapted sugar responses in the mousetaste cell. Comp. Biochem. Physiol. 92A, 181–183 (1989).

49. Berghard, A. & Buck, L. Sensory transduction in vomeronasal neurons:evidence for G alpha o, G alpha i2, and adenylyl cyclase II as major componentsof a pheromone signaling cascade. J. Neurosci. 16, 909–918 (1996).

50. Malnic, B., Hirono, J., Sato, T. & Buck, L. B. Combinatorial receptor codes forodors. Cell 96, 713–723 (1999).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Axon pathfinding during early stages of CNS development isessential to establish the fundamental connectivity of the brain.Mechanisms of axon pathfinding include receptor-mediatedrecognition of attractive and repulsive cues in the environment,which induce changes in the direction of the axonal growth cone.Alterations in growth cone behavior in response to environmentalcues require cytoskeletal rearrangements; however, the cellularcomponents that transduce signals from cell surface receptors tothe cytoskeleton remain elusive1,2. Receptor-mediated regulationof growth cone motility requires a complex of proteins includ-ing transmembrane or membrane-bound receptors, signalingproteins, scaffold proteins and cytoskeletal components1–3. Mul-tivalent binding proteins can form scaffolds to anchor proteinsand regulate their function within multiprotein signaling com-plexes4–6. Despite considerable information concerning both therole of intracellular calcium7–10 and cytoskeletal dynamics3,11 inthese events, little is known of the identity or function of scaf-fold proteins controlling growth cone behavior.

The Homer family of proteins12 (also called vesl (ref. 13),CPG22 (ref. 14), PSD-zip45 (ref. 15) and Cupidin16) have beenpostulated to act as cytosolic scaffold proteins, possibly linkingcell surface transmembrane receptors and intracellular calciumstores17–19 or regulating transmembrane receptor trafficking andanchoring20–24. Homer proteins are derived from three genes inmammals and one gene in Drosophila17. Each gene in mammalscan generate alternately spliced isoforms. Homer proteins arehighly conserved across species and are characterized by anenabled VASP homology (EVH1) domain encoded by the N-ter-minal 112 amino acids of all Homer proteins. One form ofHomer, Homer1a, is induced in response to neuronal activity inrodents12,13,15 and Xenopus (unpublished data). Homer1aencodes a truncated form of the other Homer isoforms, consist-

ing of the EVH1 domain. The expression of all other forms ofHomer is not affected by activity13,17. The constitutively expressedHomer proteins contain the EVH1 domain and a C-terminalcoiled-coil (CC) domain4,17,18,23, and will be called the long formsof Homer. The long forms of Homer multimerize through asso-ciation of their CC domains. The Homer dimers then have thecapacity to crosslink multiple interacting proteins through theirligand-binding EVH1 domains. The EVH1 domain of Homer1abinds to the same proteins as the long forms of Homer; howev-er, Homer1a cannot crosslink associated proteins because it lacksthe CC domain.

Homer can associate with a variety of proteins, includingtransmembrane receptors such as class 1 metabotropic glutamatereceptors (mGluRs), inositol triphosphate receptors (IP3R), ryan-odine receptors (RyR)12,18,23 and other anchoring proteins suchas Shank19. All the known protein–protein interactions betweenHomer proteins and their binding partners are due to the EVH1domain binding a conserved proline-rich motif in the ligands25.Although Homer proteins are characterized by their EVH1domain, Homer does not bind ligands of other EVH1 proteins,such as Ena, VASP or WASP, because of a unique structure of thebinding pocket in the Homer EVH1 domain25.

We tested whether Homer proteins are involved in axonpathfinding via their ability to crosslink interacting proteins thatgovern growth cone behaviors. We expressed both long and shortforms of Homer, as well as mutant Homer proteins that do notbind ligands. We assessed the effect of expression of the differ-ent Homer proteins on axon pathfinding in vivo by collectingtime-lapse images of optic tectal cell axonal projections in Xeno-pus tadpoles over periods of 3–5 days. The data indicate that full-length Homer proteins are essential for axon pathfinding in vivo.The data support the hypothesis that Homer proteins act as scaf-

articles

The scaffold protein, Homer1b/c,regulates axon pathfinding in thecentral nervous system in vivo

Lisa Foa1, Indrani Rajan1, Kurt Haas1, Gang-Yi Wu1, Paul Brakeman2, Paul Worley2 and Hollis Cline1

1 Cold Spring Harbor Laboratory, Beckman Building, 1 Bungtown Road, Cold Spring Harbor, New York 11724, USA2 Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, Maryland 21205, USA

The first two authors contributed equally to this work

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

Homer proteins are a family of multidomain cytosolic proteins that have been postulated to serve asscaffold proteins that affect responses to extracellular signals by regulating protein–protein interactions.We tested whether Homer proteins are involved in axon pathfinding in vivo, by expressing both wild-type and mutant isoforms of Homer in Xenopus optic tectal neurons. Time-lapse imaging demonstratedthat interfering with the ability of endogenous Homer to form protein–protein interactions resulted inaxon pathfinding errors at stereotypical choice points. These data demonstrate a function for scaffoldproteins such as Homer in axon guidance. Homer may facilitate signal transduction from cell-surfacereceptors to intracellular proteins that govern the establishment of axon trajectories.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


expressing Homer1a would interfere with protein interactionsrequired for signaling events governing axon guidance. Weexpressed Homer1a in optic tectal cells using either viral genetransfer or targeted electroporation. We injected vaccinia virus(VV) encoding Homer1a (Homer1a-VV) into the brain ventricleof stage 44/45 Xenopus tadpoles. Control stage-matched tadpoleswere either uninfected or infected with a virus expressing thereporter β-galactosidase (β-gal-VV). Homer1a-VV-infected ani-mals have higher Homer1a expression in tectal neurons com-pared to uninfected control animals (Fig. 2b) or animals infectedwith β-gal-VV (data not shown).

The axon trajectories of optic tectal neurons in VV-infected ani-mals were determined by imaging localized clusters of DiI-labeledneurons over three days, starting two days after virus injection. Ani-mals infected with β-gal-VV had axonal projections comparable tothose of uninfected control animals (Fig. 3a and b). Labeling a largepopulation of neurons in one tectal lobe revealed an occasionalaxon crossing the dorsal midline in 28.5% of uninfected and 29%of β-gal-VV-infected animals (Figs. 3a, b and 6a). Axons that crossthe dorsal midline in either uninfected animals or β-gal-VV-infect-ed animals were not observed in images taken 24 hours later (Fig. 3aand b), indicating that these aberrant projections were transient.

Animals infected with Homer1a-VV show persistent aberrantaxon projections to the contralateral optic tectum (Fig. 3c). In theexample in Fig. 3, a cluster of labeled tectal neurons (lower rightcorner of the image) projects a bundle of axons across the tectalmidline on the first day of imaging. These neurons are locatedcaudally in the optic tectum and demonstrate the midline cross-ing error seen in neurons that express Homer1a when they arefirst extending their axons. More rostrally positioned neurons,which are more mature, had already projected axons past the mid-line choice point when they were labeled with dye. These axons

fold linking cell surface receptors to intracellular signaling path-ways controlling axon pathfinding.

RESULTSTectal cell axonal projectionsWe used biocytin and DiI labeling of large clusters of optic tectalneurons to determine the normal axonal projection pattern. Ros-trally projecting tectal axons navigate six choice points along astereotyped pathway to their target in the contralateral tegmen-tum in the floor of the midbrain (Fig. 1). In Fig. 3 and subse-quent figures, axons are labeled by the number of the choice pointcorresponding to the pathfinding error they make.

Expression of Homer isoforms in optic tectumWe cloned the Xenopus homolog of Homer1 from a tadpole brainlibrary. The deduced amino acid sequence of the open readingframe of Xenopus Homer is highly homologous to murineHomer17. The N-terminal half of the Xenopus protein, includingthe EVH1 domain, through which Homer binds to other pro-teins, is 98% identical to the N-terminal of human Homer1.

Homer1b/c is expressed in Xenopus tadpole brain and is dis-tributed throughout the optic tectum, including within newlydifferentiated neurons close to the caudomedial proliferative zoneof the tectum (Fig. 2b and c). Neurons in this position within thetectum are in the process of extending their axons out of the tec-tum26. Homer1b/c is located in a punctate pattern in tectal axon-al growth cones (Fig. 2d and e). Homer1a is barely detectable inbrain homogenates (Fig. 2b).

Homer regulates tectal cell axonal pathfindingWe investigated the function of Homer proteins in axon pathfind-ing of the rostrally projecting tectal neurons. We predicted that

Fig. 2. Homer isoforms are differentially expressed in Xenopus brain. (a) Homer1 isoforms. (b) Western blots showing Homer expression in tis-sue homogenates using two Homer antibodies. The Homer1b/c antibody recognizes the coiled-coil domain of Homer1b/c. The Homer1a/b/c

antibody recognizes all forms of Homer1. Homer1a expression isincreased after infection with Homer1a-VV compared to animalsinjected with PBS. Overexposure of the blot (PBS overexposure)shows Homer1b/c and very low levels of endogenous Homer1a.(c) Confocal image of the tadpole midbrain labeled with theHomer 1b/c antibody. Homer1b/c is concentrated in tectal cellbodies and the tectal neuropil. Rostral is up. (d, e) Homer1b/c isdistributed in a granular pattern in tectal cell axonal growthcones (arrows, in fluorescence (d) and DIC (e) images). Scalebars (c), 100 µm; (d, e), 10 µm. PZ, proliferative zone; CB, tectalcell body region; NP, tectal neuropil.

Fig. 1. Axon trajectory of optic tectal neurons. (a) Biocytin-labeled cluster of optic tectal neurons revealstheir projection patterns. Confocal images collectedthrough the z-axis of the midbrain region. The right half ofthe image shows a stack of confocal images collectedthrough the dorsal part of the midbrain. The left half is astack of confocal images collected through the ventral partof the midbrain. Axons navigate six choice points as theyextend to the target. (1) As soon as the axon extends fromthe soma of a newly differentiated neuron, it makes the firstchoice to grow toward lateral or medial tectum. Most tectalaxons extend laterally within the tectal neuropil, whereas afraction of the cells extend an axon toward the medial edge of the tectum. These axons avoid the dorsal midline and extend rostrally within the medialtract of the same tectal lobe. (2) Axons turn rostrally or caudally. (3) Rostrally projecting axons descend to the ventral aspect of the brain where theyconfront the third choice point, the postoptic commissure. Axons enter the commissure and cross to the contralateral side of the brain. (4) Axons exitthe commissure and turn caudally. (5) Axons leave the ventral axon fascicle and grow slightly dorsally into the tegmentum. (6) Axons recognize tegmen-tum as target and establish a complex axonal arbor. (b) Axon trajectory of a rostrally projecting tectal cell. R, rostral; C, caudal. Scale bar, 100 µm.

articles


a b

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



exit the tectum and cross to the contralateral side of the brain inthe postoptic commissure on the floor of the midbrain. On thefirst day of imaging, a single axon tipped with a growth coneentered the left tectum rostrally (#5, Fig. 3c). Images collectedthrough the z-axis of the brain show that this axon extended fromthe cluster of labeled tectal neurons in the right optic tectum, cor-rectly navigated into the postoptic commissure, but then erro-neously projected to the contralateral optic tectum, dorsal to thenormal terminal field in tegmentum. Images collected from thesame animal on the subsequent two days revealed additional axonsthat aberrantly entered the contralateral optic tectum, after mak-ing an error at choice point 5. These aberrant projections becomemore prevalent with longer times after infection with theHomer1a-VV. This likely reflects the time course of axon out-growth from the infected neurons, because it takes 2–3 days fortectal cell axons to reach the contralateral tegmentum.

To further examine the function of Homer proteins in axonpathfinding, we used targeted electroporation to confine Homer1aexpression to one tectal lobe. Electroporation of GFP fusion pro-teins of Homer1a, Homer1c and Homer1cW24A indicates thatthey are expressed throughout neurons, including throughout theaxonal growth cones (data not shown). We compared the axonpathfinding of the Homer1a-expressing neurons to that of neu-rons expressing a point mutant that does not bind Homer ligands,Homer1aW24A25. Expression of the mutant Homer1aW24A was

not expected to cause axon pathfind-ing errors, because it cannot bindHomer ligands.

Control cells expressing GFP pro-ject axons to the contralateraltegmentum, as seen with biocytinand DiI-labeled neurons (Fig. 4a–c).Electroporation of Homer1a causedthe same types of axon pathfindingerrors seen after viral expression ofHomer1a (Figs. 4d–f and 6c–f). Inaddition, the superior imaging qual-ity afforded by GFP allowed us toobserve that some axons that makean error at the dorso-ventral choicepoint remain in the lateral tract andextend into the hindbrain (Fig. 6eand f). Homer1a expression alsocauses target recognition errors:axons enter the tegmental neuropil,but fail to arborize. Instead, theyextend dorsally into the optic tectum

(#6 in Fig. 4d–f). Electroporation of the ligand-binding mutantof Homer1a, Homer1aW24A, does not cause pathfinding errors(Figs. 4g–i and 6c–e).

Homer1b/c functions in axon pathfindingHomer dimers are thought to couple two Homer binding pro-teins within the scaffold complex. Expression of Homer1a is pos-tulated to interfere with Homer function by displacingHomer1b/c from its ligands. The studies outlined above usedHomer1a expression to interfere with endogenous Homer1b/cfunction. In the following experiments, we directly testedHomer1b/c function in axon guidance by introducing wild-typeHomer1c or the mutant Homer1cW24A into a single tectal lobeby targeted electroporation. Homer1cW24A, which carries thesame point mutation as Homer1aW24A, can dimerize withendogenous Homer1b/c via the CC domain, but does not bindHomer ligands. This construct is predicted to interfere withHomer function and cause axon projection errors. Overexpres-sion of Homer1c alters calcium signaling through associatedreceptors24, which suggests that specific levels of Homer1b/c areimportant for its function. Therefore, overexpression of Homer1citself might also be predicted to affect axon pathfinding.

Overexpression of Homer1c or expression of Homer1cW24Acauses the same axon pathfinding and target recognition errors asseen with expression of Homer1a (Fig. 4j–o). Tectal cells expressing

articles

Fig. 3. Viral expression of Homer1acauses axon pathfinding errors. Time-lapse confocal images collected at 24-hintervals over 3 days showing DiI-labeledtectal cell axon projections to the con-tralateral optic tecta in representative β-gal-VV-infected (a), uninfected (b) andHomer1a-VV-infected (c) animals. Low-magnification images of the DiI injectionsite within an outline of the midbrain areshown to the left of each series. Drawingsof the confocal image stacks are shownbelow each image. Numbers 1 and 5 referto the type of axon projection errors.Scale bar, 100 µm.

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. 4. Electroporation of Homer proteins causes axonpathfinding and target recognition errors. Confocal images col-lected through the midbrain, five days after either electropora-tion with GFP alone or co-electroporation of GFP with Homerconstructs. In each case, the complete z-series was split intodorsal (pseudo-colored red) and ventral (pseudo-coloredgreen) stacks. The merged image (right) is enlarged to illustratethe complete tectal axon trajectory through the contralateraltectum and tegmentum. (a–c) Whole-brain electroporationwith GFP illustrates the normal trajectory of tectal axons.Tectal cells and dendrites are located in the dorsal tectum (a).Tectal cell axons terminating correctly in the contralateraltegmentum are marked by arrows (b, c). (d–f) Co-electropora-tion of GFP with Homer1a caused axon pathfinding errors atthe dorsal midline (1) and at the dorsoventral choice point(data not shown). Homer1a expression also caused targetrecognition errors (6). The merged image in (f) shows singleaxons changing color from green to red, indicating that theyaberrantly extended from ventral to dorsal sections. (g–i) Co-electroporation of the ligand-binding mutant Homer1aW24Aand GFP results in tectal cell axons terminating correctly in thecontralateral tegmentum (arrows). (j–l) Overexpression ofHomer1c caused target recognition errors (6), and pathfindingerrors at the dorsal midline (1) and the dorsoventral choicepoint (5). (m–o) Expression of the ligand-binding mutantHomer1cW24A caused axon pathfinding and target recogni-tion errors at positions 1, 5 and 6. In all examples, thalamicaxons that normally cross to the contralateral side of the brainvia the posterior commissure are labeled PC. The dashed whiteoutline of the brain was traced from transmitted light images.Scale bar, 100 µm.

Homer1c or Homer1cW24A extend aberrant axon projectionsacross the dorsal tectal midline (#1 in Fig. 4j, l, m and o). Axonsalso make errors at choice point 5 in the floor of the brain con-tralateral to the electroporated tectum (#5, Fig. 4j–o) and make tar-get recognition errors in the contralateral tegmentum (#6 Fig. 4j–o).These data using electroporation of Homer constructs indicate thatHomer functions cell autonomously to affect axon pathfinding.

Single-cell analysis of axon pathfinding errorsThe above experiments suggest that the pathfinding error at the dor-sal midline is made by younger neurons, which are just extendingthe axon from the cell body. The later contralateral pathfinding erroris made by more mature neurons, which had already extended anaxon past the midline choice point at the time we induced Homer1aexpression. To test this further, we imaged single tectal neurons atdifferent positions along the rostrocaudal axis of the developing tec-tum26 in control and Homer1a-VV-infected animals. Single neu-rons were labeled either with DiI or by electroporation with GFP27.Two-photon images of a single GFP-expressing neuron show a con-trol axon projection to the contralateral tegmentum (Fig. 5a–c).

We collected time-lapse images over 24 hours from a relative-ly young DiI-labeled neuron in a Homer1a-VV-infected animal(Fig. 5d–f). At the first imaging session, the axon had grown acrossthe dorsal tectal midline. By the next day, the errant projectionlooped back across the midline to the tectum from which it orig-inated. More mature neurons had already extended an axon ros-trally out of the tectum on the first day of imaging. Axons from

these neurons correctly navigated until the dorsoventralchoice point, where they erroneously projected dorsally tothe contralateral optic tectum (Fig. 5h, i, k and l).

Axons that project to the contralateral optic tectum canmake yet another error, and cross the dorsal tectal midlineto terminate in the tectal lobe containing the cell body from

which they originated (Fig. 5 j–l). This indicates that Homer1aexpression can cause axon errors at the tectal midline either earlyin the trajectory (as in the example in Fig. 5d–f) or later in the tra-jectory (Fig. 5j–l), when growth cones would never normally con-front the dorsal midline.

Quantification of axon pathfinding errorsComparable numbers of control (uninfected, β-gal-VV-infect-ed or GFP-expressing) animals had axons that projected acrossthe dorsal midline at each day of imaging (Fig. 6a and c). Thisindicates that viral infection or electroporation did not causeaxon projection errors. Significantly more (χ2, p < 0.05)Homer1a-expressing animals had axons crossing the dorsal mid-line on all three days of imaging (Fig. 6a and c). For example, inthe Homer1a-electroporated animals, the error rate was 56% onday 1 and increased to 88% on day 3, significantly greater valuesthan for the control GFP-expressing animals (χ2, p < 0.001).These values underestimate the fraction of animals with axonpathfinding errors, because animals that did not exhibit dorsalmidline errors did exhibit errors at the dorso-ventral choice point.Furthermore, when axons crossing the tectal midline wereobserved in the control animals, they were single axons per ani-mal, whereas Homer1a-expresssing animals typically had mul-tiple errant axons (Figs. 3 and 4).

Homer1cW24A seems to act as a dominant negative byincreasing the rate of dorsal midline errors. On the first day ofimaging, 50% of Homer1cW24A-expressing animals had mid-

articles


a b c

d e f

g h i

j k l

m n o

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



line errors, significantly more errors than in the GFP controls(χ2, p < 0.001, Fig. 6c). The error rate remained significantlygreater than controls throughout the imaging period. Overex-pression of Homer1c increased midline error rates from 25% to62% over the three days of imaging, also significantly greater thancontrols (χ2, p < 0.001, Fig. 6c).

Expression of both Homer1a and Homer1cW24A significantlyincreased the relative numbers of animals with aberrant projec-tions to the contralateral optic tectum (Fig. 6d). For both con-

structs, the contralateral error rate was significantly greater thancontrols on the first day of imaging, and increased further overthe next two days. Overexpression of Homer1c also led toincreased numbers of animals with axons erroneously project-ing to the contralateral tectum. Homer1aW24A did not causeerrant projections to the contralateral optic tectum.

Homer1a, Homer1c and Homer1cW24A increased the fre-quency of each of the three error types analyzed, whereasHomer1aW24A did not increase the rate of these errors (Fig. 6eand f). This analysis suggests that Homer proteins function ateach of these choice points.

Homer functions in axon error correctionAnalysis of single DiI-labeled cells in Homer1a-VV-infected ani-mals allowed a more accurate assessment of the effect of Homer1aexpression on the frequency of axons crossing the dorsal tectalmidline (Fig. 6b). In β-gal-VV-infected animals, the fraction ofDiI-labeled neurons that extended an axon across the tectal mid-line decreased from 6% to 3% over 24 hours, comparable to thevalues from uninfected animals. In contrast, the fraction of

articles

Fig. 5. Time-lapse analysis of axon projections in single tectal cells afterHomer1a-VV infection. (a, b) Two-photon images of a single tectal neu-ron expressing GFP, collected at 24 h intervals, marked day 1 and day 2.(c) Drawing of the neuron at day 2 shows the normal tectal axon trajec-tory. The dashed line represents the ventral portion of the axon. (d–l) Single DiI-labeled tectal cell axon trajectories in three Homer1a-VV-infected animals. Each row shows a rendered stack of confocal images col-lected at daily intervals and a drawing of the neuron at day 2. Tectal cells inHomer1a-VV-infected animals make errors at the dorsal tectal midline (1)and at the dorsoventral choice point (5). Scale bar, 50 µm.

a b c

f

i

l

d

g

j k

h

e

a b

c d

e f

Fig. 6. Quantitative analysis of pathfinding and target recognition errorsafter Homer expression. (a) The percentage of animals with tectal cellaxons crossing the dorsal midline was increased in Homer1a-VV-infectedanimals compared to uninfected and β-gal-VV-infected animals. Numbersof animals with midline errors on Days 1, 2 and 3 of imaging: uninfected,10/35, 9/33, 9/33; β-gal-VV, 19/65, 17/58, 12/33; Homer1a-VV, 19/39,22/39, 12/18. (b) Homer1a expression increases the percentage of singleneurons with axons crossing the dorsal midline. Numbers of neuronswith midline errors on Days 1 and 2 of imaging: uninfected, 2/25, 0/25; β-gal-VV, 3/48, 1/36; Homer1a-VV, 8/76, 14/58. (c) The percentage of ani-mals with tectal cell axons crossing the dorsal midline consistentlyincreased after Homer1a, Homer1c or Homer1cW24A electroporation.Numbers of animals with midline errors on Days 1, 2 and 3 of imaging:GFP, 3/14, 2/14, 2/14; Homer1a, 9/16, 13/16, 14/16; Homer1a W24A,1/14, 1/14, 1/14; Homer1c, 3/12, 5/12, 8/12; Homer 1c W24A, 8/16,8/16, 10/16. (d) The percentage of animals with tectal axons making con-tralateral errors (pathfinding and target recognition errors combined)consistently increased after Homer1a, Homer1c or Homer1cW24Aexpression. Numbers of animals with contralateral errors on Days 1, 2and 3 of imaging: GFP, 0/10, 4/10, 4/10; Homer1a, 3/16, 11/16, 13/16;Homer1a W24A, 0/11, 1/11, 3/11; Homer1c, 3/11, 7/11, 10/11; Homer1c W24A, 8/16, 13/16, 15/16. (e) Analysis of each of the three types ofcontralateral errors after expression of Homer1a, Homer1c orHomer1cW24A. (f) The normal projection and three types of contralat-eral errors are schematized in a sagittal view of the brain. Axons extend-ing into the hindbrain (gray) or into the optic tectum (black) committedpathfinding errors at choice point 5. Other axons entering the tectum(white) committed a target recognition error. *p < 0.05 for (a) and (b),and *p < 0.001 for (c–e) to controls compared by χ2 analysis.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


labeled neurons in Homer1a-VV-infected animals that aberrantlyextended an axon across the tectal midline increased from 11% to24% over 24 hours. Thus, viral expression of Homer1a signifi-cantly increased (χ2, p < 0.01) the frequency with which axonscross the tectal midline.

The greater proportion of Homer1a-VV-infected animals withaxons crossing the dorsal tectal midline relative to controls (Fig. 6a and b) could arise from an increased rate of axonpathfinding errors at the midline. Alternatively, the increasederror rate could arise from failure to correct aberrant axon tra-jectories. In 48 neurons from β-gal-VV-infected animals, weobserved 3 neurons (6%) with axons projecting across the tectalmidline on the first day of imaging. The following day, two ofthe three aberrant axons had been lost, but the labeled neuronsremained alive and appeared healthy. One aberrant axon per-sisted at least 24 hours after the first observation. Comparableerror correction was seen in uninfected controls (Fig. 6b). Thesedata indicate that in control animals, aberrant axon projectionscan be corrected by loss of the errant axon. In contrast, 100% ofthe axons in Homer1a-VV-infected animals that had crossed themidline on the first day of imaging were maintained to the secondday of imaging, and elaborated branches in the contralateral tec-tum. This analysis indicates that aberrant axon trajectories canbe corrected in control animals (at least within the 24-hour obser-vation period). The analysis further indicates that this correctionmechanism is lost in Homer1a-VV-infected animals.

DISCUSSIONThe establishment of connections within the brain requires thatthe axonal growth cones navigate to the target region, recognizethe target, and elaborate an axon arbor. Axon guidance dependson the detection of environmental cues, the interpretation of thecues as attractive or repulsive, and the transduction of this sig-nal to machinery governing growth cone motility. Similarly, tar-get recognition also requires detection and response to signals tostop growing. We provide evidence that the long forms of Homerhave a function in axon pathfinding and target recognition.Homer may form part of a multiple protein scaffold complex inaxonal growth cones that links guidance receptors to machinerygoverning growth cone behaviors.

We used in vivo imaging of tectal cell axonal projections todetermine the effect of expressing wild-type and ligand-bindingmutants of Homer proteins on axon pathfinding. Homer1b/c ispresent in tectal cell axonal growth cones, whereas Homer1a isundetectable in young neurons during the period of axon out-growth. Interfering with Homer function by prematurely express-ing the naturally occurring truncated isoform Homer1a causesaxon pathfinding errors that are not seen with expression of theligand-binding mutant Homer1aW24A. Expression of the lig-and-binding mutant of the long form of Homer, Homer1cW24A,causes the same pathfinding errors as Homer1a expression.Homer1a binds to the same effectors as the long forms of Homerand can act in a dominant negative fashion to interfere withHomer function18. Homer1cW24A can also act as a dominantnegative because it can sequester endogenous Homer1b/c butcannot bind Homer ligands. This would prevent the Homerdimer from linking two ligands and reduce the efficacy of sig-naling from guidance receptors to intracellular effectors. Final-ly, errors in axon pathfinding and target recognition seen withoverexpression of the long form of Homer, Homer1c, suggestthat the expression level of Homer1b/c within the growth coneis critical for signal transduction governing these events.

Homer affects pathfinding at specific choice points Interference with Homer function resulted in axon pathfindingerrors at stereotyped choice points in the trajectory, consistentwith studies in Drosophila and zebrafish, demonstrating that spe-cific constellations of transmembrane receptors and signalingmolecules govern growth cone behaviors at particular choicepoints28,29. The location of the pathfinding error at the dorsalmidline or the dorso-ventral choice point is most likely due tothe relative position of the growth cone in the trajectory at thetime of altered Homer expression. Young neurons, which make anerror at the dorsal midline, are likely in the process of extendingthe axon toward the midline at the time of exogenous Homerexpression. Consequently, they cannot process the inhibitoryguidance information at the dorsal midline, and they cross themidline instead of avoiding it.

Axons from the more mature neurons already extended pastthe midline choice point by the time exogenous Homer proteinswere expressed. It took about two days for the axon to reach thedorso-ventral choice point, where the errant axons grew into thecontralateral tectum or hindbrain. This interpretation is furthersupported by the observation that some axons that had aber-rantly projected to contralateral optic tectum made yet anotherpathfinding error at the dorsal midline and crossed into the sametectum from which their journey started. Therefore, axons requireHomer function to avoid the dorsal midline either early or latein their trajectory. This observation also indicates that the mid-line guidance cue that Homer transduces is inhibitory1, ratherthan a positive signal that maintains axons in the tract along themedial edge of the tectum. The guidance cue at the dorso-ven-tral choice point may be either positive, drawing growth conesto the tegmentum, or inhibitory, preventing the growth conefrom straying off the correct trajectory1.

Some tectal axons correctly project to the contralateraltegmentum after expression of Homer constructs, but thenextend dorsally into the tectum. Target recognition and signalsto stop growing are made by the target and detected by afferentaxons in a developmentally regulated manner30. Our data indicatethat endogenous Homer proteins transduce target recognitionand stop-growing signals.

Many molecular mechanisms cooperate to guide axons totheir targets. Nevertheless, axon pathfinding errors do occur nor-mally. Most error correction is thought to occur by cell death;however, previous studies have suggested that error correctioncan occur during axon pathfinding31. Repeated in vivo imaging ofaxonal projections allowed us to observe directly that axons domake errors in their projection pattern, and that errors can becorrected over a period of less than a day by retraction of theerrant axon by an apparently healthy neuron. Furthermore, theincrease in numbers of aberrant projections seen in animalsexpressing Homer1a partially results from a failure to correctaberrant axon projections.

Requirement for a specific level of functional HomerExpression of both wild-type Homer1c and the ligand-bindingmutant of Homer1c resulted in the same axon pathfinding andtarget recognition errors. This suggests that the cellular process-es governed by Homer require a specific level of functionalHomer protein in the growth cone. Either too little or too muchfunctional Homer protein can interfere with signaling in thegrowth cone. Other instances in which loss of function and over-expression of a protein result in the same phenotype have beenreported28,32–36. These examples include intercellular signalingmolecules, cell surface receptors and intracellular signal trans-

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



duction molecules, indicating that events governing axon guid-ance and connectivity can be finely tuned at multiple points inthe biochemical cascade. The data presented here suggest thatsignal transduction events concerning axon guidance and targetrecognition depend on the spatial and temporal distribution ofHomer protein within the growth cone.

Possible mechanisms of Homer functionWe suggest three scenarios by which Homer may regulate axonpathfinding and target recognition. In the first scenario, Homerregulates axonal pathfinding by affecting calcium signaling ingrowth cones. Intracellular calcium concentration within a growthcone controls growth cone motility both in vivo37 and in vitro7–10.A variety of environmental cues affect the spatiotemporal distrib-ution of intracellular calcium within growth cones7,10. IP3R-medi-ated calcium increases are found in growth cones, and interferencewith IP3R or RyR-mediated calcium release prevents neurite exten-sion38 and alters growth cone response to guidance cues8. Consid-erable evidence suggests that Homer affects calcium signaling bylinking transmembrane receptors to IP3R or RyR18 or by regulat-ing the efficacy of different intracellular calcium signaling path-ways24. The long forms of Homer proteins may regulate calciumrelease from intracellular stores via the IP3R or RyR in response toguidance cues signaling through transmembrane receptors. In thisscenario, expression of Homer1a or Homer1cW24A could perturbthe calcium signaling events required to recognize and respond totarget recognition signals, or to guidance cues at the dorsal midlineor the dorso-ventral choice point. Precise regulation of signalingfrom transmembrane receptors to intracellular calcium stores maybe lost in presence of excess Homer1c.

Alternatively, the axon pathfinding errors may arise from achange in the cell surface distribution of receptor molecules onthe axonal growth cones. Disrupting either the receptor traffick-ing20–22 or clustering24 function of Homer could interfere withthe recognition of guidance cues in the environment, leading topathfinding errors. A third scenario is that Homer could direct-ly regulate actin polymerization14 by transducing signals fromcell surface receptors to the actin cytoskeleton.

Scaffold proteins in the growth coneThe data presented here suggest that the long forms of Homerfunction in axon guidance, possibly linking cell surface proteinsto intracellular components required to mediate growth cone guid-ance decisions. In mature neurons, Homer is a scaffold protein inpostsynaptic sites. The quintessential multiprotein scaffold com-plex in neurons is the postsynaptic density (PSD), which includestransmembrane receptor proteins, enzymes, scaffold proteins, cal-cium regulatory proteins, adhesion molecules and cytoskeletal pro-teins6,39. Within the PSD, Homer, Shank and PSD95 form asubmembrane scaffold that can affect signaling through mGluRsand NMDA receptors19. A large degree of overlap exists in the pro-tein components in the PSD scaffold complex and in the signal-ing machinery within the growth cone. For instance,neurotransmitter receptors, adhesion molecules, protein kinases,phosphatases and cytoskeletal regulatory proteins are just some ofthe components found in both the PSD and growth cone.Although this degree of overlap may seem surprising at first glancegiven the different functions of the growth cone and the synapse,both cellular compartments do serve to detect and respond toextracellular signals. Therefore, Homer may function both in axonguidance and at the synapse, by participating in distinct signalingpathways depending on the spatiotemporal distributions of dif-ferent binding partners and downstream effectors.

METHODSVaccinia virus construction and infection. Vaccinia virus (VV) encodingthe full length open reading frame of rat Homer1a cDNA (amino acids1–186) was generated and used to infect Xenopus CNS neurons as previ-ously described40,41. All viruses also express β-galactosidase (β-gal) behinda weaker p7.5 viral promoter41. Infection with β-gal VV does not affectmorphological or synaptic development of the retinotectal system42.

Electroporation. Tectal neurons were labeled by single-cell electropo-ration27 with purified pEGFP (Clontech, Palo Alto, California). Alter-natively, targeted electroporation was used to express genes in a singleoptic tectal lobe. DNA solution (75–125 nl, 0.5 µg/µl in 2 mM CaCl2,colored with fast green) was microinjected into the brain ventricle ofanesthetized stage 44–45 tadpoles. Platinum electrodes (1–2 mm) wereplaced on the skin, on either side of the midbrain, and 3–5 pulses of 50 V with an exponential decay of τ = 70 ms were delivered. Electropo-ration did not cause an increase in cell death (data not shown). A rangeof DNA concentrations from 0.2–2.0 µg/µl yielded comparable levelsof GFP expression. Homer constructs were either electroporated alone(GFP fusion constructs) or mixed in equal proportion with pEGFP. Theconstructs used were pEGFP (Clontech), Homer1a or Homer1aW24Adriven by CMV promoter in the pRK5 vector, and GFP-Homer1a, GFP-Homer1c, or GFP-Homer1cW24A fusion proteins in the pEGFPvector. Plasmids were coded before electroporation. All data acquisi-tion and analysis were done blind to treatment.

Dye labeling. Tectal neurons were labeled with either biocytin (0.5% inPBS) or DiI (0.02% 1,1´-dioctadecyl-3,3,3´3´-tetramethylindocarbocya-nine perchlorate, DiI; Molecular Probes, Eugene, Oregon) in 100% ethanol,using 1 to 10 nA positive current applied in 3 to 10 pulses of 1 to 200 msduration43. Higher pulse frequency, intensity and duration were used tolabel clusters of neurons, whereas lower parameters were used to obtainsingle-cell labeling. DiI labels projection axons more efficiently than GFP.Animals labeled with biocytin were anesthetized 24 h later and fixed in4% paraformaldehyde, and processed to visualize the biocytin as follows.Isolated brains were incubated in blocking solution (PBS, 0.3% Triton X-100, and 5% goat serum) overnight at 4°C. Brains were then incubated inFITC-tagged streptavidin (1:200, Amersham Pharmacia Biotech, Piscat-away, New Jersey) overnight at 4°C, rinsed extensively in PBS and mount-ed in Vectashield (Vector Labs, Burlingame, California).

Image acquisition. Confocal images were collected in 2–4 µm stepsthrough the entire z-dimension of labeled neurons with a Noran Instru-ments XL laser scanning confocal attachment mounted on an uprightNikon Optiphot through a 40× Nikon oil immersion lens (1.30 NA).Each optical section was an average of 8–16 frames.

Two-photon images were collected on a custom-built instrument mod-ified from an Olympus Fluoview confocal scan box mounted on anOlympus BX50WI microscope (Olympus America, Melville, New York),with a Tsunami femtosecond-pulsed Ti: Sapphire laser (Spectra Physics,Mountain View, California). We used an Olympus LUMPlanFl/IR 40×water immersion lens at 1–2 µm steps through the entire z-dimensionof GFP-labeled neurons. Each optical section was an average of threeframes. Animals were anesthetized with 0.01% MS222 during screeningand imaging, and recovered from anesthesia between imaging sessions.For DiI-labeled specimens, the first image was obtained one to threehours after labeling. Subsequent images were obtained at 24-h intervalsover the next two days. For GFP-labeled specimens, the first image wasobtained 24 h after labeling. Subsequent images were obtained at dailyintervals over five days. GFP-filled axons could be reliably traced threedays after electroporation.

Image analysis. Image stacks were analyzed in both two- and three-dimen-sional projections using Object Image (Norbert Vischer, University of Ams-terdam, Netherlands). Montages were constructed from two-dimensionalprojections of confocal stacks in Adobe PhotoShop. To aid in the visual-ization of dorsal versus ventral localization of axonal projections, opticalsections collected through the dorsal half (0–80 µm) of the brain werepseudocolored red, whereas optical sections collected through the ventralhalf (80–200 µm) of the brain were pseudocolored green.

articles©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


Tectal explant cultures. Tectal lobes from anesthetized (0.05% MS222)stage 48 Xenopus laevis tadpoles were dissected into sterile Ca2+ and Mg2+

free HBST saline, containing 0.3 mM EDTA (5.8 mM NaCl, 0.6 mM KCl,5.0 mM HEPES, pH 7.7). All membranes were removed, including the pia,and the lobes were cut in half. Tissue pieces were rinsed twice in fresh ster-ile HBST (including 0.3 mM CaCl2·2H2O and 0.83 mM MgSO4·7H2O),then placed on coverslips coated in laminin (50 µg/ml). The explants wereleft for several hours to settle before addition of culture medium contain-ing 50% L15, 10% fetal calf serum, 25% saline (95 mM NaCl, 1 mM KCl,0.6 mM CaCl2·2H2O, 3.9 MgSO4·7H2O, 4 mM glutamine, 8.4 mM HEPES)and 5% serum (5 ng/ml sodium selenite, 5 µg/ml transferin, 5 µg/mlinsulin, 5000U penicillin streptomycin). After four days at room temper-ature, cultures were processed for immunocytochemistry.

Immunostaining and western blot analysis. Tectal explant cultures werefixed in 4% paraformaldehyde, washed first in 0.1 M phosphate buffer(PB), second in PB containing 0.3% Triton X-100 (PBX; Sigma, St. Louis, Missouri); third in PBX containing 5% normal goat serum.Cultures were incubated for 2 h at room temperature in anti-Homer1b/c antibody (1:1000 in PB containing 0.03% Triton X-100), washed inPBX and incubated for 1 h at room temperature in FITC-tagged sec-ondary antibody. For immunostaining brain sections, stage 46–47 tad-poles were fixed in 4% paraformaldehyde in PB overnight at 4°C, rinsed,cryoprotected, and cut into 20-µm horizontal cryostat sections. Sec-tions were preincubated with blocking solution containing 5% goatserum and 0.3% Triton X-100 in PB for 1 h followed by an overnightincubation at 4°C in anti-Homer1b/c antibodies diluted 1:1000 inblocking solution17. Sections were incubated in fluorescently taggedsecondary antibody for two hours at room temperature, rinsed in PBSand mounted in Vectashield. For western blots, proteins in tissuehomogenates were separated by SDS-PAGE and transferred to nitro-cellulose. Blots were incubated in Tween, followed by a 1:100 dilution ofthe anti-Homer1a/b/c or 1:1000 of anti-Homer1b/c antibody. Blotswere rinsed and incubated in 1:1000 dilution of HRP-tagged goat-antirabbit secondary antibody. The immunoreactive bands were visualizedwith the ECL chemiluminescence kit.

ACKNOWLEDGEMENTSWe thank K. Bronson, N. Dawkins and B. Burbach for technical help, E. Ruthazer,

P. O’Brien, B. Burbach and K. Svoboda for building the two-photon microscope,

E. Ruthazer for help with the two-photon image acquisition and analysis, J.C. Tu

and B. Xiao for constructs, and members of the Cline lab for discussions.

A. Demetriades helped with experiments through the Undergraduate Research

Program at CSHL. Supported by the NIH (H.T.C., K.H. and P.W.), the Eppley

Foundation (H.T.C.) and the Helen Hoffritz Fund (H.T.C.).


1. Goodman, C. S. Mechanisms and molecules that control growth coneguidance. Annu. Rev. Neurosci. 19, 341–377 (1996).

2. Hu, S. & Reichardt, L. F. From membrane to cytoskeleton: enabling aconnection. Neuron 22, 419–422 (1999).

3. Lin, C. H., Thompson, C. A. & Forscher, P. Cytoskeletal reorganizationunderlying growth cone motility. Curr. Opin. Neurobiol. 4, 640–647 (1994).

4. Fraser, I. D. & Scott, J. D. Modulation of ion channels: a “current” view ofAKAPs. Neuron 23, 423–426 (1999).

5. Klauck, T. M. et al. Coordination of three signaling enzymes by AKAP79, amammalian scaffold protein. Science 271, 1589–1592 (1996).

6. Sheng, M. & Pak, D. T. Glutamate receptor anchoring proteins and themolecular organization of excitatory synapses. Ann. NY Acad. Sci. 868,483–493 (1999).

7. Kater, S. B. & Mills, L. R. Regulation of growth cone behavior by calcium. J. Neurosci. 11, 891–899 (1991).

8. Hong, K., Nishiyama, M., Henley, J., Tessier-Lavigne, M. & Poo, M.-m.Calcium signalling in the guidance of nerve growth by netrin-1. Nature 403,93–98 (2000).

9. Zheng, J. Q. Turning of nerve growth cones induced by localized increases inintracellular calcium ions. Nature 403, 89–93 (2000).

10. Zheng, J. Q., Poo, M.-m. & Connor, J. A. Calcium and chemotropic turningof nerve growth cones. Perspect. Dev. Neurobiol. 4, 205–213 (1996).

11. Bentley, D. & O’Connor, T. P. Cytoskeletal events in growth cone steering.Curr. Opin. Neurobiol. 4, 43–48 (1994).

12. Brakeman, P. R. et al. Homer: a protein that selectively binds metabotropicglutamate receptors. Nature 386, 284–288 (1997).

13. Kato, A., Ozawa, F., Saitoh, Y., Hirai, K. & Inokuchi, K. vesl, a gene encodingVASP/Ena family related protein, is upregulated during seizure, long-termpotentiation and synaptogenesis. FEBS Lett. 412, 183–189 (1997).

14. Nedivi, E., Fieldust, S., Theill, L. & Hevroni, D. A set of genes expressed inresponse to light in the adult cerebral cortex and regulated duringdevelopment. Proc. Natl. Acad. Sci. USA 93, 2048–2053 (1996).

15. Tadokoro, S., Tachibana, T., Imanaka, T., Nishida, W. & Sobue, K.Involvement of unique leucine-zipper motif of PSD-Zip45 (Homer 1c/vesl-1L) in group 1 metabotropic glutamate receptor clustering. Proc. Natl. Acad.Sci. USA 96, 13801–13806 (1999).

16. Shiraishi, Y. et al. Cupidin, an isoform of Homer/Vesl, interacts with the actincytoskeleton and activated rho family small GTPases and is expressed indeveloping mouse cerebellar granule cells. J. Neurosci. 19, 8389–8400 (1999).

17. Xiao, B. et al. Homer regulates the association of group 1 metabotropicglutamate receptors with multivalent complexes of homer-related, synapticproteins. Neuron 21, 707–716 (1998).

18. Tu, J. C. et al. Homer binds a novel proline-rich motif and links group 1metabotropic glutamate receptors with IP3 receptors. Neuron 21, 717–726(1998).

19. Tu, J. C. et al. Coupling of mGluR/Homer and PSD-95 complexes by theShank family of postsynaptic density proteins. Neuron 23, 583–592 (1999).

20. Ciruela, F., Soloviev, M. M., Chan, W.-Y. & McIlhinney, R. A. J. Homer-1c/vesl-1L modulates the cell surface targeting of metabotropic glutamatereceptor type 1a: Evidence for an anchoring function. Mol. Cell. Neurosci. 15,36–50 (2000).

21. Ciruela, F., Soloviev, M. M. & McIlhinney, R. A. Co-expression of metabotropicglutamate receptor type 1alpha with homer-1a/Vesl-1S increases the cellsurface expression of the receptor. Biochem. J. 795–803 (1999).

22. Roche, K. W. et al. Homer 1b regulates the trafficking of group Imetabotropic glutamate receptors. J. Biol. Chem. 274, 25953–25957 (1999).

23. Kato, A. et al. Novel members of the Vesl/Homer family of PDZ proteins thatbind metabotropic glutamate receptors. J. Biol. Chem. 273, 23969–23975(1998).

24. Kammermeier, P. J., Xiao, B., Tu, J. C., Worley, P. F. & Ikeda, S. R. Homer proteinsregulate coupling of group I metabotropic glutamate receptors to N-typecalcium and M-type potassium channels. J. Neurosci. 20, 7238–7245 (2000).

25. Beneken, J. et al. Structure of the Homer EVH1 domain-peptide complexreveals a new twist in polyproline recognition. Neuron 26, 143–154 (2000).

26. Wu, G.-Y., Zou, D. J., Rajan, I. & Cline, H. T. Dendritic dynamics in vivochange during neuronal maturation. J. Neurosci. 19, 4472–4483 (1999).

27. Haas, H., Sin, W.-C., Javaherian, A., Li, Z. & Cline, H. T. Single-cellelectroporation for in vivo neuronal gene expression. Neuron 29, 571–581(2001).

28. Luo, L., Jan, L. & Jan, Y. N. Small GTPases in axon outgrowth. Perspect. Dev.Neurobiol. 4, 199–204 (1996).

29. Karlstrom, R. O., Trowe, T. & Bonhoeffer, F. Genetic analysis of axonguidance and mapping in the zebrafish. Trends Neurosci. 20, 3–8 (1997).

30. Zhang, Q. & Mason, C. A. Developmental regulation of mossy fiber afferentinteractions with target granule cells. Dev. Biol. 195, 75–87 (1998).

31. Rajan, I. & Denburg, J. L. Error correction during guidance of pioneer axonsin the leg of the cockroach embryo. Roux’s Arch. Dev. Biol. 205, 476–485(1996).

32. Gao, F.-B., Kahwi, M., Brenman, J. E., Jan, L. Y. & Jan, Y. N. Control ofdendritic field formation in Drosophila: the roles of Flamingo andcompetition between homologous neurons. Neuron 28, 91–101 (2000).

33. Krasnow, R. E., Wong, L. L. & Adler, P. N. Dishevelled is a component of thefrizzled signaling pathway in Drosophila. Development 121, 4095–4102 (1995).

34. Krasnow, R. E. & Adler, P. N. A single frizzled protein has a dual function intissue polarity. Development 120, 1883–1893 (1994).

35. Cabelli, R. J., Shelton, D. L., Segal, R. A. & Shatz, C. J. Blockade ofendogenous ligands of trkB inhibits formation of ocular dominancecolumns. Neuron 19, 63–76 (1997).

36. Cabelli, R. J., Hohn, A. & Shatz, C. J. Inhibition of ocular dominance columnformation by infusion of NT-4/5 or BDNF. Science 267, 1662–1666 (1995).

37. Gomez, T. M. & Spitzer, N. C. In vivo regulation of axon extension andpathfinding by growth-cone calcium transients. Nature 397, 350–355 (1999).

38. Takei, K., Shin, R. M., Inoue, T., Kato, K. & Mikoshiba, K. Regulation of nervegrowth mediated by inositol 1,4,5-trisphosphate receptors in growth cones.Science 282, 1705–1708 (1998).

39. Kennedy, M. B. Signal transduction molecules at the glutamatergicpostsynaptic membrane. Brain Res. Brain Res. Rev. 26, 243–257 (1998).

40. Mackett, M., Smith, G. L. & Moss, B. in DNA Cloning: A Practical Approach(ed. Glover, D. M.) 191–210 (IRL, Oxford, 1985).

41. Wu, G.-Y., Zou, D.-J., Koothan, T. & Cline, H. T. Infection of frog neuronswith Vaccinia Virus permits in vivo expression of foreign proteins. Neuron 14,681–684 (1995).

42. Cline, H. T. Dendritic arbor development and synaptogenesis. Curr. Opin.Neurobiol. 11, 118–126 (2001).

43. Cline, H. T., Edwards, J. A., Rajan, I., Wu, G. Y. & Zou, D. J. in Imaging: ALaboratory Manual (eds. Yuste, R., Lanni, F. & Konnerth, A.) 13.1–13.12(Cold Spring Harbor Laboratory Press, Cold Spring Harbor, 1999).

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



KATP channels are found in many tissues, including skeletal andsmooth muscle, heart, pancreatic β-cells, pituitary, and brain1.These channels are thought to regulate various cellular func-tions such as hormone secretion2, excitability of neurons3 andmuscles4, and cytoprotection during ischemia5,6 by coupling cellmetabolism to membrane potential1. The KATP channels in pan-creatic β-cells are critical metabolic sensors that determine glu-cose-responsive membrane excitability in the regulation ofinsulin secretion7. In the brain, KATP channels have been foundin many regions, including substantia nigra8, neocortex9, hip-pocampus10 and hypothalamus11. In the VMH, the GR neuronsincrease their firing rate in response to elevation of extracellu-lar glucose levels12,13. Although KATP channels are present in GRneurons11,14, their molecular identity and functional role areunclear. In addition, the contribution of GR neurons to systemicglucose homeostasis is not known.

The KATP channel is an octameric protein consisting of twosubunits: the pore-forming inward rectifier K+ channel mem-ber Kir6.1 or Kir6.2, and the sulfonylurea receptor SUR1 orSUR2 (SUR2A, SUR2B or possibly other SUR2 splice vari-ants)15–18. Whereas pancreatic β-cell KATP channels compriseKir6.2 and SUR1, cardiac KATP channels consist of Kir6.2 andSUR2A15–17. For different neuronal populations, all possiblecoexpression patterns of Kir6.1 or Kir6.2 and SUR1 or SUR2Ahave been reported10,19–21.

We generated Kir6.2-deficient mice (Kir6.2–/–)22, in whichboth glucose-responsive and sulfonylurea-induced insulinsecretion were defective22. During the course of our study, we

found that recovery from hypoglycemia was impaired inKir6.2–/– mice, suggesting that the secretion of counter-regu-latory hormones in response to low glucose levels is impaired.Because the hypothalamus, especially the VMH, is importantin the control of counter-regulatory hormones23–25, we inves-tigated the molecular composition and functional role of KATPchannels in the hypothalamic GR neurons of Kir6.2–/– andwild-type mice (Kir6.2+/+).

RESULTSResponse of glucagon and epinephrine to hypoglycemiaWe found that recovery from systemic hypoglycemia induced byinsulin injection was severely impaired in Kir6.2–/– mice (Fig. 1a), suggesting a deficiency in the secretion of counter-reg-ulatory hormones such as glucagon and catecholamines. We thenevaluated the counter-regulatory hormone secretion. Althoughepinephrine secretion in response to insulin-induced hypo-glycemia was similar in Kir6.2–/– and Kir6.2+/+ mice (Fig. 1b),the glucagon secretion was markedly reduced in Kir6.2–/– mice(Fig. 1c). Because KATP channels comprising Kir6.2 are presentnot only in the insulin-secreting β-cells of pancreatic islets26,27

but also in other endocrine cells of the islets including glucagon-secreting α-cells28,29, we examined glucagon secretion from iso-lated pancreatic islets. The glucagon secretion in response tochange from high (16.7 mM) to low (1 mM) glucose concentra-tion was similar in Kir6.2–/– and Kir6.2+/+ mice (Fig. 1d). In addi-tion, the glucagon response to a synthetic choline ester, carbachol(50 µM), was not impaired but was somewhat enhanced in

articles

ATP-sensitive K+ channels in thehypothalamus are essential for themaintenance of glucose homeostasis

Takashi Miki1, Birgit Liss2, Kohtaro Minami1, Tetsuya Shiuchi3, Atsunori Saraya1, YasushigeKashima1, Masatsugu Horiuchi3, Frances Ashcroft4, Yasuhiko Minokoshi3, Jochen Roeper2 andSusumu Seino1

1 Department of Cellular and Molecular Medicine, Graduate School of Medicine, Chiba University, Chiba 260-8670, Japan2 MRC Anatomical Neuropharmacology Unit, Oxford University, Mansfield Road, Oxford OX1 3TH, UK3 Department of Medical Biochemistry, Ehime University School of Medicine, Ehime 791-0295, Japan4 University Laboratory of Physiology, Oxford University, Parks Road, Oxford OX1 3TH, UK

The first two authors contributed equally to this work

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

Glucose-responsive (GR) neurons in the hypothalamus are thought to be critical in glucose homeostasis,but it is not known how they function in this context. Kir6.2 is the pore-forming subunit of KATP channelsin many cell types, including pancreatic β-cells and heart. Here we show the complete absence of bothfunctional ATP-sensitive K+ (KATP) channels and glucose responsiveness in the neurons of the ventro-medial hypothalamus (VMH) in Kir6.2–/– mice. Although pancreatic α-cells were functional in Kir6.2–/–,the mice exhibited a severe defect in glucagon secretion in response to systemic hypoglycemia. In addition, they showed a complete loss of glucagon secretion, together with reduced food intake inresponse to neuroglycopenia. Thus, our results demonstrate that KATP channels are important in glucosesensing in VMH GR neurons, and are essential for the maintenance of glucose homeostasis.©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


Fig. 1. Blood glucose levels, epinephrine andglucagon secretion in Kir6.2+/+ and Kir6.2–/– mice.(a) Changes in blood glucose levels after exoge-nous insulin injection. Human insulin (0.5 IU/kg)was injected intraperitoneally to conscious malemice. Results are expressed as percent of initialblood glucose concentration. Open circles,Kir6.2+/+ mice; filled circles, Kir6.2–/– mice (n = 7for each). (b) Epinephrine secretion induced byinsulin-induced hypoglycemia. Plasma epineph-rine levels were measured 60 min after injectionof either human insulin (1 IU/kg) or saline (n = 5for each). (c) Glucagon secretion by insulin-induced hypoglycemia. Hypoglycemia wasinduced as described in Fig. 1b. Plasma glucagonlevels were measured before and 60 min afterinsulin injection (n = 13 for each). *p < 0.0001.(d) Glucagon secretion from pancreatic islets.Glucagon secretion from the islets incubatedwith Krebs–Ringer bicarbonate buffer containing16.7 mM glucose (Glu), 1 mM glucose or 7 mMglucose plus 50 µM carbachol (Carb) is shown.The results were obtained from 3–4 independent experiments (n = 11–17). (e) Glucagon secretion in vivo by intracerebroventricular administration of2DG. Plasma glucagon levels were measured before and 15 min after the injection of 2DG (1 mg/body; n = 6). **p < 0.02. For (b–e), open columns andfilled columns represent Kir6.2+/+ and Kir6.2–/– mice, respectively. The values are means ± s.e.m. NS, not significant.

Kir6.2–/– compared to Kir6.2+/+ mice. These results suggest thatboth the glucose sensing by the α-cells and the response of theα-cells to autonomic input30 remain unaffected in Kir6.2–/–, andthat the primary defect in these mice is upstream of the α-cells.

Response of glucagon to neuroglycopeniaIn the brain, neuroglycopenia stimulates glucagon secretionthrough activation of autonomic neurons. In the hypothalamus,2-deoxyglucose (2DG) induces neuroglycopenia31, thereby stim-

ulating glucagon secretion24,32. To determine the effect of 2DGon glucagon secretion in Kir6.2–/–, 2DG was injected into the thirdventricle. The administration of 2DG produced an increase inglucagon secretion in Kir6.2+/+ but not in Kir6.2–/– mice (Fig. 1e).

Glucose responsiveness of VMH neuronsBecause the VMH possesses the highest density of GR neuronsand is involved in glucagon secretion during hypoglycemia23–25,we examined the electrophysiological properties of VMH neu-rons in in vitro brain slices from Kir6.2+/+ and Kir6.2–/– mice.We distinguished three different neuronal populations, basedon their spontaneous firing rates and their distinct subthresholdrebound behavior (Fig. 2a), as has been reported previously13.Type A neurons quickly resumed spiking at the end of hyper-polarizing current injections (Fig. 2a, top). In contrast, type Bneurons showed rebound calcium spikes (Fig. 2a, middle),whereas type C neurons displayed a prominent delay in repo-larization from hyperpolarized membrane potentials (Fig. 2a,bottom). There were no differences in the relative abundanceof these three VMH cell types in brain slices from Kir6.2–/– and

articles


0

2

4

6

8

10

Saline Insulin0

100

200

300

400

0 60Time (min)

20

40

60

80

100

0 60 120 180 240Time (min)

0

50

100

150

200

250

NS

NS

∗

NS

0

20

40

60

80

100

120

0 15

Time (min)

∗∗

NS

Blo

od g

luco

se (

% o

f bas

al)

Glu

cago

n se

cret

ion

(pg

/ 10

isle

ts /

h)

16.7

mM

Glu

1 m

M G

lu

7 m

M G

lu

+50

µM C

arb

Pla

sma

epin

ephr

ine

(ng

/ ml)

Pla

sma

gluc

agon

(ng

/ m

l)

Pla

sma

gluc

agon

(ng

/ m

l)

Type A

Type B

Type C

500 ms

25 m

V

500 ms

25 m

V

Kir6.2+/+ Kir6.2 –/–

Cell-attached

Whole-cell

0 1 2 3 4sec

Repatch same cell

pA40

–200

20

Hz86420

∗

Hz86420

pA40

–200

20

0 1 2 3 4 s

–20

20

–60

mV

pA40

–200

20

pA40

–200

20

–20

20

–60

mV

0 1 2 3 4sec

0 1 2 3 4 s

2.5 mM glucose

25 mM glucose

2.5 mM glucose

25 mM glucose

2.5 mM glucose

25 mM glucose

2.5 mM glucose

25 mM glucose

1 s

30 m

V

1 s

30 m

V

1 s

30 m

V

500 ms

25 m

V

Fig. 2. Electrophysiological properties and glucose responsiveness ofVMH neurons. (a) Three different types of VMH neurons were identi-fied by their electrophysiological properties. Representative current-clamp recordings of spontaneous activity (left) and of responses to 50pA injection of hyperpolarizing current (right) are shown. (b) VMH neu-rons of Kir6.2–/– are defective in glucose sensing. Shown are cell-attached recordings of spontaneous activity in response to an increasein extracellular glucose concentrations from 2.5 to 25 mM (top panels).After cell-attached recordings, the same neurons were repatched forwhole-cell recordings to identify the neuronal cell type (middle).Repatching of identified GR neurons in Kir6.2+/+ demonstrated that theywere either type A or type C neurons. VMH neurons of Kir6.2–/–

already exhibited spontaneous activity with a higher frequency com-pared to Kir6.2+/+ of 2.5 mM glucose, and no further increase in activitywas observed in response to increased glucose concentration (topright). Bottom, summary cell firing rates in low and high concentrationsof glucose. The values are means ± s.e.m. Twenty-four percent of VMHneurons in Kir6.2+/+ but none in Kir6.2–/– responded to increased glu-cose concentrations. *p < 0.01.

a b c

d e

a

b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Kir6.2+/+ mice, suggesting the structural integrity of the VMHin Kir6.2–/– (Kir6.2+/+, type A, 47%, n = 16/34; type B, 32%, n = 11/34; type C, 21%, n = 7/34; Kir6.2–/–, type A, 45%, n = 21/46; type B, 35%, n = 16/46; type C 20%, n = 9/46).

To examine glucose sensing by VMH neurons under physio-logical metabolic conditions, we recorded neuronal activity inthe cell-attached patch configuration at different extracellularglucose concentrations (Fig. 2b, top), and repatched the neuronsfor subsequent cell-type identification in the whole-cell mode(Fig. 2b, middle). Consistently with previous studies33,34, 24%(n = 8/33) of Kir6.2+/+ VMH neurons showed an approximatelytwofold increase in spontaneous discharge rate (1.8 ± 0.4, p < 0.001) in response to an increase in glucose from 2.5 mM(1.8 ± 0.2 Hz) to 25 mM glucose (3.2 ± 0.3 Hz, Fig. 2b, bottomleft). Repatching these glucose responsive neurons demonstratedthat they were either type A or type C neurons. In contrast, noneof the Kir6.2–/– VMH neurons, including all identified type A ortype C neurons, displayed changes in their spiking frequency inresponse to increased glucose (2.5 mM glucose, 6.0 ± 0.6 Hz; 25 mM glucose, 6.1 ± 0.7 Hz, n = 27; Fig. 2b, bottom right). Inaddition, Kir6.2–/– VMH neurons displayed higher discharge ratesat low glucose concentrations than those of Kir6.2+/+. Takentogether, these results strongly suggest that Kir6.2-containingKATP channels are essential for glucose sensing in VMH neurons.

Functional property of the KATP channel in VMH neurons To ascertain the presence of KATP channel currents in the plas-ma membrane of VMH neurons in Kir6.2+/+, we dialyzed thecells with ATP-free pipette solution (Fig. 3a). In all three typesof Kir6.2+/+ VMH neurons (n = 20), the dialysis induced ces-sation of spontaneous activity and membrane hyperpolariza-

tion (control, –40.1 ± 1.3 mV; ATP washout, –61.5 ± 1.9 mV;n = 20, p < 0.001; Fig. 3a, left). This effect was blocked by thesulfonylurea tolbutamide. By contrast, ATP washout had noeffect on the electrical properties of any type of VMH neuronsin Kir6.2–/– mice (0/22; control, –34.1 ± 1.2 mV; ATP washout,–35.8 ± 1.6 mV, n = 22; Fig. 3a, right), suggesting the absence ofthe KATP channels in VMH neurons in Kir6.2–/–. We nextrecorded KATP channel currents using the whole-cell voltage-clamp configuration (Fig. 3b). In all three types of VMH neu-rons in Kir6.2+/+, dialysis with ATP-free pipette solutionactivated time- and voltage-independent membrane currents(48.8 ± 6.9 pA at –50 mV, n = 20; 11 type A, 3 type B, and 6 type C neurons; Fig. 3b, upper left) that reversed close to EK(–93.1 ± 2.0 mV, n = 10; Fig. 3b, middle left). The currents werecompletely blocked by tolbutamide, indicating that they flowedthrough KATP channels. In contrast to Kir6.2+/+ neurons, dial-ysis with ATP-free solution did not activate currents in VMH neurons of Kir6.2–/– mice (–0.5 ± 1.6 pA, n = 20; Fig. 3b, upper and middle right), demonstrating the completeabsence of KATP channels in the plasma membrane. The IC50(8.2 µM) for tolbutamide block is consistent with the KATPchannels containing SUR1 (ref. 35; Fig. 3b, bottom right panel).

The KATP subunits in VMH neuronsThe electrophysiological data on Kir6.2–/– clearly showed Kir6.2to be the pore-forming subunit of the plasma membrane KATPchannels in VMH neurons. In addition, the sensitivity of theKATP channels to tolbutamide strongly suggests that SUR1 isthe sulfonylurea receptor subunit. To define the molecular com-position of the KATP channel directly, the expression of the KATPchannel subunit mRNA was determined in all three types of

articles

–20

20

–60

mV

0 1 2 3 4s

Control

ATP washout ATP washout

+ 100 µM tolbutamide+ 100 µM tolbutamide

Control

ATP washout

mV0 –40 –20–60–80

–20

20

–60

mV

–20

20

–60

mV

–20

20

–60

mV

–20

20

–60

mV

–20

20

–60

mV

0 1 2 3 4s

∗

mV0 –40 –20–60–80

Control

ATP washout

Kir6.2+/+ Kir6.2–/–

1.0

0.8

0.6

0.4

0.2

0.0

0.1 1 10 100 1,000

I KATP/I max

Tolbutamide (µM)

Kir6.2+/+

0 pA

200 s

20 p

A

0 pA

–120 mV

– 40 mV

0 pA

Tolbutamide

200 s

20 p

A

Kir6.2+/+ Kir6.2–/–

0 pA

10 mV

100

pA

E rev: –90 mV

Kir6.2+/+

IKATP (pA)at –50 mV

Kir6.2–/–

∗∗60

40

20

0

Washout

Washout

IC50: 8.2 µM

Control

Fig. 3. Characterization of KATP channels in VMH neurons of Kir6.2+/+ and Kir6.2–/–. (a) Standard whole-cell current-clamp recordings of VMH neu-rons in Kir6.2+/+ and Kir6.2–/–. Traces show spontaneous activity of membrane potential in 25 mM extracellular glucose concentration (top), after 15 min dialysis with ATP-free pipette (middle), and after ATP washout plus application of 100 µM tolbutamide (bottom). Bottom, summary of changesin membrane potential after ATP washout in Kir6.2+/+ and Kir6.2–/–. *p < 0.001. (b) Standard whole-cell voltage-clamp recordings of KATP channel cur-rents in VMH neurons. Top, current responses at –50 mV, each plotted 10 s during dialysis with ATP-free pipette solution. Application of tolbutamideis indicated by the boxes (gray, 3 µM; white, 30 µM; black, 300 µM). Middle, current responses to voltage ramps from –120 mV to –40 mV before andafter ATP washout (Erev = reversal potential of the activated current). Bottom left, summary of outward K+ currents activated by ATP washout at –50 mV. *p < 0.0001. Bottom right, dose–response curve for tolbutamide inhibition of normalized KATP whole-cell currents activated by dialysis withATP-free pipette solution (IKATP/Imax) in VMH neurons of Kir6.2+/+ (n = 4). The values are means ± s.e.m. The line represents the fit of the mean databy a Hill equation with an IC50 of 8.2 µM and a Hill coefficient of 1.0.

a b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


VMH neurons in Kir6.2+/+, using the single-cell reverse tran-scription (RT)-multiplex polymerase chain reaction (PCR) tech-nique20,37. The electrophysiological phenotype of the VMHneuron was determined before harvesting the cytoplasm forexpression profiling (Fig. 4a). We analyzed 16 neurons (8 type A, 2 type B, and 6 type C neurons; Fig. 4b–d), all ofwhich expressed glutamate decarboxylase (GAD67) (16/16) butnot tyrosine hydroxylase (TH) mRNA, indicating the GABAer-gic phenotype. Both Kir6.2 and SUR1 were detected in all threetypes of VMH neurons examined (15/16). In contrast, noKir6.1, SUR2A or SUR2B was detected in single VMH neurons.

Food intake in response to 2DG, Leptin and NPYIn addition to stimulating glucagon secretion, 2DG alsoincreases food intake in normal mice, presumably by inhibit-ing the activity of GR neurons in the hypothalamus37.Kir6.2+/+ and Kir6.2–/– mice were injected with 2DG, and thefood intake was monitored for 3 h after injection. The incre-

ment in food intake of Kir6.2–/– mice (0.095 ± 0.037 g/3 h, n = 28) was significantly less than that of Kir6.2+/+ mice (0.279 ± 0.039 g/3 h, n = 27, p < 0.002; Fig. 5a), indicating afunctional role for hypothalamic KATP channels in the controlof food intake. Leptin, an adipocyte hormone that reducesfood intake38, inhibits hypothalamic neurons by activatingKATP channels39. Leptin inhibited food intake in Kir6.2+/+ andKir6.2–/– to a similar degree (Fig. 5b), demonstrating that theeffect of leptin on food intake is independent of Kir6.2-con-taining KATP channels. Neuropeptide Y (NPY), which increas-es food intake40, was also similarly effective in Kir6.2+/+ andKir6.2–/– (Fig. 5c).

DISCUSSIONThe hypothalamus regulates the secretion of counter-regulatoryhormones such as glucagon and catecholamines through theautonomic nervous system, and thus is critical in glucose home-ostasis30,41. A subset of hypothalamic neurons clustered in theVMH respond to elevated extracellular glucose levels42 by increas-ing their firing rates. Our findings in the present study demon-strate that KATP channels comprising Kir6.2 and SUR1 have a keyrole in this glucose-sensing process.

Kir6.2–/– mice exhibited impaired glucagon secretion duringsystemic hypoglycemia, despite the fact that glucagon secretion inresponse to a low concentration of glucose from isolated pan-creatic islets was normal. In addition, intracerebroventricularadministration of 2DG did not elicit glucagon secretion inKir6.2–/–. These in vivo findings indicate that the primary defectin glucagon secretion in Kir6.2–/– is not in the pancreatic α-cellitself but rather in its hypothalamic regulation. In support ofthese findings, in vitro brain slice experiments revealed the com-plete loss of glucose sensing in VMH associated with lack of func-tional KATP channels in the plasma membrane of Kir6.2–/– VMHneurons. These results demonstrate that Kir6.2 forms the pore

articles


SUR

2A/B

THKir6

.2SU

R1

GA

D 67

Kir6

.1

- 1000

- 500- 300

- 1000

- 500- 300

bp

500 ms

25 m

V0 mV

16

12

8

4

0

100

50

25

0

75

Num

ber

of c

ells

Percent detection

SUR

2 A/B

THKir6

.2

SUR

1

GA

D 67

Kir6

.1

Fig. 4. Single-cell RT-PCR of KATP channel subunits in VMH neurons inKir6.2+/+. (a) Whole-cell current-clamp recording of a VMH neuron. Fastrepolarization after injection of –50 pA hyperpolarizing current identifiedthis neuron as type A (compare with Fig. 2a). The cytoplasm of this neu-ron was harvested for single-cell RT-PCR analysis. (b) Agarose gel analy-sis of single-cell RT-PCR. The same neuron as shown in (a) was analyzed.PCR products corresponding to GAD67, Kir6.2 and SUR1 weredetected. No Kir6.1, SUR2A or SUR2B was detected in this cell. (c) Positive control for RT-PCR. Mouse hypothalamic cDNA was used asthe template for the positive control for KATP channel subunits, GAD andTH. The predicted sizes (bp) of the amplified PCR products were as fol-lows: GAD67, 701 bp; SUR2B, 337 bp; Kir6.2, 297 bp; SUR1, 400 bp;Kir6.1, 447 bp; TH, 376. (d) Summary of single-cell RT-PCR of VMH neu-rons (total 16 neurons; 8 type A, 2 type B, and 6 type C neurons).

a

b

c

d

Fig. 5. The effects of 2DG, leptin and NPY onfood intake in Kir6.2+/+ and Kir6.2–/– mice. (a) Effect of 2DG on food intake. Food intake for 3h after saline injection (x) was measured. Threedays later, food intake for 3 h after 2DG adminis-tration (y) was measured in the same mice. Theincrement in food intake in the same mice was cal-culated by subtracting x from y. The incrementswere as follows: Kir6.2+/+, 0.279 ± 0.039 g/3 h, n = 27; Kir6.2–/–, 0.095 ± 0.037 g/3 h, n = 28, *p < 0.002. (b) Effect of leptin on food intake. The decrement in food intake due to leptin was cal-culated similarly to (a). The cumulative food intake was measured over 24 h. (c) Effect of NPY on food intake. The increment in food intake dueto NPY was calculated similarly to (a). Open columns and filled columns represent Kir6.2+/+ and Kir6.2–/–, respectively. The data were obtainedfrom 7–30 samples in (a–c). The values are means ± s.e.m. NS, not significant.

0

0.1

0.2

0.3

0.4

0.5

0.6

Saline

Foo

d in

take

(g

/ 3h)

2DG Increment DecrementSaline Leptin

0

1

2

3

4

Foo

d in

take

(g

/ day

)

Foo

d in

take

(g

/ 3h)

Saline NPY Increment

0

0.5

1

1.5

∗ NSNS

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 the KATP channel in hypothalamic GR neurons, and that thischannel is an essential part of the neuronal glucose sensor thatcouples alterations in extracellular glucose levels to changes inneuronal excitability. We also provide evidence that the KATPchannels in the VMH regulate peripheral glucagon secretion inresponse to neuroglycopenia. Mutation of Kir6.2 or SUR1 caus-es familial persistent hyperinsulinemic hypoglycemia of infancy(PHHI)16. In addition to the unregulated insulin secretion, thepresent findings suggest that impaired glucagon secretion due toKATP channel dysfunction in the hypothalamus might also con-tribute to the prolonged hypoglycemia in these patients.

The molecular composition of KATP channels in GR VMHneurons is controversial21,43–45. Our results show that all threetypes of VMH neurons express both Kir6.2 and SUR1, but thatonly a subset of these neurons are glucose responsive. Thus, coex-pression of Kir6.2 and SUR1 is necessary but not sufficient forhypothalamic glucose sensing. It seems likely that the differentglucose sensitivities of VMH neurons result from differentialexpression of genes involved in glucose metabolism. Analogouslyto the pancreatic β-cell, these might include the low-affinity glu-cose transporter GLUT246 and the glucose-phosphorylatingenzyme glucokinase47. Consistent with this possibility, the expres-sion of both GLUT2 and glucokinase isoform have been report-ed in VMH neurons34,43,48.

Our results from the 2DG-induced food intake experimentsstrongly suggest that the KATP channels in the hypothalamus arealso involved in the control of appetite. Studies have shown thatfood intake is controlled by complex hormonal and neuronal sig-naling pathways49. The adipocyte hormone leptin, one of the keysignals of satiety, has been reported to activate hypothalamic KATPchannels acutely in vitro39. However, it is not known if KATP chan-nels are important in mediating the anorexic effect of leptin. Weshow here that the effects of both leptin and the orexic peptideNPY on food intake are independent of the KATP channels inVMH neurons. Thus, the mechanism of KATP channel-mediat-ed regulation of food intake is distinct from that of leptin andNPY. Because leptin and NPY affect the long-term regulation ofbody weight49, the KATP channels in VMH neurons might beinvolved in short-term regulation of food intake, by couplingglucose levels to appetite.

The present study shows that genetic disruption of Kir6.2deprives the VMH neurons of both KATP channel activity andglucose responsiveness. In addition, activation of the KATP chan-nels in the GR neurons is critical for stimulation of both glucagonsecretion and food intake when brain glucose levels fall. On theother hand, inhibition of the pancreatic β-cell KATP channels isessential for glucose-dependent insulin secretion when bloodglucose levels rise22. Thus, hypothalamic and β-cell KATP chan-nels not only share the same molecular composition but also actin concert as central and peripheral glucose sensors to coordi-nate the maintenance of glucose homeostasis.

METHODSMice. Mice lacking Kir6.2 were generated as described22. The mutantmice and wild-type mice were backcrossed for five generations to aC57Bl/6 background. All the in vivo experiments began at around8:00–10:00 a.m. All animal procedures were approved by the Chiba Uni-versity and Oxford University Animal Care Committees.

Glucagon and catecholamine secretion in vivo. Male mice 12–20 weeksold were deprived of food for 16 h before the experiments. Blood sampleswere drawn from the orbital sinus. Glucagon secretion induced byinsulin-induced hypoglycemia was examined before and 60 min after the

intraperitoneal injection of human insulin (1 U/kg). Glucagon secretioninduced by 2DG was examined before and 15 min after intracere-broventricular injection of 2DG (1 mg per body) to conscious mice thathad been stereotaxically implanted with a stainless cannula in the thirdintracerebroventricle five to seven days before the experiment. Cate-cholamine was measured 60 min after intraperitoneal injection of humaninsulin (1 U/kg) or saline.

Glucagon secretion in vitro. Glucagon secretion from isolated pancreaticislets was measured using a previously designed method50, slightly modified.Briefly, pancreatic islets were isolated by the collagenase method and werehandpicked under a stereomicroscope at room temperature. The freshlyisolated islets were preincubated at 37°C for 60 min in Krebs–Ringer bicarbonate buffer, pH 7.4, supplemented with 10 mM HEPES, 0.1% bovineserum albumin and 7 mM glucose. The islets (10 in each tube) were incu-bated for 60 min in 500 µl of the Krebs–Ringer bicarbonate buffer con-taining 7 mM glucose plus 50 µM carbachol, 16.7 mM glucose or 1 mMglucose in the presence of 500 KIU/ml of aprotinin. Immediately after incu-bation, aliquots (350 µl) of the medium were removed for assay of glucagon.

Measurements of blood glucose, glucagon and catecholamine. Bloodglucose levels were measured as described22. The glucagon levels in plas-ma or perfusate were measured by RIA kit from LINCO (Linco, Mis-souri). The plasma catecholamine levels were measured by HPLC withfluorescence detection (Tosoh, Tokyo, Japan).

Electrophysiology. Coronal brain slices (250 µm) containing theVMH were prepared from 13–16-day-old Kir6.2–/– and Kir6.2+/+ mice,and electrophysiological brain slice patch-clamp recordings were doneusing DIC-IR optics, as previously described20. Briefly, after at least 30 min recovery, midbrain slices were transferred for patch-clamprecordings into a chamber continuously perfused with artificial cere-brospinal fluid (ACSF) at 2–4 ml/min with either high (25 mM glu-cose) or low glucose (2.5 mM glucose plus 22.5 mM sucrose). ACSFwas bubbled with a mixture of 95% O2 and 5% CO2 at 37°C. Forwhole-cell recordings, patch pipettes (1–2.5 MΩ) were pulled fromthin-walled borosilicate glass (Clark, Reading, UK) and filled withinternal solution containing 120 mM K-gluconate, 20 mM KCl, 10 mM HEPES, 10 mM EGTA and 2 mM MgCl2, pH 7.3(290–300 mOsm). For cell-attached recordings, patch pipettes

(5–6 MΩ) pulled from borosilicate glass (Clark) were filled with anexternal solution containing 150 mM NaCl, 2.5 mM KCl, 5 mMHEPES, 2 mM MgCl2 and 2 mM CaCl2, pH 7.4 (290–300 mOsm).The Pulse/Pulsefit (Heka, Lambrecht, Germany) program was usedfor data acquisition and Igor (Wavemetrics, Lake Oswego, Oregon)for analysis. Recordings were digitized at 2–5 kHz and filtered witha low-pass filter Bessel characteristic of 1 kHz cutoff frequency.

Multiplex and nested single-cell RT-PCR. Single-cell RT-multiplex PCRexperiments, including primer sequences and PCR protocols, were done aspreviously described20. Briefly, after reverse transcription, the cDNAs forthe KATP channel subunits (Kir6.1, Kir6.2, SUR1, SUR2A and SUR2B),tyrosine hydroxylase (TH), and the 67 kD form of glutamate decarboxylase(GAD67) were simultaneously amplified in a first multiplex PCR, followedby a second amplification in individual nested PCR reactions. Aliquots(15 µl) of PCR products were separated and visualized in an ethidium-bromide-stained agarose gel (2%) by electrophoresis (predicted sizes ofthe PCR-generated fragments, GAD67, 702 bp; SUR2A, 513 bp; SUR2B,337 bp; Kir6.2, 297 bp; SUR1, 400 bp; Kir6.1, 447 bp; TH, 377 bp). Allindividual PCR products were verified by direct sequencing. Mouse hypo-thalamic cDNA (< 10 fmol) was used as template for positive controls.

Measurements of food intake. Mice eating ad libitum were injected with2DG (500 mg/kg), mouse leptin (4 µg) or rat NPY (2 µg) at 9:00 a.m.As a control, vehicle (saline) was injected. 2DG was administeredintraperitoneally; leptin and NPY were administered intracerebroven-tricularly. Accumulative food intake was measured for 3 h (2DG andNPY) or 24 h (leptin) after injection. The food used was normal mousechow, CE-2 (Japan Clea, Tokyo, Japan).

articles©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


ACKNOWLEDGEMENTSThis work was supported by Grants-in-Aid for Creative Basic Research

(10NP0201) and for Scientific Research from the Ministry of Education, Science,

Sports and Culture, Japan; by Research Grants from the Ministry of Health and

Welfare, Japan, by grants from Novo Nordisk Pharma, the Yamanouchi

Foundation for Research on Metabolic Disorders, the Wellcome Trust; and the

Medical Research Council. J.R. is supported by the Monsanto Senior Research

Fellowship; B.L., by the Todd-Bird Junior Research Fellowship and the Blaschko

Visiting Research Scholarship.

RECEIVED 12 FEBRUARY; ACCEPTED 9 MARCH 2001

1. Ashcroft, F. M. Adenosine 5´-triphosphate-sensitive potassium channels.Annu. Rev. Neurosci. 11, 97–118 (1988).

2. Bernardi, H. et al. ATP-modulated K+ channels sensitive to antidiabeticsulfonylureas are present in adenohypophysis and are involved in growthhormone release. Proc. Natl. Acad. Sci. USA 90, 1340–1344 (1993).

3. Ashford, M. L., Sturgess, N. C., Trout, N. J., Gardner, N. J. & Hales, C. N.Adenosine-5′-triphosphate-sensitive ion channels in neonatal rat culturedcentral neurones. Pflugers Arch. 412, 297–304 (1988).

4. Spruce, A. E., Standen, N. B. & Stanfield, P. R. Voltage-dependent ATP-sensitive potassium channels of skeletal muscle membrane. Nature 316,736–738 (1985).

5. Amoroso, S., Schmid-Antomarchi, H., Fosset, M. & Lazdunski, M. Glucose,sulfonylureas, and neurotransmitter release: role of ATP-sensitive K+

channels. Science 247, 852–854 (1990).6. Terzic, A., Jahangir, A. & Kurachi, Y. Cardiac ATP-sensitive K+ channels:

regulation by intracellular nucleotides and K+ channel-opening drugs. Am. J.Physiol. 269, C525–545 (1995).

7. Cook, D. L. & Hales, C. N. Intracellular ATP directly blocks K+ channels inpancreatic B-cells. Nature 311, 271–273 (1984).

8. Roper, J. & Ashcroft, F. M. Metabolic inhibition and low internal ATP activateK-ATP channels in rat dopaminergic substantia nigra neurones. PflugersArch. 430, 44–54 (1995).

9. Ohno-Shosaku, T. & Yamamoto, C. Identification of an ATP-sensitive K+

channel in rat cultured cortical neurons. Pflugers Arch. 422, 260–266 (1992).10. Zawar, C., Plant, T. D., Schirra, C., Konnerth, A. & Neumcke, B. Cell-type

specific expression of ATP-sensitive potassium channels in the rathippocampus. J. Physiol. (Lond.) 514, 327–341 (1999).

11. Ashford, M. L., Boden, P. R. & Treherne, J. M. Glucose-induced excitation ofhypothalamic neurones is mediated by ATP-sensitive K+ channels. PflugersArch. 415, 479–483 (1990).

12. Oomura, Y., Ono, T., Ooyama, H. & Wayner, M. J. Glucose and osmosensitiveneurones of the rat hypothalamus. Nature 222, 282–284 (1969).

13. Minami, T., Oomura, Y. & Sugimori, M. Electrophysiological properties andglucose responsiveness of guinea-pig ventromedial hypothalamic neurones invitro. J. Physiol. (Lond.) 380, 127–143 (1986).

14. Ashford, M. L., Boden, P. R. & Treherne, J. M. Tolbutamide excites ratglucoreceptive ventromedial hypothalamic neurones by indirect inhibition ofATP-K+ channels. Br. J. Pharmacol. 101, 531–540 (1990).

15. Ashcroft, F. M. & Gribble, F. M. Correlating structure and function in ATP-sensitive K+ channels. Trends Neurosci. 21, 288–294 (1998).

16. Aguilar-Bryan, L. & Bryan, J. Molecular biology of adenosine triphosphate-sensitive potassium channels. Endocr. Rev. 20, 101–135 (1999).

17. Seino, S. ATP-sensitive potassium channels: a model of heteromultimericpotassium channel/receptor assemblies. Annu. Rev. Physiol. 61, 337–362 (1999).

18. Chutkow, W. A., Makielski, J. C., Nelson, D. J., Burant, C. F. & Fan, Z.Alternative splicing of sur2 exon 17 regulates nucleotide sensitivity of theATP-sensitive potassium channel. J. Biol. Chem. 274, 13656–13665 (1999).

19. Karschin, A., Brockhaus, J. & Ballanyi, K. KATP channel formation by thesulphonylurea receptors SUR1 with Kir6.2 subunits in rat dorsal vagalneurons in situ. J. Physiol. (Lond.) 509, 339–346 (1998).

20. Liss, B., Bruns, R. & Roeper, J. Alternative sulfonylurea receptor expressiondefines metabolic sensitivity of K-ATP channels in dopaminergic midbrainneurons. EMBO J. 18, 833–846 (1999).

21. Lee, K., Dixon, A. K., Richardson, P. J. & Pinnock, R. D. Glucose-receptiveneurones in the rat ventromedial hypothalamus express KATP channelscomposed of Kir6.1 and SUR1 subunits. J. Physiol. (Lond.) 515, 439–452(1999).

22. Miki, T. et al. Defective insulin secretion and enhanced insulin action in KATPchannel-deficient mice. Proc. Natl. Acad. Sci. USA 95, 10402–10406 (1998).

23. Borg, W. P. et al. Ventromedial hypothalamic lesions in rats suppress

counterregulatory responses to hypoglycemia. J. Clin. Invest. 93, 1677–1682(1994).

24. Borg, W. P., Sherwin, R. S., During, M. J., Borg, M. A. & Shulman, G. I. Localventromedial hypothalamus glucopenia triggers counterregulatory hormonerelease. Diabetes 44, 180–184 (1995).

25. Borg, M. A., Sherwin, R. S., Borg, W. P., Tamborlane, W. V. & Shulman, G. I.Local ventromedial hypothalamus glucose perfusion blockscounterregulation during systemic hypoglycemia in awake rats. J. Clin. Invest.99, 361–365 (1997).

26. Inagaki, N., et al. Reconstitution of IKATP: an inward rectifier subunit plusthe sulfonylurea receptor. Science 270, 1166–1170 (1995).

27. Sakura, H., Ammala, C., Smith, P. A., Gribble, F. M., Ashcroft, F. M. Cloningand functional expression of the cDNA encoding a novel ATP-sensitivepotassium channel subunit expressed in pancreatic beta-cells, brain, heartand skeletal muscle. FEBS Lett. 377, 338–344 (1995).

28. Bokvist, K. et al. Characterisation of sulphonylurea and ATP-regulated K+

channels in rat pancreatic A-cells. Pflugers Arch. 438, 428–436 (1999).29. Suzuki, M., Fujikura, K., Inagaki, N., Seino, S. & Takata, K. Localization of the

ATP-sensitive K+ channel subunit Kir6.2 in mouse pancreas. Diabetes 46,1440–1444 (1997).

30. Taborsky, G. J. Jr., Ahren, B. & Havel, P. J. Autonomic mediation of glucagonsecretion during hypoglycemia: implications for impaired alpha-cellresponses in type 1 diabetes. Diabetes 47, 995–1005 (1998).

31. Muller, E. E., Cocchi, D. & Forni, A. A central site for the hyperglycemicaction of 2-deoxy-d-glucose in mouse and rat. Life Sci. 10, 1057–1067 (1971).

32. Borg, M. A. et al. Chronic hypoglycemia and diabetes impaircounterregulation induced by localized 2-deoxy-glucose perfusion of theventromedial hypothalamus in rats. Diabetes 48, 584–587 (1999).

33. Silver, I. A. & Erecinska, M. Glucose-induced intracellular ion changes insugar-sensitive hypothalamic neurons. J. Neurophysiol. 79, 1733–1745(1998).

34. Yang, X. J., Kow, L. M., Funabashi, T. & Mobbs, C. V. Hypothalamic glucosesensor: similarities to and differences from pancreatic beta-cell mechanisms.Diabetes 48, 1763–1772 (1999).

35. Gribble, F. M., Tucker, S. J., Seino, S. & Ashcroft, F. M. Tissue specificity ofsulfonylureas: studies on cloned cardiac and beta-cell K(ATP) channels.Diabetes 47, 1412–1418 (1998).

36. Bergen, H. T., Monkman, N. & Mobbs, C. V. Injection with gold thioglucoseimpairs sensitivity to glucose: evidence that glucose-responsive neurons areimportant for long-term regulation of body weight. Brain Res. 734, 332–336(1996).

37. Lambolez, B., Audinat, E., Bochet, P., Crepel, F. & Rossier, J. AMPA receptorsubunits expressed by single Purkinje cells. Neuron 9, 247–258 (1992).

38. Elmquist, J. K., Elias, C. F. & Saper, C. B. From lesions to leptin: hypothalamiccontrol of food intake and body weight. Neuron 22, 221–232 (1999).

39. Spanswick, D., Smith, M. A., Groppi, V. E., Logan, S. D. & Ashford, M. L.Leptin inhibits hypothalamic neurons by activation of ATP-sensitivepotassium channels. Nature 390, 521–525 (1997).

40. Stephens, T. W. et al. The role of neuropeptide Y in the antiobesity action ofthe obese gene product. Nature 377, 530–532 (1995).

41. Steffens, A. B., Strubbe, J. H., Balkan, B. & Scheurink, J. W. Neuroendocrinemechanisms involved in regulation of body weight, food intake andmetabolism. Neurosci. Biobehav. Rev. 14, 305–313 (1990).

42. Levin, B. E., Dunn-Meynell, A. A. & Routh, V. H. Brain glucose sensing andbody energy homeostasis: role in obesity and diabetes. Am. J. Physiol. 276,R1223–1231 (1999).

43. Lynch, R. M., Tompkins, L. S., Brooks, H. L., Dunn-Meynell, A. A. & Levin, B. E. Localization of glucokinase gene expression in the rat brain. Diabetes 49,693–700 (2000).

44. Karschin, C., Ecke, C., Ashcroft, F. M. & Karschin, A. Overlappingdistribution of K(ATP) channel-forming Kir6.2 subunit and the sulfonylureareceptor SUR1 in rodent brain. FEBS Lett. 401, 59–64 (1997).

45. Dunn-Meynell, A. A., Rawson, N. E. & Levin, B. E. Distribution andphenotype of neurons containing the ATP-sensitive K+ channel in rat brain.Brain Res. 814, 41–54 (1998).

46. Pessin, J. E. & Bell, G. I. Mammalian facilitative glucose transporter family:structure and molecular regulation. Annu. Rev. Physiol. 54, 911–930 (1992).

47. Matschinsky, F. M., Glaser, B., Magnuson, M. A. Pancreatic beta-cellglucokinase: closing the gap between theoretical concepts and experimentalrealities. Diabetes 47, 307–315 (1998).

48. Leloup, C. et al. Glucose transporter 2 (GLUT 2): expression in specific brainnuclei. Brain Res. 638, 221–226 (1994).

49. Schwartz, M. W., Woods, S. C., Porte, D. Jr., Seeley, R. J. & Baskin, D. G.Central nervous system control of food intake. Nature 404, 661–671 (2000).

50. Salehi, A., Chen, D., H. Kanson, R., Nordin, G. & Lundquist, I. Gastrectomyinduces impaired insulin and glucagon secretion: evidence for a gastro-insular axis in mice. J. Physiol. (Lond.) 514, 579–591 (1999).

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Motion is characterized by a series of positions an object traversesover time. An object presented sequentially at two positions, with atime delay between them, is sufficient for us to perceive motion1.Likewise, depth can be signaled by two positions, one in each eye;a difference in position between images in left and right eyes (binoc-ular disparity) is sufficient to produce our perception of depth2.Therefore, the extraction of motion and depth from retinal imagesinvolves similar processes that could use a common encoding mech-anism. Because the striate cortex is the first site along the centralvisual pathways at which neural selectivities for direction of motionand binocular disparity arise, neurons in this area may be involvedin joint-encoding of motion and depth. However, the encoding ofmotion and that of binocular disparity have been studied sepa-rately, and the issue of joint-encoding has not been addressed.

Here we establish that single neurons in the striate cortexencode motion and binocular disparity jointly. This joint-encod-ing is of direct relevance to a compelling and intriguing visual illu-sion, the Pulfrich effect, which is described below. For this reason,we first examine the neural basis of the Pulfrich effect to illustratethe joint-encoding. We then determine a relationship between thetuning properties of individual neurons for motion and binoculardisparity. This relationship is critical for determining whetherjoint-encoding has an advantage over independent-encoding.

A pendulum swinging back and forth along a plane perpen-dicular to an observer’s line of sight seems to follow an ellipticalpath in depth when seen binocularly with a light-attenuating fil-ter in front of one eye3. This phenomenon, the Pulfrich effect,has been investigated extensively in psychophysics4–6 but mini-mally in physiology7,8. It is known that the filter reduces thetransmission speed of neural signals in the filtered eye7–11, there-by creating an interocular difference in signal latency. This in

turn changes the perceived depth of the pendulum11. However,the neural mechanism that translates the interocular latency dif-ference into the depth change is a matter of debate12.

The classical explanation of the Pulfrich effect is as follows3.Because the neural signal from the filtered eye reaches the visu-al cortex with a slight delay compared to the signal from the unfil-tered eye, the pendulum position indicated by the signal in thefiltered eye falls slightly behind that in the other eye at a givenmoment. This creates a spatial offset between the pendulum posi-tions in the two eyes (binocular disparity), and shifts the appar-ent depth of the pendulum (Fig. 1a). One possible mechanismthat underlies this process is a set of neurons tuned to variousinterocular spatial offsets specified by signals arriving simulta-neously from the two eyes (interocular spatial offset hypothesis,Fig. 2a). This explanation is essentially based on simple stimu-lus geometry (Fig. 1a), rather than a neural mechanism that con-verts a delay in one eye to a change in the perceived depth.Although it accounts for the standard Pulfrich effect, it falls shortof explaining Pulfrich-like phenomena such as those seen indynamic noise patterns containing no coherent motion13,14 andstroboscopic stimuli with no interocular spatial offset5,6,15,16.This suggests that a specific physiological process underlies thePulfrich effect and its related phenomena.

An alternative explanation is that the visual system uses thelatency difference between the neural signals originating from cor-responding points on the two retinas to compute the depth of thependulum relative to the fixation depth6,17. An object moving eitherin front of or behind the fixation plane stimulates a point on theretina in one eye first and then its corresponding point in the othereye (Fig. 1b). Therefore, the latency difference between the neuralsignals from corresponding points may indicate a moving object

articles

Joint-encoding of motion anddepth by visual cortical neurons:neural basis of the Pulfrich effect

Akiyuki Anzai1,2, Izumi Ohzawa1,3 and Ralph D. Freeman1

1 Group in Vision Science, School of Optometry, University of California, Berkeley, California 94720-2020, USA2 Present address: Department of Anatomy and Neurobiology, Washington University School of Medicine, St. Louis, Missouri 63110, USA 3 Present address: Department of Biophysical Engineering, Graduate School of Engineering Science, Osaka University, Toyonaka, Osaka,

560-8531, Japan

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

Motion and stereoscopic depth are fundamental parameters of the structural analysis of visualscenes. Because they are defined by a difference in object position, either over time or across theeyes, a common neural machinery may be used for encoding these attributes. To examine this idea,we analyzed responses of binocular complex cells in the cat striate cortex to stimuli of various intra-and interocular spatial and temporal shifts. We found that most neurons exhibit space–time-oriented response profiles in both monocular and binocular domains. This indicates that theseneurons encode motion and depth jointly, and it explains phenomena such as the Pulfrich effect. Wealso found that the relationship between neuronal tuning of motion and depth conforms to thatpredicted by the use of motion parallax as a depth cue. These results demonstrate a joint-encodingof motion and depth at an early cortical stage.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 1. The Pulfrich effect. (a) A classical explanation of the Pulfrich effect. A light-attenuating filter (gray rectangle) placed in front of one eye causes a time delay for theneural signal in that eye. Therefore, the perceived position of the swinging pendulum inthe filtered eye (gray circle) falls slightly behind that in the other eye (open circle) for agiven moment, creating a spatial offset between the perceived images in the two eyes.Left, when a pendulum swings to the right while the right eye is covered with a filter,the spatial offset corresponds to a binocular disparity that shifts the perceived depth ofthe pendulum toward the observer (hatched circle). Right, when the pendulum swingsto the left, the spatial offset corresponds to a binocular disparity that shifts the per-ceived depth of the pendulum away from the observer (hatched circle). The amount ofthe binocular disparity depends on the pendulum position along the fronto-parallelplane because the speed of the pendulum motion varies with the position. Therefore,the pendulum appears to move along an elliptical trajectory (solid arrow) in depth as itswings back and forth on the fronto-parallel plane (dashed arrow). (b) A time delaybetween stimulation of the corresponding points in the two eyes by a moving object.An object moving to the right in front of the fixation plane F (or to the left behind thefixation plane) stimulates a retinal point in the left eye (PL) first (at t1) and then, withsome time delay (at t2), the corresponding point in the right eye (PR). Therefore, alatency difference between the neural signals from the corresponding points in the twoeyes may be used as a cue for a moving object not on the fixation plane. (c) Space–timetrajectories of an object moving along a constant depth plane at a constant speed. Thetwo solid lines indicate retinal positions of left and right eye images of the object over time, respectively. The horizontal distance between the two lines cor-responds to the binocular disparity of the object, and the tilt of the lines from vertical indicates speed of the object motion. Interocular spatial offset andtime difference can be specified between any combination of two points on the two lines. For instance, the object position in the right eye at time t2 canhave interocular spatial offsets of D1, D2 and D3, and interocular time differences of ∆T1, ∆T2 and ∆T3, with each of the object positions in the left eye attime t1, t2 and t3, respectively. There is a linear relationship between the interocular spatial offsets and time differences (inset).

not on the fixation plane. However, interocular latency differencealone is not sufficient to determine the relative depth of the object;for a given latency difference, the magnitude of the relative depthdepends on the speed of object motion, and the sign of the relativedepth is determined by the direction of object motion. To explainthe Pulfrich effect with this scheme, the visual system requiresmechanisms tuned to various interocular latency differences at cor-responding points (that is, at zero interocular spatial offset), as wellas a mechanism that converts the latency differences into depthchanges according to the speed and direction of the pendulummotion (interocular time difference hypothesis, Fig. 2b).

There is also a third possibility. Interocular spatial offset andlatency difference may both contribute to the determination ofthe relative depth of the pendulum6,16,18. Consider an objectmoving at a given depth. Left and right eye images of the objecttraverse the retinas along the trajectories illustrated by the twosolid lines in Fig. 1c. Horizontal and vertical distances betweenthe two lines correspond to the interocular spatial offset and timedifference, respectively. Instead of detecting one of the distancesfor a given space or time (as in the previous two hypotheses), thevisual system may integrate measurements of interocular spatialoffset and time difference for various combinations of two pointsalong the trajectories. Such a strategy is expected to produce arobust signal of depth, because it uses information gathered overa range of space and time. Because there is a linear relationshipbetween the interocular spatial offset and time difference (Fig. 1c, inset), this hypothesis predicts that neurons that carry

out this process exhibit a similar linear relationship in their tun-ing functions (space–time integration hypothesis, Fig. 2c).

To test these hypotheses, we used an efficient receptive fieldmapping technique19 and examined responses of neurons in thecat’s striate cortex for various interocular spatial offsets and timedifferences of stimuli. We found that most neurons exhibit tun-ing that is consistent with the space–time integration hypothe-sis (Fig. 2c). This supports the idea that motion and depth areencoded jointly. We also found that these neurons are tuned tospecific combinations of stimulus speed and binocular dispari-ty, suggesting that the joint-encoding carries information unavail-able to independent-encoding of motion and depth. These resultsindicate that motion and depth are encoded jointly at the initialstage of cortical processing.

RESULTSWe measured responses of single neurons in the cat striate cortexto dichoptic presentations of one-dimensional noise patterns thatconsisted of 16 bars in each eye20,21. From the responses, we extract-ed components due to interactions between any combination oftwo bars. The interactions were then examined according to thespatial offset (∆X) and time difference (∆T) between the two bars,for cases in which the two bars belonged to the same eye (monoc-ular interaction) or where the two bars were imaged in oppositeeyes (binocular interaction). These interactions characterize howneurons respond to motion and binocular disparity (D); interac-tions between the two bars that appear sequentially at different loca-

articles


t3

t2

t1 D1

D2

D3∆T2(=0)

∆T3

∆T1Tim

e

Retinal position

Left eye Right eye

D1D2D3

∆T1

∆T2

∆T3

PL PR

F

t1

t1

t2

t2

a

b c

0

0

+

+

Interocular spatial offset

Inte

rocu

lar

time

diffe

renc

e

0 + 0 +

0

+

0

+

Fig. 2. Hypothetical tuning profiles of neurons for interocular spatialoffset and time difference. Each oval indicates a region of responses fora single neuron. (a) Interocular spatial offset hypothesis. Neurons aretuned to an interocular time difference of zero, but to various interocu-lar spatial offsets. (b) Interocular time difference hypothesis. Neuronsare tuned to an interocular spatial offset of zero, but to various interoc-ular time differences. (c) Space–time integration hypothesis. Neuronsare tuned to various combinations of interocular spatial offset and timedifference in such a way that there is a linear relationship between them.

a b c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



tions describe responses to motion, whereas those between the twobars that appear across the two eyes represent responses to binoculardisparity. Our focus here is on binocular complex cells, becausethey are generally selective for binocular disparity and are insensi-tive to monocular spatial phase, so that their responses depend onrelative, not absolute, positions of the two bars21.

Most of the binocular complex cells we examined (52/59)exhibited both monocular and binocular interactions (Fig. 3a–c).However, we also found, for a small number of cells (7/59), strongmonocular but minimal binocular interactions (Fig. 3d). Becausesuch cells cannot encode binocular disparity, they were exclud-ed from the rest of our analysis. We never observed a binocularcell that exhibited binocular but not monocular interactions.

As previously reported22, many monocular profiles wereobliquely oriented in the ∆X – ∆T domain (Fig. 3a and b, left andright columns), indicating that these cells were selective for thedirection of stimulus motion. Binocular profiles of these cells weresimilarly oriented in the D – ∆T domain (Fig. 3a and b, middle col-umn). The tilt angle of the profiles varied from cell to cell, but for agiven cell, monocular and binocular profiles were generally simi-lar. (See below for a parametric analysis.) To quantify the degree oftilt in the binocular profile, we computed a tilt directional index(TDI) for each cell that described a response bias toward one direc-tion of tilt with respect to the opposite direction of tilt (see Methodsfor definition of TDI). A TDI of zero indicated no bias, that is, there

was no tilt, whereas a TDI of one meant the response was com-pletely accounted for by one direction of tilt. The index is equivalentto the direction selectivity index used to quantify the response biastoward the cell’s preferred direction of stimulus motion relative tothe non-preferred direction23, and is analogous to the index thatdescribes the degree of inseparability in space-time receptive fieldsof simple cells24. Indices for cells shown in Fig. 3a, b and c are 0.869,0.839 and 0.076, respectively. The distribution of TDI for our pop-ulation of cells (n = 52) was widely spread with a mean of 0.44 (Fig. 4). Therefore, there was a clear bias for the binocular profile tobe tilted from vertical.

A horizontal slice of the binocular profile represents a cell’stuning for binocular disparity, and, therefore, the obliquely ori-ented profiles indicated that disparity tuning shifted with interocular time difference of the two bars. Previously, it wasshown that placing a light-attenuating filter in front of one eyecauses a shift in tuning of neurons in the cat’s striate cortex forinterocular spatial phase disparity8. This result and ours pre-sented here provide an explanation as to why the incorporationof an interocular latency difference into the neural signals caus-es a shift of the apparent depth of the pendulum in the Pulfricheffect. Whereas it could not be determined in the previous study8

if the neurons were tuned to interocular spatial offset, interocu-lar time difference or both, our results demonstrate clearly that itis both. The approximately linear relationship between interoc-ular spatial offset and time difference, seen in the profiles, is con-sistent with the space–time integration hypothesis (Fig. 2c).

We also found that some cells exhibited profiles that were notoriented (Fig. 3c, TDI = 0.076). Such profiles seemed to be con-sistent with the prediction of the interocular spatial offset hypoth-esis (Fig. 2a). However, they were more likely to be part of acontinuum of degree of tilt. This interpretation was consistent withthe finding that monocular profiles also come in various amounts

articles

0

280

–280

∆T (

ms)

2.6 –2.6

Left eye Right eye

2.6 –2.6 2.6 –2.6

Binocular

∆XL(degrees) D(degrees) ∆XR(degrees)

0 0 0

0

280

–280 2.2 –2.2 2.2 –2.2 2.2 –2.2 0 0 0

0

280

–280 2.2 –2.2 2.2 –2.2 2.2 –2.2 0 0 0

0

280

–280 3.1 –3.1 3.1 –3.1 3.1 –3.1 0 0 0

Fig. 3. Examples of two-bar interaction profiles. Profiles of four binocularcomplex cells (a–d) are shown for the left eye (left column), right eye(right column) and binocular (middle column) domains. The x- and y-axesof each profile indicate the relative spatial offset (∆XL, ∆XR and D) andtime difference (∆T) between two bars in the stimulus, respectively.When two bars belong to different eyes, the x- and y-axes represent aninterocular spatial offset (that is, binocular disparity, D) and an interoculartime difference between the two bars, respectively. Because an interac-tion of a bar with itself cannot be obtained with the stimulus used in thisstudy (binary m-sequence noise), one data point at ∆XL(or ∆XR) = 0 and ∆T = 0 in each of the monocular profiles is missing andhas been filled with a black square. If this data point were available, thetwo green contour regions in the second and fourth quadrants of themaps would have been connected through the point to form an elongatedregion (as has been shown previously22), just like those in the binocularprofiles. Green and red contour regions indicate positive and negative val-ues, respectively. For each cell, contour lines are drawn such that theydivide the response amplitude between 0 and the greater of the positiveand negative peaks of the three profiles into 6 equally spaced levels.

a

b

c

d

Num

ber

of c

ells

Tilt directional index

10.750 0.25 0.5

12

9

6

3

0

Fig. 4. A population distribution of tilt directional index for binocularprofiles. The distribution has a mean of 0.44 (n = 52). An index of 0.25means that the response to the preferred direction of tilt is greater thanthat to the non-preferred direction of tilt by 67%. Because more than70% of the cells have an index of that magnitude or greater, there is aclear bias for the binocular profile to be tilted from vertical.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Fig. 6. An explanation of the Pulfrich effect based on neurons tuned to motion and binocular disparity. Binocular profiles of neurons that preferrightward motion (top row) and leftward motion (bottom row) are illustrated. The x- and y-axes of each profile indicate binocular disparity and inte-rocular time difference (∆T) of the stimulus, respectively. A positive value for the interocular time difference meant that the stimulus was delayed inthe right eye relative to that in the left eye, whereas a negative value indicated a delay in the left eye. White and gray oval regions indicate positiveand negative values, respectively. A horizontal slice of a profile at ∆T = 0 represents the binocular disparity tuning of a neuron. The neurons illus-trated here exhibit binocular disparity tuning curves that peak at different locations along the binocular disparity axis (indicated by X), and there-fore are tuned to different binocular disparities: zero- (left column), near (crossed)- (middle column), and far (uncrossed)-disparity (right column).When the pendulum swings to the right, it mainly activates neurons that prefer rightward motion (top row). If there is no light-attenuating filterplaced in front of either eye (and assuming that the observer is fixating onthe plane of pendulum motion), these neurons would be tuned to zerodisparity. This is because a stimulus at zero disparity and ∆T = 0 (indicatedby the black dot in the top left panel) would be excitatory for them. Whena filter is placed in front of the right eye, introducing a delay ∆TR for theneural signal in that eye, zero-disparity neurons would stop respondingbecause the stimulus at ∆TR and zero disparity is not excitatory. On theother hand, it is an excitatory stimulus for neurons tuned to a near-dis-parity (black dot, top middle panel). This could explain why the perceiveddepth shifts toward the observer in this case. Placing the filter in front ofthe left eye produces a delay ∆TL, and would excite neurons tuned to afar-disparity (black dot, top right panel). This could predict a shift of theperceived depth away from the observer. The same principle applies toneurons that prefer leftward motion when the pendulum swings to theleft. In this case, however, covering the left and right eyes with a filterexcites near- and far-disparity neurons, respectively. This would result in ashift of the perceived depth in the opposite direction to what would hap-pen when the pendulum swings to the right (compare the location ofblack dots between the top and bottom panels in the same column).

of tilt, depending on the cell’s selectivity for the direction of stim-ulus motion22.

In a previous study7 that examined the tuning of area 18 neu-rons in cats to interocular spatial offset and time difference, itwas found that some cells responded best to non-zero interocu-lar time differences. This result is consistent with the interocu-lar time difference hypothesis (Fig. 2b). However, because detailsof the tuning were not examined, it was not possible to distin-guish between different hypotheses based on the data. In fact, wefound no cells that showed profiles consistent with the interocu-lar time difference hypothesis.

The oriented profiles for monocular interactions are ideal fordetecting moving stimuli because they match expected responsesto such stimuli22. The oriented profiles for the binocular interac-tions shown here may be interpreted analogously. Human observersperceive motion in certain stimuli containing an interocular spa-tial offset and a time difference25–27, a phenomenon called dichop-tic motion. The binocular profiles in the middle column of Fig. 3aand b are consistent with expected responses to dichoptic motionstimuli. Therefore, these neurons may constitute a neural basis fordichoptic motion as well. With regard to underlying mechanisms fordichoptic motion, there has been a debate as to whether earlymotion sensors are monocular or binocular28,29. Our results sug-gest that they are binocular and are tuned to binocular disparity.

These findings demonstrate that most neurons encode motionand binocular disparity jointly. Is there a relationship between themotion and binocular disparity encoded by individual neurons? If

there is, it may indicate that the visual system selects certain com-binations of motion and binocular disparity to extract informationthat would not have been available had they been processed sep-arately. On the other hand, if there is no correlation between them,the joint-encoding at this stage seems only to preserve informa-tion about concurrency of motion and binocular disparity. Toaddress this question, we next examined the relationship betweenmotion and binocular disparity tuning of neurons.

For the parameter that represents motion tuning, we estimat-ed the optimal speed (V) as a ratio of the optimal temporal fre-quency (Ft) to the optimal spatial frequency (Fs) for each profile.These three parameters were generally matched between monoc-ular and binocular interactions for each cell. However, there wasa small tendency for binocular optimal spatial frequency to belower and binocular optimal temporal frequency to be slightlyhigher than the monocular counterpart. As a consequence, opti-mal speed tended to be slightly higher for binocular compared tomonocular interaction (t-test; p < 0.01). For the parameter thatrepresented binocular disparity tuning, we estimated optimalbinocular disparity (d) for each binocular profile. The magnitudeof optimal binocular disparity depended on the optimal spatialand temporal frequencies and the optimal speed of the binocu-lar profiles (Fig. 5). We found that the range of binocular dispar-ities encoded by a population of cells decreased with optimal

articles


0.1 0.8

2

0 1 40

2

0 0.4 7

Fs (cycles/degree) Ft (cycles/s) V (degrees/s) Opt

imal

bin

ocul

ar d

ispa

rity

1 1

2

0

1

Optimal parameters in binocular interaction

(de

gree

VA

)

0.3 2 7

dFig. 5. Relationships between motion and depth tuning. Magnitude of theoptimal binocular disparity (d) in degree visual angle (degree VA) is plot-ted as a function of optimal spatial and temporal frequencies and speedfor a population of cells (n = 52). (a) Dependency on optimal spatial fre-quency. The range of binocular disparity decreases with spatial frequency.(b) Dependency on optimal temporal frequency. The range of binoculardisparity increases with temporal frequency. (c) Dependency on optimalspeed. The range of binocular disparity increases with speed. Theseresults indicate that fine disparities are encoded by neurons tuned to var-ious spatial and temporal frequencies and speeds, whereas coarse binoc-ular disparities are encoded by neurons tuned to low spatial frequencies,high temporal frequencies and high speeds.

Binocular disparity

0

+

Inte

rocu

lar

time

diffe

renc

e ∆T

0

0

+

Near Far

Zero-disparity neuron

Near-disparity neuron

Far-disparity neuron

0 Near Far 0 Near Far

0 Near Far 0 Near Far 0 Near Far

∆TR

∆TL

∆TR

∆TL

X X X

X X X

a b c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



spatial frequency (Fig. 5a), but it increased with optimal tempo-ral frequency (Fig. 5b). As a result, the disparity range increasedwith optimal speed (Fig. 5c). In other words, fine disparities wereencoded by neurons tuned to various spatial and temporal fre-quencies and speeds, whereas coarse disparities were encoded byneurons tuned to low spatial frequencies, high temporal fre-quencies, and therefore, high speeds. These results are in accordwith human psychophysical data that indicate that stereoacuitycan be maintained at higher speeds for sinusoidal gratings of lowspatial frequencies compared with gratings of high spatial fre-quencies30, and that the upper disparity limit for coarse stereop-sis increases with stimulus speed31.

DISCUSSIONMotion and binocular disparity constitute two distinct cues forthe visual system to obtain spatial and temporal structures of ascene. Yet they are similar in that both can be described by inter-actions between two points in space and time. Therefore, it is like-ly that the visual system encodes them similarly. Our currentresults demonstrate that motion and binocular disparity areencoded jointly by single neurons in the striate cortex.

We showed that most complex cells in the cat striate cortexexhibited responses due to two-bar interactions for both motionand binocular disparity. Space–time oriented profiles of monocu-lar interaction (Fig. 3a and b) represent motion and are known asmotion energy responses, because they correspond to responsespredicted by a mechanism that detects energy in a motion stimulus(a motion energy model32,33). A similar mechanism has been pro-posed to explain responses due to interocular two-bar interactionswhen there is no time difference between the two bars34,35. Thismechanism is ideally suited for encoding binocular disparity andis called a binocular disparity energy model. The binocular profilesdescribed in this paper, which include interocular time differencesbetween the two bars, can easily be explained by incorporating amotion energy model into a binocular disparity energy model (ahybrid energy model18). Although a binocular disparity energymodel itself does not constrain binocular profiles to have a certainshape in the space–time domain (except at ∆T = 0), a motion ener-gy model provides a space–time-oriented structure to the profiles.

The structures seen in two-bar interaction, including thespace–time oriented ones, have been interpreted as receptive fieldstructures of simple-cell-like subunits that underlie individual com-plex cells21,22,36–38. Subunits are thought to combine inputs linear-ly over space and time21,33–36,39. This summation is linear regardlessof whether it occurs within one eye or across the two eyes. There-fore, binocular interactions exhibit a structure similar to that ofmonocular interactions as long as the latter are similar for the twoeyes. In other words, if monocular profiles are space–time oriented,so is the binocular profile (except for some complex cells such asthat shown in Fig. 3d). In this sense, the space–time oriented struc-ture of the binocular profile results from linear binocular subunitsthat are selective for the direction of stimulus motion. It is thisspace–time oriented binocular interaction that seems to be theunderlying neural basis for the Pulfrich effect and dichoptic motion.

In light of our findings, the Pulfrich effect can be explained asfollows. When the pendulum swings to the right, it mainly acti-vates neurons that prefer rightward motion at around the speed ofthe pendulum (Fig. 6, top row). These neurons exhibit two-barinteraction profiles that are oriented to the right in the space–timedomain. If there is no light-attenuating filter placed in front ofeither eye, then these neurons are the ones that are tuned to zerodisparity (Fig. 6, top left). Now suppose that the right eye is cov-ered with a filter, effectively delaying the neural signal in that eye

by ∆TR. Neurons tuned to zero disparity would decrease respons-es (or stop responding) to the pendulum because a stimulus at∆TR and zero disparity would not be optimal (or excitatory) forthem. On the other hand, such a stimulus is now excitatory forneurons tuned to near (crossed) disparities and would elicitresponses from them (Fig. 6, top middle). Therefore, the per-ceived depth would be expected to shift toward the observer. Onthe other hand, if the filter is placed in front of the left eye, thenneurons tuned to far (uncrossed) disparities would be excited(Fig. 6, top right). This would result in a shift of the perceiveddepth away from the observer. When the pendulum swings to theleft, the same principle applies to neurons that prefer leftwardmotion, except that the direction of shift in the perceived depthreverses (Fig. 6, bottom row). A similar explanation has been pro-posed based on predictions of a hybrid energy model18. Our dataprovide direct physiological evidence that supports the model.

Another finding reported here is that fine disparities are encod-ed by neurons tuned to various speeds, whereas coarse dispari-ties are encoded by those tuned to high speeds. Is there anadvantage of encoding motion and binocular disparity in thismanner? We propose that the relationship between optimal speedand binocular disparity signifies the strategy by which the visualsystem encodes the three-dimensional structure of a scene. Motionof an observer in a direction other than that along the line of sightcreates differences in the motion of retinal images of stationaryobjects at different depths (motion parallax). This is another cuefor depth perception40. When the moving observer fixates a point,images of the stationary objects farther away from the fixationpoint move faster than those of objects closer to the fixation point.In other words, the speed of image motion and binocular dis-parity are closely correlated. Therefore, if neurons were to encodeinformation about the three-dimensional structure of a scenebased on both motion and binocular disparity, then the relation-ship shown in Fig. 5c would be expected.

There is increasing evidence that suggests that motion anddepth are processed together in the visual system41–46. Our resultsstrongly support this notion. We show here that phenomena suchas the Pulfrich effect and dichoptic motion are likely to be theconsequences of joint-encoding at the level of single neurons inthe striate cortex. We also report that binocular complex cells inthe cat’s striate cortex encode specific combinations of motionand binocular disparity that are consistent with the relationshippredicted from motion parallax. Therefore, the joint-encodingof motion and binocular disparity is presumably a design of visu-al processing.

The visual system could process motion and binocular dis-parity separately before combining the results of the computa-tions together to obtain the three-dimensional structure of ascene. Instead, the visual system integrates them in a specificmanner. Such a strategy is not only parsimonious and efficient, itis also reliable; it uses information about the correlation betweenmotion and binocular disparity in the image, which is unavail-able if the two cues are processed separately.

METHODSElectrophysiology. Extracellular recordings were made from single neu-rons in the striate cortex of anesthetized and paralyzed cats. Details of sur-gical and experimental procedures and the physiological recording setup aredescribed elsewhere20,21. All procedures complied with the National Insti-tute of Health Guide for the Care and Use of Laboratory Animals and wereapproved by the University of California Animal Care Committee.

After orientation and spatial frequency tuning of neurons were deter-mined with drifting sinusoidal gratings, responses of binocular neurons

articles©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


to one-dimensional dichoptic white noise stimuli were measured. Eachmonocular stimulus consisted of 16 bars. The orientation of the bars wasset to the optimal for each cell. The luminance of each bar was updatedevery 40 ms according to a binary m-sequence19, and took a value thatwas either brighter (+18 cd/m2) or darker (–18 cd/m2) than the lumi-nance of the background (20 cd/m2). The visual stimulation lasts about20 min, during which spike activity is recorded.

Data analysis. Spike trains were cross-correlated with the stimulussequences of two bars at various spatial offsets and time differences toobtain response profiles for two-bar interaction. When the two bars werechosen from the same eye, the cross-correlation yielded a monocular two-bar interaction profile. On the other hand, if the two bars belonged to dif-ferent eyes, a binocular interaction profile was obtained. Both monocularand binocular profiles were considered at a common optimal correlationdelay, which is defined as the delay at which the sum of squared values ofall data points in the profile is maximum. The x-axis of the profiles wasthe relative spatial distance between the two bars (∆XL and ∆XR for theleft and right eye interactions, respectively, and D for the binocular inter-action, which corresponds to binocular disparity). The y-axis was the rel-ative temporal distance between the two bars (∆T). These profiles wereanalogous to those reported previously22,38. However, they differ fromthose of our previous study35; the y-axis was ∆T, not a correlation delay T.Because of an intrinsic property of binary m-sequences, the interactionof a bar with itself at ∆T = 0 cannot be obtained. Therefore, one data pointis missing from each monocular profile (at ∆XL or ∆XR = 0 and ∆T = 0).

Each interaction profile was transformed into the spatio-temporal fre-quency domain by means of a fast Fourier transform and optimal spatial(Fs) and temporal (Ft) frequencies are obtained at the peak. Then, the opti-mal speed (V) was computed as Ft/Fs. Although it is not intuitive to thinkof spatial and temporal frequencies and speed in the binocular domain,these terminologies are used for both monocular and binocular profilesfor simplicity. In addition, for each binocular profile, a slice at ∆T = 0 istaken, and a one-dimensional Gabor function is fit to the sliced profileusing a Levenberg–Marquardt method47. The optimal binocular disparity(d) was estimated as the phase of the best fit Gabor function, expressed indegree visual angle (degree VA) by taking the optimal spatial frequency ofthe cell into account.

To quantify the degree of tilt in the binocular profile, a tilt direction-al index (TDI) was computed as TDI = (Rp – Rn)/(Rp + Rn), where Rpand Rn are response amplitudes at (Fs, Ft) and (Fs, –Ft), respectively.

ACKNOWLEDGEMENTSWe thank G. DeAngelis for helpful comments and suggestions. This work was

supported by research and CORE grants from the National Eye Institute (EY-

01175 and EY-03176).

RECEIVED 5 SEPTEMBER 2000; ACCEPTED 28 MARCH 2001

1. Braddick, O. A short-range process in apparent motion. Vision Res. 14,519–527 (1974).

2. Wheatstone, C. Contributions to the physiology of vision: I. On someremarkable and hitherto unobserved phenomena of binocular vision. Phil.Trans. Royal Soc. Lond. 128, 371–394 (1838).

3. Pulfrich, C. Die Stereoskopie im Dienste der isochromem undherterochromen Photometrie. Naturwissenschaft 10, 553–564 (1922).

4. Julesz, B. & White, B. Short term visual memory and the Pulfrichphenomenon. Nature 222, 639–641 (1969).

5. Morgan, M. J. & Thompson, P. Apparent motion and the Pulfrich effect.Perception 4, 3–18 (1975).

6. Burr, D. C. & Ross, J. How does binocular delay give information aboutdepth? Vision Res. 19, 523–532 (1979).

7. Cynader, M., Gardner, J. & Douglas, R. in Frontiers in Visual Science (Cool, S. J. & Smith, E. L., eds.) 373–386 (Springer, Berlin, 1978).

8. Carney, T., Paradiso, M. A. & Freeman, R. D. A physiological correlate of thePulfrich effect in cortical neurons of the cat. Vision Res. 29, 155–165 (1989).

9. Lit, A. The magnitude of the Pulfrich stereophenomenon as a function ofbinocular differences in intensity at various levels of illumination. Am. J.Psych. 62, 159–181 (1949).

10. Wilson, J. A. & Anstis, S. M. Visual delay as a function of luminance. Am. J.Psych. 82, 350–358 (1969).

11. Nickalls, R. W. The rotating Pulfrich effect, and a new method of determiningvisual latency differences. Vision Res. 26, 367–372 (1986).

12. Howard, I. P. & Rogers, B. J. Binocular Vision and Stereopsis (Oxford Univ.Press, New York, 1995).

13. Tyler, C. W. Stereopsis in dynamic visual noise. Nature 250, 781–782 (1974).14. Falk, D. S. Dynamic visual noise and the stereophenomenon: interocular time

delays, depth and coherent velocities. Percept. Psychophys. 28, 19–27 (1980).15. Lee, D. N. A stroboscopic stereophenomenon. Vision Res. 10, 587–593 (1970).16. Morgan, M. J. Perception of continuity in stroboscopic motion: a temporal

frequency analysis. Vision Res. 19, 491–500 (1979).17. Ross, J. Stereopsis by binocular delay. Nature 248, 363–364 (1974).18. Qian, N. & Andersen, R. A. A physiological model for motion-stereo

integration and a unified explanation of Pulfrich-like phenomena. Vision Res.37, 1683–1698 (1997).

19. Sutter, E. E. in Nonlinear Vision: Determination of Neural Receptive Fields(Pinter, R. B. & Nabet, B., eds.) 171–220 (CRC, Boca Raton, 1992).

20. Anzai, A., Ohzawa, I. & Freeman, R. D. Neural mechanisms for processingbinocular information I. Simple cells. J. Neurophysiol. 82, 891–908 (1999).

21. Anzai, A., Ohzawa, I. & Freeman, R. D. Neural mechanisms for processingbinocular information II. Complex cells. J. Neurophysiol. 82, 909–924 (1999).

22. Emerson, R. C., Citron, M. C., Vaughn, W. J. & Klein, S. A. Nonlineardirectionally selective subunits in complex cells of cat striate cortex. J. Neurophysiol. 58, 33–65 (1987).

23. DeAngelis, G. C., Ohzawa, I. & Freeman, R. D. Spatiotemporal organization ofsimple-cell receptive fields in the cat’s striate cortex. I. General characteristicsand postnatal development. J. Neurophysiol. 69, 1091–1117 (1993).

24. DeAngelis, G. C., Ghose, G. M., Ohzawa, I. & Freeman, R. D. Functionalmicro-organization of primary visual cortex: receptive field analysis ofnearby neurons. J. Neurosci. 19, 4046–4064 (1999).

25. O. Braddick & A. Adlard in Visual Psychophysics and Physiology (Armington, J.C., Krauskopf, J. & Wooten, B. R., eds.) 417–426 (Academic, New York, 1978).

26. Georgeson, M. A. & Shackleton, T. M. Monocular motion sensing, binocularmotion perception. Vision Res. 29, 1511–1523 (1989).

27. Carney, T. & Shadlen, M. N. Dichoptic activation of the early motion system.Vision Res. 33, 1977–1995 (1993).

28. Carney, T. & Shadlen, M. N. Binocularity of early motion mechanisms:comments on Georgeson and Shackleton. Vision Res. 32, 187–191 (1992).

29. Georgeson, M. A. & Shackleton, T. M. No evidence for dichoptic motionsensing: a reply to Carney and Shadlen. Vision Res. 32, 193–198 (1992).

30. Morgan, M. J. & Castet, E. Stereoscopic depth perception at high velocities.Nature 378, 380–383 (1995).

31. Ziegler, L. R. & Roy, J.-P. Large scale stereopsis and optic flow: depthenhanced by speed and opponent-motion. Vision Res. 38, 1199–1209 (1998).

32. Adelson, E. H. & Bergen, J. R. Spatio-temporal energy models for theperception of motion. J. Opt. Soc. Am. A2, 284–299 (1985).

33. Emerson, R. C., Bergen, J. R. & Adelson, R. H. Directionally selective complexcells and the computation of motion energy in cat visual cortex. Vision Res.32, 203–218 (1992).

34. Ohzawa, I., DeAngelis, G. C. & Freeman, R. D. Stereoscopic depthdiscrimination in the visual cortex: neurons ideally suited as disparitydetectors. Science 249, 1037–1041 (1990).

35. Ohzawa, I., DeAngelis, G. C. & Freeman, R. D. Encoding of binoculardisparity by complex cells in the cat’s visual cortex. J. Neurophysiol. 77,2879–2909 (1997).

36. Movshon, J. A., Thompson, I. D. & Tolhurst, D. J. Receptive field organizationof complex cells in the cat’s striate cortex. J. Physiol. (Lond.) 283, 79–99(1978).

37. Szulborski, R. G. & Palmer, L. A. The two-dimensional spatial structure ofnonlinear subunits in the receptive fields of complex cells. Vision Res. 30,249–254 (1990).

38. Gaska, J. P., Jacobson, L. D., Chen, H. W. & Pollen, D. A. Space-time spectra ofcomplex cell filters in the macaque monkey: a comparison of results obtainedwith pseudowhite noise and grating stimuli. Vis. Neurosci. 11, 805–821 (1994).

39. Ohzawa, I. & Freeman, R. D. The binocular organization of complex cells inthe cat’s visual cortex. J. Neurophysiol. 56, 243–259 (1986).

40. Rogers, B. J. & Graham, M. E. Motion parallax as an independent cue fordepth perception. Perception 8, 125–134 (1979).

41. Poggio, G. F. & Fischer, B. Binocular interaction and depth sensitivity instriate and prestriate cortex of behaving rhesus monkey. J. Neurophysiol. 40,1392–1405 (1977).

42. Maunsell, J. H. & Van Essen, D. C. Functional properties of neurons inmiddle temporal visual area of the macaque monkey. II. Binocularinteractions and sensitivity to binocular disparity. J. Neurophysiol. 49,1148–1167 (1983).

43. Felleman, D. J. & Van Essen, D. C. Receptive field properties of neurons in areaV3 of macaque monkey extrastriate cortex. J. Neurophysiol. 57, 889–920 (1987).

44. Roy, J.-P., Komatsu, H. & Wurtz, R. H. Disparity sensitivity of neurons inmonkey extrastriate area MST. J. Neurosci. 12, 2478–2492 (1992).

45. Bradley, D. C., Qian, N. & Andersen, R. A. Integration of motion and stereopsisin middle temporal cortical area of macaques. Nature 373, 609–611 (1995).

46. DeAngelis, G. C., Cumming, B. G. & Newsome, W. T. Cortical area MT andthe perception of stereoscopic depth. Nature 394, 677–680 (1998).

47. Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. NeumericalRecipes in C 2nd edn. (Cambridge Univ. Press, New York, 1992).

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Cortical plasticity is likely involved in the normal cognitive pro-cessing of adult animals, and is important even in the function-ing of primary sensory areas1. Whereas plasticity may underliefunctional recovery following CNS lesions, it may also mediatefunctional changes accompanying experience and learning. Severalcharacteristics of perceptual learning suggest the involvement ofearly stages in sensory processing, perhaps even as early a stage asprimary sensory cortex. Evidence in support of this idea has beenfound in the somatosensory and auditory systems2–4. Primarysensory cortex is a useful model for learning because its underly-ing mechanisms—its circuitry, functional architecture and recep-tive field (RF) structure—can be accessibly studied.

Various cortical changes both are associated with improve-ment in perceptual performance and occur in primary sensorycortices2–4, which suggests that the mechanisms of learning may begeneral to the neocortex as a whole. Independent studies of plas-ticity of RF properties and functional architecture of primary visu-al cortex (V1) heighten the possibility that plasticity associatedwith perceptual learning might occur there. Whereas certain RFproperties such as ocular dominance are mutable only during alimited critical period early in postnatal development5, a numberof other properties, most notably, visual topography and RF size,can be influenced by visual experience throughout life1.

In the visual system, psychophysical evidence shows that train-ing can improve discrimination stimulus attributes, includingposition, depth, orientation, motion, texture, spatial phase andhyperacuity6–11. In a previous study, we showed that practice witha particular visual discrimination task, three-line bisection, pro-duces a substantial improvement that is specific to the trainedstimulus12. The specificity for position and orientation suggeststhat the early stages of visual processing are involved in the learn-ing of this task. To determine whether V1 is involved in the learn-ing of such visual discriminations, we trained two macaquemonkeys to perform the same bisection task used with humansubjects, and we recorded from cells in area V1. We examinedthe RF properties and the map of visual space in trained animalsfor changes that might relate to perceptual training. In addition

to its classical response properties, a cell’s response to a stimu-lus within the RF is modulated by the presence of additional stim-uli around the RF, and this modulation depends strongly on thegeometric relationship of the stimulus elements. We exploredthe tuning of cells to shifts in the lateral placement of two par-allel lines with positions analogous to those of lines in the bisec-tion task. Furthermore, we examined the influence of thebehavioral state of the animal on these interactions.

RESULTSMonkeys improve bisection discrimination with trainingBoth monkeys showed substantial improvement in bisection per-formance with training. Over the course of 30 weeks of training,monkey 1 showed a threshold reduction greater than a factor ofthree (Fig. 1c), and monkey 2 showed a 57% threshold reduc-tion (Fig. 1d).

Monkeys were trained with a series of tasks that approximat-ed the final bisection task. The initial threshold (Fig. 1b) wasmeasured during the first week of training on the bisection taskitself, and some learning may have occurred before this time.During the recording phase, the monkey had fewer opportunitiesto practice the bisection discrimination, but the threshold mea-sured several months after electrophysiological recordings beganshows that the improved performance was maintained over thisperiod (Fig. 1b).

Cortical magnification does not change with trainingOne of the most striking effects of perceptual training in thesomatosensory and auditory modalities3,4 is the remapping of thecortical representation of the sensory surface such that a muchlarger region of cortex is responsive to stimulation of the trainedregion. To examine whether a remapping of visual space accom-panied the learning of the bisection discrimination task describedabove, we constructed a topographic map of the cortex of thetrained monkeys from recordings of superficial layer cortical cells.The location of the RF center of an isolated unit was determinedby recording the response to the flashing of a small bar in a pattern

articles

Learning to see: experience andattention in primary visual cortex

Roy E. Crist, Wu Li and Charles D. Gilbert

The Rockefeller University, 1230 York Avenue, New York, New York 10021, USA

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

The response properties of neurons in primary sensory cortices remain malleable throughout life.The existence of such plasticity, and the characteristics of a form of implicit learning known asperceptual learning, suggest that changes in primary sensory cortex may mediate learning. Weexplored whether modification of the functional properties of primary visual cortex (V1)accompanies perceptual learning. Basic receptive field properties, such as location, size and orienta-tion selectivity, were unaffected by perceptual training, and visual topography (as measured bymagnification factor) was indistinguishable between trained and untrained animals. On the otherhand, the influence of contextual stimuli placed outside the receptive field showed a change consis-tent with the trained discrimination. Furthermore, this property showed task dependence, onlybeing manifest when the animal was performing the trained discrimination.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


of equally spaced locations both along and orthogonal to an axisdefined by the orientation preference of the neuron. Recordingswere made in a grid-like pattern in steps of 0.5 to 1 mm. The RFlocations determined for 144 separate penetration sites in monkey 1 were used to estimate the appropriate position of isoaz-imuth and isoelevation lines surrounding the representation ofthe visual location where the bisection stimulus was presented(Fig. 2b). An estimate of cortical magnification of the area sur-rounding the trained location was obtained by calculating thesquare root of the number of square mil-limeters representing a 1° × 1° area of visualspace17; the magnification factor was 20 min/mm for the cortical region repre-senting the visual position of the bisectionstimulus. This figure matches reported val-ues for magnification at this cortical location

Fig. 2. Representation of visual space did notchange as a result of bisection training. (a) Comparison of cortical magnification factormeasured at several locations surrounding therepresentation of the trained visual location. (b) Map of cortical area representing the regionof space where the bisection stimulus was pre-sented during training (3.25° parafoveal).Isoazimuth and isoelevation lines are drawnbased on the RF positions measured for eachpenetration site (black dots). (c) Cortical map ofuntrained hemisphere of monkey 2. (d) Corticalmap of trained hemisphere of monkey 2. Lightgray areas in (b) and (d) indicate region wherebisection task was represented. Dark gray areasin (b–d) show area over which magnification fac-tor was calculated for panel (a).

in the cortex of naive monkeys18, indicating that no substantialincrease occurred as a result of perceptual training.

A similar map was made from the trained hemisphere ofmonkey 2 (Fig. 2d). The magnification factor at the cortical locusof the bisection stimulus was 19 min/mm, closely matching thevalue obtained in monkey 1. Though the values for the two mon-keys used in our study match those reported for naive monkeys,the individual variation found in the cortical representation ofspace might permit more subtle changes to be missed in thesecomparisons. Therefore, we mapped the untrained hemisphere ofmonkey 2 (Fig. 2c). The magnification factors for various eccen-tricities in both trained hemispheres and the untrained hemi-sphere of monkey 2 were compared across the cortical regionsurrounding the bisection task; the values in all three hemisphereswere almost identical (Fig. 2a).

No change in receptive field size or orientation tuningAlong with substantial changes in the size of the cortical repre-sentation of the trained skin surface, changes in RF size have beenreported to accompany the learning of a somatosensory dis-crimination task3. Thus, we asked whether changes in RF size inV1 accompanied learning of the bisection discrimination. Weflashed a small bright bar in locations spanning the area of theRF along the orientation axis, and calculated RF size as the cen-ter-to-center distance between the outermost stimulus bar loca-tions showing response above spontaneous levels. The mean RF

Fig. 1. Monkeys improved with training on bisection task. (a) The tasksthat monkeys were trained to perform. Top, a single bisection stimuluspresentation as it was shown to a monkey. Bottom, fixation task in whichthe monkey was merely required to maintain fixation until the fixationpoint was dimmed. Beneath the images of the stimuli are traces indicatingthe time course of stimulus presentation. (b) Thresholds measured formonkey 1 for each week of training. During the recording phase, themonkey received less bisection training, but improvement was retained.(c) Before and after training thresholds measured for monkey1. (d) Before and after training thresholds measured for monkey 2.

articles


Bisectiontrials

Fixationtrials

Fixation Stimulus Delay Response

Fixationpoint

Saccadetargets

Bisectionstimulus

Response

500 ms

Lever

Fixation point

Stimulus

Reward

10

0Beforetraining

Aftertraining

16

0

Beforetraining

Aftertraining

Week of training

0

16

0 30 166

Thr

esho

ld (

min

. of a

rc)

Recordingphase

Trainingphase

monkey 2

monkey 1

Monkey 2, right (trained) hemisphereMonkey 2, left (untrained) hemisphere

L M

P

A

L

-3.5

-3.1

-2.7

-2.3 -1.9

-1.5

-2.9 -2.5

-2.1

-1.7

-1.3

-0.9

-0.5

-0.1

Monkey 1, right (trained) hemisphere

1 mm

-2.1

-1.7-1.3

-0.9

-0.5

-3.1

-2.7

-2.3

-1.9

-1.5

M

P

A

0.5 0.9

1.3

1.7

2.1

2.5 2.9

-3.1

-2.7

-2.3

-1.9

-1.5

2.7-3.1 2.3-2.7 1.9-2.3 1.5-1.9distance from fovea (degrees)

0

1

2

3

4

Mag

nific

atio

n fa

ctor

(mm

/deg

ree)

monkey 1, trained hemispheremonkey 2, trained hemispheremonkey 2, untrained hemisphere

elevation (deg)azimuth (deg)

a

b c

d

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


monkey was performing. Two effects could be distinguished.First, the response rate of a neuron was slightly reduced whenthe monkey was instructed to perform bisection trials. This find-ing was not unexpected, as we have observed previously thatwhen additional stimuli (in this case, the bisection stimulus) areplaced in the neighborhood of the RF, the response of the cell isinhibited. The most striking effect, however, was that the shape ofthe tuning curve for the modulation produced by a flanking barwas very different when the monkey was performing the bisectiontask than when it was performing the fixation task. In fixationtrials, the response of a typical neuron was reduced by the pres-ence of a flanking bar on either side of the RF (Fig. 4a, top right).When the monkey was performing the bisection discrimination,the effect of the flanking bar changed dramatically from inhibi-tion to facilitation, and the facilitatory effect depended on thedistance between the two bars (Fig. 4a, bottom). Furthermore,the extent of facilitation was asymmetric; the facilitatory effectof a flanking bar was stronger on one side of the RF.

This effect was even more dramatic for a second cell (Fig. 4b),in which presenting a flanking bar on one side of the RF duringbisection trials produced a substantial facilitation, whereas pre-

Fig. 4. Contextual interactions changed when the monkey performedthe bisection task. (a) Left, stimulus arrangement used to examine con-textual interactions during the performance of fixation and bisection tri-als. Dotted line indicates borders of RF. Upper right, tuning curve of onecell to the placement of parallel bars while the monkey was performing afixation task. Response rate of cell to all stimuli has been normalized bythe response rate obtained by placing a single optimally oriented barinside the RF. Lower right, tuning curve of the same cell for the same setof parallel bar stimuli while the monkey performed bisection trials. (b) Example of the tuning of another cell for parallel bar stimulus. (c) Example of the tuning from a third cell for parallel bar stimulus. Blacklines indicate responses obtained when the animal was performing thefixation task. Gray lines indicate tuning during bisection trials. In the tun-ing curves, the vertical lines represent the boundaries of the receptivefields determined by the minimum response field technique (seeMethods). These boundaries did not change with bisection versus fixa-tion trials. Error bars, s.e.m.


size was calculated for two trained and two untrained monkeysfor neurons at eccenticities near that of the bisection stimulusused with the trained animals (between 2° and 4°); RF size intrained and untrained animals did not differ (Fig. 3a).

Cells in V1 are also particularly sensitive to the orientation ofedges and line stimuli. Selectivity for orientation has been pro-posed to be involved in hyperacuity discriminations. However,we found similar orientation tuning bandwidth of cells in cor-tex of trained versus untrained animals (Fig. 3b).

Task dependence of contextual interactionsBecause standard properties such as magnification factor, RF sizeand orientation selectivity were unaffected by training, weexplored possible changes in higher-order properties such as con-textual tuning. Cells in V1 are sensitive to patterns more com-plex than oriented line segments19. The effect of placingadditional lines outside the classical RF of a cell depends on theprecise geometric relationships between the elements of the stim-ulus pattern. For example, a single collinear line presented outsidethe classical RF along the orientation axis often provides facili-tatory input to a unit responding to an optimally oriented linewithin its RF20; when placed in a side-by-side configuration, theline outside the RF has an inhibitory influence21,22. We thereforeexamined whether contextual patterns in the three-line bisectionstimulus uniquely affected the responses of visual cortical neu-rons in trained animals.

We centered an optimally oriented bar in the RF, and mea-sured the response to a second, parallel bar presented in a flank-ing position, similar to the side-by-side arrangement of lines inthe bisection stimulus. Because the monkeys were trained to per-form both a fixation task and a bisection task, we could exam-ine the responses of our units while an animal was performingdifferent visual discriminations. We measured changes in theresponses of neurons with RFs near the trained location of thevisual field, and compared the changes based on the task the

articles

60

30

0Trained

RF

Siz

e (a

rcm

in)

Untrained

80

40

0Trained

Tuni

ng w

idth

(de

g)

Untrained

Fig. 3. RF size and orientation tuning bandwidth did not change follow-ing bisection training. RF size was determined using the minimumresponse field technique (see Methods). (a) Comparison of RF size intrained and untrained monkeys. Left bar indicates mean RF size in mon-key trained on bisection discrimination (n = 139; monkey 1, 119 units;monkey 2, 20 units). Right bar indicates mean RF size in untrained ani-mals (n = 51). (b) Orientation tuning in trained and untrained monkeys.Left bar shows mean width of orientation tuning curves in two trainedmonkeys (n = 108; monkey 1, 87 units; monkey 2, 21 units). Right barindicates the mean value of tuning width in untrained animals (n = 63).

a

b

Bisectiontrials

Fixationtrials

8

-1 0 1

1

8

-1 0 1

1

Nor

mal

ized

resp

onse

Nor

mal

ized

resp

onse

Distance (degrees)

0

1

2

-1 0 1Distance (degrees)

Nor

mal

ized

res

pons

e

0

1

-1 0 1Distance (degrees)

Nor

mal

ized

res

pons

e 4

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. 5. Modulation depended on the task thatthe monkey performed in trained but notuntrained hemisphere. (a) Modulation duringbisection and fixation trials. Amount of modula-tion observed for different units when the mon-keys were performing the bisection task incomparison to the amount of modulation whenthe animals were performing fixation trials (n = 67). (b) Monkey performed bisection taskon trained hemisphere, and test stimuli werepresented to cells in untrained hemisphere.Modulation did not depend on the behavioralcondition for cells in the untrained hemisphere.

senting the bar on the opposite side caused a profound inhibi-tion of the unit’s response. The shape of the tuning curve forflanking interactions during bisection trials varied considerablyfrom cell to cell (Fig. 4a–c).

The wide variety of tuning shown by different cells for flank-ing interactions during bisection performance challenged us todevelop a method of comparing the responses over the record-ed population. The amount of facilitation or inhibition producedby the flanking bar was greater when the animal was performingbisection trials than when it was performing fixation trials. Todescribe this effect for the population of neurons, we defined ameasure of the strength of these modulatory interactions.

Modulation is a measure of the influence of the parallel baron the cell’s response to a bar centered in its RF. By definition,modulation ranges from zero (the response to single bar was notinfluenced by the presence of the additional bar) to one (theresponse was facilitated by an additional bar in one position andinhibited by the additional bar in another position). The amountof modulation observed for 67 cells while the monkey performedbisection trials was plotted against the modulation observed underfixation trials (Fig. 5a). The difference in the distribution in mod-ulation in the trained hemisphere between fixation and bisectiontrials was highly significant (p < 0.0002, t = 3.95, paired t-test).

Most cells examined in the experiment described above hadRFs near but not overlapping the visual location where the bisec-tion stimulus was presented during training. The appearance oftraining effects in the region surrounding the trained location is inagreement with psychophysical evidence from human observersdemonstrating that bisection discrimination training improvesperformance within a couple of degrees of the trained location12.

modulation = maximum normalized response – minimum normalized responsemaximum normalized response + minimum normalized response

Bisection performance diminishes with distance from the trainedlocation, suggesting that cells in cortical areas remote from thetrained location should not show the task-dependent changes inmodulation. Contrary to what we observed in the trained hemi-sphere, in the untrained hemisphere, the modulation of cellularresponses by a parallel line was similar during the bisection andfixation tasks (Fig. 5b). The difference in modulation under bisec-tion between the trained and untrained hemispheres was signifi-cant (p < 0.05, t = 2.22, two-sided t-test), and the difference inmodulation in the untrained hemisphere between fixation andbisection was not significant (t = 1.07, p = 0.29, paired t-test).

Contextual interactions have been shown to be influenced byspatial attention23. Clearly, the focus of attention changed fromthe center of vision to the peripheral location of the bisection

articles


fixation

bisection

0.0 0.2 0.4 0.6 0.8 1.00.0

0.2

0.4

0.6

0.8

1.0

trained hemisphere

0.0 0.2 0.4 0.6 0.8 1.00.0

0.2

0.4

0.6

0.8

1.0

untrained hemisphere

bisection

fixation

( = 67)n ( = 30)n

a b

fixation

bisection

fixation

fixation

bisection

fixation

0.0

0.2

0.4

0.6

0.8

1.0 Trained hemisphere ( = 12)n

0.0

0.2

0.4

0.6

0.8

1.0 Untrained hemisphere

0.0 0.2 0.4 0.6 0.8 1.00.0

0.2

0.4

0.6

0.8

1.0

( = 13)n

Untrained hemisphere ( = 13)nFig. 6. Changes in contextual interactions are not observed for untrainedstimulus patterns or in the untrained hemisphere. (a) Task dependence ofcontextual interactions was only seen for patterns present in the trainedstimuli. Modulation by a collinear line showed no effect of behavioral con-dition. (b) Monkey performed a fixation task while test stimuli were pre-sented in untrained hemisphere in the presence of bisection-like stimulus.The presence of a bisection-like stimulus did not lead to changes in mod-ulation. (c) Monkey performed bisection task in trained location while teststimuli and bisection-like pattern were presented to cells in untrainedhemisphere. Modulation of cellular responses in the untrained hemi-sphere did not depend on behavioral condition.

a

b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


DISCUSSIONThe suggestion that perceptual learning might rely on primarysensory cortex comes from the great specificity in the learningfor position and orientation, attributes to which early stages insensory processing are selective. The specificity of visual percep-tual learning in itself is not sufficient to indicate the involvementof the early stages of visual cortical processing, because even cellsin inferotemporal cortex can be induced to show selectivity forsuch properties24. Moreover, attention can reduce the size of acell’s RF in certain cortical areas25 and therefore heighten thespatial resolution of tasks involving these areas. One also has toaccount for the specificity of perceptual learning to the contextwithin which a discrimination is made. Because cells in V1 showselectivity for complex stimulus configurations, it becomes dif-ficult to use complexity as a clue to determine the site of learningwithin the cortical hierarchy. In this vein, we note that learning todiscriminate even very complex stimulus patterns can show somespecificity for the location in which the stimulus was presentedduring training26.

Nevertheless, the suggestion that V1 might be involved in per-ceptual learning is reinforced by the demonstration of learningeffects in other primary sensory areas, and one might expectcommon mechanisms to apply in cortical areas serving all sen-sory modalities. V1 is capable of remapping space under certainconditions—in particular, after retinal lesions27–29. No changes incortical magnification, or cortical recruitment, were found in thecurrent study to accompany training on the bisection discrimi-nation, despite large gains in perceptual ability. Furthermore, wefound the basic RF properties of size and orientation tuning tobe unaffected by bisection training, even though analogous prop-erties were modified in somatosensory and auditory cortex fol-lowing training on vibration or pitch discrimination. Theapparent difference in the mechanisms operating in different cor-tical areas may reflect the fact that our training involved emer-gent properties of cortical neurons, and showed specificity forcontext. In contrast, the changes in the somatosensory and audi-tory systems resulted from discrimination training on the inputproperties to the cortex, and therefore might not depend onintrinsic cortical circuits.

Increasing the amount of modulation invoked by visual pat-terns in the trained stimulus can easily be imagined to facilitateperceptual performance. It is noteworthy in this context that dur-

Fig. 7. Measurement of receptive field profiles during fixa-tion and bisection trials in trained and untrained hemi-sphere. (a) Example of a cell from untrained hemisphere.Responses were measured for a single bar stimulus placed invarious positions along an axis orthogonal to the RF orien-tation. Dark lines are fitted to the mean responses, graylines are the Gaussian fits. Solid lines are measurementstaken during bisection, and dashed lines are measurementstaken during fixation. (b) Example of a cell from trainedhemisphere, same conventions as in (a). (c, d) Populationanalysis of cells from trained and untrained hemispheres,showing mean difference in RF position (c) and size (d) under bisection versus fixation. For this analysis, the RFposition and width were obtained from the peak positionand one standard deviation, respectively, of the Gaussianfits. Solid bars, trained hemisphere; open bars, untrainedhemisphere. No significant difference was observed ineither trained or untrained hemispheres.


task when our monkeys performed the bisection discrimination.If the change in contextual interaction were solely due to changesin the focus of attention, we would expect the modulation of cel-lular responses to other contextual patterns. Therefore, we exam-ined the effect of performing the bisection task on the modulationproduced by a collinear line presented in the RF surround, butlittle change in the modulation index was observed (Fig. 6a).

The finding that task-dependent changes were only seen forcontextual patterns present in the trained stimulus suggests thatthe difference observed in contextual interactions described herewas the result of the training the monkeys received. In our exper-iments, however, when a monkey was performing the bisectiontask, the test pattern was presented in the RF of the cell simulta-neously with the bisection stimulus itself. This raises the possi-bility that the changes in cortical interactions we have describedmight be due to the difference in the stimulus pattern. If this werethe case, however, the response of a cell to other local contextu-al patterns, such as the collinear line used in the experimentdescribed above, would also change when the bisection patternwas present. This was clearly not the case (Fig. 6a). Further evi-dence of this fact comes from recordings in the untrained hemi-sphere, where the modulation of cellular responses was unaffectedby the simple presence of the bisection stimulus (Fig. 6b). Final-ly, to further demonstrate that cells in untrained cortex did notshow task-dependent changes in modulation, we determinedwhether the presence of a bisection-like stimulus in the untrainedhemifield during bisection discrimination in the trained loca-tion would induce changes in the modulation of cells in untrainedcortex. No such change was observed (Fig. 6c).

The above findings indicate that contextual interactions, andnot the responses of cells to simple stimuli, were affected bytraining. However, given that the contextual interactions showtask dependence, one might ask whether the standard RF prop-erties of size, position and orientation might be similarly affect-ed. We measured the RF profiles of cells under both fixation andbisection trials, and fitted these profiles with a Gaussian. TheRF profiles were nearly identical under both behavioral condi-tions and in both the trained and untrained hemispheres (Fig. 7).Between fixation and bisection trials, no significant differencewas found for RF position (n = 45, t = 0.27, p ≈ 0.79), RF size(n = 45, t = –1.52, p ≈ 0.13) or orientation preference (n = 20, t = –1.67, p > 0.1).

articles

-1.0 -0.5 0.0 0.5 1.00

40

80

120

160 Untrained hemisphere

< -0.3

-0.3to

-0.2

-0.2to

-0.1

-0.1to

0

0 to0.1

0.1to

0.2

0.2to

0.3

≥ = 0.30

5

10

15

20

25

RF position difference (degree)

Bar position (degree)

N=45

traineduntrained

< -0.3

-0.3to

-0.2

-0.2to

-0.1

-0.1to

0

0 to0.1

0.1to

0.2

0.2to

0.3

≥ = 0.30

3

6

9

12

15

RF size difference (degree)

N=45

-1.0 -0.5 0.0 0.5 1.00

40

80

120

160

Bar position (degree)

fixationbisectionfixation (gauss)bisection (gauss)

Trained hemisphere

Res

pons

e (s

pike

s/s)

Num

ber

of c

ells

a b

dc

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


ing bisection trials, many cells were strongly facilitated by a par-allel bar. In a study of contextual interactions in monkeys view-ing passively, cells in untrained animals did not show facilitationfor this pattern21. The absence of such facilitation in untrainedanimals strengthens the conclusion that these interactions werethe result of bisection training. A substrate for these interactionsexists in V1 in the axon collaterals of superficial layer pyramidalcells, which extend over several millimeters. These cells connectunits of similar orientation preference30–33 and are therefore wellpositioned to mediate interactions between the elements of thebisection stimulus used in our study. The ability of long-rangehorizontal connections to provide both excitatory and inhibito-ry input to their targets34–36 is also suggestive of a role in theinteractions described above. The implementation of perceptuallearning by a selective modulation of subsets of horizontal con-nections allows for the specificity of the learning for the detailsof stimulus configuration. A mechanism of cortical recruitmentwould not have this property, and further, might be expected tolead to a decline in performance in the untrained portions ofvisual space represented by the adjacent cortical regions. This‘robbing’ is not seen in the perceptual learning experiments.

The contextual influence that best represents the attributeinvolved in the bisection task is the modulation in the cells’responses to a second, parallel line placed at varying separationfrom a line placed centrally within the RF. The observed effectswere specific for this property, because they were not seen for adifferent contextual influence, that of a colinear line at varying off-sets. To make a stronger connection between the modulation indexand the trained task, one can ask how the scale of the sensitivityto line separation measured for an individual neuron compares tothe threshold in the bisection task. At the point at which neuronsshow their greatest sensitivity (the steepest part of the slope of thecontextual tuning curves), the amount of change in the distancebetween the parallel lines that gives one standard deviation changein the firing rate is 9.75 minutes of arc, on average. The thresholdin the task, after training, was four minutes of arc. The sensitivi-ties are therefore roughly of the same scale, but the differencesargue that some amount of pooling in the activity of the neuronsis required to achieve the level of behavioral performance.

Perhaps the most striking aspect of the findings reported hereis the apparent ability of the cortex to dynamically modify theprocessing of visual information according to immediate behav-ioral requirements. The monkeys in this study were trained toperform two different tasks: a simple dimming task and the three-line bisection discrimination. Lateral interactions in trained mon-keys depended on the task the monkey was performing at thetime. The task dependence of the contextual interaction permitsthe same neurons to mediate entirely different perceptual func-tions that may require opposing neuronal mechanisms. The inhi-bition of responses by parallel lines under the fixation task hasbeen suggested to be involved in surface segmentation21. Asshown here, this inhibition can switch to facilitation during theperformance of the three-line bisection task. One would not wantto design a system in which three-line bisection training woulddisrupt the subject’s ability to segment the visual scene. To allowthe performance of both tasks, one would either have to segre-gate the neurons mediating the task into separate functional com-partments, or allow the same neurons to multiplex their functionin a task-dependent fashion. The results of the current study sug-gest the latter solution. The further implication of this idea is thatat the same time the cells change their RF structure, they changetheir line label, such that modulation in their firing is interpret-ed differently by the rest of the nervous system.

As a mechanism underlying the improvement in the bisec-tion task, we propose a change in the strength andexcitatory/inhibitory balance of a subset of horizontal inputs toV1 neurons. This modulation of the horizontal input would thenvary according to the separation between the source and targetneurons, therefore providing a greater modulation of the tuningto the separation of parallel lines. The contextual modulationwould in turn be modulated by top-down influences, presum-ably mediated by feedback connections from higher-order cor-tical areas, to generate its task dependence. Thus, contextualinfluences within a particular cortical area may come not justfrom lateral connections within that area, but as an interactionbetween local circuits and feedback connections from higher-order cortical areas, thus providing a mechanism for both thestimulus selectivity and task dependence of the cortical respons-es to trained stimulus patterns.

METHODSTwo macaque monkeys (Macaca mulatta) were trained and used to collectthe physiological data reported here. During recording and training, mon-keys were seated in a primate chair facing a computer monitor. The mon-key’s head was restrained using a surgically implanted stainless steel post.Eye movements were monitored using a scleral search coil system13 (CNCEngineering, Seattle, Washington). During stimulus presentation, monkeyswere required to maintain fixation within a 0.75–1.0° rectangular win-dow; trials were aborted and reward was withheld if an eye movementgreater than 0.5° was made. The actual variability in eye position fromtrial to trial was much less than that allowed by the fixation window. Themean position and standard deviation for typical experiments (based ona sample of data obtained during the recording of 21 units), relative tothe fixation point, was 0.07 ± 0.03° in azimuth and –0.02 ± 0.05° in ele-vation. Quality of fixation was the same during both fixation and bisectiontrials; the mean difference in eye position between fixation and bisectiontrials for typical sessions was 0.03 ± 0.04° in azimuth and 0.01 ± 0.02° inelevation. All procedures were conducted in compliance with the Nation-al Institutes of Health Guide for the Care and Use of Laboratory Animals,and under approval of institutional review boards.

Training. Monkeys were initially trained to perform a simple fixationtask (Fig. 1a). The animal was taught to initiate trials by pulling a leverattached to the primate chair. When the lever was pulled, a small brightspot was displayed on the screen, and the animal was required to main-tain fixation on the spot until it was dimmed. If the monkey released thelever within a brief period of time following the dimming of the fixationspot, a small drop of juice was given as a reward.

To teach the monkeys to perform the bisection discrimination, we firsttrained them to perform a series of tasks designed to lead to bisection per-formance. The initial thresholds for bisection performance reflected thefirst thresholds measured for the final bisection task. In the bisection task,the monkey was presented with a set of three parallel horizontal lines (Fig. 1a). The monkey’s task was to determine whether the central linewas nearer to one or the other flank. During the presentation of the stim-ulus, the monkey was required to maintain fixation on a small spot. Fol-lowing the presentation of the bisection stimulus, the fixation spot wasextinguished and the monkey indicated its response by making an eyemovement to one of two small spots presented at the top and bottom ofthe screen. Correct responses were rewarded with a drop of juice. Theresponses of the monkey to the bisection task were recorded and used tocalculate the threshold of performance using the method of probits.

Electrophysiological recording. After training was complete, monkeyswere surgically implanted with a steel chamber (inner diameter, 22 mm)enclosing a craniotomy over a portion of V1. Surgical procedures weredone under aseptic conditions with pentobarbital sodium anesthesia.Recordings were made with glass-coated platinum iridium microelec-trodes14 with impedances between 1.0 MΩ and 3.0 MΩ. Using a step-ping motor microdrive (Narishige, Tokyo, Japan), penetrations weremade at typical intervals of 0.5 mm through the dura.

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Recordings from single units were conducted daily in 2 to 4 hour ses-sions. After penetrating the dura, a rough RF map was obtained whilethe animal performed a fixation task. All recordings were made from theopercular surface of V1 at RF eccentricities ranging from 1.5° to 5.0°.Granule layers were identified by the characteristics such as high levelsof spontaneous activity and brisk on/off response15,16. When such activ-ity was encountered, the electrode was retracted to restrict recording tothe superficial 600 µm of the cortex.

Neuronal activity was recorded over 600-ms epochs spanning the pre-sentation of the stimulus. The level of background activity was measured for200 ms, and a stimulus was then presented for 100 ms. During each trial,three to five 600-ms recording periods were conducted. For each cell, atime window was set within a range of 50–250 ms after stimulus onsetdepending on the latency and length of the response. The mean firing ratewithin this response window minus the spontaneous firing rate, calculat-ed from the number of spikes obtained in the first 200 ms of the record-ing epoch, was used to determine the magnitude of the evoked response.The t-test was used to evaluate the significance of the evoked response.

Each recording session began by characterizing the RF extent and ori-entation preference while the monkey performed fixation trials. RF extentwas determined by the minimum response field technique in which asmall bar (typically 0.2–0.25° in length and 3 inches in width) was pre-sented in steps (0.1–0.25° apart) either along the principal orientationaxis of the cell to determine its length, or orthogonal to the orientationaxis to determine its width. The distance between the outermost pointsthat elicited a significant response was defined to be the size of the RF.For the comparison between RF profiles during fixation and bisectiontrials in trained and untrained hemisphere (Fig. 7), RF extent was deter-mined by fitting the responses obtained at each position with a Gauss-ian and taking the standard deviation to be the size of the RF.Subsequently, contextual interactions were examined while the animalperformed either fixation or bisection trials. During bisection trials,experimental stimuli were presented simultaneously with the bisectionstimulus. The fixation and bisection trials were collected in separateblocks, and we collected 10 trials per stimulus condition. The tuningcurves were calculated, and are shown along with standard error bars.

ACKNOWLEDGEMENTSThis work was supported by NIH grants EY07968 (C.D.G.) and GM07524

(R.E.C.).

RECEIVED 12 OCTOBER 2000; ACCEPTED 28 FEBRUARY 2001

1. Gilbert, C. Adult cortical dynamics. Physiol. Rev. 78, 467–485 (1998).2. Weinberger, N. M. et al. Retuning auditory cortex by learning: a

preliminary model of receptive field plasticity. Concepts in Neuroscience 1,91–132 (1990).

3. Recanzone, G. H., Merzenich, M. M., Jenkins, W. M., Grajski, K. A. & Dinse,H. R. Topographic reorganization of the hand representation in cortical area 3b of owl monkeys trained in a frequency-discrimination task. J. Neurophysiol. 67, 1031–1056 (1992).

4. Recanzone, G. H., Schreiner, C. E. & Merzenich, M. M. Plasticity in thefrequency representation of primary auditory cortex followingdiscrimination training in adult owl monkeys. J. Neurosci. 13, 87–103(1993).

5. Hubel, D. H. & Wiesel, T. N. Functional architecture of macaque monkeyvisual cortex. Proc. R. Soc. Lond. B Biol. Sci. 198, 1–59 (1977).

6. Gilbert, C. D. Early perceptual learning. Proc. Natl. Acad. Sci. USA 91,1195–1197 (1994).

7. McKee, S. P. & Westheimer, G. Improvement in vernier acuity with practice.Percept. Psychophys. 24, 258–262 (1978).

8. Poggio, T., Fahle, M. & Edelman, S. Fast perceptual learning in visualhyperacuity. Science 256, 1018–1021 (1992).

9. Fahle, M. & Edelman, S. Long-term learning in vernier acuity: effects ofstimulus orientation, range and of feedback. Vision Res. 33, 397–412 (1993).

10. Kapadia, M. K., Gilbert, C. D. & Westheimer, G. A. Quantitative measure forshort-term cortical plasticity in human vision. J. Neurosci. 14, 451–457(1994).

11. Levi, D. & Polat, U. Neural plasticity in adults with amblyopia. Proc. Natl.Acad. Sci. USA 93, 6830–6834 (1996).

12. Crist, R. E., Kapadia, M. K., Westheimer, G. & Gilbert, C. D. Perceptuallearning of spatial localization: specificity for orientation, position andcontext. J. Neurophysiol. 78, 2889–2894 (1997).

13. Judge, S. J., Richmond, B. J. & Chu, F. C. Implantation of magnetic searchcoils for measurement of eye position: an improved method. Vision Res. 20,535–538 (1980).

14. Wolbarsht, M. L., MacNichol, E.F. Jr. & Wagner, H. G. Glass insulatedplatinum microelectrode. J. Physiol. (Lond.) 132, 1309–1310 (1960).

15. Von der Heydt, R. & Peterhans, E. Mechanism of contour perception inmonkey visual cortex. I. Lines of pattern discontinuity. J. Neurosci. 9,1731–1748 (1989).

16. Snodderly, D. M. & Gur, M. Organization of striate cortex of alert, trainedmonkeys (Macaque Fascicularis): ongoing activity, stimulus selectivity andwidths of receptive field activating regions. J. Neurophysiol. 71, 2100–2125(1995).

17. Van Essen, D. C., Newsome, W. T. & Maunsell, J. H. R. The visual fieldrepresentation in striate cortex of the macaque monkey: asymmetries,anisotropies and individual variability. Vision Res. 24, 429–448 (1984).

18. Dow, B. W., Synder, A. Z., Vautin, R. G. & Bauer, R. Magnification factor andreceptive field size in foveal striate cortex of the monkey. Exp. Brain. Res. 44,213–228 (1981).

19. Knierim, J. J. & Van Essen, D. C. Neuronal responses to static texture patternsin area V1 of the alert macaque monkey. J. Neurophysiol. 67, 961–980 (1992).

20. Kapadia, M. K., Ito, M., Gilbert, C. D. & Westheimer, G. Improvement invisual sensitivity by changes in local context: parallel studies in humanobservers and in V1 of alert monkeys. Neuron 15, 843–856 (1995).

21. Kapadia, M. K., Westheimer, G. & Gilbert, C. D. The spatial distribution ofexcitatory and inhibitory context interactions in primate visual cortex. J. Neurophysiol. 84, 2048–2062 (2000).

22. Kapadia, M. K., Westheimer, G. & Gilbert, C. D. Dynamics of spatialsummation in primary visual cortex of alert monkeys. Proc. Natl. Acad. Sci.USA 96, 12073–12078 (1999).

23. Ito, M. & Gilbert, C. D. Attention modulates contextual influences in theprimary visual cortex of alert monkeys. Neuron 22, 593–604 (1999).

24. Vogels, R. & Orban, G. A. Activity of inferior temporal neurons duringorientation discrimination with successively presented gratings.J. Neurophysiol. 71, 1428–1451 (1994).

25. Moran, J. & Desimone, R. Selective attention gates visual processing in theextrastriate cortex. Science 229, 782–784 (1985).

26. Dill, M. & Fahle, M. Limited translation invariance of human visual patternrecognition. Percept. Psychophys. 60, 65–81 (1998).

27. Gilbert, C. D., Hirsh, J. A. & Wiesel, T. N. Lateral interactions in visual cortex.Cold Spring Harb. Symp. Quant. Biol. 55, 663–677 (1990).

28. Kaas, J. H. et al. Reorganization of retinotopic cortical maps in adultmammals after lesions of the retina. Science 248, 229–231 (1990).

29. Heinen, S. J. & Skavenski, A. A. Recovery of visual responses in foveal V1neurons following bilateral foveal lesions in adult monkey. Exp. Brain Res. 83,670–674 (1991).

30. Ts’o, D. Y., Gilbert, C. D. & Wiesel, T. N. Relationships between horizontalinteractions and functional architecture in cat striate cortex as revealed bycross-correlation analysis. J. Neurosci. 6, 1160–1170 (1986).

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

32. Gilbert, C. D. & Wiesel, T. N. Columnar specificity of intrinsic horizontalconnections in cat visual cortex. J. Neurosci. 9, 2432–2442 (1989).

33. Das, A. & Gilbert, C. D. Receptive field expansion in adult visual cortex islinked to dynamic changes in strength of cortical connections. J. Neurophysiol. 74, 779–792 (1995).

34. McGuire, B. A., Gilbert, C. D., Rivlin, P. K. & Wiesel, T. N. Targets ofhorizontal connections in macaque primary visual cortex. J. Comp. Neurol.305, 370–392 (1991).

35. Hirsch, J. A. & Gilbert, C. D. Synaptic physiology of horizontal connectionsin the cat’s visual cortex. J. Neurosci. 11, 1800–1809 (1991).

36. Weliky, M., Kandler, K., Fitzpatrick, D. & Katz, L. C. Patterns of excitationand inhibition evoked by horizontal connections in cortex share a commonrelationship to orientation columns. Neuron 15, 541–552 (1995).

articles©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


articles


For most species, the analysis of visual motion is essential forextracting information about the surrounding environment1,2.Knowing what direction and how fast something moves is oftencritical for capturing prey and for avoiding capture. Visual motioninformation also aids in the perceptual reconstruction of the spa-tial layout of the environment around us as we navigate throughthe world3–5. Neurons in the middle temporal (MT) area of pri-mate visual cortex are specialized for visual motion extraction6,7.Research on the MT neurons’ properties has largely concentrat-ed on examining how they encode the direction of moving fea-tures (for example, refs. 8–11)—much less is known about howthey process the speed of image movement.

The visual speed signals from area MT have been implicatedin a number of important behavioral tasks, such as the visualpursuit of moving targets12,13 and the determination of self-motion5,14. Understanding the speed-tuning characteristics ofMT neurons is therefore crucial for determining how the brainperforms these tasks. The speed tuning of MT neurons has beenassessed using moving bars or random dot patterns across a rangeof speeds15–18, and the resulting tuning curves were often quitepeaked15. This was taken as evidence that these neurons are tunedfor the speed of image motion.

However, the possibility remains that MT neurons are tuned tothe temporal frequency of changes in the light intensity patternrather than specifically to the speed of the moving feature. Con-sider, for example, neurons in the primate primary visual cortex(V1). If tested with a moving bar or edge, some primate V1 neu-rons will respond more strongly to a particular speed of thebar/edge19, so one could mistakenly conclude that these neuronsare speed tuned. However, the response of these V1 neurons alsodepends on the spatial structure of the stimulus (the spatial fre-quency); they are not uniquely responding to the image speed20.

Changes to either the spatial or temporal frequency will influencetheir firing rates. A neuron truly encoding image speed willrespond to a particular speed, regardless of the spatial frequencycontent in the stimulus. It is currently not clear whether MT neu-rons have this property, although preliminary evidence suggeststhat some may fit in this category (W.T. Newsome, M.S. Gizzi &J.A. Movshon, Invest. Opthal. Vis. Sci. Suppl. 24, 106, 1983; J.A. Movshon et al., Invest. Opthal. Vis. Sci. Suppl. 29, 327, 1988).

The analysis and understanding of visual motion stimuli canbenefit from a transformation into the spatiotemporal or fre-quency domain21. Moving edges are a common part of our visu-al environment and can be analyzed in terms of their spatial andtemporal frequency content. According to Fourier theory22, astatic edge can be synthesized from a combination of two-dimen-sional intensity sine-wave components (Fig. 1a and b). Becausethe edge is not moving, all the different sinusoids have tempo-ral frequencies of 0 Hz. In a plot of spatial frequency versus tem-poral frequency (Fig. 1c), the amplitude spectrum for this staticedge would fall on a horizontal line at 0 Hz.

When the edge moves from right to left at a particular speedv, the low-frequency sine-wave components of the edge willhave a low temporal frequency. However, to ‘keep in step,’ thehigh spatial frequency components will necessarily have a hightemporal frequency23. An edge moving leftward at speed v willtherefore have a spectrum with a slope equal to v when plot-ted in spatiotemporal frequency space (Fig. 1c)21,24. The spec-trum necessarily passes through (0, 0). The overall speed ofeach sine-wave component is equal to the temporal frequen-cy divided by the spatial frequency. If the speed of the edgeincreases, the spatial frequency content remains the same, butthe temporal frequency of each component increases by anamount proportional to the spatial frequency, and thus, the

Speed skills: measuring the visualspeed analyzing properties ofprimate MT neurons

John A. Perrone1 and Alexander Thiele2,3

1 Department of Psychology, The University of Waikato, Private Bag 3105, Hamilton, New Zealand2 The Salk Institute for Biological Studies, 10010 N. Torrey Pines Rd., La Jolla, California 92037, USA 3 Present address: Department of Psychology, University of Newcastle upon Tyne, Ridley Building, Claremont Place, Newcastle upon Tyne,

NE1 7RU, UK

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

Knowing the direction and speed of moving objects is often critical for survival. However, it is poorlyunderstood how cortical neurons process the speed of image movement. Here we tested MTneurons using moving sine-wave gratings of different spatial and temporal frequencies, and mappedout the neurons’ spatiotemporal frequency response profiles. The maps typically had oriented ridgesof peak sensitivity as expected for speed-tuned neurons. The preferred speed estimate, derived fromthe orientation of the maps, corresponded well to the preferred speed when moving bars were pre-sented. Thus, our data demonstrate that MT neurons are truly sensitive to the object speed. Thesefindings indicate that MT is not only a key structure in the analysis of direction of motion and depthperception, but also in the analysis of object speed.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

slope of the edge spectrum increases (Fig. 1c). Therefore, anyneural mechanism (either a single sensor or an ensemble)encoding the particular speed v of the moving edge needs torespond selectively to the slope of the edge spectrum.

Such a mechanism would respond best to combinations ofspatial and temporal frequencies that fall along a straight line infrequency space (Fig. 1d). This mechanism would respond equal-ly well to sine-wave gratings of various spatiotemporal frequen-cy content as long as the speed of the sine wave was equal to v.The sensitivity profile of the mechanism in the spatiotemporalfrequency domain will be referred to as the ‘spectral receptivefield.’ The spectral receptive field is the amplitude part of theFourier-transformed spatiotemporal receptive field, and it givesan indication as to which spatial and temporal frequencies willcause the mechanism to respond. We will refer to the hypothet-ical speed-tuned mechanism as the ‘oriented spectral receptivefield hypothesis.’ Such oriented mechanisms are often referredto as being ‘inseparable’ because one cannot generate them bysimply multiplying together two separate spatial and temporalfrequency amplitude response functions.

If MT neurons are truly speed tuned, they should possess spec-tral receptive fields that are oriented relative to the spatial andtemporal frequency axes. Moreover, one should find a range ofspectral receptive field orientations, each corresponding to thepreferred speed of the individual MT neuron15–18. The main goalof this study was to look for any sign of orientation. We thereforemapped out the spectral receptive fields of MT neurons to lookfor evidence of oriented (inseparable) spectral receptive fields.Such evidence would confirm that these neurons are truly sensi-tive to the speed of moving edges and are not just responding tothe temporal frequency component of the motion. We found thata large proportion of our MT neurons had oriented (inseparable)spectral receptive fields, and that this orientation was closely linkedto a neuron’s speed tuning when tested with a moving bar.

RESULTSInseparable versus separable spectral receptive fieldsContour plots (Fig. 2) from four representative MT neurons inour sample (n = 84) exhibit regions of peak sensitivity that are

clearly oriented relative to the spatial and temporal frequencyaxes. Therefore, these data support the oriented spectral recep-tive field hypothesis. To quantify the degree of orientation in allof our cells, we fitted the 30 grating responses using two differenttypes of two-dimensional Gaussian functions: non-oriented andoriented (see Methods). The latter included an extra parameter,which rotated the Gaussian around its peak by an angle θ mea-sured relative to the vertical. The non-oriented fit function isequivalent to the oriented function when θ = 0°. In both cases,the best fit between the particular Gaussian and the MT-nor-malized responses (for example, Fig. 3) was carried out usingleast-squares minimization (see Methods).

The correlation coefficient (r) was calculated as an overallmeasure of the degree of fit. In Fig. 3, the non-oriented fit andoriented fit r-values were 0.75 and 0.88, respectively. Across ourwhole population, the means ± s.d. of the r-values for the non-oriented and oriented fits were0.80 ± 0.13 and 0.86 ± 0.10,respectively. For each cell, wetested the hypothesis that the θparameter in the oriented-fitmodel was 0°, using a non-lin-ear regression analysis that gen-erates a Student t-statistic25 (seeMethods). A significant t-valueindicated that we could rejectthe hypothesis that the θ para-meter value is 0°.

For the neuron in Fig. 3, thet-value was 5.9, which exceed-ed the critical value, t = 2.06(24 df, p < 0.005, two-tailed).Thus, for this neuron, we couldreject the hypothesis Ho: θ = 0°.For our population of MT neu-rons, the mean t-value was 3.12 ± 3.3 (s.d.), and the medi-an was 2.59. Most (61%) neu-rons generated t-values that

Spatial frequency (cycles/degree)

0

0

Tem

pora

l fre

quen

cy (

Hz)

0

0

10–10

10

–10

v + ∆v v

v

Oriented spectral receptivefield hypothesis

Fig. 1. Representation of a moving edge in the spatiotemporal frequencydomain. (a) Edge moving from right to left at speed v (degrees/s). (b) Stylized Fourier sine-wave components making up the edge profile.The different sinusoids generate different temporal frequencies, becausethe temporal frequency is a product of the speed (v) and the spatial fre-quency of the sinusoid. High spatial frequency sinusoids will have highertemporal frequencies than the low spatial frequency sinusoids. (c) Spatialversus temporal frequency plot, with the shaded line representing theFourier amplitude spectrum of the moving edge. The length of the line hasbeen truncated to better indicate changes to its position. The spectrumhas a slope equal to v and passes through (0, 0). If the speed of the edgeincreases by ∆v, the temporal frequency of each sinusoid increases by anamount proportional to the spatial frequency. This results in a change inthe slope or orientation of the spectrum (dashed lines). (d) A mechanismsensitive to particular orientations of the edge spectrum (that is, thespeed) would be expected to have a region of peak sensitivity that is elon-gated and oriented relative to the spatial and temporal frequency axes.

a b

c d

1 2 3 4 5


Tem

pora

l fre

quen

cy (

Hz)

2

4

6

8

10

12

14

16

70

75

80

85

90

95

100

2

4

6

8

10

12

14

16

20

25

30

35

15

2

4

6

8

10

12

14

16

0

1

2

3

4

5

6

7

2

4

6

8

10

12

14

16

810121416182022

Fig. 2. Data from four representative MT neurons in our sample. The contour plots show the responses to sine-wavegratings moving in the neurons’ preferred direction. The plots were created from the responses to 30 different grat-ing patterns based on combinations of 6 spatial and 5 temporal frequencies (see Methods). Only the upper-right quad-rant of spatiotemporal frequency space is depicted in these plots. Shaded vertical bars on right, average response ofthe neurons (see Methods) in each region of the plot. The regions of peak sensitivity (white) are oriented relative tothe spatial and temporal frequency axes, and therefore support the oriented spectral receptive field hypothesis.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


exceeded the critical level (Fig. 4). An oriented two-dimensionalGaussian provided a better description of these data than a non-oriented Gaussian. However, for a proportion of neurons (∼ 40%),the distinction between oriented and non-oriented was not dis-cernable, a result consistent with earlier reports of the prevalenceof spectral receptive field orientation in MT (W.T. Newsome,M.S. Gizzi & J.A. Movshon, Invest. Opthal. Vis. Sci. Suppl. 24,106, 1983; J.A. Movshon et al., Invest. Opthal. Vis. Sci. Suppl. 29,327, 1988). Possible reasons for this result are discussed below.

The spectral receptive fields of a large proportion of our MTneuron population tended to have a ridge of peak sensitivity ori-ented relative to the spatial and temporal frequency axes. An ori-ented Gaussian that is not aligned with the spatial or temporalfrequency axes comes under the category of an inseparable two-dimensional function. We therefore conclude that most of ourMT spectral receptive fields are better described as inseparablerather than separable. Because of the imprecision of our two-dimensional Gaussian fitting procedure (see Methods) and ourlimited sampling of the spatial and temporal frequency dimen-sions, we cannot be certain of the exact proportion of ‘oriented’versus ‘non-oriented’ spectral receptive fields in our sample.

Spectral receptive field orientation and shape analysisWe next analyzed the aspect ratio (σy/σx) and the orientation (θ)of the best-fitting oriented two-dimensional Gaussians, to see ifthey were consistent with the oriented spectral receptive fieldhypothesis. If the spectral receptive fields are elongated as sug-gested by the hypothesis, the spread of the fitted Gaussian alongthe θ direction (σy´) should be larger than the spread in the 90° + θ direction (σx´), that is, σy´/σx´ > 1. Furthermore, if MTneurons possess spectral receptive fields tuned for the orientedspectra generated by moving edges, then we would expect thereceptive field spectra in our contour plots to have a major axisthat is rotated in a clockwise direction relative to the vertical (tem-poral frequency) axis (Fig. 1d). This means that the best-fittingoriented two-dimensional Gaussian should have a θ parameterthat is positive rather than zero or negative.

For our whole sample of MT neurons, the median value ofσy´/σx´ for the oriented two-dimensional Gaussian fits was 7.2(mean ± s.d, 26.3 ± 69.5). The σy´/σx´ ratio for the oriented fit inFig. 3 was 10.01. We can reject the hypothesis that the averageshape of the spectral receptive fields was circular, that is, σy´/σx´ = 1 (t = 3.3, df = 83, p < 0.001, two-tailed). Overall, thespectral receptive fields were elongated in the direction expectedby the oriented spectral receptive field hypothesis. The mean value

for the orientation parameter (θ) was 6.1 ± 8.5°, which was sig-nificantly different from 0° (t = 6.57, 83 df, p < 0.001, two-tailed).

Under the oriented spectral receptive field hypothesis, theridges of peak sensitivity of the MT neurons’ spectral receptivefields should fall on lines of different slopes passing through theorigin (Fig. 5a). More formally, θ should equal β in each case,where β = 90° – atan (y/x). Thus, ideally, the difference betweenβ and θ should equal zero. As expected, the distribution of thesedifferences is centered around zero (Fig. 5b). The median of theerror distribution was 1.86°, and the mean was 5.4 ± 14.4°, whichwas significantly different from 0° (t = 3.43, 83 df, p < 0.001, two-tailed). This difference can be attributed to the fact that, inmany cases, the oriented two-dimensional Gaussian is not nec-essarily the best description of the MT neuron’s spectral recep-tive field. The two-dimensional Gaussian function is symmetricabout its main axes, and, yet, most of the spectral receptive fieldswe observed in our sample were asymmetric. In some cases, theoriented Gaussian fits were able to accommodate the asymme-try in the MT maps by incorporating a value of θ that was clos-er to the vertical than the true orientation of the spectral receptivefield. Although the oriented fits performed better than the non-oriented fits, they are still not optimal descriptors of the MT neu-ron spectral receptive fields, and this may be reflected in someof the β − θerrors.

Examination of the θ angle distribution in Fig. 5a revealsanother difficulty we faced in attempting to verify the orientedspectral receptive field hypothesis. Most neurons (51%) had spec-tral receptive fields that were consistent with sensitivity to highrates of edge speed (> 8°/s). This is in keeping with earlier find-ings that show that MT neurons tend to be tuned to fairly highspeeds15–17. It means that most MT neurons will have spectralreceptive fields that have a primary orientation close to the ver-tical and therefore will be hard to distinguish from non-orientedspectral receptive fields. This may explain why earlier studieshave reported mixed results in their search for oriented (insepa-

1 2 3 4 50

0.2

0.4

0.6

0.8

1 2 3 4 5

2

4

6

8

10

12

14

16

1 2 3 4 5

MT Neuron data Non-oriented fit Oriented fit


Tem

pora

l fre

quen

cy (

Hz)

Fig. 3. Example of Gaussian fitting procedure. Left, response data incontour plot form from another of the MT neurons in our sample.Data are normalized relative to the maximum response across the 30test gratings; range, 0 to 1.0 (bar on right). Middle, best-fitting non-ori-ented (NO) two-dimensional Gaussian (see Methods) for this cell.Right, best-fitting oriented (O) two-dimensional Gaussian. The (x, y)location of the peaks were 1.0 cycles/degree, 10.2 Hz and 1.6cycles/degree, 13.9 Hz for the NO and O fits respectively. The σx, σyvalues were 1.64, 8.77 for the NO fit and 1.4, 14.4 for the O fit. Thepedestal values (p) were 0.96 (NO) and 0.88 (O). For the O fit, thevalue of the orientation parameter (θ) was 6.5°.

–2 0 2 4 6 8 10 12 14 16 18 200

2

4

6

8

10

12

14

16

18

20

Num

ber

of n

euro

ns

Student t- value

Fig. 4. Contribution of the orientation parameter (θ) to the two-dimen-sional Gaussian-MT data fits. Student t-values greater than 2.06 (blackbars) indicate that the orientation parameter significantly contributed toa good fit. This was the case for most of the MT neurons (51/84 = 60.7%).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

rable) fields (W.T. Newsome, M.S. Gizzi & J.A. Movshon, Invest.Opthal. Vis. Sci. Suppl. 24, 106, 1983; J.A. Movshon et al., Invest.Opthal. Vis. Sci. Suppl. 29, 327, 1988). Fortunately, our sampleincluded sufficient neurons tuned to slower speeds, and thesehelped reveal the orientation.

MT neuron speed responses to moving barsMoving edges and bars have spectra containing a broad range ofspatial frequencies (that is, they are ‘broad-band’). The speedresponse of MT neurons to broad-band stimuli should thereforedepend on how well the edge/bar spectrum aligns with the spec-tral receptive field of the neuron. In theory, neurons tuned toslow edge/bar speeds should have a shallow spectral receptivefield orientation (for example, Fig. 2, top) and those tuned tohigher edge/bar speeds should have a steeply oriented spectralreceptive field (for example, Fig. 2, bottom). This is an obviousbut critical prediction of the oriented spectral receptive fieldhypothesis that has never been tested experimentally before.Therefore, we measured the preferred bar speed (see Methods)along with the spatiotemporal spectral receptive fields for a sub-set (n = 48) of our MT neurons.

For the four examples shown in Fig. 6, the speed tuning inresponse to a moving bar is closely related to the orientation ofthe spectral receptive field. This general trend was apparent acrossthe subset of our MT neurons tested using moving bars (Fig. 7).The fitted line is given by log V´ = 0.42 log V + 1.66 (r = 0.58, r2 = 0.34, F = 11.1, 1, 22 df, p < 0.005). A lot of the noise in thedata can be attributed to the fact that small errors in the θ esti-mate translate into large errors in the speed estimate (V´) becauseof the tangent function linking θ and V´. This is especially so forvalues of θ close to 0°, that is, the most common value of θ inour sample. Overall, however, the slope of the regression line waspositive. Thus, we confirm that the orientation of an MT neu-ron’s spectral receptive field is closely tied to the optimum speedsensitivity of the neuron when tested with broad-band stimulisuch as moving bars.

DISCUSSIONOur data show that many MT neuronshave oriented (inseparable) spectral recep-tive fields that enable them to respondselectively to particular spatiotemporal fre-quency combinations, that is, to a certainspeed of stimulus movement. By examin-ing the responses of the neurons in spa-tiotemporal frequency space, weconfirmed that these neurons have prop-erties that are closely matched to the mostcommon stimulus they encounter, that is,moving edges. In spatiotemporal frequen-cy space, a moving edge has a spectrumthat falls on a line oriented relative to thespatial and temporal frequency axes21,24.The elongated and oriented spectral recep-tive fields of the MT neurons in our sam-ple are ‘tuned’ for this type of stimulus.Although it has long been suspected thatthis could be the case (for example, W.T. Newsome, M.S. Gizzi & J.A. Movshon, Invest. Opthal. Vis. Sci.Suppl. 24, 106, 1983), our data provide thefirst clear evidence of this type of tuning.

Having demonstrated that many MTneurons possess spectral receptive fields

that are oriented and inseparable, we are still left with the ques-tion as to how these receptive fields are constructed. The neuronsat an earlier stage of the visual motion pathway (V1) do not haveoriented spectral receptive fields20. Yet somehow, by the time wereach MT, the neurons have acquired spectral receptive field prop-erties that are suitable for true speed selectivity. We are currentlytesting our MT data against a speed mechanism model based onthe combined inputs from just two V1 neurons, one with sus-tained temporal frequency tuning and the other with transienttuning (J.A. Perrone, Soc. Neurosci. Abstr. 24, 789.9,1998; J.A. Per-rone & A. Thiele, Invest. Opthal. Vis. Sci. Suppl. 41, S720, 2000).

Our MT neuron data, and the earlier speed tuning results15–18,put tight constraints on theories concerning the type of speedinformation that is available at the level of MT. As in otherdomains (such as wavelength coding or direction coding), a sin-gle neuron with band-pass tuning cannot unambiguously indicatehow much of a particular stimulus dimension is present. There-fore, the individual MT neurons cannot signal (for example, usinga rate code) that something is moving at a particular speed (forexample, 10 degrees/s). By way of illustration, as a consequence ofthe shape of its speed-tuning curve, the neuron in Fig. 6b wouldproduce an output of about 70 impulses/s in response to barspeeds of both 8 and 16 degrees/s. A change in the contrast ofthe bar would further complicate this link between firing rateand bar speed. Therefore an estimate of the actual edge speed—presumably from some sort of population code (for example, ref. 26)—would have to be derived at a neural stage after MT. Ofcourse, this limitation only applies to speed estimation systemsbased around the MT spectral receptive field properties. Visualspeed estimates could also be derived by the visual systemthrough additional means (such as feature tracking27).

The MT spectral receptive field data reported here reveal thata simple matching strategy has been adopted by the primate visu-al system to register the speed of moving edges. The individualspectral receptive fields act as ‘templates’ for a particular speed ofedge movement (for example, Fig. 6). If the edge speed matches


Tem

pora

l fre

quen

cy (

Hz)

Per

cent

age

of c

ells

-40 -20 0 20 40 60 800

5

10

15

20

25

30

35

40

Orientation error (degrees)

0 1 2 3 4 5 60

2

4

6

8

10

12

14

16

1˚/sec

2˚/sec

4˚/s8˚/s16˚/s

Fig. 5. Test of the oriented two-dimensional Gaussian alignment. (a) Location (x, y) of the peaksfor the best fitting oriented two-dimensional Gaussian for all cells in our sample (open circles).Lines through circles, orientation of the best-fitting Gaussian (θ). Dotted lines correspond tothe location of the amplitude spectra that would be generated by moving edges; speeds areshown on the right and top of figure. If the oriented spectral receptive field hypothesis is true,the best-fitting Gaussians should have main axes that pass through (0, 0). (b) Error distributionshowing how much the orientation of each best-fitting Gaussian (θ) deviates from the angle (β)specified by the oriented spectral receptive field hypothesis. Both angles are measured from thevertical. The errors are equal to (β − θ) where β = 90° – atan (y/x). The bin sizes are 5° in width.Most cells are closely aligned to the lines specified by the hypothesis.

a b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


the optimum speed tuning of the neuron, a large output results. Ifthe speed is too fast or slow for the cell, the spectrum of the mov-ing edge does not line up or match the MT spectral receptive fieldvery well, and a low response is generated. A similar template-matching scheme seems to be at work in a motion processing areabeyond MT as well. The neurons in the medial superior temporalregion (MST) of primate cortex are capable of acting as templatesfor the global patterns of image motion that occur during self-motion5,14,28,29. The template concept is a pervasive and simpleone, but in the case of MT neurons, it is not obvious that it isbeing applied; the nature of the neuron-stimulus match onlybecomes apparent in the spatiotemporal frequency domain.

METHODSAnimal subjects. We used two adult rhesus monkeys (Macaca mulatta,1 male, 1 female) in this study. Experimental protocols were approvedby the Salk Institute Animal Care and Use Committee, and conform toUSDA regulation and to NIH guidelines for the humane care and use oflaboratory animals. Details about procedures for surgery, wound main-tenance and the behavioral fixation task are provided elsewhere30,31.

Electrophysiological recordings. Details about procedures for recordingextracellular action potentials from isolated cortical neurons are routine,and have been described repeatedly elsewhere (for example, ref. 32). Fourcriteria were used to determine whether neurons were recorded from MT:selectivity for the direction of stimulus motion, consistency of retinotopicorganization with known topography, consistency of receptive fields withknown dependence on visual field eccentricity, and consistency of elec-trode positions (determined from activity pattern while advancing intoMT) with expected sulcal topography. In addition, stereotaxic MRI scansobtained before surgery were used to further confirm that our recordingswere in a region of cortex consistent with the typical location of area MT.

Apparatus. Visual stimuli were generated using a SGT Pepper Graphicsboard (Number Nine Computer Corporation, 640 × 480 pixel resolu-tion on 27 × 20.25 cm, analog RGB output, 8 bits/gun) residing in a Pen-tium-based PC. Stimuli were displayed on an analog RGB monitor (SonyGDM 2000TC, 60 Hz, non-interlaced) at a distance of 63 cm. CORTEX 5.7 (Laboratory of Neuropsychology, NIMH) was used for dataacquisition, behavioral control and stimulus generation. Monitor out-put was linearized for each of the three phosphors independently33.

Determination of basic response properties. Initially, each isolated MTneuron was mapped while monkeys fixated centrally. Thereafter, direc-tion tuning was assessed using a sinusoidal grating (0.7 cycles/degree, 4 Hz, 100% Michelson contrast) moving in each of eight different direc-tions (along cardinal and oblique axes) centered on the receptive field.

0

20

40

60

80

100

Mea

n R

espo

nse

(Im

p/se

c)

5

10

15

20

2

4

6

8

10

12

14

16

15

20

25

30

2

4

6

8

10

12

14

16

5

10

15

20

25

30

2

4

6

8

10

12

14

16

Bar speed (degrees/s)

20

40

60

80

100

2

4

6

8

10

12

14

16

1 2 3 4 5


Tem

pora

l fre

quen

cy (

Hz)

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

1 2 4 8 16 32 64

Fig. 6. Relationship between the spectral receptive field orientation ofrepresentative MT neurons and their preferred speed tuning obtainedfrom moving bar tests (see Methods). Left, spectral receptive fields fromfour neurons in our sample derived using moving sine-wave gratings. Theorientation of the ridge of peak sensitivity relative to the spatial and tem-poral frequency axes was found from the θ parameter in the best-fittingoriented two-dimensional Gaussian (see Methods). Under the orientedspectral receptive field hypothesis, θ should be directly related to the opti-mum speed tuning (V´) of the neuron, that is, V´ = tan (90° – θ) becauseV´ = temporal frequency/spatial frequency. Solid arrows in the right panelsindicate the calculated value of V´. These panels show the speed tuningcurves for the same neurons obtained using a moving bar (see Methods).The dashed arrows indicate the peak of the tuning curves. From top tobottom, the peak and V′ values (in degrees/s) were as follows: 4, 5.1; 8,7.6; 16, 11.5; 64, 301 (arrow position truncated). In general, neurons witha shallow spectral receptive field slope were tuned to slow bar speeds,and those with a steep slope were tuned to high bar speeds.

Log optimum speed tuning from bar stimuli (degrees/s)

Log

spee

d tu

ning

est

imat

e fr

omsp

ectr

al r

ecep

tive

field

orie

ntat

ion

(deg

rees

/s)

–1 0 1 2 3 4 5–1

0

1

2

3

4

5Fig. 7. Prediction of the spectral receptive field orientation of MT neu-rons from their optimum speed tuning, obtained using a moving bar. Asubset (48/84) of the neurons in our sample were tested with a movingbar in addition to the standard sinusoidal grating tests. This producedspeed tuning curves from which the optimum speed (V) of the neuronwas estimated by fitting a one-dimensional Gaussian to the speed tuningdata (see Methods). The log-transformed V estimates have been plottedagainst log V´, where V´ is the speed tuning estimate based on the ori-entation of the spectral receptive field of the neuron (see Methods). Astandard linear regression analysis was carried out to see if V´ is pre-dictable from V. This analysis was done for all neurons in our sample(24/48) that produced an oriented Gaussian fit with an r-value greaterthan 0.8 (the average for the non-oriented Gaussian fits) and a fit to thespeed-tuning data, which also exceeded r = 0.8.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

The direction of motion that yielded the largest response was termed the‘preferred’ direction, and the direction opposite to the preferred wastermed the ‘null’ direction. Strength of directional bias along the pre-ferred–null axis was quantified by a direction index (DI): DI = 1 – ND/PD,where PD and ND are firing rate changes elicited by motion in preferredand null directions, respectively (after subtraction of background activi-ty). Neurons with a DI < 0.5 were excluded from further study, that is, all84 neurons used in the current study were directionally selective.

Mapping the spectral receptive field. Stimuli consisted of moving yel-low–black sinusoidal gratings with a mean luminance of 24 cd/m2. Theywere presented on a yellow background (24 cd/m2), with CIE coordi-nates x = 0.492, y = 0.446. Stimuli were viewed within a square apertureand moved along one of the cardinal or oblique axes. Aperture widthwas 5° for neurons with foveal and parafoveal receptive fields, and 10° ifa neuron’s receptive field diameter was larger than 5°.

Thirty different spatiotemporal frequency combinations moving in thepreferred direction were used to determine the spectral receptive field(temporal frequencies, 1, 2, 4, 8 or 16 Hz; spatial frequencies, 0.2, 0.4, 0.7,1.4, 2.8 or 5.6 cycles/degree). Luminance modulation of sinusoidal stim-uli was set to 10% Michelson contrast. The choice of a relatively low con-trast resulted from initial testing with high (100%) luminance contrast.Stimuli of 100% luminance contrast activate MT neurons strongly at mostspatiotemporal frequencies, thereby potentially concealing the orientedspectral receptive field. Weaker stimuli failed to activate neurons if spa-tiotemporal frequencies were non-optimal. The neuron’s spatiotemporalfrequency tuning could therefore be recovered more easily. Neuronal datawere included into the sample if 5–8 trials per condition were recorded.

Data analysis. For each neuron, we calculated the mean activity for eachstimulus condition (n = 30) over a 1000-ms window, which began 50 msafter stimulus onset and ended 50 ms after stimulus offset. These meanswere then used to construct the spatiotemporal frequency tuning andplotted in contour plot form (for example, see Fig. 2). The means werenormalized relative to the peak value to give a 6 × 5 array of responses R.These 30 values were fitted with a two-dimensional Gaussian function:G(u, ω) = (exp (–(u´)2/σx

2)) × (exp (–(ω´)2/σy2)) + p, where

u´ = (u – x) cos θ + (ω – y) sin θ and ω´ = –(u – x) sin θ + (ω – y) cos θ.Here, u is the spatial frequency of the test grating, ω is the temporal fre-quency, (x, y) is the location of the peak of the Gaussian (in u, ωcoordi-nates), σx and σy is the spread of the Gaussian in the u´ and ω dimensions,respectively. A constant value (p) is added, and then the G values are nor-malized relative to the maximum. The values of x, y, σx, σy and p wereoptimized using fminsearch in MatLab (MathWorks, Natick, Massachu-setts) to minimize the sum of the squared deviations (see below) betweenthe thirty R and G values. Two versions of the two-dimensional Gaussianwere fitted: non-oriented, in which θ was constrained to equal 0°, andoriented, where θ was free to take on any value. For all neurons, the degreeof fit between the final optimized Gaussian function values and the 30MT responses was measured using the mean-squared error.

Here, Rj is the MT neuron response at spatial and temporal frequency(uj, ωj), and Gj is the value of the Gaussian model (non-oriented or ori-ented) at (uj, ωj). We calculated the correlation coefficient (r) as a directmeasure of the fit between the Gaussian values and the MT responses.We tested the hypothesis that θ = 0° in the oriented model case by cal-culating the 95% confidence intervals on the nonlinear least-squaresparameter estimate of θ. These confidence intervals can be used to derivea t-value that reflects the probability that the θ-value significantly con-tributed to the fit25. This confidence interval estimation process relies,to a certain extent, on the assumption of equal variance across the Rj val-ues. However, it has been repeatedly reported that the standard varia-tion of MT responses (the variance) scales linearly with the mean (forexample, ref. 34). We therefore tested the robustness of our approach byrepeating the non-linear regression analysis after we had first transformedall of the Rj responses using a square-root transform. The distributionof t-values was largely unaffected by this transformation, and the pro-

1N Σ(R – Gj j )

2N

j=1

portion of neurons for which we were able to reject the θ = 0° hypothe-sis remained about the same (62% transformed, 61% non-transformed).

Speed tuning tests with moving bars. A bar 10° long and 0.2° or 0.5° wide(depending upon size of receptive field) was moved in the preferred direc-tion of the cell. Its main axis was orthogonal to this direction and had amean luminance of 22.8 cd/m2. (That is, the Michelson contrast was 10%.)We explored various speed ranges during the course of the study to try toimprove the likelihood of including the ‘peak’ in the speed tuning curves. For11 of our neurons, the bar speeds were 4, 8, 16, 32 and 64°/s. For 15 neu-rons, the test speeds were 1, 2, 4, 8, 16 and 32°/s. One neuron was tested at1, 2, 4, 8, 16, 24, 32, 48 and 56°/s, and the remainder (21) were tested at 1, 2,4, 8, 16, 24, 32, 40, 48 and 56°/s. Because the refresh rate of our monitorwas limited to 60 Hz, the problem of temporal aliasing23 was a concern, sowe treated the data generated from the higher speeds with caution. Mostneurons had a peak that was clearly defined without reliance on the highspeed (56°/s, 64°/s) data. Usually, 5–8 trials were recorded at each speedwith each trial lasting 1000 ms. The bar started moving from a position thatensured that it was at the center of the receptive field 500 ms after the startof trial. Care must be taken in selecting an appropriate response measurewhen testing neurons for speed selectivity using non-periodic stimuli suchas bars or edges17,35. Because the bar always crossed the center of the recep-tive field at a known time, we averaged the responses over a fixed time win-dow (200 ms long) and with a constant lag (50 ms) relative to the 500 msmark. The mean responses to each bar speed were then fit with a one-dimensional Gaussian for which the amplitude, mean, standard deviationand a pedestal value were optimized to produce the best fit (in a minimumleast-squares sense). The mean of the Gaussian (V) was taken as an esti-mate of the speed tuning preference of the neuron. In the fitting procedure,V was constrained to lie between 1 and 64°/s (the maximum range of our bartest speeds). The ‘peaks’ derived from the Gaussian fitting nearly all laywithin the range of bar test speeds. Only two neurons in our regressionanalysis (Fig. 7) had a mean V lower than the minimum bar speed at whichthe neuron was tested at (4°/s). A correlation coefficient (r) was used toassess the degree of fit. Speed tuning estimates were not derived for neu-rons where the Gaussian fit produced r values less than 0.8.

ACKNOWLEDGEMENTSWe thank K. Dobkins, R. Krauzlis and G. Stoner for their comments, and

J. Costanza and K. Sevenbergen for technical assistance. This work was

supported by NASA grant NAG 2-1168 to J.P. and a Human Frontier Science

Program fellowship to A.T. Some of the research reported in this paper was done

during tenure by J.P. as a Sloan Visiting Scientist at the Salk Institute.

RECEIVED 28 NOVEMBER 2000; ACCEPTED 16 MARCH 2001

1. Gibson, J. J. The Perception of the Visual World (Houghton Mifflin, Boston,1950).

2. Nakayama, K. Biological image motion processing: a review. Vision Res. 25,625–660 (1984).

3. Koenderink, J. J. & van Doorn, A. J. Invariant properties of the motionparallax field due to the movement of rigid bodies relative to the observer.Opt. Acta. 22, 773–791 (1975).

4. Longuet-Higgins, H. C. & Prazdny, K. The interpretation of moving retinalimages. Proc. R. Soc. Lond. B Biol. Sci. 208, 385–397 (1980).

5. Perrone, J. A. & Stone, L. S. A model of self-motion estimation withinprimate extrastriate visual cortex. Vision Res. 34, 2917–2938 (1994).

6. Dubner, R. & Zeki, S. M. Response properties and receptive fields of cells inan anatomically defined region of the superior temporal sulcus. Brain Res. 35,528–532 (1971).

7. Maunsell, J. H. R. & Newsome, W. T. Visual processing in monkey extrastriatecortex. Annu. Rev. Neurosci. 10, 363–402 (1987).

8. Adelson, E. H. & Movshon, J. A. Phenomenal coherence of moving visualpatterns. Nature 300, 523–525 (1982).

9. Albright, T. D. Direction and orientation selectivity of neurons in visual areaMT of the macaque. J. Neurophysiol. 52, 1106–1129 (1984).

10. Stoner, G. R. & Albright, T. D. Neural correlates of perceptual motioncoherence. Nature 358, 412–414 (1992).

11. Newsome, W. T., Britten, K. H. & Movshon, J. A. Neuronal correlates of aperceptual decision. Nature 341, 52–54 (1989).

12. Newsome, W. T., Wurtz, R. H., Dursteler, M. R. & Mikami, A. Deficits invisual motion processing following ibotenic acid lesions of the middletemporal visual area of the macaque monkey. J. Neurosci. 5, 825–840 (1985).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


13. Lisberger, S. G. & Movshon, J. A. Visual motion analysis for pursuit eyemovements in area MT of macaque monkeys. J. Neurosci. 19, 2224–2246(1999).

14. Tanaka, K. & Saito, H. Analysis of the motion of the visual field by direction,expansion/contraction, and rotation cells clustered in the dorsal part of themedial superior temporal area of the macaque monkey. J. Neurophysiol. 62,626–641 (1989).

15. Maunsell, J. H. R. & Van Essen, D. C. Functional properties of neurons in themiddle temporal visual area of the macaque monkey. I. Selectivity forstimulus direction, speed, orientation. J. Neurophysiol. 49, 1127–1147 (1983).

16. Lagae, S., Raiguel, S. & Orban, G. A. Speed and direction selectivity ofmacaque middle temporal neurons. J. Neurophysiol. 69, 19–39 (1993).

17. Rodman, H. R. & Albright, T. D. Coding of visual stimulus velocity in areaMT of the macaque. Vision Res. 27, 2035–2048 (1987).

18. Felleman, D. J. & Kaas, J. H. Receptive-field properties of neurons in themiddle temporal visual area (MT) of owl monkeys. J. Neurophysiol. 52,488–513 (1984).

19. Orban, G. A., Kennedy, H. & Bullier, J. Velocity sensitivity and directionselectivity of neurons in areas V1 and V2 of the monkey: Influence ofeccentricity. J. Neurophysiol. 56, 462–480 (1986).

20. Foster, K. H., Gaska, J. P., Nagler, M. & Pollen, D. A. Spatial and temporalfrequency selectivity of neurones in visual cortical areas V1 and V2 of themacaque monkey. J. Physiol. (Lond.) 365, 331–363 (1985).

21. Watson, A. B. & Ahumada, A. J. in Motion: Perception and Representation (ed.Tsotsos, J. K.) 1–10 (Association for Computing Machinery, New York, 1983).

22. Bracewell, R. N. The Fourier Transform and its Applications (McGraw-Hill,New York, 1978).

23. Watson, A. B., Ahumada, A. J. Jr. & Farrell, J. E. Window of visibility: apsychophysical theory of fidelity in time-sampled visual motion displays.

J. Opt. Soc. Am. A 3, 300–307 (1986).24. Fahle, M. & Poggio, T. Visual hyperacuity: spatiotemporal interpolation in

human vision. Proc. R. Soc. Lond. B Biol. Sci. 213, 451–477 (1981).25. Draper, N. R. & Smith, H. Applied Regression Analysis (Wiley-Interscience,

New York, 1998).26. Treue, S., Hol, K. & Rauber, H. J. Seeing multiple directions of motion-

physiology and psychophysics. Nat. Neurosci. 3, 270–276 (2000).27. Del Viva, M. M. & Morrone, M. C. Motion analysis by feature tracking.

Vision Res. 38, 3633–3653 (1998).28. Perrone, J. A. Model for the computation of self-motion in biological

systems. J. Opt. Soc. Am. A 9, 177–194 (1992).29. Perrone, J. A. & Stone, L. S. Emulating the visual receptive field properties of

MST neurons with a template model of heading estimation. J. Neurosci. 18,5958–5975 (1998).

30. Dobkins, K. R. & Albright, T. D. What happens if it changes color when itmoves?: The nature of chromatic input to macaque visual area MT.J. Neurosci. 14, 4854–4870 (1994).

31. Thiele, A., Dobkins, K. R. & Albright, T. D. The contribution of color tomotion processing in macaque middle temporal area. J. Neurosci. 19,6571–6587 (1999).

32. Thiele, A., Distler, C. & Hoffmann, K. P. Decision related activity in themacaque dorsal visual pathway. Eur. J. Neurosci. 11, 2044–2058 (1999).

33. Watson, A. B. et al. Use of a Raster framebuffer in vision research. Behav. Res.Meth. Instr. Comp. 18, 587–594 (1986).

34. Britten, K. H., Shadlen, M. N., Newsome, W. T. & Movshon, J. A. Responsesof neurons in macaque MT to stochastic motion signals. Vis. Neurosci. 10,1157–1169 (1993).

35. Movshon, J. A. The velocity tuning of single neurons in the striate cortex. J. Physiol. (Lond.) 249, 445–468 (1975).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



During natural viewing, certain objects (such as faces) requiredetailed central scrutiny to perform such subtle visual tasks asdetecting facial expressions and eye gaze directions. Larger objects(such as buildings or scenes) occupy a more peripheral field loca-tion, and can be recognized by their more peripheral-shape infor-mation. This distinction is further illustrated by the tendency ofscanning eye movements to fixate face parts rather than back-ground objects1. However, the potential role of this distinctionin the organization of object representations has not beenaddressed so far.

Early visual areas of primates are retinotopically organized, sothat the visual field is mapped in each area along two orthogonalaxes: polar angle and eccentricity2–6. The center/periphery orga-nization, that is, eccentricity mapping, is one of the most strikingand robust organizational principles in the primate visual cortex.Both monkey and human cortices exhibit a meta-structure of cen-ter-periphery organization, in which similar distances from thefovea are mapped in stripes that are continuous across the entireensemble of retinotopic visual areas2–8. The center/periphery orga-nization extends into higher-order visual areas, whereas the polarangle representation in these areas is cruder, and orderly repre-sentations of the visual field meridians are absent9,10. Despite theevident importance of eccentricity maps, their possible relation-ship to object recognition has received little attention, and thepossible effect of this organization on the way different object cat-egories are represented in the human brain has not been studied.

Recently, the distinction between representation of faces andbuildings has become a central issue in human visual cortex stud-ies, due to the discovery that clearly distinct cortical regions aredifferentially activated by the two image categories: buildings acti-

vate a medial region along the collateral sulcus/parahippocam-pal gyrus11–13, whereas faces activate a neighboring, more lateralregion along the posterior fusiform gyrus13–18. The segregatedrepresentation of these object categories was attributed by someauthors11 to task- or semantics-related specialization, and by oth-ers19 to their particular geometric information.

Here we report on an association between the two functionalorganizations found in human visual cortex: eccentricity maps andobject categorization. Thus, we found that face-related regions areassociated with central visual field representations, whereas build-ing-related regions are associated with peripheral field representa-tions. Furthermore, the center–periphery organization seems toencompass the entire constellation of high-order human objectareas. Within the center–periphery maps, we found a hierarchical-like organization in that posterior regions manifested higher retino-topic bias compared to more anterior regions. Thus, our resultsunify two sets of findings in human visual cortex, eccentricity map-ping and object selectivity, into a global principle of organization.

RESULTSTo explore the potential relationship between eccentricity mapsand object selectivity, we first located face-related and building-related regions in the human visual cortex (experiment 1A). Theseregions were then superimposed onto the representation of visualfield eccentricity in each subject (experiment 1B). To increase thesensitivity of high-order object areas to the visual field mapping,we constructed the retinotopic stimuli from a variety of naturalobject images (Fig. 1, see Methods20). We also mapped the hori-zontal and vertical visual field meridians that delineate borders ofretinotopic areas4,7,20,21 and superimposed the object areas on them.

articles

Center–periphery organization ofhuman object areas

Ifat Levy1,2, Uri Hasson2, Galia Avidan1,3, Talma Hendler4,5 and Rafael Malach2

1 The Interdisciplinary Center for Neural Computation, Hebrew University of Jerusalem, Jerusalem 91904, Israel2 Department of Neurobiology, Weizmann Institute of Science, Rehovot 76100, Israel 3 Department of Neurobiology, Hebrew University of Jerusalem, Jerusalem 91904, Israel4 Functional Brain Imaging Laboratory, Wohl Institute for Advanced Imaging, Tel-Aviv Sourasky Medical Center, Tel Aviv 64239, Israel5 Faculty of Medicine, Tel Aviv University, Tel Aviv 69978, Israel

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

The organizing principles that govern the layout of human object-related areas are largelyunknown. Here we propose a new organizing principle in which object representations are arrangedaccording to a central versus peripheral visual field bias. The proposal is based on the finding thatbuilding-related regions overlap periphery-biased visual field representations, whereas face-relatedregions are associated with center-biased representations. Furthermore, the eccentricity mapsencompass essentially the entire extent of object-related occipito-temporal cortex, indicating thatmost object representations are organized with respect to retinal eccentricity. A control experimentruled out the possibility that the results are due exclusively to unequal feature distribution in theseimages. We hypothesize that brain regions representing object categories that rely on detailed cen-tral scrutiny (such as faces) are more strongly associated with processing of central information,compared to representations of objects that may be recognized by more peripheral information(such as buildings or scenes).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Typically, face-related voxels were found in two foci (Fig. 2aand b): the lateral occipital region (LO) and the posterior fusiformgyrus (pFs). LO is situated ventrally and posteriorly to MT, extend-ing into the posterior inferotemporal sulcus. Region pFs is anteriorand lateral to areas V4/V8 (ref. 22), extending into the occipito-temporal sulcus, and corresponds to the fusiform face area (FFA)described previously16. Both foci largely overlapped the repre-sentation of the visual field center (Fig. 2c, yellow). Building-relat-ed voxels were found mainly in the collateral sulcus, where theypartially overlapped an upper meridian representation and extend-ed beyond it (Fig. 2a and b). This region largely overlapped theperipheral visual field representation (Fig. 2c, green) and some-times extended to the mid visual field representation (Fig. 2c, pur-ple), but always avoided the central field representation.Building-related voxels were also found in a dorsal region, in thevicinity of V3A and V7, where they often tended to overlap theperiphery and mid representations.

In all the face-related regions, activation was significantlystronger in response to central stimuli compared to mid andperipheral stimuli (Fig. 3, LO, center versus periphery, p < 0.005,center versus mid, p < 0.005, n = 12, one-tailed paired t-test;pFs, center versus periphery, p < 0.005, center versus mid, p < 0.05, n = 11, one-tailed paired t-test). In analyzing the build-ing-related regions, we included only voxels that both wereselective to buildings compared to faces, and were anterior toareas V4/V8 (Fig. 2b). This region exhibited a high preference tothe peripheral visual field representation compared to the cen-tral and mid ones (Fig. 3, Anterior CoS, periphery versus cen-ter, p < 10–5, periphery versus mid, p < 10–5, n = 12, one-tailedpaired t-test).

To test the relationship between eccentricity and object cate-gorization directly, we conducted another experiment, in whichwe mapped both center versus periphery and buildings versusfaces during one scan (experiment 2). In the center and periph-ery conditions of this experiment, subjects viewed the exact sameobjects (see Methods), such that the two conditions only differed

in the part of the visual field stimulated by the images and notin their shape features.

Again, face-related voxels were found in LO and in the pFs,where they overlapped the representation of the visual field cen-ter to a large extent, and building-related voxels were found main-ly in the collateral sulcus, where they largely overlapped theperipheral visual field representation (Fig. 4a).

The Talairach23 coordinates of the face- and building-relatedregions (Table 1) showed that our maps were in close correspon-dence to previous reports (buildings11,13,24, faces15,16,18,25). Thewhite circle in Fig. 2b shows the approximate position of face-related regions reported in early studies.

To make sure that subjects were able to recognize the objectsin the peripheral stimuli, we conducted a behavioral experimentin which subjects were required to name the central and periph-eral stimuli from experiment 2. The results showed that underthe specific task of that experiment, there was a slight trendtoward better recognition of objects in the center (mean ± s.d.,91 ± 7% correct responses) compared to the periphery (86 ± 9%correct responses).

Thus, it is clear that a consistent association exists between therepresentation of particular object images and the central versusperipheral representation. However, it should be emphasized thatthe object representations were not homogenous: a clear indica-tion of a hierarchical trend was observed, in that more posteriorregions manifested a higher eccentricity bias compared to themost anterior regions. Thus, all face-related areas exhibited a sig-nificant central bias (Fig. 4b, LO, p < 0.0005; pFs, p < 0.05, n = 5,one-tailed paired t-test, center versus periphery). However, face-related foci located in LO showed a significantly higher centralbias than those located in pFs. The ratio between activation to the

articles


Fig. 1. Stimuli used to map object-selective areas and eccentricity rep-resentations (experiments 1A and 2). Examples of stimuli used to mapthe face- and building-related areas and the center and periphery repre-sentations (see Methods for details). The center stimulus shown herewas enlarged four times compared to the actual experiment, for pre-sentation purposes.

Fig. 2. Object-selective areas and visual field eccentricity maps. Anexample of face and building-related regions in one subject. (a) Preferential activation to faces versus buildings (red) and to buildingsversus faces (blue) obtained in Experiment 1A, shown on sagittal, coro-nal and axial slices (left) and on a three-dimensional reconstructed brain(right). The color scales indicate the statistical correlation. The three-dimensional brain is shown in a ventral view. R, right; L, left; A, anterior;P, posterior. (b) The same regions from (a) are shown on the unfoldedright hemisphere. Color scales are the same as in (a). White dotted linesdenote borders of retinotopic visual areas V1, V2, V3, VP, V3A andV4/V8. The white circle surrounds the approximate locations of face-related activations reported in early studies15,16,18,25. LO, lateral occipitalregion; pFs, posterior fusiform gyrus; Ant. CoS, anterior collateral sul-cus. (c) Borders of face-related (red) and building-related (blue) regionssuperimposed on central (yellow), mid (purple) and peripheral (green)visual field representations obtained in Experiment 1B. The face-relatedregions largely overlap the central visual field representation, whereasthe building-related regions overlap the mid and peripheral ones butavoid the central visual field representation.

a

b c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



center and periphery conditions was significantly higher in LOthan in pFs (p < 0.02, one-tailed paired t-test). Activation ratio inthe most anterior part of the face region in each subject (up to 3 voxels) was not significantly different from the ratio in the entirepFs (p = 0.1). The building-related area exhibited high preferenceto the peripheral visual field representation (Fig. 4b, p < 0.002, n = 5, one-tailed paired t-test). Comparing the center/peripheryratio between the entire area and its most anterior part (up to 3 voxels), showed no significant difference (p = 0.1).

The association of faces and buildings with central and periph-eral representations may have emerged from the retinalcenter/periphery distribution of features in face and buildingimages; for example, building images may tend to contain morelow-level visual features such as edges and corners in the periph-ery than in the center. To test this possibility we conducted anoth-er experiment (experiment 3), in which subjects viewed picturesof buildings and faces as in experiment 2 (Fig. 5, ‘regular’), butalso pictures of larger faces and smaller buildings. These imageswere aimed at increasing the density of visual features in theperiphery in the case of faces, and decreasing it in the case of build-ings (Methods, Fig. 5). We compared the spectral energies of thecentral and peripheral parts of the images in each category (Meth-ods, Fig. 5) and found that in the peripheral part of the visualfield, the big-faces spectral energy was indeed higher than theenergy of the small buildings.

As expected, in low-level retinotopic areas, which containorderly representations of vertical and horizontal meridians (dot-ted lines in Fig. 6a), the activation pattern followed the retinal fea-ture distribution in the images. Thus, ‘large-face’ selective voxelstended to overlap more peripheral field representations (green)compared to ‘small-building’ selective voxels, which activatedmore central representations (yellow). However, this trend wasinverted in more anterior regions, outside the early retinotopicareas: the large-face selective voxels here overlapped central visu-

al field representations, whereas the small-buildings were associ-ated with peripheral field representations.

Another way to analyze this experiment is to select voxelsthat were preferentially activated by regular faces compared toregular buildings and those that exhibited the opposite pref-erence, and to examine their activation in response to largefaces and small buildings (which were both ignored in the sta-tistical tests). This analysis showed that face-related voxels werealso activated by large faces (mean ± s.e.m., 1.4 ± 0.1%) morethan by small buildings (0.6 ± 0.1%; p < 0.001, n = 6, one-tailed paired t-test), whereas building-related voxels wereactivated by small buildings (0.9 ± 0.1%) more than by largefaces (0.4 ± 0.1%; p < 0.005; Fig. 6b). Overall, these resultsclearly rule out the possibility that the center/periphery biasof faces and buildings is due to a difference in the retinal dis-tribution of features in the images of these objects. However,voxels in the anterior collateral sulcus, which were preferen-tially activated by buildings compared to faces, also showedsomewhat higher activation to large faces compared to regu-lar ones. This preference can be expected from the peripheralvisual field bias observed in this region.

To what extent can the center–periphery organization beextended to other object categories? To delineate the entireexpanse of object-related cortex, we used a diverse set of objectsand compared the activation produced by it with that produced bytexture patterns (experiment 1). This contrast was shown previ-

articles

Fig. 3. Activation to different eccentricities in face- and building-related areas.Average signal from twelve subjects, experiment 1. Left, face-related voxels. Voxelswere selected by applying a statistical test that searched for preferential activationfor faces versus buildings (faces > buildings). Error bars, s.e.m. Asterisk (p < 0.05)and two asterisks (p < 0.005) denote significantly weaker activation compared tothe center condition (one-tailed paired t-test; LO, n = 12; pFs, n = 11). Right, build-ing-related voxels. Voxels were selected by applying the buildings > faces test. Onlyvoxels that were outside the retinotopic areas were included. Error bars, s.e.m.Circle denotes significantly weaker activation compared to the periphery condition(p < 10–5, n = 12, one-tailed paired t-test). Abbreviations as in Fig. 2.

a

b

Fig. 4. Simultaneous mapping of object areas and eccentricity represen-tations. (a) Activation maps obtained from experiment 2 in the righthemispheres of two subjects. Borders of face- (red) and building- (blue)related areas are superimposed on central (yellow) and peripheral(green) representations. Dotted lines, borders of retinotopic visualareas. (b) Average signal from the five subjects who participated inexperiment 2. Left, face-related voxels. Voxels were selected by applyinga statistical test that searched for preferential activation for faces versusbuildings (faces > buildings). Error bars, s.e.m. Asterisk (p < 0.05) andtwo asterisks (p < 0.0005) denote significantly stronger activationelicited by central stimuli compared to peripheral ones (one-tailedpaired t-test). A significant central bias was demonstrated in all the face-related areas, although pFs showed less bias than LO. Right, building-related voxels. Voxels were selected by applying the buildings > facestest. Only voxels outside the retinotopic areas were included. Errorbars, s.e.m. Circle denotes significantly stronger activation elicited byperipheral stimuli compared to central ones (p < 0.002, one-tailedpaired t-test). Abbreviations as in Fig. 2.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


ously to be highly effective in delineating object-related cortex(the lateral occipital complex26). To maximize the statistical sen-sitivity of the test, we averaged the maps across 13 subjects (seeMethods; Fig. 7). The entire constellation of occipito-temporalobject areas stretching from the collateral sulcus medially to LOdorsally was highlighted, including face-related voxels, and a smallregion in the superior-temporal sulcus (Fig. 7). Due to the use ofa bilateral surface coil, our mapping of more frontal and parietalregions was less certain in this figure.

To relate these areas to the eccentricity organization, we super-imposed the borders of object-related cortex, averaged across 13subjects, onto a center–periphery map obtained by averaging 12of the same subjects (Fig. 7b). As can be seen, essentially the entireextent of occipito-temporal object areas was included in the cen-ter–periphery organization. More anteriorly, toward the anteriorparahippocampal gyrus ventrally and the superior temporal sul-cus dorsally, weakly activated patches appeared to lie outside theeccentricity map. This result indicates that although at presentwe cannot identify the exact pattern of activation that is related toeach object category, we could conclude that most of its repre-sentation should be found somewhere within the bounds of thecenter–periphery global map.

DISCUSSIONCenter–periphery organization in human object areasOur results reveal an association between object images and theorganization of visual field eccentricity. Thus, in high-order objectareas, both large and small face images tended to be associated withcentral visual field representations (Figs. 4a and 6a, red), whereasboth large and small building images tended to overlap peripher-al field representations (Figs. 4a and 6a, blue). This associationcannot be attributed to irregular mapping results, because bothour maps of face and building-related regions, as well as our mapsof central versus peripheral visual field representations, closely cor-relate with previously reported maps (faces13–16,18,24,25,27, build-ings11,13,24, center/periphery5,22).

The finding of an eccentricity map in high-order object areasextends the previous report by our group of a foveal bias in theLOC20. The extension of the eccentricity maps to areas beyondthe already characterized retinotopic areas18,22 is most likelydue to the use of object stimuli in the eccentricity mapping,rather than the texture-like stimuli typically used in earlier stud-

ies. Texture stimuli have been shown to belargely ineffective in activating high-orderobject areas26.

The present result may seem to be atodds with previous work by our group,which showed substantial position andsize invariance in the LOC26,28. However,this is not the case for the central-biasedLOC, because changes in object image sizeor position, as long as they overlap thevisual field center, are not expected to sub-stantially affect the overall activation level(see also Fig. 6).

Macaque IT, which was suggested to be homologue to humanLOC, has been shown to exhibit object selectivity29,30 and to man-ifest a foveal bias10,31–34. A suggestion for a center/periphery seg-regation, compatible with the one described here, was found inposterior IT, in which the central visual field was represented moredorsally, and the peripheral visual field more ventrally10,35. How-ever, these studies did not compare the feature/object selectivity inthese regions, so it is unclear whether macaque IT actually exhibitsan association between visual field and object selectivity similarto the one found here.

Although our results clearly point to a central versus peripheralbias in object-related, high-order areas, these regions did notexhibit a well-organized visual meridian representation4,7,20 whichis characteristic of early retinotopic areas. This result is again com-patible with response properties of monkey IT neurons10,35 as wellas other neuroimaging results (for example, see ref 18).

A consequence of the physical distribution of features?The central versus peripheral bias we observed could not beexplained as a simple consequence of a center/periphery imbal-ance in the statistical distribution of visual features present in theface and building images used in our experiment. The relation-ship of faces and buildings to eccentricity maps was maintainedeven when the center/periphery balance of features was substan-tially modified by changing image size (Fig. 5; compare the spec-tral energy of the large faces and the small buildings). Theperipheral bias did manifest itself in an enhancement of the acti-vation for both buildings and faces when these were increased insize (Fig. 6b); however, this enhancement was not sufficient toovercome the shape-selective, preferential activation for buildingsover faces characteristic of this region.

articles


Table 1. Talairach23 coordinates of face-related and building-related regions.

Left hemisphere Right hemisphere

x y z x y z

FacesLO –40 ± 10 –72 ± 3 –13 ± 10 41 ± 2 –69 ± 6 –10 ± 7PFs –38 ± 6 –50 ± 7 –21 ± 6 33 ± 6 –44 ± 8 –18 ± 4BuildingsAnterior CoS –25 ± 1 –42 ± 3 –10 ± 2 25 ± 2 –39 ± 6 –12 ± 2

Values are mean ± s.d. in mm.

Fig. 5. Stimuli used in experiment 3. Average spectral energy of thecentral (top) and peripheral (bottom) parts of the images in each cate-gory of experiment 3. Energy was calculated as the sum of squares ofamplitudes in the range 0.1–9 cycles/degree, in each image part. y-axis,normalized energy (see Methods). Error bars, s.d. In the peripheral partof the visual field, the energy in large face images was higher than theenergy in small buildings.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Potential confoundsAdditional factors that could have affected the results are atten-tional effects and eye movements. Attentional level was main-tained across the various experimental conditions by using anidentical task of equal attentional demand (1-back memory task)throughout the experiment (see experiment 1A, Methods). Theclear retinotopy observed in our retinotopic and eccentricitymaps rules out major eye movements during the scans. In addi-tion, we obtained similar results using brief (250-ms) image presentations, which prevented extensive scan-ning eye movements (see experiment 2, Methods). Thus, ourresults cannot be attributed to differential eye movement in thedifferent conditions.

In summary, our results unite two seemingly unrelated orga-nizational features of human visual cortex, eccentricity maps andobject selectivity, into a global organization in high-order occip-ito-temporal cortex.

Putative sources for the center-periphery organizationSuch a center/periphery organization may have a develop-mental basis. During the layout of object representations,object categories are associated with the region of visual spacethat is attended during the establishment of these represen-tations. Because faces require central scrutiny, possibly due tothe minute differences in features that are critical for recog-

nition, they are associated with a central field bias, whereasbuildings will be associated with a peripheral bias. In relation tothis, expertise training in recognition of specific objects (forexample, birds) leads to enhanced activation in face-related(and by implication, center-biased) cortical regions25,36.

A complementary explanation is that the center/peripheryorganization allows for a more efficient allocation of process-ing resources for different object categories. Objects whoseidentification necessitates high acuity will receive more exten-sive inputs from the foveal representation, which provides theneeded spatial resolution. In contrast, objects that can be recognized at a coarser level or that require large-scale inte-gration of features will be associated with more peripheral rep-resentations. We would thus anticipate that representations ofletters and digits (for example, refs. 14, 37), which stronglydepend on foveal vision, will be associated with central fieldrepresentations. We are currently exploring this prediction.

articles

Fig. 6. Experiment 3, feature distribution experiment. (a) Results ofexperiment 3 in the right hemisphere of one subject. Red, voxels pref-erentially activated by large faces compared to small buildings; blue,voxels preferentially activated by small buildings compared to largefaces. Left, object-selective areas superimposed on retinotopic borders,which are denoted by dotted lines. Color scales indicate the degree ofstatistical correlation. Right, the same areas superimposed on theeccentricity representation (yellow, center; green, periphery). Dottedline, estimated anterior border of retinotopic areas. Outside the retino-topic areas, the large-face voxels overlapped the central visual field rep-resentation, whereas the small buildings were associated with theperipheral field representation (indicated by arrows). (b) Average signalfrom the six subjects who participated in experiment 3. Left, voxelsselected by applying a statistical test that searched for preferential acti-vation for regular faces compared to regular buildings. Error bars, s.e.m.Right, voxels selected by applying the test ‘regular buildings > regularfaces.’ Error bars, s.e.m. Large faces and small buildings were notincluded in the voxel selection test, and only voxels that were outsideretinotopic areas were included in the analysis. The charts show thatface-related voxels were also preferentially activated by large faces, andbuilding-related voxels were also activated by small buildings.

a

b

Fig. 7. Large-scale relationship of object-related cortex with center–periphery organization. (a) Preferential activation to objectsversus patterns (red) and to patterns versus objects (blue) from 13subjects (experiment 1). The results are presented on an inflatedbrain, shown in a ventral view (left) and on the unfolded hemi-spheres (right). Abbreviations as in Fig. 2; STS, superior temporalsulcus; PHG, parahippocampal gyrus. (b) Eccentricity maps from 12subjects presented on an inflated brain shown in a ventral view (left)and on the unfolded hemispheres. Yellow, center; purple, mid; green,periphery. The borders of object areas from (a) were superimposedon the unfolded eccentricity map (red). Most of the object-relatedregions, with the exception of a few anterior foci, were containedwithin the center–periphery organization. Color scales indicate sta-tistical correlation.

a

b

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Relationship to other object categories?Although we present data here regarding only two specific cate-gories, buildings and faces, our results are also relevant to otherobject categories. This conclusion stems from the finding that sub-stantial overlap occurred between the extent of object-selectiveoccipito-temporal cortex and the center/periphery eccentricitymaps. The implication of this large-scale correspondence is thatany object category will have to be mapped somewhere along theeccentricity dimension and consequently will be associated, to someextent, with a particular combination of ‘preferred’ eccentricities.

The fact that different object classes are mapped according toa center/periphery rule does not exclude the possibility that addi-tional stimulus dimensions may be mapped in an orderly man-ner within this cortical expanse13. Clearly, the face-related voxelsdo not overlap the entire center-biased regions, leaving room forother possible object categories. Similarly, various category-specificsubdivisions may occur within the periphery-biased representationof the collateral sulcus (for example, Epstein and Kanwisher19).

Hierarchical organization within human object areasThe center–periphery organization described here provides a uni-fied organizing principle for the entire extent of occipito-tempo-ral, object-related cortex. However, this cortical expanse is notuniform. In particular, the more dorsal–posterior face-relatedregions seem to show a higher degree of central-field bias com-pared to the more ventral–anterior parts in the posterior fusiformgyrus (pFs), although the pFs did show a significant central bias(Figs. 3 and 4b), which was particularly evident when comparedto the neighboring, peripherally biased collateral sulcus.

A similar hierarchical trend was also observed along the ante-rior–posterior axis of the collateral sulcus as one moves fromV4/V8 toward the more anterior part of the sulcus. These resultsare compatible with our previous reports of a differential posi-tion and size selectivity within the LOC, whereby posterior regionsshowed a higher degree of sensitivity to these changes comparedto anterior regions28.

Following the acceptance of this work, a paper appeared38

showing a center/periphery organization in dorsal LO usingchecker-board stimuli—thus providing additional confirmation tothe prevalence of this organization in high-order visual areas.

METHODSSubjects. Fourteen healthy subjects (8 women, 24–49 years old), partici-pated in one or more of the experiments. All subjects had normal or cor-rected-to-normal vision and provided written informed consent. TheTel-Aviv Sourasky Medical Center approved the experimental protocol.

MRI acquisition. Subjects were scanned on a 1.5 Signa Horizon LX 8.25GE scanner equipped with a quadrature surface coil (Nova Medical, Wake-field, Massachusetts), which covered the posterior brain regions. Bloodoxygenation level dependent (BOLD) contrast was obtained with gradient-echo echo-planar imaging (EPI) sequence (TR, 3000, TE; 55; flip angle,90°; field of view, 24 × 24 cm2; matrix size × 80 × 80). The scanned vol-ume included 17 nearly-axial slices of 4-mm thickness and 1-mm gap.T1-weighted high resolution (1 × 1 × 1 mm) anatomical images and athree-dimensional SPGR sequence were acquired for each subject to allowaccurate cortical segmentation and reconstruction, and volume-basedstatistical analysis.

Visual stimuli. Stimuli were generated on a PC, projected onto a tangentscreen positioned in front of the subject’s forehead, and viewed through atilted mirror.

Experiment 1. This experiment comprised two separate scans. In the firstscan (experiment 1A), areas that showed preferential activation to com-mon objects, faces or buildings were located (‘objects scan’), and in the

second scan (experiment 1B), eccentricity maps were obtained (‘eccen-tricity scan’). Thirteen subjects participated in this experiment. The eccen-tricity scan of one subject was excluded due to problems in dataacquisition.

In the objects scan (1A) subjects were presented with black and whitedrawings of faces, buildings, common objects and texture patterns shownin seven 9-s blocks of each category. The blocks were pseudo-randomlyordered and alternated with 6-s blanks. Each block consisted of 9 pictures,randomly ordered. The experiments included either 64 or 32 differentpictures (4 and 9 subjects respectively). Each picture was presented for800 ms followed by a blank interval of 200 ms. One or two pictures ineach block were repeated, and subjects were asked to perform a ‘one-back’matching task, while fixating on a central red point.

In the eccentricity scan (1B), subjects were presented with pictures ofdifferent objects, which were located in three eccentricities of the visualfield: center (a circle of 1.4° diameter), mid (a ring of 2.5° inner diameterand 5° outer diameter) and periphery (a ring of 10° inner diameter and20° outer diameter). Three types of central stimuli were used in separateepochs: faces, common objects (mainly animals) and written words. Pic-tures were presented in 18-s blocks, in which each picture was presentedfor 250 ms. Subjects were requested to fixate on a small fixation dot. Visu-al epochs alternated with 6-s blanks. Four cycles of the stimuli were shown.

Experiment 2. This experiment was designed to simultaneously mapobject-selective activation and center–periphery visual field bias (Fig. 1).Five subjects participated in the experiment. Line drawings of faces andbuildings were used to locate object-selective areas (black and white, visu-al angle 12° × 12°). For the center–periphery mapping we used coloreddrawings of a variety of common objects. In the ‘center’ epochs, the stim-uli were located in a circle at the center of the visual field (diameter, 1.8°).In the ‘periphery’ epochs, a number of copies (12–13) of the same objectwere placed within a ring confined to the peripheral visual field (11.5°inner diameter, 20° outer diameter, Fig. 1). Pictures of faces and build-ings were presented in six blocks of 9 s each. Each block consisted of 18different pictures. Thirty-six pictures of each type were used throughoutthe experiment. Each picture was presented for 250 ms followed by a blankinterval of 250 ms. Central and peripheral pictures were presented in five18-s blocks, in which each picture was presented for 250 ms. Seventy-twopictures of each type were used throughout the experiment. The visualstimulation blocks were ordered pseudo-randomly and alternated with 6-s blanks. A red fixation point was positioned centrally through the entireexperiment, and subjects were instructed to fixate on it.

Experiment 3: Feature distribution experiment. Six subjects participatedin this experiment. They were presented with pictures of faces and build-ings as in experiment 2 (12° × 12°) and with two additional categories:large faces (same faces, enlarged to a size of 17.5° × 17.5°) and small build-ings (same buildings reduced to a size of 5.8° × 5.8°). Sixteen pictures ofeach category were used. Presentation procedure and task were the sameas in experiment 1A.

Behavioral experiment. Six subjects participated in a behavioral experi-ment, which was conducted outside of the magnet six months after thefMRI scans. They were presented with the central and peripheral stimulifrom experiment 2, and were asked to name them, while fixating on ared dot at the center of the screen. Each picture was presented for 250 ms followed by a 1250-ms blank. Percentages of correct responseswere calculated.

Mapping borders of visual areas. The representations of vertical and hor-izontal visual field meridians were mapped in all subjects in order to delin-eate borders of retinotopic areas4,7,20,21,39. Visual stimulation was presentedin 18-s blocks. Each image was presented for 250 ms. The stimuli con-sisted of triangular wedges that compensated for the expanded foveal rep-resentation. The wedges were presented either vertically (upper or lowervertical meridians) or horizontally (left or right horizontal meridians).The wedges consisted of either gray-level natural images or black andwhite objects-from-texture pictures40. Subjects were requested to fixateon a small central cross. Visual epochs alternated with 6-s blanks. Fourcycles of the stimuli were shown.

articles


©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



Data analysis. fMRI data were analyzed with the BrainVoyager softwarepackage (R. Goebel, Brain Innovation, Masstricht, Netherlands) and withcomplementary in-house software. Each subject’s data from each scanwere analyzed separately (except for the multi-subject analysis, see below).The functional images were superimposed on two-dimensional anatom-ical images and incorporated into the three-dimensional data sets throughtrilinear interpolation. The complete data set was transformed intoTalairach23 space. Preprocessing of functional scans included three-dimen-sional motion correction and high-frequency temporal filtering. Statisti-cal analysis was based on the General Linear Model41.

The cortical surface was reconstructed from the three-dimensionalSPGR scan, unfolded, cut along the calcarine sulcus, and flattened. Theobtained activation maps were superimposed on the unfolded cortex andthe Talairach coordinates were determined for the center of each ROI.

The two-dimensional Fourier transforms (FT) of the images in exper-iment 3 were calculated using the Matlab 5.3 software (Mathworks, Nat-ick, Massachusetts, 1999) according to the following formula:

Here, X is the FT, x is the image and N × N is the image size.FT was computed separately for the central part of each image and the

peripheral part. The square amplitudes of frequencies between 0.1 and 9 cycles/degree in each image part were summed (total energy):

Here, E is the total energy and the summation is over the frequencies in theabove range.

The bar charts in Fig. 5 present the mean total energy in the centraland peripheral parts of each category, normalized by the regular facestotal energy.

Multi-subject analysis. The object-areas map in Fig. 7 was obtained from13 subjects. The eccentricity map was obtained from 12 of these subjects.To create the maps, the time courses of all subjects were transformed intoTalairach space, z-normalized and concatenated, and the statistical testswere done on the concatenated time course.

ACKNOWLEDGEMENTSThis study was funded by JSMF 99-28 CN-QUA.05 and Israel Academy 8009

grants. We thank M. Harel for help with the brain flattening, E. Okon for

technical help, and V. Levi, S. Peled, D. Ben Bashat, P. Rotshtein and D. Palti for

help with running the experiments.


1. Yarbus, A. L. Eye Movements and Vision (Plenum, New York, 1967).2. Gattass, R., Gross, C. G. & Sandell, J. H. Visual topography of V2 in the

macaque. J. Comp. Neurol. 201, 519–539 (1981).3. Gattass, R., Sousa, A. P. & Gross, C. G. Visuotopic organization and extent of

V3 and V4 of the macaque. J. Neurosci. 8, 1831–1845 (1988).4. DeYoe, E. A. et al. Mapping striate and extrastriate visual areas in human

cerebral cortex. Proc. Natl. Acad. Sci. USA 93, 2382–2386 (1996).5. Tootell, R. B. H. et al. Functional analysis of V3A and related areas in human

visual cortex. J. Neurosci. 71, 7060–7078 (1997).6. Engel, S. A., Glover, G. H. & Wandell, B. A. Retinotopic organization in

human visual cortex and the spatial precision of functional MRI. Cereb. Cortex7, 181–192 (1997).

7. Sereno, M. I. et al. Borders of multiple visual areas in humans revealed byfunctional magnetic resonance imaging. Science 268, 889–893 (1995).

8. Rosa, M. G. in Extrastriate Cortex in Primates (eds. Rockland, K. S., Kaas, J. &Peters, A.) 127–203 (Plenum, New York, 1997).

9. Desimone, R. & Ungerleider, L. G. Multiple visual areas in the caudal superiortemporal sulcus of the macaque. J. Comp. Neurol. 248, 164–189 (1986).

10. Boussaoud, D., Desimone, R. & Ungerleider, L. G. Visual topography of area

E = Σl lA(i, j) 2

X (u, v) = Σ Σ x(j, k)eN – 1 N – 1

j = 0 k = 0

– 2πiN

(ju + kv)

TEO in the macaque. J. Comp. Neurol. 306, 554–575 (1991).11. Aguirre, G. K., Zarahn, E. & D’Esposito, M. An area within human ventral

cortex sensitive to “building” stimuli: evidence and implications. Neuron 21,373–383 (1998).

12. Epstein, R. & Kanwisher, N. A cortical representation of the local visualenvironment. Nature 392, 598–601 (1998).

13. Ishai, A., Ungerleider, L. G., Martin, A., Schouten, H. L. & Haxby, J. V.Distributed representation of objects in the human ventral visual pathway.Proc. Natl. Acad. Sci. USA 96, 9379–9384 (1999).

14. Puce, A., Allison, T., Asgaei, M., Gore, J. C. & McCarthy, G. Differential sensitivityof human visual cortex to faces, letterstrings and textures: a functional magneticresonance imaging study. J. Neurosci. 16, 5205–5215 (1996).

15. Clark, V. P. et al. Functional magnetic resonance imaging of human visualcortex during face matching: a comparison with positron emissiontomography. Neuroimage 4, 1–15 (1996).

16. Kanwisher, N., McDermott, J. & Chun, M. M. The fusiform face area: amodule in human extrastriate cortex specialized for face perception. J. Neurosci. 17, 4302–4311 (1997).

17. McCarthy, G., Puce, A., Gore, J. C. & Allison, T. Face specific processing in thehuman fusiform gyrus. J. Cogn. Neurosci. 9, 605–610 (1997).

18. Halgren, E. et al. Location of human face-selective cortex with respect toretinotopic areas. Hum. Brain Mapp. 7, 29–37 (1999).

19. Epstein, R., Harris, A., Stanley, D. & Kanwisher, N. The parahippocampalplace area: recognition, navigation, or encoding? Neuron 23, 115–125 (1999).

20. Grill-Spector, K. et al. A sequence of object-processing stages revealed by fMRIin the human occipital lobe. Hum. Brain Mapp. 6, 316–328 (1998).

21. Engel, S. A. et al. fMRI of human visual cortex. Nature 369, 525 (1994).22. Hadjikhani, N., Liu, A. K., Dale, A. M., Cavanagh, P. & Tootell, R. B.

Retinotopy and color sensitivity in human visual cortical area V8. Nat.Neurosci. 1, 235–241 (1998).

23. Talairach, J. & Tournoux, P. Co-Planar Stereotaxic Atlas of the Human Brain(Thieme Medical, New York, 1988).

24. Haxby, J. V. et al. The effect of face inversion on activity in human neuralsystems for face and object perception. Neuron 22, 189–199 (1999).

25. Gauthier, I., Tarr, M. J., Anderson, A. W., Skudlarski, P. & Gore, J. C.Activation of the middle fusiform ‘face area’ increases with expertise inrecognizing novel objects. Nat. Neurosci. 2, 568–573 (1999).

26. Malach, R. et al. Object-related activity revealed by functional magneticresonance imaging in human occipital cortex. Proc. Natl. Acad. Sci. USA 92,8135–8139 (1995).

27. Hoffman, E. A. & Haxby, J. V. Distinct representations of eye gaze and identityin the distributed human neural system for face perception. Nat. Neurosci. 3,80–84 (2000).

28. Grill-Spector, K. et al. Differential processing of objects under various viewingconditions in the human lateral occipital complex. Neuron 24, 187–203 (1999).

29. Tanaka, K. Neuronal mechanisms of object recognition. Science 262, 685–688(1993).

30. Tanaka, K. Inferotemporal cortex and object vision. Annu. Rev. Neurosci. 19,109–139 (1996).

31. Gross, C. G., Bender, D. B. & Rocha-Miranda, C. E. Visual receptive fields ofneurons in inferotemporal cortex of the monkey. Science 166, 1303–1306(1969).

32. Seltzer, 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).

33. Desimone, R., Fleming, J. & Gross, C. G. Prestriate afferents to inferiortemporal cortex: an HRP study. Brain Res. 184, 41–55 (1980).

34. Baizer, J. S., Ungerleider, L. G. & Desimone, R. Organization of visual inputsto the inferior temporal and posterior parietal cortex in macaques. J. Neurosci.11, 168–190 (1991).

35. Hikosaka, K. Representation of foveal visual fields in the ventral bank of thesuperior temporal sulcus in the posterior inferotemporal cortex of themacaque monkey. Behav. Brain Res. 96, 101–113 (1998).

36. Gauthier, I., Skudlarski, P., Gore, J. C. & Anderson, A. W. Expertise for carsand birds recruits brain areas involved in face recognition. Nat. Neurosci. 3,191–197 (2000).

37. Gauthier, I. et al. The fusiform “face area” is part of a network that processesfaces at the individual level. J. Cogn. Neurosci. 12, 495–504 (2000).

38. Tootell, R.B. & Hadjikani, N. Where is ‘dorsal v4’ in human visual cortex?retinopic, topagraphic and functional evidence. Cereb. Cortex 11, 298–311(2001).

39. Grill-Spector, K., Kushnir, T., Hendler, T. & Malach, R. The dynamics ofobject-selective activation correlate with recognition performance in humans.Nat. Neurosci. 3, 837–843 (2000).

40. Grill-Spector, K., Kushnir, T., Edelman, S., Itzchak, Y. & Malach, R. Cue-invariant activation in object-related areas of the human occipital lobe.Neuron 21, 191–202 (1998).

41. Friston, J. et al. Statistical parametric maps in functional imaging: a generallinear approach. Hum. Brain Mapp. 2, 189–210 (1995).

articles©

2001

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co


articles


It seems plausible that music, like language, has a syntax: bothhave a structure based on complex rules. However, how a musi-cal syntax may be described has remained a matter of debate1–4.To investigate the processing of musical syntax, EEG studies5,6

have taken advantage of the listener’s ability to expect specificmusical events according to a preceding musical context, and todetect violations of harmonic expectancies within a musicalsequence. This ability may be an indication that a musical syn-tax exists, mainly because the specificity of harmonic expectancycorresponds to the degree of harmonic relatedness as described bymusic theory1,7–9. That is, subjects expect to hear in sequenceharmonically related but not harmonically unrelated chords.

Event-related brain potentials (ERPs) elicited by syntacticincongruities in language and music were compared in a pre-vious study5. In that study, harmonic incongruities were inter-preted as grammatical incongruity in music. It was shown thatboth musical and linguistic structural incongruities elicit pos-itivities with a latency of about 600 ms (the so-called P600)that are statistically indistinguishable. The P600 reflects moregeneral knowledge-based structural integration during the per-ception of rule-governed sequences. Additionally, a negativemusic-specific ERP component with a latency of around350 ms and an anterior right-hemisphere lateralization wasobserved. This right anterio temporal negativity (RATN) waselicited by out-of-key chords, and taken to reflect the applica-tion of music-syntactic rules.

In another EEG study6, harmonically unrelated and func-tionally inappropriate chords occurred within sequences of in-key chords. Sequences consisting of in-key chords were composedto build up a musical context, which correlates in listeners withthe buildup of strong expectancies to hear harmonically appro-priate chords in sequence7,8. The principles that form the basisof these expectancies have been described as a ‘hierarchy of har-monic stability’8, and correspond to the theory of harmony7,8.Harmonically appropriate chords are tonally related chords orchord functions that fit well at certain positions in a musical con-text (for example, a tonic chord at the end of a sequence)8. Inap-propriate chords elicited an early right-anterior negativity

(ERAN). Notably, such chords were consonant major chords; itwas only the musical context that made them sound unexpect-ed. Within the musical context, they could only be differentiatedfrom the in-key chords by the application of (implicit) musicalknowledge about the principles of harmonic relatednessdescribed by music theory. These principles or rules of musictheory may be thought of as musical syntax4–8.

Here we used the same experimental protocol as the preced-ing EEG study6. Participants (all ‘non-musicians’) were present-ed with directly succeeding chord sequences, each consisting offive chords (Fig. 1). Sequences consisting exclusively of in-keychords (cadences) established a musical context toward the endof each sequence (Fig. 1a). Due to the buildup of musical con-text, harmonic expectancies that were most specific at the endof each sequence were generated in listeners. Besides the in-keychord sequences, however, some sequences contained harmoni-cally unexpected chords: a ‘Neapolitan sixth chord’ occurred atthe third position in 25%, and at the fifth position in another25% of all sequences (Fig. 1b and c). This chord is a variation ofthe subdominant, and contains two out-of-key notes, althoughthe chord itself is major and consonant.

Compared to in-key chords, chords containing out-of-key(‘non-diatonic’) notes are, in music-theoretical terms, moredistant from the tonal center, and therefore perceived as unex-pected6,8,9,11,12. As noted before, the ability of listeners toexpect chords according to their harmonic relatedness to a pre-ceding harmonic context has been proposed to reflect the exis-tence of a musical syntax. Because the Neapolitan chordsviolated the expectancy for tonally related chords to follow,effects elicited by the Neapolitan chords were thus proposedto reflect music-syntactic processing. Because of the musicalcontext buildup, the harmonic expectancies of listeners wereviolated to a higher degree at the fifth position (where theexpectancies were most specific) compared to the third posi-tion of a sequence. Therefore, the effects of Neapolitan chordswere proposed to be larger at the fifth compared to the thirdposition. In addition, from a music-theoretical perspective,Neapolitan chords function harmonically as a subdominant

Musical syntax is processed inBroca’s area: an MEG study

Burkhard Maess, Stefan Koelsch, Thomas C. Gunter and Angela D. Friederici

Max Planck Institute of Cognitive Neuroscience, PO Box 500 355, D-04303, Leipzig, Germany

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

The present experiment was designed to localize the neural substrates that process music-syntacticincongruities, using magnetoencephalography (MEG). Electrically, such processing has beenproposed to be indicated by early right-anterior negativity (ERAN), which is elicited by harmonicallyinappropriate chords occurring within a major-minor tonal context. In the present experiment, suchchords elicited an early effect, taken as the magnetic equivalent of the ERAN (termed mERAN). Thesource of mERAN activity was localized in Broca’s area and its right-hemisphere homologue, areasinvolved in syntactic analysis during auditory language comprehension. We find that these areas arealso responsible for an analysis of incoming harmonic sequences, indicating that these regionsprocess syntactic information that is less language-specific than previously believed.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

variation; a Neapolitan chord at the third position of thesequence was, functionally, fairly suitable (because a subdom-inant in that position was appropriate), whereas a Neapolitanat the fifth position was functionally inappropriate (becauseonly a tonic chord would be appropriate in that position).

Thus, a Neapolitan chord as presented here may be taken as‘music-syntactically’ incongruous on the basis of both music-psychological (with respect to harmonic expectations) and music-theoretical reasoning (with respect to harmonic chord functionsand rules). From both perspectives, the degree of music-syntac-tic incongruity is higher for Neapolitans at the fifth compared tothe third position. In the present study, we show that the mag-netic effect elicited by the Neapolitans was stronger at the fifthcompared to the third position, indicating that this effect reflectsmusic-syntactic processing. This effect was generated in bothhemispheres in the inferior pars opercularis, known in the lefthemisphere as Broca’s area.

RESULTSIn-key chords elicited a large mean global field power (MGFP, ameasure of the strength of an evoked field), present in all sub-jects at around 200 ms (relative tostimulus onset, Fig. 2a). (This mag-netic effect will henceforth be

referred to as the P2m.) Brain responses elicited from Neapolitanand in-key chords in the fifth position clearly differed (Fig. 2b).Neapolitan chords elicited a particular early magnetic field effect,which was, at any sensor, nearly uni-modal over time, and waslargest around 200 ms (like the P2m). This effect (henceforthreferred to as the mERAN) can best be seen in the differencewaves of Fig. 2b. Virtually no magnetic effects were observableafter around 350 ms, for Neapolitans or for in-key chords.

The field maps of both P2m and mERAN reveal a dipolarpattern over each hemisphere (Fig. 3a and b). In all subjects,the fields of the mERAN had virtually an inversed ‘polarity’compared to the fields of the P2m. Moreover, the steepest fieldgradients of the mERAN are anterior to those of the P2m, indi-cating that the neural generators of the mERAN are anteriorto those of the P2m.

Effects elicited by Neapolitan chords at the third and fifthposition were very similar in distribution and time course; how-ever, the third-position effects were distinctly smaller (about halfof the strength of fifth-position effects, Figs. 2c and 3c). TheMGFP of the mERAN (in-key chord signals subtracted fromNeapolitan chord signals, Fig. 4) elicited at the third position dif-fered significantly from the MGFP of the mERAN elicited at thefifth position (paired t-test; t = 5.69, p = 0.005). (MGFP was cal-culated for third and fifth position for each subject separately inthe time window from 170–210 ms.)

Dipole solutionsDipole solutions for the P2m and the mERAN elicited at the fifthposition were obtained from each subject. (The signal-to-noiseratio (SNR) of the effects elicited by Neapolitan chords at thethird position was too low to calculate reliable dipole solutions;see Methods.) Then, locations of dipoles were transformed intoa Talairach-sized standard brain, and averaged across subjects.For the P2m, one dipole was located in each hemisphere withinthe middle part of Heschl’s gyrus (in the superior temporal

Fig. 1. Examples of chord sequences. (a) Cadences consisting exclu-sively of in-key chords. (b) Chord sequences containing a Neapolitansixth chord at the third position. (c) Chord sequences containing aNeapolitan at the fifth position; Neapolitan chords are indicated byarrows. (d) Example of directly succeeding chord sequences as pre-sented in the experiment.

a

b

c

d

Fig. 2. Time courses of magnetic fieldstrength. Data were chosen from tworepresentative subjects at four sensorslocated in the magnetic field maxima. (a) P2m time course elicited by in-keychords. (b) Signals evoked by chords atthe fifth position, plotted separately forNeapolitan (dashed lines) and in-keychords (dotted lines). The effect elicitedby Neapolitan chords (mERAN) is indi-cated by the solid lines (difference wave,Neapolitan chord signals subtractedfrom in-key chord signals); this effectwas maximal around 200 ms. (c) Signalsevoked by chords at the third position(line designations as in b).

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


Fig. 3. P2m and mERAN, magnetic field maps. Themaps of the mERAN were calculated by subtractingthe event-related magnetic fields (ERFs) elicited byin-key chords from the ERFs of Neapolitan chords.

gyrus), which corresponds to Brodmann’s area (BA) 41 (Fig. 5).The dipole solution for the mERAN indicated, in each hemi-sphere, one dipole located in the inferior part of the pars oper-cularis (in the inferior frontal gyrus, part of BA 44; Fig. 5). Theresidual normalized variance of dipole solutions was, on aver-age, 5% for the mERAN and 4% for the P2m.

The generators of the mERAN were located approximately2.5 cm anteriorly, and 1.0 cm superiorly with respect to thegenerators of the P2m (Table 1). The generators of both theP2m and the mERAN appear to have a stronger dipole momentin the right than in the left hemisphere; a right-hemisphericpredominance of the mERAN was present in four of six sub-jects. However, statistical analysis did not reveal a hemispher-ic difference of effects.

To test whether the dipole locations of mERAN and P2mdiffered significantly, y- and z-coordinates of dipoles were ana-lyzed separately using ANOVAs with condition (P2m ×mERAN) and hemisphere (left × right dipoles) as factors. BothANOVAs for y- and z-coordinates yielded an effect of condi-tion (y-coordinates, F1,5 = 37.2, p < 0.005; z-coordinates,F1,5 = 21.5, p < 0.01), indicating that the mERAN is generat-ed anteriorly and superiorly to the P2m.

Only very small P1 and N1 were elicited by all chords. This ispresumably a consequence of the continuous stimulus presenta-tion, in which one chord directly followed the other; the onsetof each chord was not an abrupt change in loudness. Particular-ly, the N1 is thought to correspond to transient detection, becausethe N1 is evoked by sudden changes in the level of energy imping-ing on the sensory receptors13,14.

DISCUSSIONIn-key chords elicited a distinct magnetic field effect, which wasmaximal around 200 ms (P2m). The P2m is suggested here asthe magnetic equivalent of the electrical P2, because of its timecourse, its ‘polarity,’ and the location and orientation of its gen-erators. (The generators would produce a positive electricalpotential over fronto-central scalp regions.) The average of thetransformed individual dipole solutions yielded two generators of

the P2m, one located in each hemisphere inthe middle of Heschl’s gyrus (that is, withinor near the primary auditory cortex, near thegenerators of the P1m15–17 and the N1m18–20).The dipole of the P2m tended to have astronger dipole moment in the right comparedto the left hemisphere. This finding mightreflect a preference of the right hemisphere forthe processing of tones and chords21–24.

Neapolitan chords at the fifth position ofthe chord sequences elicited magnetic fieldsthat differed distinctly from those elicited byin-key chords at the same position (althoughparticipants were instructed to ignore theharmonies; see Methods). Neapolitan chordselicited an early magnetic field effect that was

maximal around 200 ms, the mERAN. The mERAN is regard-ed here as the magnetic equivalent of the (electrical) ERAN.Four findings support this assumption. First, the mERAN wassensitive to harmonically inappropriate chords. Second, thetime course of the mERAN was virtually identical to the timecourse of the ERAN6. Third, in all subjects, the fields of themERAN had an inversed polarity compared to the fields of theP2m (corresponding to the ERAN and the P2). Fourth, themERAN is, like the ERAN, considerably smaller (about 50%)when elicited by Neapolitan chords at the third versus the fifthposition (see below).

In contrast to the P2m, the generators of the mERAN werenot located within the temporal lobe. The mERAN was generat-ed approximately 2.5 cm anterior to and 1.0 cm superior to theP2m in both hemispheres, namely, in each hemisphere withinthe inferior part of BA 44 (inferior part of the pars opercularis).In the left hemisphere, this is known as Broca’s area.

The mERAN, like the ERAN, is suggested to reflect thebrain’s response to a harmonic context violation. Chords pre-ceding the Neapolitan chords at the fifth position strongly estab-lished a tonal key. During such a sequence, listeners build up a‘hierarchy of harmonic stability’8, which induces strong har-monic expectations for harmonically appropriate chords to fol-low. At the fifth position of the chord sequences, a tonic chordwas harmonically most appropriate. Instead of a tonic, aNeapolitan chord occurred, which contained out-of-key notesand therefore sounded unexpected in the established tonal envi-ronment6–9,12,25–27. Moreover, the remaining in-key note of theNeapolitan chords (in C major, f ) was also unexpected, becauseit does not belong to the tonic chord. The ability to perceivedistances between chords (andkeys, respectively) and toexpect certain harmonies (andharmonic functions) to ahigher or lower degree canonly rely on a representationof the principles of harmonicrelatedness described by music

Fig. 4. Mean global field power signals of the mERAN (MGFP averaged over all MEG channels and all subjects).The MGFP was significantly stronger (shaded area) at the fifth position versus the third position.

0 200

50

ms

fT

ERANm, 5th chord ERANm, 3rd chord

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

theory. These principles, or rules, were reflected in the harmonicexpectancies of listeners and may be interpreted as musical syn-tax (see Introduction).

The mERAN was distinctly larger when elicited at the fifthcompared to the third position. This finding supports the hypoth-esis that the mERAN reflects music-syntactic processing, becausethe degree of music-syntactic incongruity was higher at the fifthcompared to the third position. Because of the musical contextbuildup, which was more specific at the end of the sequence, andbecause of the inappropriate chord function of a Neapolitan atthe fifth position (subdominant-variation instead of tonic), themusical syntax was violated to a higher degree at the fifth com-pared to the third position.

mERAN and MMNmThe mERAN is generated more anteriorly than the mismatchnegativity (MMN) or its magnetic equivalent (MMNm).Whereas the MMN receives its main contributions from gen-erators located in temporal areas28, we found that the mERANwas generated in the frontal lobe. Frontal contributions to theMMN have been reported for the frequency-MMN and EEG,but not for MEG29, Broca’s area or its homologue30–32. More-over, Neapolitan chords were not physical oddballs (no physi-cal regularity preceded the Neapolitan chords); thus nofrequency- or spectral-MMN could be elicited. Therefore,results support the hypothesis that the mERAN is not anMMN, at least not the ‘classical’ MMN6,14,42.

The Neapolitan chords at the third and fifth position werephysically identical. Therefore, it could only be the degree ofmusic-syntactic incongruity, referring only to music-theoreticalterms, that modulated the amplitude of the mERAN. That is, thefinding that the mERAN is larger when elicited at the fifth posi-tion again strongly supports the hypothesis that the mERAN isnot an MMN. Rather, the results indicate that the mERAN isspecifically correlated with the processing of auditory informa-tion within a complex rule-based context.

Inferior BA 44 and syntax processing in languageBroca’s area and its right homologue, particularly the inferiorpart of BA 44, are involved in the processing of syntactic aspectsduring language comprehension33–40, and are specialized for fast

and automatic syntactic parsing processes10. The early left ante-rior negativity (ELAN) reflecting these processes10,41 is also gen-erated, at least partially, in Broca’s area and its right-hemispherehomologue38. The dipole solution for the magnetic ELAN revealsdipoles in the left and the right inferior frontal cortex (with verysimilar locations as the mERAN) in addition to bilateral tempo-ral dipoles38. As described in the Introduction, the ERAN high-ly resembles the ELAN, though with a different hemisphericweighting.

The present results indicate that Broca’s area and its right-hemisphere homologue might also be involved in the processingof musical syntax, suggesting that these brain areas process con-siderably less domain-specific syntactic information than previ-ously believed. Like syntactic information of language, which isfast and automatically processed in Broca’s area and its right-hemisphere homologue, music-syntactic information processedin the same brain structures also seems to be processed auto-matically42. The magnetic fields of the mERAN were, in four ofsix subjects (but not in the grand average), stronger over the rightthan over the left hemisphere. This finding is consistent with theELAN, which is prevalently (although not consistently) strongerover the left hemisphere. It is thus suggested here, as a workinghypothesis, that the left pars opercularis is more involved in theprocessing of language syntax, and the right pars opercularis morein the processing of musical syntax. However, both hemispheresseem to be considerably activated in both domains.

In the present study, harmonically inappropriate chordsactivated Broca’s area and its right-hemisphere homologue.This finding is important for several reasons. First, it demon-strates that complex rule-based information is processed inthese areas with considerably less domain-specificity than pre-viously believed21,39,43. This might suggest that these areasprocess syntax, that is, complex rule-based information, in adomain other than language. This finding might lead to newinvestigations of syntax processing in the musical, or even otherauditory but non-linguistic domains. Second, it reveals froma functional-neuroanatomical view a strong relationshipbetween the processing of language and music. This relation-ship might at least partly account for influences of musicaltraining on verbal abilities44,45. Third, the present study intro-duced a new method of investigating music perception using

Fig. 5. Grand average dipole solutions for P2m and mERAN. Grand aver-age dipole solutions, yellow; P2m, top; mERAN, bottom. Each panel showsleft and right sagittal and axial (parallel to AC-PC line) view. Dipole solu-tions for both the P2m and the mERAN refer to two-dipole configurations(one dipole in each hemisphere). Blue discs, single subject solutions.

Table 1. Locations and strengths of P2m and mERANdipoles (grand average of back-transformed dipolesolutions).

Dipole coordinates (x, y, z) and dipole moments (Q)

P2m left P2m right mERAN left mERAN right(mean ± s.e.m.) (mean ± s.e.m.) (mean ± s.e.m.) (mean ± s.e.m.)

x (mm) –45 ± 2 51 ± 2 –48 ± 5 50 ± 3

y (mm) –16 ± 4 –19 ± 2 9 ± 4 6 ± 4

z (mm) 4 ± 2 4 ± 1 16 ± 1 14 ± 2

Q (nAm) 14 ± 5 22 ± 10 31 ± 15 35 ± 12

Values are given with respect to the Talairach coordinate system.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


MEG. Effects were elicited in ‘non-musicians’ (even thoughNeapolitan chords were task-irrelevant), supporting thehypothesis of an (implicit) musical ability of the human brain,and enabling a broad generalization of the present findings.

METHODSSubjects. Six right-handed and normal-hearing subjects (20 to 27 yearsold; mean, 22.5; 4 females) participated in the experiment. Subjects werenon-musicians, that is, they had never learned singing or an instrument,and they did not have any special musical education besides normalschool education.

Stimuli. The pool of stimuli consisted of 128 different chord sequences;each sequence consisted of five chords. The first chord was always thetonic of the following chord sequence; chords at the second positionwere tonic, mediant, submediant, subdominant, dominant of the dom-inant, secondary dominant of mediant, secondary dominant of sub-mediant or secondary dominant of supertonic. The third position chordwas subdominant, dominant, dominant six-four, Neapolitan sixth, or ifpreceded by a secondary dominant-mediant, the submediant or super-tonic. The fourth position chord was the dominant seventh, and thefifth position chord was either the tonic or the Neapolitan sixth. Textureof chords followed the classical theory of harmony46. From the pool of128 sequences, 1350 chord sequences were randomly chosen such thatthe secondary dominants, Neapolitan chords at the third position, andNeapolitan chords at the fifth position of a sequence occurred with aprobability of 25% each. Presentation time for chords 1 to 4 was 600 ms,and for chord 5, 1200 ms. In 10% of the sequences, an in-key chordfrom position 2–5 was played by an instrument other than piano. Chordsequences were presented in direct succession. The same stimuli wereused in experiment 1 of the preceding EEG study6.

Procedure. Three experimental sessions were conducted (each compris-ing three blocks). Participants were only informed about the deviantinstruments, not about the Neapolitan chords or their nature. Partici-pants were instructed to ignore the harmonies.

MEG recording. The continuous raw MEG was recorded using the 4D-Neuroimaging Magnes WHS 2500 whole-head system (San Diego, Cal-ifornia), which used 148 magnetometer channels, 11 magnetic referencechannels and four EOG channels. Signals were digitized with a band-width of 0.1 Hz to 50 Hz and a sampling rate of 254.31 Hz. The contin-uous MEG data were filtered off-line with a 2–10 Hz band-pass filter(1001 points, FIR). All subjects’ averaged data were transformed onto asensor-position representative for all blocks of this subject using ASA(ANT Software, Enschede, The Netherlands) and were accumulated persubject across all blocks and sessions.

Data analysis. For each participant, a realistically shaped volume con-ductor was constructed, scaled to the subject’s real head size. This wasachieved by adjusting the size of the Curry-Warped brain (an averagebrain obtained from more than 100 subjects; Neuroscan Labs, Ster-ling, Virginia, B. Maess and U. Oertel, Neuroimage, 10, A8, 1999) toeach individual head shape. This method results in independent scal-ing factors for all three spatial dimensions. The adjustment procedurethus enabled source localization with an accuracy close that achievedwith individual MR-based models. These scaling factors were also use-ful for the transformation of localization results into the Talairach-sized brain.

To achieve a higher signal-to-noise ratio, the ERFs evoked by all in-key chords were combined (The magnetic field maps of the P1, N1 andP2m virtually did not differ between in-key chords presented at differ-ent positions within the chord-sequences.) Dipole orientations were sep-arated into tangential and radial contributions for each subject. The radialcontributions were then eliminated. The criterion for an acceptable dipolesolution was the explanation of at least 90% of normalized variance foreach subject. The data of only two subjects fulfilled the criterion for themERAN elicited at the third position, so no grand-average of dipole solutions was done in this case.

ACKNOWLEDGEMENTSThis work was supported by the Leibniz Science Prize awarded to A.D. Friederici

by the German Research Foundation.

Note: Examples of the stimuli are available on the Nature Neuroscience web site

(http://www.neurosci.nature.com/web_specials).


1. Swain, J. Musical Languages (Norton, UK, 1997).2. Sloboda, J. The Musical Mind: The Cognitive Psychology of Music (Oxford

Univ. Press, New York, 1985).3. Lerdahl, F. & Jackendoff, R. A Generative Theory of Music (MIT Press,

Cambridge, Massachusetts, 1999).4. Raffmann, D. Language, Music, and Mind (MIT Press, Cambridge,

Massachusetts, 1993).5. Patel, A. D., Gibson, E., Ratner, J., Besson, M. & Holcomb, P. Processing

syntactic relations in language and music: an event-related potential study.J. Cogn. Neurosci. 10, 717–733 (1998).

6. Koelsch, S., Gunter, T., Friederici, A. D. & Schröger, E. Brain indices of musicprocessing: ‘non-musicians’ are musical. J. Cogn. Neurosci. 12, 520–541(2000).

7. Krumhansl, C. & Kessler, E. Tracing the dynamic changes in perceived tonalorganization in a spatial representation of musical keys. Psychol. Rev. 89,334–368 (1982).

8. Bharucha, J. & Krumhansl, C. The representation of harmonic structure inmusic: hierarchies of stability as a function of context. Cognition 13, 63–102(1983)

9. Bharucha, J. & Stoeckig, K. Reaction time and musical expectancy: primingof chords. J. Exp. Psychol. Hum. Percept. Perform. 12, 403–410 (1986).

10. Friederici, A. D., ed. Language Comprehension: A Biological Perspective(Springer, Berlin, 1998).

11. Bharucha, J. Anchoring effects in music: the resolution of dissonance. Cognit.Psychol. 16, 485–518 (1984).

12. Bharucha, J. & Stoeckig, K. Priming of chords: spreading activation oroverlapping frequency spectra? Percept. Psychophys. 41, 519–524 (1987).

13. Clynes, M. in Average Evoked Potentials: Methods, Results and Evaluations(eds. Donchin, E. & Lindsley, D.) 363–374 (US Government Printing Office,Washington, DC, 1969).

14. Näätänen, R. Attention and Brain Function (Erlbaum, Hillsdale, New Jersey,1992).

15. Liegeois-Chauvel, C., Musolino, A., Barier, J., Marquis, P. & Chauvel, P.Evoked potentials recorded from the auditory cortex in man: evaluation andtopography of the middle latency hypothesis. Electroencephalogr. Clin.Neurophysiol. 92, 204–214 (1994).

16. Mäkelä, J., Hämäläinen, M., Hari, R. & McEvoy, L. Whole-head mapping ofmiddle-latency auditory magnetic fields. Electroencephalogr. Clin.Neurophysiol. 92, 414–421 (1994).

17. Pantev, C. et al. Specific tonotopic organizations of different areas of thehuman auditory cortex revealed by simultaneous magnetic and electricrecordings. Electroencephalogr. Clin. Neurophysiol. 94, 26–40 (1995).

18. Hari, R., Aittoniemi, M., Jarvinen, M., Katila, T. & Varpula, T. Auditoryevoked transient and sustained magnetic fields of the human brain. Exp.Brain Res. 40, 237–240 (1980).

19. Pantev, C., Hoke, M., Lütkenhöner, B. & Lehnertz, K. Tonotopic organizationof the auditory cortex: pitch versus frequency representation. Science 246,486–488 (1989).

20. Pantev, C. et al. Identification of sources of brain neuronal activity with highspatiotemporal resolution through combination of neuromagnetic sourcelocalization (NMSL) and magnetic resonance imaging (MRI).Electroencephalogr. Clin. Neurophysiol. 75, 173–184 (1990).

21. Zatorre, R., Evans, A., Meyer, E. & Gjedde, A. Lateralization of phonetic andpitch discrimination in speech processing. Science 256, 846–849 (1992).

22. Auzou, P. et al. Topographic EEG activations during timbre and pitchdiscrimination tasks using musical sounds. Neuropsychologia 33, 25–37(1995).

23. Levänen, S., Ahonen, A., Hari, R., McEvoy, L. & Sams, M. Deviant auditorystimuli activate human left and right auditory cortex differently. Cereb.Cortex 6, 288–296 (1996).

24. Tervaniemi, M. et al. Lateralized automatic auditory processing of phoneticversus musical information: a PET study. Hum. Brain Mapp. 10, 74–79(2000).

25. Krumhansl, C., Bharucha, J. & Castellano, M. Key distance effects onperceived harmonic structure in music. Percept. Psychophys. 32, 96–108(1982).

26. Krumhansl, C., Bharucha, J. & Kessler, E. Perceived harmonic structure ofchords in three related musical keys. J. Exp. Psychol. Hum. Percept. Perform. 8,24–36 (1982).

27. Berent, I. & Perfetti, C. An on-line method in studying music parsing.Cognition 46, 203–222 (1993).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

28. Alho, K. Cerebral generators of mismatch negativity (MMN) and itsmagnetic counterpart (MMNm) elicited by sound changes. Ear Hear. 16,38–51 (1995).

29. Rinne, T., Alho, K., Ilmoniemi, R., Virtanen, J. & Näätänen, R. Separate timebehaviors of the temporal and frontal mismatch negativity sources.Neuroimage 12, 14–19 (2000).

30. Giard, M., Perrin, F. & Pernier, J. Brain generators implicated in processing ofauditory stimulus deviance. A topographic ERP study. Psychophysiology 27,627–640 (1990).

31. Alain, C., Woods, D. L. & Knight, R. T. A distributed cortical network forauditory sensory memory in humans. Brain Res. 812, 23–37 (1998).

32. Opitz, B., Mecklinger, A., von Cramon, D. Y. & Kruggel, F. Combiningelectrophysiological and hemodynamic measures of the auditory oddball.Psychophysiology 36, 142–147 (1999).

33. Caplan, D., Alpert, N. & Waters, G. Effects of syntactic and propositionalnumber on patterns of regional cerebral blood flow. J. Cogn. Neurosci. 10,541–552 (1998).

34. Caplan, D., Alpert, N. & Waters, G. PET-studies of syntactic processing withauditory sentence presentation. Neuroimage 9, 343–351 (1999).

35. Caplan, D., Alpert, N., Waters, G. & Olivieri, A. Activation of Broca’s area bysyntactic processing under condition of concurrent articulation. Hum. BrainMapp. 9, 65–71 (2000).

36. Dapretto, M. & Booheimer, S. Form and content: dissociating syntax andsemantics in sentence comprehension. Neuron 24, 427–432 (1999).

37. Ni, W. et al. An event-related neuroimaging study distinguishing form andcontent in sentence processing. J. Cogn. Neurosci. 12, 120–133 (2000).

38. Friederici, A., Wang, Y., Herrmann, C., Maess, B. & Oertel, U. Localization ofearly syntactic processes in frontal and temporal cortical areas: amagnetoencephalographic study. Hum. Brain Mapp. 11, 1–11 (2000).

39. Just, M., Carpenter, P., Keller, T., Eddy, W. & Thulborn, K. Brain activationmodulated by sentence comprehension. Science 274, 114–116 (1996).

40. Meyer, M., Friederici, A. D. & von Cramon, D. Y. Neurocognition ofauditory sentence comprehension: event related fMRI reveals sensitivityto syntactic violations and task demands. Cognit. Brain Res. 9, 19–33(2000).

41. Hahne, A. & Friederici, A. D. Electrophysiological evidence for two steps insyntactic analysis: early automatic and late controlled processes. J. Cogn.Neurosci. 11,194–205 (1999).

42. Koelsch, S., Schröger, E., Gunter, T. & Friederici, A. D. Differentiating ERANand MMN: an ERP-study. Neuroreport (in press).

43. Shaywitz, B. et al. Sex differences in the functional organization of the brainfor language. Nature 373, 607–609 (1995).

44. Chan, A. S., Ho, Y. C. & Cheung, M. C. Music training improves verbalmemory. Nature 396, 128 (1998).

45. Douglas, S. & Willatts, P. The relationship between musical ability andliteracy skills. J. Res. Reading 17, 99–107 (1994).

46. Hindemith, P. Unterweisung im Tonsatz, 1. Theoretischer Teil (Schott, Mainz,1940).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


The goal of theory of mind is to explain the ability to predict andunderstand actions of both oneself and other intelligent agents. Twotypes of approaches attempt to account for the cognitive mecha-nism that subserves such a capacity. The theory theorists maintainthat this ability is underpinned by a folk-psychological theory ofthe structure and functioning of the mind (that may be innate andmodularized or learned individually)1. On the contrary, the simu-lation theory posits that the attributer tries to covertly mimic themental activity of the target (for review, see refs. 2, 3), and postulatesshared states of mind between the attributer and the target. Thistheory has generated considerable interest among philosophers ofmind, cognitive scientists and, recently, neuroscientists. The ques-tion of agency (how a subject attributes an action to himself or toanother agent4,5) is at the core of the simulation theory.

Motor imagery can be considered a way to access motorintentions or plans, in which the representation of a givenaction is internally performed without any overt motor output.It can be used as a natural protocol to address the issue ofagency within the simulation theory (for review, see ref. 6). Sofar, motor imagery has always been studied in a first-personsubjective perspective, and several neuroimaging studies haveconsistently demonstrated a striking functional equivalencewith actual action. The first-person perspective is associatedwith activation of the inferior parietal, premotor and SMA onthe left side as well as the ipsilateral cerebellum7–9. Further evi-dence in support of shared motor representation between men-tal simulation of action and motor execution is provided byexperiments in patients with impairments in motor imageryfollowing parietal lesions10,11. Common brain regions areinvolved during action generation, action simulation andaction observation (for a meta-analysis see ref. 12).

However, there must exist, at the neural level, a distinctionbetween first-person and third-person perspective representation.

The objective of this study was to probe the effect of perspectivetaking on the neural network engaged during mental simulation ofaction. Subjects were required either to imagine themselves per-forming a given action (first-person perspective) or to imagine theexperimenter performing the same action (third-person perspec-tive). Two perceptual modalities were used to identify brain regionsstrictly involved in perspective taking during action simulationirrespective of sensory input. These two subjective perspectiveswere initiated either from photographs of familiar objects or fromsentences depicting familiar actions. We used pictures of objectsin order to have a reference situation comparable to that used inprevious neuroimaging studies, in which motor imagery was most-ly visually triggered. However, verbal auditory stimuli were alsochosen because of their ecological features. The most natural sit-uation in which one is led to use first- and third-person perspec-tives is surely linguistic communication (should it be written,spoken or heard). We found that a limited number of brain areasmay be specifically involved in self/other distinction, namely rightinferior parietal lobe, precuneus and somatosensory cortex.

RESULTSFirst-person perspective simulation versus control First-person perspective simulation, irrespective of the presen-tation modality of the stimuli, was associated with left hemi-spheric regional cerebral blood flow (rCBF) increases in theinferior parietal lobe, precentral gyrus, superior frontal gyrus(SMA proper), occipito-temporal junction (MT/V5) and ante-rior insula. The cerebellum and precuneus were activated in theright hemisphere (A1 – AC and V1 – VC; Table 1; Fig. 1).

Third-person perspective simulation versus control Third-person perspective simulation, irrespective of the presen-tation modality, was associated with bilateral rCBF increases in

Effect of subjective perspectivetaking during simulation of action:a PET investigation of agency

Perrine Ruby1 and Jean Decety1,2

1 Inserm unit 280, 151 Cours Albert Thomas, 69424 Lyon Cedex 3, France2 Cermep, 59 Bld Pinel, 69003 Lyon, France

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

Perspective taking is an essential component in the mechanisms that account for intersubjectivityand agency. Mental simulation of action can be used as a natural protocol to explore the cognitiveand neural processing involved in agency. Here we took PET measurements while subjectssimulated actions with either a first-person or a third-person perspective. Both conditions wereassociated with common activation in the SMA, the precentral gyrus, the precuneus and the MT/V5complex. When compared to the first-person perspective, the third-person perspective recruitedright inferior parietal, precuneus, posterior cingulate and frontopolar cortex. The opposite contrastrevealed activation in left inferior parietal and somatosensory cortex. We suggest that the rightinferior parietal, precuneus and somatosensory cortex are specifically involved in distinguishingself-produced actions from those generated by others.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

the precuneus. On the left side, activation foci were detected inthe precentral gyrus, superior frontal gyrus (pre-SMA) and occip-ito-temporal junction (MT/V5). The inferior parietal lobule andfrontomarginal gyrus were both activated on the right side (A3 – AC and V3 – VC; Table 2; Fig. 2).

Areas involved in first- and third-person perspectivesThe conjunction analysis (p < 0.0001, t > 1.29) calculated with thefour contrasts (A1 – AC, V1 – VC, A3 – AC, V3 – VC) revealedbilateral rCBF increase in the precuneus (x = 6, y = –68, z = 46; t-value, 2.90; p corrected, 0.000 and –8, –64, 40; 2.43; 0.001) and inthe MT/V5 complex (–58, –60, 12; 2.22; 0.006 and 52, –54, 8*;1.70; 0.292). The precentral gyrus (–22, –12, 54; 2.25; 0.004) andSMA (–10, 4, 64; 2.14; 0.011) were activated in the left hemisphere.

Third-person versus first-person perspectivesCompared to first-person perspective, third-person perspectivesimulation was specifically associated with left rCBF increase in

the posterior cingulate cortex. On the right side, activation fociwere detected in the precuneus, the inferior parietal lobule andfrontopolar gyrus ((A3 + V3) – (A1 + V1); Table 3; Fig. 3).

First-person versus third-person perspectives First-person perspective relative to third-person perspective ((A3 + V3) – (A1 + V1), p < 0.0001, t > 3.85) showed a strongrCBF increase in the inferior parietal lobule (–66, –32, 26; 6.47;0.000), the posterior insula (–42, –10, –8; 5.46; 0.006) and the post-central gyrus* (–36, –40, 40; 4.58; 0.142) in the left hemisphere. Abilateral increase was also detected in the inferior occipital gyrus(56, –54, –24; 5.62; 0.003 and –48, –50, –18*; 4.11; 0.498).

DISCUSSIONFirst- and third-person perspectives correspond to everyday lifesituations. This study explored the effect of perspective takingon the neural substrates involved in action simulation. All brainregions activated during first-person perspective conditions wereconsistent with previous neuroimaging experiments that haverevealed the neural correlates of motor imagery7–9,13. The involve-ment of these regions, namely the inferior parietal, SMA, pre-central gyrus in the left hemisphere and ipsilateral cerebellum(Table 1; Fig. 1), has been interpreted in favor of a functionalequivalence between action simulation and action execution12,14.When compared with third-person perspective, the main effect offirst-person perspective resulted in strong left hemispheric acti-vation of the inferior parietal lobule, as well as increased activa-tion in the somatosensory cortex. This can be interpreted asevidence of a prominent role of left inferior parietal lobe in pro-gramming the self ’s movements, because the programming canpotentially be transformed into execution. Detecting a

Table 1. Areas significantly activated during first-personsimulation irrespective of the modality (A1 – AC inconjunction with V1 – VC).

L/R Coordinates t-value p corrected

Brain region

x y z

Inferior parietal lobe L –64 –30 30 5.30 0.000



Superior frontal gyrus (SMA) L –12 –2 58 4.90 0.000

Occipito-temporal junction (MT/V5) L –56 –66 4 4.15 0.000

Precentral gyrus L –26 –16 58 3.81 0.001

Cerebellum R 44 –54 –32 3.19 0.046

Anterior insula* L –30 16 8 2.98 0.128

Precuneus* R 6 –68 46 2.90 0.188

p < 0.001 (corrected for whole brain), t > 1.88. x, y, z refer to MNI coordi-nates. L, left; R, right hemisphere. *Some activated clusters are reported,even though they do not survive correction for the whole brain volume,because we think they are both neurobiologically plausible and relevant inthe light of our hypotheses.

Fig. 1. Brain areas activated by first-person simulation. Foci of activa-tion (A1 – AC in conjunction with V1 – VC) have been superimposedonto the sagittal (left hemisphere) and axial top views of the single-sub-ject MRI of SPM 99.

Table 2. Areas significantly activated during third-personsimulation irrespective of the modality (A3 – AC inconjunction with V3 – VC).


Brain region

x y z

Precuneus R 6 –64 38 5.09 0.000

Precuneus L –10 –62 38 4.14 0.000

Precentral gyrus L –22 –14 54 3.70 0.002

Occipito-temporal junction (MT/V5) L –50 –64 16 3.50 0.008

Superior frontal gyrus (SMA) L –8 4 62 3.39 0.015

Inferior parietal lobe* R 48 –58 38 2.93 0.166

Frontomarginal gyrus* R 28 50 –8 2.39 0.878

See Table 1 legend.

Fig. 2. Brain areas activated by third-person simulation. Foci of activa-tion (A3 – AC in conjunction with V3 – VC) have been superimposedonto lateral (left and right hemispheres), posterior and top views of thesingle-subject MRI of SPM 99.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


somatosensory area only and precisely when first-person per-spective is compared to third-person perspective is of particularinterest, and reveals the area’s participation in distinguishing selffrom other, as previously suggested15. The activation of MT/V5complex shows that this region is involved not only in actualmotion perception, but also in imagined movement, which isconveyed by action simulation. This is consistent with other stud-ies that have demonstrated activation in MT/V5 by apparentmotion16,17, illusory motion18, imagined motion19 and staticimages with implied motion20.

According to the simulation theory, there should be an over-lap between regions involved in first- and third-person perspec-tives. Our results show that this is partly true. Imaginingsomeone else’s action is associated with activation in several areasthat are common to first-person simulation, namely, the SMA,precentral gyrus, precuneus and MT/V5 (Table 2; Fig. 2 and theconjunction analysis; A1 – AC, V1 – VC, A3 – AC, V3 – VC).However, this overlap is not complete. There were specificincreases in the parietal, cingulate and frontal cortices for third-person perspective simulation when compared to first-personperspective simulation (Table 3; Fig. 3). Left inferior parietalactivity, which was very strong for first-person perspective simu-lation, disappeared when imagining someone else’s action. Inaddition, a strong increase was detected in the right inferior pari-etal lobe during the third-person perspective experiment (Figs. 1,2 and 4). In parallel, specific rCBF augmentations were detectedin the precuneus, left posterior cingulate cortex and right fron-topolar gyrus. The specific activation of both right inferior pari-etal cortex and precuneus during third-person simulation mayaccount for a neural mechanism that is important in the determi-nation of agency. This interpretation is supported by evidencefrom clinical neuropsychology, and from brain imaging studiesin both normal volunteers and schizophrenic patients.

The right inferior parietal cortex is activated when subjectswatch other people in an effort to imitate them21,22. Moreover, apatient with an abscess in the right parietal cortex has beendescribed, in a neuropsychological case study, to have believedthat his body was being controlled by external forces. Thispatient made statements such as, “My head is empty,” “I have nothoughts,” and “I feel hypnotized”23. Schizophrenic patientsshow hyperactivation of the right inferior parietal cortex, and

experience passivity as compared to healthy subjects during theperformance of freely selected joystick movements24. It was pro-posed that such abnormal responses in the parietal lobe causethe misattribution of self-generated acts to external entities.

Furthermore, in a PET experiment exploring the neural cor-relates of hypnosis, rCBF decreases were found in the right infe-rior parietal lobule, left posterior cingulate gyrus and leftprecuneus25. Deactivation of the precuneus, in particular, wasconsidered to be an important metabolic feature of this uncon-scious state26. In agreement with those results, it has been sug-gested that the right posterior parietal lobe has a determinantrole in high-order body or self representation27,28.

From the viewpoint of cognitive psychology, having a uni-fied perspective involves keeping track of the relationshipbetween what is perceived and what is done, and hence beingaware of agency. In this sense, it has been suggested that per-spective taking already involves self-consciousness29. Thus, atthe physiological level, the brain may need to create a particu-larly vivid representation of the self to discriminate betweenself and other. During third-person perspective simulation,one needs especially to be aware of who the self is, in order tobe able to imagine another person with the same neuralresources as the self. So as not consciously to confuse third-person simulation with first-person simulation, regions thatare critical for body schema or corporeal awareness may behighly recruited. Although this interpretation is speculative,during the third-person perspective simulation, specific rCBFincreases occurred precisely in brain regions where decreaseswere found during the hypnotic state (right inferior parietallobule, posterior cingulate and precuneus). A neuroimagingstudy of self versus non-self judgments has provided furtherresults in favor of this hypothesis. Judgments about either facepictures or personality trait words were indeed associated withactivation in the precuneus only in the case of self process-ing30. In our study, although the precuneus was activated inboth perspectives, it was much more involved during third-person perspective (Fig. 4), which was consistent with ourhypothesis (that is, overactivation of regions involved in self-representation during third-person perspective). According tothose converging results, we suggest that the right inferiorparietal lobe and the precuneus are critically involved in dis-criminating the self from others, by way of their involvementin the representation of the self.

Table 3. Areas significantly and specifically activatedduring third-person simulation compared to first-personsimulation ((A3 + V3) – (A1 + V1)).


Brain region

x y z

Posterior cingulate L –12 –50 38 5.55 0.004

Precuneus L/R 0 –66 34 5.36 0.009

Parieto-occipital fissure R 8 –68 24 5.30 0.012

Inferior parietal lobe R 44 –64 24 4.94 0.042

Inferior parietal lobe* R 50 –58 30 4.68 0.105

Frontopolar gyrus* R 14 72 10 4.37 0.266

p < 0.0001 (corrected for multiple comparisons), t > 3.85. See Table 1 legend.

Fig. 3. Brain areas activated by third- versus first-person simulation.Sagittal, axial and coronal sections of the brain (x = 8, y = –68, z = 24)showing specific areas of activation associated with third-person simula-tion when compared to first-person simulation ((A3 + V3) – (A1 + V1)).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om



articles

Fig. 4. Activation profiles for clusters in theright inferior parietal and precuneus acrossactivation conditions. The histogram barsrepresent the relative adjusted regionalcerebral blood flow values. A1, first-personsimulation with auditory stimuli; A3, third-person simulation with auditory stimuli;AC, auditory control; VC, visual control;V1, first-person simulation with visual stim-uli; V3, third-person simulation with visualstimuli. Both precuneus and right inferiorparietal show stronger activation for third-person perspective simulation, less activa-tion for first-person perspective simulation,and very low activation for control situa-tions in which self-representation is notrequired for the task.

To take a third-person perspective, subjects have to be awareof what the actor intends to do before simulating the actor’saction. This awareness could be compared to a kind of theory ofmind process. The posterior cingulate activation in particularcould be associated with such processing, as several imagingstudies have demonstrated its involvement during tasks requir-ing mind-reading31,32.

The specific activation in the frontopolar gyrus duringthird-person perspective simulation could be interpreted asdemonstrating the existence of an inhibitory phenomenon dur-ing third-person perspective simulation. An ANCOVA analysis,performed using frontopolar activity as a covariate of interestand subjects as a confound, provided results that allowed us toformulate this hypothesis, because left inferior parietal lobule(56, –30, 26) was significantly negatively correlated with thefrontopolar gyrus (t = 4.06). Our assumption of an inhibitoryrole of the frontopolar region is also in accordance with neuro-anatomical deficits that have been discovered in schizophrenicpatients. Inhibitory neurons (GABA neurons) in the anteriorcingulate and in the frontopolar cortex are lacking in the brainsof patients who are susceptible to confusing the self andother33,34. In addition, patients with lesions of this part of thefrontal cortex may exhibit utilization behavior, which has beeninterpreted as a consequence of impaired inhibition35.

Our study demonstrates that it is possible, at a representa-tional level, to identify which brain regions are involved in first-person perspective, and which are involved in third-personperspective. Several cortical areas (right inferior parietal, pre-cuneus and somatosensory cortex) are proposed to be engagedin distinguishing the self from the other, and should be investi-gated further to better understand agency disorders in bothneurological and psychopathological patients.

METHODSSubjects. Ten right-handed healthy male volunteers were recruited (24.2 ± 2.9 years old). All subjects gave written informed consent accord-ing to the Helsinki declaration. The study was approved by the local eth-ical committee (CCPPRB, Centre Léon Bérard, Lyon), and subjects werepaid for their participation.

Activation protocol. Subjects were scanned during four target conditions(A1, A3, V1, V3), and two control conditions (AC, VC), which wereduplicated once and presented in a pseudorandomized order, counter-balanced across subjects (12 scans per subject). Half the conditions werecomposed of visual stimuli (V1, V3, VC), and half were composed ofauditory stimuli (A1, A3, AC). During the scanning procedure, auditoryand visual conditions were never mixed; half the subjects saw the visual

conditions block first, and half saw the auditory block first. All actionsselected for this study required the use of the right hand.

During each visual condition (V1, V3, VC), subjects were presentedwith photographs of familiar objects (for example, a razor, shovel or ball).Each stimulus was presented for 5 s on a dark background. In the V1condition, subjects were instructed to imagine themselves (that is, usingthe first-person perspective) acting with the object for as long as itappeared on the screen. In the V3 condition, subjects were instructed toimagine the experimenter acting with the object (that is, using the third-person perspective). The same set of photographs of objects (n = 14) wasused across these two conditions. In VC condition, subjects were asked topassively watch another set of photographs of objects.

During the auditory conditions (A1, A3, AC), subjects were presentedwith verbal sentences recorded onto CD in the experimenter’s voice. Eachsentence lasted approximately 2 s and was followed by a blank period of 3 s. At the end of the blank period, a beep (300 ms) warned the subjectthat the next sentence would arrive. Each auditory condition included 14sentences. In the A1 and A3 conditions, the same series of sentences usingfamiliar actions (for example, stapling sheets of paper, peeling a banana)were used. These sentences were declined at the present tense and the sub-ject of the verb was either ‘you’ in condition A1 (for example, “You arestapling a sheet of paper”) or ‘I’ in condition A3 (for example, “I am sta-pling a sheet of paper”). In those two conditions, subjects had thus toimagine what the experimenter said (that is, in A1, with first-person per-spective; in A3, with third-person perspective). In the AC condition, thesentences described landscapes that did not include humans, motion oranimals (for example, “You are seeing a field of wheat”). Subjects wereinstructed to imagine themselves contemplating these landscapes.

All subjects were extensively trained in each of the experimental con-ditions. They were familiarized with the experimental setup, the experi-menter’s voice and physiognomy, and also with first- and third-personperspectives of action simulation. For the latter, they were trained toimagine the experimenter in a three-quarters view so that no right/leftconflict could arise during imagination. The stimuli used in the trainingsession were different from those used in the PET experiment.

Scanning procedure. A Siemens CTI HR+ (63 slices, 15.2 cm axial fieldof view) PET tomograph with collimating septa retracted operating inthree-dimensional mode was used. Sixty-three transaxial images (slicethickness of each, 2.42 mm) without gaps between them were acquiredsimultaneously. A venous catheter to administer the tracer was insertedin an antecubital fossa vein in the left forearm. Correction for attenua-tion was made using a transmission scan collected at the beginning ofeach study. After a 9-mCi bolus injection of H2

15O, scanning was startedwhen the brain radioactive count rate reached a threshold value and con-tinued for 60 s. Integrated radioactivity accumulated in 60 s of scanningwas used as an index of rCBF.

The three visual conditions used a NEC projector (800 × 600 pixels)to display colored photographs on a screen located at the back of thecamera. A mirror placed in front of the subjects’ eyes allowed them tosee the projected images by reflection. The resultant distance from the

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


articles


eyes to the screen was approximately 50 cm (corresponding field of view,42° in the horizontal dimension and 32° in the vertical direction). APower Macintosh computer (Apple, Cupertino, California) with theSUPERLAB software was used to control the display processing.

Data analysis. Images were reconstructed and analyzed with the StatisticalParametric Mapping software (SPM99, Wellcome Department of Cog-nitive Neurology, UK36; implemented in MATLAB 5, Math Works, Nat-ick, Massachusetts). For each subject, images were realigned to the firstscan and then normalized into the MNI stereotaxic space. Data were con-volved using a Gaussian filter with a full-width half maximum (FWHM)parameter set to 12 millimeters.

The design for statistical analysis in SPM was defined as ‘multi-sub-jects and multi-conditions’ with 105 degrees of freedom. Global activityfor each scan was corrected by grand mean scaling. The condition (covari-ate of interest) and subject (confound, fixed effect) effects were estimat-ed voxelwise according to the general linear model. Linear contrasts wereassessed to identify the significant difference between conditions, andwere used to create an SPM t, which was transformed into an SPM zmap. The SPM z maps were thresholded at p < 0.001 (corrected forwhole brain) for conjunction analysis and at p < 0.0001 (corrected forwhole brain) for main effect analysis. Anatomical identification was doneusing atlases both of Talairach and Tournoux37 and of Duvernoy38.

Three conjunction analyses were done. The first was designed to focuson regions activated during first-person simulation compared to controlconditions, irrespective of the presentation modality (A1 – AC in con-junction with V1 – VC). The second was designed to detect brain areasinvolved in third-person simulation compared to control, irrespective ofthe presentation modality (A3 – AC in conjunction with V3 – VC). Thethird was designed to formally identify regions commonly involved infirst- and third-person perspectives (A1 – AC, V1 – VC, A3 – AC and V3 – VC).

Two main effect analyses were done to reveal the brain areas specifical-ly involved in third-person perspective simulation compared to first-per-son perspective ((A3 + V3) – (A1 + V1)) and the reverse ((A1 + V1) – (A3 + V3)).

Post hoc analysis was used to assess task-related regional activity. Theanalysis represented rCBF adjusted values in each task to demonstratethe differential involvement of a given brain area in the six experimen-tal conditions.

ACKNOWLEDGEMENTSThis research was supported by the Cognitique Programme from the French

Ministry of Education. We thank A. Goldman (University of Arizona, Tucson)

and A. Meltzoff (University of Washington, Seattle) for their comments during

the preparation of the manuscript. D. Cardebat (Inserm unit 455, Toulouse,

France) gave us advice on the experimental protocol.

RECEIVED 19 DECEMBER 2000; ACCEPTED 1 FEBRUARY 2001

1. Gopnik, A. & Meltzoff, A. N. Words, Thoughts, and Theories (MIT Press,Cambridge, Massachussetts, 1998).

2. Gallese, V. & Goldman, A. Mirror neurons and the simulation theory ofmind-reading. Trends Cogn. Sci. 2, 493–501 (1998).

3. Goldman, A. in Metarepresentations: A Multidisciplinary Perspective (ed.Sperber, D.) 171–196 (Oxford University Press, New York, 2000).

4. Daprati, E. et al. Looking for the agent: an investigation into consciousness ofaction and self-consciousness in schizophrenic patients. Cognition 65, 71–86(1997).

5. Gallagher, S. Philosophical conceptions of the self: implications for cognitivescience. Trends Cogn. Sci. 4, 14–21 (2000).

6. Decety, J. in The Imitative Mind: Development, Evolution, and Brain Bases (eds.Meltzoff, A. N. & Prinz, W.) (Cambridge Univ. Press, Cambridge, UK, in press).

7. Decety, J. et al. Mapping motor representations with positron emissiontomography. Nature 371, 600–602 (1994).

8. Stephan, K. M. et al. Functional anatomy of the mental representation ofupper extremity movements in healthy subjects. J. Neurophysiol. 73, 373–386(1995).

9. Lotze, M. et al. Activation of cortical and cerebellar motor areas duringexecuted and imagined hand movements: an fMRI study. J. Cogn. Neurosci.11, 491–501 (1999).

10. Ochipa, C. et al. Selective deficit of praxis imagery in ideomotor apraxia.Neurology 49, 474–480 (1997).

11. Sirigu, A. et al. The mental representation of hand movements after parietalcortex damage. Science 273, 1564–1568 (1996).

12. Grèzes, J. & Decety, J. Functional anatomy of execution, mental simulation,observation and verb generation of actions: a meta-analysis. Hum. BrainMapp. 12, 1–19 (2001).

13. Gerardin, E. et al. Partially overlapping neural networks for real andimagined hand movements. Cereb. Cortex 10, 1093–1104 (2000).

14. Jeannerod, M. Mental imagery in the motor context. Neuropsychologia 33,1419–1432 (1995).

15. Iacoboni, M. et al. Cortical mechanisms of human imitation. Science 286,2526–2528 (1999).

16. Goebel, R., Khorram-Sefat, D., Muckli, L., Hacker, H. & Singer, W. Theconstructive nature of vision: direct evidence from functional magneticresonance imaging studies of apparent motion and motion imagery. Eur. J.Neurosci. 10, 1563–1573 (1998).

17. Stevens, J. A., Fonlupt, P., Shiffrar, M. & Decety, J. New aspects of motionperception: selective neural encoding of apparent human movements.Neuroreport 11, 109–115 (2000).

18. Tootell, R. B. et al. Visual motion aftereffect in human cortical area MTrevealed by functional magnetic resonance imaging. Nature 375, 139–141(1995).

19. O’Craven, K. M., Rosen, B. R., Kwong, K. K., Treisman, A. & Savoy, R. L.Voluntary attention modulates fMRI activity in human MT-MST. Neuron 18,591–598 (1997).

20. Kourtzi, Z. & Kanwisher, N. Activation in human MT/MST by static imageswith implied motion. J. Cogn. Neurosci. 12, 48–55 (2000).

21. Decety, J. et al. Brain activity during observation of actions. Influence ofaction content and subject’s strategy. Brain 120, 1763–1777 (1997).

22. Grèzes, J., Costes, N. & Decety, J. Top-down effect of strategy on theperception of human biological motion: PET investigation. Cogn.Neuropsychol. 15, 553–582 (1998).

23. Mesulam, M. M. Dissociative states with abnormal temporal lobe EEG.Multiple personality and the illusion of possession. Arch. Neurol. 38, 176–181(1981).

24. Spence, S. A. et al. A PET study of voluntary movement in schizophrenicpatients experiencing passivity phenomena (delusion of alien control). Brain120, 1997–2011 (1997).

25. Rainville, P. et al. Cerebral mechanisms of hypnotic induction andsuggestion. J. Cogn. Neurosci. 11, 110–125 (1999).

26. Maquet, P. et al. Functional neuroanatomy of hypnotic state. Biol. Psychiatry45, 327–333 (1999).

27. Bisiach, E. & Berti, A. in The Cognitive Neurosciences (ed. Gazzaniga, M. S.)1331–1340 (MIT Press, Cambridge, Massachusetts, 1995).

28. Wolpert, D. M., Goodbody, S. J. & Husain, M. Maintaining internalrepresentations: the role of the human superior parietal lobe. Nat. Neurosci.1, 529–533 (1998).

29. Hurley, S. L. Non-conceptual self-consciousness and agency: perspective andaccess. Commun. Cognit. 30, 207–248 (1997).

30. Kircher, T. T. et al. Towards a functional neuroanatomy of self processing:effects of faces and words. Cogn. Brain Res. 10, 133–144 (2000).

31. Brunet, E., Sarfati, Y., Hardy-Bayle, M. C. & Decety, J. A PET investigation ofthe attribution of intentions with a nonverbal task. Neuroimage 11, 157–166(2000).

32. Fletcher, P. C. et al. Other minds in the brain: a functional imaging study of“theory of mind” in story comprehension. Cognition 57, 109–128 (1995).

33. Kalus, P., Muller, T. J., Zuschratter, W. & Senitz, D. The dendritic architectureof prefrontal pyramidal neurons in schizophrenic patients. Neuroreport 11,3621–3625 (2000).

34. Selemon, L. D. & Goldman-Rakic, P. S. The reduced neuropil hypothesis: acircuit based model of schizophrenia. Biol. Psychiatry 45, 17–25 (1999).

35. Lhermitte, F., Pillon, B. & Serdaru, M. Human autonomy and the frontallobes. Part I: Imitation and utilization behavior: a neuropsychological studyof 75 patients. Ann. Neurol. 19, 326–334 (1986).

36. Friston, K. J. et al. Statistical parametric maps in functional imaging: ageneral linear approach. Hum. Brain Mapp. 3, 189–210 (1995).

37. Talairach, P. & Tournoux, J. A. Co-Planar Stereotaxic Atlas of the HumanBrain (Thieme, Stuttgart, 1988).

38. Duvernoy, H. M. The Human Brain. Surface, Three-Dimensional SectionalAnatomy and MRI (Springer, New York, 1991).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om


Nature Neuroscience May 2001

Documents

Transcript of Nature Neuroscience May 2001