Nature Neuroscience April 1999

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

editorialWhat causes schizophrenia? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295

news and viewsExperience-dependent development of NMDA receptor transmission . . . . . . . . 297Kevin Fox, Jeremy Henley and John Isaac SEE ARTICLE, PAGE 352

A tale of two spikes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299Yves Frégnac

BMPs: time to murder and create? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301Gordon Fishell SEE ARTICLE, PAGE 339

Anandamide: a candidate neurotransmitter heads for the big leagues . . . . . . . . 303David W Self SEE ARTICLE, PAGE 358

book reviewThe brain’s past . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305BY JOHN ALLMAN

Reviewed by Russell D Fernald

scientific correspondenceNerve growth factor acts via retinoic acid synthesis to stimulate neurite outgrowth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307J Corcoran and M Maden

A molecular correlate of memory and amnesia in the hippocampus . . . . . . . . . . 309S M Taubenfeld, K A Wiig, M F Bear and C M Alberini

contents

http://neurosci.nature.com

volume 2 no 4 april 1999

Blood and colleagues demonstratethat the emotional response todissonance in a musical passagecorrelates with activity in severalparalimbic and neocortical brainregions, which are different fromthose previously implicated inother aspects of music perception.Cover illustration by Christie Albin.See page 382.

nature neuroscience • volume 2 no 4 • april 1999 i

Fornix lesions disrupt memoryand CREB expression.

Page 309.

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

Activity-dependent increases inP2X channel pore size.

Pages 315 and 322.

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

199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

http://library.neurosci.nature.com/server-java/Propub/neuro/nn0499_295.pdf








vergara

Resaltado

reviewFrontal cortex contributes to human memory formation. . . . . . . . . . . . . . . . . . . 311R L Buckner, W M Kelley and S E Petersen

articlesPore dilation of neuronal P2X receptor channels . . . . . . . . . . . . . . . . . . . . . . . . . 315C Virginio, A MacKenzie, F A Rassendren, R A North and A Surprenant

Neuronal P2X transmitter-gated cation channels change their ion selectivity in seconds. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322B S Khakh, X R Bao, C Labarca and H A Lester

G-protein-coupled receptors act via protein kinase C and Src to regulate NMDA receptors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331W-Y Lu, Z-G Xiong, S Lei, B A Orser, E Dudek, M D Browning and J F MacDonald

BMPs inhibit neurogenesis by a mechanism involving degradation of a transcription factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339J Shou, P C Rim and A L Calof SEE NEWS AND VIEWS, PAGE 301

Presynaptic depolarization facilitates neurotrophin-induced synaptic potentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346L Boulanger and M-M Poo

Rapid, experience-dependent expression of synaptic NMDA receptors in visual cortex in vivo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352E M Quinlan, B D Philpot, R L Huganir and M F Bear SEE NEWS AND VIEWS, PAGE 297

Dopamine activation of endogenous cannabinoid signaling in dorsal striatum. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358A Giuffrida, L H Parsons, T M Kerr, F Rodríguez de Fonseca, M Navarro and D Piomelli SEE NEWS AND VIEWS, PAGE 303

Involvement of striate and extrastriate visual cortical areas in spatial attention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .364A Martínez, L Anllo-Vento, M I Sereno, L R Frank, R B Buxton, D J Dubowitz, E C Wong, H Hinrichs, H J Heinze and S A Hillyard

A physiological correlate of the ‘spotlight’ of visual attention . . . . . . . . . . . . . . . 370J A Brefczynski and E A DeYoe

Attention activates winner-take-all competition among visual filters . . . . . . . . . . 375D K Lee, L Itti, C Koch and J Braun

Emotional responses to pleasant and unpleasant music correlate with activity in paralimbic brain regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382A J Blood, R J Zatorre, P Bermudez and A C Evans

nature neuroscience • volume 2 no 4 • april 1999 ii

contents

BMPs inhibit neurogenesisvia MASH1.

Pages 301 and 339.

Spatial attention andvisual cortical activation.

Pages 364 and 370.

Experience rapidly inducesNMDA receptor expression.

Pages 297 and 352.


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om













nature neuroscience • volume 2 no 4 • april 1999 295

Schizophrenia remains unexplained. None of the abnormalitiesreported in the brains of schizophrenics is clearly diagnostic for thedisease in the way that (say) plaques and tangles are for Alzheimer’sdisease. In the absence of a clear cellular pathology, the main clues asto the cause are epidemiological. There is general agreement thatgenes and environment are both involved; however, no genes have yetbeen identified, while most of the reported environmental influ-ences are tentative hypotheses at best.

A recent paper1, based on a very large cohort from Denmark,provides what may be the most comprehensive picture to date ofthe epidemiology of schizophrenia. The authors took advantage ofthe excellent civil registry and health care records in that country toanalyze data from 1.75 million people, of whom 2669 developedschizophrenia; this sample included virtually every new case of schiz-ophrenia between 1970 and 1993. The aim was to test the relativeimportance of some of the previously proposed risk factors in a largepopulation.

The data confirm the well-known tendency for schizophrenia torun in families; individuals with a schizophrenic parent or siblingwere almost ten times more likely to develop schizophrenia them-selves, and for those with two affected parents the increase in riskwas almost fifty-fold. The data also confirm a previously reportedand puzzling season-of-birth effect; people born in March had a10% elevated risk, whereas those born in September showed a cor-respondingly reduced risk. Perhaps most surprising is the effect ofplace of birth. Those born in the capital city, Copenhagen, had a 2.4-fold elevated risk compared to those born in rural areas, with inter-mediate risk factors for suburbs and provincial towns. (The risk washigher still for those born in Greenland—which belongs to Den-mark—or in other countries, although sample sizes for those cate-gories were relatively small.)

The implications become apparent when the numbers are trans-lated into population attributable risk (PAR), which takes intoaccount the number of people exposed to each risk factor. The fac-tor with the greatest impact is place of birth, which the authors esti-mate accounts for 34.6% of the total PAR; the combined effect ofplace and season accounts for 41.4%. Taking these numbers at facevalue, if the environmental risk factor(s) could be identified andeliminated, 41.4% of cases of schizophrenia could be prevented.Given that schizophrenia is estimated to affect about 1% of theworld’s population, the potential implications are dramatic indeed.

Clearly these findings will require careful scrutiny. One concernwith any registry-based study is the accuracy of the diagnosis, butDenmark has good psychiatric services, and most of the experts weconsulted felt that diagnostic errors were unlikely to undermine theconclusions. A more serious concern arises from the distinctionbetween risk attributable to family history and risk attributable togenotype as a whole. The authors conclude that family historyaccounts for only 5.5% of the total cases, far less than the 41% attrib-

uted to environmental factors. Yet the contribution of genotype as awhole may be much greater than the family history would suggest.Kenneth Kendler (Virginia Commonwealth University), whodescribes the data as “excellent”, feels that the interpretation is flawed,because most people with a genetic vulnerability will not have anaffected first-degree relative. Thus, although the authors may betechnically correct in attributing only 5.5% of cases to the effect ofparents and siblings, this is likely to substantially underestimate theimportance of genetic effects. Bernard Devlin (University of Pitts-burgh) agrees, and believes that the method used by the authors mayalso lead to an overestimate of the contribution of the environment.It is difficult to guess by how much, he says, but it would clearly bepremature to conclude that any one environmental risk factoraccounts for more cases than does genotype.

Nevertheless, the environmental effects are substantial and seemto demand explanation. One possibility is exposure to infection,either in utero or in early childhood; this would fit well with theeffects of season and urbanization, and might also explain the effectof being born abroad, if for instance the mother is exposed to for-eign pathogens to which she has less immunity. The evidence forthe infection hypothesis, however, is still weak, according to DanielWeinberger (National Institute of Mental Health), who believes thatgenetic explanations remain equally plausible; for instance, allelesthat confer risk of schizophrenia on the offspring might also affect thebehavior of their parents, making them more likely to migrate tocities, or more likely to mate in summer than in winter.

Further progress is likely to depend on the identification of sus-ceptibility genes, which—given the promising signs from linkagestudies—cannot be far away. It would be naive, however, to expect anearly explanation of the disease, particularly given that even betweenmonozygotic twins, concordance is only about 50%. Cloned genesmight provide immediate insights (if, say, their expression is restrict-ed to developing dopamine neurons), but this would be a stroke ofluck indeed. Recall that the genes that cause familial Alzheimer’s dis-ease or Huntington’s disease are ubiquitously expressed and havenot yet led to a clear understanding of either disease process, despitea well-defined cellular pathology. The absence of such signs in schiz-ophrenia may also be a problem in making animal models; how willwe recognize a schizophrenic mouse?

The immediate impact of cloned genes will be on epidemiolo-gy, specifically on the ability to stratify the patient population bygenotype to reveal environmental effects. Epidemiology in turn willprovide important clues in the search for a cellular pathology; if, forinstance, the effects of season and place of birth that are apparent inthe Danish cohort really do signify a prenatal environmental influ-ence, this should motivate an intensive study of brain development,and of the role of susceptibility genes, during the epidemiologicallydefined critical period.1. Mortensen, P. B. et al. N. Engl. J. Med. 340, 603–608 (1999).

editorial

What causes schizophrenia?


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


The maturation of cortical circuit-ry depends critically on experi-ence, as sensory deprivationprevents many of the changes incortical function that normallyoccur with age. During the devel-opment of glutamatergic trans-mission in sensory cortex, theproportion of synapses withdetectable AMPA receptor currentsincreases, and the decay kinetics ofNMDA receptors become faster1,2.Both these changes serve to reducethe relative contribution of NMDAreceptors to synaptic currents andare thought to depend on synapticactivity. The increase in AMPAreceptor currents is thought to bedue to rapid insertion of AMPAreceptors into the synaptic mem-brane under the control of NMDAreceptors3, and the change inNMDA receptor kinetics isbelieved to result from a develop-mental switch in the subunit com-position of NMDA receptors (Fig.1). The NR1 subunit combineswith various NR2A–D subunits to pro-duce receptor subtypes with differentkinetics, of which receptors containingNR2A have the fastest decay times. Onpage 352 of this issue, Quinlan and col-leagues demonstrate that rats deprived ofvisual experience by dark rearing have lowlevels of NR2A-containing receptors invisual cortex, that these receptors arerapidly increased by light exposure, andthat the process is controlled by NMDAreceptors.

Excitatory transmission is very differ-ent in the immature brain compared with

down to adult levels within a fewdays of such exposure6. Thesechanges, which seem to mimicnormal development, could beexplained by increases in AMPAreceptor currents, changes inNMDA receptor currents or both.

This raises the question ofwhether maturational changes inglutamate receptors during normaldevelopment are experiencedependent. Although the propor-tion of synapses with AMPA recep-tor currents increases duringdevelopment in visual cortex7 andin other mammalian sensory sys-tems such as barrel cortex1,3, theeffects of experience on thisprocess in vivo are not yet known.On the other hand, dark rearing oractivity block by direct applicationof tetrodotoxin prevents NMDAreceptor decay kinetics from reach-ing their mature form in rat visualcortex2. The mature kinetics(decay time constant, 100 ms) arecharacteristic of NMDA receptors

composed of NR1 and NR2A subunits,whereas the immature form (350 ms) ischaracteristic of receptors with NR1 andNR2B subunits8. The concept that thedevelopmental change in receptor kineticsdepends on subunit composition in vivogains support from the correlationbetween the timing of NR2A expressionand the switch from slow to fast kinetics9.Moreover, this correlation extends toindividual cells, as cortical neurons thatexpress higher levels of NR2A mRNAhave faster NMDA receptor kinetics10. Byshowing that NR2A expression is sup-pressed in the visual cortex in the absenceof visual experience, Quinlan and col-leagues now provide an explanation forwhy dark rearing delays the change inNMDA receptor kinetics.

So what are the possible mechanismsfor regulation of NMDA receptor sub-units? Receptor proteins are traditionally

Experience-dependent developmentof NMDA receptor transmissionKevin Fox, Jeremy Henley and John Isaac

Light exposure changes the subunit composition and kinetics of NMDA receptors in thedeveloping visual cortex. Quinlan and colleagues suggest that this may be due to rapidsynaptic insertion of receptors containing newly synthesized NR2A subunits.

the adult. In the immature brain, synap-tic transmission is weak, extremely plas-tic and mediated in large part by NMDAreceptors. Some glutamatergic synapsesin young animals have no AMPA recep-tor currents, making them functionally‘silent’ at resting membrane potentials3.In the adult, transmission is stronger, lessplastic and mainly mediated by AMPAreceptors. These developmental changesrequire visual experience, because synap-tic transmission in the visual cortex canbe preserved in an immature state bydelaying the onset of light exposure4. Ifanimals are reared in the dark, visualresponses are weaker and less orientationselective and include larger NMDA recep-tor currents than in light-reared animalsof the same age. However, visual responsesstrengthen within hours of the first expo-sure to light5, and the NMDA receptorcomponent of the visual response shrinks

news and views

Kevin Fox is at the Cardiff School of Biosciences,Cardiff University, Museum Avenue, CardiffCF1 3US, UK. Jeremy Henley and John Isaacare at the MRC Centre for Synaptic Plasticity,Department of Anatomy, Bristol University,University Walk, Bristol BS8 1TT, UK. e-mail: [email protected].

Fig. 1. Changes in postsynaptic glutamate receptors during develop-ment. Left, immature synapses contain NMDA receptors (NMDAR)and no detectable AMPA receptors (AMPAR). NMDA receptors atthese so-called ‘silent synapses’ have slow kinetics and are likely to con-sist of NR1 and NR2B subunits. Right, the mature synapse containsAMPA receptors as well as NMDA receptors containing NR2A sub-units, which have fast kinetics. The transition between the two statesdepends on visual experience. Note that NMDA receptors are shownas pentamers but may be tetramers. The exact stochiometry of thesubunits is not known, and the particular configurations shown hereare only meant to illustrate the possible state of affairs.

Visual experience

NMDAR activation

NR1 NR2B NR2A AMPAR subunits

Amy Center


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

298 nature neuroscience • volume 2 no 4 • april 1999

news and views

thought to be translated in the nucleusand delivered to the synapse via axontransport. However, several factors sug-gest that another mechanism mustaccount for these results. First, NR2A pro-tein levels increase extremely rapidly(within 1 hour) in the synaptoneurosomefraction (a synapse-enriched biochemicalpreparation), too fast to rely on axontransport. Second, the increase is blockedby the protein synthesis inhibitor cyclo-heximide. Taken together, these resultsimply that NR2A subunits are synthesizedlocally in the dendrites. Third, a func-tional change occurs within the same timespan, which suggests that synthesis is fol-lowed by rapid local assembly of subunitsand insertion of NR2A-containing recep-tors into the postsynaptic membrane (Fig.2). Finally, because all this is set in motionby visual experience and blocked byNMDA receptor antagonists, some synap-tically activated second messenger systemsuch as calcium must control the process.

Several conditions would be requiredfor such a mechanism to occur in neu-rons. First, mRNA for NR2A should bepresent in dendrites. Although there isno direct evidence in visual cortical neu-rons, mRNA for glutamate receptors ispresent in dendrites of cultured hip-pocampal neurons11. However, in thisstudy, NR2A mRNA (together withNR2C) was the scarcest of all the gluta-mate receptors. This may indicate thatNR2A mRNA very rarely reaches thedendrites or, more interestingly, that it is

NMDA receptors (Fig. 2) and hence leadsto local translation of the αCaMKIImRNA. Importantly, in this study, CPEBwas found to be enriched in postsynapticdensities, and αCaMKII mRNA was pre-sent in dendrites, so the required machin-ery for translation was present near thesynapse. Therefore, these results provide amodel for how local translation of NR2AmRNA might depend on activity.

Glycine receptors may also be insert-ed via a mechanism similar to the oneproposed for NMDA receptors by Quin-lan and colleagues. There is good evidencethat the α subunits of the glycine recep-tor are synthesized dendritically and theninserted in complexes together with βsubunits, which are produced by conven-tional somatic synthesis (see ref. 14).Interestingly, the mRNAs encoding bothglycine receptor and NMDA receptor sub-units are rather unusual in having verylong 3´ untranslated regions (3´UTRs),which is where CPEB binds on theαCaMKII mRNA. Although the role ofthese long 3´ UTRs is not fully under-stood, they could be involved in the regu-lation of translation14.

Whatever the exact mechanisms ofexpression, one can ask what purposesuch changes in NMDA receptors couldserve during development. One possi-bility is that they form part of a negativefeedback system, whereby the plasticityof the synapse is downregulated as itbecomes more potentiated. The inci-dence of silent synapses decreases withdevelopment, probably as a result ofNMDA receptor activation, which caus-es insertion of AMPA receptors. If thesame intracellular pathway were to causesynthesis and insertion of the NR2A-containing NMDA receptor, then subse-quent activation of the same synapsewould cause less calcium influx perdepolarization because of the fasterdecay kinetics of this NMDA receptorsubtype. This may either make it moredifficult to accumulate sufficient post-synaptic calcium to produce furtherpotentiation, or actually favor a processlike long-term depression, which hasbeen postulated to result from synapticactivity producing low levels of postsy-naptic calcium. Either way, the initialemphasis would be on connecting weaksynapses by a potentiation mechanism(perhaps AMPA receptor insertion trig-gered by NRB-type NMDA receptors).Subsequently, in older animals, thestrength of the newly functional synaps-es could be reduced or erroneouslyformed connections eliminated by

under tight and perhapsinducible regulation. On theother hand, PCR was usedto amplify the mRNA in thisstudy, and very low levels ofdendritic mRNA may simplyresult from a non-specific‘leak’ of somatic mRNA. Asyet, there is little informa-tion on the mechanisms bywhich mRNA may be specif-ically targeted to dendrites.

Second, the machineryfor the translation and post-translational modificationof receptor complexes andfor their assembly and pack-aging into vesicles should bepresent in dendrites close tospines. There is evidencethat dendritic shafts andspines contain polyribo-somes associated with mem-branous cisterns. These arethe so-called synapse-asso-ciated polyribosome com-

plexes or SPRCs for short; (for review,see ref. 12). Although there are no pub-lished data to show that translation actu-ally occurs at these sites, there isevidence for translation at isolatedgrowth cones, where these structures arealso found12. SPRCs are thought to becapable of translation and post-transla-tional modifications such as glycosyla-tion and phosphorylation. In this case,they would also have to assemble thevarious subunits and package them invesicles for membrane insertion. In thissense, SPRCs may act like a mini endo-plasmic reticulum and Golgi apparatusfor dendrites.

Third, synaptic activity should be ableto induce local translation. Unfortunately,there is no direct evidence for local trans-lation of NMDA receptor subunits result-ing from synaptic activity at present.However, there is some evidence for howlocal translation of another mRNA mightbe regulated by synaptic activity. An ini-tial step in translation involves mRNApolyadenylation, which in turn dependson a regulatory protein known as cyto-plasmic polyadenylation element bindingprotein (CPEB). It has been found thatCPEB-dependent polyadenylation of theα subunit of calcium/calmodulin-depen-dent protein kinase II (αCaMKII) mRNAoccurs within 30 minutes of exposingdark-reared animals to the light14. Thisvisual activity probably triggerspolyadenylation via a second messengersystem such as calcium influx through

Fig. 2. Mechanisms for rapid synthesis, assembly and insertionof NMDA receptors. Translation of NR2A mRNA is proposedto be calcium dependent. By activating NMDA receptors, visualexperience leads to translation of new NR2A, which is com-bined with other NMDA receptor subunits in a Golgi-like endo-some and inserted in the membrane. The source of the othersubunits may be recycled NMDA receptors from the postsynap-tic membrane or subunits transported from the cell body.

Internalization

Ca2 +

Assembly

Local synthesisof NR2A

Insertion

Local synthesis

of NRsubunits

Dissassembly,recycling?

NR1NR2BNR2A

Amy Center


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

news and views


Inputs from more distant synapses takelonger to reach the cell body and aremore attenuated than inputs originatingat synapses closer to the cell body. Such aneuron is predicted to behave as a spa-tiotemporal correlator; it is more likely tofire an action potential if all the EPSPsreach the axon hillock simultane-ously. For this to occur, the inputsmust arrive in an appropriate tem-poral relationship to compensatefor their different locations and theresulting delays in reaching the siteof action potential generation.

In reality, however, dendrites arenot passive cables. They express avariety of voltage-dependent ionchannels that modify their bio-physical properties, allowing themnot only to transmit synaptic inputsto the cell body, but also to performconsiderable local processing.Recently, refined techniques, suchas the use of multiple simultaneouspatch electrodes at different pointson a single neuron, have revealed acomplex picture of neuronal func-tion that differs in several impor-tant respects from the classicalmodel. First, ‘hot spots’ of excitabil-ity within the apical dendrites arethought to modify the propagationof EPSPs from synaptic sites to thecell body, so as to reduce the differ-ences of timing and amplitude thatresult from their different locationson the dendritic tree2–4. Second,sodium action potentials originat-ing at the soma–hillock region canpropagate not only along the axons,

Since the discovery that action potentialsinitiated at the soma can propagate backinto dendrites, the role of these spikes inshaping the cell’s response to its inputshas been of intense interest. In a forth-coming issue of Nature, Larkum, Zhu andSakmann1 show that backpropagatedaction potentials in cortical pyramidalneurons can interact with weak synapticinputs in the apical dendrites, triggering adendritic calcium spike. The calciumwave is in turn propagated to the soma,causing the neuron to fire a burst ofaction potentials. This mechanism allowsfor the all-or-none amplification of weakinputs, and may have important implica-tions for how information impinging ondistal dendrites is processed at the cellularlevel within the cortex.

In the classic view of synaptic integra-tion, excitatory postsynaptic potentials(EPSPs) are transmitted passively throughthe dendritic tree to the cell body andaxon hillock, where sodium action poten-tials are generated. The EPSPs aresummed at this ‘decision point’, and if thetotal depolarization reaches threshold, theneuron fires an action potential, which isthen propagated along the axon to therest of the network. In this model, whichhas often been applied to pyramidal cellsof the hippocampus and neocortex, thedendrites are treated as passive cables, andthe effect of a given synapse depends onits location within the dendritic tree.

A tale of two spikesYves Frégnac

Backpropagating action potentials amplify the response to weakdendritic inputs. A new study suggests that this may serve to linksimultaneous inputs to different dendritic compartments.

Yves Frégnac is at the Equipe Cognisciences,Institut Alfred Fessard, CNRS, 91 198, Gif sur Yvette, France. e-mail: [email protected]

but also backward into the dendritic tree5.Third, there exists at least one other sitefor action potential generation, which isdistinct from the soma–axon region. Thissecond site, described in layer-V corticalneurons, is in the tuft of the apical den-drite. This region, which is rich in volt-age-dependent calcium and sodiumchannels, gives rise to calcium spikes6,7.Calcium spikes are typically of muchlonger duration than sodium actionpotentials, but these regenerative eventsinvolving voltage-gated channels are nor-mally attenuated as they spread to thesoma.

The new study in Nature fromLarkum et al.1 is a logical continuation

Fig. 1. The association of a subthreshold distal den-dritic input (1) with a backward propagating somaticaction potential (2) generates a wide dendritic calciumspike (1+2) sufficient to reactivate the soma and causea burst of firing. Courtesy of Dr. Thierry Bal (IAF,CNRS, France).

1 and 2

1

Layers I-II Layer V

1 2

2

synaptic depression (perhaps dephos-phoryation of AMPA channels triggeredby NR2A-type NMDA receptors).

1. Crair, M. C. & Malenka, R. C. Nature 375,325–328 (1995).

2. Carmignoto, G. & Vicini, S. Science 258,1007–1011 (1992).

3. Isaac, J. T. R., Crair, M. C., Nicoll, R. A. &Malenka, R. C. Neuron 18, 269–280 (1997).

4. Fox, K., Daw, N., Sato, H. & Czepita, D. Nature350, 342–344 (1991).

5. Buisseret, P., Garey-Bobo, E. & Imbert, M.Nature 272, 816–817 (1978).

6. Fox, K. Daw, N., Sato, H. & Czepita, D. J.Neurosci. 12, 2672–2684 (1992).

7. Rumpel, S., Hatt, H. & Gottmann, K. J.Neurosci. 18, 8863–8874 (1998).

8. Williams, K., Russell, S. L., Shen, Y. M. &Molinoff, P. B. Neuron 10, 267–278 (1993).

9. Fox, K. Neuron 15, 485–488 (1995).

10. Flint, A. C., Maisch, U. S., Weishaupt, J. H.,Kriegstein, A. R. & Monyer, H. J. Neurosci. 17,2469–2478 (1997).

11. Miyashiro, K., Dichter, M. & Eberwine, J. Proc.Natl. Acad. Sci. USA 91, 10800–10804 (1994).

12. Steward, O. Neuron 18, 9–12 (1997).

13. Wu, L. et al. Neuron 21, 1129–1139 (1998).

14. Kirsh, J., Meyer, G. & Betz, H. Mol. Cell.Neurosci. 8, 93–98 (1996).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


news and views

of previous work from Sakmann and col-leagues, who for several years have beenstudying the role of backpropagatingspikes in neocortical neurons. Whenspikes initiated in the soma are propa-gated backward into the dendrites, theyevoke an activity-dependent influx of cal-cium7,8. It was suggested as early as 1995that these dendritic calcium transientsmight affect both the receptive and theintegrative properties of the dendrites8,providing the cell with a way to regulatedendritic processing, and hence its owninputs. Subsequent work9 showed that ifthe backpropagating action potentialcoincided with EPSPs in the distal den-drites, a long-lasting (several hundredms) calcium wave could be evoked with-in the dendrite, which could not be trig-gered by either the action potential or theEPSPs alone.

The new findings of Larkum and col-leagues advance this story in several ways.First, they show that this effect has a veryprecise time dependence in that the back-propagated action potential must arrivewithin about ten milliseconds of theEPSP to trigger a calcium spike withinthe dendrite. More importantly, theyshow that the dendritic calcium spike isthen propagated to the soma, where itproduces a short burst of sodium actionpotentials riding on the back of the slow-er calcium wave. The dendritic EPSPalone produces no response without thebackpropagating action potential, so thismechanism represents a large amplifica-tion of the response to the dendriticinput (Fig. 1).

What might be the functional signifi-cance of this behavior? Larkum and col-leagues suggest that it may allow layer Vpyramidal neurons to detect associationsbetween two types of cortical inputs. Theweak input to the apical dendrite (input 1in Fig. 1) originates from cortical layers I-II,which receive descending informationfrom higher cortical areas, as well asascending neuromodulatory subcorticalinputs. The stronger inputs that trigger thebackpropagating action potentials (input2 in Fig. 1) would carry sensory informa-tion, which reaches layer V cells via thedeeper cortical layers in which some of thethalamocortical sensory afferents termi-nate. Such a mechanism might allow forperceptual binding of sensory inputs basedon their modulation by contextual infor-mation from higher cortical areas.

This idea, although attractive, muststill be considered speculative. For onething, the amplification effect has onlybeen shown for inputs to the apical tuft

between a positive and negative effect. Asimilar dependence on the relative tim-ing of pre- and postsynaptic activity hasalso been described in the electrosensorylobe of the electric fish14.

The calcium spikes described byLarkum and colleagues show an obviousparallel with these examples of synapticplasticity. Both processes show a simi-larly strong dependence on the relativetiming of pre- and postsynaptic activity,and it is well known that increases inintracellular calcium are involved inmany forms of synaptic plasticity. Thegraphs for the time dependence are qual-itatively similar in both cases, with theinput being amplified/strengthened if thedendritic EPSP occurs coincident withor immediately before the backpropa-gating action potential, and depressed ifthe order is reversed, although thegraphs do not superimpose perfectly(compare Fig. 2 of Larkum et al. withFig. 3 of ref. 13).

Leaving aside the possible relation-ship to long-term synaptic plasticity, theassociative process uncovered by Larkumand colleagues has one immediate andremarkable consequence for under-standing the input–output relationshipsof pyramidal cells: it implies that thesame cell can produce two different out-puts in response to the same input—sin-gle spikes or bursts, depending onwhether the the somatic action potentialproduced by a strong proximal input isaccompanied by a distal input to the api-cal dendrite. It is even possible that den-dritic amplification underlies mostbursting behavior in cortical pyramidalneurons, given that they do not normal-ly fire bursts even in response to pro-longed somatic depolarization (and arethus very unlikely to fire bursts inresponse to a single synaptic input).What about the functional implicationsfor information coding? Much of theinformation content in neuronal spiketrains is thought to be carried by the firstspike rather than by bursts, but there issome evidence that bursts may have aspecial role in the computational process.For instance, in the primary visual cor-tex of behaving monkeys, visually evokedbursts have been found to correlate withthe oculomotor context in which thevisual stimulus occurs (S. Martinez-Conde, S. Macknik and D.H. Hubel, per-sonal communication).

At a more abstract computationallevel, the transition between singlespikes and bursts may reflect a multi-plexing process in which the cell pro-

of the dendrite, and it remains to bedetermined whether it will generalize toother inputs and dendritic compart-ments. Another caveat is that the effectwas much more robust when the authorsused a dendritic patch electrode to mimicthe EPSC (by injecting an appropriatecurrent waveform) than when they usedextracellular stimulation to produce truesynaptic input. The likely explanation forthis discrepancy is that the stimulatingelectrode activates not only excitatory butalso inhibitory inputs to the dendriteunder study, and that inhibition weakensor cancels the effect. Indeed, the authorsprovide two lines of evidence that den-dritic amplification is very sensitive toinhibition. First, amplification was onlyfound reproducibly in their slice prepa-rations when inhibition was partiallyblocked by low concentrations of GABAantagonists. Second, in an elegant exper-iment using paired recordings from inter-connected pyramidal cells and inhibitoryinterneurons, they show that a singleinhibitory PSP (IPSP) is sufficient toblock the calcium wave for 150 ms, andthat a burst of IPSPs can produce a blockthat lasts for as long as 400 milliseconds.Cortical pyramidal neurons in vivoreceive strong tonic inhibition, which sig-nificantly reduces their input resistance10

and should certainly affect the propaga-tion of both sodium and calcium spikesback and forth along the dendrite. Inaddition, excitation and inhibition arecoordinated in vivo11, and the balancebetween the two is likely to be a criticaldeterminant of whether dendritic ampli-fication occurs in any given case.

Another interesting possibility, whichthe authors surprisingly do not discuss,is that dendritic amplification may berelated to associative learning and toHebbian synaptic plasticity (see ref. 12for review). In classical long-term poten-tiation, a strong depolarizing stimulusacts as an unconditioned reinforcingstimulus, which strengthens a weakersynaptic input (corresponding to theconditioned stimulus) when the twoinputs are paired. Similarly, action poten-tials initiated at the cell body can regu-late plasticity at synapses on the distaldendrites via backpropagation13. In thislatter study, synapses between recipro-cally connected layer-V pyramidal neu-rons were strengthened or weakeneddepending on whether the backpropa-gating action potential arrived before orafter the EPSC. The process was highlysensitive to the exact relative timing, with20 milliseconds making the difference


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

news and views


epithelial cultures give rise to mixedcolonies containing both neuronal prog-enitors and differentiated ORNs; the cellsfrom which the colonies arise are thoughtto be the stem cells that give rise to newneurons in vivo6. The authors found,remarkably, that the addition of BMP2, 4or 7 to these cultures completely blocksthe appearance of colonies.

The progression from stem cell to dif-ferentiated neuron is a multi-step processwith at least two defined intermediatestages2. To determine where in thisprocess BMPs might act, the authorsadded BMPs at different times and foundthat the block to ORN productionoccurred within the first twenty-fourhours. This is at least three days beforedifferentiated ORNs begin to appear, sug-gesting that BMPs must block a relativelyearly stage in the ORN lineage. Consistentwith this, early exposure to BMPs greatlyreduced the level of proliferation (as mea-sured by incorporation of [3H]thymidineby neuronal progenitor cells), suggestingthat BMPs act on a still-proliferating pre-cursor rather than on postmitotic neu-rons. BMPs do not cause an immediateincrease in cell death, although apoptoticcell death does occur later.

How might BMPs inhibit develop-ment of ORN precursors? The authorsinvestigated the possibility that it mightact through the transcription factorMASH1 (mammalian achaete-scutehomolog 1), which is expressed at an earlystage in the olfactory receptor lineage7.MASH1 was an attractive candidatebecause mutant mice lacking this proteinshow a phenotype that is very reminiscentof the BMP-treated cultures; matureORNs are almost totally absent from theolfactory epithelium, which instead showsa high level of apoptotic cell death8. Theauthors therefore examined the effect ofBMP treatment on MASH1 expression intheir cultures. Within sixty minutes ofexposure to BMPs, the number ofMASH1-expressing cells fell by fifty per-cent, with a maximal decrease seen after

When it comes to proliferation in themature organism, neurons on the wholeare a timid bunch. During embryogene-sis, massive proliferation occurs in boththe central and peripheral nervous system,but this ends by or soon after birth, andthere is very little neurogenesis in theadult. One exception, however, is theolfactory epithelium, where new olfactoryreceptor neurons (ORNs) continue to beformed throughout life1,2. Work frommany laboratories has suggested that thegeneration of new ORNs is a dynamicallyregulated process. For instance, their rateof production is dramatically increased inresponse to injury3,4 suggesting that neu-rogenesis in the intact epithelium maynormally be repressed. The signal medi-ating this repression has so far remainedelusive, but a paper on page 339 of thisissue suggests that the culprit may be abone morphogenic protein (BMP).

BMPs are a large family of secretedgrowth factors, the original members ofwhich were identified by their ability topromote bone growth. Our view of theirmyriad functions continues to expand,and BMPs have now been shown to act onmost tissues of the body. In the develop-ing nervous system, for instance, BMPsinhibit neural induction, dorsalize thespinal cord and promote cell death in thehindbrain5.

Their presence in the olfactory epithe-lium and their inhibitory effects in otherparts of the nervous system suggested toShou and colleagues that BMPs might alsobe promising candidates for mediating theinhibition of ORN development. To testthis possibility, the authors used a neu-ronal colony-forming assay, in which thevarious steps of ORN proliferation anddifferentiation are recapitulated in culture.If left unperturbed for six days, olfactory

BMPs: time to murder andcreate?Gordon Fishell

The olfactory epithelium produces new neurons throughoutlife. Shou et al. show that BMPs can inhibit this process byinducing degradation of the transcription factor MASH1.

Gordon Fishell is in the Developmental GeneticsProgram, The Skirball Institute, NYU MedicalCenter, 540 First Ave., 4th Floor, Lab 7, New York, New York 10016, USA. e-mail: [email protected]

duces different outputs depending onwhen and where in the dendritic treethe input occurred. This in turn couldallow the cell to recognize specific inputpatterns15. When such patterns (givingrise to subthreshold inputs at the api-cal dendrites) are associated with astrong input (sufficient to trigger abackpropagating action potential), thecell would signal this by firing a burst.One can speculate that such transfor-mations—in which an irregular tempo-ral pattern of presynaptic spikes istransformed into a pattern of bursts bythe postsynaptic neuron—might formthe cellular basis for the emergence oflarge functional assemblies of neurons.The dominance of synchronized burst-ing of neurons in different layers anddifferent columns produced duringintense association of ascending anddescending information may underlieepisodic perceptual binding at the cor-tical level. Burst behavior, in turn, mayextend the temporal window duringwhich a given cortical cell can detect thearrival of a combination of inputs indistinct dendritic compartments andpromote the boosting and eventuallythe strengthening of otherwise sublim-inal synaptic influences.

1. Larkum, M. E., Zhu, J. J. & Sakmann, B.Nature (in press).

2. Adams, P. Curr. Biol. 2, 625–627 (1992).

3. Johnston, D., Magee, J. C., Colbert, C. M. &Cristie, B. R. Annu. Rev. Neurosci. 19,165–186 (1996).

4. Softky, W. Neuroscience 58, 13–41 (1994).

5. Stuart, G. J. & Sakmann, B. Nature 367,69–72 (1994).

6. Stuart, G., Schiller, Y. & Sakmann, B. J.Physiol. (Lond.) 505, 617–632 (1997).

7. Schiller, J., Schiller, Y., Stuart, G. &Sackmann, B. J. Physiol. (Lond.) 505,605–616 (1997).

8. Markram, H., Helm, P. J. & Sakmann, B. J.Physiol. (Lond.) 485, 1–20 (1995).

9. Markram, H. Cereb. Cortex 7, 523–533(1997).

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

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

12. Frégnac, Y. in Handbook of Brain Theory andNeural Networks (ed. Arbib, M.) 459–464(MIT Press, Cambridge, Massachusetts,1995).

13. Markram, H., Lübke, J., Frotscher, M. &Sakmann, B. Science 275, 213–215 (1997).

14. Bell, C. C., Han, V. Z., Sugawara, Y. & Grant,K. Nature 387, 278–281 (1997).

15. Mel, B. Neural Computation 4, 502–517(1992).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


news and views

two hours. This rapid loss ofMASH1 cannot be explainedsimply by cessation of new syn-thesis, because the half-life ofthe protein in these cells (asmeasured using cycloheximideto block protein synthesis) is atleast four hours. Instead,MASH1 must be activelydegraded; consistent with this,its disappearance can be pre-vented by pharmacologicalinhibitors of proteasome activ-ity, suggesting that BMP caus-es MASH1 protein to besomehow targeted for protea-some-mediated proteolysis.One trivial explanation for theresults might be that MASH1degradation reflects the non-specific proteolysis that pre-cedes apoptotic cell death. Thisseems unlikely, however,because cell death does notoccur until much later (20hours). A more direct testwould be to show that the dis-appearance of MASH1 is inde-pendent of caspases (whichmediate proteolysis duringapoptosis), but this experimentmay be difficult, given thediversity of caspases and thecomplexity of their regulation.

The findings of Shou andcolleagues suggest the follow-ing scenario (Fig. 1a): BMPsinduce a rapid proteasome-mediated degradation ofMASH1 protein at an earlystage in the ORN lineage. Theprogenitors that lose MASH1cease proliferation and ulti-mately undergo apoptosis, ter-minating the lineage beforedifferentiated ORNs can begenerated. To prove this model, onewould ideally like to block MASH1 pro-teolysis and show that proliferation anddifferentiation can then proceed even inthe presence of BMPs. However, thisexperiment is not straightforward;although proteasome inhibitors can beused for short-term studies on MASH1degradation, prolonged exposure is toxicto these cultures, precluding any study ofthe longer-term consequences for prolif-eration or differentiation. It may eventu-ally be possible to identify and mutate thesites in MASH1 that cause it to be target-ed for proteolysis, but this would involvesubstantial further work. Hence, althoughthe phenotype of the MASH1-deficient

MASH1 seems to have asimilar role in both lineages,which is perhaps not surpris-ing given the abundant evi-dence that similartranscription factors act todetermine neural fate andidentity in widely divergentphyla10. In both the olfactoryepithelium11 and the neuralcrest12, MASH1 is expressedat an early stage and thendownregulated before differ-ention. Moreover, mice lack-ing MASH1 show a loss bothof ORNs and of sympatheticand enteric ganglionic neu-rons (the latter two of whicharise from the neural crest)8.The more surprising result isthat BMP signaling seems tohave opposite effects in thetwo lineages; in neural crestcells, it induces MASH1 geneexpression and promotesneurogenesis, whereas inolfactory epithelial cells, itinduces the degradation ofMASH1 protein and inhibitsneurogenesis.

To make sense of this result,we must first consider the roleof MASH1 in more detail. Inthe case of neural crest cells(which have been better stud-ied), the downregulation ofMASH1 is as critical for theirterminal differentiation as itsearlier expression was for theirproliferation and survival. Twolines of evidence support thisidea. First, when neural crestcells are infected with a retro-virus that drives MASH1expression, they are induced toexpress early differentiation

markers, such as Phox2a and c-RET, evenin the absence of BMP signaling13. Yet,most of these cells do not become fullydifferentiated neurons; even though mostof the colonies that develop from cellsinfected with the virus go on to develop atleast some neurons, these still representonly a small proportion of the cells withina given colony. Moreover, the cells that dobecome neurons seem to be those thatexpress lower levels of MASH1 (perhapsdue to silencing of the viral enhancer),suggesting that the downregulation ofMASH1 may be required for their differ-entiation14. A second piece of evidencecomes from further analysis of theMASH1-deficient mutant mice15; cells

mice is certainly suggestive of a causalrelationship, it may be some time beforethis can be tested directly.

Interestingly, this is not the first timethat MASH1 and the BMPs have beenfound to cross paths. Work by Andersonand colleagues has suggested a quite dif-ferent role for BMPs on MASH1 regulationin the neural crest9. In that system, BMPsact instructively to induce MASH1 expres-sion in neural crest stem cells, causing themto adopt a neuronal fate (Fig. 1b). (Likeolfactory epithelial cells, neural crest cellsrespond to both BMP2 and BMP4,although unlike olfactory cells, they showno apparent response to BMP7; the basisfor this different selectivity is not known.)

Fig. 1. Role of BMP signaling in development of olfactory and sympatheticneuronal lineages. (a) Olfactory epithelial stem cells give rise to a MASH1-positive progenitor, which can adopt different fates. In the presence of BMP,MASH1 is rapidly degraded, and the cell subsequently dies by apopotosis. Inthe absence of BMP signaling, it goes on to proliferate, forming intermediateneuronal progenitors and eventually postmitotic olfactory receptor neu-rons. (b) Neural crest stem cells are induced by some but not all BMPs toform MASH1-positive progenitors, which later downregulate MASH1 andgive rise to sympathetic neurons. (In the absence of BMP signaling, the stemcells give rise to other cell types.) (c) The two lineages share some commonfeatures, including a requirement for MASH1 expression at an early stage,and for its downregulation at a later stage. Both lineages are regulated byBMP signaling; its different effects in the two cases may reflect subtle differ-ences in their sensitivities to different doses at different developmentalstages, rather than any fundamental difference in its mode of action.

A Olfactory receptor neurons

B Sympathetic neurons

C General model

Olfactory receptor neurons

ApoptosisMASH1degraded

ApoptosisMASH1degraded

Intermediate neural progenitor

MASH1 +veprogenitor

(BMP?)

BMP2,4,7

Olfactory epitheliumstem cell

Neural creststem cell

BMP 2,4

MASH1 +veprogentor

Sympathetic neurons

Stem cell

BMP

BMP

BMPOlfactory receptor

or sympathetic

neuron

MASH1 +veprogentor

Amy Center

a

b

c


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

news and views


involve dysfunction of dopamine signaling.The name anandamide is derived from

the Indian Sanskrit term ananda, meaning‘bliss and tranquillity’1, undoubtedly in ref-erence to psychoactive effects of cannabi-noids in humans. Anandamide belongs toa class of molecules called eicosaniods, andit was first isolated based on its hydrophobicproperties, by analogy with exogenouscannabinoids such as ∆9-THC6. It isexpressed throughout the brain, and it ismost prevalent in the hippocampus, stria-tum, cerebellum and cortex, structures thatregulate learning, movement and cognition,among other behaviors. Another endo-cannabinoid, 2-arachidonylglycerol (2-AG),which was discovered more recently, is evenmore highly expressed in the brain1. Bothmolecules fulfill at least some of the criteriafor neurotransmitter status. They both acti-vate the brain cannabinoid receptor CB1,and both have putative biosynthetic path-ways. (They are synthesized from arachi-donic acid and phospolipids.) Anandamidealso has a putative mechanism for its inac-tivation via re-uptake and intracellulardegradation. Being hydrophobic molecules,neither anandamide nor 2-AG is packagedinto synaptic vesicles (in contrast to con-ventional neurotransmitters); instead, theyare thought to be released by phospholi-pase-mediated cleavage followed by passivediffusion across the plasma membrane1.

Because of their low (micromolar)affinities for the CB1 receptor, however,many investigators were skeptical as towhether anandamide or 2-AG ever attainsufficient concentrations to activate the

Endocannabinoids are endogenous sub-stances that mimic the psychoactive effectsof marijuana on cannabinoid receptors1.The story of their discovery goes back to thelast decade, when pharmacological andmolecular studies2,3 led to the identificationof a G-protein-coupled receptor that wasactivated by ∆9-tetrahydrocannabinol (∆9-THC), the major psychoactive substance inmarijuana. Just as the existence of opioidreceptors led to the discovery of endogenousopioid neurotransmitters in the 1970s4, theidentification of the brain cannabinoidreceptor CB1 spurred a search for naturallyoccurring ligands within the brain.

Several endogenous ligands for the CB1receptor have been discovered, but none hasyet been shown to function as a neuro-transmitter. In this issue of Nature Neuro-science, Giuffrida and colleagues report thatlocal depolarization can trigger the releaseof anandamide, the first endocannabinoididentified, in the striatum of awake, freelymoving rats5. They also show that anan-damide release can be stimulated bydopamine receptors, and that this leads tothe inhibition of dopamine-mediated loco-motor behavior via cannabinoid receptors.Their findings promise to propel anan-damide from candidate status to bona fideneurotransmitter, and may also open thedoor to novel treatments for diseases that

Anandamide: a candidateneurotransmitter heads forthe big leaguesDavid W. Self

Activation of dopamine receptors triggers release ofanandamide, an endogenous cannabinoid, in vivo, leading toinhibition of dopamine-mediated locomotor behavior.

David Self is in the Division of MolecularPsychiatry, Yale University School of Medicine,Connecticut Mental Health Center, 34 Park St.,New Haven, Connecticut 06508, USA. e-mail: [email protected]

within the ventral telencephalon thatwould normally express MASH1 showpremature differentiation, again suggest-ing that the continued expression ofMASH1 may normally serve to inhibitprogression to the fully differentiated state.

How then can we explain the appar-ently opposite effects of BMPs in olfac-tory and neural crest lineages? It is ofcourse possible that BMPs might some-how produce opposite effects on MASH1in each cell type. A more attractive pos-sibility, however, is that underlying sig-naling pathways are fundamentallysimilar in the two cases (Fig. 1c), and thatBMPs can induce both the appearanceand the degradation of MASH1 in bothlineages (David Anderson, personal com-munication). This would allow BMPs toact as both promoters and inhibitors ofneuronal fates, depending on preciselywhen they act. In such a model, thechoice between differentiation and deathcould depend on the exact timing andamount of BMP signaling relative to theprogenitor cells’ changing responsivenessover time.

Clearly, the function of BMPs in theolfactory epithelium is far from resolved,and the findings of Shou and colleaguesraise a number of interesting questions.Does the BMP-mediated degradation ofMASH1 actually cause the cessation of celldivision and the onset of apoptosis? Whatis the molecular link between BMP sig-naling and the proteolysis of MASH1? IsMASH1 the only molecule targeted byBMPs for degradation, or are there oth-ers? Does the level of one or more BMPact to control the rate of olfactory neuro-genesis in vivo? No doubt, future experi-ments will soon address these issues. Inthe interim, it seems that BMPs have onceagain dropped a question on our plate.

1. Graziadei, P. P. & Graziadei, G. A. J.

Neurocytol. 8, 1–18 (1979).

2. Calof, A. L., Mumm, J. S., Rim, P. C. & Shou,J. J. Neurobiol. 36, 190–205 (1998).

3. Schwartz-Levey, S., Chikaraishi, D. M. &Kauer, J. S. J. Neurosci. 11, 3556–3564(1991).

4. Holcomb, J. D., Mumm, J. S. & Calof, A. L.Dev. Biol. 172, 307–323 (1995).

5. Hogan, B. L. Genes Dev. 10, 1580–1594(1996).

6. Mumm, J. S., Shou, J. & Calof, A. L. Proc.Natl. Acad. Sci. USA 93, 11167–11172(1996).

7. Cau, E., Gradwohl, G., Fode, C. & Guillemot,F. Development 124, 1611–1621 (1997).

8. Guillemot, F. et al. Cell 75, 463–476 (1993).

9. Shah, N. M., Grove, A. & Anderson, D. J. Cell85, 331–343 (1996).

10. Anderson, D. J. & Jan, Y. N. in Molecular andCellular Approaches to Neural Development (eds.Cowan, W. M., Jessell, T. M. & Zipursky, S. L.)26–63 (Oxford Univ. Press, New York, 1997).

11. Gordon, M. K., Mumm, J. S., Davis, R.,Holcomb, J. D. & Calof, A. L. Mol. Cell.Neurosci. 6, 363–379 (1995).

12. Lo, L., Johnson, J. E., Wuenschell, C. W.,

Saito, T. & Anderson, D. J. Genes Dev. 5,1524–1537 (1991).

13. Lo, L., Sommer, L. & Anderson, D. J. Curr.Biol. 7, 440–450 (1997).

14. Lo, L., Tiveron, M.-C. & Anderson, D. J.Development 125, 609–620 (1998).

15. Casarosa, S., Fode, C. & Guillemot, F.Development 126, 525–534 (1999).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


news and views

receptor in the brain. Although certainmemory- and anxiety-enhancing effects ofthe cannabinoid receptor antagonistSR 141716 suggest that endocannabinoidsare tonically active, the new findings of Giuf-frida and colleagues5 lay this concern to restfor anandamide.

The authors focused on the striatum,which expresses high levels of CB1 receptors.They induced depolarization in the stria-tum of awake rats and showed that this leadsto the release of anandamide. Release of 2-AG, in contrast, did not reach detectable lev-els either before or after stimulation. Thissuggests that anandamide is the main lig-and for striatal cannabinoid receptors,although it remains possible that otherendocannabinoids might also be involved.

The authors found that anandamiderelease could also be induced by pharmaco-logical activation of the D2 class ofdopamine receptors, which are known to beimportant in striatal function. D2 receptoractivation causes rats to become hyperac-tive, so the authors asked whether anan-damide release might be involved in thisbehavioral response. They found that block-ing the CB1 receptor with a pharmacological

may have important therapeutic implica-tions for the development of treatments formovement disorders. For example, drugsthat block endocannabinoid effects at theCB1 receptor could potentiate or prolongthe therapeutic efficacy of dopamine-basedtreatment strategies currently used inParkinson’s disease while having minimaleffects on their own. In contrast, drugs thatstimulate the CB1 receptor could reducedyskinesias associated with Huntington’sdisease or antipsychotic treatment, possiblyat doses with minimal psychoactive effects.Indeed, CB1 receptor binding is decreasedin target regions of striatal neurons duringthe early stages of Huntington’s disease11,12,suggesting that alterations in endocannabi-noid signaling possibly contribute to the dis-ease pathology itself.

Given their diffuse localizationthroughout the brain, it is likely that anan-damide and other endocannabinoidsinteract with multiple neurotransmittersin ways that are yet to be discovered. Inaddition to reward and anxiety, behavioralstudies suggest an interaction betweenendocannabinoid systems and appetite,pain, epilepsy and other behavioral states1.As these complex interactions are unrav-eled and understood, endocannabinoidsystems are likely to gain appreciation as aprominent signaling pathway in the brain,which could open the door to new treat-ment strategies for a variety of disordersassociated with these behaviors.

1. Felder, C. C. & Glass, M. Annu. Rev. Pharmacol.Toxicol. 38, 179–200 (1998).

2. Howlett, A. C. & Fleming, R. M. Mol.Pharmacol. 26, 532–538 (1984).

3. Matsuda, L. A., Lolait, S. J., Brownstein, M. J.,Young, A. C. & Bonner, T. I. Nature 346,561–564 (1990).

4. Kosterlitz, H. W. & McKnight, A. T. Adv. Intern.Med. 26, 1–36 (1980).

5. Giuffrida, A. et al. Nat. Neurosci. 2, 358–363(1999).

6. Devane, W. A. et al. Science 258, 1946–1949(1992).

7. Terranova, J. P. et al. Psychopharmacology 126,165–172 (1996).

8. Navarro, M. et al. Neuroreport 8, 491–496(1997).

9. Glass, M. & Felder, C. C. J. Neurosci. 17,5327–5333 (1997).

10. Van Ree, J. N., Slangen, J. F. & De Wied, D. J.Pharmacol. Exp. Ther. 204, 547–557 (1978).

11. Takahashi, R. N. & Singer, G. Pharmacol.Biochem. Behav. 11, 737–740 (1979).

12. Koob, G. F. & LeMoal, M. Science 278, 52–58(1997).

13. Glass, M., Faull, R. L. & Dragunow, M.Neuroscience 56, 523–527 (1993).

14. Richfield, E. K. & Herkenham, M. Ann. Neurol.36, 577–584 (1994).

antagonist potentiated D2-recep-tor-induced hyperactivity, where-as it had no effect on baselineactivity. Their results suggest thatanandamide can reach a suffi-cient concentration to producefunctional effects, but only afterstimulation of D2 receptors.Thus, the released anandamideseems to function as a ‘brake’ thatlimits the behavioral response toD2 receptor activation.

Although the source of thereleased anandamide in the stria-tum is not yet known, both pre-and postsynaptic elements of stri-atal architecture contain D2receptors that could trigger anan-damide release (Fig. 1). Presy-naptic D2 receptors couldstimulate anandamide releasefrom dopamine terminals to pro-vide negative feedforward regu-lation of postsynaptic D2-receptor-mediated locomotorbehavior, in conjunction with thepresynaptic D2 receptor’s nega-tive feedback effects on dopaminerelease itself. Alternatively, post-synaptic D2 receptors could stim-ulate anandamide release fromstriatal neurons as a negativefeedback mechanism on thesame CB1-receptor-containingneurons. In view of the latter pos-

sibility, it is interesting that CB1 receptorsapparently switch coupling from inhibitionto activation of adenylyl cyclase when stim-ulated concurrently with D2 receptors,thereby counteracting the inhibitory effectsof D2 receptors on the cyclase7. Yet anotherpossibility is that anandamide release is trig-gered indirectly by other neurotransmittersystems within the striatum that are mod-ulated by D2 dopamine receptors.

The pleasant psychoactive effects ofcannabinoids in humans are well known,and some studies have reported that labo-ratory animals will self-administer cannabi-noids intravenously8,9. However, otherstudies suggest that systemically adminis-tered cannabinoids produce anxiety anddysphoria and oppose reward mechanismsin rodents1. Because dysphoria and anhe-donia have been associated with reduceddopamine levels in striatal subregions10,cannabinoid-induced inhibition ofdopamine-mediated behavior may con-tribute to these aversive effects of exogenouscannabinoids.

In any event, the interaction of endoge-nous cannabinoids with dopaminergic sys-tems reported by Giuffrida and colleagues

Dopamine

terminal

Locomotor activation

?+

+

-

-

+

Anandamide

Dopamine

CB1 Receptor

D2 Receptor

?

Fig. 1. Possible pathways by which D2 dopamine receptorscould trigger anandamide release. Presynaptic D2 receptors,which inhibit dopamine release, could further reduce itsbehavioral effects by causing anandamide release, whichinhibits dopamine-induced locomotion. Alternatively, postsy-naptic D2 receptors could cause anadamide release, reducingtheir locomotor effects by negative feedback.

Amy Center


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


scientific correspondence

Nerve growth factor actsvia retinoic acid synthesisto stimulate neuriteoutgrowthJonathan Corcoran and Malcolm Maden

Developmental Biology Research Centre, The Randall Institute, King’s College London, 26–29 Drury Lane, London WC2B 5RL, UK

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

Nerve growth factor (NGF) stimulates neurite outgrowth from cul-tured adult dorsal root ganglia (DRG)1, and the vitamin A derivativeall-trans-retinoic acid (tRA) induces neurite outgrowth from variousembryonic sources, including DRG2,3. Are such similarities in effectsof NGF and tRA because they are both components of the samegenetic cascade leading to neurite outgrowth? tRA upregulates low-and high-affinity NGF receptors3,4 and induces the transcription ofNGF itself5, suggesting that tRA may be upstream of NGF, but weshow the converse, namely, that NGF is upstream of tRA.

When adult mouse DRG are cultured in the presence of NGFand an inhibitor of tRA synthesis, neurite outgrowth does not occur.Conversely, when tRA is added along with a blocking antibody toNGF, neurite outgrowth occurs as normal. We further show thatNGF induces transcription of both the retinoic-acid-synthesizingenzyme RALDH-2 and the retinoic acid receptor-β as well asdetectable release of synthesized tRA. We propose that tRA isrequired for adult DRG neurite regeneration and that NGF actsupstream of tRA to induce its synthesis.

Cellular effects of tRA are mediated by binding to nuclear recep-tors that are ligand-activated transcription factors. There are twoclasses of receptors, retinoic acid receptors (RARs) and retinoid Xreceptors (RXRs), with three subtypes of each: α, β and γ6,7. RARreceptors mediate gene expression by forming heterodimers withthe RXRs, whereas RXRs can mediate gene expression either ashomodimers or by forming heterodimers with orphan receptors.An additional mechanistic association between NGF and tRA path-ways is suggested by the findings that the nuclear receptor NGFIBheterodimerizes with the RXRs8 and that NGFIB is rapidly inducedin PC12 cells by the administration of NGF9.

Although it is clear that tRA stimulates neurite outgrowth fromembryonic DRG2,3, it is not yet known if the same occurs in adultDRG. Adult mouse DRG were cultured in the presence of NGF (100ng per ml) or tRA (100 nM) for five days. In both cases, neurite out-growth occurred (Fig. 1b and data not shown). Little or no neuriteoutgrowth occurred in control adult DRG cultured in delipidatedserum (Fig. 1a), giving significant differences in number of neuritesfrom NGF- and tRA-treated cultures (Fig. 2a). When tRA was addedtogether with NGF, there was no additive effect of the two treatments(Fig. 1c), and no significant difference was found between tRA, NGFor tRA plus NGF groups (Fig. 2a). Although it may be that bothNGF and tRA are at individual saturating concentrations, the lackof synergy may also imply that NGF and tRA act through the samepathway to cause neurite outgrowth. One could imagine either tRAinducing the production of NGF5 or NGF inducing the productionof tRA by stimulating a tRA-synthesizing enzyme.

To test which of these is most likely, we cultured adult DRG inthe presence of NGF and 10 µM disulphiram, a compound thatblocks the conversion of retinaldehyde to tRA by inhibiting theenzyme aldehyde dehydrogenase10. Addition of disulphiram com-

pletely abolished NGF-induced neurite outgrowth (Fig. 1d) com-pared to NGF alone or to NGF and DMSO (vehicle for disulphi-ram; Fig. 2a). To confirm that disulphiram did not affect cell survivalwithin the explants, we used two types of rescue. In both cases,explants were cultured for eight days in medium supplemented withdisulphiram. In the first rescue, tRA was added to the explants fromthe beginning of the experiment; in the second, tRA was added onday four. Neurite outgrowth occurred in both of the tRA-rescuedbut not in the disulphiram-alone cultures (Figs. 1e and f and 2b).

Inhibition of the inductive effect of NGF but not of tRA by disul-phiram suggests that NGF may precede tRA in the cascade leading toneurite outgrowth. To test this, we used a blocking antibody againstNGF. In the presence of NGF and the blocking antibody, virtuallyno neurite outgrowth occurred (Fig. 1g; compare to DRG culturedin the presence of NGF alone, Fig. 1b). On the other hand, DRGcultured in the presence of the NGF-blocking antibody and tRA(Fig. 1h) showed neurite outgrowth equivalent to that obtained withNGF alone (Figs. 1b and 2c).

If NGF is upstream of tRA, it should induce synthesis of tRA afteraddition to DRG cultures. To test this, we used an F9 reporter cellline that responds to the presence of tRA because of transfection

Fig. 1. Neurite outgrowth in adult mouse DRG. Cells were cultured forfive (a–d, g, h) or eight days (e, f) in the presence of delipidated serumplus (a) no addition, (b) NGF, 100 ng per ml, (c) NGF and 100 nM tRA,(d) NGF and 10 µM disulphiram, (e) disulphiram and tRA added on day0, (f) disulphiram, (g) NGF and blocking antibody or (h) NGF-blockingantibody and tRA.

a b

c d

e f

g h


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 2. Neurite numbers, tRA synthesis andgene induction in adult mouse DRG after vari-ous treatments. (a) Effects on neurite numberat five days (1, no additive; 2, NGF, 100 ng perml; 3, tRA, 100 nM; 4, NGF, 100 ng per ml andtRA, 100 nM; 5, 100 ng per ml NGF and 10 µMdisulphiram; 6, NGF, 100 ng per ml andDMSO). Error bars, s.e., n = 6. *p < 0.01;**p < 0.0001, compared to +NGF, Student’s t-test. (b) tRA rescue of DRG treated with 10

µM disulphiram (left to right, no tRA; 100 nM tRA, day 0; 100 nM tRA, day 4). Error bars’ s.e., n = 6. *p < 0.01; **p < 0.0001, compared to no tRA, Student’st-test. (c) NGF-blocking antibody on 5-day DRG cultures. Left, NGF, 100 ng per ml; center, NGF plus blocking antibody; right, blocking antibody plus 100 nMtRA. Error bars’ s.e., n = 6. *p < 0.01, compared to NGF + α-NGF, Student’s t-test. (d) Increase in percentage β-galactosidase-positive F9 cells in response tocultured DRG. Left, no additive; center, NGF, 100 ng per ml; right, NGF plus blocking antibody. Cells were counted in three separate fields, and the experi-ment repeated three times. Error bars’ s.e., n = 9. *p < 0.025 compared to no tNGF, Student’s t-test. (e) RT-PCR analysis of RALDH-2 enzyme and RARβexpression in adult DRG cultured with or without NGF (100 ng per ml) for five days. GAPDH was used to indicate presence of cDNA in both samples.

with 1.8 kb of the mouse RARβ2 gene promoter containing a retinoicacid response element linked to the lacZ gene11. In the presence oftRA, activated cells can be detected after β-galactosidase histo-chemical staining. NGF itself does not activate these cells, as therewas no labeling of the F9 cells above background in the presence ofNGF. We then cultured adult DRG in delipidated serum for five daysunder three different conditions: no NGF, with NGF or NGF plusthe NGF-blocking antibody. The DRG were then sonicated andplaced on the F9 reporter cells. NGF-treated DRG homogenates pro-duced a clear RA signal relative to untreated DRG (Fig. 2d). Thisactivation was prevented when the DRG were cultured with block-ing antibody in addition to NGF (Fig. 2d).

We next considered which tRA-synthesizing enzyme might beinduced by NGF. Retinol is converted by a two-step oxidative process,first to retinaldehyde and then to retinoic acid (for review, see ref.12). Retinaldehyde dehydrogenase type 2 (RALDH-2) is expressed inthe developing nervous system13. Using RT-PCR, we found stronginduction of RALDH-2 by NGF in cultured adult DRG (Fig. 2e).Finally, we also found upregulation of the RARβ receptor in NGF-stimulated cultures (Fig. 2e).

Our results show that tRA can stimulate neurite outgrowth froman adult neural tissue, the DRG. NGF similarly stimulates neuriteoutgrowth from this tissue, and we have demonstrated that it does soby inducing tRA synthesis via an enzyme, RALDH-2. In the pres-ence of either an NGF-blocking antibody or an inhibitor of tRA syn-thesis, NGF fails to act. Thus the most likely sequence of events inthe induction of neurite outgrowth by NGF is NGF→RALDH-2→tRA→RARβ→ neurite outgrowth. We have not yet determinedif NGF is directly responsible for inducing RALDH-2, or if someintermediary protein is required for this process. However, as NGFIBis one of the earliest genes induced by NGF9 and its product can het-erodimerize with the RXRs8, the NGFIB/RXR heterodimer may beresponsible for activating the RALDH-2 gene. Neurotrophins have

been considered as potential agents for induction of nerve regener-ation14 and treatment of neurodegenerative diseases15, but a majorproblem for their use is lack of effective modes of delivery to the siteof injury. Because tRA is required for the regenerative response andis downstream of NGF, then the problem of delivery to the lesioncould be overcome, as tRA is a low-molecular-weight lipophilic com-pound that can be administered orally. Thus, tRA may be of use inclinical neurology.

ACKNOWLEDGEMENTSThis work was supported by a project grant from The Wellcome Trust and the

BBSRC.

RECEIVED 8 NOVEMBER 1998; ACCEPTED 27 JANUARY 1999

1. Lindsay, R. J. Neurosci. 8, 2394–2405 (1988).2. Quinn, S. D. P. & De Boni, U. In Vitro Cell. Dev. Biol. 27A, 55–62 (1991).3. Haskell, B. E., Stach, R. W., Werrbach-Perez, K. & Perez-Polo, J. R. Cell Tissue Res.

247, 67–73 (1987).4. Rodriguez-Tebar, A. & Rohrer, H. Development 112, 813–820 (1991).5. Wion, D., Houlgatte, R., Barbot, N., Barrand, P., Dicou, E. & Brachet, P. Biochem.

Biophys. Res. Commun. 149, 510–514 (1987).6. Kastner, P., Chambon, P. & Leid, M. in Vitamin A in Health and Disease (ed.

Blomhoff, R.) 189–238 (Dekker, New York, 1994).7. Kliewer, S. A., Umesono, K., Evans, R. M. & Mangelsdorf, D. J. in Vitamin A in

Health and Disease (ed. Blomhoff, R.) 239–255 (Dekker, New York, 1994)8. Mangelsdorf, D. J. & Evans, R. M. Cell 83, 841–850 (1995).9. Millbrandt, J. Neuron 1, 183–188 (1988).10. McCaffery, P., Lee, M.-O., Wagner, M. A., Sladek, N. E. & Drager, U. Development

115, 371–382 (1992).11. Maden, M., Sonneveld, E., van der Saag, P. T. & Gale, E. Development 125,

4133–4144 (1998)12. Duester, G. Biochemistry 35, 12221–12227 (1996).13. Drager, U. C. & McCaffery, P. in Enzymology and Molecular Biology of Carbonyl

Metabolism Vol. 5 (eds. Weiner, H. et al.) 185–192 (Plenum, New York, 1995).14. Schnell, L., Schneider, R., Kolbeck, R., Barde, Y.-A. & Schwab, M. E. Nature 367,

170–173 (1994).15. Schatzl, H. M. Trends Neurosci. 18, 463–464 (1995).



aA

vera

ge

nu

mb

er o

fn

euri

tes

per

ad

ult

DR

Gb

c

Ave

rag

e n

um

ber

of

neu

rite

s p

er a

du

lt D

RG

Per

cen

t in

crea

se in

th

e n

um

ber

of

blu

e ce

lls

–tRA +tRA +tRA day 0 day 4

+disulph +disulph +disulph

NGF NGF + tRA +anti-NGF anti-NGF

–NGF +NGF +NGF +anti-NGF

eRALDH–2

RAR–β

GAPDH

–NGF +NGF

+NGF +tRAControl +NGF+tRA

+NGF+disulph

+NGFDMSO

d

Ave

rag

e n

um

ber

of

neu

rite

s p

er a

du

lt D

RG


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om



A molecular correlate ofmemory and amnesia inthe hippocampusStephen M. Taubenfeld1, Kjesten A. Wiig2, Mark F. Bear2

and Cristina M. Alberini1

1 Department of Neuroscience, Brown University, Providence, Rhode Island 02912, USA

2 Howard Hughes Medical Institute, Brown University, Providence, Rhode Island 02912, USA

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

Memory consolidation in humans and other species is profoundlydisrupted by lesions of either the medial temporal lobes or regions ofthe thalamus1–3. It has been proposed that these structures regulatethe neuronal gene expression necessary for long-term memory4.Evidence suggests that long-term memory formation requires theactivity of members of the cAMP response element (CRE) bindingprotein (CREB) transcription factor family5,6, and that CRE-regu-lated genes are expressed in the hippocampus in response to inhibito-ry avoidance training7,8. Here we show that lesions of the fornix, amassive fiber bundle connecting the hippocampus with the septumand hypothalamus, specifically disrupt both consolidation ofinhibitory avoidance memory and CREB-mediated responses in thehippocampus. We propose that inputs passing through the fornixregulate this memory consolidation by regulating CREB-mediatedgene expression in hippocampal neurons.

Rats were given electrolytic lesions of the fornix and were trainedon an inhibitory avoidance task one week after surgery (Fig. 1).Lesioned rats did not differ from controls in initial latency to enterthe dark chamber (p > 0.05). The two groups of animals did, how-ever, differ in their overall retention profiles. ANOVA revealed a sig-nificant group × delay interaction (F3,42 = 2.988, p < 0.04) and asignificant main effect of group (F1,14 = 20.559, p < 0.0005). At 0

and 6 hours, control and fornix-lesioned rats showed similar reten-tion, whereas at 24 and 48 hours, lesioned rats were severelyimpaired. This time course suggests that deficits in fornix-lesionedanimals reflect impaired memory consolidation.

CREB has a fundamental role in memory consolidation5,9,10,and CRE-mediated gene expression is induced in the hippocampusfollowing inhibitory avoidance training8. Because phosphorylationof CREB at Ser-133 is a necessary step for CREB-dependent tran-

Fig. 1. Fornix lesions produceimpairment on the inhibitoryavoidance task. (a) Represen-tative lesion (asterisk). Surgerywas done on male Long-Evansrats (200–250 g) as described15.In 25 of 32 lesions, the dorsalfornix was severed and the fim-bria extensively damaged. Inaddition, the anterior aspects ofthe lateral and triangular septalnuclei and the septofimbrialnucleus were at least partiallydamaged in all subjects. Theremaining seven animals hadpartial damage to the fornix andfimbria, and only minor damageto the septal nuclei. No damageto the underlying thalamic struc-tures or to the hippocampal for-mation was observed in anycase. (b) Inhibitory avoidancetraining involved placing the ratin a lighted chamber connectedwith a dark chamber. Ten seconds later, the door separating the chamberswas opened, allowing the rat to enter the dark chamber, where it received afootshock (2 s, 1.5 mA). Retention was assessed 0, 6, 24 and 48 hours laterby returning the rat to the lighted chamber and measuring latency to enterthe dark chamber. Fornix-lesioned rats performed similarly to controls ininitial training and in retention tests zero and six hours later. By 24 h, fornix-lesioned rats were severely impaired.

0

100

200

300

400

500

600

tim

e sp

ent i

n li

gh

t ch

amb

er (s

eco

nd

s)

acquisition 0 6 24 48

delay (hours)

fornix lesionsham-operated

a

*

Fig. 2. Inhibitory avoidance training increases hip-pocampal CREB phosphorylation in normal rats butnot in rats with fornix lesions. Hippocampi fromunoperated or lesioned rats were homogenized inlysis buffer (0.2 M NaCl, 0.1 M HEPES, 10% glycerol,2 mm NaF, 2 mM Na4P2O7, 5 mM EDTA, 1mMEGTA, 2 mM DTT, 0.5 mM PMSF, 1 mM benzami-dine, 10 µg per ml leupeptin, 400 U per ml aprotinin,1 µM microcystin) and resolved by SDS-PAGEbefore electroblotting. Membranes were incubatedwith anti-Ser133-PCREB (1:2000) or anti-CREB(1:1000) antisera (Upstate Biotechnology, LakePlacid, New York) and visualized with ECL(Amersham, Arlington Heights, Illinois) before den-sitometric analysis (NIH Image). (a) PCREB andCREB western blots of hippocampal extracts fromunoperated and fornix-lesioned animals. Examplesare shown for three conditions: killed withoutreceiving shock on entering chamber, killed immedi-ately after training shock in chamber (0 h) and killedsix hours after training shock (6 h). (b–d)Densitometric analysis of western blots of hip-pocampi taken from unoperated and fornix-lesioned animals at various timepoints after training and compared to ‘no-shock’ controls. (b) Unoperated ani-mals showed significant increases (p < 0.05) in PCREB over ‘no-shock’ controls (n = 8) at 0 h (n = 8), 3 h (n = 4) and 6 h (n = 8) after training. In contrast,lesioned animals showed no increase in PCREB after training (n = 4 per timepoint). (c) Total CREB is unchanged in both lesioned and unoperated rats 0, 3 or6 h after training (n = 4 per timepoint). (d) PCREB levels in unshocked rats were unchanged at 0, 3, 6 or 9 h (n = 8, 4, 8, 4) after exposure to the apparatus.

a

unoperated

no shock 0 hr 6 hr

fornix lesion

PCREB

CREB

PCREB

CREB

b

0

50

100

150

200

noshock

0 3 6 9

% o

f no

sho

ck

time (hours)

fornix lesionunoperated

PCREB d

100

150

200

0

50no shock

3 6 9

time (hours)

PCREBc

0

50

100

150

200

noshock

0 3 6

time (hours)

CREB

0

a

Delay (hours)

Tim

e sp

ent

in li

gh

t ch

amb

er (

s)

b

a

b

no shock 0 h 6 h

Unoperated

Fornix lesion

PCREB

PCREB

CREB

CREB

Time (hours)

c

Time (hours) Time (hours)

d

Per

cen

t o

f n

o s

ho

ck


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


scriptional activation11, we analyzed western blots of phosphorylat-ed CREB (PCREB) in the hippocampus following inhibitory avoid-ance training in normal and fornix-lesioned rats (Fig. 2).

In unoperated rats immediately after training, hippocampalPCREB increased to 152.3 ± 12.4% of levels in control animals thatentered the dark chamber but received no shock and were immedi-ately killed (Fig. 2a and b). PCREB was increased three, six and ninehours after training (158.6 ± 8.3%, 155.1 ± 12.4%, 150.7 ± 28.6%,respectively). One-way ANOVA revealed a significant main effect oftime (F4,26 = 3.346, p < 0.02), and Dunnet post-hoc comparisonsconfirmed that PCREB was elevated over control ‘no-shock’ levelsat 0, 3 and 6 hours after training (p < 0.05). Western blots showed nochange in CREB expression after training (Fig. 2a and c). Thus, train-ing increased phosphorylation of hippocampal CREB protein.

To determine whether increased PCREB in unoperated animalswas related to the consolidation of the inhibitory avoidance task orto other stimuli evoked by exposure to the apparatus, we measuredPCREB in hippocampi of unoperated animals that entered theinhibitory avoidance chamber but received no shock (Fig. 2d). CREBphosphorylation was unchanged from immediately killed controllevels 3, 6 or 9 hours after exposure to the apparatus (80.3 ± 6.3%,101.6 ± 8.9%, 107.5 ± 12.2%; p > 0.05). Similarly, no change inPCREB was observed in animals that received the shock only(86.5 ± 7.1%; n = 4). These data suggest that persistent elevation inPCREB was specifically associated with consolidation of inhibitoryavoidance memory.

Immunohistochemical staining in hippocampi of untrained ani-mals revealed low levels of PCREB (for example, Fig. 3a). PCREBstaining was variable in neurons of the dentate gyrus and CA3 andgenerally negative in CA1. Inhibitory avoidance training, however,produced a strong, regionally specific increase in PCREB immuno-stained neurons in CA1 and dentate gyrus (Fig. 3b) and, to someextent, in CA3 (data not shown). PCREB immunoreactivityincreased immediately after training (Fig. 3a and b) and persistedthree and six hours after training (data not shown).

We next investigated whether the PCREB increase followinginhibitory avoidance training is affected by fornix lesions. Westernblots of hippocampal extracts from untrained, lesioned rats showedbasal levels of PCREB (117.4 ± 4.6%) comparable to those found inunoperated no-shock controls. However, unlike levels in controls,the levels of PCREB in hippocampi of rats with fornix lesions didnot increase either zero or six hours after training (108.7 ± 11.2%,80.1 ± 6.7%; Fig. 2a and b). Two-way ANOVA (F5,28 = 6.713,p < 0.0003) and Student Newman Keuls post-hoc analysis revealedsignificantly lower CREB phosphorylation 0 and 6 hours after train-

ing in lesioned rats than in unoperated rats (p < 0.05).Total hippocampal CREB was unchanged after trainingin both lesioned and control groups (Fig. 2c). Immuno-histochemistry confirmed that fornix lesions preventedPCREB induction by training in CA1 and dentate gyrus(Fig. 3c and d); no induction was found in any hip-pocampal subregions in lesioned rats. Thus, hippocam-pal CREB phosphorylation induced by inhibitoryavoidance training is prevented by fornix lesions.

It is interesting that, when tested at early timepoints, fornix-lesioned animals show normal learningand memory but not hippocampal PCREB increases.Moreover, lesioned rats display significant (althoughclearly impaired) memory 24 h after training. Initiallearning and memory could reflect synaptic modifica-tions independent of new protein synthesis12. Resid-ual memory at 24 h might be explained by changesoutside the hippocampus, or by mechanisms that do

not involve increased CREB phosphorylation.How might fornix lesions disrupt both PCREB increases and

consolidation of long-term inhibitory avoidance memory? The hip-pocampus receives projections from the septum, hypothalamus andbrain stem via the fornix. One hypothesis is that these axons con-vey signals regulating CREB-dependent gene expression in hip-pocampus. Indeed, if activity in the fornix is temporarily arrestedimmediately before inhibitory avoidance training, memory deficitsare comparable to those produced by lesioning, whereas inactiva-tion 48 h after training and before testing has no effect13. Thus, activ-ity in the fornix is necessary for consolidation, but not expression,of memory. In contrast, inactivation of the dorsal hippocampus pre-vents both memory encoding and retrieval14.

Initial learning is likely to result from changes in the transmis-sion of synapses conveying information about where the animal is inspace. Whether or not these changes are made permanent dependson the timely occurrence of new gene expression. We propose thatsignals to hippocampal neurons via the fornix contribute to memoryconsolidation by modulating CREB-dependent gene expressionrequired for establishment of long-term memory. Identification ofthe critical chemical signals, their transduction pathways and thegenes regulated may suggest treatments for amnesia associated withdamage to the temporal lobe memory system.

ACKNOWLEDGEMENTS

This work was supported by the Whitehall Foundation (grant # F97-07), the Charles

A. Dana Foundation and the Howard Hughes Medical Institute. The authors thank

Deborah Brenner, Stephane Nedelec, Eric Sklar, Suzanne Meagher and Arnold

Heynen for their assistance.

RECEIVED 11 DECEMBER 1998; ACCEPTED 23 FEBRUARY 1999

1. Scoville, W. B. & Milner, B. J. Neurol. Neurosurg. Psychiatry 20, 11–21 (1957).2. de Castro, J. & Hall, T. W. Physiol. Psychol. 3, 201–204 (1975).3. Squire, L. R. Memory and Brain (Oxford, New York, 1987).4. Bear, M. F. Nature 385, 481–482 (1997).5. Yin, J. C. P. et al. Cell 79, 49–58 (1994).6. Silva, A. J., Kogan, J. H., Frankland, P. W. & Kida, S. Annu. Rev. Neurosci. 21,

127–148 (1998).7. Bernabeu, R. et al. Proc. Natl. Acad. Sci. USA 94, 7041–7046 (1997).8. Impey, S. et al. Nat. Neurosci.1, 595–601 (1998).9. Yin, J. C. P., Del Vecchio, M., Zhou, H. & Tully, T. Cell 81, 107–115 (1995).10. Bourtchuladze, R. et al. Cell 79, 59–68 (1994).11. Montminy, M. Annu. Rev. Biochem. 66, 807–822 (1997).12. Frey, U. & Morris, R. G. M. Nature 385, 533–536 (1997).13. Baldi, E., Lorenzini, C. A., Sacchetti, B., Tassoni, G. & Bucherelli, C. Brain Res.

799, 270–277 (1998).14. Lorenzini, C. A., Baldi, E., Bucherelli, C., Sacchetti, B. & Tassoni, G. Brain Res.

730, 32–39 (1996).15. Wiig, K. A., Cooper, L. N. & Bear, M. F. Learn. Mem. 3, 313–325 (1996).


Fig. 3. CREB phospho-rylation after inhibitoryavoidance learning isinduced mainly in CA1and dentate gyrus (DG).Examples of anti-PCREB(1:1000) staining in 40-µm coronal brain sec-tions from unoperated (a and b) and lesioned (c and d) rats without(no shock) and immedi-ately after footshocktraining (0 h; n = 4 pergroup).

CA1

DG

b

d

a

c

no shock 0 hb

d

Un

op

erat

edFo

rnix

lesi

on


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


Primate models of memory and multiple lines of evidence fromneuropsychology suggest that medial temporal lobe brain struc-tures are essential for memory formation. Moreover, their func-tion seems to be specific to forms of memory based on consciousreflections of the past, such as those experienced during a recol-lection of a vivid memory or of a newly learned fact. In contrast,other expressions of memory such as habit formation and skilllearning that occur without conscious awareness seem to be inde-pendent of medial temporal lobe participation1–3.

An open question has been how, and to what degree, corticalareas outside the medial temporal lobes contribute to consciousremembering. This review focuses on a set of findings emergingfrom brain-imaging studies that provide evidence for an impor-tant role of frontal regions in memory formation. Like the medialtemporal lobes, the frontal lobes seem to be involved in formingmemories that support conscious recollection of the past. Further-more, activity within specific frontal regions correlates with a widerange of behavioral factors that influence memory formation.

Frontal activity during encodingAs might be expected, intentional memorization is a highly effec-tive means of encoding words into long-term memory. Positronemission tomography (PET) and functional magnetic resonanceimaging (fMRI) studies consistently demonstrate that specific areaswithin left frontal cortex are active when subjects intentionallymemorize words4–6. Brain activity increases most often within pos-terior regions of frontal cortex near the border between motor andprefrontal cortex (located dorsally along inferior frontal gyrus, nearBrodmann’s areas 44 and 6) and also within more ventral pre-frontal regions (near Brodmann’s areas 44, 45 and 47; Fig. 1).

Most instances of memory formation in everyday life, how-ever, occur incidentally, without any intention to remember. Howdoes this happen? Cognitive psychology research suggests thatmemory formation is a byproduct of certain kinds of informa-tion processing. For example, items for which meaning and rela-tionship to other remembered items are elaborated are betterremembered than items processed in a shallow fashion whereonly surface characteristics are examined7.

Frontal cortex contributes tohuman memory formation

Randy L. Buckner, William M. Kelley and Steven E. Petersen

Departments of Psychology, Radiology, Neurology and Anatomy and Neurobiology, Washington University, Campus Box 1125, One Brookings Drive, St. Louis, Missouri 63130, USA

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

The contribution of medial temporal lobe structures to memory is well established. However recentbrain-imaging studies have indicated that frontal cortex may also be involved in human memory for-mation. Specific frontal areas are recruited during a variety of procedures that promote memory for-mation, and the laterality of these areas is influenced by the type of information contained in thememory. Imaging methods that capture momentary changes in brain activity have further shownthat the likelihood of memory formation correlates with the level of activity in these areas. Theseresults, taken in the context of other studies, suggest that memory formation depends on jointparticipation of frontal and medial temporal lobe structures.

Frontal regions active during intentional memorization arealso active during behavioral manipulations that incidentallyalter the effectiveness of memory encoding. For example, whensubjects make meaning-based judgments about words, multipleregions within left frontal cortex are activated. Those words areremembered even though the subjects make no explicit attemptat memorization8–10. By contrast, if subjects perform a surface-based task where words are judged, for example, to be in upper-case or lowercase letters, frontal activity is minimal, and the wordsare most often forgotten8–10. Dividing attentional resources alsoseems to influence memory encoding. When attention is direct-ed away from an item, that item is likely to be forgotten even if asubject is attempting to remember it11. Consistent with thisbehavioral observation, one influential study using PET demon-strated that adding a secondary distracting task during inten-tional memorization caused brain activity in frontal cortex todiminish and memory to be impaired12.

Perhaps the strongest evidence for frontal involvement inencoding comes from studies examining event-by-event vari-ance in memorization. In everyday life, some experiences areremembered, whereas others are forgotten. Although the levelof meaning-based processing or direction of attention to thesevarious experiences may account for a portion of this variabil-ity, it is often unclear what makes a particular experience mem-orable. This issue was initially addressed by electricalscalp-recording techniques. When scalp potentials of subjectswere recorded at the time of memorization, distinct neural sig-natures were noted for words that were later remembered ascompared to those for words later forgotten13,14 (reviewed inref. 15). Recent developments in fMRI methods16–20 haveallowed the same phenomenon to be examined with more pre-cise spatial localization within the brain (Fig. 3). Such proce-dures show that, on average, the level of activity within frontalcortex can predict whether an item will later be rememberedor forgotten. This is true of both words10 and picture scenes21.These findings strongly suggest that frontal activity can be influ-enced by (or influences) subtle differences between events thataffect memory encoding.

review© 1999 Nature America Inc. • http://neurosci.nature.com

© 1

999

Nat

ure

Am

eric

a In

c. •

htt

p:/

/neu

rosc

i.nat

ure

.co

m


Frontal involvement in nonverbal memorySo far, we have focused on processing of words. Left frontal cor-tex has been consistently associated with encoding of verbal mate-rials22. However, we are also able to remember aspects of eventsthat are not purely verbal in nature, such as a tune on the radio orthe appearance of a neighbor’s dog. How are these kinds of infor-mation remembered? Cognitive psychologists have long believedthat memory formation relies on multiple streams of informa-tion, most often distinguished as verbal and nonverbal codes.The most compelling behavioral evidence for this possibility isthe finding that a picture of an object such as a dog is more like-ly to be remembered than the presentation of the word “dog”.The implication is that pictures are associated with both non-verbal (image-based) and verbal codes, whereas words are asso-ciated with only a verbal code23,24.

Further evidence that multiple codes contribute to memoryprocesses comes from studies of brain-damaged patients25–28.These studies suggest that verbal and nonverbal codes may beprocessed in different hemispheres. For example, ‘split-brain’patients—epileptic individuals who have had communicationbetween their cerebral hemispheres disrupted to minimize thespread of seizure activity—perform significantly better on testsof face memorization when the faces are presented to the righthemisphere rather than to the left hemisphere26

In addition, memorization of materials associated with dif-ferent codes can activate distinct regions of left and right frontalcortex (Fig. 2). Memorization of unfamiliar faces6 and texturepatterns29, neither of which can be easily associated with a ver-bal label, activate right frontal regions. Of particular interest isthe finding that memorization of nameable objects—items thatcan be associated with both verbal- and nonverbal imagery-basedcodes—often elicits bilateral activation of frontal cortex6. Anintriguing interpretation of this finding is that distinct regionsof frontal cortex will, for a single event, code the multi-ple kinds of information available.

Lateral differences are also noted across the event-by-event studies discussed above. Left frontal activitypredicted which words would be remembered10, where-as right frontal activity predicted which picture sceneswould be remembered21. Taken together, these studiessuggest that, depending on the kind of informationbeing memorized, multiple lateralized regions withinfrontal cortex may participate in encoding.

Moreover, experimental conditions can also affectthe lateralization of regions engaged during memoryformation. For example, left frontal activity is observedduring memorization of faces under conditions wherelong intervals occur between face presentations30,31. Ver-bal encoding operations that rely on left frontal activi-

ty may be more accessible under these condi-tions, although this possibility has not yetbeen tested directly.

Frontal cortex damageDespite the support for a role of frontal cor-tex in encoding from brain-imaging studies,its implication in memory formation by stud-ies of brain-damaged patients is less clear.Memory disturbances following frontal dam-age have been noted to varying degrees,depending on the exact kind of memory test.In particular, memory difficulties are observedin frontal patients when the test requires rec-

ollection of a particular context (for example, remembering whomade a particular statement) or judgment of the timing of anevent32–34. However, patients with frontal damage often performwell on many other tests of memory, such as simple recognitionof past items. Preservation of certain memory abilities followingfrontal damage has often been contrasted to the profound amne-sia that can follow damage to the medial temporal lobes.

There are several possible explanations for this apparent incon-gruity. First, left and right frontal regions seem to be used for dif-ferent materials, strategies and contexts. Thus, for limited unilaterallesions, remaining frontal regions either in the same or oppositehemisphere may be able to overcome difficulties in many encodingsituations. Similarly, in medial temporal injury, damage usuallymust be bilateral to produce profound memory deficits.

Second, regions related to memory encoding do not exist sole-ly to encode information for later retrieval. Rather, the function ofcertain frontal regions may concurrently relate to both memoryencoding and immediate task demands. Frontal cortex is impli-cated in a variety of functions, including higher-level thought andplanning35,36. Several hypotheses propose that certain regions with-in frontal cortex participate in short-term maintenance andmanipulation of information, which are essential to many kinds ofinformation-processing tasks. One possibility is that frontal cor-tex may serve as an internal buffer, allowing manipulation of infor-mation and, in turn, selection among possible responses to thatinformation36–38. The dual nature of this processing presents a for-midable challenge to the study of memory formation. Memoryencoding may be the eventual byproduct of information process-ing engaged by the frontal lobes for other reasons. By this view, asingle processing event may have two quite distinct effects. Thefirst is observed during the event and relates to immediate taskcompletion (for example, working-memory judgment or wordgeneration). The second effect is observed tangentially but lies at

review

Fig. 1. A lateral view of the humanbrain schematically illustrates frontalregions active during tasks that pro-mote memorization of verbal mate-rials. These regions may be involvedin memory encoding. A number ofstudies exploring intentional andincidental encoding as well as theirevent-by-event variances convergeto show that activity within thesefrontal regions correlates with highlevels of memory performance.

Fig. 2. Frontal regions active during memory encoding may depend on the materialsbeing memorized. Data revealing lateralization differences across verbal (words) andnon-verbal (faces) materials. (a) Words activate left frontal cortex, whereas (c) facesactivate right frontal cortex. By contrast, (c) objects (associated with both an imageand a name) activate both right and left frontal cortex. The data6 show fMRI activity incoronal sections associated with intentional encoding of different materials.

a b c


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


the heart of long-term memory encoding: a process engaged todeal with present task demands initiates a cascade of events thatforms a memory for that episode. Thus, processing events asso-ciated with frontal activity promote memory formation eventhough much of the activity may arise for reasons other thanintentional memorization. Because memory studies that usebrain-damaged patients often exclude patients with specific verbalprocessing deficits, there may be a ‘catch-22’ in trying to associ-ate memory function with frontal injury. Given the possible dualfunction of frontal activity, frontal lesions that produce speechand verbal fluency impairments may also have important butoften-overlooked effects on memory encoding39. In two tellingstudies, patients with speech and fluency difficulties typical of leftfrontal damage did poorly on recognition tests of studiedwords40,41. Both studies also included patients with deficits likelyarising from damage to similar regions in the right hemisphere.These patients were impaired at remembering items associatedwith nonverbal codes including pictures40,41 and birdsongs40.

Relation to the medial temporal lobeOne obvious issue to be raised at this point is the relation betweenfrontal and medial temporal regions. Damage to the medial tem-poral lobes is often associated with nearly complete or partialloss of the ability to remember new experiences in the presence ofrelatively intact cognitive functioning in other domains (forexample, see refs. 1, 2 and 42). Memory disturbances noted earlyin the progression of Alzheimer’s disease are associated withpathology in this region43–45. Similarly, non-human primate mod-els of memory loss suggest the lesion location for producingmemory impairment falls within the hippocampus and adjacentcortex (within the medial temporal lobes)46,47. These data con-vincingly demonstrate that medial temporal lobe regions are crit-ical for memory formation.

Despite the accumulation of evidence from other methods,medial temporal lobe structures remain a somewhat elusive targetfor brain imaging. Whereas a number of studies observe medialtemporal lobe activity during encoding (particularly in theparahippocampal cortex48), a number do not—even in the pres-ence of robust activity within frontal cortex49. Despite numerousattempts to do so, relatively few studies have reported consistentactivity directly within the hippocampus during various task con-ditions that differ in their ability to promote memory formation.

Nonetheless, how medial temporal and frontal brain regionsmay jointly participate in memory formation is suggested by datafrom other sources. Areas within the medial temporal lobesreceive inputs from many cortical regions and may contribute tomemory formation by associating or ‘binding’ these inputs withalready-present contents of long-term memory2,50. According tothis view, damage to the medial temporal lobe disrupts memo-ry formation by preventing the construction of novel, often arbi-

trary, associations that characterize new experiences. Sev-eral investigators have noted that the few imaging studiesdetecting hippocampal activity during memory encodingexplore tasks that involve complex new items or associa-tions among items5,6,48,51–54.

So how does the frontal cortex fit into this picture? Onespeculation would be that the critical cascade drivinghuman memory formation occurs only when frontal activ-ity provides information to medial temporal lobe struc-tures. The medial temporal lobe may then function to bindtogether processed information from frontal and othercortical regions to form lasting, recollectable memorytraces. Thus, both regions would be critical to the con-

ception of a memory, and lack of participation of either brainregion would disrupt memory formation.

A consequence of this hypothesis is that activity within frontalcortex should fail to instill a memory if the medial temporal lobesare damaged. Indeed, this prediction is upheld by recent find-ings. Patients with amnesia due to medial temporal lobe dam-age can show normal frontal activity patterns associated withencoding, yet these patients fail to form new memories49,55. Cor-respondingly, as discussed above, normal subjects with intactmedial temporal lobes fail to form memories when frontal activ-ity is absent. Thus the interaction of frontal and medial tempo-ral regions, rather than the isolated contribution of either region,seems to be crucial for the effective formation of memories thatcontribute to consciousness of past events.

ACKNOWLEDGEMENTSNeal Cohen, Anthony Wagner, David Donaldson, Jessica Logan and Amy

Sanders provided comments. This work was supported by grants from the

National Institute of Mental Health (MH57506-01), the McDonnell Center for

Higher Brain Function and the Human Frontiers Science Program.

RECEIVED 8 OCTOBER 1998; ACCEPTED 13 JANUARY 1999

1. Squire, L. R. Memory and the hippocampus: a synthesis from findings withrats, monkeys, and humans. Psychol. Rev. 99, 195–231 (1992).

2. Cohen, N. J. & Eichenbaum, H. Memory, Amnesia, and the HippocampalSystem (MIT Press, Cambridge, Massachusetts, 1993).

3. Schacter, D. L. & Tulving, E. Memory Systems 1994 (MIT Press, Cambridge,Massachusetts, 1994).

4. Kapur, S. et al. The neural correlates of intentional learning of verbal materials:a PET study in humans. Brain Res. Cogn. Brain Res. 4, 243–249 (1996).

5. Dolan, R. J. & Fletcher, P. C. Dissociating prefrontal and hippocampalfunction in episodic memory encoding. Nature 388, 582–585 (1997).

6. Kelley, W. M. et al. Hemispheric specialization in human dorsal frontal cortexand medial temporal lobe for verbal and nonverbal memory encoding.Neuron 20, 927–936 (1998).

7. Craik, F. I. M. & Tulving, E. Depth of processing and the retention of words inepisodic memory. J. Exp. Psychol. Gen. 104, 168–294 (1975).

8. Kapur, S. et al. Neuroanatomical correlates of encoding in episodic memory:levels of processing effect. Proc. Natl. Acad. Sci.USA 91, 2008–2011 (1994).

9. Demb, J. B. et al. Semantic encoding and retrieval in the left inferiorprefrontal cortex: a functional MRI study of task difficulty and processspecificity. J. Neurosci. 15, 5870–5878 (1995).

10. Wagner, A. D. et al. Building memories: remembering and forgetting ofverbal experiences as predicted by brain activity Science 281, 1188–1191(1998).

11. Craik, F. I., Govoni, R., Naveh-Benjamin, M. & Anderson, N. D. The effects ofdivided attention on encoding and retrieval processes in human memory. J.Exp. Psychol. Gen. 125, 159–180 (1996).

12. Shallice, T. et al. Brain regions associated with acquisition and retrieval ofverbal episodic memory. Nature 368, 633–635 (1994).

13. Fabiani, M., Karis, M. & Donchin, E. P300 and recall in an incidentalmemory paradigm. Psychophysiology 23, 298–308 (1986).

14. Paller, K. A. Recall and stem-completion priming have differentelectrophysiological correlates and are modified differentially by directedforgetting. J. Exp. Psychol. Learn. Mem. Cogn. 16, 1021–1032 (1990).

15. Rugg, M. D. in Electrophysiology of Mind: Event-Related Brain Potentials andCognition (eds. Rugg, M. D. & Coles, M. G. H.) 132–170 (Oxford Univ. Press,Oxford, 1995).

review

Fig. 3. Activity within frontal regionscan predict which items will beremembered or forgotten. Depictedis the fMRI signal from a left dorsalfrontal region during an incidentalmemory encoding task. The data10

were divided based on subsequentmemory performance such that thebrain activity associated with later-remembered words could be sepa-rated from activity associated withlater-forgotten words. Time (s)

Raw

MR

I sig

nal

ch

ang

e

Remembered

Forgotten


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


16. Buckner, R. L. et al. Detection of cortical activation during averaged singletrials of a cognitive task using functional magnetic resonance imaging. Proc.Natl. Acad. Sci. USA 93, 14878–14883 (1996).

17. Dale, A. M. & Buckner, R. L. Selective averaging of rapidly presentedindividual trials using fMRI. Hum. Brain Mapp. 5, 329–340 (1997).

18. Josephs, O., Turner, R. & Friston, K. Event-related fMRI. Hum. Brain Mapp.5, 243–248 (1997).

19. Kim, S. G., Richter, W. & Ugurbil, K. Limitations of temporal resolution infunctional MRI. Magn. Reson. Med. 37, 631–636 (1997).

20. Zarahn, E., Aguirre, G. & D’Esposito, M. A trial-based experimental designfor fMRI. Neuroimage 6, 122–138 (1997).

21. Brewer, J. B., Zhao, Z., Glover, G. H. & Gabrieli, J. D. E. Making memories:brain activity that predicts how well visual experience will be remembered.Science 281,1185–1187 (1998).

22. Tulving, E., Kapur, S., Craik, F. I. M., Moscovitch, M. & Houle, S. Hemisphericencoding/retrieval asymmetry in episodic memory: positron emissiontomography findings. Proc. Natl. Acad. Sci. USA 91, 2016–2020 (1994).

23. Paivio, A. & Csapo, K. Picture superiority in free recall: imagery or dualcoding? Cogn. Psychol. 5, 176–206 (1973).

24. Paivio, A. Mental Representations (Oxford Univ. Press, New York, 1986).25. Milner, B. & Taylor, L. Right-hemisphere superiority in tactile pattern-

recognition after cerebral commissurotomy: evidence for nonverbal memory.Neuropsychologia 10, 1–15 (1972).

26. Gazzaniga, M. S. & Smylie, C. S. Facial recognition and brain asymmetries:clues to underlying mechanisms. Ann. Neurol. 13, 536–540 (1983).

27. Milner, B. Some cognitive effects of frontal-lobe lesions in man. Phil. Trans.R. Soc. Lond. B 298, 211–226 (1982).

28. Jones-Gotman, M. & Milner, B. Design fluency: the invention of nonsensedrawings after focal cortical lesions. Neuropsychologia 15, 653–674 (1977).

29. Wagner, A. D. et al. Material-specific lateralization of prefrontal activationduring episodic encoding and retrieval. Neuroreport 9, 3711–3717 (1998).

30. Haxby, J. V. et al. Face encoding and recognition in the human brain. Proc.Natl. Acad. Sci.USA 93, 922–927 (1996).

31. Grady, C. L. et al. Age-related reductions in human recognition memory dueto impaired encoding. Science 269, 218–221 (1995).

32. Milner, B., Petrides, M. & Smith, M. L. Frontal lobes and the temporalorganization of memory. Hum. Neurobiol. 4, 137–142 (1985).

33. Shimamura, A. P., Janowsky, J. S. & Squire, L. R. in Frontal Lobe Function andDysfunction (eds. Levin, H., Eisenberg, H. M. & Benton, A. L.) 173–198(Oxford Univ. Press, New York, 1991).

34. Schacter, D. L. Memory, amnesia, and frontal lobe dysfunction. Psychobiology15, 21–36 (1987).

35. Stuss, D. T. & Benson, D. F. The Frontal Lobes (Raven, New York, 1986).36. Fuster, J. M. The Prefrontal Cortex (Raven, New York, 1989).37. Goldman-Rakic, P. S. in The Handbook of Physiology: Section 1. The Nervous

System: Vol. V Higher functions of the brain: (eds. Plum, F. & Mountcastle, V.)Pt. 1, 373–417 (American Physiological Society, Bethesda, Maryland, 1987).

review

38. Smith, E. E. & Jonides, J. Working memory: a view from neuroimaging. Cogn.Psychol. 33, 5–42 (1997).

39. Buckner, R. L. & Tulving, E. in Handbook of Neuropsychology (eds. Boller, F. &Grafman, J.) 439–466 (Elsevier, Amsterdam, 1995).

40. Riege, W. H., Metter, E. J. & Hanson, W. R. Verbal and nonverbal recognitionmemory in aphasic and nonaphasic stroke patients. Brain Lang. 10, 60–70(1980).

41. Whitehouse, P. J. Imagery and verbal encoding in left and right hemispheredamaged patients. Brain Lang. 14, 315–332 (1981).

42. Corkin, S. Lasting consequences of bilateral medial temporal lobe lobectomy:Clinical course and experimental finding in H.M. Semin. Neurol. 4, 249–259(1984).

43. Hyman, B. T., Van Horsen, G. W., Damasio, A. R. & Barnes, C. L. Alzheimer’sdisease: cell-specific pathology isolates the hippocampal formation. Science225, 1168–1170 (1984).

44. Samuel, W., Masliah, E., Hill, L. R., Butters, N. & Terry, R. Hippocampalconnectivity and Alzheimer’s dementia: effects of synapse loss and tanglefrequency in a two-component model. Neurology 44, 2081–2088 (1994).

45. Price, J. L., Davis, P. B., Morris, J. C. & White, D. L. The distribution oftangles, plaques and related immunohistochemical markers in healthy agingand Alzheimer’s disease. Neurobiol. Aging 12, 295–312 (1991).

46. Zola-Morgan, S. & Squire, L. R. Neuroanatomy of memory. Annu. Rev.Neurosci. 16, 547–563 (1993).

47. Murray, E. A. What have ablation studies told us about neural substrates ofstimulus memory? Semin. Neurosci. 8, 13–22 (1996).

48. Schacter, D. L. & Wagner, A. D. Medial temporal lobe activations in fMRI andPET studies of episodic encoding and retrieval. Hippocampus 9, 7–24 (1999).

49. Buckner, R. L. & Koutstaal, W. Functional neuroimaging studies of encoding,priming, and explicit memory retrieval. Proc. Natl. Acad. Sci. USA 95,891–898 (1998).

50. Moscovitch, M. Memory and working-with-memory: A component processmodel based on modules and central systems. J. Cogn. Neurosci. 4, 257–267(1992).

51. Cohen, N. J. et al. Hippocampal system and declarative (relational) memory:Summarizing from functional neuroimaging studies. Hippocampus 9, 83–98(1999).

52. Martin, A., Wiggs, C. L. & Weisberg, J. Modulation of human medialtemporal lobe activity by form, meaning, and experience. Hippocampus 7,587–593 (1997).

53. Stern, C. E. et al. The hippocampal formation participates in novel pictureencoding: evidence from functional magnetic resonance imaging. Proc. Natl.Acad. Sci. USA 93, 8660–8665 (1996).

54. Tulving, E., Markowitsch, H. J., Craik, F. I. M., Habib, R. & Houle, S. Noveltyand familiarity activations in PET studies of memory encoding and retrieval.Cereb. Cortex 6, 71–79 (1996).

55. Gabrieli, J. D. E., Poldrack, R. A. & Desmond, D. E. The role of left prefrontalcortex in language and memory. Proc. Natl. Acad. Sci. USA 95, 906–913 (1998).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


P2X receptors comprise a family of ATP-gated ion channels thatare structurally distinct from the nicotinic and glutamate super-families1,2. They are widely distributed in the peripheral and cen-tral nervous systems and mediate fast excitatory transmission atnerve–muscle and nerve–nerve synapses2–6. They have also beenimplicated in presynaptic modulation of transmitter release7–9.Seven P2X cDNAs have been cloned1,2. The main properties ofneuronal P2X receptors are reproduced fairly well by heterolo-gously expressed cloned receptors; the finding that combinations ofsubunits are sometimes required indicates that the channels canform either as homomers or heteromers1,2,10. Brief application ofATP or related agonists leads to the opening of an ion channelwithin a few milliseconds, and that channel is selectively perme-able to small cations. Both heterologously expressed P2X1–P2X4and native receptor channels in neurons and smooth muscle arepermeable to sodium and potassium, but have very limited per-meability to large cations such as N-methyl-D-glucamine(NMDG+)3,11–13. The receptor channels also show a wide rangeof permeabilities to calcium11–15. Thus, P2X receptors seem toresemble their counterparts in the nicotinic and glutamate super-families fairly closely in kinetics of activation and permeability.

However, one member of the P2X receptor family differs inseveral ways from the others. The P2X7 receptor is found not inneurons, but in cells of the immune system16–18. All P2X recep-tors have intracellular amino (N) and carboxyl (C) termini andtwo membrane-spanning domains separated by a large extracel-lular domain, but the P2X7 intracellular C-terminus is much longerthan the others. The P2X7 receptor is cytolytic; activation by ATPinduces cell swelling, membrane disruption and lysis within sec-onds to minutes16,18. In this and other respects, the P2X7 receptorresembles the P2Z receptor of macrophages and mast cells16–19.Truncated receptors lacking the long C-terminus do not inducecytolysis, but still function as ligand-gated cation channels16. Fur-thermore, when ATP is applied to cells expressing the P2X7 recep-

Pore dilation of neuronal P2Xreceptor channels

C. Virginio2,3, A. MacKenzie1,2, F. A. Rassendren2,4, R. A. North1,2 and A. Surprenant1,2

1 Institute of Molecular Physiology, University of Sheffield, Sheffield S10 2TN, England2 Geneva Biomedical Research Institute, GlaxoWellcome, Geneva Switzerland3 Present address: GlaxoWellcome SPA, Via A. Fleming, 2-37100 Verona, Italy4 Present address: Institut de Genetique Humaine, CNRS UPR 1142, 34396 Montpellier, Cedex 5, France

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

P2X receptors are ligand-gated ion channels activated by the binding of extracellular adenosine 5′-triphosphate (ATP). Brief (< 1 s) applications of ATP to nodose ganglion neurons or to cellstransfected with P2X2 or P2X4 receptor cDNAs induce the opening of a channel selectivelypermeable to small cations within milliseconds. We now show that, during longer ATP application(10–60 s), the channel also becomes permeable to much larger cations such as N-methyl-D-glucamine and the propidium analog YO-PRO-1. This effect is enhanced in P2X2 receptors carryingpoint mutations in the second transmembrane segment. Progressive dilation of the ion-conductingpathway during prolonged activation reveals a mechanism by which ionotropic receptors may alterneuronal function.

tor, cytolysis is preceded by the development of high permeabilityto large cations such as NMDG+ and to organic dyes such as ethid-ium, propidium and YO-PRO-1(refs. 16,18–20). This suggests thatprogressive dilation of the channel ion-conducting pathway leadsdirectly to cell lysis.

Struck by the high variability in cation permeabilities measuredpreviously, we wondered whether such variability might beexplained by time-dependent increases in permeability to largecations similar to that observed at the P2X7 receptor. We foundthat, during sustained (40 s) ATP application, both native neu-ronal and heterologously expressed P2X2 and P2X4 receptors devel-op an increased permeability to large cations that can be as great asthat observed at the cytolytic P2X7 receptor. This high permeabil-ity state was rapidly reversible and did not lead to cell lysis. BecauseP2X4 receptors in particular are localized to presynaptic trans-mitter-release sites in the brain9, it is possible that these receptorsmay be involved in synaptic modulation by transferring high-mol-ecular-weight molecules into or out of the nerve terminal.

RESULTSCation permeability increases with sustained activationWe recorded currents from HEK cells expressing the P2X2 recep-tor during a 40-s application of ATP in bi-ionic NMDG+

o/Na+i

solution (Fig. 1a). Within milliseconds, ATP activated an out-ward current that quickly declined, becoming inward within 5 s.Current–voltage curves constructed from ramp voltage com-mands delivered at 2–4 s intervals revealed a shift in reversalpotential from approximately –70 mV at 2 s to a steady-statevalue of about –25 mV after 30 s (range –42 to –12 mV; Fig. 1a,Table 1). These results suggest large increases in permeability toNMDG+ during sustained receptor activation (Table 1).

The positive shift in reversal potential was not due to loss ofseal or whole-cell recording configuration, as current and mem-brane conductance returned to control levels on agonist washout

articles© 1999 Nature America Inc. • http://neurosci.nature.com

© 1

999

Nat

ure

Am

eric

a In

c. •

htt

p:/

/neu

rosc

i.nat

ure

.co

m


(Fig. 1a). Moreover, repeated 40-s applications separated by 10-s intervals yielded similar current–voltage curves for up to 5 min(longest time examined). To rule out other possible artifacts dueto changes in local ion concentrations during sustained receptoractivation, we did two control experiments. First, cells were heldat depolarized potentials (–20 and 0 mV) so that little or noinward current flowed during receptor activation, and ramps of10–30 mV range were applied close to the expected reversal poten-tial to minimize inward ion flux. Second, cells were held at 20 mV,and a ramp command was given only at 2 s and at 30 s. These pro-tocols did not significantly change reversal potential. Using sero-tonin as agonist, additional control experiments were done inHEK cells transfected with the mouse 5-HT3 receptor, a ligand-gated ion channel that is a cation-selective member of the nico-tinic receptor family. At this receptor, outward current with areversal potential of –88 mV gradually declined during the sus-tained application of 5-HT (presumably because of receptordesensitization21), but no inward current was observed duringthe entire 40 s of agonist application (Fig. 1b, Table 1).

The results in Fig. 1a were obtained in 40%(range 22–65%;134 cells from 13 separate transfections) of HEK cells expressingP2X2 receptors; in the remaining cells (n = 84), reversal poten-tial (–76 ± 3 mV) shifted by less than 5 mV during the 40-s acti-vation period.

Experiments on HEK cells stably expressing the P2X2 receptorgave results similar to those obtainedwith transient expressers. The pro-portion of cells stably expressing P2X2that showed increased cation perme-ability varied little over six monthsbut was significantly lower (15–22%)than that of cells from transient-expression experiments. There wereno consistent differences between thetwo cell populations either in mor-phology or in receptor expression asmeasured by current density (pA perpF) in response to maximal ATP con-centration (100 µM).

We next did the same set ofexperiments on HEK cells tran-

siently transfected with a splice variant of the P2X2 receptor(P2X2-short or P2X2(b)

22,23) and the P2X4 receptor, both ofwhich are widely expressed throughout the central nervous sys-tem9,22–24. Qualitatively similar results were obtained (Table 1).There were no significant differences between cells expressingthe two forms of the P2X2 receptor in either steady-state rever-sal potential or proportion of cells exhibiting the permeabilityincrease. Both the P2X4 and P2X2 receptors underwent signifi-cant dilation; however P2X4 showed significantly smaller dila-tion than P2X2 receptors (p < 0.02). Neither cell swelling norcell damage was associated with increased cation permeabilityduring sustained receptor activation of the wild-type P2X2 recep-tor. Results refer only to those cells in which receptor activationresulted in a time-dependent increase in cation permeability.

Pore dilation is progressiveIf the increase in pore size involves only a single transition, thenthe rate of increase should be independent of the size of the per-meant cation. Reversal potentials were measured for the smallercations DMA and TRIS, as well as for NMDG+ (see Methods fordimensions) as a function of time. For all cations, changes in rever-sal potential over time were well fit by single exponentials, butrates were faster for smaller ions (τNMDG+ = 6.3 ± 0.8 s, τTRIS =4.6 ± 0.4 s, τDMA = 2.5 ± 0.1 s at 100 µM ATP, n = 5, Fig. 2a). Thisimplies that the dilation is a progressive event, with the pore

articles

Fig. 1. Sustained P2X2-receptor activationcauses rapid increase in permeability to NMDG+.(a). Current recording in bi-ionic NMDG+/Na+

solution during a 40-s application of ATP inHEK293 cell expressing P2X2 (wild-type) recep-tor at a –60 mV holding potential. ATP-activatedcurrent is initially outward but rapidly becomesinward. Current–voltage curve for another cell inwhich the same protocol was followed but duringwhich 1-s voltage ramps were delivered at 2–4 sintervals during ATP application. Reversal poten-tial shifts from –72 mV at 2 s to a steady level of–15 mV at about 20 s. (b) Control experimentusing identical protocols as in a during recordingfrom HEK293 cells expressing the 5-HT3 recep-tor. The 5-HT-induced outward current slowlydecreases due to receptor desensitization butreversal potential does not shift from the initialvalue of –83 mV.

a

b

Table 1. Mean reversal potentials and permeability ratios at 2 and 30 s after agonistapplication for nodose neurons and HEK cells expressing P2X or 5-HT3 receptors.

2 s in ATP 30 s in ATP

Erev (mV) PNMDG+ Erev (mV) PNMDG+ Number of cellsPNa+ PNa+

P2X2 (wild-type) –71 ± 0.8 0.066 –31 ± 4 0.29 48P2X2-short –68 ± 2 0.068 –28 ± 7 0.33 12P2X, nodose neurons –78 ± 1.2 0.046 –23 ± 3 0.4 9P2X2N333A –69 ± 1 0.065 –12 ± 2 0.62 23P2X4 –71 ± 0.8 0.06 –46 ± 5 0.16 255-HT3 –85 ± 0.7 0.034 –86 ± 0.7 0.034 22

Solutions were bi-ionic NMDG+o/Na+

i. Agonist was 100-µM ATP or 30-µM 5-HT.


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

(Fig. 3). Concentration–response curves obtained by measuringpeak amplitude of the current evoked by a brief (2–5 s) applica-tion of ATP to two cell types in normal solution were comparedto concentration–response curves for NMDG+ permeability shiftsfor the P2X2 and P2X4 receptor. Half-maximal ATP concentra-tions (EC50) for P2X2- and P2X4-mediated currents in physio-logical saline were 19 ± 3 µM and 24 ± 2 µM, respectively (n = 5;n = 4). In both P2X2-expressing HEK cells and nodose neurons,2-min applications of 1–3 µM ATP, concentrations that produced< 30% of maximum current in normal solution (data not shown)did not affect permeability to NMDG+ (Fig. 3a). Threshold ATPconcentration for induction of a permeability increase was 10–20µM. The time constant (τ) for channel transition to the NMDG+-permeable state and thus, the rate constant into that state, wasestimated by exponential fitting to graphs such as those in Fig.3a. The rate constant (τ–1) increases steeply as a function of ATPconcentration (Fig. 3b).

Measurements of agonist-off rates were made using the bimorphagonist-delivery system, which allows solution changes within 4ms. In cells still attached to coverslips in NMDG+, time constants(τoff) of current recovery after 2 s and 40 s applications of maximalATP concentration (100 µM) were 1.8 ± 0.3 s and 2.2 ± 0.4 s,respectively (n = 8). Measurements in cells removed from cover-


becoming permeable first to DMA+, then to TRIS+ and, last, toNMDG+. Within this limited range of cation sizes, the conductingpore seems to dilate by approximately one angstrom per second.

The channel remains cation selectiveTo determine whether the change in reversal potential may beexplained by a shift from cation to anion selectivity, the experi-ments were repeated in a low-chloride solution. The externalsolution was changed from the standard 154 mM NMDG+ and117 mM Cl– to 34.5 mM NMDG+ and 23.4 mM Cl– (a fivefoldreduction), shifting reversal potential from –121 ± 5 mV at 2 sto –60 ± 3 mV at 40 s (n = 4, Fig. 2b). Theoretical values expect-ed if the pore remained cation selective are –124 mV and –58 mVat t of 0 and 30 s, respectively (using PNMDG+/PNa+ obtained fromthe standard NMDG+ solution and the Goldman-Hodgkin-Katzequation). In low-Cl– solution, the time constant of the shift inreversal potential (11 ± 0.6 s; Fig. 2b) did not differ from the con-trol value. These results make it unlikely that significant anionpermeation occurs during the cation permeability increase.

Rate of pore dilation is concentration dependentSteady-state PNMDG

+/PNa+ ratio and the rate of cation perme-

ability change increased as a function of agonist concentration

articles

Fig. 2. Kinetic properties and cation selectivity for P2X receptorsexpressed heterologously in H2K cells (black symbols) and native recep-tors in nodose neurons (white symbols). (a) Time course of reversalpotential change in bi-ionic solutions; data points are fit to a single expo-nential. Points were obtained from a single cell (HEK or nodose neuron)first in NMDG+, then TRIS and then DMA. (b) Reversal potential as afunction of time in standard NMDG+/Cl– solution (circles) and afterchanging to a low NMDG+/Cl– solution; points in low NMDG+/Cl– arefitted to the exponential function expected if change was solely due toan increase in cation permeability, n = 4 for each point.

Time in ATP (s)

Time in ATP (s)

Rev

ersa

l po

ten

tial

(m

V)

Rev

ersa

l po

ten

tial

(m

V)

a

b

Fig. 3. Permeability increase is concentration dependent. (a) Pointswere obtained from single cells (HEK, black symbols; nodose neuron,white symbols) in bi-ionic NMDG+

o/Na+i. Before reversal potential mea-

surements, concentration–response curves were obtained first in normalphysiological saline during brief, 2-s agonist applications. In both cases,100 µM ATP was maximal, 10 µM ATP produced 35–45% maximumresponse and 3 µM ATP produced 20–23% maximum response. (b) Summary of all experiments as illustrated in (a); rate constant ofreversal potential change (as 1/τ) is plotted as a function of ATP concen-tration on a log–log scale. Each point is mean ± s.e. of 9–12 experiments.

a

bTime in ATP (s)

ATP concentration (µM)

Rev

ersa

l po

ten

tial

(m

V)

Rat

e co

nst

ant

(s-1

)


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


slips in normal physiological saline placed directlyin the flow path yielded τoffs of 100 ms for 2 s appli-cations (R.J. Evans & R.A.N., unpublished obser-vations) and 140 ± 22 ms (n = 5) after 40 s. Similarvalues have been reported for native P2X receptorsin bullfrog dorsal root ganglion neurons25. Theseresults indicate that the channel closes at the samerate from its initial open state (at 2 s) and from itsdilated, NMDG+-permeable state (at 40 s).

Dilation may not involve second messenger pathwaysWe did a series of experiments to address the possibility that thedilation was due to activation of a second messenger pathway;namely, concomitant activation of metabotropic P2Y receptors,which are present in HEK cells26. Suramin (10 µM), which blocksP2Y-mediated increases in intracellular calcium in HEK cells(data not shown) but does not inhibit P2X4 receptors27, did notalter the increase in NMDG+ permeability at the P2X4 receptor (n= 6). Neither the proportion of P2X2-expressing cells thatincreased NMDG+ permeability nor magnitude of the increasewere changed significantly after incubation with BAPTA-AM (n= 32) or when recordings were made with pipettes containing 20mM BAPTA (n = 32) or 10 mM GTP-γ-S (n = 9). Similarly, nei-ther forskolin (20 µM, n = 4) nor phorbol 12,13-dibutyrate (20

µM, n = 6) significantly altered NMDG+ permeability increases.Rate of permeability increase, but not the steady-state value, wassignificantly decreased in P2X2-expressing cells by the additionof calcium (2 mM) to the extracellular NMDG+ solution (n = 5).Time constants for permeability increase induced by ATP (100µM) were 24 ± 1.6 s and 7.2 ± 0.5 s in NMDG+ with and withoutadded calcium, respectively. The slower kinetics are most likelyexplained by allosteric inhibition of the P2X2 receptor by externalcalcium3,12,13.

Pore dilation measured by YO-PRO-1 uptakeIt is possible that the observed dilation was an artifact of sodiumremoval, and would not occur under physiological conditions.We therefore used the uptake of a fluorescent dye, YO-PRO-1, tomeasure the increase in pore size18 (Fig. 4). Additionally, use ofYO-PRO-1 allows estimation of permeability to a cation larger

articles

Fig. 5. Kinetics of YO-PRO1 uptake at cloned and neuronal P2X recep-tors. (a) Rate of YO-PRO-1-uptake during ATP application in HEK cellsexpressing P2X2 receptor (circles), P2X4 receptor (squares) and nodoseneurons (triangles). Each point is the average from eight HEK cells in thefield of view and one nodose neuron; standard error bars are not shownfor HEK cells for clarity but are < 12% for all points (b) HEK cellsexpressing the P2X2N333A receptor show a tenfold higher rate of YO-PRO-1 uptake; points are averages of 11 cells in field of view, s.e. < 7%for each point; data for P2X2 wild type are the same as in (a). (c) Summary of steady-state YO-PRO-1 fluorescence measured at 60 sin the presence of 100 µM ATP in physiological solution and NMDG+ asindicated. Note the break and change in scale required for P2X2N333A.All data were obtained using a 40× objective.

a

Time (s)

Rat

e o

f Y

OP

RO

up

take

(d

/dt)

Rat

e o

f Y

OP

RO

up

take

(d

/dt)

b

c

YO

PR

O f

luo

resc

ence

at

60 s

(arb

itra

ry u

nit

s)

Time (s)

Fig. 4. HEK cells expressing P2X receptors take upYO-PRO-1 during 60-s application of ATP. Digital pho-tomicrographs under transmitted light of cellsexpressing 5-HT3, P2X4, P2X2 and P2X2N333A areshown at left of each series; fluorescent images showcells before (0 s) and during agonist application. YO-PRO-1 (2 µM) was present 10 min before and duringagonist application; experiments were done in externalNMDG+. Images obtained using 100× objective. Scalebar (arbitrary units), 50 to 200 for P2X2, P2X4 and 5-HT3; 200 to 800 for P2X2N333A.

0 s 10 s 20 s 60 s


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


than NMDG+ (see Methods). Rate of change in fluorescence wasused to indicate the amount of dye entering the cell per unit time;this reached a steady-state after 30–50 s (Fig. 5a and b). The rateof YO-PRO-1 uptake in normal physiological saline with maximalagonist concentration did not differ significantly between recep-tors; 10–90% rise times were 36 ± 4 s and 39 ± 5 s at P2X2 andP2X4 receptors, respectively (n = 36; n = 11). The rise times inNMDG+ solution were 19 ± 3 s and 21 ± 4 s respectively (Fig.5a). Absolute steady-state (at 60 s) YO-PRO-1 fluorescence wassignificantly higher at P2X2 receptors than at P2X4 receptors inboth normal and NMDG+ solution (Fig. 5c). Proportions of cellsthat took up YO-PRO-1 upon receptor activation were similarfor cells transiently expressing P2X2 or P2X4 (20–55%, 12 –36%,respectively) and for cells stably expressing P2X2 (7–45%). Theseresults imply that pore dilation occurs in normal physiologicalsolution as well as in the absence of extracellular sodium.

Mutations in TM2 of P2X2 alter pore dilationIf increased permeability in response to agonist exposure result-ed from a dilation of the ion-conducting pathway, then substi-tutions of amino acids contributing to the pore might be expectedto affect such changes. On the other hand, if increased NMDG+

permeability arises from alteration of physically separate regionsof the protein, such substitutions might be without effect. Cys-teine-scanning mutagenesis has been used to localize several sitesin the second transmembrane domain of the P2X2 receptor towithin the ion conducting pathway28,29. We examined the effectsof point mutations at four of these residues: N333, T336, L338and G342. The following substitutions were made at each site:A, G, N, D, E, K, S and Q. Proteins carrying alanine at any oneof the four positions yielded receptors that showed large shiftsin permeability and YO-PRO-1 uptake in all cells examined; thesedifferences were examined in detail for N333A (n = 69). Perme-ability to NMDG+ increased tenfold during 30 s of receptor acti-vation (Table 1). However, the time constant of the reversalpotential change and its concentration dependence did not differfrom the wild-type receptor (Fig. 3b). No electrophysiological orvisual indications of cell damage were observed during repeated40-s applications of ATP for up to 10 min (n = 10). Steady-stateYO-PRO-1 fluorescence and the rate of YO-PRO-1 uptake were

dramatically increased; rise time in NMDG+ solution was 10 ± 1 s(n > 100), and steady-state YO-PRO-1 fluorescence was 7–10-fold greater than for the wild-type P2X2 receptor (Figs. 4 and 5band c).

Sensory neuron P2X receptors also dilateThe stoichiometry and subunit composition of native P2X recep-tors is not yet known. It is possible that the pore dilation observedin heterologously expressed receptors results from ‘unnatural’homomeric channel formation not found in neurons. We there-fore considered it essential to determine whether similar pore dila-tion was observed in native receptors. Primary cultures of ratnodose neurons all express P2X receptors and respond to ATP withlarge, non-desensitizing inward currents2,25,30,31. All nodose neu-rons also possess cation-selective 5-HT3 receptors, at which 5-HTevokes currents similar in amplitude to those mediated by ATP(data not shown). We measured NMDG+, TRIS and DMA rever-sal potential as well as YO-PRO-1 uptake during activation of P2Xor 5-HT3 receptors in these neurons; results were essentially thesame as those observed with heterologously expressed P2X2 recep-tors (Fig. 6). Agonist application shifted the NMDG+ reversalpotential from –78 mV to –23 mV with a time constant and ATP-concentration dependence that were not significantly differentfrom those of heterologously expressed P2X2 receptors (Table 1,Figs. 2 and 6a), whereas the 5-HT3-mediated NMDG+ reversalpotential remained constant at –83 ± 0.9 mV (n = 14, Fig. 6a).Time constants for permeability increase in response to 100 µMATP were τNMDG

+ = 7.3 ± 0.5 s, τTRIS = 3.8 ± 0.6 s and τDMA = 2.1± 0.1 s, (n = 4, Fig. 2a). The ATP-mediated increase in permeabil-ity was observed in 38% of neurons examined (23 of 60 cells).Reversal potentials shifted by < 7 mV in the remaining cells; YO-PRO-1 uptake was observed in a similar proportion of cells.Repeated applications of the maximal concentration of ATP (100µM; 40–60 s duration) for up to 10 min in either solution pro-duced no biophysical or visual indications of cell damage.

DISCUSSIONA progressive shift in reversal potential during prolonged ago-nist applications can arise from ion accumulation, and this hassometimes been misinterpreted as an altered permeability. It is

articles

Fig. 6. Sustained P2X receptor activation in cul-tured nodose neurons results in increasedNMDG+ permeability as well as YO-PRO-1uptake. (a) Current–voltage curves were obtainedfrom experiment similar to that illustrated in Fig.1a from a single neuron to which ATP (100 µM)was applied for 40 s; 60 s after ATP washout, 5-HT(30 µM) was applied for 60 s. (b) Digital photomi-crograph of nodose neuron under transmittedlight shown at left and fluorescence image before(0 s), then 20 and 60 s after ATP application inNMDG+ solution.

a P2X in nodose neuronTime in ATP (s)

5-HT3 in nodose neuronTime in 5-HT (S)

b


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


often difficult to exclude ion accumulation in sequestered or dif-fusionally limited domains, but several observations indicate thatthis was not a contributing factor our experiments. First, the samechange in reversal potential was observed over a wide range ofholding potentials and when inward current flow was prevent-ed. Second, currents through 5HT3 receptors were of similar ini-tial amplitude, but showed no change in reversal potential. Boththese experiments conclusively indicate that the change in rever-sal potential signifies a change in permeability. Third, although weroutinely used calcium-free solutions to facilitate calculation ofpermeability of the monovalent cation from the reversal poten-tial, it is unlikely that absence of calcium from the extracellularsolution was causative. The steady-state change in reversal poten-tial was essentially the same when the external solution contained2 mM calcium. Fourth, and most compelling, the experimentswith YO-PRO-1 indicate that a relatively large cation (376 dal-tons) readily enters the cell in an otherwise normal extracellularsolution and without the disturbance of a patch-clamp electrode.

Several questions arise regarding the possible mechanism ofthe permeability increase. Does it result from a dilation of thesame ion-conducting pathway that opens within milliseconds ofATP application, or does a second, independent pathway openmore slowly? The large effects of several point mutations at posi-tions previously shown to reside within the ion-conducting poreof the initially NMDG+-impermeable channel27,28 are consistentwith the simple model of pore dilation but, of course, do notprove it. Testing the accessibility of cysteine residues tomethanethiosulphonates by looking for inhibition of current inan external NMDG+ solution might be a useful approach. Whenshort applications of ATP are used, the initial rate of rise of thewhole-cell current is steeply dependent on the ATP; the 10-90%rise time is a few milliseconds at 100 µM ATP and the steady-state dose–response curve saturates at this concentration25,32.This is consistent with models in which the closed–open transi-tion is very fast compared to the binding step. However, in a sim-ple linear model, the rate at which such an open poreprogressively dilates would not be expected to depend on the ATPconcentration, whereas we found that the rate of dilation con-tinued to increase up to 400 µM ATP. This suggests that at leasttwo parallel states are required to adequately model the results.Initial channel opening must result from ATP binding with anassociation rate constant (kon) on the order of 106 M–1 s–1, where-as the kon for NMDG+ conductance is about 1000 times slower.The examined point mutations seem to favor opening to the sec-ond rather than the first state. One possible interpretation is thatthe first state corresponds to a mono- or bi-liganded channel(Hill coefficients are typically two or more), and that the secondstate requires binding of ATP to additional subunits.

It is not obvious why only a proportion of cells examinedundergo the permeability increase. This was true for channelsexpressed in HEK cells as well as native channels in nodose gan-glion neurons. Although channel rundown prevented study ofsingle-channel currents in outside-out patches during sustainedATP application, permeability increases did occur when ATP wasincluded in the recording pipette in the cell-attached configura-tion (unpublished observations). It may be that additional mol-ecules are required for the dilation. We considered the possibilitythat a distinct subset of receptors (P2Y, for example) is expressedby all cells we studied, and that the slow permeability increaseresults from a modification such as phosphorylation mediatedthrough a distinct pathway. Such a mechanism would not pro-vide an obvious explanation for the cell-to-cell variability. Fur-thermore, we believe that it is unlikely because of the relatively

high ATP concentrations required, because dilation was notaltered by intracellular BAPTA or GTP-γ-S or extracellularforskolin and phorbol esters and because concentrations ofsuramin that block P2Y receptors had no effect on the perme-ability increase at P2X4 receptors.

The results must be interpreted with respect to the behavior ofthe P2X7 receptor, where similar dilation is also associated withpermeabilization and, eventually, cell lysis16,18–20. Application ofagonist similar to that used in the present experiments (40 s atmaximal concentration) invariably lyse cells expressing P2X7receptors within 60–90 s, but no cell death is ever found withP2X2 receptors or nodose ganglion neurons, even with exposuresof up to 10 min. On the other hand, the rate of development ofNMDG+ permeability is approximately the same for both recep-tors. This implies that P2X2, P2X4 and P2X7 receptors all incor-porate a pore that dilates with continued agonist application, butuniquely for the P2X7 receptor does this progress to cell lysis. TheC-terminus of the P2X7 receptor is required for cell lysis16 and,although YO-PRO1 uptake is reduced by over 90% in the C-ter-minally truncated P2X7 receptor, it is not abolished18; in fact, it iswithin the range observed in the present study for the P2X4 recep-tor. Taken together with the present mutagenesis results, this mayindicate that the C-terminal domain is not directly involved withthe initial dilation of the ion channel at any of the P2X receptors.

Ligand-gated ion channels open following the binding of anagonist to their extracellular domain. Although multiple closedand open states can often be distinguished on the basis of kinet-ic measurements, it is generally assumed that all the open stateshave the same permeation properties. The present results indi-cate that this is not the case at P2X receptors, which undergo aprogressive increase in permeability following agonist binding.Given the extensive distribution of these receptors in nervous tis-sue (particularly P2X4), this could have profound consequencesfor neuronal signaling. P2X4 receptors are found both pre- andpostsynaptically in the mammalian brain9,24, and a progressiveincrease in calcium permeability with repeated or prolongedreceptor activation could mediate synaptic plasticity. Further-more, cell death in the nervous system may result in release ofATP; subsequent prolonged activation of high-permeability P2Xreceptors might enable the transfer of large molecules across thecell membranes of neighboring cells.

METHODSCells. HEK293 cells transiently transfected with cDNA encoding the fol-lowing receptors were used for the majority of heterologous expressionexperiments: rat P2X2-long (wild type33), rat P2X2-short22, rat P2X4

27

and mouse 5-HT334. The ‘short form’ of the P2X2 receptor bearing a 69-

amino-acid deletion in the C-terminus22 is known as P2X2∆370–438 orP2X2(b)

23. Methods of transfection using Lipofectin and co-transfectionwith eGFP to identify receptor-expressing cells have been described28.All recordings and YO-PRO-1 fluorescence measurements were carriedout 16–48 h after transfection. Experiments were also conducted onHEK293 cells stably expressing the P2X2 wild-type receptor, for whichproperties have been described13; this receptor did not carry the C-ter-minal epitope tag (see below). Dissociation and culture protocols for ratnodose neurons were as described12,30 All neuronal recordings were car-ried out 4–12 days after dissociation.

Plasmids and mutagenesis. Construction and mutagenesis of P2X2 cDNA(provided by D. Julius, UCSF) carrying a C-terminus epitope has beendescribed28 The antigenic tag (EYMPME) was used for protein detec-tion by immunohistochemistry as described28. The mouse 5-HT3 recep-tor carried a hexahistidine epitope at the C-terminus constructed asdescribed34 and kindly provided by Dr. R. Hovius, Swiss Federal Insti-tute of Technology, Lausanne. This epitope-tagged receptor shows much

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


less desensitization during sustained activation than the wild-type recep-tor (A.S., unpublished data) and was, therefore, a more appropriate con-trol for our experiments.

Electrophysiology. Standard whole-cell recordings were obtained usingthe EPC-9 patch-clamp system (HEKA, Lambrecht Germany). Patchpipettes (4–7 MΩ) were filled with 160 mM NaCl, 10 mM HEPES and 11EGTA. Normal extracellular solution contained 147 mM NaCl, 2 mMKCl, 1 mM MgCl2, 2 mM CaCl2, 10 mM HEPES and 12 mM glucose.After whole-cell configuration was obtained in normal solution, the solu-tion was changed to 154 mM test cation, 10 mM HEPES and 12 mM glu-cose to measure permeability to monovalent cations. To determine anionselectivity, reversal potentials were first measured in 154 mM NMDG, 117mM chloride and then in solution containing 34.5 mM NMDG and 23.4mM chloride with glucose increased to 246 mM to maintain osmolarity.

ATP or 5-hydroxytryptamine (5-HT) was applied with a fast-flowU-tube delivery system35 unless otherwise stated. Rapid kinetics wereexamined using a piezoelectric bimorph36 (Vernitron, Bedford, Ohio)delivery system. Time to agonist equilibration as measured by 10–90%rise time of the junction potential of an open pipette tip was 1–3 ms.Liquid junction potentials were compensated by use of a salt bridge(3 M KCl) as described37. Unless otherwise stated, membrane poten-tial was held at –60 mV. Currents in the presence of agonist were notcorrected for linear leakage because currents in the absence of agonistwere < 0.5% of maximum agonist-induced currents. Current–voltagerelations were obtained by voltage ramps (0.5–1 s duration) from –90to 0 mV unless otherwise stated. Permeability ratios (PX/PNa) were cal-culated as described in detail12,13, using PX/PNa = ([Nai]exp(FV/RT))/[Xo]. Dimensions of cations were estimated using vander Waals radii in energy-minimized models constructed with DesktopMolecular Modeller (Oxford Univ. Press). These were dimethylamine(DMA) 6.1× 4.2× 3.8 Å, tris(hydroxymethyl)aminomethane (TRIS),7.5×6.5×6.4 Å; N-methyl-D-glucamine (NMDG+), 10.8×8.7×6.5 Å;YO-PRO-1, 16.8×12.8×8.2 Å. Values stated in text and tables are mean± s.e. from individual cells, whereas graphs were drawn by averagingresults from all experiments and fitting a single curve to the pooleddata. All experiments were done at room temperature.

YO-PRO-1 fluorescence. Fluorescence was measured using a ZeissAxiovert 100 and oil-immersion Fluar 40× or Plan-Neofluar 100× objec-tive and the Photonics monochromator Imaging (TLLionVISION) sys-tem (Photonics, Planegg, Germany). YO-PRO-1 (Molecular Probes,Eugene, Oregon) fluorescence was measured from single cells in the fieldof view (usually 10–40 cells with 40× objective) with excitation and emis-sion wavelengths of 491and 509 nm, respectively. YO-PRO-1 (1–5 µM)was present in all solutions before and during agonist application. Imageswere captured at 0.2–2 Hz. YO-PRO-1 fluorescence from individual cellswere averaged to obtain mean response. Because YO-PRO-1 binding isa cumulative process, fluorescence traces were differentiated to obtainthe rate of YO-PRO-1 uptake into cells. Data are mean ± s.e.

ACKNOWLEDGEMENTSWe are grateful to Daniele Estoppey and Denis Fahmi and Alison Newbolt for

assistance with cell culture, transfections and mutagenesis.

RECEIVED 19 NOVEMBER 1998, ACCEPTED 8 JANUARY 1999

1. North, R. A. & Barnard, E. A. Nucleotide receptors. Curr. Opin. Neurobiol. 7,346–357 (1997).

2. North, R. A. P2X purinoceptor plethora. Semin. Neurosci. 8, 187–194 (1996).3. Evans, R. J. & Surprenant, A. P2X receptors in autonomic and sensory

neurons. Semin. Neurosci. 8, 217–223 (1996).4. Evans, R. J. & Surprenant, A. Vasoconstriction of guinea-pig submucosal

arterioles following sympathetic nerve stimulation is mediated by the releaseof ATP. Br. J. Pharmacol. 106, 242–249 (1992).

5. Evans, R. J., Derkach, V. & Surprenant, A. ATP mediates fast synaptictransmission in mammalian neurones. Nature 357, 503–505 (1992).

6. Edwards, F. A., Gibb, A. J. & Colquhoun, D. ATP receptor-mediated synapticcurrents in the central nervous system. Nature 359, 144–147 (1992).

7. Gu, J. G. & MacDermott, A. B. Activation of ATP P2X receptors elicitsglutamate release from sensory neuron synapses. Nature 389, 749–753 (1997).

8. Khakh, B. S. & Henderson, G. ATP receptor-mediated enhancement of fastexcitatory neurotransmitter release in the brain. Mol. Pharmacol. 54, 372–378(1998).

9. Le, K. T. et al. Sensory presynaptic and widespread somatodendriticimmunolocalization of central ionotropic P2X ATP receptors. Neuroscience83, 177–190 (1998).

10. Lewis, C. et al. Coexpression of P2X2 and P2X3 receptor subunits can accountfor ATP-gated currents in sensory neurons. Nature 377, 432–435 (1995).

11. Benham, C. D. & Tsien R. W. A novel receptor-operated Ca2+-permeablechannel activated by ATP in smooth muscle. Nature 328, 275–278 (1987).

12. Virginio, C., North, R. A. & Surprenant, A. Calcium permeability and block athomomeric and heteromeric P2X2 and P2X3 receptors, and P2X receptors inrat nodose neurones. J. Physiol. (Lond.) 510, 27–35 (1998).

13. Evans, R. J. et al. Ionic permeability and divalent cation effects on two ATP-gated cation channels (P2X receptors) expressed in heterologous cells. J.Physiol. (Lond.) 497, 413–422 (1996).

14. Valera, S. et al. A new class of ligand-gated ion channel defined by P2Xreceptor for extracellular ATP. Nature 371, 516–519 (1994).

15. Garcia-Guzman, M. et al. Characterization of recombinant human P2X4receptor reveals pharmacological differences to the rat homologue. Mol.Pharmacol. 51, 109–118 (1997).

16. Surprenant, A., Rassendren, F., Kawashima, E., North, R. A. & Buell, G. Thecytolytic P2Z receptor for extracellular ATP identified as a P2X receptor(P2X7). Science 272, 735–738 (1996).

17. Collo, G. et al. Tissue distribution of the P2X7 receptor. Neuropharmacology 361277–1284 (1997).

18. Rassendren, F. et al. The permeabilizing ATP receptor (P2X7): cloning andexpression of a human cDNA. J. Biol. Chem. 272, 5482–5486 (1997).

19. Di Virgilio, F. The P2Z purinoceptor: an intriguing role in immunity,inflammation and cell death. Immunol. Today 16, 524–528 (1995).

20. Humphreys, B. D., Virginio, C., Surprenant, A., Rice, J. & Dubyak, G. R.Isoquinolines as antagonists of the P2X7 nucleotide receptor: high sensitivityfor the human versus rat receptor homologues. Mol. Pharmacol. 54, 22–32(1998).

21. Gill, C. H., Peters, J. A. & Lambert, J. J. An electrophysiological investigation ofthe properties of a murine recombinant 5-HT3 receptor stably expressed inHEK293 cells. Br. J. Pharmacol. 114, 1211–1221 (1995).

22. Brandle, U. et al. Desensitization of the P2X(2) receptor controlled byalternative splicing. FEBS Lett. 404, 294–298 (1997).

23. Simon, J. et al. Localization and functional expression of splice variants of theP2X2 receptor. Mol. Pharmacol. 52, 237–248 (1997).

24. Collo, G. et al. Cloning of P2X5 and P2X6 receptors and the distribution andproperties of an extended family of ATP-gated ion channels. J. Neurosci. 16,2495–2507 (1996).

25. Li, C., Peoples, R. W. & Weight, F. F. Ethanol-induced inhibition of a neuronalP2X purinoceptor by an allosteric mechanism. Br. J. Pharmacol. 123, 1–3(1998).

26. Harden, T. K. et al. in P2 Nucleotide Receptors (eds. Turner, J. T., Weisman, G.A. & Fedan, J. S.) 109–134 (Humana, Totowa, New Jersey, 1998).

27. Buell, G., Lewis, C., Collo, G., North, R. A. & Surprenant, A. An antagonist-insensitive P2X receptor expressed in epithelia and brain. EMBO J. 15, 55–62(1996).

28. Rassendren, F., Buell, G., Newbolt, A., North, R. A. & Surprenant A.Identification of amino acid residues contributing to the pore of a P2Xreceptor. EMBO J. 16, 3446–3454 (1997).

29. Egan, T. M., Haines, W. R. & Voigt, M. M. A domain contributing to the ionchannel of ATP-gated P2X2 receptors identified by the substituted cysteineaccessibility method. J. Neurosci. 18, 2350–2359 (1998).

30. Khakh, B. S., Humphrey, P. P. A. & Surprenant, A. Electrophysiologicalproperties of P2X-purinoceptors in rat superior cervical, nodose and guinea-pig coeliac neurones. J. Physiol. (Lond.) 484, 385–395 (1995).

31. Li, C., Peoples, R. W. & Weight, F. F. Mg2+ inhibition of ATP-activated currentin rat nodose ganglion neurons: evidence that Mg2+ decreases the agonistaffinity of the receptor. J. Neurophysiol. 77, 3391–3395 (1997).

32. Evans, R. J. et al. Pharmacological characterization of heterologouslyexpressed ATP-gated cation channels (P2X purinoceptors). Mol. Pharmacol.48, 178–183 (1995).

33. Brake, A. J., Wagenbach, M. J. & Julius, D. New structural motif for ligand-gated ion channels defined by an ionotropic ATP receptor. Nature 371,519–523 (1994).

34. Tiairi, A. et al. Ligand binding to the serotonin 5-HT3 receptor studied with anovel fluorescent ligand. Biochemistry (in press).

35. Fenwick, E. M., Marty, A. & Neher, E. A patch-clamp study of bovinechromaffin cells and of their sensitivity to acetylcholine. J. Physiol. (Lond.)331, 577–597 (1982).

36. Clements, J. D. Rapid solution exchange using a piezoelectric translator.Axobits 14, 9–11 (1993).

37. Neher, E. in Methods in Enzymology, Ion Channels vol. 207, 123–130 (1992).

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Purinergic excitatory synapses use ATP as a fast synaptic transmitterto activate P2X-receptor cation channels1–7. In addition to mediat-ing postsynaptic responses, P2X receptors also presynaptically mod-ulate fast excitatory transmission8,9. ATP is released to theextracellular space during high-frequency neuronal stimulation10,11,cellular depolarization12 or ischemia13,14; therefore, it is important toidentify the membrane receptors activated by extracellular ATP. Genecloning approaches show that P2X receptors are a new multigenefamily of ATP-gated cation channels7,15,16. Known P2X receptor-channel subunits are widely expressed in brain17–20 on the soma,dendrites and terminals of neurons21, indicating that P2X receptorchannels and purinergic transmission may be important for neu-ronal physiology. P2X4 channel subunits are especially widespread inthe brain and form functional channels that can be either homo-meric22–24 or heteromeric with P2X6 (ref. 25). However, our under-standing of the roles of distinct P2X channel subunits in purinergictransmission and their physiological importance is still primitive.

When the P2X7 channel was cloned from macrophages, it wasfound that heterologously expressed P2X7 channels, but not otherP2X receptors, change their ion selectivity when exposed to ATP26, 27.However, we found that neuronal P2X4, P2X2 and heteromericP2X2/P2X3 channels have two distinct forms of opening; here wedescribe a second mode that involves an apparent increase in porediameter and depends on previous exposure to ATP. Thus, changesin pore diameter seem to be a general property of this ion-channelfamily. Also, we find that mutations of a residue thought to lie with-in the pore result in altered patterns of changes in ion selectivity,identifying the pore itself as the site of action of such permeabilitychanges. The ability of P2X receptor channels to change their ionicselectivity may have profound consequences for encoding at centraland peripheral purinergic synapses.

RESULTSHistory-dependent increases in P2X4 channel currentsWe applied ATP at submillimolar concentrations to activate P2X4receptor channels expressed in Xenopus oocytes, using a valve-

Neuronal P2X transmitter-gatedcation channels change their ionselectivity in seconds

Baljit S. Khakh, Xiaoyan R. Bao, Cesar Labarca and Henry A. Lester

Division of Biology 156-29, California Institute of Technology, Pasadena, California 91125, USA

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

Fast synaptic transmission depends on the selective ionic permeability of transmitter-gated ionchannels. Here we show changes in the ion selectivity of neuronal P2X transmitter-gated cationchannels as a function of time (on the order of seconds) and previous ATP exposure. Heterologouslyexpressed P2X2, P2X2/P2X3 and P2X4 channels as well as native neuronal P2X channels possess various combinations of mono- or biphasic responses and permeability changes, measured byNMDG+ and fluorescent dye. Furthermore, in P2X4 receptors, this ability to alter ion selectivity canbe increased or decreased by altering an amino-acid residue thought to line the ion permeationpathway, identifying a region that governs this activity-dependent change.

operated rapid application system that allows for solution changeswithin less than a second (Fig. 1). The EC50 for ATP was 3 µM(Hill slope, 1.3). Suramin (30 µM), an ATP-receptor antagonist,did not block P2X4-channel currents (0.03 ± 0.1% change using10 µM ATP; n = 4), whereas the same concentration blocked P2X2-channel currents evoked by 10 µM ATP (91 ± 2% block; n = 4).Ten-second application of ATP produced inward currents thatdesensitized to 54 ± 1% of peak current (peak inward current,2.5 ± 0.4 µA; n = 4). After washout, the ATP-evoked currentsreturned to baseline levels. These pharmacological and kinetic dataare similar to published findings22–25. Rise time and washout forP2X4 currents evoked by brief ATP application are the same as forneuronal P2X2 receptor-channels27, and approximate the kinet-ics that we observed in the perfusion system (Fig. 1a).

For P2X4 channels during longer (up to 5 min) applicationsof ATP, the initial current (I1) was followed by a slower current(I2) that started to develop in seconds and peaked after 2–4 min-utes with an amplitude 150–300% of that of I1 (Figs. 1 and 2).During washout of ATP, I2 deactivated more slowly than the solu-tion change. In contrast, P2X2 receptor-channels show no I2 forsimilar durations of ATP application, but P2X2 channels do showdesensitization by about 5–10% of the peak28 (Fig. 1). We alsoapplied ATP to cells expressing P2X3 channels29,30, producingdesensitization of ATP-evoked inward currents by ~ 99% within10 seconds; no current resembling I2 was detected even after ATPapplications in excess of 5 min (Fig. 1d, inset). We next deter-mined if the I2 at P2X4 receptors could be evoked by repetitivebrief applications of ATP (1 s exposure to 10 or 100 µM ATPevery 10 s for 1 min). During a single one-second pulse of ATP,the evoked current reached a peak and desensitized (Fig. 1e);upon washout, the current returned nearly to baseline levels with-in a few seconds. However, during repetitive applications of ATP,I2 developed in the same cell. In contrast, no I2 was observed dur-ing repetitive applications of ATP to P2X2-expressing cells(Fig. 1f). Furthermore, in P2X4-expressing cells, I2 developedonly with continuous or repetitive application of ATP and always


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


followed I1. Thus ATP evokes a biphasic response at brain P2X4channels; the second current, I2, depends on previous exposure toATP and is much larger than the first.

We determined current–voltage relationships at the peak ofI1 and I2 (Fig. 2b). The reversal potentials of I1 and I2 were iden-tical in solutions with Na+ as the permeant cation (I1,–4.6 ± 0.9 mV; I2, –5.6 ± 2 mV; n = 6). Similar currents at themacrophage P2X7 receptor are blocked by extracellular calciumions27; therefore we examined the effects of calcium on P2X4 cur-rents (Fig. 2c–e). I2 developed more slowly and to a lower peaklevel in 0.1 mM calcium than in calcium-free solution, and wascompletely blocked by 1–5 mM calcium. Calcium affected nei-ther peak I1 nor its rise time at these concentrations (Fig. 2). Thisindicates that the occurrence and kinetics of I2 are governed bycalcium concentrations, a regulatory mechanism that may alsoexist in neurons. It is noteworthy that extracellular calcium con-centration can fall to values as low as 0.08 mM during maintainedneuronal activity or anoxia32–37. Under the same conditions, ATPis also released into the extracellular space, presumably at mil-limolar local concentrations10,11,13,14. This suggests that I2 maydevelop in neurons when two coincident signals occur: namely, afall in calcium and repetitive or sustained ATP release into theextracellular compartment.

We tested whether I1 and I2 currents also occur for P2X4channels expressed in human embryonic kidney cells (HEK 293;Fig. 2f and g)38. There were two phenotypes in nominally calci-um-free solutions. First, in 5 of 11 cells, we observed both I1 andI2 currents similar to those observed in oocytes. Second, in 6 of 11cells, we observed a monophasic response that desensitized witha t1/2 of 46 s. In 2 mM extracellular calcium, only monophasiccurrents were recorded from HEK 293 cells transfected with P2X4(n = 5). In these cells, I1 desensitized markedly faster than in cal-cium-free solutions, with a t1/2 of 6 s. Furthermore, onlymonophasic currents were recorded in calcium-free solutions

from HEK 293 cells expressing P2X2 channels; these responsesare similar to those recorded in oocytes27,28 (Fig. 2h).

Ion permeability increases for P2X4 channels in secondsThe development of I2 could be explained by an increase in P2X4channel pore diameter that allowed enhanced sodium entry dur-ing sustained or repetitive exposure to ATP27. We therefore mea-sured changes in reversal potentials26 when we substitutedpermeant extracellular ions (Na+) with NMDG+, which perme-ates poorly through most ligand-gated cation channels because itis just larger than the narrowest part of the channel pores38. Thusany increase in pore diameter could be detected in real time bytracking the permeability of NMDG+ through P2X4 channels.When ATP was applied in NMDG+ solutions, an initial outwardcurrent developed with a time course that resembled that of I1in normal recording solutions (Fig. 3). Because the holdingpotential, –60 mV, is more positive than the P2X4 channel rever-sal potential in NMDG+ (–66 mV for the cell in Fig. 3a), the cur-rent mostly consists of outward movement of intracellularpotassium through P2X4 channel pores. During longer ATP puls-es, with NMDG+ as the main extracellular cation, an inward cur-rent developed within 30 seconds with kinetics that resemble thedevelopment of I2 in sodium solutions. Thus I2 reflects enhancedNMDG+ permeability with respect to I1. The reversal potentialshifted by +18.6 ± 1.6 mV from –61.8 ± 3.8 mV to –44.2 ± 4.0mV during ATP application in NMDG+ solution (Fig. 3; n = 8),representing a 2.1 ± 0.13-fold increase in PNMDG+/PNa+, from0.097 ± 0.014 to 0.20 ± 0.028. Upon removal of ATP (after briefor long pulses), the channels returned to the non-conductingclosed state (Figs. 1 and 3).

We next tested whether I2 could develop with repetitive briefATP applications. Whereas a single 1-second ATP pulse evokedonly an outward current, repetitive ATP application to the samecell evoked a current that changed from outward to inward with-

articles

Fig. 1. Activity-dependent increase inbrain P2X4, but not P2X2 or P2X3 chan-nel currents. (a) ATP-evoked current(10 s pulse at 100 µM) recorded from aXenopus oocyte expressing homomericP2X4 channels. (b) The same experimentwas repeated with P2X2-expressingoocytes; here the ATP-evoked currentdisplays no apparent desensitization. (c) Following a pulse of ATP (100 µMapplied for ∼ 5 min), the resultant mem-brane current clearly shows two peaks.The first (I1) develops rapidly and thendesensitizes; the second peak (I2) devel-ops slowly over several minutes andreturns to baseline levels after washoutof ATP. (d) The experiment was repeatedfor P2X2-expressing oocytes; a pulse ofATP evoked only a single peak, with littledesensitization (5.7 ± 2.2%; n = 4). Theinset shows a representative trace froman experiment where 10 µM ATP wasapplied to a P2X3-expressing cell (scale,0.5 µA, 25 s; n = 5); in such cells the cur-rent desensitized by 99.1 ± 0.6% within10 s. (e) ATP was applied repetitively toP2X4-expressing cells. The current evoked by a single pulse of ATP (100 µM; 1 s) rises rapidly, reaches a peak and returns nearly to baselinelevels. If applied repetitively (6 pulses every 10 s), then each ATP pulse evoked a fast current; a sustained current developed following 2–6pulses. (f) ATP evoked only fast currents in P2X2-expressing cells following single or repetitive pulses.

a

b

c

d

e

f

Brief ATP pulses Long ATP pulses Repetitive ATP pulses


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


in two ATP pulses, shifting its reversal potential from–59.5 ± 2.0 mV at first pulse to –31.6 ± 2.2 mV at sixthpulse (Fig. 3c). This reflects a 3.02 ± 0.25-fold increase inPNMDG+/PNa+ (0.096 ± 0.007 to 0.29 ± 0.03; n = 4). Thesedata show that the development of I2 at P2X4 channels(Fig. 1) is due to a history-dependent increase in ion per-meability on the second time scale; furthermore, I2 occurswhether I1 flows inward or outward.

Although P2X4 channels are Ca2+ permeable24, and Xenopusoocytes express Ca2+-activated chloride currents, it is unlikelythat chloride channels have a role in either I1 or I2, for these rea-sons. Most experiments were done in Ca2+-free solutions, andthe reversal potentials for both I2 and I1 are close to 0 mV, where-as the reversal potential for a Ca2+-activated Cl– current is around–18 mV under our recording conditions. No slow currentsresembling I2 were observed for P2X2, which is also Ca2+ per-meable38. Extracellular calcium inhibits I2, which develops evenin NMDG+ solutions, without net inward current flow. I2 alsodoes not occur with the G347Y mutant, even though currents atthis channel are large and, presumably, would also allow calci-um entry (see below). I2 occurs at G347R and G34K mutants eventhough these channels have a much reduced, or absent I1 (seebelow). Indeed, all of the data can be explained only by an intrin-sic ability of P2X4 to detect the history of ATP exposure and torespond with increased permeability to large cations.

Other neuronal channels: P2X, 5-HT3 and nicotinicWe also studied P2X2, P2X3 and heteromeric P2X2/P2X3 chan-nels that are found natively in the central and peripheral nervoussystems2,4,7,17–19,28–31, and P2X7 channels that mediate cell lysis inmacrophages27. To compare ion channel classes, we studied 5-HT3serotonin and α4β2 nicotinic channels using methods identical tothose described for P2X4 channels (see Fig. 3). When activated forup to five minutes with agonist, 5-HT3 and nicotinic channelsshowed no detectable changes in ion selectivity (Table 1). In con-

trast, P2X2, P2X2/P2X3 and P2X7 channels showed an increase inNMDG+ permeability when they were activated with agonist, eventhough P2X2 channels showed only monophasic currents in sodi-um solutions (for comparison we show data for P2X4 in Table 1).We could not make reliable measurements of shifts in NMDG+

reversal potential for P2X3 channels because they desensitizedalmost completely by the time the P2X4, P2X2 and P2X2/P2X3channels began to display the second phase29–31 (Fig. 1).

We next used uptake of the fluorescent dye YO-PRO-1 as anindicator of pore dimension in oocytes expressing various chan-nels and exposed to ATP27 (Fig. 4). YO-PRO-1 is a propidium dyethat increases its fluorescence when it binds nucleic acids.Increased fluorescence of extracellularly applied YO-PRO-1 in thepresence of agonist indicates that YO-PRO-1 has permeated intothe cell. On the whole, these data support our electrophysiologi-cal studies and show that P2X4, P2X2 and P2X7 channels allowsignificant YO-PRO-1 entry, whereas P2X3 and 5-HT3 channelsdo not. However, one difference is also apparent: P2X2/P2X3 het-eromeric channels show clear increases in NMDG+ permeabili-ty, but only show weak fluorescence for YO-PRO-1. Thus, trackingthe reversal potential of a slightly permeant ion (NMDG+) is amore sensitive measure of changes in ion selectivity than is testingfor uptake of the larger YO-PRO-1 molecule.

Overall, each neuronal P2X receptor shows a unique combi-nation of history-dependent decreases in ion selectivity. P2X2/P2X3heteromeric channels show only the increase in NMDG+ perme-ability, P2X2 shows both increases in NMDG+ permeability andYO-PRO-1 uptake, and P2X4 (like the non-neuronal P2X7) shows

articles

Fig. 2. Properties of I1 and I2. P2X channels were expressed inoocytes (a–e) and HEK 293 cells (f–h). (a–g) P2X4; (h) P2X2.(a) I1 and I2 have the same overall dependence on ATP concen-tration, with EC50s of 3.9 and 1 µM, respectively (Hill slopes, 1.7and 1.9; n = 60). (b) Agonist-induced currents during membranepotential ramps (–120 to +60 mV), showing that I1 and I2 (thicktrace) have similar reversal potentials and that neither rectifystrongly. (c–e) Representative traces from oocytes exposed to10 µM ATP in varying extracellular [Ca2+]. (c) Nominally Ca2+-free recording solution; I1 peak amplitude –1009 ± 194 nA;40–80% rise time, 365 ± 82 ms; I2 peak amplitude, –1720 ± 620nA; rise time,160 ± 33 s (n = 5). (d) 0.1 mM Ca2+; I1 (peakamplitude, –690 ± 114 nA; rise time, 570 ± 118 ms) desensi-tized; subsequently current grew slowly to peak I2 (amplitude,–583 ± 70 nA ; rise time, 253 ± 32 s; n = 5). (e) In 5 mM Ca2+, I1peak amplitude was –1496 ± 145 nA with a rise time of350 ± 73 ms (n = 3); I2 was absent. (f) Representative tracesfrom a HEK 293 cell exposed to 100 µM ATP in 2 mM Ca2+ andin nominally Ca2+-free solutions. Traces are scaled to give thesame peak current. ATP-evoked currents desensitized muchfaster in solution containing Ca2+ (t1/2 is 5.9 ± 1.4 s from330 ± 86 pA in 2 mM Ca2+, whereas t1/2 is 45.5 ± 15.4 s from607 ± 295 pA peak current in nominally Ca2+-free solutions,n = 5). (g) Representative recording of ATP-evoked I1 and I2peaks from HEK 293 cells in Ca2+-free solution. The netincrease in current was 6.3 ± 2.9-fold from 227 ± 81 pA at I1 to675 ± 119 pA at I2 (n = 5); washout has been truncated for clar-ity. (h) Representative recording from a HEK 293 cell express-ing P2X2 channels.

a b

Cu

rren

t (µ

A)

[ATP] µM

Cu

rren

t (µ

A)

Vm (mV)I2

I1

c d e0 Ca2+ 0.1 mM Ca2+ 5 mM Ca2+

f g h

I2

I1

P2X4 P2X4 P2X2


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

increases in NMDG+ permeability, YO-PRO-1 uptake and themacroscopic sodium current (I2). We do not know whether thisspectrum of effects represents qualitative differences in the natureof history-dependent open channels or quantitative differences inthe percent of time spent in various permeability states.

Native P2X channels in neuronsWe next sought to determine whether native neuronal P2X chan-nels also show changes in ion selectivity. We studied neurons fromthe superior cervical ganglion (SCG) because SCG neurons con-tain transcripts for P2X2 and P2X4 channels7,17–20,22,28, theyexpress functional P2X and 5-HT3 channels7, and P2X4 channelswere first cloned from an SCG-cDNA library22. As previously

shown7, we found that, in extracellular Na+ solutions, 100 µMATP evoked monophasic inward currents of –297 ± 63 pA with areversal potential of 0.5 ± 3.6 mV (seven of eight neurons; oneneuron, no response), and that 100 µM serotonin evoked inwardcurrents of –116 ± 90 pA with a reversal potential of 6.4 ± 7.0 mV(three of three neurons). We next repeated these experiments withNMDG+ replacing all the extracellular Na+, and in the nominalabsence of Ca2+. In such conditions, the ATP-evoked currentreversal potential changed from –83 ± 9 to –13 ± 4 mV over ~ twominutes (four of ten neurons); often the whole-cell configurationwas lost in solutions that contained no Na+ and reduced Ca2+ (sixof ten neurons). As an alternative, we therefore used YO-PRO-1uptake into SCG neurons as a measure of pore formation (Fig. 5).

Fig. 3. Activity-dependent increasesin cation permeability of P2X4 chan-nels. (a) Record of the holding cur-rent at –60 mV from a cell bathed inrecording solution in which Na+ wasreplaced by NMDG+; ATP (10 or100 µM) was applied for the periodindicated by the solid bar. Each pointgives the holding current from amembrane potential ramp (–120 to+60 mV). (b) Current–voltage rela-tionships at the time points indi-cated. ATP evoked an initial outwardcurrent; reversal potential shiftedwithin 0.5 min by +5 mV, causingcurrent to become inward, andfinally by +27 mV. (c) Repeated ATPpulses (10 or 100 µM, 1 s) wereapplied for the periods indicated byshort bars. The first pulse of ATPevoked an initial outward current,whereas after six pulses (once every10 s) the evoked current was inwardand reversal potential had shifted by+26.5 mV (d).

a

b

c

d

Time (min)

Cu

rren

t (µ

A)

Cu

rren

t (n

A)

Time (min)

98 mM NMDG+98 mM NMDG+

a, b c

Cu

rren

t (n

A)

Fig. 4. Activation of neuronal P2X2 andP2X4 channels evokes YO-PRO-1 uptake.(a–c) False-color images; black indicateszero and white indicates maximum fluo-rescence. (a) Oocyte expressing P2X7channels and exposed to YO-PRO-1 butno ATP (background fluorescence). (b) P2X7-expressing oocyte exposed toATP and to YO-PRO-1. (c) Oocyteexposed to the cell-permeabilizing agent,saponin. (d) Summary of experimentsshown in (a) through (c) above. The fluo-rescence produced by saponin could notbe quantified with confidence because itsaturated the CCD. P2X channels wereactivated with 300–1000 µM ATP in thepresence of YO-PRO-1 (10 µM) for 10min before images were acquired. TheP2X2/P2X3 heteromeric channel wasactivated with 100 µM αβ-methyleneATP; the 5-HT3 channel with 1 mM sero-tonin. Controls from separate experi-ments have been pooled; n = 4–12 cells.

a b c

d

Fluorescence (arbitrary units)

+

+

+

+

+

+

+

–

+

+

–

+

+

+

+

+

serotonin

+

–

+


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

In 11 of 15 neurons, we found that ATP (300 µM; 2–10 min) couldevoke YO-PRO-1 entry into SCG neurons. Some neuronsresponded rapidly (< 120 s) with an increase in YO-PRO-1 fluo-rescence (for example, neurons 1 and 2 in Fig. 5), whereas othersresponded much more slowly (> 240 s, for example, neuron 6)when similarly challenged with ATP (Fig. 5g). The pooled datafrom a number of such experiments (Fig. 5h) shows that ATP canevoke marked YO-PRO-1 entry within five to ten minutes of ATPexposure; no such increase in YO-PRO-1 fluorescence wasobserved with serotonin (n = 4). We also noted that the mor-phology of SCG neurons changed dramatically after exposure toATP for 30 min (compare Fig. 5a with 5f), resulting in a granu-lar appearance similar to that of dead cells.

Mutants change the ratio of I2 to I1P2X receptors define a structural family of ligand-gated cationchannels that have little sequence similarity with other ion chan-nels7. The P2X subunit has two transmembrane regions (TM),intracellular N- and C-termini and a large extracellularloop22,39,40 (Fig. 6, inset). We chose to mutate G347 of P2X4because it is a conserved residue and substituted cysteine acces-sibility mutagenesis41,42 shows that this glycine is covalentlymodified by agents applied either outside or inside the cell, indi-cating that it is at, or near, the narrowest part of the pore, anarea referred to as the channel gate42. Glycine is also known toallow for local protein flexibility. A priori, we reasoned thatmutation of the presumed channel gate would provide useful

Table 1. P2X receptor channels show history-dependent increases in permeability to NMDG+, whereas serotonin andnicotinic channels do not.

Initial After 2–5 min of agonist

Transmitter-gated cation Na+ reversal NMDG+ reversal PNMDG+ Change NMDG+ Fold increasechannel potential (mV) potential (mV) PNa+ reversal potential (mV) PNMDG+

PNa+

P2X4 –1.7 ± 1.3 –51.8 ± 1.0 0.1 ± 0.006 +28.1 ± 0.9 * 3.1 ± 0.1P2X2 +5.5 ± 3.1 –57.0 ± 2.6 0.1 ± 0.01 +44.3 ± 3.8 * 6.0 ± 0.9P2X2/P2X3 heteromer +8.0 ± 2.0 –59.6 ± 1.4 0.1 ± 0.05 +29.8 ± 1.0 * 3.4 ± 0.7P2X3 +3.5 ± 2.1 –46.0 ± 1.2 0.2 ± 0.008 —— ——P2X7 –8.2 ± 1.8 –55.5 ± 1.1 0.1 ± 0.05 +47.0 ± 3.2 * 6.7 ± 0.85–HT3 serotonin +6.9 ± 0.6 –57.0 ± 4.6 0.1 ± 0.02 < 5 < 0.1α4β2 nicotinic –4.4 ± 2.8 –93.7 ± 4.6 0.02 ± 0.005 < 5 < 0.1

Results for oocyte expression. Shifts in NMDG+ reversal potential and fold changes in PNMDG+/PNa+ were measured 2–5 min after applying agonist in NMDG+

solutions. In the case of P2X3 channels a shift in reversal potential could not be measured with confidence because the channels desensitized within 1 s; insuch cases, we can not provide an absolute value for the change in NMDG+ permeability, although this is negligible. *Significant (p < 0.05; Student’s paired t-test) shifts in reversal potential and increases in PNMDG+/PNa+. For all data sets, n = 4–8. 5-HT3 and nicotinic channels were activated with 10 µM serotoninand acetylcholine, respectively. The P2X2/P2X3 heteromeric channel currents did not desensitize markedly over 5 min during activation with 10 µM αβ-methylene ATP. Other P2X channels were activated with 10 or 100 µM ATP.

Fig. 5. ATP-evoked YO-PRO-1 uptake into neurons. (a) Bright-field image of SCG neu-rons in the presence of YO-PRO-1 (10 µM). (b) Fluorescent image of (a) showing onlyweak YO-PRO-1 signal before the addition of 300 µM ATP. Fluorescence images 5 minafter (c), 10 min after (d) and 30 min after (e) addition of ATP. (f) Bright-field imagetaken after 30 min exposure to ATP; note granular appearance of cells. (g) Time courseof fluorescence increase in neurons labeled 1 through 6 in panels (a) through (f). Notethat neurons 1 and 2 respond with an increase in YO-PRO-1 fluorescence within 5 min,whereas neurons 5 and 6 respond only after a greater delay (8–18 min). Background fluo-rescence was measured from areas indicated as 7, 8 and 9. Circles show the average timecourse of the YO-PRO-1 signal from neurons 1–6; triangles show average background flu-orescence. (h) Mean YO-PRO-1 signal from SCG neurons after various treatments fromthree separate experiments. Fluorescence reaches significance above pooled backgroundlevels 5–10 min after ATP exposure27. Pooled data for neurons that showed no increasein YO-PRO-1 fluorescence in response to ATP are also shown (n = 4).

a b

c d

e f

g

Flu

ore

scen

ce a

bo

veb

ackg

rou

nd

(ar

bit

rary

un

its)

Time after ATP addition (min)

h

Flu

ore

scen

ce a

bo

ve

bac

kgro

un

d (

arb

itra

ry u

nit

s)


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


insight as to the molecular determinants of the observed changein ion permeability of P2X4 channels.

We mutated G347 to 13 other residues and tested for func-tional responses to pulses of ATP (10 or 100 µM, ~ 5 min dura-tion; Fig. 6). Of these mutants, six showed I1 and I2 responsewaveforms that were identical to those for the wild-type channel,and four did not express currents; deletion of G347 resulted inno expression. However, three mutants displayed clearly differ-ent response waveforms (Fig. 7b–d). Replacing glycine at posi-tion 347 by a lysine (G347K) or arginine (G347R) producedATP-evoked currents with little I1, whereas I2 remained (Fig. 7band c). We also observed a mutantthat produced the opposite effect: atyrosine (G347Y) resulted in ATP-evoked currents that completelylacked I2; instead the current desen-sitized within about 20 s to a steady-state level of 2.2 ± 0.8% of the I1 peakcurrent (Fig. 7d). Absolute values ofpeak I1 and I2 as well as differences inamplitude between I1 and I2 for wildtype and mutants, expressed as foldincrease, are given to the right of eachtrace. In wild-type channels, I2 is2–3-fold larger than I1, whereas inG347R and G347K mutants, I2 is10–30-fold larger; in G347Y mutantchannels, I2 is undetectable. Impor-tantly, both G347Y and G347Rmutant channels have the same EC50as wild-type channels for ATP toevoke I1 or, if present, I2 currents(Fig. 8), suggesting that overall pro-tein structure is unaltered.

We next tested for the ability ofNMDG+ to permeate G347Y andG347R mutant channels. G347Rmutant channels showed no outwardcurrent in NMDG+ (Fig. 8b); instead,an inward current developed immedi-ately with a reversal potential(–35.3 ± 1.0 mV; PNMDG+/PNa+ = 0.25± 0.01) that was identical to the inwardcurrent during the I2 phase of the wild-

type channel and did not detectably shift duringfive minutes of ATP application (< 3 mV shift inreversal potential; n = 7; Fig. 8b). The same exper-iment for G347Y mutant channels showed an ini-tial outward current upon ATP application inNMDG+, but there was no inward current; impor-tantly, the reversal potential of the outward cur-

rent was similar (∼ 6 mV more negative) to that of the outwardcurrent at wild-type P2X4 (–72.4 ± 0.6 mV; PNMDG+/PNa+ is 0.058± 0.002; n = 5), and this was I1 (Fig. 8f). The data showed only lowNMDG+-permeability for the G347Y mutant pore but only highpermeability for the G347K mutant pore.

DISCUSSIONP2X receptors mediate fast purinergic excitatory synaptic trans-mission in the peripheral and central nervous systems1–7; there-fore it is important to identify properties that may be predictiveof their function in synapses. This study shows that some P2X

articles

Fig. 6. Mutant P2X4 channels. The bar graph showspeak ATP-evoked currents from cells that wereinjected with either wild-type or mutant P2X4 chan-nel cRNAs. The inset cartoon (adapted from ref. 16)illustrates the presently understood membrane topol-ogy of P2X channels39,40; glycine 347 is in TM2.Mutants are grouped according to the side-chainproperties. Not all mutants were tested in the sameexperiment (n = 4–10); control data is pooled fromcells expressing wild-type P2X4 channels tested ineach experiment.

100 µM ATP-evoked current (–nA)

Nonpolar

Charged (+)

Neutral

Charged (–)

Fig. 7. Channel-gate mutations increase or decrease I2 amplitude. (a–d) Representative recordings ofATP-evoked (100 µM, ~5 min) currents from wild-type and mutant P2X4-expressing oocytes were madein the same experiment. Mean peak amplitudes ± s.e. of I1 and I2 along with relative change are given tothe right of each trace. (a) Wild type; (b) G347K; (c) G347R; (d) G347Y; note absence of I2.

a wild type b G347K

d G347Yc G347R

I1 = –851 ± 27 nAI2 = –2107 ± 420 nA(2.7 ± 0.8-fold increase)

I1 = –151 ± 26 nAI2 = –3719 ± 504 nA(26 ± 3.2-fold increase)

I1 = –145 ± 45 nAno I2

I1 = –35 ± 6 nAI2 = –354 ± 127 nA(10 ± 2.9-fold increase)


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


receptors are bi-functional: they open to a small pore in a sec-ond or less, and some P2X receptors also desensitize on a sim-ilar time scale. Additionally, during maintained or repetitiveATP applications, they open a larger pore; this is reflected as alarge activation-dependent increase in channel current in sodi-um solutions. Some P2X receptors do not show biphasic cur-rents in sodium solutions (for example, P2X2 and P2X2/P2X3),indicating that for these channels, the underlying channel states(I1 and I2) overlap in time. Overall, this study extends previ-ous findings for the macrophage P2X7 channel27 and makesthree further advances.

First, changes in cation permeability in response to ATPapplication were previously thought to be unique tomacrophage P2X7 channels27. We have shown ATP-evokedchanges in ion selectivity for neuronal P2X channels as well.Thus, P2X receptor channels represent the first examples ofneuronal transmitter-gated cation channels that can changetheir ion permeability in seconds in response to previous expo-sure to transmitter. Second, earlier studies specify a crucial rolefor the C-terminus in the ability of P2X7 channels to form largepores27. P2X2, P2X3 and P2X4 channels have much shorter C-termini (24–120 residues, versus 238 residues in P2X7); thus along tail seems unnecessary for bi-functional operation of P2Xchannels. Our mutagenesis results show that bi-functionalityof P2X4 channels is controlled at least in part by the secondtransmembrane domain in the channel pore. Third, P2X7receptor-channels have a limited distribution in neurons43. Incontrast P2X2, P2X2/P2X3 and P2X4 channels are found inpain sensors29–31, peripheral sympathetic and parasympathet-

ic as well as central neurons17–21. Native neu-ronal P2X channels are implicated in thetransduction of pain29–31, mediate ATP fastsynaptic transmission in the peripheral,enteric and central nervous systems2–7 andpresynaptically modulate fast glutamatergictransmission in the central nervous system8,9.Composition of native neuronal P2X chan-nels is presently undetermined, but someclearly comprise P2X2, P2X3, P2X4 or a com-bination of these2–9,17–25,28–31.

Mechanistic insightsThe channel pore of P2X receptors was originally thought to beformed by a re-entrant loop preceding TM2 (refs. 28,44), butsubsequent studies suggest that TM2 itself lines the pore41,42.Here, we have shown that it is possible to dramatically changethe permeability of P2X4 channels to NMDG+ by mutating aconserved glycine at position 347. This glycine is of particularinterest because it is conserved in all channels of this type15,41,42,it is thought to form the gate42, and glycine is the smallest of allresidues and provides local flexibility in proteins; this flexibilitymay favor conformational transitions that underlie the changesin pore diameter implied by our data. By altering G347, whichlies at or near the gate, we perturb these protein-conformationchanges. The structural details of such conformation changes,which take place over tens of seconds, will require further study.Interestingly, the pore-lining segment of a large-conductancebacterial mechanosensitive channel has many small amino acids,such as glycine, at its polar face45. Furthermore, mammalianacid-sensing channels are topologically similar to P2X channels;they too show a biphasic current evoked by low pH, and the sec-ond phase is accompanied by a decrease in ion selectivity46.Additionally, Shaker K+ channels undergo C-type inactivationthat is accompanied by a change in ion selectivity47.

Implications for neuronsClassically, the ion permeability of neuronal channels has beenviewed as fixed—constrained by the physical dimensions of thechannel pore. The remarkable selectivity48 of the pore forms thebasis of synaptic excitation and inhibition, as well as plasticity

articles

Fig. 8. Channel-gate mutants of P2X4 have high andlow cation selectivity. (a–c) P2X4 G347R mutants.(a) Concentration–effect curves for ATP at G347R-mutant P2X4 channels. EC50 for I2 is 6 µM (n = 15);inset shows a representative recording for 10 µMATP (scale, 500 nA, 50 s). (b) Holding currents at–60 mV from a cell bathed in a recording solutionwith Na+ replaced by NMDG+; period of ATP appli-cation indicated by bar. (c) Currents during mem-brane potential ramps (–120 to +60 mV over200 ms) at time points 1–7 indicated in (b). InNMDG+, ATP evoked only inward currents with areversal potential of –39 mV. (d–f) P2X4 G347Ymutants. (d) ATP concentration–effect curves. EC50for I1, ~4 µM (n = 15); inset shows representativerecording for 10 µM ATP (scale, 100 nA, 20 s). (e) Holding currents at –60 mV in recording solu-tion in which Na+ was replaced by NMDG+; barindicates period of ATP application. (f) Current dur-ing a membrane potential ramp (–120 to +60 mVover 200 ms) at peak ATP-evoked current;Erev = –66 mV.

a

b

c

d

e

f

G347R G347Y

Cu

rren

t (-

nA

)

Cu

rren

t (-

nA

)

Cu

rren

t (n

A)

Cu

rren

t (n

A)

Cu

rren

t (n

A)

Cu

rren

t (n

A)

Time (min) Time (min)

[ATP] µM[ATP] µM


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


resulting from calcium influx49. Therefore, binary properties ofchannels, such as whether or not certain ions permeate in par-ticular cellular settings, can have profound effects on the encod-ing properties of synapses16,49. Our data demonstrateactivation-history-dependent changes in ion selectivity for neu-ronal P2X receptors, and it is possible that similar responses mayalso occur in other neurons that express P2X recep-tors2–9,17–24,28–31 during coincident ATP release10,11,13,14 anddrops in extracellular calcium32–37, conditions that may favor thistransition. In view of the activation-dependent nature of the ionselectivity change, neuronal P2X channels may function as activ-ity/frequency detectors; for P2X4 channels, one simple conse-quence may be enhanced cellular depolarization accompanyingthe I2 phase of the biphasic current. Interestingly, biphasic ATP-evoked currents have been described in the hippocampus, an areathat is abundant in P2X4, and these currents seem to underlieATP-evoked changes in synaptic plasticity (Yamazaki et al., Soc.Neurosci. Abstr. 24, 425.3, 1998). Nevertheless, the cellular con-sequences of ion-selectivity changes in neuronal P2X channelsare largely unknown, although for macrophage P2X7 channels,these changes cause cell lysis27. It is interesting that neuronsrelease ATP during ischemia13,14, raising the possibility that P2X-mediated permeability changes have a role in pathology as well.

METHODSMolecular biology. Wild-type P2X2, P2X3, P2X7 and P2X4 cDNAscloned into pcDNA3 or pcDNA1 were obtained from GlaxoWellcome(UK) and 5-HT3 cDNA from David Julius. We obtained α4 and β2 nico-tinic cDNAs previously. cDNAs were linearized at unique restriction sitesdownstream of the poly (A) tail. The cDNAs were transcribed in vitrousing the mMESSAGE mMACHINE kit (Ambion, Austin, Texas). EachXenopus laevis oocyte was injected with 5–10 ng of cRNA in a volume of50 nl; preparation and maintenance of oocytes was as described50, andall electrophysiological recordings were made 1–4 d after injection. Site-directed mutagenesis of the P2X4 cDNA clone was done with syntheticoligonucleotides (Quick Change; Stratagene, La Jolla, California). Mutantsand original clones were propagated in Top 10F′ E. Coli, and plasmidswere purified using standard techniques. Bacterial colonies harboringmutant plasmids were screened for the loss of a restriction site producedusing the same mutagenic oligonucleotide, and positive mutant cloneswere confirmed by DNA sequencing.

Electrophysiology. Two-electrode voltage-clamp recording of oocytes wasdone using the Geneclamp 500 amplifier (Axon Instruments, Foster City,California). Micropipette electrodes were pulled from borosilicate glass(Sutter Instruments, Novato California) and back-filled with 3 M KCl toyield resistances of 1–2 MΩ. Recordings were made in solution consisting of98 mM NaCl, 5 mM HEPES and 1 mM MgCl2, pH 7.45, perfused by grav-ity flow over oocytes at a rate of ∼ 3 ml per min (chamber volume, ~ 300µl). Solutions containing ATP were applied to the oocyte using a solenoid-operated solution switcher (General Valve, Fairfield, New Jersey); completeexchange occurred within 0.5–1 s. Voltage control of oocytes was main-tained using a Digidata 200 interface and a personal computer runningpCLAMP 6 or pCLAMP 7 software (Axon Instruments). Data were filteredat 200–500 Hz and digitized at 3–5 times this rate. Voltage ramps (–120 to+60 mV over 100–500 ms) were routinely used because significant desen-sitization of I1 caused an apparent change in the rectification of the cur-rent–voltage relationship if steps were used; data were filtered at 1 kHz anddigitized at 3 kHz. All experiments were done at 20–23°C. Data in the textand graphs are shown as mean ± s.e. from n determinations as indicated.All ramps shown are leak subtracted. Ion permeability ratios were calcu-lated from shifts in reversal potentials using the function PNMDG+/PNA+ =exp(δErevF/RT); where PNMDG+ is the permeability to NMDG+, PNa+ is thepermeability to Na+ and δErev is the shift in reversal potential. We assume thereversal potential in Na+ solutions to be 0 mV (Fig. 2).

Mammalian cells were transfected using Superfect (Qiagen), and allrecordings were made 24–48 hours later using an Axoclamp 2A ampli-

fier (Axon Instruments). Recording solutions used for mammalian cellsare described38. Rat SCG neurons were dissociated and maintained incell culture30 (up to 5 days), and all recordings were made using methodsand solutions as described30. ATP (100 µM) was applied to single HEK293 cells and SCG neurons under voltage clamp (–60 mV) using a U-tube concentration-clamp system.

Imaging. YO-PRO-1 was purchased from Molecular Probes (Eugene,Oregon), dissolved and stored as recommended by the manufacturer.We used YO-PRO-1 at 10 µM as described27. Images were viewed with anOlympus IMT-2 microscope using a 40× objective and fluorescein isoth-iocyanate filter and recorded with a Quantix cooled CCD camera usingAxon Imaging Workbench software (Axon Instruments).

ACKNOWLEDGEMENTSThanks to I. Chessell (GlaxoWellcome, UK) for providing P2X cDNA clones.

The authors are grateful to H. Li, S. McKinney and J. Sydes for assistance with

preparation of oocytes and neurons and for advice on the YO-PRO-1

experiments, and to other members of the group for comments. This work was

supported by the National Institutes of Health (NS-11756), a Wellcome Trust

(UK) Prize Travelling Fellowship to B.S.K. and a Caltech Summer

Undergraduate Research Fellowship to X.R.B.

RECEIVED 30 NOVEMBER 1998, ACCEPTED 20 JANUARY 1999

1. Burnstock, G. Purinergic nerves. Pharmacol. Rev. 24, 509–581, (1972).2. Evans, R. J., Derkach, V. & Surprenant, A. ATP mediates fast synaptic

transmission in mammalian neurons. Nature 357, 503–505 (1992).3. Edwards, F. A., Gibb, A. & Colquhoun, D. ATP receptor-mediated synaptic

currents in the central nervous system. Nature 359, 144–146 (1992).4. Galligan, J. J. & Bertrand, P. P. ATP mediates fast synaptic potentials in enteric

neurons. J. Neurosci. 14, 7563–7571 (1994).5. Bardoni, R., Goldstein, P. A., Lee, J., Gu, J. G. & MacDermott, A. B. ATP P2X

receptors mediate fast synaptic transmission in the dorsal horn of the ratspinal cord. J. Neurosci. 17, 5297–5304 (1997).

6. Nieber, K., Poelchen, W. & Illes, P. Role of ATP in fast excitatory synapticpotentials in locus coeruleus neurons of the rat. Br. J. Pharmacol. 122,423–430 (1997).

7. Surprenant, A., Buell, G. & North, R. A. P2X receptors bring new structure toligand-gated ion channels. Trends Neurosci. 18, 224–229 (1995).

8. Gu, J. G. & MacDermott, A. B. Activation of ATP P2X receptors elicitsglutamate release from sensory neurone synapses. Nature 389, 749–753(1997).

9. Khakh, B. S. & Henderson, G. ATP receptor-mediated enhancement of fastexcitatory neurotransmitter release in the brain. Mol. Pharmacol. 54, 372–378(1998).

10. Wieraszko, A., Goldsmith, G. & Seyfried, T. N. Stimulation-dependentrelease of adenosine triphosphate from hippocampal slices. Brain Res. 485,244–250 (1989).

11. Cunha, R. A., Vizi, E. S., Ribeiro, J. A. & Sebastiao, A. M. Preferential releaseof ATP and its extracellular catabolism as a source of adenosine upon high-but not low-frequency stimulation of rat hippocampal slices. J. Neurochem.67, 2180–2187 (1996).

12. Terrian, D. M., Hernandez, P. G., Rea, M. A. & Peters, R. I. ATP release,adenosine formation, and modulation of dynorphin and glutamic acidrelease by adenosine analogues in rat hippocampal mossy fibersynaptosomes. J. Neurochem. 53, 1390–1399 (1989).

13. Lutz, P. L. & Kabler, S. Release of adenosine and ATP in the brain offreshwater turtle (Trachemys scripta) during long term anoxia. Brain Res.769, 281–286 (1997).

14. Braun, N., Zhu, Y., Krieglstein, J., Culmsee, C. & Zimmermann, H.Upregulation of the enzyme chain hydrolysing extracellular ATP aftertransient forebrain ischaemia in the rat. J. Neurosci. 18, 4891–4900 (1998).

15. North, R. A. Families of ion channels with two hydrophobic segments. Curr.Opin. Cell Biol. 8, 474–483 (1996).

16. Green, T., Heinemann, S. F. & Gusella, J. F. Molecular neurobiology andgenetics: investigation of neural function and dysfunction. Neuron 20,427–444 (1998).

17. Kidd, E. J. et al. Localisation of P2X purinoceptor transcripts in the ratnervous system. Mol. Pharmacol. 48, 569–573 (1995).

18. Vulchanova, L. et al. Differential distribution of 2 ATP-gated ion channels (P-2Xreceptors) determined by immunocytochemistry. Proc. Natl. Acad. Sci. USA 93,8063–8067 (1996).

19. Vulchanova, L. et al. Immunohistochemical study of the P2X2 and P2X3receptor subunits in rat and monkey sensory neurons and their central terminals.Neuropharmacology 36, 1229–1242 (1997).

20. Collo, G. et al. Cloning of P2X5 and P2X6 receptors and the distribution and

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


35. Krnjevic, K., Morris, M. E., Reiffenstein, R. J. & Ropert, N. Depthdistribution and mechanism of changes in extracellular K+ and Ca2+

concentrations in the hippocampus. Can. J. Physiol. Pharmacol. 60,1658–1671 (1982).

36. Pumain, R. & Heinemann, U. Stimulus- and amino acid-induced Ca2+ andpotassium changes in rat neocortex. J. Neurophysiol. 53, 1–16 (1985).

37. Morris, M. E. & Trippenbach, T. Changes in extracellular [K+] and [Ca2+]induced by anoxia in neonatal rabbit medulla. Am. J. Physiol. 264, R761–769(1993).

38. Evans, R. J. et al. Ionic permeability of, and divalent cation effects on, twoATP-gated cation channels (P2X receptors) expressed in mammalian cells. J.Physiol. (Lond.) 497, 413–422 (1996).

39. Torres, G. E., Egan, T. M. & Voigt, M. M. Topological analysis of the ATP-gated ionotropic P2X2 receptor subunit. FEBS Lett. 425, 19–23 (1998).

40. Newbolt, A. et al. Membrane topology of an ATP-gated ion channel (P2Xreceptor). J. Biol. Chem. 273, 15177–15182 (1998).

41. Rassendren, F., Buell, G., Newbolt, A., North, R. A. & Surprenant, A.Identification of amino acid residues contributing to the pore of a P2Xreceptor. EMBO J. 16, 3446–3454 (1997).

42. Egan, T. M., Haines, W. R. & Voigt, M. M. A domain contributing to the ionchannel of ATP-gated P2X2 receptors identified by the substituted cysteineaccessibility method. J. Neurosci. 18, 2350–2359 (1998).

43. Collo, G. et al. Tissue distribution of the P2X7 receptor. Neuropharmacology36, 1277–1283 (1997).

44. MacKinnon, R. Pore loops: an emerging theme in ion channel structure.Neuron 14, 889–892 (1995).

45. Chang, G., Spencer, R. H., Lee, A. T., Barclay, M. T. & Rees, D. C. Structure ofthe MscL homolog from Mycobacterium tuberculosis: a gatedmechanosensitive ion channel. Science 282, 2220–2226 (1998).

46. Lingueglia, E. et al. A modulatory subunit of acid sensing ion channels inbrain and dorsal root ganglia. J. Biol. Chem. 272, 29778–29783 (1997).

47. Starkus, J. G., Kuschel, L., Rayner, M. D. & Heinemann, S. H. Ion conductionthrough C-type inactivated Shaker channels. J. Gen. Physiol. 110, 539–550(1997).

48. Lester, H. A. The permeation pathway of neurotransmitter-gated ionchannels. Annu. Rev. Biophys. Biomol. Struct. 21, 267–292 (1992).

49. Berridge, M. J. Neuronal calcium signaling. Neuron 21, 13–26 (1998).50. Quick, M. W. & Lester, H. A. in Ion Channels of Excitable Cells (ed. Narahashi,

T) 261–279 (Academic, San Diego, 1994).

articles

properties of an extended family of ATP-gated ion channels. J. Neurosci. 16,2495–2507 (1996).

21. Le, K. T. et al. Sensory presynaptic and widespread somatodendriticimmunolocalisation of central ionotropic P2X ATP receptors. Neuroscience 83,177–190 (1998).

22. Buell, G., Lewis, C., Collo, G., North, R. A. & Surprenant, A. An antagonist-insensitive P2X receptor expressed in epithelia and brain. EMBO J. 15, 55–62(1996).

23. Seguela, P., Haghighi, A., Soghomonian, J. J. & Cooper, E. A novel neuronal P2xATP receptor ion channel with widespread distribution in the brain. J. Neurosci.16, 448–455 (1996).

24. Soto, F. et al. P2X4: An ATP-activated ionotropic receptor cloned from rat brain.Proc. Natl. Acad. Sci. USA 93, 3684–3688 (1996).

25. Le, K. T., Babinski, K. & Séguéla, P. Central P2X4 and P2X6 channel subunitscoassemble into a novel heteromeric ATP receptor. J. Neurosci. 18, 7152–7159(1998).

26. Hille, B. Ionic Channels of Excitable Membranes (Sinauer, Sunderland,Massachusetts, 1992).

27. Surprenant, A., Rassendren, F., Kawashima, E., North, R. A. & Buell, G. Thecytolytic P2Z receptor for extracellular ATP identified as a P2X receptor (P2X7).Science 272, 735–738 (1996).

28. Brake, A. J., Wagenbach, M. J. & Julius, D. New structural motif for ligand-gatedion channels defined by an ionotropic ATP receptor. Nature 371, 519–523 (1994).

29. Chen, C. C. et al. A P2X purinoceptor expressed by a subset of sensory neurons.Nature 377, 428–431 (1995).

30. Lewis, C. et al. Coexpression of P2X2 and P2X3 receptor subunits can account forATP-gated currents in sensory neurons. Nature 377, 432–435 (1995).

31. Cook, S. P., Vulchanova, L., Hargreaves, K. M., Elde, R. & McCleskey, E. W.Distinct ATP receptors on pain-sensing and stretch-sensing neurons. Nature 387,505–508 (1997).

32. Nicholson, C., ten Bruggencate, G., Stockle, H. & Steinberg, R. Ca2+ andpotassium changes in extracellular microenvironment of cat cerebellar cortex. J.Neurophysiol. 41, 1026–1039 (1978).

33. Krnjevic, K., Morris, M. E. & Reiffenstein, R. J. Changes in extracellular Ca2+ andK+ activity accompanying hippocampal discharges. Can. J. Physiol. Pharmacol.58, 579–582 (1980).

34. Krnjevic, K., Morris, M. E. & Reiffenstein, R. J. Stimulation-evoked changesin extracellular K+ and Ca2+ in pyramidal layers of the rat’s hippocampus.Can. J. Physiol. Pharmacol. 60, 1643–1657 (1982).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


Most excitatory synapses in the central nervous system use L-glu-tamate as a transmitter. This transmitter activates a variety of post-synaptic ionotropic receptors including the AMPA, NMDA andkainate subtypes1. At CA1 hippocampal synapses, high-frequen-cy stimulation can depolarize pyramidal neurons sufficiently topermit the activation of Ca2+-permeable NMDA channels. Thisinflux of Ca2+ stimulates PKC and Ca2+-calmodulin kinase II(CamKII), whose activity is essential for the induction of long-term potentiation (LTP)2. Calcium entering through NMDA chan-nels activates CamKII, causing a long-term enhancement ofpostsynaptic AMPA currents, perhaps by directly phosphorylatingthese receptors or by increasing the number of available AMPAreceptors at these synapses3. Tyrosine kinases also participate inthe induction of LTP at CA1 hippocampal synapses4–6. They maydo so at least in part by enhancing NMDA receptor activity7.

The serine-threonine kinase PKC directly phosphorylatesNMDA receptors8–11. Activation of this kinase has been reportedto enhance NMDA receptor currents recorded in isolated trigem-inal neurons12, in isolated and cultured hippocampal neurons13

and in hippocampal neurons in slices14, as well as the responsesof recombinant and native NMDA receptors expressed in Xeno-pus oocytes11,15. This suggests that phosphorylation of NMDAreceptor subunits is likely responsible for an upregulation of theirfunction. In contrast, there have been several reports that phor-bol esters may depress NMDA-receptor-mediated currents inhippocampal pyramidal neurons in the slice16 and in culturedhippocampal neurons (M.C. Bartlett, M.W. Salter & J.F.M. Soc.Neurosci. Abstr. 15, 218.13, 1989). Even more difficult to reconcileis the lack of correlation between the presence or absence of sitesof PKC-dependent phosphorylation in recombinant NMDAreceptor subunits and the functional potentiation of these recep-tors by phorbol esters11. The relationship between PKC-inducedphosphorylation of the NMDA receptor and changes in its func-tion is therefore uncertain.

G-protein-coupled receptors act viaprotein kinase C and Src to regulateNMDA receptors

W-Y. Lu1, Z-G. Xiong1, S. Lei1, B. A. Orser2, E. Dudek3, M. D. Browning3 and J. F. MacDonald1

1 Departments of Physiology, Pharmacology and 2Anaesthesia, University of Toronto, Toronto M5S 1A8, Canada3 Department of Pharmacology, Program in Neuroscience, University of Colorado Health Science Center, University of Colorado,

Denver, Colorado 80262, USA

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

The N-methyl-D-aspartate (NMDA) receptor contributes to synaptic plasticity in the central nervoussystem and is both serine-threonine and tyrosine phosphorylated. In CA1 pyramidal neurons of thehippocampus, activators of protein kinase C (PKC) as well as the G-protein-coupled receptor ligandsmuscarine and lysophosphatidic acid enhanced NMDA-evoked currents. Unexpectedly, this effectwas blocked by inhibitors of tyrosine kinases, including a Src required sequence and an antibodyselective for Src itself. In neurons from mice lacking c-Src, PKC-dependent upregulation was absent.Thus, G-protein-coupled receptors can regulate NMDA receptor function indirectly through a PKC-dependent activation of the non-receptor tyrosine kinase (Src) signaling cascade.

Stimulation of G-protein-coupled receptors activates non-receptor tyrosine kinases, including members of the Src as wellas the focal adhesion families17–19. In turn, activation of thesetyrosine kinases can provide a mechanism for stimulation of theMAPK kinase pathway that participates in the initiation of cellproliferation20. In hippocampal neurons, application of carba-chol initiates a PKC-dependent activation of tyrosine kinases thatcan in turn regulate the function of the Kv1.2 potassium chan-nel21,22. Therefore, G-protein-coupled receptors might also mod-ulate NMDA receptor function indirectly though activation of aPKC-dependent tyrosine-kinase signaling cascade. Here we pro-vide evidence that both muscarinic receptors and growth-pro-moting lysophosphatidic acid (LPA) receptors activate a PKC/Srcsignaling cascade that is responsible for regulating the activity ofNMDA channels at CA1 hippocampal synapses.

RESULTSWhole-cell, patch-clamp currents from acutely isolated CA1 hip-pocampal pyramidal neurons were used to assess NMDA receptorfunction, and 4β−phorbol 12-myristate 13-acetate (4β-PMA) wasused to activate endogenous PKC activity in these neurons. Appli-cation of 4β-PMA potentiated NMDA-activated currents evokedby subsaturating concentrations of NMDA and glycine (Fig. 1aand b). In contrast, the inactive phorbol ester 4α−PMA was with-out effect (Fig. 1b). The active phorbol increased the peak of theNMDA-evoked current (Ip; Fig. 1b). To confirm that this effectwas due to activation of PKC, we did a series of recordings withor without chelerythrine, a selective inhibitor of PKC23, in thepatch pipette. Chelerythrine strongly depressed the enhancementof Ip (Fig. 1b), confirming the role of endogenous PKC in theresponse to 4β-PMA. Steady-state current (Iss) was either littleaffected or slightly enhanced in amplitude following applicationof 4β-PMA (Fig. 1a and c). These results were confirmed usingcalphostin C, a different inhibitor23 of PKC (data not shown).

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


The degree of potentiation of Ip directly depended on the con-centration of NMDA and glycine used (Fig. 1c–e). In a fixed con-centration of glycine, increasing concentrations of NMDA wereassociated with a reduced potentiation of Ip (data not shown).Similar results were observed when responses to a fixed concen-tration of NMDA were compared in various concentrations ofglycine (Fig. 1e). Small potentiations of Ip were always observedin saturating concentrations of NMDA and glycine. Steady-statecurrents were only enhanced in lower concentrations of eitherNMDA or glycine, whereas a depression of steady-state currentswas apparent in saturating concentrations of this agonist and co-agonist (Fig. 1d). These results illustrate that 4β-PMA treatmentseemed to substantially enhance desensitization, assessed as theratio of steady-state to peak currents (Iss/Ip). This apparentenhancement of desensitization was accentuated at higher con-centrations of NMDA (data not shown) or glycine (Fig. 1f).

The phorbol ester may have altered the potency of the receptorfor NMDA or for glycine. Therefore the effects of 4β-PMA on theconcentration–response relationships for both glycine and NMDAwere examined. Application of 4β-PMA did not change the EC50values for glycine (control, 0.72 ± 0.11 µM; PMA, 0.56 ± 0.12 µM;n = 8 each group, p = 0.28) or NMDA (control, 94 ± 15 µM; PMA,81 ± 6.8 µM; n = 5 each group, p = 0.30) for the peak currents(Fig. 1g). Also the potentiation of peak NMDA-evoked currentsdid not depend on the holding potential (Fig. 1h).

Phosphatase inhibitors also enhance phosphorylation byendogenous kinases. We have previously shown that selectiveinhibitors of the Ca2+-independent serine-threonine phos-phatases PP1 and PP2A enhance NMDA-evoked currents record-

ed from cultured hippocampal neurons24. In isolated CA1 neu-rons, extracellular application of the phosphatase inhibitor okada-ic acid slightly enhanced Ip alone (Fig. 2c), but substantiallyaccentuated the potentiation induced by 4β-PMA (Fig. 2). Thisresult suggests that endogenous serine-threonine phosphatasesare significant in limiting the PKC-dependent potentiation ofNMDA-evoked currents in CA1 hippocampal neurons.

Stimulation of G-protein-coupled receptors acting throughPKC or activation of PKC using phorbol esters can activate non-receptor tyrosine kinases18–20. We therefore tested the hypothesisthat activation of G-protein-coupled receptors modulates NMDAreceptors though a PKC-linked tyrosine kinase signaling path-way. Pre-exposure to genistein, a broad-spectrum tyrosine kinaseinhibitor, blocked the phorbol-ester-induced potentiation of thesecurrents. In contrast, the enhancement was insensitive todaidzein, the inactive but structurally related analogue of genis-tein (Fig. 3a). To provide additional support for a tyrosine kinaseblockade, we also compared the effects of lavendustin A and itsinactive analogue lavendustin B. Lavendustin A pretreatmentblocked the phorbol-ester-induced potentiation, whereas laven-dustin B did not (Fig. 3b and c).

To determine if the phorbol-ester-induced potentiation ofNMDA responses observed in Xenopus oocytes might also bemediated via activation of tyrosine kinases, we expressed nativeNMDA receptors in these cells. Total rat brain mRNA was inject-ed into oocytes, and NMDA-evoked currents were assessed beforeand after application of 4β-PMA. The phorbol ester approxi-mately quadrupled these currents but pre-application of genis-tein substantially reduced the phorbol-ester-induced potentiation,

articles

Fig. 1. Phorbol esters enhance peakNMDA-evoked currents and seem to facil-itate desensitization in isolated CA1 hip-pocampal neurons. (a) Time-dependentenhancement of NMDA-evoked whole-cellcurrents (50 µM NMDA; 0.5 µM glycine)by 4β-PMA (100 nM). (b) Normalizedpeak currents (Ip) before and during theapplication of phorbol esters. Theresponses to NMDA were enhanced by4β-PMA (100 nM, n = 9) but not by theinactive isomer 4α-PMA (100 nM, n = 9). Inanother series of recordings (n = 6), chel-erythrine (10 µM) prevented the enhance-ment by 4β-PMA. (c) SuperimposedNMDA-evoked currents before and afterexposure to 4β-PMA. This phorbol estersubstantially enhanced Ip without alteringsteady-state currents (Iss) recorded2.5 seconds after the beginning of NMDAapplication. (d) Responses of the sameneuron using saturating concentrations ofNMDA (1 mM) and glycine (10 µM). Thepeak current was still enhanced, althoughmuch less so, but Iss was substantiallydepressed. (e) Potentiation of Ip (100 µMNMDA) by 4β-PMA normalized to controlvalues (n = 8) demonstrates that this effectdecreases as the concentration of glycine isincreased. Each set of points was fitted using the relationship y = y0 + ae–bx (y0 = 1.3, a = 0.3; b = 6.4 × 10–5 M–1). Similar results were found with increas-ing concentrations of NMDA (data not shown, glycine 3 µM; y0 = 1.2; a = 0.3; b = 5.5 × 10–3 M–1). (f) For a series of recordings, Iss/Ip versus glycine con-centration. In the presence of PMA, the degree of desensitization was enhanced (n = 8; two-way ANOVA; p < 0.005), with the limiting values of thisratio reaching yo = 0.68 for the control condition and yo = 0.41 in the presence of PMA. (g) PMA had no effect on the concentration–response rela-tionship for NMDA in the presence of 3 µM glycine or for glycine in the presence of 100 µM NMDA (data not shown). (h) The current–voltage rela-tionships for an example recording before and after PMA application.

b

f

gh

a

c d

No

rmal

ized

I Pea

k

Time course (min)

[Glycine] (µM)

[Glycine] (µM)N

orm

aliz

ed I P

eak

No

rmal

ized

I Pea

k

eI ss

/Ip

[NMDA] (mM)


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

A member of the Src family is the most likely candidate7

for the tyrosine kinase upregulating NMDA channel activi-ty. We therefore included in the whole-cell patch pipette theantibody anti-cst1, which selectively inhibits the Src familyof kinases7,25. A nonselective immunoglobulin G (IgG) frac-tion was used as a control for these experiments. The phor-bol-ester-induced potentiation was substantially depressed

in the presence of anti-cst1 (Fig. 5a). In an additional series ofrecordings, a Src-selective blocking antibody anti-Src1 (refs. 7, 25)was applied to determine if Src itself was involved. The 4β-PMA-induced potentiation was again blocked (data not shown). Addi-tional evidence that Src was the kinase involved was provided byapplying the unique domain peptide fragment Src(40–58)7,25 tothe interior of the cells. This peptide blocked the phorbol-ester-induced potentiation, whereas a control peptide composed of ascrambled sequence sSrc(40–58) was ineffective (Fig. 5b). Wethen examined neurons from mice lacking Src because of a tar-geted gene deletion26. The phorbol ester enhanced NMDA-evoked currents in neurons from control mice, but littlepotentiation was found in neurons taken from the mutant mice

Fig. 2. Applications of the phosphatase inhibitor okadaic acid fur-ther enhanced the effects of PMA on peak NMDA-activated cur-rents in isolated hippocampal neurons. In this and all subsequentfigures, NMDA was applied at a concentration of 50 µM andglycine at 0.5 µM unless otherwise indicated. (a, b) Okadaic acid(20 nM) was included in the patch pipette in one series of record-ings (n = 6) and left out in matched recordings from a secondseries of cells (n = 5). Potentiation by PMA was significantly greaterwith the phosphatase inhibitor (10 minutes in PMA, 35 ± 8% nookadaic acid, 76 ± 11% with okadaic acid, two-way ANOVA,p < 0.05). (c) Okadaic acid (20 nM) was also applied to the bathingsolution in a separate series of experiments. The application ofPMA, after stabilization of currents in the presence of okadaic acid,potentiated Ip more than twofold over what was observed in theabsence of this phosphatase inhibitor.


suggesting that at least a proportion of this effect was mediatedthrough activation of tyrosine kinases (Fig. 3d).

The effect of 4β-PMA on single NMDA channel activity incell-attached patch recordings was then examined using isolat-ed hippocampal neurons. The probability of channel openingwas increased following application of 4β-PMA (Fig. 4a–c), aswere mean open times, burst, cluster and supercluster durations(Fig. 4d). These changes were prevented by application of genis-tein (Fig. 4a, c and d). This evidence suggests that phorbol esterspotentiate NMDA channel activity by a mechanism that dependsupon activation of a tyrosine kinase. Therefore an attempt wasmade to identify the specific tyrosine kinase involved in theresponse to 4β-PMA.

articles

a

b c

Control + Okadaic acid

No

rmal

ized

I Pea

k

No

rmal

ized

I Pea

k

Time course (min) Time course (min)

Fig. 3. Non-selective blockers of tyrosine kinasesreduce the PMA-induced potentiation of NMDA-evoked currents recorded in isolated neurons andXenopus oocytes. (a) The 4β-PMA-induced potentia-tion of Ip in isolated neurons was prevented by pre-application of genistein (50 µM) but not by the inactiveanalogue daidzein (50 µM). A separate series of record-ings (data not shown) demonstrated that genistein itselfdepressed NMDA-evoked currents (18.5 ± 7.6%, n = 8),but a stable level of depression was reached 2 to 3 min-utes after application onset. (b) Lavendustin A (+Lav-A)but not lavendustin B (+Lav-B) blocked the PMA-induced potentiation of peak currents. (c) Data for aseries of recordings are shown with lavendustin A(10 µM, n = 8) or lavendustin B (10 µM, n = 9). (d) PMAinduced a more than fourfold (507 ± 105%, n = 12)potentiation of NMDA-evoked currents when wholebrain mRNA was expressed in Xenopus oocytes.Genistein depressed this potentiation (335 ± 79%,n = 11, p < 0.01).

a b

c d

Time course (min)

Time course (min)Time course (min)

No

rmal

ized

I Pea

kN

orm

aliz

ed I P

eak

No

rmal

ized

I Pea

k

L Genistein

P Daidzein


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


(Fig. 5c). Taken together, this evidence strongly supports thehypothesis that Src is required for the 4β-PMA-induced enhance-ment of NMDA-activated currents.

We anticipated that intracellular application of recombinantSrc would mimic the effects of phorbol esters on NMDA-evokedcurrents. This was confirmed when we perfused recombinantpp60c–Src into the patch pipette while simultaneously monitoringNMDA-activated currents. Intracellular perfusion of c-Srcenhanced peak currents and seemed to enhance desensitization byreducing Iss/Ip (Fig. 5d–f). To determine if activation of PKC was

also required for the Src-induced potentiation of NMDA-evokedcurrents, we co-applied c-Src together with the PKC inhibitor chel-erythrine. This inhibitor failed to alter the enhancement (Fig. 5e),demonstrating that activation of Src is downstream of PKC.

The phorbol-ester-induced modulation of Ip strongly depend-ed on the concentrations of agonist, and in saturating concen-trations of NMDA and glycine, Iss was significantly depressed.The concentrations of glutamate reaching the receptors duringtransmission are likely to be near saturating for NMDA recep-tors. Therefore, it was not obvious what effect activation of PKC

articles

Fig. 4. Cell-attached patches were also used torecord single-channel NMDA-activated currents(10 µM) in isolated CA1 neurons. (a) Example single-channel currents before (1)and after (2) application of 4β-PMA (100 nM),genistein (50 µM; 3) or genistein (50 µM) plus4β-PMA (100 nM; 4). (b) A continuous recordof channel open probability (PO) before and dur-ing the bath application of 4β-PMA. (c) Similarly,a continuous record of channel open probabilitybefore and during the application of either genis-tein or genistein plus 4β-PMA. (d) Effects of 4β-PMA (hatched bars), genistein (open bars) andgenistein plus 4β-PMA (striped bars) on single-channel parameters. Values given are mean ± s.e.Po, open probability; to, mean open time; B,burst length; C, cluster length; S, superclusterlength (**p < 0.01, *p < 0.05 one-tailed t test). Ineach case, channel amplitude was unchanged(control, 2.80 ± 0.05; 4β-PMA, 2.70 ± 0.11;genistein, 2.81 ± 0.23; genistein + 4β-PMA, 2.86± 0.26 pA, n = 6 for each).

Fig. 5. The 4β-PMA-induced potentia-tion of NMDA-evoked currentsdepends on the activation of Src. (a) The 4β-PMA-induced enhance-ment was significantly depressed byincluding anti-cst1 (10 µg/ ml, n = 11)in the patch pipette but not by thenon-selective immunoglobulin G frac-tion (IgG, 10 µg/ml, n = 7). (b) AddingSrc(40–58) (25 µg/ml, n = 11) in thepipette solution blocked the enhance-ment, but a scrambled Src7

(sSrc(40–58), 25 µg/ml, n = 5,AGSHAPFPSPARAGVAPDA) did not.(c) PMA enhanced NMDA-evoked cur-rent in hippocampal CA1 neuronsacutely isolated from control mice(n = 10) but not in neurons taken fromSrc-lacking mutants (B6, 129-Srctm1Sor,n = 10, cells from three different ani-mals). (d) Internal perfusion of thepatch pipette with pp60c-Src for fiveminutes potentiated Ip in rat CA1 hip-pocampal neurons. The Ip wasincreased to 129 ± 9% of the controlvalue (n = 6), whereas Iss wasdecreased to 94 ± 6%. Time constantsof desensitization were decreased byc-Src. (e) Co-perfusion of c-Src andchelerythrine did not alter the enhancement of NMDA-evoked currents (c-Src, n = 6; c-Src and chelerythrine, n = 7; p > 0.05). (f) The valuesIss/Ip were decreased, demonstrating an apparent enhanced desensitization of these currents (n = 6, p < 0.01).

a b

c d

Per

cen

t o

f co

ntr

ol

P0

P0

a b

c d

e f

No

rmal

ized

I Pea

k

No

rmal

ized

I Pea

kN

orm

aliz

ed I P

eak

No

rmal

ized

I Pea

k

Time course (min) Time course (min)

Time course (min)

I ss/

I p


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


might have on synaptic NMDA receptors. To address this ques-tion, we recorded miniature excitatory synaptic currents (mEPSCs) in cultured hippocampal neurons following intracel-lular application of PKM. This constitutively active form of PKCenhanced the NMDA-receptor-mediated component of sponta-neous miniature currents (Fig. 6a, c and d). In contrast, co-appli-cation of PKM and Src(40-58) was ineffective (Fig. 6b–d). Inaddition, the Src inhibitory peptide itself depressed the ampli-tude of the NMDA-receptor-mediated component, suggestingsome potential ongoing Src regulation of synaptic NMDA recep-tors in these neurons (Fig. 6b and c).

The CA1 neurons of the hippocampus have a M1 subtype ofmuscarinic receptor28, and activation of this receptor potentiatesNMDA-evoked responses in these neurons29. Muscarinic recep-tors act via non-receptor tyrosine kinases, and βγ subunitsreleased from G proteins may directly activate Src kinase18,30.Therefore, we determined whether or not muscarine potentiatesNMDA-evoked currents by activating PKC. Application of mus-carine enhanced Ip (Fig. 7a and b), and inclusion of chelerythrinein the patch pipette blocked this potentiation (Fig. 7b), demon-strating a role for endogenous PKC. We then examined whetheror not this muscarine-induced potentiation required the activa-tion of tyrosine kinases. This potentiation was blocked by appli-cation of lavendustin A, but not by lavendustin B (Fig. 7c).Blockade of the muscarine-induced potentiation of NMDA-acti-vated currents was also demonstrated using Src(40–58) but notthe scrambled sequence sSrc(40–58) (Fig. 7d). Including anti-Src1 in the pipette also blocked this enhancement (Fig. 7e). Thus,PKC and Src kinase must be activated in the upregulation ofNMDA receptors by muscarinic receptors.

We next considered the possibility that a non-transmitterG-protein-coupled receptor might also regulate NMDA recep-tors through a PKC/Src signaling cascade. For this, we chosethe lysophosphatidic acid (LPA) receptor, which stimulates cel-lular proliferation via non-receptor tyrosine kinase and theMAPK cascade20. Application of LPA enhanced NMDA-evoked

currents and occluded the phorbol-ester-induced potentiation(Fig. 7f). The effects of LPA were also blocked by applicationof chelerythrine, genistein or Src(40–58) (Fig. 7f), demon-strating that LPA receptors also modulate NMDA responses viaa PKC/Src signaling cascade.

DISCUSSIONPhorbol esters and PKC regulate NMDA channel activityWe have shown that the phorbol ester 4β-PMA selectivelypotentiated peak NMDA-evoked currents in isolated CA1 pyra-midal neurons. This effect was mediated by endogenous PKCbecause its actions were not mimicked by the inactive PMAanalogue and were blocked by application of the selective PKCinhibitors chelerythrine or calphostin C in the patch pipette.Furthermore, we recently reported similar effects on peakNMDA-activated currents following the intracellular perfusionof PKM into these neurons13. In the present study, the poten-tiation of Ip was not associated with any change in the poten-cy of either NMDA or its co-agonist glycine. This suggests thatthe affinity of the receptor for the agonists was likely unaltered.In cell-attached patches from isolated neurons, we demon-strated that PMA substantially increased the apparent openprobability of NMDA channels and altered various open timeparameters, including supercluster durations.

Although the substrate of PKC in these experiments isunknown, we anticipated that it might be subject to dephospho-rylation by serine-threonine phosphatases. We demonstrated thatokadaic acid was able to further potentiate the effects of PMA onIp. This result indicates that serine-threonine phosphatases are alsoimportant for regulating the PKC-induced potentiation of thesecurrents. This result is consistent with our previous observationsusing phosphatase inhibitors on cultured hippocampal neurons24.

We previously demonstrated that phorbol esters reversiblydepress NMDA-evoked currents in cultured hippocampal neu-rons. In contrast, we found that peak NMDA-evoked currentswere enhanced by intracellular application of PKM to either cul-

articles

Fig. 6. Intracellular PKM enhanced the NMDA-receptor-mediated component of mEPSCs recordedfrom cultured hippocampal neurons, and thisenhancement was prevented by co-application ofSrc(40–58). (a) Miniature EPSCs recorded from aneuron using a patch pipette containing PKM (1). Thecompetitive NMDA receptor antagonist AP5(20 µM) was used to block NMDA-receptor-medi-ated components (2), and subtraction revealed thiscomponent (3, bold). The n values indicate the num-ber of miniatures averaged for each trace shown.Application of PKM enhanced the peak of this

NMDA-receptor-mediated component. (b) When the pipettecontained PKM and Src(40–58), no progressive potentiation ofthe NMDA component was observed. (c) Using this protocol andpipettes lacking (control, n = 5) or containing the dialysis solutionused to collect PKM (mock, n = 5), we observed no change in theamplitude of this NMDA-receptor-mediated component over theperiod from 2 (black bars) to 8–12 minutes (gray bars) of record-ing. The inclusion of PKM was associated with a significant poten-tiation of NMDA-evoked miniatures (n = 7; p < 0.05).Co-application of PKM and Src(40–58) was associated with sub-stantially smaller NMDA miniatures and lacked the progressivepotentiation (n = 5, p < 0.05). (d) The data taken at 8–12 minutesnormalized with respect to the amplitudes recorded at 2 minutes.PKM enhanced NMDA miniature amplitude, but this effect wasblocked by Src(40–58).

a b

dc

Intracellular PKM Intracellular PKM + Src(40–58)

Am

plit

ud

e (p

A)

of

NM

DA

com

po

nen

t o

f m

EPSC

s

No

rmal

ized

NM

DA

co

mp

on

ent

of

mEP

SCs

n = 138 n = 133 n = 115 n = 129


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


tured or isolated hippocampal neurons13. We can now providea rationale for this apparent contradiction. Specifically, detectionof a PKC-induced potentiation of Ip clearly depends both on theconcentrations of NMDA and glycine present at the receptorsand on whether or not the kinetics of agonist delivery are suffi-cient to resolve Ip. In saturating or near-saturating concentra-tions of NMDA and glycine, phorbol esters enhanced peakcurrents much less while substantially depressing Iss. In our orig-inal experiments in cultured hippocampal neurons, we appliednear-saturating concentrations of agonist using a relatively slowapplication technique. Thus, the depression we observed previ-ously likely correlates with our failure to resolve Ip and our detec-tion of a phorbol-ester-induced depression of Iss.

Glycine concentrations at central synapses are unknown; how-ever, concentrations of glutamate reaching synaptic receptorsmay be near saturating for NMDA receptors. Therefore it wasnecessary to determine whether activation of PKC would poten-tiate or depress synaptic currents. Phorbol esters could not beused for this purpose because of their strong effects on the presy-naptic release of glutamate31. We found that intracellular appli-cation of PKM potentiated the NMDA receptor component ofspontaneous mEPSCs, confirming that activation of PKCenhances the function of synaptically located NMDA receptors.

PKC enhances NMDA channel activity by activating SrcWe have shown that the phorbol-ester-induced potentiation of Ipdepends on activation of tyrosine kinases. This was supported bythe findings that the PMA-induced potentiation was preventedby application of the tyrosine kinase inhibitors genistein and laven-dustin A but not by the inactive compounds daidzein and laven-dustin B. Genistein also blocked the PMA-induced enhancementof NMDA channel activity recorded in cell-attached patches.

We have provided substantial evidence that the tyrosine kinaseinvolved in the PKC-induced potentiation of NMDA currents isa member of the Src family of kinases, or Src itself. We haveshown that the inhibitory antibodies anti-cs1 and anti-Src1 pre-vented the effects of PMA on NMDA receptors. In addition, thepeptide Src(40–58) blocked the PKC-dependent modulation ofNMDA responses, whereas the scrambled sequence sSrc(40–58)was without effect. More support for this conclusion was pro-vided by our demonstration that the phorbol-ester-inducedpotentiation was absent in cells taken from Src-deficient mice.Furthermore, the PKM-induced enhancement of the NMDAreceptor component of mEPSCs recorded in cultured hip-pocampal neurons was prevented by co-application ofSrc(40–58), providing evidence that synaptic NMDA receptorscan be regulated by this mechanism. We have also demonstrat-

articles

Fig. 7. G-protein-coupled receptors mod-ulate NMDA-receptor-mediated responsesthrough activation of PKC and proteintyrosine kinase Src. (a) Application ofmuscarine (10 µM) potentiated peakNMDA-evoked currents in isolated neu-rons. (b) This potentiation was eliminatedby the PKC inhibitor chelerythrine in thepatch pipette (control, n = 6; chelery-thrine, n = 5, two-way ANOVA, p < 0.005).(c) The potentiation was prevented bypretreatment of the cells with lavendustinA (10 µM, n = 5) but not by lavendustin B(n = 4). (d) The potentiation of Ip was pre-vented by Src(40–58) (25 µg/ml, n = 7) butnot by sSrc(40–58) (25 µg/ml, n = 5). (e) The potentiation was also depressedby including anti-Src1 but not the controlIgG (n = 5). (f) Application of LPA(lysophosphatidic acid, 20 nM, n = 7)enhanced Ip and occluded the potentiationby 4β-PMA. The effects of LPA wereblocked by including chelerythrine (n = 3)in the pipette or by co-applying genistein(50 µM, n = 5). The Src(40–58) (n = 5) alsoblocked this potentiation,whereas thescrambled peptide (n = 5) failed to do so.

a b

c

d

e

No

rmal

ized

I Pea

k

No

rmal

ized

I Pea

k

No

rmal

ized

I Pea

k

No

rmal

ized

I Pea

k

Time course (min)

Time course (min)

Time course (min)

Time course (min)

f

No

rmal

ized

I Pea

k

Time course (min)


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


ed that PMA and PKM13 enhance the apparent open probabilityof NMDA channels in a manner that resembles the effects of cSrcon these channels7. The intracellular application of either cSrcor PKM also enhanced peak NMDA currents. Finally, the effectsof cSrc were insensitive to the PKC inhibitor chelerythrine, where-as the effects of PKM were blocked by a selective Src inhibitor,demonstrating that PKC sequentially activates Src in hippocam-pal neurons. Src has been reported to enhance responses ofrecombinant NMDA receptors by relieving their inhibition byextracellular zinc32. Unexpectedly, neither the potentiation of Ipby PMA or cSrc in hippocampal neurons was reduced by chelat-ing extracellular zinc (unpublished data), suggesting an alterna-tive mechanism is responsible.

There is little evidence that Src is directly phosphorylated andactivated by PKC. However, there is strong evidence that PKCcan activate CAKβ/PYK2 in hippocampal neurons21. In turn,autophosphorylated CAKβ/PKY2 binds to and potentially acti-vates Src kinase18–20. This suggests that the mechanism wherebyPKC stimulates Src in CA1 hippocampal neurons maybe throughthe co-activation of additional tyrosine kinases.

G-protein-coupled receptors signal to NMDA receptorsG-protein-coupled receptors, including muscarinic and LPAreceptors acting through non-receptor tyrosine kinases, activatethe MAPK cascade18,20,27. Our results demonstrate that mus-carinic receptors in CA1 pyramidal neurons act through the sameinitial stages of this cascade to signal a parallel upregulation ofNMDA channel function. This is supported by our demonstra-tions that the muscarine-induced potentiation of NMDA-acti-vated currents is prevented by the selective PKC inhibitorchelerythrine, by the non-selective tyrosine kinase inhibitor laven-dustin A, by the Src-specific blocking antibody anti-Src1 and bythe Src inhibitory peptide Src(40–58). Chelerythrine, genisteinand Src(40–58) also blocked the effect of LPA on peak NMDA-activated currents, and LPA itself occluded the effects of PMA,demonstrating that LPA receptors likely act through the samekinase cascade. NMDA channel activity is associated with boththe growth and development of central neurons33. Neonatal hip-pocampal pyramidal neurons are unlikely to proliferate duringactivation of the MAPK pathway. However, these neurons under-go extensive morphological changes in concert with long-termchanges in synaptic plasticity34. Therefore, coupling NMDAreceptor function directly to this signaling cascade provides ameans of controlling both the expression of growth-related genesand the influx of calcium through NMDA receptors.

METHODSIsolated neurons and recordings of NMDA-evoked currents. CA1 neu-rons were isolated from hippocampal slices taken from postnatal rats(Wistar 10–24 days) or mice (B6129F2/J wild type and B6129-Srctm1Sor,Jackson Laboratory 7–15 days) using described procedures35. We notedthat the degree of potentiation of NMDA-evoked currents was least incells taken from 7-day-old rats and greatest in those 14 to 24 days. Thisaccounts for some of the observed variations in the degree of potenti-ation. To control for such variation, we made recordings from cellstaken from the same animal (both control and drug-treated). Record-ings from control and treated cells were always made on the same day.The extracellular solution was composed of 140 mM NaCl, 1.3 mMCaCl2, 5 mM KCl, 25 mM HEPES, 33 mM glucose and 0.0005 mMtetrodotoxin, with pH 7.4 and osmolarity between 325 and 335mosmol. Only neurons that retained their pyramidal shape, includinga major primary and several secondary dendritic processes, were usedfor recordings. Whole-cell, patch-clamp recordings were done at roomtemperature (20–22°C). After formation of a whole-cell configuration,

the recorded neurons were voltage clamped and lifted into the streamof solution supplied by a computer-controlled, multi-barreled perfusionsystem (τ of exchange of about 2 ms). To monitor series resistance, weapplied a voltage-step of –10 mV before each application of NMDA.The series resistance in these recordings varied between 6 and 8 MΩ.Recordings where series resistance varied by more than 10% were reject-ed. No electronic compensation for series resistance was used. Theintracellular solution contained 140 mM CsmethylSO4 or CsF, 11 mMEGTA, 1 mM CaCl2, 2 mM MgCl2, 10 mM HEPES, 2 mM TEA and4 mM K2ATP, with pH 7.3 and osmolarity between 295 and 300mosmol. Some drugs were included in the patch pipette, whereas inthe case of cSrc, an intracellular perfusion system was used to applythe kinase and peptides to the tip of the patch pipette. The source ofdrugs used in this study are as follows: NMDA, glycine and muscarine(Sigma), 4β-PMA, 4α -PMA and PPD (Alexis), genistein (RBI),daidzein, lavendustin A and B (Calbiochem), pp60c-Src (Upstate Bio-chemicals), anti-cst1, anti-Scr1 (S. Courtneidge, SUGEN, RedwoodCity, California), Src(40–58), sSrc(40–58) (M.W. Salter, Hospital forSick Children, Toronto, Ontario). Currents were recorded using anAxopatch 1-B or Axopatch 200 amplifier (Axon Instruments), and datawere digitized, filtered (2 kHz) and acquired using the pClamp6 pro-gram (Axon Instruments). All population data are expressed asmean ± s.e. The Student’s paired t-test or the ANOVA test (one- or two-way) was used as appropriate to determine statistical significance. Sin-gle-channel activity was recorded from isolated neurons using thecell-attached patch configuration. Following formation of the patch,the cell was superfused with extracellular solution. Patch pipettes werefilled with the extracellular solution plus NMDA (10 µM) and glycine(1 µM). Patches with stable basal channel activity were used for theanalysis, and an 8–10 minute period of continuous recording was donebefore applying 4β-PMA (100 nM) to the superfusion. Probability ofopening, open times and shut times were determined off-line using the50% crossing thresholds method. Analysis of dwell-time distributionswas done using the sum of multiple exponentials and the Levenberg-Marquard least-squares method. Groupings of openings into bursts,clusters and superclusters were determined as described7.

Miniature synaptic currents in cultured hippocampal neurons. Cul-tures of fetal hippocampal neurons were prepared according todescribed techniques36. The cultures were used for electrophysiolog-ical recordings 12–17 days after plating. The electrodes (3–5 MΩ)were coated with Sylgard to improve the signal-to-noise ratio. Spon-taneous miniature EPSCs in cultured hippocampal neurons wererecorded immediately after formation of the whole-cell patch con-figuration and continuously monitored for 35–80 minutes. Each cellwas exposed to extracellular solution supplemented with tetrodotox-in (0.5 µM), strychnine (1 µM), bicuculline methiodide (10 µM) andglycine (3 µM). Miniature EPSCs were filtered at 2 kHz and storedon tape before their off-line acquisition with an event-detection pro-gram (SCAN, Strathclyde software; courtesy of J. Dempster). Fordetection, the trigger level was set approximately three times higherthan the baseline noise. False events were eliminated by subsequentinspection of the raw data. For averaging, the number of selectedevents for each group ranged from 120–450.

Oocytes injected with whole-brain mRNA. Oocytes from Xenopus lae-vis were dissected and treated with 0.2 % collagenase (Sigma) for twohours in Ca2+-free saline and subsequently defolliculated using fineforceps. Selected stage V and VI oocytes were injected with whole ratbrain mRNA (75 ng per oocyte) and maintained at 16°C. Four to sixdays after injection, oocytes were used for recordings using a conven-tional two-electrode, voltage-clamp method. The external bath wasMg2+-free Ringer’s solution containing 96 mM NaCl, 2 mM KCl,1.8 mM Ca2+ and 5 mM HEPES, with pH 7.4. Currents were evoked bybath application of NMDA (200 µM) together with glycine (10 µM). Tominimize Ca2+-activated Cl– currents, we replaced Ca2+ with Ba2+

during NMDA application. For the genistein-treated group, oocyteswere incubated in solution containing 50 µM genistein 1 hour beforeand during the recording. The initial peak amplitude of NMDA-evokedcurrents was measured and normalized to that of the control.

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


ACKNOWLEDGEMENTSWe thank M.W. Salter, S. Courtneidge and J. Roder for reagents and knockout

mice. We thank M.W. Salter for assistance with the design of some of the

experiments. We also thank X-M. Yu and L-Y. Wang for discussions. This work

was supported by grants from the MRC of Canada. W-Y.L., Z-G.X. and S.L. are

fellows of the Heart & Stroke, the MRC and the Ontario Neurotrauma Fnd.,

respectively.

RECEIVED 21 OCTOBER 1998, ACCEPTED 21 JANUARY 1999

1. Hollmann, M. & Heinemann, S. Cloned glutamate receptors. Annu. Rev.Neurosci. 17, 31–108 (1994).

2. Bliss, T. V. & Collingridge, G. L. A synaptic model of memory: long-termpotentiation in the hippocampus. Nature 361, 31–39 (1993).

3. Barria, A., Muller, D., Derkach, V., Griffith, L. C. & Soderling, T. R.Regulatory phosphorylation of AMPA-type glutamate receptors by CaM-KIIduring long-term potentiation. Science 276, 2042–2045 (1997).

4. Kojima, N. et al. Rescuing impairment of long-term potentiation in fyn-deficient mice by introducing Fyn transgene. Proc. Natl. Acad. Sci. USA 94,4761–4765 (1997).

5. O’Dell, T. J., Grant, S. G. N., Karl, K., Soriano, P. M. & Kandel E. R.Pharmacological and genetic approaches to the analysis of tyrosine kinasefunction in long-term potentiation. Cold Spring Harbor Symp. Quant. Biol.57, 517–526 (1992).

6. Lu, Y. M., Roder, J. C., Davidow, J. & Salter, M. W. Src activation in theinduction of long-term potentiation in CA1 hippocampal neurons. Science279, 1363–1367 (1998).

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

8. Tingley, W. G. et al. Characterization of protein kinase A and protein kinase Cphosphorylation of the N-methyl-D-aspartate receptor NR1 subunit usingphosphorylation site-specific antibodies. J. Biol. Chem. 272, 5157–5166(1997).

9. Tingley, W. G., Roche, K. W., Thompson, A. K. & Huganir R. L. Regulation ofNMDA receptor phosphorylation by alternative splicing of the C-terminaldomain. Nature 364, 70–73 (1993).

10. Leonard, A. S. & Hell, J. W. Cyclic AMP-dependent protein kinase andprotein kinase C phosphorylate N-methyl-D-aspartate receptors at differentsites. J. Biol. Chem. 272, 12107–12115 (1997).

11. Zukin, R. S. & Bennett, M. V. Alternatively spliced isoforms of the NMDARIreceptor subunit. Trends Neurosci. 18, 306–313 (1995).

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

13. Xiong, Z.-G. et al. Regulation of NMDA receptor function by constitutivelyactive protein kinase C. Mol. Pharmacol. 54, 1055–1063 (1998).

14. Ben-Ari, Y., Aniksztejn, L. & Bregestovski, P. Protein kinase C modulation ofNMDA currents: an important link for LTP induction. Trends Neurosci. 15,333–339 (1992).

15. Kelso, S. R., Nelson, T. E. & Leonard, J. P. Protein kinase C-mediatedenhancement of NMDA currents by metabotropic glutamate receptors inXenopus oocytes. J. Physiol. (Lond.) 449, 705–718 (1992).

16. Markram, H. & Segal, M. Activation of protein kinase C suppresses responsesto NMDA in rat CA1 hippocampal neurons. J. Physiol. (Lond.) 457, 491–501(1992).

17. Sasaki, H. et al. Cloning and characterization of cell adhesion kinase beta, a

novel protein-tyrosine kinase of the focal adhesion kinase subfamily. J. Biol.Chem. 270, 21206–21219 (1995).

18. Luttrell, L. M. et al. Role of c-Src tyrosine kinase in G protein-coupledreceptor- and Gbetagamma subunit-mediated activation of mitogen-activated protein kinases. J. Biol. Chem. 271, 19443–19450 (1996).

19. Della, R. G. J. et al. Ras-dependent mitogen-activated protein kinaseactivation by G protein-coupled receptors. Convergence of Gi- and Gq-mediated pathways on calcium/calmodulin, Pyk2, and Src kinase. J. Biol.Chem. 272, 19125–19132 (1997).

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

21. Lev, S. et al. Protein tyrosine kinase PYK2 involved in Ca2+-inducedregulation of ion channel and MAP kinase functions. Nature 376, 737–745(1995).

22. Felsch, J. S., Cachero, T. G. & Peralta, E. G. Activation of protein tyrosinekinase PYK2 by the m1 muscarinic acetylcholine receptor. Proc. Natl. Acad.Sci. USA 95, 5051–5056 (1998).

23. Hemmings, H. C. J. in Regulatory Protein Modification: Techniques andProtocols (ed. Hemmings, H. C. J.) 121–218 (Humana, Totowa, New Jersey,1997).

24. Wang, L. Y., Orser, B. A., Brautigan, D. L. & MacDonald, J. F. Regulation ofNMDA receptors in cultured hippocampal neurons by protein phosphatases1 and 2A. Nature 369, 230–232 (1994).

25. Roche, S., Fumagalli, S. & Courtneidge, S. A. Requirement for Src familyprotein tyrosine kinases in G2 for fibroblast cell division. Science 269,1567–1569 (1995).

26. Lowell, C. A. & Soriano, P. Knockouts of Src-family kinases: stiff bones,wimpy T cells, and bad memories. Genes Devel. 10, 1845–1857 (1996).

27. Gutkind, J. S. The pathways connecting G protein-coupled receptors to thenucleus through divergent mitogen-activated protein kinase cascades. J. Biol.Chem. 273, 1839–1842 (1998).

28. Levey, A. I. Muscarinic acetylcholine receptor expression in memory circuits:implications for treatment of Alzheimer disease. Proc. Natl. Acad. Sci.USA 93,13541–13546 (1996).

29. Marino, M. J., Rouse, S. T., Levey, A. I., Potter, L. T. & Conn, P. J. Activation ofthe genetically defined m1 muscarinic receptor potentiates N-methyl-D-aspartate (NMDA) receptor currents in hippocampal pyramidal cells. Proc.Natl. Acad. Sci.USA 95, 11465–11470 (1998).

30. Igishi, T. & Gutkind, J. S. Tyrosine kinases of the Src family participate insignaling to MAP kinase from both Gq and Gi-coupled receptors. Biochem.Biophys. Res. Commun. 244, 5–10 (1998).

31. Parfitt, K. D. & Madison, D. V. Phorbol esters enhance synaptic transmissionby a presynaptic, calcium-dependent mechanism in rat hippocampus. J. Physiol.(Lond.) 471, 245–268 (1993).

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

33. Constantine-Paton, M. NMDA receptor as a mediator of activity-dependentsynaptogenesis in the developing brain. Cold Spring Harbor Symp. Quant.Biol. 55, 431–443 (1990).

34. Wheal, H. V. et al. Molecular mechanisms that underlie structural andfunctional changes at the postsynaptic membrane during synaptic plasticity.Prog. Neurobiol. 55, 611–640 (1998).

35. Wang, L. Y. & MacDonald, J. F. Modulation by magnesium of the affinity ofNMDA receptors for glycine in murine hippocampal neurones. J. Physiol.(Lond.) 486, 83–95 (1995).

36. MacDonald, J. F., Mody, I. & Salter, M. W. Regulation of N-methyl-D-aspartate receptors revealed by intracellular dialysis of murine neurones inculture. J. Physiol. (Lond.) 414, 17–34 (1989).

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


Neurogenesis—the process whereby neuronal progenitor cellsproliferate and differentiate into postmitotic neurons—is tightlyregulated in the vertebrate1. Although many studies have focusedon molecular signals that stimulate neurogenesis2, signaling mol-ecules that inhibit it are just as likely to be important in regulat-ing neuron number. Strong evidence for regulation ofneurogenesis through inhibitory signals has come from studiesof the rodent olfactory epithelium. If olfactory receptor neuronsof the OE are killed, neuronal progenitors proliferate and gener-ate new ORNs3–5, a response that tissue culture studies suggestis due to loss of an ORN-derived inhibitory signal that normallysuppresses neurogenesis6. Similar ‘feedback regulation’ of neu-ron production has also been reported in larval Xenopus retina7.

The bone morphogenetic proteins (BMPs) are good candi-dates for molecules that act as negative regulators of neurogene-sis. Indeed, BMPs seem to function as inhibitory signals from theearliest stages of vertebrate neural development: during embry-onic neural induction, endogenous BMP4 promotes acquisitionof epidermal fate and suppresses neural fate in developing ecto-derm8,9. In cultures of embryonic brain, BMPs have also beenshown to inhibit proliferation10,11 and to induce apoptosis10,12

of progenitor cells.A powerful general approach for exploring molecular mech-

anisms underlying the regulation of neurogenesis is offered bytissue culture assays of mouse OE13. For example, we haveexploited such assays not only to demonstrate that fibroblastgrowth factors are positive regulators of neurogenesis, but alsoto identify the lineage stages at which they and other positive reg-ulators (such as the transcription factor MASH1) exert theireffects5,6,14,15. Because BMPs and their receptors are expressedin embryonic OE and/or olfactory placode16–19 (J.S. & A.L.C.,unpublished results), we were prompted to test BMPs for possi-ble effects on OE neurogenesis. Our findings indicate that BMPs

BMPs inhibit neurogenesis by amechanism involving degradationof a transcription factor

Jianyong Shou, Peter C. Rim and Anne L. Calof

Department of Anatomy & Neurobiology and the Developmental Biology Center, 364 Med Surge II, University of California Irvine, College of Medicine, Irvine, California 92697-1275, USA

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

Bone morphogenetic proteins (BMPs), negative regulators of neural determination in the earlyembryo, were found to be potent inhibitors of neurogenesis in olfactory epithelium (OE) cultures.BMPs 2, 4 or 7 decreased the number of proliferating progenitor cells and blocked production ofolfactory receptor neurons (ORNs). Experiments suggested that this effect was due to an action ofBMPs on an early-stage progenitor in the ORN lineage. Further analysis revealed that progenitorsexposed to BMPs rapidly (< 2 h) lost MASH1, a transcription factor known to be required for the pro-duction of ORNs. This disappearance was due to proteolysis of existing MASH1 protein, but newgene expression was required to trigger it. The data suggest a novel mechanism of BMP action,whereby the induced degradation of an essential transcription factor results in prematuretermination of a neuronal lineage.

2, 4 and 7 strongly inhibit OE neurogenesis, and that they exertthis inhibitory action on neuronal progenitor cells at a specificstage in the ORN lineage. In addition, we report a novel mecha-nism of BMP action, whereby ligand binding to target cellsinduces proteolytic degradation of a transcription factor whoseactivity is known to be required for neuronal development.

RESULTSTo assess BMP effects on neurogenesis, we first used a neuronalcolony-forming assay. When purified OE neuronal progenitorcells are cultured at clonal density on feeder layers of mitoticallyinactivated fibroblasts for over six days, colonies containing bothundifferentiated neuronal progenitors and differentiated ORNsdevelop6. Addition of BMP4 (10 ng per ml) at the time of cellplating completely blocked development of such neuronalcolonies (Table 1). Interestingly, BMP4 had no significant effecton any of the other (non-neuronal) colony types that develop inthese assays, indicating that the observed inhibitory effect ofBMP4 on neuronal colony development was not due to toxicityof the added protein. Equivalent effects were obtained withBMP2, a close homologue20 of BMP4 (Table 1). We also testedBMP7, a member of the 60A subfamily of BMPs and hence moredistantly related20 to BMP4. In embryonic brain development,BMP7 has been reported to have a stimulatory, rather than aninhibitory, effect on progenitor cell proliferation10,21. However, inour OE neuronal colony-forming assays, BMP7 was equivalentto BMPs 2 and 4 in its inhibitory effects (Table 1).

The ORN lineage is complex, with at least two stages of prog-enitor cells interposed between neuronal stem cells and post-mitotic ORNs. The neuronal colony-forming cell is hypothesizedto be the neuronal stem cell6,22. When cultured on feeder cell lay-ers, this cell can continue to produce downstream progenitorsand ORNs for up to two weeks23,24. However, in OE explant cul-

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


tures grown in defined medium, neuronalstem cells initially give rise to downstreamprogenitor cells but subsequently becomeundetectable. Stem cells may die or simplycease dividing, as factors such as fibroblastgrowth factors are known to be necessary fortheir survival and/or proliferation14. The prog-eny of the stem cell is thought to be theMASH1+ neuronal progenitor, which under-goes one to two rounds of division to give riseto MASH1-negative immediate neuronal pre-cursors5 (INPs). INPs divide one to two timesin vitro (cell cycle length, ~17 h), quantitativelygiving rise to post-mitotic ORNs14,25.

Because development of cells at each stagein the ORN developmental pathway is regu-lated5,6,14,15,26, it was important to know atwhich stage(s) BMPs exert their inhibitory effect. To address this,we used neuronal colony-forming assays with BMP4 added at dif-ferent times following cell plating, grew the cultures a total of sixdays and then counted neuronal colonies. When BMP4 was addedearly (0–2 days after plating), neuronal colonies were greatlydecreased in number (Fig. 1a). However, addition of BMP4 atthree days or later after initiation of the cultures produced no sig-nificant effect. Because early addition of BMP4 was required forinhibition of neuronal colony formation, we asked how rapidlyBMP4 acts. To accomplish this, we added BMP4 for the first 24hours in vitro; then cultures were washed and allowed to grow forthe remaining five days of the assay. Exposure of cultures to BMP4for only 24 hours following plating was sufficient to dramaticallyinhibit neuronal colony development (Fig. 1b). Thus, the effectof BMP4 on neurogenesis is very rapid. Moreover, its effect is likelyto be exerted on neuronal progenitors rather than post-mitoticORNs, because ORNs are detectable in colonies only after aboutfour days in vitro, when progenitor cells have had sufficient timeto undergo several divisions and generate neurons6. Potential tar-gets of BMP4 action thus included the neuronal colony-formingcell or the cell type thought to be its progeny, the MASH1-express-ing progenitor, or the progeny of the MASH1+ progenitor, theimmediate neuronal precursor of ORNs.

If BMPs act on neuronal progenitor cells, we reasoned that theeffects of BMPs might involve changes in progenitor cell prolif-eration. To address this we used explant cultures of OE, in whichthe proliferation kinetics of cells are easily quantified14,25. OEexplants were cultured for 20 hours in BMP4, with [3H]thymi-dine added for the last 6 hours of culture to detect cells in S phase.Addition of BMP4 dramatically decreased the number of neu-ronal cells incorporating [3H]thymidine (Fig. 2a), indicating thatBMP4 inhibits proliferation of neuronal progenitor cells.

Because late addition of BMP4 to colony-forming assays hadno effect on neuronal colony numbers in our experiments, itseemed unlikely that BMPs would have adverse effects on the sur-vival of ORNs. To test this directly, we cultured OE explants for 48hours, with BMP4 and [3H]thymidine added for the final 24hours. We focused our analysis on ORNs that had been gener-ated in the first 24 hours of the culture period; these could bedistinguished by expression of NCAM (a marker for ORNs25)and lack of [3H]thymidine labeling, indicating that they werealready post-mitotic at the time of BMP4 addition. This wasimportant, because effects of BMP4 on ORNs generated after thefirst 24 hours of culture could be secondary to its effects on prog-enitor cell proliferation (Fig. 2a). The total number of [3H]thymi-dine-negative ORNs per OE explant was virtually identical in

cultures grown in the presence or absence of BMP4 (Fig. 2band c), indicating that BMP4 had no effect on neuronal survival.

Apparently, BMPs inhibit olfactory neurogenesis not byimpairing ORN survival, but by decreasing numbers of prolifer-ating neuronal progenitor cells. Because our previous studiesshowed that expression of the transcription factor MASH1 marksan early stage of neuronal progenitor in the ORN lineage5, andstudies of a targeted mutation in the Mash1 gene showed thatMASH1 is essential for both survival of neuronal progenitors anddevelopment of ORNs in vivo22,26,27, we sought to determine ifMASH1-expressing progenitors are affected by BMPs. We cul-tured OE explants for eight hours (when the number of MASH1-expressing cells in explant cultures is high5) in the presence orabsence of BMP4. When cultures were stained with an antibodyto MASH128, MASH1 immunoreactivity was virtually abolishedby BMP4: 6.5 ± 1.3% (s.e.) of migratory cells exhibited MASH1

articles

Table 1. Inhibition of neuronal colony formation by BMPs.

Number of neuronal colonies Number of non-neuronal colonies(% control value) (% control value)

BMP4 (n = 4) 0 77.6 ± 25.7 (s.d.)BMP2 (n = 2) 0 67.6 ± 6.4 (s.d.)BMP7 (n = 2) 0 96.6 ± 20.1 (s.d.)

Neuronal colony-forming assays were done with progenitor cells plated in the absence (control) orpresence of BMP4 (10 ng per ml), BMP2 (10 ng per ml) or BMP7 (10 ng per ml) for the entire cultureperiod. Neuronal and non-neuronal colonies were counted in each condition, for a minimum of twoindependent experiments (n). Numbers of colonies were expressed as a percentage of the controlvalue for the same experiment, and percentages averaged for all experiments in a given condition (s.d.,standard deviation). No neuronal colonies were observed when BMPs were added. Actual numbers ofcolonies under control conditions: BMP4 experiments, neuronal, 26.3 ± 19.9 (s.d.); non-neuronal,71.8 ± 33.3 (s.d.). BMP2 experiments, neuronal, 15.5 ± 2.1 (s.d.); non-neuronal, 48.5 ± 7.8 (s.d.); BMP7experiments: neuronal, 57 ± 32.5 (s.d.); non-neuronal, 177 ± 50.9 (s.d.).

Fig. 1. Inhibition of neuronal colony formation requires early additionof BMP4. (a) Neuronal colony-forming assays were done as describedfor Table 1, except that BMP 4 (10 ng per ml) was added to half theplates at 0, 1, 2, 3 or 4 days after progenitor cells were plated, and waspresent for the rest of the culture period. After six days in vitro, neuronalcolonies were counted and that value expressed as a percentage of thecontrol value for the same experiment. Data are plotted as mean ±range for 2–3 independent experiments in each test condition. Asterisksindicate that no neuronal colonies ever developed under these condi-tions. (b) Neuronal colony-forming assays were done in the presence orabsence of BMP4 (10 ng per ml). After 24 h, plates were rinsed in cal-cium- and magnesium-free Hank’s balanced salt solution and re-fed withgrowth medium; BMP4 was not replenished in half of the plates to whichit had originally been added (BMP4 0–24 h). After six days in vitro, neu-ronal colonies were counted and the data expressed as described above.Error bars indicate the range of values obtained in three independentexperiments; asterisk indicates that no neuronal colonies ever devel-oped in the BMP4 0–6 days condition.

Nu

mb

er o

f n

euro

nal

colo

nie

s (%

of

con

tro

l)

a

Time of BMP4 addition(days in vitro)

b

Control BMP4 BMP40–24 h 0–6 days

0 1 2 3 4


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

nucleus; if BMPs caused MASH1 translocation into the cyto-plasm, dilution of the protein might lower the immunofluo-rescent signal below the limits of detection by eye. Control andBMP4-treated cultures were therefore processed for MASH1immunostaining in the normal manner and imaged using a dig-ital cooled CCD camera. Fluorescence over the entire cell body(not just the nucleus) was measured for individual, randomlychosen, migratory neuronal progenitor cells, and cells werebinned according to fluorescence intensity. The resulting fre-quency histogram (Fig. 4b) shows a small subpopulation ofcells with very high fluorescence intensities (> 45 units per cell)in control cultures; these cells were also judged by eye to beMASH1-immunoreactive. In BMP-treated cultures, however,

immunoreactivity in control cultures, whereas only 0.41 ± 0.19%(s.e.) of migratory cells did so in BMP4-treated cultures. To testwhether BMP4 causes MASH1+ cells to die10,12, we cultured OEexplants for eight hours with or without BMP4 (20 ng per ml)and processed them for MASH1 immunocytochemistry andDNA fragmentation in situ (TUNEL assay15) to detect apoptoticcells. Death of MASH1-expressing cells was an unlikely explana-tion for decreased MASH1 immunoreactivity in BMP4-treatedcultures, as very little apoptotic cell death occurred with or with-out BMP4: 1.97 ± 0.91% (s.e.) of migratory cells were TUNEL-positive in control cultures, and 1.92 ± 0.22% (s.e.) wereTUNEL-positive in BMP4-treated cultures. Furthermore, noMASH1+ cells were TUNEL-positive in either condition.

Further investigation showed that the effect of BMPs onMASH1 expression is very rapid. Exposing explant cultures toBMP4 for the last two hours of an eight-hour culture period wassufficient to dramatically decrease MASH1 immunoreactivity inprogenitor cells (Fig. 3a). (BMP2 and BMP7 also abolishedMASH1 immunoreactivity within two hours; data not shown.)We confirmed that effects on MASH1 expression were due specif-ically to the action of BMP4 (as opposed to, for instance, a con-taminant in the recombinant protein preparation) by showingthat the effect could be blocked by the specific BMP antagonist,noggin29(Fig. 3a). We also cultured explants for six hours andthen exposed them to BMP4 for varying periods. The numberof MASH1-immunoreactive cells was decreased by 50% after 60minutes of BMP4 treatment, and the decrease was maximal by2 hours (Fig. 3b). Photomicrographs of cultures treated for twohours with BMP4 and then immunostained to detect MASH1are shown in Fig. 4a.

The rapidity of BMP4’s effect on MASH1 suggested thatBMPs act directly on MASH1-expressing progenitor cells todecrease the levels of MASH1 protein within these cells. We didtwo types of control experiments to determine if MASH1 pro-tein levels were decreased in cultures treated for two hours withBMP4. The first was to test whether the apparent loss of MASH1immunoreactivity might be an artifact of its redistributionwithin the cell. MASH1 protein is normally concentrated in the

Fig. 2. Effects of BMP4 on OE neuronal progenitor cells and ORNs.(a) OE explants were cultured for a total of 20 h in the presence orabsence of BMP4 (10 ng per ml), with 1.5 µCi per ml [3H]thymidineadded for the final 6 h. The percentage of cells that were [3H]thymi-dine-positive ([3H]TdR+) was determined as the fraction of totalmigratory cells surrounding each explant that had > 5 silver grainsover the nucleus. Approximately 5,000 migratory cells were countedin each condition. Data are plotted as mean ± s.e. (b) OE explantswere cultured for a total of 48 h, with BMP4 (10 ng per ml) added tohalf the cultures and [3H]thymidine (0.1 µCi per ml) added to all thecultures for the last 24 h. Total cell number, the number of cellsexpressing neural cell adhesion molecule (NCAM, a marker for post-mitotic ORNs14,25), the number of [3H]thymidine-positive cells, andthe number of cells that were positive for both NCAM and[3H]thymidine were counted for each explant in both conditions; thearea of the explant body was also measured using NIH Image 1.61.Cells that were NCAM-positive but [3H]thymidine-negative wereconsidered to be ORNs that had been generated during or before thefirst 24 h of the culture period, before BMP4 addition. The number ofthese ORNs was counted for each explant in each condition; forcomparison, these values were normalized to an average explant areaof 25,000 µm2 (approximate mean explant area). Approximately4,000 migratory cells (> 95% of which were NCAM-positive ORNs) were counted in each condition. Data are plotted as the mean of values obtainedfrom two independent experiments; error bars indicate root mean square of the standard errors (c) Fluorescence photomicrographs of OE explantsfrom the experiments described in (b). Red color indicates NCAM immunostaining. Scale, 50 µm.


articles

b

c

Per

cen

t [3

H]T

dR

+ce

lls

a

Tota

l OR

Ns/

exp

ian

t

Fig. 3. BMP4 causes a rapid decrease in MASH1 immunoreactivity in OEneuronal progenitors. (a) OE explants were cultured for six hours, thenexposed to vehicle (C; culture medium), BMP4 (B, 20 ng per ml), noggin(N, 150 ng per ml, a fivefold molar excess) or BMP4 plus noggin (B + N,20 ng per ml and 150 ng per ml, respectively) for an additional 2 h (total,8 h in culture). BMP4, noggin, and the mixture were held as 10× stockson ice for 1 h before addition. The total number of migratory cells andthe number of these that were MASH1-positive were counted for eachexplant, and the percentage of MASH1-positive migratory cells wasexpressed on a per-explant basis. Data are plotted as mean ± s.e. of val-ues obtained from a minimum of 20 explants (~3,000 migratory cells) ineach condition. (b) OE explants were cultured for 6 h, then exposed toBMP4 (10 ng per ml) for 30, 60 or 120 min; control cultures in whichvehicle alone (culture medium) was added were also taken at each timepoint. Data are plotted as described in (a).

a b

Per

cen

t M

ASH

1+

cells

Exposure to BMP (min)


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


this population was absent, indicating that MASH1-immunore-active protein is indeed lost, and not simply redistributed withincells. We also assessed levels of MASH1 expression byimmunoblotting. Cultures were lysed, lysates were elec-trophoresed on SDS-polyacrylamide gels and proteins weretransferred electrophoretically to nitrocellulose membrane andprobed with an anti-MASH1 monoclonal antibody28. Theresults (Fig. 4c) indicate that addition of BMP4 to OE culturesfor two hours caused an 80% reduction in the amount ofMASH1 protein.

These findings, taken together with the rapidity of the BMPeffect (Fig. 3b), suggested that the mechanism of action of BMPsmight be at the level of MASH1 protein degradation. Indeed,cycloheximide-block experiments supported this idea by showingthat the half-life of MASH1 protein is sufficiently long (> 4 hours)in these cells that a complete cessation of Mash1 transcription ortranslation could not produce the rapid loss of MASH1 proteinobserved after treatment with BMPs. For these experiments, cul-tures were grown for six hours with cycloheximide (5 µg per ml)added to half the cultures for an additional two, four or six hours.On counting the percentage of migratory cells in each conditionthat were MASH1-positive, we found that, after 4 hours in cyclo-heximide, the number of MASH1-positive cells decreased by only20% (control, 6.40 ± 0.67%, mean ± s.e.; cycloheximide,5.16 ± 0.53%, mean ± s.e.

Because ligand-stimulated targeting of proteins for protea-some-mediated proteolytic degradation is an important meansof regulating many cellular processes30, we hypothesized that theproteasome pathway might be involved in the BMP-mediatedloss of MASH1 in OE neuronal progenitors. To test this, weassayed pharmacological inhibitors of the proteasome pathwayfor their ability to abolish BMP-induced MASH1 degradation.Lactacystin or MG132, both highly selective inhibitors that reactwith active sites of the proteasome and block proteolytic cleav-

age31,32, were used. Both inhibitors completely blocked degrada-tion of MASH1 following two-hour incubation of cultures inBMP2 (Fig. 5a). Thus, these data indicate that BMPs bind to OEneuronal progenitor cells and stimulate rapid degradation ofMASH1 protein via the proteasome pathway.

It is generally thought, however, that BMPs exert their effectson target cells at a transcriptional level, because BMP receptor acti-vation results in phosphorylation of cytoplasmic effector proteinscalled Smads that then translocate to the nucleus and initiate newgene expression33. To test whether the BMP effect on MASH1

articles

a b

Per

cen

t M

ASH

1+

cells

Fig. 5. MASH1 degradation is via proteasome-mediated proteolysis;downregulation of MASH1 immunoreactivity by BMP4 is dependent ontranscription and translation. (a) OE explants were cultured for a totalof eight hours with BMP2 (BMP, 20 ng per ml) added to half the culturesfor the final two hours. thirty minutes before BMP addition, lactacystin(L, 10 µM49), MG132 (M, 10 µM49) or control vehicle (Ctrl, 0.1% DMSOin culture medium) was added. Data were evaluated and plotted asdescribed in Fig. 3a. (b) OE explants were grown for six hours in cul-ture, then exposed to control vehicle (Ctrl, 0.05% ethanol in culturemedium), cycloheximide (CHX, 5 µg per ml) or actinomycin D (AD, 5µg per ml) for 30 minutes. BMP4 (BMP, 20 ng per ml) was then added tohalf the cultures in each condition for an additional 2 h for a total of 8.5h in vitro. Data were evaluated and plotted as described in Fig. 3a.

Fig. 4. BMP-mediated decrease in MASH1immunoreactivity is due to loss of MASH1protein. (a) Fluorescence and phase-con-trast photomicrographs of explant culturesgrown for a total of eight hours in vitro,with or without BMP4 (20 ng per ml) addedfor the final two hours. In control conditions(Ctrl), arrow indicates a cluster of migratoryneuronal progenitor cells expressingMASH1; arrowheads indicate examples ofindividual MASH1-positive cells. In BMP4(BMP4), no cells have detectable MASH1immunofluorescence. Scale, 20 µm. (b) OEexplants were grown for a total of eighthours, with half the cultures exposed toBMP4 (10 ng per ml) for the final two hours,and processed for MASH1 immunoreactivity.Fluorescence intensities of 235 randomlychosen, migratory neuronal progenitor cellsin each condition were determined. Cellswere binned according to their total fluores-cence level (expressed in integer units; x-axis), and the number of cells that fell withineach bin was graphed as the percentage of total cells imaged for a given growth condition (y-axis). (c) OE suspension cultures were prepared asdescribed15 and cultured for a total of eight hours, with or without BMP4 (20 ng per ml) for the final two hours. Sample preparation and immunoblot-ting procedures are described in Methods. Lanes 1–4 were each loaded with 120 µg of total cellular protein. MASH1 was detected as a doublet ofapproximately 33 kD. Densitometric analysis of band intensity using NIH Image 1.61 revealed a substantial reduction (80%) in the amount of MASH1-immunoreactive protein present in cultures treated with BMP4 (lanes 3 and 4) versus controls (lanes 1 and 2). 70 µg of total cellular protein fromCA77 cells (a rat thyroid C cell line expressing high levels of MASH148) was used as a positive control for MASH1 detection.

a b

c

% o

f ce

lls

Fluorescence intensity


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


degradation required the induction of new gene expression, weused pharmacological agents to inhibit RNA and protein synthesis.Either cycloheximide or actinomycin D (which by themselves hadno effect on MASH1 expression) abolished loss of MASH1 result-ing from two-hour exposure to BMP4 (Fig. 5b). Thus, although amajor consequence of BMP binding to OE neuronal progenitorcells is rapid proteolytic degradation of MASH1, new gene expres-sion is required to trigger this event.

DISCUSSIONThese results indicate that BMPs 2, 4 and 7 have a potent anti-neurogenic effect on OE cultures: BMPs markedly decrease thenumber of proliferating neuronal progenitor cells by acting onprogenitors at an early stage in the ORN lineage. A transcription-dependent yet very rapid consequence of BMP action is the pro-teolytic degradation of the essential transcription factor MASH1in early-stage progenitors.

These in vitro consequences of BMP action parallel the effectsof loss of MASH1 on the OE in vivo27 strikingly well. In micehomozygous for a targeted disruption of the Mash1 gene,extremely few ORNs arise. Cells that express transcripts from the(disrupted) Mash1 gene are found in the OE, but the OE is abnor-mally thin and exhibits an abnormally high rate of apoptosis22,26.No increase in size of any non-neuronal cell population is appar-ent in the OE26,27, suggesting that the neuronal lineage is simplyterminated by cell death at or after the Mash1-expressing stage.Consistent with this view, inhibition of apoptosis can rescue neu-rogenesis in Mash1–/– OE cells to some extent22.

In a similar manner, when OE cultures are treated with BMPs,the ORN lineage also apparently terminates. Development ofneuronal colonies is completely inhibited, but there is no increasein the numbers of any non-neuronal colony type (Table 1), sug-gesting that neuronal progenitor cells are not driven into a non-neuronal fate. As in Mash1–/– animals, there is a marked decreasein the overall number of proliferating progenitor cells (Fig. 2a).Even the cell death found in the OE of Mash1–/– animals is phe-nocopied by BMP treatment of OE cultures: although short-term(8 h) exposure to BMPs does not elicit cell death (see Results),20-h BMP4 exposure results in a substantial (57%) increase inthe number of apoptotic migratory neuronal cells in OE explantcultures (data not shown).

These similarities, taken together with the observation thatBMPs cause rapid disappearance of MASH1 from OE cultures(Fig. 3b), suggest that the destruction of MASH1 couldaccount for the anti-neurogenic effect of BMPs. Proof of thishypothesis would involve demonstrating that blockade ofMASH1 proteolysis renders OE cell resistant to inhibition ofneurogenesis by BMPs. Unfortunately, the only currently avail-able tools for blocking MASH1 proteolysis are general pro-teasome inhibitors, and proteosomes also control key elementsin cell cycle progression, such as cyclin levels30. Not surpris-ingly, we find that the lowest concentrations of MG132 andlactacystin that block BMP-induced MASH1 proteolysis alsodirectly arrest neuronal progenitor cell division in OE cul-tures, making it impossible to test for a rescue of neurogene-sis (J.S. & A.L.C., data not shown). Testing whether MASH1degradation has a causal role in the antineurogenic effect ofBMPs will most likely require other approaches, such as theidentification of mutant forms of MASH1 that fail to becometargeted for degradation by BMPs, and the introduction ofsuch mutant protein into OE progenitor cells.

Indeed, the question of how MASH1 becomes targeted forproteolysis by BMPs is an intriguing one. It is widely accepted

that BMP signaling involves the receptor-mediated phosphory-lation of Smad proteins, which translocate to the nucleus andtrigger specific gene expression33. Both cycloheximide and actin-omycin D abolished the effects of BMP4 on MASH1 expressionby OE neuronal progenitor cells (Fig. 5b), strongly suggestingthat new gene expression is required to trigger MASH1 proteol-ysis. Because degradation of MASH1 is rapid (50% complete onehour after BMP addition; Fig. 3b), it is reasonable to speculatethat a gene directly activated by BMP signaling may be the pri-mary effector of MASH1 proteolysis (for instance, a gene thatencodes an enzyme that post-translationally modifies MASH1,marking it for destruction). Differences among various cell typesin the genes that are downstream targets of BMP signaling mostlikely explain why BMPs can trigger degradation of MASH1 pro-tein in olfactory neuronal progenitor cells but can actually main-tain Mash1 expression in some neural crest cells34.

However, little is known about the genes directly activated bySmads, especially in neuronal progenitor cells, in which most ofthe known effects of BMPs (such as induction of apoptosis and/orchanges in cell fate) occur many hours or even days followingBMP exposure10,12,34–37. In several cases, it has been suggestedthat BMPs exert their effects through the induction of tran-scriptional activators, such as the homeobox-containing genesMsx-1 and Msx-2, which then elicit further changes in geneexpression and alter cell fate10,12,38. Our study indicates, how-ever, that BMP signaling can lead relatively directly to the loss oftranscriptional activators, and that this may be an importantmeans of effecting rapid changes in cell function. In this light, itis interesting that a recent study has shown a role for the post-translational inactivation (although not destruction) of the helix-loop-helix transcription factor HES-1 in mediating the actionsof nerve growth factor in neuronal differentiation39.

The present study provides the first evidence, for any cell type,that BMP signaling may be mediated via targeting of moleculesto the proteasome pathway. Recently, however, TGF-β signalingwas shown to modulate proteasome-mediated proteolysis ofRhoB40. Furthermore, it is well established that cytokines trig-ger proteasome-mediated proteolysis of cytoplasmic I-κB, whichthen releases NFκB to enter the nucleus and activate gene expres-sion41, a process essential for development42,43. Proteolytic degra-dation via the proteasome pathway appears to be a majormechanism by which β-catenin levels are regulated in the Wntsignaling pathway as well44.

One question not resolved by our work is whether BMPs aresolely or partly responsible for the feedback inhibition of neuroge-nesis in the OE that we have described in vitro and that is presumedto occur in vivo6. Current evidence suggests that this inhibition ismediated by a heat-labile macromolecule(s)23, but whether it is aBMP that has been detected in the OE (for instance, see J.S., P.C.R.& A.L.C. Soc. Neurosci. Abstr. 23, 120.7, 1997), a novel BMP or anunrelated molecule remains to be determined.

A second question is whether the mechanism underlying anti-neurogenic action of BMPs in the OE can be generalized to otherparts of the nervous system in which BMPs have been reported toinhibit cell proliferation10,11 and to stimulate apoptosis10,12. BecauseMASH1 is expressed in several regions of the developing CNS, itwill be interesting to determine whether BMPs cause MASH1degradation in neural tissues other than the OE. Unlike the OE,however, most of the CNS does not absolutely require MASH1 forneuronal development27, suggesting that MASH1 functions maybe redundantly carried out by other, related transcriptional activa-tors45. It will therefore be interesting to learn whether BMPs exerteffects on the levels of these other factors as well.

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


METHODSPurification of OE neuronal progenitor cells from E14.5–15.5 Rosa26transgenic mice (Jackson Labs) and culture methods for neuronal colony-forming assays were as described6 with two modifications: OE stromalcell feeder layers (derived from outbred CD-1 embryos) were mitoticallyinactivated by γ-irradiation (3300 rad), and calcium concentration inthe culture medium was increased to 1.4 mM. Except where noted, cul-tures were re-fed with fresh medium (with BMPs or other agents, ifappropriate) every other day, then fixed and processed for X-gal stain-ing after a total of six days. Quantitative analysis of neuronal and non-neuronal colonies was as described6. OE explant cultures using tissuepurified from E14.5–15.5 CD-1 mice (Charles River) were generated asdescribed5, and grown on polylysine/merosin-coated coverslips14 indefined, serum-free low-calcium culture medium containing 5 mg perml crystalline bovine serum albumin15 (ICN).

Explant cultures were fixed with 10% formalin in phosphate-bufferedsaline containing 5% sucrose for 15–30 minutes. Anti-MASH1 immuno-cytochemistry5 and anti-NCAM immunocytochemistry14 were done asdescribed. For cultures incubated in [3H]thymidine, coverslips were dehy-drated and dipped in NTB2 emulsion (Kodak) diluted 1:1 in water, thenexposed for 48 h (for proliferation assays) or 7 days (for ORN survivalassays) at –80oC, developed in D-19 developer, and nuclei stained withHoechst 33258 (bisbenzimide; 1 µg per ml). TUNEL staining to detect apop-totic cells was done as described15.

For quantitative analysis of immunofluorescence, cultures were processedfor MASH1 immunoreactivity and individual cells were imaged under rho-damine optics. Images for each cell (over the entire cell) were acquired asraw data files using a cooled CCD digital camera (Diagnostic InstrumentsSP100, 1315 × 1035 pixel resolution). Raw data files were imported to AdobePhotoshop 4.0 for image analysis. Fluorescence intensities of individual cellswere calculated as the sum of the pixel intensities for each cell (with back-ground pixel intensities subtracted).

For immunoblots, OE suspension cultures were collected by cen-trifugation (2000 × g, 1 min), lysed in SDS gel loading buffer46, boiledfor 10 min and centrifuged to remove debris; then urea was added to8 M. Proteins were separated on 14% SDS-PAGE and transferred tonitrocellulose membrane using standard procedures46. The membranewas incubated in undiluted A42B7 anti-MASH1 hybridoma super-natant28 overnight at 4oC, washed in phosphate-buffered saline, brieflyblocked with 1% blocking reagent (Boehringer-Mannheim) and incu-bated with affinity purified rabbit anti-mouse IgG1 (1:1,000 dilution,Dako) for 30 min at room temperature. After washing, the blot wasincubated in horseradish peroxidase-conjugated anti-rabbit antibody(1:5,000, Amersham) for one hour at room temperature. Followingextensive washes in buffer (0.1 M Tris, pH 7.5), the membrane wasdipped in ECLTM chemiluminescence substrate and exposed to Hyper-filmTM (Amersham) for 20 min at room temperature. The blot was thenstripped in 62.5 mM Tris, pH 6.8, 2% SDS, 100 mM β-mercaptoethanolfor 1 h at 55 oC, rinsed, and re-probed with a rabbit antiserum to β-tubulin47 to ensure that an equal amount of cellular protein had beenloaded on each lane (data not shown).

ACKNOWLEDGEMENTSThe authors are grateful to Youn Kim for help with experiments, and to Arthur

Lander for suggestions regarding these studies. We thank Genetics Institute for the

gift of recombinant human BMPs, Richard Harland for the gift of recombinant

Xenopus noggin, David Anderson for anti-MASH1 hybridoma and Frank

Solomon for rabbit antiserum to β-tubulin. This work was supported by a grant to

A.L.C. from the Institute on Deafness and Other Communication Disorders of the

N.I.H. (DC03583).

RECEIVED 5 NOVEMBER 1998, ACCEPTED 10 FEBRUARY 1999

1. Caviness, V. S., Takahashi, T. & Nowakowski, R. S. Numbers, time andneocortical neuronogenesis: a general developmental and evolutionary model.Trends Neurosci. 18, 379–383 (1995).

2. Temple, S. & Qian, X. bFGF, neurotrophins, and the control of corticalneurogenesis. Neuron 15, 249–252 (1995).

3. Schwartz-Levey, S., Chikaraishi, D. M. & Kauer, J. S. Characterization ofpotential precursor populations in the mouse olfactory epithelium usingimmunocytochemistry and autoradiography. J. Neurosci. 11, 3556–3564 (1991).

4. Schwob, J. E., Szumowski, K. E. M. & Stasky, A. A. Olfactory sensory neurons aretrophically dependent on the olfactory bulb for their prolonged survival. J.Neurosci. 12, 3896–3919 (1992).

5. Gordon, M. K., Mumm, J. S., Davis, R. A., Holcomb, J. D. & Calof, A. L.Dynamics of MASH1 expression in vitro and in vivo suggest a non-stem cell siteof MASH1 action in the olfactory receptor neuron lineage. Mol. Cell. Neurosci. 6,363–379 (1995).

6. Mumm, J. S., Shou, J. & Calof, A. L. Colony-forming progenitors from mouseolfactory epithelium: Evidence for feedback regulation of neuron production.Proc. Natl. Acad. Sci. USA 93, 11167–11172 (1996).

7. Reh, T. A. & Tully, T. Regulation of tyrosine hydroxylase-containing amacrinecell number in larval frog retina. Dev. Biol. 114, 463–469 (1986).

8. Hawley, S. H. B. et al. Disruption of BMP signals in embryonic Xenopusectoderm leads to direct neural induction. Genes Dev. 9, 2923-2935 (1995).

9. Wilson, P. A. & Hemmati-Brivanlou, A. Induction of epidermis and inhibitionof neural fate by Bmp-4. Nature 376, 331–336 (1995).

10. Furuta, Y., Piston, D. W. & Hogan, B. L. Bone morphogenetic proteins (BMPs)as regulators of dorsal forebrain development. Development 124, 2203-2212(1997).

11. Li, W., Cogswell, C. A. & LoTurco, J. J. Neuronal differentiation of precursors inthe neocortical ventricular zone is triggered by BMP. J. Neurosci. 18, 8853–8862(1998).

12. Graham, A., Francis-West, P., Brickell, P. & Lumsden, A. The signaling moleculeBMP4 mediates apoptosis in the rhombencephalic neural crest. Nature 372,684–686 (1994).

13. Calof, A. L., Mumm, J. S., Rim, P. C. & Shou, J. in The Neuron in Tissue Culture(ed. Haynes, L.) 23–44 (Wiley, Chichester, 1999).

14. DeHamer, M., Guevara, J., Hannon, K., Olwin, B. & Calof, A. L. Genesis ofolfactory receptor neurons: Regulation of progenitor cell divisions by fibroblastgrowth factors. Neuron 13, 1083–1097 (1994).

15. Holcomb, J. D., Mumm, J. S. & Calof, A. L. Apoptosis in the neuronal lineage ofthe mammalian olfactory epithelium: regulation in vivo and in vitro. Dev. Biol.172, 307–323 (1995).

16. Dewulf, N. et al. Distinct spatial and temporal expression patterns of two type Ireceptors for bone morphogenetic proteins during mouse embryogenesis.Endocrinology 136, 2652–2663 (1995).

17. Helder, M. N. et al. Expression pattern of Osteogenic Protein-1 (BoneMorphogenetic Protein-7) in human and mouse development. J. Histochem.Cytochem. 43, 1035–1044 (1995).

18. Zhang, D., Mehler, M. F., Song, Q. & Kessler, J. A. Development of bonemorphogenetic protein receptors in the nervous system and possible roles inregulating trkC expression. J. Neurosci. 18, 3314–3326 (1998).

19. Wu, D. K. & Oh, S.-H. Sensory organ generation in the chick inner ear. J.Neurosci. 16, 6454–6462 (1996).

20. Kingsley, D. M. The TGF-β superfamily: new members, new receptors, and newgenetic tests of function in different organisms. Genes Dev. 8, 133–146 (1994).

21. Arkell, R. & Beddington, R. S. P. BMP-7 influences pattern and growth of thedeveloping hindbrain of mouse embryos. Development 124, 1–12 (1997).

22. Calof, A. L., Mumm, J. S., Rim, P. C. & Shou, J. The neuronal stem cell of theolfactory epithelium. J. Neurobiol. 36, 190–205 (1998).

23. Calof, A. L. et al. Factors regulating neurogenesis and programmed cell death inmouse olfactory epithelium. Ann. NY Acad. Sci. 855, 226–229 (1998).

24. Pixley, S. K. CNS glial cells support in vitro survival, division, and differentiationof dissociated olfactory neuronal progenitor cells. Neuron 8, 1191–1204 (1992).

25. Calof, A. L. & Chikaraishi, D. M. Analysis of neurogenesis in a mammalianneuroepithelium: proliferation and differentiation of an olfactory neuronprecursor in vitro. Neuron 3, 115–127 (1989).

26. Cau, E., Gradwohl, G., Fode, C. & Guillemot, F. Mash1 activates a cascade ofbHLH regulators in olfactory neuron progenitors. Development 124, 1611–1621(1997).

27. Guillemot, F. et al. Mammalian achaete-scute homolog 1 is required for the earlydevelopment of olfactory and autonomic neurons. Cell 75, 463–476 (1993).

28. Lo, L. C., Johnson, J. E., Wuenschell, C. W., Saito, T. & Anderson, D. J.Mammalian achaete-scute homolog 1 is transiently expressed by spatiallyrestricted subsets of early neuroepithelial and neural crest cells. Genes Dev. 5,1524–1537 (1991).

29. Zimmerman, L. B., De Jesus-Escobar, J. M. & Harland, R. M. The Spemannorganizer signal noggin binds and inactivates bone morphogenetic protein 4.Cell 86, 599–606 (1996).

30. Weissman, A. M. Regulating protein degradation by ubiquitination.Immunol. Today 18, 189–198 (1997).

31. Rock, K. L. et al. Inhibitors of the proteasome block the degradation of mostcell proteins and the generation of peptides presented on MHC class Imolecules. Cell 78, 761–771 (1994).

32. Fenteany, G. et al. Inhibition of proteasome activities and subunit-specificamino-terminal threonine modification by lactacystin. Science 268, 726–731(1995).

33. Kretzschmar, M. & Massague, J. SMADs: mediators and regulators of TGF-βsignaling. Curr. Opin. Genet. Dev. 8, 103–111 (1998).

34. Lo, L. C., Sommer, L. & Anderson, D. J. MASH1 maintains competence forBMP2-induced neuronal differentiation in post-migratory neural crest cells.Curr. Biol. 7, 440–450 (1997).

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


35. Varley, J. E., Wehby, R. G., Rueger, D. C. & Maxwell, G. D. Number ofadrenergic and Islet-1 immunoreactive cells is increased in avian trunk neuralcrest cultures in the presence of human recombinant Osteogenic Protein-1.Develop. Dynam. 203, 434–447 (1995).

36. Gross, R. E. et al. Bone morphogenetic proteins promote astroglial lineagecommitment by mammalian subventricular zone progenitor cells. Neuron 17,595–606 (1996).

37. Shah, N. M., Groves, A. K. & Anderson, D. J. Alternative neural crest cell fatesare instructively promoted by TGFβ superfamily members. Cell 85, 331–343(1996).

38. Marazzi, G., Wang, Y. & Sassoon, D. Msx2 is a transcriptional regulator in theBMP4-mediated programmed cell death pathway. Dev. Biol. 186, 127–138(1997).

39. Ström, A., Castella, P., Rockwood, J., Wagner, J. & Caudy, M. Mediation ofNGF signaling by post-translation inhibition of HES-1, a basic helix-loop-helix repressor of neuronal differentiation. Genes Dev. 11, 3168–3181 (1997).

40. Engel, M. E., Datta, P. K. & Moses, H. L. RhoB is stabilized by transforminggrowth factor β and antagonizes transcriptional activation. J. Biol. Chem.273, 9921–9926 (1998).

41. Alkalay, I. et al. Stimulation-dependent IκBa phosphorylation marks the NF-κB inhibitor for degradation via the ubiquitin-proteasome pathway. Proc.

Natl. Acad. Sci. USA 92, 10599–10603 (1995).42. Kanegae, Y., Tavares, A. T., Belmonte, J. C. I. & Verma, I. M. Role of Rel/NF-

κB transcription factors during the outgrowth of the vertebrate limb. Nature392, 611–614 (1998).

43. Bushdid, P. B. et al. Inhibition of NF-κB activity results in disruption of theapical ectodermal ridge and aberrant limb morphogenesis. Nature 392,615–618 (1998).

44. Aberle, H., Bauer, A., Stappert, J., Kispert, A. & Kemler, R. β-catenin is atarget for the ubiquitin-proteasome pathway. EMBO J. 16, 3797–3804 (1997).

45. Lee, J. E. Basic helix-loop-helix genes in neural development. Curr. Opin.Neurobiol. 7, 13–20 (1997).

46. Sambrook, J., Fritsch, E. F. & Maniatis, T. Molecular Cloning: A LaboratoryManual. (Cold Spring Harbor Press, Cold Spring Harbor, New York, 1989).

47. Kim, S., Magendantz, M., Katz, W. & Solomon, F. Formation of the chickenerythrocyte marginal band in vivo: orgins of a cytoplasmic microtubulestructure. J. Cell Biol. 104, 51–59 (1987).

48. Clark, M. S., Lanigan, T. M., Page, N. M. & Russo, A. F. Induction of aserotonergic and neuronal phenotype in thyroid C-cells. J. Neurosci. 15,6167–6178 (1995).

49. Magae, J. et al. Transcriptional squelching by ectopic expression of E2F-1 andp53 is alleviated by proteasome inhibitors MG-132 and lactacystin. Oncogene

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


Activity-dependent synaptic modifications are crucial to the nor-mal development and function of the nervous system, but thebasic underlying mechanisms are not fully understood. Recently,it has been proposed that activity-dependent synaptic plasticitymay involve a family of secreted proteins known as neu-rotrophins1–5. Neurotrophins were initially identified based ontheir ability to promote neuronal survival and morphologicaldifferentiation6. Several lines of evidence have suggested a rolefor neurotrophins in synaptic plasticity. First, the expression ofmany neurotrophins is upregulated by electrical activity7–9, andthe secretion of neurotrophins is triggered by depolarization orsynaptic activity10–12. Second, acute application of neurotrophinsresults in potentiation of synaptic transmission at peripheral13,14

and central synapses15–18 as well as morphological changes indeveloping nerve processes19–22. Third, manipulations of the levelof neurotrophins influence activity-dependent segregation ofthalamocortical afferents in the developing visual system23,24 andinnervation of sympathetic ganglia25. Fourth, the neurotrophinBDNF is required for normal induction of long-term potentia-tion (LTP) in the CA1 region of the hippocampus26–29. Howmight neurotrophins influence activity-dependent synaptic plas-ticity? Activity can induce synthesis and secretion of neu-rotrophins, which in turn modulate synaptic structure or efficacyof synaptic transmission. To account for activity-dependentsynaptic modifications, however, it would be important to restrictthe synaptic action of secreted neurotrophins to those inputs thatare active at the time when neurotrophins are released. Restrictionmay be accomplished by a localized secretion of neurotrophin atthe site of active synapses and a limited spread of secreted neu-rotrophins30,31. Alternatively, it has been suggested that activepresynaptic terminals may be more susceptible to the modula-tory actions of secreted neurotrophins29,32,33. Such an effect ofactivity on neurons’ synaptic responsiveness to neurotrophins

has not been directly demonstrated. In the present study, weaddressed this issue by studying the effect of activity on the actionof neurotrophins at Xenopus neuromuscular synapses in culture.Our findings indicate that presynaptic depolarization in the pres-ence of BDNF greatly facilitates the potentiation of transmitterrelease induced by BDNF. These results provide direct evidencefor the activity-dependent synaptic action of neurotrophins, amechanism that may be crucial in conferring an advantage toactive inputs during refinement of developing nerve connections.

RESULTSThe synaptic efficacy of developing neuromuscular junctions wasmonitored by whole-cell recording of synaptic currents frominnervated myocytes in one-day-old Xenopus nerve-muscle cul-tures34. Evoked postsynaptic currents (EPCs) were triggered byextracellular suprathreshold stimulation of the presynaptic neu-ronal soma. Spontaneous miniature postsynaptic currents(MEPCs) were observed in the postsynaptic myocyte in theabsence of presynaptic action potentials. The frequency of MEPCswas not affected by the presence of tetrodotoxin in the medium35,indicating that they reflect spontaneous quantal release of trans-mitter. Confirming previous studies13,14, we found that acute treat-ment with exogenous BDNF at a high concentration (50 ng perml) resulted in potentiation of both spontaneous and evokedsecretion of transmitter at these developing Xenopus neuromus-cular synapses (data not shown), but had no effect on either EPCsor MEPCs at a lower concentration (10 ng per ml; Figs. 1a and2a). However, when the presynaptic neuron was depolarized by abrief (50 s) perfusion with medium containing 20 mM K+ whilethe postsynaptic myocyte was voltage-clamped at resting poten-tial (–70 mV), a marked increase in EPC amplitude was observedduring the same BDNF treatment (Fig. 1a and c). Perfusion withhigh-K+ medium alone did not result in any change in EPCs

articles

Presynaptic depolarizationfacilitates neurotrophin-inducedsynaptic potentiation

Lisa M. Boulanger and Mu-ming Poo

Department of Biology-0357, 9500 Gilman Drive, University of California at San Diego, La Jolla, California 92093-0357, USA

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

Neurotrophins have been proposed to participate in activity-dependent modifications of neuronalconnectivity and synaptic efficacy. Preferential strengthening of active inputs requires restriction ofputative neurotrophin-mediated synaptic potentiation to active synapses. Here we report thatpotentiation of synaptic efficacy by brain-derived neurotrophic factor (BDNF) is greatly facilitated bypresynaptic depolarization at developing neuromuscular synapses. Brief depolarization in thepresence of low-level BDNF results in a marked potentiation of both evoked and spontaneous synap-tic transmission, whereas exposure to either BDNF or depolarization alone is without effect. Thispotentiation depends on the relative timing of depolarization and reflects an enhancement of trans-mitter secretion from the presynaptic neuron. Thus synapses made by active inputs may beselectively strengthened by secreted neurotrophins as part of activity-dependent refinement ofdeveloping connections or of mature synapses.


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


(Fig. 1a and b). Because perfusion with high-K+ medium causesonly presynaptic depolarization when the postsynaptic myocyteis clamped at –70 mV, we tested whether the same effect can beinduced by direct electrical stimulation of the presynaptic neu-ron at the soma. Repetitive electrical stimulation (2 Hz for 7.5 s)of the presynaptic neuron in the presence of 10 ng per ml BDNFalso led to an increase in the amplitude of EPCs (Fig. 1a and d),whereas stimulation alone was ineffective (Fig. 1a and b). Thusthe presence of presynaptic activity facilitated the action of BDNFin potentiating synaptic function.

The influence of presynaptic depolarization on BDNF’s actionwas also reflected in the frequency of spontaneous MEPCs. With-out depolarization, 10 ng per ml BDNF had no effect on theMEPC frequency (Fig. 2a). Without BDNF, MEPC frequencyshowed a transient elevation immediately following perfusionwith high-K+ medium, but quickly returned to control levels fol-lowing reintroduction of normal recording medium (Fig. 2a),consistent with depolarization-induced Ca2+ elevation in thepresynaptic nerve terminal36. However, in the presence of BDNF,brief perfusion with high-K+ medium produced a marked andpersistent elevation of MEPC frequency (Fig. 2a). Similar changesin MEPC frequency were observed after depolarization by repet-itive presynaptic stimulation (Fig. 2c). In the presence of TrkB-IgG37, a recombinant scavenger protein that selectively bindsBDNF in the medium, the combined treatments of depolariza-tion and BDNF were without effect (Fig. 2b). Thus depolariza-tion-triggered synaptic potentiation by BDNF required binding,presumably to its specific membrane receptor, TrkB. Finally, wevoltage-clamped both the presynaptic neuron and the postsy-naptic myocyte at the resting membrane potential (–70 mV) dur-ing high-K+ perfusion and in the presence of BDNF. No

significant potentiation of MEPC frequency was observed(Fig. 2c), suggesting that presynaptic depolarization is necessaryfor BDNF potentiation. Presynaptic voltage-clamp efficacy at thenerve terminal was confirmed by lack of increase in MEPC fre-quency during perfusion with high K+ (Fig. 2d, compared toFig. 2a and b).

The potentiation of EPCs and MEPCs by BDNF followingbrief depolarization seems to be presynaptic in origin. First, themean and distribution of MEPC amplitudes did not changedespite a marked increase of EPC amplitude and MEPC fre-quency (Fig. 3a), nor was there any significant change in the timecourse of EPCs and MEPCs (Fig. 3a and c). If the apparent poten-tiation of MEPC frequency were the result of an increase inMEPC amplitude, which permitted detection of small events pre-viously lost in baseline noise, both the mean amplitude and shapeof the amplitude distribution would change. Thus our findingsare inconsistent with changes in properties or density of postsy-naptic transmitter receptors. Second, the fluctuation of EPCamplitudes, as reflected by the coefficient of variation (c.v.; seelegend, Fig. 3b), are consistent with a presynaptic change(Fig. 3b). Assuming that EPC amplitudes can be described by abinomial distribution, c.v. is independent of quantal size andshould remain constant if the change in EPC amplitude is dueonly to a change in postsynaptic receptor sensitivity38,39. Wefound that the ratio of squared coefficients of variation beforeand after potentiation showed a clear dependence on the extentof synaptic potentiation (Fig. 3b), which is consistent with apresynaptic increase in probability of vesicular exocytosis or innumber of available release sites38,39. Third, paired-pulse facili-tation (PPF) of the evoked responses (a phenomenon associat-ed with presynaptic changes40) was significantly reduced in

articles

Fig. 1. Effect of depolarization on BDNF-induced changesin evoked postsynaptic currents (EPCs). (a) Continuoustrace represents membrane currents recorded from aninnervated myocyte under voltage-clamp (Vc = –70 mV, fil-tered at 150 Hz). EPCs were elicited at the times marked byvertical lines. BDNF (10 ng per ml) was present in the cul-ture during the time marked by the horizontal bar. Perfusionwith high-K+ saline (20 mM, 50 s) occurred at the timemarked by ‘K+’. Samples of EPCs (averages of 7–10 events)are shown below at higher time resolution. Scale, fasttraces, 1 nA, 40 ms; slow trace, 1 nA, 90 s. Also shown aresamples of EPCs (averages of 7–10 events) from four otherexperiments, elicited at similar time points as above, before(left of arrow) or after (right of arrow) the treatment withBDNF alone (10 ng per ml), high-K+ saline alone, or repeti-tive electrical stimulation of the presynaptic neuron soma(‘STIM’) (2 Hz, 15 pulses) alone or in combination withBDNF. (b–d) Time course of changes in EPC amplitudesunder various experimental conditions. (b) Synapsestreated with BDNF (circles, n = 8), high-K + medium (trian-gles, n = 3) or repetitive electrical stimulation (squares,n = 6), respectively, at the time marked by the arrow. (c) Synapses perfused 5 min after the onset of BDNF treat-ment with high-K + medium for 50 s (n = 5). (d) Synapsesrepetitively stimulated 5 min after the onset of treatmentwith BDNF (n = 7). Each point in (b–d) represents meanEPC amplitude (± s.e.), normalized to the value during aten-minute control period (first ten min of each recording).

a

b c d

Mea

n E

PC

am

plit

ud

e (n

orm

aliz

ed)

Time (min)


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

synapses treated with a combination of depolarization and BDNF,but not in synapses treated with either alone (Fig. 3c and d).Taken together, these results support the notion that BDNF pairedwith depolarization enhanced synaptic transmission by increas-ing the level of evoked and spontaneous transmitter secretionfrom the presynaptic nerve terminal.

Effective facilitation depended on the timing of depolariza-tion relative to addition of BDNF to the culture. We found thatwhen depolarization was applied immediately before BDNF wasadded to the culture, no synaptic potentiation was induced(Fig. 4a). In contrast, depolarization coincident with or duringthe first 20 min after BDNF addition led to a marked potentia-tion (Fig. 4a). Thus depolarization must occur in the presenceof BDNF for the induction of synaptic potentiation and leavesno effect five minutes after its termination. We noted, however,that depolarization applied after the cells have been exposed toBDNF for 20 min was ineffective in facilitating synaptic poten-tiation (data not shown). This may be accounted for by desensi-tization or downregulation of TrkB receptors, mechanismsdemonstrated in other systems41,42.

An immediate and transient effect of depolarization is theopening of voltage-dependent Ca2+ channels in the presynapticneuron, raising the possibility that Ca2+ influx may be necessaryfor the facilitation of BDNF-induced synaptic potentiation by

depolarization. We addressed this question by using Ca2+-freesolution or medium containing a Ca2+ channel blocker (10 mMCo2+) during high-K+ depolarization. These experimental con-ditions were effective in preventing Ca2+ influx, as they com-pletely abolished EPCs as well as any K+-induced transientelevation of MEPC frequency (Fig. 4b). Surprisingly, neithertreatment with Co2+ nor Ca2+-free solution had any effect on thefacilitation of BDNF-induced potentiation by depolarization, sothese data are grouped together (Fig. 4b). Furthermore, perfu-sion of Ca2+-free, high-K+ solution, followed by the return ofnormal Ca2+ medium to restore synaptic transmission, inducedelevation of the EPC amplitude similar to that found after treat-ment with BDNF and depolarization in normal medium(3.2 ± 0.24 times control values at 10–40 min after the K+ treat-ment; n = 3). Thus, depolarization does not seem to act by open-ing Ca2+ channels. One intriguing possibility is that binding ofBDNF and/or subsequent signal transduction via the TrkB recep-tor, including receptor dimerization and tyrosine phosphoryla-tion, are directly influenced by membrane potential.

DISCUSSIONHere we have shown that presynaptic depolarization facilitates thesynaptic potentiation induced by BDNF. This effect was revealedby using a level of exogenous BDNF that was by itself ineffective in

Fig. 2. Effect of depolarization on BDNF-induced changes in the frequency of miniaturepostsynaptic currents (MEPCs). (a) Examples of membrane currents recorded frominnervated myocytes under three different conditions: BDNF alone (10 ng per ml), high-K+ medium alone (20 mM, 50 s) and high K+ during BDNF treatment. Samples of MEPCsare shown below at a higher time resolution. Scale, 500 pA, 2.5 min (slow traces); 500 pA,50 ms (fast traces). (b) Summary of changes in MEPC frequency with time under variousexperimental conditions: BDNF alone (squares, n = 4); high K+ alone (circles, n = 4);BDNF and high-K+ combined (solid triangles, n = 9); and BDNF and high K+ combined, inthe presence of TrkB-IgG (open triangles, n = 3). Frequency values for each synapse werenormalized to that observed during control period (first 10 min) and are presented asmean MEPC frequency per min (± s.e.). (c) Summary of effects on MEPC frequency. Barsrepresent the mean normalized MEPC frequency (± s.e.) 10–50 min after the onset of thetreatment (number of synapses examined are shown in parentheses). Stimulation or highK+ (as described above) was applied after 5 min in 10 ng per ml BDNF. Trk, TrkB-IgGthroughout experiment; VC, presynaptic voltage clamp (–70 mV) during depolarization.*Values higher than controls (p < 0.05, t-test). (d) MEPC frequency in the presence ofpresynaptic voltage clamp, BDNF and high K+. Top, sample recording of myocyte mem-brane currents before, during and after perfusion with high K+ in the presence of 10 ngper ml BDNF. Scale bar is the same as above. Graph summarizes changes in average MEPCfrequency with time from three experiments. Each point represents the mean normalizedMEPC frequency per min (± s.e.).

a

b

Time (min)

Mea

n M

EPC

fre

qu

ency

(no

rmal

ized

)

c

d

Mea

n M

EPC

fre

qu

ency

(n

orm

aliz

ed)

Time (min)

Mea

n M

EPC

fre

qu

ency

(no

rmal

ized

)K+ BDNF


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


inducing potentiation. The increase in synaptic strength dependson the timing of depolarization and seems to occur throughenhancement of transmitter secretion from the presynaptic nerveterminal. The effect of depolarization does not require Ca2+ influx,suggesting that depolarization may act directly on BDNF receptor

binding or signal transduction. These results indicate that neuronalactivity can confer rapid synaptic responsiveness to BDNF.

Studies of activity-dependent refinement of the developingnervous system have led to the hypothesis that competition forlimited amounts of target-derived factors underlies the selective

articles

Fig. 4. The importance of timing of depolarization and the role of Ca2+ influx. (a) Changes in MEPC frequency in three sets of experiments. Squares,high K+ (20 mM, 50 s) was applied 5 min before the onset of BDNF treatment (10 ng per ml, n = 9). Triangles, high K+ was applied 5 min after theonset of BDNF treatment (n = 12). Circles, only BDNF was applied to the culture (n = 4). (b) The role of Ca2+ influx: changes in MEPC frequency inthree sets of experiments. Triangles, same as in (a) Circles, high K+ was applied 5 min after the onset of BDNF treatment in the presence of Ca2+-freesaline with (n = 6) and without (n = 4) 10 mM Co2+. Squares, Ca2+-free treatment alone, with (n = 3) and without (n = 3) 10 mM Co2+. The trianglesin (a) and (b) represent the same data presented in Fig. 2b, shown here for comparison with other treatments. Each point represents mean normal-ized MEPC frequency (± s.e.).

Fig. 3. Analyses of MEPCs and EPCs. (a) Amplitude distributions of MEPCsbefore and after treatment with low-levelBDNF (10 ng per ml) and high K+ (20 mM,50 s). Cumulative probability refers to theprobability of observing events with ampli-tudes smaller than or equal to a given value.Data were obtained from 10 synapsesbefore (circles) and 10–50 min after (trian-gles) exposure to the combined treatmentof high-K+ and BDNF. Inset, superimposedaverage traces of MEPCs before and aftertreatment. Dashed line depicts MEPCs(average of 178 events) for a typical synapse,10–50 min after the treatment. The super-imposed solid line depicts MEPCs (averageof 38 events) during a control period (first10 min of recording). (b) Analysis of ampli-tude fluctuation of EPCs. The ratio of coeffi-cients of variation squared [c.v.2 =variance/(mean)2] before and after synapticpotentiation (c.v.b2/c.v.a2) was plottedagainst the extent of synaptic potentiationinduced by combined high-K+ and BDNFtreatment from six experiments.Potentiation factor is defined as the meanEPC amplitude 10–50 min after treatmentdivided by mean EPC amplitude before thetreatment. Only synapses with more than35 EPC events recorded were used for thisanalysis. The horizontal dotted line represents prediction from binomial statistics for postsynaptic changes, and the diagonal line represents the bestlinear fit of the data. (c, d) Paired-pulse facilitation of evoked responses. Shown in (c) are samples of EPCs (averages of 40 events) in response topresynaptic paired-pulse stimuli (25 ms interval) before (dotted trace) and 10–50 min after (solid trace) exposure to a combined treatment of high K+

and BDNF. Lines connecting peaks of each pair are for comparison of the degree of facilitation. Scale, 0.33 nA, 6 ms. In (d), ratio of average paired-pulse facilitation (after treatment/before treatment) is shown for synapses treated with BDNF, high K+ or stimulation alone or in combination. Foreach cell, PPF was normalized to initial PPF during a ten-minute control period; a value of less than one represents a decrease in PPF. *Treatments forwhich PPF changed significantly after application (p < 0.05, t-test). All error bars represent s.e.

a b

(CV

b/C

Va)

2

MEPC amplitude (normalized) Potentiation factor

c d

Time (min)

a

Mea

n M

EPC

fre

qu

ency

(no

rmal

ized

)

b

Mea

n M

EPC

fre

qu

ency

(no

rmal

ized

)

Time (min)

Cu

mu

lati

ve p

rob

abili

ty

PP

F a/P

PF b


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

stabilization and elimination of nerve connections43. Neu-rotrophins are candidates for such target-derived factors3. Theexpression of NGF and BDNF in central neurons is regulated byneuronal activity7,8, and the secretion of both of these factors canbe triggered by membrane depolarization10,11. In rat skeletal mus-cle, activity elevates the production of NT-4 (ref. 9). In nerve-muscle co-cultures, repetitive activity triggers the release of NT-4from postsynaptic muscle cells, resulting in potentiation of synap-tic responses12. All of these results implicate activity-dependentavailability of neurotrophins in mediating synaptic modifications.However, they do not suggest any obvious mechanism for selectivestabilization or elimination of specific synaptic inputs. In princi-ple, activity may confer an advantage to a synaptic input if theaction of the target-derived neurotrophin is regulated by activi-ty. Depolarization promotes cortical dendritic growth33 and reti-nal ganglion cell survival44 induced by neurotrophin as well asenhancing membrane insertion of neuronal TrkB receptors45.Activity could also promote the response of the presynaptic neu-ron to neurotrophins by regulating neurotrophin binding or sig-nal transduction. Voltage-dependent ligand binding has beendemonstrated for membrane receptors associated with ion chan-nels such as dihydropyridine receptors46. It is possible that, beinga transmembrane protein, TrkB also exhibits voltage-dependentligand binding. Voltage-dependent changes in the conformationof neurotrophin-receptor complexes might affect the efficiencyof receptor dimerization and autophosphorylation, although allthese possibilities have yet to be examined. In addition, depolar-ization triggers neuronal membrane recycling through increasedendocytosis. Endocytotic uptake of neurotrophin–receptor com-plexes, which may be an integral part of intracellular signaling47,could also be enhanced by depolarization.

Low levels of two neurotrophic factors, BDNF and ciliary neu-rotrophic factor (CNTF), can act in a synergistic manner toenhance synaptic efficacy at developing neuromuscular synaps-es14. It is possible that presynaptic depolarization triggers secre-tion of endogenous neurotrophin that acts synergistically withapplied BDNF to potentiate the synapse. This secreted neu-rotrophin could then act in an autocrine or paracrine fashion tosupplement low local levels of neurotrophin past the thresholdfor synaptic potentiation. Endogenous secreted neurotrophinlevels are likely to be lower than those used in vitro. Our findingsdemonstrate that a previously ineffective level of neurotrophinmay enhance synaptic transmission in the presence of brief presy-naptic depolarization. This result underscores the notion thatbiological actions of neurotrophins must be considered withinthe context of other coincident signals received by the neuron.Consideration of coincident signals is also relevant for the actionsof BDNF on neuronal growth and survival, as shown by the mod-ulatory effect of cytosolic cAMP44,48. Interestingly, cAMP levelscan be regulated by the presence of neuronal depolarizationthrough voltage-sensitive (but Ca2+-independent) adenylyl cyclaseactivity, found in several neuronal systems49.

In conclusion, we have demonstrated that presynaptic depo-larization is a critical factor in determining the synaptic actionof neurotrophins. This dependence on presynaptic activity pro-vides a natural mechanism for selective stabilization of activenerve terminals during synaptic refinement in the developingnervous system and adds a new dimension to the role of neu-rotrophins in synaptic plasticity.

METHODSCell culture. Preparation of Xenopus nerve/muscle cultures followeddescribed methods 34. The cells were plated on clean glass coverslips and

were used for experiments after 24 h incubation at room temperature(20–22°C). The culture medium consisted of (v/v) 50% Leibovitz’s medi-um (L-15, Gibco), 1% fetal bovine serum (Gibco) and 49% Ringer’s solu-tion (115 mM NaCl, 2 mM CaCl2, 2.5 mM KCl, 10 mM HEPES, pH 7.3).In zero-Ca2+ medium, 2 mM MgCl2 or 2 mM CoCl2 was substituted forCaCl2; in the first instance, 10 mM NaCl was omitted from the Ringer’ssolution and replaced with 10 mM EGTA.

Electrophysiology. Synaptic currents were recorded from singly inner-vated individual spherical muscle cells by the whole-cell recordingmethod50 using a patch-clamp amplifier (Axopatch 1D, Axon Instru-ments). The solution inside the recording pipette contained 150 mMKCl, 1 mM NaCl, 1 mM MgCl2 and 10 mM HEPES (pH 7.2). Record-ings were made at room temperature in culture medium. A patch elec-trode at the cell body under loose-seal conditions was used forextracellular stimulation of the presynaptic neuron. Recordings in whichthere was apparent cell damage or substantial changes (>20%) in theseries or input resistance were discarded. All other recordings wereincluded for analysis in this work. The typical series resistance was 10MΩ, and the input resistance of the postsynaptic myocyte was in therange of 50–200 MΩ. The estimated error in voltage clamp of themyocyte was therefore about 3–12 mV when Vc = –70 mV. The record-ed membrane currents were filtered at 10 kHz and stored by a videotaperecorder for later playback onto a storage oscilloscope (2201, Tektron-ix) or an oscillographic recorder (RS3200, Gould) and for analysis bycomputer. Evoked currents were distinguished from spontaneous MEPCsby the presence of a clear stimulation artifact immediately precedingEPCs. Synaptic currents were analyzed using the SCAN program (pro-vided by J. Dempster, Univ. of Strathclyde, Glasgow).

ACKNOWLEDGEMENTSWe thank Genentech Inc. for providing Trk-IgG fusion protein and Benedikt

Berninger and Alejandro Schinder for discussion of the manuscript. This work

was supported by grants from NIH.

RECEIVED 23 SEPTEMBER 1998, ACCEPTED 8 JANUARY 1999

1. Thoenen, H. Neurotrophins and neuronal plasticity. Science 270, 593–598(1995).

2. Lo, D. C. Neurotrophic factors and synaptic plasticity. Neuron 15, 979–981(1995).

3. Katz, L. C. & Shatz, C. J. Synaptic activity and the construction of corticalcircuits. Science 274, 1133–1138 (1996).

4. Bonhoeffer, T. Neurotrophins and activity-dependent development of thecortex. Curr. Opin. Neurobiol. 6, 119–126 (1996).

5. Berninger, B. & Poo, M.-M. Fast actions of neurotrophic factors. Curr. Opin.Neurobiol. 6, 324–330 (1996).

6. Lewin, G. R. & Barde, Y.-A. Physiology of the neurotrophins. Annu. Rev.Neurosci. 19, 289–317 (1996).

7. Gall, C. M. & Isackson, P. J. Limbic seizures increase neuronal production ofmessenger RNA for nerve growth factor. Science 245, 758–761 (1989).

8. Lu, B., Yokoyama, M., Dreyfuss, C. F. & Black, I. B. Depolarizing stimuliregulate nerve growth factor gene expression in cultured hippocampalneurons. Proc. Natl. Acad. Sci. USA 88, 6289–6292 (1991).

9. Funakoshi, H. et al. Muscle-derived neurotrophin-4 as an activity-dependenttrophic signal for adult motor neurons. Science 268, 1495–1499 (1995).

10. Blöchl, A. & Thoenen, H. Characterization of nerve growth factor (NGF)release from hippocampal neurons: evidence for a constitutive and anunconventional sodium-dependent regulated pathway. Eur. J. Neurosci. 7,1220–1228 (1995).

11. Goodman, L. J. et al. Regulated release and polarized localization of brain-derived neurotrophic factor in hippocampal neurons. Mol. Cell. Neurosci. 7,222–238 (1996).

12. Wang, X.-H. & Poo, M.-M. Potentiation of developing synapses bypostsynaptic release of neurotrophin-4. Neuron 19, 825–835 (1997).

13. Lohof, A. M., Ip, N. Y. & Poo, M.-M. Potentiation of developingneuromuscular synapses by the neurotrophins NT-3 and BDNF. Nature 363,350–353 (1993).

14. Stoop, R. & Poo, M.-M. Synaptic modulation by neurotrophic factors:differential and synergistic effects of brain-derived neurotrophic factor andciliary neurotrophic factor. J. Neurosci. 16, 3256–3264 (1996).

15. Lessmann, V., Gottmann, K. & Heumann, R. BDNF and NT-4/5 enhanceglutamatergic synaptic transmission in cultured hippocampal neurones.Neuroreport 6, 21–25 (1994).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

16. Kim, H. G., Wang, T., Olafsson, P. & Lu, B. Neurotrophin 3 potentiatesneuronal activity and inhibits gamma-aminobutyratergic synaptictransmission in cortical neurons. Proc. Natl. Acad. Sci. U S A 91, 12341–12345(1994).

17. Kang, H. & Schuman, E. M. Long-lasting neurotrophin-inducedenhancement of synaptic transmission in the adult hippocampus. Science267, 1658–1662 (1995).

18. Levine, E. S., Dreyfus, C. F., Black, I. B. & Plummer, M. R. Brain-derivedneurotrophic factor rapidly enhances synaptic transmission in hippocampalneurons via postsynaptic tyrosine kinase receptors. Proc. Natl. Acad. Sci.U S A 92, 8074–8077 (1995).

19. Connolly, J. L., Seeley, P. J. & Greene, L. A. Regulation of growth conemorphology by nerve growth factor: a comparative study by scanningelectron microscopy. J. Neurosci. Res. 13, 183–198 (1985).

20. McAllister, A. K., Lo D. C. & Katz L. C. Neurotrophins regulate dendriticgrowth in developing visual cortex. Neuron 15, 791–803 (1995).

21. Cohen-Cory, S. & Fraser, S. E. Effects of brain-derived neurotrophic factor onoptic axon branching and remodeling in vitro. Nature 378, 192–196 (1995).

22. Ming, G.-L., Lohof, A. & Zheng, J. Acute morphogenic and chemotropiceffects of neurotrophins on cultured embryonic Xenopus spinal neurons. J.Neurosci, 17, 7860–7871 (1997).

23. Maffei, L., Berardi, N., Domenici, L., Parisi, V. & Pizzorusso, T. Nerve growthfactor (NGF) prevents the shift in ocular dominance distribution of visualcortical neurons in monocularly deprived rats. J. Neurosci. 12, 4651–4662(1992).

24. 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).

25. Causing, C. G. et al. Synaptic innervation density is regulated by neuron-derived BDNF. Neuron 18, 257–267 (1997).

26. Figurov, A. et al. Regulation of synaptic responses to high-frequencystimulation and LTP by neurotrophins in the hippocampus. Nature 381,706–709 (1996).

27. Korte, M. et al. Virus-mediated gene transfer into hippocampal CA1 regionrestores long-term potentiation in brain-derived neurotrophic factor mutantmice. Proc. Natl. Acad. Sci. USA 93, 12547–12552 (1996).

28. Kang, H., Welcher, A. A., Shelton, D. & Schuman, E. M. Neurotrophins andtime: different roles for TrkB signaling in hippocampal long-termpotentiation. Neuron 19, 653–664 (1997).

29. Gottschalk, W., Pozzo-Miller, L. D., Figurov, A. & Lu, B. Presynapticmodulation of synaptic transmission and plasticity by brain-derivedneurotrophic factor in the developing hippocampus. J. Neurosci. 18,6830–6839 (1998).

30. Wang, X.-H., Berninger, B. & Poo, M.-M. Localized synaptic action ofneurotrophin-4. J. Neurosci. 18, 4985–4992 (1998).

31. Biffo, S., Offenhauser, N., Carter, B. D. & Barde, Y.-A. Selective binding andinternalization by truncated receptors restrict the availability of BDNFduring development. Development 121, 2461–2470 (1995).

32. Birren, S. J., Verdi, J. M. & Anderson, D. J. Membrane depolarization inducesp140trk and NGF responsiveness, but not p75 LNGFR, in MAH cells. Science257, 395–397 (1992).

33. McAllister, A. K., Katz, L. C. & Lo, D. C. Neurotrophin regulation of corticaldendritic growth requires activity. Neuron 17, 1057–1064 (1996).

34. Tabti, N., Alder, J. & Poo, M.-M. in Culturing Nerve Cells, 2nd edn. (eds.Banker, G. & Goslin, K.) 237–260 (MIT Press, Cambridge, Massachusetts,1998).

35. Song, H. J. et al. Expression of a putative vesicular acetylcholine transporterfacilitates quantal transmitter packaging. Neuron 18, 815–826 (1997).

36. Katz, B. The transmission of impulses from nerve to muscle, and thesubcellular unit of synaptic action. Proc. R. Soc. Lond. B 155, 455–479 (1962).

37. Shelton, D. L. et al. Human trks: molecular cloning, tissue distribution, andexpression of extracellular domain immunoadhesins. J. Neurosci. 15, 477–491(1995).

38. Bekkers, J. M. & Stevens, C. F. Presynaptic model for long-term potentiationin hippocampus. Nature 346, 724–729 (1990).

39. Malinow, R. & Tsien, R. W. Presynaptic enhancement shown by whole-cellrecordings of long-term potentiation in hippocampal slices. Nature 346,177–180 (1990).

40. Kamiya, H. & Zucker, R. S. Residual Ca2+ and short-term synaptic plasticity.Nature 371, 603–606 (1994).

41. Carter, B. D., Zirrgiebel, U. & Barde, Y.-A. Differential regulation of p21ras

activation in neurons by nerve growth factor and brain-derived neurotrophicfactor. J. Biol. Chem. 270, 21751–21757 (1995).

42. Frank, L., Ventimiglia, R., Anderson, K., Lindsay, R. M. & Rudge, J. S. BDNFdown-regulates neurotrophin responsiveness, TrkB protein and TrkB mRNAlevels in cultured rat hippocampal neurons. Eur. J. Neurosci. 8, 1220–1230 (1996).

43. Purves, D., Snider, W. D. & Voyvodic, J. T. Trophic regulation of nerve cellmorphology and innervation in the autonomic nervous system. Nature 336,123–128 (1988).

44. Meyer-Franke, A., Kaplan, M., Pfrieger, F. W. & Barres, B. A. Characterizationof the signaling interactions that promote the survival and growth ofdeveloping retinal ganglion cells in culture. Neuron 15, 805–819 (1995).

45. Meyer-Franke, A. et al. Depolarization and cAMP elevation rapidly recruitTrkB to the plasma membrane of CNS neurons. Neuron 21, 681–693 (1998).

46. McCleskey, E. W. & Almers, W. Dihydropyridine receptors in muscle arevoltage-dependent but most are not functional calcium channels. Nature 314,747–751 (1985).

47. Thoenen, H. & Barde, Y.-A. Physiology of nerve growth factor. Physiol. Rev.60, 1284–1335 (1980).

48. Song, H.-J., Ming, G.-L. & Poo, M.-M. cAMP-induced switching in theturning direction of nerve growth cones. Nature 388, 275–278 (1997).

49. Reddy, R. et al. Voltage-sensitive adenylyl cyclase activity in cultured neurons:a calcium-independent phenomenon. J. Cell Biol. 24, 14340–14346 (1995).

50. Hamill, O. P., Marty, A., Neher, E., Sakmann, B. & Sigworth, F. J. Improvedpatch-clamp techniques for high-resolution current recording from cells andcell-free membrane patches. Pflügers Arch. 391, 85–100 (1981).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

NATURE NEUROSCIENCE

CO R R I G E N D U M

Corrigendum: Presynaptic depolarization facilitates neurotrophin-induced synaptic potentiationLisa Boulanger and Mu-ming PooNat. Neurosci. 2, 346–351 (1999); corrected after print 10 January 2008

In the version of this article initially published, the first author’s middle initial was omitted. The correct name should be Lisa M. Boulanger. The error has been corrected in the PDF version of the article.

©20

08 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


articles

The neuronal response to glutamate released at excitatory syn-apses depends on the complement of glutamate receptors in thepostsynaptic membrane. Changes in receptor number, type andmolecular composition can substantially alter the properties ofsynaptic transmission. During postnatal development, excitato-ry synaptic transmission is readily modified by sensory experi-ence. Thus, it has been suggested that some forms ofexperience-dependent synaptic plasticity are associated with rapidchanges in the complement of postsynaptic glutamate receptors.Most attention has focused on changes in the synaptic expres-sion of AMPA receptors. However, NMDA receptors (NMDARs)are critical in triggering experience-dependent synaptic modifi-cations, and changes in synaptic expression of these receptorscould have a large effect on the properties of synaptic plasticityduring development. In the present study, we addressed the pos-sibility that experience rapidly regulates the synaptic expressionof NMDARs in visual cortex in vivo.

NMDARs in vivo are heteromeric ion channels composed ofNR1 and NR2 subunits. The subtype (A–D) of the NR2 subunitconfers distinct functional properties to the receptor1–7. Receptorscontaining NR2B predominate in the neonatal forebrain, andover the course of development, these are replaced or supple-mented with NR2A-containing receptors2,8. This subunit switchalters the channel properties such that the synaptic NMDAR-mediated currents shorten in duration2,9. The developmentalshortening of NMDAR currents in visual cortical neurons is post-poned when animals are deprived of vision10, suggesting that theNR2A/B subunit composition of synaptic NMDARs differsbetween dark-reared animals and light-reared controls. We test-ed this hypothesis and then examined the effects of brief lightexposure in dark-reared animals. Our results show that new

Rapid, experience-dependentexpression of synaptic NMDAreceptors in visual cortex in vivo

Elizabeth M. Quinlan1, Benjamin D. Philpot1, Richard L. Huganir2 and Mark F. Bear1

1 Howard Hughes Medical Institute, Department of Neuroscience, Brown University, Providence, Rhode Island 02912, USA2 Howard Hughes Medical Institute, Department of Neuroscience, The Johns Hopkins University School of Medicine,

Baltimore, Maryland 21205, USA

The first two authors contributed equally to this paper.

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

Sensory experience is crucial in the refinement of synaptic connections in the brain duringdevelopment. It has been suggested that some forms of experience-dependent synaptic plasticity in vivo are associated with changes in the complement of postsynaptic glutamate receptors,although direct evidence has been lacking. Here we show that visual experience triggers the rapidsynaptic insertion of new NMDA receptors in visual cortex. The new receptors have a higher propor-tion of NR2A subunits and, as a consequence, different functional properties. This effect ofexperience requires NMDA receptor activation and protein synthesis. Thus, rapid regulation of post-synaptic glutamate receptors is one mechanism for developmental plasticity in the brain. Changes inNMDA receptor expression provide a mechanism by which brief sensory experience can regulate theproperties of NMDA receptor-dependent plasticity in visual cortex.

NMDARs with a higher NR2A/B ratio are inserted into synap-tic membrane within one hour of the onset of visual experience.

RESULTSTo examine NMDAR subunit composition, we used quantitativeimmunoblotting for NR2A, NR2B and NR1 in synaptoneuro-somes prepared from the visual cortices of postnatal day (P)21–23 light-reared and dark-reared Long-Evans rats. Synap-toneurosomes are a biochemical fraction that is enriched forsynaptic proteins11. This biochemical analysis showed that synap-toneurosomal NR2A protein levels were significantly higher inlight-reared visual cortex than in dark-reared cortex (mean opti-cal density (OD) ± s.e.m., light reared, 1077 ± 224; dark reared,743 ± 152, paired t-test, p < 0.005, n = 6). In contrast, levels ofNR2B and NR1 proteins were not affected by dark rearing (NR2Blight reared, 1907 ± 175; dark reared, 1777 ± 133, paired t-test,p > 0.05, n = 6; NR1 light reared, 1656 ± 202; dark reared,1656 ± 222, paired t-test, p > 0.1, n = 8). Thus, as predicted, alower NR2A/B ratio (Fig. 1) is correlated with slower NMDARcurrent kinetics10 in dark-reared visual cortex. The effect of darkrearing in visual cortex seems to be a specific consequence of sen-sory deprivation, as there were no detectable differences inNMDAR protein levels in the hippocampi of dark-reared or light-reared animals (NR2A light reared, 1009 ± 184 ; dark reared,1067 ± 161, paired t-test, p > 0.1, n = 6; NR2B light reared,873 ± 225; dark reared, 838 ± 221, paired t-test, p > 0.1, n = 6). Toexamine the effects of visual experience on synaptoneurosomalNMDARs, we exposed dark-reared animals to a normal 12:12light:dark cycle for 24 or 48 hours. These experiments revealedthat the reduction in synaptoneurosomal NR2A content in dark-reared animals could be completely reversed within one day of


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

light exposure (Fig. 1). Again, there were no significant effects ofvisual experience on the levels of NR2B or NR1.

These biochemical data show that the ratio of NR2A/B proteinat the synapse is affected by light deprivation and subsequentvisual experience. However, it was critical to determine whetherthese differences in protein reflect differences in the compositionof functional NMDARs at visual cortical synapses. To addressthis question, we examined the ifenprodil sensitivity of synapti-cally evoked, NMDAR-mediated field potentials in slices of visu-al cortex from P21–28 rats. Ifenprodil selectively blocksNR2B-containing NMDARs12, so enhanced ifenprodil sensitiv-

ity indicates a lower NR2A/B ratio in the synaptic NMDARs. TheNMDAR component of layer 2/3 field potentials was pharmaco-logically isolated in artificial cerebrospinal fluid (ACSF) con-taining reduced Mg2+ and blockers of AMPA and GABAAreceptors (Fig. 2). Once a stable baseline was obtained, we applied3 µM ifenprodil, a concentration that selectively blocks NR2B-containing NMDARs12–14. The percent inhibition by ifenprodilwas calculated by comparing the baseline data with those col-lected 90 minutes after the onset of drug application. In our ini-tial experiments, slices from dark-reared and light-reared animalswere studied with the experimenter ‘blind’ to the rearing condi-

Fig. 1. Visual experience regulates the composi-tion of NMDARs in synaptoneurosomes fromvisual cortex. Quantitative immunoblotting forNMDAR subunit proteins of synaptoneurosomesprepared from visual cortices of P21–23 ratsraised in a normal (12:12) light:dark cycle (LR), incomplete darkness (DR) or in darkness frombirth followed by exposure to the normal lightcycle for 24 (+24) or 48 (+48) h. (a) NR2A pro-tein is reduced in visual cortex synaptoneuro-somes from dark-reared animals, and thisreduction is reversed by subsequent light expo-sure (one-way ANOVA. p = 0.004, *significant difference versus all other groups in Tukey HSD post-hoc comparison, p < 0.05). The O.D. from each sample was normalized to the light-reared control runon the same gel, and the summarized data are presented as percent of LR values (mean ± s.e.m.).Inset, representative immunoblot for NR2A protein in synaptoneurosomes from LR, DR, DR + 24and DR + 48 visual cortex. (b) Visual experience does not alter the complement of NR2B protein insynaptoneurosomes from visual cortex (one-way ANOVA, p = 0.930). The same synaptoneurosomeswere probed as in (a). Inset, representative immunoblot for NR2B protein in synaptoneurosomesfrom LR, DR, DR + 24 and DR + 48 visual cortex. (c) Visual experience does not alter the comple-ment of NR1 protein in synaptoneurosomes from visual cortex (one-way ANOVA, p = 0.176).Immunoblotting for NR1 protein was done simultaneously with the immunoblotting for NR2A pro-tein depicted in (a). Inset, representative immunoblot for NR1 protein in synaptoneurosomes fromLR, DR, DR + 24 and DR + 48 visual cortex.

50

75

100

125

150

% o

f lig

ht-

rear

ed

NR2A

A

LR DR +24 +48

*

a

Per

cen

t o

f lig

ht-

rear

ed

Fig. 2. The ifenprodil sensitivity of NMDAR-mediated field potentials in visual cortex isexperience dependent. (a) The preparation.Slices of visual cortex were prepared fromP21–28 rats, and synaptic field potentials wererecorded in layers 2/3 in response to layer-4stimulation. (b) Experimental design.NMDAR-mediated synaptic field potentialswere isolated pharmacologically. To determinethe contribution of NR2B-containingNMDARs to this response, we applied 3 µMifenprodil. The remaining synaptic field poten-tial was completely blocked by 100 µM AP5(DL-2-amino-5-phosphonovaleric acid), con-firming that the response reflects NMDAR-mediated currents. (The residual AP5-insensitive negativity is non-synaptic and is notaffected by the rearing history of the animal.)This representative example was recordedfrom a slice taken from a P26 dark-reared rat.The field potentials shown are averages of fourconsecutive responses taken at the indicatedtimes. (Trace 1 is the AMPAR-dominated field potential recorded before isolation of the NMDAR component.) Throughout the experiment, therewas no change in the non-synaptic component of the evoked response. Scale bar, 0.2 mV and 5 ms. Dashed line, 30-minute average of the NMDAR-mediated field potential amplitude before ifenprodil application. (c) NMDAR-mediated responses in dark-reared rats are more sensitive to ifenprodilthan light-reared controls, and the effect of dark rearing is reversed by 24, 48 or 96 h in a normal light cycle. Data are mean (± s.e.m.) reduction after90-minute ifenprodil application relative to the 30-minute baseline average. *Significant difference versus all other groups in Tukey HSD post-hoc com-parison, p < 0.05. (d) Partial blockade of NMDARs with AP5, which does not distinguish between NMDAR subtypes, reveals no difference betweendark-reared and light-reared groups.

0

0.1

0.2

0.3

0.4

0.5

0.6

-40 0 40 80 120 160Time from ifenprodil application (min)

FP a

mp

litud

e (m

V)

2

43

1+2 2+3 3+4

3 µM ifenprodil100 µM AP5

B

SIV

WM

II/III

A

0

10

20

30

40

50

LR DR +24 +48 +96

*

0

0.2

0.4

0.6

0.8

1.0

1.2

-40 -20 0 20 40 60 80 100

Time from AP5 application (min)

LRDR

1 µM AP5

a

b

c

d

FP a

mp

litu

de

(mV

)

Per

cen

t re

du

ctio

n b

y if

enp

rod

ilN

orm

aliz

ed N

MD

A F

P

Time from ifenprodil application (min) Time from AP5 application (min)

% o

f lig

ht-

rear

ed

50

75

100

125

150NR2B

B

LR DR +24 +48

% o

f lig

ht-

rear

ed

50

75

100

125

150NR1

C

LR DR +24 +48

b

c

Per

cen

t o

f lig

ht-

rear

edP

erce

nt

of

ligh

t-re

ared


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


tions. This experiment revealed a signifi-cant increase in the ifenprodil sensitivity ofNMDAR-mediated responses in dark-reared visual cortex (n = 12 slices from 7rats) as compared to light-reared controls(n = 15 slices from 7 rats; t-test, p < 0.05).To confirm the selectivity of this effect, werepeated the experiment using a subsatu-rating concentration of AP5 (1 µM) thatproduces a partial block of NMDARresponses, but does not distinguish amongthe NMDAR subunits. Unlike ifenprodil,the AP5-sensitivity of NMDAR responseswas comparable in the dark-reared (n = 7slices from 4 rats) and light-reared (n = 8slices from 4 rats) groups (t-test, p = 0.49).There were also no significant differencesin the absolute magnitude of the AMPA-receptor-dominated or NMDAR-mediat-ed field potentials between light-reared anddark-reared groups, or in the stimulationintensity required to elicit 80% of the max-imal response.

To test whether visual experience couldreverse the effects of dark rearing, weexamined ifenprodil sensitivity in a secondseries of slices taken from animals that werelight reared (n = 27 slices from 13 rats),dark reared (n = 26 slices from 13 rats,including those used for the ‘blind’ portionof the study) or dark reared and thenexposed to 24 (n = 14 slices from 8 rats),48 (n = 15 slices from 7 rats) or 96 hours(n = 10 slices from 5 rats) of the normallight cycle (Fig. 2). Exposing dark-rearedrats to a regular light cycle was sufficient torestore the light-reared phenotype (one-way ANOVA, F4,91 = 4.85, p = 0.001). Thedata demonstrate that 24 or more hours ofvisual experience is sufficient to abolish theeffect of dark rearing.

Both the biochemical and the electro-physiological results show that the synaptic NMDARs have alower NR2A/B ratio in dark-reared visual cortex than in light-reared cortex. These findings were not entirely unexpected, asthey are consistent with previous reports on the effect of lightdeprivation on the development of NMDAR response kineticsand ifenprodil sensitivity in the visual pathway10,15. However, itwas a surprise that NMDARs in dark-reared cortex completelyrecovered the light-reared phenotype with one day of light expo-sure. Thus, our next step was to explore the minimum visualexperience required to change NMDAR composition. Dark-reared animals were exposed to 0.5, 1, 1.5 or 2 hours of a normallighted environment, and changes in NR2A, NR2B and NR1 pro-tein were measured in visual cortical synaptoneurosomes. Theseexperiments showed that light exposure for as little as one hourinduced a significant increase in NR2A protein (Fig. 3a and b).Again, the levels of NR2B and NR1 were not significantly changedby light exposure. Thus, within two hours of light exposure, theNR2A/B ratio reached the light-reared value (Fig. 3c).

To address the question of whether the rapid change in synap-toneurosomal protein reflects an alteration in the compositionof functional synaptic NMDARs, we examined the effects of brief

light exposure on the ifenprodil sensitivity of NMDAR-mediatedresponses (Fig. 3d). Two hours of visual experience (n = 10 slicesfrom 6 rats) was sufficient to produce a significant reduction inifenprodil sensitivity compared to dark-reared controls (n = 6slices from 5 rats; t-test, p < 0.05), consistent with a change inthe NR2A/B ratio. Taken together, the data show that two hoursof visual experience is sufficient to significantly alter the molec-ular composition (and therefore function) of synaptic NMDARsin visual cortex. To our knowledge, this is the first demonstra-tion that sensory experience (versus deprivation) can induce achange in the complement of postsynaptic glutamate receptorsin vivo.

Experience-dependent modifications of synaptic responsesin visual cortex have been shown to require NMDAR activa-tion16,17. We therefore were interested to know whether the expe-rience-induced regulation of NMDAR composition was itselfdependent on NMDAR activation in the visual pathway. Toaddress this question, we injected dark-reared animals in the darkwith the competitive NMDAR antagonist CPP (3-[2-carboxyp-iperazin-4-yl]-propyl-1-phosphonic acid, 10–15 mg per kg, i.p.)30 minutes before exposing them to light for two hours. Subse-

articles

Fig. 3. Brief light exposure induces arapid change in synaptic NMDAR com-position and function in visual cortex.(a) Representative immunoblots forNR2A and NR1 proteins in synap-toneurosomes prepared from visualcortices of rats raised in completedarkness (DR), DR plus 0.5, 1, 1.5 or2 h of light, or in a normal light cycle(LR). (b) Brief light exposure increasedNR2A protein from synaptoneuro-somes of visual cortex (one-wayANOVA, p < 0.05). Each sample is nor-malized to the dark-reared control runon the same gel (dashed line), and thesummarized data are presented as per-cent of DR values (mean ± s.e.m.). (c) Brief light exposure increasesNR2A/B from synaptoneurosomes ofvisual cortex (one-way ANOVA,p < 0.05). For each sample, the signalfor NR2A/NR2B is calculated beforenormalizing to the dark-reared controlrun on the same gel (dashed line).Summarized data are presented as per-cent of DR values (mean ± s.e.m.). (d) Brief light exposure decreases thesensitivity of NMDAR potentials toifenprodil. Normalized NMDAR-medi-ated field potentials in dark-reared rats(filled circles) or dark-reared ratsexposed to light for two hours (DR + 2,open circles). Data are mean value aver-aged for the four pulses of probingstimulation. After two hours of light,NMDAR-mediated field potentials invisual cortex slices are less sensitive toifenprodil than slices from dark-rearedrats (t-test at 90 min after ifenprodil,p < 0.01). Error bars indicate s.e.m.Dashed line, normalized 30-minutebaseline.

80

100

120

140

160

% o

f dar

k-re

ared

0 0.5 1 1.5 2 LRHours of light exposure

//

B

NR2A

0

0.2

0.4

0.6

0.8

1.0

1.2

-40 -20 0 20 40 60 80 100

Time from ifenprodil application (min)

No

rmal

ized

NM

DA

FP

DR + 2DR

D

3 µM ifenprodil

ANR2ANR1

DR +0.5 +1 +1.5 +2 LR

75

100

150

200

% o

f dar

k-re

ared

0 0.5 1 1.5 2 LRHours of light exposure

C

125

175

225NR2A/B

//

a

b

c

d

Per

cen

t o

f d

ark-

rear

edHours of light exposure

Per

cen

t o

f d

ark-

rear

ed

Hours of light exposure

No

rmal

ized

NM

DA

FP



199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


quent electrophysiological analysis of theifenprodil sensitivity of NMDAR-mediat-ed responses in visual cortex revealed nosignificant differences between the CPP-injected, dark-reared group (n = 10 slicesfrom 6 rats) and the CPP-injected groupreceiving light exposure (dark-reared +2;n = 13 slices from 8 rats; t-test, p = 0.47).Thus, CPP treatment completely prevent-ed the experience-dependent modificationof synaptic NMDARs (Fig. 4a).Immunoblot analysis confirmed that thelight-induced increase in NR2A protein wasalso completely blocked in the visual cor-tex of CPP-injected animals (Fig. 4b; rawmean O.D. ± s.e.m., dark reared,1010 ± 116; dark reared + 2 h light,936 ± 120; one-tailed t-test, p = 0.34).Thus, NMDAR activity is necessary for theexperience-dependent insertion of NR2A-containing synaptic NMDARs.

In a final series of experiments, weexamined whether the experience-depen-dent increase in synaptic NR2A requiresprotein synthesis. Animals were injected inthe dark with the mRNA translationinhibitor cycloheximide (1 mg per kg, i.p.)30 minutes before receiving one hour oflight exposure. This treatment completelyblocked the experience-induced increase inNR2A protein (Fig. 4c; raw meanOD ± s.e.m., dark reared, 946 ± 46; darkreared + 1 h light, 941 ± 62; n = 4, one-tailed t-test, p = 0.50).

DISCUSSIONStudies in vitro have suggested that somedevelopmental changes in glutamate recep-tors are activity dependent. For example,growing neuronal cultures under condi-tions of heightened18 or reduced19 activity can cause a change inthe surface expression of synaptic AMPA receptors. Likewise,NMDAR subunit expression14,20 and clustering at postsynapticsites21 in vitro are regulated by presynaptic activity and postsy-naptic NMDAR activation. However, these changes invariablyhave required days for their expression, too slow to account forrapid, activity-dependent synaptic modifications22. Thus, thefinding that only one to two hours of visual experience can alterthe complement of postsynaptic glutamate receptors in visualcortex in vivo is of considerable interest. Such a change is fastenough to contribute to the rapid synaptic modifications thathave been reported in the visual cortex of dark-reared animalsexposed to light23–25.

A model consistent with the results of our study is that sen-sory experience regulates, via NMDAR activation, the synthesisand postsynaptic surface expression of NR2A-containingNMDARs in the visual cortex in vivo. This proposed mechanismmust be viewed as tentative because our manipulations ofNMDARs and protein synthesis were systemic, and might haveexerted their effects at sites other than the visual cortex. How-ever, the idea that NMDAR subunit expression can be controlledby NMDAR activity receives strong support from studies usingcultured cerebellar slices. In this preparation, postsynaptic

NMDAR activation (for several days and coupled with neureg-ulin) drives expression of the NR2C subunit, which, in turn, reg-ulates the phenotype of NMDARs in granule cells20 . There isalso a precedent for experience-dependent regulation of mRNAtranslation in the visual cortex. Specifically, brief visual experi-ence stimulates the rapid polyadenylation and translation of α-CaMKII mRNA26. The localization of NR2A and NR2B mRNAto neurites of cultured neurons27 raises the intriguing possibili-ty that experience-dependent regulation of NR2A synthesis couldoccur in dendrites28 in response to synaptic activation.

What are the functional consequences of changing the sub-unit composition of NMDARs in visual cortex? In visual cortex,as elsewhere, the amount of calcium passing through activatedNMDARs can determine whether a synapse undergoes long-termpotentiation (LTP) or long-term depression (LTD)29. The expe-rience-dependent increase in the NR2A/B ratio and the con-comitant shortening of synaptic NMDAR currents are likely tohave a significant impact on the properties of synaptic plasticity.Shortening NMDAR currents would be expected to alter theLTD-LTP ‘modification threshold’ (θm), making LTD more like-ly and LTP less likely in response to a given amount of synapticactivation30. Indeed, studies of LTD and LTP in visual cortexreveal precisely this change in dark-reared animals exposed to

articles

Fig. 4. Treatment of animals with CPP, acompetitive antagonist of NMDARs, orcycloheximide, an inhibitor of mRNAtranslation, blocks the experience-induced increase in NR2A/B. (a) Effectof ifenprodil on the magnitude of nor-malized NMDAR-mediated field poten-tials in visual cortex slices taken fromrats injected in the dark with CPP. Meanvalue averaged for the four pulses ofprobing stimulation in the DR (filled cir-cles) and DR + 2 (open circles) groups.(b) Immunoblots of NR2A and NR1proteins of synaptoneurosomes pre-pared from visual cortices of dark-reared rats treated with CPP (D) andfrom visual cortex of animals exposedto light for 2 h (+2) starting 30 min afterCPP treatment. Summary data showthat light exposure did not change thelevel of NR2A subunit following CPPtreatment (solid line). For comparison,NR2A data from visual cortex in DRand DR + 2 without CPP (dashed line)are shown. Each sample was normalizedto the mean dark-reared value, and thesummarized data are presented as per-cent of DR values (mean ± s.e.m.). (c) Immunoblots of NR2A and NR1proteins of synaptoneurosomes pre-pared from visual cortices of dark-reared rats treated with cycloheximide(D) and from visual cortex of animalsexposed to light for 1 h (+1) starting30 min after cycloheximide (CHX)treatment. Light exposure produced nochange in the level of NR2A subunit fol-lowing CHX treatment (solid line). Forcomparison, NR2A data from visualcortex in DR and DR + 1 without CHX(dashed line) are shown.


0

0.2

0.4

0.6

0.8

1.0

1.2

-40 -20 0 20 40 60 80 100

No

rmal

ized

NM

DA

FP

DR + 2DR

3 µM ifenprodil

70

85

100

115

130

145

0 1 2

% o

f d

ark-

rear

ed


CTL

CPP

NR1

D D D D +2 +2 +2 +2 NR2A

80

90

100

110

120

130

% o

f d

ark-

rear

ed

0 1Hours of light exposure

CHX

CTL

NR2ANR1

D D D +1 +1 +1 +1

B

C

a

c

b

No

rmal

ized

NM

DA

FP


Per

cen

t o

f d

ark-

rear

ed


Per

cen

t o

f d

ark-

rear

ed



199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


light31. Theoretical investigations suggest that these activity-dependent adjustments of θm are crucially involved in corticaldevelopment by maintaining the network of modifiable synaps-es within a useful dynamic range32. We suggest that one molec-ular mechanism for such experience-dependent modificationsof synaptic plasticity in visual cortex is the regulation of NMDARsubunit composition.

METHODSImmunoblot analysis. Male and female P21–23 Long-Evans rats (CharlesRiver) were anesthetized with methoxyflurane vapor either in the dark(dark-reared) or the light (light-reared and light-exposure groups) andwere decapitated following disappearance of corneal reflexes in com-pliance with the U.S. Department of Health and Human Services andBrown University guidelines. Light exposure was begun 5–7 h into thelight cycle. Synaptoneurosomes were prepared using a procedure adapt-ed from ref. 11. The primary visual cortex was rapidly dissected in ice-cold dissection buffer (212.7 mM sucrose, 2.6 mM KCl, 1.23 mMNaH2PO4, 26 mM NaHCO3, 10 mM dextrose, 1 mM MgCl2, 0.5 mMCaCl2, 0.02 mM CNQX and 0.1 mM AP5, saturated with 95% O2 and5% CO2) and immediately homogenized in ice-cold homogenizationbuffer (10 mM HEPES, 1 mM EDTA, 2 mM EGTA, 0.5 mM DTT, 0.1mM PMSF, 10 mg per liter leupeptin, 50 mg per liter soybean trypsininhibitor and 100 nM microcystin). Tissue was homogenized in aglass–glass tissue homogenizer (Kontes, Vineland, New Jersey), and thehomogenate was passed sequentially through two 100-µm-pore nylonmesh filters, followed by a 5-µm-pore filter, and centrifuged at 1000 × gfor 10 min. The resulting pellets were resuspended in boiling 1% SDSand stored at –80°C. Equal amounts of synaptoneurosome protein,determined using the BCA assay (Pierce, Rockford, Illinois), wereresolved on 7.5% polyacrylamide gels, transferred to nitrocellulose andprobed with either anti-NR2A or anti-NR2B polyclonal antibodies(1:1000; ref. 6) or anti-NR1 monoclonal antibody (1:1000, clone 54.1,Pharmingen, San Diego, California), followed by the appropriate sec-ondary antibody coupled to horseradish peroxidase (1:3500, Sigma, St.Louis, Missouri) in Tris-buffered saline, pH 7.3, containing 1% bovineserum albumin and 0.1% Triton X-100 (Sigma). Visualization ofimmunoreactive bands was produced by enhanced chemiluminescence(Amersham ECL) captured on autoradiography film (Amersham HyperECL). Digital images produced by densitometric scans of autoradi-ographs on a ScanJet IIcx (Hewlett Packard) with DeskScan II software(Hewlett Packard) were quantified using NIH Image 1.60 software. Theintensity of each band was determined relative to a baseline immedi-ately above and below the band within the same lane, and normalizedto light-reared or dark-reared controls run on the same gel.

Electrophysiology. Slice electrophysiology experiments were done asdescribed33. Briefly, rats were deeply anesthetized with inhalation anes-thetic methoxyflurane and decapitated. The brain was removed, dissect-ed and sliced in dissection buffer as described above, with the exceptionsthat the buffer contained 3 mM MgCl2, 1 mM CaCl2 and 5–10 mMkynurenic acid instead of CNQX and AP5. Slices were allowed to recov-er for 1–2 h at room temperature in artificial cerebrospinal fluid (ACSF)containing 124 mM NaCl, 5 mM KCl, 1.25 mM Na2PO4, 26 mMNaHCO3, 1 mM MgCl2, 2 mM CaCl2 and 10 mM dextrose, saturated in95% O2, 5% CO2. For recording, slices were placed in a submersionrecording chamber, maintained at 30°C and perfused with ACSF at a rateof 2 ml per min. Extracellular electrodes (filled with ACSF; 1.0 MΩ) wereused to monitor field potentials evoked with a stimulating electrode (con-centric bipolar tungsten). The magnitude of responses was monitoredby the amplitude of the field potential. Stable baseline responses wereelicited two per min at 80% of maximal response. NMDAR-mediatedresponses were pharmacologically isolated in artificial cerebrospinal fluid(ACSF) containing 3 mM CaCl2, 0.1 mM MgCl2, 0.1% DMSO, 20 µMCNQX, 1 µM glycine and 0.5 µM bicuculline methiodide. Probing stim-ulation consisting of 4 pulses delivered at 30-second intervals was givenevery 10 min to assess NMDAR-mediated field potentials. Because ofstudies demonstrating inhibition of NR2A-containing NMDARs byzinc34, we did pilot studies to demonstrate that there was no tonic inhi-

bition of NMDARs by zinc in our slice preparation (data not shown).NMDAR blockade was achieved by the bath application of 100 µM DL-2-amino-5-phosphonovaleric acid (AP5; Sigma). Ifenprodil (3 µM; RBI)was used to block NR2B-containing receptors. This concentration ofifenprodil is in a range that produces nearly maximal inhibition of NR2B-containing NMDARs, has little effect on NR2A-containing NMDARsand does not affect voltage-dependent calcium channels12–14.Dose–response curves done for this study also determined that this con-centration is specific to antagonism of the NMDAR (data not shown).As noted previously15, we found that wash-out of ifenprodil is slow andincomplete. Therefore, to calculate the percent inhibition of the NMDAR-mediated field potential by ifenprodil, we compared the 30-minute aver-age immediately before adding the drug with the average of 4 consecutivesweeps collected 90 minutes after drug application, when the ifenprodileffect had completely equilibrated. A similar difference between light-reared and dark-reared cortex was found using another NR2B-selectiveantagonist, CP-101,606-27 (5 µM, Pfizer; data not shown).

Drug injections. To block NMDA receptors or protein synthesis, we inject-ed animals in the dark (i.p.) with either CPP or CHX. CPP-injected ani-mals were awake and alert. However, a fraction of the animals displayedslightly reduced locomotor activity. CHX-injected animals were awake andalert, and displayed no overt illness or distress 1.5 h after injection. How-ever, we did note in pilot studies that animals were visibly sluggish at 2.5 hafter the injection. Therefore, we restricted analysis to the earlier time point.Synaptic transmission and neuronal excitability remain normal for manyhours following the blockade of protein synthesis with CHX35.

ACKNOWLEDGEMENTSThe authors thank D. Olstein, A. Sekhar, E. Sklar and S. Meagher for assistance.

This work was supported in part by grants from the Human Frontiers Science

Program and the National Eye Institute.

RECEIVED 11 DECEMBER 1998, ACCEPTED 24 FEBRUARY 1999

1. Wyszynski, M. et al. Differential regional expression and ultrastructurallocalization of alpha-actinin-2, a putative NMDA receptor-anchoringprotein, in rat brain. J. Neurosci. 18, 1383–1392 (1998).

2. Monyer, H., Burnashev, N., Laurie, D. J., Sakmann, B. & Seeburg, P. H.Developmental and regional expression in the rat brain and functionalproperties of four NMDA receptors. Neuron 12, 529–540 (1994).

3. Ishii, T. et al. Molecular characterization of the family of the N-methyl-D-aspartate receptor subunits. J. Biol. Chem. 268, 2836–2843 (1993).

4. Rostas, J. A. et al. Enhanced tyrosine phosphorylation of the 2B subunit of theN-methyl-D-aspartate receptor in long-term potentiation. Proc. Natl. Acad.Sci. USA 93, 10452–10456 (1996).

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

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

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

8. Sheng, M., Cummings, J., Rolden, L. A., Jan, Y. N. & Jan, L. Y. Changingsubunit composition of heteromeric NMDA receptors during developmentof rat cortex. Nature 368, 144–147 (1994).

9. Flint, A. C., Maisch, U. S., Weishaupt, J. H., Kriegstein, A. R. & Monyer, H.NR2A subunit expression shortens NMDA receptor synaptic currents indeveloping neocortex. J. Neurosci. 17, 2469–2476 (1997).

10. Carmignoto, G. & Vicini, S. Activity-dependent decrease in NMDA receptorresponses during development of the visual cortex. Science 258, 1007–1011(1992).

11. Hollingsworth, E. B. et al. Biochemical characterization of a filteredsynaptoneurosome preparation from guinea pig cerebral cortex: cyclicadenosine 3′:5′-monophosphate-generating systems, receptors, and enzymes.J. Neurosci. 5, 2240–2253 (1985).

12. Williams, K., Russell, S. L., Shen, Y. M. & Molinoff, P. B. Developmentalswitch in the expression of NMDA receptors occurs in vivo and in vitro.Neuron 10, 267–278 (1993).

13. Church, J., Fletcher, E. J., Baxter, K. & MacDonald, J. F. Blockade by ifenprodilof high voltage-activated Ca2+ channels in rat and mouse culturedhippocampal pyramidal neurones: comparison with N- methyl-D-aspartatereceptor antagonist actions. Br. J. Pharmacol. 113, 499–507 (1994).

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

14. Gottmann, K., Mehrle, A., Gisselmann, G. & Hatt, H. Presynaptic control ofsubunit composition of NMDA receptors mediating synaptic plasticity. J.Neurosci. 17, 2766–2774 (1997).

15. Ramoa, A. S. & Prusky, G. Retinal activity regulates developmental switchesin functional properties and ifenprodil sensitivity of NMDA receptors in thelateral geniculate nucleus. Dev. Brain. Res. 101, 165–175 (1997).

16. Roberts, E. B., Meredith, M. A. & Ramoa, A. S. Suppression of NMDAreceptor function using antisense DNA blocks ocular dominance plasticitywhile preserving visual responses. J. Neurophysiol. 80, 1021–1032 (1998).

17. Bear, M. F., Kleinschmidt, A., Gu, Q. & Singer, W. Disruption of experience-dependent synaptic modifications in striate cortex by infusion of an NMDAreceptor antagonist. J. Neurosci. 10, 909–925 (1990).

18. Lissin, D. V. et al. Activity differentially regulates the surface expression ofsynaptic AMPA and NMDA glutamate receptors. Proc. Natl. Acad. Sci. USA95, 7097–7102 (1998).

19. Turrigiano, G. G., Leslie, K. R., Desai, N. S., Rutherford, L. C. & Nelson, S. B.Activity-dependent scaling of quantal amplitude in neocortical neurons.Nature 391, 892–896 (1998).

20. Ozaki, M., Sasner, M., Yano, R., Lu, H. S. & Buonanno, A. Neuregulin-betainduces expression of an NMDA-receptor subunit. Nature 390, 691–694 (1997).

21. Rao, A. & Craig, A. M. Activity regulates the synaptic localization of theNMDA receptor in hippocampal neurons. Neuron 19, 801–812 (1997).

22. Craig, A. M. Activity and synaptic receptor targeting: the long view. Neuron21, 459–462 (1998).

23. Buisseret, P., Gary-Bobo, E. & Imbert, M. Ocular motility and recovery oforientational properties of visual cortical neurones in dark-reared kittens.Nature 272, 816–817 (1978).

24. Buisseret, P., Gary-Bobo, E. & Imbert, M. Plasticity in the kitten’s visualcortex: effects of the suppression of visual experience upon the orientationalproperties of visual cortical cells. Dev. Brain Res. 4, 417–426 (1982).

25. Fox, K., Daw, N., Sato, H. & Czepita, D. Dark-rearing delays the loss ofNMDA-receptor function in kitten visual cortex. Nature 350, 342–344(1991).

26. Wu, L. et al. CPEB-mediated cytoplasmic polyadenylation and the regulationof experience-dependent translation of α-CaMKII mRNA at synapses.Neuron 21, 1129–1139 (1998).

27. Miyashiro, K., Dichter, M. & Eberwine, J. On the nature and differentialdistribution of mRNAs in hippocampal neurites: implications for neuronalfunctioning. Proc. Natl. Acad. Sci. USA 91, 10800–10804 (1994).

28. Steward, O. mRNA localization in neurons: a multipurpose mechanism?Neuron 18, 9–12 (1997).

29. Bear, M. F. & Malenka, R. C. Synaptic plasticity: LTP and LTD. Curr. Opin.Neurobiol. 4, 389–399 (1994).

30. Gold, J. I. & Bear, M. F. A model of dendritic spine Ca2+ concentrationexploring possible bases for a sliding modification threshold. Proc. Natl.Acad. Sci. USA 91, 3941–3945 (1994).

31. Kirkwood, A., Rioult, M. G. & Bear, M. F. Experience-dependent modificationof synaptic plasticity in visual cortex. Nature 381, 526–528 (1996).

32. Bienenstock, E. L., Cooper, L. N. & Munro, P. W. Theory for the developmentof neuron selectivity: Orientation specificity and binocular interaction invisual cortex. J. Neurosci. 2, 32–48 (1982).

33. Kirkwood, A., Dudek, S. M., Gold, J. T., Aizenman, C. D. & Bear, M. F.Common forms of synaptic plasticity in the hippocampus and neocortex invitro. Science 260, 1518–1521 (1993).

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

35. Otani, S., Marshall, C. J., Tate, W. P., Goddard, G. V. & Abraham, W. C.Maintenance of long-term potentiation in rat dentate gyrus requires proteinsynthesis but not messenger RNA synthesis immediately post-tetanization.Neuroscience 28, 519–526 (1989).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

The striatum is a key component of the forebrain system thatcontrols planning and execution of motor behaviors. Excitatorysignals generated in sensorimotor and limbic areas of the neo-cortex and in the thalamus converge on this region, where theyare integrated and redistributed to other structures of the basalganglia and to the substantia nigra1. How the striatum integratesthese inputs, which are mediated by the fast neurotransmitterglutamate, is only partially understood. Nevertheless, it is gen-erally agreed that slow-acting modulatory substances, includingdopamine, acetylcholine and neuroactive peptides, participatein this process by influencing the excitability of striatal neurons2.In support of this idea, abnormalities in striatal neuromodula-tion have been linked to a spectrum of neuropsychiatric disor-ders, of which Parkinson’s disease3 and Tourette’s syndrome4,5

are two well-documented examples.Cannabinoid receptors, the pharmacological target of the mar-

ijuana constituent ∆9-tetrahydrocannabinol (∆9-THC), are dense-ly expressed in striatum6–8, where they are twice as numerous asD1 dopamine receptors8 and 12 times as numerous as µ opioidreceptors9. Activation of cannabinoid receptors has profound con-sequences on the electrophysiological properties of striatal neu-rons10, as well as on motor behaviors that are mediated by striatalprojection systems11. Furthermore, clinical observations suggestthat marijuana and ∆9-THC may be beneficial in psychomotordisorders associated with the basal ganglia, such as Tourette’s syn-drome12,13, pointing to an involvement of cannabinoid receptors inabnormal striatal function. Interpreting these results is made dif-ficult, however, by our inadequate knowledge of the intrinsic sig-

Dopamine activation ofendogenous cannabinoid signalingin dorsal striatum

A. Giuffrida1, L. H. Parsons2, T. M. Kerr2, F. Rodríguez de Fonseca3, M. Navarro3

and D. Piomelli1

1 Department of Pharmacology, 360 Med Surge II, University of California at Irvine, Irvine, California 92697-4625, USA2 Department of Neuropharmacology, The Scripps Research Institute, La Jolla, California 92037, USA3 Department of Psychobiology, Universidad Complutense, Madrid, 28233, Spain

The first two authors contributed equally to this work.

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

We measured endogenous cannabinoid release in dorsal striatum of freely moving rats by microdialy-sis and gas chromatography/mass spectrometry. Neural activity stimulated the release ofanandamide, but not of other endogenous cannabinoids such as 2-arachidonylglycerol. Moreover,anandamide release was increased eightfold over baseline after local administration of the D2-like(D2, D3, D4) dopamine receptor agonist quinpirole, a response that was prevented by the D2-likereceptor antagonist raclopride. Administration of the D1-like (D1, D5) receptor agonist SKF38393 hadno such effect. These results suggest that functional interactions between endocannabinoid anddopaminergic systems may contribute to striatal signaling. In agreement with this hypothesis,pretreatment with the cannabinoid antagonist SR141716A enhanced the stimulation of motor behav-ior elicited by systemic administration of quinpirole. The endocannabinoid system therefore may actas an inhibitory feedback mechanism countering dopamine-induced facilitation of motor activity.

naling system by which cannabinoid receptors are engaged. Indeed,although several endogenous cannabinoid (endocannabinoid) lig-ands, including anandamide14,15 and 2-arachidonylglycerol (2-AG)16–18, have been identified and their biosynthetic routespartially elucidated15,18–21, the physiological mechanisms that reg-ulate release of these compounds remain elusive.

RESULTSAnandamide release in vivoWe examined the occurrence and regulation of endogenouscannabinoid release in the dorsal striatum of freely moving ratsby using microdialysis combined with isotope dilution gas-chromatography/mass-spectrometry (GC/MS)22. Microdialysissamples obtained during 30-min collections under baseline con-ditions contained detectable levels of anandamide (1.5 ± 0.3pmol per sample, mean ± s.e.m, n = 60; Fig. 1),palmitylethanolamide (PEA), an acylethanolamide that acti-vates peripheral CB2-like receptors23,24 (0.7 ± 0.1 pmol/sam-ple), and oleylethanolamide, the functions of which remainunknown19,25 (1.5 ± 0.2 pmol per sample). By contrast, 2-AGwas not detectable under these conditions (data not shown). Nomeasure was taken in these analyses to prevent the uptake andenzymatic hydrolysis of anandamide and 2-AG15,18,26–30; thusthe impact of these inactivation processes on endogenouscannabinoid levels remains to determined.

To test whether endogenous cannabinoids are released byneural activity, we perfused the striatum with artificial cere-brospinal fluid (ACSF) containing a depolarizing concentration


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

of KCl (60 mM). This high-K+ pulse significantly increased anan-damide outflow (Fig. 2a), whereas it had no effect on PEA,oleylethanolamide or 2-AG (Fig. 2c and data not shown; n = 13).After reinstatement of normal ACSF, anandamide levels rapidlyreturned to basal values (Fig. 2a). The overall time course of thisresponse was identical to that of K+-induced dopamine release,measured in parallel microdialysate samples (Fig. 2b). The effectof high K+ on anandamide outflow was prevented either by theNa+-channel blocker tetrodotoxin (1 µM) or by removal of Ca2+

ions, two treatments that alone had no significant effect on basalanandamide levels (Fig. 3). These results demonstrate that anan-damide is released in the dorsal striatum of freely moving ratsduring neural activity, fulfilling an essential criterion for this lipidto be considered a neuromodulator in the central nervous sys-tem (CNS). Furthermore, the finding that 2-AG and PEA maynot be released during neural activity indicates that in striatumsuch a role is specific to anandamide.

D2 receptors stimulate anandamide releaseThe modulatory neurotransmitter dopamine regulates essentialaspects of striatal physiology2 by interacting with two pharma-

cologically distinct groups of G-protein-coupled receptors, D1-like (D1 and D5) and D2-like (D2, D3 and D4)31,32. To determinewhether activation of dopamine receptors in striatum affectsanandamide release, we locally applied by reverse dialysis selectiveD1-like and D2-like receptor ligands. Administration of the D2-like agonist quinpirole resulted in an eightfold stimulation ofanandamide outflow (Fig. 4a). Extracellular anandamide levelsremained elevated for at least two hours after quinpirole admin-istration (Fig. 4a), possibly as a result of slow clearance of thedrug from its site of application. In support of this possibility,when the D2-like antagonist raclopride (20 µM) was adminis-tered after quinpirole, baseline anandamide levels were reachedwithin 30–60 min (data not shown). As with high K+, the out-puts of 2-AG, PEA and oleylethanolamide were not affected byquinpirole (data not shown).

To investigate the receptor mechanism underlying theresponse to quinpirole, we examined the effects of the D2-likeantagonist raclopride. Raclopride (20 µM) did not affect micro-dialysate anandamide concentrations when applied alone, butcompletely prevented the stimulatory effects of quinpirole(Fig. 4b). Furthermore, the D1-like agonist SKF38393 (10 µM)

Fig. 1. Microdialysis ofendogenous cannabinoids inrat brain. (a) Striatal locationof the microdialysis probes.The stippled area indicatesthe approximate position ofthe probes in the 77 animalsincluded in this study. CTX,cortex; CP, caudate-putamen;NAc, nucleus accumbens. (b, c) Identification by gaschromatography/mass spec-trometry of anandamide inmicrodialysis perfusates of ratdorsal striatum. Anandamideand other endogenouscannabinoids were purified chromatographically from 30-min dialysate samples and analyzed simultaneously by selected ion monitoring GC/MS as bistrimethylsilylethers. For quantitation, synthetic deuterium-containing standards were added to all samples. Representative tracings for selected frag-ments characteristic of endogenous anandamide (b, mass-to-charge ratio, m/z = 404) and synthetic [2H4]-anandamide (c, m/z = 408). The arrow indi-cates the retention time of standard anandamide. Results are from one experiment and are typical of 77 independent experiments.

b

c

Ion

cu

rren

tIo

n c

urr

ent

Time (min)

a

Fig. 2. Local application of a high KCl concentration (60 mM) stimulates anandamide release in dorsal striatum of freely moving rats. Effects of intras-triatal K+ depolarization on dialysate levels of anandamide (a), dopamine (DA; b) and other acylethanolamides (AE; c): palmitylethanolamide(squares) and oleylethanolamide (diamonds). Results are means ± s.e.m. (n = 13) of the amount of compounds present in 30-min dialysate samples,expressed as percent of baseline values (a, c) or as nmol per liter (b). 2-AG was below the detection limit of this assay, which was approximately1 pmol per sample. **p < 0.001, ***p < 0.0001.

a

An

and

amid

e (%

bas

elin

e)

Dia

lysa

te D

A (

nM

)

AE

(% b

asel

ine)

Time (min) Time (min) Time (min)

bc

CTX


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

did not change the basal outflows of anandamide (Fig. 4c), PEA,2-AG or oleylethanolamide (data not shown). The lack of effectof SKF38393 underscores the differences between D1-like andD2-like receptor agonists with respect to anandamide release, butdoes not rule out the possibility that D1-like receptors may reg-ulate this process in other ways, for example by acting synergis-tically or antagonistically with D2-like receptors.

Modulation of motor activityThe results of these neurochemical experiments indicate thatdopamine acting at D2-like receptors stimulates anandamiderelease in dorsal striatum, suggesting that the endocannabinoidsystem participates in dopaminergic regulation of striatal func-tion. To test this possibility, we determined whether the behav-ioral response elicited by systemic administration of quinpirole inrats is affected by the CB1 receptor antagonist SR141716A33. Inagreement with previous results34, quinpirole (1 mg per kg)caused a biphasic motor response characterized by transient sup-pression of movement, which is thought to be caused by activa-tion of presynaptic D2-like receptors, followed by a longer-lastinghyperactivity, possibly due to activation of postsynaptic D2-likereceptors34,35. This response included changes in horizontal loco-motion, time spent in immobility and sniffing frequency (Fig. 5).As previously reported36, SR141716A had no overt effect onmotor activity when given alone at a dose of 1 mg per kg (Fig. 5).Nevertheless, when SR141716A was injected at the same dose 60min before quinpirole, the late phase of quinpirole-inducedmotor activation was markedly potentiated, whereas the initial

phase of motor suppression remainedunchanged (Fig. 5). Thus pharmacologicalblockade of CB1 receptors enhances the motorstimulation produced by activation of postsy-naptic D2-like receptors, but has little or no effecteither on basal motor activity or on presynapticD2-receptor-dependent motor inhibition.

DISCUSSIONIn striatal and cortical neurons in primary cul-ture, formation of anandamide is stimulated bymembrane depolarization, suggesting that thiscompound may be produced during neuralactivity and participate in endocannabinoid sig-naling15,18. Here we used a combination ofmicrodialysis and GC/MS techniques22 to inves-tigate the release of anandamide and otherendogenous cannabinoid substances in the dor-

sal striatum of freely moving rats. We found that neural activityevoked by a localized pulse of high K+ stimulates the outflow ofanandamide, but not 2-AG and PEA. The possibility implied bythese findings that anandamide acts as a neural mediator in stria-tum is supported by both anatomical and pharmacological evi-dence. GABAergic medium spiny neurons, which account forabout 95% of the striatal neuron population and are the source ofmost striatofugal projections1, contain large numbers of CB1cannabinoid receptors8. Activation of these receptors causespresynaptic inhibition of GABA release in vitro10 and profound-ly affects motor behaviors in vivo37,38. Indeed, certain aspects ofthe motor inhibition produced by systemically administeredcannabimimetic drugs, such as attenuation of stereotyped behav-iors, may be mediated by their ability to activate striatal CB1receptors39. By showing that anandamide is released in striatumduring neural activity, our results point to this endogenouscannabinoid lipid as a primary component of the network ofneurally active substances that regulate striatal function2.

Brain tissue contains 2-AG in amounts 170 times greater thananandamide18,40. Thus, we were surprised to find that the extra-cellular levels of this compound in striatum are undetectableboth under baseline conditions and during neural activity. Lim-itations of our isotope dilution assay are unlikely to account forthis negative result, as this method provides very similar detectionlimits for 2-AG (1 pmol per sample) and anandamide (0.4 pmolper sample, see Methods). Differences in biological inactivationare also an improbable explanation, because 2-AG and anan-damide are eliminated at comparable rates (M. Beltramo and D.

Fig. 3. K+-stimulated anandamiderelease requires membrane depo-larization and external Ca2+. Effectsof a high-K+ pulse on the release ofanandamide in artificial cere-brospinal fluid (ACSF, n = 13),ACSF containing tetrodotoxin(TTX, 1 µM, n = 6) and ACSF withzero Ca2+(Ca2+-free, n = 5).Whitebars, baseline anandamide release;black bars, K+-stimulated release.Results are expressed as describedin Fig. 2 legend.

Fig. 4. D2-like dopamine receptoractivation evokes anandamide releasein striatum. Effects on dialysate anan-damide levels of intrastriatal adminis-tration of quinpirole (QUIN, 10 µM), aD2-like agonist (a), raclopride (RACL,20 µM), a D2-like antagonist appliedalone or with quinpirole (b) orSKF38393 (SKF; 10 µM), a D1-like ago-nist (c). Results, expressed asdescribed in Fig. 2 legend, are means ±s.e.m. of six experiments for eachcondition. *p < 0.05; **p < 0.001.

ACSF TTX Ca2+-free

An

and

amid

e (%

bas

elin

e)

Time (min)Time (min)

An

and

amid

e (%

bas

elin

e)

An

and

amid

e (%

bas

elin

e)

a b

c


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Piomelli, unpublished observations). Alternatively, anandamideand 2-AG may be produced under different physiological cir-cumstances and/or in distinct regions of the CNS. Consistentwith this view, in hippocampal slices, high-frequency stimula-tion of glutamatergic Schaffer collaterals selectively increases theaccumulation of 2-AG, but not anandamide18.

Although anandamide and PEA are formed through a com-mon biosynthetic mechanism15,19, PEA does not interact witheither of the two cannabinoid receptor subtypes41, CB1 and CB2,whose genes have been isolated thus far42. Pharmacological exper-iments indicate, however, that, in peripheral tissues, PEA acti-

vates a CB2-like receptor, which mediates antinociception andanti-inflammation23,24. Our results, showing that PEA may not bereleased extracellularly during neural activity in striatum, fur-ther highlight the peripheral roles of this compound.

Unlike neurotransmitters and neuropeptides, which are releasedfrom synaptic terminals via vesicle secretion, anandamide may beproduced and released upon demand by a mechanism that involvesphospholipase-mediated cleavage of the membrane phospholipidprecursor N-arachidonylphosphatidylethanolamine15,19–21,43. Suchnonvesicular release process suggests that anandamide may act inthe CNS more as an autacoid (local mediator) substance than asa classical neuromodulator. Lipid autacoids such as the eicosanoidsand platelet-activating factor are formed by receptor-mediatedcleavage of phospholipids and act near their sites of production,where they are also rapidly inactivated (for review, see ref. 44).That anandamide may conform to this model is suggested by ourfinding that occupation of striatal D2-like dopamine receptors dra-matically stimulates anandamide outflow. A parsimonious inter-pretation of this result is that anandamide may be released fromstriatal neurons or nigrostriatal dopaminergic terminals, both ofwhich bear D2-like receptors, and may exert its effects within aconfined volume of striatal tissue.

What, if any, is the physiological function of striatal anandamiderelease? We have begun to address this question by studying theeffects of the selective CB1 receptor antagonist SR141716A on thebehavioral responses produced by quinpirole in rats. The resultsshow that, although SR141716A has no overt effect when admin-istered alone, it enhances the motor activation elicited by quinpi-role. The inverse agonist properties of SR141716A, which havebeen characterized in vitro45, cannot account for such a differentialeffect; a more plausible interpretation that is also consistent withour neurochemical data is that pharmacological blockade ofcannabinoid receptors increases quinpirole-induced hyperactivityby removing the inhibitory control of endogenously released anan-damide. According to this hypothesis, occupation of D2-like recep-tors by dopamine elicits the release of anandamide in striatum andpossibly in other regions of the CNS that contribute to movementcontrol. By engaging CB1 receptors, anandamide may act in turnto counter dopamine stimulation of motor activity, which isthought to be mediated by postsynaptic D2-like and D1-like recep-tors46. In further support of this hypothesis, anandamide inhibitsmovement when it is administered as a drug47, and cannabimimet-ic agents attenuate amphetamine-evoked hyperactivity48. The func-tional interaction between anandamide and dopaminedemonstrated in this study suggests a possible participation of theendogenous cannabinoid system in pathologies that involve dys-regulated dopamine neurotransmission. Thus, our findings mayhave implications for neuropsychiatric disorders such as schizo-phrenia, Tourette’s syndrome and Parkinson’s disease and maypoint to novel therapeutic approaches for these conditions.

METHODSDrugs. SR141716A (N-(piperidin-1-yl)-5-(4-chlorophenyl)-1-(2,4-dichlorophenyl)-4-methyl-1H-pyrazole-3-carboxamide•HCl) was pro-vided by RBI (Natick, Massachusetts) as part of the Chemical SynthesisProgram of the NIMH (N01MH30003); all other drugs were from RBIor Sigma (St. Louis, Missouri).

Microdialysis. Male Wistar rats (Charles River, Holister, California) wereanesthetized with halothane (1.0–1.5%), and stainless steel microdialysisguide cannulae (model CMA/10; Carnegie Medicine Apparatus, Solna,Sweden) were implanted in the caudate-putamen (from Bregma A + 1.0mm, L ± 2.5 mm; from dura V – 2.8 mm). A recovery period of at least5 days was allowed before the experiments. Approximately 12 hours before

Fig. 5. The cannabinoid antagonist SR141716A potentiates quinpirole-evoked hyperactivity. Effects of systemic administration of vehicle(hatched bars), SR141716A (dotted bars; 1 mg per kg, i.p.), quinpirole(open bars; 1 mg per kg, s.c.) or quinpirole plus SR141716A (closedbars; 1 mg per kg each; SR141716 was injected 60 min before quinpi-role) on horizontal locomotion (a), immobility (b) and sniffing (c).Results are means ± s.e.m. of 7–9 experiments. *p < 0.05 compared tovehicle; †p < 0.05 compared to quinpirole alone.

a

b

Loco

mo

tio

n

(nu

mb

er o

f cr

oss

ing

s)Im

mo

bili

ty (

s)Sn

iffi

ng

(n

um

ber

of

even

ts)

c

Time (min)


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

the microdialysis sessions, animals were lightly anesthetized (1–2%halothane), and microdialysis probes (model CMA/10, Carnegie Medi-cine Apparatus; 4 mm active length) were inserted into the guide cannu-lae. Anesthesia was sufficiently brief so that animals regained movementwithin 3 min of probe insertion. An artificial cerebrospinal fluid (ACSF)consisting of 145 mM NaCl, 2.8 mM KCl, 1.2 mM CaCl2, 1.2 mM MgCl2,0.25 mM ascorbic acid and 5.4 mM D-glucose (pH 7.2–7.4) was used asperfusion medium; this solution was delivered at a flow rate of 0.2 µl permin until two hours before the start of the experiment when the flow ratewas increased to 10 µl per min. Preliminary experiments indicated thatthis relatively high flow rate was necessary to collect a sufficient amount ofendogenous cannabinoids for reliable detection under baseline condi-tions. In vitro recovery of [3H]anandamide was 5.8 ± 0.2% (n = 3), whichwas comparable to that of dopamine under similar flow-rate conditions49.Dialysate concentrations were not corrected for recovery. Probe outlettubing was modified to reduce back-pressure so that no ultrafiltrationwas observable across the dialysis membrane. The ACSF was delivered tothe probes via a single-channel liquid swivel (Instech, Plymouth Meet-ing, Pennsylvania) attached to a balance arm above the animal cage toensure freedom of movement during the experiment. Microdialysate sam-ples were collected at 30-min intervals into glass vials containing internalstandards for GC/MS analysis (1.2 nmol of each [2H4]acylethanolamideand 1.0 nmol of [2H8]-2-AG) in 1 ml methanol. At the end of each exper-iment, microdialysis probe placement was verified histologically50.

Analytical procedures. Microdialysis samples were extracted with chlo-roform/methanol, fractionated by high-performance liquid chromatog-raphy (HPLC) and analyzed by GC/MS as described22, (N. Stella and D.Piomelli, unpublished results). [2H4]acylethanolamides were preparedfollowing standard procedures22 and [2H8]-2-AG was purchased fromDeva Biotech (Hartboro, Pennsylvania). The limit of detection, that is,the injected quantity that produced a signal corresponding to an aver-age blank plus 3 standard deviations, was 0.4 pmol for anandamide, 0.1pmol for PEA, 0.1 pmol for oleylethanolamide22 and 1 pmol for 2-AG(N. Stella and D. Piomelli, unpublished data). Concentrations in micro-dialysis perfusates are expressed as percent of baseline values, which werecalculated by averaging the first three samples collected before treatment.Dopamine was measured by HPLC and electrochemical detection50. Sta-tistical significance was determined by one-way analysis of variance fol-lowed by Student-Newmann-Keuls multiple comparison test.

Behavioral testing. We studied the effects of pretreatment with SR141716A(1 mg per kg intraperitoneal, i.p., 60 min before) or vehicle (10% dimethyl-sulfoxide in water, i.p., 60 min before) on the acute effects of quinpirole (1 mgper kg subcutaneous, s.c.) on spontaneous behavior and horizontal loco-motor activity. Spontaneous behavior was studied in a glass observation box(40 cm × 30 cm × 30 cm, one rat per box). Animals were placed in the boxfive minutes before the test, and tested during five minutes for various behav-iors including sniffing frequency and time spent in immobility. This proce-dure was repeated for each animal 5, 30, 60 and 120 min after theadministration of either vehicle or drugs. The animals were returned to theirhome cage at the end of each testing interval. The tests were conducted in asound-isolated room, illuminated with an indirect halogen light (125 lux).The behavior was videotaped and scored by trained observers blind to exper-imental conditions. Locomotor activity was studied in an opaque open field(100 cm × 100 cm × 40 cm), the floor of which was marked with20 cm × 20 cm squares. The field was illuminated using a ceiling halogenlight that was regulated to yield 350 lux at the center of the field. The ratswere habituated to the field of study for 10 min the day before testing. Onthe experimental day, the animals were placed in the center of the field andlocomotor activity (number of lines crossed) scored during five minutes.Behavior was tested 5, 30, 60 and 120 minutes after the injection of eithervehicle or drugs.

ACKNOWLEDGEMENTSWe thank C. Sañudo-Peña, N. Stella and M.J. Walker for comments and dis-

cussion. Part of this work was conducted at the Neurosciences Institute and

was supported by Neurosciences Research Foundation, which receives major

support from Novartis. Additional support was from the National Institute

of Drug Abuse (DA12447 and DA12413, to D.P.), CICYT and Plan Nacional

sobre Drogas (F.R.F., M.N.). F.R.F. is a research Fellow of the Jaime del Amo

Foundation.

RECEIVED 22 OCTOBER 1998; ACCEPTED 13 JANUARY 1999

1. Gerfen, C. R. The neostriatal mosaic: multiple levels of compartmentalorganization. Annu. Rev. Neurosci. 15, 285–320 (1992).

2. Graybiel, A. M., Aosaki, T., Flaherty, A. W. & Kimura, M. The basal gangliaand adaptive motor control. Science 265, 1826–1831 (1994).

3. Chase, T. N., Engber, T. M. & Mouradian, M. M. Contribution of dopaminergicand glutamatergic mechanisms to the pathogenesis of motor responsecomplications in Parkinson’s disease. Adv. Neurol. 69, 497–501 (1996).

4. Wolf, S. S. et al. Tourette syndrome: prediction of phenotypic variation inmonozygotic twins by caudate nucleus D2 receptor binding. Science 273,1225–1227 (1996).

5. Eidelberg, D. et al. The metabolic anatomy of Tourette’s syndrome. Neurology48, 927–934 (1997).

6. Herkenham, M. et al. Cannabinoid receptor localization in brain. Proc. Natl.Acad. Sci. USA 87, 1932–1936 (1990).

7. Matsuda, L. A., Lolait, S. J., Brownstein, M., Young, A. & Bonner, T. I.Structure of a cannabinoid receptor and functional expression of the clonedcDNA. Nature 346, 561–564 (1990).

8. Herkenham, M. et al. Characterization and localization of cannabinoidreceptors in rat brain: a quantitative in vitro autoradiographic study. J.Neurosci. 11, 563–583 (1991).

9. Sim, L. J., Selley, D. E., Xiao, R. & Childers, S. R. Differences in G-proteinactivation by µ- and δ-opioid, and cannabinoid receptors in rat striatum.Eur. J. Pharmacol. 307, 97–105 (1996).

10. Szabo, B., Dörner, L., Pfreundtner, C., Nörenberg, W. & Starke, K. Inhibitionof GABAergic inhibitory postsynaptic currents by cannabinoids in rat corpusstriatum. Neuroscience 85, 395–403 (1998).

11. Sañudo-Peña, M. C., Patrick, S. L., Patrick, R. L. & Walker, J. M. Effects ofintranigral cannabinoids on rotational behavior in rats: interactions with thedopaminergic system. Neurosci. Lett. 206, 21–24 (1996).

12. Hemming, M. & Yellowlees, P. M. Effective treatment of Tourette’s syndromewith marijuana. J. Psychopharmacol. 7, 389–391 (1993).

13. Müller-Vahl, K. R., Schneider, U., Kolbe, H. & Emrich, H. M. Treatment ofTourette-syndrome with delta-9-tetrahydrocannabinol. Am. J. Psychiatry (inpress).

14. Devane, W. et al. Isolation and structure of a brain constituent that binds tothe cannabinoid receptor. Science 258, 1946–1949 (1992).

15. Di Marzo, V. et al. Formation and inactivation of endogenous cannabinoidanandamide in central neurons. Nature 372, 686–691 (1994).

16. Sugiura, T. et al. 2-Arachidonoylglycerol: a possible endogenous cannabinoidreceptor ligand in brain. Biochem. Biophys. Res. Commun. 215, 89–97 (1995).

17. Mechoulam, R. et al. Identification of an endogenous 2-monoglyceride,present in canine gut, that binds to cannabinoid receptors. Biochem.Pharmacol. 50, 83–90 (1995).

18. Stella, N., Schweitzer, P. & Piomelli, D. A second endogenous cannabinoidthat modulates long-term potentiation. Nature 388, 773–778 (1997).

19. Cadas, H., di Tomaso, E. & Piomelli, D. Occurrence and biosynthesis ofendogenous cannabinoid precursor, N-arachidonoyl phosphatidyl-ethanolamine, in rat brain. J. Neurosci. 17, 1226–1242 (1997).

20. Sugiura, T. et al. Transacylase-mediated and phosphodiesterase-mediatedsynthesis of N-arachidonoylethanolamine, an endogenous cannabinoid-receptor ligand, in rat brain microsomes. Eur. J. Biochem. 240, 53–62 (1996).

21. Sugiura, T. et al. Enzymatic synthesis of anandamide, an endogenouscannabinoid receptor ligand, through N-acylphosphatidylethanolaminepathway in testis: involvement of Ca2+-dependent transacylase andphosphodiesterase activities. Biochem. Biophys. Res. Commun. 218, 113–117(1996).

22. Giuffrida, A. & Piomelli, D. Isotope dilution GC/MS determination ofanandamide and other fatty acylethanolamides in rat blood plasma. FEBSLett. 422, 373–376 (1998).

23. Facci, L. et al. Mast cells express a peripheral cannabinoid receptor withdifferential sensitivity to anandamide and palmitoylethanolamide. Proc. Natl.Acad. Sci. USA 92, 3376–3380 (1995).

24. Calignano, A., La Rana, G., Giuffrida, A. & Piomelli, D. Control of paininitiation by endogenous cannabinoids. Nature 394, 277–281 (1998).

25. Schmid, H. H. O., Schmid, P. C. & Natarajan, V. The N-acylation-phosphodiesterase pathway and cell signalling. Chem. Phys. Lipids 80,133–142 (1996).

26. Desarnaud, F., Cadas, H. & Piomelli, D. Anandamide amidohydrolaseactivity in rat brain microsomes: identification and partial characterization.J. Biol. Chem. 270, 6030–6035 (1995).

27. Ueda, N., Kurahashi, Y., Yamamoto, S. & Tokunaga, T. Partial purificationand characterization of the porcine brain enzyme hydrolyzing andsynthesizing anandamide. J. Biol. Chem. 270, 23823–23827 (1995).

28. Hillard, C. J., Wilkison, D. M., Edgemont, W. S. & Campbell, W. B.Characterization of the kinetics and distribution of N-arachidonylethanolamine (anandamide) hydrolysis by rat brain. Biochim.Biophys. Acta 1257, 249–256 (1995).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

29. Cravatt, B. F. et al. Molecular characterization of an enzyme that degradesneuromodulatory fatty-acid amides. Nature 384, 83–87 (1996).

30. Beltramo, M. et al. Functional role of high-affinity anandamide transport, asrevealed by selective inhibition. Science 277, 1094–1097 (1997).

31. Civelli, O. in Psychopharmacology: The Fourth Generation of Progress (eds.Bloom, F. E. & Kupfer, D. J.) 155–161 (Raven, New York, 1995).

32. Surmeier, D. J., Yan, Z. & Song, W. J. Coordinated expression of dopaminereceptors in neostriatal medium spin neurons. Adv. Pharmacol. 42,1020–1023 (1998).

33. Rinaldi-Carmona, M. et al. SR141716A, a potent and selective antagonist ofthe brain cannabinoid receptor. FEBS Lett. 350, 240–244 (1994).

34. Eilam, D. & Szechtman, H. Biphasic effect of D-2 agonist quinpirole onlocomotion and movements. Eur. J. Pharmacol. 161, 151–157 (1989).

35. Thorn, L., Ashmeade, T. E., Storey, V. J., Routledge, C. & Reavill, C. Evidenceto suggest that agonist modulation of hyperlocomotion is via post-synapticdopamine D2 or D3 receptors. Neuropharmacology 36, 787–792 (1997).

36. Compton, D. R., Aceto, M. D., Lowe, J. & Martin, B. R. In vivocharacterization of a specific cannabinoid receptor antagonist (SR141716A):inhibition of ∆9-tetrahydrocannabinol-induced responses and apparentagonist activity. J. Pharmacol. Exp. Ther. 277, 586–594 (1996).

37. Gough, A. L. & Olley, J. E. Catalepsy induced by intrastriatal injections of ∆9-THC and 11-OH-∆9-THC in the rat. Neuropharmacology 17, 137–144(1978).

38. Souilhac, J., Poncelet, M., Rinaldi-Carmona, M., Le Fur, G. & Soubrié, P.Intrastriatal injection of cannabinoid receptor agonist induced turningbehavior in mice. Pharmacol. Biochem. Behav. 51, 3–7 (1995).

39. Navarro, M. et al. Motor disturbances induced by an acute dose of∆9-tetrahydrocannabinol: possible involvement of nigrostriataldopaminergic alterations. Pharmacol. Biochem. Behav. 45, 291–298 (1993).

40. Kondo, S. et al. 2-Arachidonoylglycerol, an endogenous cannabinoidreceptor agonist: identification as one of the major species of

monoacylglycerols in various rat tissues, and evidence for its generationthrough Ca2+-dependent and independent mechanisms. FEBS Lett. 429,152–156 (1998).

41. Showalter, V. M., Compton, D. R., Martin, B. R. & Abood, M. E. Evaluation ofbinding in a transfected cell line expressing a peripheral cannabinoid receptor(CB2): identification of cannabinoid receptor subtype selective ligands. J.Pharmacol. Exp. Ther. 278, 989–999 (1996).

42. Matsuda, L. A. Molecular aspects of cannabinoid receptors. Crit. Rev.Neurobiol. 11, 143–166 (1997).

43. Cadas, H., Gaillet, S., Beltramo, M., Venance, L. & Piomelli, D. Biosynthesisof an endogenous cannabinoid precursor in neurons and its control bycalcium and cAMP. J. Neurosci. 16, 3934–3942 (1996).

44. Piomelli, D. in Psychopharmacology: The Fourth Generation of Progress (eds.Bloom, F. E. & Kupfer, D. J.) 595–607 (Raven, New York, 1995).

45. Landsman, R. S., Burkey, T. H., Consroe, P., Roeske, W. R. & Yamamura, H. I.SR141716A is an inverse agonist at the human cannabinoid CB1 receptor.Eur. J. Pharmacol. 334, R1–2 (1997).

46. Waddington, J. L., Molloy, A. G., O’Boyle, K. M. & Pugh, M. T. inNeurobiology of Stereotyped Behaviour (eds. Cooper, S. J. & Dourish, C. T.)64–90 (Clarendon, Oxford, 1990).

47. Fride, E. & Mechoulam, R. Pharmacological activity of the cannabinoidreceptor agonist, anandamide, a brain constituent. Eur. J. Pharmacol. 231,313–314 (1993).

48. Pryor, G. T., Larsen, F. F., Husain, S. & Braude, M. C. Interactions of delta9-tetrahydrocannabinol with d-amphetamine, cocaine, and nicotine in rats.Pharmacol. Biochem. Behav. 8, 295–318 (1978).

49. Parsons, L. H. & Justice, J. B. J. Quantitative approaches to in vivo brainmicrodialysis. Crit. Rev. Neurobiol. 8, 189–220 (1994).

50. Parsons, L. H. et al. Neurochemical evidence that postsynaptic nucleusaccumbens D3 receptor stimulation enhances cocaine reinforcement. J.Neurochem. 67, 1078–1089 (1996).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

The panoramic scenes of the world contain more informationthan we can take in with a single glance. To examine the finedetails of a visual scene, we must rely on the brain’s capability tofocus attention in a spatially selective manner and thereby facil-itate the perception of stimuli within a restricted zone of the visu-al field1,2. This covert focusing of attention has been likenedmetaphorically to a ‘spotlight’3 or ‘zoom lens’4 that can be shift-ed to relevant locations even when the eyes remain stationary.Psychophysical experiments have shown that stimuli falling with-in the spotlight of attention are detected and discriminated morerapidly and accurately than stimuli at unattended locations1–4.

The brain system that controls the attentional spotlight con-sists of an interconnected network of cortical and subcorticalstructures that modulates incoming information in the visualpathways5–7. A fundamental question that remains unresolved,however, is exactly where along the visual processing pathwaythis afferent neural activity is first modulated (either enhancedor suppressed) by spatial attention. The preponderance of evi-dence to date from single-neuron recordings in monkeys8,9 andfrom electrophysiological10,11, blood-flow neuroimaging12–16 andoptical imaging17 studies in humans indicates that neuralresponses to attended-location stimuli are enhanced in higherextrastriate cortical areas but not in the striate cortex itself.

Recent single-neuron experiments in monkeys18–20 and pre-liminary reports of fMRI studies in humans (M. Worden & W.Schneider, Soc. Neurosci. Abstr. 22, 729.7, 1996; S.P. Gandhi et al.,ARVO Meeting Abstr., 1998) have raised the possibility that spa-tial-selective attention may influence striate cortex activity dur-ing tasks that involve difficult visual discriminations. In addition,

Involvement of striate andextrastriate visual cortical areas in spatial attention

A. Martínez1, L. Anllo-Vento2, M. I. Sereno3, L. R. Frank4, R. B. Buxton4, D. J. Dubowitz5, E. C. Wong4,6, H. Hinrichs7, H. J. Heinze7 and S. A. Hillyard2

1 Departments of Psychology, 2Neurosciences, 3Cognitive Science, 4Radiology and 6Psychiatry, University of California at San Diego, 9500 Gilman Drive, La Jolla, California 92093-0608, USA

5 Division of Biology, California Institute of Technology, 200 East California Blvd., Pasadena, California, 91125 USA7 Dept. of Clinical Neurophysiology, Otto-von-Guericke University, Leipziger Strasse 44, Magdeburg, Germany

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

We investigated the cortical mechanisms of visual-spatial attention while subjects discriminated pat-terned targets within distractor arrays. Functional magnetic resonance imaging (fMRI) was used tomap the boundaries of retinotopic visual areas and to localize attention-related changes in neuralactivity within several of those areas, including primary visual (striate) cortex. Event-relatedpotentials (ERPs) and modeling of their neural sources, however, indicated that the initial sensoryinput to striate cortex at 50–55 milliseconds after the stimulus was not modulated by attention. Theearliest facilitation of attended signals was observed in extrastriate visual areas, at 70–75milliseconds. We hypothesize that the striate cortex modulation found with fMRI may represent adelayed, re-entrant feedback from higher visual areas or a sustained biasing of striate corticalneurons during attention. ERP recordings provide critical temporal information for analyzing thefunctional neuroanatomy of visual attention.

several neuroimaging studies have reported activation in or nearstriate cortex during discrimination tasks in nonselective (thatis, active versus passive) designs21–24. In the present study, neur-al activity associated with spatially focused attention was localizedto both striate and extrastriate visual areas that were positivelyidentified by retinotopic mapping techniques25. ERPs recordedin the same task provided critical information about the timecourse of stimulus-selection processes in these cortical areas.

The spatial attention task used here required subjects to dis-criminate lateralized target stimuli surrounded by distractors ina ‘cluttered’ visual field. The stimuli were 3 × 3 arrays of crossessuperimposed on a background checkerboard pattern (Fig. 1a)that were flashed with equal probability to either the right or leftvisual field in a random sequence at an average rate of two arraysper second. The central element of most (86%) of the arrays wasan upright ‘T’, which was inverted in infrequent (14%) targetarrays. The subject’s task was to maintain fixation on a centralarrow and to attend to the sequence of arrays in the visual fieldindicated by the arrow’s direction. Detections of target arrays inthe attended field were reported by a button press. Stimuli in theopposite field were to be ignored. The direction of the arrow, andthus the subject’s direction of attention, alternated every 20 sec-onds between the left and right visual field during experimentalruns lasting three minutes.

Blood oxygen level-dependent (BOLD)-weighted fMRIimages were acquired during task performance from ten con-tiguous slices extending anteriorly from the occipital pole. Atten-tion effects over the entire group were obtained by transformingeach individual’s image set into standard Talairach coordinates26


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

and averaging the time series of the BOLD signal for each co-reg-istered pixel. Functional maps of brain activity related to thedirection of attention were generated by cross-correlating eachpixel of this group-averaged time series with a model reflectingthe alternating block design of the experiment27.

RESULTSSubjects correctly detected an average of 81 ± 7% of the targets inthe attended visual field. Significant increases in the BOLD signal(reflecting increases in regional cerebral blood flow, rCBF, andhence in neural activity28) were observed in several posterior cor-tical areas in the hemisphere contralateral to the attended visualfield (Fig. 1b). These included the region of the calcarine fissure(which contains the striate cortex), the lingual, middle occipitaland fusiform gyri, and the posterior parietal cortex. To identifythe specific visual areas in which these rCBF changes occurred,the cortical surface of each subject was reconstructed and unfold-ed, and retinotopic areas were mapped onto the flattened sur-face25. Attention-related activations for individual subjects were

then projected onto their retinotopic maps (Fig. 2).In all subjects, the boundaries of the retinotopic visual areas

V1, V2, V3, VP, V3A and V4v could be clearly identified (Fig. 2).Attention-related increases in rCBF in the hemisphere con-tralateral to the attended visual field were observed in all subjectsin area V1 and in most subjects in areas V2, V3, VP and V4v(Table 1). These activations were found at parafoveal retinotopiclocations in both dorsal and ventral cortical areas, correspond-ing to the stimulus position in the visual field. In addition, mostsubjects had significant activation in weakly retinotopic areas ofthe middle occipital gyrus anterior to V3A and in the posteriorfusiform gyrus anterior to V4v, as well as in non-retinotopic pos-terior parietal cortex (Fig. 2).

To obtain converging evidence about the time course of thisattention-related neural activity, in separate sessions we record-ed ERPs time-locked to the attended and unattended stimulusarrays under identical task conditions. For both right and leftvisual field stimuli, the attended arrays elicited enlarged positiveP1 (onset at 70–75 ms) and negative N1 (onset at 130–140 ms)

Fig. 1. Experimental design and attention–related activations. (a) Experimental stimuli and design. Superimposed on the experimental blockdesign are fMRI signal changes associated with attention. Each tracing representsthe averaged time course from five significantly activated pixels in striate cortexof the right (RH) and left (LH) hemisphere. (b) Attention-related activations in arepresentative subject superimposed on corresponding anatomical images. Thelocations of the four selected coronal slices are given in Talairach26 coordinates(y-values). The left hemisphere appears on the left in all images. Intensity of col-ored regions reflects percentage signal change (difference between signal during attend-right and attend-left divided by the total signal) of significantlyactivated areas. Pixels with a time course of activation positively correlated with the task design (that is, showing greater activation during attentionto the left visual field) are shown in the red-to-yellow scale. Those with a time course negatively correlated with the design (that is, showing greateractivation during attention to the right visual field) are displayed in the dark-to-light blue scale. Only pixels correlating at r > ± 0.5 (p < 0.02, cor-rected) are shown. Spatial attention produced contralateral activation foci in the calcarine fissure (calc.), lingual gyrus (ling.), posterior fusiform gyrus(fusi.), middle occipital gyrus (mid. occ.) and posterior parietal lobe (par.).

a b

Table 1. Cortical regions activated by attention

Ventral V1 V2 VP V4v fusi. RH 7, –88, 0 (6) 7, –78, –3 (5) 9, –74, –8 (5) 19, –70, –11 (4) 33, –61, –13 (4)

[150 ± 18] [156 ± 10] [167 ± 10] [167 ± 18] [192 ± 18]

LH –9, –90, –5 (4) –12, –79, –8 (6) –16, –75, –7 (6) –26, –76, –11 (5) –31, –60, –11 (6)[180 ± 20] [150 ± 18] [156 ± 11] [188 ± 16] [211 ± 12]

Dorsal V1 V2 V3 V3A mid. occ. post. par.RH 7, –89, 1 (6) 7, –84, 5 (3) 21, –88, 13 (5) 24, –81, 19 (2) 27, –75, 13 (4) 28, –49, 57 (3)

[193 ± 15] [141 ± 16] [138 ± 8] [125 ± 7] [219 ± 13] [162 ± 11]

LH –8, –91, 0 (6) –10, –85, 0 (5) –22, –85, 14 (5) - - - (0) –29, –75, 19 (5) –29, –55, 50 (3)[162 ± 20] [151 ± 10] [141 ± 9] [219 ± 11] [150 ± 10]

Mean Talairach coordinates of fMRI activation clusters within each visual area in ventral (top) and dorsal (bottom) cortical divisions. Coordinates for clustersin the right (RH) and left (LH) hemispheres are given separately. The total number of subjects (of six) showing significant (p < 0.02, corrected) activation ineach area is given in parentheses. The total brain volume activated within each area (in cubic mm) is shown in brackets ± the standard error of the mean.


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

components over contralateral occipital scalp areas (Fig. 3a), as inprevious studies10,11,29. In contrast, spatial attention did not affectthe amplitude of the earlier C1 component (onset at 50–55 ms).Repeated measures ANOVAs showed significant amplitudeincreases with attention for both early (72–104 ms) and late(104–136 ms) phases of the P1 (F1,18 = 8.9, p < 0.01 andF1,18 = 39.4, p < 0.001, respectively) and for the N1 (F1,18 = 15.0,p < 0.002), particularly over the contralateral scalp. The C1 wasnot significantly modulated by attention (F1,18 = 3.6, n.s.).

We compared the surface voltage topographies of the midlineparieto-occipital C1 and the contralateral occipital P1 attentioneffects (attended minus unattended differences in early and latetime windows; Fig. 3b). From these grand-average topographicaldata, the neural sources of the C1 and of the P1 attention effectwere estimated by dipole modeling using the Brain ElectricalSource Analysis (BESA) algorithm30. As in previousreports10,29,31, a single dipole in each hemisphere within the cal-carine fissure in or near the primary visual cortex accounted forthe C1 component’s voltage topography over the time interval of50–80 ms (Fig. 4). Anatomical localization of dipolar sources wasachieved by projecting the BESA dipole coordinates of the group-average model (Fig 4a) onto the MRIs of seven individuals fol-lowing co-registration of the BESA sphere with the MRI images(Fig. 4b). Dipole positions were converted into Talairach coor-dinates and averaged across all subjects (Fig. 4c; see ref. 29 fordetails).The Talairach coordinates for the left hemisphere C1dipole were –9, –85, 5 and for the right hemisphere dipole were10, –85, 5, both within the calcarine fissure. In contrast, the P1attention effect required two pairs of dipoles for accurate mod-eling of its early and late phases (Fig. 4); the first dipole pair indorsal extrastriate cortex of the middle occipital gyrus (left hemi-sphere, –32, –90, 9; right hemisphere, 33, –90, 10) accounted forthe P1 over the time interval 72–96 ms, and the second pair inthe ventral fusiform area (left hemisphere, –36, –56, –11; righthemisphere, 37, –56, –11) accounted for the time interval104–136 ms.

DISCUSSIONThe fMRI data reported here provide direct evidence for theinvolvement of specific, retinotopically mapped visual corticalareas, including V1, V2, V3, VP and V4v, in spatial-selectiveattention. Whereas previous neuroimaging studies have shownactivation of extrastriate visual areas in attention to location12–16,the present finding of enhanced neural activity in retinotopical-ly mapped striate cortex in a design that separates selective fromnon-selective attention effects has not been reported previously.This engagement of primary visual cortex may be attributed toour use of a difficult discrimination task that requires narrowfocusing of the attentional spotlight in a cluttered visual field (M.Worden & W. Schneider Soc. Neurosci. Abstr. 22, 729.7, 1996).

Our combined ERP and fMRI measurements have implica-tions for the specific mechanisms by which attention to locationmodulates visual information processing in these cortical areas.Intriguingly, no attention-related changes were observed in theamplitude of the short-latency C1 component that reportedlyrepresents the initial afferent response evoked in V1 by visualstimuli29,31, despite the fMRI evidence that spatial attention wasassociated with increased neural activity in area V1. Althoughthe neural generators of surface-recorded ERPs cannot be local-ized with the same degree of certainty as can hemodynamicchanges using fMRI, the localization of the C1’s dipole to the cal-carine fissure, as well as its short onset latency (50 ms) and itsretinotopic properties10,31, are strongly indicative of a source in

area V1. Accordingly, these ERP findings argue against thehypothesis that spatial attention modulates the initial passage ofvisual input from the lateral geniculate nucleus through area V1(refs. 32, 33), even under these cluttered field conditions.

If the modulation of activity in V1 found with fMRI does notrepresent a change in the initial geniculostriate input, what thenis the role of the striate cortex in spatial attention? One hypoth-esis that draws support from both animal19,20 (A.D. Mehta et al.,Soc. Neurosci. Abstr. 23, 121.1, 1997) and human34 studies is thatattentional modulation of striate activity occurs with a longerlatency than the initial evoked response in striate cortex, and rep-resents a delayed or re-entrant feedback of enhanced visual signals

Fig. 2. Retinotopically mapped visual areas and co-localized attentionalactivations. Retinotopic visual areas (top) and regions of increased neuralactivity during spatial attention (bottom) mapped onto flattened corticalrepresentations of the left (LH) and right hemispheres (RH) for the samesubject shown in Fig. 1b. Sulcal cortex, dark gray; gyral cortex, light gray.Retinotopic (blue and yellow) areas representing upper (+) and lower (–)visual fields are located ventrally and dorsally, respectively. Uncoloredareas include retinotopic areas representing unstimulated parts of thevisual field (beyond six degrees of eccentricity) as well as non-retinotopicvisual areas. Attention-related activations (bottom) were determined bycross-correlating pixel time courses with the task block design model;positive correlations (red scale) indicate increased neural activity duringattend-left conditions and negative correlations (blue scale) increasedactivity during attend-right. Only pixels correlating at r > 0.5 (p < 0.02,corrected) are displayed. Dotted white lines on activation maps areboundaries of visual areas traced from field sign maps (top).Abbreviations of cortical regions are as in Fig. 1. The middle occipital(mid. occ.) region included the superior and inferior divisions of the mid-dle occipital gyrus and associated sulci (lateral occipital and lunate).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

back into V1 from higher extrastriate areas. Such adelayed attention effect was not evident in the presentERP recordings, but it could have escaped detection if thestriate cortex source were weak enough to be masked bythe stronger sources that were concurrently active inextrastriate cortex. An alternative hypothesis would bethat the V1 activity observed during attention with fMRIrepresents a top-down ‘bias’ signal that produces a sustainedincrease in neural activity in V1 but does not modulate the initialstimulus-evoked response (for example, refs. 8, 35). Further workis needed to distinguish among these alternative mechanisms.

The ERP results indicate that the earliest facilitation of attend-ed inputs occurs at a level beyond the striate cortex starting at70–75 ms after the stimulus (after the onset of the P1 attentioneffect). The calculated source of this early facilitation was nearthe dorsal occipital foci of fMRI activation in area V3 and moreanterior regions of the middle occipital gyrus. Similar dorsalsources for the P1 attention effect have been reported in studiesthat presented stimuli to the lower14 but not upper29,36 visualfields, suggesting that this early facilitation occurs in retinotopi-cally organized extrastriate areas. In contrast, the source of thelater phase of the P1 effect (104–136 ms) was situated in ventral

occipital cortex in the region of area V4v and posterior fusiformgyrus; this activity may be attributed to enhanced processing ofthe visual target information in ventral areas specialized for pat-tern and object recognition11,13,27,36. The activation foci observedin parietal cortex most likely reflect the engagement of the atten-tional control network that orchestrates the facilitation of attend-ed inputs in extrastriate visual cortex1,6,7.

In sum, these experiments provide evidence that the primaryvisual cortex is involved in spatial attention, but this area doesnot serve as the locus of initial sensory gain control where attend-ed visual inputs are first selectively enhanced. This essential func-tion of ‘attentional amplification’5, which improves theperceptibility of stimuli at attended locations, initially occurs inretinotopically organized extrastriate visual areas. These amplifiedsignals are then routed to higher visual areas, including those of

Fig. 4. Dipole modeling of cortical sources of ERPs. (a) Dipole mod-eling of intracranial sources of the C1 wave (left) and P1 attentioneffects (right). The symmetrical pair of dipoles shown for the C1accounted for 92% of the variance in its scalp voltage distribution inthe unattended grand average waveform over the interval 50–80 msafter stimulus onset. Source waveforms at left of head show timecourse of modeled activity for LH (1) and RH (2) dipoles. The twopairs of dipoles fitting the early (72–104 ms) and late (104–136 ms)phases of the P1 attention effects accounted for 94.8% of the variance of the scalp distribution of the grand average difference waves (attend minusunattend) over the interval 72–136 ms. Source waveforms are shown for the four P1 dipoles in response to left visual field stimuli: (1) early phase,right hemisphere, (2) early phase, left hemisphere, (3) late phase, right hemisphere, (4) late phase, left hemisphere. (b) Projections of calculated right-hemisphere dipolar sources of the C1 wave (left) and P1 attention effects (right) onto corresponding sagittal brain sections of an individual subject.Early P1 dipole, square; late P1 dipole, circle. (c) Projections of calculated dipolar sources averaged across subjects and projected on correspondingsections of the Talairach and Tournoux atlas26.

a C1 dipoles b

c

Fig. 3. Grand-averaged ERP waveforms and scalp topographies.(a) ERP waveforms averaged over all subjects in response tostandard (non-target) stimuli in the left visual field. Equivalentwaveforms were elicited by right visual field stimuli. Recordingsshown are from electrodes at occipitotemporal (TO1/TO2),temporal (T5/T6) and occipitoparietal (IPz) sites. Other sitesare indicated as dots on the head icon. (b) Spline-interpolatedvoltage maps derived from the grand averaged waveformsshown in (a). Color scales are in microvolts. Left map showsvoltage topography averaged over the time window 50–80 msfor the unattended ERPs to left visual field stimuli. DistinctiveC1 and P1 distributions are evident. Center and right mapsshow the distributions of the early and late P1 attention effectsas manifested in the difference waves formed by subtracting theERP to standard left visual field stimuli when unattended fromthe ERP to the same stimuli when attended.

a

b

400 ms

–1.0 mV

P1 attention effect


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

the occipitotemporal ventral stream, to gain preferential accessto limited-capacity stages of feature analysis and pattern recog-nition. Bringing together the anatomical specificity of fMRI map-ping and the time resolution of ERP recordings makes it possibleto characterize the functional roles of specific brain areas in cog-nitive processes such as selective attention.

METHODSfMRI procedures and data analysis. Six subjects (5 female, age range23–41 years) gave written informed consent before participating in thefMRI experiment. Subjects were selected on the basis of their ability tomaintain steady control of fixation as assessed by electro-oculographicrecordings in pilot sessions. During fMRI scanning, eye movement wasmonitored continuously using an infrared-sensitive video camera sys-tem with a sensitivity of ± 0.5 degrees of visual angle. Runs withdetectable eye movements were discarded and repeated.

The task stimuli were back-projected onto a screen at the foot of themagnet bore. Subjects viewed the stimuli via a mirror attached to thehead coil. Each stimulus subtended 5.5 degrees of visual angle, and theinnermost edge appeared 1.7 degrees to the left or right of fixation. Stim-uli were presented in randomized sequences to either the left or rightvisual field with onset asynchronies varying between 400 and 600 ms.Stimulus duration was 100 ms.

Anatomical and functional images were acquired with a 1.5-T SiemensVISION MR scanner equipped with a 26 cm-diameter circularly polar-ized head coil. BOLD-weighted images were acquired with an echo pla-nar imaging sequence (TR = 2500 ms, TE = 64 ms, flip angle = 90degrees) in the coronal plane (2.5 × 2.5 mm in-plane resolution). Sev-enty-four repetitions on each of ten 5-mm slices were acquired duringeach three-minute run; the first two repetitions were not used in dataanalysis. For anatomical localization, high-resolution (1 × 1 × 1mm)T1-weighted images were acquired using a three-dimensional magne-tization-prepared rapid gradient echo sequence (TR = 11.4 ms, TE =4.4 ms, flip angle = 10 degrees). Both anatomical and BOLD-weightedimages were transformed into the standardized coordinate system ofTalairach and Tournoux26.

Time-dependent echo planar images were post-processed with AFNIsoftware37. Following in-plane motion correction, the raw time-seriesdata from each of four runs collected from every subject were averagedindividually. Group data were obtained by averaging the time series overall subjects. A series of phase-shifted trapezoids representing the peri-odic alternation of conditions (attend-right, attend-left) in the blockdesign of the experiment were used as reference waveforms. Each trape-zoid function was correlated on a pixel-by-pixel basis with the averaged(individual or group) signal-strength time series by a least-squares fit togenerate a functional intensity map. Gram–Schmitt orthogonalizationwas used to remove linear drift in the time series 27.

Significance levels of attention-related activations were determined byusing a region of interest (ROI) analysis based on data from four pilotsubjects. We identified anatomical regions of activation in these subjectsand defined a single, large-volume ROI (28 ml) within the occipital cor-tex, which included the calcarine fissure, collateral sulcus, lingual gyrusand middle occipital gyrus. A conservative statistical correction (Bon-ferroni) based on the number of ROI voxels was applied for determin-ing significance levels of attention-related activations in both individualand group data. These activations were considered significant for pixelscorrelating with the direction of attention at r > 0.5 (corrected p < 0.02).

Retinotopic mapping of visual areas. In a separate session, we obtainedBOLD-weighted images while subjects viewed a slowly rotating checker-board wedge and a dilating checkerboard circle. The periodic activationsproduced by these stimuli were used to calculate the borders of the retino-topically organized visual areas based on whether they contain a mirror-image, or non-mirror-image representation of the visual field (see ref.25 for details).

ERP procedures and data analysis. ERPs were recorded during task per-formance from a group of 19 subjects (13 female, age range 18–41 years)including the 6 studied with fMRI. Recordings were made from 41 scalp

sites with an amplifier bandpass of 0.01–80 Hz. ERPs elicited by the sametask stimuli as in the fMRI experiment were averaged separately accord-ing to field of stimulus and direction of attention, and grand-averagedover all subjects. Trials with eye movements or other artifacts were reject-ed off-line. ERP components were quantified as mean amplitudes in spe-cific time windows relative to a 100-ms prestimulus baseline. Meanamplitudes of C1 (50–80 ms), P1 in its early (72–104 ms) and late(104–136 ms) phases, and N1 (150–180 ms) were analyzed by repeated-measures ANOVAs with factors of attention (same stimulus when attend-ed and unattended), visual field (left and right), electrode site (20 pairsin each hemisphere) and hemisphere (left and right). Estimation of thedipolar sources of early ERP components was done using the BESA algo-rithm as described29,38. Modeling of the C1 component was done joint-ly on the waveforms elicited by unattended left and right visual-fieldstandard stimuli (Fig. 3a). The interval between 50–80 ms was simulta-neously with two dipoles, one in each hemisphere, that were constrainedto have mirror-symmetrical locations and orientations. The attentionaldifference waves (attended minus unattended amplitudes) for left andright visual-field stimuli were used to fit the P1 attention effect. The early(72–104 ms) and late (104–136 ms) phases of the P1 were fited sequen-tially, each with a pair of dipoles constrained to be mirror-symmetrical inlocation but allowed to vary in orientation. Different dipole-fitting strate-gies that included relaxing symmetry constraints and using differentstarting locations yielded highly similar dipole configurations.

ACKNOWLEDGEMENTSWe thank Matt Marlow, Cecelia Kemper and Carlos Nava for technical

assistance. Supported by grants from NIMH (MH25594), ONR (N00014-93-

0942), NIH (NS36722), HMRI, and from the Deutsch Forschungsgemeinschaft

(HE 1531/3).

RECEIVED 18 OCTOBER 1998, ACCEPTED 18 FEBRUARY 1999

1. LaBerge, D. Attention Processing: The Brain’s Art of Mindfulness (HarvardUniv. Press, Cambridge, Massachusetts, 1995).

2. Luck, S. J., Hillyard, S. A., Mouloua, M. & Hawkins, H. L. Mechanisms ofvisual-spatial attention: Resource allocation or uncertainty reduction. J. Exp.Psychol. Hum. Percept. Perform. 22, 725–737 (1996).

3. Posner, M. I., Snyder, C. R. & Davidson, B. J. Attention and the detection ofsignals. J. Exp. Psychol. Gen. 109, 160–174 (1980).

4. Eriksen, C. W. & St. James, J. D. Visual attention within and around the fieldof focal attention: A zoom lens model. Percept. Psychophysics 40, 225–240(1986).

5. Posner, M. I. & Dehaene, S. Attentional networks. Trends Neurosci. 17, 75–79(1994).

6. Corbetta, M. Frontoparietal cortical networks for directing attention and theeye to visual locations: Identical, independent, or overlapping neuralsystems? Proc. Natl. Acad. Sci. USA 95, 831–838 (1998).

7. Nobre, A. C. et al. Functional localization of the system for visuospatialattention using positron emission tomography. Brain 120, 515–533 (1997).

8. Luck, S. J., Chelazzi, L., Hillyard, S. A. & Desimone, R. Neural mechanisms ofspatial selective attention in areas V1, V2, and V4 of Macaque visual cortex. J.Neurophysiol. 77, 24–42 (1997).

9. Colby, C. L. The neuroanatomy and neurophysiology of attention. J. ChildNeurol. 6, 90–118 (1991).

10. Mangun, G. R. Neural mechanisms of visual selective attention.Psychophysiology 32, 4–18 (1995).

11. Hillyard, S. A. & Anllo-Vento, L. Event-related brain potentials in the study ofvisual selective attention. Proc. Natl. Acad. Sci. USA 95, 781–787 (1998).

12. Heinze, H. J. et al. Combined spatial and temporal imaging of brain activityduring visual selective attention in humans. Nature 372, 543–546 (1994).

13. Mangun, G. R. et al. Covariations in ERP and PET measures of spatialselective attention in human extrastriate visual cortex. Hum. Brain Mapp. 5,273–279 (1997).

14. Woldorff, M. G. et al. Retinotopic organization of the early visual-spatialattention effects as revealed by PET and ERPs. Hum. Brain Mapp. 5, 280–286(1997).

15. Kastner, S., De Weerd, P., Desimone, R. & Ungerleider, L. Mechanisms ofdirected attention in the human extrastriate cortex as revealed by functionalMRI. Science 282, 108–111 (1998).

16. Mangun, G. R., Buonocore, M. H., Girelli, M. & Jha, A. ERP and fMRImeasures of visual spatial selective attention. Hum. Brain Mapp. 6, 383–389(1998).

17. Gratton, G. Attention and probability effects in the human occipital cortex:An optical imaging study. Neuroreport 8, 1749–1753 (1997).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

18. Motter, B. C. in The Attentive Brain (ed. Parasuraman, R.) 51–69 (MIT Press,Cambridge, Massachusetts, 1998).

19. Roelfsema, P. R., Lamme, V. A. & Spekreijse, H. Object-based attention in theprimary visual cortex of the macaque monkey. Nature 395, 376–381 (1997).

20. Vidyasagar, T. R. Gating of neuronal responses in macaque primary visualcortex by an attentional spotlight. Neuroreport 9, 1947–1952 (1998).

21. Shulman, G. L. et al. Top-down modulation of early sensory cortex. Cereb.Cortex 7, 193–206 (1997).

22. Dupont, P. et al. Different perceptual tasks performed with the same visualstimulus attribute activate different regions of the human brain: A positronemission tomography study. Proc. Natl. Acad. Sci. USA 90, 10927–10931 (1993).

23. Watanabe, T. et al. Attention-regulated activity in human primary visualcortex. J. Neurophysiol. 79, 2218–2221 (1998).

24. Watanabe, T. et al. Task-dependent influences of attention on the activationof human primary visual cortex. Proc. Natl. Acad. Sci. USA 95, 11489–11492(1998).

25. Sereno, M. I. et al. Borders of multiple visual areas in humans revealed byfunctional magnetic resonance imaging. Science 268, 889–893 (1995).

26. Talairach, J. & Tournoux, P. Co-Planar Stereotaxic Atlas of the Human Brain:3-Dimensional proportional system: An approach to cerebral imaging. (Thieme,New York, 1988).

27. Bandettini, P. A., Jesmanowicz, A., Wong, E. C. & Hyde, J. S. Processingstrategies for time-course data sets in functional MRI of the human brain.Magn. Reson. Med. 30, 161–173 (1993).

28. Buxton, R. & Frank, L. A model for the coupling between cerebral blood flowand oxygen metabolism during neural stimulation. J. Cereb. Blood FlowMetab. 17, 64–72 (1997).

29. Clark, V. P. & Hillyard, S. A. Spatial selective attention affects earlyextrastriate but not striate components of the visual evoked potential. J. Cogn.Neurosci. 8, 387–402 (1996).

30. Scherg, M. in Auditory Evoked Magnetic Fields and Electric Potentials (eds.Grandori, F., Hoke, M. & Roman, G. L.) 40–69 (Karger, Basel, 1990).

31. Clark, V. P., Fan, S. & Hillyard, S. A. Identification of early visually evokedpotential generators by retinotopic and topographic analysis. Hum. BrainMapp. 2, 170–187 (1995).

32. Crick, F. Function of the thalamic reticular complex: The searchlighthypothesis. Proc. Natl. Acad. Sci. USA 81, 4586–4590 (1984).

33. Skinner, J. E. & Yingling, C. D. in Attention, Voluntary Contraction and Event-Related Cerebral Potentials. (ed. Desmedt, J. E.) 30–69 (Karger, Basel, 1977).

34. Aine, C. J., Supek, S. & George, J. S. Temporal dynamics of visual-evokedneuromagnetic sources: Effects of stimulus parameters and selectiveattention. Int. J. Neurosci. 80, 79–104 (1995).

35. Rees, G., Frackowiak, R. & Frith, C. Two modulatory effects of attention thatmediate object categorization in human cortex. Science 275, 835–838 (1997).

36. Heinze, H. J., Luck, S. J., Mangun, G. R. & Hillyard, S. A. Visual event-relatedpotentials index focused attention within bilateral stimulus arrays. I.Evidence for early selection. Electroencephalogr. Clin. Neurophysiol. 75,511–527 (1990).

37. Cox, R. W. AFNI—Software for analysis and visualization of functionalmagnetic resonance neuroimages. Computers Biomed. Res. 29, 162–173(1996).

38. Anllo-Vento, L., Luck, S. J. & Hillyard, S. A. Spatio-temporal dynamics ofattention to color: Evidence from human electrophysiology. Hum. BrainMapp. 6, 216–238 (1998).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Helmholtz1 and William James2 each noted the potential disso-ciation between the point of gaze fixation and the focus of atten-tion within the field of view. More recently, this phenomenonhas been compared to a ‘spotlight’ that evokes attentionalenhancement of visual information within a circumscribedregion of visual space or within the confines of a target object3–5.Information inside the spotlight is processed more quickly ormore efficiently, whereas outside the spotlight, information isprocessed “less, or differently, or not at all”6–8. However, thephysiological basis of this spatially restricted attentional effecthas remained obscure. Both neuroimaging and single-neuronstudies have shown that visual attention can modulate respons-es in a number of visual areas9–17. It still remains controversialwhether true attentional modulation occurs at the earliest stagesof cortical processing such as primary visual cortex18. Althoughattention-related shifts of cortical activation have been observedpreviously19–21, there has not been a convincing demonstrationthat such shifts have a precise spatial metric concordant with asubject’s ability to accurately move the focus of attention. Here,we report experiments in which we used fMRI with human sub-jects to demonstrate focal enhancement of cortical activity thatmoves in precise register with covert shifts in the focus of atten-tion. We show that these shifts are metrically accurate withinthe cortical representations of the visual field found in occipi-totemporal visual cortex. (An animation of the cortical pro-gression of attentional enhancement can be viewed athttp://www.mcw.edu/cellbio/visionlab.)

RESULTSTo study visuospatial attention, we used a task in which the sub-ject’s gaze remained fixated on a central marker while spatial atten-tion was directed to a cued location (target segment) within anarray of segments (Fig. 1). Subjects detected specific color/orien-tation conjunctions (for example, blue-horizontal) within thecued segment while ignoring other uncued segments. On aver-age, subjects responded correctly on 85% of the trials. Over a peri-od of 40 seconds, the cued segment was shifted to successively

A physiological correlate of the‘spotlight’ of visual attention

Julie A. Brefczynski and Edgar A. DeYoe

Department of Cellular Biology, Neurobiology and Anatomy, Medical College of Wisconsin, 8701 Watertown Plank Road, Milwaukee, Wisconsin53226, USA

Correspondence should be addressed to E.A.D. (deyoe@ mcw.edu)

Here we identify a neural correlate of the ability to precisely direct visual attention to locations otherthan the center of gaze. Human subjects performed a task requiring shifts of visual attention (butnot of gaze) from one location to the next within a dense array of targets and distracters while func-tional MRI was used to map corresponding displacements of neural activation within visual cortex.The cortical topography of the purely attention-driven activity precisely matched the topography ofactivity evoked by the cued targets when presented in isolation. Such retinotopic mapping of atten-tion-related activation was found in primary visual cortex, as well as in dorsomedial and ventraloccipital visual areas previously implicated in processing the attended target features. These resultsidentify a physiological basis for the effects of spatially directed visual attention.

greater eccentricities (Fig. 1). The complete sequence of shifts wasrepeated 5 times within each 200-second fMRI scan run.

Repeated shifts of attention to targets at greater eccentricitiesin the right visual field produced cyclic cortical enhancementthat was spatially mapped in both striate (V1) and extrastriatecortex of the left occipital lobe (Fig. 1). As attention shifted fromthe perifovea to the periphery, the locus of cortical enhancementshifted anteriorly away from the occipital pole. Although theamplitude of response varied from position to position, the pro-gression was reminiscent of the retinotopic mapping of visualfield eccentricity observed previously in human visual cortex22–25.Such spatially mapped attentional modulation was observed in allfive subjects.

The region of attentional modulation extended throughoutmedial occipital cortex as well as ventrally into, and surround-ing, the collateral sulcus. Based on previous retinotopic map-ping24, this swath traversed portions of V1 and proposedextrastriate visual areas V2, V3, VP, V4v and sometimes cortexanterior to V4v. Attentional modulation was also seen deep with-in the left calcarine fissure (Fig. 3e), verifying that V1 as well asextrastriate cortex was modulated by the shifting focus of atten-tion. Mean Talairach coordinates26 for the anterior–posteriorextremes of the swath of enhancement in medial cortex (V1/V2)were (+x is left, +y is posterior, +z is superior) 15.4, 75.0, 10.0mm and 20.4, 96.8, –4.2 mm, respectively. For ventral cortex, theswath of enhancement extended from 19.8, 62.8, –8.0 mm to24.6, 89.0, –16.8 mm. (In the following analysis, we have simplycompared responses in medial occipital cortex versus ventraloccipitotemporal cortex, and have used the terms medial andventral cortex to refer to these composite regions.)

To determine conclusively if the pattern of attentionalenhancement followed the cortical retinotopy of single cued seg-ments, we repeated the experiment but with only one cued seg-ment present at a time (Fig. 1). On this task, subjects respondedcorrectly on 88% of the trials. The resulting spatial pattern ofactivation closely matched the pattern produced when only atten-tion was shifted from segment to segment (Fig. 1, compare left


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


to right). To provide a quantitative measure of the match betweenthe two data sets, we compared the response phase (represent-ing eccentricity) for individual voxels pooled across all subjects.Temporal phases were highly correlated for attention-only ver-sus single-segment responses in both medial and ventral cortex(Fig. 2). The correlation coefficient (r) was 0.94 for medial cortexand 0.96 for ventral cortex, thereby indicating a very close matchbetween the locus of attentional enhancement and the retino-topy (visual field topography) of the individual segments pre-sented in isolation. Together, these data show that the attentionalmapping can be readily observed in individual subjects, but isalso consistent across all subjects.

To test these results further, three subjects performed a taskin which the attentional sequence was directed into the oppositevisual field. A comparison between rightward and leftward atten-tional sequences (Fig. 3a and b) shows that the regular patternof phase mapping switched accordingly from the left to the righthemisphere. If instead, attention remained directed toward thefixation point throughout the scan, no consistent cyclic activa-tion was detected (Fig. 3c). In the leftward and rightwardsequences, we noted that activation was sometimes apparent at

the homotopic location in the opposite hemisphere, though thiswas weaker and poorly phase mapped. This may reflect a mildsuppressive effect at the mirror-imaged attentional focus.

To explicitly identify stimulus-related activation rather thanattentional modulation, we presented the whole stimulus arrayfor 20-second blocks alternated with comparable blocks con-sisting of the fixation point alone on an isoluminant gray field.This evoked strong activation uniformly throughout occipital

articles

Fig. 1. Retinotopic attentional mod-ulation compared to activationevoked by cued targets alone. Cuedsegment (left column) showsschematic sequence of target seg-ments cued for attentional scrutiny.Stimulus (bottom left) is actual targetarray. FMRI signal (left) shows signalmodulation of individual voxels atsites indicated on adjacent correla-tion maps. Temporal phase shift ofthe signal at each site identifies thecorresponding locus of attention.Correlation maps show sites wheretiming of modulation was positivelycorrelated (red) or anti-correlated(blue) with the timing of attentionalshifts. Displayed segment (right col-umn) shows schematic sequence ofsingle segments presented duringotherwise identical control experi-ment. Composite of single segmentsshown in stimulus bottom right. FMRISignal and correlation map on rightshow results of control experiment.Structural MRI (bottom) is aparasagittal section (13.6 mm left ofmidline) through occipital lobe insame plane as correlation maps.Sulcal landmarks; CaS, calcarine sul-cus; CoS, collateral sulcus; POS; pari-eto-occipital sulcus.

Fig. 2. Comparison of visual field topography (coded by temporal phaseof fMRI response) for attentional foci (y-axis) versus single segments (x-axis). Each circle represents a single responsive voxel. Data are pooledacross subjects. Left column, medial occipital cortex consisting primarilyof V1 and V2. Right column, ventral occipitotemporal cortex within andsurrounding the collateral sulcus. Top row, shifts in eccentricity. Bottomrow, shifts in polar angle.

Medial occipital Ventral occipital

Ecce

ntr

icit

yP

ola

r an

gle

Att

enti

on

al p

has

eA

tten

tio

nal

ph

ase

Single seg. phase Single seg. phase

Attention Single segment

fMRIsignal

Correlationmap

Correlationmap

fMRIsignal

Displayed segment

Cued segment

fMRI correlation

200 s200 s

r = 0.94 r = 0.96

r = 0.97 r = 0.87

1′

2′

3′

4′

1

2

3

4


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

visual cortex, illustrated in a ventral slice (Fig. 3d). This dif-fuse activation was in marked contrast to the focal, orderedpattern of activation caused by the shifts of visual attention.In sum, activation produced by pure attentional shifts (Fig. 3aand b) consisted of a modulation of the ongoing, stimulus-related activation (Fig. 3d).

We were concerned that uncontrolled eye movements, espe-cially toward the cued segment, might produce an artifactualresponse modulation that could be mistaken for attentionalretinotopy. To test this, we used an infrared eye tracker to mon-itor subjects’ eye movements during performance of the atten-tional task outside the scanner. There was no detectable patternof eye movements that correlated with the sequence of attentionalshifts. Although tiny eye movements below the resolution of oureye tracker could not be ruled out, such instabilities could not

produce the large-scale retinotopicorganization observed here.

We next compared the strengthof attentional modulation to themagnitude of response for the iso-lated single segments. The atten-tional enhancement was modestlybut significantly stronger in ventralcortex compared to medial cortex(Fig. 4a). We averaged the meanfMRI signals across comparableregions-of-interest in five subjects(Fig. 4b). Although the mean atten-tional modulation for all active vox-els was less than half thesingle-segment response, the ampli-tude in select voxels of ventral cor-tex could be as large as the responseevoked by the isolated single seg-ments (for example, site 1, 1′ inFig. 1). This was not the case formedial cortex. This suggests that, insome restricted brain locations, theattentional effects may be compara-ble to the stimulus effects themselves.

Finally, we repeated the atten-tional-shift experiment using a cir-cumferential cue pattern thatrequired the subject to shift atten-

tion to segments at successive polar angles. The observed pat-tern of activation (Fig. 5a) matched the known corticalrepresentation of polar angle in which the inferior quadrant isrepresented dorsally and the superior quadrant is representedventrally in medial occipital cortex24. Again the topography ofthe attentional shifts closely matched the topography of the iso-lated cued segments (compare Fig. 5a and b). Accordingly, thetemporal phase of the fMRI response (representing polar angle)for individual voxels was highly correlated in both medial andventral occipital cortex (r = 0.91 and 0.86 respectively; Fig. 2).

DISCUSSIONOur results show that attention directed to a specific target loca-tion in the visual field produced multiple foci of cortical enhance-ment in occipital visual cortex. The positions of these foci within

Fig. 3. Further characterization of attentionaleffects. (a) Retinotopically mapped activation due toshifts of attention into the right visual field for a sin-gle subject. Colors of activated voxels (right panel)correspond to attentional focus (left schematic) thatproduced greatest modulation. (b) Activation pro-duced by attentional shifts into the left visual field.(c) Activation produced by attention maintained at fixation point. (d) Activation produced by presentationof whole target array alternated every 20 seconds with fixation point alone. Note diffuse activation in bothhemispheres that contrasts with focal, retinotopic activation produced in (a) and (b) (denoted by whitebrackets). (e) Example of attentional modulation within the depths of the calcarine sulcus, unequivocallyassociated with V1.

a

b

c

d

e

Right hemifield

Left hemifield

Passive fixation

Block design

V1 modulation

Fig. 4. Amplitude and timecourse of attentional modulation. (a) Averagemodulation of fMRI signals produced by shifts of attention versus singlesegments for medial versus ventral occipital cortex. *Significant differencebetween medial and ventral cortex (t = 3.5, p < 0.01, two-tail). (b) Waveforms of fMRI signals with phase delays corresponding to themost peripheral cued segment, averaged across subjects for medial versusventral occipital cortex. Amplitude is in normalized fMRI units.

Modulation amplitude b

No

rmal

ized

am

plit

ud

e

a Average waveforms

Medial Ventral

Sin

gle

seg

men

tA

tten

tio

n o

nly

Am

plit

ud

eA

mp

litu

de

Time (s) Time (s)

**


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

the cortex corresponded precisely with the cortical representa-tions of the attended target presented in isolation. When the focusof attention shifted, the cortical enhancement shifted in precisecorrespondence. Because this study used a dense array of visualtargets filling the field of view, the resulting focal activationreflected the spatial characteristics of the attentional modulationitself, not a ‘filtering’ of the modulation through a pattern of spa-tially isolated stimulus features. In sum, these results not onlydemonstrated a physiological correlate of spatial attention, butalso showed that it is accurately described using a retinotopicmetric. In principle, it should be possible to use these results todetermine the locus of attentional scrutiny from the pattern ofbrain activation alone.

Attentional enhancement was strongest for a small subset ofvoxels in ventral cortex encompassing a region that is active inprocessing the color and orientation attributes that are selectedfor attentional scrutiny27–31. However, retinotopically mappedattentional effects were also seen at the earliest stages of corticalprocessing in the medial occipital lobe, including V1. The expe-rience of shifting one’s focus of attention could therefore reflectneural events occurring at a variety of cortical sites, even prima-ry visual cortex. (We did not examine the lateral geniculate nucle-us.) However, recent psychophysical work indicates that thespatial resolution of focal attention is not as fine as the spatialresolution of cells in V1, at least for certain tasks32. This suggeststhat the attentional experience is more likely to be linked to neur-al events at cortical stages beyond V1. Indeed, our own observa-tions showed that the largest attentional effects were found in a

small subset of voxels in ventral occipitotemporal cortex. Theseconclusions are generally in accord with those of a recent report33.

The present findings do not necessarily imply that the per-ceived attentional spotlight must be two-dimensional, or that itcorresponds to an undifferentiated region of space. Within a givenretinotopic zone of cortex, distinct populations of cells repre-senting different distances, objects, features or surfaces may beselectively modulated depending on the attentional task. Thus,it is possible for attentional effects to be both retinotopic and‘object-based’5, especially for two-dimensional displays such asthose used in this study. Under other circumstances, the perceivedwindow of attention, as well as the properties of the cortical acti-vation, could depart significantly from the analogy of an atten-tional spotlight.

METHODSSubjects, stimulus and task. Seven subjects (four male and three female)drawn from the Medical College of Wisconsin faculty and students par-ticipated in this study. (Not all subjects participated in all tests.) Informedconsent was obtained from all subjects in accordance with proceduresand protocols approved by the Medical College of Wisconsin internalreview committee.

Subjects used a custom-designed optical system34 to view a computergraphics stimulus array generated with a Cambridge Instruments VSGvideo board driving a modified Sharp XG-2000U video projector. Thestimulus array depicted in Fig. 1 consisted of either 6 (eccentricity map-ping) or 8 (polar angle mapping) sectors, each containing 4 target seg-ments, altogether subtending 56° of visual angle. The sizes and stripeperiods of the segments were varied to compensate for correspondingchanges in cortical magnification factor and spatial acuity at increasingeccentricities from the center of gaze. Color differences (blue versusorange) were clearly evident even at the largest eccentricities. Every twoseconds, the color and/or stripe orientation of each segment could changerandomly. The subject’s task was to fixate the central white cross butmonitor the color/orientation pattern of a segment designated by a pre-arranged audio cue (“one”, “two”, “three” or “four”) presented every tenseconds via custom electrostatic headphones (Koss Inc.). No visual aspectof the stimulus array could be used to identify the cued segments. Thesubject pressed one of two buttons to indicate the observed conjunction(blue-horizontal or orange-vertical versus blue-vertical or orange-hori-zontal). This feature-conjunction task ensured that focal attention wasengaged and directed toward the cued segment35.

To map the effects of shifting focal attention, the target segment wascued in one of two sequential patterns, either along the horizontal merid-ian at successively increasing eccentricities or along a circumference atsuccessive counterclockwise positions. At each cued location, the subjectmade five successive judgments (one every two seconds). The completesequence of four cued locations was then repeated five times within each200-second fMRI scan.

Imaging and data analysis. Gradient-recalled, echo-planar, functionalimaging was done with a General Electric (Milwaukee, Wisconsin) Signa1.5-tesla MRI scanner equipped with a custom RF/gradient head coilacquiring 102 gradient-recalled (TE = 40 ms,TR = 2 s, FA = 90o) echo-planar images with 3.75 × 3.75 × 6.0 mm resolution. Twelve slices spanningoccipitotemporal cortex were collected in the axial plane. (Two of thepolar angle experiments were collected in the coronal plane with identicalslice thickness.) Anatomical images were obtained using a T1-weightedspoiled GRASS (gradient recalled at steady state) pulse sequence at a res-olution of 256 × 192 × 1.0–1.2 mm depending on subject brain size.

After coregistering successive fMRI images to reduce motion artifacts,activated voxels were identified by cross-correlation with an idealizedresponse waveform36 based on a smoothed and delayed version of thecue timing sequence. The temporal phase of the fMRI response was deter-mined using a custom algorithm based on the Hilbert transform37.Response amplitude for each voxel was estimated as the covariance ofthe fMRI response with the idealized response waveform. Phase mapswere constructed in one of two ways. For Fig. 1, voxels were pseudo-col-

Fig. 5. Example of retinotopic mapping of cortical activation in an indi-vidual subject. (a) Activity produced by shifts of attention to cued tar-gets increasing in polar angle or (b) by same targets presented inisolation. Anatomical slice is parasagittal, 17 mm left of midline.Conditions matched those for the eccentricity mapping experiment(Fig. 1) except that targets were cued in a circumferential sequence(within the third ring of segments) proceeding from inferior to superiorwithin the left visual field. Resulting fMRI activation in the right hemi-sphere shifted sequentially from dorsal to ventral occipital cortex (red,target at 202.5o; yellow, target at 247.5o; green, target at 292.5o; blue,target at 337.5o; zero reference at superior vertical meridian).Quantitative comparison of retinotopy for attentional shifts versus sin-gle segments is illustrated in Fig. 2. POS, parieto-occipital sulcus.

Attention only Single segmenta b

POSPOS


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

ored using a color scale representing the magnitude and sign of the cor-relation coefficient and then subsampled, smoothed and interpolatedusing a gaussian filter of half-width equal to the original voxel size. Anoutline drawing of the primary sulcal landmarks in the section of inter-est was then overlaid on the resulting correlation map. Such maps areadvantageous in that no arbitrary thresholding is applied to the data,thereby retaining maximum sensitivity. However, responses for only onetemporal phase can be shown on each map. In Figs. 3 and 5, differentcolors were used to code different temporal phase ranges for voxels whosecorrelation exceeded a minimum threshold (r > 0.35). This allowed con-struction of composite figures showing responses to different phases onthe same map.

It is important to note that the phase-mapping technique used hereto identify retinotopic organization does not distinguish cyclic enhance-ment (increased activation) from cyclic suppression (decreased activa-tion) or from alternating enhancement and suppression. However, directinspection of typical response waveforms indicated that the primaryattentional effect was enhancement.

ACKNOWLEDGEMENTSOur thanks to Jon Wieser for technical assistance. Supported by NIH grants

EY10244 and MH51358 to EAD and a Keck Foundation grant to the Medical

College of Wisconsin.


1. von Helmholtz, H. Treatise on Physiological Optics Vol. III (Dover, New York,1910).

2. James, W. The Principles of Psychology Vol. I (Classics of Psychiatry andBehavioral Sciences Library, Birmingham, 1890).

3. LaBerge, D. Spatial extent of attention to letters and words. J. Exp. Psychol.Hum. Percept. Perform. 9, 371–379 (1983).

4. Crick, F. Function of the thalamic reticular complex: The searchlighthypothesis. Proc. Natl. Acad. Sci. USA 81, 4586–4590 (1984).

5. Duncan, J. Selective attention and the organization of visual information. J.Exp. Psychol. Gen. 113, 501–517 (1984).

6. Posner, M. I. & Petersen, S. E. The attention system of the human brain.Annu. Rev. Neurosci. 13, 25–42 (1990).

7. Crick, F. The Astonishing Hypothesis: The Scientific Search for the Soul(Scribner, New York, 1994).

8. Luck, S. J. & Hillyard, S. A. The role of attention in feature detection andconjunction discrimination: an electrophysiological analysis. J. Neurosci. 80,281–297 (1995).

9. Desimone, R., Wessinger, M., Thomas, L. & Schneider, W. Attentional controlof visual perception: cortical and subcortical mechanisms. Cold SpringHarbor Symp. Quant. Biol. 55, 963–971 (1990).

10. Corbetta, M. et al. Selective and divided attention during visualdiscrimination of shape, color, and speed: functional anatomy by positronemission tomography. J. Neurosci. 11, 2383–2402 (1991).

11. Corbetta, M., Miezin, F. M., Shulman, G. L. & Petersen, S. E. A PET study ofvisuospatial attention. J. Neurosci. 13, 1202–1226 (1993).

12. Motter, B. C. Neural correlates of attentive selection for color or luminance inextrastriate area V4. J. Neurosci. 14, 2178–2189 (1994).

13. Treue, S. & Maunsell, J. H. R. Attentional modulation of visual motionprocessing in cortical areas MT and MST. Nature 382, 539–541 (1996).

14. Beauchamp, M. S., Cox, R. W. & DeYoe, E. A. Graded effects of spatial andfeatural attention on human area MT and associated motion processingareas. J. Neurophysiol. 78, 516–520 (1997).

15. O’Craven, K. M. et al. Voluntary attention modulates fMRI activity in humanMT-MST. Neuron 18, 591–598 (1997).

16. Kastner, S., De Weerd, P., Desimone, R. & Ungerleider, L. G. Mechanisms ofdirected attention in the human extrastriate cortex as revealed by functionalMRI. Science 282, 108–111 (1998).

17. McAdams, C. J. & Maunsell, J. H. R. Effects of attention on orientation-tuning function of single neurons in macaque cortical area V4. J. Neurosci.19, 431–441 (1999).

18. Hillyard, S. A. Combining steady-state visual evoked potentials and fMRI tolocalize brain activity during selective attention. Hum. Brain Mapp. 5,287–292 (1997).

19. Heinze, H. J. et al. Combined spatial and temporal imaging of brain activityduring visual selective attention in humans. Nature 372, 543–546 (1994).

20. Mangun, G. R. et al. Covariations in ERP and PET measures of spatialselective attention in human extrastriate visual cortex. Hum. Brain Mapp. 5,273–279 (1997).

21. Woldorf, M. G. et al. Retinotopic organization of early visual spatial attentioneffects as revealed by PET and ERPs. Hum. Brain Mapp. 5, 280–286 (1997).

22. Schneider, W., Noll, D. C. & Cohen, J. D. Functional topographic mapping ofthe cortical ribbon in human vision with conventional MRI scanners. Nature365, 150–153 (1993).

23. Sereno, M. I. et al. Borders of multiple visual areas in humans revealed byfunctional MRI. Science 268, 889–893 (1995).

24. DeYoe, E. A. et al. Mapping striate and extrastriate visual areas in humancerebral cortex. Proc. Natl. Acad. Sci. USA 93, 2382–2386 (1996).

25. Engel, S. A., Glover, G. H. & Wandell, B. A. Retinotopic organization inhuman visual cortex and the spatial precision of functional MRI. Cereb.Cortex 7, 181–192 (1997).

26. Talairach, J. & Tournoux, P. Co-Planar Stereotaxic Atlas of the Human Brain(Thieme, New York, 1988).

27. Desimone, R. & Schein, S. J. Visual properties of neurons in area V4 of themacaque: Sensitivity to stimulus form. J. Neurophysiol. 57, 835–868 (1987).

28. Lueck, C. J. et al. The colour centre in the cerebral cortex of man. Nature 340,386–389 (1989).

29. Schein, S. J. & Desimone, R. Spectral properties of V4 neurons in themacaque. J. Neurosci. 10, 3369–3389 (1990).

30. Beauchamp, M. S., Haxby, J. V., Jennings, J. & DeYoe, E. A. An FMRIadaptation of the Farnsworth-Munsell 100 hue test reveals human color-selective areas. Cereb. Cortex (in press).

31. Hadjikhani, N. et al. Retinotopy and color sensitivity in human visual corticalarea V8. Nat. Neurosci. 1, 235–241 (1998).

32. He, S., Cavanagh, P. & Intriligator, J. Attentional resolution and the locus ofvisual awareness. Nature 383, 334–337 (1996).

33. Tootell, R. B. H. et al. The retinotopy of spatial attention. Neuron 21,1409–1422 (1998).

34. DeYoe, E. A., Neitz, J., Miller, D. & Wieser, J. Functional magnetic resonanceimaging (FMRI) of visual cortex in human subjects using a unique videographics stimulator. Proc. Soc. Magn. Reson. Med. 3, 1394 (1993).

35. Treisman, A. M. Perceptual grouping and attention in visual search forfeatures and for objects. J. Exp. Psychol. Hum. Percept. Perform. 8, 194–214(1982).

36. Bandettini, P. A., Jesmanowicz, A., Wong, E. C. & Hyde, J. S. Processingstrategies for functional MRI of the human brain. Magn. Reson. Med. 30,161–173 (1993).

37. Saad, Z. S., DeYoe, E. A. & Ropella, K. M. in Proceedings of the 19thInternational Conference - IEEE.EMBS 460–463 (Chicago, Illinois, 1997).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


Although attention profoundly alters visual perception1, it is notequally important to all aspects of vision. For example, attentionis of little or no help to many detection tasks (for example, detect-ing a luminance increment2), and the degree to which it benefitsdiscrimination tasks varies widely with the discriminated attribute(for example, discriminating color, orientation, form3,4). Herewe report how attention alters thresholds for discriminating con-trast, orientation and spatial frequency of simple patterns. Basedon earlier work, we expected markedly different effects on dif-ferent thresholds5. In addition, we describe how attention changesthresholds for detecting one pattern in the presence of another,superimposed pattern of different orientation or spatial fre-quency. Together, these measurements characterize the visualmechanisms that underlie basic pattern vision.

Perceptual thresholds for stimulus contrast, orientation andspatial frequency have been studied for several decades6–8. Quan-titative accounts of these thresholds have become increasinglyrefined and usually involve a population of ‘noisy filters’ tunedto different orientations and spatial frequencies. Although ear-lier models postulated filters that are independent of each other,there are serious shortcomings to this approach9,10. More recentmodels postulate an interaction between filters with spatiallyoverlapping receptive fields11–13, specifically, the normalizationof individual filter responses relative to the total response of thelocal filter population (‘divisive inhibition’14). This normaliza-tion accounts naturally for several otherwise puzzling observa-tions, among them the initial decrease and later increase ofcontrast discrimination thresholds with increasing stimulus con-trast12,13 (‘dipper function’6,8) and the relative constancy of ori-entation and spatial frequency thresholds over a wide range ofstimulus contrasts9,15.

An intriguing parallel to these perceptual accounts can befound in certain models of visual cortical responses to stimuluscontrast and orientation16,17. Despite marked differences in detail,the models in question consider a population of neurons withoverlapping receptive fields, broadly tuned to a range of differ-ent orientations, and normalize individual responses relative to

Attention activates winner-take-allcompetition among visual filters

D.K. Lee, L. Itti, C. Koch and J. Braun

Computation and Neural Systems 139–74, California Institute of Technology, Pasadena, California 91125, USA

The first two authors contributed equally to this study.

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

Shifting attention away from a visual stimulus reduces, but does not abolish, visual discrimination per-formance. This residual vision with ‘poor’ attention can be compared to normal vision with ‘full’attention to reveal how attention alters visual perception. We report large differences between resid-ual and normal visual thresholds for discriminating the orientation or spatial frequency of simple pat-terns, and smaller differences for discriminating contrast. A computational model, in which attentionactivates a winner-take-all competition among overlapping visual filters, quantitatively accounts forall observations. Our model predicts that the effects of attention on visual cortical neurons includeincreased contrast gain as well as sharper tuning to orientation and spatial frequency.

the population response. The normalization, which in some casesis implemented as a divisive inhibition, sharpens orientation tun-ing16 and renders it less dependent on stimulus contrast17. Thus,both perceptual and neuronal sensitivity to contrast and orien-tation seem to involve response normalization.

Here we report that attention modulates the response nor-malization that seems to underlie basic pattern vision. We reachthis conclusion by comparing attentional changes in humanthresholds to predictions of a computational model based onresponse normalization. Our model is similar to several oth-ers11–13,17 and comprises three stages: a population of overlap-ping filters responsive to different orientations and spatialfrequencies at one visual location, non-linear interactions amongthis population to carry out the normalization and an ‘idealobserver’ decision that discriminates between stimulus alterna-tives on the basis of the maximum likelihood and is limited onlyby noise. Our observations are consistent with an attentional mod-ulation of the second, but not the first or third, stage of the model.

RESULTSPsychophysicsAlthough visual thresholds are usually measured when stimuliare fully attended, here we use a concurrent task to establishthresholds when stimuli are at best poorly attended4,18,19. Theconcurrent task in question forces observers to withdraw atten-tion from peripheral stimuli and to focus on stimuli near fixa-tion (Fig. 1, Methods). This psychophysical manipulation ishighly effective and causes substantial perceptual deficits in theperiphery similar to the deficits obtained after a lesion in visualcortical area V4 of the monkey19. However, the perception ofperipheral stimuli is not entirely abolished. Practiced observersenjoy a significant residual vision outside the focus of attentionand render reliable threshold judgments about peripheral stim-uli, especially when the display is uncluttered and contains onlya few salient stimuli4.

Observers discriminated contrast, orientation or spatial fre-quency of a luminance-modulated pattern appearing at varying

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


locations of 4° eccentricity (peripheral target; Fig. 1a). To drawattention away from this pattern, we asked observers to discrim-inate whether five shapes near fixation (central targets) were the“same” or “different”. When observers carried out both tasks,they concentrated attention on the central task, which they wereinstructed to consider the primary task, and thus left the periph-eral target poorly attended (double-taskthresholds). In contrast, when observersviewed the same display but performedonly the peripheral task, they fully attendedto the peripheral target (single-task thresh-olds). The comparison of single- and dou-ble-task thresholds reveals if and howattention alters visual perception.

We compared five types of thresholdsunder single- and double-task conditions(Fig. 2a–e). When peripheral targets werefully attended, contrast detection thresholds(zero mask contrast) were about 20% lower,and contrast discrimination thresholds(mask contrast greater than zero) about40–50% lower than when peripheral targetswere poorly attended (Fig. 2a). In addition,the decrease of the discrimination thresholdas mask contrast increases from zero (dip-per) was evident only when targets werefully attended. Note that the target positionvaried from trial to trial (to forestall eyemovements) and that positional uncertain-ty of this kind is known to reduce the dip-per20–22. Therefore, it is possible that ourdata underestimate the depth of the dipper.

The effects of attention on spatial fre-quency and orientation discrimination wereeven more pronounced (Fig. 2b and c).Spatial frequency thresholds were about60% lower and orientation thresholdsabout 70% lower when peripheral targetswere fully attended compared to when theywere poorly attended. Note that both typesof thresholds remained essentially constantfor contrast values above 20%.

Interactions between superimposedstimuli of different orientation or spatial fre-

quency (target and mask; Fig. 2d and e) were also altered by atten-tion. When target and mask had similar orientation or spatial fre-quency, attention lowered the maximal threshold by about 50%(consistent with Fig. 2a, mask contrast 0.5). As target and maskbecame progressively more different, fully and poorly attendedthresholds decreased toward the same baseline level. The baseline

articles

Peripheral response

Stimulus MaskFixation

Centralresponse100-150 ms 200 ms 200 ms

Stimulus

Mask

Fixation

A

B

C

Peripheral response

a

b

c

Fig. 1. Measurement of visual thresholds with either full or poor atten-tion. (a) Sequence of fixation, stimulus and mask displays (schematic).Observers fixate the center of all displays. The stimulus comprises acentral and a peripheral component, which appear at varying locationsof constant eccentricity. The central component consists of 5 Ts and/orLs (central targets) and observers report “same” (that is, 5 Ts or 5 Ls)or “different” (that is, 4 Ts + 1 L or 4 Ls + 1 T). The peripheral compo-nent consists of the luminance-modulated patterns shown in Fig. 2a–e(peripheral target). For example, the peripheral component might be agrating pattern of vertical or tilted orientation, in which case observerswould report “vertical” or “tilted.” The mask display limits visual persis-tence of central targets. (b) Single task (peripheral target ‘fullyattended’), observers fixate the center but respond only to the periph-eral task (see Fig. 2). (c) Double task (peripheral target ‘poorlyattended’), observers fixate the center and respond first to the centraltask and second to the peripheral task.

Fig. 2. Single- and double-task thresholds compared. Five types of thresholds were measured. Ineach case, observers discriminate between two alternative forms of the peripheral (4° eccentric-ity) target. Filled and open symbols represent fully attended (single-task) and poorly attended (dou-ble-task) thresholds, respectively (mean and standard error of two observers). Solid and dashedcurves represent the corresponding model predictions. (a) Contrast detection and discrimination.Observers report the presence (arrows) or absence of a vertical target stripe from a circularmasking pattern (contrast range 0.0–0.5). (b, c) Spatial frequency and orientation discrimination.Observers report whether a circular target grating (contrast 0.02–0.8) exhibits higher or lowerspatial frequency (b) or whether its orientation is vertical or tilted clockwise (c). (*) indicates fur-ther data points off scale. (d, e) Orientation and spatial frequency masking. Observers report thepresence (arrows) or absence of a vertical target stripe from circular masking patterns (contrast0.5) of different orientation (difference range 0–90; d) or different spatial frequency (differencerange –1 to +1 oct; e). (f) Model parameters of plausible fits computed separately for single- anddouble-task data. Although all 10 parameters are permitted to differ, most parameters do not dif-fer significantly (n.s.).

a b c

d e f

Targ

et c

on

tro

lTa

rget

co

ntr

ol

Targ

et c

on

tro

l∆ω

(oct

aves

)

∆θ(°

)

Mask contrast Contrast Contrast

Mask θ – Target θ (°) Mask ω – Target ω (°)


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


was comparable to thresholds without mask (Fig. 2a, mask con-trast 0), indicating minimal interactions between targets andmasks of very different orientation or spatial frequency.

ModelThe visual thresholds measured here are thought to reflect theactivity of a population of ‘noisy filters’ selective for stimuli ofdifferent orientations and spatial frequencies7,8. We define a fil-ter tuned to orientation θ and spatial frequency ω by

where Eθ,ω is the linear response, cs, θs and ωs are the contrast,orientation and spatial frequency of the stimulus, A is the con-trast gain, B is the background activity, and σθ and σω are thesharpness of tuning (for a sinusoidal grating stimulus). Whenthe properties of such filters are inferred from behavioral thresh-old measurements, they tend to match the response propertiesof neurons in visual cortical areas V1 and V2 (refs. 23–25).Accordingly, each visual filter is thought to correspond to a pop-ulation of visual cortical neurons tuned to a particular orientationand spatial frequency.

Can the observed effects of attention be understood simply asa change in the properties of individual visual filters? To answerthis question, we first examined the case in which filters are inde-pendent, so that the output of each filter, Rθ,ω, is a monotonic (andperhaps non-linear) function of its linear response, Eθ,ω. We alsoassume that the variance of the filter output, V2

θ,ω, is given by

where β is the ‘light noise’ and ε is the ‘dark noise’. This approx-imates the response variance of visual cortical neurons25. Giventhese assumptions, the observed 20% difference between con-trast detection thresholds with full and poor attention implies(see Methods) that either the gain A decreases or the light noiseβ increases by about 20%. The 60% to 70% difference in spatialfrequency and orientation thresholds, on the other hand, cannotbe explained by a 20% change in A and/or β. To account for allobservations, we must therefore assume that attention alters notonly the contrast gain or noise level of visual filters, but also theirtuning widths, σθ and σω, for orientation and spatial frequency.

Next we analyzed a more complex model, which also proves tobe consistent with attentional changes of both gain and tuning(Fig. 3). The first stage of this model consists of a population ofoverlapping linear filters responsive to different orientations andspatial frequencies at one visual location (Eq. 1). The second stageassumes that filters are not independent but interact so as to nor-

malize individual responses relative to the filter population.Specifically, second-stage responses, Rθ,ω, are obtained by sub-jecting first-stage responses, Eθ,ω, to a power law followed by divi-sive inhibition:

The exponents of the power law, δ and γ, are of particularconsequence: their absolute values govern the strength of theinteraction between filters, and their difference determines thesaturation of responses at high contrast. The semi-saturationconstant, S, determines the response at low stimulus contrast.The distribution of weight factors, W θθ′,ωω′,

whose Gaussian widths are given by Σθ and Σω, determines whetherthe ‘inhibitory pool’ includes the entire filter population or onlyfilters tuned to similar orientations and spatial frequencies.

The third stage of the model discriminates between stimulusalternatives based on the maximum likelihood of second-stageresponses. This corresponds to an ideal observer whose perfor-mance is limited only by the variance (noise) of second-stageresponses (Eq. 2). Further details about the decision are givenelsewhere15 (also see Methods).

When we fit this model (10 free parameters: γ, δ, σ θ, σω, Σθ,Σω, S, B, β, ε) separately to single- or double-task data, weobtained good agreement between predicted and observedthresholds with physiologically plausible parameter values (solidcurves in Fig. 2). Note in particular the realistic widths of filtertuning, with half-widths at half-maximum between 12° and 15°for orientation and 0.42 octaves (oct) and 0.52 oct for spatial fre-quency, compared to 20 ± 9o and 0.76 ± 0.30 oct for neurons inmonkey visual cortex25. Note also that orientation and spatialfrequency thresholds remain constant for contrast values above20% (Fig. 2b and c) and that the curves for full and poor atten-tion appear displaced vertically rather than horizontally. Thisshows clearly that attention changes more than contrast gain,because a difference in gain of the linear filter stage would mere-ly produce a horizontal displacement. The main discrepancybetween model and data is that the model predicts a more pro-nounced dipper for contrast discrimination thresholds than isactually observed (Fig. 2a). Because our data may underestimatethe dipper (see above), this prediction may be correct.

That a single set of parameter values accounts for all thresh-olds observed with full attention is not a matter of course. Onemight have expected that attending to stimulus orientation would

articles

Fig. 3. Three-stage model of visual filters and their interac-tions (schematic). Each stimulus location is analyzed by lin-ear filters sensitive to different orientations and spatialfrequencies (Eq. 1, first-stage responses, Eθ,ω). Filterresponses are subjected to excitatory and inhibitory inter-actions in the form of amplification and divisive normaliza-tion (Eqs. 3, 4, second-stage responses, Rθ,ω). The decisionstage assumes that first-stage responses show a variancesimilar to that of cortical neurons (Eq. 2) and choosesbetween stimulus alternatives on the basis of maximal likeli-hood. See ref. 15 and Methods for details. Our results sug-gest that attention strengthens non-linear interactionsbetween filters (gray box), but does not affect other parts ofthe model.

+

Stimulus

Noise

Noise

Noise

Statisticallyefficientdecision

+

-

+-

--

Linear filters Non-linearinteractions

Eθ, ω

R θ, ω

Decision

V 2θ,ω = β(Rθ,ω + ε) (2)

Rθ,ω = (3)(Eθ,ω)γ

Sδ + ∑θ′,ω′

Wθθ′,ωω′(Eθ′,ω′)δ

Wθθ′,ωω′= e e (4)–(θ – θ′)2

2∑2θ

–(ω – ω′)2

2∑2ω

Eθ,ω= Acse e + B (1)–(θs – θ)2

2σ2θ

–(ωs – ω)2

2σ2ω


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


affect visual processing differently than, say, attending to spatialfrequency. Instead, our results are consistent with the possibilitythat attention alters visual processing in the same way for all exam-ined tasks. However, a strict test of task independence wouldrequire that model parameters be determined independently fordifferent tasks. Unfortunately, such a test is not feasible becausethe data from any one task do not constrain all model parameters.

Although there are several differences between the parame-ters obtained with full and poor attention, the change in the expo-nents of the power law, γand δ, is particularly significant (Fig. 2f).We assess the significance of a difference in the values of a givenparameter by determining how rapidly the quality of fit deteri-orates when this value is changed (see Methods). To study therole of γ and δ, we fit the model simultaneously to both single-and double-task data, while allowing only γ and δ to take differ-ent values depending on attention. In other words, γ and δ taketwo values, whereas all other parameters take a single value (12free parameters total). Once again, we obtained acceptable fitswith physiologically plausible parameter values (‘12-dimension-al joint fits’, solid curves in Fig. 4a–e, left columns in Fig. 4f).

In contrast, when we allow all parameters except γ and δ totake different values depending on attention (18 free parameters,total), there are no acceptable fits with plausible parameter values.The optimal fit under these assumptions predicts neither the dip-per in the contrast discrimination thresholds (Fig. 4a) nor themaximal extent of contrast masking (Fig. 4d and e). To obtainthis poor fit, the tuning widths for orientation and spatial fre-quency and the size of the inhibitory pool have to change dra-matically (σθ from 17o to 5o, σω from 0.7 oct to 0.3 oct, and Σ θfrom 0.6 σ θ to 5 σω). It seems unlikely that attention would altercortical interactions so profoundly.

DISCUSSIONWe measured thresholds for discriminating the contrast, orienta-tion and spatial frequency of simple patterns that are either fully orpoorly attended. We observed differences of 20% in contrast detec-

tion thresholds, 40–50% in contrast discrimination thresholds (andappearance of the dipper), 60–70% in orientation and spatial fre-quency discrimination thresholds, and up to 50% in contrast mask-ing thresholds. These observations tightly constrain any effectattention may have on the visual filters that are thought to underliebasic pattern vision. Comparison with a computational model showsthat the observed effects of attention are consistent with strongerinteractions among filters, but not with a change in noise parameterswithout change in interactions, as is sometimes thought26,27. Essen-tially, the effects of attention on different thresholds are too disparateto be accommodated by a single change in noise parameters.

articles

Fig. 4. Predicted thresholds when atten-tion changes some model parametersbut not others. Format and experimentaldata are identical to Fig. 2. The thickcurves represent a simultaneous fit toboth single- and double-task data (solidand dashed, respectively), in which onlythe exponents δ and γ take different val-ues depending on attention (12 freeparameters). Observed and predictedthresholds agree reasonably well, andparameter values are physiologicallyplausible (two leftmost columns in f).The thin curves represent the optimaljoint fit to single- and double-task data(solid and dashed, respectively) when allparameters except the exponents takedifferent values depending on attention(18 free parameters). Neither the dipper(a) nor the maximal extent of contrastmasking (d, e) are predicted, and para-meter values are biologically implausible(two rightmost columns in f).

Stimulus contrast

Resp

onse

0 0.5 10

0.5

1A

Stimulus orientation

Resp

onse

0 10 20 30 400

0.5

1C

Spatial frequency ratio

Resp

onse

0

0.5

1

1 2 3

D

Resp

onse

0

0.4

Stimulus contrast0 0.05 0.1

0.2

B

Fig. 5. Effect of attention on early visual processing. Predictions basedon 12-dimensional joint fit in Fig. 4f. Attention increases the contrastgain (3.3-fold, a), causes the contrast response to assume sigmoidalshape at low contrast (b) and sharpens orientation tuning by 40% (c)and spatial-frequency tuning by 30% (d). To the extent that the visual fil-ters of our model reflect individual neurons in visual cortex, this pre-dicts that attention both increases the gain and sharpens the tuning ofsuch neurons.

a b c

d e f

Targ

et c

on

tras

tTa

rget

co

ntr

ast

Targ

et c

on

tras

t∆ω

(oct

aves

)

∆θ(°

)

ContrastContrastMask contrast

Mask θ – Target θ (°) Mask ω/Target ω

b

Stimulus contrast Stimulus contrast

Stimulus orientation Spatial frequency ratio

Res

po

nse

dc

Res

po

nse

a

Res

po

nse

Res

po

nse


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


In the framework of our model, the strength of interactionsamong filters is controlled by the exponents of a power law, γandδ. The immediate reasons why larger exponents account for theobserved effects of attention are as follows: for small stimuluscontrasts, higher exponents reduce background firing and thesigmoidal shape of the contrast response (Fig. 5a and b), whichexplains the improved contrast detection thresholds and theenhanced dipper of the contrast discrimination curve. For larg-er stimulus contrasts, higher exponents entail a 3.3-fold increasein contrast gain (Fig. 5b), which accounts for lower contrast dis-crimination and contrast-masking thresholds. Additionally, high-er exponents sharpen the tuning for orientation by 40%, and forspatial frequency by 30% (Fig. 5c and d), lowering thresholds fordiscriminating orientation and spatial frequency still further. Tothe extent that visual filters can be identified with individual neu-rons in visual cortex, our model thus predicts that attentionchanges both the gain and tuning of such neurons.

The more fundamental reason, however, is that larger expo-nents activate what is best described as a winner-take-all com-petition among visual filters. Attention (larger exponents) shiftsthe distribution of responses across the population of filters(Fig. 6). Attention accentuates existing differences between filterresponses, boosting filters that respond relatively well to a givenstimulus, while suppressing filters that respond relatively poorly.This explains the perceptual advantage conferred by attention:attention enhances the sensory representation by restrictingresponses to the filters that are tuned best to the stimulus at hand.

Previous studies of attentional changes in visual thresholds arebroadly consistent with our results, even though our effects arelarger. This includes reports that attention reduces contrast thresh-olds by 17% (ref. 27), orientation acuity by 15% (for an individualtarget without distractors)28, and size acuity by 20% (ref. 29). How-ever, these studies manipulated attention with a spatial cue ratherthan with a concurrent task, which complicates quantitative com-parison. We believe that a concurrent task detains attention moreconsistently than spatial cueing; certainly concurrent tasks inducesubstantially larger changes in thresholds. An effect we have notconsidered here is that attention is able to improve perceptual deci-

sions in the face of positional uncertainty20–22. We estimate thateliminating spatial uncertainty about the target location wouldimprove contrast detection thresholds by 19%, but would have noappreciable effect on any thresholds at stimulus contrasts higherthan 10% (see Methods). Thus, reduced uncertainty cannotexplain the pattern of attentional effects we have observed.

Our model is also consistent with recent findings in the visu-al cortex of humans and monkeys. Attentional changes in neu-ronal activity have been reported in several early visual corticalareas, including areas V1, V2, V4 and MT/MST30–35 (see alsoS.P. Gandhi, D.J. Heeger & G.M. Boynton, Inv. Ophth. Vis. Sci.(Suppl.) 39, 5194, 1998). Furthermore, the notion that attentionmodulates a local competition in visual cortex has been proposedindependently based on theoretical36,37 and single-neuron stud-ies38. In the macaque, attentional modulation of responses in thevisual cortex is weak or absent if only a single stimulus is presentin the receptive field, suggesting that attention modulates inter-actions between neurons with overlapping receptive fields30,33.Finally, our model is consistent with reports that attentionincreases contrast gain in areas V2 and V4 of the macaque39 (seealso J. Reynolds, T. Pasternak & R. Desimone, Inv. Ophth. Vis.Sci. (Suppl.) 38, 3206, 1997). Whether attention sharpens the ori-entation tuning of visual cortical neurons remains controver-sial39,40. Our model predicts that fully focused attention sharpensorientation tuning in the parts of visual cortex that mediate basicpattern vision (presumably areas V1 and/or V2). In area V4,increased competition would presumably sharpen tuning alongother, more complex, stimulus dimensions.

Finally, we do not wish to claim that attention is restricted tolocal interactions at one particular level of visual cortex. Morethan likely, attention has additional effects on long-range inter-actions at the same level and, indeed, at all levels of visual cor-tex. Nevertheless, our results show that the activation of awinner-take-all competition among overlapping visual filtersexplains many basic perceptual consequences of attention.

METHODSPsychophysics. Stimuli were displayed on an SGI Indigo (1024 × 1286pixels RGB). Viewing was binocular at 120 cm distance (1° corre-sponds to 80 pixels). Mean luminance was 30 cd/m2, with linear incre-ments of 0.07 cd/m2 (obtained by gamma correction and ‘color bitstealing’41), and room luminance was 3 cd/m2. Central targetsappeared at 0–0.8° eccentricity and measured 0.4° across. Peripheraltargets appeared at 4°eccentricity, in a circular aperture of 1.5° (tim-ing as shown in Fig. 1). They were either sinusoidal gratings (Fig. 2band c) or vertical stripes whose luminance profile was given by the6th derivative of a Gaussian (Fig. 2a, d and e). Mask patterns weregenerated by superimposing 100 Gabor filters, positioned randomlywithin the circular aperture (Fig. 2a, d and e). The spatial frequencywas 4 cpd (vertical stripes in Fig. 2a, d and e; sinusoidal gratings inFig. 2b; superimposed Gabors in Fig. 2a and d), and the mask con-trast was 0.5 (Fig. 2d and e). All thresholds were established with stan-dard adaptive staircase methods (80 trials per block). The values givenare averages from between 12 and 20 blocks of trials and 2 naiveobservers. Standard deviations were computed separately for eachobserver, and error bars represent the average value. In the double-task situation, observers were required to match or exceed a certainlevel of central performance. Approximately 15% of double-taskblocks were discarded because of poor central performance. In bothsingle- and double-task situations, observers fixated the display cen-ter, ensuring identical visual stimulation. The brief presentation effec-tively precludes saccades toward the peripheral target.

Concurrent-task method. An important concern in concurrent-taskexperiments is the level of processing at which the two tasks interfere.In general, interference can reflect limitations of attention, memory

articles

Fig. 6. Attentional change in the response distribution. Predictionsbased on 12-dimensional joint fit in Fig. 4f. Responses Rθ,ω of filterstuned to orientations between –20o to +20o, to a grating stimulus oforientation 0o and contrasts between 0 and 5% (threshold regime).Responses to fully and poorly attended stimuli are represented by thered and blue surfaces, respectively (shown interleaved for clarity). Bystrengthening a winner-take-all competition among visual filters, atten-tion restricts responses to the filters tuned best to the stimulus at hand.


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


and/or response generation42–44. In the present experiments, observershave ample time to respond to each task in turn, so that limitations ofresponse generation are not likely. Because interference disappears whencentral and peripheral targets are presented successively (for example,with an onset asynchrony of 200 ms or more)4,45,46, a limitation ofmemory is also unlikely. Further evidence that the critical factor is atten-tion is that interference does not depend on the nature of the centraltask (as long as it poses a sufficient demand on attention). For example,central tasks based on form, color or motion discrimination47, as wellas on an ‘attentional blink’48, produce comparable interference.

Independent-filter model. If the ‘transducer’ function Rθ,ω= t(Eθ,ω) is linearover small ranges of contrast, and if the decision between stimulus alterna-tives is based on maximum likelihood, one can derive simple proportional-ity relationships for the contrast detection threshold, ∆cdet, and theorientation and spatial frequency discrimination thresholds, ∆θdis and ∆ωdis:

These relations indicate how thresholds depend on filter parameters suchas β/Α, σθ and σω. Note that not all thresholds depend on all parameters.

Interacting-filter model. The first stage of the model comprises 150 filterswith Fourier representations in the shape of a 2-dimensional Gaussiancentered on 30 orientations (0 ≤ θi < π) and 5 spatial frequencies (2 cpd≤ ωi < 8 cpd; Eq. 1). Increasing the number of filters does not alter modelpredictions substantially. From each Fourier representation, a filter pairin quadrature phase is reconstructed, and linear responses Eθ,ωare com-puted numerically. The effective tuning widths are σθ and σθ for sinu-soidal gratings, 1.1 σθ and 1.3 σω for vertical stripes, and 1.7 σθ and 1.3σω for mask patterns. The second stage of the model is given by Eqs. 3and 4. For γ, δ >>1, only the largest first-stage responses produce sig-nificant second-stage responses. For γ, δ = 1, first- and second-stageresponses are proportional. At high stimulus contrast, second-stageresponses follow a power law with exponent γ – δ. The transducer func-tion becomes increasingly sigmoidal with higher exponents γ, δ. Thethird stage of the model assumes that second-stage filter responses exhib-it Gaussian-distributed noise (Eq. 2), and uses maximum-likelihoodprinciples to predict ideal observer thresholds from the means and vari-ances of these responses. Specifically, the Fisher information for each fil-ter and for the entire population provides a lower limit for the variance ofany unbiased estimate of a stimulus parameter such as contrast, orien-tation or spatial frequency15,48,49.

Model fits. Fits involve downhill simplex error minimization50, withsimulated annealing overhead. For ‘separate fits’, the 10 free modelparameters were fit either to single- or double-task data (each dataset comprised 32 values from 5 experiments). Fig. 2 shows a plausiblefit, whose fit error is 8% larger than the optimal fit. For the ‘12-dimen-sional joint fit’ in Fig. 4, two parameters take different values for sin-gle- and double-task data. Thus, 12 free parameters are fit to 64 datavalues. Fig. 4 shows a plausible fit, whose fit error is 12% larger thanthe optimal fit. For the ‘18-dimensional joint fit’ in Fig. 4, eight para-meters assume different values for single- and double-task data. Thus,18 free parameters are fit to 64 data values. Fig. 4 shows the optimalfit, in which attention has a number of physiologically implausibleeffects, among them a 6-fold increase in contrast gain, 70% sharperorientation tuning and 60% sharper spatial frequency tuning.

To assess the significance of different parameter values, a ‘tolerancerange’ was computed for each parameter, within which the error of fitincreases by no more than 10% (a small but noticeable degradation ofthe quality of fit). The difference between single- and double-task val-ues of a given parameter was considered significant if it fell outside thistolerance range (Figs. 2f and 4f).

Positional uncertainty. As an alternative to the present model, we con-sidered the possibility that attention lowers thresholds by reducing posi-

tional uncertainty. We used the formalism of ref. 22 and chose a noiselevel that reproduced ‘fully attended’ thresholds without positionaluncertainty. Then we introduced positional uncertainty among the eightpossible target locations to predict the ‘poorly attended’ thresholds. Posi-tional uncertainty increased contrast detection thresholds by 19%, butleft all other thresholds almost unchanged. For example, contrast dis-crimination thresholds at 10% pedestal contrast increased only by 1.5%.Thus, positional uncertainty cannot explain the pattern of attentionaleffects we have observed.

ACKNOWLEDGEMENTSSupported by NSF, NIMH, ONR and the NSF-ERC at Caltech. We thank T.

Albright and T. Sejnowski for access to facilities and J. Gallant, A. Manwani, S.

Shimojo, K. Watanabe and B. Zenger for comments and discussions.

RECEIVED 24 NOVEMBER 1998; ACCEPTED 24 FEBRUARY 1999

1. Pashler, H. The Psychology of Attention (MIT Press, Cambridge,Massachusetts, 1997).

2. Mueller, H. J. & Findlay, J. M. Sensitivity and criterion effects in the spatialcueing of visual attention. Percept. Psychophys. 42, 383–399 (1987).

3. Treisman, A. & Gormican, S. Feature analysis in early vision: evidence fromsearch asymmetries. Psychol. Rev. 12, 97–136 (1988).

4. Braun, J. & Julesz, B. Withdrawing attention at little or no cost: Detection anddiscrimination tasks. Percept. Psychophys. 60, 1–23 (1998).

5. Lee, D. K., Koch, C. & Braun, J. Spatial vision thresholds in the near absenceof attention. Vision Res. 37, 2409–2418 (1997).

6. Nachmias, J. & Sansbury, R. V. Letter: Grating contrast: discrimination maybe better than detection. Vision Res. 14, 1039–1042 (1974).

7. Wilson, H. R. A transducer function for threshold and suprathresholdhuman vision. Biol. Cybern. 38, 171–178 (1980).

8. Legge, G. E. & Foley, J. M. Contrast masking in human vision. J. Opt. Soc. Am.70, 1458–1471 (1980).

9. Bowne, S. F. Contrast discrimination cannot explain spatial frequency,orientation or temporal frequency discrimination. Vision Res. 30, 449–461(1990).

10. Wilson, H. R. & Wilkinson, F. Evolving concepts of spatial channels in vision:from independence to nonlinear interactions. Perception 26, 939–960 (1997).

11. Wilson, H. R. & Humanski, R. Spatial frequency adaptation and contrast gaincontrol. Vision Res. 33, 1133–1149 (1993).

12. Foley, J. M. Human luminance pattern-vision mechanisms: maskingexperiments require a new model. J. Opt. Soc. Am. 11, 1710–1719 (1994).

13. Zenger, B. & Sagi, D. Isolating excitatory and inhibitory nonlinear spatialinteractions involved in contrast detection. Vision Res. 36, 2497–2513 (1996).

14. Carandini, M. & Heeger, D. J. Summation and division by neurons in primatevisual cortex. Science 264, 1333–1336 (1994).

15. Itti, L., Braun, J., Lee, D. K. & Koch, C. in Advances in Neural InformationProcessing Systems, Vol. 9 (eds. Mozer, M. C., Jordan, M. I. & Petsche, T.)173–179 (MIT Press, Cambridge, Massachusetts, 1997).

16. Somers, D. C., Nelson, S. B. & Sur, M. An emergent model of orientationselectivity in cat visual cortical simple cells. J. Neurosci. 15, 5448–5465 (1995).

17. Carandini, M., Heeger, D. J. & Movshon, J. A. Linearity and normalization insimple cells of the macaque primary visual cortex. J. Neurosci. 17, 8621–8644(1997).

18. Braun, J. & Sagi, D. Vision outside the focus of attention. Percept. Psychophys.48, 45–58 (1990).

19. Braun, J. Visual search among items of different salience: removal of visualattention mimics a lesion in extrastriate area V4. J. Neurosci. 14, 554–567(1994).

20. Palmer, J. Attention in visual search: distinguishing four causes of set-sizeeffects. Curr. Dir. Psychol. Sci. 4, 118–123 (1995).

21. Solomon, J. A., Lavie, N. & Morgan, M. J. Contrast discrimination function:spatial cuing effects. J. Opt. Soc. Am. 14, 2443–2448 (1997).

22. Foley, J. M. & Schwarz, W. Spatial attention: effect of position uncertainty andnumber of distractor patterns on the threshold-versus-contrast function forcontrast discrimination. J. Opt. Soc. Am. 15, 1036–1046 (1998).

23. DeValois, R. L. & DeValois, K. K. Spatial Vision (Oxford Univ. Press, NewYork, 1988).

24. Wilson, H. R., Levi, D., Maffei, L., Rovamo, J. & DeValois, R. in VisualPerception: The Neurophysiological Foundations (eds. Spillmann, L. & Werner,J. S.) 231–272 (Academic, San Diego, California, 1990).

25. Geisler, W. S. & Albrecht, D. G. Visual cortex neurons in monkeys and cats:detection, discrimination, and identification. Vis. Neurosci. 14, 897–919(1997).

26. Bonnel, A.-M. & Miller, J. Attentional effects on concurrent psychophysicaldiscriminations: investigations of a sample size model. Percept. Psychophys.55, 162–179 (1994).

27. Lu, Z. L. & Dosher, B. A. External noise distinguishes attention mechanisms.Vision Res. 38, 1183–1198 (1998).

articles

∆cdet∝ β–A (5)

∆θdis ∝ σ θ √ β—Acs∆ωdis ∝ σ ω √ β—Acs

(6,7)


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


28. Morgan, M. J., Ward, R. M. & Castet, E. Visual search for a tilted target: testsof spatial uncertainty models. Q. J. Exp. Psychol. 51, 347–370 (1998).

29. Yeshurun, Y. & Carrasco, M. Spatial attention improves performance in spatialresolution task. Vision Res. 39, 293–305 (1999).

30. Moran, J. & Desimone, R. Selective attention gates visual processing in theextrastriate cortex. Science 229, 782–784 (1985).

31. Motter, B. C. Focal attention produces spatially selective processing in visualcortical areas V1, V2, and V4 in the presence of competing stimuli. J.Neurophysiol. 70, 909–919 (1993).

32. Treue, S. & Maunsell, J. H. Attentional modulation of visual motionprocessing in cortical areas MT and MST. Nature 382, 539–541 (1996).

33. Luck, S. J., Chelazzi, L., Hillyard, S. A. & Desimone, R. Neural mechanisms ofspatial selective attention in areas V1, V2, and V4 of macaque visual cortex. J.Neurophysiol. 77, 24–42 (1997).

34. Roelfsema, P. R., Lamme, V. A. & Spekreijse, H. Object-based attention in theprimary visual cortex of the macaque monkey. Nature 395, 376–381 (1998).

35. Brefczynski, J. A. & DeYoe, E. A. A physiological correlate of the ‘spotlight’ ofvisual attention. Nat. Neurosci. 2, 370–374 (1999).

36. Niebur, E. & Koch, C. A model for the neuronal implementation of selectivevisual attention based on temporal correlation among neurons. J. Comput.Neurosci. 1, 141–158 (1994).

37. Tsotsos, J. K. et al. Modeling visual attention via selective tuning. Artif. Intell.78, 507–547 (1995).

38. Desimone, R. Visual attention mediated by biased competition in extrastriatevisual cortex. Phil. Trans. R. Soc. Lond. B 353, 1245–1255 (1998).

39. McAdams, C. J. & Maunsell, J. H. R. Effects of attention on orientation-tuning

functions of single neurons in macaque cortical area V4. J. Neurosci. 19,431–441 (1999).

40. Spitzer, H., Desimone, R. & Moran, J. Increased attention enhances bothbehavioral and neuronal performance. Science 240, 338–340 (1988).

41. Tyler, C. W. Colour bit-stealing to enhance the luminance resolution ofdigital displays on a single pixel basis. Spatial Vis. 10, 369–377 (1997).

42. Duncan, J. Selective attention and the organization of visual information. J.Exp. Psychol Gen. 113, 501–517 (1984).

43. Pashler, H. Dual-task interference in simple tasks - Data and theory.Psychology B 116, 220–244 (1994).

44. Dosher, B. A. & Sperling, G. in Handbook of Perception and Cognition,Perception and Cognition at Century’s End: History, Philosophy, Theory (ed.Hochberg, J.) 201–254 (Academic, New York, 1998).

45. Duncan, J., Ward, R. & Shapiro, K. Direct measurement of attentional dwelltime in human vision. Nature 369, 313–315 (1994).

46. Braun, J. Vision and attention: the role of training. Nature 393, 424–425(1998).

47. Lee, D. K., Koch, C. & Braun, J. Visual attention is undifferentiated:concurrent discrimination of form, color, and motion. Percept. Psychophys.61 (in press).

48. Seung, H. S. & Sompolinsky, H. Simple models for reading neuronalpopulation codes. Proc. Natl. Acad. Sci. USA 90, 10749–10753 (1993).

49. Pouget, A., Zhang, K., Deneve, S. & Latham, P. E. Statistically efficientestimation using population coding. Neural Comput. 10, 373–401 (1998).

50. Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. NumericalRecipes in C, 2nd ed. (Cambridge Univ. Press, Cambridge, 1992).

articles


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


Emotional responses to pleasantand unpleasant music correlate withactivity in paralimbic brain regions

Anne J. Blood, Robert J. Zatorre, Patrick Bermudez and Alan C. Evans

Neuropsychology/Cognitive Neuroscience Unit, Montreal Neurological Institute, McGill University, 3801 rue University, Montreal, PQ H3A 2B4, Canada

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

Neural correlates of the often-powerful emotional responses to music are poorly understood. Here weused positron emission tomography to examine cerebral blood flow (CBF) changes related toaffective responses to music. Ten volunteers were scanned while listening to six versions of a novelmusical passage varying systematically in degree of dissonance. Reciprocal CBF covariations wereobserved in several distinct paralimbic and neocortical regions as a function of dissonance and of per-ceived pleasantness/unpleasantness. The findings suggest that music may recruit neural mechanismssimilar to those previously associated with pleasant/unpleasant emotional states, but different fromthose underlying other components of music perception, and other emotions such as fear.

articles

Music has an extraordinary ability to evoke powerful emotions.This ability is particularly intriguing because, unlike most otherstimuli that evoke emotion, such as smell, taste or facial expres-sion, music has no obvious intrinsic biological or survival value.Although changes in certain physiological processes have beencharacterized in response to music1,2, neural correlates of emo-tional responses to music, their relation to music perception andtheir relation to other forms of emotion have not been well stud-ied. Here we present a novel approach to the study of music andemotion, using positron emission tomography (PET) to mea-sure cerebral correlates of affective and perceptual responses tomusical dissonance. Although music may often evoke positiveemotions, individual differences in musical preference mayobscure any systematic neural correlates. Instead, we opted tostudy the negative affective reactions elicited by dissonance, whichappear to be relatively consistent and stable. Listeners who havebeen exposed to the Western tonal idiom typically respond read-ily to dissonance, even in the absence of formal musical training.This phenomenon presumably indicates that listeners have inter-nalized the tonal rules of music in their culture and react to vio-lations of these rules3.

Many aspects of melodic processing depend on the integrity ofsuperior temporal and frontal cortices. More specifically, bothlesion and functional imaging studies indicate that regions ofauditory cortex within the right superior temporal gyrus arespecifically involved in analysis of pitch and timbre4–8, and thatworking memory for pitch entails interactions between temporaland frontal cortices6,9. Few data are available concerning the affec-tive component of musical processing. A recent case study of apatient with amusia, however, has suggested that perceptual andemotional analysis of music may be dissociated10. In parallel withvisual face processing, judgments of affective content of a melody(happy versus sad) can be made even in the complete absence ofany ability to identify or recognize a melody. Here our purposewas to investigate whether such dissociations could be under-

stood in terms of distinct neural mechanisms engaged by musi-cally induced affect and to establish their functional anatomy.

Identifying neural correlates of affective responses to musicmay also prove advantageous in coming to a more general under-standing of emotion. Unlike many other stimuli, music can oftenevoke emotion spontaneously, in the absence of external associa-tions11–14. Our choice of the dissonance paradigm was predicat-ed on this aspect, because dissonance directly elicits feelings ofunpleasantness in novel passages without any prior associations.

In the present study, we used a novel melody, which was madeto sound more or less consonant or dissonant by varying the har-monic structure of its accompanying chords (Fig. 1a). PET scanswere obtained while subjects with no more than amateur musi-cal training listened to six versions of this stimulus (termed Diss0through Diss5), designed based on pilot studies to spontaneouslyelicit a continuum of pleasant to very unpleasant emotionalresponses. In addition, acoustically matched noise stimuli werepresented as a sensory control condition, allowing us to exam-ine the complete activation pattern related to processing of thestimuli9. Regression analysis was used to correlate rCBF withdegree of dissonance; this analysis was complemented by con-ventional subtraction analysis. Regional covariation analyses werethen used to identify possible functional interactions betweenspecific cerebral structures. Subjects also rated the emotionalquality of the music, using a rating scale with eight pairs of adjec-tives. We hypothesized that the variations in affective quality ofthe stimuli would correlate with rCBF changes in regions involvedin emotional processes, and that these regions would differ fromthose involved in perceptual analysis of music.

RESULTSRegional CBF changes in paralimbic and neocortical areas wereassociated with both increasing dissonance and increasing con-sonance. However, distinct structures were activated by disso-nance versus consonance. Activity in right parahippocampal


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


gyrus and precuneus regions correlated with increasing disso-nance, whereas activity in orbitofrontal, subcallosal cingulate andfrontal polar cortex correlated with decreasing dissonance (equiv-alent to increasing consonance). Regional covariation analysesdemonstrated inverse correlations between these two sets ofregions. Right parahippocampal, right orbitofrontal and medi-al subcallosal cingulate activity was confirmed by subtractionanalysis. The unpleasantness of increasing dissonance was con-firmed by subjects’ analog ratings; these ratings also covariedwith rCBF changes in similar paralimbic regions.

An analysis examining rCBF changes as a function of increas-ing dissonance (see Methods) identified significant positive cor-relations in right parahippocampal gyrus and right precuneusregions (Table 1 and Fig. 2a). Significant negative correlations,corresponding to increasing consonance, were found in largeareas of orbitofrontal cortex bilaterally, medial subcallosal cin-gulate region (3 mm to the right of the midline) and right frontalpole (Table 1 and Fig. 2b).

Subtraction of images obtained during the most consonant(Diss0) version from those obtained during the most dissonant(Diss5) version confirmed activity in right parahippocampalgyrus (Table 2). Decreases in activity (corresponding to increasedconsonance) were found in right orbitofrontal and medial sub-callosal cingulate regions (Table 2), also confirming findings inthe regression. No activity was observed in precuneus or frontalpolar regions with subtraction analysis, nor were any addition-al regions identified.

To investigate the entire pattern of activity elicited by the stim-uli (as opposed to that related specifically to the variation in dis-sonance), we compared rCBF in conditions Diss0 and Diss5 tothe matched noise control condition. This subtraction revealed,

as expected, activity in superior temporal cortices bilater-ally (Table 2). These regions were similar across both dis-sonance conditions and were found outside of primaryauditory areas, consistent with the use of the noise stimu-lus as a baseline, which would be expected to control fornonspecific auditory stimulation. In addition, this com-parison yielded patterns of activity very similar to thosefound in regression analyses, including reciprocal activa-tion of the parahippocampal gyrus, also observed in theregional covariation analyses described below. Increasedorbitofrontal cortex rCBF was observed in the Diss0 com-parison, whereas decreased rCBF was found in medial sub-callosal cingulate in the Diss5 condition.

Regional covariation analyses (Table 3) were used to deter-mine whether activity in any other brain regions correlatedeither positively or negatively with activity in right parahip-pocampal, right orbitofrontal and medial subcallosal cingu-late regions identified in the original regression (Table 1).Bilateral orbitofrontal and frontal polar activity were found tocovary negatively with activity in right parahippocampalgyrus, whereas right parahippocampal activity was found tocovary negatively with activity in right orbitofrontal andmedial subcallosal cingulate regions (Table 3). Activity inright and medial precuneus regions covaried positively withactivity in right parahippocampal gyrus. Bilateral frontal polaractivity covaried positively with activity in both rightorbitofrontal and medial subcallosal regions.

Using a one-way ANOVA, behavioral ratings of unpleas-ant versus pleasant, tense versus relaxed, irritated versusunirritated, annoying versus unannoying, dissonant versusconsonant and angry versus calm were found to interactsignificantly (p < 0.05) with dissonance level. These ratings

also had positive correlation coefficients with dissonance level,ranging from 0.62 to 0.39 (Fig. 1b). Higher levels of dissonancewere correlated with higher average ratings of adjectives associ-ated with negative emotions (that is, unpleasant, tense, irritat-ed, annoying, dissonant and angry), whereas higher levels ofconsonance were correlated with higher average ratings of adjec-tives associated with positive emotions (pleasant, relaxed, unir-ritated, unannoying, consonant and calm). Ratings of boredversus interested and sad versus happy did not show significantinteractions with amount of dissonance and had correlation coef-ficients of 0.18 and 0.33, respectively (Fig. 1b).

articles

Fig. 1. Examples of music stimuli and average subject ratings of unpleasant ver-sus pleasant and happy versus sad for each version. (a) Excerpts from the mostconsonant version (major triads, Diss0) and the most dissonant version (flatted13th triads, Diss5) of music stimuli used in the PET study. (b) Line graphsdemonstrating averaged subject ratings following scans for each of the six ver-sions, Diss0 through Diss5. Ratings of very pleasant (+5) versus very unpleasant(–5) demonstrated significant interactions (ANOVA; p < 0.001) and a high corre-lation coefficient (r = 0.57) with dissonance level . Ratings of very sad (+5) versusvery happy (–5) did not demonstrate significant interactions and had a lowercorrelation coefficient (r = 0.33) with dissonance level.

Major triads (Diss0)

Flatted 13th triads (Diss5)

Pleasant/Unpleasant Sad/Happy

Rat

ing

s

Rat

ing

s

Dissonance levelDissonance level

Table 1. Regression of rCBF with dissonance level.

Region Brodmann Coordinates t ValueArea x y z

Positive CorrelationsR. parahippocampal gyrus 28/36 25 –28 –21 4.73R. precuneus 7 17 –52 59 4.01R. precuneus 7 8 –57 53 3.92

Negative CorrelationsR. orbitofrontal cortex 14* 13 30 –18 –6.84M. subcallosal cingulate 25 3 18 –15 –6.81L. orbitofrontal cortex 13* –24 32 –14 –4.00L. orbitofrontal cortex 14* –5 41 –21 –3.75R. frontal pole 10 13 65 14 –3.53

Positive correlations denote increasing dissonance; negative correlationsdenote increasing consonance. Coordinates refer to location in stereotaxicspace15. *Nomenclature following ref. 16.


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

Individual subject ratings of unpleasantness versus pleasant-ness of the music were also regressed against rCBF changes(Table 4 and Fig. 3a and b). Ratings of increasing unpleasantnesscorrelated, albeit weakly, with rCBF activity in right parahip-pocampal gyrus, but in the identical location to that found in theregression with dissonance level. Increasing unpleasantness alsocorrelated with activity in posterior cingulate, whereas ratings ofincreasing pleasantness correlated with activity in rightorbitofrontal and medial subcallosal cingulate cortex.

DISCUSSIONThe data presented here demonstrate that rCBFchanges in specific paralimbic and neocortical areasknown to be involved in affective processing correlatewith increasing dissonance or consonance. Theseinclude right parahippocampal gyrus, right precuneus,bilateral orbitofrontal, medial subcallosal cingulateand right frontal polar regions. Different groups ofstructures correlated positively with dissonance asopposed to consonance. The presence of inverseregional covariations suggests that reciprocal func-tional interactions exist between parahippocampaland frontal regions. Furthermore, dissonance wasassociated with certain positive or negative emotion-al ratings, suggesting that the regions in question areinvolved specifically in response to these emotions,rather than other emotions that did not change as afunction of dissonance. The paralimbic and neocor-tical regions identified are distinct from areas of sec-ondary auditory cortex that emerged in controlsubtractions. Activity in these auditory cortical areaslikely represents perceptual processes, including oper-ations related to processing consonance and disso-nance; however, activity in these regions was verysimilar for Diss0 and Diss5 conditions, suggesting thatsimilar perceptual processes were elicited.

Right parahippocampal gyrus activity was foundconsistently throughout regression, subtraction and

covariation analyses. Although the parahippocampal gyrus hasbeen traditionally associated with learning and memory process-es17–19, it has strong reciprocal connections with the amygdala20,suggesting involvement of this region in emotional processes aswell. Previous findings support this conjecture; for example, rCBFincreases in parahippocampal gyrus are associated with unpleasantemotions evoked by pictures with negative emotional valence21.The precuneus region is active in response to a variety of stimuli,including memory-related and selective attention processes22,23,

Table 2. Subtraction analyses.


Diss5 – Diss0rCBF increases

R. parahippocampal gyrus 28/36 25 -28 -21 4.73rCBF decreases

R. orbitofrontal cortex 14* 12 32 -18 -5.82M. subcallosal cingulate 25 0 18 -15 -4.99

Diss5 – noiserCBF increases

R. superior temporal gyrus 22 50 12 –9 3.55L. superior temporal gyrus 41/22 –46 –9 5 3.69R.parahippocampal gyrus 28/36 28 –37 –18 2.86rCBF decreases

M. subcallosal cingulate 25 4 18 –21 –4.06

Diss0 – noiserCBF increases

R. superior temporal gyrus 22 50 10 –3 3.70L. superior temporal gyrus 41/22 –47 –13 5 3.42R. orbitofrontal cortex 14* 12 32 –15 5.42rCBF decreases

R. parahippocampal gyrus 28/36 28 –25 –20 –3.48

Coordinates refer to location in stereotaxic space15. *Nomenclature following ref. 16

a

b

c

d

e

f

Fig. 2. Cortical regions demonstrating significant rCBF correlationswith dissonance level. Regression analyses were used to correlaterCBF from averaged PET data with dissonance level (Diss0 – Diss5).Correlations are shown as t-statistic images superimposed on corre-sponding averaged MRI scans (see Table 1). The t-statistic ranges foreach set of images are coded by color scales below each column, cor-responding to images (a–c) and (d–f). (a–c) Positive correlationswith increasing dissonance demonstrated rCBF activity in rightparahippocampal gyrus (a, sagittal section, x = 25 mm; b, coronal sec-tion, y = –28 mm) and right precuneus (c, sagittal section, x = 8 mm).(d–f) Negative correlations with increasing dissonance (equivalent topositive correlations with increasing consonance) demonstrated rCBFactivity in bilateral orbitofrontal cortex (d, horizontal section,z = –18 mm; also shows medial subcallosal cingulate), medial subcal-losal cingulate (e, sagittal section, x = 3 mm) and right frontal polarregions (f, sagittal section, x = 13 mm).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

suggesting that this region may be related toprocesses not specific to emotion. The find-ing that emotional responses, particularlynegative ones, activate regions related to bothmemory and attention may indicate thatthese processes are interrelated24. However,subject ratings of ‘bored’ versus ‘interested’in the present study did not vary systemati-cally with dissonance level, suggesting thatany differences in arousal between conditionswere minimal.

Orbitofrontal cortex, subcallosal cingu-late and frontal polar regions have all beenimplicated in emotional processing21,25–35.For example, orbitofrontal cortex damagein monkeys disinhibits control of affectiveprocessing30. Identification of emotionalexpression is impaired in patients withlesions in subcallosal and other ventralmedial prefrontal regions31; these impair-ments often occur independently of per-ceptual impairments in recognition ofstimuli presented. The ventromedial por-tion of prefrontal cortices is proposed to beinvolved in making judgments about stim-uli based on their emotional valence28,29.Finally, the subcallosal cingulate regionexhibits decreased baseline rCBF indepressed patients compared to normals32;the present data agree with this finding,because activity in the subcallosal regiondecreased with unpleasantness, whereas itincreased with pleasantness of the stimuli.

Regional covariations, both positive and negative, suggestthere may be functional interactions between regions associ-ated with negative versus positive emotions. Reciprocal region-al activation was found between increasing versus decreasingdissonance conditions, as well as in control subtractions. Thus,increasing activity in certain regions during negative emotionsseems to be associated with a corresponding decrease inactivity in regions that are active during positive emotions.Positive correlations of orbitofrontal and subcallosal cingu-late activity with activity in bilateral frontal polar cortexduring more consonant conditions also suggest functionalinteractions between regions within a given type of affec-tive response.

The regions activated in this study differ from those activatedduring perceptual analysis of music4–8, supporting the hypothesisthat there may be a dissociation between perceptual and emo-tional responses to music. This hypothesis is supported by thelack of significant activity in these regions in regression analyses,as well as by the anatomical similarity of auditory cortical regions

Table 3. Inter-regional covariation of rCBF with regions of interest from Table 1.

Region of Interest Regions of Brodmann Coordinates t ValueCovariation Area x y z

R. Parahippocampal GyrusPositive correlations

R. precuneus 7 7 –54 53 5.18M. precuneus 7 0 –49 36 4.85

Negative correlations

R. orbitofrontal 14* 8 20 –17 –12.83L. orbitofrontal 14* –7 20 –17 –10.64R. frontal pole 10 23 61 –3 –7.31L. frontal pole 10 –28 56 –5 –5.70

R. Orbitofrontal CortexPositive correlations

R. frontal pole 10 17 67 5 6.73L. frontal pole 10 –4 68 5 5.20


R.parahippocampal g 28/36 32 –28 –20 –6.89M. precuneus 7 0 –45 39 –6.79

M. Subcallosal CingulatePositive correlations

R. frontal pole 10 11 67 11 7.93L. frontal pole 10 –4 68 5 6.19


M. precuneus 7 1 –42 39 –6.69R.parahippocampal g 28/36 31 –28 –20 –6.29

Coordinates refer to location in stereotaxic space15. *Nomenclature following ref. 16

Fig. 3. Cortical regions demonstrating significant rCBF correlationswith ratings of increasing unpleasantness and increasing pleasantness.Regression analyses were used to correlate rCBF from averaged PETdata with individual subject ratings of unpleasantness versus pleasant-ness. Correlations are shown as t-statistic images superimposed oncorresponding averaged MRI scans (see Table 2). The t-statistic rangesare coded by color scales below each column, corresponding to images(a–c) and (d, e). (a–c) Positive correlations with increasing unpleas-antness (equivalent to negative correlations with increasing pleasant-ness) demonstrated rCBF activity in right parahippocampal gyrus (a, sagittal section, x = 25 mm; b, coronal section, y = –28 mm; alsoshows left posterior cingulate) and left posterior cingulate (c, sagittalsection, x = –3 mm). (d, e) Positive correlations with increasing pleas-antness demonstrated rCBF activity in right orbitofrontal cortex (d, horizontal section, z = –17 mm) and medial subcallosal cingulate (e, sagittal section, x = –1 mm).

a

b

c

d

e


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

activated in the control stimulus (noise) subtraction for Diss0 andDiss5 (Table 2). Although the specific structures involved differbetween perceptual and emotional responses to music, both thepresent study and some previous studies of music perception5–7

have suggested a relative hemispheric specialization favoring righttemporal and frontal structures. This suggests that circuitry relat-ed to the emotional components of music may be anatomicallyproximal to that used during more perceptual processes. It is alsopossible that right hemisphere dominance of responses in thisstudy is due to a general dominance of right hemisphere activityunderlying emotional processes36,37.

The present study also suggests dissociations between neuralcorrelates of different emotions. Many studies of emotion haveexamined fear perception and conditioning. The amygdala, par-ticularly in the left hemisphere, has been clearly implicated as akey structure in fear processing38–41. In the present study, amyg-dala activation was not detected, and activated structures werefound primarily in the right hemisphere. In addition, sites ofrCBF change in the present study differed between negative ver-sus positive emotions. These findings suggest that the process-ing or experience of different emotions is associated withdistributed activity in different cerebral structures. Because dis-sonance is only one way of eliciting emotional responses to music,it is possible that music that induces different types of emotionswould recruit different neural substrates. This may be especial-ly likely if emotion is elicited through memory or association,rather than spontaneously.

In summary, the findings in this study identify activity in para-limbic and neocortical regions correlated with degree of musi-cal dissonance, and thus begin to characterize the neural basisfor emotional responses to music. These regions have been pre-viously shown to be associated with certain emotional process-es. However, these regions differ from those that are active duringperceptual aspects of music processing, as well as from thoseattributed to processing different emotions. The findings of thisstudy not only begin to define a neural network associated specif-ically with emotional responses to music, but also demonstratedissociations from other important cognitive processes.

METHODSSubjects. Ten normal, right-handed volunteers, five male and five female,participated in this study. Subjects were screened to verify that they had nomore than amateur musical training. Before the PET studies, 11 addition-al subjects (not used in the PET study) were used in pilot studies to verifythat the stimuli produced the desired emotional responses.

Stimuli. A novel melody was made to sound more or less consonant or dis-sonant by varying the harmonic structure of its accompanying chords, pro-

ducing six otherwise-identical versions of a music passage.Each version consisted of a melody, which remained con-stant between versions, and a three-note chord accompa-niment (major triads, dominant 7ths, 9ths, 11ths, 13ths orflatted 13ths, corresponding to dissonance levels 0 through5, denoted as Diss0 through Diss5), used to produce thevarying amounts of dissonance (Fig. 1a). Dissonance levelincreased incrementally between versions, and was uniformthroughout a given version, such that there was no resolu-tion of dissonance in any music passage. The pieces werecomposed specifically for this experiment so that they wereunfamiliar to all subjects, thus eliminating the possibilityof personal associations contributing to the emotional valueof the music. Acoustically matched noise bursts9 were playedas a baseline control stimulus; to be comparable to the musicstimuli, they were constructed to approximate the duration,intensity and onset-offset shape of the melody.

Musical stimuli were created as General MIDI files on aPC platform with Cubase VST 3.5 by Steinberg Software and were pre-sented via a Kurzweil® MASS[ies]® synth engine using the default patch set.At a tempo of 160 beats per minute, the complete melody lasted approxi-mately one minute, 13 seconds. The acoustically matched noise burst stim-ulus was created with Cool Edit Pro 1.1 by Syntrillium Software and Mitsynby WLH. Music stimuli were played for pilot subjects in both a piano and anorgan version to determine which timbre produced the strongest correlationbetween dissonance level and ratings of unpleasantness.

Pilot studies. Eleven pilot subjects listened to each of the six tunes and ratedtheir responses to each tune on an 11-point scale for 12 pairs of contrastingadjectives (see below). Ratings were evaluated using ANOVA and linearregression analysis to identify significant effects of the six different versionson response valence/strength and their correlation coefficients, respective-ly. These preliminary studies determined that dissonance stimuli producedthe desired emotional responses and were thus suitable for use in the PETstudy. The piano version elicited more incremental unpleasant/pleasantratings than the organ version (that is, higher correlation coefficient) andwas thus chosen as the timbre to be used in the PET study. The six pairs ofadjectives that demonstrated significant interactions with dissonance levelin the piano version were used in the PET study to similarly identify subjects’emotional responses to the music (see below). Two pairs of non-significantadjectives also were used in the PET study to demonstrate that certain emo-tions were not elicited by the stimuli.

PET study. PET scans were done on a Siemens HR+ scanner, using the[15O] bolus water technique to measure regional cerebral blood flow(rCBF)42 without arterial blood sampling43. Each subject also received anMRI scan for anatomical registration of PET data44 and resampling into astandardized stereotaxic coordinate system15.

The music and matched noise stimuli described above were used in thePET study. Stimuli were calibrated at 75 dB SPL, and were presented bin-aurally. The melody alone was played twice for each PET subject beforeentering the scanner to familiarize the subject and thereby avoid responsesspecific to hearing novel stimuli. To avoid repetition effects, however, sub-jects did not hear the six different versions of the music passage until PETscans were performed. Before scanning, subjects were also familiarized withanalog rating scales and instructed that we were interested in examiningtheir emotional responses to the music.

Ten subjects were scanned during each of the six versions of the musicpassage, plus the noise control stimulus, for a total of seven scans per sub-ject. Each scan lasted one minute. Stimulus onset occurred approximately10 seconds before scan onset to establish and stabilize subjects’ responsesto stimuli before each scan began. Thus, each stimulus was played only oncethrough the duration of a scan. Scan order was pseudo-randomized betweensubjects to minimize ordering effects. Subjects were instructed to listencarefully to each piece of music as it was played. Subjects used a bipolaranalog scale of –5 to +5 immediately following each scan to rate emotion-al valence and intensity of stimuli. Adjectives selected for ratings in the PETstudy were unpleasant versus pleasant, tense versus relaxed, irritated ver-sus unirritated, annoying versus unannoying, dissonant versus consonant,and angry versus calm, all of which varied significantly with dissonance

Table 4. Covariation of rCBF with ratings of stimulus pleasantness.


Positive CorrelationsM. subcallosal cingulate 25 –1 17 –15 6.53R orbitofrontal cortex 14* 12 32 –17 5.76

Negative CorrelationsL posterior cingulate 23/31 –3 –33 32 –3.69R. parahippocampal gyrus 28/36 25 –28 –21 –2.68

Positive correlations denote increasing pleasantness; negative correlations denote increasingunpleasantness. Coordinates refer to location in stereotaxic space15. *Nomenclature followingref. 16.


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om


articles

level in pilot testing, and bored versus interested and happy versus sad,which did not vary significantly in pilot testing.Data analysis. Regression maps45 were calculated to assess the significanceof the relationship between dissonance level and rCBF, to detect co-vari-ance of regional brain activity, and to assess the significance of the rela-tionship between individual subject ratings and rCBF. Regression analysisinvolves correlation of incremental changes in a specific experimental vari-able, such as stimulus rate, with rCBF. This analysis can also be used to makeinferences of functional connectivity, by identifying correlations betweenactivity in a given volume of interest and activity in the rest of the brain,and is complementary to conventional subtraction analysis. Regressionsused an analysis of co-variance (ANCOVA)46. Values equal to or exceedinga criterion of t = 3.53 were considered significant (p < 0.01, two-tailed),yielding a false-positive rate of 0.58 in 182 resolution elements (each ofwhich has dimensions 14 × 14 × 14 mm), if the volume of brain gray mat-ter is 500 cm3.

As a complementary method of analysis, we also used subtractions47 toidentify regions that may have responded nonlinearly to dissonance. Imagesobtained during Diss0 were subtracted from images obtained during Diss5to isolate rCBF changes due solely to the change in dissonance. Subtractionof the noise control condition from the most dissonant (Diss5) and mostconsonant (Diss0) versions was used to identify rCBF responses due to gen-eralized music perception.

ACKNOWLEDGEMENTSWe thank Christine Beckett for assistance in composing the music stimuli used in this

experiment, and Pierre Ahad for his expertise in sound technology and computer

programming. We also thank the technical staff of the McConnell Brain Imaging

Unit and of the Medical Cyclotron Unit for their assistance, and Sylvain Milot for his

technical expertise. This work was supported by Grants MT11541 and GR13972

from the Medical Research Council of Canada, by the Jeanne Timmins Costello

Fellowship in Neuroscience awarded to A.J.B. by the Montreal Neurological

Institute, and by the McDonnell-Pew Cognitive Neuroscience Program.


1. Davis, W. B. & Thaut, M. H. The influence of preferred relaxing music onmeasures of state anxiety, relaxation, and physiological responses. J. Music Ther.26, 168–187 (1989).

2. Krumhansl, C. L. An exploratory study of musical emotions andpsychophysiology. Can. J. Exp. Psychol. 51, 336–352 (1997).

3. Krumhansl, C. L. Cognitive Foundations of Musical Pitch; Oxford PsychologySeries No. 17 (Oxford Univ. Press, New York, 1990).

4. Milner, B. A. in Interhemispheric Relations and Cerebral Dominance (ed. V.Mountcastle) 177–195 (Johns Hopkins Univ. Press, Baltimore, Maryland, 1962).

5. Zatorre, R. J. Pitch perception of complex tones and human temporal-lobefunction. J. Acoust. Soc. Am. 84, 566–572 (1988).

6. Zatorre, R. J. & Samson, S. Role of the right temporal neocortex in retention ofpitch in auditory short-term memory. Brain 114, 2403–2417 (1991).

7. Zatorre, R. J., Evans, A. C., Meyer, E. & Gjedde, A. Lateralization of phoneticand pitch processing in speech perception. Science 256, 846–849 (1992).

8. Binder, J. R. et al. Human brain language areas identified by functional MRI. J.Neurosci. 17, 353–362 (1997).

9. Zatorre, R. J., Evans, A. C. & Meyer, E. Neural mechanisms underlying melodicperception and memory for pitch. J. Neurosci. 14, 1908–1919 (1994).

10. Peretz, I., Gagnon, L. & Bouchard, B. Music and emotion: perceptualdeterminants, immediacy and isolation after brain damage. Cognition 68,111–141 (1998).

11. Goldstein, A. Thrills in response to music and other stimuli. Physiol. Psychol. 8,126–129 (1980).

12. Dowling, W. J. & Harwood, D. L. Music Cognition 62–89; 202–224 (Academic,Orlando, Florida, 1986).

13. Sloboda, J. A. Music structure and emotional response: some empiricalfindings. Psychol. Music 19, 110–120 (1991).

14. Robazza, C., Macaluso, C. & D’Urso, V. Emotional reactions to music bygender, age, and expertise. Percept. Mot. Skills 79, 939–944 (1994).

15. Talairach, J. & Tournoux, P. Co-Planar Stereotaxic Atlas of the Human Brain(Thieme, New York, 1988).

16. Petrides, M. & Pandya, D. N. Comparative architectonic analysis of the humanand the macaque frontal cortex. Handbook Neuropsychol. 9, 17–58 (1994).

17. Zola-Morgan, S., Squire, L. R., Amaral, D. G. & Suzuki, W. A. Lesions ofperirhinal and parahippocampal cortex that spare the amygdala andhippocampal formation produce severe memory impairment. J. Neurosci. 9,4355–4370 (1989).

18. Bunsey, M. & Eichenbaum, H. Critical role of the parahippocampal region forpaired-associate learning in rats. Behav. Neurosci. 107, 740–747 (1993).

19. Aguirre, G. K., Detre, J. A., Alsop, D. C. & D’Esposito, M. Theparahippocampus subserves topographical learning in man. Cereb. Cortex 6,823–829 (1996).

20. Stefanacci, L., Suzuki, W. A. & Amaral, D. G. Organization of connectionsbetween the amygdaloid complex and the perirhinal and parahippocampalcortices in macaque monkeys. J. Comp. Neurol. 375, 552–582 (1996).

21. Lane, R. D. et al. Neuroanatomical correlates of pleasant and unpleasantemotion. Neuropsychologia 35, 1437–1444 (1997).

22. Berthoz, A., Parietal and hippocampal contribution to topokinetic andtopographic memory. Phil. Trans. R. Soc. Lond. B 352, 1437–1448 (1997).

23. Le, T. H., Pardo, J. V. & Hu, X. 4T-fMRI study of nonspatial shifting of selectiveattention: cerebellar and parietal contributions. J. Neurophysiol. 79, 1535–1548(1998).

24. LeDoux, J. E. Emotional memory systems in the brain. Behav. Brain Res. 58,69–79 (1993).

25. Wheeler, R. E., Davidson, R. J. & Tomarken, A. J. Frontal brain asymmetry andemotional reactivity: a biological substrate of affective style. Psychophysiology30, 82–89 (1993).

26. Rolls, E. T., Hornak, J., Wade, D. & McGrath, J. Emotion-related learning inpatients with social and emotional changes associated with frontal lobedamage. J. Neurol. Neurosurg. Psychiatry 57, 1518–1524 (1994).

27. George, M. S. et al. Brain activity during transient sadness and happiness inhealthy women. Am. J. Psychiatry 152, 341–351 (1995).

28. Bechara, A., Tranel, D., Damasio, H. & Damasio, A. R. Failure to respondautonomically to anticipated future outcomes following damage to prefrontalcortex. Cereb. Cortex 6, 215–25 (1996).

29. Damasio, A. R., The somatic marker hypothesis and the possible functions ofthe prefrontal cortex. Phil. Trans. R. Soc. Lond. B 351, 1413–1420 (1996).

30. Dias, R., Robbins, T. W. & Roberts A. C. Dissociation in prefrontal cortex ofaffective and attentional shifts Nature 380, 69–72 (1996).

31. Hornak, J., Rolls, E. T. & Wade, D. Face and voice expression identification inpatients with emotional and behavioural changes following ventral frontal lobedamage. Neuropsychologia 34, 247–261 (1996).

32. Drevets, W. C. et al. Subgenual prefrontal cortex abnormalities in mooddisorders. Nature 386, 824–827 (1997).

33. Lane, R. D., Reiman, E. M., Ahern, G. L., Schwartz, G. E. & Davidson, R. J.Neuroanatomical correlates of happiness, sadness, and disgust. Am. J.Psychiatry 154, 926–933 (1997).

34. Paradiso, S. et al. Emotional activation of limbic circuitry in elderly normalsubjects in a PET study. Am. J. Psychiatry 154, 384–389 (1997).

35. Zald, D. H. & Pardo, J. V. Emotion, olfaction, and the human amygdala:amygdala activation during aversive olfactory stimulation. Proc. Natl. Acad. Sci.USA 94, 4119–4124 (1997).

36. Lane, R. D., Kivley, L. S., Du Bois, M. A., Shamasundara, P. & Schwartz, G. E.Levels of emotional awareness and the degree of right hemispheric dominancein the perception of facial emotion. Neuropsychologia 33, 25–38 (1995).

37. Erhan, H., Borod, J. C., Tenke, C. E. & Bruder, G. E. Identification of emotion ina dichotic listening task: event-related brain potential and behavioral findings.Brain Cogn. 37, 286–307 (1998).

38. Adolphs, R., Tranel, D., Damasio, H. & Damasio, A. R. Fear and the humanamygdala. J. Neurosci. 15, 5879–5891 (1995).

39. Hugdahl, K. et al. Brain mechanisms in human classical conditioning: a PETblood flow study. Neuroreport 6, 1723–1728 (1995).

40. Morris, J. S. et al. A differential neural response in the human amygdala tofearful and happy facial expressions. Nature 383, 812–815 (1996).

41. Rogan, M. T. & LeDoux, J. E. Emotion: systems, cells, synaptic plasticity. Cell85, 469–475 (1996).

42. Raichle, M. E., Martin, W. R. W., Herscovitch, P., Mintun, M. A. & Markham, J.Brain blood flow measured with intravenous H2(15)O. II. Implementation andvalidation. J. Nucl. Med. 24, 790–798 (1983).

43. Fox, P. T. & Raichle, M. E. Stimulus rate dependence of regional cerebral bloodflow in human striate cortex, demonstrated by positron emission tomography.J. Neurophysiol. 51, 1109–1120 (1984).

44. Collins, D. L., Neelin, P., Peters, T. M. & Evans, A. C. Automatic 3Dintersubject registration of MR volumetric data in standardized Talairachspace. J. Comput. Assist. Tomogr. 18, 192–205 (1994).

45. Paus, T., Perry, D. W., Zatorre, R. J., Worsley, K. J. & Evans, A. C. Modulationof cerebral blood flow in the human auditory cortex during speech: role ofmotor-to-sensory discharges. Eur. J. Neurosci. 8, 2236–2246 (1996).

46. Sokal, R. R. & Rohlf, F. J. Biometry 2nd edn (Freeman, San Francisco,California, 1981).

47. Worsley, K. J., Evans, A. C., Marrett, S. & Neelin, P. A three-dimensionalstatistical analysis for CBF activation studies in human brain. J. Cereb. BloodFlow Metab. 12, 900–918 (1992).


199

9 N

atu

re A

mer

ica

Inc.

• h

ttp

://n

euro

sci.n

atu

re.c

om

Documents

Nature Neuroscience April 1999