editorialThe public face of neuroscience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

news and viewsTwo faces for an opioid peptide—and more receptors for pain research . . . . . . 185Frédéric Simonin and Brigitte L. KiefferSEE ARTICLE, PAGE 201

Right place at the right time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187Dan H. SanesSEE ARTICLE, PAGE 247

Cold emerging from the fog . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189Amy B. MacDermott and C. Justin LeeSEE ARTICLE, PAGE 254

The frontal cortex: does size matter?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190Richard E. PassinghamSEE ARTICLE, PAGE 272

Trust in the brain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192Ralph AdolphsSEE ARTICLE, PAGE 277

book reviewA lighthouse for neural modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195Theoretical Neuroscience: Computational and Mathematical Modeling of Neural Systemsby Peter Dayan and Larry AbbottReviewed by Christof Koch

contents

http://neurosci.nature.com

volume 5 no 3 march 2002

Winston and colleagues usedevent-related fMRI to examinebrain activation while subjectsassessed either trustworthiness orapparent age of unknown faces.Explicit trustworthinessjudgments evoked enhancedactivity in the right superior tem-poral sulcus, whereas increasedactivity in bilateral amygdala andright insula was seen during pre-sentation of faces rated asuntrustworthy regardless of thetask. The findings suggest a func-tional dissociation between inten-tional and automatic judgmentsof trustworthiness.See pages 192 and 277.

nature neuroscience • volume 5 no 3 • march 2002 i

A family of G-protein-coupledreceptors in nociceptive neurons.

Pages 185 and 201.

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature America Inc., 345 Park Avenue South, New York, NY 10010-1707. Editorial Office: 345 Park Avenue South, New York, NY10010-1707. Tel: (212) 726 9200, Fax: (212) 696 9635. Annual subscription rates: USA/Canada: US$199/US$213 (personal), US$99/US$106 (student), Canada add 7% for GST: 140911595RT001;U.K./Europe: £185 (personal), £105 (student); Rest of world (excluding China, Japan, Korea): £235 (personal), £110 (student); Japan: Contact Nature Japan K.K., MG Ichigaya Building 5F, 19-1 Haraikatamachi, Shinjuku-ku, Tokyo 162-0841. Tel: 81 (03) 3267 8751, Fax: 81 (03) 3267 8746. Authorization to photocopy for internal or personal use, or internal or personal use of specif-ic clients, is granted by Nature Neuroscience to libraries and others registered with the Copyright Clearance Center (CCC) Transactional Routing Service, provided the base fee of $9.00 an article (or$1.00 a page) is paid direct to CCC, 27 Congress Street, Salem, MA 01970, USA. Back issues: US$45, Canada add 7% for GST; Application to mail periodicals postage rate is pending at New York,NY. CPC PUB AGREEMENT #40032744. POSTMASTER: Send address changes to Nature Neuroscience Subscription Department, P.O. Box 5054, Brentwood, TN 37024-5054. Printed by Publishers Press,Inc., Lebanon Junction, KY, USA. Copyright © 2002 Nature America Inc.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

contents

nature neuroscience • volume 5 no 3 • march 2002 ii

TrkB and GABAergic interneurondevelopment in the cerebellum.

Page 225.

Rapid actin turnover in dendritic spines.

Page 239.

Inhibitory synapses and auditory coincidence detection.

Pages 187 and 247.

brief communicationsLow glucose–sensing cells in the carotid body . . . . . . . . . . . . . . . . . . . . . . . . . . 197R Pardal and J López-Barneo

Gender-specific induction of enhanced sensitivity to odors . . . . . . . . . . . . . . . . 199P Dalton, N Doolittle and P A S Breslin

articlesProenkephalin A gene products activate a new family of sensory neuron–specific GPCRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201P M C Lembo, E Grazzini, T Groblewski, D O’Donnell, M Roy, J Zhang, C Hoffert, J Cao, R Schmidt, M Pelletier, M Labarre, M Gosselin, Y Fortin, D Banville, S H Shen, P Ström, K Payza, A Dray, P Walker and S AhmadSEE NEWS AND VIEWS, PAGE 185

Differential modulation of Cav2.1 channels by calmodulin and Ca2+-binding protein 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210A Lee, R E Westenbroek, F Haeseleer, K Palczewski, T Scheuer and W A Catterall

Ventricle-directed migration in the developing cerebral cortex . . . . . . . . . . . . . 218B Nadarajah, P Alifragis, R O L Wong and J G Parnavelas

TrkB receptor signaling is required for establishment of GABAergic synapses in the cerebellum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225B Rico, B Xu and L F Reichardt

Rhythmic gene expression in pituitary depends on heterologous sensitization by the neurohormone melatonin . . . . . . . . . . . . . . . . . . . . . . . . . . 234Ce von Gall, M L Garabette, C A Kell, S Frenzel, F Dehghani, P Schumm-Draeger, D R Weaver, H Korf, M H Hastings and J H Stehle

Rapid turnover of actin in dendritic spines and its regulation by activity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239E N Star, D J Kwiatkowski and V N Murthy

Experience-dependent refinement of inhibitory inputs to auditory coincidence-detector neurons. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247C Kapfer, A H Seidl, H Schweizer and B GrotheSEE NEWS AND VIEWS, PAGE 187

Specificity of cold thermotransduction is determined by differential ionic channel expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254F Viana, E de la Peña and C BelmonteSEE NEWS AND VIEWS, PAGE 189

Boosting of neuronal firing evoked with asynchronous and synchronous inputs to the dendrite. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261H Oviedo and A D Reyes

Reduced prefrontal activity predicts exaggerated striatal dopaminergic function in schizophrenia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267A Meyer-Lindenberg, R S. Miletich, P D Kohn, G Esposito, R E Carson, M Quarantelli, D R Weinberger and K F Berman

Humans and great apes share a large frontal cortex . . . . . . . . . . . . . . . . . . . . . . 272K Semendeferi, A Lu, N Schenker and H DamasioSEE NEWS AND VIEWS, PAGE 190

Automatic and intentional brain responses during evaluation of trustworthiness of faces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277J S Winston, B A Strange, J O’Doherty and R J DolanSEE NEWS AND VIEWS, PAGE 192

classifieds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . see back pages

Dopamine and prefrontal cortexfunction in schizophrenia.

Page 267.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

nature neuroscience • volume 5 no 3 • march 2002 183

If the research enterprise is to thrive, it must not only deliverpractical benefits, but also capture the popular imagination. ANew York-based public television station has just released anambitious new series, The Secret Life of the Brain, which aims toconvey the excitement and achievements of modern neuroscienceto a broad general audience. Made by Emmy-winning producerDavid Grubin, with an accompanying book by the popular neu-roscience writer Richard Restak, the series is likely to be widelyviewed, and may have a significant influence on how neuro-science is perceived by the public.

In many ways, the series provides an excellent snapshot of thefield. Each of the five episodes presents one chapter in the lifestory of the brain, from birth to childhood, through the teenageyears to adulthood and old age. The science is accurate and up-to-date, thanks in part to the many prominent neuroscientists whowere interviewed. Perhaps wisely, the producers avoid introduc-ing too much detail. Television is often better at conveyingimpressions and emotions than at presenting complex logicalarguments, and the producers play to these strengths, never lettingtechnical issues detract from the human drama of the subject. Inthe fourth episode, on the adult brain, Antonio Damasio arguesagainst the traditional separation of thought and emotion, andhis dictum “We are not thinking machines, we are feelingmachines that think” might have been the producers’ motto.

The series opens with Emily Dickinson’s poem The brain iswider than the sky, and the final episode ends with a moving recita-tion by the 95-year-old American poet laureate Stanley Kunitz. Thewhole series is overtly poetic in its intentions, beautifully producedand full of evocative stories and images. Some are tragic—a womanholding back tears as she describes how her husband lost his capac-ity for emotional understanding following a stroke 23 years ago.Others are uplifting—children whose determination helped themrecover after losing half their cortex to epilepsy; or writer LaurenSlater, who overcame suicidal depression to write a widelyacclaimed book describing her experiences. The brain’s life-longplasticity offers the hope of new therapies for many conditions pre-viously regarded as untreatable, and this optimistic message is thecentral theme of the entire series. Although intended for a lay audi-ence, basic neuroscientists who seldom encounter patients will findhere a powerful reminder of the human dimension to their work.

Nevertheless, although it is carefully balanced in some ways—notably the many interviews with prominent womenresearchers—The Secret Life of the Brain suffers from severalflaws. First, it is unremittingly ‘America-centric’. The scientistswho appear here are almost exclusively US-based, and the con-tributions, past and present, of Europe, Japan, Australia andother countries are never acknowledged. Similarly, to referrepeatedly to the “millions of Americans” suffering from vari-ous brain disorders is to ignore many more who live elsewhere.

It is true that the US invests more in neuroscience research thanany other country, and that it represents the largest potentialmarket for the new treatments that the series promises. It is per-haps understandable that the program’s sponsors—which includethe National Science Foundation and Pfizer among others—wantto reach American voters, consumers and (with luck) philan-thropists. But the US surely has enough isolationist tendencieswithout having them echoed in its science documentaries.

The series also downplays the importance of animal research.There is one striking moment when the narrator describes Mriganka Sur’s experiments on the rewired ferret cortex as thecamera moves across Leonardo’s portrait Lady with an Ermine(the animal is actually a ferret), inviting viewers to ponder thesimilarities and differences between the sitter and her pet. Butapart from this scene, animal experiments are hardly mentioned.The emphasis on human-interest stories is understandable, butthe naive viewer will have no idea that animal experiments havebeen critical in almost every discovery described here. At a timewhen public opposition to animal research poses a major threatto the future of the field, this seems like an important omission.

Finally, the focus on cognitive neuroscience and its practicalimplications means that many of neuroscience’s greatest achieve-ments and most important questions are never mentioned. Stud-ies of language, emotion and drug craving make good stories,but some of the hypotheses explored here—for instance, the linkbetween teenage waywardness and late maturation of prefrontalcortex—are tentative at best, and the program does not conveyhow much the frontiers of cognitive neuroscience depend on arigorous foundation of cellular and molecular research. Thegraphics depict neural pathways as pipes and sprinklers that con-vey happy, sad or fearful signals across the brain, but althoughthe program pays the usual lip service to the brain’s vast com-plexity, it offers no insight into how networks of neurons mightoperate. The neural underpinnings of perception, decision-mak-ing and consciousness, with their attendant philosophical ques-tions, are not discussed at all. A lack of respect for reductionismis reflected in the synapse animations, which are graphicallysophisticated but fundamentally misleading—neurotransmittermolecules are shown migrating purposefully as if animated bysome unseen intelligence, rather than diffusing according to thefamiliar laws of physics. An opportunity was lost here to conveythe idea of neurons as machines that obey natural laws, and thusto suggest continuity between mind and inanimate matter.

The idea that 100 billion neurons can give rise to humanmental life is, as Francis Crick has said, an astonishing hypoth-esis. Perhaps the producers of Secret Life of the Brain thoughtit was too much for their audience, but this would have beenan even better series if it had tried to confront the central mys-teries of neuroscience head-on.

editorial

The public face of neuroscience

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

nature neuroscience • volume 5 no 3 • march 2002 185

Nociceptive sensory neurons conveynoxious stimuli from the periphery tothe central nervous system and are keyplayers in both acute and persistent pain.Their cell bodies are located in dorsalroot ganglia (DRG), and their axonsproject to several laminae of the dorsalhorn of the spinal cord. These pain sen-sors express a variety of signaling molecules with pronociceptive orantinociceptive activity, notable exam-ples being substance P and opioid pep-tides, respectively. Because theyrepresent primary sites for pain process-ing, DRG neurons have been the focusof intense research to identify moleculartargets of pain neurotransmission, withthe aim of developing potent analgesiccompounds. In this issue, Lembo et al.1

of AstraZeneca report the cloning of afamily of G-protein-coupled receptors(GPCRs) expressed solely in DRGs. Theauthors propose that BAM22, an opioidpeptide, is the endogenous ligand forthese receptors. Most intriguingly, thispreproenkephalin product uses its non-opioid part to activate these receptors.BAM22 therefore has dual opioid andnon-opioid activities.

Using degenerate PCR with primerscorresponding to highly conservedregions of GPCRs, Lembo et al. isolat-ed a novel GPCR-encoding cDNA fromrat DRG. Using this rat cDNA as aprobe, they further cloned a family ofsix homologous genes in humans. Thereceptors showed a unique expressionpattern restricted to dorsal root andtrigeminal ganglia in rat and human,and therefore the authors named them‘SNSR’ (sensory neuron specific recep-tors). Double-labeling experiments fur-ther narrowed down SNSR distributionmainly to isolectin B4-positive, small-diameter nociceptors, which are one oftwo major classes of unmyelinated

terminal opioid sequence followed by aseventeen amino-acid C-terminal exten-sion. This peptide was isolated frombovine adrenals and, like other opioidpeptides, binds to all three opioid recep-tors with high affinity. Its biological rolehas not been explored extensively.

The authors’ analysis of shorter ver-sions of BAM22 indicated that aminoacids 8–22 were necessary for calciummobilization via SNSRs, but the opioidpart, amino acids 1–7, were notrequired, suggesting a complete disso-ciation of opioid and SNSR agonistactivities in the peptide. Furtherdose–response analyses confirmed thepreviously known activity of BAM22 atthe delta opioid receptor, and, using aprototypal opioid antagonist (nalox-one), the authors showed that there was no opioid component in theBAM22–SNSR3 interaction. Finallythey synthesized a tritiated Bam 8–22peptide and confirmed direct high-affinity binding at SNSRs, as well as alack of competition by various opioids.Taken together, the data define SNSRs

Two faces for an opioid peptide—andmore receptors for pain researchFrédéric Simonin and Brigitte L. Kieffer

A new study reports the cloning of a family of ‘orphan’ G-protein-coupled receptors that arelocalized in human and rat small sensory neurons. These receptors are activated by a peptidederived from preproenkephalin A and may be involved in modulating nociception.

fibers (C fibers) mediating noxiousstimuli (Fig. 1).

The authors’ next goal was to find aligand for this class of orphan receptors.Using an assay based on intracellular cal-cium mobilization (widely used inGPCR research), they tested a vast panelof commercially available compounds incells stably expressing SNSR 3 or SNSR 4. The opioid peptide BAM22potently activated the two receptors,whereas many related peptides wereinactive. Opioid peptides are a family ofneuropeptides that are processed fromlarge precursor proteins known as pro-opiomelanocortin, preproenkephalinand preprodynorphin2. All opioid pep-tides share a common N-terminalsequence (Tyr-Gly-Gly-Phe-Leu/Met),which interact with mµ, δ and κ recep-tors. The three receptors were identifiedas a subfamily of highly homologousGPCRs3 and are expressed in the centralnervous system, including the DRG andthe dorsal horn of the spinal cord4.BAM22, one of many preproenkephalinproducts5, consists of the canonical N-

news and views

The authors are at the IGBMC, 1 rue L:aurentFires, 67400 Illkirch, France.e-mail: briki@esbs.u-strasbg.fr

Dorsal root ganglion

trkA, p75CGRP, SPBDNF

RET, IB4GFRα1/2TMPP2X3

Spinal cord

Pain fibers in skin and deep tissue

Thalamus and neocortex

Brainstem

SNSRs/mrgsmRNAs

Bam22

Fig. 1. Schematic diagram of pain pathways from the periphery to the CNS. DRG neurons are het-erogenous and show distinct phenotypic and cellular features. Because they have distinct conductionvelocities, Aδ fibers (myelinated) are assumed to mediate acute sharp pain, whereas C fibers (unmyeli-nated) convey dull delayed pain12. C fibers only are represented. Two major classes of C fibers havebeen identified11. One class consists of neurons that express the nerve growth factor receptor TrkA,as well as the neuropeptides substance P and CGRP, and project to lamina I and the outer part of lam-ina II of the dorsal horn of the spinal cord. The second class of C fibers binds isolectin B4 and projectsto the inner part of lamina II. Both classes of neuron are critical in pain pathogenesis, including inflam-matory and neuropathic pain13–15. SNSR/mgr receptors are mainly expressed in isolectin B4–positivecells and are proposed to be activated endogenously by the preproenkephalin-derived peptide BAM22.

Amy Center

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

186 nature neuroscience • volume 5 no 3 • march 2002

as a non-opioid class of receptors andBAM22 as a peptide with two distinctmolecular targets, opioid and SNSRreceptors. This ‘de-orphanization’ ofSNSR receptors is a most intriguingaspect of Lembo’s report. Many orphanGPCRs have been identified in therecent years6, and the search for ligandshas intensified7, particularly in thepharmaceutical industry, to broadenpanels of therapeutic targets. The con-cept of a dual activity for an opioidpeptide has been proposed previously,although not for BAM22. δ-endorphin,a major opioid peptide produced fromthe prepro-opiomelanocortin precur-sor, binds to immune cells via its C-ter-minal portion8 and enhances immunefunction in a non-opioid manner9. Theputative receptor was never character-ized at the molecular level, and mech-anisms have remained elusive. Thecurrent findings on BAM22 are muchmore open to further explorationbecause the receptor is cloned.

Lembo’s findings nicely parallel therecent study of Dong et al.10. Using sub-tractive cloning, the authors of thatstudy identified a novel family of GPCRsin mice and humans and demonstratedthat some members of this family arespecifically expressed in primary senso-ry neurons. The receptors were calledmrgs (Mas-related genes) because oftheir relatively high homology with theGPCR Mas oncogene. The mrg familycan be divided into three major homol-ogy groups (mrgA, mrgB and mrgC)and comprises 31 murine and 8 humangenes with intact coding sequences,along with related pseudogenes. Threeof the six human SNSR sequences iden-tified by Lembo et al.1 are identical tohuman mrgs, whereas the three othersequences are close paralogues. Thehuman SNSR sequences seem most sim-ilar to the seventeen murine mrgAsequences, although it was not possibleto define clear orthologous pairs. Strik-ingly, despite repeated attempts, Lemboand coworkers identified only one SNSRin rat. This rat sequence shares highesthomology with the mrgC subfamily,which otherwise seems to consist ofpseudogenes in mouse. Altogether, datafrom the two groups highlight notablespecies differences in the expression ofmrg/SNSR receptors. These discrepan-cies may suggest important functional

unknown. SNSR and opioid activities ofBAM22 may well interact either syner-gistically or in opposition. It is thereforecritical to clarify the non-opioid biolog-ical activity of BAM22 in vivo and tolook for endogenous peptides showingonly SNSR activity. Mice lacking pre-proenkephalin or all three opioid recep-tors, both of which exist, may contributeto understanding these issues.

In conclusion, pain research hasrevealed a limited number of opioid recep-tor genes3. The discovery of an entirelynew and distinct receptor gene family foran opioid peptide opens exciting possibil-ities for unanticipated roles of endogenousopioid peptides in nociception. SNSRcloning may also lead to new pain modu-latory peptides and will contribute tounderstanding the role of isolectin B4-positive cells in the induction of pain.Finally the cloning of SNSRs has impor-tant clinical implications, as SNSR ligandscould potentially be developed for thetreatment of chronic pain conditions,some of which are resistant to opiates. Inaddition, the highly restricted distributionof SNSRs is of great advantage in devel-oping drugs with limited side effects.

1. Lembo, P. M. C. et al. Nature Neurosci. 5,201–209 (2002).

2. Akil, H. et al. Annu. Rev. Neurosci. 7,223–255 (1984).

3. Kieffer, B. L. Cell. Mol. Neurobiol. 15,615–635 (1995).

4. Mansour, A., Fox, C. A., Akil, H. & Watson,S. J. Trends Neurosci. 18, 22–29 (1995).

5. Rossier, J. in Opioids I. Handbook ofExperimental Pharmacology (eds. Herz, A.)423–441 (Springer, Berlin, 1993).

6. Lee, D. K., George, S. R., Evans, J. F., Lynch,K. R. & O’Dowd, B. F. Curr. Opin.Pharmacol. 1, 31–39 (2001).

7. Civelli, O. et al. Trends Neurosci. 24, 230–237(2001).

8. Woods, J. A., Shahabi, N. A. & Sharp, B. M.Life Sci. 60, 573–586 (1997).

9. Van den Bergh, P., Rozing, J. & Nagelkerken,L. Lymphokine Cytok. Res. 13, 63–69 (1994).

10. Dong, X., Han, S., Zylka, M. J., Simon, M. I.& Anderson, D. J. Cell 106, 619–632 (2001).

11. Snider, W. D. & McMahon, S. B. Neuron 20,629–632 (1998).

12. Julius, D. & Basbaum, A. I. Nature 413,203–210 (2001).

13. Mantyh, P. W. et al. Science 278, 275–279(1997).

14. Malmberg, A. B., Chen, C., Tonegawa, S. &Basbaum, A. I. Science 278, 279–283 (1997).

15. Nichols, M. L. et al. Science 286, 1558–1561(1999).

differences in the processing of noci-ceptive inputs across species, and raiseinteresting questions with regard towhich animals should be studied for thedevelopment of novel analgesics.

On another front, both studies strong-ly suggest that members of the mrg/SNRSGPCR family are receptors for neuropep-tides. Dong et al. screened a series ofknown neuropeptides on two of theirmouse mrgs, using an assay similar toLembo’s. They did not test BAM22, butthey found agonist activity for RF-amidepeptides10, one of which—called NPFF—is known to exhibit anti-opioid proper-ties. RF-amide peptides show extremelyhigh potency at two other GPCRs thathave been identified recently as their gen-uine receptors (unrelated to mrg/SNSRs).One can speculate that BAM22, ratherthan RF-amide peptides, may be theendogenous ligand for the mrgs identi-fied by Dong et al.10. Based on the multi-plicity of mrg/SNSR receptors, it ispossible that— beyond BAM22— novelendogenous peptides or binding activitiesfor already-known neuropeptides will beidentified for this new receptor family.

Finally a unique feature of SNSRs istheir expression pattern, restricted toDRG nociceptors. Both studies1,10 showthat all the genes tested in adult DRG(mrg1, 3, 4, mrgD and rat SNSR) areexpressed essentially only in isolectinB4-positive cells. Specific expression insensory neurons is not unique to SNSRs.Other pain signaling molecules, such asthe P2X3 purinoceptor, a tetrodotoxin-resistant voltage-gated sodium channeland the vanilloid VR1 receptor, are alsoexpressed solely in those neurons11. Asproposed by the authors, restriction ofSNSR expression to DRG nociceptorssuggests a possible role for this receptorfamily in pain control. At present, how-ever, the physiological role of theBAM22/SNSRs system is unknown. Pre-cise localization of receptor proteins andpeptide ligand(s) could aid in deter-mining their function.

The obvious next experiment is totest whether Bam 8–22 has analgesicproperties in vivo. Although the authorssuggest this, there is no basis yet forspeculation about whether SNSRs acti-vation is pro- or antinociceptive. Thisraises the interesting issue of dual opi-oid/SNSR activity for BAM22, the for-mer inhibiting pain and the latter being

news and views©

2002

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co

m

nature neuroscience • volume 5 no 3 • march 2002187

When I took a driver’s education course asa teenager, I recall that the car wasequipped with an extra set of brakes belowthe front passenger’s seat where ourinstructor sat. This arrangement generat-ed some interesting results: for example, anaggressive student applied the acceleratorto avoid hitting a bounding squirrel and,in near synchrony, our formerly calminstructor hammered the brakes. Unfortu-nately, the integration of signals was notfavorable for the squirrel. However, we stu-dents did learn an important lesson: thebrakes are optimally placed at a single posi-tion, preferably adjacent to the accelerator.A new study by Kapfer and coworkers1 sug-gests that inhibitory synapses (the ‘brakes’)are also best located at a single position oncentral auditory neurons when temporalprocessing is of great importance. Themost intriguing aspects of this anatomicalspecificity are that it emerges duringinhibitory synapse elimination and that itrequires normal auditory experience.

For decades we have known that ani-mals can locate a sound based on thesmall difference in its arrival times at thetwo ears. These interaural time differ-ences (ITDs) are created by a soundsource located to one side of an animal’shead. Low sound frequencies are themost useful signals for detecting ITDs,and animals with good sensitivity below1,500 Hz tend to perform best on thistask2. For example, consider a gerbil thatis under attack from an eagle owl, a cred-ible event on the steppes of Mongolia.Although both ears hear the owl perfect-ly well, the signal obviously arrives firstat the closer ear (Fig. 1a). The ITD isdetermined by the distance between theanimal’s two ears, which produces a dif-ference of no more than 150 µs for smallrodents such as the gerbil.

A group of brainstem neurons, knowncollectively as the medial superior olivary(MSO) nucleus, are well suited to detecting these small time differences. The

direct evidence to resolve this question,the new observations by Kapfer et al.1 lendserious anatomical and phylogenetic sup-port for a selective role of inhibitorysynapses during ITD processing.

In adult MSO neurons, inhibitorysynapses are located chiefly on the soma8

(Fig. 2). Kapfer et al.1 counted synapticcontacts at the light microscopic levelusing both presynaptic (glycine) and post-synaptic (glycine receptor and gephyrin)markers. They found that for every fivesynapses located on the adult MSO cellbody, there is only one on the dendrites.Surprisingly, the inhibitory terminals areequally distributed along the soma anddendrite at 10 days after birth. The adultinnervation pattern does not emerge fromthe addition of inhibitory terminals to thesoma, but rather is created by the selec-

MSO neuron receives two sets of excita-tory inputs, with each set activated by oneear (Fig. 1b). When both excitatory path-ways are activated at precisely the sametime, the MSO neuron detects the coin-cident excitatory potentials and firesaction potentials. When the pathways areactivated asynchronously, the MSO neu-rons do not respond. Thus, the dischargerate of an MSO neuron varies with therelative latency of the two inputs and,therefore, the position of a soundalong the horizon3.

The cellular and theoretical basesof coincidence detection havereceived a great deal of attention, andunsurprisingly these studies havefocused on excitatory connectionsand voltage-gated channels. Forexample, glutamate receptors andpotassium channels with very rapidkinetics are implicated in temporalprecision4. Furthermore, sophisticat-ed models can largely account forMSO neuron behavior with only twosets of excitatory inputs5.

Despite the simple elegance of anexcitatory coincidence detector, thereis a nagging need to account for thepresence of inhibitory terminals onMSO neurons. There are two inde-pendent sets of inhibitory glycinergicprojections to the MSO arising fromthe medial and lateral nuclei of the lat-eral lemniscus (MNTB and LNTB),each set being driven by one ear (Fig. 1c). One possibility is thatinhibitory synapses are always active,and they simply decrease the mem-brane resistance and membrane timeconstant. This would tend to decreasethe duration and amplitude of all exci-tatory potentials and shorten the win-dow for temporal summation. Asecond possibility is that inhibitorysynaptic activity occurs with tempo-ral precision and participates directlyin ITD coding. Indeed, the inhibitorysynapses on MSO neurons producerapid postsynaptic currents6, andthese same inhibitory afferents canaccurately follow stimuli up to severalhundred Hertz7. Although there is no

news and views

Right place at the right timeDan H. Sanes

A puzzle of systems neuroscience, how the CNS encodes timeintersects with a question that fascinates developmentalbiologists, the elimination of synaptic connections.

The author is in the Center for Neural Science,New York University, 6 Washington Place, NewYork, New York 10003, USA. e-mail: sanes@cns.nyu.edu

b Excitatory model of MSO

ITD

c Excitatory-inhibitory model of MSO

a

BOB CRIMI

MSO

MNTB

LNTB

MSO

Cochlea

Cochlearnucleus

Fig. 1. Time difference analysis in the gerbil brainstem. (a) A schematic of interaural time differencedetection by a gerbil. The ITD is produced by theongoing time of arrival of an acoustic signal at thetwo ears. (b) Neurons of the medial superior oli-vary (MSO) nucleus receive excitatory projectionsfrom each cochlear nucleus. MSO neurons respondto sounds that produce coincident excitatory activ-ity. This excitatory circuit explains much of MSOneuron behavior. (c) MSO neurons actually receiveinhibitory projections from two nuclei. Glycinergicprojections from the medial nucleus of the trape-zoid body (MNTB) are activated by the contralat-eral ear, and glycinergic projections from the lateralnucleus of the trapezoid body (LNTB) are activatedby the ipsilateral ear. The inhibitory terminals arelocalized to the MSO somata, and are hypothesizedto increase the precision of ITD coding.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

188 nature neuroscience • volume 5 no 3 • march 2002

tive elimination of inhibitory synapsesfrom dendritic membrane.

The elimination of excitatory synap-tic connections has been closely studied,and the process is known to be influencedby the activity of the synapses themselves.The thinking is that molecular cues cre-ate accurate projection patterns, but thatthe final location and number of excita-tory contacts is influenced by synapticactivity. Thus, macroscopic innervationpatterns such as topography and columnsare not determined by synaptic activi-ty9,10. However, auditory and visual expe-rience can affect the coding properties ofindividual central neurons by restrictingtheir anatomical connections11,12.

Given this evidence from excitatorysynapses, might the developmental reor-ganization of inhibitory synapses in theMSO also be activity dependent? Kapferet al. first noticed that inhibitory synaps-es are eliminated during the period ofpostnatal development when gerbilsbegin to experience airborne sound. Thissuggested that acoustic experience mightbe involved. When they surgically elim-inated one cochlea before the onset ofhearing, the glycinergic terminalsremained on the MSO dendrites intoadulthood (Fig. 2). Indeed, there weremore glycinergic terminals than in con-trol juvenile neurons, possibly because ofsprouting. Alternatively, the extra termi-nals may be present at the time ofcochlea removal (postnatal day 7), butalready eliminated at the juvenile agewhen synapse counts were first obtained(postnatal day 10).

The functional effect of unilateralcochlear ablation is not entirely clear

of ITD information, it is notoriously dif-ficult to record from this structure in vivo.A simpler approach might be for model-ers to explore the effect of compartmen-talizing inhibitory conductances on MSOneuron coding properties.

If inhibitory terminals are refinedthrough a process that is largely depen-dent on experience with ITDs, then itshould be possible to convert a rat MSOinto a gerbil MSO (at least with regardto the location of its glycinergic termi-nals) by designing an acoustic environ-ment that activates excitatory andinhibitory afferents in a temporally pre-cise manner. Amplitude-modulatedtones presented through miniature ear-phones at many ITDs might do the trick.Finally, the experience-dependent reor-ganization of inhibitory synapses with-in the MSO suggests that activity mightadjust their strength, thus leading totheir elimination. In fact, MNTB synaps-es do display long-term inhibitorysynaptic depression, particularly in juve-nile animals15. No matter how thesequestions are resolved, it is now clearthat MSO provides a compelling linkbetween sensory coding properties andinhibitory synaptic development. Per-haps someone should study those mys-terious excitatory connections.

1. Kapfer, C., Seidl, A. H., Schweizer, H. &Grothe, B. Nat. Neurosci. 5, 247–253 (2002).

2. Yost, W. A. & Gourevitch, G. DirectionalHearing (Springer, New York, 1987).

3. Spitzer, M. W. & Semple, M. N. J. Neurophysiol. 73, 1668–1690 (1995).

4. Trussell, L. O. Annu. Rev. Physiol. 61, 477–496(1999).

5. Brughera, A., Stutman, E. R., Carney, L. H. &Colburn, H. S. Aud. Neurosci. 2, 219–233(1996).

6. Smith, A. J., Owens, S. & Forsythe, I. D. J. Physiol. (Lond.) 529, 681–698 (2000).

7. Joris, P. X. & Yin, T. C. J. Neurophysiol. 73,1043–1062 (1995).

8. Henkel, C. K. & Brunso-Bechtold, J. K. J. Comp. Neurol. 354, 470–480 (1995).

9. Harris, W. A. J. Comp. Neurol. 194, 303–317(1980).

10. Katz, L. C. & Crowley, J. C. Nat. Rev. Neurosci.3, 34–42 (2002).

11. Guo, Y. & Udin, S. B. J. Neurosci. 20,4189–4197 (2000).

12. DeBello, W. M., Feldman, D. E. & Knudsen, E. I.J. Neurosci. 21, 3161–3174 (2001).

13. Kitzes, L. M., Hageyama, G. H., Semple, M. N.& Kil, J. J. Comp. Neurol. 353, 341–363 (1995).

14. Grothe, B. J. Neurophysiol. 71, 706–721(1994).

15. Kotak, V. C. & Sanes, D. H. J. Neurosci. 20,5820–5826 (2000).

because it causes a great deal ofafferent sprouting. For example,MNTB neurons lose excitatoryafferents from the contralateralcochlear nucleus when the con-tralateral cochlea is removed, butthey gain a set of excitatory affer-ents from the ipsilateral cochlearnucleus due to local sprouting.Furthermore, the excitatorycochlear nucleus afferents do notremain restricted to one MSOdendrite, but sprout to the inner-vate them both13. Given thisanatomical tangle, it is likely thatthe entire MSO neuron willreceive synchronous excitationand inhibition whenever sound ispresent at the remaining ear. Thatis, any temporal precision of exci-tatory and inhibitory potentials

would be lost. To further test whetheracoustic experience is involved, gerbilpups were reared in an omnidirectionalwhite noise environment, effectivelyreducing experience with ITDs duringdevelopment. Although the effect wasmodest, this manipulation also reducedinhibitory synapse elimination from theMSO dendrites.

Finally, Kapfer and colleagues asked aninteresting phylogenetic question. Notingthat ITD processing is performed best byanimals with low-frequency hearing, suchas gerbils, they asked whether the selectivepattern of inhibitory innervation was cor-related with an animal’s ability to use thiscue. High-frequency sounds cannot beused for ITD processing because the wave-length is shorter than the animal’s head,and the signal’s phase becomes ambigu-ous. Thus, mammals with only high-fre-quency hearing cannot localize a soundusing ITD coding. Interestingly, Kapfer etal. found that in animals with high-fre-quency hearing (rats, opossums and bats),the glycinergic terminals are not restrictedto the MSO soma.

It remains possible that inhibitorysynapses on MSO operate in an identicalfashion whether or not they are restrict-ed to the soma. For example, the phar-macological blockade of glycinergicinhibition to the bat MSO does perturb atemporal coding property (e.g., a selec-tive response to amplitude modulationfrequency), even though inhibitory ter-minals are not confined to the soma14.Although one hopes that a pharmacolog-ical blockade of inhibition will be exam-ined in the gerbil or cat MSO todetermine whether there is a specific loss

news and views

JuvenileCochlear ablation

MSO neuron

Glycinergicbouton

Adult

BOB CRIMI

Fig. 2. Inhibitory glycinergic terminals are initiallyfound on both the MSO soma and dendrites in juvenileanimals. During normal maturation, inhibitory termi-nals are selectively eliminated from the dendrite, lead-ing to a specific somatic localization in adults. Whengerbils are unilaterally deafened at postnatal day 7, theelimination of inhibitory terminals does not occur dur-ing development. Instead, there are more terminalsthan normal on both the soma and dendrite.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

nature neuroscience • volume 5 no 3 • march 2002 189

news and views

Cold is detected by a small subpopulationof peripheral sensory nerves that enter thecentral nervous system in the superficialdorsal horn of the spinal cord and respondto cold with bursts of action potentials1.Despite extensive research into the mecha-nism of cold transduction, there is noagreement on the specifics. Mostly, investi-gators have been searching for a single-channel mechanism analogous to that fortransduction of noxious heat: a strongstimulus (>43°C) activates vanilloid recep-tor-1 (VR1), a recently identified protein inthe transient receptor potential (TRP) fam-ily that provides one of the major mecha-nisms for noxious heat transduction byperipheral nerves2. VR1 channels open,allowing depolarizing current to flow intothe heat-sensitive nerves and evoke actionpotential firing. However, it may be thatwith cold, the mechanism is not so simple.In this issue, Viana et al.3 suggest a new wayof thinking about transduction of a coldstimulus into an excitatory signal. Usingtrigeminal ganglion neurons from newbornmice, they show that cold transduction isnot simply due to activation of a singlechannel but rather is an emergent propertydependent on the expression, density andactivation of several different channelsexpressed in cold-sensitive neurons.

The peripheral fibers of some sensoryneurons respond to gentle cooling, and oth-ers respond to noxious cold. These fibers firesingle or multiple bursts of action potentialsin a pattern that reflects the intensity orrapidity of cooling4. Cold-sensitive nervesare generally considered to be a small sub-set of the small-diameter fibers, althoughwith extreme cold temperatures, a muchlarger proportion of sensory nerves may becold sensitive5. To identify cold-sensitivetrigeminal neurons, Viana et al.3 put disso-ciated neurons into short-term cell culture,

VR1, CMR1 is in the TRP family of pro-teins, reinforcing the idea that these TRPproteins are important in transduction oftemperature.

According to Viana et al.3, the coldresponse is determined not only by theprominent expression of certain channelsin cold-sensitive neurons, but also by theprominent expression of another K+ cur-rent, termed IKD (Fig. 1a), in cold-insen-sitive neurons. In these neurons, the stronghyperpolarizing influence provided by IKDprevents the depolarizing action of coldfrom eliciting action potentials (Fig. 1b).In cold-sensitive neurons, however, the ini-tial depolarization is not prevented by IKD(Fig. 1a) because these channels areexpressed at low levels. The importance ofIKD in preventing cold response in cold-insensitive neurons was tested by addingthe IKD blocker 4-aminopyridine (4-AP)to the bath and re-testing for sensitivity tocold stimulus. Under these conditions,some previously cold-insensitive neuronsbecome responsive to cold (Fig. 1c), indi-cating that a potassium-channel brake pre-vents these neurons from responding. This,in turn, raises the interesting possibility thatsmall changes in the balance of channelexpression or properties in cold-insensitiveneurons could transform cold-insensitivefibers into cold-sensitive fibers in vivo.

The study by Viana et al.3 brings us astep closer to understanding the complexmechanism of cold transduction, but muchremains to be clarified. Peripheralsomatosensory nerves are responsible forreporting the occurrence and location oftouch, temperature, pain and position.Detection of the stimuli for these sensationsoccurs in the peripheral terminals of thesensory nerves. It is important to bear inmind that Viana et al. performed all of theirexperiments at the level of the soma, whichmay not reflect the density and distributionof channels found in the terminals of thesame cells. Nevertheless, by identifyingmechanistic elements of cold transduction,Viana et al. make it possible to hypothesizenew mechanisms contributing to the set-ting of cold threshold. Many of the chan-nels contributing to the cold response aremodulated either directly or indirectly bychanges in intracellular second messengers,particularly cyclic nucleotides, makingthem good candidates for dynamic regula-tion of cold threshold. For example, theactivation curve of HCN channels shifts tothe right as cyclic nucleotides bind to theintracellular binding site11. Under theseconditions, HCN channels will provide amore powerful depolarizing influence onmembrane potential and contribute to a

then used a cold-evoked change in intra-cellular calcium concentration to rapidlyscreen for the small subpopulation (9%)that were sensitive to cold. Each responsivecell was studied using electrophysiologicaltechniques to probe the mechanism gener-ating cold sensitivity. Some cold-sensitivecells depolarized transiently to drive a burstof action potentials, and others showedoscillations in membrane potential thatevoked repeated bursts of action potentialssimilar to the activity of cold fibers record-ed in the skin. The current driving thedepolarization, referred to as Icold, was pro-posed to reflect activation and de-activationof multiple channels with different time,temperature and voltage dependencies,which produce a net decrease in conduc-tance that is usually transient.

Viana et al.3 found that as the tempera-ture drops, the leak channels that normal-ly conduct potassium ions (K+) close,causing membrane depolarization (Fig. 1a).As this occurs, the depolarizing influence ofthe inwardly rectifying Ih current mediatedby HCN channels (hyperpolarization-acti-vated cyclic nucleotide–gated K+ channel6)begins to diminish as a result of the decreasein membrane potential and the cold stim-ulus itself. While Icold is on, the net depo-larization causes action potential firing andassociated calcium entry through voltage-gated calcium channels. This complex inter-play of channels that all contribute to thecold response is in contrast to previouslyproposed mechanisms that have focused oncold modulation of single channels orpumps. These hypotheses include a simpledecrease in leak conductance, such as mightbe mediated by the two-pore potassiumchannel TREK-1 (ref. 7), modulation ofNa+–K+ ATPase activity or modulation ofchannels in the ENac\DEG family8. It wasrecently proposed that cold-induced cur-rent is due to activation of a relatively non-selective cation current, activated with alower threshold in the presence of menthol9.A cold- and menthol-sensitive receptor(CMR1) with the same characteristics hasvery recently been cloned by McKemy etal.10, adding weight to the hypothesis thatthis cation channel is an important elementin cold transduction (Fig. 1a). Similar to

Cold emerging from the fogAmy B. MacDermott and C. Justin Lee

Proposed mechanisms for the sensation of cold have focusedon single proteins. A paper in this issue reports that cold trans-duction depends on a complex interplay among ion channels.

Amy MacDermott is in the Dept. of Physiology& Cellular Biology and the Center forNeurobiology and Behavior, 630 West 168thStreet, Rm. 1109BB, New York, New York10032, USA. Justin Lee is in the Departmentt ofPharmacology, Emory University, 1510 CliftonRoad, Atlanta, Georgia 30322, USA.e-mail: abm1@columbia.edu

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

interesting to test its response pattern toa noxious heat stimulus.

Although Viana et al.3 and Reid and oth-ers9,10 come to different conclusions aboutthe nature of Icold, it may be that they are infact looking at different components of thesame complex response. The novelty of theViana study is that cold responsiveness is acomplex interplay among multiple channels,and that these basic elements are present inmany sensory neurons, even those that arenot normally sensitive to cold.

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

2. Caterina, M. J., Schumacher, M. A., Tominaga,M., Rosen, T. A., Levine, J. D. & Julius, D.Nature 389, 816–824 (1997).

3. Viana, F., de la Pena, E. & Belmonte, C. NatureNeurosci. 5, 254–260 (2002).

4. Kenshalo, D. R. & Duclaux, R. J. Neurophysiol. 40,319–332 (1977).

5. Simone, D. A. & Kajander, K. C. J. Neurophysiol.77, 2049–2060 (1997).

6. Clapham, D. E. Neuron 21, 5–7 (1998).

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

8. Askwith, C. C., Benson, C. J., Welsh, M. J. &Snyder, P. M. Proc. Natl Acad. Sci. USA 98,6459–6463 (2001).

9. Reid, G. & Flonta, M. L. Nature 413, 480 (2001).

10. McKemy, D. D., Neuhausser, W. M. & Julius, D.Nature (in press).

11. Ludwig, A., Zong, X., Jeglitsch, M., Hofmann, F.& Biel, M. Nature 393, 587–591 (1998).

12. Okazawa, M., Terauchi, T., Shiraki, T.,Matsumura, K. & Kobayashi, S. Neuroreport 11,2151–2155 (2000).

Historically there has been a long searchfor specializations of the human brain thatmight account for our intellectual pre-eminence. It has often been claimed that

our frontal lobes, and in particular the pre-frontal cortex, are especially enlarged rela-tive to other animals. The evidence that ismost frequently cited comes from the clas-sic work of Brodmann1, who measured thesize of the prefrontal cortex and neocor-tex in man and non-human primates. Onecan use these values to plot the size of thehuman prefrontal cortex against that ofthe entire neocortex, and then perform aregression analysis to ask what value forprefrontal cortex one would expect for aneocortex as large as it is in the humanbrain. From Brodmann’s data, the pre-frontal cortex is roughly two times as largeas expected. The possible functional rele-

190 nature neuroscience • volume 5 no 3 • march 2002

was sensitive to menthol. Thus it is likelythat the recently cloned menthol receptor,CMR1, is a critical molecular mediator ofcold and menthol transduction. Otherstudies, however, including the one byViana et al.3, indicate that unlike noxiousheat sensation, in which there is a clearthreshold for cellular excitation (around43°C), the threshold for responses toinnocuous cold (15–28°C) or noxious cold(<15°C) are broadly distributed. This indi-cates a more complex cellular response thanis most probably mediated by CMR1 alone.

The effect of menthol is analogous tothe action of capsaicin, the spicy compo-nent of hot chili peppers and an agonistfor the noxious heat receptor, VR1. Inter-estingly, in both the Viana et al.3 andMcKemy et al.10 studies, about half of thecold- and menthol-sensitive neurons werealso sensitive to capsaicin. This suggeststhat some cold-sensitive neurons are alsoresponsive to noxious heat, consistentwith studies of the properties of coldfibers4. This observation raises an inter-esting question of how cold signals aredistinguished from those of noxious heatin the same sensory fiber that is respon-sive to both signals. The answer might liein the differential firing pattern encodedby each stimulus. For example, Viana etal. reported that some cold-sensitive neu-rons display a rhythmic bursting firingpattern in response to cold stimuli. If thistype of cold-sensitive neuron is alsoresponsive to noxious heat, it would be

stronger cold response. Another way thatcold sensitivity of sensory neurons may beregulated is by changing relative channeldensity through transcriptional regulationof channel expression.

Menthol, a cyclic terpene alcohol foundin mints, induces the sensation of coolingin the mouth and on the skin. Thus it isintriguing that it turns out to be a usefultool for studying cold sensitivity in indi-vidual sensory neurons. Menthol has beenshown to enhance cold-induced influx ofcalcium9 and to activate menthol receptorsthat are highly calcium permeable10,12. Allthe cold-sensitive neurons in the Viana etal. study3 either were activated directly bymenthol or responded to less cool temper-atures in the presence of menthol. In con-trast, none of the cold-insensitive neurons

news and views

x

a

b

c

Cold sensitive

Cold insensitive

Cold insensitive

mV

mV

mV

33oC

20oC

20oC

20oC

33oC

33oC

+4-AP

Na+

Na+HCN (Ih)

Ca2+

VGCCVGSC

LeakTREK-1(?)

Ikd

K+

CMRI

Na+

HCN (Ih)

Ca2+

VGCCVGSC

LeakTREK-1(?)

Ikd

K+

Na+HCN (Ih)

VGCCVGSC

LeakTREK-1(?)

Ikd

K+

x

x

x

x xNa+

x

The frontal cortex: does sizematter?Richard E. Passingham

The human frontal cortex has been reported to be proportion-ally larger than in other primates. Magnetic resonance scans ofhumans, apes and monkeys now cast doubt on this idea.

The author is in the Department ofExperimental Psychology, Oxford University,South Parks Road, Oxford OX1 3UD, UK. e-mail: dick.passingham@psy.ox.ac.uk

Fig. 1. Multiple ion channels contribute to cold-induced firing of action potentials in cold-sensitive neurons andpharmacologically modified cold-insensitive neurons. (a) A representation of how a combination of channels mightcontribute to the cold response. Closing of leak channels (possibly TREK-1) and opening of CMR1 channels fol-lowed by closing of HCN channels causes cold-sensitive neurons to fire a transient burst of action potentials dur-ing a cold stimulus. (b) Activation of IKD exerts a brake on any depolarizing influences, inhibiting the firing of acold-insensitive neuron. (c) In the presence of 4-AP, a blocker of IKD, a cold-insensitive neuron fires action poten-tials in response to a cold stimulus. VGSC, voltage-gated sodium channels; VGCC, voltage-gated calcium channels.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

humans1,6. Two possible reasons exist. Thefirst is the difference in sample size. Theauthors showed that variation in valueswithin a species can determine whether ornot one obtains differences betweenhuman and ape brains. Thus, any differ-ence might depend on which specimensare chosen for comparison. A second pos-sibility is that there are differences in theprefrontal cortex, but not in the frontallobes as a whole. In support of this, theauthors referred to their own cytoarchitec-tural analysis of the prefrontal cortex inhuman and non-human primate brains12,in which they measured the size of themost anterior portion of the prefrontal cor-tex (also called the ‘frontal pole’ or cytoar-chitectural area 10) and found it to beproportionately larger in humans than inapes or monkeys. By their own argument,however, this comparison cannot be reliedupon until measurements are taken fromseveral specimens for each species. Fur-thermore, there is considerable variationbetween human brains in the cytoarchi-tectural borders within the prefrontal cor-tex, and regression analysis showed that thesize of the frontal pole in the human brainwas no larger than expected for a primatewith as large a brain12.

Given their conclusion that thehuman and ape frontal lobe do not dif-fer in gross anatomy, the authors consid-er the possibility that they might differin organization or in local circuitry. Theycite a previous finding13 that there arelarge, spindle shaped cells in the medialfrontal cortex (anterior cingulate cortex)of man and the great apes, but not in anyother primate species. However, onemight question the assumption that theremust be microstructural features thatuniquely characterize the human brain.The fact that an area of the brain is nobigger than expected for a primate withas big a brain does not mean that theabsolute difference in the size of the areais not crucial for function. It means onlythat the human brain evolved accordingto the rules for primates. It was arguedlong ago that what may be critical forintelligence is the absolute amount of tis-sue (or number of neurons) above thataccounted for by the general relationbetween brain size and body size14. Larg-er bodies require larger brains, but forany particular body size the different

vance of this finding is that imaging stud-ies have shown the prefrontal cortex to beactivated when subjects plan2 or solve thesorts of problems that make demands ongeneral intelligence3. Furthermore, thereis a significant correlation between thevolume of frontal grey matter and intelli-gence as measured on such tests4.

In this issue, Semendeferi and col-leagues5 cast doubt on the idea thatour frontal lobes are disproportion-ately large. Previous studies usuallytook measurements from only a singleindividual for each species, and this ledto disagreement: some argued for alarger-than-expected prefrontal cortexin humans6, whereas others arguedagainst a difference7. Studies also mea-sured the proportion of the prefrontalcortex that lies within the frontal sulci(‘gyrification’)8,9 and claimed that thisvalue for the human brain was signifi-cantly greater than predicted from datafor monkeys and apes. However,although gyrification is correlated withvolume, it is not a direct measure offrontal cortical volume.

Semendeferi and colleagues used mag-netic resonance imaging (MRI) to direct-ly measure the volume of frontal cortexand neocortex in 15 great apes (chim-panzees, bonobos, gorillas, orangutans), 4lesser apes (gibbons) and 5 monkeys(rhesus, cebus). For comparison, theyscanned 10 normal human subjects. Theauthors compared the proportion of theneocortex that is made up of the frontallobes, that is, the tissue from the anteri-or bank of the central sulcus to the frontalpole. The values for the human brain(36.4–39.3%) overlapped with those forthe great apes (35–38.7%), but the frontallobes formed a significantly smaller proportion of the neocortex in the lesserapes and monkeys (27.5–32.3%). Theauthors also performed a regressionanalysis in which they plotted the volumeof the frontal lobes against the volume ofthe neocortex. The values for the frontallobes in the human brain lay within therange predicted from the data for non-human primates (Fig. 1).

To examine the frontal lobes in greaterdetail, the authors divided them into threesectors. The first consisted of the cortex inthe anterior bank of the central sulcus. Thissector includes much, but not all, of themotor cortex (area 4) as assessed by cytoar-chitecture. The second consisted of the pre-central gyrus, which is the cortex anterior tothe central sulcus and extending rostrally tothe precentral sulcus. This would includemuch but not all of the premotor cortex(area 6) as assessed by cytoarchitecture. Thefinal sector was the cortex anterior to theprecentral sulcus. This would include all ofthe prefrontal cortex, but also some of theanterior part of the premotor cortex. Thevalues for the human precentral gyrus(5.4–7.9%) fell largely within the range forthe great apes and gibbons (5.5–10.4%). Thevalues for the cortex anterior to the precen-tral gyrus ranged from 28.8% to 33% in thehuman brains. Again these values over-lapped with the values for the great apes(25.5–29.7%), though lying outside therange for the lesser apes (22–23.8%).

These findings support the argumentthat humans do not show a dispropor-tionately larger frontal cortex. However,the authors are cautious in their conclu-sions and acknowledge that the divisionsused in the study do not exactly corre-spond to the cytoarchitectural divisionsbetween motor, premotor and prefrontalcortex. They argue that it would be a for-bidding task to identify, from multiplespecimens and species, the relevantcytoarchitectural borders. At the momentit is not possible to distinguish these bor-ders on MRI scans, though further devel-opments in MRI scanning at highmagnetic field strengths may make itpossible in the future. One possibility, solong as several postmortem specimensare available for each species, would beto use an ‘observer-independent method’of identifying cytoarchitectural bound-aries by statistical analysis of the densityof neurons over the different corticallaminae10. Until such studies are per-formed, however, we can draw no certainconclusions concerning the proportion-ate evolutionary development of the pre-frontal cortex in the human brain.

There remains an apparent conflictbetween the results reported in this paperand previous studies that suggested a larg-er proportionate frontal cortex in

nature neuroscience • volume 5 no 3 • march 2002 191

news and views

Human

Chimpanzee

Gorilla

Orangutan

GibbonMacaque

Fig. 1. A comparison of the frontal lobes (colored) in human and several non-human primatespecies. The evolutionary relationships among the species are indicated by the connecting lines.Semendeferi and colleagues5 found that human frontal lobes are not disproportionately largerthan predicted for a primate brain of its size. (Figure courtesy of K. Semendeferi and H. Damasio).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

The author is in the Dept. of Neurology,University of Iowa College of Medicine, 200 Hawkins Street, Iowa City, Iowa 52242, USA e-mail: ralph-adolphs@uiowa.edu

Our observations of others provide a con-tinual source of cues that we use to guideour social behavior: some people we avoid,others we approach, some we find inter-esting, some boring, some we trust, andsome we do not (Fig. 1a)1. What happensin the brain when we make these judg-ments? Diverse findings from cognitiveneuroscience have begun to provide someanswers. When we look at someone’s face,the brain first needs to construct a per-ceptual representation that provides infor-mation about the features of the face andtheir configuration, a process that drawson extrastriate visual cortices in thefusiform and superior temporal gyri2. Per-ceptual processing of facial features canthen be linked to the generation of judg-ments about the person. Previous studieshave suggested that social judgmentsinvolve the amygdala, regions of prefrontalcortex and regions of somatosensory-relat-ed cortices3. But what are the precise con-tributions of these different brain regions?And what features of the stimuli providethe relevant information?

In this issue, Winston and colleagues4

examined these two questions by usingthe power of event-related fMRI to studya particular class of social judgments:those of trustworthiness. Subjects wereshown pictures of faces and asked to per-form two tasks while in the scanner. One

task investigated the neural regions thatmight subserve overt processing by hav-ing subjects indicate whether the facelooked trustworthy or not. The other taskexamined automatic, implicit processingof trustworthiness by having subjects per-form an unrelated cognitive evaluation(indicate whether the face looked like ahigh school student or a college student).

To analyze the brain regions activatedby seeing trustworthy or untrustworthyfaces, the authors used an event-relateddesign that allows the responses to be ana-lyzed on a trial-by-trial basis. A subject firstsaw the faces while in the scanner andpushed a button to indicate how each faceshould be categorized (trustworthy/untrustworthy or high school/college,depending on the task). After the scanningsession, the subject saw the faces again andrated trustworthiness on a numerical scale.Because these judgments were presumed toclosely correlate with those made (explicit-ly or implicitly) in the scanner, only thepost-scan ratings were used for analysis.The event-related design permitted thefMRI signal recorded for each face in thescanning session to be correlated with thesubsequent rating given to that face (whichwas somewhat idiosyncratic for each sub-ject). This analysis revealed that the amyg-dala was significantly more activated whensubjects viewed faces that they later ratedas most untrustworthy than when theyviewed those rated as trustworthy, irre-spective of whether an explicit judgmentwas required during the scan. A similar pat-tern of activation was observed for the right

insula and the fusiform gyrus. The report-ed amygdala activation fits neatly with alesion study that found impaired judgmentof untrustworthiness from faces followingbilateral amygdala damage5, and it is alsoin line with prior findings demonstratingamygdala activation in response to facestimuli (expressions of fear) regardless ofthe task6, even when shown subliminally7.This suggests that the amygdala participatesin relatively automatic, obligatory process-ing of certain categories of stimuli.

To analyze the brain regions that wouldbe activated by explicit rather than implicittrustworthiness judgments, the authors con-trasted the fMRI signal between the twotasks. Right superior temporal sulcus wasactivated more when subjects performed thetrustworthy/untrustworthy categorizationin the scanner than when they performedthe high school/college categorization, irre-spective of the subject’s later ratings of trust-worthiness. This activation fits well withprior demonstrations that attention to eyecontact and other social information in aface stimulus leads to preferential activationof this region8,9. A more complicated pat-tern was observed in orbitofrontal cortex:its activation depended both on the ratedtrustworthiness of the face and on the taskperformed in the scanner.

The findings provide support for amodel of social cognition in which regionsof extrastriate visual cortex process per-ceptual information about socially rele-vant visual stimuli, and the amygdala andorbitofrontal cortex then orchestrate emo-tional reactions to such stimuli. Theobserved patterns of activation suggestthat the amygdala contributes to rapidand automatic emotional responses,whereas the orbitofrontal cortex con-tributes to emotional responses only inthe context of a particular conscious eval-uation. Emotional responses couldinclude both cognitive and somaticchanges (for example, changes in atten-tion or in autonomic response, respec-

192 nature neuroscience • volume 5 no 3 • march 2002

mammalian groups differ in brain size,with the primates having especiallyenlarged brains. The human neocortex isover three times as large as expected fora primate matched for body size15. Eventhough the present study shows that thehuman frontal lobes do not differ as aproportion of the neocortex, they are overthree times larger than would be expect-ed for a hypothetical great ape of thesame body weight. Such a difference mustbe of immense consequence for ourcapacity to plan and reason2,3.

8. Zilles, K. et al. Anat. Embryol. 179, 173–179(1988).

9. Rilling, J. K. & Insel, T. R. J. Human Evol. 37,191–223 (1999).

10. Schleicher, A. et al. Neuroimage 9, 165–177(1999).

12. Semendeferi, K. et al. Am. J. PhysicalAnthropol. 114, 224–241 (2001).

13. Nimchinsky, E. A. et al. Proc. Natl. Acad. Sci.USA 96, 5268–5273 (1999).

14. Jerison, H. J. Evolution of the Brain andIntelligence (Academic, New York, 1973).

15. Passingham, R. E. The Human Primate(Freeman, Oxford, 1982).

1. Brodmann, K. Anat. Anz. Suppl. 41, 157–216(1912).

2. Dagher, A., Owen, A. M., Boecker, H. &Brooks, D. J. Brain 122, 1973–1987 (1999).

3. Duncan, J. et al. Science 289, 457–460 (2000).

4. Thompson, P. M. et al. Nature Neurosci. 4,1253–1258 (2001).

5. Semendeferi, K., Lu, A., Schenker, N. &Damasio, H. Nature Neurosci. 5, 272–276(2002).

6. Deacon, T. The Symbolic Species (Allen Lane,London, 1997).

7. Uylings, H. B. M. & van Eden, C. G. Prog.Brain Res. 85, 31–62 (1990).

news and views

Trust in the brainRalph Adolphs

A new event-related fMRI study suggests that decisions abouttrustworthiness involve structures that process emotions, andraises intriguing questions about cues used for such judgments.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 1. A neurological basis for judging trustworthiness. (a) What facial features aid in our deci-sions about trustworthiness in others? Winston and colleagues4 showed images of unknown peo-ple (not the President or former Vice-President) to subjects during fMRI scanning to determinewhat brain regions are activated while people make decisions about trustworthiness. (b) A modelof how the brain processes socially relevant stimuli. Some of the regions activated in the study byWinston and colleagues are indicated, along with arrows showing the major routes of informationprocessing. The fusiform gyrus (FG, green) and superior temporal sulcus (STS, red) process fea-tures of the face stimulus. The amygdala (AM, blue) associates perception of the face with an emo-tional response to the face. The insula (INS, purple) participates in representing this emotionalresponse as a feeling about the person whose face we view. Activation in STS can also be modu-lated by the task, demonstrating top-down influences and suggesting that most information flowsin both directions along this circuit. (Photos of Bush and Gore provided by the Associated Press.)

nature neuroscience • volume 5 no 3 • march 2002 193

tively), and these changes in turn could beperceived by viewers as a feeling that theyhave about the person whose face theysee10. The reported activation in the insu-la (a visceral somatosensory cortex) mightreflect such a perception of one’s ownemotional response to the stimuli (Fig. 1b). It seems plausible that viewingpeople who look untrustworthy wouldproduce emotional responses and changesin feeling in the perceiver, and that suchfeelings might be used, in part, to makethe judgments. This could be tested in thefuture by also measuring (or indepen-dently manipulating) the viewer’s emo-tional state during the experiment.

The present study raises an intriguingquestion: what makes different faces lookdifferentially trustworthy to a viewer? It isoften assumed that complex social judg-ments must be reducible to a combinationof simpler judgments. Perhaps we judgetrustworthiness by the relative amounts ofnegative facial expression, averted gaze, andso on. The authors addressed this issue byeliminating as many of these ingredients aspossible: the faces were all unfamiliar, Cau-casian, male, front-facing, and with directgaze. Although this assured that the brainactivations observed could not be attrib-uted simply to these other factors, it alsoeliminated many of the cues on which wenormally base judgments of trustworthi-ness. The subjects were thus making trust-worthiness judgments on the basis of cuesthat are very impoverished compared tothe wealth of information typically avail-able in real life: for instance, we often areacquainted with the person whom wejudge, and much of our judgment is based

er stimulus category that is correlated withjudgments of trustworthiness. As we sawabove, the authors of the present study select-ed their stimuli so that they were judged toshow large differences in trustworthiness, butonly slight differences in emotional expres-sion and no differences in ethnicity. Howev-er, other studies demonstrate that theamygdala is also activated by viewing facesthat show certain emotional expressions,notably fear11, and that it is activated whensubjects view faces of people from anotherrace12,13. One might therefore ask what it isthat the stimuli in these different studies sharein common, in virtue of which they are ableto drive amygdala activation. If one under-took such an investigation, one might findthat the amygdala is specialized not for pro-cessing untrustworthiness, fearfulness, or racemembership, but for a different category thatis correlated with all of these. Neurosciencemay be showing us patterns in the brain thatcorrespond to stimulus categories for whichwe as yet have no precise definition.

1. Macrae, C. N. & Bodenhausen, G. V. Annu.Rev. Psychol. 51, 93–120 (2000).

2. Haxby, J. V., Hoffman, E. A. & Gobbini, M. I.Trends Cogn. Sci. 4, 223–233 (2000).

3. Adolphs, R. Behav. Cogn. Neurosci. Rev. 1,21–61 (2002).

4. Winston, J. S., Strange, B. A., O’Doherty, J. &Dolan, R. J. Nature Neurosci. 5, 277–283 (2002).

5. Adolphs, R., Tranel, D. & Damasio, A. R.Nature 393, 470–474 (1998).

6. Vuilleumier, P. et al. Neuron 30, 829–841(2001).

7. Whalen, P. J. et al. J. Neurosci. 18, 411–418(1998).

8. Hoffman, E. A. & Haxby, J. V. Nature Neurosci.3, 80–84 (2000).

9. Narumoto, J. et al. Cogn. Brain Res. 12,225–231 (2001).

10. Damasio, A. R. The Feeling of What Happens:Body and Emotion in the Making ofConsciousness (Harcourt Brace, New York,1999).

11. Morris, J. S. et al. Nature 383, 812–815 (1996).

12. Phelps, E. A. et al. J. Cogn. Neurosci. 12,729–738 (2000).

13. Hart, A. J. et al. Neuroreport 11, 2351–2355(2000).

not so much on their appearance as onwhat we know about them.

Despite the careful selection of facesused in the study by Winston and col-leagues, it was impossible to eliminate allvariance in expression, and the faces dif-fered slightly in terms of the emotion theyshowed. The authors therefore asked sub-jects to rate the faces in terms of their emo-tional expressions, to determine whetherthese ratings might account for any of theother findings. Indeed, expressions of angeror sadness were negatively correlated withtrustworthiness ratings, and happiness waspositively correlated. Furthermore, therated emotional expression of the face con-tributed to some of the brain activationsobserved in the experiment. However,when this effect was taken into account,there still remained a significant activationin the right amygdala that could be attrib-uted to the trustworthiness ratings of thestimuli independent of their emotion rat-ings. Judgments of trustworthiness andjudgments of certain emotions thus seemto rely on a partly overlapping set of fea-tures, but the overlap is not complete.

A second, complementary question iswhy different viewers assign differentialtrustworthiness to a given face. It is notonly the features of the stimuli that are dri-ving the judgments made in the task(although they of course provide one trig-ger), but also the personalities and autobi-ographies of the viewers. To the extent thatviewers agree on certain ratings, one canreveal stereotypes in the judgments wemake. But it would be most interesting toextend such a study, not by varying thestimuli in terms of how subjects rate them,but by varying the subjects’ individualpropensity to make the attributions—thatis, by measuring personality traits in theparticipants. Do people who are prone tojudge others as untrustworthy have a hyper-active amygdala, for instance?

It remains an open question whether acti-vations in structures such as the amygdala arespecific to processing ‘trustworthiness’, orwhether they result from processing anoth-

news and views

FG

AM

STS INS

Visual stimuli

Emotional response

a

b

Amy Center

AP/Wide World Photos

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Low glucose–sensing cellsin the carotid bodyRicardo Pardal and José López-Barneo

Laboratorio de Investigaciones Biomédicas, Departamento de Fisiología andHospital Universitario Virgen del Rocío, Universidad de Sevilla, E-41013,Seville, Spain

Correspondence should be addressed to J.L.-B. (lbarneo@hvr.sas.junta-andalucia.es)

Published online: 19 February 2002, DOI: 10.1038/nn812

Decreased plasma glucose concentration elicits a complex neu-roendocrine response that prevents or rapidly corrects hypoglycemiaas required to preserve brain function1–4; however, where and howlow glucose is sensed is unknown4–6. Here we show that low glu-cose increases secretion from glomus cells in the carotid bodies,sensory organs whose stimulation by hypoxia produces sympatheticactivation, by a process that depends on extracellular Ca2+ influxand is paralleled by inhibition of voltage-gated K+ channels. Wepropose a new glucose-sensing role for the carotid body glomuscell that serves to integrate information about blood glucose andO2 levels and to activate counterregulatory responses.

The carotid body is composed of innervated clusters of glomuscells that, when activated by reduced O2 tension, release dopamineand other transmitters to stimulate afferent sensory fibers7–9. Usinga new thin-slice preparation that retains the structure of the carotidbody and preserves the characteristic response of glomus cells tohypoxia10, we examined the effect of decreased extracellular glu-cose. Exposure of glomus cells in these slices to a glucose-free (0 mMglucose) solution induced a secretory activity as monitored byamperometry9,10 (Fig. 1a). The average rate of secretion during thelast minute of exposure to low glucose, calculated from cumulativesecretion traces (Fig. 1a), was over 20 times that in the control con-dition of 5 mM glucose (Fig. 1b). The size distribution and meanarea of quantal events triggered by low glucose (Fig. 1c) were sim-ilar to those previously observed in glomus cells activated by hypox-ia10, suggesting that the two conditions (glucopenia and hypoxia)induced the release of a common vesicle pool. The effect of low glu-cose on glomus cells was concentration dependent. At normal air O2tension (PO2 = 150 mmHg), control cells showed almost no secre-tory activity. When glucose was lowered to 2 mM, catecholaminerelease increased in proportion to the glucopenia (Fig. 1d and e).

At air PO2, glomus cells are inhibited as a result of suppres-sion of the O2-sensitive activation pathway9. The typical responseto hypoxia (PO2 ≈ 25 mmHg) of glomus cells in the slices10 wasmarkedly enhanced in 0 mM glucose (Fig. 2a), and the meansecretory rate of cells exposed to hypoxia and glucose-free solu-tions was two- to threefold greater than to either stimulus sepa-rately (Fig. 2b). We therefore studied the sensitivity to glucoseof glomus cells maintained at 12% O2 (≈90 mmHg), an O2 ten-sion comparable to that in the arterial blood irrigating the ratcarotid body in vivo. At this PO2, 5 mM glucose induced cate-cholamine release, and this activity was reversibly reduced onexposure to higher (10 mM) glucose concentration (Fig. 2c).The dependence of glomus cell secretion on glucose concentra-tion at PO2 of 90 mmHg was displaced toward glucose levelshigher than those that triggered secretion at normal air O2 ten-sion (Fig. 2d and e). Therefore, at the PO2 level of arterial blood,glomus cell secretory activity is modulated by glucose in a phys-iological concentration range.

Catecholamine secretion elicited by glucose-free solutions wasabolished by blockade of voltage-gated Ca2+ channels with 0.2 mMcadmium7–10 (n = 3; Fig. 3a), suggesting that this response dependson depolarization-evoked Ca2+ influx. Modulation of glomus cellsecretion by low glucose was maintained in slices treated withglibenclamide (5 µM), a blocker of ATP-regulated K+ (KATP) chan-nels. In addition, input resistance measured in patch-clamped cellsheld at –80 mV (286 ± 11 MΩ, mean ± s.d., n = 9), although rel-atively low (perhaps due to the high temperature or to electricalcoupling), was not altered by removal of glucose (267 ± 11 MΩ,n = 9). These data suggested that leakage and KATP channels werenot appreciably influenced by the changes in glucose concentra-tion. In contrast, glucose deficiency produced a reversible reductionof outward K+ current amplitude in the entire range of membranepotentials (Fig. 3b and c). At +20 mV, exposure to 0 mM glucosedecreased peak outward current amplitude by 38 ± 12% (mean ±s.d., n = 7). These effects of low glucose were independent ofchanges in intracellular ATP concentration, because in mostexperiments cells were dialyzed with an internal solution con-taining 4 mM MgATP. In addition, low glucose–induced glomuscell secretion was not prevented by 10 mM sodium pyruvate (n =3; data not shown), a condition that supports the intracellular pro-duction of ATP. Low glucose regulated voltage-dependent K+ chan-nels selectively, as it had no effect on the small inward currentcharacteristic of rat glomus cells (Fig. 3d). In addition, inhibitionof the K+ current by low glucose was maintained after blockade ofthe Ca2+-activated maxi-K+ channels with iberiotoxin (Fig. 3e).

Our findings strongly suggest that carotid body glomus cellsare physiological glucose detectors that transduce glucose levelsinto variable rates of transmitter release, which stimulate afferentsensory fibers to evoke sympathoadrenal activation. Blockers of

brief communications

nature neuroscience • volume 5 no 3 • march 2002 197

Fig. 1. Secretory response of rat glomus cells to low glucose. (a) Top,amperometric signal showing increased secretory activity in a glomus cellexposed to 0 mM extracellular glucose. Bottom, cumulative secretion sig-nal (in femtocoulombs, fC) resulting from the time integral of the ampero-metric recording. The slice was continuously perfused with a standardsolution containing 117 mM NaCl, 4.5 mM KCl, 23 mM NaHCO3, 1 mMMgCl2, 2.5 mM CaCl2, 5 mM glucose, 5 mM sucrose. The low-glucose solu-tions were obtained by replacing glucose with equimolar amounts ofsucrose. (b) Average secretion rate in cells exposed to 5 mM (88 ± 45 fC/min, mean ± s.d., n = 14) and 0 mM (1,870 ± 386 fC/min, n = 14 cells) glucose (p < 0.001, Student’s t-test). (c) Distribution of thearea of exocytotic events in low glucose. A Gaussian fit to the data issuperimposed. (d) Secretory response of a single glomus cell to variouslevels of low glucose. (e) Secretory activity as a function of extracellularglucose concentration. PO2 in the experiments was 150 mmHg and tem-perature was 35–37ºC. Each data point is the average of 3–5 measure-ments in 3 cells (mean ± s.d.). The curve in Fig. 1e was drawn by eye. Use ofanimals was approved by the institutional Laboratory Animal Committee.

a

b c

d

e

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

198 nature neuroscience • volume 5 no 3 • march 2002

brief communications

voltage-dependent K+ channels evoke depolarization8,11 and cate-cholamine secretion10 in these cells; it therefore is likely that, like thehypoxia transduction cascade8–10, low-glucose signaling is initiat-ed by inhibition of K+ channel activity, which facilitates membranedepolarization, Ca2+ influx through voltage-gated Ca2+ channelsand transmitter release. These observations help to explain reportsthat anesthetized animals show a rapid increase in the output ofhepatic glucose after infusion of the carotid body region withNaCN5,12, alterations of carbohydrate metabolism in acute hypox-ia13, or impairment of insulin-induced counterregulatory responseto mild hypoglycemia in dogs after resection of the carotid bodyand surrounding tissues6. Although peripheral glucose controlmight also be accomplished at the liver or portal vein4,14,15, thestrategically located carotid bodies are of special importance forbrain homeostasis, as neurons are particularly vulnerable to thesimultaneous lack of glucose and oxygen1,2. The function of glomuscells as combined O2 and glucose sensors is surely advantageousin facilitating activation of the counterregulatory measures inresponse to small reductions of either variable.

AcknowledgementsWe thank A. Alvarez-Buylla, P. Ortega-Sáenz, G. Gasic and A. Konnerth.

Research was supported by grants from the Spanish Ministry of Science and

Technology (1FD97-1614) and Fundaciones La Caixa and Ramón Areces. J.L.-B.

received the “Ayuda a la investigación 2000” of the Juan March Foundation.

Competing interests statementThe authors declare that they have no competing financial interests.

RECEIVED 4 DECEMBER; ACCEPTED 28 DECEMBER 2001

1. Auer, R. Stroke 17, 699–708 (1986).2. Martin, R. L., Lloyd, H. G. & Cowan, A. I. Trends Neurosci. 17, 251–257 (1994).3. Gerich, J. E. & Campbell, P. J. Diabetes Metab. Rev. 4, 93–111 (1988).4. Cane, P., Artal, R. & Bergman, R. N. Diabetes 35, 268–277 (1986).5. Alvarez-Buylla, R. & de Alvarez-Buylla, E. R. Resp. Physiol. 72, 347–360 (1988).6. Koyama Y. et al., Diabetes 49, 1434–1442 (2000).7. López-Barneo, J., López-López, J. R., Ureña, J. & González, C. Science 241,

580–582 (1988).8. Wyatt, C. N. & Peers, C. J. Physiol. 483, 559–565 (1995).9. Ureña, J., Fernández-Chacón, R., Benot, A. R., Alvarez de Toledo, G. &

López-Barneo, J. Proc. Natl. Acad. Sci. USA 91, 10208–10211 (1994).10. Pardal, R., Ludewig, U., García-Hirschfeld, J. & López-Barneo, J. Proc. Natl.

Acad. Sci. USA 97, 2361–2366 (2000).11. Pérez-García, T. et al. J. Neurosci. 20, 5689–5695 (2000).12. Alvarez-Buylla, R. & de Alvarez-Buylla, E. R. Brain Res. 654, 167–170 (1994).13. Zinker, B. A., Nandaran, K., Wilson, R., Lacy, D. B. & Wasserman, D. H. Am. J.

Physiol. 266, E921–E929 (1994).14. Donovan, C. M., Hamilton-Wessler, M., Halter, J. B. & Bergman, R. N. Proc.

Natl. Acad. Sci. USA 91, 2863–2867 (1994).15. Havener, A. L., Bergman, R. N. & Donovan, C. M. Diabetes 46, 1521–1525

(1997).

Fig. 2. Increase in sensitivity to glucose in low PO2. (a) Augmentation ofthe secretory response to 0 mM glucose during hypoxia (PO2 ≈ 25 mmHg).Secretion rates during 3 responses to hypoxia are 1,837, 5,816 and1,729 fC/min. Resetting of the integrator used to calculate the cumula-tive secretion signal is indicated by dotted lines. (b) Average secretionrates (in fC/min) in cells exposed to a PO2 of ≈25 mmHg (1,780 ± 439, n = 8), 0 glucose (1,870 ± 386, n = 14) and both simultaneously (4,400 ± 1,300, n = 8; p < 0.005). (c) Modulation of secretory activity byglucose in the 5–10-mM range in a glomus cell exposed to 12% O2(PO2 = 90 mmHg). Secretory rates are 1,670 (0 mM glucose, 20% O2),505 (5 mM glucose, 12% O2), 8 (10 mM glucose, 12% O2) and 595 (recov-ery in 5 mM glucose, 12% O2). (d) Secretory activity versus extracellularglucose concentration. PO2 was 90 mmHg. Each data point is the mean ±s.d. of 3–5 measurements in 3 cells. The curve of Fig. 1e is shown for com-parison (dotted line). (e) Logarithm of maximal secretion rate (ordinate)versus glucose concentration at 2 different PO2 values. Values of secre-tion are from Figs. 1e () and 2d (). The control solution was bubbledwith 5% CO2, 20% O2 and 75% N2 (PO2 ≈ 150 mmHg). The 2 low PO2 levels used were obtained by continuously bubbling the solution in one reservoirwith either 5% CO2 and 95% N2 (hypoxia, PO2 in the chamber ≈25 mmHg) or 5% CO2, 12 % O2 and 83% N2 (PO2 in the chamber ≈90 mmHg).Equilibration of solutions in the chamber, as determined with an O2 electrode, required <30 s. The pH of all solutions was 7.4, osmolality ≈300 mOsm/kgand temperature 35–37°C. Curves/lines in 2d, e were fitted by eye.

Fig. 3. Low glucose-induced extracellular Ca2+ influx and inhibition of voltage-gatedoutward K+ currents. (a) Reversible suppression of low glucose–evoked secretoryactivity by 0.2 mM Cd2+. (b) Outward K+ currents from a patch-clamped glomus celldepolarized to 0 and +20 mV and exposed to 0 mM glucose. The control (c) and recov-ery (r) external solutions contained 5 mM glucose. (c) Peak outward current ampli-tude–voltage relation in the same cell bathed in control solutions before () and after() exposure to 0 mM glucose (). Holding potential, –80 mV. Pipette solution, 125 mMKCl, 4 mM MgCl2, 4 mM MgATP, 10 mM HEPES, 0.1 mM EGTA, pH = 7.2 and osmolality285–290 mOsm/kg. PO2, 150 mmHg; 35–37ºC. (d) Inward and outward currentsrecorded from a patch-clamped glomus cell depolarized to –10 mV and exposed to0 mM glucose. The control external solution contained 5 mM glucose. (e) Outward K+

currents recorded from a patch-clamped glomus cell depolarized to +20 mV andexposed to 0 mM and 5 mM (control) glucose. The external solutions contained 200 nMiberiotoxin (IBTX), a blocker of the Ca2+-activated maxi-K+ channels.

a

b

c

d e

a

b c

d e

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Gender-specific inductionof enhanced sensitivity toodorsPamela Dalton, Nadine Doolittle and Paul A.S. Breslin

Monell Chemical Senses Center, 3500 Market Street, Philadelphia, Pennsylvania19104, USA

Correspondence should be addressed to P.D. (pdalton@pobox.upenn.edu)

Published online: 4 February 2002, DOI: 10.1038/nn803

Induction of olfactory sensitivity in humans was first illustratedwhen men and women who were initially unable to smell thevolatile steroid androstenone (5α-androst-16-en-3-one) devel-oped that ability after repeated, brief exposures1. Because thisfinding has not been replicated with other compounds inhumans, it has been assumed that olfactory induction is a nar-rowly constrained phenomenon, occurring only in individualswith specific anosmias, perhaps only to androstenone (compareref. 2). Here we show that induction of enhanced olfactory sen-sitivity seems to be a more general phenomenon, with markedchanges in olfactory acuity occurring during repeated test expo-sures to several odorants among people with average baselinesensitivity to these compounds. This increased sensitivity (aver-aging five orders of magnitude) was observed only among femalesof reproductive age. These observations provide convincing evi-dence that female olfactory acuity to a variety of odorants canvastly improve with repeated test exposures. They also suggest asensory basis for the anecdotal observation of greater olfactorysensitivities among females and raise the possibility that the olfac-tory-induction process may be associated with female reproduc-tive behaviors such as pair bonding and kin recognition.

Robust gender differences in olfactory ability are largelyrestricted to aspects of olfactory processing that require higher-level cognition, such as odor identification or odor memory3,which lends credence to the view that human olfactory sensitiv-ity is relatively unaffected by neuroendocrine influences. Mostolfactory sensitivity comparisons between men andwomen, however, have measured a single odorthreshold (compare ref. 4), whereas we made our ini-tial finding serendipitously while studying odor–tasteintegration. We measured olfactory (benzaldehyde,cherry–almond) and taste (saccharin, sweet) thresh-olds in the same individuals on 30 occasions, eachseparated by ∼ 2 days5 (for detailed methods, see Sup-plementary Methods on the supplementary infor-mation page of Nature Neuroscience online). Menand women were equally sensitive at the start of theexperiment; women, but not men, showed markedincreases (3–6 log units) in sensitivity to benzalde-hyde across the test sessions. Sensitivity to the tasteof saccharin did not increase among either men orwomen, suggesting that the phenomenon was sen-sory-modality specific and was not based on thresh-old measurement familiarity (Fig. 1b).

We verified the magnitude and gender specificityof olfactory induction in the same six subjects and insix naive subjects. The tests were bracketed by thresh-old assessments of two control odorants to discernwhether this phenomenon represented practice gen-

eralization or specific exposure-induced olfactory sensitization. Ascontrols, we selected two odorants that varied in their perceptualsimilarity to benzaldehyde: 5-methylfurfural, which has a cherryodor similar to that of benzaldehyde, and isoamyl acetate, whichhas a banana odor that is qualitatively different although categori-cally similar (fruity). Because the previous data indicated that sig-nificant differences in sensitivity between men and women emergedin as few as 6 test sessions (Fig. 1a inset), we tracked changes in sen-sitivity to benzaldehyde across only 10 sessions (20 thresholds).

Sensitivity to benzaldehyde increased substantially (range, 3–11log units) for both naive (mean, 2.6 log units) and experienced(mean, 3.9 log units beyond first sensitivity increase) women, asshown by an F-test (F(1,9) = 9.2; p < 0.0001), but not for men (p > 0.1; Fig. 2). Increases of this magnitude were especially notableamong experienced females, whose benzaldehyde thresholds at thestart of the second study (mean, 8.8 × 10–5 mM) were equivalent tothose measured at the end of the first study (mean, 8.1 × 10–5 mM),despite a 3-month interval between studies. Although femalesshowed a slight cross-facilitation of sensitivity to 5-methylfurfur-al, there was no change in threshold response to amyl acetate. Thus,the effect of repeated threshold testing on women was replicableand odorant specific, providing evidence that these effects werenot due to mere practice with the olfactory threshold task.

We extended the generality of these observations by testing10 new participants (5 males and 5 females) using citralva, a com-pound structurally different from benzaldehyde that elicits alemon–orange smell, and benzaldehyde, the control odorant. Cit-ralva thresholds for women decreased markedly (mean, 6 logunits; F(1,7) = 13.5; p = 0.0001; Fig. 3), whereas thresholds formen were virtually unchanged (p > 0.1). There were no changesin sensitivity to benzaldehyde in either gender. Thus, sensitiza-tion may be induced with very different compounds, and, unlikewhat was presumed in androstenone induction, initial insensi-tivity to the compound was not a prerequisite.

To explore the plausible explanation that sex hormones medi-ate the gender differences in sensitivity (compare refs. 6, 7), wetested females whose average estrogen levels were presumed tobe substantially lower than those of the females in the first stud-ies: 4 girls (ages 8–10) and 4 postmenopausal women (ages49–61) not taking hormone replacement therapy at present. We

brief communications

nature neuroscience • volume 5 no 3 • march 2002 199

Fig. 1. Gender effects of repeated test exposures on odor and taste sensitivity. Meanthresholds (± s.e.m.) are expressed as log millimolar concentration of benzaldehyde orsaccharin in liquid phase. (a) A 2-way analysis of variance (ANOVA), with gender as thebetween-group factor and test session (1–30) as the within-group factor, showed thatbenzaldehyde thresholds declined significantly among females but not males (F(29,116) = 3.84; p = 0.007). Women’s thresholds declined significantly in only 6 tests(see inset; F(1,5) = 4.20; p = 0.02), although men’s did not (F(1,5) = 0.42; p = 0.82). (b) Oral thresholds for saccharin were unchanged for either group.

a b

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

200 nature neuroscience • volume 5 no 3 • march 2002

brief communications

also tested an equal number of age-matched boys and men. Wetested females and males from both groups across sessions 2–8for sensitivity to benzaldehyde and in sessions 1 and 9 to citralva(control). Among the younger subjects, sensitivity to benzalde-hyde did not change with time for either boys or girls as shown byan ANOVA (group × test: F(7,42) = 0.15; p = 1.0), and there wereno gender differences for citralva thresholds between the first andthe last sessions (group × test: F(7,12) = 0.55, p = 0.79). Similar-ly, there were no significant gender differences across sessions inthe older group (group × test: F(7,42) = 2.07, p = 0.07). Notably,postmenopausal women showed no significant reductions inthresholds over time for either odor (F(1,7) = 0.90, p = 0.52).Though preliminary, these results raise the possibility that theamounts of female sex hormones present between onset ofmenarche and menopause have a role in enabling exposure-induced increases in odor sensitivity, and they generate severaltestable hypotheses.

This potential hormonal influence does not conform in anysimple fashion with previous descriptions of the action ofestrogenic compounds on olfactory sensitivity6,8. For exam-ple, in testing carried out two or three times weekly, there wasno evidence of threshold fluctuations among females, indicat-ing no cyclical influence of estrogen as found previously forhumans and other mammals6. If future studies confirm therole of neuroendocrine factors (such as estrogen) in activatingneural systems responsible for sensitivity induction, one mightexpect synthetic hormones in women (e.g., hormone replace-ment therapy in postmenopausal or ovarectomized women)and antitestosterone therapy in men to produce olfactory sen-sitization as well.

Gender-specific changes in olfactory sensitivity may be morelikely to occur with biologically relevant odors. Androstenone,for example, occurs at higher concentrations in men’s axillarysecretions9 and could serve as a marker for the presence of poten-tial mates, as it does among pigs. Whereas specific anosmia toandrostenone occurs in ≤50% of adult humans, females are morelikely to smell androstenone. In addition, among adults who cansmell it, females have lower thresholds than men10, possibly sig-nifying induction of sensitivity in some women who are regu-larly exposed to male axillary volatiles.

The demonstration that women’s olfactory sensitivity increasesfaster and to a much greater degree than that of men provides aconfirmation of the many anecdotal observations in which femalesensitivity to ambient odors seems to exceed male sensitivity. Aftersensitivity had been induced, females who evaluated benzaldehydeat concentrations corresponding to their initial threshold could eas-ily detect it and often provided appropriate quality descriptors(fruity, cherry). As a possible negative consequence, this phenom-enon may account for the greater prevalence of odor-related envi-ronmental complaints from females, such as those that occur withsick-building or chemical intolerance syndromes11, where repeti-tive exposure coupled with directed attention may underlie differ-ences in sensitivity and response. On the positive side, the potentialfor vast improvements in detection sensitivity after repeated expo-sures to olfactory stimuli may also confer specific advantages toreproductive females, such as increased ability for olfactory-basedkin recognition12 and mother–infant bonding13, as well as height-ened ability to identify nutritive food sources and avoid toxicants.

Note: Supplementary Information can be found on the Nature Neuroscience

website (http://neurosci.nature.com/web_extras).

RECEIVED 10 SEPTEMBER; ACCEPTED 27 NOVEMBER 2001

1. Wysocki, C. J., Dorries, K. M. & Beauchamp, G. K. Proc. Natl Acad. Sci. USA86, 7976–7978 (1989).

2. Voznessenskaya, V. V., Parfyonova, V. M. & Wysocki, C. J. Adv. Biosci. 93,399–406 (1994).

3. Cain, W. S. Chem. Senses 7, 129–142 (1982).4. Stevens, J. C., Cain, W. S. & Burke, R. J. Chem. Senses 13, 643–653 (1988).5. Dalton, P., Doolittle, N., Nagata, H. & Breslin, P. A. S. Nature Neurosci. 3,

431–432 (2000).6. Doty, R. L., Huggins, G. R., Snyder, P. J. & Lowry, L. D. J. Comp. Physiol.

Psychol. 95, 45–60 (1981).7. Pietras, R. J. & Moulton, D. G. Physiol. Behav. 12, 475–491 (1974).8. Dhong, H. J., Chung, S. K. & Doty, R. L. Brain Res. 824, 312–315 (1999).9. Gower, D. B., Holland, K. T., Mallet, A. L., Rennie, P. J. & Watkins, W. J.

J. Steroid Biochem. Mol. Biol. 48, 409–418 (1994).10. Dorries, K. M. in The Science of Olfaction (eds Serby, M. J. & Chobor, K. L.)

245–278 (Springer, New York, 1992).11. Fiedler, N. & Kipen, H. Environ. Health Perspect. 105, 409–415 (1997).12. Wedekind, C. & Furi, S. Proc. R. Soc. Lond. B 264, 1471–1479 (1997).13. Porter, R. H., Cernoch, J. M. & McLaughlin, F. J. Physiol Behav. 30, 151–154

(1983).

Fig. 2. Gender effects of repeated test exposures to benzaldehyde onmean (± s.e.m.) benzaldehyde and control thresholds. Although initialthresholds for naive and experienced volunteers differed, the changesover trials did not, and thus data from the 2 groups were combined.Two-way, repeated-measures ANOVAs with gender as the between-group factor showed a significant interaction between group and time(F(9,90) = 3.29; p < 0.001).

Fig. 3. Gender effects of repeated test exposures to citralva. Meanthresholds (2 per subject per session; ± s.e.m.) for citralva obtainedfrom 5 males and 5 females across 8 test sessions, with benzaldehydetested as the control odorant before and after last session, respectively.A repeated-measures ANOVA showed a significant main effect of group(F(1,8) = 21.6; p = 0.001) and a significant interaction between groupand time (F(7,56) = 17.1; p < 0.0001).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 201

GPCRs belong to a superfamily of membrane-bound proteinsand, after agonist binding, transmit signals across the plasmamembrane mainly through heterotrimeric G proteins1,2. GPCRscan be activated by a variety of ligands including neuropeptides,lipids, biogenic amines, light, chemokines and growth factors1–3.Homology cloning and bioinformatic analyses of sequence data-bases have identified a large number of GPCRs for which the cog-nate ligands remain to be assigned and that are collectivelyreferred to as ‘orphan’ GPCRs4. In an effort to understand thesignaling components involved in the pathway and the patho-genesis of pain, we have focused our research on GPCRs specifi-cally expressed in the somatosensory system, as this axis is knownto be pivotal in sensory perception and analgesia5.

Another very important component of the nociceptive sen-sory neurotransmission is the opioidergic system. All mammalianendogenous opioid peptides are derived from three large pre-cursors: pro-opiomelanocortin, the precursor for β-endorphinand melanocyte-stimulating hormone (MSH)-related peptides;proenkephalin A, the precursor for Met- and Leu-enkephalins;and prodynorphin, the precursor for dynorphin-related pep-tides6. These opioid peptides bind and activate µ, δ and κ opi-oid receptors and also a distant relative of the opioid-receptorfamily, the orphanin FQ or nociceptin receptor7,8. The opioid

Proenkephalin A gene productsactivate a new family of sensoryneuron–specific GPCRs

Paola M.C. Lembo1*, Eric Grazzini1*, Thierry Groblewski1*, Dajan O’Donnell1*, Marie-Odile Roy1, Ji Zhang1, Cyrla Hoffert1, Jack Cao1, Ralf Schmidt1, Manon Pelletier1, Maryse Labarre1, Mylene Gosselin1, Yves Fortin2, Denis Banville2, S.H. Shen2, Peter Ström3, Kemal Payza1, Andy Dray1, Philippe Walker1 and Sultan Ahmad1

1 AstraZeneca R&D Montreal, 7171 Frederick-Banting, Ville Saint-Laurent, Quebec H4S 1Z9, Canada2 Pharmaceutical Sector, Biotechnology Research Institute, 6100 Royalmount Avenue, Montreal, Quebec H4P 2R2, Canada3 AstraZeneca R&D Södertälje, Sweden, S-151 85

*These authors contributed equally to this work.

Correspondence should be addressed to P.M.C.L. (paola.lembo@astrazeneca.com)

Published online: 19 February 2002, DOI: 10.1038/nn815

Several peptide fragments are produced by proteolytic cleavage of the opioid peptide precursorproenkephalin A, and among these are a number of enkephalin fragments, in particular bovineadrenal medulla peptide 22 (BAM22). These peptide products have been implicated in diverse bio-logical functions, including analgesia. We have cloned a newly identified family of ‘orphan’ G pro-tein–coupled receptors (GPCRs) and demonstrate that BAM22 and a number of its fragments bindto and activate these receptors with nanomolar affinities. This family of GPCRs is uniquelylocalized in the human and rat small sensory neuron, and we called this family the sensoryneuron–specific G protein–coupled receptors (SNSRs). Receptors of the SNSR family are distinctfrom the traditional opioid receptors in their insensitivity to the classical opioid antagonistnaloxone and poor activation by opioid ligands. The unique localization of SNSRs and their activa-tion by proenkephalin A peptide fragments indicate a possible function for SNSRs in sensoryneuron regulation and in the modulation of nociception.

peptide precursors are expressed throughout the CNS9,10 and inseveral peripheral tissues and, as stated earlier, give rise to a num-ber of active peptides.

Proenkephalin A can be proteolytically cleaved to produce anumber of peptide products including peptide F, peptide E, Met-enkephalin, BAM22, BAM20 and BAM12 (ref. 11). The functionof many of these peptide fragments has remained obscure,although they have been shown to affect a number of physiolog-ical processes, particularly nociception12. BAM22 has the classi-cal opioid YGGFM (Met-enkephalin) motif and binds with highaffinity to all three known opioid receptors, µ, δ and κ13,14, indi-cating a possible function in pain transmission; however, its pre-cise function remains elusive.

In this report, we describe the discovery of a family of GPCRsuniquely expressed in small nociceptive sensory neurons, theSNSRs15, and the finding that BAM22 binds and activates SNSRsby an opioid-independent mechanism.

RESULTSIdentification of a family of GPCRsDuring a search for GPCRs using RNA isolated from a primary cul-ture of rat dorsal root ganglia (DRG), we identified a partial GPCRcDNA fragment of 650 base pairs (bp), called 3b-32. Subsequent

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

202 nature neuroscience • volume 5 no 3 • march 2002

cloning and analysis of the full-length rat 3b-32 receptor demon-strated an open reading frame of 1,014 bp encoding a protein of337 amino acids with 35% identity to the Mas oncogene receptor16.Using rat 3b-32 as a probe, we isolated six different intronless genesfrom a human genomic library. The rat 3b-32 receptor shares50–55% identity and 60–70% homology with the human 3b-32

subtypes (1, 2, 3, 4, 5 and 6) at the amino acid level (Fig. 1a), where-as the human 3b-32 homologs (SNSRs) share 80–98% identity witheach other (as determined using the algorithm ClustalW).

Although the homology between rat and human SNSRs is50–55%, we addressed the question of orthologs by using a degen-erate-oligonucleotide PCR approach on rat cDNA derived from

articles

Fig. 1. Sequences of the rat and human SNSRs. (a) Amino-acid sequence alignment of six human and one rat SNSR receptor subtypes. TM1–TM7,predicted transmembrane regions. (Continued on next page.)

a

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

DRG/brain/spinal cord and consistently isolated the same rat recep-tor subtype. In addition, we isolated the mas oncogene and rat tho-racic aorta receptors from other tissues using the same degenerateoligonucleotides, indicating their ability to isolate receptors shar-ing <50% homology to SNSRs. We have repeatedly screened manyhuman cDNA/genomic libraries using the rat and human SNSRsand have isolated the same sequences (Fig. 1a).

The phylogenetic analysis of SNSRs showed that they are dis-tinct from the mas and rat thoracic aorta receptor families shar-ing a 35% identity (Fig. 1b). During the preparation of thismanuscript, Dong et al.17 reported a family of GPCRs called MRGs(mas-related genes). The relationship of MRGs with SNSRs is asfollows: SNSR1 is identical to hMRGX3 described by Dong et al.,and SNSR2 is its closest subtype; SNSR4 is identical to hMRGX1,

articles

nature neuroscience • volume 5 no 3 • march 2002 203

Fig. 1. (con’t) (b) Phylogenetic analysis of SNSRs and homologs using the ClustalW program.

b

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

and SNSR3 is its closest relative; and SNSR6 is identical tohMRGX4, and SNSR5 is its closest homolog (Fig. 1b). The humanSNSRs cluster with the mouse MRGA family, whereas rat SNSRshares the highest homology with the MRGC subfamily. The ratand human SNSRs constitute a previously unknown family ofGPCRs with the seven predicted α-helical transmembrane domainscontaining highly conserved motifs such as the NPXXY sequencelocated in transmembrane 7, as reported for other members of therhodopsin class I family18.

The unique distribution of SNSRsInitial in situ hybridization (ISH) analysis indi-cated that 3b-32 mRNA expression was con-fined to the DRG neurons in both fetal (Fig. 2a)and adult (Fig. 2c) rats. Microscopic examina-tion of emulsion-processed rat DRG sectionsshowed that 3b-32 expression was in factrestricted to a subset of small-diameter neurons,as evidenced by a dense and specific accumula-tion of silver grains over some but not all small-diameter neurons; hence, we called this familySNSRs (Fig. 2f). We also detected rat SNSRmRNA transcripts in a subset of small neurons

Fig. 3. SNSRs are expressed uniquely in a subset ofnociceptive neurons in human sensory ganglia.(a–d) Low-power magnification brightfield anddarkfield photomicrographs, demonstrating cellularlocalization of human SNSR mRNA in adult DRG(a, b) and trigeminal ganglia (c, d). (e) High-powermagnification brightfield photomicrograph of a humanDRG, showing specific accumulation of silver grainsover some but not all small-diameter neurons; no spe-cific labeling is present over large-diameter neuronsand satellite cells. (f) RT-PCR analysis of 25 human tis-sues cDNAs; the only positive result is with humanDRG cDNA. Lanes: 1, 1-kb ladder; 2, DRGs; 3, brain;4, heart; 5, kidney; 6, spleen; 7, liver; 8, colon; 9, lung;10, small intestine; 11, muscle; 12, stomach; 13, testis;14, placenta; 15, salivary gland; 16, thyroid; 17, adrenalgland; 18, pancreas; 19, ovary; 20, uterus; 21, prostate;22, skin; 23, plasma blood leukocytes; 24, bone mar-row; 25, fetal brain; 26, fetal liver. (g) β−actin controls,showing equal loading.

within trigeminal ganglia (Fig. 2h), sensory gangliathat are analogous to DRG, but not in the sympathet-ic superior cervical ganglion (Fig. 2j) or in the nodoseganglion (Fig. 2l), indicating SNSR is uniquely asso-ciated with somatosensory afferents in the rat. With

the exception of dorsal root and trigeminal ganglia, all periph-eral and CNS rat tissues we examined by ISH were devoid of spe-cific SNSR labeling. We obtained similar ISH results for humandorsal root and trigeminal ganglia (Fig. 3a, c and e) using SNSR3or SNSR5 as probes. Moreover, ISH of monkey tissues withSNSR3 as a probe demonstrated similar distribution, indicatingthat the function of this receptor is likely to be the same acrossspecies (data not shown). Furthermore, RT-PCR analyses of apanel 25 human tissues using oligonucleotides designed to con-served regions of SNSR1-6 confirmed that SNSR mRNA expres-sion is detected exclusively in human DRG (Fig. 3f).

Fig. 2. SNSR is expressed in a subset of nociceptive neu-rons in rat sensory ganglia. (a, c) ISH autoradiograms, show-ing expression of SNSR mRNA in a rat embryo atembryonic day 17 (a) and an adult rat brain sagittal sectionand spinal cord cross-section flanked by two DRG (c).Specific labeling is associated solely with DRG in both thefetal and adult rat; all other tissues are devoid of SNSRmRNA expression. (b, d) Cresyl violet–stained sections, foranatomical reference. (e–l) Brightfield and darkfield pho-tomicrographs of rat SNSRs, showing specific accumulationof silver grains over a subset of small-diameter neurons inDRG (e, f) and trigeminal ganglion (TG) (g, h); superior cer-vical ganglion (SCG) (i, j) and nodose ganglion (k, l) aredevoid of SNSR expression.

204 nature neuroscience • volume 5 no 3 • march 2002

articles

a b c

d

k

Nodose

g ie

DRG TG

f h l

SCG

j

c ea

b d

DR

Gs

Bra

in

Hea

rtK

idn

ey

Liv

er

Co

lon

Lu

ng

Sm

all i

nte

stin

e

Mu

scle

Sp

leen

Sto

mac

h

Tes

tis

Pla

cen

ta

Sal

ivar

y g

lan

d

Th

yro

id

Ad

ren

al g

lan

d

Pan

crea

s

Ova

ry

Ute

rus

Pro

stat

e

Ski

n

PB

L

Bo

ne

mar

row

Fet

al b

rain

Fet

al li

ver

1.5

f

g

kb

1.0 kb800 bp600 bp

400 bp

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 4. SNSRs are expressed predominantly in IB4-positive DRGneurons in the rat. (a–d) ISH detection of SNSR mRNA (silvergrains), using 35S-labeled riboprobes in combination with SP (a),CGRP (b), IB4 (c) or VR1 (d) immunohistochemistry. Only a smallportion of SNSR mRNA–expressing cells show substance P (7%) orCGRP (7%) immunoreactivity. Most SNSR-positive cells bind IB4(76%) and many express VR1 (56%) receptors. (a, b) Toluidine bluecounterstaining shows immunohistochemically undetectableanatomical landmarks; the small blue cells (a) are satellite cells.Arrows, double-labeled cells; arrowheads, SNSR single–labeled cells.Scale bar represents 50 µm.

Most small-diameter somatosensory afferents (also knownas nociceptors) respond to noxious mechanical, thermal andchemical stimuli and can be divided into two main classes:the substance P/calcitonin gene–related peptide/TrkA popu-lation and the lectin IB4–positive population19. To identifythe neuronal phenotype(s) of SNSR-expressing cells, we useddouble-labeling studies combining ISH (using the rat SNSRas a probe) with immunohistochemical detection of classical neu-ronal markers on rat DRG tissue sections. Only a very small pro-portion of SNSR mRNA–expressing cells (∼ 7%) containedimmunoreactivity to substance P (Fig. 4a) or calcitoningene–related peptide (Fig. 4b), whereas most SNSR-positive cells(∼ 80%) were IB4-positive (Fig. 4c), indicating that SNSRs areassociated preferentially with the IB4 class of nociceptors.Approximately half of the SNSR-positive neurons co-localizedwith the heat-responsive vanilloid receptor VR1 (Fig. 4d), indi-cating that SNSR-positive neurons may differ in their sensitivi-ty to capsaicin and thermal stimuli.

Functional characterization of SNSRsTo identify putative ligands for SNSRs, we screened HEK293scells stably expressing either human SNSR3 or SNSR4 in a cal-cium-mobilization assay using fluorescence-imaging plate read-er technology (FLIPR). We initially tested a panel of knownpeptides, lipids, biogenic amines and nucleosides at 1 µM.SNSR3 was preferentially activated by opioid-related peptides,including γ1-MSH, γ2-MSH, γ3-MSH and dynorphin A, witheffector concentration for half-maximum response (EC50) valuesof 457 ± 50, 372 ± 95, >1,000 and >1,000 nM, respectively (Table 1). Further characterization showed that BAM22 was themost potent compound and evoked a large and dose-dependentrelease of intracellular calcium in cells stably expressing thehuman SNSR3, with an EC50 value of 13 ± 5 nM (n = 10) (Table 1). We obtained similar results with BAM22 and otheropioid-type ligands for cells stably expressing SNSR4 (EC50value, 16 ± 5; n = 5; Table 1), whereas we found no response toBAM22 in nontransfected cells.

Structure–function studies with BAM22 fragments and otherproenkephalin A–derived peptides such as peptide E (BAM3200or BAM (1–25)) demonstrated a very different structural require-ment for the activation of opioid receptor and SNSR3 (Table 1).Many of the classical opioid peptides13,14 potent at δ opioid recep-tors (DORs) were inactive at SNSRs and, conversely, many BAM-related peptides potent at SNSRs were inactive at DORs. TheMet-enkephalin motif YGGFM was not required for the func-tional activation of SNSR3 (Table 1); in fact, the entire Met-enkephalin motif at the N-terminal region of BAM22 waseliminated with no apparent loss in the potency for the peptide(Table 1). Similarly, the C-terminal fragments of BAM22, BAM(8–22) and BAM (8–25), which do not contain the YGGFMmotif, were full and potent agonists at the human SNSRs

(Table 1). We obtained similar results with HEK293s cells express-ing SNSR4 (Table 1). Collectively, the data indicate that SNSRsbelong to a distinct non-opioid class of receptors, as these tworeceptor families demonstrate very distinct structure–activityrelationships with little overlap in their activation profiles.

We found further distinction between opioid receptor andSNSRs with respect to their sensitivity to classical opioid antag-onists. The opioid-receptor antagonist naloxone abolished thecalcium response elicited by both BAM22 and SNC80 (a δ opi-oid–selective agonist) at a concentration of 10 µM in cells co-expressing the human DOR (hDOR) in the presence of Gαqi5,a chimeric G protein20 (Fig. 5a). In contrast, the BAM22-medi-ated calcium response in cells expressing SNSR3 (in the absenceof Gαqi5) was not affected by the presence of either naloxone(Fig. 5b) or naltrindole (data not shown) at concentrations thatabolished the calcium response mediated by the DOR. Theinsensitivity to naloxone and natrindole was reproducible withSNSR4-expressing HEK293s cells (data not shown). The dataindicate that SNSRs represent a family of receptors distinct fromthe opioid-receptor family. The BAM22-mediated calciumresponse was not abolished by pretreatment with pertussistoxin, indicating that Gαq proteins are involved in the calci-um-signaling pathway (Fig. 5c).

Binding profile of SNSRsWe further assessed the specificity of BAM22 for human SNSRwith whole-cell binding analyses. Using [3H]BAM (8–22) as aradioligand on HEK293s cells stably expressing SNSR4, we iden-tified a single population of binding sites with a maximum boundvolume (Bmax) of 1.3 ± 0.2 pmol/mg and a high-affinity value ofKd = 10.2 ± 1.4 nM (n = 3) (Fig. 6a and b). We obtained similarresults with SNSR3 (Kd = 9.6 ± 1.8 nM; Bmax = 0.5 ± 0.1pmol/mg). In addition, [3H]BAM (8–22) specific binding onhSNR4 was not displaced by naloxone at concentrations of up to10 µM (Fig. 6c). Moreover, the N-terminal fragments of BAM22,BAM12 and other opioid RF amide-related peptides did notinhibit the specific binding of [3H]BAM (8–22). Thus, the N-ter-minal motif of BAM-related peptides does not confer function-al or binding activities on SNSRs. Finally, BAM22 and BAM(8–22) inhibited the specific binding of [3H]BAM (8–22), withKi values of 21.0 ± 4.5 nM (n = 3) and 17 ± 2.2 nM (n = 3),respectively (Fig. 6c). The rank order of potency for BAM22and BAM (8–22) was similar for binding and functional assays(Fig. 6c and Table 1). We confirmed these results in cells

articles

nature neuroscience • volume 5 no 3 • march 2002 205

a b

c d

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Moreover, we have also demonstrated that SNSRs are potent-ly activated by proenkephalin A gene products including BAM22and BAM (8–22), thus reinforcing the idea that SNSRs may mod-ulate pain transmission. BAM22 is derived from proenkephalin Agene and its mRNA is localized in the rat DRG and the dorsalhorn of the spinal cord, areas known to be involved in the controlof nociception22,23. Biochemical studies have demonstrated thepresence of the proenkephalin A–derived peptide products in ratDRG neurons24, indicating that in addition to targeting the opi-oid receptors, these peptides could possibly target SNSRs. How-ever, the presence and physiological function of BAM (8–22)remain to be determined.

The post-translational processing of proenkephalin A involvesdistinct proteases giving rise to several products including Met-and Leu-enkephalins, peptide F, peptide E (BAM3200/BAM(1–25)), BAM18 and BAM20 (ref. 11). Tandem mass spectrom-etry analysis using bovine chromaffin cells and mouse brain hasdemonstrated that the processing of the proenkephalin A andother neuropeptides is not limited to the conventional basicamino-acid residues but could also result from cleavage at othersites, thereby providing support for the idea of the existence ofBAM (8–22) and BAM (8–25)25,26.

expressing SNSR3, and obtained Ki values for BAM (8–22) of2 and 4 nM (n = 2). As expected, [3H]BAM (8–22) showed nobinding activity on the human opioid receptors µ, δ and κ, as theMet-enkephalin motif is absent (data not shown). The data agreewith the functional results with respect to the structure–activityrelationship of BAM-related peptides.

DISCUSSIONWe have cloned a family of GPCRs, the SNSRs, that is unique-ly localized in a subset of small dorsal root and trigeminal sen-sory neurons in both rat and human. To our knowledge, noother GPCRs except those of this family (described in this studyand that of Dong et al.17) demonstrate such a localized distrib-ution exclusive to the sensory ganglia. As small-diameter sen-sory neurons are believed to mediate nociceptive transmissionin both acute pain5 and chronic pain states associated with nerveinjury or inflammation19, it is highly probable that SNSRs areinvolved in the function of nociceptive neurons. Indeed, otherproteins such as tetrodotoxin (TTX)-insensitive sodium chan-nels, vanilloid receptors and P2X3 receptors with relatively selec-tive distribution in small sensory neurons have all beenindicated to be important in nociception21.

206 nature neuroscience • volume 5 no 3 • march 2002

articles

Table 1. The EC50 values of opioid-related peptides as determined by FLIPR analyses using HEK293s cells expressinghuman SNSR3 SNSR4 or DORs.

Sequence DOR SNSR3 SNSR4(EC50, nM) (EC50, nM) (EC50, nM)

γ1-MSH amide YVMGHFRWDRF Inactive 457 ± 50 600;100γ2-MSH YVMGHFRWDRFG Inactive 372 ± 95 333 ± 130γ3-MSH YVMGHFRWDRFGRRQGSSSSGVGGQ ND >1,000 >1,000Met-Enk YGGFM 3; 5 >1,000 NDMet-Enk-RF-amide YGGFMRF 9 ± 3 >1,000 >1,000Met-Enk-Arg-Phe YGGFMRF 10 ± 3 >1,000 NDMet-Enk-Gly-Leu YGGFMRGL 30; 22 >1,000 NDLeu-Enk YGGFL 6; 10 >1,000 NDDynorphin-A YGGFLRRIRPKLKWDNQ 40; 29 >1,000 >1,000Dynorphin-A amide YGGFLRRIRPKLKWDNQ 14 ± 4 >1,000 >1,000Dynorphin A (1–13) YGGFLRRIRPKLK 13 ± 2 >1,000 >1,000Dynorphin A (2–12) GGFLRRIRPKL Inactive Inactive NDDynorphin A (2–17) GGFLRRIRPKLKWDNQ Inactive Inactive NDNPFF FLFQPQRF Inactive >10,000 >10,000Nociceptin FGGFTGARYSARYLANQ ND Inactive NDBAM3200 (1–25) YGGFMRRVGRPEWWMDYQKRYGGFL 369 ± 30 565 ± 150 1,000; 500BAM22 (1–22) YGGFMRRVGRPEWWMDYQKRYG 16 ± 7 13 ± 5 16 ± 5BAM (1–20) YGGFMRRVGRPEWWMDYQKR 30; 21 360; 220 350; 200BAM (1–12) YGGFMRRVGRPE 6 ± 2 Inactive InactiveBAM (2–22) GGFMRRVGRPEWWMDYQKRYG Inactive 25 ±10 33 ±12BAM (6–22) RRVGRPEWWMDYQKRYG Inactive 21 ± 4 8; 25BAM (8–22) VGRPEWWMDYQKRYG Inactive 28 ± 8 14 ±7BAM (13–22) WWMDYQKRYG Inactive 102 ± 20 150; 50BAM (15–22) MDYQKRYG Inactive >1,000 >1,000BAM (16–22) DYQKRYG Inactive >1,000 >1,000BAM (18–22) QKRYG Inactive Inactive InactiveBAM (8–25) VGRPEWWMDYQKRYGGFL 163; 230 27 ± 8 25 ± 7BAM (8–20) VGRPEWWMDYQKR Inactive 170 ± 40 1,000; 400BAM (8–18) VGRPEWWMDYQ Inactive Inactive Inactive

EC50 values were determined by FLIPR assays using HEK293s cells expressing SNSR3, SNSR4 or hDOR (in the presence of Gαqi5). ND, not determined.Data represent n = 3 ± s.e.m., or n = 2 where two EC50 values are given.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Peptide E and BAM22 have been identified either chemical-ly or immunohistochemically in brain, gastrointestinal tissue,plasma and chromaffin granules of the bovine adrenalgland10,26–30 and have been implicated in the inhibition of gas-trointestinal transit, micturition reflex and analgesia2,10,12,26–32.The complete sequences of peptide E, and hence its productsBAM22, BAM (8–22) and BAM (8–25), are phylogenetically con-served, indicating that they may share similar physiological func-tions. The actions of these peptides on nociception are presumedto be mediated through their interactions at opioid receptors,which may well be true9,12,29,30,32–34. In addition to activatingopioid receptors, however, BAM-related peptides might also acti-vate SNSRs in vivo, as they show high affinity for SNSRs. Giventhe low homology between the SNSRs and opioid receptors(10–15% identity), their insensitivity to opioid antagonists andtheir different pharmacology, the SNSRs are distinct from theopioid-receptor family.

Opioids have been used clinically as effective analgesics formany pain conditions, but their use is limited by their consid-erable CNS-mediated side effects. The unique localization ofSNSRs to a subset of sensory neurons in the DRG and thetrigeminal ganglion might indicate a lower potential for sideeffects compared to other, more widely distributed targets.

Recently Dong and colleagues reported the cloning andcharacterization of a family of mouse receptors, calledMRGs17. The family was classified into four main subgroups:MRGA, MRGB, MRGC and MRGD. The study showed thatMRGA1-A8 and MRGD were specifically expressed in a sub-population of sensory neurons. Furthermore, MRGA1 andMRGA4 were activated by neuropeptides belonging to the RF-amide family, in particular NPFF (EC50, ∼ 200 nM) and NPAF(EC50, ∼ 60 nM), respectively.

We have cloned rat and human receptors uniquely localized inDRG and trigeminal ganglia and have called them SNSRs. Thecloned SNSRs belong to the MRG family; the rat SNSR sharesthe highest homology with the MRGC subfamily, whereas thehuman SNSRs cluster with the MRGA subdivision. The rat andhuman SNSRs cluster in different subfamilies despite sharing asimilar mRNA distribution. Further studies may be required todetermine if the rat SNSR is the ortholog of the human SNSR;however, we have expended extensive effort to find sequencesmore similar to human SNSR and have found none so far. It ispossible that within the rat species there exists only the MRGCsubfamily and the MRGA subfamily is absent.

In contrast to Dong et al., we have identified and characterizedpharmacologically proenkephalin A gene products and BAM22 as

potent ligands for the human SNSR receptors; in contrast, NPFFand related peptides activated the human SNSRs with an affinity of>10 µM. The mouse MRGA family consists of 17 members, andit is possible that BAM22 may activate one of the MGRAs. Ourobservation that the family of SNSRs is activated by the opioid-type ligand BAM22 by a non-opioid mechanism may provide fur-ther understanding in the mechanisms of pain transmission andopportunities for therapeutic development in analgesia.

METHODSCloning of rat and human SNSRs. The partial rat SNSR was cloned bydegenerate PCR using oligonucleotides corresponding to highly con-served regions of G protein–coupled receptors with the followingnucleotide sequences: A (sense), 5′-GGCCGTCGACTTCATCGTC(A/T)(A/C)(T/C)CTI(G/T)CI(TC)TIGC(A/C/G/T)G-3′ ; B (antisense), 5′-(A/G)(C/A/T)(A/T)(A/G)CA(A/G)TAIATIATIGG(A/G)TT-3′.

Poly(A)+ mRNA was isolated from cultured fetal rat DRG (Sprague-Dawley rats). A cDNA PCR fragment corresponding to 650 bp was iso-lated after reverse transcription using a Pharmacia Biotech kit and thefollowing PCR conditions: 3 min at 94°C and 40 cycles of 1 min at 94°C,1 min at 45°C and 1 min at 72°C. The full-length rat SNSR sequence wasobtained from rat genomic DNA using the 650-bp fragment and the Pro-moter Finder DNA Walking kit (Clontech, Palo Alto, California). Thefull-length 1,154-bp sequence (337 amino acids) was used to screen ahuman genomic library in the λ vector Fix II (Stratagene, La Jolla, Cali-fornia). Using the strategies described above, six human clones were iso-lated and subcloned into pcDNA 3.0 for mammalian expression. SNSR3

articles

nature neuroscience • volume 5 no 3 • march 2002 207

Fig. 5. Effect of naloxone on the calcium response mediated by BAM22in HEK293s cells expressing human SNSRs. (a) HEK293s cells express-ing hDORs. Intracellular calcium mobilization response induced byBAM22 () or SNC80 () in nontreated cells. Effect of naloxone pre-treatment (10 µM for 3 min) on the BAM22-mediated () and SNC80-mediated () calcium response. (b) HEK293s cells expressing humanSNSR3. Intracellular calcium mobilization response induced by BAM22() or SNC80 () in nontreated cells. Effect of naloxone pretreatment(10 µM for 3 min) on BAM22-mediated () and SNC80-mediated ()calcium response. Data represent a single experiment representative ofthree independent experiments. BAM22 stimulates a dose-dependentcalcium response in HEK293s cells expressing human SNSR. (c) BAM22dose-dependently stimulates intracellular mobilization of calcium inHEK293s cells stably expressing SNSR3 in the absence () or presence() of pertussis toxin treatment (24 h; 50 ng/ml). Data represent themean ± s.e.m of 10 () and 4 () independent experiments.

a

b

c

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

208 nature neuroscience • volume 5 no 3 • march 2002

and SNSR4 were transfected into HEK293s cells using LipofectaminePlus (Invitrogen, Carlsbad, California).

Measurement of intracellular calcium. Intracellular calcium assays weredone as described before35. Cells were plated without selection markersin a 96-well plate and ‘loaded’ with 3 µM Fluo-3 AM (TEF LABS, Austin,Texas) for 1 h. After the incubation step, cells were washed five times inHanks’ medium plus 20 mM HEPES, pH 7.4, and 0.1% BSA and wereanalyzed using the FLIPR system to measure the mobilization of intra-cellular calcium in response to various ligands.

Whole-cell binding. HEK293s cells stably expressing the human SNSRreceptor (subtypes 3 and 4) were grown in DMEM medium supple-mented with 10% FBS, 2 mM L-glutamine, 100 units/ml penicillin G,0.25 µg/ml amphotericin B, 100 µg/ml streptomycin and 600 µg/mlneomycin. Cells were plated in 24-well tissue culture plates coated withpoly-D-lysine (1 × 105 cells/well) and were grown for 48 h. After removalof the culture medium, cells were washed with binding buffer (Earle’sbuffer supplemented with 0.8 mM phenanthroline, 0.09% glucose, 0.1%BSA and 0.1 mM PMSF, pH 7.40) and incubated for 2 h at 4°C with

150 µl binding buffer containing various concentrations of [3H]BAM(8–22) (65 Ci/mmol) in the presence or in absence of 1 µM unlabeledBAM (8–22). For competition studies, the concentration of [3H]BAM(8–22) was fixed at 10 nM. The reaction was terminated by removal ofbinding medium followed by a rapid wash with ice-cold binding buffer(600 µl/well). The cells were collected after the addition of 200 µl of0.1 M NaOH to each well. Cell lysate was neutralized by addition of50 µl of 2.5 M acetic acid and the associated radioactivity was mea-sured after addition of 1 ml liquid scintillation fluid (Scintisafe Plus50%; Fisher Scientific, Pittsburgh, Pennsylvania).

In situ hybridization. Frozen human tissues were obtained from theBrain and Tissue Bank for Developmental Disorders (University ofMaryland at Baltimore). Rat tissues were collected from adult maleSprague-Dawley rats (Charles River, St-Constant, Quebec, Canada).Frozen tissues were sectioned (14 µm in thickness) and thaw-mountedonto slides. Antisense and sense riboprobes were transcribed in vitrofrom linearized DNA fragment of the rat (660 bp) and human SNSR3and SNSR5 (600 bp) sequences using SP6 or T7 RNA polymerase in thepresence of [α-35S]UTP (∼ 800 Ci/mmol; Amersham, Piscataway, NewJersey). ISH was done on rat embryo sections and on several adult ratand human tissues, including CNS and various ganglia, as describedbefore35. Sections were exposed to Kodak Biomax MR film for 14–17 dand subsequently dipped in Kodak NTB2 emulsion diluted 1:1 withwater and were exposed for 6–8 weeks at 4°C before development andcounterstaining with either cresyl violet acetate or hematoxylin and eosin(Sigma, St. Louis, Missouri). All animal tissues were treated according toCanadian Council on Animal Care guidelines and approved by the localAstraZeneca Animal Care Committee.

Human gene expression profiling. Rapid-Scan Gene Expression Panels(OriGene Technologies, Rockville, Maryland) were used to confirm thehighly restricted tissue distribution of the human SNSR orthologs. TheHuman Rapid-Scan panel consists of a 24-well PCR plate containing dried,normalized to β-actin first-strand cDNA isolated from 24 human tissuesarrayed. Primers were designed to conserved regions of all six humanclones (sense, 5′-GC(C/T)(G/T)TCTCCATCTACATCCTCAAC-3′; anti-sense, 5′-A(G/T)CCCT(C/T)TGGAGAACCAGCT-3′). The RT-PCR reac-tion was performed using a Light Cycler (Roche Diagnostics, Branchburg,New Jersey) and the following conditions for human SNSRs: 10 min at96°C and 45 cycles of 1 s at 96°C, 15 s at 57°C and 30 s at 72°C (similarconditions were used for β-actin amplification except primer annealingwas done at 55°C). Human cDNA derived from DRG (Analytical Biolog-ical Services, Wilmington, Delaware) was used as a positive control forPCR amplification. The 1-kb ladder used to assess molecular sizes waspurchased from Invitrogen (Carlsbad, California).

Double-labeling studies. Rats anesthetized with 65 mg/kg sodium pento-barbital were perfused with 4% paraformaldehyde in 0.1 M sodium phos-phate buffer, and their cervical DRG together with spinal cords wereremoved, frozen and cut on a microtome into sections 30 µm in thickness.Free-floating DRG sections were processed immunohistochemically toassess the presence of substance P (SP), calcitonin gene-related peptide(CGRP, 1:1,000 dilution; Peninsula Laboratories, Belmont, California),vanilloid receptor (VR1) (1:10,000 dilution; Neuromics) and IB4 lectinbinding (1:100 dilution; Sigma), using the avidin–biotin amplificationmethod with the Vectastain ABC kit (Vector Laboratories, Burlingame,California) according to manufacturer’s specifications. Tissues were thenmounted onto slides, fixed in 4% paraformaldehyde and digested withproteinase K (10 µg/ml). In situ hybridization was done as described above.

articles

Fig. 6. [3H]BAM (8–22) selectively binds to SNSR. (a) Saturation-bind-ing isotherm for [3H]BAM (8–22); () specific binding, () nonspecificbinding. (b) Scatchard analysis of [3H]BAM (8–22)–binding isotherm.(c) Competition of [3H]BAM (8–22)’s binding to intact HEK293s cellsstably expressing SNSR4 using BAM22 (), BAM12 (), BAM (8–22)(), naloxone () or Met-enkephalin-RF-amide (). Data representthe average of three independent experiments.

a

b

c

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Double-labeled cells were identified as those showing brown cytoplasmicstaining (SP, CGRP, IB4 and VR1) and a specific accumulation of silvergrains (SNSR mRNA) over the cell body. Cervical ganglia from three ratswere quantified for each marker (five randomly selected sections/rat). Atleast 310 cells were counted for each probe–protein combination.

GenBank accession numbers. Rat SNSR, AF474986; human SNSR1,AF474987; human SNSR2, AF474988; human SNSR3, AF474989; humanSNSR4, AF474990; human SNSR5, AF474991; human SNSR6, AF474992.

AcknowledgementsWe thank R. Panetta, A. Beaudet and M. Perkins for critical review of the

manuscript, and M. Valiquette, H.-V. Khang, L. Meury, M. Coupal, J.

Butterworth and M. Duchesne for technical expertise.

Competing interests statementThe authors declare that they have competing financial interests: see the Nature

Neuroscience website (http://neuroscience.nature.com) for details.

RECEIVED 25 OCTOBER 2001; ACCEPTED 23 JANUARY 2002

1. Marinissen, M. J. & Gutkind, J. S. G-protein-coupled receptors andsignaling networks: emerging paradigms. Trends Pharmacol. Sci. 22,368–376 (2001).

2. Stadel, J. M., Wilson, S. & Bergsma, D. J. Orphan G protein–coupledreceptors: a neglected opportunity for pioneer drug discovery. TrendsPharmacol. Sci. 18, 430–437 (1997).

3. Rohrer, D. K. & Kobilka, B. K. G protein–coupled receptors: functional andmechanistic insights through altered gene expression. Physiol. Rev. 78, 35–52(1998).

4. Civelli, O. et al. Novel neurotransmitters as natural ligands of orphan G-protein-coupled receptors. Trends Neurosci. 24, 230–237 (2001).

5. Rang, H. P., Bevan, S. & Dray, A. in Textbook of Pain 3rd ed. (eds. Wall, P. D. &Melzack, R.) 201–215 (Churchill Livingstone, Edinburgh, UK, 1994).

6. Loh, Y. P., Brownstein, M. J. & Gainer, H. Proteolysis in neuropeptideprocessing and other neural functions. Annu. Rev. Neurosci. 7, 189–222(1984).

7. Henderson, G. & McKnight, A. T. The orphan opioid receptor and itsendogenous ligand–nociceptin/orphanin FQ. Trends Pharmacol. Sci. 18,293–300 (1997).

8. Mansour, A., Hoversten, M. T., Taylor, L. P., Watson, S. J. & Akil, H. Thecloned µ, δ and κ receptors and their endogenous ligands: evidence for twoopioid peptide recognition cores. Brain Res. 700, 89–98 (1995).

9. Hollt, V. et al. Pro-enkephalin intermediates in bovine brain and adrenalmedulla: characterization of immunoreactive peptides related to BAM-22Pand peptide F. Life Sci. 31, 1883–1886 (1982).

10. Pittius, C. W., Seizinger, B. R., Pasi, A., Mehraein, P. & Herz, A. Distributionand characterization of opioid peptides derived from proenkephalin A inhuman and rat central nervous system. Brain Res. 304, 127–136 (1984).

11. Dores, R. M., McDonald, L. K., Steveson, T. C. & Sei, C. A. The molecularevolution of neuropeptides: prospects for the ‘90s. Brain Behav. Evol. 36,80–99 (1990).

12. Hollt, V., Tulunay, F. C., Woo, S. K., Loh, H. H. & Herz, A. Opioid peptidesderived from proenkephalin A but not that from pro-enkephalin B aresubstantial analgesics after administration into brain of mice. Eur. J.Pharmacol. 85, 355–356 (1982).

13. Boersma, C. J., Pool, C. W., Van Heerikhuize, J. J. & Van Leeuwen, F. W.Characterization of opioid binding sites in the neural and intermediate lobe

of the rat pituitary gland by quantitative receptor autoradiography. J.Neuroendocrinol. 6, 47–56 (1994).

14. Quirion, R. & Weiss, A. S. Peptide E and other proenkephalin-derivedpeptides are potent kappa opiate receptor agonists. Peptides 4, 445–449(1983).

15. Ahmad, S., Banville, D., Fortin, Y., Lembo, P. & O’Donnell, D. Novel G-protein coupled receptor. WO 99/32519 (1998).

16. Young, D., Waitches, G., Birchmeier, C., Fasano, O. & Wigler, M. Isolationand characterization of a new cellular oncogene encoding a protein withmultiple potential transmembrane domains. Cell 45, 711–719 (1986).

17. Dong, X., Han, S., Zylka, M. J., Simon, M. I. & Anderson, D. J. A diversefamily of gpcrs expressed in specific subsets of nociceptive sensory neurons.Cell 106, 619–632 (2001).

18. Bockaert, J. & Pin, J. P. Molecular tinkering of G protein–coupled receptors:an evolutionary success. EMBO J. 18, 1723–1729 (1999).

19. Snider, W. D. & McMahon, S. B. Tackling pain at the source: new ideas aboutnociceptors. Neuron 20, 629–632 (1998).

20. Coward, P., Chan, S. D., Wada, H. G., Humphries, G. M. & Conklin, B. R.Chimeric G proteins allow a high-throughput signaling assay of Gi-coupledreceptors. Anal. Biochem. 270, 242–248 (1999).

21. Caterina, M. J. & Julius, D. Sense and specificity: a molecular identity fornociceptors. Curr. Opin. Neurobiol. 9, 525–530 (1999).

22. Harlan, R. E., Shivers, B. D., Romano, G. J., Howells, R. D. & Pfaff, D. W.Localization of preproenkephalin mRNA in the rat brain and spinal cord byin situ hybridization. J. Comp Neurol. 258, 159–184 (1987).

23. Weihe, E. Neurochemical anatomy of the mammalian spinal cord: functionalimplications. Anat. Anz. 174, 89–118 (1992).

24. Pohl, M. et al. Expression of preproenkephalin A gene and presence of Met-enkephalin in dorsal root ganglia of the adult rat. J. Neurochem. 63,1226–1234 (1994).

25. Che, F. Y. et al. Identification of peptides from brain and pituitary ofCpe(fat)/Cpe(fat) mice. Proc. Natl. Acad. Sci. USA 98, 9971–9976 (2001).

26. Goumon, Y. et al. Processing of proenkephalin-A in bovine chromaffin cells.Identification of natural derived fragments by N-terminal sequencing andmatrix-assisted laser desorption ionization-time of flight mass spectrometry.J. Biol. Chem. 275, 38355–38362 (2000).

27. Boarder, M. R., Evans, C., Adams, M., Erdelyi, E. & Barchas, J. D. Peptide Eand its products, BAM 18 and Leu-enkephalin, in bovine adrenal medullaand cultured chromaffin cells: release in response to stimulation. J.Neurochem. 49, 1824–1832 (1987).

28. Davis, T. P., Hoyer, G. L., Davis, P. & Burks, T. F. Proenkephalin A-derivedpeptide E and its fragments alter opioid contractility in the small intestine.Eur. J. Pharmacol. 191, 253–261 (1990).

29. Mizuno, K., Minamino, N., Kangawa, K. & Matsuo, H. A new family ofendogenous “big” Met-enkephalins from bovine adrenal medulla:purification and structure of docosa- (BAM-22P) and eicosapeptide (BAM-20P) with very potent opiate activity. Biochem. Biophys. Res. Commun. 97,1283–1290 (1980).

30. Swain, M. G., MacArthur, L., Vergalla, J. & Jones, E. A. Adrenal secretion ofBAM-22P, a potent opioid peptide, is enhanced in rats with acute cholestasis.Am. J. Physiol. 266, G201–G205 (1994).

31. Baird, A., Klepper, R. & Ling, N. In vitro and in vivo evidence that the C-terminus of preproenkephalin-A circulates as an 8500-dalton molecule. Proc.Soc. Exp. Biol. Med. 175, 304–308 (1984).

32. Dray, A., Nunan, L. & Wire, W. Proenkephalin A fragments exhibit spinal andsupraspinal opioid activity in vivo. J. Pharmacol. Exp. Ther. 235, 670–676 (1985).

33. Sanchez-Blazquez, P., Garzon, J., Lee, N. M. & Hollt, V. Opiate activity ofpeptides derived from the three opioid peptide families on the rat vasdeferens. Neuropeptides 5, 181–184 (1984).

34. Sanchez-Blazquez, P. & Garzon, J. Opioid activity of pro-enkephalin-derivedpeptides in mouse vas deferens and guinea pig ileum. Neurosci. Lett. 61,267–271 (1985).

35. Cao, J. et al. Cloning and characterization of a cDNA encoding a novelsubtype of rat thyrotropin-releasing hormone receptor. J. Biol. Chem. 273,32281–32287 (1998).

articles

nature neuroscience • volume 5 no 3 • march 2002 209

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

210 nature neuroscience • volume 5 no 3 • march 2002

articles

Calcium entry into cells through voltage-gated Ca2+ channelsinitiates a wide range of cellular processes including proteinphosphorylation, gene expression and neurotransmitter release1.Neuronal Ca2+ channels consist of a pore-forming α1 subunitand auxiliary β, α2δ and sometimes γ subunits2, and their func-tion depends considerably on interactions with additional reg-ulatory factors. For example, the activation ofG-protein-coupled receptors by neurotransmitters inhibitsCav2.1 and Cav2.2 channels, which mediate P/Q-type and N-type Ca2+ currents, respectively, through the binding of G-pro-tein βγ subunits to distinct sites on the Ca2+ channel α 1subunit3–5. These channels are also inhibited by direct interac-tions with synaptic SNARE (soluble NSF attachment proteinreceptor proteins)—a process that may optimize couplingbetween Ca2+ entry and synaptic vesicle fusion6–8. Character-izing the functional interactions between Ca2+ channels andother signaling molecules is therefore crucial to understandinghow many Ca2+-dependent processes in neurons are regulated.

We have shown previously that the prominent Ca2+ sensorCaM binds to a CaM-binding site (CBD) in the carboxy-ter-minal domain of the α12.1 subunit and mediates the dual feed-back regulation of Cav2.1 channels by Ca2+ ions9,10. A secondsite, located amino-terminal to the CBD, is analogous to theIQ domain that mediates Ca2+/CaM-dependent inactivationof Cav1 (L-type) channels11–13. The IQ domain of α12.1 inter-acts with CaM in vitro and also contributes to the regulationof Cav2.1 channels by CaM14,15. Ca2+/CaM mediates both facil-itation and enhanced inactivation of Cav2.1 channels in trans-fected cells during trains of depolarizations10,15. PresynapticCav2.1 channels in the brain undergo similar forms of Ca2+-

Differential modulation of Cav2.1channels by calmodulin and Ca2+-binding protein 1

Amy Lee1, Ruth E. Westenbroek1, Françoise Haeseleer2, Krzysztof Palczewski1–3, Todd Scheuer1

and William A. Catterall1

Departments of 1Pharmacology, 2Opthalmology and 3Chemistry, University of Washington School of Medicine, Seattle, Washington 98195-7280, USA

Correspondence should be addressed to W.A.C. (wcatt@u.washington.edu)

Published online: 4 February 2002, DOI: 10.1038/nn805

Cav2.1 channels, which mediate P/Q-type Ca2+ currents, undergo Ca2+/calmodulin (CaM)-dependent inactivation and facilitation that can significantly alter synaptic efficacy. Here we reportthat the neuronal Ca2+-binding protein 1 (CaBP1) modulates Cav2.1 channels in a manner that ismarkedly different from modulation by CaM. CaBP1 enhances inactivation, causes a depolarizingshift in the voltage dependence of activation, and does not support Ca2+-dependent facilitation ofCav2.1 channels. These inhibitory effects of CaBP1 do not require Ca2+, but depend on the CaM-binding domain in the α1 subunit of Cav2.1 channels (α12.1). CaBP1 binds to the CaM-bindingdomain, co-immunoprecipitates with α12.1 from transfected cells and brain extracts, and colocalizeswith α12.1 in discrete microdomains of neurons in the hippocampus and cerebellum. Our resultsidentify an interaction between Ca2+ channels and CaBP1 that may regulate Ca2+-dependent formsof synaptic plasticity by inhibiting Ca2+ influx into neurons.

dependent modulation that can lead to both synaptic facilita-tion and depression16–18. Because Cav2.1 channels are essen-tial to neurotransmitter release at most central synapses19,20,regulation by CaM may contribute widely to mechanisms ofactivity-dependent synaptic plasticity.

Calmodulin is the best characterized member of a superfam-ily of Ca2+-binding proteins that exhibit four EF-hand motifs, oneor more of which may be nonfunctional in the coordination ofCa2+ (ref. 21). Included in this superfamily are the neuronal Ca2+-binding proteins (NCBPs) that, unlike CaM, are localized pri-marily in neurons22. Some NCBPs can substitute for CaM in vitro23,24, which suggests that NCBPs may regulate effectorsthat are typically thought to be modulated by CaM. Here we havestudied the interaction of CaBP1, an NCBP located in the retinaand brain25, with Cav2.1 channels. We show that CaBP1 binds tothe CBD of the α12.1 subunit, but with properties and function-al consequences that are different from those of CaM. Our findingsexpand the repertoire of modulatory interactions that take placebetween Ca2+ channels and Ca2+-binding proteins and indicatethat NCBPs, in addition to CaM, may have a role in the activity-dependent regulation of neuronal Ca2+ influx.

RESULTSCaBP1 interacts with the CBD of α12.1Although CaBP1 is a neuron-specific Ca2+-binding protein thatshares nearly 56% amino acid sequence identity with CaM25, itdiffers in having a consensus site for N-terminal myristoylation,an alternatively spliced region, inactivating amino acid substitu-tions in the second of the four EF-hand motifs, and an extra turnin the helical domain that links the N- and C-terminal lobes

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 211

(Fig. 1a). To determine whether CaBP1 can substitute for CaMin interactions with Cav2.1 channels, we tested the ability ofCaBP1 to interact with various intracellular domains of the α12.1subunit. In yeast two-hybrid assays, CaBP1 activated transcriptionof HIS3 and lacZ reporter genes only in yeast that had beencotransformed with α12.1 constructs that included the CBD (Fig. 1b and c). CaBP1 did not interact with the IQ domain orwith a control plasmid that lacked the CaBP1 coding region.These results indicated that CaBP1 may modulate Cav2.1 channelfunction through interactions with the CBD.

To confirm that CaBP1 associated with the CBD in theintact channel, we tested whether CaBP1 co-immunoprecipi-tated with Cav2.1 channels from cotransfected tsA-201 cells.In this assay, CaM co-immunoprecipitates with α12.1 in aCa2+-dependent manner only when the cells are exposed toCa2+ ionophore9. Under these conditions CaBP1 also co-immunoprecipitated with α12.1. When Ca2+ was buffered with10 mM EGTA, however, the association of CaBP1 with thechannel was not affected (Fig. 2a). This co-immunoprecipita-tion of CaBP1 was specific because CaBP1 was not immuno-precipitated with control IgG or with α12.1-specific antibodies

in cells transfected with CaBP1 alone, and CaBP1 did not co-immunoprecipitate with α12.1 subunits that lacked the CBD.Thus, despite its Ca2+ independence, the interaction betweenCaBP1 and Cav2.1 channels requires the same intracellulardomain of Cav2.1 that binds CaM.

CaBP1 associates with neuronal Cav2.1 channelsTo determine whether CaBP1 associated with endogenous Cav2.1Ca2+ channels, co-immunoprecipitation experiments were donewith extracts from rat cerebellum, which contains high concen-trations of α12.1 and CaBP1 mRNA25,26. Immunoblots of CaBP1showed two proteins (28 and 36 kDa) that specifically co-immunoprecipitated with α12.1 (Fig. 2b, left) but not with controlIgG (Fig. 2b, middle). The 36-kDa species might represent calden-drin, a larger isoform of CaBP1 that is produced from alternativesplicing25,27. The 28-kDa species was consistent in size with thepredicted molecular mass of the long CaBP1 isoform that we usedin transfected cells (Fig. 2b, right), in support of a physiologicalinteraction between neuronal Cav2.1 channels and CaBP1.

To identify the potential cellular sites of interaction betweenCaBP1 and α12.1, we immunostained rat brain sections withantibodies specific for both proteins. Compared with theimmunostaining of α12.1, the immunostaining for CaBP1 wasgenerally far more restricted within the brain and more com-monly associated with somatodendritic regions than with nerve

Fig. 1. CaBP1 binds specifically to the CBD of α12.1. (a) Diagram ofCaBP1 and CaM. The four Ca2+-binding EF-hand motifs are shown asboxes, and key structural differences between CaBP1 and CaM are indi-cated by arrows. (b) Diagram of the rat brain α12.1 subunit (rbA) show-ing the intracellular domains that were tested for interaction withCaBP1 in yeast two-hybrid assays. The amino acid boundaries of theindicated constructs are given in parentheses. (c) β-galactosidase assaysof yeast cotransformed with the α12.1 constructs shown in (b) andeither CaBP1 or control vector (pACT2).

Fig. 2. CaBP1 associates with the α12.1 subunit in tsA-201 cells and ratbrain. (a) Lysates from cells transfected with Cav2.1 plus CaBP, CaBP1alone or Cav2.1∆CBD plus CaBP1 were subjected to immunoprecipitation(i.p.) with affinity-purified α12.1-specific antibodies or control IgG as indi-cated. Experiments were done with 10 mM EGTA (lanes 1 and 2) or 2 mM Ca2+ (lanes 3–6). Blots were probed with α12.1- (top) or CaBP1-specific antibodies (bottom). (b) Rat cerebellar proteins immunoprecipi-tated with α12.1-specific antibodies (CNA5) or control IgG wereimmunoblotted with α12.1- (top) or CaBP1-specific antibodies (bottom).Lysate from tsA-201 cells transfected with CaBP1 was used as a control.

a

b

c

a

b

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

212 nature neuroscience • volume 5 no 3 • march 2002

articles

inactivation of ICa caused by Ca2+/CaM proceeded witha single exponential time course (τ = 852.3 ± 63.7 ms at+20 mV, n = 18) that was relatively insensitive to the testvoltage (Fig. 4a and b). By contrast, CaBP1 caused ICa todecay significantly faster than in cells that were trans-fected with only Cav2.1.

In almost all of the cells that were cotransfected withCaBP1, the decay of ICa evoked by +20- and +30-mV pulseswas best fit by a double exponential function, with a slowcomponent similar to control and a fast component com-prising 30–40% of the peak current (Fig. 4c). With a +10-mV test pulse, however, biphasic inactivation was detected

in only 11 out of 20 cells that had been cotransfected with CaBP1.At this test voltage, which elicits the peak inward ICa, Ca2+/CaM-dependent inactivation is maximal10. Therefore, the absence of afast phase of inactivation in some cells cotransfected with CaBP1might have resulted from more effective competition by Ca2+/CaM.

Competition between CaM and CaBP1 for Cav2.1 channelswas supported further by the observation of a marked reduc-tion in Ca2+-dependent inactivation in cells cotransfected withCaBP1 (Fig. 5a and b). Enhanced inactivation caused by CaMresults in a significant reduction in the residual current at theend of a 1-second depolarizing test pulse normalized to the peakcurrent (Ires/Ipk) for ICa as compared with IBa (refs. 9, 10). Bycontrast, in cells cotransfected with CaBP1, Ires/Ipk was alreadyreduced when Ba2+ was the charge carrier and was not signifi-cantly different for ICa and IBa (Fig. 5a and b). Ca2+-dependentinactivation was not affected in the same way by cotransfection

Fig. 3. CaBP1 colocalizes with α12.1 in rat brain sections. Ratbrain sections were double-labeled with antibodies specific forCaBP1 and α12.1. Labeling for CaBP1 is shown in green (a, d)and for α12.1 in red (b, e). In the merged images (c, f), double-labeled structures appear yellow. Representative examples areshown from the molecular layer of the cerebellum (a–c) andthe CA1 region of the hippocampus (d–f). Scale bars, 5 µm(a–c) and 50 µm (d–f).

terminals. However, CaBP1 and α12.1 showed similar patternsof punctate staining in the CA1 region of the hippocampus andin the molecular layer of the cerebellum (Fig. 3a–f). As a largeproportion of punctate labeling of α12.1 in the cerebellum colo-calizes with that of syntaxin28, it is likely that CaBP1 and Cav2.1channels coexist in at least some presynaptic nerve terminals.Immunostaining for CaBP1 and for α12.1 also overlapped in clus-ters along the dendrites of cerebellar Purkinje neurons and instructures that resembled dendritic spines (data not shown), how-ever, which suggests that CaBP1 may associate with Cav2.1 chan-nels in the post- as well as in the presynaptic membrane.

CaBP1 enhances inactivation of Cav2.1 channelsTo elucidate the functional consequences of the interactionbetween CaBP1 and Cav2.1 channels, we determined the effectof CaBP1 on Ca2+ currents (ICa) in whole-cell patch-clamprecordings of transfected tsA-201 cells. We firstcompared the effects of transfected CaBP1 andendogenous CaM on inactivation of ICa. Wehave shown previously that Ca2+/CaMenhances the inactivation of ICa during stepdepolarizations when intracellular recordingsolutions contain 0.5 mM EGTA9,10. Here,

Fig. 4. CaBP1 enhances the inactivation of ICa in tsA-201 cells transfected with Cav2.1 channels. (a) Representative traces of ICa from cells transfectedwith Cav2.1 either alone (bottom) or with CaBP1 (top).Currents were evoked by 1-s pulses to the indicatedvoltages from a holding potential of –80 mV and werescaled for comparison. (b) Time constants for the inac-tivation of Cav2.1 channel currents in the absence ofCaBP1. Test currents were evoked by pulses to theindicated voltages as described in (a) and fit with a sin-gle exponential function. Data were averaged from6–18 cells. (c) Time constants for inactivation of ICa incells cotransfected with CaBP1. Test currents wereevoked by the same voltages as in (a) and (b), but cur-rent traces were fit with a double exponential function.Fast (τfast, filled bars) and slow time constants (τslow,open bars) were averaged from 7–20 cells.

a b c

d e f

a

b c

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 213

with CaM instead of CaBP1, which indicated that the faster,Ca2+-independent inactivation was a specific consequence ofthe modulation of Cav2.1 channels by CaBP1.

To clarify the effects of CaBP1 on fast inactivation of Cav2.1channels, we measured the amplitude of ICa at the 200-ms timepoint during a 1-s test pulse and normalized this to the peak cur-

rent (I200/Ipk, Fig. 5c and d). We used more positive test voltagesto limit Ca2+ entry, thus minimizing the contribution of endoge-nous CaM in these experiments. With 0.5 mM EGTA, faster inac-tivation of ICa in cells with CaBP1 caused a significant decrease inI200/Ipk (0.45 ± 0.06 for CaBP1 versus 0.79 ± 0.03 for control, p < 0.01). CaBP1 significantly enhanced fast inactivation of IBa

(I200/Ipk of 0.43 ± 0.09 for CaBP1 versus 0.74 ± 0.07 forcontrol, p < 0.05) and also of ICa with 10 mM of theintracellular calcium chelator BAPTA (I200/Ipk of 0.56 ± 0.08 for CaBP1 versus 0.85 ± 0.03 for control, p < 0.02). The CBD was essential for these effects on inac-tivation, because CaBP1 had no effect on channels inwhich this domain had been deleted (Fig. 5c and d,Cav2.1∆CBD; p > 0.3). Together with biochemical analyses,these results support a Ca2+-independent association ofCaBP1 with the CBD, which mediates a strong accelera-tion of Cav2.1 channel inactivation that does not requireCa2+ influx or intracellular accumulation of Ca2+.

Fig. 5. Fast, Ca2+-independent inactivation ofCav2.1 channels by CaBP1 differs from themodulation of Cav2.1 channels by CaM. (a) Cav2.1 channel currents recorded with Ca2+

or Ba2+ as the permeant ion. Test pulses wereapplied from a holding voltage of –80 mV to+10 mV (Ca2+) or 0 mV (Ba2+) for Cav2.1 eitheralone or cotransfected with CaM, or to +20 mV(Ca2+) or +10 mV (Ba2+) for cells cotransfectedwith CaBP1, to account for the positive shift involtage-dependent activation caused by CaBP1.The intracellular solution contained 0.5 mMEGTA. (b) The residual current amplitude atthe end of a test pulse (Ires, indicated in a) wasnormalized to the peak current (Ipk) for cellstransfected with Cav2.1 either alone or withCaBP1 or CaM. (c) Representative currentsevoked by a test pulse to +30 mV (+20 mV forIBa) in cells transfected with wild-type ormutant Cav2.1 lacking the CBD (Cav2.1∆CBD)either alone or cotransfected with CaBP1.Intracellular solutions contained 0.5 mM EGTAexcept where 10 mM BAPTA is indicated andextracellular solutions contained 10 mM Ca2+

except where Ba2+ is indicated. (d) Currentamplitudes at 200 ms (I200, indicated in c) werenormalized to the peak current (Ipk) and plottedfor the different conditions. Recordings werefrom tsA-201 cells transfected with Cav2.1 or Cav2.1∆CBD either alone (open bars) or with CaBP1 (filled bars). Results represent averages of5–13 cells. Asterisks indicate statistically significant differences between the paired groups (p ≤ 0.05).

Fig. 6. CaBP1 alters the voltage dependence of Cav2.1 activa-tion. Tail current–voltage curves from tsA-201 cells trans-fected with Cav2.1 (a–c) or Cav2.1∆CBD (d) either alone (opencircles) or with CaBP1 (filled circles). Test pulses (10 ms) tothe indicated voltages were applied from a holding voltage of–80 mV and peak tail currents were measured upon the repo-larization of cells to –40 mV, normalized to the largest tail cur-rent in the series, and plotted against test voltage. Test pulseswere held for 10 ms, as activation of currents was completebut inactivation was minimal during this time. Bath solutionscontained 10 mM Ca2+ (a, c, d) or Ba2+ (b), and intracellularsolutions contained 0.5 mM EGTA (a, b, d) or 10 mM BAPTA(c). Each point represents the mean of 7–20 cells.

a b

c d

a b

c d

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

214 nature neuroscience • volume 5 no 3 • march 2002

articles

larizations, which should initially minimize the impact ofvoltage-dependent inactivation and reveal facilitation ofICa early in the train. With 0.5 mM intracellular EGTA,Cav2.1 Ca2+ currents undergo a sustained facilitationand gradually inactivate below initial current amplitudesafter 800 ms of repetitive pulses (Fig. 7c), an effect thatdepends on Ca2+/CaM10. In cells cotransfected withCaBP1, facilitation of ICa was reduced markedly, withcurrent amplitudes rapidly inactivating below initial val-ues only 200 ms into the train (Fig. 7c).

The maximum facilitated ICa amplitude at 50 ms in cellscotransfected with CaBP1 (1.04 ± 0.02, n = 13) was not signifi-cantly different from the Ca2+-independent facilitation of Ba2+

currents in cells transfected with Cav2.1 alone (1.04 ± 0.03, n = 5,p = 0.80) or cotransfected with CaBP1 (1.02 ± 0.02, n = 11, p = 0.52; Fig. 7d), which indicated that CaBP1 does not supportCa2+-dependent facilitation of Cav2.1 channels. Together with theenhanced inactivation and positive shifts in activation caused byCaBP1, the absence of Ca2+-dependent facilitation would strong-ly limit voltage-dependent Ca2+ entry through Cav2.1 channels.These results highlight further the different modulation of theseCa2+ channels by CaBP1 and CaM.

DISCUSSIONWe have shown that the neuronal Ca2+-binding protein CaBP1interacts with and modulates Cav2.1 channels in a manner that ismarkedly different from that of CaM. CaBP1 bound to the CBD ofα12.1 but caused significantly faster inactivation of Cav2.1 channelcurrents than that caused by CaM. CaBP1 also positively shiftedtail current-activation curves and did not support Ca2+-dependentfacilitation of Cav2.1 currents. Neither the association of CaBP1with the CBD nor the inhibitory modulation by CaBP1 requiredCa2+, in contrast to the effects of CaM on Cav2.1 channels, whichare strictly dependent on Ca2+. The observed association and colo-calization of CaBP1 and Cav2.1 channels in neurons in the brainsuggest that Ca2+ channel regulation by CaBP1 may be an impor-tant determinant of Ca2+ signaling pathways in neurons.

Ca2+-independent binding and modulation by CaBP1The Ca2+ independence of the interaction between CaBP1 and

Fig. 7. CaBP1 does not support Ca2+-dependent facilitation ofCav2.1 channels. (a, b) Voltage dependence of Cav2.1 Ca2+

currents evoked before (P1, filled circles) and after (P2, opencircles) a depolarizing prepulse. Tail currents were measuredby repolarizing cells to –40 mV for 5 ms after variable testvoltages and normalized to the largest tail current evoked byP1. Inset, representative currents evoked by a test pulse to+10 mV before (filled circles) and after (open circles) the pre-pulse. Intracellular recording solution contained 0.5 mMEGTA. Results were obtained from cells transfected withCav2.1 either alone (a, n = 7) or with CaBP1 (b, n = 10). (c, d) Cav2.1 channel currents elicited by repetitive depolariza-tions. Test pulses (+20 mV (c) or +10 mV (d) to account forvoltage shifts cause by Ba2+ substitution) at a frequency of 100Hz were applied to cells transfected with Cav2.1 either alone(open circles) or along with CaBP1 (filled circles). Peak currentamplitudes were normalized to the first pulse in the series andplotted against time during the train. Every second data pointis shown. Intracellular recording solutions contained 0.5 mMEGTA, and bath solutions contained 10 mM Ca2+ (c) or Ba2+

(d). In (c), n = 9 for open circles; n = 13 for closed circles. In(d), n = 5 for open circles; n = 11 for closed circles.

CaBP1 shifts voltage dependence of Cav2.1 activation In cells cotransfected with CaBP1 and Cav2.1, the normalized tailcurrent–voltage curve was shifted positively and was shallowerthan in cells transfected with Cav2.1 alone (Fig. 6a). CaBP1 causedsignificant increases in the half-activation voltage, V1/2 (12.8 ± 1.3 mV for CaBP1 versus 4.5 ± 0.9 mV for control, p < 0.01), and slope factor of the tail current–voltage curve (–8. 7 ± 0.5 mV for CaBP1 versus –5.2 ± 0.6 mV for control, p < 0.01). Similar to the other actions of CaBP1 on Cav2.1 channels,these effects on ICa activation were essentially reproduced withintracellular BAPTA and extracellular Ba2+ but were not observedwith Cav2.1∆CBD channels (Fig. 6b–d), which indicates that theCa2+-independent association of CaBP1 with the CBD results ina newly identified, multifaceted regulation of Cav2.1 channels.

Ca2+-dependent facilitation is not supported by CaBP1Activity-dependent increases in intracellular Ca2+ cause an initialfacilitation of ICa owing to the interaction of Ca2+/CaM with Cav2.1channels (refs 10, 15). This Ca2+-dependent facilitation was evidentwith 0.5 mM intracellular EGTA in paired-pulse protocols, in whichCa2+ influx during a short prepulse induced a significant increasein the tail current elicited by a subsequent test pulse (Fig. 7a). Withthe same voltage protocol, no facilitation of ICa was observed in cellscotransfected with CaBP1 (Fig. 7b). Because of the strong voltage-dependent enhancement of ICa inactivation caused by CaBP1, it waspossible that paired-pulse facilitation in cells cotransfected withCaBP1 might have been obscured by the onset of inactivation dur-ing the conditioning prepulse. Alternatively, CaBP1, unlike CaM,might not support Ca2+-dependent facilitation of ICa.

To distinguish between these possibilities, we analyzed theproperties of ICa during trains of short (5-ms) repetitive depo-

a

b

c

d

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 215

rat cerebellum and that their subcellular distributions overlap inthis brain region, which indicates that CaBP1 may have a physiological role in the regulation of Cav2.1 channels. Becauseboth CaM and CaBP1 interacted with the same site on the α12.1subunit, an important issue is whether Cav2.1 would interactfunctionally with CaM and/or CaBP1 in neurons in which bothCa2+-binding proteins are expressed.

Although we do not know whether CaM and CaBP1 bindsimultaneously to Cav2.1 channels, our electrophysiological stud-ies suggested that CaM and CaBP1 might competitively regulatethe channel. CaBP1 more strongly enhanced inactivation of ICawhen the influence of Ca2+/CaM was suppressed either withextracellular Ba2+ or intracellular BAPTA, or with test voltagesthat elicited submaximal Ca2+ influx. These results imply thatwhen intracellular Ca2+ concentrations are high Cav2.1 channelsmay be facilitated predominantly by CaM, and that the inacti-vating effects of CaBP1 become most prominent when cytoplas-mic Ca2+ concentrations decline. In this way, CaM and CaBP1may coordinately act as a molecular switch to intensify neuronalCa2+ influx in response to activity-dependent alterations in intra-cellular concentrations of Ca2+.

NCBPs and synaptic transmissionEmerging evidence supports a role for NCBPs in the regulation ofsynaptic transmission. In particular, neuronal Ca2+ sensor-1(NCS-1), which is more distantly related to CaM than is CaBP1,regulates neurotransmitter release35, synapse formation36 andneuronal circuits that control associative learning37. Notably,NCS-1 has been implicated in the negative regulation of Ca2+

channels in chromaffin cells38, which suggests that Cav2.1 chan-nels may be modulated by NCBPs in addition to CaBP1.

Given the widespread distribution of Cav2.1 channelsthroughout the nervous system, the cell type–specific modula-tion of Cav2.1 by CaBP1, CaM or other NCBPs may fundamen-tally determine the nature of presynaptic and postsynaptic Ca2+

signals and the functional consequences of synaptic activity.

METHODSYeast two-hybrid assays. We amplified cDNAs encoding the long iso-form of human CaBP1 (ref. 25) and the cytoplasmic domains of α12.1by polymerase chain reaction and subcloned them into the yeast two-hybrid vectors pACT2 and pAS2-1, respectively (Clontech, Palo Alto,California). To test for interactions between CaBP1 and specific domainsof α12.1, the corresponding plasmids were cotransformed into yeast strainY190. We assayed growth on medium lacking histidine and β-galactosi-dase to identify interacting proteins as described9.

Cell culture and transfection. We grew tsA-201 cells to ∼ 70% conflu-ency and transfected them by the calcium phosphate method with anequimolar ratio of cDNAs encoding the rat brain Ca2+ channel sub-units α12.1 (rbA), β2a and α2δ (ref. 26). The α12.1 construct that lacksamino acids 1969–2000 (α12.1∆CBD) has been described10. The longisoform of human CaBP1 (ref. 25) was subcloned into the BamHI sitesof pcDNA3.1+ (Invitrogen, Carlsbad, California) and transfected at a5:1 molar excess with Ca2+ channel subunits. For electrophysiologicalexperiments, we plated cells on 35-mm dishes and transfected themwith 5 µg of total DNA, including 0.3 µg of a CD8 expression plasmidto allow the detection of transfected cells. For co-immunoprecipitationassays, we plated cells on 150-mm dishes and transfected them with 50 µg of total plasmid DNA.

Co-immunoprecipitation assays. At least 48 h after transfection, tsA-201cells were homogenized in ice-cold lysis buffer (1% Nonidet P-40 in TBS(20 mM Tris-HCl, pH 7.3, 150 mM NaCl), 10 mM EGTA and proteaseinhibitors) and centrifuged at 1,000g for 5 min. To maintain Ca2+-

Cav2.1 was unexpected given the previously observed Ca2+-dependent association of CaBP1 with other CaM targets25. It ispossible that very local rises in Ca2+ might have escaped buffer-ing by BAPTA in our experiments, which could have been suffi-cient for binding to CaBP1 and for causing Ca2+-dependentmodulation of ICa. This possibility seems unlikely, however,because the modulation by CaBP1 did not change appreciablywhen Ba2+ was the permeant ion; Ba2+ ions bind to EF-handmotifs with relatively low affinity and so should not reproduceCa2+-dependent regulation of target molecules29.

Calcium-free CaM can associate with and regulate several tar-gets, including the ryanodine receptor RyR1, cyclic GMP kinaseand a CaM-dependent adenylyl cyclase from Bordetella pertus-sis30–32. In addition, GCAPs—photoreceptor Ca2+-binding pro-teins—activate guanylyl cyclases in their Ca2+-free forms33,34.Thus, CaBP1 might have a similar flexibility and interact withand regulate some effectors without binding Ca2+. Although wecannot exclude the possibility that CaBP1 may modulate someaspects of Ca2+ channel function in a Ca2+-dependent manner,we found no evidence to support a requirement for Ca2+ in theeffects of CaBP1 on the activation and inactivation of Cav2.1channels. Thus, despite the Ca2+-sensing capability of CaBP1, wepropose that CaBP1 itself does not mediate Ca2+-dependent reg-ulation of Cav2.1 channels, but might indirectly influence feed-back regulation by Ca2+ by competing with CaM. Thus, CaBP1may act more like auxiliary Ca2+ channel β-subunits by alteringthe intrinsic properties of Cav2.1 channels to fine-tune voltage-gated Ca2+ entry in specific classes of neurons.

Distinct modulation of Cav2.1 by CaBP1 and CaMWe have shown that both CaM and CaBP1 interact with the CBD ofthe α12.1 subunit and that this site is essential for full channel reg-ulation by both proteins. Conflicting evidence indicates that the IQdomain—a sequence that is N-terminal to the CBD—is involvedin the modulation of Cav2.1 channels by Ca2+/CaM15. Our resultsdo not support the importance of the IQ domain in modulationby CaBP1 because, first, CaBP1 interacted with the CBD but notthe IQ domain in yeast two-hybrid assays (Fig. 1b and c); second,deleting the CBD prevented the co-immunoprecipitation of CaBP1with α12.1 (Fig. 2a); and third, the fast inactivation and shifts inthe voltage dependence of activation caused by CaBP1 were abol-ished in channels that lacked the CBD (Figs. 5 and 6). Notably,removing the CBD from α12.1 eliminated regulation by CaBP1more completely than it eliminated regulation by CaM10. Together,our results indicate that the CBD may be the primary determinantfor the functional effects of CaBP1 on Cav2.1 channels.

If both CaM and CaBP1 interact with the CBD, how is it thatCaBP1 causes Ca2+-independent fast inactivation and positive-ly shifted activation, whereas CaM causes Ca2+-dependent facil-itation and inactivation of Cav2.1 channels? One possibility isthat key structural features that distinguish CaBP1 from CaM,such as its extra-long central helical domain and N-terminalmyristoylation (Fig. 1a), may permit Ca2+-independent bindingof CaBP1 to the CBD, which might then lead to its uniqueinhibitory modulation of ICa. Future experiments that determinehow such differences between CaBP1 and CaM contribute to spe-cific forms of Cav2.1 regulation may reveal how ion channels andother signaling molecules are differentially modulated by CaMand related Ca2+-binding proteins.

Modulation of neuronal Cav2.1 channels by CaBP1 Our immunoprecipitation and immunofluorescence studiesshowed that CaBP1 and α12.1 associate physically in extracts of

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

216 nature neuroscience • volume 5 no 3 • march 2002

articles

dependent interactions, we pretreated some groups with 5 µM A23187and 2 mM CaCl2 for 15 min but did not include EGTA in the lysis buffer.The postnuclear supernatant (300–400 µg of membrane protein) was incu-bated with 15 µg of α12.1-specific antibodies (raised against CNA5)39 for2 h at 4°C. Immune complexes were separated on protein A–Sepharose,resolved by SDS–PAGE and transferred to nitrocellulose. For immunoblot-ting, we blocked nitrocellulose filters for 30 min in 5% milk/TBS and incu-bated them with the CaBP1 antiserum UW72 (ref. 25; 1:1,000 dilution)or with CNA5-specific antibodies (2.5 µg/ml) for 1 h. Blots were washedthree times in TBS with 0.05% Tween 20 (TBST) and incubated with horse-radish peroxidase–linked protein A (Amersham, Piscataway, New Jersey;1:2,000) for 40 min. We used ECL western blotting reagent (Amersham)for detection of chemiluminescence.

For co-immunoprecipitations from rat brain, we homogenized cere-bellar tissue from two adult male rats in 0.3 M sucrose, 75 mM NaCl, 10 mM Tris-HCl, pH 7.4, and 10 mM EGTA. We included proteaseinhibitors in the homogenization buffer and in buffers used at all sub-sequent steps. Homogenates were centrifuged for 10 min at 1,000g, andmembrane fractions were separated from the postnuclear supernatantat 100,000g for 30 min. Membrane proteins were solubilized with 4 mlof buffer A (1% Triton X-100, 10 mM Tris, pH 7.4, and 10 mM EGTA)and insoluble material was removed by further centrifugation (100,000gfor 30 min). Ca2+ channels were immunoprecipitated with 15 µg ofCNA5-specific antibodies per ml of solubilized membrane protein. Weisolated immune complexes on protein A–Sepharose and detected theassociated CaBP1 by immunoblotting as described above.

Immunocytochemistry. Anesthetized adult Sprague–Dawley rats were per-fused intracardially with 4% paraformaldehyde in 0.1 M sodium phosphatebuffer, pH 7.4. The brain was post-fixed and cryoprotected in 30% (w/v)sucrose, and tissue sections (35 µm) were cut on a sliding microtome in 0.1 M phosphate buffer. Tissue sections were rinsed with 0.1 M Tris-bufferedsaline (TBS) and blocked sequentially with 2% avidin and 2% biotin. For thedouble-labeling of CaBP1 and α12.1, we incubated tissue sections in UW72antiserum (diluted 1:100) for 36 h at 4°C, biotinylated goat antibody againstrabbit IgG (Vector Laboratories, Burlingame, California; 1:300) for 1 h at37°C, and avidin D–fluorescein (Vector Laboratories; 1:300) for 1 h at 37°C,with rinsing between each step. The tissue was then blocked with 5% nor-mal rabbit serum in TBS for 1 h and incubated with affinity-purified Fabfragments for 1 h at 37°C. After rinsing, tissue sections were incubated withantibodies specific for CNA5 (1:15) for 36 h at 4°C, biotinylated goat anti-body against rabbit IgG (1:300) for 1 h at 37°C, and avidin D–Texas Red(1:300) for 1 h at 37°C. Tissue sections were mounted on gelatin-coatedslides, protected with coverslips, and viewed with a Bio-Rad MRC 600microscope in the W.M. Keck Imaging Facility at the University of Wash-ington. All procedures conformed to protocols approved by the AnimalWelfare Committee of the University of Washington.

Electrophysiology and data analysis. At least 48 h after transfection,tsA-201 cells were incubated with CD8-specific antibody–coatedmicrospheres (Dynal, Oslo, Norway) to permit detection of transfect-ed cells. We recorded whole-cell Ca2+ currents with a List EPC-7 patch-clamp amplifier and filtered them at 5 kHz. Leak and capacitivetransients were subtracted using a P/–4 protocol. Extracellular record-ing solutions were composed of 150 mM Tris, 1 mM MgCl2 and 10 mMCaCl2 or BaCl2; intracellular solutions were composed of 120 mM N-methyl-D-glucamine, 60 mM HEPES, 1 mM MgCl2, 2 mM Mg-ATPand 0.5 mM EGTA or 10 mM BAPTA. The pH of all solutions wasadjusted to 7.3 with methanesulfonic acid.

The time course of ICa decay was fit by either A[exp(–t/τ)] orAslow[exp(–t/τslow)] + Afast[exp(t/τfast)], where t is time; Aslow and Afast are theamplitudes of the slow and fast exponentials, respectively, at t = 0; and τslowand τfast are the time constants of the decay of the two processes. Normalizedtail current–voltage curves were fit with a single Boltzmann function: A/1+ exp[(V – V1/2)/k] + b, where V is test pulse voltage, V1/2 is the midpointof the activation curve, k is a slope factor, A is the amplitude and b is thebaseline. Curve fits and data analysis were done with Igor Pro software(Wavemetrics, Lake Oswego, Oregon). All averaged data are the mean ±s.e.m. We determined the statistical significance of differences betweengroups by Student’s t-test (SigmaPlot, SPSS Science, Chicago, Illinois).

AcknowledgementsThis work was supported by NIH Research Grant R01 NS22625 to W.A.C, a

NSRA postdoctoral research fellowship from NIH (F32 NS10645) to A.L., NIH

Research Grant R01 EY08061 to K.P. and research grants from Research to

Prevent Blindness, Inc., the Alcon Research Institute and the E.K. Bishop

Foundation to K.P.

RECEIVED 24 SEPTEMBER; ACCEPTED 21 DECEMBER 2001

1. Miller, R. J. Multiple calcium channels and neuronal function. Science 235,46–52 (1987).

2. Catterall, W. A. Structure and function of neuronal Ca2+ channels and theirrole in neurotransmitter release. Cell Calcium 24, 307–323 (1998).

3. Ikeda, S. R. Voltage-dependent modulation of N-type calcium channels by G-protein βγsubunits. Nature 380, 255–258 (1996).

4. Herlitze, S. et al. Modulation of Ca2+ channels by G protein βγ subunits.Nature 380, 258–262 (1996).

5. De Waard, M. et al. Direct binding of G-protein βγ complex to voltage-dependent calcium channels. Nature 385, 446–450 (1997).

6. Bezprozvanny, I., Scheller, R. H. & Tsien, R. W. Functional impact of syntaxinon gating of N-type and Q-type calcium channels. Nature 378, 623–626(1995).

7. Zhong, H., Yokoyama, C. T., Scheuer, T. & Catterall, W. A. Reciprocalregulation of P/Q-type Ca2+ channels by SNAP-25, syntaxin andsynaptotagmin. Nat. Neurosci. 2, 939–941 (1999).

8. Wiser, O., Bennett, M. K. & Atlas, D. Functional interaction of syntaxin andSNAP-25 with voltage-sensitive L- and N-type Ca2+ channels. EMBO J. 15,4100–4110 (1996).

9. Lee, A. et al. Ca2+/calmodulin binds to and modulates P/Q-type calciumchannels. Nature 339, 155–159 (1999).

10. Lee, A., Scheuer, T. & Catterall, W. A. Ca2+/calmodulin-dependentfacilitation and inactivation of P/Q-type Ca2+ channels. J. Neurosci. 20,6830–6838 (2000).

11. Qin, N., Olcese, R., Bransby, M., Lin, T. & Birnbaumer, L. Ca2+-inducedinhibition of the cardiac Ca2+ channel depends on calmodulin. Proc. Natl.Acad. Sci. USA 96, 2435–2438 (1999).

12. Peterson, B. Z., DeMaria, C. D. & Yue, D. T. Calmodulin is the Ca2+ sensor forCa2+-dependent inactivation of L-type calcium channels. Neuron 22,549–558 (1999).

13. Zühlke, R. G., Pitt, G. S., Deisseroth, K., Tsien, R. W. & Reuter, H.Calmodulin supports both inactivation and facilitation of L-type calciumchannels. Nature 399, 159–161 (1999).

14. Pate, P. et al. Determinants for calmodulin binding on voltage-dependentCa2+ channels. J. Biol. Chem. 275, 39786–39792 (2000).

15. DeMaria, C. D., Soong, T., Alseikhan, B. A., Alvania, R. S. & Yue, D. T.Calmodulin bifurcates the local Ca2+ signal that modulates P/Q-type Ca2+

channels. Nature 411, 484–489 (2001).16. Forsythe, I. D., Tsujimoto, T., Barnes-Davies, M., Cuttle, M. F. & Takahashi, T.

Inactivation of presynaptic calcium current contributes to synapticdepression at a fast central synapse. Neuron 20, 797–807 (1998).

17. Cuttle, M. F., Tsujimoto, T., Forsythe, I. D. & Takahashi, T. Facilitation of thepresynaptic calcium current at an auditory synapse in rat brainstem. J. Physiol. (Lond.) 512, 723–729 (1998).

18. Borst, J. G. & Sakmann, B. Facilitation of presynaptic calcium currents in therat brainstem. J. Physiol. (Lond.) 513, 149–155 (1998).

19. Dunlap, K., Luebke, J. I. & Turner, T. J. Exocytotic Ca2+ channels inmammalian central neurons. Trends Neurosci. 18, 89–98 (1995).

20. Wheeler, D. B., Randall, A. & Tsien, R. W. Roles of N-type and Q-type Ca2+

channels in supporting hippocampal synaptic transmission. Science 264,107–111 (1994).

21. Polans, A., Baehr, W. & Palczewski, K. Turned on by Ca2+! The physiologyand pathology of Ca2+-binding proteins in the retina. Trends Neurosci. 19,547–554 (1996).

22. Burgoyne, R. D. & Weiss, J. L. The neuronal calcium sensor family of Ca2+-binding proteins. Biochem. J. 353, 1–12 (2001).

23. Sallese, M. et al. The G-protein-coupled receptor kinase GRK4 mediateshomologous desensitization of metabotropic glutamate receptor 1. FASEB J.14, 2569–2580 (2000).

24. Schaad, N. C. et al. Direct modulation of calmodulin targets by the neuronalcalcium sensor NCS-1. Proc. Natl. Acad. Sci. USA 93, 9253–9258 (1996).

25. Haeseleer, F. et al. Five members of a novel Ca2+-binding protein (CABP)subfamily with similarity to calmodulin. J. Biol. Chem. 275, 1247–1260 (2000).

26. Stea, A. et al. The localization and functional properties of a rat brain α1Acalcium channel reflect similarities to neuronal Q- and P-type channels. Proc.Natl. Acad. Sci. USA 91, 10576–10580 (1994).

27. Seidenbecher, C. I. et al. Caldendrin, a novel neuronal calcium-bindingprotein confined to the somato-dendritic compartment. J. Biol. Chem. 273,21324–21331 (1998).

28. Westenbroek, R. E. et al. Immunochemical identification and subcellulardistribution of the α1A subunits of brain calcium channels. J. Neurosci. 15,6403–6418 (1995).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

35. McFerran, B. W., Graham, M. E. & Burgoyne, R. D. Neuronal Ca2+ sensor 1,the mammalian homologue of frequenin, is expressed in chromaffin andPC12 cells and regulates neurosecretion from dense-core granules. J. Biol.Chem. 273, 22768–22772 (1998).

36. Chen, X. L. et al. Overexpression of rat neuronal calcium sensor-1 in rodentNG108-15 cells enhances synapse formation and transmission. J. Physiol.(Lond.) 532, 649–659 (2001).

37. Gomez, M. et al. Ca2+ signaling via the neuronal calcium sensor-1regulates associative learning and memory in C. elegans. Neuron 30,241–248 (2001).

38. Weiss, J. L., Archer, D. A. & Burgoyne, R. D. Neuronal Ca2+ sensor-1/frequenin functions in an autocrine pathway regulating Ca2+ channels inbovine adrenal chromaffin cells. J. Biol. Chem. 275, 40082–40087 (2000).

39. Sakurai, T., Westenbroek, R. E., Rettig, J., Hell, J. & Catterall, W. A.Biochemical properties and subcellular distribution of the BI and rbAisoforms of α1A subunits of brain calcium channels. J. Cell Biol. 134, 511–528(1996).

articles

nature neuroscience • volume 5 no 3 • march 2002 217

29. Chao, S. H., Suzuki, Y., Zysk, J. R. & Cheung, W. Y. Activation of calmodulinby various metal cations as a function of ionic radius. Mol. Pharmacol. 26,75–82 (1984).

30. Rodney, G. G., Williams, B. Y., Strasburg, G. M., Beckingham, K. &Hamilton, S. L. Regulation of RYR1 activity by Ca2+ and calmodulin.Biochemistry 39, 7807–7812 (2000).

31. Yamaki, T. & Hidaka, H. Ca2+-independent stimulation of cyclic GMP-dependent protein kinase by calmodulin. Biochem. Biophys. Res. Commun.94, 727–733 (1980).

32. Greenlee, D. V., Andreasen, T. J. & Storm, D. R. Calcium-independentstimulation of Bordetella pertussis adenylate cyclase. Biochemistry 21,2759–2764 (1982).

33. Rudnicka-Nawrot, M. et al. Changes in biological activity and folding ofguanylate cyclase-activating protein 1 as a function of calcium. Biochemistry37, 248–257 (1998).

34. Haeseleer, F. et al. Molecular characterization of a third member of the guanylylcyclase-activating protein subfamily. J. Biol. Chem. 274, 6526–6535 (1999).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

218 nature neuroscience • volume 5 no 3 • march 2002

articles

It is widely thought that neurons of the cerebral cortex are gen-erated in the ventricular zone (VZ) that lines the dorsal telen-cephalic ventricles1,2. Normal development of the cortex requiresthe orchestrated migration of postmitotic neurons from the ger-minal VZ to the overlying cortical plate (CP). Cell birth–datingstudies have shown that the generation of the neurons of the CPfollows an ‘inside-out’ sequence, such that early-generated cellsform the deeper layers whereas later-born cells migrate past theexisting layers to reside more superficially3–5. Upon completionof migration, neurons become organized in six layers, each con-taining a complement of pyramidal cells, the excitatory projec-tion neurons, and nonpyramidal cells, the inhibitoryinterneurons. Although radial migration is the predominantmode of movement by cortical neurons6, a substantial propor-tion migrate tangentially7–11. A direct association between neu-ronal phenotype and dispersion pattern within the cortex hasbeen seen, with pyramidal neurons arranged radially and GABA-containing interneurons dispersed tangentially12.

The majority of cortical interneurons are generated in theventral telencephalon. Different experimental approaches haveshown that cells arising in the ganglionic eminence (GE), theprimordium of the basal ganglia, transgress the corticostriatalboundary and follow tangential migratory routes to take uppositions in the developing cerebral wall13–17. Work on mutantswith genetic deletions of Dlx1 and Dlx2 has shown that bothgenes are required for the migration of these cells from the ven-tral to the dorsal telencephalon14,18. In addition, experimentson slice cultures have indicated that the neural adhesion mole-cule TAG-1, expressed on developing corticofugal axons, medi-ates this process11,19. To investigate how interneurons migrateand integrate within the developing cortical circuitry, we usedtime-lapse imaging of acute brain slices to follow the migrato-ry behavior of neurons traversing tangentially through the cor-tical anlage. These experiments showed that populations of

Ventricle-directed migration in thedeveloping cerebral cortex

Bagirathy Nadarajah1, Pavlos Alifragis1, Rachel O. L. Wong2 and John G. Parnavelas1

1Department of Anatomy and Developmental Biology, University College London, Gower Street, London WC1E 6BT, UK2Department of Anatomy and Neurobiology, Washington University School of Medicine, St. Louis, Missouri 63110, USA

Correspondence should be addressed to J.G.P. (j.parnavelas@ucl.ac.uk)

Published online: 19 February 2002, DOI: 10.1038/nn813

It is believed that postmitotic neurons migrate away from their sites of origin in the germinal zonesto populate distant targets. Contrary to this notion, we found, using time-lapse imaging of brainslices, populations of neurons positioned at various levels of the developing neocortex that migratetowards the cortical ventricular zone. After a pause in this proliferative zone, they migrate radially inthe direction of the pial surface to take up positions in the cortical plate. Immunohistochemicalanalysis together with tracer labeling in brain slices showed that cells showing ventricle-directedmigration in the developing cortex are GABAergic interneurons originating in the ganglioniceminence in the ventral telencephalon. We speculate that combinations of chemoattractant andchemorepellent molecules are involved in this ventricle-directed migration and that interneuronsmay seek the cortical ventricular zone to receive layer information.

postmitotic neurons, located at various zones of the develop-ing cortex, migrate towards the cortical VZ. We have definedthis mode of movement as ‘ventricle-directed migration’, andhave shown that neurons that enter the cortical proliferativezone subsequently migrate towards the pial surface. Further,using tracer-labeling techniques and immunohistochemistry,we have shown that neurons showing ventricle-directed migra-tion are indeed GABAergic interneurons arising in the ventraltelencephalon. These observations indicate that the VZ con-tains permissive cues for cells of noncortical origin, and mayprovide layer information not only to pyramidal neurons20,21

but also to interneurons.

RESULTSVentricle-directed migration: time-lapse imagingAcute brain slices taken from mouse or rat embryos and labeledwith Oregon Green BAPTA 488 were used to follow the migra-tory behavior of cortical cell types. Labeled cells were observedin all layers of the developing cortex after a 2-hour incubationin the dye. They were labeled in their entirety, such that tips ofgrowth cones and thin trailing processes were clearly visible(Figs. 1 and 2). Similar to earlier reports7,22, cells migrating inthe direction of the pia were oriented radially or at an angle(Fig. 1a). There were also labeled cells migrating tangentially(parallel to the pial surface) within the VZ, intermediate zone(IZ) (Fig. 1a) and CP. In these acute slice preparations, the VZmay also include portions of the subventricular zone, as thesetwo zones cannot be clearly delineated. In addition to radialand tangential modes of movement, we also observed a popu-lation of labeled cells actively migrating in the direction of theventricle from various levels of the developing cortex (Fig. 1a).We refer to this pattern of movement as ‘ventricle-directed migra-tion’. Examination of acute brain slices taken from mice at embry-onic day (E) 13–16 (n = 65 slices, 40 embryos, 320 moving cells)

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 219

Fig. 2. Time-lapse sequence of a cell undergoing ventricle-directed migra-tion in an acute mouse cortical slice labeled with Oregon Green BAPTA 488.The arrows point to the tip of the leading process that advances rapidlytowards the ventricular surface (bottom). As soon as it contacts the ventri-cle (t = 66, t = 70), the cell ‘waits’ before rapidly retracting the process(arrows, t = 72) to resume movement in the opposite direction. The arrow-heads point to the trailing process of the cell as it moves initially towards theventricle; it subsequently grows into the leading process when the cellchanges direction and moves towards the pial surface. CP, cortical plate; IZ,intermediate zone; VZ, ventricular zone. Scale bar, 20 µm. (For a time-lapseview of ventricle-directed migration, see Web Movie 2 on the supplemen-tary information page of Nature Neuroscience online.)

Fig. 1. Time-lapse imaging illustrating the typical patterns of cell movement inan acute mouse cortical slice labeled with Oregon Green BAPTA 488. (a) A–Dare labeled cells in the IZ migrating in the direction of the pial surface (A), tan-gentially through the IZ (B) and towards the ventricle (C, D) over a period of70 min. The cell bodies are highlighted with asterisks, and the arrows point tothe direction of the leading processes of cells C and D. (b) Illustration of thetrajectories and direction (arrows) of cells A–D; time interval between pointsis 10 min. (c) Somal displacement, plotted as a function of time, shows thesaltatory pattern of movement of cells C and D. CP, cortical plate; IZ, inter-mediate zone; VZ, ventricular zone. Scale bar, 15 µm. (For a time-lapse view ofneuronal migration, see Web Movie 1 on the supplementary informationpage of Nature Neuroscience online.)

showed that the earliest age at which ventricle-directed migra-tion took place was E14 (17% of moving cells). Such movementwas more prevalent, however, in the later stages of corticogen-esis (22% at E15 and 26% at E16). Cells that underwent ven-tricle-directed migration had distinct morphological features.Their somata were located in the CP, IZ or VZ, and their lead-ing processes were oriented in the direction of the ventricle(Figs. 1a and 2). The leading processes were often branched andshowed growth cone–like structures at the tips, indicating activemovement. In addition, a short, thin trailing process pointingtowards the pial surface was sometimes evident in the courseof migration (Fig. 2).

Time-lapse sequences showed that cells that migratetowards the ventricle move at average speeds of 50 µm/hour.We often observed cells that had unbranched leading process-es at the start of imaging, but gave rise to branches with time.In such cases, the soma moved rapidly (1–3 µm/min) up tothe branch point, and paused for an extended period to retractone of the processes before resuming movement in the direc-tion of the remaining branch (Fig. 1a, cell C, t = 20 min).Thus, ventricle-directed migration seemed to be saltatory, withrapid movements punctuated by short periods of relativelyslow advancement or stationary phases (Fig. 1c). Previousstudies in a variety of culture systems have indicated that cellsshowing saltatory patterns of movement are closely associated

with radial glia7,23,24. Notably, our time-lapse sequences haveshown that cells that undergo ventricle-directed migrationmove faster than do those glial-guided neurons that migratetowards the pia (Fig. 1b; compare the displacement of cell Awith cells C, D).

To further characterize the behavior of cells that undergoventricle-directed migration, we followed the movement ofthose that were located in the lower IZ at the start of imaging.These recordings showed that some cells actively migratedtowards the VZ until their leading processes, with a growthcone–like structure at the tip, reached the ventricular surface.The soma of these cells then paused for an extended period(∼ 45 min) while a thin trailing process appeared (Fig. 2). Withtime, the trailing process became thicker and extended in thedirection of the pia to become the new leading process. Sub-sequently, the old leading process retracted from the ventric-ular surface (Fig. 2; t = 70 min) and eventually disappeared asthe soma resumed movement in the direction of the pia.

a

b c

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 3. Characterization of cells with ventricle-directedprocesses in fixed brain sections. (a and b) Section taken froman E17 rat brain and double immunostained for Tuj1 (red) andGABA (green). A high proportion of Tuj1-positive neurons inthe VZ of the LCX that have their leading processes orientedtowards the ventricular surface show strong expression ofGABA (arrows). A weakly labeled cell is indicated with anarrowhead. (c and d) Section from an E16 rat brain that hadreceived a single pulse of BrdU and was fixed 12 h later. It showsthat Tuj1-positive neurons with ventricle-directed morphologies(green, arrows) in the VZ of the MCX do not contain BrdU(red). (e and f) Section from an E16 rat brain that had receivedthree pulses of BrdU and fixed after 18 h shows that GABAergicneurons (green, arrows) with ventricle-directed features in theVZ of the MCX are not positive for BrdU (red). (g) Illustrationof calbindin-positive neurons in the CP that have their leadingprocesses oriented in the direction of the ventricle (arrows) in asection taken from an E18 rat brain. (h) Schematic illustration ofthe spatio-temporal pattern of Tuj1- or GABA-positive neuronsthat show ventricle-directed morphologies in the cortical VZ.The earliest cohort was observed in the LCX at E15 and contin-ued towards the MCX over time, maintaining a lateral to medialgradient. VZ, ventricular zone; CP, cortical plate; IZ, intermedi-ate zone; MCX, medial cortex; LCX, lateral cortex. Scale bars:(a–f), 25 µm; (g), 12 µm.

220 nature neuroscience • volume 5 no 3 • march 2002

articles

Interneurons show ventricle-directed morphologyTo investigate whether cells with leading processes orient-ed in the direction of the ventricle were present in thedeveloping brain, we stained sections of fixed embryonicmouse (E13–16) and rat (E16–18) brains with a panel ofantibodies that label early neuronal populations. Theseexperiments showed the presence of Tuj1-positive cells inthe VZ whose features were similar to those of cells thatunderwent ventricle-directed migration in real-time imag-ing (compare Figs. 3a and b with Fig. 2). Further, a numberof cells with ventricle-directed features in the VZ, IZ andCP were also immunopositive for GABA or calbindin,markers of cortical interneurons. To determine the pro-portion of neurons with ventricle-directed processes in the cor-tical VZ that are GABAergic, we double immunolabeled sectionsfor both Tuj1 and GABA. Our analysis showed that, at all agesexamined, nearly all (90%) Tuj1-positive neurons with ventri-

cle-directed processes also expressed GABA (Figs. 3a and b).Notably, although the intensity of Tuj1 labeling was fairly con-stant in these neurons (Fig. 3b), the intensity of immunostain-ing for GABA varied considerably (Fig. 3a). Further, to determinethe spatio-temporal pattern of distribution of neurons with ven-tricle-directed features, we examined brain sections that werestained for Tuj1 or GABA. The earliest group of ventricle-direct-ed cells appeared in the lateral cortex of rats around E15 (E13 in

Fig. 4. Cells that undergo ventricle-directed migration arise in ven-tral telencephalon. Placement of CMTMR-coated particles in the LGEof slices obtained from embryonic rat brains showed dye-labeled cellsin the cortex after 1–2 DIV. (a) Horizontally and radially orientedlabeled cells present in the VZ, IZ and CP of an E17 slice after 2 DIV.(b, c) Labeled cells with leading process oriented towards the ventri-cle were present in the VZ of an E16 cortical slice (b) and in the CPand IZ of an E18 slice (c; arrows) after 1 DIV. In addition, cells thathad leading processes oriented in the direction of the pial surfacewere also observed in the VZ (indicated by * in c). (d) Schematic dia-gram illustrating the distribution of dye-labeled cells that had ventri-cle-directed processes in regions of the cortical VZ along therostro-caudal axis of the brain. These cells were more prevalent inthe VZ of slices that were obtained from middle regions of the brain.VZ, ventricular zone; IZ, intermediate zone; CP, cortical plate; LV, lat-eral ventricle. Scale bars: (a, c), 100 µm; (b), 15 µm.

b

c d

f

a

g

e

h

a

b c d

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 221

mouse) and continued to progress towards the medial cortex(Fig. 3h). To ascertain whether this population of cells alsoincluded migrating neurons that were generated in the corticalVZ, we carried out birth-dating experiments with bromod-eoxyuridine (BrdU). For these experiments, pregnant rats (E16)were given a single injection of BrdU and the embryos harvested12 hours or, in some cases, 18 or 24 hours after three pulses ofBrdU (at 2-hour intervals). Examination of brain sections thatwere double immunolabeled for Tuj1 or GABA with BrdU showedthat none of the Tuj1- or GABA-positive neurons in the VZ, eitherpositioned horizontally or with ventricle-directed processes, hadincorporated BrdU at 12-, 18- (Figs. 3c to f) or 24-hour time-points. These observations indicate that neurons that lay scatteredin the VZ with ventricle-directed processes must have becomepostmitotic at least 12 hours before to the injection of BrdU and,hence, probably are not cortical VZ–derived neurons. Takentogether, these observations corroborate our time-lapse data andillustrate that neurons with ventricle-directed features are a sub-set of interneurons that are probably of noncortical origin.

Ventricle-directed neurons arise in the basal forebrainRecent studies indicate that the vast majority of cortical interneu-rons migrate into the cortex from the GE in the ventral telen-cephalon14,16. To confirm the origin of interneurons that showventricle-directed migration, tungsten particles coated withCMTMR were placed in the subventricular zone of the lateral gan-glionic eminence (LGE) of slices taken from embryonic rat ormouse brains. Examination of slices after 2 days in vitro (DIV)showed that dye-labeled cells had migrated across the corticos-

triatal boundary, with the earliest cohort observed inthis region 12 hours after dye placement. The greatmajority of labeled cells had leading processes orient-ed horizontally, indicating active tangential movement.In addition, dye-labeled cells with ventricle-directedmorphology were observed in the VZ, IZ and CP at allembryonic ages examined (Figs. 4b and c). Further, inthe VZ, cells with such morphology often had leadingprocesses that contacted the ventricular surface in amanner similar to that seen in time-lapse sequences(compare Fig. 4b with Fig. 2). Immunohistochemicalanalysis of cultured slices obtained from E16–18 ratbrains showed that the majority of dye-labeled cellswith ventricle-directed morphology were GABA posi-tive (Figs. 5a and b). Cell counts in slices (30 slices from10 embryos) taken from E17 brains and maintained for2 DIV showed that ∼ 70% of dye-labeled cells (n = 190cells) located in the VZ that had ventricle-directedprocesses were strongly immunoreactive for GABA. Theremainder were either unlabeled or weakly labeled.

Recent reports have indicated that interneuronsarising in the MGE and LGE show temporal differencesin their migratory patterns25. To investigate whetherneurons that emanate from both eminences undergoventricle-directed migration, the subventricular zonesof the two regions were labeled each with different flu-orescent dyes in slices taken from E17 rat embryos andmaintained for 2 DIV. These experiments indicatedthat neurons emanating from both the MGE and LGEundergo ventricle-directed migration (Figs. 5c to e).

Previous experiments in the rostral migratorystream have indicated that migrating neurons have theability to divide in situ after being committed to theneuronal phenotype26. To investigate whether some

GABAergic neurons seek the cortical VZ for mitosis, slices fromE16 rat embryos were incubated with BrdU after the applicationof dye for 24–48 hours. Immunohistochemical analysis showedthat none of the dye-labeled cells in the neocortex had incorpo-rated BrdU (Figs. 6a and b). To further investigate the time ofterminal division of these cells, slices were prepared from E17rat embryos that had received a single injection of BrdU 24 hoursbefore the fetuses were harvested. Examination of slices that weremaintained for 1–3 DIV showed that, although some dye-labeledcells were positive for BrdU after 1 DIV, most, including thosewith ventricle-directed processes, were positive only after 2–3DIV (Figs. 6c and d). Thus, these ventrally derived neurons musthave become postmitotic during the 24-hour period after BrdUinjection in vivo. Taken together, these observations show thatcells undergoing ventricle-directed migration are postmitoticand committed to the neuronal phenotype at the time of enter-ing the dorsal telencephalon.

Prevalence of ventricle-directed migrationTo investigate whether the ventricle-directed mode of movementis prevalent in all regions of the cortical VZ, we placed CMTMRin the rostral or caudal portions of the ganglionic eminences inslices of E16–18 rat brains. Slices obtained from the anteriorregions of brains (sectioned at the level of the septum) and main-tained for 2 DIV showed rostral GE cells with ventricle-directedprocesses in the cortical VZ. Similarly, slices taken from caudalparts of brains (sectioned at the level of the ventral thalamus)contained cells from the caudal GE with ventricle-directedprocesses. Comparison of slices from rostral, middle and caudal

Fig. 5. Characterization of labeled cells with ventricle-directed features in vitro. (aand b) Immunohistochemical characterization of an E17 cortical slice after 2 DIVshowed that the majority of dye-labeled cells (a) were positive for GABA (greenstaining in b; colocalization appears yellow, as indicated by arrows). (c–e) Placementof red- and green-dye-coated particles in the subventricular zones of the LGE andMGE, respectively (c), showed that neurons emanating from both eminences undergoventricle-directed migration (d, e); large arrows point to the green cells from MGE,whereas the small arrows indicate the red cells that emanated from LGE. VZ, ventric-ular zone; IZ, intermediate zone; CP, cortical plate. Scale bar: 60 µm.

a b

d ec

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

222 nature neuroscience • volume 5 no 3 • march 2002

labeled neurons with their main processes oriented towards theventricle in sections of fixed embryonic cortex lends support tothese in vitro observations. Further, our double immunolabelingexperiments have shown that the vast majority of Tuj1-positiveneurons in the cortical VZ with horizontal- or ventricle-directedleading processes also contain GABA. An earlier study, usingcumulative BrdU labeling, had shown that neurons migratingtangentially through the cortical VZ are postmitotic27. The lack ofBrdU in Tuj1- or GABA-positive neurons in our experimentsindicates that these neurons must have become postmitotic atleast 12 hours (the shortest survival time analyzed) before theembryos were harvested and, hence, are probably not generatedin the cortical VZ. In this context, it is noteworthy that neuronsgenerated in the cortical VZ show pial-directed leading process-es during their radial ascent7.

Our CMTMR tracer-labeling experiments have shown that asubstantial proportion of all dye-labeled cells (30–40%) thatmigrated into the cortex had ventricle-directed leading process-

articles

cortical regions showed, however, that neurons withventricle-directed features were more prevalent in thecortex halfway in the rostro-caudal dimension (Fig. 4d).Immunohistochemical analysis indicated that themajority of dye-labeled cells from the rostral and caudalGE that had populated the anterior and posterior cor-tical regions, respectively, were also GABA positive(data not shown). To investigate whether ventricle-directed migration is widespread during corticogen-esis, slices were prepared from E16–18 rat brains andthe LGE labeled with CMTMR. Examination of slicesafter 2 DIV showed that a substantial percentage ofall labeled cells in the cortex of these slices had mor-phologies suggestive of ventricle-directed migration(30%, 33%, 40% at E16, E17 and E18, respectively)(Fig. 7). Although the fraction of labeled cells with ven-tricle-directed processes in the VZ and IZ did notchange significantly, the population of neurons thatappeared to descend from the CP increased markedlywith the progression of cortical development (Fig. 7).In addition to cells with ventricle-directed processes,the VZ also contained populations of labeled cells withother orientations. Characterization of these dye-labeled cells indicated that, although a substantialnumber (30–50%) were positioned horizontally, oth-ers had their leading processes oriented radially in thedirection of the pia (Fig. 4c; 8%, 21%, 26% at E16, E17and E18, respectively).

DISCUSSIONOur time-lapse study provides the first direct evidencethat postmitotic cells in different layers of the devel-oping neocortex actively seek the proliferative zone.Using tracer labeling together with immunohistochemistry, wehave identified the cells that undergo ventricle-directed migra-tion as populations of interneurons that arise in the ventral telen-cephalon. We suggest that these neurons actively enter the VZ toreceive layer information that is essential for their correct inte-gration into the developing cortex.

Recent studies have shown that cortical interneurons arisingin the GE follow tangential migratory routes to the developingcortex14,16. On entering the cortex, they show a tangential ori-entation and appear predominantly in the IZ but also in otherlayers of the cortical anlage. Our tracer-labeling experiments andtime-lapse recordings presented here have shown that a subsetof interneurons coursing through the IZ move initially towardsthe ventricular wall before migrating radially to their destina-tions in the CP. The presence of Tuj1-, GABA- or calbindin-

articles

Fig. 7. Neurons with ventricle-directed features are present at all stagesof corticogenesis. Analysis of E16–18 rat cortical slices after 2 DIVshowed CMTMR-labeled cells with ventricle-directed features in theVZ, IZ and CP. The number of cells with ventricle-directed processes isplotted as a fraction of all labeled cells in a given region and after nor-malizing for the thickness of the corresponding region. The total popula-tion of CMTMR-labeled cells in a given region includes those withventricle-directed features, those that were oriented horizontally, cellsthat were radially with pial-directed features, and those that could notbe classified with certainty due to insufficient labeling. Numbers refer tothe total number of labeled cells.

Fig. 6. Cells with ventricle-directed features are postmitotic. (a and b) E16 rat brainslices were treated with BrdU for 24 h after placement of CMTMR-coated particles inthe LGE. Examination of sections that were stained for BrdU (green) showed that thedye-labeled cells (a) were negative for BrdU (b, arrows), indicating that these cells werepostmitotic. (c and d) Cortical slices obtained from E17 rat embryos that had receiveda single injection of BrdU 24 h before the fetuses were harvested. Examination of slicesafter 2 DIV showed that a number of CMTMR-labeled neurons (c) were positive forBrdU (green staining in d; arrows), indicating that these cells became postmitotic duringthe 24-h period after injection. VZ, ventricular zone. Scale bar, 30 µm.

a b

c d

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

es at all embryonic ages examined. These cells closely resembledthose seen in our time-lapse imaging, strongly suggesting thatthe cells showing ventricle-directed migration in our time-lapserecordings were indeed cortical interneurons. Although the frac-tion of dye-labeled cells with ventricle-directed processes in theVZ and IZ did not vary with time, their proportion in the CPincreased significantly. It is pertinent to note that calbindin-pos-itive cells with ventricle-directed processes were more abundantin the CP of fixed brain sections in late neurogenesis (Fig. 3g).

In addition to neurons with ventricle-directed processes inthe VZ, we also saw a substantial number of dye-labeled cells ori-ented parallel to the ventricular wall. We speculate that althoughsome of the labeled cells oriented tangentially may have descend-ed from the IZ, others may have transgressed the LGE-cortex ven-tricular zones, particularly during mid- and late corticogenesis.An earlier report28 indicated that cell movements between telen-cephalic proliferative zones are restricted; however, we have seenin fixed brain sections GABA- or Tuj1-stained neurons with lead-ing process oriented tangentially in the VZ of the cortico-striatalboundary. It is possible that GE neurons continue to migrate tan-gentially through the cortical VZ as reported earlier27, but theappearance of a subset of labeled cells with pial-directed mor-phologies indicates that these cells probably do exit the VZ.

Little is known about the molecular mechanisms that guideinterneurons or the cellular elements that may provide a sub-stratum for their migration from ventral to dorsal telencephalon.It has been suggested that axons may provide a substratum fornon-radial neuronal migration29,30, and there is recent evidencethat cortical interneurons migrate along axonal bundles of thecorticofugal fiber system to reach the developing cortex11,19. Ourtime-lapse recordings have shown that interneurons that under-go ventricle-directed migration show a saltatory pattern of move-ment. Such movement suggests the involvement of glia in themigration of these cells, as previous studies23 have associatedsaltatory motion with glial guidance. Nonetheless, the possibili-ty of neurophilic interactions between axonal fibers and descend-ing interneurons cannot be ruled out.

Why ventricle-directed migration?Earlier birth-dating studies have shown that pyramidal neuronsare disposed in an ‘inside-out’ pattern within the CP. According tothe protomap hypothesis, the cortical VZ contains intrinsic posi-tional information and serves as a blueprint for the organizationof the developing cerebral cortex20. In agreement with this hypoth-esis, transplantation studies have shown that cortical neurons—presumptive pyramidal cells—obtain their laminar informationfrom the VZ before their terminal division21. Despite accumulat-ing evidence that the majority of cortical interneurons are gener-ated in the ventral telencephalon, relatively little is known abouthow these neurons integrate into specific cortical layers. Earlierthymidine autoradiography studies have shown that corticalinterneurons also have an inside-out pattern of disposition with-in the CP31,32. It is, therefore, likely that cortical interneurons mayalso require layer information that enables them to integrate with-in the CP in an inside-out gradient. Based on these earlier obser-vations, we hypothesize that a subset of cortical interneuronsenters the proliferative zone, guided by a combination of chemoat-tractant and chemorepellent molecules, in order to acquire layerinformation. Our finding that ventricle-directed migration isprevalent in all regions of the cortical VZ at all stages of cortico-genesis lends support to this notion. It is possible that theseinterneurons obtain cues from the local environment or frompyramidal cells through neural–neural interactions.

The present finding of ventricle-directed migration, togeth-er with the recent demonstration of two modes of radial migra-tion, locomotion and somal translocation22, indicates that youngneurons may use different distinct modes of cell movement toreach their positions in the developing cortex.

METHODSPreparation of brain slices. Brain slices were prepared from embryonicmice (E13–16, where E1 = day vaginal plug was found) or rats (E16–19)as described previously22. Briefly, brains embedded in 3% low-melting-point agarose (Sigma, London, UK), were sectioned in ice-cold oxy-genated artificial cerebrospinal fluid (ACSF), pH 7.4, at 300 µm using aVibroslice. Coronal slices obtained from the anterior half of the cerebralhemispheres were mounted onto porous nitrocellulose filters (0.45 µm;Millipore, London, UK) and transferred to 12-well culture plates. Sliceswere allowed to recover for 1 h in defined medium with 5% CO2 at 37°Cand then either were incubated with Oregon Green BAPTA 488 AM orreceived focal applications of fluorescent dyes. The culture medium con-tained DMEM (Sigma), 5% heat-inactivated fetal bovine serum (Gibco,UK), 1× N-2 (Gibco), 100 µM L-glutamine, 2.4 g/liter D-glucose andpenicillin/streptomycin (1:1,000, Sigma). To follow the migration oflabeled cells from the GE, slices were cultured for 2 DIV and, for prolif-eration assays, pulsed with BrdU (10 µg/ml) for varying lengths of time.

Time-lapse confocal imaging. Time-lapse imaging of migrating cells wasdone as described previously22. Briefly, acute brain slices were incubatedwith Oregon Green BAPTA 488 AM (10 µg/ml in DMSO; MolecularProbes, Eugene, Oregon) and pluronic acid (0.0025%) at 37°C for 2 h.They were then transferred to a temperature-controlled (35–37°C) glasschamber fitted onto a Bio-Rad confocal microscope stage and perfusedthroughout the recording period with oxygenated medium (40–50 ml/h).Images of labeled cells from the dorsomedial neocortex were collectedusing 488-nm excitation and 522/535-nm emission filters. Cells wereselected for imaging only if their somata and processes were clearlylabeled. To follow the migratory movements of cells over substantiallylonger periods (>3 h), stacks of images were collected in the z-plane every15 min through regions encompassing several labeled cells of interest.The speed of migration and the trajectory of migrating cells were sub-sequently analyzed using Metamorph software (Universal Imaging, WestChester, Pennsylvania).

Application of fluorescent dyes. To label the population of corticalinterneurons arising in the ventral telencephalon11, tungsten particlescoated with fluorescent markers 4,4-chloromethylbenzoylaminotetram-ethylrhodamine (CMTMR) or 5-chloromethylfluorescein diacetate(CMFDA; Molecular Probes) were applied to the subventricular zone ofthe LGE of slice cultures using micropipettes. To coat tungsten particles,stock solution of CMTMR or CMFDA (10 mM) in dimethyl sulfoxidewas diluted in ethylene dichloride to yield a final concentration of 1 mM;50 µg of tungsten particles were then spread evenly on a glass slide towhich 100 µl of the fluorescent dye was added.

Immunohistochemistry. Cultured brain slices were fixed with 4%paraformaldehyde in 0.1 M phosphate buffer (pH 7.4), embedded in 3%agar, and sectioned at 50–70 µm with a vibrotome before processing forimmunohistochemistry as described previously22. Brains of embryonicmice or rats, fixed in 4% paraformaldehyde, were cut with a cryostat at15 µm and processed for immunohistochemistry. To characterize the phe-notype of migrating cells, sections were incubated overnight with prima-ry antibodies against Tuj1 (mouse monoclonal, 1:1,000, DevelopmentalHybridoma Bank, Iowa City, Iowa; rabbit polyclonal, 1:1,000, RDIResearch Diagnostics, Flanders, New Jersey), GABA (rabbit polyclonal,1:750, Sigma, St. Louis, Missouri) or calbindin (rabbit polyclonal, 1:750,Swant, Bellinzona, Switzerland). After washing, sections were incubatedwith FITC-conjugated secondary antibodies (against mouse or rabbit,1:500, Molecular Probes) at room temperature for 2 h. For BrdU label-ing, sections were treated with 2 M HCl at room temperature for 1 h,rinsed in 0.1 M sodium borate buffer and processed for immunohisto-chemistry (using mouse monoclonal antibody, 1:500, Sigma). Labeled

articles

nature neuroscience • volume 5 no 3 • march 2002 223

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

224 nature neuroscience • volume 5 no 3 • march 2002

10. Mione, M. C., Cavanagh, J. F., Harris, B. & Parnavelas, J. G. Cell fatespecification and symmetrical/asymmetrical divisions in the developingcerebral cortex. J. Neurosci. 17, 2018–2029 (1997).

11. Parnavelas, J. G. The origin and migration of cortical neurones: new vistas.Trends Neurosci. 23, 126–131 (2000).

12. Tan, S. S. et al. Separate progenitors for radial and tangential cell dispersionduring development of the cerebral neocortex. Neuron 21, 295–304 (1998).

13. De Carlos, J. A., Lopez-Mascaraque, L. & Valverde, F. Dynamics of cellmigration from the lateral ganglionic eminence in the rat. J. Neurosci. 16,6146–6156 (1996).

14. Anderson, S. A., Eisenstat, D. D., Shi, L. & Rubenstein, J. L. Interneuronmigration from basal forebrain to neocortex: dependence on Dlx genes.Science 278, 474–476 (1997).

15. Tamamaki, N., Fujimori, K. E. & Takauji, R. Origin and route of tangentiallymigrating neurons in the developing neocortical intermediate zone.J. Neurosci. 17, 8313–8323 (1997).

16. Lavdas, A. A., Grigoriou, M., Pachnis, V. & Parnavelas, J. G. The medialganglionic eminence gives rise to a population of early neurons in thedeveloping cerebral cortex. J. Neurosci. 19, 7881–7888 (1999).

17. Wichterle, H., Turnbull, D. H., Nery, S., Fishell, G., & Alvarez-Buylla, A. Inutero fate mapping reveals distinct migratory pathways and fates of neuronsborn in the mammalian basal forebrain. Development 128, 3759–3771 (2001).

18. Anderson, S., Mione, M., Yun, K., & Rubenstein, J. L. Differential origins ofneocortical projection and local circuit neurons: role of Dlx genes inneocortical interneuronogenesis. Cereb. Cortex 9, 646–654 (1999).

19. Denaxa, M., Chan, C.-H., Schachner, M., Parnavelas, J. G. & Karagogeos, D.The adhesion molecule TAG-1 mediates the migration of corticalinterneurons from the ganglionic eminence along the corticofugal fibersystem. Development 128, 4635–4644 (2001).

20. Rakic, P. Specification of cerebral cortical areas. Science 241, 170–176 (1988).21. McConnell, S. K. & Kaznowski, C. E. Cell cycle dependence of laminar

determination in developing neocortex. Science 254, 282–285 (1991).22. Nadarajah, B., Brunstrom, J. E., Grutzendler, J., Wong, R. O. L. & Pearlman,

A. L. Two modes of radial migration in early development of the cerebralcortex. Nature Neurosci. 4, 143–150 (2001).

23. Edmondson, J. C. & Hatten, M. E. Glial-guided granule neuron migration invitro: a high-resolution time-lapse video microscopic study. J. Neurosci. 7,1928–1934 (1987).

24. Komuro, H. & Rakic, P. Dynamics of granule cell migration: a confocalmicroscopic study in acute cerebellar slice preparations. J. Neurosci. 15,1110–1120 (1995).

25. Anderson, S. A., Marin, O., Horn, C., Jennings, K. & Rubenstein, J. L. Distinctcortical migrations from the medial and lateral ganglionic eminences.Development 128, 353–363 (2001).

26. Luskin, M. B. Restricted proliferation and migration of postnatally generatedneurons derived from the forebrain subventricular zone. Neuron 11, 173–189(1993).

27. O’Rourke, N. A., Chenn, A. & McConnell, S. K. Postmitotic neurons migratetangentially in the cortical ventricular zone. Development 124, 997–1005(1997).

28. Neyt, C., Welch, M., Langston, A., Kohtz, J. & Fishell, G. A short-range signalrestricts cell movement between telencephalic proliferative zones. J. Neurosci.17, 9194–9203 (1997).

29. Rakic, P. Principles of neural cell migration. Experientia 46, 882–891 (1990).30. Gray, G. E., Leber, S. M. & Sanes, J. R. Migratory patterns of clonally related

cells in the developing central nervous system. Experientia 46, 929–940(1990).

31. Miller, M. W. Cogeneration of retrogradely labeled corticocortical projectionand GABA-immunoreactive local circuit neurons in cerebral cortex. Dev.Brain Res. 23, 187–192 (1985).

32. Cavanagh, M. E. & Parnavelas, J. G. Development of somatostatinimmunoreactive neurons in the rat occipital cortex: a combinedimmunocytochemical–autoradiographic study. J. Comp. Neurol. 268, 1–12(1988).

articles

sections were examined using a confocal microscope, and the images weresubsequently reconstructed using Metamorph imaging software.

Quantitative analysis of CMTMR-labeled cells. For quantitative analysis,slices were fixed after 2 DIV and imaged using the confocal microscope. Toascertain the position of labeled cells within the cortical anlage, the offsetsettings of the confocal microscope were adjusted to enhance tissue back-ground. Stacks of images were collected in sequence from the corticostri-atal boundary to the dorso-medial cortex, thus covering the largest aspect ofneocortex. Each stack of images, consisting a number of optical sectionscollected in the z-plane through a depth of 100 µm tissue thickness, werethen collapsed into single images using Metamorph imaging software. Amontage of the neocortex was subsequently assembled from the series ofcollapsed images and the labeled cells located in the VZ, IZ and CP werecounted. Labeled cells, positioned at various levels of the cortical anlage,with leading processes oriented towards the ventricle, were considered asthose showing ventricle-directed migration. The fraction of labeled cellsthat showed ventricle-directed features was subsequently normalized to thethickness of the zones appropriate to the stage of development.

Note: Supplementary Web Movies can be found on the Nature Neuroscience

website (http://neurosci.nature.com/web_specials).

AcknowledgementsThe work was supported by grants by the Wellcome Trust to B.N. and J.G.P.

(grant number 050325) and by the US National Eye Institute to R.O.L.W.

Competing interests statementThe authors declare that they have no competing financial interests.

RECEIVED 3 DECEMBER 2001; ACCEPTED 16 JANUARY 2002

1. Rakic, P. Mode of cell migration to the superficial layers of fetal monkeyneocortex. J. Comp. Neurol. 145, 61–83 (1972).

2. Rakic, P. Neuronal migration and contact guidance in the primatetelencephalon. Postgrad. Med. J. 54 Suppl 1, 25–40 (1978).

3. Angevine, J. B., Jr. & Sidman, R. L. Autoradiographic study of the cellmigration during histogenesis of cerebral cortex in the mouse. Nature 192,766–768 (1961).

4. Berry, M. & Rogers, A. W. The migration of neuroblasts in the developingcerebral cortex. J. Anat. 99, 691–709 (1965).

5. Rakic, P., Stensas, L. J., Sayre, E. & Sidman, R. L. Computer-aided three-dimensional reconstruction and quantitative analysis of cells from serialelectron microscopic montages of foetal monkey brain. Nature 250, 31–34(1974).

6. Hatten, M. E. Central nervous system neuronal migration. Annu. Rev.Neurosci. 22, 511–539 (1999).

7. O’Rourke, N. A., Dailey, M. E., Smith, S. J. & McConnell, S. K. Diversemigratory pathways in the developing cerebral cortex. Science 258, 299–302(1992).

8. Tan, S. S. & Breen, S. Radial mosaicism and tangential cell dispersion bothcontribute to mouse neocortical development. Nature 362, 638–640 (1993).

9. Reid, C. B., Liang, I. & Walsh, C. Systematic widespread clonal organizationin cerebral cortex. Neuron 15, 299–310 (1995).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Four neurotrophins have been identified in mammals: nervegrowth factor (NGF), brain-derived neurotrophic factor (BDNF),neurotrophin-3 (NT-3) and neurotrophin-4 (NT-4/5). NGF acti-vates TrkA, BDNF and NT-4/5 activate TrkB, and NT-3 activatesTrkC primarily, although it can also activate TrkA and TrkB lessefficiently in some cell types (reviewed in ref. 1).

Analyses of mutant mice show that neurotrophins are impor-tant in the survival, differentiation and maintenance of neuronsin the PNS, but not in the CNS. Nevertheless, the neurotrophinsare attractive candidates for modulating several processes dur-ing postnatal development of the CNS, including regulation ofdendritic arborization, axonal sprouting and synaptic transmis-sion2–8. Despite progress in understanding the functions of neu-rotrophins in regulating dendritic arborization6,7, little is knownabout their role in synaptogenesis. In vitro studies implicateBDNF-mediated activation of TrkB receptors in the developmentof GABAergic neurons9–13. In addition, in vivo studies show thatoverexpression of BDNF accelerates the maturation of cerebel-lar and cortical inhibitory circuits14,15.

The cerebellum is ideal for studies of CNS synaptogenesis.For example, synapses in the granule-cell layer are organized inhighly stereotyped structures called glomeruli, which greatly facil-itates their analysis at the ultrastructural level. TrkB and BDNFare expressed at high levels in the cerebellum during the postna-tal period, including during synaptogenesis16. BDNF is expressedin the granule cells and deep cerebellar nuclei17, whereas TrkB isexpressed at different concentrations in Purkinje cells, granulecells, interneurons and glia cells18,19. TrkB and its ligands areinvolved in the development of granule cells4,20–23, Purkinjecells4,23,24 and pontine mossy fibers25. The early postnatal mor-tality of trkB and BDNF knockout mice prevents studies on the

TrkB receptor signaling is requiredfor establishment of GABAergicsynapses in the cerebellum

Beatriz Rico, Baoji Xu and Louis F. Reichardt

Howard Hughes Medical Institute and Department of Physiology, University of California, San Francisco, California 94143, USA

Correspondence should be addressed to L.F.R. (lfr@cgl.ucsf.edu)

Published online: 11 February 2002, DOI: 10.1038/nn808

Neurotrophins are essential to the normal development and maintenance of the nervous system.Neurotrophin signaling is mediated by Trk family tyrosine kinases such as TrkA, TrkB and TrkC, as wellas by the pan-neurotrophin receptor p75NTR. Here we have deleted the trkB gene in cerebellarprecursors by Wnt1-driven Cre–mediated recombination to study the function of the TrkB in the cere-bellum. Despite the absence of TrkB, the mature cerebellum of mutant mice appears similar to that ofwild type, with all types of cell present in normal numbers and positions. Granule and Purkinje celldendrites appear normal and the former have typical numbers of excitatory synapses. By contrast,inhibitory interneurons are strongly affected: although present in normal numbers, they expressreduced amounts of GABAergic markers and develop reduced numbers of GABAergic boutons andsynaptic specializations. Thus, TrkB is essential to the development of GABAergic neurons andregulates synapse formation in addition to its role in the development of axon terminals.

roles of TrkB signaling in cerebellar synaptogenesis, because thisprocess reaches its peak in the second and third postnatal weeks26.

To overcome this problem, trkB conditional-mutant mice havebeen generated using the Cre/LoxP recombination system7,8,27.Using this system, we have generated mice that lack TrkB expres-sion in the cerebellum by crossing a trkB conditional strain7 witha line of transgenic Wnt1Cre mice in which Wnt1 regulatory ele-ments direct the expression of Cre recombinase to the neuralcrest and primordial midbrain, including the precursors cells ofthe cerebellum and precerebellar system28–30. The Wnt1Cre/trkBconditional-mutant mice survive to adulthood with normal num-bers of all cerebellar cell types, but have marked deficits inGABAergic enzymes and synapses.

RESULTSPattern of Wnt1Cre-mediated deletion in the cerebellumThe Wnt1 promoter/enhancer is active at embryonic day 8.5(E8.5) in the region of the neuroepithelium from which all cellsin the cerebellum are derived28. To determine the pattern ofWnt1-driven Cre-mediated recombination in the cerebellum, wecrossed Wnt1Cre transgenic mice with mice carrying the Crereporter R26R. In R26R mice, expression of β-galactosidase iden-tifies all cells in which Cre-mediated recombination has occurred.Analysis of adult Wnt1Cre;R26R mice showed that recombina-tion had occurred in all types of cell in the cerebellum and inmost cells in the midbrain, which indicated that progenitor cellscommitted to these regions expressed the Wnt1Cre transgeneduring development. Some recombination also occurred in otherregions of the CNS and PNS (Fig. 1a and b).

We used antibodies against β-galactosidase to examineexpression of β-galactosidase in specific cerebellar cell types

articles

nature neuroscience • volume 5 no 3 • march 2002 225

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 2. Loss of TrkB receptor in the cerebellum of trkB conditional-mutant mice. (a–g) Sagittal sections showing the expression of TrkB inwild-type (a–d) and Wnt1;fBZ/fBZ (e–g) miceat P60. (b, f) Higher-magnification images cor-responding to the boxed regions in (a) and(e), respectively. (c, d, g) Higher-magnifica-tion images corresponding to the boxedregions in (b) and (f). (b–d) In wild-type con-trols, expression of TrkB is prominent inPurkinje (pl) and Golgi cells (arrows), but lowin granular cells (star) and molecular-layerinterneurons (arrowheads). (h) Immunoblotanalysis of TrkB receptor protein in the cere-bellum of 2-month-old wild-type, fBZ/fBZ andWnt1;fBZ/fBZ mice. β-Tubulin protein wasused for normalization. igl, internal granule-cell layer; ml, molecular layer; nu, interneuronnucleus; pl, Purkinje cell layer; TrkB, full-length TrkB receptor; TrkB-T, truncated TrkBreceptor. Scale bars, 500 µm (a, e), 100 µm(b, f), 50 µm (c, g) and 10 µm (d).

To confirm that the pattern of recombination found inWnt1Cre;R26R mice was also found in our mutant mice, weexamined the expression of β-galactosidase in Wnt1Cre;fBZ/fBZmice carrying two conditional trkB alleles. The conditional fBZallele was generated by inserting a loxP-flanked full-length com-plementary DNA of trkB, followed by a tau-lacZ cDNA, into thetrkB locus8. Cre-mediated recombination thus eliminated expres-sion of TrkB and activated expression of the tau-lacZ reporterunder the control of the trkB promoter.

Expression of Tau–β-galactosidase indicated that Cre-medi-ated recombination had occurred in the progenitors of severalcell types in the adult cerebellum (Fig. 1g–j). As predicted by

(Fig. 1c–f). We found that all GABAergic interneurons, Purk-inje and granule cells were labeled by β-galactosidaseimmunostaining (Fig. 1c–f and Supplementary Table 1, avail-able on the supplementary information page of Nature Neu-roscience online). Although there were examples in whichβ-galactosidase–containing inclusion bodies were not seen in aconfocal section through a cell, inclusions bodies could alwaysbe found in other optical sections of that cell.

Fig. 1. Wnt1-driven Cre-mediated dele-tion of the R26R and trkB alleles, identi-fied by expression of the reporters lacZand tau-lacZ, respectively. (a–f) R26Rallele; (g–j) trkB allele. (a) Sagittal view ofthe brain showing the expression of lacZ.(b) Higher magnification of (a). For moredetailed information about the nuclei in(a) and (b), see the expanded version ofFig. 1a and b (Supplementary Fig. 1)on the supplementary information pageof Nature Neuroscience online. (c–f) Singleconfocal plane images using monoclonal(c–h) or polyclonal (i, j) antibodiesagainst β-galactosidase and diverse cell-type markers. (c, g) Colocalization of β-galactosidase (green puncta) and GABA(red) in Golgi cells. (d, h) Colocalizationof β-galactosidase (green puncta) andparvalbumin (red) in interneurons of themolecular layer. (e) Colocalization of β-galactosidase (green puncta) and the α6subunit of the GABAA receptor(α6GABAA; red) in granule cells. Notethe amount of α6GABAA receptor in theglomerulus (asterisks), where granule-cell dendrites are located. Receptorexpression is also localized in the thincytoplasm of the granule cell (inset). (f) Colocalization of β-galactosidase(green puncta) and calbindin (red) inPurkinje cells. Note the characteristic expression of β-galactosidase in the space occupied by the interneurons (white triangles). (i, j) Localization of β-galactosidase in Bergman glia (i) and astroglia (j), identified by morphology (i) or expression of GFAP (j). Arrows indicate double-labeled neurons orprocesses. nu, granule-cell nucleus. Scale bars, 1 mm (a), 500 µm (b), 50 µm (j), 15 µm (c, f), 10 µm (d, e, g–i).

226 nature neuroscience • volume 5 no 3 • march 2002

articles

a b

c d e f

g h i

a b c

e f g

h

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

the above results, recombination occurred in most interneuronsin the cerebellum (Fig. 1g and h and Supplementary Table 1).In contrast, none of the antibodies detected immunoreactivityto β-galactosidase in Purkinje and granule-cell bodies, mostprobably because the Tau–β-galactosidase fusion protein is trans-ported into dendrites and axons31. The results obtained withWnt1Cre;R26R mice indicated, however, that the Wnt1Cre trans-genic strain was driving recombination in all cerebellar cell types.Because activity of the Wnt1 promoter precedes the initialappearance of TrkB at postnatal day 7 (P7)21, cerebellar neuronsalmost certainly developed without exposure to signaling mediated by TrkB.

Loss of TrkB in the cerebellum of trkB mutant miceIn agreement with previous studies18,19, in situ hybridizationshowed that expression of TrkB varied among different cell types ofthe cerebellum (Fig. 2a–d). Expression was high in Purkinje cellsand Golgi interneurons (Fig. 2a–c), but more moderate in basketand stellate interneurons and in granule cells (Fig. 2c and d).

Because the Wnt1Cre transgene was expressed in the prog-enitors of all cerebellar cell types, we anticipated that trkBexpression would be lost in Wnt1Cre;fBZ/fBZ conditional-mutant mice. In situ hybridization showed that there was acomplete loss of trkB messenger RNA in the cerebellum of adultWnt1Cre;fBZ/fBZ conditional mutants (Fig. 2a–g). An affini-ty-purified antibody against the extracellular domain of TrkB32

did not detect TrkB protein in immunoblots of the cerebellumof Wnt1Cre;fBZ/fBZ conditional mutants (Fig. 2h). Consistentwith previous results8, fBZ/fBZ mice expressed roughly 25%of the normal amount of full-length TrkB and no truncatedisoforms of this protein (Fig. 2h).

Normal cerebellar architecture without TrkBTo determine the effect of the lack of trkB on the cerebellum,we first examined the gross anatomical organization in 2-month-old mutant mice. Analyses of sections stained by Nissland antibodies specific for glial fibrillary acidic protein (GFAP)indicated that there were no apparent differences in folia devel-opment, or in laminar and glial scaffold organization, betweenwild-type control and conditional-mutant mice (Fig. 3a–h).

Because our study focused on lobule IV (see below), we exam-ined the cross-sectional areas of the molecular and the granule-celllayers in this lobule. Significant reductions in the molecular (24%)

and granule cell (16%) layers were observed (Table 1). To deter-mine whether the reductions in cross-sectional areas were causedby reductions in cell number, we quantified the main cell types ofthe cerebellum. In the granule-cell layer of the conditional mutant,the density of granule cells (the main cell type in this layer) wasincreased by 12%, whereas the density of Golgi cells was unchanged(Table 1 and Fig. 3f and i). The average diameter of granule cells inmutants was not smaller than in control littermates (Table 1), whichsuggested that the loss of cross-sectional area was caused by a reduc-tion in the number or size of fibers between these cells. A decreasein fiber number or size would also explain why granule-cell densi-ty was increased slightly in mutant animals.

Because the molecular layer contains a very low density of cells,the 24% reduction in its area was almost certainly caused by a lossof dendritic or axonal volume. The cell densities in the molecu-lar layer were similar in all three genotypes (Table 1). We detect-ed no significant changes in the diameter or number of parallelaxons profiles (Table 1). A loss of Purkinje cells or reduced vol-ume of Purkinje cell trees might also potentially explain thereduced area of the molecular layer. We observed no significantchange in the density of Purkinje cells in Wnt1Cre;fBZ/fBZ mutantmice (Table 1 and Fig. 3g and j).

Granule and Purkinje cell development without TrkBAs in previous studies of BDNF and trkB mutants4,22,23, weobserved a modest delay in granule-cell migration in the trkB con-ditional-mutant mice (not shown). Thus, by the end of the thirdpostnatal week, cells had completed migration to the internal gran-ule-cell layer, and both wild-type control and mutant external gran-ule-cell layers appeared identical (not shown). Despite the absenceof TrkB, granule-cell development seemed normal. First, the den-sity of granule cells was not reduced in conditional-mutant mice(Table 1). Second, the morphology of granule cells, as observed inGolgi-impregnated sections, appeared normal (Fig. 4a and c).Third, analysis of the ultrastructure of granule cells by electronmicroscopy indicated that the organization of these cells was nor-mal (Fig. 4b and d). Fourth, the diameters of granule-cell bodiesand the density and diameter of parallel fiber profiles were simi-lar in all three genotypes (Table 1). Last, granule cells seemed toexpress normal amounts of the α6 subunit of the GABAA recep-tor (Fig. 4e–h), a molecular marker of maturity and postsynapticfunction. Together, these findings suggested that the developmentof granule cells occurred normally in the absence of TrkB signaling.

articles

nature neuroscience • volume 5 no 3 • march 2002 227

Fig. 3. Cerebellar architecture in trkB conditional-mutant mice at P50–P80. (a–d) Nissl-stained sagittal sections of wild-type (a, b) andWnt1Cre;fBZ/fBZ (c, d) mice. (b, d) Higher-magnification images from lobule IV. (e–j) Immunohistochemistry of GFAP in wild-type (e) andWnt1Cre;fBZ/fBZ (h) mice, and immunohistochemistry of GABA (f, i) and calbindin (g, j) in wild-type and Wnt1;fBZ/fBZ mice. I–X, cerebellar lobules;igl, internal granule-cell layer; ml, molecular layer; pl, Purkinje cell layer. Scale bars, 500 µm (a, c), 100 µm (b, d–f, h, i) and 200 µm (g, j).

a b

c d

e f g

h i j

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

biosynthetic enzyme glutamic aciddecarboxylase). GAD65 is localizedprimarily in presynaptic boutons34 andthe onset of GAD65 expression in thecerebellum seems to follow synaptoge-nesis35.

To rule out the possibility that aphenotype could result from a gener-al delay in the development of themutant mice, we analyzed the adultcerebellum (P50–P80). Comparedwith wild-type and fBZ/fBZ controllittermates, Wnt1Cre;fBZ/fBZ condi-tional-mutant mice showed marked-ly reduced immunoreactivity toGAD65 in both the granule-cell layerand the molecular layer (Fig. 5a–c,g–i). Because the organization ofGAD65-containing boutons in thegranule-cell layer made the accuratecounting of terminals impractical, wequantified the total labeling found ina defined area (8,600 µm2).

Compared with the wild-type con-trol, a marked reduction (∼ 80%) inthe area labeled by GAD65 immunos-taining was observed in the granule-cell layer of trkB conditional mutants(Fig. 6a). For quantitative analyses,the intensities of the immunofluores-cent signal for GAD65 were measuredand expressed as gray levels. In thegranule-cell layer, there was a morethan twofold reduction in the num-ber of GAD65-containing particleswith high gray levels (>100 units),which suggested that the expressionof GAD65 in boutons was alsoreduced (Fig. 6d). Quantification ofthis phenotype in the molecular layer,where individual boutons can beidentified, showed that, comparedwith in control mice (wild-type andfBZ/fBZ), there was a roughly 75%reduction in the number of GAD65-

containing boutons in mutant mice (Fig. 6b). The intensity ofGAD65 expression in these boutons was also reduced (Fig. 6e),and the loss of the large-sized boutons (>1.5 µm) was almostcomplete (Fig. 6g–i).

To examine whether TrkB function is required only for main-tenance and not for initially establishing GABAergic function inthe cerebellum, we examined expression of GAD65 at P21—astage when the process of synaptogenesis in the cerebellum isconcluding26. At P21, expression of GAD65 was reduced marked-ly in both the granular and the molecular layer of the mutantcerebellum (not shown).

To confirm a reduction of GABAergic innervation in themutant cerebellum, we analyzed the expression of the high-affin-ity plasma membrane GABA transporter GAT-1 in the adult cere-bellum36. Expression of GAT-1 in the cerebellum is restricted toaxon terminals of GABAergic cells (basket, stellate and Golgi cells),and its onset of expression is simultaneous with the establishmentof GABAergic synapses36. In agreement with the expression of

228 nature neuroscience • volume 5 no 3 • march 2002

Our observations also indicated that the density and distribu-tion of Purkinje cells were normal in adult mutant mice (Fig. 3gand j). In agreement with a previous study of BDNF knockout ani-mals22, a gross analysis of the Purkinje cell dendritic tree usingantibodies specific for calbindin showed that the primary and sec-ondary dendrites were well differentiated (Fig. 4i and j).

Decreased GAD65 and GAT-1 in trkB mutantsBecause BDNF mediates the maturation and strength of GABAer-gic synapses in the cerebellum and visual cortex14,15, we extendedour analyses to GABAergic cells and fibers. A preliminary analy-sis of GABAergic innervation in the cerebellum of trkB conditionalmutants suggested that anterior lobules (such as lobule IV) weremore perturbed than posterior lobules (such as lobule VIII), con-sistent with the suggestion that NT-3 signaling through TrkC maycompensate for the lack of TrkB signaling in more posterior lob-ules33. Concentrating on an anterior lobule (lobule IV), we firstanalyzed the expression of GAD65 (an isoform of the GABA

articles

Table 1. Analysis in the cerebella of trkB conditional-mutant mice

Measurement Genotype n (mean ± s.e.m.) Percentage of wild type

Molecular layer area (IV) Wild type 3 0.154 ± 0.009 mm2 –fBZ/fBZ 3 0.161 ± 0.008 mm2 –Wnt1Cre;fBZ/fBZ 3 0.117 ± 0.004 mm2*, **a 76*, **a

Internal granule-cell layer area (IV) Wild type 3 0.110 ± 0.005 mm2 –fBZ/fBZ 3 0.121 ± 0.003 mm2 –Wnt1Cre;fBZ/fBZ 3 0.092 ± 0.005 mm2 *, **a 84*, **a

No. of granule cells per mm2 (IV) Wild type 3 1,710.62 ± 41.52 –fBZ/fBZ 3 1,711.10 ± 50.71 –Wnt1Cre;fBZ/fBZ 3 1,918.04 ± 74.01*, **b 112 *, **b

No. of Golgi cells per mm2 (IV) Wild type 3 21.60 ± 3.42 –fBZ/fBZ 3 23.27 ± 4.09 –Wnt1Cre;fBZ/fBZ 3 27.25 ± 7.57 126

No. of interneurons in molecular layer per mm2 (IV) Wild type 3 126.37 ± 4.84 –

fBZ/fBZ 3 123.55 ± 10.26 –Wnt1Cre;fBZ/fBZ 3 122.03 ± 6.64 –

No. of Purkinje cells per mm2 (IV) Wild type 3 7.44 ± 0.40 –fBZ/fBZ 3 8.44 ± 0.22 –Wnt1Cre;fBZ/fBZ 3 7.55 ± 0.48 –

Granule-cell size (diameter, IV, em) Wild type 3 7.6 ± 0.29 µm –fBZ/fBZ 3 7.4 ± 0.21 µm –Wnt1Cre;fBZ/fBZ 3 7.9 ± 0.18 µm –

Parallel fiber diameter (IV, em) Wild type 3 0.223 ± 0.006 µm –fBZ/fBZ 3 0.218 ± 0.007 µm –Wnt1Cre;fBZ/fBZ 3 0.232 ± 0.002 µm –

No. of parallel fiber profiles per 100 µm2 (IV, em) Wild type 3 233.15 ± 30.34 –

fBZ/fBZ 3 318.30 ± 42.235 –Wnt1Cre;fBZ/fBZ 3 287.410 ± 22.474 –

Length of the symmetric synaptic specializations (IV, em) Wild type 3 0.22 ± 0.012 µm –

fBZ/fBZ 3 0.22 ± 0.02 µm –Wnt1Cre;fBZ/fBZ 3 0.20 ± 0.034 µm –

Asterisks denote significant differences between the genotype and wild-type control (*) or fBZ/fBZ control (**)mice. em, electron microscopy; IV, lobule IV.aOne-way ANOVA, p < 0.05bOne-way ANOVA, p < 0.01.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

GAD65, immunoreactivity to GAT-1 was reduced markedly inthe molecular and, more prominently, in the granule-cell layer inlobule IV of conditional-mutant mice (Fig. 5d–f and j–l).

Because the inhibitory input from GABAergic interneuronswas impaired markedly in the mutant mice, we examined whetherGABAergic projections from the cerebellum were also disturbed.We focused our analysis on the fastigial nucleus, which is the maintarget of the GABAergic axons of Purkinje cells located in anteri-or lobules of the cerebellum. A roughly 70% reduction in the areaof GAD65 immunostaining was found in the fastigial nucleus ofWnt1Cre;fBZ/fBZ mice (Figs. 5m–o and 6c). A decrease in theintensity of the remaining GAD65 terminals was also observed inthe fastigial nucleus of the mutant mice (Fig. 6f).

Reductions in symmetric synapses in trkB mutantsThe reduction of GABAergic markers in cerebella lacking TrkBsuggested that TrkB is required for establishing inhibitory neu-rotransmission. To examine inhibitory synapses more directly,we analyzed the ultrastructure of GABAergic terminals in thegranule-cell layer, where the exceptionally regular synaptic orga-nization of the glomerulus made it possible to compare theorganization of the inhibitory input in the presence and absenceof TrkB (Fig. 7a). The organization of glomeruli seemed to besimilar in all three genotypes examined (Fig. 7b–d); however,quantification of inhibitory (symmetric) synapses indicated aroughly 70% decrease in the number of inhibitory synapses perglomerulus in Wnt1cre;fBZ/fBZ mutant mice (Fig. 7e and i).The number of inhibitory synapses per Golgi neuron terminalwas also reduced by about 40% in these mice (Fig. 7f–h, j).Notably, the number of inhibitory synapses per terminal wasreduced by 20% in fBZ/fBZ mice (Fig. 7f, g, j). As comparedwith wild-type controls, there was no significant difference inthe length of symmetric synaptic densities in fBZ/fBZ orWnt1cre;fBZ/fBZ mice (Table 1).

In vitro studies have suggested that BDNF may act as a tar-get-derived trophic factor for basilar pontine mossy fibers25.Because Cre was also expressed in the pontine nucleus progeni-tors (Fig. 1a), we investigated whether the density of glomeruliwas affected. A significant 27% reduction in glomeruli densitywas found in Wnt1Cre;fBZ/fBZ mutant mice as compared withwild-type control mice (Fig. 7k). We also quantified the numberof excitatory (asymmetric) synapses in the remaining glomeruli(mossy fiber–granule cell synapses). No significant differencesin the number of excitatory synapses were found between geno-types, which suggested that the mossy fiber–granule cell dendritesynapses were normal (Fig. 7l). Loss of GABAergic inputs andglomeruli might explain the 16% reduction found in the cross-

sectional area of the granule-cell layer. Consistent with theanatomical phenotypes, Wnt1Cre;fBZ/fBZ mice were ataxic andhad severe deficits in motor coordination (SupplementaryFig. 2, available on the supplementary information page of NatureNeuroscience online).

DISCUSSIONWe have examined the role of TrkB in development of the cere-bellum by using mice that lack TrkB expression in this structure.In contrast to conventional knockout mice, the Wnt1Cre;fBZ/fBZconditional mutants developed into healthy adults. Absence ofTrkB did not reduce the survival of any cerebellar cell popula-tion but did reduce the volumes of the granule-cell and molec-ular layers. Notably, there were significant reductions inexpression of the GABA biosynthetic enzyme GAD65 and theGABA transporter GAT-1 within the terminals of GABAergicinterneurons and Purkinje cells. Interneurons also formedreduced numbers of nerve terminals and, in the granule-cell layer,fewer inhibitory synapses. The decrease in the number ofinhibitory synapses was independent of axon terminal develop-ment. By contrast, the number of excitatory synapses formed onthese same glomeruli by mossy fiber afferents appeared normal.

articles

nature neuroscience • volume 5 no 3 • march 2002 229

Fig. 4. Granular and Purkinje cell morphologies in trkB conditional-mutant mice at P50–P80. (a, c) Golgi-impregnated sections showingnormal body size and typical claw-like telodendria, which are character-istic of differentiated granule cells, in wild-type and Wnt1Cre;fBZ/fBZmice. (b, d) Electron microphotographs showing the normal structurein granule cells of wild-type and Wnt1Cre;fBZ/fBZ mice. Note thatnuclear chromatin is distributed similarly in both genotypes. (e–h) Immunohistochemistry of the α6 subunit of the GABAA receptorin lobule IV of wild-type and Wnt1Cre;fBZ/fBZ mice. (f, h) Higher-magni-fication images of the internal granule-cell layer shown in (e) and (g),respectively. (i, j) Purkinje cells at high magnification stained for cal-bindin. Note the well-differentiated primary dendrites present in bothwild-type and Wnt1Cre;fBZ/fBZ mice (arrows). igl, internal granule-celllayer; ml, molecular layer. Scale bars, 20 µm (a, c, f, h, i, j), 2 µm (b, d)and 200 µm (e, g).

a b

c d

e f

g h

i j

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

elevated apoptosis observed in the BDNF mutantsmay reflect their poor health. Although our resultsindicate that TrkB signaling is not necessary forgranule-cell survival, TrkB signaling may promotesurvival in the absence of other factors. Five- to six-fold increases in the apoptosis of granule cells areobserved at P14 in mice with compound deficien-cies in TrkB and TrkC, but not in animals that aredeficient in only one receptor23.

Although data suggest that BDNF-mediated TrkBsignaling has a direct effect on the differentiation ofPurkinje cell primary dendrites in conventionalmutants at P8 and P12 (refs. 4, 23), these dendritesappeared normal in adult Wnt1Cre;fBZ/fBZ animals.Because normal primary and secondary dendritesare observed in P17 and older BDNF mutants22, wethink that delays in development might explain theobservations in P8–P14 conventional mutants.

In conditionally targeted mice, the molecularlayer area was reduced by 24% in lobule IV despitethe apparently normal development of granule andPurkinje cells. It is possible that the parallel fibersmay have been shorter in the TrkB-deficient cere-bellum. Our analysis measured only their densityand diameter and not their length. In addition,although the Purkinje cell dendrites appeared nor-mal in our mutant mice, there might have beensubtle effects on the higher-order branches of thesedendrites that contributed to the reduction.

It has been suggested that basket, stellate andgranule cells are essential to the proper formation of Purkinjedendritic arbors and their planar orientations40,41. We haveshown here that the absence of TrkB has marked effects on Golgi,basket and stellate interneurons. As a result, the inhibitory cir-cuit mediated by these interneurons is almost certainly impaired.Purkinje and granule cell–specific Cre lines are needed to exam-ine the potential subtle roles of TrkB on Purkinje dendritic treesor granule-cell axons independently of this circuit.

Development of inhibitory synapsesThe absence of TrkB resulted in significant deficits in the termi-nals of interneurons that expressed GABA within the cerebellarcortex and in the terminals of Purkinje cells in the deep cerebel-lar nuclei. Fewer inhibitory terminals were seen and these hadlower amounts of GAD65. In addition, fewer inhibitory synaps-es were formed by these terminals. Consistent with this, treat-ment with BDNF enhances inhibitory synaptic transmission in vitro9–13, and overexpression of BDNF accelerates the matu-ration of GABAergic inputs in both the visual cortex and cere-

Fig. 5. Loss of GABAergic markers in the cerebellum oftrkB conditional mutant at P50–P80. (a–f) Large clustersof GAD65-containing (a–c) and GAT-1-containing (d–f)boutons derived from Golgi interneurons in the granule-cell layer of lobule IV in wild-type (a, d), fBZ/fBZ (b, e)and Wnt1;fBZ/fBZ (c, f) mice. (g–l) Homogeneous neu-ropil of individual GAD65-positive (g–i) and GAT-1-pos-itive (j–l) boutons derived from basket and stellateinterneurons in the molecular layer of lobule IV in wild-type (g, j), fBZ/fBZ (h, k) and Wnt1;fBZ/fBZ (i, l) mice.(m–o) Localization of GAD65 in the fastigial nucleus ofwild-type (m), fBZ/fBZ (n) and Wnt1;fBZ/fBZ (o) mice.Scale bar, 25 µm (a–o).

230 nature neuroscience • volume 5 no 3 • march 2002

Development of cerebellar morphologyIn targeted animals, the absence of TrkB did not significantly alterthe overall morphology, layering or foliation of the adult cerebel-lum. Deficits in the foliation pattern have been observed previouslyin BDNF mutants at P14 (ref. 4). It seems possible that this pheno-type is not a direct result of the BDNF mutation because postnatalBDNF mutants fail to thrive and are stunted in development owingto poor breathing and cardiac performance37,38.

Development of granule and Purkinje cellsThe absence of TrkB did not reduce the number or normal dif-ferentiation of granule cells in targeted animals. By contrast, pre-vious studies have suggested that BDNF signaling to TrkBregulates the survival of granule cells at P8 (ref. 4). It is possiblethat compensatory mechanisms might result in a normal finaldensity of granule cells in adult animals, despite elevated apop-tosis during development. Alternatively, because malnutritionresults in a decreased number of cerebellar granule cells39, the

articles

a b c

d e f

g h i

j k l

m on

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

reduced markedly in both the molecular and granule-cell lay-ers. In the molecular layer, only 25–30% of the normal num-ber of GABAergic boutons were detected; in the granule-celllayer, there was a fivefold reduction in the area occupied byGABAergic boutons. Analysis by electron microscopy also sug-gested that there were fewer GABAergic boutons in the gran-ule-cell layer. We measured 70% fewer inhibitory synapticspecializations per glomerulus, but only 40% fewer inhibitory

bellum in vivo14,15. A recent study suggests that TrkB-mediatedsignaling may promote the formation of excitatory synapses insome brain regions. In this study, overexpression of BDNF wasshown to increase both the complexity of retinal ganglion cellaxonal arbors and the number of GFP-synaptobrevin clusters inthese arbors in the optic tectum of Xenopus tadpoles42.

In adult Wnt1Cre;fBZ/fBZ mice, the number of terminalsand the total area labeled by antibodies specific for GAD65 were

Fig. 6. Quantitative analysis of the loss ofGABAergic markers in lobule IV and in thefastigial nucleus of trkB conditional-mutantmice at P50–P80. (a) Quantification of thetotal area labeled with GAD65-specific antibodies in the granule-cell layers of wild-type, fBZ/fBZ and Wnt1;fBZ/fBZ mice. (b) Quantification of the number of GAD65-labeled boutons in the molecular layer ofwild-type, fBZ/fBZ and Wnt1;fBZ/fBZ mice.(c) Quantification of the total area labeledwith GAD65-specific antibodies in the fasti-gial nucleus of wild-type, fBZ/fBZ andWnt1;fBZ/fBZ mice. (d–f) Histograms com-paring the intensities of GAD65 immunos-taining (indicated in gray levels) in thegranule-cell layer (d), molecular layer (e) orfastigial nucleus (f) of wild-type, fBZ/fBZ andWnt1;fBZ/fBZ mice. (g) Quantification ofsize of GAD65-containing boutons in themolecular layer of wild-type, fBZ/fBZ andWnt1;fBZ/fBZ mice. (h, i) Sample showingthe diverse size of boutons in wild-type (h)and Wnt1;fBZ/fBZ (i) mice. Asterisk denotesa significant difference from the wild-type control (n = 3, p < 0.03 (a) and p < 0.04 (c) by one-way analysis of variance; ANOVA). Data shown arethe mean ± s.e.m. (f, g) Immunohistochemistry of GAD65 in the molecular layer of wild-type (f) and Wnt1;fBZ/fBZ (g) mice. Open triangles indicatesmall boutons, arrows indicate large boutons. Scale bar, 20 µm (f, g).

articles

nature neuroscience • volume 5 no 3 • march 2002 231

Fig. 7. Reduction in the number of symmetric synapses inthe granule-cell layer of trkB conditional-mutant mice atP50–P80. (a) Diagram of the normal organization of theglomerulus. The large mossy fiber (mf) constitutes the cen-ter, which is surrounded by dendrites (gcd) of granular cells(gc) and, more externally, by GABAergic Golgi axon vari-cosities (gav), which synapse onto the granule-cell den-drites. (b–d) Fine structure of three representativeglomeruli in wild-type, fBZ/fBZ and Wnt1Cre;fBZ/fBZ mice.Dotted lines indicate the contour of the mossy fiber. (e, g,h) Higher-magnification images corresponding to the boxedregion in (b–d), respectively, showing symmetric synapsesbetween Golgi axons and granule-cell dendrites (arrowheads). (f) Fine structure of a representativeglomerulus in wild-type mice, showing that the Golgi axonvaricosity forms several synapses in the same terminal. (i, j) Quantification of the number of symmetric synapsesper glomerulus (i) or per Golgi axon varicosity (j) in wild-type, fBZ/fBZ and Wnt1Cre;fBZ/fBZ mice. (k) Quantificationof the number of glomeruli in 1,000 µm2. (l) Quantificationof the number of asymmetric synapses per glomerulus.Asterisks denote significant differences between the geno-type where the asterisk is located and wild-type (*) orfBZ/fBZ (**) mice (n = 3, p < 0.01 (i) and p < 0.05 (j) by one-way ANOVA). as, asymmetric synapses; ss, symmetricsynapses. Data shown are the mean ± s.e.m. Scale bars, 1 µm (b–d) and 0.5 µm (e–h).

a b c d

e f g h

i j k l

a b c

d e f

g

h i

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

232 nature neuroscience • volume 5 no 3 • march 2002

synaptic specializations per GABAergic bouton. This discrep-ancy suggests that there was a twofold reduction in the num-ber of boutons contacting each glomerulus.

Our findings show that TrkB signaling has a direct role in con-trolling the number of GABAergic synapses in healthy animals.First, loss of synaptic contacts occurred in the absence of pre- orpostsynaptic neuronal loss. Second, not only was the total num-ber of synapses reduced (which could directly reflect a reductionin the number of axonal branches and terminals), but there wasalso a reduction in the number of synaptic specializations perterminal. This indicates that TrkB may be involved directly incontrolling the number of GABAergic synaptic specializations.

In vitro studies suggest that TrkB ligands promote survivaland neurite outgrowth by cultured basilar pontine nuclei—themain source of cerebellar mossy fibers25. Consistent with this,we observed a modest 27% reduction in the density of glomeruliin lobule IV of trkB conditional mutants. The remaining mossyfiber boutons looked morphologically normal and had normalnumbers of excitatory synapses. The reduction in mossy fiberbouton number was clearly insufficient to explain the fivefoldreduction in the area of GABAergic boutons, as well as the four-to fivefold reduction of inhibitory synapses per glomeruli in thegranule-cell layer.

In summary, our results show that TrkB signaling in vivo isnecessary for several aspects of GABAergic neurotransmission inthe cerebellum. TrkB seems to be functioning at two differentlevels: first, in maintaining the synthesis and uptake of GABA;and second, in regulating the number of terminals per axon andthe morphological specializations in synaptic contacts. Deficitsto these processes caused by a lack of TrkB almost certainly dis-rupt cerebellar function.

METHODSTransgenic mouse strains. Mice lacking TrkB in the cerebellum were pro-duced by breeding mice carrying a loxP-flanked trkB allele (fBZ)7 withtransgenic mice in which Wnt1 regulatory elements drive Cre recombi-nase expression (Wnt1Cre)28. In most experiments, trkB conditionalmutants were compared with fBZ/fBZ and wild-type control littermates.

To analyze the distribution of recombination, we generated mice het-erozygous for both the Wnt1Cre transgene and the R26R reporter43. Ani-mal procedures were approved by the University of California SanFrancisco Committee on Animal Research.

Immunoblot analysis and immunohistochemistry. Protein extracts wereprepared from the cerebellum of wild-type, fBZ/fBZ and Wnt1;fBZ/fBZmice. We used 10 µg of protein per lane for SDS-PAGE and immunoblots.Antibodies against the TrkB extracellular domain32 and against β-tubu-lin (Sigma, St. Louis, Missouri; 1:400) were used sequentially on the sameblot.

For immunohistochemistry, animals were deeply anesthetized andperfused with PBS (pH 7.4), followed by 4% paraformaldehyde in PBSand a series of sucrose-PBS solutions (15–30%). We cut 40-µm serialsagittal sections in a sliding microtome. Littermates were processed inparallel in each experimental group (n = 3).

For double immunohistochemistry, free-floating sections were prein-cubated in 5% bovine serum albumen (BSA), 0.3% Triton X-100 dilutedin PBS for 1 h at room temperature, and then incubated for 36 h at 4 °Cwith primary antisera diluted in 1% BSA, 0.3% Triton X-100 in PBS.

Cocktails included monoclonal (Promega, Madison, Wisconsin;1:1,000) or polyclonal (5 prime–3 prime, Boulder, Colorado; 1:5,000)antibodies against β-galactosidase and one of the following antisera:mouse calbindin-specific antibody (Sigma; 1:1,000), rabbit calbindin-specific antibody (Swant, Bellinzona, Switzerland, 1:1,000), mouse par-valbumin-specific antibody (Sigma; 1:1,000), rabbit parvalbumin-specificantibody (Swant; 1:1,000), rabbit GABA-specific antibody (Sigma;1:2,000), mouse GFAP-specific antibody (Chemicon, Temecula, Califor-

nia; 1:1,000), mouse phosphorylated neurofilament–specific antibody(NF200, Sigma; 1:2,000), mouse GAD65-specific antibody (Roche, Indi-anapolis, Indiana; 1:500), rabbit GAT-1–specific antibody (N. Brecha,Univ. California Los Angeles; 1:1,000) and an affinity-purified rabbitantibody against the α6 subunit GABAA receptor (A. Stephenson, Uni-versity of London; 0.5 µg/ml).

Both monoclonal and polyclonal antibodies against β-galactosidaselabeled the same cell types. We then rinsed sections and incubated themin the appropriate secondary antibodies: goat antibodies against eithermouse or rabbit Alexa 488 and Texas Red (Molecular Probes) diluted inthe same solution as the primary antibodies.

Quantification of fluorescence intensity and cell counts. We exam-ined GAD65-immunoreactive boutons with a Bio-Rad MRC 1000 con-focal microscope (100×, n.a. 1.30 Plan-Neofluor). The same fields wereselected for all genotypes. Two images spaced 1-µm apart were usedto quantify the fluorescence intensity from each sample (n = 3). Allimages were taken by Kalman averaging (four times) by excitation at568 nm (laser power 3%, iris 2.4). To obtain binary images, we alsoused the background threshold of the wild-type sections for the cor-responding sections from fBZ/fBZ and Wnt1Cre;fBZ/fBZ littermates.For particle analysis, the threshold was set at 13 pixels, which was theestimated minimum size of a GAD65 bouton in the Wnt1Cre;fBZ/fBZmutant mice. We measured the fluorescent intensities (gray levels) andthe immunolabeled regions in an area of 8,600 µm2, using NIH Imagesoftware as described44.

Quantification of the cell counts is described in the SupplementaryMethods (available on the supplementary information page of NatureNeuroscience online).

Golgi staining and in situ hybridization. We stained brains by the rapidGolgi method45. In situ hybridization was done with digoxigenin-labeledriboprobes on 100-µm sections as described46. We used antisense ribo-probes to both the extracellular and tyrosine kinase domains of trkB.Specificity of the probes was established by using sense probes as controls.

Electron microscopy. Mice were perfused with 0.9% NaCl, followed by2.5% glutaraldehyde and 1% paraformaldehyde in 0.1 M sodium cacody-late buffer, pH 7.4, for 20–30 min. The heads were removed and storedovernight at 4 °C, and then each cerebellar lobule was dissected out andpostfixed for 1 h in 2% osmium tetroxide, 0.1 M sodium cacodylate, pH 7.4. The lobules were then dehydrated using a series of ethanol dilu-tions and flat-embedded in an Epon-Araldite mixture. We used semithinsections stained with toluidine blue to identify and trim the medial sagit-tal plane of lobule IV. Ultrathin sections were cut and stained with uranylacetate and lead citrate. Quantitative analyses were done blind to geno-type. We took 15 electron micrographs from the glomeruli at a final mag-nification of 35,000× for each genotype in each experiment (n = 3).Inhibitory Golgi cell synapses were identified by the criteria given in ref. 47. The total numbers of symmetric synapses per glomerulus and perGolgi axon varicosity were counted. We measured the diameter of eachgranule cell in its longest axis (10 cells per genotype per experiment; n = 3).We used 10 electron micrographs of the molecular layer at a final magni-fication of 35,000× for each genotype to analyze parallel fibers in eachexperiment (n = 3). The diameter of the fiber was measured in its longestaxis (10 fiber profiles per micrograph). For counting numbers of glomeruliand excitatory synaptic specializations (asymmetric synapses), we used 20 electron micrographs from the granule-cell layer at a final magnifica-tion of 7,500× for each genotype in each experiment (n = 3).

Note: See supplementary information on the Nature Neuroscience website

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

AcknowledgementsWe thank A. P. McMahon and P. Soriano for the Wnt1Cre and R26R transgenic

mice; O. Marín, M. Stryker, S. Bamji, L. Elia, T. Elul, U. Fünfschilling and J.

Zhu for comments on the manuscript; S. Huling, and I. Hsie for assistance with

articles©

2002

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co

m

electron microscopy; N. Brecha for the antibody against GAT-1; and A.

Stephenson for the antibody against the α6 subunit of the GABAA receptor. This

work was supported by a grant from the USPH and by the HHMI. B.R. was

supported by a postdoctoral fellowship from the Ministerio de Educación, Spain.

L.F.R. is an Investigator of the HHMI.

Competing interests statementThe authors declare that they have no competing financial interests.

RECEIVED 8 NOVEMBER 2001; ACCEPTED 14 JANUARY 2002

1. Huang, E. J. & Reichardt, L. F. Neurotrophins: roles in neuronal developmentand function. Annu. Rev. Neurosci. 24, 677–736 (2001).

2. 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).

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

4. Schwartz, P. M., Borghesani, P. R., Levy, R. L., Pomeroy, S. L. & Segal, R. A.Abnormal cerebellar development and foliation in BDNF–/– mice reveals arole for neurotrophins in CNS patterning. Neuron 19, 269–281 (1997).

5. Martínez, A. et al. TrkB and TrkC signaling are required for maturation andsynaptogenesis of hippocampal connections. J. Neurosci. 18, 7336–7350(1998).

6. McAllister, A. K., Katz, L. C. & Lo, D. C. Neurotrophins and synapticplasticity. Annu. Rev. Neurosci. 22, 295–318 (1999).

7. Xu, B. et al. Cortical degeneration in the absence of neurotrophin signaling:dendritic retraction and neuronal loss after removal of the receptor TrkB.Neuron 26, 233–245 (2000).

8. Xu, B. et al. The role of brain-derived neurotrophic factor receptors in themature hippocampus: modulation of long-term potentiation through apresynaptic mechanism involving TrkB. J. Neurosci. 20, 6888–6897 (2000).

9. Mizuno, K., Carnahan, J. & Nawa, H. Brain-derived neurotrophic factorpromotes differentiation of striatal gabaergic neurons. Dev. Biol. 165,243–256 (1994).

10. Widmer, H. R. & Hefti, F. Stimulation of GABAergic neuron differentiationby NT-4/5 in cultures of rat cerebral cortex. Dev. Brain Res. 80, 279–284(1994).

11. Vicario-Abejón, C., Collín, C., McKay, R. D. G. & Segal, M. Neurotrophinsinduce formation of functional excitatory and inhibitory synapses betweencultured hippocampal neurons. J. Neurosci. 18, 7256–7271 (1998).

12. Seil, F. J. & Drake-Baumann, R. TrkB receptor ligands promote activity-dependent inhibitory synaptogenesis. J. Neurosci. 20, 5367–5373 (2000).

13. Marty, S., Wehrle, R. & Sotelo, C. Neuronal activity and brain-derivedneurotrophic factor regulate the density of inhibitory synapses inorganotypic slice cultures of postnatal hippocampus. J. Neurosci. 20,8087–8095 (2000).

14. Bao, S. W., Chen, L., Qiao, X. X. & Thompson, R. F. Transgenic brain-derivedneurotrophic factor modulates a developing cerebellar inhibitory synapse.Learn. Memory 6, 276–283 (1999).

15. Huang, Z. J. et al. BDNF regulates the maturation of inhibition and thecritical period of plasticity in mouse visual cortex. Cell 98, 739–755 (1999).

16. Lindholm, D., Hamnér, S. & Zirrgiebel, U. Neurotrophins and cerebellardevelopment. Persp. Dev. Neurol. 5, 83–94 (1997).

17. Rocamora, N., Garcialadona, F. J., Palacios, J. M. & Mengod, G. Differentialexpression of brain-derived neurotrophic factor, Neurotrophin-3, and low-affinity nerve growth factor receptor during the postnatal development of therat cerebellar system. Mol. Brain Res. 17, 1–8 (1993).

18. Klein, R. et al. Targeted disruption of the TrkB neurotrophin receptor generesults in nervous system lesions and neonatal death. Cell 75, 113–122(1993).

19. Yan, Q. et al. Immunocytochemical localization of TrkB in the centralnervous system of the adult rat. J. Comp. Neurol. 378, 135–157 (1997).

20. Segal, R. A., Takahashi, H. & McKay, R. D. G. Changes in neurotrophinresponsiveness during the development of cerebellar granule neurons.Neuron 9, 1041–1052 (1992).

21. Gao, W. Q., Zheng, J. L. & Karihaloo, M. Neurotrophin-4/5 (NT-4/5) and

brain-derived neurotrophic factor (BDNF) act at later stages of cerebellargranule cell differentiation. J. Neurosci. 15, 2656–2667 (1995).

22. Jones, K. R., Fariñas, I., Backus, C. & Reichardt, L. F. Targeted disruption ofthe BDNF gene perturbs brain and sensory neuron development but notmotor neuron development. Cell 76, 989–999 (1994).

23. Minichiello, L. & Klein, R. TrkB and TrkC neurotrophin receptors cooperatein promoting survival of hippocampal and cerebellar granule neurons. GenesDev. 10, 2849–2858 (1996).

24. Shimada, A., Mason, C. A. & Morrison, M. E. TrkB signaling modulates spinedensity and morphology independent of dendrite structure in culturedneonatal Purkinje cells. J. Neurosci. 18, 8559–8570 (1998).

25. Rabacchi, S. A. et al. BDNF and NT4/5 promote survival and neuriteoutgrowth of pontocerebellar mossy fiber neurons. J. Neurobiol. 40, 254–269(1999).

26. Altman, J. & Bayer, S. A. Development of the Cerebellar System. In Relation toits Evolution, Structure, and Functions (CRC, Boca Raton, Florida, 1997).

27. Minichiello, L. et al. Essential role for TrkB receptors in hippocampus-mediated learning. Neuron 24, 401–414 (1999).

28. Danielian, P. S., Muccino, D., Rowitch, D. H., Michael, S. K. & McMahon, A. P.Modification of gene activity in mouse embryos in utero by a tamoxifen-inducible form of Cre recombinase. Curr. Biol. 8, 1323–1326 (1998).

29. Chai, Y. et al. Fate of the mammalian cranial neural crest during tooth andmandibular morphogenesis. Development 127, 1671–1679 (2000).

30. Rodriguez, C. I. & Dymecki, S. M. Origin of the precerebellar system. Neuron27, 475–486 (2000).

31. Callahan, C. A. & Thomas, J. B. Tau–β-galactosidase, an axon-targeted fusionprotein. Proc. Natl. Acad. Sci. USA 91, 5972–5976 (1994).

32. Huang, E. J. et al. Expression of Trk receptors in the developing mousetrigeminal ganglion: in vivo evidence for NT-3 activation of TrkA and TrkB inaddition to TrkC. Development 126, 2191–2203 (1999).

33. Tojo, H. et al. Neurotrophin-3 is expressed in the posterior lobe of mousecerebellum, but does not affect the cerebellar development. Neurosci. Lett.192, 169–172 (1995).

34. Esclapez, M., Tillakaratne, N. J. K., Kaufman, D. L., Tobin, A. J. & Houser, C. R.Comparative localization of two forms of glutamic acid decarboxylase andtheir mRNAs in rat brain supports the concept of functional differencesbetween the forms. J. Neurosci. 14, 1834–1855 (1994).

35. Greif, K. F., Erlander, M. G., Tillakaratne, N. J. K. & Tobin, A. J. Postnatalexpression of glutamate decarboxylases in developing rat cerebellum.Neurochem. Res. 16, 235–242 (1991).

36. Morara, S., Brecha, N. C., Marcotti, W., Provini, L. & Rosina, A. Neuronaland glial localization of the GABA transporter GAT-1 in the cerebellar cortex.Neuroreport 7, 2993–2996 (1996).

37. Balkowiec, A. & Katz, D. M. Brain-derived neurotrophic factor is required fornormal development of the central respiratory rhythm in mice. J. Physiol.(Lond.) 510, 527–533 (1998).

38. Donovan, M. J. et al. Brain derived neurotrophic factor is an endothelial cellsurvival factor required for intramyocardial vessel stabilization. Development127, 4531–4540 (2000).

39. Ito, M. The Cerebellum and Neural Control (Raven, New York, 1984).40. Altman, J. Experimental reorganization of the cerebellar cortex. VI. Effects of

X-irradiation schedules that allow or prevent cell acquisition after basket cellsare formed. J. Comp. Neurol. 165, 49–64 (1976).

41. Altman, J. Experimental reorganization of the cerebellar cortex. VII. Effectsof late X-irradiation schedules that interfere with cell acquisition after stellatecells are formed. J. Comp. Neurol. 165, 65–76 (1976).

42. Alsina, B., Vu, T. & Cohen-Cory, S. Visualizing synapse formation inarborizing optic axons in vivo: dynamics and modulation by BDNF. Nat.Neurosci. 4, 1093–1101 (2001).

43. Soriano, P. Generalized lacZ expression with the ROSA26 Cre reporter strain.Nat. Genet. 21, 70–71 (1999).

44. Fukuda, T., Aika, Y., Heizmann, C. W. & Kosaka, T. GABAergic axonterminals at perisomatic and dendritic inhibitory sites show differentimmunoreactivities against two GAD isoforms, GAD67 and GAD65, in themouse hippocampus: a digitized quantitative analysis. J. Comp. Neurol. 399,424–426 (1998).

45. Valverde, F. The rapid Golgi technique for staining CNS neurons. Lightmicroscopy. Neurosci. Protocols 1, 1–9 (1993).

46. Nieto, M. A., Patel, K. & Wilkinson, D. G. In situ hybridisation analysis ofchick embryos in whole mount and in tissue sections. Methods Cell Biol. 51,219–235 (1996).

47. Palay, S. L. & Chan-Palay, V. Cerebellar Cortex: Cytology and Organization(Springer, New York, 1974).

articles

nature neuroscience • volume 5 no 3 • march 2002 233

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

234 nature neuroscience • volume 5 no 3 • march 2002

articles

Temporally precise gene expression is crucial to normal devel-opment and physiology; in contrast, the temporal dysregulationof gene expression is a principal factor in cellular malfunctionand disease. One of the best-characterized organizers of an inter-nal temporal order that partitions incompatible metabolic func-tions across the 24-hour cycle in mammals is the central clock inthe hypothalamic suprachiasmatic nucleus (SCN)1. Rhythm gen-eration in clock cells of the SCN is regulated by interlocked tran-scriptional/translational feedback loops that comprise theproducts of canonical clock genes, of which the cardinal exam-ple in humans is PERIOD (PER)2–6.

Circadian expression of canonical clock genes is not limited tocells of the SCN but is found in many peripheral tissues, albeit witha delayed phase relative to expression in the SCN2,4,6. More signif-icantly, the peripheral rhythms of clock-gene expression dampen in vitro after a few cycles7–9, emphasizing their dependence on neu-ronal and/or neuroendocrine output from the SCN. Maintainingsynchrony between the SCN and clock-controlled genes in periph-eral tissues is essential to normal physiology; their desynchronyunderlies pathologies associated with shift work and jet-lag10.

Both neural and endocrine pathways can sustain circadiangene expression in peripheral tissues. For example, rhythms ofthe rodent clock gene Period (Per) in the liver can be regulatedby both glucocorticoids7 and cycles of restricted feeding11. In therodent pineal gland, circadian Per1 expression is sustained bynoradrenergic activation of the cyclic AMP signaling pathway12,13.The same pathway drives the rhythm in the synthesis of pineal

Rhythmic gene expression in pituitarydepends on heterologous sensitizationby the neurohormone melatonin

Charlotte von Gall1, Martine L. Garabette2, Christian A. Kell1, Sascha Frenzel1, Faramarz Dehghani1, Petra-Maria Schumm-Draeger3, David R. Weaver4, Horst-Werner Korf1,Michael H. Hastings5 and Jörg H. Stehle1

1 Institute of Anatomy II and 3Institute of Internal Medicine, Johann Wolfgang Goethe-University, Theodor-Stern-Kai 7, D-60590 Frankfurt, Germany2 Department of Anatomy, University of Cambridge, Downing Street, Cambridge CB2 3DY, UK4 Department of Neurobiology, University of Massachusetts Medical School, 55 Lake Avenue North, Worcester, Massachusetts 01655-0126, USA5 Laboratory of Molecular Biology, Neurobiology Division, Medical Research Council Centre, Hills Road, Cambridge CB2 2QH, UK

Correspondence should be addressed to J.H.S. (stehle@em.uni-frankfurt.de)

Published online: 11 February 2002, DOI: 10.1038/nn806

In mammals, many daily cycles are driven by a central circadian clock, which is based on the cell-autonomous rhythmic expression of clock genes. It is not clear, however, how peripheral cells areable to interpret the rhythmic signals disseminated from this central oscillator. Here we show thatcycling expression of the clock gene Period1 in rodent pituitary cells depends on the heterologoussensitization of the adenosine A2b receptor, which occurs through the nocturnal activation ofmelatonin mt1 receptors. Eliminating the impact of the neurohormone melatonin simultaneouslysuppresses the expression of Period1 and evokes an increase in the release of pituitary prolactin. Ourfindings expose a mechanism by which two convergent signals interact within a temporal dimensionto establish high-amplitude, precise and robust cycles of gene expression.

melatonin14–16, representing a neuroendocrine hand of the cen-tral circadian clock17, and is also a major coordinator of circadi-an and seasonal physiology18. The mechanisms by which thecentral oscillator is able to drive rhythmic gene expression inperipheral target tissues and by which long-term physiologicalchanges are achieved, however, are not known.

The hypophyseal pars tuberalis (PT) provides an excellentmodel with which to investigate how rhythmic gene expression inthe periphery is temporally gated, because it contains a high den-sity of receptors for the neurohormone melatonin and it is impli-cated in the expression of both circadian and seasonal endocrinecycles19. This structure also expresses clock genes in rhythms fol-lowing the time of day and the time of year2,20.

Here we present results that define the signaling mechanismsby which the central pacemaker in the SCN regulates circadiangene expression in a peripheral target tissue. We show that theneurohormone melatonin serves as a pivotal link between the cen-tral circadian pacemaker and the PT, which results in long-termmodulation of prolactin levels.

RESULTSIn the PT of wild-type C3H/HeN mice, a sharp peak in levels ofmPer1 mRNA occurred shortly after the dark–light transition(Fig. 1a), followed 6 hours later by an increase in nuclear con-centrations of mPER1 protein (Fig. 1b and d). These rhythmswere generated endogenously by circadian output, rather thanby the light/dark cycle, because the mPer1 and mPER1 rhythms

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 235

Fig. 1. Rhythmic expression of mPer1 mRNA andmPER1 protein in the mouse pars tuberalis (PT) is dri-ven by endogenous melatonin. (a–c) Semiquantitativeanalyses of mPer1 mRNA (a), examined by in situhybridization, and mPER1 protein (b, c), examined byimmunocytochemistry in the PT. Zeitgeber time 00(ZT00) is defined as lights on (a, b); circadian time 00(CT00) is defined as subjective lights on (c). Note thatthe mPER1 rhythm was identical in wild-type (WT)mice kept under conditions of 12 h light/12 h dark (b, filled squares) or in constant darkness (c, filled tri-angles). Values shown are the mean ± s.e.m. of resultsfrom 3–5 animals for each time point and areexpressed as the percentage of the maximum value.Data at ZT18 and ZT22 are double-plotted. (d–f) Representative coronal sections, showing mPER1immunoreaction in the PT of wild-type (d, f) ormt1R–/– (e) mice that were maintained under 12 hlight/12 h dark conditions. Wild-type mice in (f) wereeither sham-operated (Sham) or pinealectomized(PinX). Animals were killed at ZT06 or ZT22, respec-tively. Scale bar, 50 µm.

persisted in the PT of wild-type mice maintained under constantdarkness (Fig. 1c). It has been shown in hamster that the pinealgland acts as a link between the SCN and the PT20,21. In pinealec-tomized (PinX) mice, both expression of mPer1 mRNA andexpression of mPER1 protein remained at a nadir in the PT at alltime points investigated (Fig. 1a, c and f).

We next examined cycles of gene expression in the PT cells ofmice carrying a deletion of the melatonin mt1 receptor (mt1R–/–).In these mice, levels of mPER1 were consistently low in the PT(Fig. 1b and e), despite the fact that the mice had been bred in amelatonin-proficient C3H/HeN background. This melatonin-dependent regulation of clock-gene expression was tissue specif-ic, because the circadian expression of mPer1 and mPER1 in theSCN did not differ between wild-type, PinX wild-type andmt1R–/– mice (data not shown).

Despite their dependence on endogenous melatonin, neitherexpression of mPer mRNA nor expression of mPER1 protein inthe PT is elicited by single or repeated injections of physiologicalmelatonin into long-term PinX Syrian hamsters21. This suggeststhat a prolonged absence of melatonin results in a loss of sensi-tivity to the hormone in the PT. To confirm this finding and fur-ther test the desensitization hypothesis, we administered serialdaily injections of melatonin to PinX Siberian hamsters, begin-ning the night after surgery. This treatment preserved the expres-sion of Per1 mRNA and PER1 protein in the PT (Fig. 2a).Together, these results show that the pineal neurohormone mela-tonin, acting through the mt1R on PT cells (Fig. 2b), is respon-sible for long-term regulation of PER1 in the PT. These resultsalso show that melatonin is involved in regulating the sensitivity ofthe PT to receptor activation.

The best characterized cellular effect of melatonin is its inhibi-tion of adenylyl cyclase through the mt1R19,22. But this acuteinhibitory action cannot explain why, in the absence of melatonin,expression of the mPer1 gene is downregulated. We thereforethought that melatonin might act in the PT to sensitize the adeny-lyl cyclase signaling pathway23, thereby setting the ‘gain’ for an

unidentified activating stimulus. We focused onadenosine as a potential candidate for this stimulusbecause it increases cAMP levels via the adenosineA2b receptor24. In brain and pituitary, the adeno-sine A2b receptor is expressed most prominently

in the PT24 (Fig. 2b) and in the pineal gland25. A role for adenosinein PT signaling has been also suggested, because adenosine canincrease phosphorylation of the activating transcription factorCREB (cyclic AMP–responsive element binding protein) in PTcells from sheep26.

To examine the contribution of the adenosine A2b receptor torhythmic expression of clock genes in the rodent PT, we treatedPinX Siberian hamsters with the adenosine A2b receptor NECA(5′-N-ethylcarboxamidoadenosine) for 6 days, beginning imme-diately after pinealectomy, to augment adenosinergic stimula-tion. Similar to immediate treatment with melatonin, NECArestored the PER1 protein peak in the PT of long-term PinX ham-sters (Fig. 2a). By contrast, expression of Per1 mRNA and PER1protein in the SCN was unaffected by pinealectomy or by treat-ment with melatonin or NECA (data not shown), confirmingthat the link between melatonin, adenosine A2b receptor signalingand clock-gene expression in PT cells is tissue specific.

The adenosine A2b receptor is positively linked to adenylylcyclase24, whereas melatonin signaling in the PT causes an acutesuppression of this enzyme22. To define the biochemical inter-action that occurs between melatonin and adenosine duringregulation of clock-gene expression in the PT, we examined thedynamics and mechanisms underlying regulation of the mPER1signal using mouse hypothalamic slices. The basal level ofmPER1 in control tissues, maintained for 2–14 hours in vitro,was low. Exposure to NECA, but not to various other cAMP-elevating agents including VIP (vasoactive intestinal peptide; 1 µM), PACAP (pituitary adenylate cyclase-activating polypep-tide; 100 nM), norepinephrine (1 µM), estrogen (10 nM), EGF(epidermal growth factor; 10 nM) and cGMP (1 µM), causeda time-dependent increase in mPER1, which peaked after 6 hours (Fig. 2c). This was preceded by an acute upregulation ofboth the production of cAMP (control 0.09 ± 0.03 pmol/ml (n = 6) versus NECA 0.70 ± 0.20 pmol/ml (n = 8); p < 0.01)and the phosphorylation of CREB, which peaked 40 minutesafter application of NECA (Fig. 2c).

a

b

c

d

e

f

WT

mt1R–/–

ZT22

ZT06 ZT22

WT

Sham

ZT06

PinX

ZT06

ZT060 3618 22 02 06 10 14 18 220

100 WTWT PinX

Zeitgeber time (ZT)

0 3618 22 02 06 10 14 18 220

100 WTmt1R-/-

Zeitgeber time (ZT)

0 3618 22 02 06 10 14 18 220

100WT PinXWT

Circadian time (CT)

mP

ER

1 m

RN

A(%

of

max

imu

m)

mP

ER

1 p

rote

in(%

of

max

imu

m)

mP

ER

1 p

rote

in(%

of

max

imu

m)

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

236 nature neuroscience • volume 5 no 3 • march 2002

articles

Fig. 2. Melatonin sensitizes adenosinergicand/or cAMP-mediated expression of mPER1in rodent PT cells. (a) Expression ofimmunoreactive mPER1 in the Siberian ham-ster PT in vivo, with representative coronalimages and the corresponding semiquantita-tive densitometric analysis. Injections of eithermelatonin or the adenosinergic agonist NECAreversed the loss of immunoreactive mPER1 inthe PT of pinealectomized (PinX) animals (*, p < 0.05; ***, p < 0.001). Tissues were col-lected at ZT06 on day 6 after pinealectomy.Each data point represents the mean ± s.e.m.of results from 5 animals. Scale bar, 50 µm. (b) RT-PCR analyses show that adenosine A2band melatonin mt1 receptors (mt1R) are pre-sent in the PT of wild-type mice. The adeno-sine A2b receptor was found in the PT but notin the SCN of wild-type mice (histone H3.3was used as an internal control). (c) In PTexplants of wild-type mice, the adenosine ago-nist NECA led to a peak in phosphorylatedCREB (pCREB) within 40 min of stimulation,whereas maximum concentrations of mPER1protein were reached after 6 h. Under contin-uous stimulation with NECA, pCREB andmPER1 immunoreaction returned to basal lev-els within 80 min and 12 h, respectively. ThecAMP antagonist Rp8CPT-cAMPS blocked theNECA-induced phosphorylation of CREB andthe increase in mPER1 levels in PT explants ofwild-type mice. Scale bar, 50 µm. (d) The con-centration of mPER1 in PT explants of wild-type mice is insensitive to melatonin (Mel, 1nM) alone, but is significantly elevated afterapplication of the adenosinergic agonist NECA(1 µM for 6 h; +, p < 0.05 versus control, Co);a representative mPER1 immunoreaction isshown at top right. Pretreatment of PTexplants for 6 h with melatonin significantly potentiates the NECA-induced immunoreaction of mPER1 at concentrations >10 nM (*, p < 0.05; **, p <0.01 versus NECA alone); a representative mPER1 immunoreaction is shown at bottom right. Each data point represents the mean ± s.e.m. of resultsfrom 5 animals. Representative coronal images are shown on the right. Scale bar, 25 µm.

We confirmed the contribution of cAMP-dependent signal-ing to the adenosinergic induction of CREB phosphorylation andmPER1 protein expression in the PT, by pretreating the cells withthe protein kinase A inhibitor H89 (1 µM) and the cAMP antag-onist Rp8CPT-cAMPS (1 mM), which completely abolishedNECA-induced phosphorylation of CREB and induction ofmPER1 (Fig. 2c). Treatment with melatonin alone had no effecton the concentrations of mPER1 protein in PT explants of wild-type mice (Fig. 2d); however, pretreating PT explants with mela-tonin for 6 hours (mimicking the endogenous nocturnalmelatonin surge) before stimulation with NECA potentiated the

induction of mPER1 protein by NECA (Fig. 2d), confirming in vitro the melatonin-dependent sensitization of adenosine A2bsignaling observed in vivo. This sensitization effect of the cAMPsignaling pathway by melatonin in PT cells was also observedwhen slices were preincubated with melatonin plus NECA beforeadenosinergic stimulation. The melatonin effect occurred at con-centrations higher than 10 pM (Fig. 2d), which corresponds wellwith the physiological concentrations of the pineal hormonefound in mouse blood at night27.

It was important to identify the effect on the endocrine systemof melatonin-dependent temporal signaling in the PT. Cells fromthe PT are known to determine the release of prolactin from theadenohypophysis through an unidentified paracrine factor19, andin rodents melatonin attenuates the release of prolactin throughhigh-affinity mt1Rs that are present within the pituitary only onPT cells22,28. To examine the impact of the dysregulation of mela-tonin-dependent signaling in PT cells, we measured the 24-hour

pC

RE

B

0 min

0 min

b

mt1R 480

H3.3 217

A2b 485

bpSCN PT

***

*

*

c

a

d

0Sham PinX PinX + mel PinX + NECA

a

b

0 0 –13 –12 –11 –10 –9 –80

50

100**

** *

NECA

Mel

Co

+

100

mP

ER

1(%

of

max

imu

m)

mP

ER

1(%

of

max

imu

m)

Mel + NECA

20 min

180 min

40 min

360 min

80 min

720 min

40 min+Rp8PT-cAMPS

360 min+Rp8PT-cAMPS

mP

ER

1

Log [mel (M)]

0 922 02 06 10 14 18 22 020

1,000

2,000

3,000

WT

mt1R–/–

Zeitgeber time (ZT)

Pro

lact

in (

pg

/ml)

p ≤ 0.05

Fig. 3. Disinhibition of prolactin concentrations in the blood of mela-tonin mt1 receptor knockout (mt1R–/–) mice. Amounts of circulatingprolactin are significantly higher in mt1R–/– mice than in wild-type litter-mates (p ≤ 0.05). Each data point represents the mean ± s.e.m. of resultsfrom 5–8 animals.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 237

profile of prolactin secretion by wild-type and mt1R–/– mice. Atabout the middle of the dark period, we observed a marked upreg-ulation of prolactin concentrations in the blood of the mutant miceas compared with their wild-type littermates (Fig. 3).

DISCUSSIONThe hierarchical organization of the mammalian circadian sys-tem asks for cues, released by the central pacemaker in the SCN,to entrain the circadian patterning of gene expression in theperiphery. We have shown here that, in the model system of therodent pituitary, rhythmic gene expression is timed by two inter-locked factors: adenosine, an activator of cAMP, and the neuro-hormone melatonin, an inhibitor of cAMP.

Central to this timed activation of gene expression is the noc-turnally restricted impact of melatonin, which defines the durationof the night as a hormonal hand of the central oscillator. Noctur-nally released melatonin acutely suppresses circadian expression ofmPer1 in the PT by inhibiting the cAMP-dependent signalingpathway. Simultaneously, the hormone sensitizes the adenylylcyclase signaling pathway. As dawn approaches, melatonin disap-pears from the circulation and PT cells are released from tran-scriptional suppression, thereby facilitating the acute inductionof Per1 gene expression by circulating or locally released adeno-sine29. (For additional details, see Supplementary Fig. 1 on thesupplementary information page of Nature Neuroscience online.)

This dual cellular impact of the pineal hormone in vivo—acute suppression and delayed heterologous sensitization—enhances the contrast between night and day and confersstability to rhythmic transcription in PT cells. This mechanismensures a temporally and spatially precise extension of the cir-cadian output of the central clock in the SCN, which targetsthe secretion pattern of prolactin from cells of the pars distal-is of the pituitary. We have shown also that rhythmic expres-sion of the Per1 gene forms an intrinsic part of rhythmic PTsignaling events. Notably, the tissue-specific and cycling expres-sion of genes in PT cells that depend on the dual impact ofadenosine and melatonin is not restricted to clock genes, butapplies also to expression of the inhibitory transcription fac-tor of the cyclic AMP signaling pathway, ICER (induciblecAMP early repressor14–17,20,25) (to be presented elsewhere).

Our observations provide a cellular explanation for phe-nomena that have been well-documented at the ‘system’ level;that is, the PT can translate nocturnal exposure to the neuro-hormone melatonin into a signal that regulates the amplitude ofprolactin secretion from the pituitary18,19,30–32. The regulationof mPER1 protein through the tissue-specific integration of twoconvergent but opponent signals may constitute a general mech-anism for peripheral non-clock cells that require the unique phas-ing of output signals. This dual-signal model may have widerapplications, not only for the temporal actions of melatonin on itstarget cells but also for the long-term maintenance of precise,high-amplitude cycles of gene expression in other tissues.

METHODSAnimals and experimental procedure. Care of animals and all experi-ments were conducted in accordance with Institutional Guidelines andwith Local Ethical Approval. Adult male C3H/HeN mice (Charles River,Margate, UK) and male Siberian hamsters (Phodopus sungorus) wereentrained to a photoperiod of 12 h light and 12 h dark, or 16 h light and8 h dark, respectively (lights on, Zeitgeber time 00, ZT00). Some animalswere pinealectomized, or sham-operated under deep anesthesia asdescribed21. We monitored locomotor activity using passive infraredmovement detectors (Racal-Guardall IR77, MKII, Racal Ltd., Slough,UK), linked to a computerized activity recording system (Dataquest IV,

Data Sciences, Frankfurt, Germany)33, with the onset of free-runningactivity defined as circadian time 12 (CT12). We treated pinealectomizedanimals with melatonin (1 µg per g of body weight; Calbiochem, BadSoden, Germany) at ZT20, or NECA (1 µg per g of body weight intraperi-toneally (i.p.); Tocris, Bristol, UK) at ZT00, for 6 consecutive days. Thetreatments started either immediately or 6 d after surgery. We generat-ed melatonin mt1 (Mel1a)–receptor knockout (mt1R–/–) mice in aC3H/HeN genetic background by backcrossing founders as described28

to C3H/HeN mice for 10 generations. At the 10th generation, heterozy-gotes were bred together to establish a homozygous knockout line andthe wild-type control line.

Brain slice preparation and stimulation. Coronal slices of mouse brain(400–600 µm) encompassing the caudal basal hypothalamus and theadjacent region of the hypophyseal PT were prepared and maintainedas described22,33,34. After preincubation (2 h), slices were stimulatedwith NECA (10 µM; Tocris) or melatonin (1 nM; Calbiochem) for upto 12 h. Additional sets of slices were pretreated with the protein kinaseA inhibitor H89 (1 µM) or the cyclic AMP antagonist Rp8CPT-cAMPS(1 mM; Biolog, Bremen, Germany) for 15 min before NECA was added.We pretreated some slices with melatonin (1 nM) or melatonin (1 nM)plus NECA (10 µM) for 6 h before stimulation with NECA (10 µM).Stimulation with NECA was done for a further 6 h after melatoninremoval. We purchases all chemicals from Sigma (Deisenhofen, Ger-many) unless otherwise indicated.

In situ hybridization and RT-PCR. We carried out in situ hybridizationfor mPer1 mRNA as described14,24,35. For RT-PCR, RNA was extractedfrom PT and SCN tissue as described14,15,24,35. First-strand complemen-tary DNA was generated using Superscript II reverse transcriptase (Invit-rogen, Karlsruhe, Germany). Primers and RT-PCR conditions for theamplification of mt1R and histone H3.3 have been described28. Wedesigned primers for RT-PCR of the mouse adenosine A2b receptoraccording to the published sequence24,36: forward, 5′-GAGCTC-CATCTTTAGCCTCCT-3′ , nucleotides 303–324; reverse, 5′-GCATG-CACGGGGAGCCAACAC-3′ , nucleotides 783–804. PCR reactionsconsisted of 25 cycles of amplification.

Immunocytochemistry. At selected times animals were anesthetized withchloral hydrate (500 mg per kg of body weight i.p.) and transcardiallyperfused with 4% paraformaldehyde, and then coronal hypothalamicbrain sections (40 µm) including the SCN or the PT region wereprocessed for mPER1 immunocytochemistry as described4,13. Alterna-tively, PT explants from the in vitro experiments were immersion-fixed in4% paraformaldehyde for 2 h at room temperature, cryoprotected with20% sucrose, cut on a cryostat (16 µm) and processed for phosphory-lated CREB (New England BioLabs, Frankfurt/Main, Germany) or forimmunoreactivity to mPER1 as described4,16,33,34.

Measurement of cAMP and hormone levels. Concentrations of cAMP,released from PT slices into the culture medium, were measured byenzyme-linked immunosorbent assay (Institute of Hormone and Fertil-ity Research, Hamburg, Germany) as described15. To measure concen-trations of prolactin (Amersham Pharmacia, Freiburg, Germany), weanesthetized male adult wild-type and mt1R–/– mice with chloral hydrate(500 mg per kg of body weight i.p.) and took blood from the right atriumwithin 30 s of removing the animal from the cage. The blood was col-lected in EDTA-coated tubes and processed according to the supplier’sinstructions. For a given experiment all samples were assayed simulta-neously in duplicate in a single assay.

Analysis of mPer1 and PER1 signals and statistics. We analyzed sectionsdensitometrically using a computerized image analysis system asdescribed33,34. Values are expressed as a percentage of the maximal inten-sity. Significant differences between groups were determined with Stu-dent’s t-test, or a one-way ANOVA followed by Dunnet’s post-hoc testusing GraphPad Prism (GraphPad, San Diego, California). Values of p < 0.05 were considered to be significantly different.

GenBank accession number. Mouse adenosine A2b receptor, GI 6680655.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

238 nature neuroscience • volume 5 no 3 • march 2002

articles

Note: Supplementary Information can be found on the Nature Neuroscience

website (http://neurosci.nature.com/web_specials).

Acknowledgements

We thank S. M. Reppert for help with generating the melatonin mt1R–/– mice

and for the antibody to mPER1; H. Zimmermann for discussion; M. L.

Eifländer, C. Illickovic, D. Kärger, I. Schneider-Hüther and S. Schüßler for

technical support; and E. Colnago, M. Holub, M. Kock, S. Leslie, S. Schotten and

C. Schultz for help. This work was supported by grants from the Deutsche

Forschungsgemeinschaft (to H.W.K. and J.H.S.), the Paul und Ursula Klein-

Stiftung and the Heinrich und Fritz Riese-Stiftung (to J.H.S.), the Biotechnology

and Biological Sciences Research Council and the Medical Research Council (to

M.H.H.) and the NIH (AG09975 to D.R.W.).

Competing interest statementThe authors declare that they have no competing financial interests.

RECEIVED 9 NOVEMBER 2001; ACCEPTED 9 JANUARY 2002

1. Klein, D. C., Moore, R. Y. & Reppert, S. M. (eds.) Suprachiasmatic Nucleus:The Mind’s Clock (Oxford Univ. Press, New York, 1991).

2. Sun, Z. S. et al. RIGUI, a putative mammalian ortholog of the Drosophilaperiod gene. Cell 90, 1003–1011 (1997).

3. Shigeyoshi, Y. et al. Light-induced resetting of a mammalian circadian clockis associated with rapid induction of the mPer1 transcript. Cell 91, 1043–1053(1997).

4. Shearman, L. P. et al. Interacting molecular loops in the mammaliancircadian clock. Science 288, 1013–1019 (2000).

5. Bae, K. et al. Differential functions of mPer1, mPer2, and mPer3 in the SCNcircadian clock. Neuron 30, 525–536 (2001).

6. Reppert, S. M. & Weaver, D. R. Molecular analysis of mammalian circadianrhythms. Annu. Rev. Physiol. 63, 647–676 (2001).

7. Balsalobre, A. et al. Resetting of circadian time in peripheral tissues byglucocorticoid signalling. Science 289, 2344–2347 (2000).

8. Balsalobre, A., Damiola, F. & Schibler, U. A serum shock induces circadiangene expression in mammalian tissue culture cells. Cell 93, 929–937 (1998).

9. Yamazaki, S. et al. Resetting central and peripheral circadian oscillators intransgentic rats. Science 228, 682–685 (2000).

10. Rajaratnam, S. M. & Arendt, J. Health in a 24-h society. Lancet 358, 999–1005(2001).

11. Stokkan, K. A., Yamazaki, S., Tei, H., Sakaki, Y. & Menaker, M. Entrainmentof the circadian clock in the liver by feeding. Science 291, 490–493 (2001).

12. Takekida, S., Yan, L., Maywood, E. S., Hastings, M. H. & Okamura, H.Differential adrenergic regulation of the circadian expression of the clockgenes Period1 and Period2 in the rat pineal gland. Eur. J. Neurosci. 12,4557–4561 (2000).

13. von Gall, C. et al. Clock gene protein mPER1 is rhythmically synthesized andunder cAMP control in the mouse pineal organ. J. Neuroendocrinol. 13,313–317 (2001).

14. Stehle, J. H. et al. Adrenergic signals direct rhythmic expression oftranscriptional repressor CREM in the pineal gland. Nature 365, 314–320(1993).

15. Maronde, E. et al. Transcription factors in neuroendocrine regulation:

rhythmic changes in pCREB and ICER levels frame melatonin synthesis. J. Neurosci. 19, 3326–3336 (1999).

16. von Gall, C. et al. Transcription factor dynamics and neuroendocrinesignalling in the mouse pineal gland: a comparative analysis of melatonin-deficient C57BL mice and melatonin-proficient C3H mice. Eur. J. Neurosci.12, 964–972 (2000).

17. Korf, H. W., Schomerus, C. & Stehle, J. H. The pineal organ, its hormonemelatonin, and the photoneuroendocrine system. Adv. Anat. Embryol. Cell.Biol. 146, 1–100 (1998).

18. Hastings, M. H. & Follett, B. K. Toward a molecular biological calendar? J. Biol. Rhythms 16, 424–430 (2001).

19. Morgan, P. J. The pars tuberalis: the missing link in the photoperiodicregulation of prolactin secretion? J. Neuroendocrinol. 12, 287–295 (2000).

20. Messager, S., Ross, A. W., Barrett, P. & Morgan, P. J. Decoding photoperiodictime through Per1 and ICER gene amplitude. Proc. Natl. Acad. Sci. USA 96,9938–9943 (1999).

21. Messager, S., Garabette, M. L., Hastings, M. L. & Hazlerigg, D. G. Tissue-specific abolition of Per1 expression in the pars tuberalis by pinealectomy inthe Syrian hamster. NeuroReport 12, 1–4 (2001).

22. Carlson, L. L., Weaver, D. R. & Reppert, S. M. Melatonin signal transductionin hamster brain: inhibition of adenylyl cyclase by a pertussis toxin–sensitiveG protein. Endocrinology 125, 2670–2676 (1989).

23. Hazlerigg, D. G., Gonzalez-Brito, A., Lawson, W., Hastings, M. H. & Morgan,P. J. Prolonged exposure to melatonin leads to time-dependent sensitizationof adenylate cyclase and down-regulates melatonin receptors in pars tuberaliscells form ovine pituitary. Endocrinology 132, 285–292 (1991).

24. Stehle, J. H. et al. Molecular cloning and expression of the cDNA for a novelA2-adenosine receptor subtype. Mol. Endocrinol. 6, 384–393 (1992).

25. Stehle, J. H. Pineal gene expression: dawn in a dark matter. J. Pineal Res. 18,179–190 (1995).

26. McNulty, S., Ross, A. W., Shiu, K. Y., Morgan, P. J. & Hastings, M. H.Phosphorylation of CREB in ovine pars tuberalis is regulated both by cyclicAMP-dependent and cyclic AMP-independent mechanisms. J. Neuroendocrinol.8, 635–644 (1996).

27. Li, X. M. et al. Relationship of atypical melatonin rhythm with two circadianclock outputs in B6D2F(1) mice. Am. J. Physiol. Regul. Integr. Comp. Physiol.278, 924–930 (2000).

28. Liu, C. et al. Molecular dissection of two distinct actions of melatonin on thesuprachiasmatic circadian clock. Neuron 19, 91–102 (1997).

29. Porkka-Heiskanen, T. et al. Adenosine: a mediator of the sleep-inducingeffects of prolonged wakefulness. Science 276, 1265–1268 (1997).

30. Stirland, J. A. et al. Photoperiodic regulation of prolactin gene expression inthe Syrian hamster by a pars tuberalis–derived factor. J. Neuroendocrinol. 13,147–157 (2001).

31. Hazlerigg, D. G. What is the role of melatonin within the anterior pituitary? J. Endocrinol. 170, 493–501 (2001).

32. Lincoln, G. A. & Clarke, I. J. Role of the pituitary gland in the development ofphotorefractoriness and generation of long-term changes in prolactinsecretion in rams. J. Neuroendocrinol. 7, 637–643 (1994).

33. von Gall, C. et al. CREB in the mouse SCN: a molecular interface coding thephase adjusting stimuli of light, glutamate, PACAP and melatonin forclockwork access. J. Neurosci. 18, 10389–10397 (1998).

34. von Gall, C., Weaver, D. R., Kock, M., Korf, H.-W. & Stehle, J. H. Melatoninlimits transcriptional impact of phosphoCREB in the mouse SCN via theMel1a receptor. NeuroReport 11, 1803–1807 (2000).

35. Pfeffer, M., Kühn, R., Krug, L., Korf, H.-W. & Stehle, J. H. Rhythmic variationin beta1-adrenergic receptor mRNA levels in the rat pineal gland: circadianand developmental regulation. Eur. J. Neurosci. 10, 2896–2904 (1998).

36. Marquardt, D. L., Walker, L. L. & Heinemann, S. Cloning of two adenosinereceptor subtypes from mouse bone marrow–derived mast cells. J. Immunol.152, 4508–4515 (1994).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Dendritic spines are the receptive sites of a majority of the glu-tamatergic synapses in the cortex, and have been objects of the-oretical and experimental studies for over a century1,2. Spinesvary in size and shape, and this morphological heterogeneitymay lead to functional diversity3. Changes in spine numberand shape have been documented under a variety of condi-tions2,4–6. These morphological changes in spines are mediat-ed, in part, by actin that is concentrated in spines2,7,8.Fluorescence microscopy of neurons expressing enhanced greenfluorescent protein (EGFP) or EGFP–actin has shown thatspines undergo changes in shape over a matter of seconds9–11.Such changes, which are thought to reflect a dynamic cytoskele-ton, have been suggested to be involved in synaptic plastici-ty12. The mechanisms in transduction of synaptic signals tospine motility remain largely unknown.

At cellular locations that are motile, such as the leading edgeof migrating cells, actin is in a dynamic equilibrium between fil-amentous and monomeric form13,14. Net polymerization at theplus end and net depolymerization at the minus end leads tocycling of actin, or ‘treadmilling’13,15. Cellular signals, by modi-fying the availability of free actin, polymerization dynamics andmembrane coupling, control cellular protrusion and motility. Incontrast to the detailed knowledge of the mechanisms in regula-tion of actin dynamics at motile cells13, similar information aboutdendritic spines is lacking. Most studies have focused on the reg-ulation of spine morphology over longer periods of time usingoverexpression of regulatory proteins such as the Rho-familyGTPases16,17. Acute, short-term regulation of signaling pathwaysand their activity dependence have not been examined in detail.

Although new spines have been observed after stimuli thatstrengthen synapses4,5, there is currently little evidence sugges-tive of a link between changes in the morphology of existingspines and regulation of synaptic strength. Pharmacological or

Rapid turnover of actin in dendriticspines and its regulation by activity

Erin N. Star1, David J. Kwiatkowski2 and Venkatesh N. Murthy1*

1Department of Molecular & Cellular Biology, Harvard University, 16 Divinity Ave., Cambridge, Massachusetts 02138, USA2Genetics Laboratory, Hematology Division, Brigham & Women’s Hospital, LM-302, 221 Longwood Ave., Boston, Massachusetts 02115, USA

Correspondence should be addressed to V.N.M. (vnmurthy@fas.harvard.edu)

Published online: 19 February 2002, DOI: 10.1038/nn811

Dendritic spines are motile structures that contain high concentrations of filamentous actin. Usinghippocampal neurons expressing fluorescent actin and the method of fluorescence recovery afterphotobleaching, we found that 85 ± 2% of actin in the spine was dynamic, with a turnover time of44.2 ± 4.0 s. The rapid turnover is not compatible with current models invoking a large population ofstable filaments and static coupling of filaments to postsynaptic components. Low-frequency stimu-lation known to induce long-term depression in these neurons stabilized nearly half the dynamicactin in the spine. This effect depended on the activation of N-methyl-D-aspartate (NMDA) receptorsand the influx of calcium. In neurons from mice lacking gelsolin, a calcium-dependent actin-bindingprotein, activity-dependent stabilization of actin was impaired. Our studies provide new informationon the kinetics of actin turnover in spines, its regulation by neural activity and the mechanismsinvolved in this regulation.

electrical stimulation of excitatory synapses has been shown tomodulate actin in dendritic spines11,18,19. These observations ofexternal morphology, as intriguing as they are, do not reveal theconsequences of activity on intracellular dynamics of actin. Forexample, decreased motility could result from the stabilizationof the entire pool of intracellular actin, or from decoupling offilamentous actin from the plasma membrane.

We used fluorescence recovery after photobleaching (FRAP) asa method to investigate the intracellular dynamics of actin in den-dritic spines and its regulation by physiological stimuli. We foundthat over 85% of actin in dendritic spines was in dynamic equilib-rium, with an average turnover time of 44 s. The turnover timewas not correlated with the size of the synapse or its age in culture.Stimuli that are known to induce long-term depression (LTD) atthese synapses stabilized a large fraction of the dynamic actin, per-haps by increasing filament lifetime. This stabilization was causedby the activation of NMDA receptors and seemed to involve gel-solin, a calcium-dependent actin-binding and -severing protein.

RESULTSEGFP–actin did not alter synaptic physiologyWe transfected cultured hippocampal neurons with a plasmidencoding EGFP–actin. As reported previously9, EGFP–actin local-izes to spines, which were clearly visible in transfected neurons(Fig. 1). The frequency and amplitude of miniature excitatorypostsynaptic currents (mEPSCs) in transfected neurons (1.6 Hz,15.81 ± 0.84 pA) were not different from those in control neu-rons (1.5 Hz, 15.51 ± 0.52 pA). In addition, synaptic vesicledynamics of presynaptic boutons terminating on spines labeledwith EGFP–actin were indistinguishable from those of controlsynapses (data not shown). These observations indicate thatexpression of EGFP–actin did not affect synaptic transmission,and allowed us to investigate actin dynamics in spines.

articles

nature neuroscience • volume 5 no 3 • march 2002 239

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

240 nature neuroscience • volume 5 no 3 • march 2002

articles

FRAP revealed rapid dynamics of actin in spinesCorroborating recent reports of spine motility9,10, we docu-mented the rapid movement of spines of hippocampal neurons(see Supplementary Movie 1, available on the supplementaryinformation page of Nature Neuroscience online). To study theintracellular dynamics of actin within spines, however, we turnedto the method of FRAP20–24.

In our version of FRAP, an individual spine was rapidly pho-tobleached using high-intensity laser illumination. The timecourse of the subsequent fluorescence recovery in the photo-bleached spine can be used to make inferences regarding actindynamics. Monomeric actin—both free G-actin and G-actinbound to small actin-binding proteins—will be exchangedbetween the spine and the parent dendrite over a time scale deter-mined by the effective diffusion coefficient of EGFP–actin. Mostof the actin in the spine is in the filamentous form; if filamen-

tous actin remains stable, recovery of fluorescence in short peri-ods of time will reflect exchange of mobile actin (filament diffu-sion will be very slow), and will be negligible. If, however, actinfilaments within the spine are in a dynamic equilibrium, new flu-orescent monomers (that have replaced the bleached monomers)will be continuously incorporated into filaments. The recoveryof fluorescence will then reflect the rate of turnover of actin (Fig. 2a; Supplementary Methods, available on the supplemen-tary information page of Nature Neuroscience online). As derivedin the Supplementary Methods, the net recovery of fluorescenceafter photobleaching is given by the equation

(1)

where fs is the fraction of total actin that is stable, ff is the frac-tion of total actin that turns over rapidly and λ is the time con-stant (inverse of the turnover rate). We have assumed that themonomeric actin recovers rapidly and that the fraction of totalactin that is in the monomeric form is (1 – fs – ff). We report val-ues in percentages rather than fractions for clarity.

Our essential finding was that the recovery of fluorescenceafter photobleaching occurred in two clearly discernible compo-

F (t) = 1 – fs – ff e – —t

λ

10 µmFig. 1. Fluorescence image of a neuron expressing EGFP–actin. Brightfluorescence is observed in dendritic spines, which are visualized athigher resolution in the inset. Neuron was transfected at day 7 in vitroand imaged on day 20.

b

c1.0 s0.2 s 40 s 140 sPre

0 40 80 120 140

1

0

Time (s)

Nor

mal

ized

fluo

resc

ence

0 1 2 3

0. 1

0. 2

0

Bleach↓

Bleach↓

Fig. 2. Fluorescence recovery after photobleaching reveals dynamicactin. (a) Illustration of the FRAP method. Fluorescent actin monomers(green barbs) and filaments are free or bound to mobile or immobileproteins (black rectangles and circles), and the binding affinities can beregulated. Selective photobleaching of actin in the spine is followed byrapid exchange of mobile monomers (free and bound) with the den-dritic region and recovery of fluorescence. If actin filaments do not turnover, filamentous actin will remain nonfluorescent (top panels) and thespine will not recover its fluorescence. On the other hand, if filamentsare turning over rapidly, fluorescent monomers will be continuouslyincorporated into them and the bleached actin monomers will beexchanged out of the spine. The spine will thus recover its fluorescence(bottom panels). (b) An example of a spine that recovered its fluores-cence after photobleaching. The image at 0.2 s after bleaching indicatesessentially complete loss of fluorescence in the bleached spine (a neigh-boring spine is unaltered), and subsequent images show gradual recov-ery of fluorescence. Boxes in the first panel indicate 2 regions wherefluorescence intensity was measured and shown below. (c) Fluorescence intensities in the 2 boxes drawn in panel b, normalizedto resting values. The spine that was bleached (solid circles) shows 2components of recovery—a rapid component with a time constant of∼ 1 s, and a slower component with a time constant of ∼ 40 s. The insetshows the early component on an expanded time scale. The continuousline is given by equation 1. Fluorescence recovery was up to ∼ 90% ofthe prebleach level. A neighboring spine that was not bleached(squares) showed no systematic changes in intensity.

a©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 241

nents, with time constants of about 1 s and 40 s (Fig. 2). Togeth-er these two components accounted for >90% of recovery of flu-orescence in the spine (Figs. 2c and 3a). The average timeconstant for recovery of EGFP–actin after photobleaching was44.2 ± 4.0 s (n = 20 spines). The average amount of rapidlymobile actin in dendritic spines was 14 ± 2%, and essentially allof the remaining actin (81.4 ± 2%) was in a dynamic state; only4.6% of total actin was stable (Table 1).

Our interpretation of the two phases of fluorescence recov-ery was supported by several experiments. First, we obtained anindependent estimate of the time constant of recovery of a mobileprotein, whose size is of the same order of magnitude as that ofEGFP–actin. We used EGFP, which is half the size of EGFP–actin.Fluorescence recovery in EGFP spines was rapid and almost com-plete, with an average time constant of 0.70 ± 0.052 s (n = 11;

Fig. 3b). This rate was similar to the fast component of recoveryof EGFP–actin, but two orders of magnitude faster than the dom-inant time constant of recovery of EGFP–actin (44.2 ± 4.0 s). Wealso carried out experiments designed to distinguish among threeprocesses that might contribute to the second phase of recovery:(i) slow diffusion of filaments (ii) slow unbinding of G-actin fromimmobile actin binding proteins, and subsequent exchangebetween the spine and parent dendrite, and (iii) the polymeriza-tion dynamics of filaments, which will be governed by many fac-tors including the affinity of filaments for actin binding proteins.The results of these experiments, described below, strongly favorthe last possibility and argue against the first two.

We used pharmacological agents that perturb actin dynam-ics to show that the fluorescence recovery of EGFP–actin wasdependent on the presence of dynamic actin. First, we treatedneurons with latrunculin A (10 µM), which sequestersmonomeric actin and causes depolymerization, and found agradual decrease in the fluorescence intensity of spines. Thissteady loss of fluorescence, presumably due to filament disas-sembly, prevented measurements using FRAP. Therefore, wetreated neurons with 5 µM cytochalasin D for 15–30 minutesand measured FRAP parameters. Cytochalasin D is a barbed-end capping molecule that prevents actin polymerization25, andit substantially slowed the recovery of EGFP–actin fluorescenceafter photobleaching (Fig. 3c). The recovery could be describedwell by assuming that cytochalasin D increased the percent ofmobile actin to 19 ± 3% (compared to 14 ± 2% in control) andconverted 56 ± 8% (p < 0.001 compared to control) of spineactin into a stable pool; this left only 24% of actin for rapidturnover. This is exactly what is expected if the barbed ends arecapped and exchange of monomeric actin occurs only at thepointed end13,25. In a complementary experiment, addition ofthe cell-permeable actin filament–stabilizing compound jas-plakinolide26 converted nearly all actin (97 ± 1%) to a stableform (presumably filaments) and completely prevented recov-

Fig. 3. Parameters of actin turnover. (a) Average plots of fluorescenceintensities (normalized to resting levels) before and after photobleachingfor 20 resting spines. Recovery was nearlycomplete by 150 s after bleaching, indicatingthat essentially all actin in spines was dynamic,with a turnover time of 44.2 ± 4.0 s (n = 20).(b) Recovery of cytosolic EGFP fluorescencewas rapid, occurring with a time constant of0.70 ± 0.052 s (n = 11 spines). Example imagesfrom a single experiment are shown at right.(c) Spines treated with cytochalasin D (Cyto)recovered much less after photobleaching.The data could be described well by assuming55.7 ± 7.6% (n = 9) of the actin was stable. Inspines treated with the actin stabilizing agentjasplakinolide (Jas), recovery was negligible(97.2 ± 1.0% stable filaments, n = 6). The con-tinuous lines are best fits of equation 1 to theaverage data. Data from control spines frompanel (a) is reproduced for comparison. (d) Turnover time and the percent of dynamicactin are not related to the level of expressionof EGFP–actin. Expression level for eachexperiment was estimated by averaging thefluorescence intensity of many spines in thevicinity of the particular spine that was photo-bleached. The correlation coefficient betweenexpression level and the time constant was0.069 (p > 0.5); that between expression leveland percent dynamic actin was 0.074 (p > 0.5).The value for the percentage of dynamic actinwas constrained to be no greater than 100%in the fitting procedure.

Table 1.

Turnover % mobile actin % stable % dynamictime (s) (1 – fs – ff) × actin actin

100 fs ×100 ff ×100

Control 44.2 ± 4.0 14.0 ± 1.6 4.6 ± 1.6 81.4 ± 1.6

Cytochalasin 59.8 ± 10.0 18.6 ± 3.4 55.7 ± 7.6 25.7 ± 7.6

Jasplakinolide — 1.7 ± 0.5 97.2 ± 1.0 1.1 ± 1.0

TTX 32.8 ± 4.4 12.0 ± 1.2 2.6 ± 3.9 85.4 ± 3.9

NMDA 42.8 ± 5.7 9.0 ± 1.7 57.3 ± 7.6 33.7 ± 7.6

Stimulation 52.2 ± 7.9 14.5 ± 1.9 40.5 ± 4.2 45.0 ± 4.2

Stim. + APV 40.5 ± 5.1 15.0 ± 2.3 13.9 ± 4.1 71.1 ± 4.1

Gsn–/– rest 48.3 ± 3.8 12.1 ± 3.5 4.5 ± 4.5 83.4 ± 1.6

Gsn–/– stim. 45.5 ± 5.3 11.1 ± 1.1 15.7 ± 3.5 73.2 ± 3.5

0

1

Nor

mal

ized

fluo

resc

ence

Nor

mal

ized

fluo

resc

ence

Nor

mal

ized

fluo

resc

ence

0 10050 150

Time (s)

Time (s)

Time (s)

a

Dyn

amic

act

in (

%)

0 60 120

50

0

Tur

nove

r tim

e co

nsta

nt (

s) 100d

50

0

100

0

1

Control

Cyto

Jas

0 10050 150

c

0 5

1

0

EGFP

b

0.34 s

–0.3 s

0.62 s

1.45 s

0.01 s

EGFP–actin expression (arbitrary units)

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

242 nature neuroscience • volume 5 no 3 • march 2002

articles

FM4-64 to label actively recycling presynapticvesicles, and examined whether EGFP–actinlabeled spines had functional presynaptic termi-nals (Fig. 4b). A majority of spines had function-al presynaptic partners. There was no correlationbetween the size of the presynaptic vesicle pool—an index of the morphological size of thesynapse27—and the dynamics of actin (Fig. 4c).

Large synapses, which are expected to be stable and mature, hadthe same actin turnover rate as small synapses (Fig. 4c).

Effect of activity on actin dynamicsTo determine if actin dynamics was affected by activity, we exam-ined the effect of blocking spontaneous activity on the parametersof actin turnover. Spontaneous action potentials occur at a lowrate (0.5 Hz on average) in our cultures, and we found that 1 µMtetrodotoxin (TTX) increased the rate of turnover of actin (timeconstant 32.8 ± 4.4 s, significantly faster than in control spines,p < 0.05). The amount of mobile and stable actin was not sub-stantially different from control values (Table 1). Next, we test-

Fig. 4. Actin turnover is not systematically related toage or size of spines. (a) The turnover of actin in spinesfrom neurons grown for 14–16 days in vitro (DIV) wasindistinguishable from those grown for 22–24 DIV.(b) An example illustrating that most spines have func-tionally active presynaptic partners. Presynaptic vesicleswere labeled with the styryl dye FM4-64 (red). FM4-64puncta without corresponding EGFP–actin spines cor-respond to synapses made onto postsynaptic targetsfrom other non-transfected cells. (c) The parametersof actin turnover did not correlate with the size of thepresynaptic recycling pool, which has been shown pre-viously to correlate well with the morphological size ofthe synapse.

ery of actin after photobleaching (Fig. 3c). As the major knownactions of cytochalasin and jasplakinolide are on filamentdynamics, our results do not support a model in which recoveryof fluorescence in untreated spines is governed by diffusion offilaments, or by slow unbinding of actin from immobile actinbinding proteins. Rather, our data supports a model in whichfilament turnover determines the time course of fluorescencerecovery (Supplementary Methods).

Another potential concern is that overexpression ofEGFP–actin could alter the dynamics of actin. If this were thecase, one would predict a systematic relation between the levelof expression of EGFP–actin and the parameters of actin dynam-ics. We measured EFGP–actin expression level for each neuronby averaging the fluorescence intensity of many spines of thatneuron. Intensities were normalized with respect to illuminationintensity, detector gain and confocal scanning parameters.Although expression levels varied by more than an order of mag-nitude, there was no systematic relation between the expressionlevel and the parameters of actin turnover in spines (Fig. 3d).

Actin turnover independent of age or synapse sizeThe relation between the dynamics of actin in spines and thematurity of synapses is unclear9–11. In our experiments, no dif-ferences in the kinetics of recovery was observed between spinesfrom neurons grown for 14–16 days in vitro and those grown for22–24 days in vitro (Fig. 4a). We next used the membrane dye

0 10050 1500

1

Time (s)

Nor

mal

ized

fluo

resc

ence

14–16 DIV 22–24 DIV

FM4–64 EGFP–actin Overlay

a

b

c

0 200100

0

100

50

Functional size of synapse (arbitrary units)

Per

cent

age

mon

omer

(%

)

Per

cent

age

stab

le

filam

ent (

%)

Tur

nove

r tim

e co

nsta

nt (

%)

10 µm

0

1

Nor

mal

ized

fluo

resc

ence

Nor

mal

ized

fluo

resc

ence

a

0 10050 150

Time (s)

Control

NMDA

1 Hz, 900 APs

Imaging

↑Bleach

0 10050 1500

1

Time (s)

Control

Stim+ APV

Stim

bStim

TTX

Fig. 5. Regulation of actin turnover in spines by activity. (a) Blockingspontaneous activity with 1 µM TTX results in a speeding up of actinturnover (time constant 32.8 ± 4.4, n = 11). Application of 10 µM NMDAcaused a significant slowing of fluorescence recovery after photobleach-ing, best described by stabilization of 57.3 ± 7.6% of the filaments (n = 6;p < 0.01 when compared to controls). (b) Actin dynamics were mea-sured after a stimulus of 900 action potentials delivered at 1 Hz.Measurements were made within 5 min of termination of stimulus. Asshown by the average recovery curve, stimulation led to a slowing ofrecovery, which could be explained by the stabilization of 40.5 ± 4.2% ofactin (n = 15; p < 0.01 when compared to controls). If 50 µM APV waspresent during the 900 AP stimulus, fluorescence recovery was notslowed down as much, 13.9 ± 4.1% of actin being stabilized (n = 8, signif-icantly different from stimulation in the absence of APV, p < 0.01).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 243

ed the effect of direct application of glutamate agonists, whichhave been previously shown to lead to a loss of actin filamentsin the spine7 as well as to spine motility18. Application of 10 µMNMDA resulted in a slight reduction of fluorescence in the spine,suggestive of a loss of filamentous actin (data not shown). Therewas also a significant slowing of the fluorescence recovery with-in 1 minute of the addition of NMDA to the perfusion media(Fig. 5a). In line with recent findings that NMDA reduces themotility of spines18, we found that the predominant effect ofNMDA is the stabilization of actin (57.3 ± 7.6% stable actin com-pared to 5 ± 2% in control spines, p < 0.001; Table 1). Incubatingneurons with 50 µM 2-amino phosphonovalerate (APV), anNMDA-receptor antagonist, before and during the addition ofNMDA largely prevented the loss of spine fluorescence and theslowing of fluorescence recovery.

Although there is evidence for modification of spines afterinduction of long-term potentiation4,5, changes in spine dynam-ics after induction of LTD are unknown. We measured actinturnover at the end of a stimulus protocol of 900 action potentialsat 1 Hz, which leads to a robust and consistent induction of LTDin dissociated cultures28,29 (E.N.S. and V.N.M., unpublisheddata). The parameters of actin dynamics were measured at theend of the stimulation protocol. We found that low-frequencystimulus resulted in a pronounced slowing of fluorescence recov-ery after photobleaching (Fig. 5b). The slower recovery could beaccounted for by an increase in the fraction of actin that was inthe stable form (41 ± 2% compared to 5 ± 2% in control, p < 0.001); the turnover rate of the remaining dynamic actin wasnot altered significantly (52.2 ± 7.9 s compared to 44.2 ± 4.0 s incontrol spines). We also examined the fluorescence intensity inspines before and after stimulation. If net depolymerizationoccurred within spines, there would be a decrease in resting flu-orescence intensity after stimulation (as monomeric actin result-ing from depolymerization would be lost to the adjacent dendriticshaft due to diffusional equilibrium). We saw little reduction influorescence intensity after stimulation, consistent with minimaldepolymerization of actin in spines. Therefore, the most parsi-monious explanation is that low-frequency stimuli mainly led tothe stabilization of a subpopulation of actin filaments.

Fig. 6. Activity-dependent stabilization of actin is impaired inneurons lacking gelsolin. (a) Actin dynamics were measured inspines of gsn–/– neurons. Average recovery curve for gsn–/–

spines (n = 12) is shown superimposed on data for controlspines, and they are statistically indistinguishable. (b) Low-fre-quency stimulation does not reduce actin turnover in gsn–/–

neurons as much as in control neurons—only 15.7 ± 3.5% ofactin was stabilized (n = 12; significantly less than controlspines after stimulation, p < 0.01). Shown are average recoverycurves for gsn–/– spines at rest and after stimulation (stim), andcontrol spines after stimulation. (c) Presynaptic vesicle releaseduring 1-Hz stimulation, measured using FM4-64 labeling, isnot significantly impaired in gsn–/– neurons. The amount ofrelease per action potential was measured as a fraction of thetotal recycling pool of vesicles, and was 0.35 ± 0.13% for con-trol synapses and 0.34 ± 0.18% for gsn–/– synapses (n = 400synapses from 4 experiments each, p > 0.1). The total recy-cling pool of vesicles was also measured by labeling synapseswith FM4-64 using a saturating stimulus of 900 action potentialat 10 Hz. No significant difference was seen between gsn–/–

and control synapses. (d) Postsynaptic responses in gsn–/– neu-rons, measured using dual component (AMPA and NMDA)mEPSCs, were not different from control neurons. Averagesof mEPSCs from 5 cells each are shown.

Slowing of actin turnover depends on NMDA receptorsThe activity-dependent increase in the turnover time of actin fil-aments was dependent on the activation of NMDA receptors.When 50 µM APV was present during the LTD stimulus, actinturnover was not slowed very much; the stable filament fractionincreased to only 14 ± 4% (compared to 41 ± 4% without APV, p < 0.001, Fig. 5b). The residual effect of LTD stimulation onactin stabilization was abolished by a cocktail of blockers: APV, 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) and low con-centrations of Ni2+ (which blocks T-type and R-type voltage-gated calcium channels). This indicates that during physiologicalstimulation, calcium entry through NMDA receptors may leadto slower turnover of actin, with some minor additional effectmediated through voltage-gated calcium channels.

Role of gelsolin in actin dynamics in spinesOur results indicate that a calcium-sensitive molecular compo-nent may be responsible for regulating actin dynamics in spine.Gelsolin is known to sever actin and cap its barbed end in a cal-cium-dependent way30–32, and is present in hippocampal neu-rons33. To test whether gelsolin is involved in stabilizing actinafter low-frequency stimulation, we studied actin dynamics inneurons from gelsolin null (gsn–/–) mice34. First, we examinedthe turnover of actin filaments in the resting spines. FRAP exper-iments indicated that the dynamics of actin in spines of gsn–/–

neurons were normal (Fig. 6a). The average amount of mobileactin in the spine (12.1 ± 3.5), the turnover time (48.3 ± 3.8 s)and the percentage of dynamic actin (83.4 ± 4.5%) were indis-tinguishable from control spines (Table 1).

Next, we repeated the LTD experiments by stimulating 900APs at 1 Hz. In control cells, this had led to a stabilization ofactin and a pronounced slowing of fluorescence recovery (Fig. 5b). In gsn–/– neurons, however, activity did not affectturnover to the same extent. Stimulation led to stabilization ofonly 15.7 ± 3.5% of the actin, which was significantly less thanthe corresponding value for control spines (45.0 ± 2.0%, p < 0.01). The percentage of mobile actin (11.1 ± 1.1) and theturnover time (45.5 ± 5.3 s) were not significantly differentfrom control spines.

0

1

Nor

mal

ized

fluo

resc

ence

Nor

mal

ized

fluo

resc

ence

0 10050 150

Time (s) Time (s)

ControlGsn–/–

0

1

0 10050 150

Gsn–/– rest

Control stim

Gsn–/– stim

0.4

0.2

0Flu

ores

cenc

e re

leas

ed p

er A

P (

%)

200

100

0

Tot

al p

ool o

f ves

icle

s(a

rbitr

ary

units

)

Ctl Gsn–/– Ctl Gsn–/–

10 ms

10 p

A

dc

a b

Control

Gsn–/–

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

244 nature neuroscience • volume 5 no 3 • march 2002

articles

The loss of stimulus-induced slowing of actin turnover inspines of gsn–/– neurons could simply be due to impairmentin neurotransmitter release, or activation of NMDA receptors.We compared the rate of vesicle exocytosis in control andgsn–/– synapses by FM4-64 imaging35. In response to the 1 Hzstimulation used for induction of LTD, there was no differ-ence in the rate of dye release between gsn–/– and controlsynapses (Fig. 6c). There was also no significant difference inthe size of the total recycling pool of vesicles between the twosynapses (Fig. 6c). We also measured the glutamate receptormediated currents at single synapses by recording dual-com-ponent mEPSCs from gsn–/– mice. The average size of unitaryAMPA and NMDA currents were very similar in control andgsn–/– neurons (Fig. 6d). We conclude that the neither releasenor NMDA currents is significantly reduced in the gsn–/– mice,and the impairment in the activity-dependent slowing of actinturnover is due to a signaling pathway downstream of NMDAreceptor activation.

DISCUSSIONThe morphological plasticity of dendritic spines has attracted muchattention recently, in part because of its potential relation to synap-tic modification2,3. Recent data indicate that even after they areformed, spines retain some motility9–11, perhaps to allow func-tional modification under the right circumstances. In this paper,we have used FRAP to assess the dynamics of actin in spines.

The first unexpected result of our experiments was that vir-tually all actin in the spine was in a dynamic state. Several mod-els of spine dynamics proposed recently18,36 invoke a stable coreof actin capped by a corona of dynamic actin. Such an organi-zation was inferred from observations that the presence of agentssuch as cytochalasins or latrunculins does not result in a rapidloss of filamentous actin in dendritic spines37,38. These phar-macological manipulations are, however, incomplete indicatorsof the dynamic state of actin because of their complex effects.For example, cytochalasin D is a barbed end-capping moleculethat prevents polymerization at that end. In its presence, how-ever, filaments can continue to exist, albeit with different poly-merization dynamics25. Similarly, latrunculins are thought tosequester monomeric actin with a 1:1 stoichiometry; therefore,at concentrations much lower than that of free actin (expected tobe ∼ 100 µM), they may have submaximal effect. Analysis usingFRAP allows direct examination of the actin turnover withinspines and provides quantitative estimates of filament stability.As recovery after photobleaching was nearly complete within 2 minutes, constant turnover of actin must occur within spines.This turnover presumably involves treadmilling of existing fil-aments, nucleation of new filaments and severing of existing fil-aments. The relative contribution of these and other processesis difficult to assess at present.

Our measurements of turnover rates of ∼ 40 s imply that actinfilaments are not irreversibly attached to immobile binding part-ners, and that there is dynamic exchange of monomers betweenfilaments and monomers. This raises questions about how fila-ments are linked to each other, and to other postsynaptic com-ponents8. A static model whereby a specific monomer in thefilament, perhaps the barbed end, maintains contact with theactin-binding domain of another protein is not supported by ourdata. Instead, our data indicate that the association of actin fila-ments with actin binding proteins must be dynamic andreversible, to allow monomer exchange. A dynamic equilibriumalso predicts that either a large number of free barbed ends are

available for actin monomer addition, or actin filament nucle-ation occurs. If a rapid burst of polymerization occurs near themembrane, this will lead to the membrane protrusions observedin resting spines9,10. If barbed ends were capped by endogenouscapping proteins or by exogenously applied cytochalasins, fila-ment turnover would be slowed.

A second notable finding of our study is that actin turnoverrate was not related to the age or maturity of the spine. A recentstudy indicated that spine motility might be a function of agein some areas of the brain, but not in the hippocampus10. Wefound that none of the parameters of actin dynamics was alteredbetween 2 and 3 weeks in vitro, but it remains unknown if spinedynamics are altered past 3 weeks. It is also of interest that evenat a specific age, spines that were associated with larger synaps-es (which might be expected to be more mature) were no lessdynamic than those associated with smaller synapses. Therefore,it seems that nearly all spines contained a large amount ofdynamic actin. In addition, blocking spontaneous activity ledto speeding up of actin turnover; thus, resting spines in vivo mayactually contain more stable actin, as higher levels of restingactivity may occur in vivo.

A link between actin dynamics and synaptic plasticity seemsself evident and there is some experimental evidence forthis39–41. As low-frequency stimulation has been used exten-sively to induce LTD depression in vitro28,29, we considered itseffect on actin turnover. Sustained low-frequency synapticactivity led to the stabilization of actin, presumably in the fil-amentous form. Although it is possible that the stabilizationof spine actin by low-frequency stimulation is unrelated to theinduction of LTD, the common dependence on NMDA recep-tor activation argues for a connection between the two. A con-nection was also proposed by a recent study that examined therole of dynamic actin in LTD41. In that study, treatment withlatrunculin A led to increased AMPA-receptor endocytosis. Theauthors interpreted this to mean that actin depolymerizationwas a trigger for endocytosis of AMPA receptors, leading toLTD. This finding is apparently at odds with our experiments aswell as those of others18, which find that glutamate applica-tion or low-frequency stimulation actually lead to stabilizationof actin. It is possible that distinct populations of actin areinvolved in the different effects, and different calcium concen-trations may have different effects on actin dynamics. In ourexperiments, short-duration cytochalasin treatment slowsdown turnover rate of actin, without significant depolymer-ization (Fig. 3c). Further experiments using more specific per-turbations of actin dynamics, combined with physiologicalstimulation, are necessary to resolve potential discrepancies.

What are the mechanisms by which activity stabilizes actinfilaments in spines? Relatively little is known about molecules inthe spine that regulate actin dynamics over short time scales.We first established that for low-frequency stimulation, calci-um entry through NMDA receptors was necessary to stabilizefilamentous actin. The fact that action potentials alone at lowfrequencies (without coincident activation of NMDA receptors)were insufficient to cause stabilization is suggestive of a coop-erative mechanism, perhaps involving the amount of calciumentering the spine. Several calcium-dependent signaling path-ways exist in the spine, and these might have indirect effects onactin dynamics. We chose to examine the role of gelsolinbecause it has been implicated in calcium-dependent regula-tion of actin dynamics in other cells42.

We found that actin turnover in resting synapses made bygsn–/– neurons was similar to that in wild-type neurons. Low-fre-

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 245

quency stimulation, however, failed to increase the fraction ofstable actin in gsn–/– neurons, indicating that gelsolin may nor-mally serve to stabilize actin in spines. This might seem coun-terintuitive, as in other circumstances gelsolin has been proposedto increase the resting turnover42. The reason for the discrepan-cy might be the capping effect of gelsolin. If the severed filamentsare tightly capped by gelsolin, polymerization will be impaired.The rate of depolymerization induced by gelsolin depends oncalcium concentration32. At a calcium concentration of 1 µM orless, which might be expected to occur under the conditions ofour experiments, depolymerization induced by gelsolin is quiteslow. Therefore, the predominant effect of gelsolin activation isprobably tight binding to the barbed end, without significantdepolymerization43. Under some other conditions, displacementof the gelsolin cap by modulators such as phosphatidylinositol-4,5-bisphosphate (PIP2) might create new barbed ends for fur-ther polymerization44. Of course, it is also possible that theimpaired activity-dependence of actin dynamics in gsn–/– miceis due to an indirect effect not involving the filament severingfunction of gelsolin.

What of the other regulatory proteins within the spine?Other actin binding proteins present in the spine includedrebrin, myosin V and α-actinin-2 (reviewed in ref. 8), as wellas SPAR, Homer and Shank, shown recently to affect spineshape45,46. Though many of these are not directly regulated bycalcium, they might still be involved in governing actin dynam-ics through intermediate pathways. The Rho family GTPaseshave recently taken center stage in the regulation of dendriticand spine morphology47. It is noteworthy that gelsolin wasrecently shown to be a downstream effector of Rac in fibrob-lasts and neutorphils48,49. Future experiments exploiting thesimple assay of FRAP can help in understanding the role of thesecomponents in regulating actin dynamics.

In summary, we found that nearly all actin within dendrit-ic spines was dynamic, with an average turnover time of ∼ 40 s. Physiological stimuli that lead to long-term depressionof synaptic strength stabilized a large fraction of spine actin,presumably in the filamentous form. This slowing was depen-dent on the activation of NMDA receptors and a signalingpathway involving gelsolin.

METHODSCultures and transfection. Hippocampal neurons were dissociated from1–2-day rats using methods described previously35. Vectors for expres-sion of EGFP–actin and EGFP were obtained from Clontech (Palo Alto,California). Neurons were typically transfected at 6–7 d in vitro usingthe calcium phosphate method50. Transfected cultures were allowed togrow for another week, allowing synapses to mature. Homozygous gsn–/–

mice have been described before34, and were obtained from a colonymaintained in one of our laboratories (D.J.K.). Actin dynamics in den-dritic spines of BALB/c mouse neurons that were used as controls weresimilar to those from rats, and data were combined. Culture and trans-fection protocols were essentially identical to those used for rats. Exper-iments were done after 13–15 d in vitro (except where noted) at roomtemperature (20–22°C). All animal experiments were approved by Har-vard University’s standing committee on the use of animals in researchand training.

Treatment with pharmacological agents. For experiments using cytocha-lasin (5 µM, 0.1% DMSO) and jasplakinolide (10 µM, 0.1% methanol),we incubated neurons in HEPES-buffered physiological saline (compo-sition given below) for 15 min in the recording chamber on the micro-scope stage and experiments were done immediately afterwards.Compounds were obtained from the following sources: cytochalasin D,jasplakinolide and TTX from Sigma (St. Louis, Missouri); CNQX, APVand picrotoxin from RBI (Natick, Massachusetts).

Electrophysiology. Whole-cell patch-clamp recordings from culturedhippocampal neurons were done with an Axopatch 200B (Axon Intru-ments, Union City, California) amplifier. Electrodes, whose tip resistancewere 4–7 MΩ, contained (in mM): 130 potassium gluconate, 10 NaCl,1 EGTA, 0.133 CaCl2, 2 MgCl2, 10 HEPES, 3.5 MgATP, and 1 NaGTP.Extracellular perfusion medium contained (in mM): 136 NaCl, 2.5 KCl,10 HEPES, 10 D-glucose, 2 CaCl2, 1.3 MgCl2, 0.050 picrotoxin and 0.5 µM TTX. Voltage-clamp recordings were obtained at a holding poten-tial of –70 mV. Dual-component mEPSCs were measured in nominallyMg2+-free external medium, supplemented with 10 µM glycine. Record-ings were filtered at 2 kHz and acquired at 10 kHz. Analysis of mEPSCswas done using custom-written programs in Microsoft Visual Basic.

Imaging. For the imaging experiments, we used a closed, small-volume,custom-built laminar flow chamber. Neurons were viewed using a con-focal microscope (Zeiss LSM510) with a water-immersion lens (×40, 0.8 NA, Zeiss, Thornwood, New York). The extracellular perfusion medi-um was the same as above. A Grass SD9 stimulator (Astro-Med, WestWarwick, Rhode Island) was used to evoke action potentials, using briefvoltage pulses (1 ms, 20–50 V, bipolar) applied to platinum wires sepa-rated by ∼ 8 mm.

Synaptic vesicles were labeled with the styryl dye FM4-64 (MolecularProbes, Eugene, Oregon) by bathing the cultures in high-potassium medi-um (40 mM KCl replacing an equal molarity of NaCl) containing 10 µMdye for 90 s. Cells were then perfused with dye-free solution for 10 min toremove surface dye. For experiments examining the rate of vesicle exo-cytosis, synapses were labeled using electrical stimulation. Loading stim-ulus was 300 action potentials delivered at 10 Hz, and dye was presentfor an additional 60 s. After a 10-min wash in dye-free solution, the rateof exocytosis was measured by imaging synapses before and after deliveryof 50 action potentials at 1 Hz. The total releasable dye was measured bydelivering 3,000 action potentials at 10 Hz.

We used the 488-nm argon line to excite EGFP, and emitted light wascollected in the 500–550-nm band. Excitation laser power was adjusted toa low level that still gave us a good signal-to-noise ratio; the power exit-ing the objective was ∼ 5 µW. Scan speed was set to achieve a pixel dwelltime of around 2 µs. Images were typically 512 × 512 pixels, encompass-ing an area of 29 × 29 µm2. The pinhole was set to be around 5 Airy disksin diameter to obtain a bigger depth of field. Photobleaching was achievedby transiently increasing the excitation power to 100% (∼ 2 mW), andscanning only the spine of interest 5 times. To examine rapid changes influorescence after photobleaching (for example, in experiments withEGFP), small regions (about 5 µm × 5 µm) were selected and scannedat ∼ 20 Hz.

Data analysis. Images were analyzed using custom-written routines inMATLAB (MathWorks Inc, Natick, Massachusetts). Regions of interest(ROI) were drawn around spines and the average intensity was calculat-ed. Background fluorescence was measured from neighboring regionsand subtracted. Data from each spine was then normalized to the pre-bleach levels. FM4-64 analysis was identical to that in previously pub-lished work35. All data are presented as mean ± standard error of themean.

Note: Supplementary Movie 1 and Supplementary Methods are available on the

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

AcknowledgementsWe thank K. Heiberger for maintaining the gsn–/– mouse colony, M. Meister and

A. M. Craig for their comments on an early version of the manuscript, and Z. Li

for assistance in some of the experiments. The work was supported by startup

funds from Harvard University and a grant from the NIH. VNM is a Sloan

Foundation Fellow, a Pew Scholar, an EJLB Foundation Scholar and a NARSAD

Young Investigator.

Competing interest statementThe authors declare that they have no competing financial interests.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

246 nature neuroscience • volume 5 no 3 • march 2002

articles

RECEIVED 26 OCTOBER 2001; ACCEPTED 24 JANUARY 2002

1. Cajal, S. R. Y. On the structure of the cerebral cortex in certain mammals. LaCellule 7, 125–176 (1891).

2. Matus, A. Actin-based plasticity in dendritic spines. Science 290, 754–758(2000).

3. Harris, K. M. Structure, development, and plasticity of dendritic spines. Curr.Opin. Neurobiol. 9, 343–348 (1999).

4. Engert, F. & Bonhoeffer, T. Dendritic spine changes associated withhippocampal long-term synaptic plasticity. Nature 399, 66–70 (1999).

5. Maletic-Savatic, M., Malinow, R. & Svoboda, K. Rapid dendriticmorphogenesis in CA1 hippocampal dendrites induced by synaptic activity.Science 283, 1923–1927 (1999).

6. Segal, M. Rapid plasticity of dendritic spine: hints to possible functions? Prog.Neurobiol. 63, 61–70 (2001).

7. Halpain, S., Hipolito, A. & Saffer, L. Regulation of F-actin stability indendritic spines by glutamate receptors and calcineurin. J. Neurosci. 18,9835–9844 (1998).

8. Rao, A. & Craig, A. M. Signaling between the actin cytoskeleton and thepostsynaptic density of dendritic spines. Hippocampus 10, 527–541 (2000).

9. Fischer, M., Kaech, S., Knutti, D. & Matus, A. Rapid actin-based plasticity indendritic spines. Neuron 20, 847–854 (1998).

10. Dunaevsky, A., Tashiro, A., Majewska, A., Mason, C. & Yuste, R.Developmental regulation of spine motility in the mammalian centralnervous system. Proc. Natl. Acad. Sci. USA 96, 13438–13443 (1999).

11. Korkotian, E. & Segal, M. Spike-associated fast contraction of dendriticspines in cultured hippocampal neurons. Neuron 30, 751–758 (2001).

12. Matus, A. Postsynaptic actin and neuronal plasticity. Curr. Opin. Neurobiol. 9,561–565 (1999).

13. Pollard, T. D., Blanchoin, L. & Mullins, R. D. Molecular mechanismscontrolling actin filament dynamics in nonmuscle cells. Ann. Rev. Biophys.Biomol. Struct. 29, 545–576 (2000).

14. Condeelis, J. How is actin polymerization nucleated in vivo? Trends Neurosci.11, 288–293 (2001).

15. Wegner, A. Head-to-tail polymerization of actin. J. Mol. Biol. 108, 139–150(1976).

16. Nakayama, A. Y., Harms, M. B. & Luo, L. Small GTPases Rac and Rho in themaintenance of dendritic spines and branches in hippocampal pyramidalneurons. J. Neurosci. 20, 5329–5338 (2000).

17. Tashiro, A., Minden, A. & Yuste, R. Regulation of dendritic spine morphologyby the rho family of small GTPases: antagonistic roles of Rac and Rho. Cereb.Cortex 10, 927–938 (2000).

18. Fischer, M., Kaech, S., Wagner, U., Brinkhaus, H. & Matus, A. Glutamatereceptors regulate actin-based plasticity in dendritic spines. Nat. Neurosci. 3,887–894 (2000).

19. Colicos, M. A., Collins, B. E., Sailor, M. J. & Goda, Y. Remodeling of synapticactin induced by photoconductive stimulation. Cell 107, 605–616 (2001).

20. Axelrod, D., Knoppel, J., Schlessinger, E., Elson, E. & Webb, W. W. Mobilitymeasurement by analysis of fluorescence photobleaching recovery kinetics.Biophys. J. 16, 1055–1069 (1976).

21. Wang, Y. L. Exchange of subunits at the leading edge of living fibroblasts:possible role of treadmilling. J. Cell Biol. 101, 597–602 (1985).

22. Theriot, J. A. & Mitchison, T. J. Actin microfilament dynamics in locomotingcells. Nature 352, 126–131 (1991).

23. McGrath, J. L., Tardy, Y., Dewey, C. F. Jr., Meister, J. J. & Hartwig, J. H.Simultaneous measurements of actin filament turnover, filament fraction,and monomer diffusion in endothelial cells. Biophys. J. 75, 2070–2078(1998).

24. Sund, S. E. & Axelrod, D. Actin dynamics at the living cell submembraneimaged by total internal reflection fluorescence photobleaching. Biophys. J.79, 1655–1669 (2000).

25. Cooper, J. A. Effects of cytochalasin and phalloidin on actin. J. Cell Biol. 105,1473–1478 (1987).

26. Bubb, M. R., Senderowicz, A. M., Sausville, E. A., Duncan, K. L. & Korn, E. D.Jasplakinolide, a cytotoxic natural product, induces actin polymerization and

competitively inhibits the binding of phalloidin to F-actin. J. Biol. Chem. 269,14869–14871 (1994).

27. Murthy, V. N., Schikorski, T., Stevens, C. F. & Zhu, Y. Inactivity producesincreases in neurotransmitter release and synapse size. Neuron 32, 673–682(2001).

28. Goda, Y. & Stevens, C. F. Long-term depression properties in a simple system.Neuron 16, 103–111 (1996).

29. Carroll, R. C., Lissin, D. V., von Zastrow, M., Nicoll, R. A. & Malenka, R. C.Rapid redistribution of glutamate receptors contributes to long-termdepression in hippocampal cultures. Nat. Neurosci. 2, 454–460 (1999).

30. Yin, H. L. & Stossel, T. P. Control of cytoplasmic actin gel–sol transformation bygelsolin, a calcium-dependent regulatory protein. Nature 281, 583–586 (1979).

31. Yin, Y. L., Albrecht, J. H. & Fattoum, A. Identification of gelsolin, a Ca2+-dependent regulatory protein of actin gel–sol transformation and itsintracellular distribution in a variety of cells and tissues. J. Cell Biol 91,901–906 (1981).

32. Kinosian, H. J. et al. Ca2+ regulation of gelsolin activity: binding and severingof F-actin. Biophys. J. 75, 3101–3109 (1998).

33. Furukawa, K. et al. The actin-severing protein gelsolin modulates calciumchannel and NMDA receptor activities and vulnerability to excitotoxicity inhippocampal neurons. J. Neurosci. 17, 8178–8186 (1997).

34. Witke, W. et al. Hemostatic, inflammatory, and fibroblast responses areblunted in mice lacking gelsolin. Cell 81, 41–51 (1995).

35. Murthy, V. N., Sejnowski, T. J. & Stevens, C. F. Heterogeneous releaseproperties of visualized individual hippocampal synapses. Neuron 18,599–612 (1997).

36. Halpain, S. Actin and the agile spine: how and why do dendritic spines dance?Trends Neurosci. 23, 141–146 (2000).

37. Allison, D. W., Gelfand, V. I., Spector, I. & Craig, A. M. Role of actin inanchoring postsynaptic receptors in cultured hippocampal neurons:differential attachment of NMDA versus AMPA receptors. J. Neurosci 18,2423–2436 (1998).

38. Zhang, W. & Benson, D. L. Stages of synapse development defined bydependence on F-actin. J. Neurosci. 21, 5169–5181 (2001).

39. Kim, C. H. & Lisman, J. E. A role of actin filament in synaptic transmissionand long-term potentiation. J. Neurosci. 19, 4314–4324 (1999).

40. Krucker, T., Siggins, G. R. & Halpain, S. Dynamic actin filaments are requiredfor stable long-term potentiation (LTP) in area CA1 of the hippocampus.Proc. Natl. Acad. Sci. USA 97, 6856–6861 (2000).

41. Zhou, Q., Xiao, M. Y. & Nicoll, R. A. Contribution of cytoskeleton to theinternalization of AMPA receptors. Proc. Natl. Acad. Sci. USA 98, 1261–1266(2001).

42. McGrath, J. L., Osborn, E. A., Tardy, Y. S., Dewey, C. F. Jr. & Hartwig, J. H.Regulation of the actin cycle in vivo by actin filament severing. Proc. Natl.Acad. Sci. USA 97, 6532–6537 (2000).

43. Janmey, P. A. et al. Interactions of gelsolin and gelsolin–actin complexes withactin. Effects of calcium on actin nucleation, filament severing and endblocking. Biochemistry 24, 3714–3723 (1985).

44. Janmey, P. A. & Stossel, T. P. Modulation of gelsolin function byphospatidylinositol-4,5-bisphosphate. Nature 325, 362–364 (1987).

45. Pak, D. T., Yang, S., Rudolph-Correia, S., Kim, E. & Sheng, M. Regulation ofdendritic spine morphology by SPAR, a PSD-95-associated RapGAP. Neuron31, 289–303 (2001).

46. Sala, C. et al. Regulation of dendritic spine morphology and synapticfunction by Shank and Homer. Neuron 31, 115–130 (2001).

47. Scott, E. K. & Luo, L. How do dendrites take their shape? Nat. Neurosci 4,359–365 (2001).

48. Arcaro, A. The small GTP-binding protein Rac promotes the dissociation ofgelsolin from actin filaments in neutrophils. J. Biol. Chem. 273, 805–813(1998).

49. Azuma, T., Witke, W., Stossel, T. P., Hartwig, J. H. & Kwiatkowski, D. J.Gelsolin is a downstream effector of rac for fibroblast motility. EMBO J. 17,1362–1370 (1998).

50. Xia, Z., Dudek, H., Miranti, C. K. & Greenberg, M. E. Calcium influx via theNMDA receptor induces immediate early gene transcription by a MAPkinase/ERK-dependent mechanism. J Neurosci 16, 5425–5436 (1996).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Neurons represent complex computing structures with numerousdendritic compartments that interact in a dynamic fashion1. Acrucial factor in the compartmentalization of a neuron is the spa-tial distribution of its different excitatory synaptic inputs2. Spe-cific spatial arrangements on single cells have also been reportedfor inhibitory inputs, even though their functional implicationson a systems level are obscure3,4. A model system to study thefunctional significance of compartmentalization, includinginhibitory components, is the neural circuit that allows temporalprocessing in the microsecond range in the context of soundlocalization. High-frequency sounds produce substantial soundpressure differences at the two ears. In contrast, low frequenciesdo not produce significant interaural sound pressure differencesand can be localized only by means of interaural time differences(ITD)5, which are in the range of just microseconds. In mam-mals with well developed low-frequency hearing, neurons of themedial superior olive (MSO) adapted to encode microsecondITDs6,7, a temporal resolution beyond that observed in othermammalian brain circuits. Accordingly, the MSO of these ani-mals shows specific structural characteristics not seen in mam-mals with only high-frequency hearing8. For instance, neuronsin the MSO of mammals that can localize low-frequency soundsby means of ITD coding (primates, carnivores and many desertrodents) are spatially highly ordered so that the cell bodies arealigned in one parasagittal plane (Fig. 1). In mammals such asrats and bats, which are not adapted to hear low frequencies, theMSO does not show such a highly ordered anatomical arrange-ment of neurons and their ITD resolution is comparably low9,10.

MSO neurons receive binaural excitatory inputs from thecochlear nuclei11 and binaural glycinergic inhibitory inputs viathe medial and lateral nuclei of the trapezoid body8,12,13 (MNTBand LNTB, respectively; Fig. 1a). ITD processing relies on precisecoincidence detection of the binaural excitatory inputs. Com-partmentalization of excitatory synaptic inputs has been sug-

Experience-dependent refinementof inhibitory inputs to auditorycoincidence-detector neurons

Christoph Kapfer1, Armin H. Seidl1, Hermann Schweizer2 and Benedikt Grothe1

1 Max-Planck-Institute of Neurobiology, Am Klopferspitz 18a, 82152 Martinsried, Germany2 Zoologisches Institut der Universität München, Luisenstrasse 14, 80333 München, Germany

Correspondence should be addressed to B.G. (bgrothe@neuro.mpg.de)

Published online: 19 February 2002, DOI: 10.1038/nn810

The spatial arrangement of inputs on to single neurons is assumed to be crucial in accurate signalprocessing. In mammals, the most precise temporal processing occurs in the context of sound local-ization. Medial superior olivary neurons can encode microsecond differences in the arrival time oflow-frequency sounds at the two ears. Here we show that in mammals with well developed low-frequency hearing, a spatial refinement of ionotropic inhibitory inputs occurs on medial superior oli-vary neurons during development. This refinement is experience dependent and does not developin mammals that do not use interaural time differences for sound localization.

gested to be crucial for achieving the high temporal precision ofcoincidence detection for ITD processing14,15. MSO neurons incats and gerbils, both low frequency–hearing mammals, show ahigh ITD resolution7,16. Indeed, their excitatory inputs are spa-tially segregated in that ipsilateral inputs innervate lateral den-drites and contralateral inputs innervate medial dendrites17,18.Less is known about the function and the topographic arrange-ment of inhibitory projections to the MSO. However, MNTBand LNTB, which give rise to the binaural glycinergic inputs,show unique morphological and physiological specializationsthat allow for highly precise temporal transmission19,20. Thereis also evidence that their accurate timing is important for tem-poral processing in the MSO10,13. This raises the question ofwhether inhibitory inputs on MSO neurons in mammals thatuse ITDs for localization of low-frequency sounds also show aspecific spatial arrangement. Therefore, we compared the dis-tribution of glycinergic inputs on MSO neurons and their devel-opment in the Mongolian gerbil (Meriones unguiculatus), adesert rodent that uses ITDs to localize low-frequencysounds16,21, to that of mammals that hear only high frequencies(rats, bats and short-tailed opossums).

RESULTSOn MSO neurons, we analyzed the distribution of presynapticglycine, postsynaptic glycine receptors and the glycine-receptoranchoring protein gephyrin, which occurs only at inhibitorypostsynaptic sites22. To reliably identify the dendrites of MSOneurons, we double-labeled sections stained with antibodiesagainst glycine (gerbil), against glycine receptor (gerbil and rat)and against gephyrin (rat) using an antibody against a dendrit-ic marker, microtubule-associated protein-2 (MAP2). We alsoinvestigated the existence of inhibitory synapses labeled bygephyrin antibody staining on gerbil MSO neurons at the elec-tron microscopic (EM) level.

articles

nature neuroscience • volume 5 no 3 • march 2002 247

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 2. Glycinergic inputs are confined to somata in adultgerbil MSO but not in juvenile and cochlear-ablated gerbils.(a) In adult gerbils, glycine (red) is restricted to the somataand proximal parts of dendrites (MAP2, blue). (b, c) Adultgerbil MSO: gephyrin is restricted to the somata and themost proximal dendrites. The dendritic neuropil is devoidof staining. (d) Juvenile MSO (before hearing onset): glycinesurrounds somata and dendrites. (e, f) Juvenile MSO:gephyrin is uniformly distributed on somata and dendrites.(g) Unilateral cochlear-ablated animals (UCA; ablatedbefore hearing onset): glycine surrounds cell bodies anddendrites. (h) Gephyrin is present on somata and dendritesin UCA animals. Filled arrows, somatic labeling; openarrows, dendritic labeling. Scale bars, 20 µm.

248 nature neuroscience • volume 5 no 3 • march 2002

Distribution in adult and juvenile gerbilsIn adult gerbils, glycine-positive staining was prominent onthe cell bodies and the proximal regions of the dendrites butwas absent on more distal regions of the dendrites (Fig. 2a;quantifications see below). Similarly, MSO cell bodies andproximal dendrites were outlined by gephyrin-positive puncta,whereas the surrounding neuropil was almost devoid of stain-ing (Fig. 2b). Semi-thin sections showed the concentration ofgephyrin puncta on the soma membranes of single cells butonly few puncta on dendrites, mainly on the most proximaldendritic compartments (Fig. 2c). Glycine receptor puncta alsoappeared in high concentration on the cell bodies and the mostproximal dendritic compartments (Figs. 1b and 3a) but not onmore distal dendrites. Also, EM sections showed many stainedsynapses on the MSO cell bodies (Fig. 4a) but virtually nostained synapses on dendrites (Fig. 4b).

We found a very different pattern of glycinergic innerva-tion in juvenile gerbils at postnatal day 10 (P10), 2 days beforehearing onset. At this age, glycinergic transmission in the supe-rior olive is functional and inhibitory23,24. Intensestaining for presynaptic glycine was visible on thecell bodies and on the entire dendrites (Fig. 2d).Similarly, both cell bodies and dendrites were heav-ily labeled by gephyrin-positive puncta (Fig. 2e and f). EM sections also showed many gephyrin-stained synapses on cell bodies and dendrites (Fig. 4c). Typically, excitatory (unstained) andinhibitory (stained) synapses were found on thesame cross-sections of single dendrites (Fig. 4d).

Experience-dependent developmentNext we asked whether the observed ontogenetic refinement ofglycinergic innervation requires an active selection of inputs and,thus, whether it depends on the experience of binaural auditorystimulation. Therefore, we unilaterally ablated one cochlea (UCA)in gerbils 5 days before hearing onset (P7), preventing MSO neu-rons from experiencing correlated binaural inputs25. Cochlearablations at this age have been shown to cause some cell loss inthe contralateral MNTB but simultaneously induce invasion ofcochlear nucleus fibers from the intact side18. Thus, both MNTBsare driven by the same ear, ‘monauralizing’ the system. Animalswere killed at P25 or older. Successful ablations were confirmedanatomically. In these animals, labeling against glycine and MAP2(Fig. 2g), gephyrin (Fig. 2h), and glycine receptors and MAP2(Fig. 3b and c) all resulted in prominent staining on the cell bod-ies, as well as on the proximal and distal dendrites, as seen in juve-nile animals. This was independent of the survival time (P25 up toseveral months; Fig. 3b and c, respectively) and the location ofthe neurons ipsilateral or contralateral to the ablation. To testwhether these effects are due to a change in the overall activity ofthe system or to the lack of experience of specific binaural cues,we raised gerbils from P10 to P25 in a sound-attenuated cham-ber and exposed them to omnidirectional white noise of moder-ate intensity (60 dB SPL). The noise does not eliminate all binauralcues (particularly in the near field) but highly reduces them26. Ingerbils raised in noise (NRA), we found a moderately but signif-icantly (see below) higher density of glycine receptor puncta atthe dendrites compared to control animals (Fig. 3d). EM con-firmed the existence of fully developed inhibitory synapses notonly on somata (Fig. 4e) but also on dendrites (Fig. 4f).

articles

Higher auditory centers

Ipsilateral

MSO

LNTBMNTB

CN

CN

ContralateralCochlea Cochlea

LNTB

MNTB

MSO

Adult gerbil

a

b

Fig. 1. Arrangement of functional inputs to the MSO. (a) Binaural exci-tation (black arrows) derives from cochlear nuclei (CN), binaural glycin-ergic inhibition (red arrows) from MNTB and LNTB. Dashed linerepresents midline. (b) Outline of auditory nuclei in the gerbil brain-stem. Inset: adult gerbil MSO, labeled with antibodies against MAP2(blue) and glycine receptor (yellow). Scale bar, 100 µm.

Adult gly/MAP2 Juvenile gly/MAP2 UCA gly/MAP2

Adult gephyrin Juvenile gephyrin UCA gephyrin

Juvenile gephyrinAdult gephyrin

a

b

c

d

e

f

g

h

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 4. Electron micrographs showing ultrastructural localizationof inhibitory, glycinergic synapses on gerbil MSO cell bodies (CB)and cross-sectioned dendrites (D). Glycinergic synapses arerevealed by postsynaptic gephyrin staining. (a) In sections of adultgerbil MSO cell bodies, numerous synaptic contacts appear to beglycinergic (white arrows). (b) On dendrites, virtually none of thepostsynaptic sites is stained for gephyrin; terminals are presum-ably excitatory (black arrows). (c) As in adults, synapses on cellbodies of juvenile animals are predominantly glycinergic. (d) Onjuvenile MSO, dendrites excitatory and inhibitory synapses areequally prevalent. (e, f) Sections of NRA animals show gephyrin-stained synapses on (e) cell bodies and (f) dendrites. Blackarrows, unstained (excitatory) synapses; white arrows, gephyrin-stained (inhibitory) synapses. Scale bars, 1 µm.

QuantificationsWe quantified the number of presynapticglycinergic boutons on individual cells bynormalizing them to the number of punctaon the soma membrane of each cell. In adultgerbils, the ratio dropped significantly (p < 0.001) within thefirst 20 µm of the proximal dendrites and decreased to evenlower values further distal (Fig. 5a). In contrast, for juvenile andUCA animals, the ratio remained at similar levels throughoutthe dendrites (there was no difference between MSO cells ipsi-and contralateral to the ablation; data not shown). The func-tions differed significantly (p < 0.001) for adult versus juvenileand UCA animals. Note that the ratios are independent of pos-sible changes in dendritic arborization. To quantify gephyrin(Fig. 5b) and glycine-receptor staining (Fig. 5c), puncta werecounted along the mediolateral axis of the MSO. For adult ani-mals, there was a high density of gephyrin puncta in areas thatincluded the cell bodies, but decreased for areas medial and lat-

Fig. 3. Glycine receptor distribution on gerbil MSOneurons. (a) Adults: glycine receptors (yellow) areconcentrated on somata; on dendrites (anti-MAP2,blue) only few puncta appear. (b, c) Cochlear-ablated animals (UCA): glycine receptors areprominent on somata and dendrites. (d) Animalsraised in omnidirectional noise (NRA): glycinereceptors cover somata and dendrites. Filledarrows, somatic labeling; open arrows, dendriticlabeling. Scale bars, 20 µm.

eral to that of the cell bodies. Even though there was a declinein the density functions for juvenile and UCA animals, thisdecline was less pronounced. Functions for adults differed sig-nificantly from those for juvenile and UCA animals (p < 0.001).Similarly, significant differences were found for the glycine recep-tor distributions from adult and UCA animals (p < 0.001). Thedifference from adult to NRA animals was, as expected, muchsmaller but still significant (p < 0.002). Counting glycine recep-tor puncta on individual dendrites resulted in similar, signifi-cant differences (data not shown).

We also compared the absolute density of glycinergicsynapses on adult, juvenile and UCA gerbil MSO somata indiaminobenzidine (DAB)-stained tissue. Comparing juveniles

and normal adults, there was no significant proliferationof glycinergic puncta (Fig. 5d). In UCA animals, however,the number of puncta per µm of cell membrane signifi-cantly increased (p < 0.001). The cell perimeter of MSOcells was similar in normal adults and UCA and only slight-ly smaller in juvenile (P10) animals (Fig. 5e). Thus, changesof puncta density on the somata cannot account for thechanges in synapse distributions described above.

Mammals with only high-frequency hearingIf the observed refinement has functional significance forITD coding, one might not expect to see it in mammalswith poor or no low-frequency hearing. Hence, we exam-ined the MSO in three such species: rats, short-tailedopossums and bats. In all these animals, the arrangementof MSO neurons is less structured; also, dendrites aremuch thinner, particularly in bats, and the MSO is verysmall in the opossum. This makes it difficult to quantifythe distribution of glycinergic inputs in the MSO in opos-sums and bats. For all of the MSO neurons that included

articles

nature neuroscience • volume 5 no 3 • march 2002 249

d

a Adult

UCA P25 NRAc UCA 18 months

b UCA P25

a b

c d

e f

CB

D

D

DCB

CB

CB

D

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

250 nature neuroscience • volume 5 no 3 • march 2002

the cell body and the bipolar dendrites, however, staining forglycine, gephyrin and glycine receptors was prominent onsomata and dendrites (Fig. 6a–c; for bats compare ref. 27). Wequantified the distribution of presynaptic glycine in adult ratsand postsynaptic gephyrin puncta in adult and juvenile rats(P10, 2 days before hearing onset). The absolute density ofglycine puncta on adult rat MSO somata was 0.32 puncta perµm cell membrane (s.e.m. 0.013), statistically indistinguish-able from that found in adult gerbils. Soma perimeters in therat were also similar to the gerbil MSO (69.0 µm; s.e.m. 1.6).However, the relative glycine and gephyrin puncta density inrats was even higher on dendrites than on somata (Fig. 6d).The distribution of postsynaptic gephyrin closely matches thatof the presynaptic glycine. Hence, inhibitory inputs to the ratMSO do not show the restriction to the somata foundin adult gerbils (Fig. 6d). The distribution of gephyrinin juvenile rats does not significantly differ from thatin adult rats (Fig. 6d). Thus, the spatial restriction ofglycinergic inputs to MSO cell bodies occurs in thegerbil, which uses ITDs, but not in rats and, most like-ly, also not in bats (data not shown) or in short-tailedopossums (Fig. 6e), the latter three all species that donot use ITDs for sound localization.

DISCUSSIONOur results show that inhibitory inputs to gerbil MSO neuronsare confined to the somata. This is consistent with the distribu-tion of flat vesicles in presynaptic terminals in the MSO of cats28

and chinchillas29. All three species are well adapted for low-fre-quency hearing, and their MSO neurons show a high ITD reso-lution8,16,30. Our data show, however, that this confinement doesnot occur in rats or, apparently, in other mammals that are alsonot specialized for low-frequency hearing and are unlikely to useITDs (such as bats and opossums). Thus, the confinement ofinputs seems to represent a specific adaptation for ITD codingin the submillisecond range. Such a refinement could minimizetemporal summation of inhibition and thereby sharpens thekinetics of the inhibition28. Direct proof of a difference in thekinetics of the glycine-mediated inhibition in these differentspecies is not available; pharmacological evidence that the tim-ing of glycinergic inhibition matters for ITD processing in ger-bil MSO neurons comes only from an in vitro study13; and howexactly well timed inhibition could be involved in shaping ITDfunctions has been assessed only theoretically10. Thus, how therefinement of inhibitory inputs contributes to better ITD cod-ing remains to be shown. Notably, the missing confinement ofinputs correlates not only with audiograms but also with a sig-

articles

Fig. 6. Glycinergic inputs on MSO neurons of only high fre-quency–hearing animals. (a) Rat MSO neuron exhibitinggephyrin labeling (green) on soma (filled arrows) and den-drites (open arrows). MAP2 counterstaining in blue. (b) Anti-gephyrin puncta (DAB staining) on soma and den-drites of an opossum bipolar neuron in the medial portion ofthe superior olivary complex. (c) Glycine-receptor staining(yellow) on soma and dendrites of a rat MSO neuron (coun-terstained with MAP2, blue). (d) Quantification of the spatialarrangement of presynaptic glycine and postsynapticgephyrin in adult and juvenile (P10) rats. Relative to thesomata, the density of glycine and gephyrin is higher andhomogeneous throughout the extent of the dendrites inadult rats. The distribution of gephyrin is identical in juvenileand adult rats. (e) Arrangement of excitatory and inhibitorysynapses on MSO cells of only high frequency–hearing com-pared to low frequency–hearing mammals. Scale bars, 20 µm.

Fig. 5. Quantification of the spatial arrangement of glycinergic inputs ongerbil MSO neurons. (a) Glycine puncta ratio per µm cell membrane ofindividual neurons drops considerably within the first proximal 20 µm ofthe dendrites in adult, but not in juvenile and cochlear-ablated gerbils. (b) Density of gephyrin in the dendritic neuropil declines significantly withincreasing distance from the soma in adults. Functions of juvenile and UCAanimals remain on an overall higher level. (c) Density of glycine receptorpuncta drops significantly after the first dendritic segment in adults but notin animals with altered auditory experience (UCA, NRA). (d) Glycinepuncta per µm cell membrane on MSO cell bodies. Rations are similar forjuvenile and adult gerbils but increased in cochlear-ablated animals. (e) Averaged soma perimeters do not significantly differ for juvenile, adultand cochlear ablated animals. Error bars, standard error of mean.

Nor

m.

punc

ta /a

rea

Distance from soma (µm)

c

20 40 60 80 1000.0

0.5

1.0Glycine receptor

120140

a

0.0

0.5

1.0

1.5 Glycine

0 20 40 60 80 100

Nor

m.

punc

ta / µ

mb

Gephyrin

0.0

0.5

1.0

26 52 78 104 130

Adult gerbil

Juvenile gerbil

UCA gerbil

NRA gerbil

Nor

m.

punc

ta /a

r ea

0.0

0.5

1.0

Adul

tJu

veni

le

UC

A

Av.

no.

of

punc

ta/µ

m

d e

0

50

100

Adul

tJu

veni

le

UC

AAv.

som

aper

imte

r (µ

m)

c

C

a b

Opossum

Rat

Rat

MSO

Rat, opossum, bat, juvenile gerbil

MSO

Adult gerbil, cat, chinchilla

ed

Distance from soma (µm)0 20 40 60 80 100 120

0.0

0.5

1.0

1.5

2.0

Nor

m.

punc

ta /

µm

Adult rat (glycine)

Juvenile rat (gephyrin)

Adult rat (gephyrin)

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

nificantly lower ITD resolution that has been shown in vivo inMSO neurons of rats9 and bats10. How somatic inhibition couldsharpen the coincidence detection of excitatory inputs has recent-ly been shown in hippocampal pyramidal cells31. There, feed-forward GABAergic inhibition curtails the time window allowedfor summation of excitation. Compared to ITD-related coinci-dence detection, however, the timescale in the hippocampus isorders of magnitudes larger.

The present study also shows that the spatial distribution ofglycinergic inputs to gerbil and rat MSO neurons is initially dif-fuse. In gerbils, but not in rats, it undergoes a substantial refine-ment within the first days after hearing onset. Neither synapseproliferation on the somata nor changes in dendritic arboriza-tion can account for the observed refinement. Thus, the devel-opmental changes are due to synapse elimination. Therefinement does not occur after cochlear ablations that preventsynchronous bilateral input to the MSO and is substantiallyreduced by partial masking of spatial acoustic cues by omnidi-rectional white noise. We do not know what the absolute timewindow for the experience-dependent development might be.The data from two adult gerbils (data not shown) exposed toomnidirectional white noise indicate, however, that the noisehas no effect during adulthood. Also, unilateral ablations in adultgerbils do not show specific effects on the distribution of glycin-ergic inputs to the MSO32. Thus, the effects of the ablation orthe masking of binaural cues by noise exposure seem to inter-fere with specific developments after hearing onset, indicatingthat coherence of binaural inputs is the driving force for therefinement.

Activity-dependent elimination of excitatory synaptic inputsis well known during development of other systems such as theneuromuscular junction33,34, the visual cortex35,36, the cerebel-lum37,38 and the somatosensory system39. In contrast, much lessis known about activity-dependent selection of inhibitory inputs.In the auditory system, activity-dependent selection of inhibito-ry inputs has been shown for the glycinergic MNTB projectionsto the lateral superior olive25,40–42. This selection, however, con-cerns connections to entire populations of neurons but not dis-tinct compartments of single cells. Refined inhibitory inputs atthe single-cell level have been reported for other systems, suchas the hippocampus43, visual cortex44 and spinal cord45, but therethe significance of the refinement remains unclear.

A possible mechanism underlying the confinement duringdevelopment could simply be a removal of inhibitory inputscaused by competition with excitatory inputs. However, thiswould explain only why inhibitory synapses are removed on thedendrites but not why and how they are maintained on the cellbody. In addition, this explanation does not hold if the exact tem-poral relationship of inhibitory and excitatory synaptic activa-tion is important. Thus, explaining the confinement of inhibitoryMSO inputs by an active mechanism that selectively strengthensappropriately timed synapses and eliminates those activated tooearly or too late is compelling. For depolarizing excitatory synaps-es, several mechanisms based on coincident pre- and postsynap-tic activity have been proposed46,47. For fast ionotropic inhibitorysynapses, long-term depression has been suggested as a possiblemechanism that weakens synaptic strength48 and, in conjunctionwith mechanisms yet unknown, may finally cause synapse elim-ination.

METHODSGlycine immunohistochemistry. Mongolian gerbils (M. unguiculatus):n = 8 adult, 4 juvenile and 8 UCA animals. Rat (Wistar): n = 2 adult

animals. Free-tailed bats (Tadarida brasiliensis): n = 2 adult animals.Animals were perfused with Ringers solution containing 0.02% heparin(2 min) followed by 4% paraformaldehyde and 0.1% glutaraldehyde(30 min). For fluorescent staining and double labeling with MAP2,brains were cryoprotected, frozen in dry ice and cut at 25 µm on a cryo-stat. Sections were incubated in 0.38% sodium borohydride solution,permeabilized with 0.3% Triton X-100 and immunostained with a poly-clonal antiserum specific to glycine (SFRI Laboratoire, Saint Jean D’Il-lac, France) and a monoclonal antibody specific to MAP2 (clone HM-2,Sigma, Deisenhofen, Germany). Secondary antibodies included rabbit-specific antibody conjugated to Cy3 (Dianova, Hamburg, Germany) andmouse-specific antibody conjugated to Alexa 488 (Molecular Probes,Eugene, Oregon). For avidin-biotin-DAB staining, 25-µm sections werecut on a vibratome and treated as described above; the secondary anti-body was rabbit-specific antibody conjugated to biotin (Dianova).

Glycine receptor immunohistochemistry. Gerbil: n = 5 adult, 3 juvenile,3 UCA and 7 NRA animals. Rat (adult): n = 3. After perfusion with arti-ficial cerebrospinal fluid, brains were removed and frozen in liquid nitro-gen. Cryostat sections (25 µm) were thaw-mounted on slides and fixed in4% paraformaldehyde (5 min). Sections were permeabilized with 0.5%Triton X-100 and immunostained with a polyclonal antibody recogniz-ing the glycine receptor α1 subunit (Chemicon International, Temecula,California) and the monoclonal antibody to MAP2 (Sigma). Secondaryantibodies were as described for glycine immunohistochemistry.

Gephyrin immunohistochemistry. Gerbil: n = 16 adult, 11 juvenile, 8UCA and 1 NRA animal. Rat: n = 4 adult and 4 juvenile animals. Short-tailed opossum (Monodelphis domestica): N = 1 adult animal. For fluo-rescent staining, animals were perfused with Ringers solution containing0.02% heparin (2 min) followed by 4% paraformaldehyde (30 min).After cryoprotection, brains were frozen in dry ice and cut at 25 µm ona cryostat. Sections were permeabilized with 0.05% Triton X-100. Sec-tions from gerbil brains were immunostained with a monoclonal anti-body to gephyrin (MAb 7a, Alexis Biochemicals, San Diego, California).Secondary antibodies were anti-mouse–biotin and streptavidin–Cy3(Dianova). To increase signal strength, a tyramide signal-amplificationsystem (NEN Life Science Products, Boston, Massachusetts) was applied,following the suggested protocol. Sections from rat brains wereimmunostained with the monoclonal antibody to gephyrin and a poly-clonal antibody to MAP2 (Santa Cruz Biotechnology, Santa Cruz, Cal-ifornia). Secondary antibodies were anti-mouse–biotin, streptavidin–Cy3(Dianova) and anti-goat–Alexa 488 (Molecular Probes). For nonfluo-rescent techniques, animals were perfused with Ringers solution fol-lowed by 4% paraformaldehyde, 0.2% glutaraldehyde and 0.2% picricacid (30 min). In gerbils, pre-embedding immunohistochemistry on75-µm vibratome sections using the monoclonal antibody againstgephyrin followed by avidin-biotin-DAB staining and silver–gold inten-sification was performed. Epoxy resin–embedded sections were resec-tioned at 8 µm on an ultramicrotome. In juvenile (P10) and adult ratsand the opossum, 25-µm vibratome sections were immunostained withthe gephyrin antibody and avidin-biotin-DAB.

Electron microscopy. Animals were perfused with Ringers solution con-taining 0.02% Heparin (2 min) followed by 4% paraformaldehyde and0.4% glutaraldehyde (30 min). Brains were post-fixed for 1 h in the samefixative and kept in 0.5 M phosphate buffer overnight at 4°C. For pre-embedding immunohistochemistry on 100 µm vibratome sections we usedmonoclonal antibodies to gephyrin and avidin-biotin-DAB labeling. Sec-tions were fixed for 7 days in 2% glutaraldehyde in PBS, osmicated, dehy-drated and embedded in araldite (Fluka, Basel, Switzerland). For lightmicroscopy, semithin sections (1 µm) were counterstained with toluidineblue. Ultrathin sections (100 nm) were counterstained with uranyl acetateand lead citrate and examined in a Zeiss EM 10 electron microscope.

Fluorescent and light microscopy. Fluorescent specimens were ana-lyzed using confocal microscopy (Leica, Wetzlar, Germany). Per spec-imen, up to 25 optical sections with 1-µm increment were analyzed.

articles

nature neuroscience • volume 5 no 3 • march 2002 251

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

252 nature neuroscience • volume 5 no 3 • march 2002

Images were digitized, adjusted, merged and false color coded usingAdobe Photoshop software (Adobe Systems, San Jose, California).DAB-stained sections were analyzed using standard light microscopy.

The specificity of all antibodies was controlled by omitting primaryor secondary antibodies. Additionally, specificity was confirmed by react-ing tissue not containing the respective antigen—for example, neocor-tex—and by reacting tissue known for strong expression of theantigen—for example, lateral superior olive, medial nucleus of the trape-zoid body and cerebellum.

Quantification. Glycine puncta density per µm cell membrane of sin-gle neurons was obtained for the cell body and compared to that ofthe dendrites (in 10-µm segments). Data were derived from analyzing87 MSO cells of 5 adult gerbils, 88 MSO cells of 2 juvenile gerbils, 81MSO cells of 4 UCA gerbils, and 50 MSO cells of 2 adult rats. All ana-lyzed cells were drawn with the aid of a drawing tube attached to themicroscope. Drawings were digitized and the somaperimeter was mea-sured with DigiTrace software (Imatec, Miesbach, Germany). Simi-larly, gephyrin puncta on individual rat MSO neurons were obtainedfrom another 2 adult and 4 juvenile rats, 50 cells each. Density of punc-ta on gephyrin– and glycine receptor–stained sections from gerbils wasobtained from rectangles of at least 40 µm height (dorso–ventral),including the area of cell bodies, and crossing the entire medial or lat-eral extension of the MSO. Rectangles were mediolaterally subdivid-ed into 26 µm (gephyrin) or 20 µm (glycine receptor) segments. Thequantity of glycine receptor puncta in the segments was counted by acomputer-based algorithm using DigiTrace software (Imatec). To com-pare tissue from many experiments with variations in staining densi-ty, data were normalized to the density of the segment that includedthe MSO somata (gephyrin) or the area with the most proximal den-dritic sections (glycine receptors). Somata showed high concentra-tions of intracellular glycine receptors and were therefore excludedfrom the analysis (however, counting puncta on individual neuronsincluding somata gave similar results). Gephyrin: 6 MSO nuclei of 3adult, 6 MSOs of 3 juvenile, 4 MSOs of 2 UCA animals. Glycine recep-tors: 8 MSOs of 4 adult, 6 MSOs of 3 UCA, 8 MSOs of 4 NRA animals.

Distributions were statistically analyzed using the Mann–Whitney U-test for two independent samples.

Noise-box. A cage with mother and litter was placed in a 100 × 80 ×80 cm3 sound-attenuated box in a quiet room. White noise from twonoise generators (Rhode & Schwarz) were presented via 12 sets ofspeakers so that each noise came from each direction simultaneously.Sound level never increased to 80 dB SPL (peak to peak). Animals werecontinuously video monitored. Noise presentation lasted from P10–25.Animals were killed on P25. All experiments were approved (Reg. Obb,AZ 2112531-40/01) according to the German Tierschutzgesetz.

AcknowledgementsWe thank U. Koch and O. Gleich for technical advice; C. Schulte, G. Breutel and

D. Büringer for technical help; and M. Götz, T. Bonhoeffer, M. Hübener, A.

Kossel, G. Neuweiler, T. Park and H. Thoenen for critical comments on the

manuscript. Supported by Max-Planck-Gesellschaft and Deutsche

Forschungsgemeinschaft (GR1205/10-1).

RECEIVED 17 SEPTEMBER 2001; ACCEPTED 22 JANUARY 2002

1. Häusser, M., Spruston, N. & Stuart, G. J. Diversity and dynamics of dendriticsignaling. Science 290, 739–744 (2000).

2. Craig, A. M., Blackstone, C. D., Huganir, R. L. & Banker, G. The distributionof glutamate receptors in cultured rat hippocampal neurons: postsynapticclustering of AMPA-selective subunits. Neuron 10, 1055–1068 (1993).

3. Lin, B., Martin, P. R., Solomon, S. G. & Grunert, U. Distribution of glycinereceptor subunits on primate retinal ganglion cells: a quantitative analysis.Eur. J. Neurosci. 12, 4155–4170 (2000).

4. Nyiri, G., Freund, T. F. & Somogyi, P. Input-dependent synaptic targeting ofα2-subunit-containing GABAA receptors in synapses of hippocampalpyramidal cells of the rat. Eur. J. Neurosci. 13, 428–442 (2001).

5. Erulkar, S. D. Comparative aspects of sound localization. Physiol. Rev. 52,237–360 (1972).

6. Goldberg, J. M. & Brown, P. B. Response of binaural neurons of dog superiorolivary complex to dichotic tonal stimuli: some physiological mechanisms ofsound localization. J. Neurophysiol. 32, 613–636 (1969).

7. Yin, T. C. & Chan, J. C. Interaural time sensitivity in medial superior olive ofcat. J. Neurophysiol. 64, 465–488 (1990).

8. Grothe, B. The evolution of temporal processing in the medial superior olive,an auditory brainstem structure. Prog. Neurobiol. 61, 581–610 (2000).

9. Inbody, S. B. & Feng, A. S. Binaural response characteristics of single neuronsin the medial superior olivary nucleus of the albino rat. Brain Res. 210,361–366 (1981).

10. Grothe, B. & Park, T. J. Sensitivity to interaural time differences in the medialsuperior olive of a small mammal, the Mexican free-tailed bat. J. Neurosci. 18,6608–6622 (1998).

11. Joris, P. X., Smith, P. H. & Yin, T. C. Coincidence detection in the auditorysystem: 50 years after Jeffress. Neuron 21, 1235–1238 (1998).

12. Cant, N. B. & Hyson, R. L. Projections from the lateral nucleus of thetrapezoid body to the medial superior olivary nucleus in the gerbil. Hear. Res.58, 26–34 (1992).

13. Grothe, B. & Sanes, D. H. Synaptic inhibition influences the temporal codingproperties of medial superior olivary neurons: an in vitro study. J. Neurosci.14, 1701–1709 (1994).

14. Agmon-Snir, H., Carr, C. E. & Rinzel, J. The role of dendrites in auditorycoincidence detection. Nature 393, 268–272 (1998).

15. Simon, J. Z., Carr, C. E. & Shamma, S.A. A dendritic model ofcoincidence detection in the avian brainstem. Neurocomputing 26–27,263–269 (1999).

16. Spitzer, M. W. & Semple, M. N. Neurons sensitive to interaural phasedisparity in gerbil superior olive: diverse monaural and temporal responseproperties. J. Neurophysiol. 73, 1668–1690 (1995).

17. Smith, P. H., Joris, P. X. & Yin, T. C. Projections of physiologicallycharacterized spherical bushy cell axons from the cochlear nucleus of the cat:evidence for delay lines to the medial superior olive. J. Comp. Neurol. 331,245–260 (1993).

18. Russell, F. A. & Moore, D. R. Afferent reorganisation within the superiorolivary complex of the gerbil: development and induction by neonatal,unilateral cochlear removal. J. Comp. Neurol. 35, 607–625 (1995).

19. Smith, P. H., Joris, P. X. & Yin, T. C. Anatomy and physiology of principalcells of the medial nucleus of the trapezoid body (MNTB) of the cat.J. Neurophysiol. 79, 3127–3142 (1998).

20. Taschenberger, H. & von Gersdorff, H. Fine-tuning an auditory synapse forspeed and fidelity: developmental changes in presynaptic waveform, EPSCkinetics, and synaptic plasticity. J. Neurosci. 20, 9162–9173 (2000).

21. Heffner, R. S. & Heffner, H. E. Sound localization and use of binaural cuesby the gerbil (Meriones unguiculatus). Behav. Neurosci. 102, 422–428(1988).

22. Kneussel, O. M. & Betz, H. Receptors, gephyrin and gephyrin-associatedproteins: novel insights into the assembly of inhibitory postsynapticmembrane specializations. J. Physiol. 525, 1–9 (2000).

23. Sanes, D. H. The development of synaptic function and integration in thecentral auditory system. J. Neurosci. 13, 2627–2637 (1993).

24. Kandler, K. & Friauf, F. Development of glycinergic and glutamatergicsynaptic transmission in the auditory brainstem of perinatal rats. J. Neurosci.15, 6890–6904 (1995).

25. Sanes, D. H., Markowitz, S., Bernstein, J. & Wardlow, J. The influence ofinhibitory afferents on the development of postsynaptic dendritic arbors.J. Comp. Neurol. 321, 637–344 (1992).

26. Withington-Wray, D. J., Binns, K. E., Dhanjal, S. S., Brickley, S. G. & Keating,M. J. The maturation of the superior collicular map of auditory space in theguinea pig is disrupted by developmental auditory deprivation. Eur. J.Neurosci. 2, 693–703 (1990).

27. Vater, M. Ultrastructural and immunocytochemical observations on thesuperior olivary complex of the mustached bat. J. Comp. Neurol. 358,155–180 (1995).

28. Clark, G. M. The ultrastructure of nerve endings in the medial superior oliveof the cat. Brain Res. 14, 293–305 (1969).

29. Perkins, R. E. An electron microscopic study of synaptic organization in themedial superior olive of normal and experimental chinchillas. J. Comp.Neurol. 148, 387–415 (1973).

30. Langford, T. L. Responses elicited from medial superior olivary neurons bystimuli associated with binaural masking and unmasking. Hear. Res. 15,39–50 (1984).

31. Pouille, F. & Scanziani, M. Enforcement of temporal fidelity in pyramidalcells by somatic feed-forward inhibition. Science 293, 1159–1163 (2001).

32. Suneja, S. K., Benson, C. G. & Potashner, S. J. Glycine receptors in adultguinea pig brain stem auditory nuclei: regulation after unilateral cochlearablation. Exp. Neurol. 154, 473–488 (1998).

33. Lichtman, J. W. & Colman, H. Synapse elimination and indelible memory.Neuron 25, 269–278 (2000).

34. Keller-Peck, C. R. et al. Asynchronous synapse elimination in neonatal motorunits: studies using GFP transgenic mice. Neuron 31, 381–394 (2001).

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

36. Chapman, B., Jacobson, M. D., Reiter, H. O. & Stryker, M. P. Oculardominance shift in kitten visual cortex caused by imbalance in retinalelectrical activity. Nature 324, 154–156 (1986).

articles©

2002

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co

m

37. Kakizawa, S., Yamasaki, M., Watanabe, M. & Kano, M. Critical period foractivity-dependent synapse elimination in developing cerebellum.J. Neurosci. 20, 4954–4961 (2000).

38. Ichise, T. et al. mGluR1 in cerebellar Purkinje cells essential for long-termdepression, synapse elimination, and motor coordination. Science 288,832–1835 (2000).

39. O’Leary, D. D., Ruff, N. L. & Dyck, R. H. Development, critical periodplasticity, and adult reorganizations of mammalian somatosensory systems.Curr. Opin. Neurobiol. 4, 535–544 (1994).

40. Sanes, D. H. & Takacs, C. Activity-dependent refinement of inhibitoryconnections. Eur. J. Neurosci. 5, 570–574 (1993).

41. Rietzel, H. J. & Friauf, E. Neuron types in the rat lateral superior olive anddevelopmental changes in the complexity of their dendritic arbors. J. Comp.Neurol. 390, 20–40 (1998).

42. Sanes, D. H. & Rubel, E. W. The ontogeny of inhibition and excitation in thegerbil lateral superior olive. J. Neurosci. 8, 682–700 (1988).

43. Craig, A. M., Blackstone, C. D., Huganir, R. L. & Banker, G. Selective

clustering of glutamate and γ-aminobutyric acid receptors opposite terminalsreleasing the corresponding neurotransmitters. Proc. Natl. Acad. Sci. USA 91,12373–12377 (1994).

44. Komatsu, Y. GABAB receptors, monoamine receptors, and postsynapticinositol trisphosphate–induced Ca2+ release are involved in the induction oflong-term potentiation at visual cortical inhibitory synapses. J. Neurosci. 16,6342–6352 (1996).

45. Alvarez, F. J., Dewey, D. E., Harrington, D. A. & Fyffe, R. E. Cell-type specificorganization of glycine receptor clusters in the mammalian spinal cord. J. Comp. Neurol. 379, 150–170 (1997).

46. Bi, G. Q. & Poo, M. M. Synaptic modifications in cultured hippocampalneurons: dependence on spike timing, synaptic strength, and postsynapticcell type. J. Neurosci. 18, 10464–10472 (1998).

47. Feldman, D. E. Timing-based LTP and LTD at vertical inputs to layer II/IIIpyramidal cells in rat barrel cortex. Neuron 27, 45–56 (2000).

48. Kotak, V. C. & Sanes, D. H. Long-lasting inhibitory synaptic depression isage- and calcium-dependent. J. Neurosci. 20, 5820–5826 (2000).

articles

nature neuroscience • volume 5 no 3 • march 2002 253

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

254 nature neuroscience • volume 5 no 3 • march 2002

articles

Rapid cooling of the skin elicits cold temperature sensations inhumans1,2. This class of somatic sensibility is mediated by theactivation of specific peripheral terminals of sensory fibers andrelayed to the CNS by anatomically distinct pathways3. Impulseactivity originating in these sensory terminals has been record-ed extracellularly in nerve filaments innervating the skin, tongue,nose and eye of different animals including primates4–9.

In contrast to recent advances in understanding the cellularbasis of sensations evoked by heat10–12, the mechanisms mediat-ing cold sensing remain poorly defined. Postulated mechanismsinclude modulation of the Na+–K+ ATPase13–15, differential tem-perature sensitivity of Na+ and K+ channels16,17 and closure ofK+ channels18,19. These temperature-dependent processes andthe postulated target molecules are present in many neurons,however, and thus can not explain the selective thermosensitivi-ty of cooling receptors. Recent preliminary evidence hints at theopening of a nonselective cation channel in cold-sensitive neu-rons during cooling20. Technical problems have also hinderedadvances in the field of cold transduction. Cold nerve endingsare very small21 and not directly accessible to intracellular record-ing22. In addition, the cell somas of cold-sensitive neurons aresparse and difficult to identify, which has prevented the charac-terization of their physiological properties.

Here we describe a subpopulation of primary sensory neu-rons excited by cooling and by the application of menthol. Exci-tation depends on closure of a background K+ conductance.These neurons express a unique set of voltage-gated ionic chan-nels, different from those found in cold-insensitive neurons.We show that this differential ion-channel expression is cru-cial to the selective excitation of these neurons during tem-perature reductions.

Specificity of cold thermotransductionis determined by differential ionicchannel expression

Félix Viana*, Elvira de la Peña and Carlos Belmonte

Instituto de Neurociencias, Universidad Miguel Hernández-CSIC, Apartado 18, San Juan de Alicante 03550, Spain

Correspondence should be addressed to F.V. (felix.viana@umh.es)

Published online: 11 February 2002, DOI: 10.1038/nn809

Sensations of cold are mediated by specific thermoreceptor nerve endings excited by lowtemperature and menthol. Here we identify a population of cold-sensitive cultured mouse trigeminalganglion neurons with a unique set of biophysical properties. Their impulse activity during coolingand menthol application was similar to that of cold thermoreceptor fibers in vivo. We show that cool-ing closes a background K+ channel, causing depolarization and firing that is limited by the slowerreduction of a cationic inward current (Ih). In cold-insensitive neurons, firing is prevented by a slow,transient, 4-AP-sensitive K+ current (IKD) that acts as an excitability brake. In addition, pharmacolog-ical blockade of IKD induced thermosensitivity in cold-insensitive neurons, a finding that may explaincold allodynia in neuropathic pain. These results suggest that cold sensitivity is not associated to aspecific transduction molecule but instead results from a favorable blend of ionic channels expressedin a small subset of sensory neurons.

RESULTSTo identify cold-sensitive primary sensory neurons, we used intra-cellular Ca2+ imaging23 during rapid reductions in temperatureto 15 ± 1°C from an adapting temperature of 33 ± 1°C (Fig. 1aand c). Only 9.3% (n = 483) of cultured mouse trigeminal gan-glion neurons loaded with Fura-2 AM dye showed an increase incytoplasmic calcium concentration ([Ca2+]i) during cooling. Incontrast, about 60% of all neurons responded to capsaicin (1 µM). Capsaicin is the compound responsible for the ‘hot’ tasteof chili peppers and is known to activate nociceptors selective-ly24. Cold sensitivity was restricted to small neurons (Fig. 1b);the average cell body diameter of cold-sensitive neurons (14.9 ± 0.2 µm) was smaller than that of cold-insensitive neu-rons (17.4 ± 0.3 µm) (p < 0.001). The distribution of thresholdtemperature evoking a [Ca2+]i response was very broad (Fig. 1d).Many neurons were excited by modest (<5°C) cooling below theresting temperature of 33 ± 1°C, whereas others required tem-perature decreases >10°C. We hypothesized that these subtypesmay correspond respectively to cooling-sensitive fibers of the skinthat respond maximally to innocuous temperature changes5,7

and to polymodal nociceptive fibers that are activated by nox-ious cold25. To test this, we investigated the effects of capsaicinapplication on cold-sensitive neurons and compared their tem-perature threshold during cooling (Fig. 1f). Among cold-sensi-tive neurons, 48% (15/31) also responded vigorously to 1 µMcapsaicin, suggesting that they are indeed cold-sensitive poly-modal nociceptors. The threshold temperature to coolingresponses did not correlate with the sensitivity to capsaicin, how-ever (p > 0.4).

Menthol is a specific activator of peripheral cold receptors26,enhancing the cooling sensation of mild low temperatures27,28. We

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 255

exposed cold-sensitive neurons to 100 µM menthol at 33°C andfound that 10 of 16 cells responded with a robust [Ca2+]i elevation(∆Ca2+ 379 ± 64 nM) (Fig. 1c). In the remaining six neurons, men-thol markedly sensitized the [Ca2+]i response to cooling, shiftingthe threshold to higher temperatures (from 24.4 ± 1.0°C to 31.5 ±1.0°C, p < 0.001). Neurons that responded to menthol at 33°C had,on average, lower response thresholds during cooling (30.5 ± 1.0°C)compared to those neurons that were only sensitized by menthol (24.4 ± 1.0°C, p < 0.001). Menthol had no effect on [Ca2+]i,

even during strong cooling, in 16 neurons that we had previ-ously identified as cold insensitive (Fig. 1c, n2), thus validatingour identification assay.

We next investigated whether cold-evoked [Ca2+]i elevationswere dependent on spike firing by simultaneously monitoring[Ca2+]i and electrical activity in the cell-attached mode (Fig. 1c).This approach minimized alterations of the intracellular milieu orneuronal resting membrane potential. Cold- and menthol-induced spike activity exhibited the bursting pattern typical of

15 20 25 30 3515

20

25

30

35

Thr

esho

ld o

f [C

a2+

] i re

spon

se (

°C)

∆[Ca2+]i

Threshold AP firing (°C)

32 30 28 26 24 22 20 18 16 140

4

8

12

Num

ber

of n

euro

ns

Threshold (°C)

6 9 12 15 18 21 24 27 30 33 360

30

60

90

Num

ber

of n

euro

ns Cold sensitiveCold insensitive

Neuron diameter (µm)

20 pA1 s

20 pA5 s

20 pA

4°C

33°C

n1

n2

100 µM menthol

200 nM

20°C

Zero Na+

20°C

20°C

20°C

2 min

N1

N2

15 µm

∆[C

a2+

] i (n

M)

400

0

33°C 20°C

33°C

60 s

∆[Ca2+]i250 nM

18°C Capsaicin

2 min

∆[Ca2+]i500 nM

a b

c

d

e

f

g

Fig. 1. Identification and characterization of cold-sensitive (CS) neurons. (a) CS neurons (N1) were identified using microscopic ratiometric fluorescentcalcium imaging (Fura-2) by their marked increases in cytoplasmic calcium during reductions in bath temperature (average increase 611 ± 82 nM, n = 45).(b) Histogram of cell diameters for CS and cold-insensitive (CI) neurons sampled randomly in cultures of trigeminal ganglia. (c) Simultaneous recording ofintracellular Ca2+ (top trace), bath temperature (middle trace) and cell-attached spike activity (bottom trace) during cooling and menthol (100 µM) appli-cation in a CS neuron (n1). In the same optical field, a second neuron (n2) did not show a Ca2+ response to either cooling or menthol application. Theinsets are expansions of the electrical records showing the bursting pattern of activity. (d) Histogram of temperature thresholds for [Ca2+]i elevationsduring cooling. (e) Correlation between temperature threshold for action potential firing, detected in cell-attached mode, and threshold for Ca2+ eleva-tion during cooling (r = 0.97, n = 19). The dotted line is the identity line. (f) [Ca2+]i response of a neuron to cooling (threshold = 31°C) and capsaicin (1 µM) application. (g) Abolition of CS [Ca2+]i responses during replacement of extracellular Na+ with choline. Representative result of n = 4.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

256 nature neuroscience • volume 5 no 3 • march 2002

articles

cellular discharges composed of bursts of 4–10 action potentialsat ∼ 50 Hz with silent interburst periods of 5–10 seconds (Fig. 2a).It has been speculated that bursting in cold-sensitive fibers dependson an oscillating generator potential operational at membrane-potential values subthreshold to spike generation30,31. Indeed, wefound an oscillating membrane potential in three cold-sensitiveneurons during cooling, even in the absence of spikes (Fig. 2b).Among the cold-sensitive neurons tested, 76% (n = 17) depolar-ized and reached action potential threshold during cooling (meandepolarization 19.9 ± 2.8 mV). In contrast, membrane-potentialresponses to cold in cold-insensitive neurons were highly variable:on average, they hyperpolarized by 9 ± 7 mV, and none reachedfiring threshold during cooling steps (n = 6) (Fig. 2c). This hyper-polarization occurred despite a marked increase in input resistance(Fig. 2c, inset). In five cold-insensitive neurons, input resistanceincreased by 37 ± 4% during cooling to 20°C.

We compared membrane properties, action potentials andfiring characteristics in cold-sensitive and -insensitive neurons ofsimilar size and found striking differences (Table 1). Cold-sen-

cold-receptor fibers26 (Fig. 1c, inset). During cooling, we founda tight correlation between the threshold temperature for [Ca2+]ielevation and the firing of action potentials (Fig. 1e). The averagepeak Ca2+ elevation evoked by cooling to 15°C was larger (519 ± 66 nM versus 248 ± 98 nM, p < 0.05) in those neuronsthat were very sensitive to cooling (threshold >28°C) comparedto neurons with higher thresholds (threshold temperature<24°C). To test the role of Na+-dependent action potentials onthe [Ca2+]i signal during cooling, we replaced external Na+ withcholine. This substitution almost abolished the rise in [Ca2+]i(Fig. 1g). Altogether, these data suggest that [Ca2+]i elevationduring cooling is secondary to spike activity.

We used whole-cell recording to further characterize mem-brane responses induced by temperature changes on cold-sensi-tive and -insensitive neurons. In cold-sensitive neurons, coolingproduced a rapid depolarization leading to an initial tonic dis-charge followed by a strong firing adaptation (Fig. 2a and 3b), apattern of response typical of mouse trigeminal thermosensitivefibers29. At steady temperatures of 25–20°C, we saw rhythmic intra-

Fig. 2. Electrophysiological proper-ties of cold-sensitive (CS) and -insen-sitive (CI) neurons. (a) Simultaneousrecording of membrane potential(top trace) and bath temperature(bottom trace) during cooling in aCS neuron, showing the typicaltonic-burst firing pattern of spikeactivity. (b) Recordings in a differentCS neuron showing membrane-potential oscillations during cooling.(c) Intracellular recording in a CIneuron, showing the hyperpolariza-tion in membrane potential.Membrane input resistance wasmonitored with repetitive injectionof a 20-pA current pulse. The insetshows the reversible increase inresistance upon cooling from 34°Cto 20°C. (d) CS neurons are charac-terized by a strong voltage- and time-dependent rectification (sag) duringhyperpolarizing pulses (arrowhead)and fast tonic firing during injection ofdepolarizing pulses. 62% (n = 21) also had a rebound spike (arrow). (e) In the same CS neuron, depolarization from more negative potentials revealsa low-threshold spike. (f) Superimposed action potentials of a CS and CI neuron (same cells as in d and g), showing the typical differences in ampli-tude and duration. The dotted line marks the 0-mV voltage level. (g) Small-diameter CI neurons have less sag during hyperpolarizing pulses and lack arebound depolarization (arrowhead). (h) Firing in CI neurons is markedly inhibited by hyperpolarizing prepulses.

Coldinsensitive

Coldsensitive

2 ms

250 ms

200 pA

20 mV

–63 mV–61 mV

–91 mV

–61 mV

20 mV

–65 mV –65 mV–62 mV

25 s25 s

20°C

20°C

150 ms5 mV

34°C

34°C20°C

34°C

15 s

27°C34°C

a b c

d e fg h

Table 1. Electrophysiological properties of cold-sensitive and cold-insensitive neurons.

Neuron Resting Input Rheobase Spike Spike Inward No. spikes Tonicdiameter MP resistance current threshold duration rectification at 1.5× firing

(µm) (mV) (MΩ) (pA) (mV) (ms) index rheobase pattern

CSneurons(n = 22) 15.2 ± 0.5 –48 ± 2 415 ± 39 54 ± 8 –34 ± 2 1.1 ± 0.1 41 ± 3 % 10 ± 1 91% (20/22)CIneurons (n = 18) 15.9 ± 0.6 –50 ± 1 382 ± 40 121 ± 21 –25 ± 2 1.8 ± 0.1 25 ± 3 % 4 ± 1 39% (7/18)t-test p = 0.37a p = 0.43a p = 0.55a p < 0.01 p < 0.001 p < 0.001 p < 0.001 p < 0.01 p < 0.001b

CI, cold-insensitive; CS, cold-sensitive; MP, membrane potential. Input resistance was measured from the slope of the peak voltage response to a series of nega-tive current steps. Spike duration was measured at half amplitude. Inward rectification index (%) was measured as 100 × (Vpeak – Vsteady-state)/Vpeak during hyper-polarizing voltage responses that reached a Vpeak value around –120 mV. The number of spikes at 1.5× rheobase current were counted in a 500-ms period.aNonsignificant. bZ-test.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 257

larizing response (Fig. 2e). In contrast, in the majority (88%) ofcold-insensitive neurons firing was strongly inhibited by a hyper-polarizing prepulse (Fig. 2h).

We also assessed the effects of temperature upon ionic con-ductances and [Ca2+]i in voltage clamp. In 11/13 cold-sensitiveneurons voltage clamped at –60 mV, no increase in [Ca2+]i wasseen during cooling (Fig. 3a, middle trace), suggesting that lowtemperature does not open Ca2+-permeable channels directly.Resting membrane conductance of cold-sensitive (Fig. 3a, continuous line) and cold-insensitive neurons (data notshown) decreased reversibly during cooling steps from 34°C to20°C. On average, this reduction was twofold larger in cold-sen-sitive neurons than in cold-insensitive neurons (p < 0.05) (Table 2). Both the transient time course of the holding current at–60 mV (Fig. 3a) and the changes in reversal potential (measuredwith slow voltage ramps) of the temperature-sensitive current (Fig. 3c and 3d) indicated that the effects of cooling upon ionicconductance are complex. Thus, in cold-sensitive neurons, thereduction in resting membrane conductance was accompaniedinitially by the development of an inward current (Fig. 3a) with anegative reversal potential (mean reversal = –82.6 ± 5.2 mV, n = 7), suggesting the closure of a K+ current. In most cases, how-

Fig. 3. Effect of cold temperature on ionic conduc-tances. (a) Simultaneous recording of membranecurrent (top trace, circles) and conductance (toptrace, continuous line) (Vhold= –60 mV), [Ca2+]i(middle trace) and bath temperature (bottom trace)during a cooling step in a cold-sensitive (CS) neuron.The insets are current responses to –10-mV steps(250-ms duration) at the times indicated by the filledcircles. Note the transient time course of the inwardcurrent, despite a sustained decrease in conductanceand the lack of a [Ca2+]i response. (b) Effect of cool-ing on the [Ca2+]i, membrane potential and spikeactivity in the same CS neuron as in (a). Note thetransient elevation in [Ca2+]i during spiking (middletrace). (c) Changes in whole-cell ionic current dur-ing cooling, monitored in a CS neuron with a 5-svoltage ramp (the voltage waveform is shown at thebottom). The control trace (34°C) is in gray. (d) Estimation of the net cold-dependent I–V curveby digital subtraction of control ramp currents(34°C) from test currents (25°C and 20°C) showingthe shift in reversal potential. (e) Changes in ioniccurrent in a cold-insensitive (CI) neuron during cool-ing (same protocol as in c). (f) Estimation of the netcold-dependent I–V curve for the CI neuron shownin (e) by the same protocol as in (d).

sitive neurons required injection of smaller positive currents tofire (lower rheobase) and also had more negative voltage spikethresholds, indicating that they are highly excitable. In addition,these neurons had a marked time-dependent inward rectifica-tion during hyperpolarizing current pulses and showed reboundfiring at the end of a hyperpolarizing pulse (Fig. 2d). In con-trast, cold-insensitive neurons had less time-dependent inwardrectification (Fig. 2g), and none (0/18) showed rebound firing(p < 0.001). The action potentials of cold-sensitive neurons werenarrower and smaller in amplitude than those of cold-insensitiveneurons (Fig. 2f). The broad somatic spikes we found in cold-insensitive neurons of small to medium size, and the high inci-dence of capsaicin responses, suggests that a majority of theseneurons are nociceptors32. Nearly all cold-sensitive neurons(91%) fired tonically at regular high frequencies in response todepolarizing pulses (Fig. 2d). In contrast, a majority (62%) ofcold-insensitive neurons either were phasic (one or two actionpotentials) or their tonic firing was strongly adapting. Finally,the firing pattern of cold-sensitive neurons was dependent onthe resting membrane potential. From a negative membranepotential (approximately –90 mV), most (12/20) cold-sensitiveneurons generated a prominent subthreshold transient depo-

30 s

20 mV

–58 mV

–750

–120

–50

–500

–250

0

Cur

rent

(pA

) 20°C

25°C

25°C

20°C

34°C

–120 –80 –40

–200

0

200

400

–200

0

200

400Icold (pA)

–1200

–800

–400

0

20°C

34°C

–120 –80 –40

Icold (pA)

V (mV)V (mV)

V (

mV

)

a

c d e f

b

–110

–90

–70

–50

20

35

0

80

∆ [C

a2+

] iT

emp

(°C

)H

oldi

ng c

urre

nt (

pA)

20°C 20°C 34°C34°C

250 ms

20 pA

3.7

2.7

1.7

Con

duct

ance

(nS

)30 s

Table 2. Biophysical properties of cold-sensitive and cold-insensitive neurons.

Membrane Resting Conductance Icold Ih IKD Conductancecapacitance conductance reduction by density density density ratioa

(pF) (nS) cold (pA/pF) (pA/pF) (pA/pF)

CS 10.0 ± 1.5 2.2 ± 0.4 46 ± 5% –6.0 ± 2.0 30.4 ± 4.3 1.4 ± 0.4 1.1 ± 0.10neurons n = 15 n = 15 n = 9 n = 9 n = 15 n = 15 n = 15CI 11.9 ± 1.1 2.5 ± 0.5 21 ± 7% –0.9 ± 0.8 14.0 ± 3.7 3.4 ± 0.6 1.67 ± 0.2neurons n = 20 n = 20 n = 11 n = 11 n = 20 n = 20 n = 20

t-test p = 0.11b p = 0.62b p = 0.01 p < 0.05 p < 0.01 p < 0.05 p < 0.05CI, cold-insensitive; CS, cold-sensitive. Resting conductance was calculated from the current response to a –10 mV voltage step from a holding of –60 mV. Theconductance reduction was estimated from the decrease in resting conductance upon temperature reduction from 33°C to 20°C. The amplitude of Ih currentwas measured as the time-dependent current upon stepping from –50 to –120 mV for 500 ms. The amplitude of IKD current was measured 1 sec after returningto –50 mV from –120 mV for 500 ms.aG (–50 mV) / G (–60 mV), estimated from the ratio of currents (I-60 – I-50) and (I-70 – I-60). bNonsignificant.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

258 nature neuroscience • volume 5 no 3 • march 2002

articles

ever, the net inward current, here termed Icold, declined duringsteady cooling, despite a sustained decrease in conductance. Therewas a tight correspondence between the time course of Icold andthe time window during which the same neuron fired action poten-tials (Fig. 3b). The amplitude of Icold was significantly larger incold-sensitive neurons (Table 2). Notably, 58% of cold-insensitiveneurons developed no net inward current (Icold) during cooling(Fig. 3e and f). In addition, in most cold-insensitive neurons thetemperature-sensitive current showed strong inwardly rectifyingproperties analogous to the Ih current33 (see below) (Fig. 3f).

Our data imply that excitation during cooling involves clo-sure of a resting K+ conductance that brings cold-sensitive neu-rons to firing threshold, and that adaptation in firing involvesadditional ionic mechanisms that do not require calcium influxor action potential firing. We examined the subthreshold inwardand outward currents active near firing threshold and foundmarked differences between cold-sensitive and -insensitive neu-rons (Table 2). The current responsible for time-dependentinward rectification during hyperpolarization, also know as Ih(ref. 33), was twofold larger in cold-sensitive (Fig. 4a) than incold-insensitive neurons (Fig. 4b and Table 2). This current wasblocked by the extracellular application of 3 mM Cs+ or 20 µMZD-7288 (Fig. 4a, inset), blockers of hyperpolarization-activatedcyclic nucleotide–gated channels33. In contrast, cold-insensitiveneurons presented much higher amplitudes of transient outwardcurrents after the hyperpolarizing voltage step (Fig. 4b and c).These currents contributed an additional resting conductance at–50 mV, quantified as a –50 mV versus –60 mV conductance ratio(Table 2). Two components of transient outward current werediscernible, based on inactivation time course and sensitivity tothe K+-channel blocker 4-aminopyridine (4-AP). In about 70%(11/16) of cold-insensitive neurons, the principal subthresholdtransient outward current had fast inactivation kinetics (τ = 58 ± 13 ms at –40 mV) and relatively low sensitivity to 4-AP(Fig. 4d). In the other 30% the main subthreshold outward cur-rent had much slower inactivation kinetics (τ = 448 ± 119 ms at–40 mV) and was fully inhibited by low concentrations of 4-AP

Fig. 4. Differential expression of ionic currents incold-sensitive (CS) and -insensitive (CI) neurons andinduction of cold thermosensitivity by K+-channelblockers. (a) Whole-cell currents in a CS neuron (12-pF cell capacitance) during 500-ms voltage stepsfrom –50 to –120 mV. The inset shows the blockingeffect of 20 µM ZD-7288 (n = 2) on the time-depen-dent inward current evoked by a step to –120 mV in adifferent CS neuron. Cs+ (3 mM) blocked 91 ± 2% ofthe time-dependent inward current (n = 4). (b) Whole-cell currents in a CI neuron (15 pF cellcapacitance) during same protocol showing the slowtransient outward tail current (IKD). The current andtime calibration also apply to the main traces in (a).(c) Amplitude of IKD current in CS and CI neurons.The current was measured at –50 mV, 1 s after a 500-ms conditioning pulse to –120 mV. (d) Low sensitivityof the fast transient K+ current to 4-AP in a CI neu-ron. (e) High sensitivity of the slow transient K+ cur-rent to 4-AP in a different CI neuron. (f) Simultaneousrecording of [Ca2+]i and bath temperature in 2 neu-rons. In n1 the application of 100 µM 4-AP unmaskeda [Ca2+]i response during the second cooling step,whereas in n2 this produced no change. The insetshows a summary of effects of 4-AP on [Ca2+]iresponses during cooling (n = 8).

(100 µM) (Fig. 4e). In previous studies, this slowly inactivating K+

current has been named IKD (ref. 34).Because of the overlapping activation and inactivation curves

(not shown) and the very slow inactivation time course, we rea-soned that IKD could have a role in reducing the excitability ofcold-insensitive neurons, preventing their discharge during cool-ing. To test this hypothesis, we applied low doses of 4-AP to cold-insensitive neurons and monitored their [Ca2+]i duringtemperature reductions. Indeed, the application of 100 µM 4-APtransformed 40% (8/20) of cold-insensitive neurons into cold-sensitive neurons (Fig. 4f), with an average temperature thresh-old of 26 ± 2°C. The percentage of transformed neurons increasedto 58% (21/36) when cold-insensitive neurons were tested with acocktail of 100 µM 4-AP and 1 mM tetraethylammonium (TEA).In addition, in neurons that became thermosensitive in 100 µM4-AP, whole-cell recordings revealed the expression of a promi-nent slow transient outward current that was blocked by 100 µM4-AP (6/8), whereas neurons (3/4) that did not become ther-mosensitive with 100 µM 4-AP showed a fast transient outwardcurrent that was insensitive to this concentration of drug. Thelow but detectable expression of IKD in cold-sensitive neurons(Fig. 4c) may be functionally significant in modulating their tem-perature threshold; the neurons most sensitive to cooling (thresh-old >28°C) had a lower expression of IKD (0.89 ± 0.34 pA/pFversus 2.26 ± 0.6 pA/pF, p = 0.05) than did those cold-sensitiveneurons excited only by strong cold (threshold <24°C).

DISCUSSIONTemperature sensing is a fundamental biological process requiredfor animal survival and adaptation to variable habitats. Here wereport that mouse primary sensory neurons responding to coolingform a well-defined subpopulation, characterized by distinct elec-trophysiological properties and specific activation by menthol. Thereliable identification of cold-sensitive neurons in culture, withmenthol sensitivity identical to that of cold-sensitive fibers in vivo,is an important step toward the characterization of the cellularmechanism involved in cold transduction. The fact that many cold-

d

caCS

bCI

–120 mV

–50 mV

–120 mV

Control

0

2

4

CICS

n2

n1

4-AP ∆Ca2+

100 nM

100 pA

250 ms

2 min

0.1 mM 4-AP

10

40–40 mV

fe

Control

300 ms500 pA

0.1 mM 4-AP 0

200

400

2 mM 4-AP

Control

0.1 mM 4-AP

Control

100 pA

250 ms IKD

cur

rent

(pA

/pF

)

∆[C

a2+]i

(nM

)T

emp

(°C

)

ZD-7288

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 259

sensitive neurons were also sensitive to capsaicin indicated that thepopulation analyzed may include cold-sensitive polymodal noci-ceptors in addition to cooling receptors. In contrast to resultsobtained in vivo, threshold temperature was not a sufficient criterionto discriminate between both types of thermoreceptors. Neverthe-less, in vivo recordings from second-order lamina I spinal neuronsalso indicate a broad and overlapping range of temperature thresh-olds for cool cells and cold-sensitive polymodal nociceptors35.

Cold-sensitive neurons respond to cooling with closure of aresting K+ conductance, membrane depolarization and firing.This result is in full agreement with a previous finding in dorsalroot ganglion neurons19. The mismatch between the time courseof the inward current and the conductance change, and the shift inreversal potential during cooling, indicated that the ionic mecha-nisms leading to excitation during cooling are more complex. Wealso found strong inhibition of Ih during cooling, which shoulddecrease excitability. We were unable to confirm the transientopening of a non-selective cation channel by cold, as suggestedrecently20. The molecular identity of the background K+ channelinhibited during cooling in primary sensory neurons awaits futurestudies. TREK1 (also known as KCNK2), a member of the two-pore-domain K+ channel subfamily36,37, has been proposed as apossible candidate18. The expression of TREK1 is not restrictedto primary sensory neurons, however; on the contrary, it has apreferential distribution in the central nervous system38.

We also show that cold-insensitive neurons have a large 4-AP-sensitive voltage-gated K+ current (IKD), active near threshold. Akinetically and pharmacologically similar current has beendescribed in nodose ganglion39 and hippocampus34, where it playsan important role in repetitive firing and synaptic integration.Expression of IKD prevents the excitation of cold-insensitive neu-rons during temperature reductions. In cold-insensitive neurons,IKD channels function as excitability brakes during cooling. Thehigh expression of IKD and the lower expression of a backgroundK+ current explain why most primary sensory neurons do notdepolarize to firing threshold despite the closure by cold of theresting K+ conductance. Accumulating evidence indicates thataltered expression of voltage-gated K+ channels may be crucial inthe development of sensory afferent hyperexcitability followinginjury40. We hypothesize that downregulation of IKD may play arole in the abnormal development of cold-induced pain (cold allo-dynia). Modulation of Ih, a current present in the majority of sen-sory neurons41,42 and sensory axons43,44, appears to be anotherkey element in the cellular response to cold. The hyperpolariza-tion induced by inhibition of Ih during cooling may limit the rangeof firing in the population of cold-sensitive neurons and furtherreduce the excitability of cold-insensitive neurons.

Our data indicate that cold sensitivity is an emergent func-tional property resulting from a unique combination of ionicchannels expressed in a small subset of sensory neurons. Theinduction of cold sensitivity in a large percentage of previouslyinsensitive cells by subtle pharmacological manipulation of ionconductances strongly supports this view. Thus, the transduc-tion mechanism for cold appears to be unique in comparisonwith those used for the detection of heat, chemical or mechanicalstimuli, where excitation is linked to activation of a specific trans-duction channel that opens a cation-permeable pore11,12,45–47.

METHODSCell culture. All experiments were conducted according to EuropeanCommunity animal use guidelines. Methods were identical to those usedpreviously23. Trigeminal ganglia were isolated from anesthetized new-born Swiss OF1 mice (postnatal day (P) 2–5), incubated with 0.25% col-lagenase, and cultured in medium consisting of 45% DMEM, 45% F-12

and 10% fetal calf serum (Invitrogen S.A., Barcelona, Spain), supple-mented with 4 mM L-glutamine (Gibco), 100 µg/ml each penicillin andstreptomycin, 20 mM glucose, nerve growth factor (NGF, mouse 7S, 100 ng/ml; Sigma-Aldrich, Madrid, Spain). Cells were plated on poly-L-lysine-coated glass coverslips and used after 1–3 d in culture.

TG electrophysiology. Cell-attached and whole-cell voltage or currentrecordings were performed simultaneously with [Ca2+]i measurementsand temperature recordings. The bath solution contained 124 mM NaCl,5 mM KCl, 1.2 mM KH2PO4, 2 mM CaCl2, 1 mM MgCl2, 26 mMNaHCO3 and 10 mM glucose, pH 7.4 (oxygenated with 95% O2 + 5%CO2). Standard patch pipettes (3–6 MΩ) contained 140 mM KCl, 10 mMNaCl, 4 mM magnesium ATP, 0.4 mM sodium GTP, 0.2 mM Fura-2pentapotassium salt (Molecular Probes, Leiden, The Netherlands) and10 mM HEPES (300 mOsm/kg), pH 7.3 (adjusted with KOH). Currentand voltage signals were recorded with an EPC-8 patch-clamp amplifier(Heka, Elektronik, Lambrecht, Germany). Stimulus delivery and dataacquisition were done with pClamp 8 software (Axon Instruments, UnionCity, California). Reversal potentials of cold-induced currents were cal-culated from currents induced by slow (16 mV/s) depolarizing ramps(from –120 to –40 mV) applied at different temperatures.

Temperature stimulation. Coverslip pieces with neurons were placed ina microchamber and continuously perfused (2–3 ml/min) with solutionswarmed at 33 ± 1°C. Temperature was adjusted with a water-cooled Pelti-er device placed at the inlet of the chamber and controlled by a feedbackdevice. Cold sensitivity was tested with a 100-s temperature ‘step’ to 15 ± 1°C. Temperature decreased and recovered in a quasi-exponentialfashion with a time constant of ∼ 10 s.

Ca2+ imaging. Neurons were incubated with 5–10 µM Fura-2 AM (Mol-ecular Probes) for 30 min at 37°C. Fluorescence measurements weremade with a Zeiss Axioskop FS upright microscope fitted with a Sensys(Roper Scientific, Tucson, Arizona) CCD camera. Fura-2 was excited at357 nm (isosbestic) and 380 nm with a Lambda 10-2 filter wheel (Sut-ter Instruments, Novato, California), and emitted fluorescence filteredwith a 510-nm long-pass filter. Calibrated ratios were displayed onlinewith AIW software (Axon Instruments). Calcium measurements weretemporally synchronized with temperature signals and electrophysio-logical records, acquired simultaneously on a second computer. Thresh-old temperature for action potential initiation (or [Ca2+]i elevation) wasestimated from the temperature level reached during the exponentiallydecaying phase of the temperature step at the time of firing of the firstaction potential.

Data are reported as mean ± standard error of the mean. Statisticalsignificance (p < 0.05) was assessed by the Student’s t-test or Z-test inthe case of evaluating proportions.

AcknowledgementsWe thank C. Robert for performing the experiment shown in Fig. 1f; R. Velasco,

E. Quintero and A. Perez for technical assistance; S. Ingham for illustrations;

and M. Domínguez, R. Gallego, F. Moya, M. Sánchez-Vives, R. Schmidt and M.

Valdeolmillos for critical comments. This work was supported by funds from the

Spanish MICYT (SAF99-0066-C02-01) and FIS (01/1162).

RECEIVED 10 SEPTEMBER 2001; ACCEPTED 16 JANUARY 2002

1. Yarnitsky, D. & Ochoa, J. L. Warm and cold specific somatosensory systems.Psychophysical thresholds, reaction times and peripheral conductionvelocities. Brain 114, 1819–1826 (1991).

2. Craig, A. D., Chen, K., Bandy, D. & Reiman, E. M. Thermosensory activationof insular cortex. Nat. Neurosci. 3, 184–190 (2000).

3 Han, Z. S., Zhang, E. T. & Craig, A. D. Nociceptive and thermoreceptivelamina I neurons are anatomically distinct. Nat. Neurosci. 1, 218–225 (1998).

4. Hensel, H. & Zotterman, Y. The response of the cold receptors to constantcooling. Acta. Physiol. Scan. 22, 96–113 (1951).

5. Hensel, H. & Iggo, A. Analysis of cutaneous warm and cold fibres in primates.Pflugers Arch. 329, 1–8 (1971).

6. Kenshalo, D. R. & Gallegos, E. S. Multiple temperature-sensitive spotsinnervated by single nerve fibers. Science 158, 1064–1065 (1967).

7. Hensel, H., Iggo, A. & Witt, I. A quantitative study of sensitive cutaneousthermoreceptors with C afferent fibers. J. Physiol. 153, 113–126 (1960).

8. Gallar, J., Pozo, M. A., Tuckett, R. P. & Belmonte, C. Response of sensory units

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

260 nature neuroscience • volume 5 no 3 • march 2002

30. Braun, H. A., Bade, H. & Hensel, H. Static and dynamic discharge patterns ofbursting cold fibers related to hypothetical receptor mechanisms. PflugersArch. 386, 1–9 (1980).

31. Iggo, A. & Young, D. W. Cutaneous thermoreceptors and thermalnociceptors. in The Somatosensory System (ed. Kornhuber, H. H.) 5–22(Thieme, Stuttgart, 1975).

32. Koerber, H. R. & Mendell, L. M. Functional heterogeneity of dorsal rootganglion cells. in Sensory Neurons Diversity, Development, and Plasticity. (ed.Scott, S. A.) 77–96 (Oxford University Press, New York, 1992).

33. Pape, H. C. Queer current and pacemaker: the hyperpolarization-activatedcation current in neurons. Annu. Rev. Physiol. 58, 299–327 (1996).

34. Storm, J. F. Temporal integration by a slowly inactivating K+ current inhippocampal neurons. Nature 336, 379–381 (1988).

35. Craig, A. D., Krout, K. & Andrew, D. Quantitative response characteristics ofthermoreceptive and nociceptive lamina I spinothalamic neurons in the cat.J. Neurophysiol. 86, 1459–1480 (2001).

36. Lesage, F. & Lazdunski, M. Molecular and functional properties of two-pore-domain potassium channels. Am. J. Physiol. Renal Physiol. 279, F793–F801(2000).

37. Goldstein, S. A., Bockenhauer, D., O’Kelly, I. & Zilberberg, N. Potassium leakchannels and the KCNK family of two-P-domain subunits. Nat. Rev.Neurosci. 2, 175–184 (2001).

38. Hervieu, G. J. et al. Distribution and expression of TREK-1, a two-pore-domain potassium channel, in the adult rat CNS. Neuroscience 103, 899–919(2001).

39. Stansfeld, C. E., Marsh, S. J., Halliwell, J. V. & Brown, D.A. 4-Aminopyridineand dendrotoxin induce repetitive firing in rat visceral sensory neurones byblocking a slowly inactivating outward current. Neurosci. Lett. 64, 299–304(1986).

40. Rasband, M. N. et al. Distinct potassium channels on pain-sensing neurons.Proc. Natl. Acad. Sci. USA 98, 13373–13378 (2001).

41. Scroggs, R. S., Todorovic, S. M., Anderson, E. G. & Fox, A. P. Variation in IH,IIR, and ILEAK between acutely isolated adult rat dorsal root ganglion neuronsof different size. J. Neurophysiol. 71, 271–279 (1994).

42. Moosmang, S. et al. Cellular expression and functional characterization offour hyperpolarization-activated pacemaker channels in cardiac andneuronal tissues. Eur. J. Biochem. 268, 1646–1652 (2001).

43. Grafe, P., Quasthoff, S., Grosskreutz, J. & Alzheimer, C. Function of thehyperpolarization-activated inward rectification in nonmyelinatedperipheral rat and human axons. J. Neurophysiol. 77, 421–426 (1997).

44. Takigawa, T., Alzheimer, C., Quasthoff, S. & Grafe, P. A special blocker revealsthe presence and function of the hyperpolarization-activated cation currentIH in peripheral mammalian nerve fibres. Neuroscience 82, 631–634 (1998).

45. Waldmann, R. & Lazdunski, M. H(+)-gated cation channels: neuronal acidsensors in the NaC/DEG family of ion channels. Curr. Opin. Neurobiol. 8,418–424 (1998).

46. Garcia-Anoveros, J. & Corey, D. P. The molecules of mechanosensation.Annu. Rev. Neurosci. 20, 567–594 (1997).

47. Hamill, O. P. & Martinac, B. Molecular basis of mechanotransduction inliving cells. Physiol. Rev. 81, 685–740 (2001).

articles

with unmyelinated fibres to mechanical, thermal and chemical stimulation ofthe cat’s cornea. J. Physiol. 468, 609–622 (1993).

9. Brock, J., Pianova, S. & Belmonte, C. Differences between nerve terminalimpulses of polymodal nociceptors and cold sensory receptors of the guinea-pig cornea. J. Physiol. 533, 493–501 (2001).

10. McCleskey, E. W. Thermoreceptors: recent heat in thermosensation. Curr.Biol. 7, R679–R681 (1997).

11. Cesare, P. & McNaughton, P. A novel heat-activated current in nociceptiveneurons and its sensitization by bradykinin. Proc. Natl. Acad. Sci. USA 93,15435–15439 (1996).

12. Caterina, M. J. et al. The capsaicin receptor: a heat-activated ion channel inthe pain pathway. Nature 389, 816–824 (1997).

13. Pierau, F. K., Torrey, P. & Carpenter, D. Effect of ouabain and potassium-freesolution on mammalian thermosensitive afferents in vitro. Pflugers Arch. 359,349–356 (1975).

14. Hensel, H. Functional and structural basis of thermoreception. Prog. BrainRes. 43, 105–118 (1976).

15. Spray, D. C. Metabolic dependence of frog cold receptor sensitivity. Brain Res.72, 354–359 (1974).

16. Carpenter, D. O. Ionic and metabolic bases of neuronal thermosensitivity.Fed. Proc. 40, 2808–2813 (1981).

17. Spray, D. C. Cutaneous temperature receptors. Annu. Rev. Physiol. 48,625–638 (1986).

18. Maingret, F. et al. TREK-1 is a heat-activated background K(+) channel.EMBO J. 19, 2483–2491 (2000).

19. Reid, G. & Flonta, M. Cold transduction by inhibition of a backgroundpotassium conductance in rat primary sensory neurones. Neurosci. Lett. 297,171–174 (2001).

20. Reid, G. & Flonta, M. L. Physiology. Cold current in thermoreceptiveneurons. Nature 413, 480 (2001).

21. Hensel, H., Andres, K. H. & von During, M. Structure and function of coldreceptors. Pflugers Arch. 352, 1–10 (1974).

22. Brock, J. A., McLachlan, E. M. & Belmonte, C. Tetrodotoxin-resistantimpulses in single nociceptor nerve terminals in guinea-pig cornea. J. Physiol.512, 211–217 (1998).

23. Viana, F., de la Pena, E., Pecson, B., Schmidt, R. F. & Belmonte, C. Swelling-activated calcium signalling in cultured mouse primary sensory neurons. Eur.J. Neurosci. 13, 722–734 (2001).

24. Bevan, S. & Szolcsanyi, J. Sensory neuron–specific actions of capsaicin:mechanisms and applications. Trends Pharmacol. Sci. 11, 330–333 (1990).

25. LaMotte, R. H. & Thalhammer, J. G. Response properties of high-thresholdcutaneous cold receptors in the primate. Brain Res. 244, 279–287 (1982).

26. Schafer, K., Braun, H. A. & Isenberg, C. Effect of menthol on cold receptoractivity. Analysis of receptor processes. J. Gen. Physiol. 88, 757–776 (1986).

27. Eccles, R. Menthol and related cooling compounds. J. Pharm. Pharmacol. 46,618–630 (1994).

28. Green, B. G. The sensory effects of l-menthol on human skin. Somatosens.Mot. Res. 9, 235–244 (1992).

29. Schafer, K., Braun, H. A. & Kurten, L. Analysis of cold and warm receptoractivity in vampire bats and mice. Pflugers Arch. 412, 188–194 (1988).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Pyramidal neurons are the primary means by which signalsprocessed in the cortex are transmitted to other parts of the CNS.Their dendrites can span all layers of cortex and contain thousandsof synaptic inputs. The EPSPs from the many dendritic branchespropagate to, and are integrated at, the soma and axon hillock,where they are transduced into action potentials1–4. Passive-cablemodels of dendrites predict that EPSPs reaching the action poten-tial initiation region are severely attenuated because of electroton-ic filtering5–6. This, in turn, suggests that the more distal synapsesplay a relatively minor role in influencing the cell’s firing.

Attenuation of EPSPs is partly alleviated by voltage-dependentNa+ and Ca2+ conductances in the dendrites7–12. Whether thissubthreshold boosting extends to the suprathreshold range is notknown. The uncertainty arises because many EPSPs are neededfor the membrane potential to exceed the threshold for actionpotentials. The increase in the average membrane potential cou-pled with the presence of action potentials during a barrage ofEPSPs alters the activation states not only of Na+ and Ca2+ con-ductances but also of various K+ and hyperpolarization-activatedcation conductances in the dendrites, which may decrease fir-ing13–18. Boosting of firing rate occurs only if the sum of all theionic currents activated in the suprathreshold range is net inward.

Because dendritic conductances are time dependent, theireffect on neuronal firing depends also on the timing of the presy-naptic action potentials. Asynchronously firing presynaptic cellswill evoke EPSPs that sum randomly to generate tonic depolar-ization. By contrast, synchronously firing presynaptic cells willevoke EPSPs that sum within a small time window to generatelarge voltage transients. Tonic depolarization will affect con-ductances with long time constants of activation or inactivation,whereas transients will affect primarily conductances with shorttime constants19. This property has been proposed to enhancethe ability of neurons to detect synchronized inputs. The pres-ence of strong K+ conductances, for example, would shorten the

Boosting of neuronal firing evokedwith asynchronous andsynchronous inputs to the dendrite

Hysell Oviedo and Alex D. Reyes

Center for Neural Science, New York University, 4 Washington Place, New York, New York 10003, USA

Correspondence and requests for materials should be addressed to A.D.R. (e-mail: reyes@cns.nyu.edu)

Published online: 11 February 2002, DOI: 10.1038/nn807

Dendritic conductances have previously been shown to boost excitatory postsynaptic potentials(EPSPs). To determine whether this boosting translates to an increase in the efficacy for evokingaction potentials, we injected barrages of EPSPs that simulate the inputs generated by a populationof presynaptic cells into either the dendrite or the soma of pyramidal neurons in vitro. Although theindividual dendritic and somatic EPSPs were identical, barrages delivered to the dendrite generatedmuch higher firing rates. Boosting occurred when the simulated cells fired asynchronously andsynchronously. This Na+-mediated boosting, which was manifested during repetitive firing, maycompensate functionally for electrotonic attenuation of EPSPs.

effective membrane time constant and the integration time win-dow such that only events that occur simultaneously wouldevoke action potentials20–21.

We directly measured the effectiveness of dendritic and somat-ic inputs in cortical pyramidal neurons in vitro. Neurons weredriven to fire with stimuli that mimicked inputs from a popula-tion of presynaptic cells firing repetitively. The simulated presy-naptic cells were made to fire either asynchronously orsynchronously to examine the response to a broad range of tem-porally correlated inputs and to test for specializations in thesoma or dendrite20 that might enhance the ability of neurons toencode input rate or timing20,22–25. As a result of Na+ conduc-tances, asynchronous and synchronous inputs delivered to thedendrite evoked substantially higher firing rates than those deliv-ered to the soma. The greater effectiveness of dendritic inputsmay offset electrotonic attenuation of EPSPs.

RESULTSWe performed whole-cell recordings in the somata and apicaldendrites of layer 5 pyramidal neurons in slices of sensorimotorcortex taken from postnatal day (P) 21–40 rats (unless otherwisespecified). To simulate excitatory postsynaptic current (EPSC)generated by a single presynaptic cell, we injected time-varyingcurrent (Fig. 1a) into either the apical dendrite (Id) or soma (Is)through the recording electrodes21. We manually adjusted theparameters of the EPSCs (see Methods) so that when these wereinjected into the cell, the resultant voltage deflections at the soma(Ed→s, Es→s) resembled unitary EPSPs (amplitude, 300–600 µV)recorded in layer 2/3 and layer 5 pyramidal neurons26.

To mimic the inputs generated by a population of neurons,we simulated the activities of a specified number of presynapticcells (Npre; see Methods). Each simulated presynaptic cell firesrepetitively at a specified rate (Fpre) to generate a train of EPSCs(Fig. 1b)21. The EPSC trains were summed and subsequently

articles

nature neuroscience • volume 5 no 3 • march 2002 261

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

262 nature neuroscience • volume 5 no 3 • march 2002

injected through either recording electrode. Because the EPSCswere summed linearly and were injected under current clamp, theprotocol mimics best the condition in which synaptic inputs fromelectrotonically distant and spatially disparate locations through-out the dendritic tree converge at a common site at either a moth-er branch (at the site of the dendritic recording) or the soma. Underthis condition, mutual shunting and changes in the driving force ofthe EPSCs are minimal and summation is linear5. Consequently,any nonlinearities introduced by dendritic conductances at thesites of convergence can be examined exclusively.

To examine the responses of neurons to inputs with a broadrange of temporal correlations, we made the simulated presynap-tic cells fire asynchronously and synchronously. In the asynchro-nous mode, we added jitter to the simulated action-potential trainsto remove temporal correlation between the presynaptic cells (seeMethods). This generated a predominantly steady current (Fig. 2a,lower trace), which was then injected into the soma. Increasingthe EPSC rate (Npre × Fpre) led the neuron eventually to fire repet-itively (Fig. 2a, upper trace). The mean (±s.d.) firing rate (takenfrom 20 trials) plotted against the EPSC rate (Fig. 2b) shows thatidentical firing rates were obtained whether the EPSC rate wasvaried by changing Npre and keeping Fpre constant or vice versa.The firing rate can therefore be approximated by equation (1):

F = k × Fpre × Npre × A = k × EPSC rate × A (1)

where k is the slope and A is the total charge (equal to the areaunder an individual EPSC). Because the product of EPSC rate

and A gives the average current (Fig. 2b, upper abscissa), k is theslope of a frequency–current plot obtained by injecting a seriesof current steps into the soma1,4.

In the synchronous mode, a specified subset of the simulat-ed cells fired identically during the second half of the stimulustrain (Fig. 2c, synch). This resulted in large composite EPSCs thatoccurred repetitively at a rate equal to Fpre. Both Npre and Fprewere kept constant. Systematically increasing the proportion ofsynchronized Npre cells increased the coefficient of variation ofthe injected current (Fig. 2d, upper abscissa) and had pronouncedeffects on neuronal firing. Synchronizing a small number ofpresynaptic cells (Fig. 2c, upper trace; 20% of 150 Npre) causedthe firing to become more irregular (compare asynchronous withsynchronous). With a further increase in synchrony, one actionpotential occurred per EPSC cycle (data not shown). In 5 out of13 neurons, 2 to 3 action potentials were evoked per cycle at 100%synchrony (Fig. 2c, lower traces). These changes in patterns wereaccompanied by changes in firing rate (n = 13). The firing rate(for the neuron shown in Fig. 2d) dipped to 20 Hz (Fpre) as theaction potentials became perfectly phase locked to the EPSCs andpeaked at 40 Hz with 100% synchrony. Although the firing ratesof all cells tested varied with the level of synchrony, the shape ofthe curve depended on the values of Npre and Fpre (H.O. andA.D.R., unpublished observations).

To determine whether inputs to other compartments gener-ated comparable responses, we made simultaneous recordings atthe soma and the apical dendrite (100–300 µm from the soma).We manually adjusted the EPSCs injected into the dendrite (Id;Fig. 1a) and soma (Is) so that the resultant average EPSPs mea-sured with the somatic electrode (Ed→s, Es→s) were identical. Ingeneral, larger and briefer EPSCs injected into the dendrite were

articles

Σ

Id

Is

Ed s

a

b

Ed d

Es s

Fig. 1. Stimulus protocol. (a) EPSPs recorded at the dendrite (Ed→d)and soma (Es→s) after current injection into the dendrite (Id) and soma(Is), respectively. Ed→s is Ed→d recorded with the somatic electrode.(b) EPSC trains were summed and injected into the cell. Vertical scalebars, 0.2 mV, 0.02 nA. Horizontal scale bar, 20 ms.

60

40

20

0210

Percent synchrony

Firi

ng r

ate

(Hz)

Firi

ng r

ate

(Hz)

EPSC rate (kHz)

0.40.20Avg. current (nA)

CV

Asynch. Synch.

25

35

45

1520 40 60 800 100

0.060.14

0.220.36

0.490.60Fig. 2. Response to somatic injection. (a) Injected current (bottom) and

associated firing (top) when the EPSCs were delivered asynchronously.(b) Plot of average firing rate (±s.d.) versus EPSC rate. EPSC rate wasvaried either by changing Npre (filled diamonds) or Fpre (open squares).Average injected current is shown on the top abscissa. (c) Firing evokedwhen 20% (*) and 100% (‡) of the EPSCs were introduced synchronously(synch.) during the last half of the stimulus train. (d) Average firing rate(±s.d.) plotted against percent synchrony. The coefficient of variation ofthe injected current is shown on the upper abscissa. Representative volt-age traces for points marked by * and ‡ are shown in (c). Npre = 150;Fpre = 20 Hz. Vertical scale bars for (a, c): 20 mV, 0.2 nA. Horizontalscale bars: 200 ms for (a), 100 ms for (c).

a b

c d

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 3. Response to asynchronous dendritic stimulation.(a) Current injected into the dendrite (third trace) and actionpotentials recorded simultaneously with the dendritic (firsttrace) and somatic (second trace) electrodes. (b) Currentinjected into the soma (lower trace) and action potentialsrecorded with the somatic electrode (upper trace). (c) The firingrate (mean ± s.d.) evoked during dendritic stimulation is plottedagainst that evoked during somatic stimulation. (d) Plot of average(±s.d.) firing rate versus EPSC rate for one cell. Stimuli weredelivered at either the dendrite (filled squares) or the soma(open circles). (e) Firing evoked (upper trace) when the cur-rent injected into the soma (lower trace) was equal to thatinjected into the dendrite (third trace in a). The bottomgraph shows the percentage of dendritic boosting (100 × (Fd –Fs)/Fs) for 9 cases where the average currents injected intothe dendrite and soma were equal. (f) Percent boosting plot-ted against the differences between the peak amplitudes (top)and between the areas (bottom) of Ed→s and Es→s. Scale barsfor (a, b, e): 20 mV, 0.2 nA, 200 ms.

necessary to compensate for the filtering of dendriticEPSPs (Ed→d)6. With this normalization procedure, thedepolarizations near the soma and action potential–initia-tion region were equal for both dendritic and somatic injec-tions; only the location of the inputs differed. A similarprocess occurs naturally in hippocampal pyramidal neu-rons27. In these cells, the magnitudes of synaptic conduc-tances increase systematically along the somatodendriticaxis such that the EPSPs recorded at the soma have com-parable amplitudes independent of their dendritic origin.

Asynchronous dendritic stimulation (Fig. 3a) consis-tently evoked higher firing rates than somatic stimulation(Fig. 3b). A plot of dendritically evoked versus somati-cally evoked firing rates (Fig. 3c; n = 23 cells) showed thatmost data points were above the unitary slope line. Acomparison of somatic and dendritic voltage traces indi-cated that action potentials were initiated in the dendritesin only 3 out of 23 neurons (data not shown). Boostingof firing rate did not vary systematically with the age ofthe rats (P21–40) and occurred in 6 of 8 neurons fromrats as young as P12–14.

A plot of firing rate versus EPSC rate for one cell(Fig. 3d) showed that the greatest difference in firingrates between dendritic (filled squares) and somatic (open cir-cles) stimulation occurred at low EPSC rates; at higher rates,each curve asymptotically approached a maximum firing rate.In general, the curves did not superimpose if one was shiftedalong the horizontal axis, but the difference between the slopesof the linear portion of the curves was not significant (p = 0.11;n = 14; paired t-test).

Boosting of firing rate was not simply due to the fact thatlarger-amplitude currents were injected into the dendrite. Thefiring rate evoked with dendritic injection (Fd) was consistent-ly higher than that evoked with somatic injection (Fs), evenwhen the average currents injected into both compartmentswere equal (Fig. 3a and e). We calculated the percentage increasein firing rate attributable to dendritic processes (equal to 100× (Fd – Fs)/Fs) for nine cases in which the average currentsinjected into the dendrite and soma were equal (Fig. 3e, bot-tom graph). The difference in firing rate between dendritic andsomatic injection was significant (p = 0.005; paired t-test).

To exclude the possibility that the observed boosting was dueto imperfect matching of Ed→s and Es→s, we plotted the percentboosting against the differences between the peak amplitudes and

between the areas of Ed→s and Es→s (Fig. 3f). The percent boost-ing was not significantly correlated with either parameter (peak,r = 0.237, p = 0.327; area, r = 0.185, p = 0.448; n = 19). Weobtained similar results when we placed a third electrode thatwas dedicated to voltage recording at the soma (n = 5; data notshown). This eliminated measurement errors associated withusing one electrode for voltage recording and current injection.

The firing rate during synchronous stimulation at the dendritewas similarly enhanced (n = 7; Fig. 4a). As with somatic stimula-tion, the average firing rate evoked with dendritic stimulation var-ied systematically with the level of synchrony (Fig. 4b). There wereno obvious differences in the degree to which synchronous orasynchronous inputs were boosted. The difference in firing ratesbetween dendritic and somatic stimulation (Fd – Fs) at 0% syn-chrony was not significantly different from that at 100% synchrony(p = 0.10; n = 8; paired t-test). The depth of the modulation(equal to the difference between the maximum and minimumfiring rates as synchrony was increased) tended to be greater fordendritic (17 ± 15 Hz; mean ± s.d.) than for somatic (8 ± 6 Hz)stimulation, although again the difference was not statisticallysignificant (p = 0.12; n = 8; paired t-test). These results indicate

articles

nature neuroscience • volume 5 no 3 • march 2002 263

30

20

0

10

0.80 0.4 1.2EPSC rate (kHz)

Firi

ng r

ate

(Hz)

0

20

40

60

0 4020Somatic firing rate (Hz)

60Den

d. fi

ring

rate

(H

z)

Per

cen

t b

oo

stin

g

0

40

80

120

1 2 3 4 5 6 7 8 9Neuron number

160

Per

cent

boo

stin

gP

erce

nt b

oost

ing

Ed s

– Es s

: area

200

400

0–0.2 –0.1 0 0.1 0.2

200

400

00 105–5–10–15

Stim. dend. rec. dend.

Stim. dend. rec. soma

Stim. soma rec. soma

Stim. soma rec. soma

600

a

b

c

d

e

f

600

Ed s

– Es s

: peak

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Fig. 4. Responses to synchronous dendritic stimula-tion. (a) Average (±s.d.) firing rate evoked with den-dritic stimulation plotted against that evoked withsomatic stimulation. (b) Plot of average (±s.d.) firingrate versus percent synchrony for stimuli delivered atthe dendrite (filled squares) and soma (open circles)for one neuron. Npre = 50; Fpre = 20 Hz.

264 nature neuroscience • volume 5 no 3 • march 2002

that the responses to synchronous inputs at the dendrites are notenhanced preferentially over asynchronous inputs.

To determine whether persistent Na+ conductances in the den-drites contributed to boosting7,9,28, we delivered the Na+-chan-nel blocker tetrodotoxin (TTX, 100 nM) exclusively to the dendritethrough a pipette. Ed→s and Es→s were adjusted to be equal beforethe application of TTX. In the presence of TTX, the firing ratesevoked with asynchronous dendritic current injection decreased(Fig. 5a and b, left traces; n = 10) to a level equal to the somati-cally evoked firing rate (Fig. 5c). We confirmed that TTX blockedexclusively the Na+ conductances in the dendrite by recording atthe soma and dendrite the action potential that was evoked withsomatic current injection (Fig. 5a and b, right traces; n = 10). Inall cases, only the dendritic action potential was blocked. A com-parison of dendritic and somatic firing both before (control; FC)and during (FTTX) TTX application showed that only the den-dritically evoked firing was affected by TTX (quantified as the per-centage change in firing rate or 100 × (FC – FTTX)/FC; Fig. 5d).Thus, boosting of firing rates can be accounted for mainly by acti-vation of local dendritic Na+ conductances.

DISCUSSIONAs a result of dendritic Na+ conductances, the firing evoked withinputs at the dendrite is greater than that of inputs at the soma.The dendritic boosting of firing occurred for inputs with a widerange of temporal correlations and occurred despite the fact thatthe dendritically and somatically injected EPSCs were adjusted sothat the EPSPs at the soma were of equal amplitudes. It is unlikelythat boosting of dendritic input resulted from the way we per-formed the EPSP normalization (Fig. 1a). A parsimonious expla-nation is that the dendrites act as a low-pass filter so that transientinputs (such as individual EPSPs) are attenuated more than tonic

inputs (as would occur during a barrage of EPSPs;Fig. 3a)5–6. As a result, the total drive to the somaduring a barrage of inputs into the dendrite maybe greater than that predicted from the sum of theindividual EPSPs reaching the soma. Three lines

of evidence suggest, however, that at the sites of the dendriticrecordings this effect constitutes a relatively small portion of theoverall boosting. First, boosting occurred even when the averagecurrent injected into the dendrite was equal to that injected intothe soma (Fig. 3e). In such a case, the average current reaching thesoma from the dendrite (in a passive neuron) should be less thanor equal to the average current that was injected into the soma.Second, boosting occurred even during synchronous stimulation(Fig. 4). During synchrony, the current injected into the dendriteconsists of transients (Fig. 2c). Finally, application of TTX exclu-sively to the dendrite blocked boosting of firing (Fig. 5).

One of the more notable results is that the effectiveness ofsynaptic inputs at evoking firing cannot be reliably predicted fromthe parameters of individual subthreshold EPSPs. In anotherstudy28, Na+ conductances in the dendrites contributed relative-ly little to the amplitudes of EPSPs injected individually. Theapparent discrepancy between those results and ours can beaccounted for by the fact that a population of EPSPs generateslarger and longer-lasting depolarizations, both of which wouldincrease activation of the persistent Na+ conductance20. There-fore, the boosting effects of Na+ may become more prominentwhen a sufficient number of EPSPs arrive to evoke repetitive firing.

Increased firing rates with dendritic stimulation occurred forboth asynchronous and synchronous stimulation, indicating thatthe result can be generalized for a broad class of inputs. Theevoked firing during asynchronous and synchronous inputsreflects different aspects of the input signal. During asynchronousstimulation, the evoked firing rate is monotonically related to,and hence encodes, input rate (equation 1). During synchronousstimulation, the firing rate varied with the level of synchrony anddid not accurately reflect input rate because Npre and Fpre werekept constant. At the sites of our dendritic recordings, we found no

articles

10 3020 40 5010

20

30

40

50

Somatic firing rate (Hz)

Den

d. fi

ring

rate

(H

z)

7550250 100

45

25

15

35

Firi

ng r

ate

(Hz)

Percent synchrony

a b

TTX

S

D

S

D100 5 15 20

20

15

10

5

0

Somatic firing rate (Hz)

ControlTTX

Den

d. fi

ring

rate

(H

z)

100

× (

Fc

– F

TT

X)/

Fc

1 2 3 4 5

Fs

Fd

0

20

40

60

80Controla

b

c d

Neuron number

Fig. 5. Role of dendritic Na+ conductances. (a) Firing evoked at soma (left) when an asynchronous stimulus was delivered at the dendrite. (b) Firingevoked during application of TTX at the dendrite. Traces at right show action potentials recorded simultaneously at the soma (S) and dendrite (D)when the stimulus was delivered at the soma. (c) Plot of average (±s.d.) firing rates evoked with dendritic stimulation versus those evoked withsomatic stimulation (n = 10). Dendritic firing rates are shown before (filled circles) and during (open squares) application of TTX. (d) Percent changein firing rate for dendritic (open squares) and somatic (filled squares) stimulation in the presence of TTX. Vertical scale bars for (a, b), 20 mV.Horizontal scale bars, 200 ms and 20 ms for left and right traces, respectively.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

evidence that the dendrites are specialized for detecting synchro-nous events. We cannot exclude the possibility that such special-izations exist at other parts of the dendritic tree, however.

Boosting of firing rate by dendritic Na+ conductances mayoccur as follows. When a population of presynaptic cells becomesactive, the synaptic current that is generated at the dendrites prop-agates to the soma and the action potential initiation region toinitiate a sequence of events leading to repetitive firing1,4. At thesites of the dendritic stimulation (100–300 µm from the soma),the action potentials are probably evoked first at the axon hillockand then at the soma and dendrite2–3. The firing rate is propor-tional to the magnitude of the synaptic current reaching thesoma4. The depolarization caused by the EPSP barrage activatesthe persistent Na+ conductances in the dendrite and the resultantinward current sums with the synaptic current to increase theoverall drive to the soma. In cases where the dendritically andsomatically injected currents are equal, the higher input resistanceof the dendrite probably causes a larger local synaptic depolar-ization and hence allows greater activation of Na+ conductances.

In addition to Na+ conductances, the dendrites of pyrami-dal neurons contain Ca2+, K+ and hyperpolarization-activat-ed cation conductances12–18. All of these affect the shape ofsubthreshold EPSPs. As with Na+ conductances, their activa-tion states are likely to change with a barrage of inputs and arethus likely to influence the resultant firing. At the site of ourdendritic recordings, Na+ conductances predominate andboosting occurs. However, the fact that there are gradients inthe distribution and biophysical properties of conductancesalong the somatodendritic axis14–16,29–30 suggests that boost-ing of firing rate, and hence the effectiveness of dendritic input,may vary throughout the entire dendritic tree.

METHODSSurgical and slicing techniques2,26 followed guidelines set forth by NewYork University’s Animal Welfare Committee. Briefly, we anesthetized Wis-tar rats with halothane and decapitated them. We excised one hemisphereof the brain, glued it to a slicing chamber and immersed it in ice-cold, oxy-genated artificial cerebrospinal fluid (125 mM NaCl, 2.5 mM KCl, 25 mMglucose, 25 mM NaHCO3, 1.25 mM NaH2PO4, 2 mM CaCl2 and 1 mMMgCl2). We used a vibratome slicer to make parasaggital (300 µm thick)slices cut at a 30° angle from the horizontal plane. We stored the slices in aholding chamber at 35°C for 30 min and at room temperature thereafter.We transferred individual slices to a recording chamber mounted on anupright microscope and perfused them with artificial cerebrospinal fluidheated to 33–34°C. We identified layer 5 pyramidal neurons with the aid ofinfrared, differential interference contrast videomicroscopy. We performedwhole-cell current-clamp recordings using borosilicate microelectrodespulled to diameters of 2 µm and 1 µm for somatic and dendritic recordings,respectively. Somatic and dendritic electrodes had direct current resistancesof 5–20 MΩ and 30–40 MΩ , respectively, when filled with 100 mMpotassium gluconate, 20 mM KCl, 4 mM MgATP, 10 mM phosphocre-atine, 0.3 mM GTP and 10 mM HEPES. We filtered voltage and currentsignals at 10 kHz and digitized them at 2–10 kHz.

We stimulated neurons with inputs designed to mimic the net synap-tic current generated when a population of presynaptic cells fire. We useda computer program to simulate the activities of a specified number ofpresynaptic cells (Npre). Each simulated cell fired repetitively for 1 s at aspecified average rate (Fpre). We added jitter to the interspike intervalssuch that they were distributed normally about a mean interval with astandard deviation of ±10%. In the asynchronous mode, all the neuronsfired independently with respect to each other. To eliminate temporalcorrelation between the neurons, we distributed the start times of thespike trains uniformly within one interspike interval. In the synchronousmode, we removed the random delay in the start time and the jitter inthe interspike intervals such that the discharge patterns of a specifiednumber of simulated neurons were identical.

Each time a simulated cell fired an action potential, an associated synap-tic current was calculated. The time course of the current (Fig. 1a) wasdescribed by I(t) = m(1 – e–t/τ0)e–t/τ1, where m is the amplitude and τ0 andτ1 are time constants. The following procedure was used to match Ed→swith Es→s. We injected Is into the soma and compiled an average of Es→sfrom 20–30 sweeps. We manually adjusted the three free parameters andre-injected Is until the average Es→s resembled unitary EPSPs recorded inprevious experiments26. We then injected Id into the dendrite and compiledthe average Ed→s (recorded with the somatic electrode) and overlaid it withthe average Es→s. We adjusted the parameters until the average Ed→smatched Es→s (Fig. 1). We convolved the current with the spike trains ofeach presynaptic cell21. We summed the current trains from all the presy-naptic cells, converted the summed current to an analog signal and inject-ed it into the cell through the amplifier and recording electrode. We deliveredstimuli at >3-s intervals to ensure that the cells reached resting conditionsafter each stimulus.

AcknowledgementsThe authors thank L. Abbott, F. Chance, A. Movshon and J. Rinzel for providing

helpful comments. This work was supported by NSF grant IBN-0079619

(A.D.R.) and by an NSF Minority Fellowship (H.O.).

Competing interests statementThe authors declare that they have no competing financial interests.

RECEIVED 11 DECEMBER 2001; ACCEPTED 14 JANUARY 2002

1. Granit, R., Kernell, D. & Lamarre, Y. Algebraical summation in synapticactivation of motoneurones firing within the ‘primary range’ to injectedcurrents. J. Physiol. (Lond.) 187, 379–399 (1966).

2. Stuart, G. & Sakmann, B. Active propagation of somatic action potentialsinto neocortical pyramidal cell dendrites. Nature 367, 69–72 (1994).

3. Colbert, M. & Johnston, D. Axonal action potential initiation and Na+

channel densities in the soma and axon initial segment of subicularpyramidal neurons. J. Neurosci. 16, 6676–6686 (1996).

4. Schwindt, P. & Crill, W. Equivalence of amplified current flowing fromdendrite to soma measured by alteration of repetitive firing and by voltageclamp in layer 5 pyramidal neurons. J. Neurophysiol. 76, 3731–3739 (1996).

5. Rall, W. Theoretical significance of dendritic trees for neuronal input-outputrelations. in Neural Theory and Modeling (ed. Reiss, R. F.) 73–97 (StanfordUniv. Press, Palo Alto, California, 1964).

6. Stuart, G. & Spruston, N. Determinants of voltage attenuation in neocorticalpyramidal neuron dendrites. J. Neurosci. 18, 3501–3510 (1998).

7. Schwindt, P. & Crill, W. Amplification of synaptic current by persistentsodium conductance in apical dendrite of neocortical neurons.J. Neurophysiol. 74, 2220–2224 (1995).

8. Magee, J. C. & Johnston, D. Synaptic activation of voltage-gated channels inthe dendrites of hippocampal pyramidal neurons. Science 268, 301–304(1995).

9. Lipowsky, R., Gillessen, T. & Alzheimer, C. Dendritic Na+ channels amplifyEPSPs in hippocampal CA1 pyramidal cells. J. Neurophysiol. 76, 2181–2191(1996).

10. Gillessen, T. & Alzheimer, C. Amplification of EPSPs by low Ni2+- andamilioride-sensitive Ca2+ channels in apical dendrites of rat CA1 pyramidalneurons. J. Neurophysiol. 77, 1639–1643 (1997).

11. Schiller, J., Schiller, Y., Stuart, G. & Sakmann, B. Calcium action potentialsrestricted to distal apical dendrites of rat neocortical pyramidal neurons.J. Physiol. (Lond.) 505, 605–616 (1997).

12. Zhu, J. J. Maturation of layer 5 neocortical pyramidal neurons: amplifyingsalient layer 1 and layer 4 inputs by Ca2+ action potentials in adult rat tuftdendrites. J. Physiol. (Lond.) 526, 571–587 (2000).

13. Kang, J., Huguenard, J. R. & Prince, D. A. Development of BK channels inneocortical pyramidal neurons. J. Neurophysiol. 76, 188–198 (1996).

14. Magee, J. C. Dendritic hyperpolarization-activated currents modify theintegrative properties of hippocampal CA1 pyramidal neurons. J. Neurosci.18, 7613–7624 (1998).

15. Poolos, N. P. & Johnston, D. Calcium-activated potassium conductancescontribute to action potential repolarization at the soma but not thedendrites of hippocampal CA1 pyramidal neurons. J. Neurosci. 19,5205–5212 (1999).

16. Bekkers, J. M. Distribution and activation of voltage-gated potassiumchannels in cell-attached and outside-out patches from large layer 5 corticalpyramidal neurons of the rat. J. Physiol. (Lond.) 525, 611–620 (2000).

17. Korngreen, A. & Sakmann, B. Voltage-gated K+ channels in layer 5

articles

nature neuroscience • volume 5 no 3 • march 2002 265

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

266 nature neuroscience • volume 5 no 3 • march 2002

neocortical pyramidal neurones from young rats: subtypes and gradients.J. Physiol. (Lond.) 525, 621–639 (2000).

18. Williams, S. R. & Stuart, G. J. Site independence of EPSP time course ismediated by dendritic Ih in neocortical pyramidal neurons. J. Neurophysiol.83, 3177–3182 (2000).

19. Reyes, A. D. Influence of dendritic conductances on the input-outputproperties of neurons. Annu. Rev. Neurosci. 24, 653–675 (2001).

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

21. Reyes, A. D., Rubel, E. W. & Spain, W. J. In vitro analysis of optimal stimuli forphase-locking and time-delayed modulation of firing in avian nucleuslaminaris neurons. J. Neurosci. 16, 993–1007 (1996).

22. Ferster, D. & Spruston, N. Cracking the neuronal code. Science 270, 756–757(1995).

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

24. Borst, A. & Theunissen, F. E. Information theory and neural coding. Nature

Neurosci. 2, 947–957 (1999).25. Stevens, C. F. & Zador, A. M. Input synchrony and the irregular firing of

cortical neurons. Nature Neurosci. 1, 210–217 (1998).26. Reyes, A. D. & Sakmann, B. Developmental switch in the short-term

modification of unitary EPSPs evoked in layer 2/3 and layer 5 pyramidalneurons of rat neocortex. J. Neurosci. 19, 3827–3835 (1999).

27. Magee, J. C. & Cook, E. P. Somatic EPSP amplitude is independent of synapselocation in hippocampal pyramidal neurons. Nature Neurosci. 3, 895–903(2000).

28. Stuart, G. & Sakmann, B. Amplification of EPSPs by axosomatic sodiumchannels in neocortical pyramidal neurons. Neuron 15, 1065–1076 (1995).

29. Colbert, C. M., Magee, J. C., Hoffman, D. A. & Johnston, D. Slow recoveryfrom inactivation of Na+ channels underlies the activity-dependentattenuation of dendritic action potentials in hippocampal CA1 pyramidalneurons. J. Neurosci. 17, 6512–6521 (1997).

30. Mickus, T., Jung, H. Y. & Spruston, N. Properties of slow, cumulative sodiumchannel inactivation in rat hippocampal CA1 pyramidal neurons. Biophys. J.76, 846–860 (1999).

articles©

2002

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co

m

The importance of dopaminergic mechanisms in the patho-physiology of schizophrenia was inferred from the link betweenthe antipsychotic efficacy of neuroleptic drugs and their affinityfor the dopaminergic D2 receptor1. Frontal cortex dysfunctionin this disorder has been posited even longer, since the modernconceptualization of schizophrenia2. Neuroimaging and basicresearch provide ample evidence for abnormalities in both thesedomains in schizophrenia3,4.

A crucial question has been whether and how these patho-physiological phenomena interact. Because interference with pre-frontal cortex dopamine inputs or intrinsic prefrontal efferentscan lead to disinhibited striatal dopamine function in the rat5,6,it has been proposed that exaggerated striatal dopaminergic neu-rotransmission in schizophrenia might result from dorsolateralprefrontal cortical dysfunction7,8. This proposal, which has beenextended by several investigators9,10, leads to the specific hypoth-esis that, in patients, elevated striatal dopaminergic functionshould be predicted by the degree to which PFC function is dis-turbed, whereas no such relationship should exist in control sub-jects without PFC pathophysiology.

The present study was designed to test this hypothesis. Weused positron emission tomography (PET) to measure bothregional cerebral blood flow (rCBF, with [15O]H2O) and presy-naptic dopaminergic function using the tracer 6-[18F]DOPA (6-FD) in the same session. Schizophrenic subjects withdrawnfrom medication for four weeks and matched normal controlswere studied during the Wisconsin Card Sorting Test, an abstract

Reduced prefrontal activity predictsexaggerated striatal dopaminergicfunction in schizophrenia

Andreas Meyer-Lindenberg1, Robert S. Miletich2, Philip D. Kohn1, Giuseppe Esposito1, Richard E. Carson3, Mario Quarantelli2, Daniel R. Weinberger4 and Karen Faith Berman1

1 Unit on Integrative Neuroimaging, National Institute of Mental Health, National Institutes of Health, 10-4C101, 9000 Rockville Pike, Bethesda, Maryland 20892-1365, USA

2 National Institute of Neurological Disorders and Stroke, National Institutes of Health, 9000 Rockville Pike, Bethesda, Maryland 20832, USA 3 PET Department, Clinical Center, National Institutes of Health, 9000 Rockville Pike, Bethesda, Maryland 20832, USA4 Clinical Brain Disorders Branch, National Institute of Mental Health, National Institutes of Health, 9000 Rockville Pike,

Bethesda, Maryland 20832, USA

Correspondence should be addressed to A.M.L. (andreasml@nih.gov)

Published online: 28 January 2002, DOI: 10.1038/nn804

Both dopaminergic neurotransmission and prefrontal cortex (PFC) function are known to be abnormalin schizophrenia. To test the hypothesis that these phenomena are related, we measured presynapticdopaminergic function simultaneously with regional cerebral blood flow during the Wisconsin CardSorting Test (WCST) and a control task in unmedicated schizophrenic subjects and matched controls.We show that the dopaminergic uptake constant Ki in the striatum was significantly higher for patientsthan for controls. Patients had significantly less WCST-related activation in PFC. The two parameterswere strongly linked in patients, but not controls. The tight within-patient coupling of these values,with decreased PFC activation predicting exaggerated striatal 6-fluorodopa uptake, supports thehypothesis that prefrontal cortex dysfunction may lead to dopaminergic transmission abnormalities.

reasoning, working memory task commonly used to investigatePFC abnormalities11. Because the vast majority of studies indi-cate that hypoactivation is the signature of PFC dysfunction inPET during this task12, we predicted that decreased activationduring the WCST should be inversely correlated with striatal 6-FD uptake in the patient group. Using the outlined multitracerimaging approach, we found decreased PFC blood flow inpatients during the WCST, and an increase in striatal 6-FD uptakein schizophrenics relative to healthy subjects. Confirming ourmain hypothesis, patients showed highly significant inverse cor-relation of these two measures, but controls did not.

RESULTSBehaviorPatients and control subjects completed the same number of tri-als during the WCST (119.3 versus 116.2, respectively; Z = 1.28, p > 0.2). However, patients made significantly more perseverativeerrors (8.6 versus 27.8; Z = –2.56, p < 0.02) and attained fewer cat-egories (9.1 versus 4.1; Z = 2.56, p < 0.02) than control subjects.

Regional cerebral blood flow and dopamine uptakeAnalysis of rCBF subtraction maps showed that the WCST sig-nificantly activated a region in the right dorsolateral prefrontalcortex (x, y, z; 40, 8, 44 mm; p < 0.001, T = 6.98, Z = 4.12) inboth groups (Fig. 1a). Additional foci of activation were observedin the right inferior parietal lobule (32, –58, 24 mm; p < 0.001,T = 4.08, Z = 3.06), occipital cortex (–30, –86, –8 mm; p < 0.001,

articles

nature neuroscience • volume 5 no 3 • march 2002 267

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

nificant difference was found between patients(average Ki, 0.00008 per min) and controls(0.00010 per min, p = 0.34).

Striatal Ki showed significant correlations withbehavioral performance in patients, but not con-trols. In the patient group, the higher the Ki, thehigher the percentage of perseverative errors(Spearman’s r = 0.829, p < 0.05) and the fewer cat-

egories attained (Spearman’s r = –0.829, p < 0.05). To investigatewhether this could be ascribed to a secondary correlation effect,we performed a stepwise multiple regression entering the per-formance parameters and prefrontal blood flow activation as theexplanatory variables. The independent variable was Ki. Indeed,we found that the task-related prefrontal blood flow was signifi-cant, but none of the performance parameters was significant (β for task-related rCBF, –0.81, p < 0.01; highest β for behavioralmeasures, categories attained, β = –0.22, p > 0.3). No correlationwas found with the number of trials in patients, or with any para-meter in the control group.

Finally, to test whether the observed correlation of prefrontalcortex activation with dopamine uptake was specific for thisbrain region, we calculated average blood flow in an ROI com-prising voxels in another region significantly activated by thetask in the occipitotemporal cortex (outside the reference regionused for the input function). Here, no correlation with striatalKi was found in either group. We also analyzed an ROI in theinferior parietal lobule on the right, an area that does projectheavily upon the striatum and is involved in working memoryperformance. Again, no significant correlation with striatal Kiwas found in either group.

DISCUSSIONIn this study, striatal F-DOPA uptake was found to be signifi-cantly increased in schizophrenic patients, in agreement withfour previous reports13–16 as opposed to one equivocal17 andone negative study18. Because the accumulation of this tracer isdue to dopa decarboxylase, an enzyme whose activity, althoughnot rate limiting, reflects a regulated aspect of presynapticdopamine synthesis19,20, these results show that presynapticdopaminergic function is exaggerated in the striatum of schizo-phrenic subjects. This agrees with compelling data indicatingincreased amphetamine-induced striatal dopamine release21.Increased dopaminergic neurotransmission in the striatum ofschizophrenics can also be inferred from elevated baseline D2receptor occupancy, a phenomenon proposed in the mid-eight-ies22 that has recently been demonstrated23.

The key purpose of the present study was to test for the exis-tence of an inverse correlation between PFC activation and stri-atal dopaminergic function in schizophrenia. In accordancewith this a priori hypothesis8, we indeed demonstrated that themeasured indicator of neuronal activation, the increase inblood flow during the WCST task, was tightly coupled withstriatal Ki in patients, but not in controls. As predicted, the less

Fig. 1. Statistical maps of regional cerebral blood flow.(a) Conjunction analysis showing voxels with signifi-cantly (p < 0.01, voxel level) higher rCBF during thetask than the control task. (b) Computer screen show-ing the Wisconsin Card Sorting Test stimuli. (c) Voxelsshowing significantly (p < 0.05) higher rCBF in the task-minus-control contrast in the frontal lobes of controlsas compared to patients.

268 nature neuroscience • volume 5 no 3 • march 2002

T = 7.87, Z = 4.35) and cerebellum (–22, –94, –12 mm; p < 0.001,T = 6.70, Z = 4.04) and, to a lesser degree, left inferior parietallobule (–36, –58, 36 mm; p < 0.005, T = 3.29, Z = 2.56).

This system was less activated in patients, with the rightDLPFC showing the maximal prefrontal between-group differ-ence in rCBF activation (Fig. 1c; 34, 36, 40 mm, BA 9; p < 0.005,voxel-level uncorrected; p < 0.001, cluster-level corrected). WhenrCBF data during the WCST were analyzed separately, values inthe region of interest were also significantly lower in patients thanin control subjects (relative cerebral blood flow, 60.3 versus 55.2;brain average, 50; Z = 2.08, p < 0.04).

In the striatal region of interest, dopamine uptake, measuredby F-DOPA Ki, was significantly higher for patients (0.0100 permin) than for normal controls (0.0084 per min; Fig. 2a; Z = –2.40, p < 0.02).

Correlation between PFC rCBF and striatal KiPatients showed a highly significant negative correlation betweentask-related activation (task minus control) in the right DLPFCROI and striatal Ki (Spearman’s r = –0.943, p < 0.005; Fig. 2b),which was absent in controls (Spearman’s r = –0.086, p > 0.85).A similar result was found for striatal Ki and rCBF during theWCST task itself, which also was negatively correlated only inpatients (Spearman’s r = –0.829, p < 0.05; Fig. 2c). No correla-tion was found in the control group (Spearman’s r = 0.371, p > 0.45; Fig. 2d). This between-groups difference in correlationof rCBF with striatal Ki itself was significant (William-Pearsontest, p = 0.027 for task-related blood flow, p = 0.0376 for WCSTminus control activation). There was no significant correlationwith blood flow during the control task (p = 0.17).

Post-hoc analysesAfter testing the main hypothesis, we did several post-hoc analy-ses for further clarification. Voxel-by-voxel Ki maps were com-pared between groups using statistical parametric mapping.This confirmed the results from the ROI analysis in showingstriatal voxels with significantly increased Ki in the patient groupand further demonstrated that the increase was most markedon the left side (maximum –28, –10, –4 mm; p < 0.001, voxel-level uncorrected; p < 0.001, cluster-level corrected). No sig-nificant laterality effects in striatal Ki were found, and similarcorrelations were observed in both hemispheres. There were nocortical areas in which 6-FD function differed significantlybetween the groups in this analysis. Average Ki in the dorsolat-eral prefrontal ROI was also examined separately; again, no sig-

articles

a b

c

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

activation found in the DLPFC, the more abnormal theincrease in Ki, demonstrating strong coupling of cortical anddopaminergic dysfunction in the disorder. Our comparisonwith an occipital region of interest, where no such correlationexisted, argues for the specificity of this finding for the dorso-lateral prefrontal region. The finding that a correlation existedonly in the patient population but not in the control groupsuggests that the observed interaction was related to the pathol-ogy present in patients, as hypothesized.

As stated, we believe that the observed tight correlation pro-vides evidence for primary dysfunction of the prefrontal cortex inschizophrenia leading to pathologically increased presynapticdopaminergic function. This concept is also consistent with ear-lier data demonstrating a correlation, again only in patients,between decreased N-acetyl-aspartate, a magnetic resonance spec-troscopy marker of neuronal integrity in the DLPFC and exag-gerated amphetamine-induced dopamine release24. The currentdata suggest that the presence of increased presynaptic DA storescould interact with the pharmacological action of amphetaminein releasing presynaptic DA7.

Basic research strongly supports the concept that the activ-ity of dopamine terminals in the striatum is under control ofthe PFC. Although differing in details about the mechanism

or directionality, a number of animal studies show that stimu-lation or inhibition of PFC function affects firing rates of sub-cortical neurons as well as dopamine release25,26. Themechanism of regulation is complex, involving direct and indi-rect projections from the cortex to brainstem and striatum.This feedback is primarily exerted through glutamatergic effer-ents from the PFC27. In humans, disruption of glutamatergicneurotransmission with ketamine leads to increased ampheta-mine-induced dopamine release in the striatum28. Important-ly, in the rodent29, PFC afferents to midbrain neuronsprojecting to the striatum are selectively located on GABAergic(inhibitory) neurons, suggesting a direct anatomical mecha-nism by which a pathological decrease in PFC excitatory outputwould lead to disinhibition of striatal dopaminergic function.This anatomical account is also consistent with the proposal30

that prefrontal cortical glutamatergic projects are a ‘brake’ onthe striatal dopamine system.

Our data may also have implications for understanding thenature of the neuropathology underlying these pathophysiolog-ical relationships. Although it is unclear whether all disturbancesin PFC function would be associated with increased striataldopamine metabolism, studies in animals have demonstratedthat developmental disruption of temporolimbic connections to

articles

nature neuroscience • volume 5 no 3 • march 2002 269

Fig. 2. Striatal dopamine uptake (Ki) and relationship with DLPFC blood flow. (a) Significantly (Z = –2.40, p < 0.02) increased Ki in schizophrenicpatients. (b) Significant correlation of Ki with blood flow change, task minus control, in patients (right-sided DLPFC ROI, bilateral striatal ROI,Spearman’s r = –0.943, p < 0.005, regression line with 95% confidence bands). (c) Significant correlation of Ki with blood flow during the WisconsinCard Sorting task in patients (Spearman’s r = –0.829, p < 0.05). (d) No significant correlation in controls (Spearman’s r = 0.371, p > 0.45).

a b

c d

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

270 nature neuroscience • volume 5 no 3 • march 2002

the PFC do produce such effects. Indeed, exaggerated striataldopamine release was found in monkeys with a neonatal tem-porolimbic lesion31. Analogous developmental abnormalities inPFC connectivity have been implicated in schizophrenia7.

Because correlative measures cannot establish causality, otherexplanations for the observed within-patient correlation betweenstriatal 6-FD uptake and cortical function also need to be consid-ered. The data might reflect a primary disturbance at the level of thebasal ganglia, in which excessive dopaminergic function, via dis-turbed gating32, could lead to functionally deficient prefrontal acti-vation. However, direct support for such a mechanism on theprefrontal–striatal level is lacking. Moreover, this interpretationwould not explain the finding in schizophrenia that decreased NAAin DLPFC predicts the striatal dopamine release after ampheta-mine24, which argues for intrinsic cellular pathology. It is temptingto speculate whether the prefrontal cortical activation deficitobserved is secondary to alterations of dopamine function at thelevel of the cortex, via the mesocortical tract, which mirrors thealteration seen at the striatum. However, available evidence sug-gests cortical dopaminergic tone is decreased in schizophrenia33.This evidence includes improved PFC function in the disease afteradministration of dopamine agonists34, positive correlationbetween cerebrospinal fluid homovanillic acid and PFC rCBF dur-ing the WCST in patients8, and D1- receptor density shown byimaging studies in the cortex of patients35. In agreement with thesestudies, our data do not show increased presynaptic dopamineturnover in the cortex, a finding that must however be seen in thecontext of low signal-to-noise ratio in the cortex due to the lowerdensity of dopaminergic innervation as compared to the striatum.

Aside from the small sample size, a potential confound in ourstudy is that our patients were previously medicated and with-drawn from medication before scanning, which even after amonth might have residual effects on striatal dopamine metab-olism. However, both prefrontal dysfunction and increased stri-atal Ki have been observed in drug-naive as well as previouslymedicated patients, making this potential confound a less likelybasis of our findings. F-Dopa uptake may also be altered in smok-ers, a condition present in two of the patients studied; however,their striatal Ki and prefrontal blood flow did not differ signifi-cantly from those of the other patients (p > 0.16–0.35).

Abnormal Ki predicted impaired task performance as well asDLPFC hypoactivation. Because of the strong linkage betweenKi and PFC dysfunction in our patient group, this is explainedas a secondary correlation, consistent with the multiple regres-sion analysis. Moreover, direct manipulation of striatal dopaminein the rat does not influence working memory performance36,and most primate data on the dopaminergic regulation of work-ing memory-related activation of the prefrontal cortex implicateprefrontal dopamine37,38.

In summary, using a multi-tracer imaging approach, wedemonstrated a tight within-patient coupling of decreased PFCactivation and exaggerated striatal 6-FD uptake in patients withschizophrenia. This provides direct evidence for a commonpathophysiological mechanism linking these abnormalities39 andcan be a model for other neuropsychiatric disorders in which aprimary disturbance based in cortical function or structure leadsto neurochemical counteradaptation.

METHODSSix patients (age 25–43 years, mean 35 years, 1 female) with DSM-IIIR-diagnosed schizophrenia40 participated in this study. They had been treat-ed with typical neuroleptics (haloperidol) in the past and were withdrawnfrom all medication four weeks before the experiment. Six healthy sub-jects (without neuropsychiatric signs or history and not on medication),

each individually chosen (before scanning) to be of the same sex andhandedness and within three years of age of a patient, were studied as acontrol group (age 24–41 years, mean 34 years, 1 female). All subjectsparticipated after giving informed consent as approved by the NationalInstitute of Mental Health Institutional Review Board and the RadiationSafety Committee. Subjects abstained from caffeine and nicotine for 4 hours before the scanning session, and they were pretreated with 100 mg of Carbidopa to reduce peripheral metabolism of 6-FD and, thus,increase availability of this tracer in the brain41.

Participants first underwent a computerized version of the WCST anda matched sensorimotor control task42 while regional cerebral blood flowwas measured after a bolus of 40 mCi [15O]-H2O. Fifteen contiguoustomographic slices (one volume) for each condition were acquired on aScanditronix PC2048-153 camera (FWHM 6.5 mm). The rCBF data werealigned43, normalized to a template image and smoothed with a 12 mmisotropic Gaussian filter. We calculated rCBF relative to a whole brainmean of 50. Subtraction images of rCBF during the WCST compared tothe control task were created and analyzed using SPM99. Coordinates ofstatistically significant brain activations are reported according to thesystem described in the Talairach-Tournoux atlas44.

Approximately 12 minutes after the rCBF scans, 4.5 mCi of 6-FD wasinfused over 45 seconds, and images were acquired from the time of infu-sion up to 120 minutes later (27 images total), while subjects performedthe WCST task. The 6-[18F]DOPA data were aligned in-plane and regis-tered43, coregistered to the rCBF scans, and affine normalized. Irregularregions of interest were drawn around the basal ganglia (roughly corre-sponding to a threshold of 3 times the mean activity), resulting in a bilat-eral region of interest. Using the time–activity curve in an occipitalreference region as the input function45, the kinetic rate constant Ki forstriatal dopaminergic uptake was calculated voxel-by-voxel using a lin-ear fit based on the Patlak method46.

To test the hypothesis that prefrontal function was coupled to 6-FDuptake, we used a conjunction analysis to define a region of interest(ROI) comprising all voxels in the prefrontal cortex significantly (p < 0.01, voxel level) activated by the task in both groups, derived bymanually selecting all voxels in prefrontal cortex significant at the cho-sen threshold. This resulted in a right-sided DLPFC region (Fig. 1a).The average rCBF in this region was then correlated with the Ki. Groupdifferences were tested nonparametrically using the Mann-Whitney Utest. Spearman’s r was used for correlations, and the Williams-Pearsontest was used to assess whether correlations differed significantlybetween the patient and the control groups.

AcknowledgementsWe would like to acknowledge the support of the staff of the PET Department,

Clinical Center, NIH, in the execution of this study and the help of Timothy

Ellmore, in data analysis.

RECEIVED 24 SEPTEMBER; ACCEPTED 26 DECEMBER 2001

1. Bennett, M. R. Monoaminergic synapses and schizophrenia: 45 years ofneuroleptics. J. Psychopharmacol. 12, 289–304 (1998).

2. Kraepelin, E., Barclay, R. M. & Robertson, G. M. Dementia Præcox andParaphrenia (E. & S. Livingstone, Edinburgh, 1919).

3. Lewis, D. A. & Anderson, S. A. The functional architecture of the prefrontalcortex and schizophrenia. Psychol. Med. 25, 887–894 (1995).

4. Laruelle, M. & Abi-Dargham, A. Dopamine as the wind of the psychotic fire:new evidence from brain imaging studies. J. Psychopharmacol. 13, 358–371(1999).

5. Jaskiw, G. E., Karoum, F. K. & Weinberger, D. R. Persistent elevations indopamine and its metabolites in the nucleus accumbens after mildsubchronic stress in rats with ibotenic acid lesions of the medial prefrontalcortex. Brain. Res. 534, 321–323 (1990).

6. Pycock, C. J., Kerwin, R. W. & Carter, C. J. Effect of lesion of corticaldopamine terminals on subcortical dopamine receptors in rats. Nature 286,74–76 (1980).

7. Weinberger, D. R. Implications of normal brain development for thepathogenesis of schizophrenia. Arch. Gen. Psychiatry 44, 660–669 (1987).

8. Weinberger, D. R., Berman, K. F. & Illowsky, B. P. Physiological dysfunction ofdorsolateral prefrontal cortex in schizophrenia. III. A new cohort and evidencefor a monoaminergic mechanism. Arch. Gen. Psychiatry 45, 609–615 (1988).

articles©

2002

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co

m

9. Grace, A. A. Cortical regulation of subcortical dopamine systems and itspossible relevance to schizophrenia. J. Neural Transm. Gen. Sect. 91, 111–134(1993).

10. Deutch, A. Y. The regulation of subcortical dopamine systems by theprefrontal cortex: interactions of central dopamine systems and thepathogenesis of schizophrenia. J. Neural Transm. Suppl. 36, 61–89 (1992).

11. Weinberger, D. R., Berman, K. F. & Zec, R. F. Physiologic dysfunction ofdorsolateral prefrontal cortex in schizophrenia. I. Regional cerebral bloodflow evidence. Arch. Gen. Psychiatry 43, 114–124 (1986).

12. Weinberger, D. R. & Berman, K. F. Prefrontal function in schizophrenia:confounds and controversies. Phil. Trans. R. Soc. Lond. B Biol. Sci. 351,1495–1503 (1996).

13. Hietala, J. et al. Depressive symptoms and presynaptic dopamine function inneuroleptic- naive schizophrenia. Schizophr. Res. 35, 41–50 (1999).

14. Hietala, J. et al. Presynaptic dopamine function in striatum of neuroleptic-naive schizophrenic patients. Lancet 346, 1130–1131 (1995).

15. Lindstrom, L. H. et al. Increased dopamine synthesis rate in medial prefrontalcortex and striatum in schizophrenia indicated by L-(β-11C) DOPA and PET.Biol. Psychiatry 46, 681–688 (1999).

16. Reith, J. et al. Elevated dopa decarboxylase activity in living brain of patientswith psychosis. Proc. Natl. Acad. Sci. USA 91, 11651–11654 (1994).

17. Dao-Castellana, M. H. et al. Presynaptic dopaminergic function in thestriatum of schizophrenic patients. Schizophr. Res. 23, 167–174 (1997).

18. Elkashef, A. M. et al. 6-(18)F-DOPA PET study in patients withschizophrenia. Positron emission tomography. Psychiatry Res. 100, 1–11(2000).

19. Cumming, P., Kuwabara, H., Ase, A. & Gjedde, A. Regulation of DOPAdecarboxylase activity in brain of living rat. J. Neurochem. 65, 1381–1390(1995).

20. Gjedde, A. et al. Dopa decarboxylase activity of the living human brain. Proc.Natl. Acad. Sci. USA 88, 2721–2725 (1991).

21. Abi-Dargham, A. et al. Increased striatal dopamine transmission inschizophrenia: confirmation in a second cohort. Am. J. Psychiatry 155,761–767 (1998).

22. Wong, D. F. et al. Positron emission tomography reveals elevated D2dopamine receptors in drug-naive schizophrenics. Science 234, 1558–1563(1986).

23. Abi-Dargham, A. et al. Increased baseline occupancy of D2 receptors bydopamine in schizophrenia. Proc. Natl. Acad. Sci. USA 97, 8104–8109 (2000).

24. Bertolino, A. et al. The relationship between dorsolateral prefrontal neuronalN-acetylaspartate and evoked release of striatal dopamine in schizophrenia.Neuropsychopharmacology 22, 125–132 (2000).

25. Jackson, M. E., Frost, A. S. & Moghaddam, B. Stimulation of prefrontal cortexat physiologically relevant frequencies inhibits dopamine release in thenucleus accumbens. J. Neurochem. 78, 920–923 (2001).

26. Murase, S., Grenhoff, J., Chouvet, G., Gonon, F. G. & Svensson, T. H.Prefrontal cortex regulates burst firing and transmitter release in ratmesolimbic dopamine neurons studied in vivo. Neurosci. Lett. 157, 53–56(1993).

27. Christie, M. J., Bridge, S., James, L. B. & Beart, P. M. Excitotoxin lesionssuggest an aspartatergic projection from rat medial prefrontal cortex toventral tegmental area. Brain Res. 333, 169–172 (1985).

28. Kegeles, L. S. et al. Modulation of amphetamine-induced striatal dopamine

release by ketamine in humans: implications for schizophrenia. Biol.Psychiatry 48, 627–640 (2000).

29. Carr, D. B. & Sesack, S. R. Projections from the rat prefrontal cortex to theventral tegmental area: target specificity in the synaptic associations withmesoaccumbens and mesocortical neurons. J. Neurosci. 20, 3864–3873(2000).

30. Carlsson, A. et al. Interactions between monoamines, glutamate, and GABAin schizophrenia: new evidence. Annu. Rev. Pharmacol. Toxicol. 41, 237–260(2001).

31. Saunders, R. C., Kolachana, B. S., Bachevalier, J. & Weinberger, D. R.Neonatal lesions of the medial temporal lobe disrupt prefrontal corticalregulation of striatal dopamine. Nature 393, 169–171 (1998).

32. Swerdlow, N. R. & Geyer, M. A. Using an animal model of deficientsensorimotor gating to study the pathophysiology and new treatments ofschizophrenia. Schizophr. Bull. 24, 285–301 (1998).

33. Grace, A. A. Phasic versus tonic dopamine release and the modulation ofdopamine system responsivity: a hypothesis for the etiology of schizophrenia.Neuroscience 41, 1–24 (1991).

34. Daniel, D. G. et al. The effect of amphetamine on regional cerebral blood flowduring cognitive activation in schizophrenia. J. Neurosci. 11, 1907–1917(1991).

35. Okubo, Y. et al. Decreased prefrontal dopamine D1 receptors inschizophrenia revealed by PET. Nature 385, 634–636 (1997).

36. Napier, T. C. & Chrobak, J. J. Evaluations of ventral pallidal dopaminereceptor activation in behaving rats. Neuroreport 3, 609–611 (1992).

37. Muller, U., von Cramon, D. Y. & Pollmann, S. D1- versus D2-receptormodulation of visuospatial working memory in humans. J. Neurosci. 18,2720–2728 (1998).

38. Goldman-Rakic, P. S. The physiological approach: functional architecture ofworking memory and disordered cognition in schizophrenia. Biol. Psychiatry46, 650–661 (1999).

39. Weinberger, D. R. & Lipska, B. K. Cortical maldevelopment, anti-psychoticdrugs, and schizophrenia: a search for common ground. Schizophr. Res. 16,87–110 (1995).

40. American Psychiatric Association. Diagnostic and Statistical Manual ofMental Disorders: DSM-III-R (American Psychiatric Association,Washington, DC, 1987).

41. Miletich, R. S. et al. 6-[18F]fluoro-L-dihydroxyphenylalanine metabolism andpositron emission tomography after catechol-O-methyltransferase inhibitionin normal and hemiparkinsonian monkeys. Brain Res. 626, 1–13 (1993).

42. Berman, K. F. et al. Physiological activation of a cortical network duringperformance of the Wisconsin Card Sorting Test: a positron emissiontomography study. Neuropsychologia 33, 1027–1046 (1995).

43. Woods, R. P., Mazziotta, J. C. & Cherry, S. R. MRI-PET registration withautomated algorithm. J. Comput. Assist. Tomogr. 17, 536–546 (1993).

44. Talairach, J. & Tournoux, P. Co-planar Stereotaxic Atlas of the Human Brain(Thieme, Stuttgart New York, 1988).

45. Brooks, D. J. et al. Differing patterns of striatal 18F-dopa uptake inParkinson’s disease, multiple system atrophy, and progressive supranuclearpalsy. Ann. Neurol. 28, 547–555 (1990).

46. Patlak, C. S., Blasberg, R. G. & Fenstermacher, J. D. Graphical evaluation ofblood-to-brain transfer constants from multiple-time uptake data. J. Cereb.Blood Flow Metab. 3, 1–7 (1983).

articles

nature neuroscience • volume 5 no 3 • march 2002 271

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

272 nature neuroscience • volume 5 no 3 • march 2002

Solid evidence exists today about the complexity of the organiza-tion of the primate frontal cortex1,2, but how this organizationhas evolved remains largely unexplored, as has been noted repeat-edly3–5. Few studies have addressed the organization of this part ofthe brain from a comparative perspective, and thus most evolu-tionary reconstructions still rely on the notion of a dispropor-tionately enlarged frontal cortex in humans. It is thisdisproportionate enlargement that is considered to be largelyresponsible for the uniqueness of human cognitive specialization6.

The evidence in favor of the traditional notion that thefrontal cortex has enlarged disproportionately in human evo-lution comes primarily from two sources7,8. An importantproblem with those earlier sources, as well as more recent stud-ies9–11, is that they are based on the analysis of one or twohemispheres of a few primates and other nonprimate mam-mals, usually to the exclusion of apes, our closest relatives. Asa further complication, the nomenclature used in this litera-ture is inconsistent3. On occasion the term ‘frontal’ is used torefer to the entire sector of the hemisphere in front of the cen-tral sulcus (the frontal lobe); on other occasions it is used torefer only to the cortex of the frontal lobe or, even more restric-tively, to the prefrontal cortex alone.

The final and incontrovertible analysis of the subsectors ofthe frontal lobe in terms of comparative neuroanatomy will comeonly from detailed cytoarchitectonic studies. The prohibitivecost of such studies and the scarcity of the great ape material,however, make it unlikely that these will be done in the foresee-able future. In the meantime, structural magnetic resonanceimaging offers the possibility of conducting a cross-species, mul-tiple-subject study of this problem. Though it is by no meansdefinitive, we regard this approach as capable of yielding a pro-visional assessment of the problem. In a set of earlier studies, wehave shown that the values for the relative size of the frontal lobeas a whole overlap in great apes and humans12,13. Nevertheless,those studies leave open the possibility that the overlap might

Humans and great apes share alarge frontal cortex

K. Semendeferi1, A. Lu1, N. Schenker1 and H. Damasio2

1Department of Anthropology, University of California at San Diego, La Jolla, California 920932Department of Neurology, University of Iowa, Iowa City, Iowa 52242

Correspondence should be addressed to K.S. (ksemende@ucsd.edu)

Published online: 19 February 2002, DOI: 10.1038/nn814

Some of the outstanding cognitive capabilities of humans are commonly attributed to adisproportionate enlargement of the human frontal lobe during evolution. This claim is basedprimarily on comparisons between the brains of humans and of other primates, to the exclusion ofmost great apes. We compared the relative size of the frontal cortices in living specimens of severalprimate species, including all extant hominoids, using magnetic resonance imaging. Human frontalcortices were not disproportionately large in comparison to those of the great apes. We suggest thatthe special cognitive abilities attributed to a frontal advantage may be due to differences inindividual cortical areas and to a richer interconnectivity, none of which required an increase in theoverall relative size of the frontal lobe during hominid evolution.

be due to a disproportionate enlargement of subcortical struc-tures in the apes (such as the basal ganglia that largely underliefrontal cortices), which could obscure differences in the size ofthe cortex of the frontal lobe. In the hope of assessing this pos-sibility, we undertook a new and comprehensive comparativestudy of the cortical mantel in humans and apes. We used mag-netic resonance scans of the brains of living subjects: humans,apes of all extant species and selected monkeys. Our sample wasthus larger than any used in previous comparative neu-roanatomical studies. We examined the overall volume of thefrontal cortex, and of two of its subdivisions that are reliablyidentifiable on the basis of gross anatomical markers in humansand apes: the cortex of the precentral gyrus, and the remainingfrontal cortex on the dorsal, mesial and orbital surfaces of thefrontal lobe (Fig. 1). We found that some primates (lesser apesand monkeys) had a relatively smaller frontal cortex thanhumans, but great apes did not.

RESULTSAs expected, humans had the largest frontal cortex. Individualvalues in humans ranged from 238.8 cm3 to 329.8 cm3 and in thegreat apes from 50.4 cm3 (in a chimpanzee) to 111.6 cm3 (in anorangutan). For the lesser apes (gibbons) the values variedbetween 13.2 cm3 and 16 cm3, and for the monkeys (rhesus,cebus) they varied between 13.3 cm3 and 15.1 cm3. Humans alsohad the largest values for the cortex of the precentral gyrus(43.3–53.8 cm3), followed by the great apes (ranging from 11.2cm3 in one chimpanzee to 28.4 cm3 in one orangutan) and thenthe gibbons (2.6–3.9 cm3). (As noted in Methods, the separationof the cortex belonging to the precentral gyrus and to the rest ofthe frontal lobe was not performed for monkeys.) The rest of thefrontal cortex rostral to the precentral gyrus had a similar distri-bution. Humans ranged from 206.8 cm3 to 283.4 cm3, great apesfrom 39.1 cm3 (in one chimpanzee) to 76.7 cm3 (in one orang-utan) and gibbons from 10.6 cm3 to 12.2 cm3.

articles©

2002

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co

m

articles

nature neuroscience • volume 5 no 3 • march 2002 273

Fig, 2. Logarithmic plot of the volume of the frontal cortexagainst the total volume of the hemispheres minus the frontalcortex. A best-fit regression line is drawn through all nonhu-man primate subjects. Dotted lines represent the predictioninterval. The human observed values are added on the plot.

Nevertheless, the human frontal cortex was not larger thanexpected for a primate brain of human size (Fig. 2). When thevolume of the frontal cortex was regressed against the volume ofthe entire hemisphere and a best-fit line was drawn through allnonhuman primates, the observed human values did not fallabove the prediction interval.

In addition, calculating the frontal cortex volume as a per-centage of the cortex volume of the hemispheres gave values forhumans on a par with those for the great apes (Fig. 3). The rangeof individual values for the frontal cortex of humans, which var-ied between 36.4% and 39.3%, overlapped with those of theorangutans (36.6–38.7%), chimpanzees (32.4–37.5%) and goril-las (35% and 36.9%). Values for gibbons (27.5–31.4%) and mon-keys (29.4–32.3%) fell outside the range of great apes andhumans, and their mean relative values were significantly small-er (standard two-sample Student’s t-test) than those of the greatapes (t = 8.5444, d.f. = 22, p ≤ 0.001) and humans (t = 13.8717,d.f. = 17, p ≤ 0.0001).

In relation to the cortex of the precentral gyrus as a percentof hemispheric cortex the two orangutans stood out, with thelargest relative values (10.2% and 10.4%); the remainingindividual values across hominoid species overlappedfrom 9% in one chimpanzee to 5.5% in one gibbon.Humans fell within this hominoid range (5.4–7.9%).The relative size of the remaining frontal cortex rangedfrom 28.8% to 33% in humans, 25.5% to 29.7% in thegreat apes and 22.0% to 23.8% in the gibbons. A few

individual great apes showed values that overlapped with those ofhumans. Specifically, two chimpanzees, one bonobo and onegorilla were within the lower end of the human range. There wasextensive overlap among the great apes themselves, but notbetween gibbons and the rest of the hominoids. In neither sub-sector of the frontal cortex did humans have values larger thanexpected for an ape of their brain size.

DISCUSSIONAlthough the absolute differences in size of the frontal cortexbetween humans and other primates were large, the frontal cor-tex in humans and great apes occupied a similar proportion ofthe cortex of the cerebral hemispheres. The range of relative val-ues did overlap in the different hominoid species. The smallerprimates (gibbons and monkeys) had a smaller percentage oftheir total cortex occupied by frontal cortices, and the range oftheir values did not overlap with that of the larger hominoids. Inrelation to overall brain size, the human frontal cortex was aslarge as expected for a primate brain of human size.

As noted, with respect to the subsectors of the frontal lobe,any definitive statements can only come from comparative cytoar-chitectonic studies of the brains of extant ape species andhumans, which would be so costly as to preclude their being car-ried out anytime soon. Therefore, our finding that the two sub-sectors of the frontal cortex (cortex of the precentral gyrus andcortex rostral to the precentral sulcus) are not disproportionate-ly larger in humans, though provisional, should prove useful untilmore definitive data become available.

How can we account for the discrepancy between earlier find-ings (Table 1) and ours? Sample size is the most likely factor. Pre-vious studies have had sample sizes of only one or twohemispheres per species and only one or two of the ape specieshave been included7,8. Brodmann7 reports a 6% increase in the‘total frontal lobe surface’ of the human brain compared to thechimpanzee brain. In our sample, we could find a comparabledifference, but only when selected humans and chimpanzees were

Fig. 1. Three-dimensional reconstruction of the brain of (a) a commonchimpanzee and (b) a human. The lateral and mesial surfaces of the lefthemisphere are shown (not to scale). The central and precentral sulciare marked in yellow and cyan, respectively. The posterior end of thefrontal cortex is marked by the yellow line (central sulcus). The whitesegments correspond to connections drawn between discontinuoussegments of these sulci. On the mesial surface, the posterior end of thefrontal cortex is marked by the solid green line and the blue line marksthe end of the orbitofrontal cortex. The red lines mark the position ofthe coronal sections through the hemisphere shown below. On thecoronal sections, cyan indicates frontal cortices anterior to the precen-tral sulcus, yellow indicates the cortex of the precentral gyrus and redthe rest of the cortex of the hemisphere.

a

b

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

274 nature neuroscience • volume 5 no 3 • march 2002

articles

compared (for example, human 2 versus chimpanzee 4 in Fig. 3). Removal of some of the ‘high-end’ great ape subjects or ofthe ‘lower-end’ human subjects in our study would have yield-ed results similar to those reported earlier. The inclusion of onlyselected primate species and an underrepresentation of the greatapes are other possible reasons for the discrepant results. Evenusing our larger sample, if the comparison had been restrictedto humans and smaller primates (Fig. 3) and had excluded mostof the great apes, the results would have been more consonantwith the older results, because gibbons and monkeys do, in fact,have a relatively smaller frontal cortex than humans. In addition,we calculated volumes of the cortical regions of interest, whileearlier studies7,8 rely on surface estimates. We doubt that thiscould account for most of the variance, however, given that thecortical volume is highly correlated with the surface of the cor-tical sheet. Finally, earlier studies use postmortem specimens asopposed to the living subjects included here, which leaves openthe possibility that differential shrinkage could haveaffected the results.

An issue that needs further investigation is the sizeof the prefrontal cortex proper (after exclusion ofmotor, premotor and limbic cortices). We do notthink that this can be addressed by gross anatomicalstudies. No sulcal landmarks can be used reliably toestablish the transition of prefrontal cortex to pre-motor cortex or the borders of individual corticalareas, and thus this issue can only be resolved on thebasis of quantitative cytoarchitectonic studies. Recentreports of such studies suggest that the size and orga-nization of individual cortical areas in the hominoidprefrontal cortex may set humans apart from thegreat apes. Specifically, in terms of size, Brodmann’s

area 10 is reported to be relatively larger in the human brain14

when compared to the ape brain, while Brodmann’s area 13 issmaller15. In the present study, we excluded the cortex of theprecentral gyrus from the total frontal cortex in order to exam-ine a subsector of the frontal cortex that includes all of the pre-frontal cortices. Brodmann’s area 6 and some limbic frontalcortices on the mesial surface were also included, however. Wefound that this region is as large as expected for an ape brainof human size, and that the relative values among individualhumans and individual great apes (but not lesser apes or mon-keys) overlapped. This result goes against the large relative dif-ferences in the prefrontal cortex between humans and great apesreported in previous publications (Brodmann7, who reports asmuch as a 12% enlargement in relation to the chimpanzee, andDeacon6, who sets it at 202% more than expected for a brainof the human size), even presuming the possibility of a muchsmaller premotor and limbic sector.

Fig. 3. Individual values of the relative volume of the frontal cortex as a percent of the volume of the total hemispheric cortex.

Table 1. The relative size of the frontal cortex.

Brodmann (1909)7* Blinkov & Glezer (1965)8* Present study**Human 36.3 32.8 37.7 (± 0.9)Chimpanzee 30.5 22.1 35.4 (± 1.9)Bonobo NA NA 34.7 (± 0.6)Gorilla NA NA 35.0 and 36.9Orangutan NA 21.3 37.6 (± 1.1)Gibbon 21.4 21.2 29.4 (± 1.8)Macaque NA NA 30.6 (± 1.5)Cebus 22.5 NA 29.6 and 31.5

*Surface of frontal cortex in % of surface of cortex of cerebral hemispheres. **Volume offrontal cortex in % of volume of cortex of cerebral hemispheres. NA, not available.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

articles

nature neuroscience • volume 5 no 3 • march 2002 275

The fact that the relative size of the frontal cortex is similarin humans and great apes does not mean that the frontal corticesare less significant to hominid cognitive specialization than hasbeen suggested1,3,16. The frontal cortices could support the out-standing cognitive capabilities of humans without undergoing adisproportionate overall increase in size. This region may haveundergone a reorganization that includes enlargement of select-ed, but not all, cortical areas to the detriment of others. The sameneural circuits might be more richly interconnected within thefrontal sectors themselves and between those sectors and otherbrain regions. Also, subsectors of the frontal lobe might haveundergone a modification of local circuitry, expressed, forinstance, in the form of distinct cytoarchitectonic patterns. Inaddition, microscopic or macroscopic subsectors might have beenadded to the mix or dropped.

There is already some evidence bearing on these possibilities.Findings from the comparative histological studies of the frontalcortex discussed above14–15 and from other brain regions17–18

suggest that the internal organization and size of individual cor-tical areas are specialized among the hominoids. In a previousstudy, we found that the relative volume of white matter under-lying prefrontal association cortices is larger in humans than ingreat apes (Semendeferi, K., Damasio, H. & Van Hoesen, G.W.,Soc. Neurosci. Abstr., 20, 587.7, 1994). This is compatible withthe idea that neural connectivity has increased in the humanbrain19. More recently we have shown that orangutans have asmaller orbitofrontal sector than other apes or humans(Semendeferi, K., Lu, A., Desgouttes, A.M. & Damasio, H., Am. J.Phys. Anthr., Suppl. 30, 278, 2000), which suggests that some dif-ferences can be found in small subsectors of the frontal lobe at agross level. Thus, it seems possible, and even likely, that eithersubsectors of the frontal lobes or individual frontal cortical areashave become specialized during hominoid and hominid evolu-tion. Cognitive specialization in each hominoid species wouldbe related to mosaic evolution20 and reorganization of specificareas in this and other parts of the brain.

In conclusion, it appears that the traditional notion that dis-proportionately large frontal lobes and frontal cortices are thehallmark of hominid brain evolution is not supported. Homo-geneity in the scaling relations of some of the major sectors in theprimate brain has been suggested before21–22, and our resultsshowed that such homogeneity also exists with respect to the sizeof the frontal cortex. Our data also point to the possibility thatthe inclusion of closely related species and larger numbers of indi-viduals per species may reveal that certain areas of the brain pre-sent slight modifications expressed even at the gross level20. Thefindings on gibbons, macaques and cebus monkeys reported here,and results from another study involving humans and baboons11,point to a possible great ape/human specialization in regard to anenlarged frontal cortex that may set larger hominoids apart fromother anthropoid primates. The enlargement of the frontal cor-tex in great apes and humans parallels the emergence in this samegroup of a unique morphological type of neurons in the anteri-or cingulate cortex, which has extensive connections with the pre-frontal cortex23. Relatively large frontal lobes and frontal corticeswould be shared hominoid features that would have appearedprior to the Plio-Pleistocene along with certain reorganizationalfeatures reported to be present in the precursors of hominoids24.

METHODSWe obtained scans of the brains of 24 nonhuman primate subjects (fromthe Yerkes Regional Primate Research Center, Emory University, Atlanta,Georgia) and of ten normal human subjects (from the Department ofNeurology, University of Iowa, Iowa City). Nonhuman primate experi-

mental use was approved by the Emory University IACUC (protocol num-ber 062-95Y) and human subjects provided informed consent in accordwith the policies of IRB of the University of Iowa College of Medicine.The subjects included adult individuals of both sexes. The nonhuman pri-mates included 15 great apes (6 chimpanzees, 3 bonobos, 2 gorillas, 4orangutans), 4 lesser apes (gibbons) and 5 monkeys (3 rhesus and 2cebus). One chimpanzee (chimpanzee 1), one bonobo (bonobo 1) andthe four gibbons were wild caught; the other nonhuman primates wereborn in captivity. Thin-cut magnetic resonance images were obtained forall human subjects in a General Electric 4096 Plus scanner operating at1.5 Tesla, using the following protocol: SPGR/50, TR 24, TE 5, NEX 2,FOV 24 cm, matrix 256 × 192, 124 contiguous coronal slices, 1.6 mmthick. The nonhuman primates were scanned in a Phillips Gyroscan NT,and thickness of axial slices varied between 1.2 and 1.4 mm with a sliceoverlap of 0.6 mm (FOV 124–230 cm).

All brains were reconstructed in three dimensions using Brainvox25,26,an interactive family of programs designed to reconstruct, segment andmeasure brains from magnetic resonance–acquired images. After eachbrain was segmented from all non-brain tissue and reconstructed in threedimensions, its contours were traced on each coronal slice. The pons,medulla and much of the midbrain were excluded using a straight lineconnecting the two transverse fissures on each coronal slice where thesestructures were visible. The two hemispheres were separated. The cor-tex, including the hippocampus and the amygdala, was manually tracedat the gray–white juncture on each individual section in both hemispheresthroughout the brain and in all subjects. The central sulcus marked theposterior border of the frontal cortex on the dorsolateral surface. When-ever the central sulcus did not reach the Sylvian fissure, it was prolongedso as to intersect this fissure. The posterior end of the orbital and ven-tromesial frontal cortex was identified on the appropriate coronal sec-tion. On the mesial surface, the end of the central sulcus was connected,in a straight line, with the most posterior segment of the ventromesialcortex. Above the corpus callosum, this line was considered the posteri-or limit of frontal cortex on the mesial surface. These demarcations wereapplied in a consistent manner across all species.

In order to separate the precentral gyrus (mostly primary motor cortex,although intersubject variability is present27) from the rest of frontal cor-tex, we used the precentral sulcus as its anterior limit. This separation wasapplied only on the hominoids (humans and apes) where, unlike the mon-keys, the precentral sulcus is large and clearly present. Also, because ofthe quality of the scans in two of the four orangutan and one of the threebonobo brains, the identification of the precentral sulcus could not beachieved convincingly; thus, only two representatives of each of thesespecies were included in the analysis of the two subdivisions of the frontalcortex, the cortex of the precentral gyrus and the frontal cortex anterior tothe precentral sulcus. On the lateral surface, when the precentral sulcuswas discontinuous, formed by two or more segments, those segmentswere connected with straight lines. The lateral and inferior end of the pre-central sulcus was connected with the Sylvian fissure using a line perpen-dicular to this fissure. On the mesial surface, the end of the precentralsulcus was connected with a straight line to the junction between the cin-gulate sulcus and the posterior limit of frontal cortex.

K.S. identified the surface landmarks in consultation with H.D. Thehuman, great ape and gibbon regions of interest (ROIs) were traced byA.L. and the monkey ROIs were traced by N.S. A reliability study wasundertaken comparing the tracings of the cortex (chosen as the mostarbitrary of the ROIs) by A.L. and N.S. to a sample of subjects traced byK.S. Five subjects (human, orangutan, gibbon, rhesus, cebus) were ran-domly chosen. Nonparametric correlation (Spearman’s rho) between theresults obtained by A.L. and K.S. was r = 0.950 (p ≤ 0.0001) and thatbetween NS and KS was r = 0.9833 (p ≤ 0.0001).

The absolute volume of all sectors in both hemispheres was calculatedusing Brainvox. The volumes of the regions of interest were calculatedfrom an automatic count of the number of voxels within a region multi-plied by the volume of a single voxel in mm3. We calculated the total vol-ume of the hemispheres, the total volume of the cortex of thehemispheres, and the volumes of frontal cortex, the cortex of the pre-central gyrus and the rest of the frontal cortex.

We analyzed our data in terms of allometry by regressing the volumeof the frontal cortex against the volume of the hemispheres minus the vol-ume to be regressed. We did this because of concern raised in the literature

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

276 nature neuroscience • volume 5 no 3 • march 2002

9. Uylings, H. B. M. & Van Eden, C. G. in Progress in Brain Research vol. 85 (eds.Uylings, H. B. M., Van Eden, C. G., De Bruin, J. P. C., Corner, M. A. &Feenstra, M. G. P.) 31–62 (Elsevier, New York, 1990).

10. Zilles, K., Armstrong, E., Schleicher, A. & Kretschmann, H. The humanpattern of gyrification in the cerebral cortex. Anat. Embryol. 179, 173–179(1988).

11. McBride, T., Arnold, S. E. & Gur, R. C. A comparative volumetric analysis ofthe prefrontal cortex in human and baboon MRI. Brain Behav. Evol. 54,159–166 (1999).

12. Semendeferi, K., Damasio, H., Frank, R. J. & Van Hoesen, G. W. Theevolution of the frontal lobes: a volumetric analysis based on three-dimensional reconstructions of magnetic resonance scans of human and apebrains. J. Hum. Evol. 32, 375–388 (1997).

13. Semendeferi, K. & Damasio, H. The brain and its main anatomicalsubdivisions in living hominoids using magnetic resonance imaging. J. Hum.Evol. 38, 317–332 (2000).

14. Semendeferi, K., Armstrong, E., Schleicher, A., Zilles, K. & Van Hoesen, G. W.Prefrontal cortex in humans and apes: a comparative study of area 10. Am. J.Phys. Anthropol. 114, 224–241 (2001).

15. Semendeferi, K., Armstrong, E., Schleicher, A., Zilles, K. & Van Hoesen, G. W.Limbic frontal cortex in hominoids: a comparative study of area 13. Am. J.Phys. Anthropol. 106, 129–155 (1998).

16. Damasio, A. R. Descartes’ Error (Grosset/Putnam, New York, 1994).17. Buxhoeveden, D. P., Switala, A. E., Roy, E., Litaker, M. & Casanova, M. F.

Morphological differences between minicolumns in human and nonhumanprimate cortex. Am. J. Phys. Anthropol. 115, 361–371 (2001).

18. Preuss, T. M., Qi, H. & Kaas, J. H. Distinctive compartmental organization ofhuman primary visual cortex. Proc. Natl. Acad. Sci. USA 96, 11601–11606(1999).

19. Zhang, K. & Sejnowski, T. J. A universal scaling law between gray matter andwhite matter of cerebral cortex. Proc. Natl. Acad. Sci. USA 97, 5621–5626(2000).

20. de Winter, W. & Oxnard, C. E. Evolutionary radiations and convergences inthe structural organization of mammalian brains. Nature 409, 710–714(2001).

21. Finlay, B. L. & Darlington, R. B. Linked regularities in the development andevolution of mammalian brains. Science 268, 1578–1583 (1995).

22. Clark, D. A., Mitra, P. P. & Wang, S. S.-H. Scalable architecture in mammalianbrains. Nature 411, 189–193 (2001).

23. Nimchinsky, E. A. et al. A neuronal morphologic type unique to humans andgreat apes. Proc. Natl. Acad. Sci. USA 96, 5268–5273 (1999).

24. Radinsky, L. The fossil evidence of anthropoid brain evolution. Am. J. Phys.Anthropol. 41, 15–28 (1974).

25. Damasio, H. & Frank, R. J. Three dimensional in vivo mapping of brainlesions in humans. Arch. Neurol. 49, 137–143 (1992).

26. Frank, R. J., Damasio, H. & Grabowski, T. J. Brainvox: an interactive,multimodal, visualization and analysis system for neuroanatomical imaging.Neuroimage 5, 13–30 (1997).

27. Roland, P. E. & Zilles, K. Brain atlases—a new research tool. Trends Neurosci.17, 458–467 (1994).

28. Smith, R. J. Current events: regression models for prediction equations. J. Hum. Evol. 26, 239–244 (1994).

29. Aiello, L. C. Allometry and the analysis of size and shape in human evolution.J. Hum. Evol. 22, 127–147 (1992).

articles

regarding the effects of regressing a large segment of the brain against thewhole brain6. We fit a restricted-range log–log regression line through allprimates excluding humans using least squares28. We added the humanvalues to the plot and compared expected versus observed values withrespect to the regression line and the prediction interval (95%). A test ofproportions was applied to evaluate the position of the human values withrespect to the prediction interval. In addition to the least-squares analysis,we applied the reduced major axis as a best-fit line29, and we found onlyminor differences in the relation between observed and expected valuesbetween the two approaches. We analyzed our data in both logarithmicand linear plots. The logarithmic transformation provides the best fit, asreflected in R2 values that are closer to 1. We also calculated the size of thefrontal cortex as a percentage of the volume of the hemispheric cortex.We used the same approaches to analyze our data on the precentral gyrusand the remaining portion of the frontal cortex.

AcknowledgementsWe thank T. Wolfson and D. Politis for statistical consulting and J. Spradling

and N. Xenitopoulos for technical and graphic support.

Competing interests statementThe authors declare that they have no competing financial interests.

RECEIVED 19 DECEMBER 2001; ACCEPTED 22 JANUARY 2002

1. Goldman-Rakic, P. S. The frontal lobes: uncharted provinces of the brain.Trends Neurosci. 7, 425–429 (1984).

2. Barbas, H. & Pandya, D. N. Architecture and intrinsic connections of theprefrontal cortex in the rhesus monkey. J. Comp. Neurol. 286, 353–375 (1989).

3. Fuster, J. M. The Prefrontal Cortex. Anatomy, Physiology, and Neuropsychologyof the Frontal Lobe (Lippincott-Raven, Philadelphia, 1997).

4. Holloway, R. L. The evolution of the primate brain: some aspects ofquantitative relations. Brain Res. 7, 121–172 (1968).

5. Jerison, H. J. in Development of the Prefrontal Cortex: Evolution, Neurobiology,and Behavior (eds. Krasnegor, N. A., Lyon, R. & Goldman-Rakic, P. S.) 9–26(Brooks, Baltimore, 1997).

6. Deacon, T. W. The Symbolic Species (Norton, New York, 1997).7. Brodmann, K. Neue Ergebnisse über die vergleichende histologische

Lokalisation der Grosshirnrinde mit besonderer Berücksichtigung desStirnhirns. Anat. Anzeiger 41, 157–216 (1912).

8. Blinkov, S. M. & Glezer, I. I. Das Zentralnervensystem in Zahlen und Tabellen(Fischer, Jena, 1968).

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

It is conjectured that human survival has depended to a largeextent on accurate social judgments and that, as an evolution-ary consequence, modular cognitive processes are devoted tosuch functions1. Neuropsychological studies and human func-tional imaging provide partial support for this idea of a dedi-cated ‘social intelligence’, particularly studies that addressperception of facial expression2–7. However, facial emotionalexpression is only one aspect of social judgment made aboutothers. In many situations, individuals must also decide whetheranother person is someone to approach or avoid, trust or dis-trust. Preliminary evidence regarding the neural underpinningsof this sort of evaluative judgment comes from studies in whichpatients with bilateral amygdala lesions make abnormal socialjudgments about others based on facial appearance8. Theseabnormalities are most pronounced in relation to faces thatreceived the most negative ratings by control subjects. Notably,such deficits are not apparent in subjects with unilateral amyg-dala lesions8. Patients with damage to ventromedial prefrontalcortex also have difficulties with trustworthiness decisions9,10.

The most influential neurobiological model of social cogni-tion11, based on inferences largely from neurophysiological record-ings in non-human primates, postulates that the superior temporalsulcus acts as association cortex for processing conspecifics’ behav-ior and that socially relevant information is subsequently labeled bythe emotional systems, such as amygdala and orbitofrontal cortex.More recent models of human social cognition also include sen-sory regions such as the face-processing area in fusiform gyrus andsomatosensory cortex (including insula, SI and SII)12–14.

Here we used event-related functional magnetic resonanceimaging (fMRI) to ascertain the neural substrates mediating eval-

Automatic and intentional brainresponses during evaluation oftrustworthiness of faces

J.S. Winston1, B.A. Strange2, J. O’Doherty1 and R.J. Dolan1,3

1 Wellcome Department of Imaging Neuroscience, 12 Queen Square, London WC1N 3BG, UK2 Institute of Cognitive Neuroscience, 17 Queen Square, London WC1N 3AR, UK3 Royal Free and University College Medical School, Roland Hill Street, London NW3 2PF, UK

Correspondence should be addressed to J.S.W. (j.winston@fil.ion.ucl.ac.uk)

Published online: 19 February 2002, DOI: 10.1038/nn816

Successful social interaction partly depends on appraisal of others from their facial appearance. Acritical aspect of this appraisal relates to whether we consider others to be trustworthy. Wedetermined the neural basis for such trustworthiness judgments using event-related functional mag-netic resonance imaging. Subjects viewed faces and assessed either trustworthiness or age. In aparametric factorial design, trustworthiness ratings were correlated with BOLD signal change toreveal task-independent increased activity in bilateral amygdala and right insula in response to facesjudged untrustworthy. Right superior temporal sulcus (STS) showed enhanced signal change duringexplicit trustworthiness judgments alone. The findings extend a proposed model of social cognitionby highlighting a functional dissociation between automatic engagement of amygdala versus inten-tional engagement of STS in social judgment.

uative social judgment. Processing of facial emotion can beimplicit, occurring when subjects make judgments about facialattributes unrelated to emotion (for example, refs. 5–7, 15, 16). Toestablish whether trustworthiness judgments might be similarlyprocessed, we used a task in which subjects viewed faces whilemaking either explicit judgments whether an individual was trust-worthy or an unrelated age assessment. To account for individ-ual differences in trustworthiness judgment, we acquired ratingsof trustworthiness for each stimulus from each subject after scan-ning and used these ratings as parametric covariates in our sub-sequent analysis. Based on the models of social cognition outlinedabove11–14, along with the neuropsychological findings8, we pre-dicted that discrete brain regions, the amygdala, orbitofrontalcortex, fusiform gyrus and superior temporal sulcus, would beimplicated in trustworthiness assessments. Consequently, theseareas formed regions of interest in our statistical analysis. Ourdata indicate that social judgments about faces involve such anetwork and that this network is differentially modulated byimplicit and explicit evaluations.

RESULTSBehavioralAfter scanning, on average subjects labeled more than half of the120 faces as having ‘neutral’ emotional expressions (mean, 65).Labeled emotional expression interacted significantly with trust-worthiness score across the group of subjects (Kruskal-Wallis test,p < 0.001). Mann-Whitney U tests showed that the trustworthi-ness scores (from 1, least trustworthy, to 7) did not differ signif-icantly between ‘disgusted,’ ‘fearful’ and ‘surprised’ faces and‘neutral’ faces (p > 0.05 in all cases). ‘Happy’ faces (mean trust-

articles

nature neuroscience • volume 5 no 3 • march 2002 277

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

worthiness rating, 4.0) were rated as significantly more trust-worthy than ‘neutral’ faces (mean rating, 3.9), and ‘angry’ (meanrating, 2.7) and ‘sad’ (mean rating, 3.6) faces as significantly lesstrustworthy (p < 0.01 in all cases). Mean trustworthiness scores(Fig. 1a) were significantly correlated with mean scores for anger,happiness and sadness from the second group of subjects (seeMethods) (p < 0.01 for each, two-tailed; Fig. 1b–e).

NeuroimagingLinear contrasts were performed to produce statistical paramet-ric maps (SPMs) of the main effect of task (explicit or implicitprocessing of trustworthiness), the main effect of trustworthi-ness and the interaction between these two factors. An addition-al model in which the effects of facial emotion of the stimuli wereincluded as covariates of no interest was used to generate an SPMrelated to the main effect of trustworthiness independent ofeffects of facial emotional expression.

Fig. 2. Main effect of explicit social judgments. (a) Random-effects SPM overlaid on a normalized structural scan froma single subject showing activation in right superior tem-poral sulcus region (x, y, z = 56, –44, 4; Z = 4.27; p < 0.05small volume corrected) when making judgments abouttrustworthiness compared to age. For illustration, usingthreshold p < 0.001 uncorrected, extent threshold of5 voxels. (b) BOLD signal measure categorized by taskand trustworthiness of faces. ‘Low,’ ‘med’ and ‘high’ referto the least trustworthy third, median third and mosttrustworthy third of faces calculated in the second modeldescribed in Methods. The y-axis represents mean (acrosssubjects) percentage signal change relative to whole brainmean over scanning session for each event type. There isno clear pattern of response to the faces according totrustworthiness. Note that statistical inference is drawnonly from the parametric model described in Methods,and not from the illustrative model in (b).

278 nature neuroscience • volume 5 no 3 • march 2002

A significant activation in the explicit compared to implicittask, independent of trustworthiness, was found in the right pos-terior superior temporal sulcus (x, y, z coordinates, 56, –44, 4;Z = 4.27; p < 0.05, corrected for multiple comparisons across asmall volume of interest; Fig. 2; Table 1). Additionally, primaryvisual cortex was significantly activated in this contrast. Attentionaland emotional manipulations are known to alter neural respons-es in early visual cortex17, and we propose that similar processesengendered by the explicit task account for this latter activation.

As predicted, significant bilateral amygdala activation was evi-dent in the contrast of untrustworthy to trustworthy faces (right,–18, 0, –24; Z = 4.29; left, –16, –4, –20; Z = 3.92; both p < 0.05,corrected for multiple comparisons across a small volume of inter-est; Fig. 3a). This examination of parametric data based on eachsubject’s ratings of faces indicates that more untrustworthy facesevoke greater BOLD responses in the amygdala (Fig. 3c and d).

articles

Fig. 1. Trustworthiness and emotion ratings for stimuli. (a) Means and standard deviations of trustworthiness scores of stimuli, rank-ordered bytrustworthiness score. (b–e) Mean emotion scores (from second cohort of sixteen subjects) and mean trustworthiness scores (from cohort of sub-jects scanned with fMRI) for anger (b), fear (c), happiness (d) or sadness (e). Lines of best fit are derived by linear regression. Both rating scalesranged from 1 (low degree of emotion or highly untrustworthy) to 7 (highly emotional or highly trustworthy).

ab

d

c

e

a

b

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

Further areas showing increased response to untrustworthyfaces included left superior temporal sulcus (–50,–58,10; Z = 4.15)and a region of the right superior middle insula (42, –4, 12; Z = 3.48;Fig. 3a and b). Additionally, bilateral activation in the fusiformgyrus was evident in this contrast (right, 44, –46, –24, Z = 3.58;left, –48, –48, –24; Z = 3.60; both p < 0.05, corrected for multiplecomparisons across a small volume of interest; Fig. 4). Table 2 pre-sents regions highlighted by this contrast as well as regions high-lighted as more responsive to faces rated as trustworthy.

To ensure that the main effect of untrustworthiness was not dri-ven by a highly significant activation in just one of the tasks alone,a masked conjunction of simple effects of trustworthiness underimplicit and explicit task conditions was carried out (Methods).This analysis confirmed that bilateral amygdala,fusiform gyrus and right insula showed signifi-cant responses to untrustworthy faces indepen-dent of task. Notably, left STS activation was notobserved in this contrast, and a post-hoc testrevealed that the effects in this region were dri-ven principally by trustworthiness judgmentsunder the explicit task.

The contrast pertaining to the interactionof task and trustworthiness demonstrated anarea in the lateral orbitofrontal cortex (–28, 42,10; Z = 3.73, p < 0.0001, uncorrected) respon-sive to untrustworthy faces in the implicit taskand to trustworthy faces in the explicit task.However, this activation failed to survive cor-rection for multiple comparisons across theentire volume of orbitofrontal cortex. No otherareas about which we had a prior hypothesiswere revealed in this contrast or in the reverseinteraction term.

Using an additional model that partialedout effects from facial expression of basic emo-tions in the stimulus set, we performed a ran-dom effects analysis across the 14 subjects.

Even under these stringent criteria, rightamygdala activation was still evident in thismodel at both uncorrected (p < 0.001) andsmall-volume corrected (p < 0.05 correct-ed for multiple comparisons across bilater-al amygdala volume) thresholds (Fig. 5).This activation (peak at 22, 2, –18; Z = 4.06)overlapped with that reported in our pri-mary model. At lower thresholds (p < 0.005,uncorrected), there was additional activa-tion in left amygdala.

DISCUSSIONThe question addressed in this study was whether the dimensionof trustworthiness in faces and the process of making social judg-ments are associated with distinct patterns of brain activation.By implication, the study is an explicit test of a proposed neuro-biological model11. The principal findings of activation in amyg-dala, orbitofrontal cortex and STS are highly consistent with thismodel. We also extend previous lesion data8 by showing amyg-dala activity in response to untrustworthy faces regardless ofwhether subjects were explicitly making trustworthiness judg-ments. This finding echoes earlier studies of obligatory threat-related processing in the amygdala18–21. In contrast to manyimaging studies of facial emotion (for example, ref. 5), the amyg-

Fig. 3. Main effect of trustworthiness in amyg-dala and insula. (a) Significant increases in BOLDsignal to untrustworthy faces in the right and leftamygdalae and right insula (right amygdala, –18,0, –24; Z = 4.29; p < 0.01 corrected; left amyg-dala, –16, –4, –20; Z = 3.92; p < 0.025 corrected;right insula, 42, –4, 12; Z = 3.48; p < 0.001uncorrected). (b–d) Responses to faces as afunction of degree of individually rated trust-worthiness for right insula (b), left amygdala (c)and right amygdala (d). Note greater responsesto less trustworthy faces across all theseregions. The y-axis is as in Fig. 2.

articles

nature neuroscience • volume 5 no 3 • march 2002 279

a b

c d

Table 1. Cerebral foci of activation in main effect of task.

Coordinates of peak activation (mm) Z scoreBrain region x y z

Explicit versus implicitPrimary visual cortex 2 –98 8 4.49Right posterior STS* 56 –44 4 4.27Right superior frontal gyrus 10 14 70 3.99Left premotor cortex –48 –2 26 3.84Left extrastriate cortex –24 –76 32 3.75Right cuneus 12 –40 56 3.62Left primary sensory cortex –30 –28 72 3.62Supramarginal gyrus –62 –40 34 3.50Right anterior insula 48 34 –6 3.37Left superior frontal sulcus –42 12 40 3.34Left pre-SMA –6 10 54 3.34Implicit versus explicitLeft fusiform gyrus –36 –36 –18 3.59Right cuneus 4 –64 12 3.49

All values, p < 0.001 uncorrected. *p < 0.05 corrected for multiple comparisons across a smallvolume of interest.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

but no differential activity according to trustworthi-ness. In other words, the right STS was activatedwhen subjects made explicit judgments about trust-worthiness. In this regard, the STS showed activitywhen subjects were required to make inferences con-cerning the likely intentionality of others. This regionhas been implicated in functional imaging studies onbiological motion35 and biological-like motion36.More critically, activity in posterior STS and adja-cent regions at the temporo-parietal junction isobserved when subjects make theory of mind infer-ences37–39. This region is suggested to be involved inintention detection40,41, rather than biologicalmotion processing per se. Intention detection is a crit-

ical component in determining whether or not to trust an indi-vidual, which may explain the activity in this region in our study.

Evidence from human patients with discrete lesions oforbitofrontal cortex indicate that this region is critical for complexsocial judgment9,10. Unlike the amygdala, this region showed task-dependent activation. When subjects made explicit judgments oftrustworthiness, this region responded more strongly to facesdeemed trustworthy. By contrast, when judging age, this regionshowed greater responses to untrustworthy individuals. Other stud-ies have reported similar task-dependent responses in lateralorbitofrontal cortex. For example, responses in a region of lateral

280 nature neuroscience • volume 5 no 3 • march 2002

dala response to untrustworthy faces was bilateral, supportingneuropsychological evidence that patients with unilateral amyg-dala lesions can successfully make trustworthiness judgments8.To our knowledge, no previous study has demonstrated an auto-maticity of amygdala response during complex social judgments.

In addition to the amygdala, the right insula was also activat-ed by faces that subjects considered untrustworthy regardless oftask. The insula is activated in a wide variety of functional imag-ing studies of emotion (for example, refs. 22–25). One suggestedrole for the insula is the mapping of autonomic changes as theyaffect the body where such mappings form the basis of ‘gut feel-ings’ about emotive stimuli26,27. Thus, apossible explanation for the insula activa-tion that we observed is that a conse-quence of amygdala activation is thegeneration of autonomically mediatedchanges in bodily states, which are thenre-mapped to the insula.

Differential activation in face-respon-sive regions of the fusiform gyrus wasobserved in relation to trustworthiness inface stimuli. Increased activity is found inmodality-specific cortical areas inresponse to stimuli with emotional con-tent relative to non-emotional stimuli (forexample, refs. 20, 21, 28–31). Enhancedextrastriate activation in response to emo-tional stimuli has been attributed to mod-ulatory influences from the amygdala32,possibly mediated by anatomical back-projections33. Indeed, a human lesionstudy highlights a possible role for theamygdala in enhancing perceptual pro-cessing of threat stimuli34. We suggest thatsuch processes extend to faces represent-ing potential threat at the social level andthat a neural consequence is enhancedfusiform activation.

The right STS showed task-related acti-vation in the explicit judgment condition

articles

Fig. 4. Main effect of trustworthiness in fusiform gyrus.(a) Significant increases in BOLD signal to untrustworthyfaces in the fusiform gyrus bilaterally (right, 44, –46, –22;Z = 3.58; p < 0.05, small volume corrected; left, –48, –48,–24; Z = 3.60; p < 0.05, small volume corrected). Thisactivation is independent of task in both the left (b) andright (c) fusiform gyrus. The y-axis is as in Fig. 2.

a

b c

Table 2. Cerebral foci of activation in differential effects of trustworthiness offaces.

Coordinates of peak activation (mm) Z scoreBrain region x y z

Untrustworthy versus trustworthyRight amygdala* 18 0 –24 4.29Left superior temporal sulcus –50 –58 10 4.15Right intraparietal sulcus 22 –54 48 4.00Left extrastriate cortex –34 –90 24 3.94Left amygdala* –16 –4 –20 3.92Right pre-SMA 8 8 62 3.83Left parahippocampal gyrus –18 –30 –18 3.81Right auditory cortex 66 –18 4 3.75Left inferior temporal gyrus –60 –14 –30 3.64Left fusiform gyrus* –48 –48 –24 3.60Right fusiform gyrus* 44 –46 –22 3.58Thalamus 4 –12 14 3.52Right insula 42 –4 12 3.48Left superior temporal gyrus –58 –32 14 3.42Trustworthy versus untrustworthyLeft insula –36 4 –4 3.65Right dorsolateral prefrontal/frontopolar cortex 34 52 6 3.23

All values, p < 0.001 uncorrected. *p < 0.05 corrected for multiple comparisons across a small volume ofinterest.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

orbitofrontal cortex vary between preference and recognition judg-ment tasks on the same stimuli42. A dissociation between implic-it/automatic social judgment and explicit (laboratory-tested) socialjudgment has also been reported in a patient with orbitofrontalcortex damage9. This patient remained able to evaluate social sit-uations under explicit task instructions but was impaired in day-to-day (“automatic”9) social judgments. Note that activation in thisregion did not survive correction for multiple comparisons in ourstudy, and we emphasize effects in this region based on its knowninvolvement in social judgments9,10.

Several regions of interest in this study (amygdala, orbitofrontalcortex and insula) are activated in processing specific facial expres-sions. Facial expressions of fear consistently activate the amyg-dala5,6,43, whereas facial expressions of disgust activate the anteriorinsula6,7. Additionally, we demonstrate correlations between thetrustworthiness scores and scores for facial emotions attributedto our stimulus set (Fig. 1b–e). Consequently, one possibility isthat differential patterns of activation seen in this study reflectinfluences from one or more emotional expressions alone. Weassessed this possibility by analyzing the fMRI data with addi-tional nuisance covariates pertaining to the degree of emotionalexpression of each of four basic emotions (anger, fear, happiness,sadness). A significant right amygdala response to untrustworthyfaces persisted, even after this secondary analysis accounted forthe variance attributed to facial emotion. These results suggestthat facial expressions of emotion provide a constituent element inmaking trustworthiness judgments but that amygdala responsesalso were independent of these effects. Notably, patients with bilat-eral amygdala lesions show deficits in making social judgmentsin the context of maintained ability to use information about emo-tional expression of face stimuli8.

It is interesting to speculate how the results of this study mightgeneralize to social judgments about stimuli in other modalities.Patients with bilateral amygdala lesions are able to make accu-rate trustworthiness judgments based on verbal reports8. It isplausible therefore that amygdala involvement in trustworthi-ness decisions may be modality-specific. This hypothesis couldbe tested in follow-up experiments involving trustworthinessjudgments about vocal stimuli, or scenarios about individualsbased on written descriptions. Our prediction would be thatsuperior temporal sulcus activation would remain in these othercontexts and that fusiform modulation would be substituted bymodality-specific cortical responses, for example, auditory cortexin the case of vocal stimuli.

In conclusion, we present functional brain imaging evidencefor a neural substrate of social cognition that conforms to a pre-viously proposed neurobiological model11. Our data extends thismodel by highlighting a dissociation between automatic andintentional engagement within this proposed circuitry. Thus,social judgments about faces reflect a combination of brainresponses that are stimulus driven, in the case of the amygdala,and driven by processes relating to inferences concerning theintentionality of others, in the case of STS.

METHODSSubjects. Informed consent to partake in a study approved by the JointNational Hospital for Neurology and Neurosurgery/Institute of Neurol-ogy Ethics Committee was obtained from 16 right-handed Caucasianvolunteers (8 male, 8 female; age range 18–30 years; mean age 23.3 years).Two subjects (both females) were excluded from the analysis; one revealedpsychiatric history after scanning and another provided extreme trust-worthiness ratings. (Spearman’s rho of correlation of ratings with meanof all other subjects, –0.445; for all remaining subjects, Spearman’s rhovalues were over 0.3.) All remaining subjects were free from psychiatric orneurological history. All subjects except one had completed more thantwo years of post-16 education, and mean length of post-16 formal edu-cation was 4.8 years.

Stimuli. Grayscale frontal images of 120 Caucasian male faces were select-ed from a larger selection of images following a pilot study outside the scan-ner. The images were selected to cover a range of trustworthiness scoresrated by the subjects in the pilot study (n = 30; 13 females, 17 males, ages17–32, mean age 23.5), but to score as low as possible on ratings of ‘happi-ness’ and ‘anger’. Gaze direction of all stimuli was directly forward. Stimuliwere adjusted to be of approximately equal size and luminance and manip-ulated such that each face was centered on a gray background in a 400 ×400 pixel image. Of the 120 stimuli used in the imaging study, 60 were highschool student photographs and 60 photographs of university students.There was no significant difference in average trustworthiness score betweenthe two groups (Mann-Whitney U test, p > 0.90).

Psychological task. The scanning session for each participant was divid-ed into two parts. In one half of the session, 60 faces were presentedsequentially, and participants made a judgment, indicating with a push-button response, whether the face was a high school or university stu-dent. In the other half of the session, they judged whether the face wastrustworthy or untrustworthy. The order of tasks was counterbalancedbetween participants. At the start of each task, a word appeared on screeninforming the subject of the task requirement (“School/Uni” or “Trust-worthiness”).

Stimuli were presented on a gray background once each in randomorder, randomly interspersed with 60 null events. Each stimulus was pre-sented for 1 s with an inter-trial interval of 2 s. Between faces, a fixationcross was presented. Null events were of 3 s duration, during which timea fixation cross remained on screen. Stimuli subtended visual angles ofapproximately 10° vertically and 5° horizontally.

Image acquisition. Subjects were scanned during task performance usinga Siemens VISION system (Erlangen, Germany) at 2 Tesla to acquire gra-dient-echo, echoplanar T2*-weighted images with BOLD (blood oxy-genation level dependent) contrast. Each volume comprised 33 × 2.2 mmaxial scans with 3-mm in-plane resolution, and volumes were continu-ously acquired every 2.5 s. Subjects were placed in light head restraintwithin the scanner to limit head movement during acquisition. Each runbegan with 5 ‘dummy’ volumes (subsequently discarded) to allow for T1equilibration effects. Additionally, a T1-weighted structural image wasacquired in each subject.

articles

nature neuroscience • volume 5 no 3 • march 2002 281

Fig. 5. Main effect of trustworthiness in amygdala independent of facialemotion. Significant increases in BOLD signal in response to untrust-worthy faces in right amygdala even when scores for four basic facialemotions are additionally used as parametric covariates in the analysis.This activation is significant at p < 0.05, corrected for multiple compar-isons across the volume of bilateral amygdala. Activation peak at 18, 2,–22 (Z = 4.06), but overlaps with right amygdala activation focus shownin Fig. 2. At lower threshold of p < 0.005 uncorrected, activation is evi-dent in left amygdala.

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om

282 nature neuroscience • volume 5 no 3 • march 2002

All functional volumes were realigned44 and slice timing corrected(R. Henson et al., Neuroimage 9, 125, 1999), normalized into a standardspace45 to allow group analysis, and smoothed with an 8-mm FWHMGaussian kernel to account for residual intersubject differences.

Debriefing. After scanning, participants undertook a self-paced task inwhich they rated all the faces on a scale of trustworthiness from 1 (high-ly untrustworthy) to 7 (highly trustworthy). When all 120 faces had beenrated, a second task was performed, in which participants named emo-tions that they perceived in the faces by means of a seven-way forcedchoice procedure (neutral, happy, sad, angry, disgust, fear, surprise). Toassist subjects with this task, we gave them a printed sheet with pho-tographs of one face from the Ekman and Friesen series46 expressing eachof these seven emotions.

Emotion ratings for stimuli. An additional set of 16 subjects (10 males,6 females; age range 19–34 years; mean age 23.7 years) undertook a taskin which they rated the degree of emotional expression within each faceon each of four basic emotions (anger, fear, happiness, sadness) in turn.Ratings were from 1 (neutral for this particular emotion) to 7 (highestdegree of this particular emotion).

Data analysis. Imaging data were analysed with SPM99 using an event-related model47. The experimental design allowed a parametric factori-al analysis whereby trustworthiness was a parametric regressor and thetask (age or trustworthiness judgment) the second factor.

The presentation of each face was modeled by convolving a delta func-tion at each event onset with a canonical hemodynamic response function(HRF) and its temporal derivative to create regressors of interest. Theseregressors were then parametrically modulated to model subject-specifictrustworthiness judgments: that is, the height of the HRF for stimuli wasmodulated as a function of the trustworthiness score assigned to that stim-ulus by the subject. Subject-specific parameter estimates pertaining to eachregressor were calculated for each voxel48. Contrast images were calculat-ed by applying appropriate linear contrasts to the parameter estimates forthe parametric regressor of each event. These contrast images were thenentered into a one-sample t-test across the 14 subjects (that is, a randomeffects analysis). In regions about which we had a prior hypothesis, weapplied a correction for multiple comparisons across a small volume ofinterest to the p-values in this region49. We report predicted regions sur-viving this correction at p < 0.05. Volumes of interest for amygdala,orbitofrontal cortex and STS were defined by drawing a mask around theregions bilaterally on a normalized T1 structural image with reference to anatlas of human neuroanatomy50 using the software package MRIcro(http://www.psychology.nottingham.ac.uk/staff/cr1/mricro.html). Totalvolume of the amygdala mask was approximately 10 cm3, volume of theorbitofrontal mask approximately 50 cm3, and volume of the STS maskapproximately 20 cm3. In the case of the fusiform gyri, small-volume cor-rection was based upon a sphere of 10 mm radius centered on coordinatesderived from a previous study21. We report descriptively activations outsideregions of interest surviving a threshold of p < 0.001 uncorrected with anextent threshold of 5 contiguous voxels.

To ensure that the main effect of trustworthiness did not arise from ahighly significant activation in just one of the simple effects (i.e., thatactivation was task independent), we created a mask from random-effectsSPMs for the simple effect of untrustworthiness under both tasks (eachthresholded at p < 0.05, uncorrected). This was used to mask the maineffect of trustworthiness. Activations surviving this masking procedurereflect responses during both implicit and explicit judgments.

For the purposes of illustration, a second model was constructed by divid-ing the events for each subject into three groups by rank score for individ-ual stimuli (that is, the least trustworthy third of faces as one event type,the median third as a second, and the most trustworthy third as a third).This model is used in Figs. 1–3 to demonstrate the direction of BOLD sig-nal change with respect to trustworthiness score. Note that statistical infer-ences are drawn solely from the parametric model described above.

The mean ratings of facial emotion derived from a second set of 16age-matched subjects (see above) were used to construct anothermodel for the data. In this model, subject-specific ratings for trust-worthiness were entered as parametric covariates, as before. Addi-

tionally, mean ratings for each of the four emotions (anger, fear, hap-piness, sadness) were entered as nuisance parametric covariates. Theparameter estimates for trustworthiness are therefore rendered inde-pendent of the effects of the four facial expressions, and variance bet-ter explained by the effects of a given facial expression will be attributedto the regressor modeling that facial expression. Contrast images fortrustworthiness derived from this model were then entered into a ran-dom-effects analysis.

AcknowledgementsThis work was supported by a programme grant to R.J.D. from the Wellcome Trust.

Competing interest statementThe authors declare that they have no competing financial interests.

RECEIVED 5 DECEMBER; ACCEPTED 26 DECEMBER 2001

1. Humphrey, N. Consciousness Regained: Chapters in the Development of Mind(Oxford Univ. Press, New York, 1983).

2. Adolphs, R., Tranel, D., Damasio, H. & Damasio, A. Impaired recognition ofemotion in facial expressions following bilateral damage to the humanamygdala. Nature 372, 669–672 (1994).

3. Hornak, J., Rolls, E. T. & Wade, D. Face and voice expression identification inpatients with emotional and behavioural changes following ventral frontallobe damage. Neuropsychologia 34, 247–261 (1996).

4. Adolphs, R., Damasio, H., Tranel, D., Cooper, G. & Damasio, A. R. A role forsomatosensory cortices in the visual recognition of emotion as revealed bythree-dimensional lesion mapping. J. Neurosci. 20, 2683–2690 (2000).

5. Morris, J. S. et al. A differential neural response in the human amygdala tofearful and happy facial expressions. Nature 383, 812–815 (1996).

6. Phillips, M. L. et al. A specific neural substrate for perceiving facialexpressions of disgust. Nature 389, 495–498 (1997).

7. Sprengelmeyer, R., Rausch, M., Eysel, U. T. & Przuntek, H. Neural structuresassociated with recognition of facial expressions of basic emotions. Proc. R.Soc. Lond. B Biol. Sci. 265, 1927–1931 (1998).

8. Adolphs, R., Tranel, D. & Damasio, A. R. The human amygdala in socialjudgment. Nature 393, 470–474 (1998).

9. Eslinger, P. J. & Damasio, A. R. Severe disturbance of higher cognition afterbilateral frontal lobe ablation: patient EVR. Neurology 35, 1731–1741 (1985).

10. Damasio, A. R. Descartes’ Error: Emotion, Reason and the Human Brain(Putnam, New York, 1994).

11. Brothers, L. The social brain: a project for integrating primate behavior andneurophysiology in a new domain. Concepts Neurosci. 1, 27–51 (1990).

12. Adolphs, R. Social cognition and the human brain. Trends. Cogn. Sci. 3,469–479 (1999).

13. Adolphs, R. The neurobiology of social cognition. Curr. Opin. Neurobiol. 11,231–239 (2001).

14. Allison, T., Puce, A. & McCarthy, G. Social perception from visual cues: roleof the STS region. Trends Cogn. Sci. 4, 267–278 (2000).

15. Dolan, R. J. et al. Neural activation during covert processing of positiveemotional facial expressions. Neuroimage 4, 194–200 (1996).

16. Blair, R. J., Morris, J. S., Frith, C. D., Perrett, D. I. & Dolan, R. J. Dissociableneural responses to facial expressions of sadness and anger. Brain 122,883–893 (1999).

17. Lane, R. D., Chau, P. M. & Dolan, R. J. Common effects of emotional valence,arousal and attention on neural activation during visual processing ofpictures. Neuropsychologia 37, 989–997 (1999).

18. Morris, J. S., Ohman, A. & Dolan, R. J. Conscious and unconscious emotionallearning in the human amygdala. Nature 393, 467–470 (1998).

19. Whalen, P. J. et al. Masked presentations of emotional facial expressionsmodulate amygdala activity without explicit knowledge. J. Neurosci. 18,411–418 (1998).

20. Strange, B. A., Henson, R. N., Friston, K. J. & Dolan, R. J. Brain mechanismsfor detecting perceptual, semantic, and emotional deviance. Neuroimage 12,425–433 (2000).

21. Vuilleumier, P., Armony, J. L., Driver, J. & Dolan, R. J. Effects of attention andemotion on face processing in the human brain. An event-related fMRI study.Neuron 30, 829–841 (2001).

22. Buechel, C., Morris, J., Dolan, R. J. & Friston, K. J. Brain systems mediatingaversive conditioning: an event-related fMRI study. Neuron 20, 947–957(1998).

23. Buechel, C., Dolan, R. J., Armony, J. L. & Friston, K. J. Amygdala-hippocampal involvement in human aversive trace conditioning revealedthrough event-related functional magnetic resonance imaging. J. Neurosci.19, 10869–10876 (1999).

24. Casey, K. L. Forebrain mechanisms of nociception and pain: analysis throughimaging. Proc. Natl. Acad. Sci. USA 96, 7668–7674 (1999).

25. Critchley, H. D., Mathias, C. J. & Dolan, R. J. Neural activity in the humanbrain relating to uncertainty and arousal during anticipation. Neuron 29,537–545 (2001).

articles©

2002

Nat

ure

Pu

blis

hin

g G

rou

p

htt

p:/

/neu

rosc

i.nat

ure

.co

m

26. Damasio, A. The Feeling of What Happens: Body and Emotion in the Making ofConsciousness (Harcourt Brace, New York, 1999).

27. Critchley, H. D., Mathias, C. J. & Dolan, R. J. Neuroanatomical basis for first-and second-order representations of bodily states. Nat. Neurosci. 4, 207–212(2001).

28. Breiter, H. C. et al. Response and habituation of the human amygdala duringvisual processing of facial expression. Neuron 17, 875–887 (1996).

29. Isenberg, N. et al. Linguistic threat activates the human amygdala. Proc. Natl.Acad. Sci. USA 96, 10456–10459 (1999).

30. Morris, J. S., Buchel, C. & Dolan, R. J. Parallel neural responses in amygdalasubregions and sensory cortex during implicit fear conditioning. Neuroimage13, 1044–1052 (2001).

31. Dolan, R. J., Morris, J. S. & de Gelder, B. Crossmodal binding of fear in voiceand face. Proc. Natl. Acad. Sci. USA 98, 10006–10010 (2001).

32. Morris, J. S. et al. A neuromodulatory role for the human amygdala inprocessing emotional facial expressions. Brain 121, 47–57 (1998).

33. Amaral, D. G. & Price, J. L. Amygdalo-cortical projections in the monkey(Macaca fascicularis). J. Comp. Neurol. 230, 465–496 (1984).

34. Anderson, A. K. & Phelps, E. A. Lesions of the human amygdala impairenhanced perception of emotionally salient events. Nature 411, 305–309(2001).

35. Bonda, E., Petrides, M., Ostry, D. & Evans, A. Specific involvement of humanparietal systems and the amygdala in the perception of biological motion.J. Neurosci. 16, 3737–3744 (1996).

36. Castelli, F., Happe, F., Frith, U. & Frith, C. Movement and mind: a functionalimaging study of perception and interpretation of complex intentionalmovement patterns. Neuroimage 12, 314–325 (2000).

37. Fletcher, P. C. et al. Other minds in the brain: a functional imaging study of‘theory of mind’ in story comprehension. Cognition 57, 109–128 (1995).

38. Gallagher, H. L. et al. Reading the mind in cartoons and stories: an fMRIstudy of ‘theory of mind’ in verbal and nonverbal tasks. Neuropsychologia 38,11–21 (2000).

39. Brunet, E., Sarfati, Y., Hardy-Bayle, M. C. & Decety, J. A PET investigation ofthe attribution of intentions with a nonverbal task. Neuroimage 11, 157–166(2000).

40. Frith, C. D. & Frith, U. Interacting minds—a biological basis. Science 286,1692–1695 (1999).

41. Frith, C. & Frith, U. in Understanding Other Minds: Perspectives FromDevelopmental Cognitive Neuroscience (eds. Baron-Cohen, S., Tager-Flusberg,H. & Cohen, D. J.) 334–356 (Oxford Univ. Press, New York, 2000).

42. Elliott, R. & Dolan, R. J. Neural response during preference and memoryjudgments for subliminally presented stimuli: a functional neuroimagingstudy. J. Neurosci. 18, 4697–4704 (1998).

43. Phillips, M. L. et al. Neural responses to facial and vocal expressions of fearand disgust. Proc. R. Soc. Lond. B Biol. Sci. 265, 1809–1817 (1998).

44. Friston, K. et al. Spatial registration and normalization of images. Hum. BrainMapp. 2, 165–189 (1995).

45. Talairach, J. & Tournoux, P. Co-planar Stereotaxic Atlas of the Human Brain(Theime, Stuttgart, Germany, 1988).

46. Ekman, P. & Friesen, W. V. Pictures of Facial Affect (Consulting PsychologistsPress, Palo Alto, California, 1975).

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

48. Friston, K. et al. Statistical parametric maps in functional imaging: a generallinear approach. Hum. Brain Mapp. 2, 189–210 (1995).

49. Worsley, K. et al. A unified statistical approach for determining significantsignals in images of cerebral activation. Hum. Brain Mapp. 4, 58–73 (1996).

50. Duvernoy, H. M. The Human Brain (Springer, Vienna, 1999).

articles

nature neuroscience • volume 5 no 3 • march 2002 283

©20

02 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://n

euro

sci.n

atu

re.c

om