Nature Neuroscience April 2004

VOLUME 7 NUMBER 4 APRIL 2004

E D I TO R I A L

315 Testing a radical theory

B O O K R E V I E W

317 The Birth of the Mind: How a Tiny Number of Genes Creates the Complexities ofHuman Thoughtby Gary MarcusReviewed by Charles Jennings

N E W S A N D V I E W S

319 Unpredictable primates and prefrontal cortexMichael L Platt see also p 404

321 Synaptic vesicles really do kiss and runR Mark Wightman & Christy L Haynes see also p 341

322 Stiffening the spinesAnnette Markus see also p 357

323 Choices, choices, choicesJerold Chun

325 Imaging gender differences in sexual arousalTurhan Canli & John D E Gabrieli see also p 411

R E V I E W

327 Neural activity and the dynamics of central nervous system developmentJackie Yuanyuan Hua & Stephen J Smith

B R I E F COM M U N I C AT I O N S

333 Reverse propagation of sound in the gerbil cochleaT Ren

335 Cholecystokinin-mediated suppression of feeding involves the brainstemmelanocortin systemW Fan, K L J Ellacott, I G Halatchev, K Takahashi, P Yu & R D Cone

i

CCK-mediated satietyand brainstem melanocortin

(p 335)

In competitive games, the outcome ofone player’s choices often depends on

the strategy chosen by eachopponent. Lee and colleagues nowshow that activity in the prefrontal

cortex may provide signals to updateestimates of expected reward inmonkeys playing a simple game

against a computer opponent. Suchsignals could underlie the generation

of random behavior for strategicpurposes. The authors also used a

reinforcement-learning algorithm topredict the monkeys’ choices.

(pp 319 and 404)

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature Publishing Group, a trading name of Nature America Inc. located at 345 ParkAvenue South, New York, NY 10010-1707. Editorial Office: 345 Park Avenue South, New York, NY 10010-1707. Tel: (212) 726 9200, Fax: (212) 6969635. Annual subscription rates: USA/Canada: US$199 (personal), US$99 (student), US$129 (postdoc). Canada add 7% GST #104911595RT001;Euro-zone: € 289 (personal), € 163 (student), € 196 (postdoc); Rest of world (excluding China, Japan, Korea): £175 (personal), £99 (student), £119(postdoc); 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 material for internal or personal use, or internal or personal use of specific clients, is granted byNature Publishing Group to libraries and others registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided the rel-evant copyright fee is paid direct to CCC, 222 Rosewood Drive, Danvers, MA 01923, USA. Identification code for Nature Neuroscience: 1097-6256/04.Back issues: US$45, Canada add 7% for GST; Periodicals postage rate paid at New York, NY and additional mailing offices. 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 © 2004 Nature Publishing Group. Printed in USA.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

VOLUME 7 NUMBER 4 APRIL 2004

337 A psychophysical test of the vibration theory of olfactionA Keller & L B Vosshall

339 Parietal somatosensory association cortex mediates affective blindsightS Anders, N Birbaumer, B Sadowski, M Erb, I Mader, W Grodd & M Lotze

A R T I C L E S

341 Dopamine neurons release transmitter via a flickering fusion poreR G W Staal, E V Mosharov & D Sulzer see also p 321

347 Tenascin-R mediates activity-dependent recruitment of neuroblasts in the adultmouse forebrainA Saghatelyan, A de Chevigny, M Schachner & P Lledo

357 Stability of dendritic spines and synaptic contacts is controlled by αN-cateninK Abe, O Chisaka, F van Roy & M Takeichi see also p 322

364 The X-linked mental retardation protein oligophrenin-1 is required for dendriticspine morphogenesisE Govek, S E Newey, C J Akerman, J R Cross, L Van der Veken & L Van Aelst

373 Local structural balance and functional interaction of excitatory and inhibitorysynapses in hippocampal dendritesG Liu

380 Synaptic dynamics mediate sensitivity to motion independent of stimulus detailsH Luksch, R Khanbabaie & R Wessel

389 Differential control over cocaine-seeking behavior by nucleus accumbens coreand shellR Ito, T W Robbins & B J Everitt

398 Glutamatergic activation of anterior cingulate cortex produces an aversiveteaching signalJ P Johansen & H L Fields

404 Prefrontal cortex and decision making in a mixed-strategy gameD J Barraclough, M L Conroy & D Lee see also p 319

411 Men and women differ in amygdala response to visual sexual stimuliS Hamann, R A Herman, C L Nolan & K Wallen see also p 325

N AT U R E N E U R O S C I E N C E C L A S S I F I E D

See back pages.

NATURE NEUROSCIENCE i i i

αN-catenin regulates dendritic and synaptic stability

(p 357)

Oligophrenin-1 is required for dendritic spine morphogenesis

(p 364)

Gender differences in amygdalaresponses to erotic images

(pp 325 and 411)

Tenascin-R mediates radialmigration in the olfactory bulb

(p 347)

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004 315

E D I TO R I A L

T he paper by Keller and Vosshall on page 337 of this issue isunusual; it describes a refutation of a theory that, whileprovocative, has almost no credence in scientific circles. The

only reason for the authors to do the study, or for Nature Neuroscienceto publish it, is the extraordinary—and inappropriate—degree ofpublicity that the theory has received from uncritical journalists.

The theory, from Luca Turin (formerly of University CollegeLondon), concerns the mechanism of olfactory transduction.Olfaction is not well understood compared to the other senses, butmost experts believe that odorant molecules bind to specific receptorsthrough conventional molecular interactions, causing a conforma-tional change in the receptor that leads to activation of intracellularsignals. Admittedly there are no clear demonstrations (apart from onestudy1 in C. elegans) that a specific receptor binding to an odorantmediates the perceptual response to that odorant, and there are someanomalies, such as molecules that smell the same despite their lack ofchemical similarity. However, this could be explained through subse-quent neural processing (if for example receptors with different speci-ficity were to activate common targets within the brain).

Turin proposed a very different theory, namely that olfactoryreceptors act like a spectroscope to detect intramolecular vibrationswithin the odorant molecule. According to this idea, the perceptualquality of an odorant is determined not by its shape but by its vibra-tional spectrum. This would be of great importance if true, but radi-cal ideas require strong evidence, which Turin did not provide. Nordid he provide a detailed explanation of how these molecular vibra-tions could lead to neural activation.

The magician James Randi, debunker of paranormal claims, oncesaid that if you claim to have a goat in your back yard, people willprobably believe you, but if you say you have a unicorn, you mustexpect closer scrutiny. The editors at Nature used to classify manu-scripts on a ‘zoological scale’ that ranged from goats to unicorns, andTurin’s paper was toward the far end of that scale. Despite the force-fulness of his assertions, most scientists in the field were unconvincedby his proposal. Thus his paper was rejected by Nature, and it waseventually published (without review, according to Turin’s ownaccount) by Chemical Senses in 1996.

Turin’s theory would probably have vanished into obscurity but fortwo coincidences. First, one of his former students had become a pro-ducer for the BBC, and she decided to make a TV documentary abouthim. Second, he had a chance encounter with writer Chandler Burr,who was so taken with the theory that he wrote a popular book aboutit. The Emperor of Scent, which appeared in 2002, is effectively amouthpiece for Turin’s views, and it is intensely hostile to the scien-tific establishment. It has attracted wide attention, and with theexception of a scathing review in this journal from Avery Gilbert2, thereviews have been almost uniformly favorable. The book is seduc-tively written, and it was recently reissued in paperback, completewith a readers’ guide to promote book club discussions.

The villains of Burr’s book include many of the leading figures inthe olfactory community, who are portrayed as ignorant and incom-petent reactionaries, along with the journal editors who rely on theiradvice. Burr’s overall verdict is that Turin’s failure to convince the sci-entific establishment of his views reflects “scientific corruption…inthe most mundane and systemic and virulent and sadly human senseof jealousy and calcified minds and vested interests.”

Many olfactory researchers were dismayed by the book and by theapparent willingness of the media to accept Burr’s verdict. Kellerand Vosshall were sufficiently upset that they decided to put Turin’stheory to an experimental test. As described in their paper, theytested three claims of the vibration theory, all of which featureprominently in Burr’s book. The experiments were conducted double-blind, and in all three cases the results were negative. Turinhimself had no role in designing the study, and one could debate (asTurin probably will) whether this study constitutes a definitive refu-tation of his theory. A conservative statement would be that Turin’sclaims are not reproducible based on the information provided inhis own publications. At the least, the burden of proof for confirma-tion of his theory is now unambiguously transferred to Turin, whereit should have been all along.

In some sense it does not matter whether the public believes in thevibrational theory of olfaction; the truth will eventually come out.But of course this is not just about olfaction. It is about the publiccredibility of the scientific process and the biases that affect sciencereporting in the popular press. It is disturbing that Emperor of Scentreceived so much favorable publicity from reviewers who were illqualified to judge its scientific content. The New York Times and The Washington Post, for instance, assigned it to their movie critic andfashion critic, respectively.

The media loves controversy, and ever since David and Goliath,the story of a lone hero taking on the establishment has had endur-ing appeal. Of course, radical ideas from outside the mainstream dooccasionally turn out to be right. Of course scientists are some-times excessively attached to conventional ideas. But in science atleast, the mainstream view is usually based on the accumulation ofevidence over many years. Journalists are trained to report bothsides of any argument, but this can be misleading when both sidesare not equally credible.

A mature body of scientific theory is like a large building, and theimpulse to demolish it is often little more than a form of intellec-tual vandalism, an expression of frustration by those who did notsucceed as architects. Some buildings outlive their usefulness, ofcourse, but the threshold for knocking them down should be high.We hope that the paper from Keller and Vosshall will serve as areminder of why that’s so.

1. Sengupta, P., Chou, J.H. & Bargmann, C.I. Cell 84, 899–909 (1996).

2. Gilbert, A.N. Nat. Neurosci. 6, 335 (2003).

Testing a radical theory

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Unpredictable primates and prefrontal cortexMichael L Platt

In many competitive games, players need to behave unpredictably so that their opponents cannot anticipate the next move. Newrecordings from monkeys playing a computer game support the idea that neurons in the prefrontal cortex may control thisbehavior and imply there is a cost for generating random behavior, which monkeys avoid unless the opponent is sophisticated.


When the United States and China played toa 0–0 draw at the end of regulation time inthe 1999 Women’s World Cup soccer finals,the tie-breaking shoot-out became one ofthe most memorable moments in sports. Tobe successful, kickers must be unpredictableor risk exploitation by the goalie1. As it onlytakes about a half second for the ball toreach the goal, goalies must anticipatewhether the ball will be kicked left or right,and leap to block the kick before they canvisually determine its path. With the 1999shoot-out tied at 4 apiece, right-footedkicker Brandi Chastain unpredictably left-footed a laser shot to the left of goalie GaoHong for the US win.

In this issue, Baraclough and colleagues2

demonstrate that unpredictable behavioralresponses in such dynamic competitive con-texts may be driven by signals carried by neu-rons in the prefrontal cortex. These cellsapparently carry sufficient informationabout prior choices and their outcomes toguide the production of strategically unpre-dictable behavior.

Despite great progress in understandingthe neural mechanisms responsible for per-ception and movement, the neural basis forsimple decision making has been poorlyunderstood until recently. In the past fewyears, however, neural correlates of stimulusstrength have been reported for a number ofsensory brain areas in subjects making sim-ple perceptual judgments3,4. Moreover, neu-ral correlates of movement value have beenfound in several cortical and subcorticalbrain areas in subjects choosing betweenunequally rewarded alternatives5,6. Such sig-

nals are prerequisites for making informedand economically advantageous choices.

In all these studies, the relationshipsbetween stimulus, response and outcomewere fixed. That is, choosing the single move-ment with the highest expected value wasalways the optimal course of action. In socialcontexts, however, decision outcomes are notdeterministic but vary depending on thechoices made by other individuals, makingprediction difficult. Game theory has beendeveloped in the social sciences to predictand explain behavior under these circum-stances7. Game theoretic models posit thatplayers evaluate the costs and benefits of eachalternative to themselves and their oppo-nents and then adopt a behavioral strategy.Typically, these behavioral strategies com-prise a probabilistic distribution of responsesthat settles at an equilibrium for all players.Equilibria of this sort are often known asNash equilibria, after the Nobel prize-win-

ning mathematician, and the resultantbehavioral strategies dominate all others. Forexample, in the children’s game ‘rock-paper-scissors’, the best response for all players is tounpredictably play each alternative one-thirdof the time. Any more predictable responsecan be easily exploited by other players.

Baraclough and colleagues2 have signifi-cantly advanced the study of the neural basisof decision making by applying such a game-theoretic approach. The authors recorded theactivity of neurons in the dorsolateral pre-frontal cortex of monkeys playing a simplegame against a computer opponent. Neuronsin this area are sensitive to stimulus location,movement preparation, working memory8

and reward expectation9. Moreover, neuro-physiological10 and neuroimaging11 studieshave implicated prefrontal cortex in encod-ing information used to make decisions.

In the new study2, two monkeys played anoculomotor version of ‘matching pennies,’ a


The author is in the Department of Neurobiology,

Duke University Medical Center, Durham,

North Carolina 27710, USA.

e-mail: [email protected]

Figure 1 Experimental design. (a) Payoff matrix for a strategic eye movement game played by monkeysagainst a computer opponent. Each monkey stared at a central circle (not shown) on a computer monitor,two peripheral green circles were illuminated, and then the central circle was turned off. The monkeythen looked to either the left or right circle. A red ring around the target revealed the computer’sselection after the monkey looked. If the computer and the monkey selected the same target, a squirt offruit juice was delivered. (b) Action potentials were recorded extracellularly from neurons in the monkeyprefrontal cortex (PFC), near the principal sulcus and anterior to the frontal eye fields.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


standard game in which the optimal strategyis to behave randomly and unpredictablyfrom trial to trial. The monkey looked at acomputer screen, saw two targets, and thenmoved its eyes to either target (Fig. 1). If themonkey chose the target selected by the com-puter, he received a squirt of fruit juice. If hechose the other target, he received nothing.

Monkeys playing this game developed different decision strategies depending on the algorithm implemented by the computeropponent. When the computer played ran-domly without regard to the monkeys’ choices,they developed spatial biases favoring one tar-get over another. As the computer rewardedtargets at random, this strategy was perfectlyreasonable. When the computer tracked onlythe monkeys’ choices, however, they adopted awin–stay, lose–shift strategy. Thus, if a monkeychose left and was not rewarded, he chose theother target on the next trial. Most impor-tantly, when the computer tracked both themonkeys’ choices and rewards, the monkeysdeveloped a strategy of choosing randomly oneach trial—the optimal strategy in the classic‘matching pennies’ game.

The authors then determined whether areinforcement learning model could accountfor the monkeys’ choices. This modelassumes that choices are made based on dif-ferences in the value functions associatedwith each alternative, which are determinedby the prior history of rewards received forchoosing each target. The model parametersdiffered in predictable ways depending onthe monkeys’ strategies. Most importantly,when monkeys played against the mostsophisticated computer program, the valuedifferences were very small, indicating thatchoices were weakly, but systematically, influ-enced by the outcome of prior choices. Thus,the monkeys behaved in a way that made itdifficult for the computer opponent to pre-dict their choices reliably. These results sug-gest that the monkeys may have converged onthe optimal strategy using a reinforcementlearning algorithm.

The authors found that many neurons inprefrontal cortex were systematically modu-lated by prior reward outcomes as well as byprior choices. Most importantly, many pre-frontal neurons were modulated by a con-junction of these two factors. For example,one prefrontal neuron in this study firedpreferentially when the monkey had selectedthe right-hand target on the previous trialand was not rewarded. Other prefrontal neu-rons were sensitive to different conjunctionsof choice and reward outcome.

The authors compared the activity of pre-frontal neurons on the strategic decision task

with activity evoked in a control task. In thistask, one target turned red, cuing the monkeythat shifting gaze to the second target wouldbe rewarded. Comparing neuronal activity inthe two tasks dissociated responses relatedpurely to movement or reward from thoserelated to making decisions. Such analysesrevealed prefrontal signals specific to strate-gic decision making, which were absent oreven reversed in the control task.

This study is an elegant and novel application of game theory to understand-ing strategic decision making in monkeys.The data imply that there is a cost to gener-ating random behavior, which monkeysavoid unless confronted with a sophisticatedopponent. Indeed, normal human subjectsappear to be quite poor at generating ran-dom sequences, but with extensive practicebecome more adept12, much like the simianjuice experts in this study.

The results also suggest that prefrontalneurons carry sufficient information to guidethe behavioral choices made by monkeys inthis task. The authors posit that neural cir-cuits responsible for maintaining persistentactivity in prefrontal cortex during standardworking memory tasks may also serve tointegrate reward and choice history in thissimple strategic game. Theoretically, thesecircuits could provide signals needed toupdate value functions using a reinforcementlearning model, and thereby guide the gener-ation of optimally random choices.

This study raises several intriguing ques-tions. First, how are the prefrontal signalsobserved in this strategic game related todecision signals observed in other corticaland subcortical areas? After all, neurons inparietal cortex, anterior cingulate cortex,posterior cingulate cortex, superior collicu-lus and basal ganglia are sensitive to thevalue of a particular movement, and thisvalue is often predicated on the prior his-tory of choices13,14. It would be interestingto know whether neurons in these otherareas continue to signal the value of a par-ticular movement in these strategic con-tests. Indeed, the results of one such studysuggest that parietal neurons encode thevalue of a particular eye movement, inde-pendent of movement probability, in mon-keys playing a strategic game at variousNash equilibria (Dorris, M.C. & Glimcher,P.W., Soc. Neurosci. Abstr., 27, 58.10, 2001).

A second question regards the necessity ofdorsolateral prefrontal cortex for strategicdecision making. Would reversible deactiva-tion of this area, for example, render mon-keys unable to generate random behavior inthe oculomotor matching-pennies task?

Moreover, would such deactivation result inthe degeneration of value signals at othernodes in the decision network, such as pari-etal cortex? This would suggest that the valuesignals observed in these areas are computedfrom the prefrontal signals recorded byBaraclough and colleagues2.

Finally, game theoretic approaches wereoriginally developed to predict and explainthe choices made by individuals interactingwith others. The unpredictable randombehavior of the monkeys in the Baracloughstudy is the optimal strategy when playingagainst an intelligent opponent. But did themonkeys actually treat the task as a com-petitive struggle with a savvy adversary? Wemight expect different outcomes if mon-keys played this game face to face withanother monkey. Under those conditions,social rules and conventions—such asdominance rank or a sense of fairness15—might exert a powerful influence on deci-sion making. The influence of culturalnorms on strategic decision making inhumans is well documented12.

The new paper2 will likely be a fundamen-tal contribution to the literature. The appli-cation of game theory to theneurophysiology of decision making is newand noteworthy, and accomplished with ele-gance and finesse. Baraclough and colleagueshave shown in their landmark study that for-malisms developed in the social sciences topredict and explain strategic behavior offer apowerful tool for understanding the neuralbasis of decision making.

1. Palacios-Huerta, I. Brown University working paper,cited in Camerer, C.F. Behavioral Game Theory(Princeton Univ. Press, Princeton, New Jersey,2003).

2. Baraclough, D.J., Conroy, M.L. & Lee, D. Nat.Neurosci. 7, 404–410 (2004).

3. Schall, J.D. & Hanes, D.P. Nature 366, 467–469(1993).

4. Shadlen, M.N. & Newsome, W.T. Proc. Natl. Acad.Sci. USA 93, 628–633 (1996).

5. Platt, M.L. Curr. Opin. Neurobiol. 12, 141–148(2002).

6. Glimcher, P.W. Decisions, Uncertainty, and the Brain(MIT Press, Cambridge, Massachusetts, 2003).

7. Kreps, D. Game Theory and Economic Modeling(Oxford Univ. Press, Oxford, UK, 1990).

8. Funahashi, S., Bruce, C.J. & Goldman-Rakic, P.S. J. Neurophysiol. 61, 1–19 (1989).

9. Leon, M.I. & Shadlen, M.N. Neuron 24, 415–425(1999).

10. Kim, J.N. & Shadlen, M.N. Nat. Neurosci. 2,176–185 (1999).

11. Sanfey, A.G., Rilling, J.K., Aronson, J.A., Nystrom,L.E. & Cohen, J.D. Science 300, 1673–1675(2003).

12. Camerer, C.F. Behavioral Game Theory (PrincetonUniv. Press, Princeton, New Jersey, 2003).

13. Platt, M.L. & Glimcher, P.W. Nature 400, 233–238(1999).

14. Coe, B., Tomihara, K., Matsuzawa, M. & Hikosaka, O.J. Neurosci. 22, 5081–5090 (2002).

15. Brosnan, S.F. & De Waal, F.B. Nature 425, 297–299(2003).

320 VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


Modern consumers realize that recycling isa good practice but that reusing products iseven better. Scientists have long debated therelative contributions of vesicle recyclingand reuse after neurotransmitter release aswell. In the classic mechanism, dubbed ‘all-or-none’ exocytosis (Fig. 1a), a synapticvesicle fuses with the presynaptic mem-brane and releases its contents into thesynapse; the vesicle membrane is then recy-cled. Alternately, a synaptic vesicle can forma transient fusion pore in the presynapticmembrane and release only part of its con-tents; in such ‘kiss-and-run’ exocytosis, thevesicle is then reused. In this issue, Staal andcoworkers1 clearly establish that smallsynaptic vesicles in dopaminergic neuronsalmost exclusively use the kiss-and-runmechanism of exocytosis.

Exocytosis is almost universally accepted asthe primary means of chemical communica-tion between neurons. Electron microscopy offreeze-fractured tissue has captured ‘omega’structures at the surface of stimulated neu-rons2. Vesicular contents are released in con-centration ratios that reflect their storedamounts3. Single-cell capacitance changes dur-ing exocytosis, indicating an increase in the cel-lular membrane area4. Finally, discrete packetsof released chemicals can be detected withamperometry5, in which the number of easilyoxidized molecules released from a vesicle ismeasured with carbon-fiber microelectrodes.

Much of the current view of exocytosisfrom small synaptic vesicles has beenextrapolated from results obtained withlarge, dense-core vesicles. Cells with largevesicles primarily use all-or-none exocyto-sis, and only occasionally kiss-and-run. All-or-none exocytosis of large vesicles issupported by whole-cell capacitance meas-urements4 as well as by amperometry,which shows that the amount of transmitterreleased from a variety of cell types corre-sponds well with known intravesicularamounts6. Video microscopy of chromaffin

cells loaded with a fluorescent dye thataccumulates in vesicles also illustrates all-or-none exocytosis: the fluorescent vesiclescan be seen to approach the plasma mem-brane and then completely lose their fluo-rescence7 (Fig. 1a).

The kiss-and-run mechanism was clearlyshown in mast cells by capacitance measure-ments with a patch-clamp electrode andsimultaneous amperometry8. The correlateddata showed that increases in membranearea are not always accompanied by fullextrusion of the vesicle contents. In a studyusing fast-scan cyclic voltammetry, an inter-mediate ‘kiss-and-hold’ state was induced inboth mast cells and chromaffin cells byincreasing the osmolarity of the extracellularsolution9. This removed the normal osmoticgradient between the vesicle interior and theextracellular space, preventing efflux of theintravesicular contents.

The applicability of principles derivedfrom large synaptic vesicles to small vesiclesin neurons has been unclear. Theoreticalconsiderations predicted that the modestsurface tension changes that accompanyfusion of a small synaptic vesicle wouldincrease the likelihood of the kiss-and-runmode10. However, the restricted populationof small vesicles within neurons—along withtheir extremely small size (diameters of 50nm or less) and the limited number of mole-cules they contain—have made it very difficult to experimentally characterize themode of small-vesicle exocytosis. Traditionalcapacitance measurements have been unsat-isfactory because the fusion of a small sur-face-area vesicle has been undetectable. Onlyrecently did Klyachko and Jackson manage todrastically reduce the noise level, enablingthem to detect the fusion of secretory vesiclessimilar in size and shape to small synapticvesicles using capacitance measurements11.In this proof-of-concept study, estimates ofthe fusion pore diameter were smaller thanthe neurotransmitter molecules that passthrough the pore. Nevertheless, this tech-nique now shows promise for the study ofsmall vesicles.

Fluorescence imaging of single neuronalvesicles is also difficult because the size ofsynaptic vesicles is smaller than the diffrac-

tion limit of standard fluorophore emissionwavelengths. Furthermore, fluorescencechanges are difficult to interpret becausethey do not necessarily correlate with the neurotransmitter efflux12. However,Gandhi and Stevens13 used a photobleach-ing process to show small but discernabledifferences in fluorescence traces of indi-vidual dye-loaded vesicles that suggestedkiss-and-run accompanying other forms ofexocytosis. Direct efflux experiments arealso difficult because neuronal vesicles typi-cally contain only 3,000 to 30,000 neuro-transmitter molecules, far fewer than themillion or so contained in the large vesiclesdescribed above. Amperometry, however,can be used to detect even such a smallnumber of molecules, and it allows real-time measurement of exocytotic events.

Staal and colleagues have now clarified thesituation by using amperometry with carbon-fiber microelectrodes to directly monitor exo-cytosis of endogenous dopamine fromcultured ventral midbrain neurons1.Measuring K+-stimulated dopamine releasefrom individual neurons, the authors identi-


Synaptic vesicles really do kiss and runR Mark Wightman & Christy L Haynes

A new study demonstrates that small synaptic vesicles exocytose dopamine through a flickering fusion pore almost exclusively, aprocess known as ‘kiss-and-run’ exocytosis. This process is driven by the need for efficient use of few synaptic vesicles.

The authors are in the Department of Chemistry,

University of North Carolina, Chapel Hill,

North Carolina 27599-3290, USA.


Time

Time

Time

c

b

a

Figure 1 Exocytotic mechanisms for smallsynaptic vesicles. Schematic drawings of themechanisms (above) and amperometric currenttraces (below). (a) Full fusion of a small synapticvesicle after initially forming a small fusion pore.After secretion, the vesicle is temporarilyincorporated into the plasma membrane. (b) Asimple kiss-and-run event. (c) A complex kiss-and-run event with three subunits, eachdecreasing in amplitude.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


fied two classes of release events. In simpleevents, all the dopamine released from a givensynaptic vesicle-membrane fusion site wasmeasured in a single amperometric peak. Incomplex events, the dopamine released from agiven synaptic vesicle-membrane fusion sitewas measured as a series of discrete peaks. Theauthors’ interpretation is that a simple eventconsists of a small synaptic vesicle generating afusion pore in the presynaptic membrane,partially discharging its contents into thesynaptic cleft, and then disconnecting fromthe membrane (Fig. 1b). A complex eventoccurs when the fusion pore flickers rapidlybetween an open and closed form, allowingrepeated partial release of vesicle contents(Fig. 1c). Comparison of the amperometrictraces from simple and complex events sup-ports this interpretation: the number ofdopamine molecules oxidized in a simpleevent is roughly equivalent to the number ofdopamine molecules oxidized in the first sub-unit of a complex event. Thus, both simpleand complex events seem to reflect kiss-and-run exocytosis.

The authors found that small synaptic vesi-cles in midbrain dopaminergic neuronsundergo kiss-and-run exocytosis almostexclusively. Kiss-and-run exocytosis is advan-tageous because it leads to increased longevityof a synaptic vesicle, thereby decreasing theimportance of the relatively slow process ofvesicle recycling through the endosomal com-partment. The authors suggest that such effi-cient vesicle use is necessary because of therelatively small number of synaptic vesiclespresent in these midbrain neurons. Kiss-and-run exocytosis also avoids inefficient use ofdopamine at synapses that lack well-definedactive zones, as is typical of dopaminergicneurons.

The complex form of kiss-and-run mayrepresent a particularly economical form ofexocytosis, which may be advantageous iftransmitter-loaded vesicles are in short supply.To test this hypothesis, the researchersexposed the cultured neurons to pharmaco-

logical agents affecting the secondary messen-gers that regulate synaptic vesicle cycling. Onaddition of a phorbol ester, an agent thatincreases the number of releasable synapticvesicles, amperometric traces revealed a rela-tive decrease in the number of complex eventsfrom 20% to 6%. After inhibition of proteinkinase C, reducing the number of releasablevesicles, the total number of exocytotic eventsper stimulus was decreased by 82%, butamperometric traces showed a relativeincrease from 20% to 40% in the number ofcomplex events. Thus complex events appearto be favored when fewer releasable vesiclesare available.

If the nature of exocytotic mechanisms isdetermined by the number of vesicles and thenature of the synapse, comparison of the newdata1 with similar data collected from cellswith large dense-core vesicles8 should revealsignificant variation. Kiss-and-run occurs inboth cases, but there are some notable differ-ences. First, the amperometric trace subunitduration is approximately 200 times shorter insmall synaptic vesicles. Second, the fusionpore flickering occurs with a ten-fold increasein frequency in small synaptic vesicles com-pared to the large dense-core vesicles. Third,the small synaptic vesicles release 25–30% oftheir dopamine cargo with each flicker of thefusion pore, whereas the large dense-core vesi-cles release <1% of their dopamine. Clearly,although the same exocytotic mechanism is atwork, the fusion pore flickering characteristicsare greatly influenced by the size of the vesicleand the function of the cell.

Some questions remain. The new research1

suggests that kiss-and-run exocytosis is drivenby the need for efficient use of a relativelysmall number of synaptic vesicles. Thishypothesis can be further tested by measuringthe relative number of full fusion and kiss-and-run events at presynaptic terminals inneurons with a larger number of vesicles, andin neurons that use other neurotransmitters.Amperometry can only detect easily oxidizedneurotransmitters such as dopamine, so new

strategies will need to be developed for othertransmitters such as glutamate. Extracellularcalcium is central in regulating exocytosis andrelease probabilities, and it will be fascinatingto explore its influence on the characteristicsof kiss-and-run and the flickering pore.Ideally, this would entail amperometric meas-urements and simultaneous calcium imaging.Because the small synaptic vesicles releasesuch a large percentage of their total neuro-transmitter concentration with each flicker ofthe fusion pore, it would also be interesting tomanipulate the intravesicular contents to seewhether the kiss-and-hold state can beinduced in neurons. Large vesicles forced intoa kiss-and-hold state through increased extra-cellular osmotic pressure undergo massiverelease when returned to isotonic conditions.Will similar manipulations force the smallsynaptic vesicles from kiss-and-run to fullfusion exocytosis? Future work will tell uswhether non-dopaminergic neurons also use anearly exclusive kiss-and-run mechanism ofexocytosis and will explore the implications ofkiss-and-run vesicle re-use for the synapticvesicle recycling mechanism.

1. Staal, R.G.W., Mosharov, E.V. & Sulzer, D. Nat.Neurosci. 7, 341–346 (2004).

2. Heuser, J.E. Q. J. Exp. Physiol. 74, 1051–1069(1989).

3. Viveros, O.H. in Handbook of Physiology Vol. 6 (eds.Blaschko, A. & Smith, A.D.) 389–426 (AmericanPhysiological Society, Washington, D.C., 1975).

4. Neher, E. & Marty, A. Proc. Natl. Acad. Sci. USA 79,6712–6716 (1982).

5. Wightman, R.M. et al. Proc. Natl. Acad. Sci. USA 88,10754–10758 (1991).

6. Finnegan, J.M. et al. J. Neurochem. 66, 1914–1923(1996).

7. Steyer, J.A. & Almers, W. Biophys. J. 76, 2262–2271(1999).

8. Alvarez de Toledo, G., Fernandez-Chacon, R. &Fernandez, J.M. Nature 363, 554–558 (1993).

9. Troyer, K.P. & Wightman, R.M. J. Biol. Chem. 277,29101–29107 (2002).

10. Amatore, C., Bouret, Y., Travis, E.R. & Wightman, R.M.Angew. Chem. Int. Ed. Engl. 39, 1952–1955 (2000).

11. Klyachko, V.A. & Jackson, M.B. Nature 418, 89–92(2002).

12. Aravanis, A.M., Pyle, J.L., Harata, N.C. & Tsien, R.W.Neuropharmacology 45, 797–813 (2003).

13. Gandhi, S.P. & Stevens, C.F. Nature 423, 607–613(2003).


Stiffening the spines

The ability of dendritic spines to change shape in response to synaptic activity is crucial for synaptic plasticity.This motility is regulated by αN-catenin, report Abe et al. on page 357. Overexpression of αΝ-catenin (green;red is PSD95) stabilized spines in cultured neurons, reducing turnover and thereby increasing their number.Lack of αN-catenin increased spine motility, even at established synaptic contacts. Spine αN-catenin was reg-ulated by synaptic activity: blocking activity with tetrodotoxin reduced αN-catenin staining (and increasedspine motility), whereas blocking inhibitory neurotransmission increased αN-catenin.The catenins link cad-herin cell adhesion molecules to the cytoskeleton, so αN-catenin is well placed to regulate spine dynamics.

Annette Markus

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


Mammalian olfactory sensory neurons havea difficult decision to make. From over athousand possible choices, each sensory neu-ron must pick only one type of olfactoryreceptor (OR) gene to express1. How this isaccomplished is still unclear. In the immunesystem, the diversity of immunoglobulinsand T-cell receptors arises through a processcalled ‘VDJ recombination’ (Fig. 1a)2. Themechanism (more generally called somaticDNA rearrangement) involves cutting seg-ments of DNA from non-germline tissue andrejoining the segments to form compositegenes, producing permanent changes inDNA and gene expression that are not passedon to future generations. The size of the ORgene family3 and its genomic organization,with over 1,000 OR genes dispersed in linearclusters and on different chromosomes4–6,raised the possibility of DNA rearrangementas a mechanism for receptor gene choice inthe olfactory system, too. If such a mecha-nism were found in mammalian neurons, itmight help to explain the brain’s complexityand diversity of connections and cell types.Now, in the first rigorous test of this hypoth-esis, two independent reports in Nature7,8

have cloned mice from single olfactory recep-tor neurons: neither finds evidence for DNArearrangements.

Of course, models for explaining how ORsare expressed need not invoke DNA rearrange-ment (Fig. 1b,c)1,4, and at least two key differ-ences exist between OR genes and immune cellreceptors. First, the entire OR is encoded by onecontiguous stretch of DNA (a single exon),negating a need for combining gene segments.Second, there are no obvious rearrangement-recognition markers flanking OR genes (recom-bination signal sequences, heptamer/nonamercis-elements), which are required for VDJrearrangement. Thus, the mechanism for DNArearrangement of OR genes would have to be

distinct from that observed in the immune sys-tem. One possible alternative mechanism mightinvolve insertion of a transposable element with promoter/enhancer activity, which might directexpression of one OR from a basal promoter(Fig. 1d).

To examine whether DNA rearrangementoccurs in OR genes, one must first devise amethod for detecting the rearrangement.Historically, clonal cell lines have been used,as they can be grown indefinitely to generatelarge amounts of DNA with the samerearrangement that, once amplified, can bedetected by standard techniques. Such anapproach was used to first identify

immunoglobulin DNA rearrangements2.However, no clonal cells lines expressing ORgenes exist. Theoretically, single-cell PCRapproaches might also work, but in theabsence of defined OR genes and specificDNA sequences to target, PCR is not techni-cally feasible. To get around this problem,Eggan et al.7 and Li et al.8 used the new—ifinvolved—strategy of expanding a single ORneuronal genome by mouse cloning9. Indeed,cloning of lymphocytes that have alreadyundergone immunological DNA rearrange-ments produce mice that maintain the origi-nal rearrangements in all tissues, yet alsoyield viable mice, validating this strategy for


Choices, choices, choicesJerold Chun

Individual olfactory sensory neurons express only one of more than a thousand different odorant receptors, suggesting that DNArearrangement may be involved. Based on a clever new technical approach, two groups now conclude that this is not the case.

The author is in the Department of Molecular

Biology at the Helen L. Dorris Institute for

Neurological and Psychiatric Disorders, The

Scripps Research Institute, ICND118, 10550 North

Torrey Pines Road, La Jolla, California 92037, USA.


Figure 1 Models for regulating gene expression in the immune and olfactory systems. (a) Classical‘VDJ’ recombination that occurs in the immune system to generate immunoglobulins. Component genesegments that do not themselves encode mature immunoglobulins are brought together to form acomposite coding region that serves as the antigen recognition portion of an antibody2. In the olfactorysystem, receptor expression controlled by short promoter (P) elements (b) or by distant loci (locuscontrol region, LCR) (c) could provide sufficient information to allow appropriate expression of ORgenes in the presence of appropriate transcription factors (TCF)4. (d) Olfactory receptor expressioncontrolled by DNA rearrangement3, in which a distant segment of DNA with promoter/enhanceractivities is placed, through rearrangements, in proximity to a basal promoter to provide specificexpression4; other variations are also possible.

b

a

c

d

PA PB

PB

PB

PB

ORA ORB

ORB

ORB

ORB

ORA

TFCB

TFCB

TFCB

LCR

ORA

ORA EA

PA

PA

PA

EA

V D

Germ line arrangement

Site-specific DNA cleavage

Non-homologous end joining

J

V D J

V D J

5′ 3′

3′

3′

3′

3′

3′

3′

3′

5′

5′

5′

5′

5′

5′

5′

Kam

al M

asut

a

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


revealing DNA rearrangements10. To assessthe neuronal genome of a single, identifiedOR gene, the two research groups extendedthe cloning approach with elegant variationsthat blended a range of other moleculargenetic techniques including embryonic stem(ES) cells (semi-immortal cells that canreconstitute a mouse), targeted knock-out/knock-ins and lineage tracing.

One way to think about cloned mice is as agenome magnifier, as a single neuronalnucleus produces an entire mouse in whichall non-immune cells are classically expectedto be genomically identical. Whereas cloningof mice was reported several years ago usingnon-neural cumulus cells9, attempts to pro-duce mice from CNS neurons (Fig. 2a) hadproven unsuccessful9,11. Unlike earliercloning reports9, Eggan et al. and Li et al.used a two-step cloning process (Fig. 2b),whereby donor nuclei are first transferredinto enucleated oocytes, which are allowed toform blastocysts from which ES cells arederived. These ES cells are then transferred toa modified recipient embryo (a tetraploidrather than diploid blastocyst) before trans-fer to a host mouse for in utero development;the end result is that cloned mice are actually

produced from the ES cells rather thandirectly from a neuronal nucleus itself.

Both groups first used permanently taggedolfactory sensory neuron nuclei to produce EScells and then cloned mice, allowing identifica-tion of the tag in subsequent steps. The resultsshowed that at least some olfactory neuronalnuclei were competent to produce cloned,apparently normal mice. If restricted expres-sion of a single OR gene in the differentiatedolfactory sensory neuron was permanent, thenthe researchers might have expected to seemice expressing only one OR type, along witholfactory sensory neurons showing a single,stereotyped neuroanatomical projection pat-tern. However, further analyses of these miceindicated normal OR gene expression, withexpression of multiple receptors and normalneuroanatomy, including spatial distributionof olfactory sensory neurons.

However, this first approach could notidentify which single OR subtype was beingused by the neuron from which a mousewas cloned. Thus, the two groups went fur-ther by permanently tagging individualneurons of defined OR identity followed bycloning. As in the first experiment, OR neu-rons in the mice again showed a normal

range of expressed ORs, despite havingoriginated from a neuron expressing anidentified OR subtype.

Finally, the researchers used ES cell DNAderived from a tagged OR neuron to searchby classical means for possible rearrange-ments surrounding that receptor—norearrangements were identified. Therefore,DNA rearrangements are not necessary forOR expression. That the two researchgroups used different ORs also strengthensthis conclusion.

The results of Eggan et al.7 and Li et al.8 refo-cus attention on epigenetic mechanisms of ORexpression. In this active field, novel interac-tions relevant to OR expression are being iden-tified, such as intracellular negative feedbackby ORs themselves, which may account for theone-receptor/one-neuron rule12. Eggan et al.look beyond olfaction per se, by providing fur-ther data on the totipotentiality of a neuron.They cloned mice using direct transfer of aneuron-derived ES cell nucleus into an oocyte(Fig. 2b). The normal-appearing mice thatresulted from these apparently totipotentnuclei suggested that even a postmitotic neu-ronal nucleus could be reprogrammed to pro-duce an entire organism.

As with most scientific studies, somecaveats may be worth considering. AlthoughEggan et al. validly concluded that theircloned mice did not contain DNA rearrange-ments that interfered with development of aviable mouse, it is notable that biologicallyimportant DNA rearrangements of theimmune system are maintained in, and arefully compatible with, normal cloned mice10.Therefore, ‘clonablity’ cannot be equatedwith an absence of DNA rearrangement. Italso remains formally possible that a subset


Figure 2 Different mouse-cloning strategies. Inall approaches, a nucleus from a single neuron isisolated and transferred to a recipient enucleatedegg, which further develops in culture into ablastocyst that, following intrauterineimplantation, could become a mouse. (a) Nucleifrom CNS neurons were unable to generate viablemice9,11. (b) In the new studies7,8, permanentlylabeled nuclei (shown in green) from olfactorysensory neurons were used in a two-step cloningapproach10 in which totipotential ES cells derivedfrom the cloned blastocyst were injected intospecially modified tetraploid blastocysts. Theapproach generated viable cloned mice derivedonly from the transplanted ES cells (versuscontributing to the placenta, which is of distinctembryological origin). To determinetotipotentiality, Eggan et al. also used an ES cellnucleus that itself had been derived from anolfactory neuron. The resulting cloned mouse is aclearer indication of the totipotential state of theneuronally derived ES cell nucleus.

Ofactory sensory neuron

Nuclear transfer

ES cellnucleus

Enucleatedoocyte

Enucleatedoocyte

Blastocyst

CNS neuron

Nuclear transfer

Enucleatedoocyte

Blastocyst

Blastocyst

Tetraploidblastocyst

EScells

Viablemouse

Viablemouse

No mice

a b

Kam

al M

asut

a

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


of ORs might still use a rearrangement mech-anism: although there is no evidence for this,these two studies only analyzed two of morethan 1,000 expressed ORs.

An unresolved issue, which may be technicaland/or biological in nature, is that no cloneshave yet been reported using direct transfer ofa neuronal nucleus into an oocyte (Fig. 2),despite expert attempts to do so with nucleifrom other neuronal populations9,11. Evenwith the use of an ES cell intermediate, theoverall success rate of cloning with neuronalnuclei seems to be ∼ 1%. Neuronal nuclear‘reprogramming’ (might it also include someforms of DNA repair?) seems to require the EScell intermediate step, although precisely whatthis step might do to the clonability of neu-ronal nuclei is currently unclear. The state ofthe remaining 99% of neuronal nuclei thatcannot be cloned remains unknown. It is con-ceivable that DNA rearrangements exist in

some of these neurons, although the nature ofsuch rearrangements remains purely specula-tive and, as noted above, might not be expectedto hamper cloning. By contrast, this 99% mostcertainly contains nuclei with global changesin chromosome number (aneuploidy) thatexist among developing and postmitotic neu-rons11,13–15. Although the function and totalextent of this aneuploidy have yet to be clari-fied, it could in part account for the low per-centage of successful clones. It could alsoaccount for the developmental failuresobserved by Eggan et al. and Li et al., as well asplace limits on the percentage of totipotentialneurons identified by Eggan et al.

That said, none of these considerationsdetracts from these first glimpses into a singleOR neuronal genome, and these impressivetechnical and scientific achievements will nodoubt yield further insights into both olfactionand other neural systems in the near future.

1. Reed, R.R. Cell 116, 329–336 (2004).2. Jung, D. & Alt, F.W. Cell 116, 299–311 (2004).3. Buck, L. & Axel, R. Cell 65, 175–187 (1991).4. Kratz, E., Dugas, J.C. & Ngai, J. Trends Genet. 18,

29–34 (2002).5. Lane, R.P. et al. Proc. Natl. Acad. Sci. USA 98,

7390–7395 (2001).6. Zhang, X. & Firestein, S. Nat. Neurosci. 5, 124–133

(2002).7. Eggan, K. et al. Nature 428, 44–49 (2004).8. Li, J., Ishii, T., Feinstein, P. & Mombaerts, P. Nature

428, 393–399 (2004).9. Wakayama, T., Perry, A.C., Zuccotti, M., Johnson,

K.R. & Yanagimachi, R. Nature 394, 369–374(1998).

10. Hochedlinger, K. & Jaenisch, R. Nature 415,1035–1038 (2002).

11. Osada, T., Kusakabe, H., Akutsu, H., Yagi, T. &Yanagimachi, R. Cytogenet. Genome Res. 97, 7–12(2002).

12. Serizawa, S. et al. Science 302, 2088–2094(2003).

13. Rehen, S.K. et al. Proc. Natl. Acad. Sci. USA 98,13361–13366 (2001).

14. Kaushal, D. et al. J. Neurosci. 23, 5599–5606(2003).

15. Yang, A.H. et al. J. Neurosci. 23, 10454–10462(2003).


“A man falls in love through his eyes, awoman through her ears,” wrote WoodrowWyatt in 1918. In this issue, Hamann and col-leagues1 use functional magnetic resonanceimaging to test whether males and femalesindeed differ in their brain responses to sexu-ally arousing images. The authors findgreater activation in males than females inthe amygdala, a brain region involved inemotional arousal, and in the hypothalamus,a brain region central to reproductive func-tions. What distinguishes this study from aprevious effort2 is that the investigators wentto great lengths to select stimuli and subjectsthat would ensure similar degrees of self-reported arousal in both sexes. Thus, theobserved brain differences are less likely toreflect sex differences in arousal; instead they

reflect sex differences in the processing ofsexually arousing stimuli.

Hamann and colleagues scanned 28healthy, heterosexual volunteers, an equalnumber of males and females. Participantspassively viewed neutral images of couplesinteracting in nonsexual ways (such as wed-dings, dancing or therapeutic massage), nudephotographs of opposite-sex individuals inmodeling poses (opposite-sex stimuli) andphotographs of couples engaged in explicitsexual acts (couples stimuli), as well as a fixa-tion cross condition to establish brain activa-tion at baseline. Participants subsequentlyrated their sexual attraction and physicalarousal in response to each image on a three-point scale. Analysis of the imaging data con-trasted brain activation to the couples stimuliversus activation to neutral or fixation stim-uli, thus revealing regions of significant acti-vation for each sex separately, as well assignificant differences between, and com-monalities across, the sexes (Fig. 1).

Both sexes reported comparable sexualattraction and physical arousal in response tothe images; both groups found the couplesstimuli to be the most attractive and arousing.The most sensitive direct comparison

between males and females looked at the con-trast in brain activation between the couplesand neutral stimuli. Both classes of stimulidepicted couples interacting, differing only inthe sexual aspect of the interaction. In thiscontrast, males showed significantly greateractivation than females in the amygdala. Thisdifferential activation in the amygdala standsin striking contrast to many brain regions thatwere commonly activated for both males andfemales—regions associated with visual pro-cessing, attention, motor and somatosensoryfunction, emotion and reward.

Several additional observations are note-worthy. First, brain activation data remainedunchanged when the one female subject whoreported low sexual arousal was excludedfrom the analysis. Removal of this subjectcaused the average arousal of the females tosignificantly exceed that of the males, yet itwas the males who exhibited greater amyg-dala activation. This is perhaps the strongestindicator that amygdala activation is notrelated to sexual arousal per se.

Second, the average differences betweenthe sexes were striking. Not only did menshow greater activation than women inresponse to sexually explicit couple images in

Imaging gender differences in sexual arousalTurhan Canli & John D E Gabrieli

Men tend to be more interested than women in visual sexually arousing stimuli. Now we learn that when they view identicalstimuli, even when women report greater arousal, the amydala and hypothalamus are much more strongly activated in men.

Turhan Canli is at the Department of Psychology,

SUNY Stony Brook, Stony Brook, New York

11794-2500, USA.


John D.E. Gabrieli is at the Department of

Psychology, Stanford University, Jordan Hall,

Stanford, California 94305, USA.


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


the left amygdala, right amygdala and hypo-thalamus, but also women did not show anygreater activation in these regions for the sex-ually explicit stimuli than for the neutralscenes. It is unclear, therefore, which neuralsystem mediates the sexual arousal reportedby the women in this study.

Third, males, but not females, showed sig-nificant activation in another region associ-ated with sexual behavior, the hypothalamus,when viewing neutral stimuli depicting cou-ples (albeit at a lower level of statistical sig-nificance). The authors speculate that thismay represent the male’s propensity to vieweven neutral interactions with females asvaguely sexual, a point that is unlikely to bemissed by late-night comedians.

Fourth, males and females differed greatlyin their amygdala responses to couples andopposite-sex stimuli. Males showed greateractivation for those stimuli that generated thegreatest arousal: there was no significant acti-vation to the nudes depicted in the opposite-sex set, but highly significant activation to thesexually explicit couples, relative to the neu-tral pictures. Females showed the oppositepattern: they had significantly greater activa-tion to the less arousing opposite-sex stimuli,but no significant activation to the copulatingcouples. The authors speculated that greateramygdala activation in males may representtheir propensity for varied, explicit sexualactivity, but the paper offers no explanationfor women’s amygdala responses. The distinc-tion between males’ and females’ amygdalareactivity appears to map onto that of ‘hard’versus ‘soft’ pornography and is likely to invitecommentary from many different schools ofthought on human sexuality.

The benefit of recruiting males andfemales matched in their ability to experienceand express sexual arousal comes at the costof differential recruitment across the sexes.For example, participants were pre-screened

to respond similarly to sexually explicit mate-rial. Intuition (and general life experience)suggests that this process generated a greateryield for males than females. Indeed, none ofthe males reported lack of arousal to visualerotica, whereas 16% of the female prospectswere excluded because of insufficient self-reported arousal. This suggests that the datareported here may not necessarily generalizeto all women.

Another noticeable sex difference thatemerged during the screening of prospectiveparticipants was related to self-reportedsame-sex desire or experience. Only 12% ofprospective males, but 36% of prospectivefemales, were excluded from the study forthis reason. The basis of this differenceremains unclear.

The only other study to directly comparebrain responses to sexual images betweenmales and females failed to detect any sex dif-ference2. In that study, males reported greatersexual arousal than females, and no signifi-cant activation differences were noted whencontrolling for arousal. Two other studieslooking only at males reported conflictingdata3,4. It is possible that the extent to whichmales show amygdala activation to sexuallyexplicit stimuli varies as a function of otherfactors, such as personality. Indeed, amygdalaactivation to positive stimuli such as pleasantscenes or happy faces varies as a function ofthe personality trait of extraversion5,6.Whether this trait may also predict individ-ual differences in amygdala activation to sex-ual stimuli is unknown.

Asymmetries in left versus right amygdalafunction are of interest, but poorly under-stood at present. Hamann and colleaguesreport greater activation in the left than rightamygdala of males for the explicit images.The only other study to report male amyg-dala activation to sexual stimuli observed itin the right hemisphere3. Consistent with theresults of Hamann et al.1, left amygdala acti-vation has been reported to be a function ofemotional arousal to non-sexual emotionalstimuli7,8, although one of these experimentsinvolved highly negative stimuli8. Studies ofthe encoding of emotional scenes into long-term memory have consistently reported astronger relation between successful encod-

ing and left amygdala activation for femalesversus right amygdala activation for males8,9.Although the specific patterns of lateralityare difficult to synthesize, amygdala activa-tion often seems to depict some sort of sexdifference in the context of emotionallyprovocative visual stimulation.

It is natural to question whether suchbrain activation differences reflect geneticor social influences on the human brain andmind. That is a question, however, thatbrain imaging cannot answer. Men andwomen are, by definition, genetically differ-ent. Men and women are also powerfullysocialized into gender roles, a socializationthat begins shortly after birth. Both geneticand social influences shape brain functionand its consequent behavior, so imaging dif-ferences could arise from either nature ornurture or both.

Hamann et al.1 have reported a thought-fully controlled study of one aspect of humansexuality. Human sexuality, however, wouldremain unfulfilled without a climax.Psychologists have long distinguishedbetween ‘appetitive’ and ‘consummatory’sexual behaviors, that is, those that lead up to,and those that conclude the sexual act. Oneimaging study went right to the point, imag-ing the male brain during ejaculation10. Theauthors of this study were not only intrepidin their choice of research topic and subjectparticipation, but they were also undeterredby concerns about motion artifacts.Remarkably, ejaculation in males was associ-ated with decreased amygdala activation.Thus, the appetitive phase of sexual arousalseems to coincide with increased amygdalaactivation that is then reversed during theconsummatory phase. This activation changeparallels the rise and rapid fall in sexualexcitement from one phase to the other. Itremains to be seen whether decreased amyg-dala activation associated with ejaculation iscausally linked to males’ subsequent unwill-ingness to snuggle.

1. Hamann, S., Herman, R.A., Nolan, C.L. & Wallen, K.Nat. Neurosci. 7, 411–416 (2004).

2. Karama, S. et al. Hum. Brain Mapp. 16, 1–13(2002).

3. Beauregard, M., Levesque, J. & Bourgouin, P. J. Neurosci. 21, RC165 (2001).

4. Redoute, J. et al. Hum. Brain Mapp. 11, 162–177(2000).

5. Canli, T. et al. Behav. Neurosci. 115, 33–42 (2001).6. Canli, T., Sivers, H., Whitfield, S.L., Gotlib, I.H. &

Gabrieli, J.D. Science 296, 2191 (2002).7. Hamann, S.B., Ely, T.D., Hoffman, J.M. & Kilts, C.D.

Psychol. Sci. 13, 135–141 (2002).8. Canli, T., Desmond, J.E., Zhao, Z. & Gabrieli, J.D.E.

Proc. Natl. Acad. Sci. USA 99, 10789–10794(2002).

9. Cahill, L. et al. Neurobiol. Learn. Mem. 75, 1–9(2001).

10. Holstege, G. et al. J. Neurosci. 23, 9185–9193(2003).


Figure 1 Gender differences in sexual arousal.When viewing sexually arousing visual stimuli,men show greater activation in the amygdala(blue), a brain region involved in emotionalarousal, and in the hypothalamus (green), a regioninvolved in reproductive function. Men showedgreater activation in these regions even whenwomen reported equal or greater sexual arousal.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

R E V I E W

The construction of neural circuits is a dynamic process consisting ofboth formation and elimination events1,2. Stereotyped pruningguided by molecular cues contributes to axon targeting3. Axonal anddendritic arborization in target areas similarly consists of both for-mation and elimination of neuronal branches and synapses. How theformation and elimination are coordinated and what role neuralactivity has in this process have been subjects of intensive study.

Synapse elimination is perhaps best characterized at the verte-brate neuromuscular junction (NMJ) and at the cerebellar climbingfiber synapses. Abundant functional and morphological evidencesuggests that in these synapses, postsynaptic targets initially receiveinput from multiple axons. As a result of synapse elimination andaxon retraction, only one of all axons maintains its input on a givenpostsynaptic cell in the adult animal1,4,5. In most parts of the centralnervous system (CNS), the heterogeneity of neural connectionsmakes it difficult to identify and study multiple axon inputs ontothe same target cell. Nevertheless, functional evidence suggests thata common rule in much of CNS development is that postsynapticcells initially receive widespread inputs, and input eliminationsharpens neural response during maturation. For instance, Hubeland Wiesel noted that neurons in the binocular zone of the visualcortex initially respond weakly to visual stimuli applied to eithereye, and the visual response becomes crisper and dominated by oneeye as the visual system matures6. Immature LGN neurons similarlyreceive multiple retinal inputs, only 1–3 of which are maintainedinto adulthood7,8. Developmental refinements of receptive fieldshave also been observed in visual systems of non-mammalianspecies9 and in auditory systems10.

Observations of developmental refinement of functional connec-tions naturally lead to questions regarding their structural basis. Is theelimination of selected synaptic inputs a result of synapse disassemblyor the change of synaptic strength at stable synaptic sites? If non-selective synapses and neuronal branches are produced and then

eliminated during development, then how are the processes of forma-tion and elimination choreographed; specifically, are there separatephases of formation and elimination? To one extreme, a ‘sequential’formation-elimination model may be proposed, where a given neu-ron initially forms excessive and unconstrained branches andsynapses, and then the inappropriate connections are subsequentlyremoved during a phase dominated by elimination. On the otherhand, CNS development may proceed through a ‘concurrent’ mode,where the formation and elimination of branches and synapses occurwithin the same time frame and are nearly balanced (Fig. 1).

Anatomical studies provide strong evidence for elimination ofstructural synapses in CNS. Cajal originally noted that pyramidalneuron spine density is higher during early postnatal developmentthan it is in adulthood11. Later, electron microscopy studies con-firmed that synapse density in the primate cortex decreases duringlate childhood and adolescence12–14. Great efforts to reconstruct allthe synaptic connections formed between one retinal ganglion celland its postsynaptic partners in the adult cat LGN resulted in the esti-mate that a single retinal ganglion cell axon arbor forms synapseswith only four postsynaptic cells15. This estimate indicates that struc-tural synapse elimination must underlie the functional input elimina-tion observed during development7,8.

Morphological reconstructions also provided clues about how for-mation and elimination are coordinated during development.Individual retinogeniculate and geniculocortical arbors recon-structed from brain tissues of young animals have sparse and nonse-lective branches, while axon arbors reconstructed from older animalshave more restricted projection areas and higher branch densi-ties16–18. Thus, the axon arborization process must involve elimina-tion of a limited number of immature connections coupled with theelaboration of remaining ones. Crowley and colleagues recentlyreported that geniculate fibers innervate proper visual cortical targetsbefore local branch elaboration19, raising concerns that elimination

Neural activity and the dynamics of centralnervous system developmentJackie Yuanyuan Hua1,2 & Stephen J Smith1

Recent imaging studies show that the formation of neural connections in the central nervous system is a highly dynamic process.The iterative formation and elimination of synapses and neuronal branches result in the formation of a much larger number oftrial connections than is maintained in the mature brain. Neural activity modulates development through biasing this process offormation and elimination, promoting the formation and stabilization of appropriate synaptic connections on the basis offunctional activity patterns.

1Department of Molecular and Cellular Physiology, Beckman Center B141, Stanford University, Stanford, California 94305, USA. 2Neurosciences Program, StanfordUniversity, Stanford, California 94305, USA. Correspondence should be addressed to S.J.S. ([email protected]).

Published online 26 March 2004; doi:10.1038/nn1218


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

R E V I E W

events may not contribute to the development of geniculocorticalconnections. Although the early geniculate afferents are mostly con-fined to columns in the Crowley et al. study, sparse afferents extendoutside these columns, in keeping with earlier observations of uncon-strained axon branch formation by immature geniculate fibers16. Thefunctional and anatomical data taken together support a concurrentmodel of neural development: synaptic connections in immatureCNS are widespread but sparse, and the elimination of inappropriateconnections is coupled to the ramification of appropriate connec-tions to refine the circuit to functional levels as seen in adults. Theconcurrence of formation and elimination ensures that connectionsdestined for elimination are promptly removed without further elab-oration, so that the refinement of neural circuits proceeds efficiently.

Given the concurrence of formation and elimination processesduring development, it is necessary to ask a quantitative question:How many times more synaptic connections are made and eliminatedduring development relative to those maintained in the adult neuralcircuit? In a system undergoing both active formation and elimina-tion, the number of trial branches and synapses formed and theneliminated within a given time period may far exceed those main-tained at the end of that period. Therefore, static anatomicalapproaches are inadequate to address this question; time-lapse imag-ing is needed. Time-lapse imaging studies increasingly suggest thatCNS development is indeed a highly dynamic process of rapid andnearly balanced formation and elimination.

Developing neurons undergo a rapid turnover of axon and den-drite branches (Fig. 2a,b). For instance, retinal ganglion cell axons inliving Xenopus laevis larvae add 150% and retract 137% of the initialbranch tip number in 2 h. Their postsynaptic partners, the tectalneuron dendrites, add 180% and retract 124% of the initial branchtip number during the same period20,21. The higher rate of branchaddition leads to a gradual increase in arbor complexity duringdevelopment. Rapid extension and retraction of branch tips havealso been observed during the development of the zebrafish retino-tectal system22,23. In addition, many studies have described age-

related, activity-dependent motility of neuronal filopodia24–29,which are fine, motile branch protrusions enriched in the immaturenervous system. Dendritic filopodia have average lifetimes on theorder of minutes24–26. Widely thought to be potential synapse pre-cursors30, motile filopodia may have a crucial role in regulating therate of synapse turnover during development.

Time-lapse imaging studies also suggest that synapses remodel rap-idly in developing neural circuits (Fig. 2c,d). Some of these studiescharacterize the dynamics of dendritic spines as proxies for synapsesand report extensive remodeling of dendritic spines in both develop-ing and adult nervous systems31,32. However, not all synapses form onspines, and not every spine bears a synapse32,33. Thus, spines are notperfect proxies for synapses. A more direct approach to visualize thedynamics of synaptic organization is to use synaptic proteins taggedwith genetically encoded fluorescent indicators as ‘synaptic markers’.Postsynaptic densities marked by green fluorescent protein (GFP)-tagged PSD-95, a postsynaptic density protein, appear, disappear andmove along the dendritic shafts within minutes of imaging in neona-tal brain slices34. In cultured hippocampal neurons, more than 20%of the postsynaptic densities turn over in a 24-h period35. Theseobservations likely underestimate the synaptic dynamics in vivo. Forcomparison, 37% of the presynaptic accumulations marked by GFP-tagged VAMP/synaptobrevin form, whereas 27% of the existingVAMP-GFP accumulations disperse, during 2 h of imaging in retinalganglion cells of X. laevis tadpoles, resulting in a moderate accumula-tion of new synapses36.

Genetically encoded synaptic markers have their own limitations.Whereas fluorescently tagged synaptic proteins are enriched insynapses, aggregations of these proteins are not guaranteed to pre-cisely represent synapses. For example, aggregates of VAMP-GFP mayalso represent transport packets37,38. A more generic problem is thatsynaptic proteins are thought to be sequentially recruited to nascentsynaptic contacts39. Thus, the presence of no single protein representsthe presence of a synapse. Rather, the aggregation of synaptic adhe-sion proteins may mark the formation of nascent synaptic contacts,whereas recruitment of different neurotransmitter receptor typesmay mark different stages of synapse maturation and stabilization.For this reason, imaging studies of multiple synaptic proteins will beneeded to fully understand synaptic dynamics.

Such caveats notwithstanding, time-lapse imaging studies haveprovided strong evidence that the construction of CNS neural cir-cuits proceeds through concurrent and nearly balanced branchgrowth and retraction and synapse formation and elimination. Nogross temporal division of phases dominated by formation or elimi-nation as proposed by the sequential model of development hasbeen observed. Furthermore, the contrast between neuronal branchand synapse remodeling on the time scales of hours or even minutesand the functional connection refinement over days or weeksimplies that the refinement of CNS neural circuits is a ‘trial-and-error’ process consisting of rapid, iterative sampling of a large num-ber of tentative synaptic contacts. Only a fraction of the trialbranches and synapses are maintained, resulting in relatively slowaccumulation of stable synaptic connections and refinement of thefunctional connectivity. Determining the quantitative differencebetween the numbers of synaptic connections formed during devel-opment and those maintained into adulthood remains an impor-tant task for future experiments.

Activity dependence of neural dynamics during developmentPioneered by the classic work of Hubel and Wiesel on mammalianvisual cortex development, abundant evidence supports that neural


Figure 1 Two models of neural development. (a) A sequential formation-elimination model: neural development proceeds through first a phase ofunconstrained and excessive formation of branches and synapses, and thena phase dominated by branch and synapse elimination. (b) A concurrentformation-elimination model: the formation and elimination of synapticconnections both proceed actively during development. As a result, thenumber of transient branches and synapses formed and eliminated withinany given time period exceeds the number of those stabilized at the end ofthat period.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

R E V I E W

activity modulates CNS development40–44. In a system where forma-tion and elimination are concurrent, activity may modulate develop-ment by regulating either the rate at which new synaptic connectionsform or the stability of existing connections. There is evidence sup-porting both types of modulation.

Before discussing such evidence in detail, we should emphasizethat it is no trivial matter to determine whether an activity effect onneuronal dynamics reflects an underlying action on the formation ofnew synaptic connections or an action on the stabilization of existingconnections. Time-lapse imaging experiments cannot distinguishthese two possibilities if the length of image-sampling intervalsexceeds the lifetimes of transient branches and synapses. However,high sampling frequency leads to higher risks in phototoxicity andphotobleaching, and encumbers image analysis. As a result, under-sampling is a common practice in the field. For example, more thanhalf of the retinal ganglion cell axon branch tips in X. laevis larvaehave a lifetime of less than an hour21, whereas most past work on thissystem used sampling intervals on the scales of hours. The use oftwo-photon or spinning-disk confocal microscopy to reduce photo-damage, and improvement in algorithms for automatic morphologi-cal tracing will likely improve our ability to carry out high-frequencytime-lapse imaging, and lead to a deeper understanding of the activity-dependence of neural dynamics.

Neural activity modulates synapse stabilizationAlthough synaptogenesis occurs in the absence of neural activ-ity45,46, it can also be modulated by neural activity32,47.Neurotransmission, especially NMDA receptor activation by coin-cident pre- and postsynaptic activity, can promote synapse matura-tion and stabilization. First, NMDA receptor activation regulatesthe organization of the postsynaptic cytoskeleton. ModerateNMDA receptor activation slows the turnover of actin filamentswithin the spine’s actin pool48 and increases the actin filament con-tent in spines49. Such regulation of the actin cytoskeleton isthought to underlie the changes in spine morphology in responseto NMDA receptor activation50. Second, NMDA receptor activa-tion recruits AMPA receptors to the postsynaptic membrane51.AMPA receptor activation in mature cultured hippocampal neu-rons reduces spine motility50 and maintains dendritic spines52.AMPA receptor overexpression has also been observed to increasesspine size and density independent of receptor activation53. Third,NMDA receptor activation may regulate the expression of struc-tural proteins to promote synapse maturation54. Homer, an imme-diate early gene that encodes a synaptic scaffolding protein,promotes the growth and enlargement of mushroom spines whencotransfected with its binding partner Shank in hippocampal cul-ture55. It is possible that similar mechanisms modify the composi-tion of adhesion proteins at the synapse, resulting in strongersynaptic adhesion and higher synapse stability as a result of appro-priate synaptic inputs. Finally, NMDA receptor activation inducesthe synthesis and release of neurotrophins, particularly brain-derived neurotrophic factor (BDNF)56,57. BDNF treatment pro-motes the maturation of presynaptic terminals in dissociatedneuronal culture58,59 and increases the number of synapses peraxon terminal in vivo36. Thus neurotrophins could serve as media-tors for activity-dependent synapse stabilization.

Contrary to the stabilizing effects of activity discussed above, Luthiand colleagues observed an increase in spine density in cultured hip-pocampal neurons induced by long-term NMDA receptor blockage60.How can these findings be reconciled? The long-term pharmacologi-cal blockage in the Luthi et al. study may activate homeostatic mecha-

nisms, which differ from the mechanisms discussed in the previousparagraph based on coincident activity detection. The effects of activ-ity blockage on development have been shown in many cases to bedifferent depending on whether relative activity level or the absoluteamount of activity is changed, and whether the change is acute orchronic61–64. To study coincident activity–dependent mechanisms ofdevelopment, other experimental protocols may be more preferablethan global activity blockage, such as selective activity suppres-sion64,65 or patterned synaptic stimuli66,67.


Figure 2 Time-lapse imaging experiments suggest that neural developmentis a highly dynamic process of concurrent formation and elimination.(a) Development of a retinal ganglion cell axonal arbor in X. laevis larvae.Top panel, projected confocal images. Lower panel, reconstructed axonarbors. Formation of new branches (red in lower panel) and elimination ofexisting branches (green in lower panel) occur concurrently. (b) Develop-ment of a tectal cell denritic arbor in X. laevis larvae reconstructed fromconfocal images. The dendritic arbor remodels extensively during 30-minimaging intervals. Branch tip formation (red) and elimination (green) bothoccur. Arrowhead indicates the axon. (c) Presynaptic remodeling of a X.laevis retinal ganglion cell axon in vivo. Red: DsRed labeled axon. Yellow:VAMP-GFP puncta. Synapse formation (arrows) and elimination(arrowheads) are both observed. (d) Remodeling of the postsynaptic densityin a dendrite segment from a cultured hippocampal neuron expressingPSD-95:GFP. Left panels, images of the dendrite branch taken before andafter a 24-h interval. Middle panels, binary images showing PSD-95:GFPclusters. Right panel, superimposition of the binary images showselimination (green) and formation (red) of synaptic clusters. Panel a isreproduced, with permission, from ref. 80; panel b is reproduced, withpermission, from ref. 20; panel c is reproduced, with permission, from ref. 36; panel d is reproduced, with permission, from ref. 35.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

R E V I E W

Neural activity modulates synapse formationBecause of the lack of high temporal resolution imaging studies, therehas been no solid evidence showing that activity directly regulates therates of synapse formation as opposed to synapse stability. However,activity is widely thought to regulate synapse formation through theregulation of local filopodial dynamics30. Not only axonal, but alsodendritic filopodia may initiate synaptic contacts24,33,68; therefore,activity-dependent increases in filopodial dynamics may effectivelyincrease the frequency of synapse formation in developing neurons.

Regulation of filopodial dynamics by neurotransmission has beenwidely observed in slice explant preparations. Kainate receptor activa-tion differentially modulates the motility of axonal filopodia onmossy fibers depending on the developmental stage27. NMDA recep-tor activation induces the formation of new dendritic protrusions inhippocampal slices69. The growth and maintenance of dendriticfilopodia on retinal ganglion cells is regulated by acetylcholine duringthe peak of cholinergic synapse formation, and subsequently by glu-tamate during the peak of glutamatergic synapse formation28. Thusneurotransmission mediated by established synapses may promotethe growth and maintenance of nearby filopodia and facilitate furthersynapse formation where functional synaptic contacts exist.

Lendvai et al.70 examined the significance of activity-dependentfilopodial dynamics in developing rat somatosensory cortex in vivo.Sensory deprivation induced by whisker trimming was found to causea concomitant reduction of cortical dendritic protrusions and degra-dation of the sensory receptive field. Interestingly, sensory depriva-tion did not affect the density or the morphology of spines, suggestingthat sensory input does not regulate synapse density per se, but ratherbiases the sites of synaptogenesis to favor synapse formation whereappropriate synaptic inputs have been received.

Neural activity modulates branch stabilityThe precise patterning of neuronal arbors contributes to the precisepatterning of synaptic connections. Activity is known to regulate neu-ronal arbor growth through regulating branch stability. In the retino-tectal systems of X. laevis and zebrafish larvae, glutamate receptor

blockade markedly decreases the branch lifetime of retinal ganglioncell axons and tectal dendrites20,21,71.

Neural activity can modulate the organization of the localcytoskeleton to stabilize neuronal branches. Lohmann et al. recordedlocal, spontaneous rises in calcium in retina explants72. These risesdepend on neurotransmission and calcium-induced calcium release(CICR) from intracellular stores in retinal ganglion cell dendrites,and they serve to maintain the retinal ganglion cell dendrites, as per-turbing the spontaneous local calcium activity by pharmacologicalblockage of the CICR signaling pathway leads to dendrite retractionwithin a few minutes. Furthermore, dendrite retraction can be pre-vented by focal calcium uncaging, which raises intracellular calciumlocally. Because the stabilizing effect of calcium observed byLohmann et al. is rapid and remains within the dendritic segmentswhere the calcium level is elevated, it likely results from local signalingthat modifies the branch cytoskeleton.

One molecular pathway by which neural activity may modify thelocal cytoskeleton is through Rho GTPase signaling. The RhoGTPases are key regulators of the actin cytoskeleton73. In X. laevistadpoles, optical nerve stimulation catalyzes the switch of RhoGTPases between active and inactive forms within seconds and pro-mote the growth of tectal cell dendrites74.

Other evidence suggests that activity may govern branch stabilityby regulating synapse turnover. A synapse may stabilize the branchbearing it through either providing physical adhesion or activatingsignalling pathways such as calcium influx-induced cytoskeletal mod-ulation. At the vertebrate NMJ, the activity-dependent differentialstabilization of neuromuscular synapses is a key determinant of theeventual motor axon branching pattern1. It may be proposed by anal-ogy that activity-dependent synapse stabilization in the CNS directsaxonal and dendritic arbor growth. In support of this idea, Vaughnand colleagues observed predominate dendritic growth of motorneurons in regions of the developing mouse spinal cord where affer-ents are enriched and synaptic density is high75. In addition, they wereable to identify synapses on dendritic filopodia and growth cones.Based on such histological evidence, Vaughn formulated the ‘synap-totropic hypothesis’ of dendrite growth, stating that the maturationand stabilization of a nascent synapse on a given nascent dendritebranch might stabilize that branch, and thus direct arbor growth68

(Fig. 3). According to this hypothesis, new synapses form predomi-nantly on dendritic filopodia. Most filopodia and nascent synapsesare promptly eliminated, but a small fraction of nascent synapsesreceiving appropriate synaptic input persists, leading to the preserva-tion of the corresponding filopodia and their conversion into stabledendrite branch segments.

Vaughn further proposed that the synaptotropic hypothesis mightalso apply to axon development. The relationship between synaptoge-nesis and axon development was recently examined by Alsina et al.36

by time-lapse imaging of X. laevis retinal ganglion cells expressingVAMP-GFP, a presynaptic marker, and the red fluorescent proteinDsRed. The authors observed marked synapse and axon branchremodeling over 24-h recording periods. In comparison to stablebranches, branches that were eventually eliminated had, on average,fewer synapses per axon terminal36, suggesting that synapse forma-tion and stabilization leads to the stabilization of axon branches.Thus, regulation of synaptogenesis could be a key link between neu-rotransmission and the patterning of neuronal arbors.

Finally, activity-dependent gene expression potently modulatesneuronal arbor growth43, sometimes through stabilizing branches.The expression and enzymatic activity of calcium/calmodulin-dependent protein kinase II (CaMKII) is regulated by activity76. In an


Figure 3 Schematic diagram of the growth of a dendrite branch illustratingthe synaptotropic hypothesis. The dendrite branch (blue) extends filopodiato form synapses (green dot) with passing axons (brown). Activity-dependent synapse stabilization (upper-right panels) leads to thestabilization of the corresponding filopodium and the formation of a newbranch; the new branch then grows through successive rounds of selectivefilopodial stabilization. Synapse elimination (lower-right panel) results inretraction of the corresponding filopodium.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

R E V I E W

elegant series of in vivo imaging experiments, Wu and colleaguesshowed that CaMKII expression is necessary and sufficient for thetransition of tectal cell dendrites from the immature, dynamic stateinto the mature, stabilized state77. As another example, overexpres-sion of the activity-regulated candidate plasticity gene 15 (cpg-15) inX. laevis tectal cells concomitantly stabilizes tectal cell dendrites andthe retinal ganglion cell axons that contact them78,79. WhetherCAMKII and CPG-15 regulate branch stability directly or indirectlythrough regulating synapse turnover remians a subject for furtherinvestigation. Another intriguing theme calling for further investiga-tion is the significance of activity-dependent local protein synthesis inregulating arbor growth80.

Could neural activity destabilize branches to cause selectivebranch retraction? Segregation of retinal ganglion cell axons fromopposite eyes in doubly innervated X. laevis tectum is mainlyaccounted for by the preferential retraction of axon branches fromterritories dominated by the other eye, and NMDA receptor block-ade reduces branch retraction81, suggesting that NMDA receptoractivation selectively destabilizes axon branches. Furthermore, thepresence of a ‘branch destabilization signal’ and the lack of this signalin neural circuits under impulse activity blockage could underlie theparadoxical axon sprouting induced by tetrodotoxin treatment82.Nitric oxide is released in an activity-dependent fashion83,84 andcauses the collapse of growth cones in vivo85, making it an attractivecandidate for such a signal.

Neural activity modulates neuronal branch formationDefinitive evidence for activity-dependent regulation of the rate ofneuronal branch formation is absent. But such a role of activity issupported by the stimulatory effects neurotrophins have on branchformation. Nerve growth factor (NGF)-soaked beads induce de novobranch formation from cultured sensory neurons86. Neurotrophininfusion elicits dendritic form changes in cortical slices87 and pro-motes RGC axon growth in vivo57. Cortical neurons overexpressingBDNF sprout massive numbers of basal dendrites and induce den-drite sprouting from their close neighbors88,89. The newly formeddendrite branches are highly unstable, suggesting that BDNF pro-motes dendritic arbor growth by accelerating branch dynamics ratherthan by promoting branch stability. Thus activity-dependent neu-rotrophin signaling can coordinate the exploratory behavior of axonand dendrite branches to favor the formation of new branches whereappropriate synaptic contacts have been established.

CONCLUSIONRecent advances in biological fluorescence imaging have providedunprecedented opportunities to observe CNS development in realtime. One striking feature of CNS development evident from watch-ing neurons develop is that the formation of CNS neural circuits is ahighly dynamic process of rapid and concurrent formation and elim-ination. In immature neural circuits, branches extend and retract;synapses make and break. Only a limited fraction of new connectionsare maintained in the mature neural circuitry. Neural activity maymodulate the formation, as well as the stability, of synapses and neu-ronal branches to regulate the continual remodeling of synaptic con-nections in immature neural circuits. Understanding the relationshipbetween synaptogenesis and arbor growth is likely to be a key step infurthering our knowledge of the mechanisms of activity-dependentneural development.

ACKNOWLEDGMENTSWe thank J.D. Jontes, L.C. Katz, E.I. Knudsen, L. Luo and J.T. Schmidt fordiscussions, members of the Smith lab for critical reading of the manuscript, and

the US National Institutes of Health and the Vincent Coates Foundation forfinancial support. Y.H. was supported by a Stanford Graduate Fellowship and aCoates Foundation Fellowship.

COMPETING INTERESTS STATEMENTThe authors declare that they have no competing financial interests.

Received 29 August 2003; accepted 21 January 2004Published online at http://www.nature.com/natureneuroscience/

1. Lichtman, J.W. & Colman, H. Synapse elimination and indelible memory. Neuron25, 269–278 (2000).

2. Goda, Y. & Davis, G.W. Mechanisms of synapse assembly and disassembly. Neuron40, 243–264 (2003).

3. Kantor, D.B. & Kolodkin, A.L. Curbing the excesses of youth: molecular insights intoaxonal pruning. Neuron 38, 849–852 (2003).

4. Crepel, F. Maturation of climbing fiber responses in the rat. Brain Res. 35, 272–276(1971).

5. Hashimoto, K. & Kano, M. Functional differentiation of multiple climbing fiberinputs during synapse elimination in the developing cerebellum. Neuron 38,785–796 (2003).

6. Hubel, D.H. & Wiesel, T.N. J. Neurophysiol. 26, 994 (1963).7. Chen, C. & Regehr, W.G. Developmental remodeling of the retinogeniculate synapse.

Neuron 28, 955–966 (2000).8. Tavazoie, S.F. & Reid, R.C. Diverse receptive fields in the lateral geniculate nucleus

during thalamocortical development. Nat. Neurosci. 3, 608–616 (2000).9. Gaze, R.M., Keating, M.J. & Chung, S.H. The evolution of the retinotectal map dur-

ing development in Xenopus. Proc. R. Soc. Lond. B Biol. Sci. 185, 301–330(1974).

10. Jackson, H. & Parks, T.N. Functional synapse elimination in the developing aviancochlear nucleus with simultaneous reduction in cochlear nerve axon branching. J. Neurosci. 2, 1736–1743 (1982).

11. Cajal, R.y. La Textura del Sistema Nerviosa del Hombre y los Vertebrados(Moya,–Madrid, 1899).

12. Huttenlocher, P.R. Synaptic density in human frontal cortex - developmentalchanges and effects of aging. Brain Res. 163, 195–205 (1979).

13. Huttenlocher, P.R., de Courten, C., Garey, L.J. & Van der Loos, H. Synaptogenesis inhuman visual cortex—evidence for synapse elimination during normal development.Neurosci. Lett. 33, 247–252 (1982).

14. Rakic, P., Bourgeois, J.P., Eckenhoff, M.F., Zecevic, N. & Goldman-Rakic, P.S.Concurrent overproduction of synapses in diverse regions of the primate cerebralcortex. Science 232, 232–235 (1986).

15. Hamos, J.E., Van Horn, S.C., Raczkowski, D. & Sherman, S.M. Synaptic circuitsinvolving an individual retinogeniculate axon in the cat. J. Comp. Neurol. 259,165–192 (1987).

16. Antonini, A. & Stryker, M.P. Rapid remodeling of axonal arbors in the visual cortex.Science 260, 1819–1821 (1993).

17. Sretavan, D.W. & Shatz, C.J. Prenatal development of retinal ganglion cell axons:segregation into eye-specific layers within the cat’s lateral geniculate nucleus. J. Neurosci. 6, 234–251 (1986).

18. Campbell, G. & Shatz, C.J. Synapses formed by identified retinogeniculate axonsduring the segregation of eye input. J. Neurosci. 12, 1847–1858 (1992).

19. Crowley, J.C. & Katz, L.C. Early development of ocular dominance columns. Science290, 1321–1324 (2000).

20. Rajan, I. & Cline, H.T. Glutamate receptor activity is required for normal develop-ment of tectal cell dendrites in vivo. J. Neurosci. 18, 7836–7846 (1998).

21. Rajan, I., Witte, S. & Cline, H.T. NMDA receptor activity stabilizes presynapticretinotectal axons and postsynaptic optic tectal cell dendrites in vivo. J. Neurobiol.38, 357–368 (1999).

22. Kaethner, R.J. & Stuermer, C.A. Dynamics of terminal arbor formation and targetapproach of retinotectal axons in living zebrafish embryos: a time-lapse study of sin-gle axons. J. Neurosci. 12, 3257–3271 (1992).

23. Kaethner, R.J. & Stuermer, C.A. Dynamics of process formation during differenti-ation of tectal neurons in embryonic zebrafish. J. Neurobiol. 32, 627–639(1997).

24. Ziv, N.E. & Smith, S.J. Evidence for a role of dendritic filopodia in synaptogenesisand spine formation. Neuron 17, 91–102 (1996).

25. Dailey, M.E. & Smith, S.J. The dynamics of dendritic structure in developing hip-pocampal slices. J. Neurosci. 16, 2983–2994 (1996).

26. Jontes, J.D., Buchanan, J. & Smith, S.J. Growth cone and dendrite dynamics inzebrafish embryos: early events in synaptogenesis imaged in vivo. Nat. Neurosci. 3,231–237 (2000).

27. Tashiro, A., Dunaevsky, A., Blazeski, R., Mason, C.A. & Yuste, R. Bidirectional regu-lation of hippocampal mossy fiber filopodial motility by kainate receptors: a two-stepmodel of synaptogenesis. Neuron 38, 773–784 (2003).

28. Wong, W.T. & Wong, R.O. Changing specificity of neurotransmitter regulation ofrapid dendritic remodeling during synaptogenesis. Nat. Neurosci. 4, 351–352(2001).

29. Chang, S. & De Camilli, P. Glutamate regulates actin-based motility in axonal filopo-dia. Nat. Neurosci. 4, 787–793 (2001).

30. Jontes, J.D. & Smith, S.J. Filopodia, spines, and the generation of synaptic diversity.Neuron 27, 11–14 (2000).


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

R E V I E W

31. Bonhoeffer, T. & Yuste, R. Spine motility. Phenomenology, mechanisms, and func-tion. Neuron 35, 1019–1027 (2002).

32. Trachtenberg, J.T. et al. Long-term in vivo imaging of experience-dependent synapticplasticity in adult cortex. Nature 420, 788–794 (2002).

33. Fiala, J.C., Feinberg, M., Popov, V. & Harris, K.M. Synaptogenesis via dendriticfilopodia in developing hippocampal area CA1. J. Neurosci. 18, 8900–8911(1998).

34. Marrs, G.S., Green, S.H. & Dailey, M.E. Rapid formation and remodeling of postsy-naptic densities in developing dendrites. Nat. Neurosci. 4, 1006–1013 (2001).

35. Okabe, S., Kim, H.D., Miwa, A., Kuriu, T. & Okado, H. Continual remodeling of post-synaptic density and its regulation by synaptic activity. Nat. Neurosci. 2, 804–811(1999).

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

37. Ahmari, S.E., Buchanan, J. & Smith, S.J. Assembly of presynaptic active zones fromcytoplasmic transport packets. Nat. Neurosci. 3, 445–451 (2000).

38. Ahmari, S.E. & Smith, S.J. Knowing a nascent synapse when you see it. Neuron 34,333–336 (2002).

39. Ziv, N.E. & Garner, C.C. Principles of glutamatergic synapse formation: seeing theforest for the trees. Curr. Opin. Neurobiol. 11, 536–543 (2001).

40. Zhang, L.I. & Poo, M.M. Electrical activity and development of neural circuits. Nat.Neurosci. 4 (Suppl.), 1207–1214 (2001).

41. Weisel, T.N. Nature 299, 583 (1982).42. Katz, L.C. & Shatz, C.J. Synaptic activity and the construction of cortical circuits.

Science 274, 1133–1138 (1996).43. Wong, R.O. & Ghosh, A. Activity-dependent regulation of dendritic growth and pat-

terning. Nat. Rev. Neurosci. 3, 803–812 (2002).44. Cline, H.T. Dendritic arbor development and synaptogenesis. Curr. Opin. Neurobiol.

11, 118–126 (2001).45. Augustin, I., Rosenmund, C., Sudhof, T.C. & Brose, N. Munc13-1 is essential for

fusion competence of glutamatergic synaptic vesicles. Nature 400, 457–461(1999).

46. Verhage, M. et al. Synaptic assembly of the brain in the absence of neurotransmittersecretion. Science 287, 864–869 (2000).

47. De Paola, V., Arber, S. & Caroni, P. AMPA receptors regulate dynamic equilibrium ofpresynaptic terminals in mature hippocampal networks. Nat. Neurosci. 6, 491–500(2003).

48. Star, E.N., Kwiatkowski, D.J. & Murthy, V.N. Rapid turnover of actin in dendriticspines and its regulation by activity. Nat. Neurosci. 5, 239–246 (2002).

49. Fukazawa, Y. et al. Hippocampal LTP is accompanied by enhanced F-actin contentwithin the dendritic spine that is essential for late LTP maintenance in vivo. Neuron38, 447–460 (2003).

50. Fischer, M., Kaech, S., Wagner, U., Brinkhaus, H. & Matus, A. Glutamate receptorsregulate actin-based plasticity in dendritic spines. Nat. Neurosci. 3, 887–894(2000).

51. Malinow, R. & Malenka, R.C. AMPA receptor trafficking and synaptic plasticity.Annu. Rev. Neurosci. 25, 103–126 (2002).

52. McKinney, R.A., Capogna, M., Durr, R., Gahwiler, B.H. & Thompson, S.M. Miniaturesynaptic events maintain dendritic spines via AMPA receptor activation. Nat.Neurosci. 2, 44–49 (1999).

53. Passafaro, M., Nakagawa, T., Sala, C. & Sheng, M. Induction of dendritic spines byan extracellular domain of AMPA receptor subunit GluR2. Nature 424, 677–681(2003).

54. Steward, O. & Schuman, E.M. Protein synthesis at synaptic sites on dendrites.Annu. Rev. Neurosci. 24, 299–325 (2001).

55. Sala, C. et al. Regulation of dendritic spine morphology and synaptic function byShank and Homer. Neuron 31, 115–130 (2001).

56. Poo, M.M. Neurotrophins as synaptic modulators. Nat. Rev. Neurosci. 2, 24–32(2001).

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

58. Collin, C. et al. Neurotrophins act at presynaptic terminals to activate synapsesamong cultured hippocampal neurons. Eur. J. Neurosci. 13, 1273–1282 (2001).

59. Tyler, W.J. & Pozzo-Miller, L.D. BDNF enhances quantal neurotransmitter releaseand increases the number of docked vesicles at the active zones of hippocampalexcitatory synapses. J. Neurosci. 21, 4249–4258 (2001).

60. Luthi, A., Schwyzer, L., Mateos, J.M., Gahwiler, B.H. & McKinney, R.A. NMDA

receptor activation limits the number of synaptic connections during hippocampaldevelopment. Nat. Neurosci. 4, 1102–1107 (2001).

61. Balice-Gordon, R.J. & Lichtman, J.W. Long-term synapse loss induced by focalblockade of postsynaptic receptors. Nature 372, 519–524 (1994).

62. Zhu, J.J. & Malinow, R. Acute versus chronic NMDA receptor blockade and synapticAMPA receptor delivery. Nat. Neurosci. 5, 513–514 (2002).

63. Zhao, H. & Reed, R.R. X inactivation of the OCNC1 channel gene reveals a role foractivity-dependent competition in the olfactory system. Cell 104, 651–660 (2001).

64. Burrone, J., O’Byrne, M. & Murthy, V.N. Multiple forms of synaptic plasticity trig-gered by selective suppression of activity in individual neurons. Nature 420,414–418 (2002).

65. Johns, D.C., Marx, R., Mains, R.E., O’Rourke, B. & Marban, E. Inducible geneticsuppression of neuronal excitability. J. Neurosci. 19, 1691–1697 (1999).

66. Colicos, M.A., Collins, B.E., Sailor, M.J. & Goda, Y. Remodeling of synaptic actininduced by photoconductive stimulation. Cell 107, 605–616 (2001).

67. Engert, F., Tao, H.W., Zhang, L.I. & Poo, M.M. Moving visual stimuli rapidly inducedirection sensitivity of developing tectal neurons. Nature 419, 470–475 (2002).

68. Vaughn, J.E. Fine structure of synaptogenesis in the vertebrate central nervous sys-tem. Synapse 3, 255–285 (1989).

69. Maletic-Savatic, M., Malinow, R. & Svoboda, K. Rapid dendritic morphogenesis inCA1 hippocampal dendrites induced by synaptic activity. Science 283, 1923–1927(1999).

70. Lendvai, B., Stern, E.A., Chen, B. & Svoboda, K. Experience-dependent plasticity ofdendritic spines in the developing rat barrel cortex in vivo. Nature 404, 876–881(2000).

71. Schmidt, J.T., Fleming, M.R. & Leu, B. J. Neurobiol. 58, 328 (2004). 72. Lohmann, C., Myhr, K.L. & Wong, R.O. Transmitter-evoked local calcium release sta-

bilizes developing dendrites. Nature 418, 177–181 (2002).73. Hall, A. Rho GTPases and the actin cytoskeleton. Science 279, 509–514 (1998).74. Li, Z., Aizenman, C.D. & Cline, H.T. Regulation of rho GTPases by crosstalk and neu-

ronal activity in vivo. Neuron 33, 741–750 (2002).75. Vaughn, J.E., Barber, R.P. & Sims, T.J. Dendritic development and preferential

growth into synaptogenic fields: a quantitative study of Golgi-impregnated spinalmotor neurons. Synapse 2, 69–78 (1988).

76. Lisman, J., Schulman, H. & Cline, H. The molecular basis of CaMKII function insynaptic and behavioural memory. Nat. Rev. Neurosci. 3, 175–190 (2002).

77. Wu, G.Y. & Cline, H.T. Stabilization of dendritic arbor structure in vivo by CaMKII.Science 279, 222–226 (1998).

78. Cantallops, I., Haas, K. & Cline, H.T. Postsynaptic CPG15 promotes synaptic matu-ration and presynaptic axon arbor elaboration in vivo. Nat. Neurosci. 3, 1004–1011(2000).

79. Nedivi, E., Wu, G.Y. & Cline, H.T. Promotion of dendritic growth by CPG15, an activ-ity-induced signaling molecule. Science 281, 1863–1866 (1998).

80. Ouyang, Y., Rosenstein, A., Kreiman, G., Schuman, E.M. & Kennedy, M.B. Tetanicstimulation leads to increased accumulation of Ca2+/calmodulin-dependent proteinkinase II via dendritic protein synthesis in hippocampal neurons. J. Neurosci. 19,7823–7833 (1999).

81. Ruthazer, E.S., Akerman, C.J. & Cline, H.T. Control of axon branch dynamics by cor-related activity in vivo. Science 301, 66–70 (2003).

82. Chapman, B. Necessity for afferent activity to maintain eye-specific segregation inferret lateral geniculate nucleus. Science 287, 2479–2482 (2000).

83. Garthwaite, J., Charles, S.L. & Chess-Williams, R. Endothelium-derived relaxing fac-tor release on activation of NMDA receptors suggests role as intercellular messengerin the brain. Nature 336, 385–388 (1988).

84. Brenman, J.E. & Bredt, D.S. Synaptic signaling by nitric oxide. Curr. Opin.Neurobiol. 7, 374–378 (1997).

85. Renteria, R.C. & Constantine-Paton, M. Exogenous nitric oxide causes collapse ofretinal ganglion cell axonal growth cones in vitro. J. Neurobiol. 29, 415–428(1996).

86. Gallo, G., Lefcort, F.B. & Letourneau, P.C. The trkA receptor mediates growth coneturning toward a localized source of nerve growth factor. J. Neurosci. 17,5445–5454 (1997).

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

88. Horch, H.W. & Katz, L.C. BDNF release from single cells elicits local dendriticgrowth in nearby neurons. Nat. Neurosci. 5, 1177–1184 (2002).

89. Horch, H.W., Kruttgen, A., Portbury, S.D. & Katz, L.C. Destabilization of corticaldendrites and spines by BDNF. Neuron 23, 353–364 (1999).


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

2f1 – f2 was more than 30 dB below f1 and f2, which is consistent withdata measured from a single point on the basilar membrane5,7. Nearthe apical end, such as at ∼ 2,800 µm, the 2f1 – f1 magnitude was evengreater than f2 by ∼ 20 dB.

In the bottom graph of Figure 1a, phases are plotted against longi-

Reverse propagation of sound inthe gerbil cochleaTianying Ren

It is commonly believed that the cochlea emits sounds throughbackward-traveling waves. In the present experiment using ascanning-laser interferometer, I detected forward-traveling butnot backward-traveling waves and found that the stapes vibratesearlier than the basilar membrane. These results contradict thecurrent theory and show that the ear emits sounds through thecochlear fluids as compression waves rather than along thebasilar membrane as backward-traveling waves.

It is widely thought that otoacoustic emissions—sounds generated bythe cochlea—propagate along the basilar membrane as backward-traveling waves to the middle ear and become acoustic emissions inthe ear canal1–4(see Supplementary Fig. 1 online). Some preliminaryresults have been reported5 (Narayan, S.S., Recio, A. & Ruggero, M.A.,Abstract 723, Twenty-first Midwinter Research Meeting of theAssociation for Research in Otolaryngology, St. Petersburg Beach, USA,February 15–19, 1998), however, that are inconsistent with the abovetheory, and the possibility of an alternative fluid wave has been dis-cussed (see Supplementary Note online).

I tested the backward-traveling wave theory by measuring the basilarmembrane vibration at the frequency of the otoacoustic emission as afunction of longitudinal location, using a newly developed scanning-laser interferometer6. The propagation direction of the basilar mem-brane vibration was determined by the slope of the phase-versus-longitudinal location curve. A negative slope in the phase dataindicates a dominative forward-traveling wave, and a positive slope indi-cates a backward-traveling wave. The delays of the basilar membrane andstapes vibrations were derived from the phase transfer functions at emis-sion frequencies (Supplementary Methods online). Animal procedureswere approved by the Oregon Health & Science University IACUC.

Data presented in Figure. 1a were evoked by two 60 dB SPL (0 dBSPL = 20 µPa) tones at 15.455 kHz (f1) and 17.000 kHz (f2), with afrequency ratio (f2/f1) of 1.1. The velocity magnitude curves inFigure 1a (top) clearly show the frequency-dependent longitudinalpatterns of basilar membrane vibration. At 17.000 kHz (dashed line),the maximum vibration was located on the basal side, whereas themaximum vibration at 15.455 kHz (dotted line) was on the apex side.The greatest overlap of f1 and f2 was near 2,200 µm. The maximumvibration at 2f1 – f2 (thick solid line in Fig. 1a) was not located at thearea of maximum overlap of f1 and f2. Rather, it was at ∼ 2,700 µm,which is close to the apical end where there is little vibration at f2. Atthe most overlapped area of f1 and f2 near the basal side (∼ 2,200 µm),

Oregon Hearing Research Center, Department of Otolaryngology and Head & Neck Surgery, Oregon Health & Science University, 3181 SW Sam Jackson Park Road,NRC 04, Portland, Oregon 97239-3098, USA. Correspondence should be addressed to T.R. ([email protected]).


Figure 1 Magnitudes (shown in logarithmic scale) and phases of basilarmembrane vibration at f1, f2 and 2f1 – f2 frequencies. (a–d) The f2/f1ratios were as follows: 1.1 in a and d, 1.05 in b and 1.2 in c; F2 was 17kHz in a–c and 12 kHz in d. Despite the change in f2/f1 and f2, all phasecurves show a negative relationship with the distance from the cochlearbase, indicating forward-traveling waves. All magnitude curves in a–c (topgraphs) show the frequency-dependent longitudinal pattern. That is,vibrations at high frequencies occurred on the basal side and those at lowfrequencies on the apical side.



©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


tudinal location. The phases of f1 and f2 show a negative relationshipwith distance from the cochlear base, indicating forward-travelingwaves. However, the phase curve at 2f1 – f2 does not show a backward-traveling wave. Indeed, the 2f1 – f2 phase (thick solid line)shows a negative relationship with distance from the base, whichclosely matches the phase response to an externally given 35 dB SPLtone at 13.909 kHz (thin solid line). The phase data show that thepropagation direction of the basilar membrane vibration at the 2f1 –f2 frequency is dominated by a forward-traveling wave.

To observe the effects of the f2/f1 ratio (the smaller the ratio, themore overlap between the two traveling waves at f1 and f2) on thepropagation direction of the basilar membrane vibration, I used vari-ous f2/f1 ratios to evoke the emissions. For example, I recordedresponses to 60 dB SPL tones at 16.190 kHz and 17 kHz (f2/f1 ratio of1.05; Fig. 1b), as well as responses evoked by 60 dB SPL tones at14.167 kHz and 17 kHz (f2/f1 = 1.2; Fig. 1c). The responses with thesmaller 1.05 ratio (Fig. 1b) indicated less separation between f1 andf2, more basal location of 2f1 – f2, and smaller phase slope differenceamong f1, f2 and 2f1 – f2 (compare bottom graphs in Fig. 1a and b).With the larger ratio (Fig. 1c) opposite effects were found. These dataconfirm that not only f1 and f2, but also 2f1 – f2, vibrate mainly asforward-traveling waves8–12.

It is possible that the putative backward-traveling wave at 2f1 – f2may be at a longitudinal location basal to the observed region inFigure 1a–c. To test this possibility, I moved the emission generationsite toward the apex using low-frequency primary tones. Data inFigure 1d were evoked by 70 dB SPL tones at 10.909 and 12 kHz (f2/f1= 1.1). The negative phase slope of 2f1 – f2 (9.818 kHz) indicates aforward-traveling wave.

The absence of a detectable backward-traveling wave was con-firmed by the finding that the stapes vibrates earlier than the basilarmembrane. I recorded magnitude and phase responses of the basilarmembrane at the f2 site, the stapes footplate and the emission to two70 dB SPL tones (f1 and f2) in a sensitive cochlea (Fig. 2). The fre-quency of f1 was stepped from 9.0 kHz to 16.8 kHz by 200-Hz incre-ments, whereas f2 was fixed at 17 kHz, resulting in a change in 2f1 – f2from 1 kHz to 16.6 kHz. The emission (thick solid line in Fig. 2a)increased with 2f1 – f2 frequency and reached a maximum near 12kHz, then decreased near the f2 frequency. The magnitude transferfunction of the stapes footplate vibration showed a similar pattern tothat of the emission except for the low frequency range. The magni-tude of basilar membrane vibration increased with the emission fre-quency and reached a maximum near the f2 frequency of 17 kHz. Thephase curve of the stapes vibration (Fig. 2b) showed the shallowestslope. The calculated delay based on phases (Fig. 2c andSupplementary Methods online) as a function of the emission fre-quency (Fig. 2d) shows that the stapes vibrates ∼ 50 µs earlier than thebasilar membrane.

The facts that the basilar membrane vibration at the emissionfrequency is dominated by a forward-traveling wave and that thestapes vibrates earlier than the basilar membrane contradict thetheory that the cochlea emits sounds through a backward-travelingwave that propagates on the basilar membrane1–4. The resultsreported here, however, are consistent with Békésy’s paradoxicalbasilar membrane vibration13, which propagates toward the heli-cotrema independently of the location of the vibration sourcealong the cochlear length, which has been theoretically demon-strated14. The present data demonstrate that the inner ear emitssound through the cochlear fluids as a compression wave, whichvibrates the stapes, launching a forward-traveling wave along thebasilar membrane (see Supplementary Fig. 2 and SupplementaryNote online).

Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTSI thank A.L. Nuttall, P. Gillespie and K. Grosh for comments on an earlier versionof the manuscript, E.V. Porsov for writing software, and S. Matthews for technicalhelp. Supported by the National Institute on Deafness and other CommunicationDisorders (NIDCD), and the National Center for Rehabilitative Auditory Research(NCRAR), Portland Veteran’s Administration Medical Center.

COMPETING INTERESTS STATEMENTThe author declares that he has no competing financial interests.

Received 20 November 2003; accepted 17 February 2004Published online at http://www.nature.com/natureneuroscience/

1. Kemp, D.T. Hear. Res. 22, 95–104 (1986).2. Probst, R., Lonsbury-Martin, B.L. & Martin, G.K. J. Acoust. Soc. Am. 89,

2027–2067 (1991).3. Shera, C.A. & Guinan, J.J. Jr. J. Acoust. Soc. Am. 105, 782–798 (1999).4. Knight, R.D. & Kemp, D.T. J. Acoust. Soc. Am. 109, 1513–1525 (2001).5. Robles, L., Ruggero, M.A. & Rich, N.C. J. Neurophysiol. 77, 2385–2399 (1997).6. Ren, T. Proc. Natl. Acad. Sci. USA 99, 17101–17106 (2002).7. Cooper, N.P. & Rhode, W.S. J. Neurophysiol. 78, 261–270 (1997).8. Rhode, W.S. J. Acoust. Soc. Am. 49 (Suppl. 2), 1218–1231 (1971).9. Khanna, S.M. & Leonard, D.G. Science 215, 305–306 (1982).10. Russell, I.J. & Nilsen, K.E. Proc. Natl. Acad. Sci. USA 94, 2660–2664 (1997).11. Narayan, S.S., Temchin, A.N., Recio, A. & Ruggero, M.A. Science 282, 1882–1884

(1998).12. Olson, E.S. Nature 402, 526–529 (1999).13. von Békésy, G. Experiments in Hearing (McGraw-Hill, New York, 1960).14. Zwislocki, J.J. J. Acoust. Soc. Am. 25, 986–989 (1953).


Figure 2 Frequency responses of the basilar membrane, stapes and ear-canal sound pressure to constant f2 (17 kHz) and frequency-varied f1.Magnitudes (a), phases (b,c) and delay (d) at 2f1 – f2 of the emission(thick solid line), basilar membrane (BM; dotted line) and the stapes(dashed line) vibrations. (a) The magnitude transfer function of the stapesfootplate vibration shows a pattern similar to that of the emission butdifferent from that of the BM vibration at the F2 place. (b,c) The phasecurve of the stapes vibration shows the shallowest slope, indicating theshortest delay. (d) The delay of stapes vibration is approximately 50 µsshorter than that of the BM vibration.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


Cholecystokinin-mediatedsuppression of feeding involvesthe brainstem melanocortinsystemWei Fan, Kate L J Ellacott, Ilia G Halatchev, Kanji Takahashi,Pinxuan Yu & Roger D Cone

Hypothalamic pro-opiomelanocortin (POMC) neurons helpregulate long-term energy stores. POMC neurons are also foundin the nucleus tractus solitarius (NTS), a region regulatingsatiety. We show here that mouse brainstem NTS POMCneurons are activated by cholecystokinin (CCK) and feeding-induced satiety and that activation of the neuronalmelanocortin-4 receptor (MC4-R) is required for CCK-inducedsuppression of feeding; the melanocortin system thus providesa potential substrate for integration of long-term adipostaticand short-term satiety signals.

Hypothalamic POMC neurons tonically inhibit food intake1 and areregulated by the long-term adipostatic factor leptin2–4. However, thecentral melanocortin system is also important in the acute regulationof satiety; in particular, central administration of melanocortinsreduces food intake by decreasing meal size, a hallmark of satiety5,6.These hypothalamic neurons send fibers to MC4-R target sites inboth the hypothalamus and brainstem, and melanocortin agonistsadministered to either region inhibit feeding7. Notably, in addition toexpression in the arcuate nucleus of the hypothalamus (ARC), POMCis also expressed in the caudal aspect of the NTS8, the primary site ofsynapse of vagal afferent fibers transmitting satiety information fromthe gastrointestinal system. NTS neurons are activated by either elec-trical or CCK-induced stimulation of vagal afferent fibers.Furthermore, leptin and CCK act synergistically to inhibit feedingand activate NTS neurons9. Yet regulation of POMC cells in the NTSby metabolic state has not been reported. Here, we test the hypothesesthat (i) brainstem POMC neurons are activated by satiety signals and(ii) central melanocortin signaling is required for the action of spe-cific signals that acutely inhibit feeding.

Intraperitoneal (i.p.) injection of CCK-8s (the sulfated 8-amino-acid form of cholecystokinin) significantly increased c-Fosimmunoreactivity in the NTS (saline 3 ± 1 cells per section, n = 6;CCK-8s 3.5 µg/kg, 54 ± 11 cells per section, n = 4; 10 µg/kg, 80 ± 11cells per section, n = 4; P < 0.001) (Fig. 1; compare Fig. 1a,d), as shownpreviously9. Immunohistochemical experiments using a previouslycharacterized transgenic mouse that expresses enhanced green fluores-cent protein (EGFP) under the control of the POMC promoter4 showedno significant difference in the number of POMC-EGFP–immunoreac-tive (IR) neurons in the NTS between saline- (Fig. 1b) and CCK-8s-treated mice (Fig. 1e). Of the NTS POMC-EGFP neurons, >30%coexpressed c-Fos immunoreactivity after CCK-8s treatment (Fig.1f,g). c-Fos expression in the ARC did not differ significantly betweengroups treated with i.p. saline or CCK-8s (data not shown). We alsoexamined a model of feeding-induced satiety. We gave POMC-EGFPmice a 5-d training in which they were allowed access to food for two

periods totaling 5 h (9:00–10:00 h, 14:00–18:00 h) and examined c-Fosimmunoreactivity in the ARC and NTS at 11:00 h (see SupplementaryFig. 1 online; fed n = 8; fasted, n = 3, *** P < 0.001). Feeding activatedc-Fos expression in ∼ 21% of ARC POMC neurons but also in 13% ofNTS POMC neurons (Supplementary Fig. 1). Although a majority ofc-Fos-IR cells in the ARC were POMC positive, only a few percent ofthose in the NTS were, showing the complexity of cells in the NTS


Vollum Institute, Oregon Health and Science University, Portland, Oregon 97239-3098, USA. Correspondence should be addressed to R.D.C. ([email protected]) orW.F. ([email protected]).


Figure 1 CCK-8s activates POMC neurons in the NTS. (a) Saline (i.p.)activates c-Fos (red) in only a few NTS neurons (arrows). Scale bar = 70 µm.(b) Anti-GFP antibodies detect POMC neurons (green) in NTS of the EGFP-POMC mouse. (c) POMC neurons are not activated by saline treatment. (d) CCK-8s (10 µg/kg, i.p.) activates c-Fos (red) in NTS neurons. (e) CCK-8s(10 µg/kg, i.p.) does not alter expression of POMC in NTS (compare b,e). (f) CCK-8s (10 µg/kg, i.p.) activates c-Fos in NTS POMC neurons (red, c-Fos;green, GFP; yellow-orange, c-Fos + GFP). (g) ∼ 30% of NTS POMC neuronsare activated by i.p. CCK-8s (3.5 or 10 µg/kg; ***, P < 0.001 vs. saline,statistical test done by one-way ANOVA with Dunnett’s post hoc test). (h) Receipt of long-term adipostatic signals and acute satiety signals byPOMC neurons in ARC and NTS, respectively. Blue, nuclei containing POMCneurons; yellow, nuclei containing MC4-R neurons that may serve tointegrate adipostatic and satiety signals. Red arrows, adipostatic signaling;green arrows, satiety signaling. BST, bed nucleus of stria terminalus; CEA,central nucleus of amygdala; PVN, paraventricular nucleus of hypothalamus;LH, lateral hypothalamic area; LPB, lateral parabrachial nucleus; AP, areapostrema; DMV, dorsal motor nucleus of vagus. All studies followed the NIHGuide For the Care and Use of Laboratory Animals and were approved by theOregon Health and Sciences University Animal Care and Use Committee.See Supplementary Methods online for details.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


involved in satiety. Previous work has shown that both catecholamin-ergic and glucagon-like peptide-1 (GLP-1)-positive cells in the NTSare involved in satiety10,11. Dual immunohistochemical analysisshowed that although POMC-EGFP–IR cells and tyrosine hydroxy-lase–IR cells are found in the same region of the NTS, they are notcoexpressed in the same neurons (see Supplementary Fig. 2 online).Likewise, POMC and GLP-1 do not colocalize in NTS neurons:POMC-EGFP–IR neurons are focused more medially than GLP-1-IRneurons (Supplementary Fig. 2).

To test whether feeding suppression by CCK-8s was dependent oncentral melanocortin signaling, we examined the ability of CCK-8s toinhibit food intake after a fast in three different mouse lines, two ofwhich carry deletions of the genes encoding melanocortin receptors 3and 4, respectively: C57BL/6J, C57BL/6J Mc3r–/–12 and C57BL/6JMc4r–/–13. Administration of CCK-8s i.p. after a 16-h fast produced a≥50% inhibition of food intake in the first 30 min in both wild-typeand Mc3r–/– mice (Fig. 2a) and continued to inhibit food intake for upto 180 min in each strain. We then administered CCK-8s to femaleMc4r–/– mice and age-matched female wild-type mice. CCK-8s signifi-cantly reduced food intake in wild-type mice, but not in Mc4r–/– mice

(Fig. 2b), at time points from 30 to 180 min. Next we examined the siteof action of endogenous melanocortins. We administered the MC3-Rand MC4-R antagonist SHU9119 (ref. 14) to rats via either the 3rd or4th ventricle to assess the relative contributions of forebrain and brain-stem MC4-R target sites in CCK-mediated inhibition of feeding. Weused subthreshold doses of SHU9119 previously determined not tostimulate food intake by these routes. Third-ventricular injection ofSHU9119, expected to access both forebrain and brainstem MC4-Rsites, partially attenuated the inhibition of food intake induced by i.p.injection of CCK-8s (Fig. 2c). Fourth-ventricle injection, which dye-injection tests had shown primarily accesses brainstem sites, completelyattenuated the CCK-8s-induced inhibition of food intake (Fig. 2d).

Both CCK-8s and normal food-induced satiety activated a smallgroup of NTS POMC neurons. These brainstem POMC cells are dis-tinct from previously characterized GLP-1-positive and cate-cholaminergic NTS neurons. CCK-8s-induced inhibition of feedingalso seems to depend on MC4-R signaling. These findings support amodel in which brainstem MC4-R neurons, and possibly NTS POMCneurons, contribute to the satiety effects of CCK and other meal-related satiety signals. Recently, electrical activation of cranial visceralafferents in the solitary tract was reported to activate POMC NTSneurons (Appleyard, S.M. et al., Soc. Neurosci. Abstr. 29, 231.11, 2003);however, the role of NTS POMC neurons in the perception of meal-related satiety has not been established. The distribution of POMCneurons in the ARC, where they are sensitive to the adipostatic hor-mone leptin, and the NTS, where they are responsive to vagally medi-ated satiety signals, makes the central melanocortin system ideallysuited for the integration of acute regulation of feeding behavior withthe long-term control of energy stores (Fig. 1h) Resistance to factorssuch as CCK may explain, in part, the profound hyperphagia andincreased meal size seen in obese subjects with mutations in Mc4r15.


ACKNOWLEDGMENTSSupported by US National Institutes of Health grants DK55819 (R.D.C.) andDK62179 (W.F.), and a grant from the Wellcome Trust (K.L.J.E.). POMC-EGFPmice were a kind gift of M. Low (Oregon Health and Science University).

COMPETING INTERESTS STATEMENTThe authors declare competing financial interests; see Nature Neuroscience websitefor details.

Received 15 September 2003; accepted 12 February 2004Published online at http://www.nature.com/natureneuroscience/

1. Fan, W., Boston, B.A., Kesterson, R.A., Hruby, V.J. & Cone, R.D. Nature 385,165–168 (1997).

2. Cheung, C.C., Clifton, D.K. & Steiner, R.A. Endocrinol. 138, 4489–4492 (1997).3. Elias, C.F. et al. Neuron 23, 775–786 (1999).4. Cowley, M.A. et al. Nature 411, 480-484 (2001).5. Williams, D.L., Grill, H.J., Weiss, S.M., Baird, J.P. & Kaplan, J.M.

Psychopharmacology 161, 47–53 (2002).6. Azzara, A.V., Sokolnicki, J.P. & Schwartz, G.J. Physiol. Behav. 77, 411–416

(2002).7. Grill, H.J., Ginsberg, A.B., Seeley, R.J. & Kaplan, J.M. J. Neurosci. 18,

10128–10135 (1998).8. Joseph, S.A., Pilcher, W.H. & Bennet-Clarke, C. Neurosci. Lett. 38, 221–225

(1983).9. Wang, L., Martinez, V., Barrachina, M.D. & Tache, Y. Brain Res. 791, 157–166

(1998).10. Rinaman, L. Am. J. Physiol. 277, R582–R590 (1997).11. Luckman, S. J. Neuroendocrinol. 4, 149–152 (1992).12. Butler, A.A. et al. Endocrinol. 141, 3518–3521 (2000).13. Huszar, D. et al. Cell 88, 131–141 (1997).14. Hruby, V.J. et al. J. Med. Chem. 38, 3454–3461 (1995).15. Farooqi, I.S. et al. N. Engl. J. Med. 348, 1160–1163 (2003).


10

8

6

4

2

0

Cu

mu

lati

ve fo

od

inta

ke(g

)

40 60 120Time (min)

**

***

**

*

*

*

Saline saline n = 6Saline CCK n = 8SHU9119 CCK n = 8SHU9119 saline n = 6

15

10

5

0

Cu

mu

lati

ve fo

od

inta

ke(g

)

40 60 120Time (min)

****

*

Saline saline n = 4Saline CCK n = 5SHU9119 CCK n = 5SHU9119 saline n = 4

****

2.0

1.5

1.0

0.5

0.0Cu

mu

lati

ve fo

od

inta

ke(g

)

30 60 120 180Time (min)

C57 saline n = 8C57 CCK n = 7MC4-R – / – saline n = 7MC4-R – / – CCK n = 7

***

***

**

2.0

1.5

1.0

0.5

0.0Cu

mu

lati

ve fo

od

inta

ke(g

)C57 saline n = 7C57 CCK n = 8MC3-R – / – saline n = 10MC3-R – / – CCK n = 9

30 60 120 180Time (min)

*** *

*

***

*** *

*

** *

*

ba

dc

Figure 2 Brainstem MC4-R signaling is required for CCK-8s-inducedfeeding inhibition. (a) Mc3r–/– mice are fully responsive to CCK-inducedinhibition of feeding. After a 16-h fast, 5–10-month wild-type C57BL/6J(C57) and Mc3r–/– mice (MC3-RKO) were injected i.p. with saline or CCK-8s (3 nmol/kg); the strains showed a comparable anorexigenic response toCCK-8s 30–180 min after treatment. (b) MC4-R is required for CCK-induced inhibition of feeding. After a 16-h fast, 9-week wild-type andMc4r–/– mice (MC4-RKO) were injected i.p. with saline or CCK-8s (3 nmol/kg). CCK-8s significantly reduced food intake in wild-type but notMc4r–/– mice. (c) Pharmacological blockade of central melanocortinreceptors in rats partially blocks CCK-induced inhibition of feeding. Ratsreceived 3rd-ventricle injections of a subthreshold dose of SHU9119(0.375 nmol/4 µl) 10–15 min before i.p. injection of CCK-8s (3 nmol/kg).(d) Pharmacological blockade of brainstem melanocortin receptors in ratsfully blocks CCK-induced inhibition of feeding. Rats received 4th-ventricleinjections of a subthreshold dose of SHU9119 (0.2 nmol/4 µl) just beforei.p. injection of CCK-8s (3 nmol/kg). Data given as mean ± s.e.m.Statistical analyses were done using one-way ANOVA (a,b) or unpaired t-test (c,d). Data presented as mean ± s.e.m. *, P < 0.05; **, P < 0.01; ***, P < 0.001. See Supplementary Methods online for details.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

(1/100 dilution; Fig. 1a) and a lower concentration (1/10,000 dilu-tion; Supplementary Fig. 1 online, panel a). At neither concentra-tion did the mixture of benzaldehyde and guaiacol have a strongervanilla character than that of its individual components. A similarresult was obtained when odor pairs were rated on an odor similar-ity rating scale (Supplementary Fig. 1 online, panel b).

A second prediction of vibration theory as proposed by Turin isthat aldehydes with an even number of carbon atoms have a differentodor than those with an odd number7. Subjects rated pairs of aldehy-des (1/10 dilution) that differed in chain length by up to six carbonatoms. Subjects rated the two aldehydes as smelling more dissimilar asthe difference in carbon atom number increased (Fig. 1b). Similarresults were obtained with pure aldehydes (Supplementary Fig. 2online). Contrary to Turin’s prediction, pairs consisting of two odd ortwo even numbered aldehydes were not perceived as more similarthan pairs consisting of an odd and an even numbered aldehyde (Fig. 1c). We found instead, as suggested in previous studies, that thecarbon chain length of these molecules is the salient feature sensed bythe olfactory system8.

A third prediction of Turin’s vibration theory is that acetophenone(AP) and completely deuterated acetophenone (AP-d8), which havethe same shape but different molecular vibrations, should have dis-tinguishable smells9. First, subjects rated paired odors for similarityusing a 10-point scale (0 = same; 10 = very different). Similarityscores for the AP versus AP-d8 pairing were no different from those ofthe identical-odorant pairings (Fig. 2a). In addition, we used a trian-gle test in which subjects were asked to identify the odd stimulus fromamong three vials (two of which contained the same substance).

A psychophysical test of thevibration theory of olfactionAndreas Keller & Leslie B Vosshall

At present, no satisfactory theory exists to explain how a givenmolecule results in the perception of a particular smell. Onetheory is that olfactory sensory neurons detect intramolecularvibrations of the odorous molecule. We used psychophysicalmethods in humans to test this vibration theory of olfactionand found no evidence to support it.

A book about the physiologist Luca Turin1, reviewed previously inNature Neuroscience2 and elsewhere3,4, has generated new interest inthe theory that the smell of a molecule is determined by intramolecu-lar vibrations rather than by the molecule’s shape. Vibration theorywas introduced in the 1930s5 and was later extended6, but no biologi-cal mechanism to convert molecular vibrations into neuronal activa-tion was proposed. As a result, the theory has been largely neglected inthe research community. In the 1990s, Turin proposed a transductionmechanism involving inelastic electron tunneling7. Whether becauseof skepticism or ‘scientific conspiracy’ (as alleged in the book andechoed in most reviews), his predictions have failed to generateempirical tests by other researchers. In the present study, we testedvibration theory’s key psychophysical predictions.

All subjects gave informed consent to participate in this study andwere tested in a well-ventilated examination room of the RockefellerUniversity hospital. Procedures were approvedby the university’s Institutional Review Board.To minimize observer bias, we used a double-blind protocol such that neither the subjectsnor the test administrator knew the identity ofthe odorant in a given vial (see SupplementaryMethods online).

Turin predicts that the smell of a mixtureof guaiacol and benzaldehyde has a vanillacharacter not found in its componentsbecause the combined molecular vibrationsof benzaldehyde and guaiacol approximatethe vibrations of vanillin7. To test this pre-diction, we asked subjects to rate the vanillacharacter of benzaldehyde, guaiacol and a1:1 mixture of both. Subjects were firstfamiliarized with the individual stimuli attwo different concentrations under non-blind conditions. In a subsequent test,vanillin at both concentrations was identi-fied with an accuracy of 84%. After beingfamiliarized with the 13-point rating scale(1 = no vanilla, 13 = extremely vanilla),subjects rated the vanilla character of theindividual components and the two- andthree-component mixtures, presented inrandom order. This procedure was done attwo concentrations: a higher concentration

Laboratory of Neurogenetics and Behavior, The Rockefeller University, 1230 York Avenue, Box 63, New York, New York 10021, USA. Correspondence should beaddressed to A.K. ([email protected]).




Figure 1 Additive synthesis and homologous series. (a) Subjects rated (on a 13-point scale13) thevanilla character of stimuli (1/100 dilutions) presented with an inter-trial interval of 30 s. Thebenzaldehyde/guaiacol mixture did not have a vanilla character stronger than either of its components(horizontal black line on each bar indicates median, boxed regions indicate 25–75% quantiles,whiskers indicate 10–90% quantiles; n = 24 subjects, 12 female; P > 0.05; Newman-Keuls test formultiple comparisons after Friedman’s test). The olfactory sensation produced by vanillin issuppressed by trigeminal stimulation14, but at the stimulus concentration used here there was nosuch interference, as is evident by the high score of the three-component mixture. Equivalent resultswith the same subjects were obtained at a 1/10,000 dilution (Supplementary Fig. 1a online). Purity ofodors: benzaldehyde >99%, guaiacol 99.7%, vanillin 99.9%. (b) Odor dissimilarity of pairs ofaldehydes was rated on a scale from 0 (same) to 10 (very different). Each subject (n = 24, 12 female)rated three randomly picked pairs from each of the seven groups (∆0, ∆1, ∆2, ∆3, ∆4, ∆5 and ∆6).Odor solutions were 1/10 dilutions. (c) The data shown in b are replotted to compare the mediansimilarity rating for pairs of aldehydes consisting of two odd, two even, or an odd and an even chainlength. No difference between groups was found; see Supplementary Methods online for details.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


ACKNOWLEDGMENTSWe thank A. Gilbert for expert advice, members of the Vosshall laboratory forcomments on the manuscript, and E. Gotschlich, B. Coller and the staff of theRockefeller University Hospital. A.K. is an M.S. Stoffel Fellow in Mind, Brain andBehavior.


Received 18 December 2003; accepted 21 January 2004Published online at http://www.nature.com/natureneuroscience/

1. Burr, C. The Emperor of Scent (Random House, New York, 2002).2. Gilbert, A.N. Nat. Neurosci. 6, 335 (2003).3. Maslin, J. The New York Times February 6 (2003), p. E8.4. Givhan, R. The Washington Post February 16 (2003), p. X–T8.5. Dyson, G.M. Chem. Ind. 647–651 (1938).6. Wright, R.H. J. Theor. Biol. 64, 473–502 (1977).7. Turin, L. Chem. Senses 21, 773–791 (1996).8. Laska, M. & Teubner, P. Chem. Senses 24, 263–270 (1999).9. Turin, L. & Yoshii, F. in Handbook of Olfaction and Gustation (ed. Doty, R.L.)

275–294 (Marcel Dekker, New York, 2003).10. Laska, M. & Teubner, P. Chem. Senses 24, 161–170 (1999).11. Livermore, A. & Hummel, T. Int. Arch. Occup. Environ. Health 75, 305–313

(2002).12. Haffenden, L.J.W., Yaylayan, V.A. & Fortin, J. Food Chem. 73, 67–72 (2001).13. Watson, W.L., Laing, D.G., Hutchinson, I. & Jinks, A.L. Dev. Psychobiol. 39,

137–145 (2001).14. Kobal, G. & Hummel, C. Electroencephalogr. Clin. Neurophysiol. 71, 241–250

(1988).

To verify that subjects understood the task, we included enantiomers(r-carvone and s-carvone) that are readily discriminable10 and differin shape but not vibration. Subjects easily distinguished the enan-tiomers but could not distinguish AP from AP-d8 (Fig. 2b). Finally,we used a duo-trio test in which two stimuli were presented and thesubject was asked to identify the one identical to a third referencesmell. In a separate session, we tested six subjects who had successfullydistinguished AP from AP-d8 to determine whether their correctselections reflected chance performance or true discrimination ofthese two odorants. None of the six subjects was able to distinguishthe two smells. The proportion of correct choices ranged from 43% to67% (mean, 53%; standard error ± 14%; Fig. 2c).

To rule out interference of the trigeminal chemosensory systemwith olfactory perception seen at high stimulus concentrations11, weused duo-trio tests to show that AP and AP-d8 were not distinguishedat a wide range of concentrations (Fig. 2d). It has recently beenreported that naive subjects perceive a difference between the odors ofdeuterated and regular benzaldehyde, but this previous study12 wasnot run double-blind and used an anomalous version of the duo-triotest. Taken as a whole, our results provide no evidence that regularand deuterated acetophenone smell different to naive subjects. Wecannot exclude the possibility, however, that olfactory training orexperience could alter the outcome of the tests done here.

After testing a variety of psychophysical predictions of vibrationtheory, as formulated by Turin, we conclude that molecular vibrationsalone cannot explain the perceived smell of an odorous molecule.

Figure 2 Isotope substitution. (a) The similarity between the smells of regular acetophenone (AP) and deuterated acetophenone (AP-d8) was rated on ascale from 0 (same) to 10 (very different) (horizontal black line on each bar indicates median, boxed regions indicate 25–75% quantiles, whiskers indicate10–90% quantiles; n = 108, 36 and trials, respectively, for the 3 comparisons shown left to right; 36 subjects, 22 female). (b) Subjects easilydistinguished r-carvone (r-CAR) from s-carvone (s-CAR) in triangle tests (one test per subject; n = 36 subjects, 22 female), but not AP from AP-d8 (twotests per subject; n = 72). (c) In duo-trio tests, two odors were presented and the subject was asked to identify the one identical to a third reference smell.Each of six subjects (1 female) took this test 30 times over the course of three days (n = 180 trials). (d) Duo-trio tests were performed with differentdilutions of both r-carvone/s-carvone and AP/AP-d8 (n = 24 subjects, 12 female). (b–d) The percentage of correct choices and the 95% confidenceintervals are shown. The dashed lines indicate chance performance. Chi-square tests were used to compare observed and expected frequencies. ***P < 0.001; *P < 0.05. Purity of odors: AP 99.3%, AP-d8 99.9%, r-carvone/s-carvone >99%.



©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


Parietal somatosensoryassociation cortex mediatesaffective blindsightSilke Anders1, Niels Birbaumer1, Bettina Sadowski2, Michael Erb3,Irina Mader3, Wolfgang Grodd3 & Martin Lotze1

To investigate the neural substrates underlying emotionalfeelings in the absence of a conscious stimulus percept, wepresented a visual stimulus in the blind field of partiallycortically blind patients and measured cortical activity (byfunctional magnetic resonance imaging, fMRI) before and afterthe stimulus had been paired with an aversive event. Afterpairing, self-reported negative emotional valence and bloodoxygen level–dependent (BOLD) responses in somatosensoryassociation areas were enhanced, whereby somatosensoryactivity predicted highly corresponding reported feelings andstartle reflex amplitudes across subjects. Our data providedirect evidence that cortical activity representing physicalemotional states governs emotional feelings.

In this study, we aimed to identify the neural substrate underlyingemotional feelings in the absence of a conscious stimulus percept1.Emotional stimuli presented in a cortical visual field defect modulatereflexive emotional responses via extrastriate projections to the amyg-dala2–5. In the absence of a cortical stimulus representation, emo-tional experiences might rely on cortical activity representing thephysiological state of the organism. Based on clinical6 and neu-roimaging studies7, somatosensory association areas within the ante-rior parietal cortex have been suggested to represent internal statesduring emotional processing. Thus, we hypothesized that neuralactivity in this region should be increased when patients report emo-tional feelings in the absence of a conscious stimulus percept, and thatit should furthermore predict the degree of correspondence betweenself-reported emotional valence and reflexive emotional responses.

Nine patients with postgeniculate lesions resulting in partial dis-connection or destruction of the left (n = 6) or right (n = 3) primaryvisual cortex participated in the study. Written informed consent wasobtained from all participants. High resolution T1 and diffusion ten-sor magnetic resonance images confirmed lesions corresponding tovisual field defects identified by Tübingen Perimetry in all patients(see Supplementary Fig. 1 and Supplementary Table 1 online). Whilesubjects focused on a central fixation cross, a male face with a neutralexpression was randomly presented in the left and right visual hemi-field. After 8 habituation trials on either side, 8 of 16 face presenta-tions in each hemifield were paired with an aversive human scream(Fig. 1). Thus, all patients received an equal number of paired andunpaired left and right hemifield stimulations, but only blind-fieldstimulations of each patient were included in the analyses. The size ofthe facial stimulus was individually adjusted to assure that blind-fieldstimulation was confined to the absolute visual field defect (approxi-mately 6 × 8.5°). Average luminance of the visual stimulus and thebackground were matched. Fixation was controlled during the entireexperiment8, and trials with saccades leading to an overlap between

the visual stimulus and the intact part of the visual field were dis-carded. fMRI data were processed with a linear model approach usingSPM99 software9. Conjunction10 in random-effects analysis revealedvoxels commonly activated in patients with left and right visual fielddefects. Bonferroni correction was used to correct for multiple com-parisons within the search volume, comprising cortical gray mattervoxels of a standard brain11 (19,449 voxels). Startle eyeblink ampli-tudes and skin conductance responses (SCR) were recorded duringscanning, and after blocks of four blind-field stimulations, subjectsrated12 the valence of their emotional feelings during two additionalblind-field stimulations, intermixed with two blank-screen presenta-tions. For all measures, responses during blind-field stimulation werecontrasted with responses during the blank-screen baseline, sepa-rately for pre-pairing and pairing blocks, and then subtracted.

All patients denied any visual sensation during blind-field stimula-tion, and analysis of fMRI data showed no activity in the region corre-sponding to the calcarine sulcus during blind-field stimulation.Startle eyeblink amplitudes of five patients (the remaining four sub-jects did not tolerate startle probe intensities that reliably elicited thereflex), and negative emotional feelings of all nine patients were sig-nificantly enhanced during blind field stimulation, relative to blankscreen baseline, after the face had been paired with the aversivescream (startle, t = 2.39; valence, t = –1.93; P < 0.05; Fig. 2). BOLDresponses of eight patients (one subject made saccades during 30% ofthe face presentations during scanning and was not included in thefMRI analysis) were significantly increased in the left anterior parietalcortex (t = 5.04, P < 0.05; MNI coordinates11 (x, y, z) = –42, –42, 45;region SII; Fig. 3a). BOLD responses in the right anterior parietal cor-tex were also increased, but this effect did not reach statistical signifi-cance (t = 1.64; (x, y, z) = 39, –48, 42). No significant changes of skinconductance responses were observed.

To further evaluate the relation among anterior parietal activity,peripheral physiological responses and reported feeling, we compared


1Institute of Medical Psychology and Behavioral Neurobiology, University of Tübingen, Gartenstrasse 29, 72074 Tübingen, Germany. 2Department of Pathophysiologyof Vision and Neuroophthalmology, University Eye Hospital Tübingen, Schleichstrasse 12-16, 72076 Tuebingen, Germany. 3Section for Experimental MR of the CNS,Department of Neuroradiology, University Hospital Tübingen, Hoppe-Seyler-Strasse 3, 72076 Tübingen, Germany. Correspondence should be addressed to S.A.([email protected]).


Figure 1 Experimental design. After a baseline of 32 ± 8 s, the face waspresented for 12 s. Left and right hemifield presentations were randomlyintermixed. After 16 consecutive presentations of the face without aversivestimulation, 8 out of 16 presentations of the face in either hemifield were paired with an aversive scream (2.2 s). White-noise startle probes (50 ms) occurred during half of the baseline intervals and half of the facepresentations (not shown).

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


activity in the highest activated voxel with startle potentiation andnegative valence ratings across subjects. Parietal activity was moreincreased in patients who showed a stronger startle potentiation, buta large increase in startle activity did not necessarily lead to strongparietal activity. Although there was a positive relation between pari-etal activity and reported negative valence (r = 0.60), parietal activitywas best predicted by the level of correspondence between reportednegative valence and startle reflex potentiation, computed as the rankdifference between the two measures (r = 0.79; Fig. 3b).

Our data confirm that aversive responses to visual stimuli can beelicited after destruction of the primary visual cortex5,13. Theseresponses have been ascribed to an extrastriate pathway to the amyg-dala, comprising the superior colliculus of the tectum and the pul-vinar of the thalamus3. In humans, amygdala activity has beenobserved in response to aversive visual stimuli presented in a post-geniculate visual field defect, and this activity covaried with neuralactivity in the posterior thalamus4. In both animals14 and humans15,startle reflex amplitudes are modulated by direct projections from thecentral nucleus of the amygdala to the nucleus reticularis pontis cau-dalis in the brain stem.

Neural activity in cortical areas involved in the representation ofthe internal state of an organism such as the startle response may con-stitute part of the basis of emotional feelings6. The anterior parietalcortex with its numerous inputs from the internal milieu representssuch an area7. The results of the present study provide direct evidencethat anterior parietal activity is linked to the level of the correspon-dence between reported emotional experiences and startle reflexpotentiation. We propose that the neural circuit that mediates emo-tional experience in the absence of conscious stimulus perceptionincorporates the previously described subcortical pathway to theamygdala2–4. This pathway may modulate reflexive responses and, viaperipheral or direct feedback, cortical activity that constitutes part ofthe neural basis for emotional feelings.


ACKNOWLEDGMENTSWe thank H. Flor, K. Mathiak, R. Veit, N. Weiskopf, L. Weiskrantz and D. Wildgruber for helpful discussions, B. Newport, M. Hülsmann and B. Wietek fortechnical support, and H.O. Karnath, P. Stoerig and U. Schiefer for permitting us toinclude patients from their wards. This study was partly supported by theVolkswagen Foundation and the Junior Science Program of the HeidelbergerAcademy of Sciences and Humanities.



1. Weiskrantz, L. Brain 126, 265–266 (2003).2. LeDoux, J.E. Curr. Opin. Neurobiol. 2, 191 (1992).3. Linke, R., De Lima, A.D., Schwegler, H. & Pape, H.C. J. Comp. Neurol. 403,

158–170 (1999).4. Morris, J.S., DeGelder, B., Weiskrantz, L., & Dolan, R.J. Brain 124, 1241–1252

(2001).5. Hamm, A.O. et al. Brain 126, 267–275 (2003).6. Adolphs, R., Damasio, H., Tranel, D., Cooper, G. & Damasio, A.R. J. Neurosci. 20,

2683–2690 (2000).7. Damasio, A.R. et al. Nat. Neurosci. 3, 1049–1056 (2000).8. Kimmig, H., Greenlee, M.W., Huethe, F. & Mergner, T. Exp. Brain Res. 126,

443–449 (1999).9. Friston, K.J. et al. Hum. Brain Mapp. 2, 189–210 (1995).10. Price, C.J. & Friston, K.J. Neuroimage 5, 261–270 (1997).11. Collins, D.L., Neelin, P., Peters, T.M. & Evans, A.C. J. Comput. Assist. Tomogr. 18,

192–205 (1994).12. Bradley, M.M. & Lang, P.J. J. Behav. Ther. Exp. Psychiatry 25, 49–59 (1994).13. Rosen, J.B. et al. J. Neurosci. 12, 4624–4633 (1992).14. Hitchcock, J.M. & Davis, M. Behav. Neurosci. 105, 826–842 (1991).15. Pissiota, A., Frans, O., Fredrikson, M., Langstrom, B. & Flaten, M.A. Eur. J.

Neurosci. 15, 395–398 (2002).


Figure 2 Startle eyeblink amplitudes and reported emotional valenceduring blind-field stimulation, relative to blank screen baseline, before(pre) and after (post) the visual stimulus had been paired with an aversivescream. Symbols distinguish individual patients.

Figure 3 Cortical activity during blind-field stimulation after the visualstimulus had been paired with the aversive event. (a) A statisticalparametric map (SPM) superimposed on coronal (top) and horizontal(bottom) sections of a standard brain11, contrasting BOLD responsesduring blind field stimulation, relative to blank screen baseline, beforeand after pairing. BOLD responses were significantly increased in the leftanterior parietal cortex. (b) Scatter plots showing the relation betweenleft anterior parietal activity and startle potentiation (top), reportednegative valence (middle), and the level of correspondence betweenreported negative valence and startle reflex potentiation (bottom),computed as rank difference between the two measures. Numbers refer to individual patients.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

Several studies suggest that small synaptic vesicles (SSVs) releaseneurotransmitter by full fusion as well as through transient fusionpores (‘kiss-and-run’ exocytosis)1–3. Capacitance recordings thatmonitor changes in plasma membrane surface area indicate thatSSV endocytosis at the calyx of Held occurs 50–100 ms after fusion4

and that 5% of fusion events by pituitary SSV-like microvesicles arefollowed (within 2 s) by endocytosis5. These relatively rapidinstances of vesicle endocytosis are consistent with formation oftransient fusion pores. In the neuromuscular junction, the kinaseinhibitor staurosporine attenuates the release of the amphipathicfluorescent dye FM1-43 more than the release of acetylcholine dur-ing SSV fusion6. This suggests that PKC inhibition reduces the SSVfusion pore aperture and may inhibit full fusion. In hippocampalneurons, kiss-and-run, full fusion and an intermediate mode ofendocytosis can be observed by labeling SSVs with a pH-sensitivefluorescent protein7. Also in some hippocampal neurons, some SSVfusion events result in partial loss of FM1-43 fluorescence8, sug-gesting that the vesicles close before FM1-43 release is complete.Neurotransmitters, however, have far higher diffusion coefficientsthan FM1-43 (refs. 6,9), and it is not known whether transient poreopenings are sufficient for release of the entire neurotransmittercontent of an SSV. In addition, the kinetics of kiss-and-run exocy-tosis cannot be determined due to the insufficient temporal resolu-tion of current approaches. We therefore adapted carbon fiberamperometry to record dopamine release from synaptic terminalsof cultured rat ventral tegmental area neurons. This techniquedirectly measures dopamine flux with a time resolution that is 2–5orders of magnitude greater than capacitance, imaging and postsy-naptic recordings (Methods). We found that small synaptic vesiclesregulate the release of neurotransmitter via rapid flickering of thefusion pore.

RESULTSDopamine release from midbrain neuronsSSVs are the predominant synaptic vesicles in cultured dopamineneurons from the ventral midbrain of rats (>99%)10,11. Neuronswere stimulated with either 40 mM K+ (Fig. 1a) or a combination of80 mM K+ and 20 nM α-latrotoxin (K+/α-LTX; Fig. 1b).α-Latrotoxin inserts into the plasma membrane in a manner facili-tated by neurexin-1 and CIRL/latrophilin receptors and forms acation channel that enables Ca2+ to enter the cell, thereby increasingthe number of exocytotic events12 (Table 1 legend10).

Both secretagogues elicited a variety of amperometric peaks (Fig. 1c,d). The average number of dopamine molecules recordedper amperometric event was similar for both secretagogues (K+,15,800 ± 4,000 molecules; K+/α-LTX, 11,600 ± 1,700 molecules,P > 0.1), although events obtained by K+/α-LTX stimulation hadsmaller amplitudes (maximum current; Imax) (K+, 35.4 ± 3.4 pA;K+/α-LTX, 18.4 ± 2.5 pA, P < 0.05, Mann-Whitney U-test) andincreased durations (width at half-height; t1/2) (K+, 108 ± 12 µs;K+/α-LTX, 178 ± 29 µs, P < 0.05).

Simple and complex amperometric eventsThe shapes of 80–85% of amperometric events induced by eithersecretagogue closely resembled those previously reported fordopamine and serotonin release during SSV exocytosis10,11,13,14

(Fig. 1c). Such peaks, which we refer to as ‘simple’ events, consistedof a single rising and a single falling phase (Fig. 1e; Methods). Ourrandom walk simulations of dopamine release indicated that theminimum fusion pore diameter consistent with the flux ofdopamine observed in simple events was 1.5–3.5 nm (Methods).Surprisingly, this calculated diameter of the fusion pore indopaminergic SSVs is nearly identical to estimates of the initial

Departments of 1Neurology and 2Psychiatry, Black 305, 650 West 168th St, Columbia University, New York, New York 10032, USA. 3Department of Neuroscience,New York State Psychiatric Institute, 722 West 168th Street, New York, New York 10032, USA. Correspondence should be addressed to D.S. ([email protected]).

Published online 29 February 2004; doi:10.1038/nn1205

Dopamine neurons release transmitter via a flickeringfusion poreRoland G W Staal, Eugene V Mosharov & David Sulzer1–3

A key question in understanding mechanisms of neurotransmitter release is whether the fusion pore of a synaptic vesicleregulates the amount of transmitter released during exocytosis. We measured dopamine release from small synaptic vesicles ofrat cultured ventral midbrain neurons using carbon fiber amperometry. Our data indicate that small synaptic vesicle fusion poresflicker either once or multiple times in rapid succession, with each flicker releasing ∼ 25–30% of vesicular dopamine. Theincidence of events with multiple flickers was reciprocally regulated by phorbol esters and staurosporine. Thus, dopamineneurons regulate the amount of neurotransmitter released by small synaptic vesicles by controlling the number of fusion poreflickers per exocytotic event. This mode of exocytosis is a potential mechanism whereby neurons can rapidly reuse vesicleswithout undergoing the comparatively slow process of recycling.


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

fusion pore formed by large dense-core vesicles (LDCVs) ineosinophils, neutrophils and chromaffin cells15–17 even though thevolume of LDCVs is 3–4 orders of magnitude larger.

The remaining events, which we refer to as ‘complex’ events (Fig. 1d), contained multiple, well-defined rising and falling phases(Fig. 1f; see Methods for flicker detection protocol). Complex eventsconsisted of 2–5 ‘flickers’ that occurred at a mean frequency of ∼ 4 kHz(Table 1) and decreased in amplitude from the first to the last flicker(Table 2 and Fig. 2a). Complex events also showed significantly longerdurations and released a greater number of molecules than simpleevents (P < 0.05; Fig. 1g,h and Table 1). As durations of consecutiveflickers did not increase (Table 2), both the distance between the siteof release and the recording electrode, and the diameter and opentime of the fusion pore, were apparently unchanged.

Pharmacological regulation of eventsPrevious studies have suggested that phorbol esters can increase thenumber of secretory events via activation of Munc-13 (refs. 18,19). Inaddition, they can enhance the calcium sensitivity of transmitterrelease20,21 and modulate the kinetics of fusion pore formation22,23

via protein kinase C (PKC). The nonspecific kinase inhibitor stau-rosporine is reported to promote kiss-and-run exocytosis, possiblythrough inhibition of PKC6,24.

We found that phorbol 12,13-dibutyrate (PDBU; 3 µM,15–30 min) increased the total number of exocytotic events per stim-ulus (Table 1 legend), consistent with the ability of phorbol esters toincrease the size of the readily releasable pool of SSVs17,18,25. In thepresence of staurosporine (5 µM, 15–30 min), fewer amperometricevents were recorded upon stimulation with K+/α–LTX, and noevents were detected when K+ alone was applied. PDBU decreased the


Figure 1 Dopamine release from axonal varicosities of rat ventral midbrain dopamine neurons. (a,b) Representative segment of current trace showingdopamine release from neurons stimulated with K+ alone (a) or with K+/α-LTX (b). The stimulus was given earlier in a portion of the trace that has beenomitted because of the paucity of events. (c,d) Representative examples of simple events (c) and complex events (d). Simple events each have a singlerising and falling slope, whereas complex events have multiple flickers, each with distinct rising and falling phases (Methods). (e,f) The upper panels showexamples of amperometric current traces; the lower panels show the first derivative (dI/dt) of the currents. In the amperometric traces, the meanbackground current is indicated by a solid line (upper panel). To be considered an ‘event,’ the dI/dt must cross a 4.5 × r.m.s. threshold (solid line, lowerpanel). (e) Events with derivatives that cross the 3.0 × r.m.s. threshold (dotted line) only once in a rising trajectory are ‘simple’. (f) Events that cross the3.0 x r.m.s. threshold multiple times are ‘complex’. The corresponding flickers (1–3) are indicated in the current trace. (g,h) Histograms of simple versuscomplex event characteristics obtained from amperometric recordings after K+/ α-LTX stimulation (n = 532 simple events and n = 130 complex eventsfrom eight sites; see Table 1 for statistics).

Figure 2 Amplitudes of flickers within complex events. (a) The amplitude(mean ± s.e.m.) of each flicker (Imax) is plotted against the flicker n. Onlyflickers from complex events with an Imax > 20 pA are shown because of theincreased contribution of background noise to smaller flickers (∼ 3 pAr.m.s., hatched box; n = 21 complex events for untreated, 23 for PDBU-treated and 9 for staurosporine-treated; all were K+/α-LTX-stimulatedevents). (b) Dependence of flicker Imax on the fraction of totalneurotransmitter released per pore opening. The mean values of theexperimental data from untreated neurons (solid line, gray circles) yielded aslope of –4.7 ± 1.7 pA/flicker (mean ± s.d.), corresponding to the decreasepredicted for release of 26 ± 9% of transmitter content (mean ± s.d.; r2 = 0.97).

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

incidence of complex events from 15% to 10% (K+ alone) and from20% to 6% (K+/α–LTX, Fig. 3), whereas staurosporine doubled theincidence of complex events to 40% (K+/α–LTX). The number offlickers per complex event did not significantly change according totreatment or secretagogue used (Table 1).

DISCUSSIONOver a decade ago, the amperometric detection of catecholaminesreleased during exocytosis was first applied to chromaffin cells thatcontain LDCVs26. Subsequent studies used amperometry to detectneurotransmitter release from SSVs in neuronal cell bodies13,14,27,28

and central synaptic terminals10. Our measurements of dopaminereleased from terminals of cultured midbrain neurons show at leasttwo types of amperometric events, which we labeled simple and com-plex. In contrast to simple events, which consisted of single ampero-metric peaks, complex events comprised 2–5 flickers that decreasedsequentially in amplitude. Although it has long been remarked thatexocytosis is not always an all-or-none event29, flickers have not beenpreviously described for SSV exocytosis because other recording tech-niques do not provide sufficient time resolution to resolve flickerswithin complex events.

Mechanisms of SSV exocytosisIn Figure 4, we illustrate several possible mechanisms of exocytosisthat could produce complex events. One scenario is that two or moreSSVs may release their contents simultaneously, but at different dis-tances from the recording electrode, thus producing overlapping sim-ple events (Fig. 4a). Such an overlap would need to be wellcoordinated, as the incidence of complex events in untreated neuronswas 200-fold greater than the probability that any two simple eventswould occur randomly within the duration of a complex event(Methods). In addition, the apparent duration of the events releasedfarther from the electrode would be longer as a result of diffusionalfiltering (that is, t1/2 of flicker 1 ≠ t1/2 of flicker 2). Alternatively, an

overlap of simple events could be due to thefusion of clustered SSVs where exocytosis ofa single vesicle within a cluster would triggerthe fusion of other vesicles (Fig. 4b). In thiscase, the average Imax for flickers within com-plex events would be identical regardless oforder (that is, Imax of flicker 1 = Imax offlicker 2). In a variation of this model, exocy-tosis could occur via the fusion of an SSV toanother SSV that had already fused with theplasma membrane (Fig. 4c); in this case, laterflickers would show an increased duration asa result of diffusional filtering and the dilu-tion of neurotransmitter inside the two fusedvesicles. As has been demonstrated forLDCVs30, multiple SSVs might fuse witheach other before exocytosis, allowing thevesicle contents to mix (compound exocyto-sis). The resulting vesicle would either pro-duce a single peak or, if the SSV matrices orcores remained intact after SSVs fusion, mul-tiple peaks with similar amplitudes and dura-tions (Fig. 4d). All of the above hypotheseswere contradicted by our experimentallydetermined Imax and t1/2 values of complexevent flickers, which decreased sequentially(Fig. 2a and Table 2).

In contrast to the above models of multiple SSV fusion, complexevents could result from the fusion of a single SSV that forms a rap-idly flickering fusion pore (Fig. 4e). In this case, the Imax of each sub-sequent flicker would decrease because of the reduced SSVneurotransmitter concentration after each pore opening (Imax offlicker 1 > Imax of flicker 2). This hypothesis was supported by theobserved decrease in the average Imax of successive flickers withincomplex events (Fig. 2a and Table 2), which corresponded to therelease of ∼ 25–30% of the SSV neurotransmitter content per flickerfor neurons in all treatment groups (Fig. 2b). The slight decrease inthe t1/2 of flickers within complex events (Table 2) is consistent withrandom walk simulations, suggesting that this decrease is due to thefiltering applied to the data (data not shown).

Although some studies show transient flickering of the fusion porein LDCVs31,32, the duration of SSV flickers observed here was consid-erably shorter than reported for LDCVs (100–150 µs vs.


Table 1 Characteristics of simple and complex events elicited by K+ and K+/α-LTX

Simple events Number of Imax (pA) t1/2 (µs)dopamine molecules

K+ control 10,400 ± 1,000 35.2 ± 3.3 92 ± 6

PDBU 8,300 ± 1,100 30.3 ± 2.9 76 ± 6*

K+/α-LTX control 10,200 ± 1,500 17.7 ± 2.3 156 ± 30

Staurosporine 14,000 ± 1,900 26.1 ± 5.1 164 ± 38

PDBU 8,200 ± 1,000 25.7 ± 4.4 91 ± 4

Complex events

Number of Imax (pA) t1/2complex (µs) Inter-flicker Flickers/event

dopamine molecules interval (µs)

K+ control 23,700 ± 8,000‡ 26.9 ± 4.8 380 ± 43‡ 240 ± 27 2.06 ± 0.06

PDBU 25,800 ± 6,400*‡ 26.4 ± 2.2 461 ± 83‡ 293 ± 71 2.38 ± 0.22

K+/α-LTX control 18,200 ± 4,900‡ 20.2 ± 4.2 507 ± 48‡ 261 ± 46 2.32 ± 0.03

Staurosporine 26,400 ± 5,500‡ 33.2 ± 6.4 573 ± 61‡ 322 ± 24* 2.51 ± 0.19

PDBU 14,200 ± 2,300‡ 19.1 ± 3.2 291 ± 41**†‡ 165 ± 28** 2.19 ± 0.12

For K+-stimulated neurons, the data are presented from untreated neurons (8 sites, 13 ± 4 events per site; mean ±s.e.m.) and PDBU-treated neurons (7 sites, 77 ± 56 events per site). No events were detected from neurons treatedwith staurosporine when K+ alone was used as a secretagogue. For K+/α-LTX-stimulated neurons, the data arepresented from untreated (8 sites, 85 ± 50 events per site), staurosporine-treated (11 sites, 15 ± 7 events per site)and PDBU-treated neurons (9 sites, 94 ± 41 events per site). Data in the table are shown as mean ± s.e.m. for eachrecording site: *P < 0.05 and **P < 0.005 for PDBU or staurosporine versus control, †P < 0.05 for PDBU versusstaurosporine, ‡P < 0.05 for complex versus simple events by two-way ANOVA . Inter-flicker intervals are shown forcomplex events with Imax > 20 pA.

K+ K+/ α-LTX0

20

40 *

*

ControlPDBUStaurosporine

* 16104

48499

130662

52843

64160

#

Co

mp

lex

even

ts (

%)

Figure 3 Pharmacological regulation of the incidence of complex events.The percentages of complex events are shown for each experimentalcondition. The numbers of complex events/total number of events areindicted within the bars. *P < 0.05 vs. control, #P < 0.005 vs. PDBU, bychi-square test. No events were detected after K+ stimulation ofstaurosporine-treated neurons.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

10,000–500,000 µs, respectively). We also found that they occurred ata much higher frequency than in LDCVs (4,000 Hz vs. 170 Hz)32 andreleased a far greater fraction of the vesicle’s neurotransmitter(25–30% vs. <1%)32.

Relevance to vesicle recyclingFor all treatment groups, the number of molecules in complexamperometric events was significantlygreater than in simple events (1.7–3.1 fold; P> 0.05; Table 1 and Fig. 1g). Interestingly, thefirst flicker within complex events was simi-lar to simple events in amplitude (18.4 ± 2.2vs. 17.7 ± 2.3 pA), number of molecules(10,800 ± 800 vs. 10,200 ± 1,500 molecules)and t1/2 (129 ± 13 vs. 156 ± 30 µs; Tables 1and 2). These data suggest that simple eventsmay generally represent neurotransmitterrelease through short-lived pores that are notopen long enough to release an SSV’s entireneurotransmitter content, implying kiss-and-run exocytosis. Complex events appearto be exocytotic events in which the fusionpore either flickers (opens and closes) orfluctuates (enlarges and constricts) severaltimes in rapid succession, resulting in therelease of a larger fraction of an SSV’s neuro-

transmitter content. Whereas full fusion of SSVs has been clearlydemonstrated33, data from the present study and others7,8 suggestthat some synapses primarily use kiss-and-run exocytosis. The presy-naptic terminals of midbrain dopamine neurons contain a relativelysmall number of SSVs34 with an apparently high probability of exocy-tosis for any given vesicle35. Fusion pore flickering and kiss-and-runexocytosis may be particularly important for such synapses to preventthe loss of SSVs during full fusion and the relatively slow process ofendocytosis and recycling.

Relevance for dopamine signalingIn contrast to fast-acting neurotransmitter systems with well-definedpre- and postsynaptic structures, dopamine neurons form ‘social’synapses that often lack well-defined active zones35–38. The dopaminereuptake transporters are located some distance away from the releasesite, enabling the neurotransmitter to diffuse and act on receptorsseveral microns away37,38. Modeling of the diffusion of dopaminereleased at social synapses in the striatum suggests that the number ofmolecules released per exocytotic event (quantal size) determineshow far dopamine diffuses and how many receptors are acti-vated37–39. Because the number of molecules released during complexevents is greater than in simple events, dopamine will diffuse througha larger volume (Fig. 5a,b), activating more receptors for a longerduration (Fig. 5c).

In summary, our data provide evidence that second-messenger sys-tems modulate the mode of SSV exocytosis by regulating the numberof fusion pore flickers per exocytotic event. Fusion pore flickeringmay provide neurons with a means to recycle vesicles more efficientlyand to control quantal size, thus regulating the spillover of neuro-transmitter from a social synapse.


Table 2 Characteristics of complex event flickers

Flicker Number of Imax (pA) t1/2 (µs)dopamine molecules

1st 10,800 ± 800 18.4 ± 2.2 129 ± 13

2nd 7,500 ± 1,200* 14.2 ± 1.5* 110 ± 9

3rd 6,000 ± 2,400* 8.4 ± 1.9†* 91 ± 8*

Data are for flickers 1–3 from complex events with Imax > 20 pA in untreated neuronsstimulated with K+/a-LTX (mean ± s.e.m.). *P < 0.05 compared with 1st flicker and †P < 0.05 compared with 2nd flicker.

Figure 4 Mechanisms that may explain complex events (left column) andpredicted averaged amperometric event shape (right column). (a) Overlap ofsimple events with spatial separation of release sites. Either vesicle couldrelease first, resulting in either of the scenarios depicted on the right. Thecarbon fiber electrode (CFE) is 100 times wider than the diameter of theSSVs, and has been omitted in b–e. (b) Exocytosis of clustered vesicles. (c) Vesicle fusion with another vesicle that has already fused to themembrane. (d) If the vesicular matrices remain intact, compoundexocytosis would occur without mixing of vesicular contents. If the matricesare labile, the contents would mix, resulting in a single amperometric peak.(e) Transmitter release from a single SSV via a flickering fusion pore.

Figure 5 Simulated dopamine spillover in the striatum. (a) Diffusion profiles of dopamine releasedfrom an SSV following one or three openings of the fusion pore (10,000 and 20,000 dopaminemolecules released, respectively; Table 1) at 4 µm from the release site as determined by randomwalk simulations. (b) Maximum dopamine concentrations reached at various distances from therelease site. The straight dashed line at 10 nM indicates the EC50 for the activation of dopaminereceptors (1–20 nM)48. (c) The duration that receptors are exposed to dopamine levels > 10 nM.Dopamine spillover from a single flicker activates receptors in a 73,500 µm3 sphere (26 µm radius),with nearby receptors activated for 480 ms. If the SSV flickers three times, then the volume of thesphere is 1.7-fold larger (∼ 125,000 µm3, 31 µm radius) and the duration of receptor activation is620 ms. These calculations are based on vesicular dopamine concentrations in L-DOPA pretreatedneurons, which are 3–5 fold higher than in untreated cultures10.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

METHODSRat ventral midbrain neuronal cultures. Postnatally derived ventral midbrainneurons were cultured as previously described40. Neurons were preincubatedwith 100 µM L-DOPA for 30 min prior to recording11. The secretagogues (92 mM NaCl, 40 mM KCl, 10 mM HEPES, 1 mM Na2HPO4, 2 mM MgCl2,1.2 mM CaCl2, ∼ 300 mosm and pH 7.4 or 52 mM NaCl, 80 mM KCl, 10 mMHEPES, 1 mM Na2HPO4, 2 mM MgCl2, 1.2 mM CaCl2 and 20 nM α-LTX)were applied by local perfusion through a glass micropipette (Picospritzer,General Valve) for 6 s at 10 p.s.i. and ∼ 30 µm from the recording site.

Amperometric recordings. A 5-µm diameter carbon fiber electrode held at+700 mV was positioned over a potential release site (Newport micromanipu-lator MX300R) and lowered until the tissue was slightly depressed10. At thispotential, dopamine is oxidized, resulting in the donation of two electrons tothe electrode. Thus, the number of molecules reaching the electrode can beestimated from the current38. The current was filtered using a 4-pole 10 kHzBessel filter built into an Axopatch 200A amplifier (Axon Instruments), sam-pled at 100 kHz (PCI-6052E, National Instruments) and digitally filtered usinga binomial 10 routine (Igor Pro, Wave Metrics) with a –3 dB cut-off of∼ 15 kHz. This yielded an overall –3 dB cut-off frequency of >8 kHz and essen-tially no time distortion for the t1/2 of amperometric events with durations>50 µs. It also broadened events of shorter duration toward 50 µs41. Traceswith root mean square (r.m.s.) noise less than 3 pA r.m.s. were analyzed. Thebackground noise was normally distributed with no maxima for any frequencycomponent between 0.6 and 33 kHz. No events were recorded when theapplied voltage was adjusted to 0 mV or when the electrode was transientlylifted from an active recording site.

Peak detection and flicker analysis. Raw amperometric data were collectedand analyzed using a locally written routine in Igor Pro. The first derivativeof the current trace (dI/dt) was used to detect amperometric events. Ther.m.s. of the dI/dt noise was first measured in a segment of the trace that didnot contain peaks. Then, dI/dt was used to detect events that were 4.5-foldlarger than the r.m.s. noise. These spikes represented the total population ofamperometric events. The beginning and the maximum of each event wereat dI/dt = 0 (Fig. 1e,f). The end of an event was defined as the point when thecurrent returned to the baseline value. If there was more than one maximumwithin an event and the dI/dt of these maxima (flickers) was three-fold largerthan the r.m.s. noise, then the event was classified as ‘complex’ (Fig. 1f).Events that included a single peak with one rising and one falling phase orfor which the dI/dt of flickers was less than three times the r.m.s. noise werecategorized as ‘simple’ events. This approach was relatively conservative inidentifying flickers, but the same rules were applied for each treatment. Wefound that the same flickers were identified independently of their orderwithin a complex event.

Due to the shape of complex events, the typical t1/2 value does not accu-rately reflect the event’s duration. Thus, the duration of complex events wascalculated as:

(1)

where t(f1) and t(fn) are the times at Imax, and t1/2(f1) and t1/2(fn) are the dura-tions of the first and the last flickers of complex events. The number of mole-cules in the first flicker was estimated by subtracting the integral of thesubsequent flickers from the integral of the entire complex event. The baselinefor the subsequent flickers was estimated as a line from the beginning of thesecond event to the end of the complex event. Although this approach mayslightly overestimate the number of molecules in the first flicker as a result ofthe nonlinear decay of events, this error would be <2%.

Statistical analysis. The data in Table 1 are reported as averages of the meanvalues from each recording site42, each of which generally represents a singlepresynaptic terminal10. Data were analyzed by ANOVA of the means unlessindicated otherwise42. As reported in several studies, the numbers of mole-

(t(fn) – t(f1) + t1/2(f1) + t1/2(fn)t complex =21/2

cules (n) released during exocytotic events are distributed as a function ofvesicle volumes so that the cube roots of n result in a normal distribution(Fig. 1g; for review, see ref. 38). As previously reported10, occasional largeevents (3 of 772 in untreated cultures) were >5 standard deviations greaterthan the geometric mean of the cube roots of n and were excluded from thedata analysis in Table 1. The data in Figure 4 are nonparametric and wereanalyzed by chi-square test.

Estimation of the expected random overlap of simple events. The probabilitythat complex events resulted from the random overlap of simple events can beestimated from the exponential decay of interspike intervals38. The time con-stant was 545 ms (R2 = 0.997 from untreated cultures; Supplementary Fig. 1online). The probability of observing an interspike interval less than 0.5 ms(two overlapping events within the duration of complex events) is P =(1– e–0.5/545) × 100 = 0.09%.

Simulation of dopamine release from the vesicle. Random walk simulations38

(finite-difference model) of molecular diffusion to an amperometric (‘con-suming’) electrode was performed using Excel software (Microsoft). Duringeach time bin (tbin), the flux (J, molecules/s) of dopamine molecules from thevesicle through a fusion pore was calculated as:

(2)

where a is the area, b is the length and Rpore is the radius of a cylindricalpore43. Cv and Nv are the concentration and the number of molecules of neu-rotransmitter in the vesicle with radius Rv. D is the diffusion coefficient,which is 6.9*10–6 cm2/s for dopamine in aqueous solution44. We used elec-tron micrographs of tyrosine hydroxylase–immunolabeled cultures to deter-mine SSV diameters under the conditions used in the recordings. SSVdiameters were similar to those of previous reports11; 50.7 ± 1.4 nm (mean ±s.e.m., n = 49 vesicles in 7 terminals; data not shown). The length of thefusion pore was estimated as 7.5–15 nm, twice the membrane thickness of 5-hydroxydopamine-labeled SSVs in this preparation11. The distance betweenthe release site and electrode was varied from 50–400 nm, beyond which theamperometric currents would be too low in amplitude to be identified. Thenumber of molecules (N) encountering the surface of the amperometric elec-trode during tbin was converted to units of amperometric current (I) usingthe following formula38:

(3)

For simulations, the diameter and the length of the fusion pore as well asthe number of molecules inside the vesicle were varied until the amplitudeand duration of simulated exocytotic events (after resampling and filtering)were the same as the average t1/2 and Imax of the amperometric peaks that hadbeen recorded experimentally (Table 1). Simulated data were re-sampled at10-µs intervals and filtered using binomial 10 smoothing to mimic the exper-imental conditions.

Given that dopamine is present in high millimolar concentrations within theSSV, it is possible that dissociation of dopamine from a lumenal core is slowerthan the diffusion of free molecules through the fusion pore. Additionally, thepore length may be longer than the thickness of the plasma membrane45. Thus,the calculated pore diameters represent a minimum estimate.

Simulation of dopamine diffusion in striatum. Dopamine spillover was mod-eled using the diffusion coefficient of dopamine in the striatum,2.7*10–6 cm2/s (ref. 46). The dopamine concentration next to the fusion porewas calculated as:

I[pA/s] =N[molecules] · 106[µs]

tbin [µs] · 3.121 · 106 [molecules · s/pA]

J =a · Cv · D

b b[cm]=

π · (Rpore [cm])2 · · D[cm2/s]Nv [molecules]

43

π · (Rv[cm])3


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

(4)

where rbin is the radius of the sphere next to the fusion pore derived as rbin =

√(tbin/2D). Dopamine uptake via the dopamine transporter was assumed tofollow Michaelis-Menten kinetics, with Vmax = 4.88 µM/s and Km = 0.77 µM(ref. 47).

A tutorial on random walk and finite difference simulations and the Excelspreadsheets used for these calculations are available at our laboratory website(http://www.columbia.edu/∼ ds43/pore_RW.html).


ACKNOWLEDGMENTSWe thank Q. Al-Awqati, K. Larsen, M. Nirenberg and Y. Schmitz for critique of themanuscript, and A. Petrenko for α-latrotoxin. Supported by the National Alliancefor Research on Schizophrenia and Depression, the Lowenstein Foundation, theParkinson’s Disease Foundation, the National Institute on Drug Abuse and theNational Institute of Neurological Disorders and Stroke.


Received 23 October 2003; accepted 27 January 2004Published online at http://www.nature.com/natureneuroscience/

1. Valtorta, F., Meldolesi, J. & Fesce, R. Synaptic vesicles: is kissing a matter of com-petence? Trends Cell Biol. 11, 324–328 (2001).

2. Heuser, J.E. & Reese, T.S. Evidence for recycling of synaptic vesicle membrane dur-ing transmitter release at the frog neuromuscular junction. J. Cell Biol. 57,315–344 (1973).

3. Ceccarelli, B., Hurlbut, W.P. & Mauro, A. Turnover of transmitter and synaptic vesi-cles at the frog neuromuscular junction. J. Cell Biol. 57, 499–524 (1973).

4. Sun, J.Y., Wu, X.S. & Wu, L.G. Single and multiple vesicle fusion induce differentrates of endocytosis at a central synapse. Nature 417, 555–559 (2002).

5. Klyachko, V.A. & Jackson, M.B. Capacitance steps and fusion pores of small andlarge-dense-core vesicles in nerve terminals. Nature 418, 89–92 (2002).

6. Henkel, A.W. & Betz, W.J. Staurosporine blocks evoked release of FM1-43 but notacetylcholine from frog motor nerve terminals. J. Neurosci. 15, 8246–8258 (1995).

7. Gandhi, S.P. & Stevens, C.F. Three modes of synaptic vesicular recycling revealed bysingle-vesicle imaging. Nature 423, 607–613 (2003).

8. Aravanis, A.M., Pyle, J.L. & Tsien, R.W. Single synaptic vesicles fusing transientlyand successively without loss of identity. Nature 423, 643–647 (2003).

9. Stiles, J.R., Van Helden, D., Bartol, T.M. Jr., Salpeter, E.E. & Salpeter, M.M.Miniature endplate current rise times less than 100 microseconds from improveddual recordings can be modeled with passive acetylcholine diffusion from a synapticvesicle. Proc. Natl. Acad. Sci. USA 93, 5747–5752 (1996).

10. Pothos, E.N., Davila, V. & Sulzer, D. Presynaptic recording of quanta from midbraindopamine neurons and modulation of the quantal size. J. Neurosci. 18, 4106–4118(1998).

11. Pothos, E.N. et al. Synaptic vesicle transporter expression regulates vesicle pheno-type and quantal size. J. Neurosci. 20, 7297–7306 (2000).

12. Bittner, M.A. Alpha-latrotoxin and its receptors CIRL (latrophilin) and neurexin 1alpha mediate effects on secretion through multiple mechanisms. Biochimie 82,447–452 (2000).

13. Bruns, D. & Jahn, R. Real-time measurement of transmitter release from singlesynaptic vesicles. Nature 377, 62–65 (1995).

14. Hochstetler, S.E., Puopolo, M., Gustincich, S., Raviola, E. & Wightman, R.M. Real-time amperometric measurements of zeptomole quantities of dopamine releasedfrom neurons. Anal. Chem. 72, 489–496 (2000).

15. Hartmann, J., Scepek, S. & Lindau, M. Regulation of granule size in human andhorse eosinophils by number of fusion events among unit granules. J. Physiol.(Lond.) 483, 201–209 (1995).

16. Lollike, K., Borregaard, N. & Lindau, M. The exocytotic fusion pore of small granuleshas a conductance similar to an ion channel. J. Cell Biol. 129, 99–104 (1995).

17. Albillos, A. et al. The exocytic event in chromaffin cells revealed by patch amperom-etry. Nature 389, 509–512 (1997).

18. Betz, A. et al. Munc13-1 is a presynaptic phorbol ester receptor that enhances neu-rotransmitter release. Neuron 21, 123–136 (1998).

C[M] =N[molecules] · 6.022 · 1023[molecules/mole]

43

π · (rbin[dm])3

19. Rhee, J.S. et al. Beta phorbol ester- and diacylglycerol-induced augmentation oftransmitter release is mediated by Munc13s and not by PKCs. Cell 108, 121–133(2002).

20. Zhu, H., Hille, B. & Xu, T. Sensitization of regulated exocytosis by protein kinase C.Proc. Natl. Acad. Sci. USA 99, 17055–17059 (2002).

21. Yang, Y., Udayasankar, S., Dunning, J., Chen, P. & Gillis, K.D. A highly Ca2+-sensi-tive pool of vesicles is regulated by protein kinase C in adrenal chromaffin cells.Proc. Natl. Acad. Sci. USA 99, 17060–17065 (2002).

22. Wang, P., Wang, C.T., Bai, J., Jackson, M.B. & Chapman, E.R. Mutations in theeffector binding loops in the C2A and C2B domains of synaptotagmin I disrupt exo-cytosis in a nonadditive manner. J. Biol. Chem. 278, 47030–47037 (2003).

23. Graham, M.E., Fisher, R.J. & Burgoyne, R.D. Measurement of exocytosis by amper-ometry in adrenal chromaffin cells: effects of clostridial neurotoxins and activationof protein kinase C on fusion pore kinetics. Biochimie 82, 469–479 (2000).

24. Scepek, S., Coorssen, J.R. & Lindau, M. Fusion pore expansion in horse eosinophilsis modulated by Ca2+ and protein kinase C via distinct mechanisms. EMBO J. 17,4340–4345 (1998).

25. Gillis, K.D., Mossner, R. & Neher, E. Protein kinase C enhances exocytosis fromchromaffin cells by increasing the size of the readily releasable pool of secretorygranules. Neuron 16, 1209–1220 (1996).

26. Wightman, R.M. et al. Temporally resolved catecholamine spikes correspond to sin-gle vesicle release from individual chromaffin cells. Proc. Natl. Acad. Sci. USA 88,10754–10758 (1991).

27. Chen, G., Gavin, P.F., Luo, G. & Ewing, A.G. Observation and quantitation of exocy-tosis from the cell body of a fully developed neuron in Planorbis corneus. J. Neurosci. 15, 7747–7755 (1995).

28. Jaffe, E.H., Marty, A., Schulte, A. & Chow, R.H. Extrasynaptic vesicular transmitterrelease from the somata of substantia nigra neurons in rat midbrain slices. J. Neurosci. 18, 3548–3553 (1998).

29. Girod, R., Correges, P., Jacquet, J. & Dunant, Y. Space and time characteristics oftransmitter release at the nerve-electroplaque junction of Torpedo. J. Physiol. 471,129–157 (1993).

30. Hafez, I., Stolpe, A. & Lindau, M. Compound exocytosis and cumulative fusion ineosinophils. J. Biol. Chem. 278, 44921–44928 (2003).

31. Alvarez de Toledo, G., Fernandez-Chacon, R. & Fernandez, J.M. Release of secretoryproducts during transient vesicle fusion. Nature 363, 554–558 (1993).

32. Zhou, Z., Misler, S. & Chow, R.H. Rapid fluctuations in transmitter release from sin-gle vesicles in bovine adrenal chromaffin cells. Biophys. J. 70, 1543–1552 (1996).

33. Cremona, O. & De Camilli, P. Synaptic vesicle endocytosis. Curr. Opin. Neurobiol. 7,323 (1997).

34. Pickel, V.M., Nirenberg, M.J. & Milner, T.A. Ultrastructural view of central cate-cholaminergic transmission: immunocytochemical localization of synthesizingenzymes, transporters and receptors. J. Neurocytol. 25, 843–856 (1996).

35. Garris, P.A., Ciolkowski, E.L., Pastore, P. & Wightman, R.M. Efflux of dopamine fromthe synaptic cleft in the nucleus accumbens of the rat brain. J. Neurosci. 14,6084–6093 (1994).

36. Schmitz, Y., Benoit-Marand, M., Gonon, F. & Sulzer, D. Presynaptic plasticity ofdopaminergic neurotransmission. J. Neurochem. 87, 273–289 (2003).

37. Gonon, F. et al. Geometry and kinetics of dopaminergic transmission in the rat stria-tum and in mice lacking the dopamine transporter. Prog. Brain Res. 125, 291–302(2000).

38. Sulzer, D. & Pothos, E.N. Presynaptic mechanisms that regulate quantal size. Rev.Neurosci. 11, 159–212 (2000).

39. Cragg, S.J., Nicholson, C., Kume-Kick, J., Tao, L. & Rice, M.E. Dopamine-mediatedvolume transmission in midbrain is regulated by distinct extracellular geometry anduptake. J. Neurophysiol. 85, 1761–1771 (2001).

40. Pothos, E.N., Przedborski, S., Davila, V., Schmitz, Y. & Sulzer, D. D2-Like dopamineautoreceptor activation reduces quantal size in PC12 cells. J. Neurosci. 18,5575–5585 (1998).

41. Colquhoun, D. & Sigworth, F.J. Fitting and statistical analysis of single-channelrecords. in Single-Channel Recording (eds. Sakmann, B. & Neher, E.) 483–587(Plenum, New York, 1995).

42. Colliver, T., Hess, E., Pothos, E.N., Sulzer, D. & Ewing, A.G. Quantitative and statis-tical analysis of the shape of amperometric spikes recorded from two populations ofcells. J. Neurochem. 74, 1086–1097 (1999).

43. Berg, H.C. Random Walks in Biology 152 (Princeton Univ. Press, Princeton, 1983).44. Nicholson, C. Diffusion of ions and macromolecules in brain tissue. in Monitoring

Molecules in Neuroscience Vol. 8 (eds. Rollema, H., Abercrombie, E., Sulzer, D. &Zackheim, J.) 71–73 (Rutgers Press, Newark, New Jersey, 1999).

45. Jeremic, A., Kelly, M., Choo, S.J., Stromer, M.H. & Jena, B.P. Reconstituted fusionpore. Biophys. J. 85, 2035–2043 (2003).

46. Rice, M.E. et al. Direct monitoring of dopamine and 5-HT release in substantia nigraand ventral tegmental area in vitro. Exp. Brain Res. 100, 395–406 (1994).

47. Schmitz, Y., Lee, C.J., Schmauss, C., Gonon, F. & Sulzer, D. Amphetamine distortssynaptic dopamine overflow: effects on D2 autoreceptors, transporters, and synapticvesicle stores. J. Neurosci. 21, 5916–5924 (2001).

48. Levant, B. The D3 dopamine receptor: neurobiology and potential clinical relevance.Pharmacol. Rev. 49, 231–252 (1997).


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

The olfactory bulb is one of the few structures in the adult forebrainin which there is a continuous supply of newborn neurons1,2. Theneural progenitors, which originate from stem cells located in thesubventricular zone (SVZ) of the lateral ventricles, follow an intricatemigration path before reaching their final position in the olfactorybulb. First, they move tangentially, in chains, along the entire extent ofthe rostral migratory stream (RMS); once in the bulb, they turn tomove radially out of the RMS into the outer layers, where they differ-entiate into inhibitory interneurons1,2.

Despite increasing knowledge about the origin, proliferation andtangential migration of neuroblasts, how they achieve their radialmigration to integrate functionally into the bulbar circuitry remainselusive. Notably, radial glia, which are central to axonal guidance andradial migration during development, are no longer present in the adultolfactory bulb2. This implies that neuroblasts arriving in the rostralextension of the RMS of the adult forebrain follow unique migratorypathways quite distinct from those seen at perinatal stages. A recentreport has provided evidence that the olfactory bulb–derived extracel-lular matrix (ECM) molecule reelin affects detachment of neuroblastsfrom chains. However, the cues instigating the processes that occuronce the neuroblasts reach the olfactory bulb3—halting tangentialmigration, initiating detachment of neuroblasts and facilitating theirradial migration—have not yet been characterized. The effect of sen-sory experience on these processes also still needs to be examined.

Here we investigated the possibility that reoriented migration ofneuroblasts in the core of the olfactory bulb is orchestrated by a gradi-ent of extracellular cues surrounding the RMS within the olfactorybulb. We found that the expression pattern of the ECM moleculetenascin-R is potentially compatible with such a functional role:tenascin-R is detectable exclusively in the deep layers of the olfactory

bulb, around the most anterior extension of the RMS (RMSOB), butnot within the RMS itself.

Tenascin-R is a member of the tenascin gene family and contains acysteine-rich amino terminal region, epidermal growth factor–likedomains, fibronectin type III homologous repeats and a domainhomologous to fibrinogen4. Tenascin-R appears to be restricted tothe CNS and is expressed by differentiating oligodendrocytes andsome inhibitory interneurons at late embryonic stages4,5. The func-tions of tenascin-R are manifold: the protein binds to voltage-dependent Na+ channels6,7, and the conduction velocity of actionpotentials is reduced in tenascin-R-deficient mice8. In addition,tenascin-R is an important constituent of perineuronal nets sur-rounding many, but not all, inhibitory interneurons9 and organiza-tion of perineuronal nets is perturbed in tenascin-R-deficient mice.Tenascin-R-deficient mice have alterations in the organization ofperisomatic synapses10 as well as synaptic transmission and plasticityin the CA1 region of the hippocampus11.

Combining in vitro and in vivo approaches, we identified a previ-ously unknown function of tenascin-R in the adult olfactory bulb.We show here that (i) tenascin-R is a key player in directing neurob-lasts into their prospective target area, (ii) the extent of its expres-sion correlates strongly with olfactory sensory activity and (iii)grafting tenascin-R-secreting cells into regions that do not receiveprogenitor neurons reroutes migrating neuroblasts to these areas.Thus, the activity-dependent recruitment of neuroblasts bytenascin-R represents a fundamental mechanism through whichneurogenesis in the adult olfactory bulb is regulated and adapted tothe level of sensory input. The tenascin-R signaling pathway mightalso provide a new approach to cell replacement therapies based onrerouting migrating cells.

1Laboratory of Perception and Memory, CNRS URA 2182, Pasteur Institute, 25 rue du Dr. Roux, 75015 Paris Cedex, France. 2Zentrum für Molekulare Neurobiologie,Universität Hamburg, Martinistrasse 52, D-20246 Hamburg, Germany. 3These authors contributed equally to this study. Correspondence should be addressed toP.M.L. ([email protected]).


Tenascin-R mediates activity-dependent recruitmentof neuroblasts in the adult mouse forebrainArmen Saghatelyan1,3, Antoine de Chevigny1,3, Melitta Schachner2 & Pierre-Marie Lledo1

Neuroblasts arising in the adult forebrain that travel to the olfactory bulb use two modes of migration: tangentially, along therostral migratory stream, and radially, in the core of the olfactory bulb where they start to ascend to the outer layers. Although themechanisms of tangential migration have been extensively studied, the factors controlling radial migration remain unexplored.Here we report that the extracellular matrix glycoprotein tenascin-R, expressed in the adult mouse olfactory bulb, initiates boththe detachment of neuroblasts from chains and their radial migration. Expression of tenascin-R is activity dependent, as it ismarkedly reduced by odor deprivation. Furthermore, grafting of tenascin-R-transfected cells into non-neurogenic regions reroutesmigrating neuroblasts toward these regions. The identification of an extracellular microenvironment capable of directingmigrating neuroblasts provides insights into the mechanisms regulating radial migration in the adult olfactory bulb and offerspromising therapeutic venues for brain repair.


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

the deep layers of the olfactory bulb surrounding the RMS, and notwithin it, indicates that tenascin-R may be involved in recruiting new-born neurons to the adult olfactory bulb.

Fewer newborn cells in tenascin-R-deficient olfactory bulbTo investigate the protein’s potential role in bulbar neurogenesis, wefirst used tenascin-R-deficient mice. We gave adult mutant and wild-type mice four pulses of BrdU, a marker for DNA synthesis, andprocessed their brains for BrdU immunohistochemistry 21 d later(Fig. 2a–d). Most of the BrdU+ nuclei were found scattered through-

Figure 1 Immunohistological detection oftenascin-R in the SVZ-OB pathway of adultmice. (a) Low-magnification image of SVZ-OBpathway immunostained for tenascin-R (TNR;red) and PSA-NCAM (green). (b–d) Sagittalsections of SVZ (b) and RMS (c) and coronalsection of olfactory bulb (OB; d) stained forTNR (left) and PSA-NCAM (right). Note absenceof staining for TNR in the SVZ, RMS andRMSOB and its presence in the GCL and IPL. (e) High-magnification images of TNR and PSA-NCAM staining in the olfactory bulb. Scalebars: a, 500 µm; b,c, 50 µm; d, 200 µm; e,100 µm. AOB, accessory olfactory bulb; CC,corpus callosum; CTX, cortex; EPL, externalplexiform layer; GCL, granule cell layer; IPL,internal plexiform layer; LV, lateral ventricle;RMS, rostral migratory stream; RMSOB, rostralmigratory stream of the olfactory bulb; ST,striatum; SVZ, subventricular zone.

A R T I C L E S

RESULTSTenascin-R expression in adult olfactory bulbTo examine the expression pattern of tenascin-R in the adult mousesubventricular zone–olfactory bulb (SVZ-OB) system, we combinedimmunofluorescence labeling with confocal microscopy. At variousmagnifications, scanning through the successive sections from theSVZ to the olfactory bulb (Fig. 1a–d) showed strong tenascin-R-positive staining that was restricted to the granule cell and internalplexiform layers of the olfactory bulb (Fig. 1d,e). The SVZ and RMS,identified using antibodies against the polysialylated form of the neu-ral cell adhesion molecule (PSA-NCAM), a marker for immatureneural cells (Fig. 1a–c), were devoid of tenascin-R labeling in all cases.We did not observe staining for tenascin-R in the olfactory bulb oftenascin-R-deficient mice (data not shown). The immunostainingpattern was consistent with previous in situ hybridization data show-ing that granule cells are the principal source of tenascin-R synthesisin the adult olfactory bulb5. The pattern of expression exclusively in


Figure 2 Reduced density of newborn cells in the olfactory bulb oftenascin-R-deficient mice at 21 d after BrdU injection. (a) BrdU+ nuclei inthe GCL of control (left) and tenascin-R-deficient mutant mice (right)showed a pronounced reduction in the number of newborn cells in themutant. (b) Mean density of newborn granule cells in control (+/+) andtenascin-R-deficient mice (–/–). (c) BrdU+ nuclei in the GL of control (left)and tenascin-R-deficient mice (right). (d) Mean density of newbornperiglomerular cells in control (+/+) and mutant (–/–) mice. **, P < 0.01.(e) Confocal 3D reconstruction of BrdU+ cells (red) in control (left) andtenascin-R-deficient mice (right) stained for the neuronal marker NeuN(green). Reconstructed orthogonal projections are presented as viewed inthe x-z (top) and y-z (right) planes. (f) Percentage of BrdU+ cells double-labeled with NeuN in control (+/+) and tenascin-R-deficient mice (–/–). (g) Double immunostaining for BrdU (red) and the astrocytic marker GFAP(green) in the GCL of control (left) and tenascin-R-deficient mice (right).(h) Percentage of BrdU+ cells double-labeled with GFAP in control andtenascin-R-deficient mice. In total, 940 (from three control mice) and 972(from four tenascin-R-deficient mice) randomly chosen BrdU+ cells wereinspected for NeuN immunostaining. Similarly, 836 and 1292 BrdU+ cellswere checked for GFAP immunopositivity. GCL, granule cell layer; GL,glomerular layer. Scale bars: a,b, 100 µm; e, 10 µm; g, 50 µm. Values inhistograms are means ± s.e.m.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

out the granule cell layer (Fig. 2a). The mean density of BrdU+ nucleiin the granule cell layer was significantly lower in mutants than incontrol mice (241 ± 16 versus 421 ± 42 cells/mm2, respectively;n = 4 for each genotype, P < 0.01; Fig. 2b). A similar effect was alsoseen in the glomerular layer (90 ± 12 versus 148 ± 2 cells/mm2,P < 0.01; Fig. 2c,d).

To test whether the 40% reduction seen in tenascin-R-deficientmice was accompanied by alterations in the fate of newborn cells, wedetermined the proportion of cells double-labeled for BrdU andeither the neuronal marker NeuN (Fig. 2e,f) or the glial marker GFAP(Fig. 2g,h). Orthogonal projections through three-dimensionally(3D) reconstructed BrdU+ cells showed that, in both genotypes,about 80% of newborn cells were neurons (n = 4; Fig. 2f). Similarly,although the numbers of BrdU+ GFAP+ cells were very small (only∼ 1% of all BrdU+ cells), no difference was seen between genotypes (n = 4; Fig. 2h). Thus, the reduction in the density of BrdU+ cells wassimilar in the different layers of the mutant olfactory bulb but the fateof newly generated cells remained unchanged.

Altered migration in tenascin-R-deficient miceThe reduced number of newborn cells in the olfactory bulb couldresult from decreased cell proliferation and rate of tangential migra-tion, altered chain organization, distorted radial migration and/orreduced survival of newborn neurons. To assess proliferation, we

quantified the number of mitotically active cells in the SVZ and theRMS 4 h after a BrdU injection. The distribution of BrdU+ cells in theSVZ-OB pathway did not differ between tenascin-R-deficient miceand control mice (Fig. 3a–c). The densities of BrdU+ cells in the SVZ(Fig. 3b,d) and RMS (Fig. 3c,d) were undistinguishable between thetwo genotypes (2,414 ± 156 versus 2,721 ± 90 cells/mm2 in the SVZand 2,992 ± 221 versus 3,298 ± 138 cells/mm2 in the RMS for con-trol and mutant mice, respectively; n = 3, P > 0.05).

To assess whether tenascin-R deficiency affects tangential migra-tion, we first examined the cytoarchitecture of the SVZ and RMS. Thechains of neuroblasts visualized by PSA-NCAM immunostaining inwhole-mount preparations of the SVZ were similar in tenascin-R-deficient and control mice (Fig. 4a). At both low and high magnifica-tions, an extensive network of tangential pathways was readily visible.Most of the chains in this network formed a longitudinal array thatwas not altered in tenascin-R-deficient mice. Similarly, the organiza-tion of PSA-NCAM+ chains along the migratory pathway to the olfac-tory bulb was the same in both genotypes (Fig. 4b). To characterizethe distribution of newborn cells in these chains, mice were given aBrdU injection and were killed 2 h later. In both genotypes, doublelabeling for PSA-NCAM and BrdU showed dividing precursor cellsintegrated in chains along the entire pathway (Fig. 4b,c), thus demon-strating that the organization of neuroblast chains is unaffected in theabsence of tenascin-R.

We then cultured SVZ explants on Matrigel to assess the rate of neu-roblast migration. This technique has previously been adapted for thestudy of tangential cell migration in vitro3,12,13. When SVZ explantsfrom postnatal day 7 (P7) control and tenascin-R-deficient mice werecultured for 20 h, an extensive network of chains formed around theexplants (Fig. 4d). There was no difference in the general organizationof the network of chains (Fig. 4d,e) or in their length (210 ± 12 µmversus 209 ± 11 µm, for control and mutant mice, respectively,12 explants from 4 control mice and 8 explants from 3 mutants, P >0.05; Fig. 4f). Altogether, these results demonstrate that tenascin-R isnot involved in proliferation and tangential migration of neuroblasts.This is consistent with our immunohistological observations thattenascin-R is not detectable in the SVZ and RMS. By contrast, highexpression of tenascin-R in the adult olfactory bulb might have a roleeither in the recruitment of neuroblasts from the RMS to the olfactorybulb or in their survival within the olfactory bulb.

To determine whether granule cells in the olfactory bulb oftenascin-R-deficient mice might be dying at a higher rate than in con-trols, we performed terminal deoxynucleotidyl transferase–mediatedbiotinylated UTP nick-end labeling (TUNEL) to evaluate the extentof apoptosis and quantified TUNEL+ cells in both the granule celllayer and the RMSOB (Fig. 5a,b). Numbers of TUNEL+ cells did notdiffer between genotypes (3.4 ± 0.4 versus 3.3 ± 0.4 cells per slice inthe granule cell layer and 1.2 ± 0.3 versus 1.7 ± 0.5 in the RMSOB ofcontrol and mutant mice, respectively; n = 3, P > 0.05; Fig. 5b). Thecombined data indicate that the reduced density of BrdU+ cells in themutant olfactory bulb may result from altered radial migration. If thiswere the case, the reduced density of newborn cells in the olfactorybulb should be accompanied by an accumulation of neuroblasts inthe RMSOB. We therefore examined whether BrdU+ cells wouldappear ‘trapped’ in the RMSOB of tenascin-R-deficient mice. Micewere killed 2 d after BrdU injection, and BrdU+ cells were quantifiedthroughout the entire SVZ-OB pathway. Newborn cells accumulatedexclusively in the RMSOB of mutant mice (Fig. 5c,d), but not in theRMS or the SVZ (Fig. 5d), confirming that tenascin-R does not affectproliferation and tangential migration. The accumulation of neurob-lasts in the RMSOB of tenascin-R-deficient mice was greater, to a


Figure 3 Normal proliferation in the SVZ and RMS of tenascin-R-deficientmice. (a) BrdU immunostaining in sagittal sections through the forebrain ofcontrol (+/+, left) and tenascin-R-deficient mice (–/–, right) showing thedistribution of BrdU+ cells in the SVZ-RMS pathway. (b,c) High-magnification images of the SVZ (b) and RMS (c) show similar numbers ofmitotically active cells. (d) Quantification of BrdU+ nuclei in the SVZ andthe RMS in control (+/+) and tenascin-R-deficient mice (–/–). CC, corpuscallosum; LV, lateral ventricle; ST, striatum. Data are presented as means ±s.e.m. Scale bars: a, 500 µm; b,c, 100 µm.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

highly significant degree, than for controls (3,646 ± 97 and 2,470 ±214 cells/mm2, respectively; n = 4; P < 0.01; Fig. 5d) and was accom-panied by a reduced density of BrdU+ cells in the olfactory bulb (Fig. 5e,f). The greatest decreases were measured in the external plexiform layer (18 ± 0.8 versus 29.4 ± 4.7 BrdU+ cells/mm2 formutant and control mice, respectively; P < 0.05) and the glomerularlayer (75.1 ± 11.9 versus 104.7 ± 15.3 BrdU+ cells/mm2; P < 0.05;Fig. 5e). A lesser, nonsignificant decrease occurred in the granule celllayer (20.1 ± 2.8 versus 24 ± 2 BrdU+ cells/mm2; P > 0.05).

To estimate whether the excess of BrdU+ cells in the RMSOB oftenascin-R-deficient mice is due to the accumulation of cells that havenot migrated further, we compared the number of BrdU+ cells in theRMSOB and the olfactory bulb in tenascin-R-deficient and wild-type

mice. Tenascin-R-deficient mice had 30.1 ± 7.7 more BrdU+ cells perslice in the RMSOB (105.1 ± 7.7 versus 75 ± 7.1 cells per slice fortenascin-R-deficient and control mice, respectively; P < 0.05) and 20.8± 9.4 fewer BrdU+ cells per slice in the olfactory bulb, includingglomerular, external plexiform and granule cells layers, (62.5 ± 4 ver-sus 83.3 ± 9.4 cells per slice for tenascin-R-deficient and control mice,respectively; P < 0.05). Notably, the total number of BrdU+ cellscounted in the olfactory bulb and RMSOB was not significantly differ-ent in tenascin-R-deficient as compared to wild-type mice (167.6 ±10.5 versus 158.3 ± 15.3 BrdU+ cells per slice for tenascin-R- deficientand control mice, respectively; P > 0.05). These results demonstratethat the excess of BrdU+ cells in the RMSOB of tenascin-R-deficientmice is due to the accumulation of cells in that region.

Notably, in both genotypes, the density of BrdU+ cells in theglomerular layer was similar at 2 and 21 d after BrdU injection (com-pare with Fig. 2d). This contrasts sharply with the density of BrdU+

cells counted in the granule cell layer, which increased 20-foldbetween 2 and 21 d after injection (compare with Fig. 2c). Theseresults imply that neuroblasts first migrate predominantly to theglomerular layer before populating the granule cell layer. Thus, thismight explain why the density of newborn cells in tenascin-R-deficient mice 2 d after BrdU injection was decreased mainly in theglomerular and external plexiform layers and only slightly in thegranule cell layer.

As cell proliferation, tangential migration and cell death were notaffected in tenascin-R-deficient mice, the reduced density of newbornneurons in the olfactory bulb and increased number of neuroblasts inthe RMSOB indicate that newborn progenitors are impeded in leavingthe RMSOB. We therefore tested the kinetics of accumulation of new-born cells in the RMSOB. When mice were given a pulse of BrdU andkilled 4 h later, the density of BrdU+ cells in the RMSOB was higher intenascin-R-deficient mice than in the controls (129.2 ± 11.7 versus100.7 ± 3 cells/mm2; P < 0.05; n = 5; Fig. 5g). This increase was evengreater 2 d after BrdU injection (Fig. 5g). Indeed, when the data wereexpressed as the ratio of mutant to control BrdU+ cell densities in theRMSOB, the increase was significantly smaller 4 h after BrdU injectionthan 2 d after injection (respectively, 1.3 ± 0.09 versus 1.5 ± 0.04,P < 0.05; Fig. 5h). These results strongly imply that the absence oftenascin-R leads to an accumulation of newborn cells in the RMSOB.

To rule out the possibility that elevated numbers of neuroblasts inthe RMSOB of tenascin-R-deficient mice result from increased localproliferation of neuroblast precursors, we carried out two sets ofexperiments. First, we injected BrdU and killed the mice 2 h later tospecifically label dividing progenitors in the SVZ-OB bulb pathway.The number of BrdU+ cells in the RMSOB stained with PSA-NCAMantibodies did not differ between genotypes (55.0 ± 10.1 versus 49.5 ± 8 cells/mm2 for control and tenascin-R-deficient mice,respectively; P > 0.05; n = 4; see Supplementary Fig. 1 online), indi-cating similar local proliferation rates. Second, we assessed the num-ber of cells marked by the endogenous cell division marker Ki67 (ref. 14,15). We found that, in agreement with the first experiment,these numbers did not differ in the RMSOB of control and tenascin-R-deficient mice (85.5 ± 50 versus 61.3 ± 28.2 Ki67+ cells/mm2; P >0.05; n = 4; see Supplementary Fig. 1). These results show that theabsence of tenascin-R reduces the number of newborn bulbar neu-rons as a result of their accumulation in the RMSOB. This highlightsthe pivotal role of tenascin-R in initiating radial migration.

Tenascin-R fosters neuroblast detachment and migrationBefore invading the olfactory bulb, neuroblasts halt their tangentialmigration, detach from their migrating chains and leave the RMSOB.


Figure 4 Normal chain organization and tangential migration of neuroblastsin tenascin-R-deficient mice. (a) Whole-mount immunostaining of PSA-NCAM+ chains in the SVZ of control (+/+, left) and tenascin-R-deficientmice (–/–, right). Insets, high magnification of the PSA-NCAM+ network ofchains. (b,c) PSA-NCAM (green) and BrdU (red) immunostaining in theRMS of control (left) and tenascin-R-deficient mice (right) at low (b) andhigh magnifications (c). (d) Phase-contrast images of SVZ explants ofcontrol (left) and tenascin-R-deficient mice (right) cultured in Matrigel for20 h. (e) High-magnification images of boxed areas in d, showing chainorganization of neuroblasts migrating out of the SVZ explants. (f) Quantification of migration distance, showing no difference betweencontrol (+/+) and tenascin-R-deficient explants (–/–). Data are presented asmeans ± s.e.m. Scale bars: a, 100 µm, inset 40 µm; b,d, 100 µm; c, 30 µm; e, 40 µm.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

To examine which of these steps might be regulated by tenascin-R, wecombined in vitro and in vivo approaches. Inspection of PSA-NCAM+

staining at the interface between the RMSOB and granule cell layershowed that most neuroblasts migrated individually in control mice,whereas many clustered in the tenascin-R-deficient mice (Fig. 6a). Toassess the role of tenascin-R in halting tangential migration anddetaching neuroblasts from chains, we cocultured SVZ explants withtenascin-R-expressing or control BHK cells (Fig. 6b–d). When SVZexplants from P7 mice were cocultured for 20 h with BHK cells, anextensive network of neuroblasts migrated out of SVZ explants (Fig. 6c,d). Quantification of the number of individualized neurob-lasts per SVZ explant showed that there were significantly more suchcells (90% more; P < 0.001) in the presence of BHK cells secretingtenascin-R (Fig. 6e). Notably, the migration distance did not differbetween control explants (248.5 ± 7.1 µm; 24 explants from 8 mice)and those exposed to tenascin-R-secreting cells (260.9 ± 6.5 µm;29 explants from 8 mice; P = 0.2; Fig. 6f). These results indicate thatthe expression of tenascin-R is instrumental in detaching neuroblastsfrom chains but not in halting their tangential migration.

This in vitro assay, however, did not allow to determine whethertenascin-R is also involved in the reorientation of tangentially migrat-ing neuroblasts as seen in vivo in the core of the olfactory bulb. Toexamine whether tenascin-R is necessary and sufficient to reroutetangentially migrating neuroblasts, we introduced tenascin-R intoforebrain regions that neither are populated by progenitor cells norexpress tenascin-R (Fig. 7a). Tenascin-R-expressing or control BHKcells prestained with PKH26 (red labeling in Fig. 7b,c) were graftedinto the striatum (Fig. 7b) or just above the horizontal limb of theRMS (hlRMS) (Fig. 7c). Whereas mice receiving control cells showedunaltered migration (upper panels in Fig. 7b,c; n = 8), all mice graftedwith transfected cells (n = 7) had neuroblasts (green labeling in Fig. 7b,c) entering the tenascin-R-containing area (blue staining in

lower panels of Fig. 7b,c). Ectopic expression of tenascin-R resulted ina 4- to 6-fold greater number of neuroblasts migrating out of the SVZ(Fig. 7d,e) and hlRMS (Fig. 7f,g; 24.5 ± 6.2 versus 86.7 ± 19.0 and15.3 ± 10.5 versus 91.9 ± 16.0 for control and tenascin-R-secretingcells injected, respectively, into the striatum and close to the hlRMS;P < 0.005). Notably, although 28.8 ± 5.8% and 35.5 ± 9.3% of neu-roblasts that migrated out of SVZ and hlRMS, respectively, weredetached from chains (arrows in Fig. 7d,f), many of the rerouted neu-roblasts still assembled in the chains (arrowheads in Fig. 7d,f). Theseresults show that tenascin-R not only promotes detachment of neu-roblasts from chains but also has an important role in the reorienta-tion of tangentially migrating neuroblasts.

Sensory input regulates tenascin-R expression in the bulbOur results indicated that tenascin-R was not only required, but alsosufficient, to initiate radial migration in the adult forebrain. We thusdecided to investigate whether this important function is regulated inan activity-dependent manner. Because the olfactory bulb is the firstcentral relay that receives direct inputs from sensory neurons, we usedunilateral odor deprivation, achieved by occlusion of one nostril, totest whether expression of tenascin-R is sensitive to the level of sen-sory activity. Nostril occlusion resulted in a small, but significant,decrease in tenascin-R mRNA in the ipsilateral bulb as early as 1–2 dafter occlusion, reaching a maximum at 4–30 d after occlusion (Fig. 8a,b). Notably, the reduction was reversible, because reopeningthe nostril for 5 d after 20 d of occlusion resulted in upregulation oftenascin-R mRNA to the level seen in the control bulb (Fig. 8b).Concomitantly, unilateral nostril occlusion also resulted in lower lev-els of tenascin-R protein in the granule cell layer by 27 ± 6%, asassessed 20 d after occlusion (Fig. 8c,d; n = 6). Sham-operated controlmice showed no changes in tenascin-R mRNA and protein levels (Fig. 8b,d). Sensory deprivation–induced reduction of both mRNA


Figure 5 Abnormal radial migration in the olfactory bulb of tenascin-R-deficient mutant mice. (a) TUNEL+ nuclei (arrows) in the GCL of control (+/+) andmutant mice (–/–). (b) Quantification of TUNEL+ nuclei in the GCL and the RMSOB of control and tenascin-R-deficient mice. (c) BrdU immunostaining incoronal sections of control (left) and tenascin-R-deficient mice (right) 2 d after BrdU injection. Note the higher density of BrdU+ nuclei in the RMSOB oftenascin-R-deficient mice. (d) Quantification of BrdU+ nuclei in coronal sections from RMSOB to SVZ (120-µm bin size) in control (black squares) andtenascin-R-deficient mice (gray circles) indicates a significant increase in the density of newborn cells exclusively in the RMSOB of tenascin-R-deficientmice. (e) Conversely, a reduced number of newborn cells is seen in the olfactory bulb of tenascin-R-deficient mice. (f) The total number of BrdU+ nuclei inthe olfactory bulb, excluding the RMSOB, is 53 ± 7 versus 38 ± 5 BrdU+ cells/mm2 in control and tenascin-R-deficient mice, respectively. (g) Time-dependent accumulation of neuroblasts in the RMSOB of tenascin-R-deficient as compared to control mice 4 h and 2 d after BrdU injection. (h) Significantincrease in the ratio of BrdU+ cells in the RMSOB of tenascin-R-deficient versus control mice measured 2 d, as compared to 4 h, after BrdU injection. Dataare presented as means ± s.e.m. *, P < 0.05; **, P < 0.01. Scale bars: a, 50 µm; c,e, 100 µm.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

and protein was further confirmed by quantitative PCR and westernblotting. Tenascin-R mRNA was reduced by 31.6 ± 15% after 7 d ofnostril occlusion and tenascin-R protein by 54.9 ± 12.5% after 20 dof occlusion (see Supplementary Fig. 2 online). These results showthat tenascin-R expression is regulated by activity.

To assess whether this activity-dependent regulation of tenascin-Rexpression could have a functional implication for the recruitment ofnewborn cells to the olfactory bulb, we inspected the density ofBrdU+ cells in the RMSOB of control and occluded bulbs 2 d afterBrdU injection. Similar to the observations with tenascin-R-defi-cient mice, downregulation of tenascin-R expression by unilateralodor deprivation for 30 d led to an accumulation of neuroblasts inthe RMSOB of the occluded as compared to the control bulbs(4,023.9 ± 209.9 versus 3,258.5 ± 228.3 cells/mm2 for odor-deprived and control bulbs, respectively; P < 0.05; n = 5; Fig. 8e,f).These results indicate that regulation of tenascin-R expression bysensory input may be crucial in the recruitment of newborn neu-rons to the olfactory bulb. Notably, the accumulation of neurob-lasts in the RMSOB was less pronounced in occluded mice than intenascin-R-deficient mice (123.5 ± 6.4% versus 147.5 ± 4.1%,respectively; P < 0.05), showing that the level of tenascin-R expres-sion in the olfactory bulb correlates with the degree of neuroblastaccumulation in the RMSOB.

DISCUSSIONBulbar neurogenesis offers a unique model for investigating themechanisms of neuronal recruitment in the adult brain. An impor-tant issue concerns the nature of the molecular cues involved in thecorrect targeting of newborn neuronal precursor cells. Prompted bythe restricted and functionally meaningful expression pattern in theadult olfactory bulb, we have identified the ECM molecule tenascin-Ras a molecular cue that induces neuroblasts to detach from chains andbegin a radial migration away from the RMS into the outer layers ofthe olfactory bulb. In addition, we have shown that ectopic expressionof tenascin-R in non-neurogenic regions reroutes neuroblasts fromtheir normal migratory pathway, thus offering a promising tool forcell replacement therapies.

Tenascin-R enhances or decreases neurite outgrowth and neuronaland glial adhesion depending on the cell type and on its associationwith other ECM molecules16. Furthermore, tenascin-R defasciculatesaxons of cerebellar granule cells in vitro, thus possibly allowinginhibitory interneurons to invade the tightly fasciculating bundles ofaxonal processes17. These observations are pertinent in view of thefact that tenascin-R induces neuroblast detachment and radial migra-tion. In this regard the function of tenascin-R may be similar to thatof reelin, which has been reported to influence this process in theadult olfactory bulb3. However, whereas reelin affects only neuroblastdetachment from chains3, tenascin-R also acts as a directional cuethat reroutes neuronal progenitors from their tangential migratorypathway. In addition, it is not clear whether tenascin-R and reelinshare the same signaling pathway for inducing neuroblast detach-ment. It will therefore be interesting to explore the effects on SVZexplants of tenascin-R alone and in combination with reelin.

What tenascin-R receptor(s) allows neuroblasts to migrate radiallyin the adult brain? Tenascin-R may act directly by interacting withparticular cellular receptors present on migrating neuroblasts, orindirectly, along with other ECM-associated molecules, by capturingand presenting some growth factors18 necessary for radial migration.Among cell surface receptors for tenascin-R are contactin (alsoknown as F3)19, acetylated gangliosides that influence phosphoryla-tion of focal adhesion kinases and affect integrin function20, andreceptor protein-tyrosine phosphatases belonging to the family ofchondroitin sulfate proteoglycans (CSPGs)21,22. Although theinvolvement of contactin and CSPGs (the latter highly expressed inthe adult RMS23) in adult neurogenesis remains to be explored, itwas recently shown that integrins are not involved in the radialmigration of neuroblasts to the olfactory bulb24. Notably, initiationof radial migration in the adult olfactory bulb correlates with the


Figure 6 Tenascin-R acts as a detachment signal for tangentially migratingneuroblasts. (a) PSA-NCAM immunostaining in coronal sections of the GCLof control (+/+, left) and tenascin-R-deficient mice (–/–, right) show thatPSA-NCAM+ cells migrate individually (arrows) in control mice, whereas theyappear clustered (arrowheads) in tenascin-R-deficient mice. (b–d) Phase-contrast image showing cocultures of SVZ explants (left) with an aggregateof BHK cells (right). Examples of SVZ explants cocultured with BHK cellsnot secreting (c) and secreting (d) tenascin-R (TNR). Right panels are highermagnifications of boxed areas in left panels. Note the increased number ofindividualized cells in cocultures of SVZ explants with BHK cells secretingTNR. (e) Significantly higher numbers of individualized cells in cocultures ofSVZ explants with BHK cells secreting TNR (41.5 ± 5.8 versus 77 ± 5.9).**, P < 0.001. (f) Quantification of migration distance of neuroblasts fromSVZ explants cocultured with BHK cells not secreting (Ctrl) or secreting(TNR) TNR. Data are presented as means ± s.e.m.; 21 and 27 explants from7 mice were analyzed in cultures without and with TNR, respectively. Scalebars: a,d (right), 50 µm; b,d (left), 100 µm.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

appearance of NMDA receptor–mediated currents25. These recep-tors are involved in neuronal migration during early developmentalstages26. It is thus possible that tenascin-R, through its cell surfacereceptor(s), trigger(s) the expression of functional NMDA receptorsand, as a consequence, radial migration. Alternatively, tenascin-Rmight directly interact with NMDA receptors through a mechanismsimilar to that described for GABAB receptors27 and thus controlradial migration more directly. Whatever its mode of action, we pro-vide here the first demonstration that tenascin-R has a specific andimportant role in the initiation of radial migration of newborn neu-rons in the adult forebrain.

It is believed that target structures provide attractive and/or sur-vival factors for developing neuronal networks28,29 and that these fac-tors can be regulated in an activity-dependent manner. For instance,

in the olfactory system, modulation of odor information flow hasbeen reported to affect the survival of newborn neurons: sensory dep-rivation by nostril occlusion reduces the number of granule cells inthe developing30–32 and adult olfactory bulb33, whereas reopening ofthe nostril after early occlusion34 as well as olfactory enrichment inadults35 increase the number of newly formed bulbar interneurons.Here, we show that as well as causing reduced survival of newbornneurons, odor deprivation also impairs radial migration of neurob-lasts from the RMS to the olfactory bulb, correlating with downregu-lation of tenascin-R expression. We therefore propose that tenascin-Ris an important mediator relaying network activity to the recruitmentof newborn neurons into the olfactory bulb.

The ability to generate neurons in the adult brain is relevant to thedevelopment of therapeutic strategies aimed at directing the migra-


Figure 7 Ectopic expression of tenascin-R reroutes migrating neuroblasts.(a) BrdU-immunostained sagittal section showing the place of BHK cellgrafts (red circles). (b) Control (upper, Ctrl) or tenascin-R-secreting (lower,TNR) BHK cells prelabeled with PKH26 (red staining) were placed into thestriatum neighboring the SVZ. The rerouted neuroblasts were quantified bycounting PSA-NCAM+ cells (green staining) in the 400-µm-diameter areacalculated from the perimeter of the graft. Note the TNR-immunopositiveareas (blue staining) in mice injected with BHK cells secreting TNR (lower).(c) As in b, but control and TNR-transfected BHK cells were grafted intothe area just above the horizontal limb of the RMS (hlRMS). Right, highmagnifications of the boxed areas shown on the left. Arrowheads indicatechains of neuroblasts diverted from the SVZ (b) or RMS (c). (d) High-magnification images of striatum injected with BHK cells secreting TNR.Arrows and arrowheads indicate, respectively, the individual neuroblastsand chains of progenitor cells rerouted from their normal migratorypathway. (e) Quantification of the number of neuroblasts rerouted from theSVZ by nontransfected (Ctrl) and TNR-transfected BHK cells (TNR). (f,g) Asin d and e, but control and tenascin-R-transfected BHK cells were graftedinto the area above hlRMS. CC, corpus callosum; CTX, cortex; LV, lateralventricle. **, P < 0.001. Values in histograms are means ± s.e.m. Scalebars: b,c, 200 (left) and 50 µm (right); d,f, 20 µm.

Figure 8 Activity-dependent expression of tenascin-R in the olfactory bulb.(a) Pseudocolor images showing expression of tenascin-R mRNA in acoronal section through the olfactory bulb receiving inputs from the opennostril (Ctrl) and the nostril closed for 10 d (Occl). Note the decreased TNRin situ hybridization signal in the odor-deprived olfactory bulb. (b) Theeffect of sensory deprivation on tenascin-R (TNR) mRNA levels wascalculated by relating the mean radioactivity (in cpm/mm2) of the occludedbulb to that of the control bulb. (c) TNR immunostaining of a coronalsection through the olfactory bulb receiving inputs from open (Ctrl) andclosed (Occl) nostrils 20 d after unilateral odor deprivation. Note decreasedimmunostaining for TNR in the GCL of the odor-deprived bulb. (d) Ratiobetween mean TNR immunofluorescence intensity (per mm2) in the GCL ofthe occluded and control bulbs. (e) BrdU+ cells in the RMSOB of the controland occluded bulbs 2 d after BrdU injection. Mice were subjected tounilateral sensory deprivation 30 d before BrdU injection. The sectionswere counterstained with methyl green; arrowheads indicate BrdU+ cells inthe GCL. (f) Quantification of BrdU+ nuclei in the RMSOB of control (Ctrl)and occluded (Occl) bulbs shows a significant increase in the density ofnewborn cells after sensory deprivation. AOB, accessory olfactory bulb;GCL, granule cell layer; RMSOB, rostral migratory stream of the olfactorybulb. *, P < 0.05; **, P < 0.01. Values in histograms are means ± s.e.m.Scale bars: c, 200 µm and e, 50 µm.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

tion of endogenous and/or grafted progenitor cells and their detach-ment from each other. Recently, it has been elegantly demonstratedthat endogenous adult stem cells can regenerate functional neurons innon-neurogenic diseased areas36–38. Neuronal progenitors migrate tothe damaged areas from the neurogenic source localized in the SVZ,implying that not only increased proliferative activity following braindamage37,38 but also changes either in the migratory capabilities ofneuroblasts and/or the microenvironment of the target regions mayrecruit the newly generated neurons for repair. Finally, our resultshave potential therapeutic implications, because they indicate thatrecruitment of neuronal progenitors in diseased brain areas can beenhanced by exogenous application of tenascin-R.

METHODSMice. Tenascin-R-deficient mice8 2–4 months of age and age-matched wild-typelittermates were derived from heterozygous parents with a mixedC57BL/6J×129Ola background. Mice were kept on a 12-h light-dark cycle atconstant temperature (22 °C) with food and water ad libitum. At 3–4 months ofage, the mice were subjected to unilateral olfactory deprivation through cauteri-zation of one nostril39. All experimental procedures were in accordance with theSociety for Neuroscience and European Union guidelines, and were approved bythe institutional animal care and use committees of the Pasteur Institute.

BrdU injections. The DNA synthesis marker 5-bromo-2′-deoxyuridine(BrdU; Sigma) was dissolved in a sterile solution of 0.9% NaCl and 1.75% NaOH (0.4 N). This solution was injected intraperitoneally at a concen-tration of 50 mg per kg body weight. BrdU containing cells were detected byimmunohistochemistry after different survival times: a single dose of BrdUwas given 2 h before killing the mouse to assess proliferation or 4 h beforekilling to assess both proliferation and initial stages of migration. Two injec-tions of BrdU spaced by 4 h were done 2 d before the mice were killed to eval-uate proliferation and migration of neuroblasts. Finally, four injectionsrepeated every 2 h were administrated to mice 21 d before they were killed toevaluate proliferation, migration and survival of newborn cells.

Immunohistochemistry. For all histological analyses requiring tissue fixationby perfusion, mice were deeply anesthetized with an overdose of sodium pen-tobarbital (100 mg per kg body weight; Sanofi) and perfused intracardiallywith saline solution (0.9% NaCl) containing heparin (5 × 103 units/ml) at 37 °C, followed by 4% paraformaldehyde in 0.1 M sodium phosphate buffer(pH 7.3). The brains were dissected out and immersed overnight in the samefixative at 4 °C. Immunohistochemistry was carried out on 40-µm-thick free-floating coronal or sagittal sections cut with a vibrating microtome (VT1000S,Leica) and collected in PBS. Sections were first incubated overnight at 4 °Cwith the following monoclonal antibodies: mouse anti-tenascin-R (clone 619;ref. 40), mouse anti-PSA and mouse anti-NeuN (Chemicon), mouse anti-Ki67(Novocastra), rabbit anti-GFAP (Dako) and rat anti-BrdU (AccurateScientific, Harlan Sera-Lab). For BrdU immunostaining, the sections werepretreated with 0.2% Triton X-100 for 2 h and DNA was denatured with 2 NHCl for 30 min at 37 °C. An overnight incubation with anti-BrdU at 4 °C wasfollowed by a 3-h incubation at room temperature (19–24 °C) with biotiny-lated donkey anti–rat IgG, 1-h incubation with avidin-biotin complex (ABCKit, Vectastain Elite, Vector Laboratories), and development with diaminoben-zidine (DAB, 0.05%) to which 0.005% H2O2 was added. Double- and triple-labeling immunofluorescence was carried out with the following fluorescentsecondary antibodies: Alexa 568–labeled goat anti–rat IgG, Alexa 488–labeledgoat anti–mouse IgG or IgM, and Cy5 anti–mouse IgG (Molecular Probes).Sections were analyzed using either a standard microscope (BX51; Olympus)for peroxidase staining or a Zeiss confocal microscope (Carl Zeiss S.A.S.)equipped with Ar 488, HeNe1 543 and HeNe2 633 lasers, using the LSM-510software package for image acquisition and data analysis.

TUNEL staining was carried out in 8-µm-thick coronal sections of theolfactory bulb to detect DNA fragmentation in situ. After deparaffinizationand rehydration, the tissue was treated with 0.5% Triton X-100 for 10 min andthen incubated 15 sec in equilibrium buffer (Serological Corporation).Sections were then incubated for 1 h in a humidified chamber at 37 °C with a

solution containing terminal deoxynucleotidyl transferase and digoxigeninnucleotides (Serological Corporation). After vigorous washing, a peroxidaselabeled anti-digoxigenin antibody (Serological Corporation) was added for 30 min at room temperature, and staining was revealed by DAB.

The forebrains of unilaterally occluded mice were embedded in gelatin andcut in 40-µm-thick coronal sections. Sections containing both bulbs, thus ipsi-and contralateral to the occluded nostril, were immunostained for tenascin-Rand immunoreactivity was quantified by laser confocal microscopy usingidentical acquisition parameters for both bulbs.

In situ hybridization. The fresh-frozen brains of sham and unilaterally odor-deprived mice were cut in 20-µm-thick coronal slices. The sections containingcontrol and occluded bulbs were thawed onto Superfrost Plus slides and storedat –80 °C until used. Hybridization was done with the synthetic oligonu-cleotide (Eurogentec) 5′-AAG CCC CTC CTT CCT CCT CCA CAG TTT GTCTCT GAG CCC TTT CTG-3′, complementary to nucleotides 720–764 of themRNA encoding Mus musculus tenascin-R. The sequence of the probe waschecked in a GenBank database search to exclude significant homology withother genes. The probe was labeled with [32P]dATP (Perkin Elmer) by the ter-minal deoxynucleotidyl transferase (Roche Diagnostics) reaction followingthe manufacturer’s instructions. The sections were hybridized overnight at 42 °C in a hybridization mixture containing 50% formamide, 10% dextransulfate, 4× SSC (1× SSC is 0.15 M NaCl, 0.015 M sodium citrate). Afterhybridization, the sections were washed for 30 min in 1× SSC prewarmed to60 °C, rinsed in 0.1× SSC and dehydrated in an ascending series of ethanolconcentrations. The digitalized autoradiograms were obtained by exposingsections in a β-imager (Biospace). This real-time imager provides rapid car-tography of 32P labeling (in cpm/mm2) in tissue sections.

Real-time PCR. Real-time PCR was carried out in a LightCycler (RocheDiagnostics) using DNA Master SYBR Green I dye (Roche Diagnostics). Sevendays after odor deprivation, total RNA was extracted from three control and threeoccluded bulbs using FastRNA Pro Green Kit (Qbiogene). cDNA synthesis wascarried out for 1 h at 37 °C in 50-µl reactions containing 2 µg total RNA, reversetranscriptase and random hexamer primers. The PCR protocol used to amplifythe sequences for tenascin-R and actin (used as a housekeeping-gene control) werean initial denaturation at 95 °C for 10 min followed by 40 cycles each consisting ofdenaturation at 95 °C for 15 s, annealing at 62 °C for 5 s and extension at 72 °C for10 s, respectively. The oligonucleotide primers used for PCR were as follows: fortenascin-R, 5′-TGCCAGGACTGAACTTGACA-3′ and 5′-CACAGTGACTTCG-GAGGAGA-3′; for actin, 5′-CTAAGGCCAACCGTGAAAAGATG-3′ and 5′-AGATGGGCACAGTGTGGGTGACC-3′. Melting-curve analysis was performedafter each PCR to confirm the specificity of the reaction and identify the peaks ofinterest in all samples.A relative standard curve for each gene-specific primer pairwas generated with tenfold serial dilutions of cDNAs derived from control andmanipulated bulbs to validate that the dynamic and amplification efficiency oftarget and control genes were approximately equal. The dilution curves were thenused for a relative quantification of target gene expression based on the individualCt (the number of cycles to reach threshold). Results were normalized accordingto the expression level of actin mRNA from the same sample.

Western blot analysis. The control and odor-deprived bulbs were homoge-nized in a glass-Teflon homogenizer in (50 mM Tris-HCl, pH 7.5, 1 mMEDTA, 1 mM EGTA, 1 mM sodium orthovanadate, 50 mM sodium fluoride,5 mM sodium pyrophosphate, 10 mM sodium β-glycerophosphate, 0.1% 2-mercaptoethanol, 1% Triton X-100) lysis buffer supplemented withProtease Inhibitor Cocktail Set III (Calbiochem). The homogenates were cen-trifuged at 13,000g at 4 °C for 20 min to remove insoluble material, and thesupernatants were assayed for protein concentration. Protein samples (15–20µg) were separated on NuPage 4–12% Bis-Tris Gel (Invitrogen) and trans-ferred to nitrocellulose membranes (Amersham Biosciences). Tenascin-R-immunoreactive bands were detected using 619 antibody, horseradishperoxidase–conjugated goat anti-mouse antibodies (Bio-Rad) and anenhanced chemiluminescence substrate mixture (ECL Plus, AmershamBiosciences). The level of expression of tenascin-R in the control and odor-deprived bulbs was normalized to that of NeuN.

Tenascin-R-secreting cells. A 6.2-kb cDNA fragment containing nucleotides1–4070 of the coding sequence of the 180-kDa rat tenascin-R was cloned into


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

the XhoI site of the pcDNA3 vector (Invitrogen) for expression under the con-trol of the CMV promoter. For expression of the protein, the BHK cell line wastransfected with 2 µg of this pcDNA3-TNR plasmid per well in a 6-well plate,using the Lipofectamine kit (Invitrogen) according to the manufacturer’sinstructions. At 24 h after transfection, tenascin-R was detected by immunocy-tochemistry at the cell surface of the transfected but not the untransfected cells(data not shown). In addition, culture supernatants were collected and ana-lyzed for secretion of tenascin-R by western blotting. A specific immunoreac-tive band was seen at 180 kDa in the supernatants of transfected but notuntransfected BHK cells (data not shown).

SVZ explant cultures. Cultures of SVZ explants were prepared as previouslydescribed12. Briefly, brains from P7 mice were dissected out and immersed inice-cold HBSS medium (Gibco). The brains were cut in 200-µm-thick sectionsand those containing the SVZ were selected for further manipulations. Undera surgical microscope, the SVZ was dissected along the lateral wall of lateralventricles and then cut into small pieces (100–200 µm in diameter) that weretransferred to 70% Matrigel (BD Bioscience). After polymerization of Matrigelat 37 °C for 10 min, Neurobasal medium containing B27 supplement,L-glutamine (0.5 mM) and penicillin-streptomycin (1:1,000) (all from Gibco)was added. Cultures were maintained in a humidified incubator at 5% CO2

and 37 °C. In some experiments, SVZ explants were cocultured with aggre-gates of control or tenascin-R-transfected BHK cells. Aggregates of BHK cellswere prepared under low-serum conditions using the hanging-drop method41.

Grafting experiments. Control and tenascin-R-expressing BHK cells werelabeled with the PKH26 red fluorescent cell linker kit (Sigma) following themanufacturer’s instructions, resulting in the staining of about 90% of all cells.Cells were grafted to the part of the striatum neighboring the SVZ (frombregma: anterioposterior, 1.5; mediolateral, 1.0; dorsoventral, 2.6) and justabove the horizontal limb of RMS (anterioposterior, 3.35; mediolateral, 0.82;dorsoventral, 3.0) using a Kopf stereotaxic apparatus (Harvard Apparatus).On the day of transplantation, BHK cells were resuspended by trypsinizationand collected after centrifugation (10 min, 475g, 4 °C). For transplantation,mice were anesthetized with a ketamine-xylazine mixture (Sigma) andapproximately 2 × 105 BHK cells in 0.3 µl of solution were injected over aperiod of 3 min using a very thin glass electrode. The electrode was then left inplace for an additional 3 min before being slowly withdrawn. Mice were killed4 d later and the number of PSA-NCAM+ cells exiting the SVZ or RMS wascounted in sagittal sections containing the grafts. The same sections were alsoprocessed for tenascin-R immunolabeling.

Quantification and statistical analyses. All quantifications were done blind tothe experimental conditions. For analysis of cell migration distance in vitro,explants were examined using phase-contrast microscopy after 20 h in culture.Migration distance was quantified by measuring the maximum distance thatcells had moved away from the perimeter of each explant.

BrdU-immunostained nuclei were quantified in every third 40-µm sectionalong the entire SVZ-OB pathway. To assess the number of newborn neuronsin the olfactory bulb, BrdU+ nuclei, observed with ×20 and ×40 objectives,were counted for the entire granule cell and glomerular layers. The numbers ofBrdU+ profiles were then related to the surface of granule cell (includingmitral cell and internal plexiform layers) and glomerular layers. For RMS andSVZ, the density of BrdU+ cells was evaluated by relating the number of BrdU+

cells to the surfaces occupied by these cells. To count TUNEL+ cells in thegranule cell layer and RMSOB, these areas were delineated manually on sec-tions counterstained with Methyl Green (Vector Laboratories). The percent-ages of BrdU + NeuN and BrdU + GFAP doubly immunostained cells wereobtained by analyzing 3D-reconstructed BrdU+ nuclei in the x-z and y-zorthogonal projections for the presence of NeuN or GFAP.

Neuroblasts migrating towards the grafted BHK cells were quantified bycounting the PSA-NCAM+ cells that had detached from the SVZ or RMS. Inmice grafted with BHK cells secreting tenascin-R, tenascin-R-immunoreactiveareas were as large as 100–400 µm in diameter. To assess the number ofrerouted neuroblasts in BHK-injected mice, PSA-NCAM+ cells outside of theirnormal migratory pathway were counted in the 400-µm-diameter area, calcu-lated from the perimeter of the graft. Counting was done along the entire

z axis, and data are presented as the total number of PSA-NCAM+ cellsrerouted from the SVZ or the RMS.

To quantify the impact of sensory deprivation on tenascin-R mRNA,hybridized (32P) radioactivity (in cpm/mm2) in the granule cell layer of theoccluded bulb was related to that of the control bulb in the same section.Measurements were carried out from 20–30 coronal sections per mouse (2–3 mice per group). To assess the abundance of the tenascin-R protein, meanfluorescence intensity (per mm2) in the granule cell layer of the occluded bulbwas related to that of the control bulb of tenascin-R-immunostained sections.

All statistical comparisons between groups were made using Student’s t-test.


ACKNOWLEDGMENTSThis work was supported by the Pasteur Institute, CNRS, The Annette Gruner-Schlumberger Foundation and grants from the French Ministry of Research andEducation (ACI ‘Biologie du Développement et Physiologie Intégrative’ and GIS‘Infections à Prions’) and the Gemeinnuetzige Hertie Stiftung. We are grateful to R. Grailhe and N. Mechawar for help with in situ hybridization, F.-A. Weltzien forhelp with quantitative PCR analysis, P. Roux at the ‘Plate-form d’ImagerieDynamique’ of the Institut Pasteur for help with confocal microscopy, A. Cardonafor help with the β-imager and M.-M. Gabellec for excellent technical assistance.We thank S. Freitag and F. Morellini for providing tenascin-R-deficient mice andM. Sibbe and M. Kutsche for the pcDNA3-TNR construct.


Received 10 December 2003; accepted 24 February 2004Published online at http://www.nature.com/natureneuroscience/

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

2. Alvarez-Buylla, A. & Garcia-Verdugo, J.M. Neurogenesis in adult subventricularzone. J. Neurosci. 22, 629–634 (2002).

3. Hack, I., Bancila, M., Loulier, K., Carroll, P. & Cremer, H. Reelin is a detachmentsignal in tangential chain-migration during postnatal neurogenesis. Nat. Neurosci.5, 939–945 (2002).

4. Jones, F.S. & Jones, P.L. The tenascin family of ECM glycoproteins: structure, func-tion, and regulation during embryonic development and tissue remodeling. Dev.Dyn. 218, 235–259 (2000).

5. Fuss, B., Wintergerst, E.S., Bartsch, U. & Schachner, M. Molecular characterizationand in situ mRNA localization of the neural recognition molecule J1-160/180: amodular structure similar to tenascin. J. Cell Biol. 120, 1237–1249 (1993).

6. Srinivasan, J., Schachner, M. & Catterall, W.A. Interaction of voltage-gated sodiumchannels with the extracellular matrix molecules tenascin-C and tenascin-R. Proc.Natl. Acad. Sci. USA 95, 15753–15757 (1998).

7. Xiao, Z.C. et al. Tenascin-R is a functional modulator of sodium channel beta sub-units. J. Biol. Chem. 274, 26511–26517 (1999).

8. Weber, P. et al. Mice deficient for tenascin-R display alterations of the extracellularmatrix and decreased axonal conduction velocities in the CNS. J. Neurosci. 19,4245–4262 (1999).

9. Bruckner, G. et al. Postnatal development of perineuronal nets in wild-type mice andin a mutant deficient in tenascin-R. J. Comp. Neurol. 428, 616–629 (2000).

10. Nikonenko, A., Schmidt, S., Skibo, G., Bruckner, G. & Schachner, M. Tenascin-R-deficient mice show structural alterations of symmetric perisomatic synapses in theCA1 region of the hippocampus. J. Comp. Neurol. 456, 338–349 (2003).

11. Saghatelyan, A.K. et al. Reduced perisomatic inhibition, increased excitatory trans-mission, and impaired long-term potentiation in mice deficient for the extracellularmatrix glycoprotein tenascin-R. Mol. Cell. Neurosci. 17, 226–240 (2001).

12. Wichterle, H., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Direct evidence for homo-typic, glia-independent neuronal migration. Neuron 18, 779–791 (1997).

13. Chazal, G., Durbec, P., Jankovski, A., Rougon, G. & Cremer, H. Consequences ofneural cell adhesion molecule deficiency on cell migration in the rostral migratorystream of the mouse. J. Neurosci. 20, 1446–1457 (2000).

14. Tanapat, P., Hastings, N.B., Reeves, A.J. & Gould, E. Estrogen stimulates a transientincrease in the number of new neurons in the dentate gyrus of the adult female rat.J. Neurosci. 19, 5792–5801 (1999).

15. Kee, N., Sivalingam, S., Boonstra, R. & Wojtowicz, J.M. The utility of Ki-67 andBrdU as proliferative markers of adult neurogenesis. J. Neurosci. Methods 115,97–105 (2002).

16. Pesheva, P. & Probstmeier, R. The yin and yang of tenascin-R in CNS developmentand pathology. Prog. Neurobiol. 61, 465–493 (2000).

17. Xiao, Z.C. et al. Defasciculation of neurites is mediated by tenascin-R and its neu-ronal receptor F3/11. J. Neurosci. Res. 52, 390–404 (1998).

18. Boudreau, N. & Bissell, M.J. Extracellular matrix signaling: integration of form andfunction in normal and malignant cells. Curr. Opin. Cell Biol. 10, 640–646 (1998).

19. Pesheva, P., Gennarini, G., Goridis, C. & Schachner, M. The F3/11 cell adhesion


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

molecule mediates the repulsion of neurons by the extracellular matrix glycoproteinJ1-160/180. Neuron 10, 69–82 (1993).

20. Probstmeier, R., Michels, M., Franz, T., Chan, B.M. & Pesheva, P. Tenascin-R inter-feres with integrin-dependent oligodendrocyte precursor cell adhesion by a ganglio-side-mediated signalling mechanism. Eur. J. Neurosci. 11, 2474–2488 (1999).

21. Xiao, Z.C. et al. Isolation of a tenascin-R binding protein from mouse brain mem-branes. A phosphacan-related chondroitin sulfate proteoglycan. J. Biol. Chem. 272,32092–32101 (1997).

22. Milev, P. et al. High affinity binding and overlapping localization of neurocan andphosphacan/protein-tyrosine phosphatase-zeta/beta with tenascin-R, amphoterin,and the heparin-binding growth-associated molecule. J. Biol. Chem. 273,6998–7005 (1998).

23. Thomas, L.B., Gates, M.A. & Steindler, D.A. Young neurons from the adultsubependymal zone proliferate and migrate along an astrocyte, extracellular matrix-rich pathway. Glia 17, 1–14 (1996).

24. Murase, S. & Horwitz, A.F. Deleted in colorectal carcinoma and differentiallyexpressed integrins mediate the directional migration of neural precursors in the ros-tral migratory stream. J. Neurosci. 22, 3568–3579 (2002).

25. Carleton, A., Petreanu, L.T., Lansford, R., Alvarez-Buylla, A. & Lledo, P.M.Becoming a new neuron in the adult olfactory bulb. Nat. Neurosci. 6, 507–518(2003).

26. Komuro, H. & Rakic, P. Modulation of neuronal migration by NMDA receptors.Science 260, 95–97 (1993).

27. Saghatelyan, A. et al. Recognition molecule associated carbohydrate inhibits post-synaptic GABAb receptors: a mechanism for homeostatic regulation of GABA releasein perisomatic synapses. Mol. Cell. Neurosci. 24, 271–282 (2003).

28. Kennedy, T.E. & Tessier-Lavigne, M. Guidance and induction of branch formation indeveloping axons by target-derived diffusible factors. Curr. Opin. Neurobiol. 5,83–90 (1995).

29. Svendsen, C.N. & Sofroniew, M.V. Do central nervous system neurons require target-derived neurotrophic support for survival throughout adult life and aging? Perspect.

Dev. Neurobiol. 3, 133–142 (1996).30. Frazier, L.L. & Brunjes, P.C. Unilateral odor deprivation: early postnatal changes in

olfactory bulb cell density and number. J. Comp. Neurol. 269, 355–370 (1998).31. Brunjes, P.C. Unilateral naris closure and olfactory system development. Brain Res.

Brain Res. Rev. 19, 146–160 (1994).32. Cummings, D.M. & Brunjes, P.C. The effects of variable periods of functional depri-

vation on olfactory bulb development in rats. Exp. Neurol. 148, 360–366 (1997).33. Henegar, J.R. & Maruniak, J.A. Quantification of the effects of long-term unilateral

naris closure on the olfactory bulbs of adult mice. Brain Res. 568, 230–234(1991).

34. Cummings, D.M., Henning, H.E. & Brunjes, P.C. Olfactory bulb recovery after earlysensory deprivation. J. Neurosci. 17, 7433–7440 (1997).

35. Rochefort, C., Gheusi, G., Vincent, J.D. & Lledo, P.M. Enriched odor exposureincreases the number of newborn neurons in the adult olfactory bulb and improvesodor memory. J. Neurosci. 22, 2679–2689 (2002).

36. Magavi, S.S., Leavitt, B.R. & Macklis, J.D. Induction of neurogenesis in the neocor-tex of adult mice. Nature 405, 951–955 (2000).

37. Arvidsson, A., Collin, T., Kirik, D., Kokaia, Z. & Lindvall, O. Neuronal replacement fromendogenous precursors in the adult brain after stroke. Nat. Med. 8, 963–970 (2002).

38. Nakatomi, H. et al. Regeneration of hippocampal pyramidal neurons after ischemicbrain injury by recruitment of endogenous neural progenitors. Cell 110, 429–441(2002).

39. Meisami, E. Effects of olfactory deprivation on postnatal growth of the rat olfactorybulb utilizing a new method for production of neonatal unilateral anosmia. BrainRes. 107, 437–444 (1976).

40. Morganti, M.C., Taylor, J., Pesheva, P. & Schachner, M. Oligodendrocyte-derived J1-160/180 extracellular matrix glycoproteins are adhesive or repulsive depending onthe partner cell type and time of interaction. Exp. Neurol. 109, 98–110 (1990).

41. Fan, C.M. & Tessier-Lavigne, M. Patterning of mammalian somites by surface ecto-derm and notochord: evidence for sclerotome induction by a hedgehog homolog.Cell 79, 1175–1186 (1994).


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

Formation of excitatory synapses is established by sequential cellularevents, including the protrusion of filopodia from dendrites, the con-tact between filopodia and axons1,2, and morphological changes ofthe filopodia into mature spines3,4. Filopodia are highly motile, andthis motility is thought to facilitate the initial formation of dendrite-axon contacts. Even after the establishment of these contacts, thespines are still motile, dynamically changing their shape5–7. Thisstructural flexibility of spines is thought to be crucial for remodelingof neural circuits8. Uncovering the cellular and molecular mecha-nisms that regulate spine motility and stability is thus important forunderstanding complex synapse functions.

Several molecules have been identified as regulators of dendriticspine formation and shape. These include the Rho family of smallGTPases9,10, scaffold proteins such as Shank and Homer (ref. 11),SPAR (ref. 12), drebrin (ref. 13), PSD95 (ref. 14) and Syndecan 2 (ref. 15). Furthermore, EphrinB-EphB interaction induces spine for-mation16, and does GluR2, even on inhibitory neurons17. Concerningthe physiological regulation of spine motility and stability, TTXenhances spine motility6; conversely, activation of either AMPA orNMDA receptors inhibits the actin-based protrusive activity of spineheads18. Sensory deprivation due to whisker trimming was found toreduce spine protrusion in the barrel cortex7. These observations sug-gest that neural activity controls spine motility. Despite such exten-sive studies, however, the molecular and cellular bases of synapsestability still remain obscure.

Classic cadherins and associated proteins are localized in synapticjunctions19–24. Our previous studies showed that when cadherin func-tion is blocked by a dominant-negative form of cadherin, synapticorganization is impaired25. Mutation of αN-catenin, a cadherin-asso-ciated protein, also affected dendritic spine morphology; although themutant spines could establish synaptic contact with axons. Based onthese and other observations26, we have postulated that the cad-

herin/catenin complex may function as a morphological regulator ofsynaptic plasticity. In the present study, we analyzed the role of αN-catenin in synapse dynamics and found that, in the absence of αN-catenin, dendritic spine heads became unusually motile and could notmaintain their stable contacts with axons. Conversely, overexpressionof this molecule caused an increase in the number of mature spines.These results suggest that αN-catenin functions as a critical agent toregulate the stability of synaptic contacts.

RESULTSDendritic spines on αN-catenin-deficient neurons are unstableTo examine the role of αN-catenin in synapse dynamics, we culturedhippocampal neurons collected from homozygous mutant mice thatlacked the gene encoding αN-catenin (Catna2–/–), as well as fromtheir heterozygous (Catna2+/–) or wild-type (Catna2+/+) littermates.In these cultures, the density of dendritic spines was not statisticallydifferent between Catna2–/– and Catna2+/+ neurons (Fig. 1).Immunostaining for a presynaptic marker, synapsin, showed that thedensity of synapsin puncta present on both dendritic spines andshafts was also not significantly different between these neurons.However, the ratio of synapsin-positive spines to the total number ofspines was slightly reduced in Catna2–/– neurons (Fig. 1a), with asmall increase in synapsin puncta density on the dendritic shafts.

The above hippocampal neurons were transfected with an enhancedgreen fluorescent protein (EGFP)-actin expression vector at the timeof plating to visualize the activities of filopodia and dendritic spines.At 14–15 d.i.v. (days in vitro), fluorescently labeled neurons were sub-jected to time-lapse observations. In Catna2+/– or Catna2+/+ neurons,dynamic movement of actin-labeled dendritic spines was observed, asreported previously27. Catna2–/– spines were also motile, but unlikeheterozygous or wild-type (Fig. 2a) neurons, they showed rapid pro-trusion and retraction of filopodia from most of the spines (Fig. 2b,

1RIKEN Center for Developmental Biology, 2-2-3 Minatojima-Minamimachi, Chuo-ku, Kobe 650-0047, Japan. 2Graduate School of Biostudies, Kyoto University,Kitashirakawa, Sakyo-ku, Kyoto 606-8502, Japan. 3Department for Molecular Biomedical Research, VIB-Ghent University, B-9000 Ghent, Belgium. Correspondenceshould be addressed to M.T. ([email protected]).


Stability of dendritic spines and synaptic contacts iscontrolled by αN-cateninKentaro Abe1,2, Osamu Chisaka2, Frans van Roy3 & Masatoshi Takeichi1,2

Morphological plasticity of dendritic spines and synapses is thought to be crucial for their physiological functions. Here we showthat αN-catenin, a linker between cadherin adhesion receptors and the actin cytoskeleton, is essential for stabilizing dendriticspines in rodent hippocampal neurons in culture. In the absence of αN-catenin, spine heads were abnormally motile, activelyprotruding filopodia from their synaptic contact sites. Conversely, αN-catenin overexpression in dendrites reduced spine turnover,causing an increase in spine and synapse density. Tetrodotoxin (TTX), a neural activity blocker, suppressed the synapticaccumulation of αN-catenin, whereas bicuculline, a GABA antagonist, promoted it. Furthermore, excess αN-catenin renderedspines resistant to the TTX treatment. These results suggest that αN-catenin is a key regulator for the stability of synaptic contacts.


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

they were equally motile in wild-type, heterozygous and homozygousneurons. These findings indicate that Catna2–/– synaptic contacts arenot stable, allowing extra filopodial protrusion from the spine heads.

αN-Catenin overexpression causes an increase in spine densityTo further explore the role of αN-catenin, we examined the effect ofits overexpression on synaptogenesis in hippocampal neurons.Freshly isolated neurons were transfected with either αN-catenin andGFP or GFP alone, and then cultured for various periods of time.First we transfected Catna2–/– neurons with αN-catenin, and foundthat the normal morphology of the spines was restored (data notshown). When wild-type neurons were used for transfection, the den-sity of the spines on dendrites significantly increased in αN-catenin-overexpressing neurons at 20 and 30 d.i.v. (Fig. 3a,b); this tendencywas already detectable at 10 d.i.v. (Fig. 3b). The width of spine headsalso showed some increase in these neurons (Fig. 3c,e). The length ofspines, on the other hand, did not differ between the control andexperimental samples, except at 10 d.i.v. (Fig. 3d,f). The difference at10 d.i.v. suggests that the conversion of filopodia into spines, occur-ring in such early cultures, was facilitated by αN-catenin overexpres-sion. These effects were observed only in excitatory neurons; that is,no ectopic spine formation was induced on inhibitory neurons.Immunostaining for synapsin showed that the above increase in spine

arrows; see also Supplementary Videos 1 and 2); this movement wasrecorded as high degrees of oscillation in the length of each spine (Fig.2d; compare to c). Many of these spines also showed unusuallydynamic deformation of their heads (Fig. 2b, arrowhead). The activefilopodial protrusions in Catna2–/– neurons likely occurred from thespine heads that were in contact with axons. For example, the spinemarked ‘d’ in Figure 2b was highly motile but appears to be associatedwith an axon (ax) visualized by accidental EGFP-actin labeling. Forfurther confirmation, we stained these neurons for PSD95, whichaccumulates only in sites of synaptic contact. We found that spineswith filopodial protrusions were PSD95-positive ones (Fig. 1b). InCatna2+/– or Catna2+/+ spines (Fig. 2a,c), we could detect some filopo-dial protrusion and retraction from the already-established synapticcontacts, but movement was restricted mostly to the head body itself,resulting in a narrower range of spine length oscillation (Fig. 2e,f).With regard to filopodia directly protruding from the dendritic shafts,

Figure 2 Enhanced motility in dendritic spines of Catna2–/– neurons. (a,b) Time-lapse movies of 15 d.i.v. neurons transfected with EGFP-tagged actin,obtained from Catna2+/+ or Catna2–/– embryos. Arrows point to representative filopodia repeating protrusion and retraction from the spine heads.Arrowhead, an example of actively deforming spines. Some axons (ax) are also labeled with EGFP-tagged actin; other axons associated with spines are notlabeled in these images. See also Supplementary Videos 1 and 2. Scale bars = 5 µm. (c,d) Tracing of the length of six representative spines indicated in a and b during a 95-min incubation period. (e,f) Mean changes in the length of spines undergoing protrusion and retraction during a 95-min period, andcumulative frequency plot for the changes. The length of spines was traced for 5 min, and the minimum length was subtracted form the maximum lengthfor each spine. More than 70 spines from six dendrites were measured. The spine-length change was significantly enhanced in neurons without αN-catenin, and it was suppressed in those overexpressing αN-catenin. *P < 0.001.

A R T I C L E S


Figure 1 Profiles of dendritic spines in Catna2+/+ and Catna2–/– neurons.(a) Mean density of total spines, synapsin-positive puncta and synapsin-positive spines. GFP-transfected neurons were immunostained for synapsinat 20 d.i.v. and subjected to analysis. The density of total spines (P = 0.11)and synapsin-positive puncta (P = 0.34) was not significantly differentbetween the wild-type and mutant samples, but that of synapsin-positivespines was significantly different (P < 0.001). (b) Immunostaining for PSD95in dendrites of Catna2+/+ or Catna2–/– neurons expressing EGFP at 16 d.i.v.Arrows show filopodial protrusion from spine heads. Scale bar = 20 µm.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

not survived during the long culture periods used in our experiments.Thus, our observations were limited to spine morphogenesis.

αN-catenin overexpression stabilizes spinesThe above observations suggest that αN-catenin functions to stabilizespines and synapses. For confirmation, we took time-lapse movies ofαN-catenin-overexpressing neurons and found that their spines wereindeed more quiescent than those of the controls in terms of filopodia-protruding activity (Fig. 2e,f). The increase in spine density in αN-catenin-overexpressing neurons suggests the possibility that spineturnover was suppressed in these neurons, resulting in their accumu-

Figure 4 Reduced turnover of dendritic spines inαN-catenin-overexpressing neurons. Time-lapseGFP-fluorescence images were collected fromrepresentative dendrites of Catna2+/+ andCatna2–/– neurons transfected with GFP, and alsoof neurons co-transfected with GFP and αN-cat-flag, every 3 d during the period from 18 to 27d.i.v. (a–c) Examples of the spines thatdisappeared (open triangles), formed anew (darktriangles) or became deformed (arrows) duringthe observations. The ratio of these spines is alsoshown (d). Data were collected from more than1,000 spines on 18 dendrites of seven neuronsfor each group in the experiment.

density in αN-catenin-overexpressing neurons was accompanied bythat in the density of synapsin-positive puncta (1.29 ± 0.10 per µm inoverexpressing neurons versus 0.75 ± 0.02 per µm in wild type,P < 0.001, n = 13, at 20 d.i.v.), as well as in the ratio of spinesimmunoreactive for synapsin to the total spines (91.3 ± 1.5% in over-expressing neurons versus 75.5 ± 1.7% in wild-type neurons,P < 0.001, n = 13, at 20 d.i.v.). On the other hand, the staining inten-sity for synapsin, GluR2 and PSD95 in individual synaptic puncta wasnot particularly altered by the αN-catenin overexpression (Fig. 3h–j).

There are two isoforms of αN-catenin, I and II; both had similareffects on spine density (Fig. 3g). Overexpression of two other sub-types of α-catenin, namely αE-catenin and αT-catenin, also inducedexcess spine formation, indicating that they share the same spine-stabilizing activity. On the other hand, overexpression of N-cadherinor β-catenin had no effect on spine morphology or density (Fig. 3g),which indicates that αN-catenin has a unique activity regarding spineand synapse formation among the molecules constituting the cadherin/catenin complex. We could detect the above effects ofαN-catenin only when postsynaptic neurons overexpressed this mole-cule. We then confined this overexpression to presynaptic sites, that is,to axons. This experiment was technically difficult, however, becausetransfected αN-catenin molecules were only faintly localized to axons;therefore we could not accurately identify the spines in contact withthe labeled axons in crowded cultures. Recent studies show that over-expression of not only β-catenin or N-cadherin, but also that of αN-catenin in hippocampal neurons enhanced their dendriticarborization28. In our transfection protocol, however, we did notobserve such an effect of αN-catenin overexpression on dendritebranching, when examined at 7 d.i.v. (Supplementary Fig. 1 online) orat later culture stages. Likewise, Catna2 mutation had no effect on thisphenomenon (Supplementary Fig. 1). Neurons that received cDNAsat the levels required for stimulation of dendrite branching might have

A R T I C L E S


Figure 3 Increased spine density in αN-catenin-overexpressing neurons. (a) Neurons at 18 d.i.v. expressing EGFP (GFP) or EGFP-tagged αN-catenin(αN-cat-GFP) immunostained for GFP. Lower panes, close-up views of arepresentative portion of each neuron. The differences in thickness andbranching of dendrites seen between the control and experimental samples aredue to a variation in morphology of individual neurons and not to αN-cateninoverexpression. (b–d) Mean spine density, head width and length in 10, 20and 30 d.i.v. neurons transfected with GFP or αN-cat-GFP. More than 820spines from 12 neurons were measured. (e,f) Cumulative frequency plots forspine length and head width in 30 d.i.v. neurons. (g) Effects of overexpressionof α-catenin subtypes, N-cadherin and β-catenin on spine density. More than470 spines from seven neurons were measured for each construct. (h–j) Tripleimmunostaining for PSD95, synapsin and GluR2 in 20 d.i.v. neuronstransfected with GFP + Flag-tagged αN-catenin (αN-cat-flag) or with GFP only.Scale bars = 5 µm. *P < 0.001, compared with GFP only.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

bicuculline treatment resulted in an increase in the αN-catenin fluores-cence intensity at the synapses. In these cultures, β-catenin signal inten-sity in synapses was only slightly decreased in the TTX-treated neurons,whereas it was significantly increased in the bicuculline-treated ones(Fig. 6b). On the other hand, synapsin signals were not particularlyaltered after these treatments, suggesting that presynaptic structureswere normally maintained. Under these experimental conditions, theentire expression levels of N-cadherin, αN-catenin and β-catenin didnot differ between the non-treated and TTX-treated cultures. In the caseof bicuculline treatment, however, the levels of these proteins wereslightly upregulated (Fig. 6c). These results suggest that the TTX treat-ment altered only the distribution of these proteins, whereas bicucullinemay have enhanced stabilization or expression of these proteins. Thereappeared to be a pool of αN-catenin, diffusely distributed and not asso-ciated with synapses, as seen in the merged images in Figure 6a. Thispool of αN-catenin probably increased in TTX-treated neurons.

In the above TTX experiments, it remained unclear whether thereduction in αN-catenin in the synapses had occurred as a result ofthe TTX-induced morphological changes in the spines or whetherTTX-dependent signals actively suppressed the synaptic αN-cateninaccumulation, secondarily leading to changes in spine shape. To testthese possibilities, we compared αN-catenin-overexpressing and con-trol hippocampal neurons for their response to TTX. In control neu-rons, their spines were changed into filopodia-like processes after theTTX treatment, whereas αN-catenin-overexpressing neurons did notclearly respond to TTX (Fig. 6d–f). Thus, excess αN-catenin interfereswith the TTX action, supporting the idea that αN-catenin may func-tion as a mediator of neural activity–dependent signals for spine mor-phological changes.

A R T I C L E S

lation. To test this possibility, we recorded theimages of identical dendrites for 9 d from 18to 27 d.i.v. (Fig. 4a–c), as it has previouslybeen reported that turnover can be observedin this time window29. We analyzed morpho-logical changes in the spines every 3 d duringthis 9-d period and recorded the means ofthe data collected from the three differentperiods in Figure 4d. During each 3-d period,in control neurons expressing GFP only,15.0% of the spines disappeared, 18.9%formed anew and the other spines remainedunchanged, although many showed a changein morphology. In Catna2–/– neurons, 19.8%of the spines disappeared, whereas in αN-catenin-overexpressing neurons, only 9.7%disappeared. These data support the idea that the overexpression ofαN-catenin suppressed spine turnover.

αN-catenin has various domains that interact with a number ofmolecules30,31. To identify the domain(s) responsible for the aboveactivities, we designed a series of deletion mutants (Fig. 5a) and usedthem for transfection (Fig. 5b). An amino (N)-terminal deletion thatremoved the β-catenin binding site (αN-277-954) abolished the spinedensity–increasing (SDI) activity, indicating that αN-catenin needs toassociate with the cadherin/β-catenin complex for its action. A shortcarboxy (C)-terminal deletion (αN-1-870) did not affect the SDIactivity; whereas molecules with longer C-terminal deletions (i.e.,αN-1-681, 1-406 and 1-262) no longer exhibited this activity (Fig.5c). Nevertheless, these mutants caused elongation of spines (Fig. 5d),as seen in Catna2–/– neurons, suggesting that they had a dominant-negative effect. These results suggest that the C-terminal domain,known to bind various cytoskeletal proteins (Fig. 5a), is essential forthe SDI activity of αN-catenin.

Involvement of αN-catenin in physiological plasticity of spinesDendritic spines and synapses are structurally and functionally modi-fied under various physiological conditions. We asked whether αN-catenin is involved in such processes. TTX, a neural activity blocker,converts dendritic spines into filopodial processes in hippocampal cul-tures32. We treated rat hippocampal neurons at 18–20 d.i.v. with 1.5 µMTTX for 3 d, and then stained them for αN-catenin. The fluorescenceintensity of αN-catenin signals associated with synapsin puncta signifi-cantly decreased in the TTX-treated neurons (Fig. 6a,b). Next, we exam-ined the effect of 40 µM bicuculline, which antagonizes GABA-receptorsand thus increases neural activity. Contrary to the TTX treatment, the


Figure 5 Requirement of the C-terminal regionof αN-catenin for the increase in spine density.(a) Deletion series of αN-catenin. The regionsrequired for association with other proteins,identified for αE-catenin, are depicted.(b) Representative dendrites expressing eachconstruct. Scale bar = 5 µm. (c,d) Mean spinedensity and length/width in 20 d.i.v. neuronstransfected with each construct. Among thedeletion mutants, only αN1-870 is effective inincreasing spine density. αN-1-681, αN-1-406and αN-1-262 cause an increase in spine length,suggesting that they have a dominant-negativeaction. More than 500 spines from sevenneurons were measured for each construct. *P <0.001, as compared to GFP only.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

DISCUSSIONThe loss of αN-catenin causes deformation of dendritic spines25. Ourtime-lapse observations revealed more dynamic features of the spinesin Catna2–/– neurons than anticipated from fixed samples. Thesespines were not simply abnormal in their static morphology; they alsoshowed an unusually high activity of filopodial protrusion from thespine heads. Although filopodial protrusion is an essential process fornew synapse formation in early neurons, this occurs from dendriticshafts but barely from established synaptic contacts. These observa-tions indicate that αN-catenin is responsible for suppressing themotile activity of filopodia/spines. We also showed that overexpres-sion of αΝ-catenin induced the formation of extra spines andsynapses. Synapses are not static; their addition and elimination on adendrite during prolonged periods have been observed29,33,34. Suchturnover of synapses was reduced in αΝ-catenin-overexpressing neu-rons. Collectively, we can conclude that αΝ-catenin contributes tosynapse stability from two aspects: suppression of spine motilityactivity and suppression of spine turnover.

Synapse stability and remodeling are essential for a variety of neu-ral activities35. We propose that αΝ-catenin may function as a regu-lator for synaptic remodeling, if external signals can modulateαΝ-catenin activities. There are several possible pathways to affectαΝ-catenin function. For example, signals that alter the binding of

αΝ-catenin to β-catenin can affect αΝ-catenin function. The inter-actions between α-catenin and β-catenin can be altered by variousfactors, such as Fer and Fyn tyrosine kinases or protein kinase CKII,which affect their binding36,37. Notably, neural activity induces redis-tribution of β-catenin into synapses38, as was confirmed by our bicu-culline experiment. Facilitating the β-catenin binding to cadherinwould be expected to recruit more αN-catenin molecules tosynapses, and this could be a process to stabilize synapses. However,we found no effect of β-catenin overexpression on synapse stabiliza-tion, suggesting that the β-catenin level itself is not sufficient to altersynapse stability. The requirement of the C-terminal domain of αΝ-catenin for synapse stabilization suggests other potential regulatorysystems. For example, the C-terminal domain is known to bind F-actin (ref. 39), ZO-1 (ref. 40) and vezatin (ref. 41), and the interac-tion of αΝ-catenin with these proteins may be involved incontrolling spine motility and stability. It is intriguing to note thatprofilin, a regulator of actin polymerization, is targeted to spines inan NMDA receptor activity–dependent manner, concomitantly sup-pressing changes in spine shape. Furthermore, blocking this profilintargeting destabilizes spine structure42. There may be interplaybetween the profilin/actin and cadherin/αΝ-catenin systems in sucha way that the former controls the latter activity, or the latter pro-duces signals for actin reorganization. Given that neural activity canmodify any of these molecular interactions, such processes areexpected to contribute to changes in spine shape.

We found that the treatment of neurons with TTX, which inducesthe conversion of spines into filopodia-like processes, caused a reduc-tion in the αΝ-catenin concentration in synapses. Furthermore, spinesbecame resistant to TTX when αΝ-catenin was overexpressed. Theseobservations suggest that neural activity blockade directly or indirectlycauses a release of αN-catenin from synapses, leading to filopodia-likespine formation, and that αN-catenin overexpression blocks this sys-tem. In the TTX experiments, we also detected a subtle reduction in thefluorescence intensity of β-catenin, but this change was not as extensiveas the change in αN-catenin, suggesting that the redistribution of αN-catenin had a dominant role in these phenomena. Importantly, overex-pression of N-cadherin and β-catenin did not have apparent effects onspine stability. This finding suggests that the size of the cytoplasmicpool of αΝ-catenin, but not that of other molecules of thecadherin/catenin complex, may be a limiting factor for the modulationof synaptic contacts. Thus, synapse stability seems not to be regulatedby a simple increase in the number of cadherin molecules recruited tosynaptic membranes but rather by an αΝ-catenin-specific signaling


Figure 6 Effects of TTX and bicuculline on αN-catenin distribution andspine morphology in rat hippocampal neurons. (a) Double immunostainingfor αN-catenin and synapsin I after 3-d incubation with a control solution,1.5 µM TTX or 40 µM bicuculline. Close-up views are also shown. (b) Normalized intensity of immunofluorescent signals of synapsin I, αN-catenin and β-catenin that have accumulated in individual synapses. *P < 0.001, as compared to control. (c) Western blots for N-cadherin, αN-catenin and β-catenin after 3-d TTX or bicuculline treatment. Relativeintensity of the αN-catenin doublet bands was also quantified (right). Thebicuculline-treated samples showed more intense signals than the control(*P < 0.001). (d–f) GFP-transfected or αN-catenin/GFP-cotransfectedneurons were treated with 1.5 µM TTX for 3 d, and then stained with anti-GFP antibody to visualize their morphology. TTX converted spines intofilopodia-like processes in control GFP-transfected neurons, whereas it failedto do so in αN-catenin-overexpressing neurons. Mean spine length after TTXtreatment and cumulative frequency plot of spine length are also shown.Scale bar = 5 µm. More than 600 spines from eight neurons were measuredfor each construct. *P < 0.001, as compared with GFP-only control.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

tor of 4×. For cells co-transfected with GFP and another gene, cells were fixedafter imaging; and then the transgene expression was verified by immunos-taining.

Image acquisition and quantification. Confocal images of immunostainedneurons were obtained with the LSM510 using a Zeiss 63× objective. Eachimage was a z-series projection taken at depth intervals of 0.36 µm. More thanten transfected neurons were chosen randomly for quantification from 5–10coverslips derived from 4–7 independent experiments for each construct. Formeasurement of spine morphology, spines located within the proximal 75-µmregion of two or three of the largest dendrites were chosen, and manuallytraced. Their length and head width were measured automatically withLSM510 software (Zeiss). To distinguish between filopodia and spines, wetriple-stained neurons with anti-PSD95 antibodies together with anti-GFPand anti-flag tag antibodies. Only the processes that contained PSD95 punctawere defined as spines in this study. For quantification of spine-length changesin time-lapse movies, we measured the length of each spine every 5 min duringa 95-min period and subtracted the minimum length from the maximallength. For quantification of long-term time-lapse analysis, we collectedimages of spines every 3 d, and traced their morphologies.

For quantification of fluorescence intensities, we triple-stained neuronswith synapsin I, αN-catenin and β-catenin. More than 20 coverslips were usedfor each condition. Confocal images of immunostained neurons wereobtained as described above, by using the LSM510 (Zeiss). The same settingsfor pinhole size, brightness and contrast were used. Fluorescence intensityanalysis was conducted as follows: catenin signals overlapping with synapsinsignals were collected by using Subtraction Macro, and these signals weredefined as the synaptic components of the catenin signals. The threshold wasset to a modest one to increase the discrimination of particles26. The samethreshold was used within the same experiments. Measured data wereexported to Excel software (Microsoft), and the data were compared by usingStudent’s t-test. Histograms showing the mean ± s.e.m. were constructed.

Immunostaining and antibodies. Immunocytochemistry of neurons was per-formed as described25 by use of the following antibodies: mouse anti-PSD-95(6G6-1C9, ABR), rabbit anti-synapsin I (Chemicon), mouse anti-GFP(7.1+11.1, Roche), rabbit anti-GFP (Chemicon), rat anti-αN-catenin (NCAT2or NCAT520,50), mouse anti-β-catenin (5H10, a gift from M.J. Wheelock,University of Nebraska Medical Center), rabbit anti-flag tag (Sigma), mouseanti-GluR2 (MAB397, Chemicon) and mouse anti-N-cadherin(Pharmingen).

Western blot analysis. Neurons cultured on each coverslip were individuallyscraped into a sample buffer and subjected to SDS-PAGE. Proteins were trans-ferred onto a nitro-cellulose membrane, and detection was performed byusing ECL Plus (Amersham Biosciences). Signals were quantified by usingScion Image (Scion). Data obtained from 26 coverslips were used for statisticalanalysis of each experimental condition. Histograms showing the mean ±standard error (s.e.m.) were constructed.


ACKNOWLEDGMENTSWe thank T. Manabe for TTX and H. Ishigami for maintaining the mice. This workwas supported by the program Grants-in-Aid for Specially Promoted Research ofthe Ministry of Education, Science, Sports, and Culture of Japan to M.T.


Received 20 January; accepted 23 February 2004Published online at http://www.nature.com/natureneuroscience/

1. Ziv, N.E. & Smith, S.J. Evidence for a role of dendritic filopodia in synaptogenesisand spine formation. Neuron 17, 91–102 (1996).

2. Jontes, J.D., Buchanan, J. & Smith, S.J. Growth cone and dendrite dynamics inzebrafish embryos: early events in synaptogenesis imaged in vivo. Nat. Neurosci. 3,231–237 (2000).

3. Okabe, S., Miwa, A. & Okado, H. Spine formation and correlated assembly of presy-naptic and postsynaptic molecules. J. Neurosci. 21, 6105–6114 (2001).

4. Marrs, G.S., Green, S.H. & Dailey, M.E. Rapid formation and remodeling of postsy-naptic densities in developing dendrites. Nat. Neurosci. 4, 1006–1013 (2001).

A R T I C L E S

system, perhaps including its interaction with cytoskeletal proteins.The overexpression of αΝ-catenin on the postsynaptic side was suffi-cient for synapse stabilization. Therefore, the above putative signalingsystem is assumed to reside on the postsynaptic side. We also shouldstress that αΝ-catenin overexpression was not accompanied byincreases in other synaptic components. In synaptic junctions, the cad-herin/catenin complex is localized in compartments separable from theactive zone20,43. It is thus possible that the αΝ-catenin-dependent sys-tem might be independent of other regulatory systems for synapseremodeling, localized in the central portion of the postsynaptic density.

Morphological plasticity in synapses often correlates with theirphysiological plasticity. For example, stimuli that cause long-termpotentiation (LTP) concomitantly induce filopodial/spine formationor their shape changes44,45, and synaptic activity regulates the numberof spines and synapses8,46,47. Because of the postnatal lethality of ourαΝ-catenin knockout mice, we could not examine whether αΝ-catenin is involved in the physiological plasticity of synapses invivo. By using different genetic methods, we should be able to uncoverthe in vivo roles of αΝ-catenin in future experiments.

METHODSMolecular construction. αE-catenin expression vector carrying the flag tag atits carboxyl end was constructed as follows: αE-catenin cDNA was amplifiedby use of a standard PCR method and subcloned into pCA-sal-flag (D. Fushimi and M.T., unpublished data) by use of a SalI linker to generatepCA-αE-catenin-flag, which expresses the C-terminal flag-tag fusion proteinunder the control of the CAG promoter48. Expression vectors for full-length ormutant αN-catenins were made as for αE-catenin, by using αN-catenincDNA, pBNCAT1b or pBNCATII (ref. 49) as a template, and the followingprimers: for full-length αN-catenin, 5′-GATATCGCCACCATGACTTCGGCAACTTCA-3′ (αN-N primer) and 5′-GTCGACGAAGGAATCCATTGCCTTG-3′ (αN-C primer); for αN1-262, αN-N primer and 5′-GTCGACGGAGGTGGCCTGAGCAGCA-3′; for αN277-954, 5′-GATATCGCCACCATGCAGCCCTGAATGAGT-3′ and αN-C primer; for αNk, αN-N primer and5′-CGGCGTCGACGAGCACACGGACTTGCTT-5′; for αN1-681, αN-Nprimer and 5′-CGGCGTCGACTGCTTTCTCCTCCTGTGGT-3′; and forαN1-870, αN-N primer and 5′-CGGCGTCGACTTTGGCTGCCTGGATGAC-3′. Expression vectors for αT-catenin were constructed as above by usingαT-catenin cDNA (PGEMTeasy-mαT-catenin (1-2979) clone7) as a template.All the constructs were checked by sequencing.

Cell culture and transfection. Hippocampal neuron cultures were prepared aspreviously described25. In brief, hippocampal neurons from E17 mice or E18Sprague-Dawley rats were plated on poly-L-lysine coated cover glasses at thedensity of 10,000 cells/cm2 and maintained in NeuroBasal medium(Invitrogen) with B27 supplements. To culture neurons from Catna2–/– mice,we collected neurons from E17 pups generated by crossing Catna2+/– parents.Those from individual hippocampi were separately cultured, and genotypingwas performed afterwards to determine the genotype in each pool of the cul-tures. Comparisons of mutant and wild-type samples were generally madeamong the cultures derived from a single littermate. cDNA transfection wasperformed by using Effectine (QIAGEN), following the manufacturer’sinstructions with optional modifications. Freshly dispersed neurons wereplated, and immediately subjected to the transfection treatments. Successfullytransfected neurons were detected by EGFP fluorescence. For TTX or bicu-culline treatment of neurons, 18 d.i.v. rat cultures were treated with 1.5 µMTTX (Wako BioProducts) or 40 µM bicuculline (Sigma) for 3 d.

Time-lapse analysis. For time-lapse imaging, live cells in culture medium weremounted in a chamber at 37 °C through which was passed a continuous flowof 5% CO2. Images were obtained with an LSM510 (Zeiss) confocal micro-scope using a 40× water immersion lens, and an additional electronic zoomingfactor of 6×. The laser power was attenuated to 0.1% to reduce phototoxicity.Z-stacked images at 0.8-µm depth intervals were taken every 5 min. For long-term time-lapse imaging, live cell images of the same dendrite were obtainedevery 3 d by using a 40× water immersion lens with an additional zooming fac-


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

5. Dunaevsky, A., Tashiro, A., Majewska, A., Mason, C. & Yuste, R. Developmental reg-ulation of spine motility in the mammalian central nervous system. Proc. Natl. Acad.Sci. USA 96, 13438–13443 (1999).

6. Korkotian, E. & Segal, M. Regulation of dendritic spine motility in cultured hip-pocampal neurons. J. Neurosci. 21, 6115–6124 (2001).

7. Lendvai, B., Stern, E.A., Chen, B. & Svoboda, K. Experience-dependent plasticity ofdendritic spines in the developing rat barrel cortex in vivo. Nature 404, 876–881(2000).

8. Yuste, R. & Bonhoeffer, T. Morphological changes in dendritic spines associatedwith long-term synaptic plasticity. Annu. Rev. Neurosci. 24, 1071–1089 (2001).

9. Tashiro, A., Minden, A. & Yuste, R. Regulation of dendritic spine morphology by therho family of small GTPases: antagonistic roles of Rac and Rho. Cereb. Cortex 10,927–938 (2000).

10. Nakayama, A.Y., Harms, M.B. & Luo, L. Small GTPases Rac and Rho in the mainte-nance of dendritic spines and branches in hippocampal pyramidal neurons. J. Neurosci. 20, 5329–5338 (2000).


12. Pak, D.T., Yang, S., Rudolph-Correia, S., Kim, E. & Sheng, M. Regulation of den-dritic spine morphology by SPAR, a PSD-95-associated RapGAP. Neuron 31,289–303 (2001).

13. Hayashi, K. & Shirao, T. Change in the shape of dendritic spines caused by overex-pression of drebrin in cultured cortical neurons. J. Neurosci. 19, 3918–3925(1999).

14. El-Husseini, A.E., Schnell, E., Chetkovich, D.M., Nicoll, R.A. & Bredt, D.S. PSD-95involvement in maturation of excitatory synapses. Science 290, 1364–1368(2000).

15. Ethell, I.M. & Yamaguchi, Y. Cell surface heparan sulfate proteoglycan syndecan-2induces the maturation of dendritic spines in rat hippocampal neurons. J. Cell Biol.144, 575–586 (1999).

16. Penzes, P. et al. Rapid induction of dendritic spine morphogenesis by trans-synapticephrinB-EphB receptor activation of the Rho-GEF kalirin. Neuron 37, 263–274(2003).

17. Passafaro, M., Nakagawa, T., Sala, C. & Sheng, M. Induction of dendritic spines byan extracellular domain of AMPA receptor subunit GluR2. Nature 424, 677–681(2003).

18. Fischer, M., Kaech, S., Wagner, U., Brinkhaus, H. & Matus, A. Glutamate receptorsregulate actin-based plasticity in dendritic spines. Nat. Neurosci. 3, 887–894(2000).

19. Yamagata, M., Herman, J.P. & Sanes, J.R. Lamina-specific expression of adhesionmolecules in developing chick optic tectum. J. Neurosci. 15, 4556–4571 (1995).

20. Uchida, N., Honjo, Y., Johnson, K.R., Wheelock, M.J. & Takeichi, M. Thecatenin/cadherin adhesion system is localized in synaptic junctions bordering trans-mitter release zones. J. Cell Biol. 135, 767–779 (1996).

21. Fannon, A.M. & Colman, D.R. A model for central synaptic junctional complex for-mation based on the differential adhesive specificities of the cadherins. Neuron 17,423–434 (1996).

22. Benson, D.L. & Tanaka, H. N-cadherin redistribution during synaptogenesis in hip-pocampal neurons. J. Neurosci. 18, 6892–6904 (1998).

23. Huntley, G.W. & Benson, D.L. Neural (N)-cadherin at developing thalamocorticalsynapses provides an adhesion mechanism for the formation of somatopically organ-ized connections. J. Comp. Neurol. 407, 453–471 (1999).

24. Iwai, Y. et al. DN-cadherin is required for spatial arrangement of nerve terminals andultrastructural organization of synapses. Mol. Cell. Neurosci. 19, 375–388 (2002).

25. Togashi, H. et al. Cadherin regulates dendritic spine morphogenesis. Neuron 35,77–89 (2002).

26. Bozdagi, O., Shan, W., Tanaka, H., Benson, D.L. & Huntley, G.W. Increasing num-bers of synaptic puncta during late-phase LTP: N-cadherin is synthesized, recruitedto synaptic sites, and required for potentiation. Neuron 28, 245–259 (2000).

A R T I C L E S

27. Fischer, M., Kaech, S., Knutti, D. & Matus, A. Rapid actin-based plasticity in den-dritic spines. Neuron 20, 847–854 (1998).

28. Yu, X. & Malenka, R.C. β-catenin is critical for dendritic morphogenesis. Nat.Neurosci. 6, 1169–1177 (2003).

29. Ebihara, T., Kawabata, I., Usui, S., Sobue, K. & Okabe, S. Synchronized formationand remodeling of postsynaptic densities: long-term visualization of hippocampalneurons expressing postsynaptic density proteins tagged with green fluorescent pro-tein. J. Neurosci. 23, 2170–2181 (2003).

30. Provost, E. & Rimm, D.L. Controversies at the cytoplasmic face of the cadherin-based adhesion complex. Curr. Opin. Cell Biol. 11, 567–572 (1999).

31. Nagafuchi, A. Molecular architecture of adherens junctions. Curr. Opin. Cell Biol.13, 600–603 (2001).

32. Papa, M. & Segal, M. Morphological plasticity in dendritic spines of cultured hip-pocampal neurons. Neuroscience 71, 1005–1011 (1996).

33. Trachtenberg, J.T. et al. Long-term in vivo imaging of experience-dependent synapticplasticity in adult cortex. Nature 420, 788–794 (2002).

34. Grutzendler, J., Kasthuri, N. & Gan, W.B. Long-term dendritic spine stability in theadult cortex. Nature 420, 812–816 (2002).

35. Matus, A. Actin-based plasticity in dendritic spines. Science 290, 754–758(2000).

36. Bek, S. & Kemler, R. Protein kinase CKII regulates the interaction of β-catenin withα-catenin and its protein stability. J. Cell Sci. 115, 4743–4753 (2002).

37. Piedra, J. et al. p120-catenin-associated Fer and Fyn tyrosine kinases regulate β-catenin Tyr-142 phosphorylation and β-catenin-α-catenin interaction. Mol. Cell Biol.23, 2287–2297 (2003).

38. Murase, S., Mosser, E. & Schuman, E.M. Depolarization drives β-catenin into neu-ronal spines promoting changes in synaptic structure and function. Neuron 35,91–105 (2002).

39. Rimm, D.L., Koslov, E.R., Kebriaei, P., Cianci, C.D. & Morrow, J.S. α1(E)-cateninis an actin-binding and -bundling protein mediating the attachment of F-actin tothe membrane adhesion complex. Proc. Natl. Acad. Sci. USA 92, 8813–8817(1995).

40. Itoh, M., Nagafuchi, A., Moroi, S. & Tsukita, S. Involvement of ZO-1 in cadherin-based cell adhesion through its direct binding to α-catenin and actin filaments. J. Cell Biol. 138, 181–192 (1997).

41. Kussel-Andermann, P. et al. Vezatin, a novel transmembrane protein, bridges myosinVIIA to the cadherin-catenins complex. Embo J. 19, 6020–6029 (2000).

42. Ackermann, M. & Matus, A. Activity-induced targeting of profilin and stabilization ofdendritic spine morphology. Nat. Neurosci. 6, 1194–1200 (2003).

43. Mizoguchi, A. et al. Nectin: an adhesion molecule involved in formation of synapses.J. Cell Biol. 156, 555–565 (2002).


45. Engert, F. & Bonhoeffer, T. Dendritic spine changes associated with hippocampallong-term synaptic plasticity. Nature 399, 66–70 (1999).

46. Hering, H. & Sheng, M. Dendritic spines: structure, dynamics and regulation. Nat.Rev. Neurosci. 2, 880–888 (2001).

47. Nimchinsky, E.A., Sabatini, B.L. & Svoboda, K. Structure and function of dendriticspines. Annu. Rev. Physiol. 64, 313–353 (2002).

48. Niwa, H., Yamamura, K. & Miyazaki, J. Efficient selection for high-expression trans-fectants with a novel eukaryotic vector. Gene 108, 193–199 (1991).

49. Uchida, N. et al. Mouse αN-catenin: two isoforms, specific expression in the nerv-ous system, and chromosomal localization of the gene. Dev. Biol. 163, 75–85(1994).

50. Hirano, S., Kimoto, N., Shimoyama, Y., Hirohashi, S. & Takeichi, M. Identification ofa neural α-catenin as a key regulator of cadherin function and multicellular organi-zation. Cell 70, 293–301 (1992).


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

MRX is a disorder characterized by cognitive impairment without anyother distinctive clinical features. A major challenge has been touncover the molecular causes of MRX and the underlying cellularmechanisms responsible for reduced cognitive function. Eleven genesinvolved in MRX have been identified to date and notably, three ofthese encode regulators or effectors of the Rho subfamily of smallGTP-binding proteins1–3. Members of this family, including RhoA,Rac and Cdc42, are key regulators of the actin cytoskeleton and affectmany aspects of neuronal development and morphogenesis3–7. Thethree Rho-linked MRX genes encode (i) oligophrenin-1, a Rho-GTPase activating protein (Rho-GAP)8, (ii) PAK3 (p21-activatedkinase-3), a serine/threonine kinase downstream of Rac and Cdc42(ref. 9) and (iii) ARHGEF6, a Rac GTPase exchange factor also knownas αPIX or Cool-2 (ref. 10). The association between mutations inRho-linked genes and MRX highlights the importance of Rho pro-teins in neuronal function and has led to the hypothesis that abnor-malities in Rho signaling may be a cause of MRX3. Studies examiningthe effects of these mutations on neuronal signaling and developmentare therefore critical for the elucidation of cellular mechanismsunderlying normal cognitive function and disease.

Here we focus on the functional characterization of oligophrenin-1(encoded by the OPHN1 gene in humans; Ophn-1 in mice), a proteinwith a Rho-GAP domain shown to negatively regulate RhoA, Rac andCdc42 in vitro and in non-neuronal cells8,11. OPHN1 was identifiedby the analysis of a balanced translocation t(X;12) observed in afemale patient with mild mental retardation. Its involvement in MRXwas established by the identification of a mutation within the OPHN1coding sequence in a family with MRX (MRX 60). In these two cases,the OPHN1 mutation is associated with a loss of, or dramatic reduc-tion in, mRNA product8. Recent studies show that oligophrenin-1 is

present in neuronal and astroglial cells and that it colocalizes withactin at the tip of growing neurites11. However, the function ofoligophrenin-1 in the brain is unknown and it remains to be seen howmutations in OPHN1 affect neuronal development and function, andcontribute to MRX.

To begin to understand how oligophrenin-1 deficiency affects neu-ronal function, we examined the effects of reducing oligophrenin-1levels on the morphology of developing hippocampal (CA1) neuronsin organotypic slices. We focused on dendritic spines, the main sitesof excitatory synapses in the brain12, because changes in spine dimen-sions and density have been associated with synaptic plasticity13–16

and learning17, as well as with neurological disorders including men-tal retardation18–20. Using RNA interference (RNAi) and antisenseRNA approaches, we found that downregulation of oligophrenin-1 inCA1 neurons resulted in a significant shortening of dendritic spines.We showed that this spine length phenotype was mediated by theRhoA/Rho-kinase signaling pathway, acting downstream ofoligophrenin-1. Furthermore, we identified an interaction betweenoligophrenin-1 and Homer, placing oligophrenin-1 within a post-synaptic complex that potentially links oligophrenin-1 to glutamatereceptor signaling. Our findings suggest how OPHN1 mutations maycompromise cognitive function.

RESULTSOligophrenin-1 in the brainWe began the characterization of oligophrenin-1 by determining itspresence, distribution and subcellular localization in the brain.Immunoblot experiments of rat tissues showed that oligophrenin-1protein levels were highest in the brain, but were also detectable tovariable extents in other tissues (Fig. 1a). Oligophrenin-1 levels were

1Cold Spring Harbor Laboratory, 1 Bungtown Road, Cold Spring Harbor, New York 11724, USA. 2Molecular and Cellular Biology Program, State University of New Yorkat Stony Brook, Stony Brook, New York 11794, USA. 3These authors contributed equally to this work. Correspondence should be addressed to L.V.A.([email protected]).


The X-linked mental retardation protein oligophrenin-1is required for dendritic spine morphogenesisEve-Ellen Govek1–3, Sarah E Newey1,3, Colin J Akerman1, Justin R Cross1, Lieven Van der Veken1 & Linda Van Aelst1,2

Of 11 genes involved in nonspecific X-linked mental retardation (MRX), three encode regulators or effectors of the Rho GTPases,suggesting an important role for Rho signaling in cognitive function. It remains unknown, however, how mutations in Rho-linkedgenes lead to MRX. Here we report that oligophrenin-1, a Rho-GTPase activating protein that is absent in a family affected withMRX, is required for dendritic spine morphogenesis. Using RNA interference and antisense RNA approaches, we show thatknock-down of oligophrenin-1 levels in CA1 neurons in rat hippocampal slices significantly decreases spine length. Thisphenotype can be recapitulated using an activated form of RhoA and rescued by inhibiting Rho-kinase, indicating that reducedoligophrenin-1 levels affect spine length by increasing RhoA and Rho-kinase activities. We further demonstrate an interactionbetween oligophrenin-1 and the postsynaptic adaptor protein Homer. Our findings provide the first insight into how mutations ina Rho-linked MRX gene may compromise neuronal function.

A R T I C L E S


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

synaptophysin (Fig. 1f). Further complemen-tary biochemical experiments confirmed thata proportion of endogenous oligophrenin-1 islocalized at the synapse. We isolated synapto-somal plasma membranes (SPM) from rathippocampi, and subsequently prepared PSDfractions, which showed that oligophrenin-1 isconcentrated in the insoluble PSD pellets aftertwo rounds of Triton extraction. Furtherextraction of Triton-insoluble complexes withN-lauroylsarcosine solubilized approximately50% of the oligophrenin-1 present, leaving50% in the PSD core. As expected, PSD-95 isconcentrated in all insoluble, detergent-extracted PSD pellets, whereas synaptophysin

is solubilized upon treatment of SPM with Triton (data shown below inFig. 7b). Together, these data reveal a postsynaptic localization foroligophrenin-1 and its presence in synapses on dendritic spines.

Our studies also showed that oligophrenin-1 is present at presynap-tic sites. Endogenous oligophrenin-1 immunolabeling was observed asnumerous puncta along axons and overlapped with synaptophysin atsome presynaptic terminals (Fig. 1e). Furthermore, T7-oligophrenin-1was concentrated in axonal synaptic boutons that stained positive forsynaptophysin (Fig. 1f, bottom row). Our findings that oligophrenin-1is present both pre- and postsynaptically in neurons suggest an impor-tant role for oligophrenin-1 in synaptic function.

Knock-down of oligophrenin-1 in neuronsBecause the mutation in OPHN1 associated with MRX results in a dra-matic reduction in OPHN1 mRNA product8, we wanted to assess theeffects of oligophrenin-1 knock-down on neuronal development andsignaling. We developed two strategies to interfere with oligophrenin-1expression in primary neurons: RNAi and an antisense approach. ForRNAi, two siRNA (small interfering RNA) duplexes were designed, onefrom the cDNA coding region (Ophn1#1) and the other from the 3′untranslated region (Ophn1#2) of rat and mouse Ophn-1. The efficacy

similar in all brain regions examined and the protein was present inboth embryonic and adult tissue (Fig. 1b,c). Furthermore, virtually allneuronal populations, including pyramidal neurons of the hip-pocampus and cerebral cortex, stained positive for oligophrenin-1(Fig. 1d). In these neurons, oligophrenin-1 was concentrated in thecell bodies and extended out into the dendrites. It was also detected inblood vessels (Fig. 1d, arrows) where it localized to vascular endothe-lial cells (data not shown). These results show that oligophrenin-1 ispresent in neurons in major regions of the brain, including those per-tinent to learning and memory.

To determine the subcellular localization of oligophrenin-1 in neu-rons, we examined the distribution of endogenous and ectopicallyexpressed (T7-tagged) oligophrenin-1 in cultured hippocampal neu-rons that have mature synapses. We found high levels of both in the cellbody, and in abundant puncta in axons, dendrites and dendritic spines(Fig. 1e,f). The presence of oligophrenin-1 at postsynaptic sites wasconfirmed by its colocalization with F-actin, which accumulates inspines (Fig. 1e,f), and its overlapping localization with PSD-95, a majorcomponent of the postsynaptic density (PSD; Fig. 1e,f). Consistentwith a postsynaptic localization, T7-oligophrenin-1 in dendritic spinesclosely juxtaposed immunolabeling with the presynaptic marker

A R T I C L E S


Figure 1 Distribution and localization ofoligophrenin-1 in the brain. (a) Multiple-tissuewestern blot from a postnatal day 30 (P30) ratprobed with the oligophrenin-1 antibody 1296.Oligophrenin-1 has a molecular mass ofapproximately 92 kDa. B, brain; SM, skeletalmuscle; H, heart; Li, liver; Lu, lung; Ki, kidney.(b) Immunoblot of P30 rat brain regions: Cer,cerebellum; Hip, hippocampus; Thal, thalamus;Flob, frontal lobes; Cor, sensory cortex; Olf,olfactory bulb. (c) Immunoblot of embryonic day18 (E18), P2, P6, P10, P30 and adult (>8-weekold) rat brains. (d) Parasagittal brain sectionsfrom a P30 rat double-immunolabeled withantibody 1296 (OPHN1, red staining) and anantibody to the neuronal marker NeuN (greenstaining). Arrows indicate immunolabeling ofblood vessels. Scale bar, 50 µm. (e) Left,hippocampal neuron at 21 d.i.v. immunolabeledwith the oligophrenin-1 antibody 1296 (OPHN1).Right, these neurons coimmunostained for actin,PSD-95 and synaptophysin. (f) Left,hippocampal neuron (21 d.i.v.) expressing T7-oligophrenin-1 (T7-OPHN1) labeled with a T7-specific antibody. Right, these neuronscoimmunostained for actin, PSD-95 andsynaptophysin. Scale bars (e,f), 10 µm.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

of these siRNAs in reducing oligophrenin-1 levels was first tested in ratembryonic fibroblast (REF52) cells, an easily transfectable cell line, andboth were successful in reducing oligophrenin-1 levels compared to anegative control duplex siRNA (Fig. 2a). The Ophn-1 and controlsiRNAs were subsequently transfected into dissociated hippocampalneurons at 5 or 6 days in vitro (d.i.v.), and 48 h after transfection, bothOphn-1 siRNAs, but not the control siRNAs, dramatically reducedoligophrenin-1 levels in these cells (Fig. 2b,c). These observations wereconfirmed by immunofluorescence labeling of cultured hippocampalneurons transfected at 6 d.i.v. with either Ophn1#2 siRNA or controlsiRNA. Neurons were double-labeled for oligophrenin-1 and a cellmarker, β-tubulin. Transfection of Ophn1#2 siRNA, but not controlsiRNA, led to a severe reduction in the intensity of oligophrenin-1immunolabeling in the cell body and almost completely abolishedoligophrenin-1 expression in neuronal processes (Fig. 2e). This effectwas specific, as β-tubulin expression was unaffected in both cases.Similarly, we observed a 30% reduction in oligophrenin-1 levels inmature neuronal cultures (15 d.i.v.) transfected with Ophn1#2 siRNA(Fig. 2c). These experiments demonstrate that the selected Ophn-1siRNAs are effective in reducing oligophrenin-1 levels in primary hip-pocampal neurons.

For the antisense approach, the full-length mouse Ophn-1 cDNA wascloned in the antisense direction into a mammalian expression vector.This construct (termed ‘antisense’) successfully knocked downoligophrenin-1 levels in REF52 cells, as seen in our immuno-blot analy-sis (Fig. 2a). Subsequent transfection of the antisense construct intohippocampal neurons also resulted in a significant reduction inoligophrenin-1 levels 48 h post-transfection (Fig. 2b,d). Taken together,we established two independent approaches to effectively knock downoligophrenin-1 expression in primary hippocampal neurons.

Oligophrenin-1 knock-down affects spine morphologyGiven the importance of the structure and shape of dendritic spines forsynaptic function12–16, and the presence of oligophrenin-1 in spines, we

assessed what effect oligophrenin-1 knock-down has on dendritic spinemorphology in CA1 pyramidal cells of rat organotypic hippocampalslices. Hippocampal slices were biolistically transfected with a GFPexpression vector alone, or with a GFP expression vector and one of thetwo Ophn-1 siRNAs, Ophn1#1 or Ophn1#2, or the control siRNA. Usingthis method, we observed that on average, two or three CA1 neurons,and a similar number of CA3 neurons, are transfected (SupplementaryFig. 1 online). GFP labeled dendrites and spines from transfected CA1neurons were imaged 48 hours post-transfection, using two-photonlaser scanning microscopy. These cells have already acquired their char-acteristic dendritic branching pattern and display numerous protru-sions from their dendrites (Fig. 3a). A minority of these protrusions arefilopodial (long and headless), however, a large fraction have a well-defined neck and head structure, characteristic of mature spines21.

We measured the length and number of spines (all protrusionsincluding filopodia) on primary and secondary dendrites (seeMethods). Both Ophn-1 siRNAs were found to significantly decreasespine length when compared to control siRNA transfected neurons (P < 0.0001 for both). The mean spine lengths for Ophn1#1 andOphn1#2 siRNA–transfected neurons were 21% and 18% smaller,respectively, than that for control siRNA-transfected neurons (Fig. 3c,d). There was no significant difference in spine lengthbetween cells transfected with a GFP expression vector alone and con-trol siRNA (P = 0.63; Fig. 3b). Similar results were obtained for spinelengths analyzed from primary and secondary dendrites separately(Supplementary Table 1 online). Thus, two different siRNAs targetedto two different regions of the Ophn-1 transcript reduced spinelength, whereas the control siRNA did not. In contrast to the effect onspine length, we found that reduced oligophrenin-1 levels did notaffect the density of spines (including filopodia), nor the density offilopodia (defined as headless protrusions longer than 2 µm) whenanalyzed separately (Supplementary Table 2 online).

To confirm these findings using an independent approach, we biolis-tically transfected the Ophn-1 antisense construct or an empty plasmid,

A R T I C L E S


Figure 2 Knock-down of oligophrenin-1 levelsusing RNAi and antisense RNA technologies.(a) Western blot of oligophrenin-1 levels inREF52 cells transfected with Ophn-1 siRNAs(Ophn1#1 and Ophn1#2), control siRNA (control)or no siRNA (none), and with either control vectoror antisense. The immunoblot was probed with1296 and anti-ERK2 as a loading control. (b) Western blots of oligophrenin-1 levels inyoung primary hippocampal neurons transfectedwith Ophn1#1 siRNA, Ophn1#2 siRNA, controlsiRNA, β-tubulin siRNA (β-tub) or no siRNA, andwith either control vector or antisense. Blots wereprobed with 1296 and anti-β-tubulin as a loadingcontrol. (c) Mean oligophrenin-1 levels in younghippocampal neurons transfected with Ophn-1siRNAs or antisense expressed as a percentage ofcontrol transfections. Oligophrenin-1 levels werenormalized to β-tubulin levels in the samesample. Ophn1#1 and Ophn1#2 siRNAssignificantly decreased oligophrenin-1 levels (t-test, P < 0.002, n = 5 in both cases), as didthe antisense construct (t-test, P = 0.003, n = 5;*denotes statistical significance). Error barsindicate standard error of the mean (s.e.m.). (d) Double immunofluorescence labeling ofneurons transfected with Ophn1#2 siRNA (bottom) or control siRNA (top). The right panel shows oligophrenin-1 immunostaining detected with 1724 (anti-OPHN1); the left panel shows the same cells labeled for β-tubulin (anti-β-tub) as a cell marker. Scale bar, 10 µm. (e) Western blots of oligophrenin-1 levelsin mature hippocampal neurons transfected with Ophn1#2 siRNA or no siRNA. Blots were probed as in b.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

together with a GFP expression vector, into CA1 neurons in hippocam-pal slices. Consistent with the Ophn-1 siRNA data, neurons expressingantisense RNA showed significantly shorter spines compared to cellstransfected with control vector (P < 0.0001). The mean spine length forneurons expressing antisense RNA was 12% smaller than for neuronsexpressing control vector (Fig. 4a). No significant difference in spinelength was observed between cells transfected with the GFP expressionvector alone and cells transfected with control vector (P = 0.14; Fig. 4b).Similar results were obtained for spines analyzed from primary and sec-ondary dendrites separately (Supplementary Table 1). As in the siRNAexperiments, mean spine and filopodia density did not differ signifi-cantly between neurons transfected with control vector and thosetransfected with antisense or GFP expression vector alone(Supplementary Table 2). Thus, by two independent approaches, wefound that knocking down oligophrenin-1 levels in hippocampal CA1neurons causes a decrease in dendritic spine length. Importantly, achange in spine length of comparable magnitude, albeit in the oppositedirection, is reported in a mouse model of fragile X syndrome22,23,indicating that morphological changes of this degree can ultimatelylead to deficits in cognitive function.

Oligophrenin-1 affects the RhoA/Rho-kinase signaling pathwayOligophrenin-1 contains a Rho-GAP domain that negatively regulatesthe activity of Rho GTPases8,11. Therefore, a likely mechanism by which

loss of oligophrenin-1 could reduce dendritic spine length is by increas-ing the activity of one or more of the Rho-GTPase family members. Todetermine which Rho GTPase family member(s) oligophrenin-1 actsupon in a neuronal context, we expressed full-length oligophrenin-1 inPC12 cells and found that oligophrenin-1 decreased global levels ofactive GTP-bound RhoA, Rac1 and Cdc42 (Fig. 5a). Although theseGTPase pull-down assays suggest that oligophrenin-1 can act as a GAPfor all three Rho GTPases, they do no take into account spatial regula-tion of the Rho GTPases that likely occurs in fully differentiated neu-rons. The GAP activity of oligophrenin-1 may be restricted in dendritesand spines by the local availability or activity of the target GTPaseand/or the presence of additional factors.

A R T I C L E S


Figure 3 Oligophrenin-1 siRNAs cause areduction in spine length in CA1 pyramidal cellsin hippocampal slices. (a) Example of abiolistically transfected hippocampal CA1pyramidal cell expressing GFP. The first panelshows the typical morphology of the CA1 cells,the second panel is an enlargement of the basaldendrites, and the third panel (the inset) showsa digital zoom of spines (indicated by arrows)used to count spine numbers and lengths. (b–d) The line graphs show the cumulativefrequency distribution of spine lengths forneurons transfected with control siRNA (black)and neurons transfected with a GFP expressionvector alone, Ophn1#1 siRNA or Ophn1#2siRNA (white). Insets show the mean spinelength for the two groups. Error bars indicates.e.m. Spine length was significantly shorter forOphn1#1 siRNA–transfected neurons than forcontrol siRNA–transfected neurons (P < 0.0001, 1,533 control siRNA spines,2,805 Ophn1#1 siRNA spines), and forOphn1#2 siRNA–transfected neurons comparedto control siRNA–transfected neurons (P <0.0001, 1,533 control siRNA spines, 3,170 Ophn1#2 siRNA spines). No significant difference in spine length was observed between cells transfectedwith the GFP expression vector alone and control siRNA–transfected cells (P = 0.63, 1,083 GFP spines, 1,533 control siRNA spines). Scale bar, 5 µm.

Figure 4 Ophn-1 antisense RNA causes a reduction in spine length in CA1pyramidal cells in hippocampal slices. (a,b) The line graphs show thecumulative frequency distribution of spine lengths for control vectortransfected neurons (black) and neurons transfected with antisense or a GFPexpression vector alone (white). Insets show the mean spine length for thetwo groups. Error bars indicate s.e.m. Spine length was significantly shorterfor antisense transfected neurons than for control vector transfected neurons(P < 0.0001, 2,508 vector spines, 2,011 antisense spines). No significantdifference in spine length was observed between cells transfected with aGFP expression vector alone and control vector transfected cells (P = 0.14;1,083 GFP spines, 2,508 vector spines). Scale bar, 5 µm.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

(P < 0.0001; Fig. 6a). Y-27632 treatment of hippocampal slices trans-fected with control duplex siRNA did not significantly increase spinelength compared to untreated slices (P = 0.11; Fig. 6a). These findingsindicate that the RhoA/Rho-kinase signaling pathway mediates theaction of oligophrenin-1 knock-down on dendritic spine length inCA1 pyramidal neurons.

The lack of phenotype in control slices treated with Y-27632implies that the RhoA/Rho-kinase pathway is repressed under physio-logical conditions6. Examination of the effects of ectopic expressionof full-length oligophrenin-1 on spine length yielded further supportfor this repression model. We found that expression ofT7-oligophrenin-1 in CA1 cells in hippocampal slices did not result inan increase in spine length compared to neurons expressing a controlvector (P = 0.14; Fig. 6b). Thus we propose that oligophrenin-1 nor-mally acts to repress the RhoA/Rho-kinase pathway to maintain spinelength. Upon removal of oligophrenin-1, inhibition of RhoA isrelieved, resulting in activation of Rho-kinase and a concomitantdecrease in spine length.

Oligophrenin-1 interacts with HomerIn addition to determining the signaling pathways downstream ofoligophrenin-1, we uncovered an interaction that provides a possibleconnection between oligophrenin-1 and receptor signaling at postsy-naptic sites. Within the amino acid sequence of oligophrenin-1, wenoticed the sequence PPLEF (residues 4–8) that corresponds to theconsensus motif found in proteins that bind to the EVH1 domain ofHomer proteins26 (Fig. 7a). Homer proteins are adaptor molecules thatlink glutamate receptors to multiple intracellular targets and influencedendritic spine morphogenesis and synaptic transmission27,28. Theidentification of a Homer binding motif, together with our finding thatoligophrenin-1 and Homer cofractionate in the PSD core (Fig. 7b),suggested a potential interaction between these two molecules. Wetherefore used pull-down assays to investigate whether oligophrenin-1and Homer associate biochemically. Beads loaded with Homer1b- andHomer1c-GST fusion proteins and incubated with lysates from rat hip-pocampi and cortices successfully pulled down oligophrenin-1 (Fig. 7c). We further used this assay to show that mutation of the con-

As a first step toward addressing the relationship betweenoligophrenin-1 and the Rho GTPases in developing hippocampal neu-rons, we expressed constitutively active Rho GTPases in CA1 neuronsin hippocampal slices and compared the resulting spine phenotypeswith that obtained upon oligophrenin-1 suppression. Expression of aconstitutively active RhoA mutant (RhoAV14) reduced spine lengthand density, whereas constitutively active Rac1 (Rac1V12) caused theformation of numerous abnormal lamellipodia-like protrusions. Aconstitutively active Cdc42 mutant (Cdc42V12) did not have anobservable effect on spine morphology (Fig. 5b). These findings areconsistent with studies by other groups6,7. It is striking that a knock-down of oligophrenin-1 most closely mimicked the effect of an acti-vated RhoA mutant with regard to changes in spine length, supportingthe idea that oligophrenin-1 acts predominantly upon RhoA withrespect to spine morphogenesis. Although reduced oligophrenin-1 lev-els did not decrease spine density as seen for constitutively activeRhoA6,7 (unpublished data), it is likely that levels of RhoA-GTP arehigher in neurons overexpressing a potent activated RhoA mutant thanin cells with reduced oligophrenin-1 levels, resulting in a more severephenotype. Alternatively, oligophrenin-1 may be just one of severalnegative regulators acting on RhoA in dendritic spines, and thereforeloss of only one such regulator may result in a less extreme phenotype.

To confirm that oligophrenin-1 affects spine length by acting onthe RhoA signaling pathway, we tested whether inhibiting this path-way can rescue the reduced spine length resulting from oligophrenin-1 knock-down. For these experiments, we made use of the Rho-kinaseinhibitor Y-27632 (ref. 24). Rho-kinase is a major downstream targetof RhoA that is involved in neurite retraction and axonogenesis25.Recent studies revealed that treatment of hippocampal neurons withY-27632 rescued dendritic simplification and reduced spine densityinduced by RhoAV14, whereas treatment of control neurons with Y-27632 did not have a pronounced effect6. Importantly, we foundthat Y-27632 treatment of hippocampal slices transfected with anOphn-1 duplex siRNA largely rescued the oligophrenin-1 knock-down effect on spine length. The mean spine length of neurons trans-fected with Ophn1#2 siRNA and treated with Y-27632 was 15% largerthan that for untreated Ophn1#2 siRNA transfected neurons

A R T I C L E S


Figure 5 Rho GTPase activation assays andactivated Rho GTPase spine phenotypes. (a)Rhotekin RhoA GTPase and Pak Rac/Cdc42GTPase activation assays show that oligophrenin-1is capable of acting as a Rho GAP for all threeGTPases in PC12 cells. Serum starved cellsexpressing oligophrenin-1/mycHis (OPHN1) orvector were left untreated or were treated with 1µM lysophosphatidic acid (LPA) for 3 min or 200ng/ml epidermal growth factor (EGF) for 1 min asindicated. Pulled-down GTP-bound GTPase wascompared to total GTPase in the cell lysates.GTPase activity for oligophrenin-1 transfectedcells is represented as a percentage of theGTPase activity of the control vector transfectedcells. Error bars indicate standard deviation(s.d.). (b) Activated Rho GTPase spinephenotypes in rat hippocampal CA1 pyramidalcells in slices expressing constitutively active(CA) Rho GTPase mutants. CA RhoA (RhoAV14)results in reduced protrusion length, whereas CARac1 (Rac1V12) causes the formation ofabnormal lamellipodia/veil-like protrusions. CACdc42 (Cdc42V12) did not show any majoreffects on spine morphology. Scale bar, 5 µm.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

sensus Homer binding motif in oligophrenin-1 disrupts Homer bind-ing. Using HEK 293 cells, we expressed either wild-type or mutant T7-oligophrenin-1, which contained mutations P4L and F8C in the Homerbinding motif. Mutant oligophrenin-1 showed markedly less binding toimmobilized Homer1b than did wild-type oligophrenin-1 (Fig. 7d).Finally, to demonstrate an in vivo association between oligophrenin-1and Homer, we coimmunoprecipitated the endogenous proteins fromrat hippocampal lysates using antibodies against both Homer andoligophrenin-1 (Fig. 7e). Together with our findings above, whichimplicate the RhoA/Rho-kinase pathway downstream of oligophrenin-1, the association between oligophrenin-1 and Homer may provide aconnection between glutamate receptor signaling and actin cytoskeletalrearrangements necessary for morphological spine changes.

DISCUSSIONMutations in single genes that cause cognitive deficits provide aunique opportunity to uncover the molecular and cellular processesthat contribute to normal brain function. To date, no information isavailable as to how mutations in Rho-linked MRX genes impact theneuronal morphology of affected individuals, and data from knock-out mouse models have not been reported. OPHN1 was the first Rho-linked MRX gene identified, and mutations in the gene are found toresult in reduced OPHN1 expression. Here we show that knockingdown oligophrenin-1 significantly decreased dendritic spine length inCA1 pyramidal neurons. Importantly, spine morphological changesof the same magnitude have been reported for a mouse model of frag-ile X22,23, indicating that such changes can compromise synaptic plas-ticity29 and potentially lead to learning and memory deficits.

We have determined that the RhoA/Rho-kinase signaling pathwaymediates the action of oligophrenin-1 knockdown on dendriticspine length. Our data suggest that this pathway, although intact, isrepressed in pyramidal neurons under physiological conditions6,30

and that endogenous oligophrenin-1 acts to repress the RhoA sig-naling pathway to maintain spine length. When this repression isalleviated by loss of oligophrenin-1, there is a subsequent increase in

RhoA and Rho-kinase activities, causing a reduction in spine length.This reduction in spine length is likely brought about by phospho-rylation of myosin light chain, either directly by Rho-kinase or indi-rectly through Rho-kinase phosphorylation and inactivation ofmyosin light chain phosphatase, leading to an increase in acto-myosin contractility25,30.

We further uncovered an interaction between oligophrenin-1 andHomer, a protein involved in dendritic spine morphogenesis andsynaptic transmission27,28. Homer proteins are key adaptor proteinsat the PSD where they organize glutamate receptor signaling com-plexes. The EVH1 domain of Homer proteins interacts with a con-served Homer binding motif, PPxxF, in a variety of binding partners,including type-I metabotropic glutamate receptors (mGluRs), inosi-tol 1,4,5 tris-phosphate receptors (IP3Rs) and Shank, a scaffoldingmolecule that indirectly links Homer to NMDA-type glutamatereceptors27,28. Interestingly, Homer cooperates with Shank to inducespine enlargement, while a dominant-negative form of Homer(Homer 1a) decreases the size and length of spines and interferes withsynaptic transmission31,32. An intriguing possibility is thatoligophrenin-1, as a regulator of the RhoA/Rho-kinase pathway indifferentiated neurons, provides a crucial link between postsynapticreceptors (via Homer) and the actin cytoskeleton to regulate den-dritic spine morphogenesis. This is consistent with extensive evidencethat glutamate receptor activation affects the stability of actin, spinemorphology and spine maintenance33–36. Additionally, glutamatergicsynaptic activity has been shown to regulate RhoA activity, whichaffects dendritic arbor stabilization37. It is therefore tempting to spec-ulate that oligophrenin-1 may act downstream of glutamatergicreceptors to regulate RhoA activity in spines, and thus contribute totheir stabilization.

Although our data are consistent with a model in which loss ofoligophrenin-1 elicits spine length changes by regulating the actincytoskeleton through the RhoA/Rho-kinase pathway, it is possiblethat calcium dynamics and signaling contribute to the spine lengthchange we observed. This is particularly relevant given that Homer

A R T I C L E S


Figure 6 Oligophrenin-1 affects the RhoA/Rho-kinase signaling pathway. (a) Inhibition of Rho-kinase rescues the decrease in spine lengthresulting from a knock-down of oligophrenin-1levels in CA1 neurons in hippocampal slices. Leftgraph shows the mean spine length for controlsiRNA transfected neurons (con), Ophn1#2transfected neurons and Ophn1#2 transfectedneurons in slices treated with 100 µM Rho-kinase inhibitor Y-27632. Error bars indicates.e.m. Spine length was significantly shorter forOphn1#2 siRNA transfected neurons than forcontrol siRNA transfected neurons (P < 0.0001;2,946 control siRNA spines, 1,668 Ophn1#2siRNA spines) and significantly longer forOphn1#2 siRNA transfected neurons treatedwith Y-27632 compared to untreated Ophn1#2siRNA transfected neurons (P < 0.0001; 1,668Ophn1#2 siRNA spines, 1,562 Ophn1#2 siRNA+ Y-27632 spines). Right graph shows the meanspine length for control siRNA transfectedneurons and control siRNA transfected neuronsin Y-27632 treated slices. There was nosignificant difference between the two groups (P = 0.11, 1,551 control siRNA spines, 1,022 control siRNA + Y-27632 spines). Scale bar, 5 µm. (b) Expression of T7-oligophrenin-1 in CA1 neurons does not affect spine length. The line graph shows the cumulativefrequency distribution of spine lengths for control vector expressing neurons (black) and neurons expressing T7-oligophrenin-1 (OPHN1; white). Insetshows the mean spine length for the two groups. Error bars indicate s.e.m. No significant difference in spine length was observed between these twogroups (P = 0.14, 1,400 control vector spines, 1,328 OPHN1 spines). Scale bar, 5 µm.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

can physically link type-I mGluRs and TRP calcium channels withIP3 receptors27,38, and that calcium release from intracellular storescan affect spine length39. A calcium increase in spines may promotethe remodeling of actin filaments and enhance actomyosin contractil-ity, causing alterations in spine structure40. Importantly, spine lengthchanges, whether or not they are dependent on calcium-mediatedactin cytoskeletal rearrangements, can themselves influence post-synaptic signaling. Alterations in spine length have been shown toaffect the diffusion rate and decay kinetics of calcium in the spinehead relative to those of the dendritic shaft41,42, and even smallchanges in spine length can cause large changes in calcium decaykinetics41. Thus, it is conceivable that interfering with oligophrenin-1function may cause alterations in calcium dynamics, likely by disrupt-ing its association with Homer and/or RhoA signaling. Given that cal-cium is a mediator of input-specific forms of synaptic plasticity42,such changes could ultimately affect learning and memory.

In summary, this study demonstrates the successful use of siRNAsin organotypic hippocampal slices to examine the function of a dis-ease-related molecule. We found that loss of oligophrenin-1 causeschanges in spine morphology, which are believed to be important forsynaptic function. Our findings provide a potential explanation as tohow loss of this protein may result in cognitive impairment underly-ing MRX. It should be noted that in addition to the involvement ofoligophrenin-1 in MRX, very recent studies implicate a role foroligophrenin-1 in mental impairment in individuals that also have

epilepsy and/or cerebellar hypoplasia43,44. It will therefore be of inter-est to investigate additional roles for oligophrenin-1 in neuronaldevelopment and disease.

METHODSOligophrenin-1 antibodies. Antibodies 1296 and 1724 were raised in rabbitsagainst the C-terminal peptide CETASRKTGSSQGRLPGDES and amino acids(aa) 635–802, respectively.

Constructs and siRNAs. Full-length mouse Ophn-1 cDNA was subcloned inreverse into the T7-epitope tagged expression vector, pCGT45, to generate theantisense construct. pCGT/GTPases were made as described45. GST-Rhotekin(aa 7–89) and GST-Pak3 CRIB (aa 65–137) plasmids were gifts from M. Schwartz (University of Virginia, Charlottesville, Virginia, USA) and R. Cerione (Cornell University, Ithaca, New York, USA). Full-length humanOPHN1 cDNA was subcloned into pCGT (pCGT/OPHN1) andpcDNA3.1/MycHisA (pcDNA3.1/MycHisA/OPHN1) (Invitrogen). Full-lengthOPHN1 P4L/F8C mutant was PCR generated and cloned into pCGT (OPHN1PF). Homer1b/1c cDNAs were subcloned into pGEX-4T-1 (Amersham).siRNAs were from Dharmacon. siRNA target sequence for Ophn1#1 was 5′ AAAGGGATCAAGACAGAAGGG 3′, and 5′ AAGAGCAGCTCTTTCTG-GCCT 3′ for OPHN1#2. Control siRNA was caspase 8 or caspase 2 duplexsiRNA (gift from Y. Lazebnik, Cold Spring Harbor Laboratory, Cold SpringHarbor, New York, USA).

Cell culture and transfection. REF52 or HEK 293 cells were transfected usingLipofectamine 2000 (Invitrogen) and harvested after 48 h. Medium-density

A R T I C L E S


Figure 7 Oligophrenin-1 interacts with Homer. (a) Sequence alignment of oligophrenin-1 with known Homer-interacting proteins containing the consensusHomer (P)PxxF binding motif. (b) Immunoblots of hippocampal lysates (10 µg per lane), SPM (1.5 µg per lane) and PSD fractions probed with antibodiesagainst oligophrenin-1 (1296), Homer1b/c, PSD-95 and synaptophysin. SPM were extracted in 0.5% Triton X-100 to yield the PSD Triton-1 pellet (P) andsupernatant (S). The PSD Triton-1-P was divided and extracted in 0.5% Triton X-100 or 3% N-lauroyl sarcosine to yield PSD Triton-2 and sarcosyl pelletsand supernatants, respectively. (c) Homer1b- or Homer1c-GST fusion proteins, or GST alone, were incubated with extracts from rat hippocampi andcortices. Bound oligophrenin-1 was detected by immunoblotting with 1296. Input lane was loaded with 15% of the extract used for the assay. (d) Lysatesfrom HEK 293 cells expressing either wild-type T7-oligophrenin-1 (OPHN1 WT) or mutant T7-oligophrenin-1 containing mutations P4L and F8C (OPHN1 PF) were incubated with Homer1b-GST or GST alone. Immunoblots showing bound oligophrenin-1 were from the same experiment and exposed tofilm for the same amount of time. Input lane was loaded with 20% of the lysates used for the assay. (e) Extracts from rat hippocampi wereimmunoprecipitated with 1296 (OPHN1), a polyclonal Homer antibody (Homer), a monoclonal control antibody against CD8 (con Ab) or no antibody (noAb). Immune complexes were subject to immunoblotting and probed with either a monoclonal Homer antibody or 1296 as indicated. Input lane wasloaded with 5% of the total lysate used.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

primary hippocampal neuron cultures were prepared from E18 rat embryos asdescribed46. For immunofluorescence (IF), hippocampal neurons (19 d.i.v.)were transfected with pCGT/OPHN1 using Lipofectamine 2000, fixed after 48 h and permeabilized. Neurons were transfected before plating with anti-sense and control vector using the Mouse Neuron Nucleofector kit (AmaxaBiosystems) and harvested 3 d later for western blotting. Transfection efficien-cies were 40–70%. For siRNA transfections, 5/6 or 12 d.i.v. neurons were trans-fected with siRNAs using Lipofectamine 2000 and harvested 48 and 72 h later,respectively, and then subjected to western blotting. For IF, neurons (6 d.i.v.)were transfected with siRNAs and methanol fixed 96 h later. Transfection effi-ciency was > 90% at 5/6 d.i.v. using a rhodamine-labeled siRNA.

Hippocampal slices were made as previously described47. Slices were gener-ated from P4 rat pups and biolistically transfected after 4 d in culture using aHelios Gene Gun (Bio-Rad). The following amounts of plasmids or siRNAswere precipitated onto 12.5 mg of 1.6-µm gold particles: 50 µg of pEGFPN3(Clontech) and 50 µg of antisense construct, or pCGN; 20 µg of pEGFPN3alone; 20 µg of pEGFPN3 and 160 µl of 20 µM duplex siRNA; 50 µg ofpEGFPN3 and 50 µg of pCGT/OPHN1, or pCGN; and 50 µg of pEGFPN3 and50 µg of pCGT/GTPase, or pCGN. To assess effective coating of siRNAs ontogold particles, a rhodamine-labeled siRNA was successfully precipitated ontobeads as examined by fluorescence microscopy (data not shown). For the Rho-kinase inhibitor experiments, 100 µM Y-27632 dihydrochloride (Alexis) wasadded to the medium at the time of transfection. Slices were incubated 48 hbefore the 1.5-h fixation in 4% PFA, 4% sucrose, PBS. All animal care proto-cols were approved by Cold Spring Harbor Laboratory.

Western blotting. Rat tissues, REF52 cells and hippocampal neurons wereprepared in 75 mM Tris (pH 6.8), 3.8% SDS, 4 M urea, 20% glycerol and sub-jected to western blotting using standard methods. We used the followingprimary antibodies: anti-oligophrenin-1 1296, anti-neuronal class III β-tubulin (Covance) and anti-ERK2 (Santa Cruz). Oligophrenin-1 levelswere normalized to β-tubulin or ERK2 levels and expressed as a percentageof control transfections.

Immunofluorescence. Parasagittal frozen brain sections (10 µm) from a P30rat were methanol-fixed, and dissociated hippocampal neurons were PFA-fixed. We used the following primary antibodies: anti-oligophrenin-1 (1296,1724), anti-neuronal class III β-tubulin, anti-PSD-95 (IgG2A; AffinityBioreagents), anti-synaptophysin clone SVP-38 (IgG1; Sigma), anti-T7 tag(IgG2B; Novogen) and anti-NeuN MAB377 (Chemicon). We used the follow-ing secondary antibodies and toxins: Alexa Fluor 488 goat anti-mouse (H+L,IgG2a and IgG1 specific), Alexa Fluor 594 goat anti-mouse (IgG2b specific) oranti-rabbit (H+L), Alexa Fluor 488-labeled phalloidin (Molecular Probes).Cells were imaged using an Axioscope fluorescence microscope (Zeiss) with a63× Plan Apochromat objective.

Two-photon imaging and image analysis. Two-photon images were obtainedusing an Olympus Fluoview laser-scanning microscope with a Ti-Sapphirelaser (Mira 900F; Coherent) at 910 nm and a LUMPlanF1/IR 40×, 0.75 NAwater immersion lens. Basal dendrites of CA1 pyramidal cells were imagedwith 5× zoom. Optical sections were spaced 1.0 µm apart and each was anaverage of three scans. For the RNAi and antisense experiments, four cellswere analyzed for pEGFPN3 alone, five for control siRNA, nine for Ophn1#1siRNA, twelve for Ophn1#2 siRNA, ten for control vector and eight for anti-sense construct. For the Rho-kinase inhibitor control experiment, five cellswere analyzed for control siRNA and five cells for control siRNA + Y-27632.For the Rho-kinase inhibitor rescue experiment, ten cells were analyzed forcontrol siRNA, five for Ophn1#2 siRNA, and five for Ophn1#2 siRNA + Y-27632. For the T7-oligophrenin-1 experiment, six cells were analyzed for con-trol vector and five for pCGT/OPHN1. Dendritic spines (protrusionsincluding filopodia) were measured from an image stack projection usingcustom NIH Object Image macros. Protrusion lengths were measured fromthe protrusion’s tip to the point where it met the dendritic shaft. Individualoptical section images were used to verify each protrusion, and protrusions ofall lengths were measured. Statistical differences remained for all groupswhen a minimum protrusion length threshold of 0.3 µm was applied. At least100 µm of primary dendrite and 200 µm of secondary dendrite were analyzed

for each cell. Spine and filopodia densities were calculated per cell. Spine sta-tistics were performed using the Kolmogorov-Smirnov two-sample test.Cumulative frequency plots (Figs. 3, 4 and 6) indicate the fraction of spines(y-axis) equal to or less than a certain length (x-axis). For illustrative pur-poses, data were binned (0.3 µm).

Rhotekin and Pak CRIB GTPase activation assays. The Rhotekin and PakCRIB assays were performed as previously described48, using PC12 cells trans-fected with pcDNA3.1/MycHisA/OPHN1 or vector.

PSD extractions. SPM were prepared from two adult rat hippocampi asdescribed49. PSD fractions were prepared and detergent-extracted, essentiallyas described50. Equal volumes of PSD pellets and supernatants were subjectedto western blotting. We used the following primary antibodies: 1296,Homer1b/c (clone D3; Santa Cruz), PSD-95 (Affinity Bioregents) or synapto-physin (clone SVP-38; Sigma).

Pull-down assays and coimmunoprecipitations. Homer1b/1c-GST fusion pro-teins and GST alone were immobilized onto glutathione-Sepharose beads(Amersham). Rat hippocampi and cortices were homogenized in 50 mM TrispH 7.4, 1 mM EDTA and 1% CHAPS (TEC); then they were sonicated and cen-trifuged. Supernatants were incubated with Homer1b/Homer1c-GST fusionproteins or GST. HEK 293 cells expressing OPHN1 WT or OPHN1 PF wereextracted in TE, 1% Triton X-100. Lysates were incubated with Homer1b-GSTfusion protein or GST. All immunoblots were probed with antibody 1296. Forimmunoprecipitations, rat hippocampi were homogenized in 50 mM Tris, pH7.4, 150 mM NaCl and 1% Triton X-100, essentially as described38. Homogenatewas centrifuged and supernatant incubated with 2 µg of antibody: 1296, poly-clonal Homer FL-354 (Santa Cruz), or monoclonal CD8. Antibody complexeswere captured using Protein A agarose (Roche). Immunoblots were probed with1296 or monoclonal Homer D3 antibody (Santa Cruz).


ACKNOWLEDGMENTSWe thank A. Piccini, E. Ruthazer, B. Burbach, C. Kopec, K. Jensen and H. Hsieh fortechnical assistance. We also thank H. Cline, R. Malinow, J. Skowronski,K. Svoboda, J. Rodriguez and members of the Van Aelst Laboratory for discussionsand/or critical reading of the manuscript. This work was supported by theNational Institutes of Health and Dana Foundation (to L.V.A.), the Wellcome Trust(to S.E.N. and C.J.A.) and the National Institutes of Health (to E.E.G).


Received 28 October 2003; accepted 30 January 2004Published online at http://www.nature.com/natureneuroscience/

1. Chelly, J. & Mandel, J.L. Monogenic causes of X-linked mental retardation. Nat. Rev.Genet. 2, 669–680 (2001).

2. Frints, S.G., Froyen, G., Marynen, P. & Fryns, J.P. X-linked mental retardation: van-ishing boundaries between non-specific (MRX) and syndromic (MRXS) forms. Clin.Genet. 62, 423–432 (2002).

3. Ramakers, G.J. Rho proteins, mental retardation and the cellular basis of cognition.Trends Neurosci. 25, 191–199 (2002).

4. Van Aelst, L. & D’Souza-Schorey, C. Rho GTPases and network signaling. Genes Dev.11, 2295–2322 (1997).

5. Luo, L. Rho GTPases in neuronal morphogenesis. Nat. Rev. Neurosci. 1, 173–180(2000).

6. Nakayama, A.Y., Harms, M.B. & Luo, L. Small GTPases Rac and Rho in the mainte-nance of dendritic spines and branches in hippocampal pyramidal neurons. J. Neurosci. 20, 5329–5338 (2000).

7. Tashiro, A., Minden, A. & Yuste, R. Regulation of dendritic spine morphology by therho family of small GTPases: antagonistic roles of Rac and Rho. Cereb. Cortex 10,927–938 (2000).

8. Billuart, P. et al. Oligophrenin-1 encodes a rhoGAP protein involved in X-linked men-tal retardation. Nature 392, 923–926 (1998).

9. Allen, K.M. et al. PAK3 mutation in nonsyndromic X-linked mental retardation. Nat.Genet. 20, 25–30 (1998).

10. Kutsche, K. et al. Mutations in ARHGEF6, encoding a guanine nucleotide exchangefactor for Rho GTPases, in patients with X-linked mental retardation. Nat. Genet.26, 247–250 (2000).

11. Fauchereau, F. et al. The RhoGAP activity of OPHN1, a new F-actin-binding proteinis negatively controlled by its amino-terminal domain. Mol. Cell. Neurosci. 23,574–586 (2003).

A R T I C L E S


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


12. Hering, H. & Sheng, M. Dendritic spines: structure, dynamics and regulation. Nat.Rev. Neurosci. 2, 880–888 (2001).

13. Engert, F. & Bonhoeffer, T. Dendritic spine changes associated with hippocampallong-term synaptic plasticity. Nature 399, 66–70 (1999).


15. Toni, N., Buchs, P.A., Nikonenko, I., Bron, C.R. & Muller, D. LTP promotes formationof multiple spine synapses between a single axon terminal and a dendrite. Nature402, 421–425 (1999).

16. Matsuzaki, M. et al. Dendritic spine geometry is critical for AMPA receptor expres-sion in hippocampal CA1 pyramidal neurons. Nat. Neurosci. 4, 1086–1092 (2001).

17. Comery, T.A., Stamoudis, C.X., Irwin, S.A. & Greenough, W.T. Increased density ofmultiple-head dendritic spines on medium-sized spiny neurons of the striatum inrats reared in a complex environment. Neurobiol. Learn. Mem. 66, 93–96 (1996).

18. Purpura, D.P. Dendritic spine “dysgenesis” and mental retardation. Science 186,1126–1128 (1974).

19. Kaufmann, W.E. & Moser, H.W. Dendritic anomalies in disorders associated withmental retardation. Cereb. Cortex 10, 981–991 (2000).

20. Fiala, J.C., Spacek, J. & Harris, K.M. Dendritic spine pathology: cause or conse-quence of neurological disorders? Brain Res. Rev. 39, 29–54 (2002).

21. Harris, K.M. Structure, development, and plasticity of dendritic spines. Curr. Opin.Neurobiol. 9, 343–348 (1999).

22. Comery, T.A. et al. Abnormal dendritic spines in fragile X knockout mice: Maturationand pruning deficits. Proc. Natl. Acad. Sci. USA 94, 5401–5404 (1997).

23. Nimchinsky, E.A., Oberlander, A.M. & Svoboda, K. Abnormal development of den-dritic spines in FMR1 knock-out mice. J. Neurosci. 21, 5139–5146 (2001).

24. Uehata, M. et al. Calcium sensitization of smooth muscle mediated by a Rho-associ-ated protein kinase in hypertension. Nature 389, 990–994 (1997).

25. Amano, M., Fukata, Y. & Kaibuchi, K. Regulation and functions of Rho-associatedkinase. Exp. Cell Res. 261, 44–51 (2000).

26. Beneken, J. et al. Structure of the Homer EVH1 domain-peptide complex reveals anew twist in polyproline recognition. Neuron 26, 143–154 (2000).

27. Fagni, L., Worley, P.F. & Ango, F. Homer as both a scaffold and transduction mole-cule. Sci. STKE 137, 1–7 (2002).

28. Xiao, B., Cheng, J. & Worley, P.F. Homer: a link between neural activity and gluta-mate receptor function. Curr. Opin. Neurobiol. 10, 370–374 (2000).

29. Huber, K.M., Gallagher, S.M., Warren, S.T. & Bear, M.F. Altered synaptic plasticityin a mouse model of fragile X mental retardation. Proc. Natl. Acad. Sci. USA 99,7746–7750 (2003).

30. Billuart, P., Winter, C.G., Maresh, A., Zhao, X. & Luo L. Regulating axon branch sta-bility: the role of p190 RhoGAP in repressing a retraction signaling pathway. Cell107, 195–207 (2001).


32. Sala, C. et al. Inhibition of dendritic spine morphogenesis and synaptic transmissionby activity-inducible protein Homer1a. J. Neurosci. 23, 6327–6337 (2003).

33. Fischer, M., Kaech, S., Wagner, U., Brinkhaus, H. & Matus, A. Glutamate receptorsregulate actin-based plasticity in dendritic spines. Nat. Neurosci. 3, 887–894 (2000).

34. Star, E.N., Kwiatkowski, D.J. & Murthy, V.N. Rapid turnover of actin in dendriticspines and its regulation by activity. Nat. Neurosci. 5, 239–246 (2002).

35. Ackermann, M. & Matus, A. Activity-induced targeting of profilin and stabilization ofdendritic spine morphology. Nat. Neurosci. 6, 1194–1200 (2003).

36. McKinney, R.A., Capogna, M., Dürr, R., Gähwiler, B.H. & Thompson, S.M. Miniaturesynaptic events maintain dendritic spines via AMPA receptor activation. Nat.Neurosci. 2, 44–49 (1999).

37. Sin, W.C., Haas, K., Ruthazer, E.S. & Cline, H.T. Dendrite growth increased by visualactivity requires NMDA receptor and Rho GTPases. Nature 419, 475–480 (2002).

38. Yuan, J.P. et al. Homer binds TRPC family channels and is required for gating ofTRPC1 by IP3 receptors. Cell 114, 777–789 (2003).

39. Korkotian E. & Segal, M. Release of calcium from stores alters the morphology ofdendritic spines in cultured hippocampal neurons. Proc. Natl. Acad. Sci. USA 96,12068–12072 (1999).

40. Hayashi, K. et al. Modulatory role of Drebrin on the cytoskeleton within dendriticspines in the rat cerebral cortex. J. Neurosci. 16, 7161–7170 (1996).

41. Majewska, A., Tashiro, A. & Yuste, R. Regulation of spine calcium dynamics by rapidspine motility. J. Neurosci. 20, 8262–8268 (2000).

42. Yuste, R., Majewska, A. & Holthoff, K. From form to function: calcium compartmen-talization in dendritic spines. Nat. Neurosci. 3, 653–659 (2000).

43. Bergmann, C. et al. Oligophrenin-1 (OPHN1) gene mutation causes X-linked mentalretardation with epilepsy, rostral ventricular enlargement and cerebellar hypoplasia.Brain 126, 1537–1544 (2003).

44. Philip, N. et al. Mutations in the oligophrenin-1 gene (OPHN1) cause X linked con-genital cerebellar hypoplasia. J. Med. Genet. 40, 441–446 (2003).

45. Joneson, T., McDonough, M., Bar-Sagi, D. & Van Aelst, L. RAC regulation of actinpolymerization and proliferation by a pathway distinct from Jun kinase. Science274, 1374–1376 (1996).

46. Pak, D.T., Yang, S., Rudolph-Correia, S., Kim, E. & Sheng, M. Regulation of den-dritic spine morphology by SPAR, a PSD-95-associated RapGAP. Neuron 31,289–303 (2001).

47. Stoppini, L., Buchs, P.A. & Muller, D. A simple method for organotypic cultures ofnervous tissue. J. Neurosci. Methods 37, 173–182 (1991).

48. Yamaguchi, Y., Katoh, H., Yasui, H., Mori, K. & Negishi, M. RhoA inhibits the nervegrowth factor-induced Rac1 activation through Rho-associated kinase-dependentpathway. J. Biol. Chem. 276, 18977–18983 (2001).

49. Lau, L.F. et al. Interaction of the N-methyl-D-aspartate receptor complex with a novelsynapse-associated protein, SAP102. J. Biol. Chem. 271, 21622–21628 (1996).

50. Cho, K.O., Hunt, C.A. & Kennedy, M.B. The rat brain postsynaptic density fractioncontains a homolog of the Drosophila discs-large tumor suppressor protein. Neuron9, 929–942 (1992).

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

Neural computation is accomplished by interactions between synapticinputs and membrane channels that translate synaptic input into mean-ingful temporal patterns of action potentials. Biophysical mechanismsunderlying this computation are still largely unknown. Several theoreticand experimental studies of computation in functional neural networkssuggest that neural computation results from a concurrent interplay ofexcitatory and inhibitory (E/I) synaptic inputs1–8. Although E/I interac-tions at the cell body have been studied9, a large proportion of inhibitorysynapses are instead distributed on the dendritic tree10–14. These den-dritic E/I synaptic inputs may interact locally to perform neural compu-tation on the dendrite branches themselves1,4–6,15,16. If so, E/I synapseswould need to be organized on dendrites to permit their meaningfulinteraction. The aim of this study was to understand the structuralorganization and functional interaction of excitatory and inhibitorysynapses within individual dendritic branches, as well as to explore themechanisms that maintain an optimal E/I balance. Our results indicatethat E/I synapses are evenly distributed on dendritic trees to maintain aconstant ratio among all dendritic branches. Furthermore, we foundthat E/I inputs to individual dendritic branches were more effective thanthose outside of dendritic branches. Therefore, this structural arrange-ment seems to facilitate meaningful E/I interactions on dendriticbranches. Finally, we found that the balance of E/I synapses was estab-lished and maintained by a push-pull regulatory mechanism.

RESULTSStructural balance of E/I synapses in dendritesTo explore the rules that control the arrangement of excitatory andinhibitory (E/I) synapses, we analyzed the distribution of functional

E/I synapses along dendritic trees by simultaneously labeling the den-dritic surface and active presynaptic terminals in cultured hippocam-pal neurons (Fig. 1). We found the number of functional synapses perunit length of dendrite to be higher in thicker branches than in thin-ner ones (Fig. 1a). Since thicker branches have a larger surface areaper unit length of dendrite than thinner ones do, maintaining ahigher number of synapses in thicker branches might serve to main-tain a constant number of synapses across a given area of dendriticsurface. To demonstrate this relationship quantitatively, we plottedthe number of synapses found along a fixed length of dendrite versusthe diameter of the dendrite (Fig. 1b). These two parameters were lin-early correlated, suggesting an even distribution of functionalsynapses across dendritic surfaces17. Next, as FM dye staining cannotdistinguish excitatory from inhibitory synapses, we used both excita-tory and inhibitory synapse-specific marker proteins to distinguishthe two types of synapses (Fig. 1c). Inhibitory synapses were distrib-uted unevenly across the entire neuronal surface, with the highestdensities found near somata, similar to the distribution pattern ofinhibitory synapses on hippocampal pyramidal neurons in vivo12,13.Despite the lower density of inhibitory synapses in the dendritic tree,the majority of inhibitory synapses (86%; Fig. 1d) were located in thedendritic tree, owing to the substantially larger surface area of den-drites versus somata. Our analysis of the distribution pattern of E/Isynapses on the dendritic tree showed that the relative numbers of E/Isynapses within dendritic branches were highly correlated (Fig. 1c,e).The ratios of E/I synapses, both among dendritic branches of individ-ual neurons and between neurons at the same stage of development,were constant (Fig. 1f). After synaptic maturation, the ratio of excita-

Picower Centre for Learning and Memory, RIKEN–MIT Neuroscience Research Center, Departments of Brain & Cognitive Sciences and Biology, MIT, Cambridge,Massachusetts 02139, USA. Correspondence should be addressed to G.L. ([email protected]).


Local structural balance and functional interaction ofexcitatory and inhibitory synapses in hippocampaldendritesGuosong Liu

Theoretical and experimental studies on the computation of neural networks suggest that neural computation results from a dynamicinterplay of excitatory and inhibitory (E/I) synaptic inputs. Precisely how E/I synapses are organized structurally and functionally tofacilitate meaningful interaction remains elusive. Here we show that E/I synapses are regulated across dendritic trees to maintain aconstant ratio of inputs in cultured rat hippocampal neurons. This structural arrangement is accompanied by an E/I functional balancemaintained by a ‘push-pull’ feedback regulatory mechanism that is capable of adjusting E/I efficacies in a coordinated fashion. We alsofound that during activity, inhibitory synapses can determine the impact of adjacent excitatory synapses only if they are colocalized onthe same dendritic branch and are activated simultaneously. These fundamental relationships among E/I synapses provide organizationalprinciples relevant to deciphering the structural and functional basis for neural computation within dendritic branches.


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

tory to inhibitory synapses was approximately 4:1 across the dendriticsurface. We checked whether such a correlation could result from arandom insertion of E and I by simulating insertion of E and Isynapses with a Poisson model. This simulation placed the chanceprobability of observing the degree of E/I correlation seen in Figure 1e at <0.001. These data suggest that the distribution of E/Isynapses on the dendritic tree is highly organized to maintain a bal-ance of excitatory and inhibitory inputs on each dendritic branch.

Functional balance of E/I synaptic inputsThis uniform distribution of E/I synapses along the dendritic treesuggested that E/I synaptic function may also be highly balanced. Wetherefore examined the relationship between the number of func-tional synapses and the relative overall strength of E/I synapses duringearly stages of synaptogenesis. As the overall strength of synaptic con-nections is determined by the quantal size (q), defined by the inte-grated conductance during quantal transmission and the frequency(f) of quantal transmission, we used G = f*q to represent synapticstrength (Fig. 2a). As expected, both excitatory and inhibitory synap-tic strengths (GE and GI) increased in proportion to the number offunctional synapses (Fig. 2b). The GI/GE balance was maintainedamong neurons at the same stage of development, although GI/GEincreased from 1 to 3 during neural network maturation (Fig. 2c).

These data suggest that coordinated insertionof E/I synapses resulted in a functional bal-ance of E/I synaptic strength similar to thestructural balance of E/I synapses (Fig. 1).

Given this coordination in synapticstrength, our results led us to ask how thesesynapses might interact temporally.Neurons in cultured networks make mutualconnections that give rise to periodic burstsof synaptic potentials (Fig. 2d), similar togiant depolarizing potentials observed inhippocampal slices18 and in vivo19. Burstingresults from the synchronous activation ofhundreds of E/I synapses. To study the com-position of these events, we separated thecontributions of E/I inputs by alternatelyclamping the membrane potential at –65 mV or 0 mV to record excitatory orinhibitory postsynaptic currents (EPSCs orIPSCs, respectively; Fig. 2d). To estimateGI/GE during bursts, we varied the clampmembrane potential to determine the burstreversal potential (Vrev) of the compoundsynaptic inputs (Fig. 2d, middle trace),which can then be used to calculate the rela-tive strength of GE versus GI (see Methods).The Vrev varied from –30 to –45 mV, corre-sponding to a GI/GE ratio that varied from0.9 to 2.2 with a mean value of 1.3 (n = 4),consistent with the GI/GE ratio calculatedfrom miniature synaptic events (mEPSCsand mIPSCs) described above. These resultsindicate that a constant ratio of E/I inputswas maintained both in general and dynam-ically across discrete time intervals.

The concurrent E/I inputs during a burstis interesting because most excitatory synap-tic inputs will be cancelled by concurrent

inhibitory inputs, such that neural output is determined by thedynamic balance of active E/I synapses (Fig. 2d,e). Despite the largeEPSCs during bursts (peak size ∼ 0.9 nA, mean 0.3 nA), the synapticpotential at the soma only reached 10–15 mV, resulting in few actionpotentials. In contrast, delivering only 0.1 nA of current into thesoma through a patch pipette induced a similar depolarization andproduced the same number of action potentials (Fig. 2e, left). Thissuggests that, as is observed in intact preparations3–8, most excitatoryinputs were cancelled by concurrent inhibitory inputs. As a result,neuronal output is controlled not only by the size of excitatoryinputs, but also by the dynamic difference between excitatory andinhibitory inputs.

Dynamic interplay of E/I inputs on dendritic treeGiven that most E/I synapses are located on the dendritic tree, we pos-tulate that most E/I cancellations take place locally within dendritebranches. To test this hypothesis, it is important to show that inhibitorysynapses on dendritic branches are active during bursting periods andmake significant contributions to the generation of bursting inhibitoryinputs. To quantify the contribution of dendritic inhibitory synapses,we recorded the inhibitory synaptic inputs while blocking functionalityof somatic inhibitory synapses with locally applied picrotoxin. Thepressure of the drug application system was carefully adjusted to limit

Figure 1 The structural balance of E/I synapses is maintained throughout hippocampal dendrites. (a) Distribution of functional presynaptic terminals on the dendritic tree. Active synapses were labeledby FM4-64 (red), and Alexa 488 delivered during whole-cell recording highlighted the correspondingdendritic surface in cultured hippocampal neurons (13 d.i.v.). (b) Number of synapses/unit length ofdendrite (D) increases linearly with enlarging radius (r) of dendritic branch (D = 0.092 + 0.095r; R2 = 0.8911; P < 0.0001). (c) Visualization of E/I synapses arranged across a single dendriticbranch. Excitatory synapses (green) were labeled with the vesicular glutamate transporter VGLUT1,whereas inhibitory synapses (red) were labeled with the synthetic enzyme GAD65 (20 d.i.v.). (d) Left:ratio of E/I in somatic and dendritic regions. A high density of inhibitory synapses occurs near thesoma. Right: distribution of inhibitory synapses in somatic and dendritic regions (86% of inhibitorysynapses located at dendritic region). (e) Ratios of E/I synapses across individual dendritic brancheswere constant (data from 31 branches in 8 neurons). (f) Ratio of E/I synapses in individual neuronsduring different stages of development. Each group of symbols represent the ratios of E/I synapses ondifferent dendritic branches from a single neuron.

A R T I C L E S


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

the picrotoxin to the proximal dendrite and somatic regions; the drugapplication pipette contained the fluorescent dye FM4-64 to monitorthe spatial spread of solution. We confirmed the effectiveness of thesomatic blockade by monitoring the size of GABAA receptor–mediatedcurrents induced by the local application of GABA to the somaticregion. The local application of picrotoxin to somata completelyblocked somatic GABAA currents (Fig. 3a), suggesting the effectiveremoval of somatic inhibitory input. When somatic inhibition wassilenced, bursting IPSCs were only reduced by 20.6 ± 2.9% (n = 3). Thispercentage of reduction is similar to the fraction of inhibitory synapsesfound at somatic locations (Fig. 1d), suggesting that somatic and den-dritic inhibitory synapses are activated in proportion to their availabil-ity, and the majority of inhibitory inputs during bursts originate atinhibitory synapses located on the dendritic tree. Hence, active E/Isynaptic inputs might interact locally on the dendritic branches.

To test this possibility, we compared somatic input conductance(Gin) in the presence and absence of bursting inputs by determiningthe change in membrane potential in response to a multi-step currentinjection (Fig. 3b). The membrane potential was recorded under cur-rent clamp using the perforated patch technique to preserve the elec-trotonic properties of the dendritic tree. Gin at rest was 2.8 nS, whereasGin observed during a bursting volley was 3.7 nS. Thus, Gin at the somaonly increased by ∼ 30% during bursting input. However, the sum ofGE and GI during bursts, measured under voltage clamp, was ∼ 13 nS(measured from a separate set of neurons, n = 10). If all of these synap-tic conductances had reached the soma, Gin would have been >15 nS.Thus, only ∼ 10 % of synaptic conductance induced by activation of E/Isynapses at the dendrite appeared to reach the soma (n = 3), suggestingthat most E/I synaptic interactions resolve within the dendritic tree.

Local interactions of E/I inputs on dendritic branchesTo understand the biophysical features of E/I interactions, we tried todetermine the temporal and spatial interactions of individual excita-tory and inhibitory synaptic inputs at dendrite branches. To overcomethe difficulty of precisely triggering synchronous E/I synaptic transmis-sion at adjacent synapses, we applied the neurotransmitters glutamateand GABA locally at E/I synapses to effectively mimic the magnitudeand time course of E/I synaptic transmission (Fig. 4a). Our previousstudies suggest that brief (0.5 ms), highly localized transmitter deliveryby iontophoresis can be used to selectively activate receptors from indi-

vidual synapses, thus mimicking endogenous synaptic transmis-sion20,21, although interpreting the results of the present manipulationwould still be valid if receptors from more than one synapse were acti-vated. Synaptic potentials evoked by the local application of either exci-tatory or inhibitory neurotransmitter were recorded at the soma. Wefirst studied the temporal interaction of E/I inputs at two adjacent E/Isites (distance less than 1 µm). The timing of glutamate receptor activa-tion was varied from 120 ms before to 120 ms after GABA receptor acti-vation (Fig. 4b). The size of the resultant excitatory postsynapticpotential (EPSP) was smaller when E/I inputs were stimulated concur-rently than when there was a time delay between the activations. Thedegree of this attenuation decayed with increasing intervals betweenEPSP and IPSP, resulting in an attenuation time constant of ∼ 18 ms(Fig. 4b,c; n = 4). The precise match of this constant with IPSC timecourse (∼ 20 ms) suggests that a significant portion of EPSP attenuationby IPSPs is associated with the opening of GABAA channels. The char-acteristics of this inhibition are similar to the shunting inhibition pre-dicted by modeling1. The time window of dendritic inhibition wasslightly longer when an inhibitory input followed an excitatory input.

We next studied the effects of space on interactions between E/Iinputs. Glutamate and GABA were released simultaneously while the


Figure 2 Functional balance of E/I synaptic inputs in a single neuron. (a) Left,distribution of active synaptic terminals on the dendrite of a recorded neuron(11 d.i.v.). Right, mEPSCs and mIPSCs were recorded from the same neuronby holding the membrane potential at –65 and 0 mV, respectively. Scale bars:10 pA /0.1 nS (vertical), 1 s/0.1 s (horizontal). (b) The strength of excitatory(GE = qE*fE) and inhibitory inputs (GI = qI*fI) are correlated and scaled linearlyagainst the number of functional synapses (four neurons). (c) The ratio of E/Iinputs in neurons with a variety of input strengths is constant (for 11 d.i.v.group: GI/GE = 1.35 (n = 4, R2 = 0.99, P < 0.0001); for 16 d.i.v. group:GI/GE = 2.55 (n = 7, R2 = 0.9, P < 0.01); for 18 d.i.v. group: GI/GE = 3.28 (n = 8, R2 = 0.21, P < 0.0001)). The dash lines are the linear fits to data withconstraint to pass zero. 18 d.i.v. data were obtained at 33 °C. (d) BurstingEPSCs and IPSCs recorded by setting membrane potential at the EPSC/IPSCreversal potential. Top trace is the synaptic input and action potential evokedat soma under current clamp. Middle and bottom traces show synaptic currentrecorded at different holding potentials to record EPSCs (Vm = VI = –65 mV)and IPSCs (Vm = VE = 0 mV). At –35 mV, E/I inputs cancel one another (Vrev).The calculated GI/GE was 1.16. Average GI/GE from four cells was 1.35. Scalebars: 600 pA (vertical), 1 s (horizontal). (e) The relationship between somaticcurrent injection and frequency of resulting action potentials. Top panels arethe action potentials evoked by somatic injection of 100 pA (left) and 200 pA(right). Bottom panel is the I/F relationship.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

distance between the glutamate and GABA delivery sites wasincreased in 1-µm steps (Fig. 4d). The influence of a given inhibitoryinput over an adjacent excitatory input was attenuated with increas-ing distance between the E/I synapses, at an overall space constant of∼ 10 µm (Fig. 4e,f). Notably, this attenuation did not seem to dependon the relative position of E/I synapses, as inhibitory synapses werecapable of attenuating more proximal excitatory inputs. In contrast,when an IPSP was evoked at a different, but adjacent, dendriticbranch, the IPSP had a substantially smaller impact on EPSP size,indicating that the influence of inhibitory synapses is local and largely

Figure 4 Spatial and temporal interactionsbetween E/I inputs. (a) Experimental set-up.Potentials evoked by local application ofglutamate or GABA were recorded at the cellsoma, when relative distance (XE to XI) and timing(TE to TI) of glutamate and GABA release weresystematically varied. (b) Temporal interactionsbetween elicited EPSPs and IPSPs. IPSC fromthe same synaptic site was superimposed to showthe relationship between the time course ofGABAA receptor opening and the time window ofEPSP attenuation (blue, Vm = –90 mV). Scalebars: 2.5 mV (vertical), 50 ms (horizontal). (c) Time window of E/I interaction. Highestattenuation occurs when E/I inputs coincide, andattenuation decays rapidly with a time constant of∼ 20 ms. (E + I)/E is the ratio of EPSP + IPSPintegral over EPSP integral as measured at thesoma. The dash lines are the exponential fits tothe data with time constants of 25 (left) and 18 ms (right). (d) Spatial interactions of EPSPsand IPSPs. XI and XE were the locations of GABAand glutamate delivery. Scale bar: 2 µm. (e) Attenuation of EPSPs by concurrent IPSPsoccurs only when both E/I synapses are locatedwithin the same dendritic branch (black traces:EPSP alone; blue traces: EPSP + IPSP). The sizeof IPSPs in XI(0) and XI(1) were similar (data notshown). Scale bars: 5 mV (vertical), 100 ms(horizontal). (f) Spatial relationship between thedistance of E/I synapses (from soma) and theefficacy of inhibition. Solid dots are percent ofinhibition when both synapses on the samedendritic branch. (1 – (E + I)/E). Open circlesrepresent cases where E/I synapses on separatebranches were activated.

A R T I C L E S

limited to individual dendritic branches. Such results also supportaforementioned models of dendritic shunting1. Overall, it seems thatE/I interactions occur locally within a dendritic branch, whereas theuniform spread of E/I synapses discussed earlier may provide a struc-tural arrangement that makes these local interactions possible.

A rule for E/I organization in hippocampal dendritesWe began to search for a simple rule to relate our observations of theuniform distribution and balance of E/I synapses throughout the den-dritic tree (Figs. 1 and 2) to the localized nature of dendritic inhibition


Figure 3 Activation of dendritic inhibitory synapses during bursts. (a) Majority of inhibitory inputs during bursts originate at inhibitorysynapses located on dendritic tree. Top panel shows the experimentalarrangement. Release of picrotoxin from pipette at left side was used toblock GABAA receptors at somatic region (region in yellow was the areaaffected). The somatic GABAA receptors were activated by local applicationof GABA through electrode at right side. The black trace is the burstinginhibitory inputs flanked by GABA currents initiated by somatic GABAreceptor activation. Red trace is the remaining bursting inhibitory inputsafter blockade of somatic inhibitory inputs. Scale bars: 0.5 s (horizontal),100 pA (vertical). (b) Change in input conductance (Gin) induced bysynaptic inputs. Currents ranging –50 to –150 pA were injected, and theevoked change in membrane potentials at rest (black trace) and duringbursting synaptic input (red trace) were compared. The resulting I/Vrelationship yields Gin at rest (2.8 nS, black spots) and during the burstperiods (3.7 nS, red spots). Scale bars: 0.5 s (horizontal), 25 mV (vertical).

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

(Fig. 4). As the interaction of E/I inputs, com-bined with the resting dendritic conductance(GR), determines the size of the compoundsynaptic potential (VS) on each dendrite, E/Isynapses on a dendritic tree might be organ-ized to maintain a constant VS within individ-ual dendritic branches. To maintain a properlevel of VS, any perturbation to VS wouldnecessitate feedback regulation of GE and GI (Fig. 5a) simultaneously. Itis likely that the robust feedback regulation required under this formula-tion to maintain a suitable level of excitatory inputs takes manyforms22,23. Candidates include the inverse correlation between excita-tory synaptic density on the dendritic tree and quantal size24, theincrease in quantal size25 and probability of release26 after action poten-tial blockade, and the increase of presynaptic transmitter release afterpostsynaptic potassium channel overexpression27,28. This model alsoadds two predictions: (i) in addition to excitatory inputs, inhibitoryinputs and resting dendritic conductance are critical in controlling VSand (ii) the balance of E/I synapses within the dendritic tree may be nec-essary for tuning VS to an appropriate level. Furthermore, it predicts thatthe absolute level of excitatory inputs should be less critical in control-ling VS than the balance of E/I inputs (Fig. 2d,e). If our assumptions arevalid, the strengths of E/I synapses and resting membrane conductanceshould be tuned in a coordinated fashion, with any offset in one direc-tion resulting in a restorative change in the other (Fig. 5a).

Maintaining the balance of E/I input strengthTo verify this model, we pharmacologically modified the efficacy ofeither excitatory or inhibitory inputs, and then assessed the extent towhich the neurons compensated for this perturbation by adjusting theirE/I efficacies to restore the original balance. The GABAA receptor antag-onist bicuculline was used to block inhibition, and the GABAA receptorpotentiator flunitrazepam was used to prolong the response of theGABAA receptors, thus enhancing functional inhibition. Similarly,NBQX was used to block the binding of glutamate to AMPA receptors,resulting in the elimination of excitation. NBQX and flunitrazepamserved to reduce VS, whereas bicuculline had the opposite effect. Werecorded EPSCs and IPSCs under the various drugs used to perturb thefunctional E/I balance (Fig. 5b, top) and the feedback responses of E/Isynapses to the perturbations over 48 h (Fig. 5b, bottom). Removinginhibition led to a reduction of excitatory synaptic strength and anenhancement of inhibitory synaptic strength. The opposite effects wereobserved after a reduction of excitatory inputs or, conversely, an

Figure 5 Feedback regulation of E/I synapsesafter perturbation of E/I balance. (a) Workingmodel of E/I organization and regulation. (b) Top panels, reducing or enhancing GABAAreceptor–mediated inhibition by the antagonistbicuculline (bicu) or potentiator flunitrazepam(fluni). Bottom panels are the compensatoryresponses of E/I synapses to perturbations.Scale bars: 500/20 pA (vertical), 0.5 s(horizontal). (c) Push-pull regulation of E/Isynaptic strength in response to perturbation ofE/I balance. Number of neurons in eachtreatment: 30 (ctrl); 14 (NBQX); 14 (fluni); 14 (bicu); 8 (TTX) (P < 0.01 for all treatments,t-test). (d) The effects of action potentialblockade on the E/I balance. (e) Gin increasesin compensation for the increase in thefunctional E/I ratio.

A R T I C L E S

enhancement of inhibitory inputs. To quantify these results, we calcu-lated the total GE and GI from the sum of mEPSCs and mIPSCs fromsomatic recording (Methods) and plotted them by treatment (Fig. 5c).In response to the imbalance, E/I synapses adjusted their synapticstrength in the opposite direction (Fig. 5c), exhibiting push-pull regula-tion. The GE and GI responses matched the predictions of our model(Fig. 5a). We also tested the effects of action potential blockade on E/Ibalance (Fig. 5d). Although GE increased slightly, the major effect ofneural activity blockade was a dramatic reduction of inhibitory synaptictransmission, similar to results reported previously29. This result sug-gests that a reduction of inhibitory inputs might be the primary mecha-nism of compensating for the loss of excitatory synaptic inputs. It isworth mentioning that because the drugs were applied to the culturemedium, the E/I balances of all dendritic branches were perturbed uni-formly. The post-treatment changes in GE and GI represent the averagedresponses from synapses in the entire dendritic tree. Thus, these resultsdo not provide direct evidence that the various regulatory responsesoccur in a dendritic branch–specific fashion. To test whether E/I balancecan be regulated in a dendrite specific fashion, new experimentalapproaches need to be developed that allow perturbation and recordingof E/I balance at individual dendritic branches.

We also attempted to determine whether the modification of thesize of GR is coordinated to maintain VS at a suitable level. Althoughit would be desirable to measure changes in GR across treatments,the size of GR at or near individual synapses is difficult to obtain.Instead, we determined the change in Gin resulting from each treat-ment to infer the relative strength of GR. Notably, Gin also changedafter treatments, following the same trend as GI (Fig. 5e). However,the magnitude of change of Gin is much smaller than that of GI, sug-gesting that synaptic sites, rather than regulation of dendritic leakchannels, for example, are the primary regulators of VS.

DISCUSSIONOur results show that E/I synapses were distributed evenly alongdendritic branches, that the functional ratio of E/I inputs was deli-


NBQXFlun

i0

5

10

15 Excitatory

0

20

40

60Inhibitory

Ctrl Fluni ( ↑ I)Bicu ( ↓ I)

Perturbation

Response

EPSCs

mEPSCs

mIPSCs

IPSCs

aGE GI

VS

+

GR

– +–

NBQXCtrl

Bicu0

2

4

6

GIN

(nS

)GI (nS

/s)

GI (nS

/s)

↑E/I ratio during treatment

b

TTX

0

5

10

15

0

20

40

60

c

Ctrl

GE (

nS/s

)

GE (

nS/s

)

CtrlBicu

d e

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

cately balanced to evenly distribute input along the dendritic sur-face, and that this E/I balance was actively maintained by push-pullregulatory mechanisms (Fig. 5a). The consequence of this arrange-ment is that the compound synaptic potential (VS) at individualdendritic branches remains constant, although the exact level ofthe VS might be different during early stages of development. Sincethe amount of synaptic input that impacts processing in the somadepends on VS, the ratio of E/I inputs, in conjunction with thenumber of voltage-dependent and independent channels within adendrite, ultimately controls neural output. As the firing of a neu-ron is determined by the dynamic interplay of E/I inputs (Fig. 2d),the constant VS might be the underlying mechanism by which aneuron reduces fluctuations in its firing rate when the level ofsynaptic input varies25. This new view emphasizes the significantrole of inhibitory synapses within dendrites in shaping synapticintegration, and the necessity of E/I balance for neural computa-tion. This idea is further supported by recent studies in vivo. In thebarrel cortex of intact rats, for example, when an increase in thenumber of excitatory synapses is elicited by whisker stimulation,inhibitory synapses are also added to preserve the 4/1 ratio of E/Isynapses observed across the dendritic trees of this system30.Inversely, reduction of synaptic input in the visual cortex in light-deprived animals prevents an increase in both excitatory synapticstrength31 and GABAergic innervation32, thus maintaining the E/Ibalance when synaptic inputs are reduced. Finally, experimentalanimal models of temporal lobe epilepsy often involve impaireddendritic inhibition33, suggesting the significance of the E/I bal-ance in preserving the functionality of neural networks.

This local regulatory mechanism might act alone or be com-plementary to an alternative mechanism that relies on actionpotential back-propagation to normalize the efficacy of dendriticsynapses23,34. The presence of local and subthreshold regulatorymechanisms could be important for two reasons: (i) as the backpropagation of an action potential cannot reach distal dendriticbranches35, not all synapses on the dendritic tree can be normal-ized by action potential frequency and (ii) for a neuron with avery low firing rate, the number of action potentials might not bethe most appropriate parameter for the adjustment of its overallsynaptic strength. For example, only 1–2% of CA1 hippocampalneurons fire action potentials during behavior. The average firingrate of these neurons is about 1 Hz36. The latest results fromwhole-cell patch-clamp recording in awake animals show that thespontaneous firing rate of barrel cortex neurons is only 0.05 Hz,50-fold less than what has been reported by extracellular unitrecordings37. The few milliseconds of membrane depolarizationover a 20-s time window might not contain as much informationfor representation of overall synaptic inputs as do continuoussubthreshold synaptic inputs. Further experiments are needed totest whether the local voltage change is sufficient to trigger adjust-ment of the strength of E/I synapses.

Another important component of the present study is the evi-dence of the localized nature of E/I interactions on dendrites. The effi-cacy of inhibitory inputs appears to be locally restricted and criticallydependent on their spatial and temporal proximity to excitatoryinputs (on the same dendritic branch, within ∼ 20 ms). The local E/Iinteractions that result are computationally equivalent to the ‘veto’operation that is thought to occur in dendritic computation1. Theuniform spread of E/I synapses could well provide a structural basisfor distributing this type of computation over entire dendritebranches. Two elegant studies of direction selectivity in retinal gan-glion cells reveal biological evidence for such computation4,6. E/I

synapses on dendrites are activated simultaneously, but the degree oftheir relative activation depends on the direction of movement ofvisual stimuli. This E/I interaction occurs on dendrites. Similar inter-actions between E/I synaptic inputs are also observed in neurons ofvarious regions of cortex7,8,38,39. Thus, the ratio of E/I inputs mightconstitute an elementary unit for neural computation across multipleregions and systems. If this is the case, the activity of inhibitorysynapses, both in their intensity and precise timing, must be ‘tuned’properly to sponsor meaningful patterns of neural activity. This pos-sibility is further supported by recent studies in visual system develop-ment showing that inhibitory and excitatory synapses are modified inconjunction40. These cases offer evidence from across the nervous sys-tem that E/I interactions in dendrites might be an important form ofneural computation. Our results provide new biophysical details ofthe balance of E/I synapses, how this balance is established through-out the dendritic tree and the regulatory principles that maintain it.This synaptic arrangement instantiated within the dendrite provides afunctional scaffold over which to conduct neuronal computation.

METHODSWhole-cell patch-clamp recording. The procedures for culturing hippocam-pal neurons, FM1-43 dye staining and whole-cell recording were essentially asdescribed previously21. Whole-cell recordings were obtained at room temper-ature (∼ 22 °C) from spindle-shaped pyramidal neurons kept in culture for11–20 d. The small experimental chamber (0.25 ml) was continuously per-fused (0.25 ml min–1) with Tyrode solution containing 145 mM NaCl, 3 mMKCl, 2.6 mM CaCl2, 1.3 mM MgCl2, 10 mM glucose and 10 mM HEPES (pH 7.4 with NaOH). For mEPSC and mIPSC recording from the same cell,the intracellular solution contained 120 mM CsMeSO3, 1 mM CaCl2, 10 mMNaCl, 10 mM EGTA, 2 mM Mg-ATP, 0.3 mM Na-GTP and 10 mM HEPES (pH 7.25 with CsOH). TTX (1 µM) was added to extracellular solution forrecording mEPSCs and mIPSCs. We found that damage during the prepara-tion for whole-cell recording could induce a high frequency of mEPSCs andmIPSCs, leading to inaccuracy in the calculated synaptic strength. Therefore,only recordings with low access resistance (< 10 MΩ) and holding current lessthan –60 pA (Vm = –70 mV) were analyzed. As pyramidal cells and interneu-rons differ substantially in their synaptic properties and their responses tostimulation that trigger synaptic plasticity, only pyramidal neurons wereincluded in this study. Pyramidal neurons were selected based on the followingselection criteria: pyramidal shape of soma with basal and apical dendrites,linear I/V relationship for the AMPA component of EPSCs, and presence ofNMDA receptors. For all experiments carried out under current-clamp (Figs. 2d,e and 3b), perforated patch clamp was used to obtain a long-lastingrecording with minimum perturbation of the intracellular environment. Thepatch pipette (1.2–2.2 MΩ resistance) contained 130 mM potassium glu-conate, 2 mM KCl, 8 mM NaCl, 0.2 mM EGTA, 1 mM MgCl2 and 10 mMHEPES (pH 7.2). Under these conditions, the normal cable properties of thedendritic tree are undisturbed. This allowed us to monitor the interplay of E/Isynaptic inputs under physiological conditions.

Calculation of the overall strength of synaptic inputs (GE and GI). GE and GI

are determined by

during bursting periods, or

during miniature synaptic events, where q is the charge transfer during quan-tal transmission. The following formula was used in calculating the ratio of E/Iinput strength (GI/GE) during bursting synaptic inputs. At the reversal poten-

t t

GE = Σ qE / t(Vm – VE) and GI = Σ qI / t(Vm – VI) 0 0

t t

GE = Σ EPSCs / t(Vm – VE) and GI = ΣIPSCs / t(Vm – VI) 0 0


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

tial for bursting inputs, excitatory and inhibitory inputs cancel each other:GE(Vrev – VE) + GI(Vrev – VI) = 0. Thus, the ratio of their conductances can becalculated by GI/GE = (Vrev – VE)/(VI – Vrev). Given the intracellular and extra-cellular solutions used in the present study, VE will be close to 0 and VI close to–65 mV. Because the kinetic properties of synaptic receptors and synapticrelease probability are temperature-sensitive, the ratio of GI/GE determined atroom temperature might differ markedly from the ratio at physiological tem-perature. We have compared the ratios of GI/GE at 22 and 33 °C (Fig. 2c).Raising the temperature in the recording chamber increased the amplitudeand sped up the decay of synaptic events. However, the integral of synapticconductance did not change markedly, resulting in a similar E/I ratio meas-ured at room temperature. Finally, to get an accurate measure of GE and GI, allactive synapses were voltage-clamped.

Evaluation of the degree of space clamp. Two criteria were used. First, ifsynaptic events had been attenuated due to poor space clamping, the rise timeof synaptic events originating at synapses located at distant locations on thedendritic tree would have been much slower than those from proximal loca-tions. The median values of 20–80% rise time of all of our recordings are <0.5 ms, suggesting that those synapses that generated miniature synapticevents were measured under a reasonable degree of voltage clamp(Supplementary Fig. 1 online, panel a). This analysis could not rule out thepossibility that some distant synapses are electronically so far away fromsomatic recording sites that their synaptic currents are attenuated to the pointthat they could not be detected. To exclude this possibility, we compared thereversal potential and kinetics of the synaptic currents evoked at various loca-tions on the dendritic tree (Supplementary Fig. 1 online, panels b–d). Neitherreversal potentials nor time courses of synaptic current changed significantlywith increased distance from the soma, demonstrating that synaptic conduc-tances from distal synapses can be detected at a somatic recording site. As thearea of synaptic conductance, rather than peak amplitude, is used to representquantal synaptic strength, errors associated with inadequate space clamp arefurther reduced.

Perturbation of E/I balance by pharmacological agents. For experiments inFigure 5, neurons were treated with various pharmacological agents (fluni-trazepam (Sigma), NBQX (Tocris), bicuculline (Tocris) and TTX (Biotium))for 48 h before recording. We determined the time course for the restorativeE/I response to perturbations. Generally, it took 12 h of treatments to alter E/Isynapses, and the responses stabilized after 24 h (data not shown). Synapticstrengths (GE and GI) were determined from mEPSCs and mIPSCs, respec-tively. Gin was calculated by plotting membrane potential as a function ofinjected current size.


ACKNOWLEDGMENTSI thank E. Hueske and B. Li for participating in part of the experiments andcontributing to Figure 1; M. Wilson, X.J. Wang, N. Wilson, S. Sadeghpour,B. Krupa and T. Emery for comments on the manuscript. This work wassupported by grants from National Institutes of Health and the RIKEN–MITNeuroscience Research Center.



1. Koch, C., Poggio, T. & Torre, V. Nonlinear interactions in a dendritic tree: localiza-tion, timing, and role in information processing. Proc. Natl. Acad. Sci. USA 80,2799–2802 (1983).

2. Bush, P.C. & Sejnowski, T.J. Effects of inhibition and dendritic saturation in simu-lated neocortical pyramidal cells. J. Neurophysiol. 71, 2183–2193 (1994).

3. Ferster, D. & Miller, K.D. Neural mechanisms of orientation selectivity in the visualcortex. Annu. Rev. Neurosci. 23, 441–471 (2000).

4. Fried, S.I., Munch, T.A. & Werblin, F.S. Mechanisms and circuitry underlying direc-tional selectivity in the retina. Nature 420, 411–414 (2002).

5. Gabbiani, F., Krapp, H.G., Koch, C. & Laurent, G. Multiplicative computation in avisual neuron sensitive to looming. Nature 420, 320–324 (2002).

6. Taylor, W.R., He, S., Levick, W.R. & Vaney, D.I. Dendritic computation of directionselectivity by retinal ganglion cells. Science 289, 2347–2350 (2000).

7. Schummers, J., Marino, J. & Sur, M. Synaptic integration by V1 neurons depends onlocation within the orientation map. Neuron 36, 969–978 (2002).

8. Shu, Y., Hasenstaub, A. & McCormick, D.A. Turning on and off recurrent balancedcortical activity. Nature 423, 288–293 (2003).

9. Chance, F.S., Abbott, L.F. & Reyes, A.D. Gain modulation from background synapticinput. Neuron 35, 773–782 (2002).

10. Beaulieu, C., Kisvarday, Z., Somogyi, P., Cynader, M. & Cowey, A. Quantitative dis-tribution of GABA-immunopositive and -immunonegative neurons and synapses inthe monkey striate cortex (area 17). Cereb. Cortex 2, 295–309 (1992).

11. Beaulieu, C. & Somogyi, P. Targets and quantitative distribution of GABAergicsynapses in the visual cortex of the cat. Eur. J. Neurosci. 2, 296–303 (1990).

12. Megias, M., Emri, Z., Freund, T.F. & Gulyas, A.I. Total number and distribution ofinhibitory and excitatory synapses on hippocampal CA1 pyramidal cells.Neuroscience 102, 527–540 (2001).

13. Miles, R., Toth, K., Gulyas, A.I., Hajos, N. & Freund, T.F. Differences betweensomatic and dendritic inhibition in the hippocampus. Neuron 16, 815–823 (1996).

14. McBain, C.J. & Fisahn, A. Interneurons unbound. Nat. Rev. Neurosci. 2, 11–23(2001).

15. Hausser, M., Spruston, N. & Stuart, G.J. Diversity and dynamics of dendritic signal-ing. Science 290, 739–744 (2000).

16. Poirazi, P. & Mel, B.W. Impact of active dendrites and structural plasticity on thememory capacity of neural tissue. Neuron 29, 779–796 (2001).

17. Larkman, A.U. Dendritic morphology of pyramidal neurones of the visual cortex ofthe rat: III. Spine distributions. J. Comp. Neurol. 306, 332–343 (1991).

18. Ben-Ari, Y., Cherubini, E., Corradetti, R. & Gaiarsa, J.L. Giant synaptic potentials inimmature rat CA3 hippocampal neurones. J. Physiol. 416, 303–325 (1989).

19. Leinekugel, X. et al. Correlated bursts of activity in the neonatal hippocampus invivo. Science 296, 2049–2052 (2002).

20. Murnick, J. G., Dube, G., Krupa, B. & Liu, G. High-resolution iontophoresis for sin-gle-synapse stimulation. J. Neurosci. Methods 116, 65–75 (2002).

21. Liu, G., Choi, S. & Tsien, R. W. Variability of neurotransmitter concentration andnonsaturation of postsynaptic AMPA receptors at synapses in hippocampal culturesand slices. Neuron 22, 395–409 (1999).

22. Davis, G.W. & Goodman, C.S. Genetic analysis of synaptic development and plastic-ity: homeostatic regulation of synaptic efficacy. Curr. Opin. Neurobiol. 8, 149–156(1998).

23. Turrigiano, G.G. & Nelson, S. B. Hebb and homeostasis in neuronal plasticity. Curr.Opin. Neurobiol. 10, 358–364 (2000).

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

25. Turrigiano, G.G., Leslie, K.R., Desai, N.S., Rutherford, L.C. & Nelson, S.B. Activity-dependent scaling of quantal amplitude in neocortical neurons. Nature 391,892–896 (1998).

26. Murthy, V.N., Schikorski, T., Stevens, C.F. & Zhu, Y. Inactivity produces increases inneurotransmitter release and synapse size. Neuron 32, 673–682 (2001).

27. Paradis, S., Sweeney, S.T. & Davis, G.W. Homeostatic control of presynaptic release istriggered by postsynaptic membrane depolarization. Neuron 30, 737–749 (2001).

28. Burrone, J., O’Byrne, M. & Murthy, V.N. Multiple forms of synaptic plasticity trig-gered by selective suppression of activity in individual neurons. Nature 420,414–418 (2002).

29. Kilman, V., van Rossum, M.C. & Turrigiano, G.G. Activity deprivation reduces minia-ture IPSC amplitude by decreasing the number of postsynaptic GABAA receptorsclustered at neocortical synapses. J. Neurosci. 22, 1328–1337 (2002).

30. Knott, G.W., Quairiaux, C., Genoud, C. & Welker, E. Formation of dendritic spineswith GABAergic synapses induced by whisker stimulation in adult mice. Neuron 34,265–273 (2002).

31. Desai, N.S., Cudmore, R.H., Nelson, S.B. & Turrigiano, G.G. Critical periods forexperience-dependent synaptic scaling in visual cortex. Nat. Neurosci. 5, 783–789(2002).

32. Morales, B., Choi, S.Y. & Kirkwood, A. Dark rearing alters the development ofGABAergic transmission in visual cortex. J. Neurosci. 22, 8084–8090 (2002).

33. Cossart, R. et al. Dendritic but not somatic GABAergic inhibition is decreased inexperimental epilepsy. Nat. Neurosci. 4, 52–62 (2001).

34. Rumsey, C.C. & Abbott, L.F. Equalization of synaptic efficacy by activity- and tim-ing-dependent synaptic plasticity. J. Neurophysiol. (2003).

35. Johnston, D. et al. Active dendrites, potassium channels and synaptic plasticity.Philos. Trans. R. Soc. Lond. B Biol. Sci. 358, 667–674 (2003).

36. Wilson, M.A. & McNaughton, B.L. Dynamics of the hippocampal ensemble code forspace. Science 261, 1055–1058 (1993).

37. Margrie, T.W., Brecht, M. & Sakmann, B. In vivo, low-resistance, whole-cell record-ings from neurons in the anaesthetized and awake mammalian brain. Pflugers Arch.444, 491–498 (2002).

38. Anderson, J.S., Carandini, M. & Ferster, D. Orientation tuning of input conductance,excitation, and inhibition in cat primary visual cortex. J. Neurophysiol. 84, 909–926(2000).

39. Zhang, L.I., Tan, A.Y., Schreiner, C.E. & Merzenich, M.M. Topography and synapticshaping of direction selectivity in primary auditory cortex. Nature 424, 201–205(2003).

40. Hensch, T.K. et al. Local GABA circuit control of experience-dependent plasticity indeveloping visual cortex. Science 282, 1504–1508 (1998).


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Objects in our visual environment are normally distinguishable by anumber of cues such as luminance, shape, color and texture, yet thepercept of motion is qualitatively invariant across different cues1,2.This perceptual invariance, known as form-cue invariance, is paral-leled by a similar response invariance of many apparently motion-sensitive neurons in primate cortical middle temporal visual area3,4,in primate cortical dorsal middle superior temporal area5 and inavian tectum6–8. Form-cue invariance may be a useful operationalprinciple. To perform optimally in a variable environment, a motionanalyzer should generalize across stimulus cues, thus encoding themotion of a stimulus regardless of the cue that enables it to be seen.Despite the apparent importance of form cue–invariant motion pro-cessing, the underlying cellular mechanisms for form-cue invarianceare poorly understood.

The avian tectal slice (Fig. 1a) provides an ideal preparation to studythe central cellular mechanisms for the analysis of dynamic spatiotem-poral stimuli9. Neurons in deep layers of the avian tectum respond tosmall moving objects largely independently of object-defining attrib-utes, but they do not respond to static stationary stimuli6,8,10–12. Deeptectal neurons of the morphologically and physiologically identifiedstratum griseum centrale (SGC-I) type (Fig. 1b) are particularly suit-able for cellular studies of spatiotemporal processing. Neurons of thisapparently motion-sensitive subpopulation have somata in layer 13,have large circular dendritic fields, extend their dendrites radially andterminate with specialized dendritic endings in layer 5b9,13. Here theSGC-I dendritic endings make monosynaptic contact with axons14

that are derived from a population of small retinal ganglion cells(RGCs)15. These RGC axons form a topographic map on the tectal sur-face and penetrate the outer tectal layers radially16.

The fact that a tectal neuron, which is only one synapse away fromthe retina, apparently displays form cue–invariant motion sensitivity

suggested to us the hypothesis that form cue–invariant motion sensi-tivity could be mediated by cellular rather than network mechanisms.To explore this hypothesis, we obtained in vitro whole-cell recordingsfrom SGC-I type neurons in combination with a series of spatiotem-poral stimulation experiments with stimulus electrodes at differentlocations in the retinotectal signal pathway. We then analyzed thefunctional role of the measured cellular properties with a computersimulation of an SGC-I model in response to assumed retinal repre-sentations of various dynamic spatiotemporal visual stimuli.

RESULTSSGC-I response to localized synaptic stimulationTo investigate the signal transfer from the RGC axon to the SGC-Isoma, we locally stimulated a small group of RGC axons with shortcurrent pulses that were delivered with a stimulus electrode in layers2–4 and recorded the response in the SGC-I soma (Fig. 1a,b). In allSGC-I cells tested, single-pulse stimulation resulted in an all-or-nonesharp-onset cellular response consisting of either one to three actionpotentials riding on a broader depolarization or no response (Fig. 1c).A previous investigation9 indicates that the sharp-onset response tosynaptic stimulation may be generated remotely from the soma, pre-sumably at the dendritic ending.

To characterize this response, we carried out one-site regular pulsetrain synaptic stimulation experiments (Fig. 2a). The sharp-onsetresponse to each stimulus pulse was probabilistic. For all stimulationintervals tested, the response probability reached a steady state afterthe second stimulus pulse (Fig. 2b). Therefore, we pooled the datafrom pulses 2–10 for the same stimulation interval to derive the meansteady-state response probability for that stimulation interval (Fig. 2c). This probability showed an exponential time dependence ofthe form P(∆t) = Pmax (1 – e–∆t/t0), where ∆t is the stimulation interval

1Institute of Biology II, Rheinisch-Westfälische Technische Hochschule Aachen, 52074 Aachen, Germany. 2Department of Physics, CB 1105, Washington University,St. Louis, Missouri 63130, USA. Correspondence should be addressed to R.W. ([email protected]).

Published online 29 February 2004; doi:10.1038/nn1204

Synaptic dynamics mediate sensitivity to motionindependent of stimulus detailsHarald Luksch1, Reza Khanbabaie2 & Ralf Wessel2

Humans and other animals generally perceive motion independently of the cues that define the moving object. To understand theunderlying mechanisms of this generalization of stimulus attributes, we have examined the cellular properties of avian wide-fieldtectal neurons that are sensitive to a variety of moving stimuli but not to static stationary stimuli. This in vitro study showedphasic signal transfer at the retinotectal synapse and binary dendritic responses to synaptic inputs that interact in a mutuallyexclusive manner in the postsynaptic tectal neuron. A model of the tectal circuitry predicts that these two cellular propertiesmediate sensitivity to a wide range of dynamic spatiotemporal stimuli, including moving stimuli, but not to static stationarystimuli in a tectal neuron. The computation that is independent of stimulus detail is initiated by tectal neurons and is completedby rotundal neurons that integrate outputs from multiple tectal neurons in a directionally selective manner.

A R T I C L E S


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

and Pmax = 0.87 and t0 = 2,025 ms for the two fitting parameters.Thus, the signal transfer from the RGC axon to the SGC-I soma wasphasic in a time-dependent probabilistic manner.

The phasic response could have originated either at the synapse orin the dendritic pathway and soma. To localize the site of phasic signaltransfer, we stimulated dendritic endings directly with a stimuluselectrode in layer 5, thus bypassing the synapse (Fig. 1b). This directelectrical stimulation of dendritic endings led to a sharp-onsetresponse that was essentially identical to the response to synapticstimulation but with a shorter latency, as described previously9. Wecarried out one-site paired-pulse direct electrical stimulation of den-dritic endings (Fig. 2d) and measured the SGC-I response probabilityfor the second pulse for different stimulation intervals (Fig. 2f). Theprobability shows an exponential time dependence of the form Pdirect (∆t) = Pmax2 (1 – e (t1 – ∆t)/t2) , where ∆t is the stimulation inter-val and the fitting parameters are Pmax2 = 1.0, t1 = 4 ms and t2 = 12 ms.We refer to the sum of the time shift, t1, and the exponential time con-stant, t2, as τdirect = t1 + t2 = 16 ms. In conclusion, signal transferwithin SGC-I neurons is tonic at time scales that are two orders ofmagnitude shorter than signal transfer at the retinotectal synapse.Therefore, the phasic signal transfer (Fig. 2c) originates at the retino-tectal synapse.

To investigate the cell’s response to spatiotemporal synaptic inputsand to test for potential distance dependence of the interaction, weplaced two stimulus electrodes in layers 2–4 at distances of250–1,500 µm apart, thus stimulating two separate groups of RGCaxons (Fig. 2e). Because of the sparse spatial distribution of dendriticendings for one SGC-I neuron, each group of stimulated RGC axonstypically activated only one dendritic ending of an SGC-I neuronunder consideration9. We recorded from an SGC-I cell that receivedinputs from both groups of axons and stimulated the two sites intemporal sequence with varying stimulation intervals. The measuredresponse probabilities to the second stimulus pulse for one stimula-tion interval showed no statistically significant distance dependencefor the two-site synaptic stimulation (Fig. 2f). Reversing thesequence in which the sites were stimulated had no effect on theresponse (data not shown). Thus we pooled the data from the two-site synaptic stimulation from different distances to derive the mean

response probability to the second pulse for each stimulation inter-val (Fig. 2f). The probability shows an exponential time dependenceof the form P2-site (∆t) = Pmax3 (1 – e (t3 – ∆t)/t4), where ∆t is the stim-ulation interval and the fitting parameters are Pmax3 = 1.0, t3 = 16 msand t4 = 14 ms. We refer to the sum of the time shift, t3, and the expo-nential time constant, t4, as the ‘interaction time’, τ = t3 + t4 = 30 ms. Although the exponential time constants for the two stimulussituations are similar, there seems to be a difference in the time shiftfor direct stimulation at one site, t1 = 4 ms, and for sequential synap-tic stimulation at two sites, t3 = 16 ms (Fig. 2f). At present, it is notknown whether this difference in time shift is mediated by theretinotectal synaptic latency of 8 ms (ref. 9), by intracellular den-dritic events, by the presynaptic horizontal network in layer 5 (refs. 14,17,18) or by combinations thereof.

In summary, this series of experiments showed a number of non-linear cellular properties: (i) the SGC-I cell responded to synapticstimulation in a binary manner; (ii) the signal transfer from the RGCaxon to the SGC-I soma was phasic in a time-dependent probabilisticmanner over large time scales; (iii) the site of the phasic signal trans-fer was the retinotectal synapse; (iv) the phasic synaptic signal trans-fer largely reached a steady state after the second stimulus pulse;(v) synaptic inputs at two locations typically interacted in a mutuallyexclusive manner when delivered within the interaction time ofapproximately 30 ms, a time scale that is two orders of magnitudeshorter than the time scale for phasic transfer of synaptic signals and(vi) the postsynaptic interaction was independent of the stimuluselectrode distances for the distances examined.

Functional role of the nonlinear cellular propertiesTo examine the role of the observed nonlinear cellular properties inshaping the response of SGC-I neurons to assumed retinal represen-tations of dynamic spatiotemporal visual stimuli, we constructed amodel of the retinal inputs and the SGC-I cell (Fig. 3). Although anumber of mechanisms may contribute to the tectal analysis ofdynamic spatiotemporal visual stimuli, in this model we focusedexclusively on the observed binary response, the phasic and proba-bilistic signal transfer and the mutually exclusive response to multipleinputs within the interaction time. This SGC-I model allowed us todetermine the limits of what the observed cellular nonlinear proper-ties can explain and thereby establish whether they are important ele-ments in the tectal analysis of assumed retinal representations ofdynamic spatiotemporal visual stimuli.

First, we investigated the model’s response to the assumed retinalrepresentation (Fig. 4a) of a static stationary luminance-defined barstimulus (Fig. 3c). An RGC spike arriving at a dendritic ending causeda dendritic spike with probability P(∆t) = Pmax (1 – e–∆t/t0). For physi-ological RGC spike rates, the time interval, ∆t, between two RGCspikes was typically much smaller than the measured parameter t0 =

A R T I C L E S


Figure 1 SGC-I morphology, location of stimulus electrodes and response tosynaptic stimulation. (a) Schematic view of the tectum slice and the positionof the SGC. Cer, cerebellum. (b) Reconstruction of an SGC-I neuron labeledwith biocytin after whole-cell patch recording. The characteristics of this celltype include a large dendritic field, the position of the soma in the upperhalf of the SGC and the arrangement of the bottlebrush dendritic endings inthe retinorecipient layer 5b, as indicated by the shaded layer. Note thepositioning of the stimulation electrodes above and within theretinorecipient layer 5b. For clarity, only a small subset of RGC axonschematics are shown. (c) Single-pulse stimulation results in either a sharp-onset response consisting of one to three action potentials riding on abroader depolarization (main graph) or no response at all (inset). The restingmembrane potential is indicated below the recording trace.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

ously spiked within the interaction time of 30 ms. As a result of thisnonlinear interaction of dendritic spikes, the SGC-I spike train wasmore regular (Fig. 4g) than the spike train of dendritic endings. Onaverage, the SGC-I firing rate in response to the moving bar stimuluswas 18.7 ± 1.1 Hz (mean ± s.d.; n = 5 trials of 3-s duration). Withoutthe phasic synaptic signal transfer, P(∆t) = Pmax, the SGC-I modelresponded to a moving bar with a rate of 32.6 ± 0.3 Hz (mean ± s.d.;n = 5 trials of 3-s duration; Fig. 4h).

The response of the SGC-I model was not limited to moving stim-uli alone; other dynamic spatiotemporal stimuli caused strongresponses. For instance, the same bar jumping to random locationswithin the receptive field every 10 ms resulted in a model SGC-I rateof 29.5 ± 0.6 Hz (mean ± s.d.; n = 5 trials of 3-s duration).

In conclusion, the SGC-I firing rate in response to a dynamic lumi-nance-defined spatiotemporal stimulus was significantly larger thanthe firing rate in response to a stationary bar. Thus, the SGC-I modelwas able to differentiate between a static stationary and a dynamicspatiotemporal stimulus such as a moving small stimulus bar. In con-trast, without phasic synaptic signal transfer, the model SGC-I cellresponded strongly to both static stationary stimuli and dynamic spa-tiotemporal stimuli and thus failed to distinguish between these twostimulus classes. This result demonstrates that the phasic signal trans-fer is a crucial biophysical element for this stimulus classification.

2,025 ms. Therefore, P(∆t) was small, leading to sparse dendritic spik-ing (Fig. 4b), which in turn led to sparse SGC-I soma spiking (Fig. 4c). Because of the sparse dendritic spiking, there was typicallyno dendritic spike interaction. On average, the simulated SGC-I firingrate in response to the stationary bar stimulus was 1.5 ± 0.5 Hz (mean ± s.d.; n = 5 trials of 3-s duration). To evaluate the functionalrole of the phasic retinotectal synaptic signal transfer, we repeated thesimulation with a stationary bar but now with a constant probability,P(∆t) = Pmax (without phasic signal transfer). In this case, more thanone dendritic ending generated spikes simultaneously at every timestep and, as a result of the nonlinear spike interaction, which wasquantified with the interaction time of 30 ms, the SGC-I modelresponded with a regular rate of 33.3 Hz to a stationary bar (Fig. 4d).

The model response to the assumed retinal representation of amoving luminance-defined bar stimulus (Fig. 3c) was completely dif-ferent. With each time step, the assumed retinal representation (Fig. 4e) of the moving stimulus bar passed over a small number ofnew dendritic endings. These new dendritic endings initially had alarge probability, P(∆t → ∞) = Pmax, of responding with a dendriticspike to the arrival of an RGC spike. Therefore, the sum of dendriticspikes per simulation time step was often one (Fig. 4f). When morethan one dendritic ending spiked simultaneously, only one spike wasgenerated in the SGC-I soma and only if the SGC-I had not previ-

A R T I C L E S


Figure 2 Response of SGC-I neurons to synaptic and direct stimulation of dendritic endings. (a) Pulse train synaptic stimulation at one location. Top trace:typical response to one trial of ten pulses. The resting membrane potential was –53 mV. Center traces: response raster plot of the same neuron for ten trials.Each tick mark indicates the occurrence of a sharp-onset response. Bottom trace: Pulse train of stimulus pulses delivered to one location in layers 2–4. Thestimulation interval was 1,500 ms. (b) Response probability as a function of pulse number for various stimulation intervals. The response probability dropssignificantly from pulse 1 to pulse 2, but for pulses 2–10 the response probability remains largely constant. (c) Mean response probability to pulse numbers 2to 10 versus the interval between pulses at one location. The line was obtained by fitting the data points with an exponential function (see text). (d) Directpaired-pulse stimulation of dendritic endings at one location for two different time intervals between stimulus pulses. The response to the second stimuluspulse failed at a stimulus interval of 15 ms. Note the different time scale in a and d. (e) Paired-pulse synaptic stimulation at two locations for two differenttime intervals between stimulus pulses. The response to the second stimulus pulse failed at a stimulus interval of 15 ms. Note the different time scale in a ande. (f) Response probability to the second pulse versus stimulation interval for paired-pulse direct stimulation at one location (open circle) and paired-pulsesynaptic stimulation at two locations (filled square). The lines were obtained by fitting the data points with an exponential function (see text). Inset: responseprobability to the second pulse versus stimulus electrode distance for paired-pulse synaptic stimulation at two locations for different stimulation intervals.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Tectal neurons are also sensitive to second-order motion in vivo7,raising the question of whether the SGC-I model contains all the bio-physical elements to respond to second-order dynamic spatiotem-poral stimuli. Second-order motion stimuli are not luminance-defined but rather correspond to regions of a common moving‘process,’ for instance in a field of random dots19–21. On any singleframe of such a second-order motion stimulus, nothing but a patternof random dots is present, but in successive frames the pixel valueswithin a moving area, such as a rectangle, are updated according to arule that leaves the average dot density constant and equal to the sta-tionary background (Fig. 3d). In a second-order motion stimulus,neither average luminance nor individual dots are moving. What ismoving is the process of selecting a new set of dots.

First, we investigated the response of the SGC-I model to theassumed retinal representation (Fig. 5a) of a static stationary whole-

field pattern of random dots (Fig. 3d). As discussed before, an RGCspike caused an SGC-I dendritic spike with a probability P(∆t) thatwas small for physiological RGC spike rates. The sparse dendriticspiking (Fig. 5b) caused sparse SGC-I soma spiking (Fig. 5c). Onaverage, the SGC-I firing rate in response to the stationary pattern ofrandom dots was 1.9 ± 0.7 Hz (mean ± s.d.; n = 5 trials of 3-s dura-tion). In contrast, without the phasic synaptic signal transfer, P(∆t) =Pmax, the SGC-I model responded with a regular rate of 33.3 Hz to thestationary dot stimulus (Fig. 5d).

Next, we investigated the SGC-I model response to the retinalrepresentation (Fig. 5e) of an uncorrelated second-order motionstimulus (Fig. 3d), which has been used in an electrophysiologicalstudy3. On any single frame, a whole-field pattern of random dotsis present, but in successive frames the pixel values within a movingrectangle (Fig. 3d) are replaced with uncorrelated random dots.The resultant percept in humans is that of a moving, twinkling rec-tangle. For most of the dendritic endings that were contacted byexcited RGC axons, the probability P(∆t) for dendritic spiking wassmall. For dendritic endings within the moving rectangle, the situ-ation was different. Because the random dots within the rectanglewere replaced by an uncorrelated set at every time step, most den-dritic endings had a large probability, P(∆t → ∞) = Pmax. Thuswhen an RGC spike arrived, dendritic endings within the rectanglewere likely to fire, leading to a rate of dendritic spikes that was

A R T I C L E S


Figure 4 Response of the model SGC-I cell to a static stationary (a−d) ormoving (e−h) luminance-defined bar stimulus (Fig. 3c). (a) The summedspikes per simulation time step (10 ms) in response to a stationary barstimulus of those excited RGC axons that terminated on dendritic endings.(b) The summed spikes per simulation time step of all dendritic endings(BBE) of one SGC-I model cell in response to a stationary bar stimulus. (c) The spikes per simulation time step of one SGC-I model cell in responseto a stationary bar stimulus. (d) The spikes per simulation time step of oneSGC-I model cell in response to a stationary bar stimulus without phasicsynaptic signal transfer. (e) The summed spikes per simulation time step ofthose excited RGC axons that terminated on dendritic endings in responseto a moving bar stimulus. (f) The summed spikes per simulation time stepof all dendritic endings of one SGC-I model cell in response to a moving barstimulus. (g) The spikes per simulation time step of one SGC-I model cell inresponse to a moving bar stimulus. (h) The spikes per simulation time stepof one SGC-I model cell in response to a moving bar stimulus withoutphasic synaptic signal transfer.

Figure 3 Structure of the SGC-I model and the visual stimuli. (a) Schematic of the model structure in cross-section. A visual stimulus(black square) activates RGC axons. A fraction of the RGC axons (blackvertical arrows) terminate on the dendritic endings of one SGC-I neuron.The remaining RGC axons (gray vertical arrows) terminate on other SGC-Ineurons and are not considered further in the simulation. The dendriticendings (BBE) are directly connected to the SGC-I soma. (b) Spatialdistribution of model SGC-I dendritic endings for one model neuron,corresponding to a top view of a. The spatial extensions of the squaredendritic field are 3,000 × 3,000 µm. (c) Representation of the barstimulus shown in the same spatial dimensions as in b. In the simulations,the stimulus bar was used for static stationary or dynamic spatiotemporalstimuli including first-order motion. (d) Representation of the random-dotstimulus shown in the same spatial dimensions as in b. In the simulations,the random-dot stimulus was used for static stationary or dynamicspatiotemporal stimuli including second-order motion.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

larger than that in the static stationary stimulus case (Fig. 5b,f).The large dendritic spike rate caused interactions between den-dritic spikes and a large SGC-I soma spike rate (Fig. 5g). On aver-age, the firing rate of the SGC-I model in response to theuncorrelated second-order motion stimulus was 14.0 ± 1.9 Hz(mean ± s.d.; n = 5 trials of 3-s duration). Without the phasicsynaptic signal transfer, P(∆t) = Pmax, the SGC-I model respondedwith a regular rate of 33.3 Hz to the uncorrelated second-ordermotion stimulus (Fig. 5h).

In addition, the SGC-I model responded to a variety of otherdynamic spatiotemporal second-order stimuli. Smaller responses ofthe SGC-I model were obtained for the correlated second-ordermotion stimulus, in which a set of random dots moved across a back-ground of random dots of equal density7. On average, the firing rateof the SGC-I model in response to the correlated second-ordermotion stimulus of a rectangle of the same size was 9.7 ± 1.1 Hz(mean ± s.d.; n = 5 trials of 3-s duration; data not shown). Responsesof the SGC-I model were also obtained from second-order dynamicspatiotemporal stimuli that were not moving. For instance, consider awhole-field pattern of stationary static random dots, but in successiveframes the pixel values within a stationary rectangle are replaced withuncorrelated random dots (Fig. 3d). The resultant percept in humansis that of a stationary twinkling rectangle. The resultant response inthe SGC-I model was 11.1 ± 0.5 Hz (mean ± s.d.; n = 5 trials of 3-sduration; data not shown). The model response increased to 16.0 ±1.4 Hz (mean ± s.d.; n = 5 trials of 3-s duration; data not shown)when the same number of dots (six dots) was replaced at randomfrom anywhere within the receptive field rather than just within therectangle. For a whole-field pattern of random dots in which the set ofrandom dots within the receptive field is replaced with a new set ofuncorrelated random dots in successive frames, the SGC-I modelresponse increased to 31.8 ± 0.2 Hz (mean ± s.d.; n = 5 trials of 3-sduration; data not shown).

In conclusion, the SGC-I firing rate in response to a dynamic spa-tiotemporal second-order stimulus, such as second-order motion,was significantly larger than the firing rate in response to a static sta-tionary set of random dots. Thus for the non-luminance-definedobject, the SGC-I model was able to differentiate between a static sta-tionary and a dynamic spatiotemporal stimulus but failed to do sowithout the phasic synaptic signal transfer, that is, a constant proba-bility P(∆t) = Pmax. This demonstrates again that phasic signal trans-fer is a crucial biophysical element for stimulus classification. Further,

the SGC-I representation of a dynamic spatiotemporal second-orderstimulus is similar to the SGC-I representation of a dynamic spa-tiotemporal first-order stimulus. Hence, the SGC-I differentiationbetween static stationary and dynamic spatiotemporal stimuli, suchas motion, is independent of the details of the stimulus.

To what extent do the model results depend on the assumptionsmade for the retinal representation of the visual stimuli? We assumedthat the RGC axons were silent in the absence of a stimulus and firedrandomly with an average rate of 80 Hz when stimulated. Althoughthis choice corresponds to biologically plausible RGC responses22,23,there may be large variations in the parameters for different subpopu-lations of RGCs in different animals24. To quantify to what extent theSGC-I stimulus classification as static stationary or dynamic spa-tiotemporal stimuli depends on the stimulated mean rate of RGC fir-ing, we repeated the simulations for different stimulated mean RGCfiring rates (while keeping the spontaneous rate at 0). Because of theimportance of motion stimuli in nature and in the available experi-mental literature, we restricted this analysis of parameter sensitivityto the static stationary and moving stimuli previously examined(Figs. 4 and 5) and considered the SGC-I model ‘motion-sensitive’(but see Discussion) when the firing rates of the SGC-I model inresponse to static stationary or moving stimuli were significantly dif-ferent. The motion sensitivity of the SGC-I model to first- and sec-ond-order motion proved to be largely independent of the stimulatedRGC rate over a wide range of physiological RGC firing rates (Fig. 6a). At low mean RGC rates, the model predicted that the SGC-Imotion sensitivity breaks down first for second-order motion (<20 Hz) and then for first-order motion (<1 Hz). Given that thestimulated RGC mean firing rate can be modulated by light inten-sity22, this model result predicts that the SGC-I motion sensitivity waslargely independent of the brightness of the stimulus over a wide

A R T I C L E S


Figure 5 Response of the SGC-I model to a static stationary (a−d) anduncorrelated second-order motion (e−h) random-dot pattern (Fig. 3d). (a) The summed spikes per simulation time step (10 ms) in response to astatic stationary random-dot stimulus of those excited RGC axons thatterminated on dendritic endings. (b) The summed spikes per simulationtime step of all dendritic endings (BBE) of one SGC-I model cell inresponse to a static stationary random-dot stimulus. (c) The spikes persimulation time step of one SGC-I model cell in response to a staticstationary random-dot stimulus. (d) The spikes per simulation time step ofone SGC-I model cell in response to a static stationary random-dot stimuluswithout phasic synaptic signal transfer. (e) The summed spikes persimulation time step of those excited RGC axons that terminated ondendritic endings in response to uncorrelated second-order motion. (f) Thesummed spikes per simulation time step of all dendritic endings of oneSGC-I model cell in response to uncorrelated second-order motion. (g) Thespikes per simulation time step of one SGC-I model cell in response touncorrelated second-order motion. (h) The spikes per simulation time stepof one SGC-I model cell in response to an uncorrelated second-order motionwithout phasic synaptic signal transfer.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

range of light intensities.In contrast to the marked independence of the motion sensitivity

of the model to the stimulated RGC mean firing rate, it was extremelydependent on the spontaneous RGC mean firing rate (the RGC activ-ity in the absence of any stimulus within its receptive field). Indeed,the motion sensitivity of the SGC-I model broke down at a sponta-neous RGC rate of 0.5 Hz or higher (Fig. 6b). Most interestingly, thismodel prediction is consistent with the experimental observation thatavian RGCs have little or no spontaneous activity25, a fact that distin-guishes avian RGCs from primate RGCs, which have a spontaneousactivity of 20 Hz26.

To what extent do the model results depend on the numerical val-ues of Pmax and t0? This question is important for two reasons. First,the numerical values for Pmax and t0 were measured in an in vitropreparation at room temperature, and the in vivo extrapolation mayhave a large error. Second, the retinotectal synapse in layer 5 is embed-ded in a dense network of horizontal17,18 and feedback27 connections.Although our knowledge of the cellular details of these connections islimited14, it is biologically plausible that these connections modulatethe two parameters of phasic retinotectal signal transfer, Pmax and t0,potentially in a stimulus- or context-dependent manner. To quantifythese parameter dependences, we repeated the simulations for differ-ent parameter values. The SGC-I model cell remained motion sensi-tive over a large range of values of Pmax around the measured value ofPmax = 0.87 (Fig. 6c) and over a large range of values of t0 around themeasured value of t0 = 2,025 ms (Fig. 6d).

The speed tuning of the SGC-I model allowed an interesting com-parison with experimental data. Qualitatively, within our model, thefollowing speed tuning to a small moving object was expected.Because of the spike generation in individual dendritic endings9 and

the spatial distribution of dendritic endings13, on average the SGC-Ispike rate should increase linearly with increasing speed at low speed.At higher speed, the mutually exclusive interaction of dendritic spikeskicks in and thus the SGC-I spike rate will not increase further withincreasing speed. To quantify the speed-tuning curve, we conductedsimulations with a moving bar stimulus (Fig. 3c) at various speedsbetween 0 and 100 deg/s. This produced a linear increase in the SGC-I rate with increasing speed in the range of 0–10 deg/s (Fig. 6e).For higher speeds, the SGC-I rate increased sublinearly and reachedits maximum rate of 33 Hz at 60 deg/s. In the range of 0–10 deg/s, thetotal number of spikes generated by an object moving through thesame area in the visual field was independent of the speed but,because of the mutually exclusive interaction, decreased for increas-ing speed (Fig. 6f). Both the range of linear speed tuning and the con-stancy of the total number of spikes within this speed range wereconsistent with in vivo experimental results (N. Troje & B. Frost, Soc.Neurosci. Abstr. 24, 642.9, 1998). The model failed, however, to repro-duce the experimentally observed decrease in the SGC-I rate forspeeds above 20–40 deg/s7,28, thus suggesting the existence of apresently unknown additional suppressive mechanism for largerspeeds that was not included in the model. Of note, the simulationresults predict that for a second-order motion stimulus the SGC-Irate is largely independent of the speed (Fig. 6e), but the total numberof spikes generated by second-order motion moving through thesame area in the visual field decreased with increasing speed (Fig. 6f).Because the spikes were mostly caused by the random update of thedots within the rectangle, the number of spikes decreased withdecreasing travel time (increasing speed). At present, no experimentalspeed tuning data are available for the second-order motion stimulusin avian tectum.

A R T I C L E S


Figure 6 Parameter dependence of the SGC-I model response to stationary stimuli and first- and second-order motion. (a) SGC-I rate versus stimulatedRGC rate. (b) SGC-I rate versus spontaneous RGC rate. (c) SGC-I rate versus Pmax. The SGC-I rate increases monotonically with increasing Pmax. (d) SGC-I rate versus t0. The SGC-I rate decreases with increasing values of t0 up to approximately t0 = 1,000 ms and for larger values remains largelyindependent of t0. (e) SGC-I rate versus stimulus speed for a moving bar and a second-order motion stimulus. (f) Number of SGC-I spikes per degree ofstimulus movement versus stimulus speed for a moving bar and a second-order motion stimulus. Data points were calculated from simulation results in eusing (spikes/deg) = (spike rate [1/s])/(stimulus speed [deg/s]). Inset: same as in f at expanded stimulus speed scale.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Does the ensemble of the model SGC-I neurons, with their ran-domly distributed dendritic endings, carry information about thedirection of motion? This is an important question, because rotundalneurons integrate outputs from multiple tectal neurons29 in a direc-tionally sensitive manner30–32. To address this question, we simulatedspike trains in response to moving stimuli for model SGC-I neuronswith spatially offset but overlapping receptive fields. We then applieda motion-sensing algorithm that was cross-correlation based33 toextract an estimate of the direction of motion from the ensemble ofsimulated spike trains. The output of this motion-sensing algorithmis a ‘net motion signal,’ the sign of which indicates the direction ofmotion. Because of the random distribution of dendritic endings andbecause of the statistical nature of spike generation, the net motionsignal varies from trial to trial, yielding a distribution of net motionsignal. A natural measure of the fidelity of the ensemble motion sensi-tivity is the SNR, which is defined as the mean of the distribution ofnet motion signals divided by the s.d. of the distribution. For theparameters chosen, the moving bar and the second-order motionyielded an SNR of 3.8 and 2.0, respectively. In contrast, for dynamicspatiotemporal stimuli that were not moving, such as the update ofuncorrelated random dots within a stationary rectangle or the wholefield, the SNR was 0.0 and 0.1, respectively. The analysis thus showsthat the ensemble of model SGC-I neurons (i) carries information todistinguish between moving and stationary dynamic spatiotemporalstimuli and (ii) carries information about the direction of motion forboth first- and second-order motion stimuli.

DISCUSSIONThis in vitro study revealed phasic signal transfer at the retinotectalsynapse and binary dendritic responses to synaptic inputs that inter-acted in a mutually exclusive manner in the postsynaptic tectal neu-ron. A model of the tectal circuitry predicts that the two observedcellular properties mediate sensitivity in the tectal SGC-I neuron to awide range of dynamic spatiotemporal stimuli, including movingstimuli, but not to static stationary stimuli. Further, the SGC-I modelrepresentation of the dynamic spatiotemporal second-order stimuluswas similar to the SGC-I representation of the dynamic spatiotempo-ral first-order stimulus. Hence, SGC-I distinction between static sta-tionary stimuli and dynamic spatiotemporal stimuli, such as motion,is independent of the details of the stimulus. The stimulus classifica-tion remained robust over a wide range of model parameters includ-ing the brightness of the stimulus.

This study indicates that phasic signal transfer at the retinotectalsynapse, presumably mediated by synaptic depression, may be a cru-cial biophysical element of tectal stimulus classification. In general,synaptic depression can endow networks with new and unexpecteddynamic properties34–37. It has recently been suggested that thalamo-cortical synaptic depression, rather than the cortical network, mayaccount for important response properties of cortical neurons38–40.Further, synaptic depression at the input synapses of the nucleus lam-inaris in chick provides an adaptive mechanism to compensate forintensity variations in the localization of sound41. The present studyassigns a similar importance to phasic signal transfer at the retinotec-tal synapse for the analysis of dynamic spatiotemporal stimuli. Thisindicates that phasic signal transfer at an incoming synapse may be aconserved feature in the dynamic regulation of neuronal sensitivityduring rapid changes in sensory input.

It has long been appreciated that it is not the light intensity itselfbut rather the pattern of local variation in intensity that is typicallythe exciting factor for neurons in visual pathways42. Often these localvariations in light intensity are mediated by motion; however, motion

is only a subset of the visual stimuli that lead to spatiotemporal varia-tions in intensity. Perhaps because most previous studies of avian tec-tal neurons and primate cortical area middle temporal neurons havefocused on the coding of motion itself, the neurons were typicallytested with static stationary stimuli to which they respond weakly ver-sus moving stimuli to which they respond strongly8,43. Based on theseresults, the neurons have been interpreted as motion-sensitive. Whentested with a broader range of stimuli, however, as was done in thissimulation study of model SGC-I neurons and in the more recent invivo studies on primate middle temporal neurons43–45, these neuronsalso respond to dynamic spatiotemporal stimuli that are not moving.Based on the examined SGC-I cellular properties, the SGC-I modelreproduced the in vivo observations that the tectal SGC-I neuron doesnot respond to static stationary stimuli but responds to moving stim-uli independently of the cues that define the moving object. In addi-tion, the model predicts strong SGC-I responses to stationarydynamic spatiotemporal stimuli that have not previously been testedin vivo. This model prediction suggests an extended interpretation ofSGC-I signal processing. Because of the phasic synaptic signal trans-fer, individual SGC-I neurons do not respond to a static stationarystimulus but do respond vigorously to a dynamic spatiotemporalstimulus that activates new dendritic endings sequentially. Becausethe SGC-I response is independent of the particular sequence of acti-vation of its dendritic endings, the individual SGC-I neuron respondsto most dynamic spatiotemporal stimuli, whether they are moving ornot. The interpretation of a single SGC-I response is thus ambiguousin this respect. This is a typical problem in breaking ambiguity bypopulation coding. The individual SGC-I neuron is sensitive to mostdynamic spatiotemporal stimuli. The spatiotemporal pattern ofspikes in the ensemble of SGC-I neurons, however, contains addi-tional spatiotemporal information about the stimulus, such asmotion parameters. In the avian brain, rotundal neurons receiveinputs from an ensemble of tectal SGC-I neurons29. Apparently, thetectal population activity is then decoded by postsynaptic rotundalneurons that are sensitive to the direction of motion and loom-ing30–32. In summary, this work indicates that a specific cellularmechanism may exist to distinguish between static stationary stimuliand dynamic spatiotemporal stimuli and further clarifies the func-tional interpretation of individual SGC-I neurons as form-cue invari-ant change-sensitive and of ensembles of SGC-I neurons as form-cueinvariant motion-sensitive. It is of note that this interpretation doesnot exclude SGC-I responses to more complex dynamic spatiotempo-ral stimuli such as relative motion11, responses that are presumablymediated by tectal network properties17,27.

The sensitivity of the SGC-I model to second-order motion19 was adirect consequence of its sensitivity to dynamic spatiotemporal stim-uli in general. Previous theoretical studies of second-order motionsensitivity have pointed out the necessity of a nonlinearity, such as arectification, followed by a summation of the preprocessed signal19,46.Both stages of processing second-order signals have long beenascribed to higher-level processing in the visual cortex21. Evidence forlow-level processing of second-order motion has been sparse47–49.Here we report that, in the avian tectal circuit, the required rectifica-tion stage is implemented by the all-or-none binary response of theSGC-I dendritic endings, which mediate a receptive field with a finestructure of rectifying subunits. The array of rectified responses issubsequently integrated by the SGC-I soma, which in turn projects tothe nucleus rotundus13,15,29,50. The computation of the direction ofsecond-order motion could be completed by rotundal cells, which areknown to process different aspects of visual information includingtranslational motion30–32.

A R T I C L E S


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


METHODSExperiment. Thirty-nine White Leghorn chick hatchlings (Gallus gallus; <5 dold) were used in this study. All procedures used in this study were approvedby the local authorities and conform to the guidelines of the NationalInstitutes of Health on the Care and Use of Laboratory Animals. Tectal slicepreparation, SGC-I soma whole-cell recording, electrostimulation with bipo-lar tungsten electrodes and SGC-I labeling were carried out as described previ-ously9. We obtained stable whole-cell patch recordings from a total of 67neurons in the chick SGC-I. The series resistance of the recordings was 10 ±2 MΩ (mean ± s.d.) and was routinely compensated. We analyzed only neu-rons that were sufficiently labeled to allow the unequivocal classification as theSGC-I cell type (Fig. 1b)13. SGC-I neurons usually have their somata in theouter aspects of the SGC-I and respond with characteristic rhythmic burstingto somatic current injection9. The cells had a stable resting potential of –66 ±4 mV (mean ± s.d.) and an input resistance at rest of 72 ± 17 MΩ (mean ± s.d.,n = 67). Pulse train stimuli (Fig. 2a–c) were repeated ten times for each stimu-lation interval in a pseudorandom sequence of stimulation intervals with awaiting time of 5 min between pulse train stimuli. The response probabilityfor a sharp-onset response for each stimulus pulse within the pulse train wasderived from the number of responses divided by the number of trials. Paired-pulse stimuli (Fig. 2d–f) were repeated five times for each stimulation intervalin a pseudorandom sequence of intervals with a waiting time of 5 min.

Model. In the bird, the visual field of each eye is approximately 100° and projectsonto a tectal circumference of approximately 10 mm. Assuming for simplicity ahomogeneous spatial distribution of RGCs in the retina and a homogeneousspatial distribution of RGC axon terminals on the tectal surface, we estimatedthat a typical SGC-I dendritic field of 3 mm in diameter corresponds to a visualfield of 30 °. In our model, we therefore considered a region of visual space thatis 30 × 30 °, which corresponds to a region of tectal surface that is 3,000 × 3,000µm. Because of this correspondence of visual space and tectal surface, it wasmore convenient to express the stimulus space in terms of tectal surface unitsrather than visual space units. In the model, the stimulus space is represented byan array of 300 × 300 squares of 10 × 10 µm, corresponding to 0.1 × 0.1° in visualspace. A binary stimulus was represented within this space and moved with aspeed in multiples of 10 µm per simulation time step (10 ms). For all simula-tions (except those for Fig. 6e,f) the stimulus speed was 10 deg/s.

We assumed that the described stimulus space is sampled by an array of100 × 100 RGC axons (Fig. 3a). Thus one RGC samples exclusively 3 × 3 stimulusspace units. We assumed no spontaneous activity for the RGC, but the RGC pro-duces a Poisson spike train while a stimulus is present within the 3 × 3 space, witha mean firing rate of 80 Hz, a rate that is typically observed experimentally23.

Each SGC-I soma is connected to approximately 500 dendritic endings.These dendritic endings are randomly distributed within the 100 × 100 array(Fig. 3b). Each dendritic ending makes a synaptic contact with one RGC axon.Thus the 100 × 100 RGC axons are sampled sparsely; because this SGC-Imodel considered only one SGC-I cell, most RGC axons (approximately 9,500axons) do not terminate on a dendritic ending.

The signal transfer from one RGC axon to one dendritic ending is phasic ina time-dependent probabilistic manner. An RGC spike causes one dendriticspike with probability P(∆t) = Pmax (1 – e–∆t/t0), with ∆t the time interval sincethe previous RGC spike and Pmax = 0.87 and t0 = 2,025 ms the parametersdetermined from pulse train stimulation experiments (Fig. 2c).

For simplicity, we assumed that dendritic spikes from all dendritic endingsarrive at the SGC-I soma without delay. Dendritic spikes were counted at thesoma. If no dendritic spike was generated in any of the dendritic endings, noSGC-I spike was generated in that simulation time step. If one or more den-dritic spikes were generated, one SGC-I spike was generated if the previousSGC-I spike occurred more than 30 ms before (outside the interaction time).No SGC-I spike was generated if the previous SGC-I spike occurred within theinteraction time of 30 ms. The SGC-I rate was calculated as the number ofspikes during 3-s simulations with five repetitions.

The luminance-defined bar stimulus of 5 × 1° in visual space corresponds to500 × 100 µm on the tectal surface and 50 × 10 space units in the model (Fig. 3c). Because each model RGC axon has a receptive field of 3 × 3 nonover-lapping space units, the bar stimulus excited 56 RGC axons. A model SGC-Ineuron has approximately 500 dendritic endings that are randomly distrib-

uted within its receptive field (Fig. 3b) and a total of 100 × 100 RGC axons ter-minating within this field. Therefore, on average, 3 ≅ (56 × 500)/(100 × 100) ofthe 56 excited RGC axons terminated on an equal number of dendritic endings(one RGC axon per dendritic ending). When the bar moved with a speed of10 deg/s in the visual space, this corresponded to a speed of one space unit persimulation time step in the model.

The stationary whole-field random-dot stimulus is 300 × 300 space units(corresponding to 30 × 30° in visual space and 3,000 × 3,000 µm on the tectalsurface). Because each model RGC axon has a receptive field of 3 × 3 nonover-lapping space units, and the dot pattern is of low density, the number ofexcited RGC axons was approximately equal to the number of dots. This is onaverage the number of space units (300 × 300) times the dot density (0.002),which yields 180 dots. Using the same values as above, on average 9 = (180 ×500)/(100 × 100) of the 180 excited RGC axons terminated on an equal num-ber of dendritic endings.

For second-order motion, we chose a rectangle of 300 × 10 space units (cor-responding to 30 × 1° in visual space and 3,000 × 100 µm on the tectal surface).Following the estimate in the previous paragraph, we expected on average 6 dots (= 300 × 10 × 0.002) within the moving rectangle and therefore an equalnumber of excited RGC axons. A model SGC-I neuron has on average 17 =(300 × 10 × 500)/(300 × 300) dendritic endings within the rectangle. Becausethe model has a total of 100 × 3.3 = 330 RGC axons within the rectangle, onaverage only 0.3 = (6 × 17)/330 of the 6 excited RGC axons terminated on anequal number of dendritic endings (one RGC axon per dendritic ending). Therectangle moved with a speed of one space unit per simulation time step (cor-responding to 10 deg/s in the visual space). On each temporal frame the ran-dom-dot pattern within the moving rectangle was replaced with a differentuncorrelated random-dot pattern of equal density and equal mean luminance.

For the tectal ensemble decoding analysis, we considered pairs of modelSGC-I neurons, labeled A and B, with each neuron having the same propertiesas described above. The receptive fields of the two neurons were offset, how-ever, by 9° in the visual space (B to the right of A), so the dendritic fields withthe randomly distributed dendritic endings were offset by 900 µm at the tectalsurface. For each simulation trial we generated spike trains, sA(t) and sB(t), forboth model SGC-I neurons in response to a dynamic spatiotemporal visualstimulus. The stimuli considered were a moving bar (Fig. 3c), second-ordermotion (Fig. 3d) and a dynamic spatiotemporal stimulus that was not moving.The latter stimulus consisted of a pattern of random dots, in which the pixelvalues within a stationary rectangle (Fig. 3d) or the whole field were replacedwith uncorrelated random dots in successive frames. The moving stimuli weremoving from left to right. The spike trains, sA(t) and sB(t), were analyzedaccording to an algorithm that was cross-correlation based33. In brief, thespike trains were convolved with an exponential function, f(t), to create low-pass-filtered responses, rA(t) = sA(t) * f(t) and rB(t) = sB(t) * f(t), where f(t) =e–t/σ for t ≥ 0, f(t) = 0 for t < 0 and σ = 10 ms. A signal, R, indicative of right-ward motion was obtained by delaying the filtered response of cell A by anamount δ, multiplying pointwise by the filtered response of cell B and sum-ming the result:

where the summation is over all the time points in the trial, and rA(t − δ) is cir-cularly shifted to match the duration of rB(t). The delay value, δ, is the spatialreceptive field offset divided by the stimulus speed. A signal, L, indicative ofleftward motion was obtained correspondingly by:

For one pair of cells and one trial, the net motion signal is N = R − L, the sign ofwhich indicates the direction of motion. Because the problem under investiga-tion was exactly symmetric in space, we considered stimuli that were movingfrom left to right, thus typically yielding positive values for N. For motion inthe opposite direction, the N values would typically be negative. For each trial,we chose a new set of randomly distributed dendritic endings. The repetitionof trials (n = 30) yielded a distribution of net motion signals, N. It can be

L = Σ rB (t – δ) rA (t).t

R = Σ rA (t – δ) rB (t),t

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


shown that the pairwise computation of the net motion signal described aboveis equivalent to an approach that combines responses of an ensemble of cellssimultaneously33.

ACKNOWLEDGMENTSThe authors thank H.J. Karten and D. Kleinfeld for support during the collectionof preliminary data, A. Mahani for comments and W.B. Kristan, H. Wagner,G. DeAngelis, M. Ariel, P. Lukasiewicz, J. Sanes and A. Carlsson for critical readingof the manuscript. The work was supported by grants from DeutscheForschungsgemeinschaft to H.L. and Whitehall Foundation and McDonnellCenter for Higher Brain Function to R.W.


Received 3 November 2003; accepted 9 January 2004Published online at http://www.nature.com/natureneuroscience/

1. Albright, T.D. Cortical processing of visual motion. in: Visual Motion and its Role inthe Stabilization of Gaze (eds. Miles, F.A. & Wallman, J.) 177–201 (Elsevier,Amsterdam, 1993).

2. Croner, L.J. & Albright, T.D. Seeing the big picture: integration of image cues in theprimate visual system. Neuron 24, 777–789 (1999).

3. Albright, T.D. Form-cue invariant motion processing in primate visual cortex.Science 255, 1141–1143 (1992).

4. Olavarria, J.F., DeYoe, E.A., Knierim, J.J., Fox, J.M. & VanEssen, D.C. Neuralresponses to visual texture patterns in middle temporal area of the macaque mon-key. J. Neurophysiol. 68, 164–181 (1992).

5. Geesaman, B.J. & Andersen, R.A. The analysis of complex motion patterns byform/cue invariant MSTd neurons. J. Neurosci. 16, 4716–4732 (1996).

6. Jassik-Gerschenfeld, D. & Guichard, J. Visual receptive fields of single cells in thepigeon’s optic tectum. Brain Res. 40, 303–317 (1972).

7. Frost, B.J., Cavanagh, P. & Morgan, B. Deep tectal cells in pigeons respond to kine-matograms. J. Comp. Physiol. A 162, 639–647 (1988).

8. Frost, B.J. Subcortical analysis of visual motion: Relative motion, figure-ground dis-crimination and self-induced optic flow. in: Visual Motion and its Role in theStabilization of Gaze (eds. Miles, F.A. & Wallman, J.) 159–175 (Elsevier,Amsterdam, 1993).

9. Luksch, H., Karten, H.J., Kleinfeld, D. & Wessel, R. Chattering and differential sig-nal processing in identified motion sensitive neurons of parallel visual pathways inchick tectum. J. Neurosci. 21, 6440–6446 (2001).

10. Jassik-Gerschenfeld, D., Minois, F. & Conde-Courtine, F. Receptive field propertiesof directionally selective units in the pigeon’s optic tectum. Brain Res. 24,407–421 (1970).

11. Frost, B.J. & Nakayama, K. Single visual neurons code opposing motion independ-ent of direction. Science 220, 744–745 (1983).

12. Frost, B.J. Moving background patterns alter directionally specific responses ofpigeon tectal neurons. Brain Res. 151, 599–603 (1978).

13. Luksch, H., Cox, K. & Karten, H.J. Bottlebrush dendritic endings and large dendriticfields: motion-detecting neurons in the tectofugal pathway. J. Comp. Neurol. 396,399–414 (1998).

14. Tömböl, T. & Németh, A. Direct connections between dendritic terminals of tectalganglion cells and glutamate-positive terminals of presumed optic fibres in layers4–5 of the optic tectum of Gallus domesticus. Neurobiology (Bp) 7, 45–67(1999).

15. Karten, H.J., Cox, K. & Mpodozis, J. Two distinct populations of tectal neurons haveunique connections within the retinotectorotundal pathway of the pigeon (Columbalivia). J. Comp. Neurol. 387, 449–465 (1997).

16. Hunt, S.P. & Webster, K.E. The projection of the retina upon the optic tectum of thepigeon. J. Comp. Neurol. 162, 433–445 (1975).

17. Hardy, O., Leresche, N. & Jassik-Gerschenfeld, D. Morphology and laminar distribu-tion of electrophysiologically identified cells in the pigeon’s optic tectum: an intra-cellular study. J. Comp. Neurol. 233, 390–404 (1985).

18. Luksch, H. & Golz, S. Anatomy and physiology of horizontal cells in the optic tectumof the chick. J. Chem. Neuroanatomy 25, 185–194 (2003).

19. Chubb, C. & Sperling, G. Drift-balanced random stimuli: a general basis for studyingnon-Fourier motion perception. J. Opt. Soc. Am. A 5, 1986–2007 (1988).

20. Lu, Z.L. & Sperling, G. Three-systems theory of human visual motion perception:review and update. J. Opt. Soc. Am. A 18, 2331–2370 (2001).

21. Baker, C.L. Jr. & Mareschal, I. Processing of second-order stimuli in the visual cor-tex. Prog. Brain Res. 134, 171–191 (2001).

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

23. Meister, M. & Berry, M.J. II. The neural code of the retina. Neuron 22, 435–450(1999).

24. Tabata, T. & Kano, M. Heterogeneous intrinsic firing properties of vertebrate retinalganglion cells. J. Neurophysiol. 87, 30–41 (2002).

25. Nalbach, H.O., Wolf-Oberhollenzer, F. & Remy, M. Exploring the image. in Vision,Brain, and Behavior in Birds (eds. Zeigler H.P. & Bischof, H.J.) 25–46 (MIT Press,Cambridge, Massachusetts, 1993).

26. Troy, J.B. & Lee, B.B. Steady discharges of macaque retinal ganglion cells. Vis.Neurosci. 11, 111–118 (1994).

27. Luksch, H. Cytoarchitecture of the avian optic tectum: neuronal substrate for cellu-lar computation. Rev. Neurosci. 14, 85–106 (2003).

28. Frost, B.J., Scilley, P.L. & Wong, S.C.P. Moving background patterns reveal double-opponency of directionally specific pigeon tectal neurons. Exp. Brain Res. 43,173–185 (1981).

29. Marin, G. et al. Spatial organization of the pigeon tectorotundal pathway: An inter-digitating topographic arrangement. J. Comp. Neurol. 458, 361–380 (2003).

30. Revzin, A.M. Functional localization in the nucleus rotundus. in: NeuralMechanisms of Behavior in the Pigeon (eds. Granda, A.M. & Maxwell, J.H.)165–175 (Plenum, New York, 1981).

31. Wang, Y. & Frost, B.J. Time to collision is signalled by neurons in the nucleus rotun-dus of pigeons. Nature 356, 236–238 (1992).

32. Sun, H. & Frost, B.J. Computation of different optical variables of looming objects inpigeon nucleus rotundus neurons. Nat. Neurosci. 1, 296–302 (1998).

33. Chichilnisky, E.J. & Kalmar, R.S. Temporal resolution of ensemble visual motion sig-nals in primate retina. J. Neurosci. 23, 6681–6689 (2003).

34. Abbott, L.F., Varela, J.A., Sen, K. & Nelson, S.B. Synaptic depression and corticalgain control. Science 275, 220–223 (1997).

35. Tsodyks, M.V. & Markram, H. The neural code between neocortical pyramidal neu-rons depends on neurotransmitter release probability. Proc. Natl. Acad. Sci. USA94, 719–723 (1997).

36. Chance, F.S., Nelson, S.B. & Abbott, L.F. Synaptic depression and the temporalresponse characteristics of V1 cells. J. Neurosci. 18, 4785–4799 (1998).

37. Goldman, M.S., Maldonado, P. & Abbott, L.F. Redundancy reduction and sustainedfiring with stochastic depressing synapses. J. Neurosci. 22, 584–591 (2002).

38. Freeman, T.C.B., Durand, S., Kiper, D.C. & Carandini, M. Suppression without inhi-bition in visual cortex. Neuron 35, 759–771 (2002).

39. Carandini, M., Heeger, D.J. & Senn, W.A Synaptic explanation of suppression invisual cortex. J. Neurosci. 22, 10053–10065 (2002).

40. Chung, S., Li, X. & Nelson, S.B. Short-term depression at thalamocortical synapsescontributes to rapid adaptation of cortical sensory responses in vivo. Neuron 34,437–446 (2002).

41. Cook, D.L., Schwindt, P.C., Grande, L.A. & Spain, W.J. Synaptic depression in thelocalization of sound. Nature 421, 66–70 (2003).

42. Lettvin, J.Y., Maturana, H.R., McCulloch, W.S. & Pitts, W.H. What the frog’s eye tellsthe frog’s brain. Proc. IRE 47, 1940–1951 (1959).

43. Palanca, B.J.A. & DeAngelis, G.C. Macaque middle temporal neurons signal depthin the absence of motion. J. Neurosci. 23, 7647–7658 (2003).

44. Newsome, W.T., Britten, K.H. & Movshon, J.A. Neuronal correlates of a perceptualdecision. Nature 341, 52–54 (1989).

45. Croner, L.J. & Albright, T.D. Segmentation by color influences responses of motion-sensitive neurons in the cortical middle temporal visual area. J. Neurosci. 19,3935–3951 (1999).

46. Wilson, H.R. The role of second-order motion signals in coherence and transparency:higher-order processing in the visual system. Ciba Foundation Symposium 184,227–244 (1994).

47. Orger, M.B., Smear, M.C., Anstis, S.M. & Baier, H. Perception of Fourier and non-Fourier motion by larval zebrafish. Nat. Neurosci. 3, 1128–1133 (2000).

48. Roeser, T. & Baier, H. Visuomotor behaviors in larval zebrafish after GFP-guidedlaser ablation of the optic tectum. J. Neurosci. 23, 3726–3734 (2003).

49. Demb, J.B., Zaghloul, K. & Sterling, P. Cellular basis for the response to second-order motion cues in Y retinal ganglion cells. Neuron 32, 711–721 (2001).

50. Hellmann, B. & Güntürkün, O. Structural organization of parallel information pro-cessing within the tectofugal visual system of the pigeon. J. Comp. Neurol. 429,94–112 (2001).

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

Much evidence supports the hypothesis that the dopamine innervationof the nucleus accumbens (NAc) is a key neural substrate mediating theprimary reinforcing and psychomotor stimulant effects of drugs ofabuse. Intravenous cocaine self-administration is reduced by 6-hydrox-ydopamine (6-OHDA)-induced dopamine depletion from the NAc1–3.Infusions of dopamine receptor agonists and antagonists directly intothe NAc alter rates of intravenous drug self-administration, as if ratsare compensating for changes in the reinforcing effects of the drug4,5.Furthermore, extracellular dopamine in the NAc is consistentlyincreased in response to experimenter-delivered or self-administeredcocaine, amphetamine, nicotine, opiates and ethanol in rats and inprimates6–10. The glutamatergic innervation of the NAc has also beenimplicated in modulating drug self-administration behavior, thoughless clearly so11,12.

Given the evidence that various neural systems innervating theNAc contribute to drug self-administration, it is perhaps surprisingthat excitotoxic lesions of the NAc itself, which destroy its mediumspiny neuron output and other intrinsic neurons, have variable,and often no, effects on drug self-administration13–16. This leavessome measure of doubt about the NAc having an essential role indrug reinforcement.

In resolving this issue, it is important to bear in mind that drug self-administration probably depends on a complex interaction of severaldistinct behavioral processes. It involves conditioning as well as theunconditioned effects of the drug itself—these factors may underliedifferent aspects of drug-seeking and drug-taking behavior17,18.Thus, data from humans and animals indicate that environmentalstimuli previously associated with self-administered drugs maypotently affect subjective (i.e., craving) as well as behavioral measuresof drug-seeking behavior and relapse19–25. Moreover, these

conditioned and unconditioned effects of drugs are neurally dissocia-ble. For example, lesions of the basolateral amygdala do not affectdrug self-administration, but they do prevent the acquisition ofcocaine-seeking behavior under the control of drug-associated stim-uli26 and reinstatement of drug-seeking after extinction27.

Distinguishing among these conditioned and unconditionedprocesses in self-administration protocols is especially important inlight of evidence that the NAc itself is a heterogeneous structure withat least two distinct regions, shell and core28,29, that may contribute indifferent ways to drug self-administration30–33. There is strong evi-dence that the NAc core is involved in the control of goal-directedbehavior by associative processes, consistent with its central positionwithin limbic cortical-ventral striatal circuitry34. For example, selec-tive dopaminergic or excitotoxic lesions of the NAc core, but not theshell, disrupt learnt Pavlovian influences on appetitive behavior35–37.One of the important ways in which Pavlovian conditioned stimuliinfluence behavior is as conditioned reinforcers that can, by them-selves, support instrumental behavior such as drug-seeking23.Excitotoxic lesions of the NAc core disrupt the capacity of food-associated conditioned reinforcers to control behavior, whereaslesions of the NAc shell specifically impair the mechanism by whichdrugs such as cocaine and d-amphetamine enhance the effects of suchstimuli38. Moreover, intra-NAc shell infusions of amphetamineenhanced the motivational effects of Pavlovian conditioned stimulion instrumental behavior (Pavlovian-to-instrumental transfer, aform of Pavlovian arousal39).

The aim of the present study was to investigate the extent to whichthe NAc core and shell contribute to conditioned, as well as uncondi-tioned, influences that govern drug-seeking and drug-taking. To thisend, we not only studied drug self-administration under standard

Department of Experimental Psychology, University of Cambridge, Downing Street, Cambridge CB1 1BB, UK. Correspondence should be addressed to B.J.E.([email protected]).


Differential control over cocaine-seeking behavior bynucleus accumbens core and shellRutsuko Ito, Trevor W Robbins & Barry J Everitt

Nucleus accumbens (NAc) dopamine is widely implicated in mediating the reinforcing effects of drugs of abuse. However, theprecise function of the NAc itself in drug self-administration has been difficult to establish. Here we show a neural double-dissociation of the behavioral processes that underlie cocaine self-administration in rats. Whereas selective excitotoxic lesions of the NAc core had only a minor effect on the acquisition of responding for cocaine under a standard schedule of continuousreinforcement, these lesions profoundly impaired the acquisition of drug-seeking behavior that was maintained by drug-associated conditioned reinforcers and assessed using a second-order schedule of cocaine reinforcement. In contrast, selectiveexcitotoxic lesions of the NAc shell did not impair drug self-administration or the acquisition of cocaine-seeking, but they didattenuate the psychostimulant effects of cocaine. These results further our understanding of how the NAc controls drug-seekingand drug-taking behavior.


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

conditions of continuous reinforcement, where every instrumentalresponse is followed by a contingent cocaine infusion, but we alsoused a procedure in which drug-seeking becomes increasingly underthe control of drug-associated conditioned reinforcers (in a so-calledsecond-order schedule of reinforcement23). We found that selectiveexcitotoxic lesions of the NAc core profoundly disrupted the acquisi-tion of cocaine-seeking behavior when this behavior was substantiallyunder the control of drug-associated conditioned reinforcers.Although lesions of the NAc shell did not impair drug self-adminis-tration or the acquisition of cocaine-seeking, they did attenuate theresponse rate–enhancing (or psychostimulant) effects of cocaine.

These findings imply a neural dissociation between the mechanismsunderlying the associative control of drug-seeking and those underly-ing the psychomotor stimulant effects of cocaine. The present resultsenhance our understanding of how the NAc controls drug-seekingand drug-taking behavior.

RESULTSCocaine self-administration under continuous reinforcementIntravenously catheterized rats with selective excitotoxic lesions ofthe NAc core or shell regions, and their sham-operated controls (Figs. 1 and 2), were initially trained daily for 2 h to acquire cocaineself-administration under a continuous reinforcement schedule.Response on one of two identical levers (active lever) led to a contin-gent infusion of cocaine (0.25 mg per infusion). They were deemed tohave acquired the task when stable responding with less than 10%variance across three consecutive days was achieved.

All three treatment groups (core, shell and sham) reached criterionlevels of responding within 10 d of beginning cocaine self-administrationunder a continuous reinforcement schedule (Fig. 3a). Three-wayANOVA of square-root transformed lever presses during acquisitionrevealed a significant group × lever × day interaction (F18,315 = 1.62,P < 0.05) and a significant main effect of lesion group (F2,65 = 6.24,P < 0.005). Separate analyses of the pattern of responding on the activeand inactive levers using two-way ANOVA followed by Newman-Keulsmultiple comparisons revealed that the core-lesioned rats responded ata significantly higher rate on the active lever on days 1 and 2 only (P <0.01) compared to control rats (group × day interaction, F9,261 = 2.61,P < 0.01; Fig. 3b). There was no difference between sham and shellgroups in responding on the active lever (F1,20 = 0.09, P = 0.76).Inactive lever responses in the core-lesioned group were slightly, butsignificantly, increased overall compared with those in the sham group(group effect, F1,29 = 7.01, P < 0.01; Fig. 3a), whereas there was no sig-nificant group effect on inactive lever responding between shell andsham groups (F1,21 = 1.91, P < 0.18). Nevertheless, significantly moreresponses were made on the active lever, as compared to the inactivelever, in all groups (lever, F1,35 = 276.03, P < 0.0001).


Figure 1 Excitotoxic lesions of the NAc core and shell. Schematicrepresentation of quinolinic acid lesions of the NAc core (top) and ibotenicacid lesions of the NAc shell (bottom). Areas shaded in gray and blackrepresent the largest and smallest extent of neuronal damage in a singleanimal, respectively. Coronal sections are +2.2 mm anterior through +0.48 mm posterior to bregma49.

Figure 2 Representative photomicrographs showing Cresyl Violet-stainedand NeuN-stained coronal sections through the NAc in rats with NAc shellor core lesions and sham-operated control subjects. (a) Nissl-stainedsection through the NAc of a control subject, showing the region of theshell and core and other markers at this antero-posterior level (island of Calleja, lateral ventricle and anterior commissure). (b) Nissl-stainedsection of a NAc shell lesion, showing the marked loss of staining in theshell region and preservation of neurons in the core region, as well as theinfusion cannula tract. (c) Nissl-stained section of a NAc core lesion;marked gliosis can be seen around the medial surface of the anteriorcommissure. Note the apparent medial shift of the anterior commissure,which is the result of the loss of neurons in the core region. The integrity of the shell region is preserved. (d,e) NeuN-stained sections through theshell region of sham (d) and shell-lesioned subjects. Note the completedisappearance of NeuN-immunoreactivity in the shell region (compare ewith d), indicating the loss of neurons in this region of the NAc followinginfusion of ibotenic acid. Abbreviations: AC, anterior commissure; Core,NAc core; Shell, NAc shell; LV, lateral ventricle; IC, Isle of Calleja.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

Typical response records on the last day of acquisition (day 10) insham, core and shell-lesioned rats (Fig. 3c) showed that response pat-terns in both sham and shell groups were characterized by an initial,rapidly occurring burst of cocaine self-administration (‘loading’phase) followed by a period of stable ‘titrating pattern,’ evenly spacedresponses on the active lever. Responding in core-lesioned rats, how-ever, had the following characteristics: (i) higher response rates, suchthat core-lesioned rats gained the maximum number of cocaine infu-sions (50) significantly earlier in the session than the shell and shamlesioned groups across the 10 d of acquisition; (ii) for the post-reinforcement pause (PRP; the period after each cocaine infusion andbefore the first response made subsequently, which provides a measureof the impact of cocaine reinforcement), there were no significant dif-ferences, indicating unimpaired control over instrumental respondingby cocaine (sham, 2.71 (mean PRP (min) on day 10) ± 0.19 (s.e.m.);core, 2.73 ± 0.26; shell, 3.07 ± 0.25; F2,37 = 0.60, nonsignificant, n.s.)

In summary, core-lesioned rats showed only minor changes in theacquisition of cocaine self-administration. They had somewhathigher rates of responding on the active lever, compared with thesham and shell groups, on the first two days of acquisition only. Inaddition, responding on the inactive lever by core-lesioned rats wasinconsistently elevated, thereby reducing discrimination between theactive and inactive levers on some days of testing. By contrast, therewas no difference between the sham and shell groups in the pattern ofacquisition of cocaine self-administration.

Cocaine dose-response functionIn all three groups, variations in the dose of cocaine produced orderly,monotonic changes in the rates of responding on the active lever, withevidently more persistent responding in extinction (i.e., under saline)in the core-lesioned rats (Fig. 4). ANOVA with repeated measures onresponse rates revealed significant main effects of dose (F4,120 = 41.24,P < 0.0001), lever (F1,30 = 171.21, P < 0.0001) and lesion (F2,30 = 5.59,P < 0.01), with significant group × lever × dose (F8,120 = 2.14,P < 0.04) and dose × lesion (F8,120 = 5.36, P < 0.001) interactions.Separate ANOVA on response rates on the active lever at each doserevealed significant main effects of lesion group in the saline (F2,17 = 12.31, P < 0.001), 0.25 mg/infusion (F2,17 = 7.40, P < 0.005)and 0.5 mg/infusion (F2,17 = 9.05, P < 0.003) conditions. Newman-Keuls pairwise comparisons revealed that core-lesioned ratsresponded at significantly higher rates than sham and shell group ani-mals on the active lever in extinction (i.e., on saline substitution) andat marginally, yet significantly, higher rates at doses of 0.25 and 0.5mg/infusion. One-way ANOVAs on response rates on the inactive leverrevealed no significant lesion group effect in any of the conditions.

Cocaine-seeking under a second-order schedule of reinforcementOnce cocaine self-administration under continuous reinforcementhad been acquired, training began under the second-order schedules,

where the response requirement for cocaine and the conditionedreinforcer was progressively increased (see Methods). It was deter-mined that each rat would have to satisfy a certain criterion in orderto move on to the next stage of training. This criterion was to obtainat least ten cocaine infusions per session at each schedule stage forthree consecutive days. Significantly more core-lesioned animalsfailed to reach criterion at each stage beyond FR10(FR4:S), the coregroup size diminishing from 15 at FR10(FR1:S) to 8 at FR10(FR10:S)(Fig. 5). Responding on the active lever increased in all groups as theschedule requirement was progressively increased across days (day;F1,35 = 713.41, P < 0.0001). However, overall ANOVAs showed thatthe groups responded at significantly different rates (group, F2,113 =4.00, P < 0.0001; group × day F2,35 = 10.80, P < 0.0002). Further analy-ses by lever and training stage separately revealed significantly lowerresponding on the active lever in the core-lesioned group comparedto controls under FR10(FR2:S) (F1,124 = 6.04, P < 0.02), FR10(FR4:S)(F1.116=7.28, P < 0.02), FR10(FR7:S) (F1,104=11.69, P < 0.002) andFR10(FR10:S) (F1,96 = 17.06, P < 0.0004) schedules. Responding onthe active lever in the shell-lesioned group was significantly lowerthan the controls under the FR10(FR10:S) schedule (F1,88 = 10.40,P < 0.004), but not at other stages. By contrast, responding on theinactive lever in the core group was significantly more elevated thanthat in the sham group (F1,1056 = 7.48, P < 0.01), at all stages of the


Figure 3 Acquisition of intravenous cocaine self-administration under a continuous reinforcement schedule. (a) Mean (±s.e.m.) number ofresponses on the active and inactive lever during each 2 hr session. *P <0.05, **P < 0.01, compared to sham. (b) The rate of responding (per min)on the active (drug-paired) lever in each session, after sham or excitotoxiclesions of the NAc core and shell regions. **P < 0.01 compared to sham.(c) Representative response records of individual rats from NAc sham, core and shell lesion groups on the last day of acquisition of cocaine self-administration under a continuous reinforcement schedule. Each bar abovethe horizontal line represents an individual response on the active lever,whereas each bar below the line represents a response on the inactive lever.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

second-order schedules of reinforcement. Responding on the inactivelever in the shell group, however, was not significantly different fromthat of the sham group (F1,786 = 1.24, n.s.).

Pattern of responding under FR10(FR10:S)The response patterns of the core-lesioned rats differed from those ofthe sham- and shell-lesioned rats in three main ways (Fig. 6a,b):(i) whereas sham- and shell-lesioned rats showed rapid burst-likepatterns of responding on the active lever, core-lesioned rats exhib-ited temporally dispersed patterns of responding; (ii) core-lesionedrats took much longer to complete the FR10(FR10:S) responserequirement before and even after the first cocaine infusion than didsham- and shell-lesioned rats; (iii) the sham- and shell-lesioned ratsshowed a regular ‘titrating’ pattern of responding, wherein a period ofinactivity followed each cocaine infusion (post-reinforcement pause,PRP); no such regular pauses were observed in core-lesioned rats.

Post-reinforcement pauseThe PRP is the period following the reinforcer delivery and before thestart of the next ratio of responding, often taken to represent therewarding impact of the drug (Fig. 7a). ANOVA revealed significantmain effects of group (F2,1836 = 569.39, P < 0.0001) and training stage(F4,1824 = 7.83, P < 0.0001) as well as a significant group × traininginteraction (F8,1824 = 41.64, P < 0.0001) in the mean duration of thePRP within a 2-h self-administration session at each of the differentsecond-order schedule requirements in sham, core- and shell-lesioned rats. In both sham- and shell-lesioned groups, the durationof the PRP increased as the response requirement rose at each stage ofthe second-order schedule. Post-hoc analysis showed the PRP dura-tion in the core-lesioned rats to be significantly shorter than in thesham controls and shell groups at all second-order schedules ofcocaine reinforcement tested (P < 0.01).

Post-conditioned reinforcement pauseThe post-conditioned reinforcement pause (PCRP) is the periodbetween the brief presentation of the conditioned reinforcer and thefirst response made subsequently. It provides a measure of the impactof the conditioned stimulus acting as a conditioned reinforcer

(Fig. 7b). Two-way ANOVA with lesion as the between-group factorand cocaine state (pre-cocaine versus post-cocaine) as the repeatedmeasure showed significant main effects of group (F2,280 = 3.26, P <0.04) and state (F1,277 = 15.36, P < 0.0001) but no interaction betweenthese factors (F2,277 = 2.56, P < 0.08, n.s.) in the mean PCRP durationof the first 200 responses under the FR10(FR10:S) schedule of cocainereinforcement in the three groups. Post hoc analysis revealed thatPCRP pauses of core-lesioned rats in the drug-free state were signifi-cantly shorter compared with the shell and sham groups (P < 0.01). Inaddition, the PCRP duration during responding for the first infusionwas significantly higher compared to that during responding underthe influence of cocaine for the second infusion, in shell and shamgroups, but not in the core-lesioned group. Thus, core-lesioned ratsshowed significantly shorter PCRP pauses before a cocaine infusion,compared to the sham- and shell-lesioned rats, but the PCRP dura-tion remained unchanged after a cocaine infusion.

Rate of responding on the active leverThe effects of NAc lesions on the mean rate of responding before andafter the first cocaine infusion during three sessions under theFR10(FR10:S) schedule were also investigated (Fig. 7c). Two-wayANOVA showed significant main effects of group (F2,28 = 10.3,P < 0.0001) and cocaine state (F1,28 = 103.2, P < 0.0001), and signifi-cant interaction between these factors (F2,28 = 7.6, P < 0.002).Separate between-subject one-way ANOVAs revealed that the rate ofresponding during the pre-cocaine period was not significantly dif-ferent between the lesion groups (F2,29 = 2.42, P = 0.11), whereas itwas for the post-cocaine state (F2,29 = 12.08, P < 0.0001). Separatewithin-subject one-way ANOVAs showed the difference in the rate ofresponding pre- and post-cocaine to be significant only for the coreand sham groups (core, F1,5 = 14.85, P < 0.006; sham, F1,15 = 303,P < 0.0001; shell, F1,5 = 4.36, P = 0.10, n.s.). Therefore, significantcocaine-induced increases in response rate were evident in sham andcore-lesioned, but not in shell-lesioned rats.

In summary, core-lesioned rats were profoundly impaired in theacquisition of drug cue–controlled cocaine-seeking behavior. Onlyhalf of the core-lesioned group completed the training, and the over-all number of responses made on the active lever as well as the rates ofresponding were significantly reduced at all stages of training underthe second-order schedule in these rats, as compared to sham-lesioned rats. Moreover, the pattern of responding of core-lesionedrats showed a marked absence of PRPs and significantly shorter PCRPpauses compared to the sham rats. In contrast, lesions of the NAc shelldid not significantly affect the acquisition of cocaine-seeking behav-ior. However, shell-lesioned rats showed significantly smallerincreases in the rate of responding in the periods after cocaine intake,as compared to sham- and core-lesioned rats.

Cocaine-induced locomotor activityWe also recorded locomotor activity during cocaine self-administra-tion sessions under the FR1 schedule on days 1–3 and 10 (Fig. 8).Two-way ANOVA revealed a significant main effect of group (F2,17 =4.75, P < 0.02) but no significant interaction between the day ofacquisition and group. A subsequent post-hoc repeated measuresANOVA showed that cocaine-induced activity measures of the


Figure 4 Cocaine dose-response function. Between-sessions dose-responsefunction and effects of substituting saline for cocaine (responding inextinction) in sham-lesioned and NAc core- and shell-lesioned rats. Errorbars, ±s.e.m. *P < 0.05.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

core-lesioned rats (P < 0.023), but not shell-lesioned rats, were signif-icantly higher than those of the shams.

DISCUSSIONNAc core and cocaine-seekingThe present data provide evidence for an important role of the coreregion of the nucleus accumbens in cocaine-seeking behavior andhelp to resolve previous data showing only variable effects of lesionsof this structure on drug self-administration behavior. NAc core-lesioned rats were impaired in the conditioned control over cocaine-seeking, similar to the disruptive effects of NAc core lesions onPavlovian approach behavior35, Pavlovian-to-instrumental transfer36

and conditioned reinforcement38. The results are also consistent withour previous finding that lesions of the NAc core impair the capacityof a food-related conditioned reinforcer to acquire discriminativecontrol over instrumental behavior38. Taken together with the evi-dence that lesions of the basolateral amygdala (BLA) also impair theability of a food-associated conditioned reinforcer to support theacquisition of a new instrumental response40,41, the NAc core mayalso be a component of the neural circuitry involved in behavioralselection based on reward-related information derived from condi-tioned reinforcers, mediated via amygdaloid or other limbic corticalafferents to the NAc34. Indeed, selective lesions of the BLA produce asimilar pattern of relative sparing of continuously reinforced cocaineself-administration, but a failure to acquire drug cue–controlledcocaine-seeking behavior26.

The deficits observed in cocaine-seeking by core-lesioned rats wereunlikely to have been due to an inability to form effective stimulus-reward associations, as post-training NAc core lesions also impairperformance under a second-order schedule of cocaine reinforce-ment (data not shown), indicating that the deficits observed in thepresent study are not specific to the acquisition of this behavior.

It is important to emphasize that core-lesioned rats in the presentstudy were able to acquire instrumental responding for cocaine undera continuous reinforcement schedule, and indeed they respondedmore than controls under certain conditions. Similarly, NAc core-lesioned rats are also able to learn to respond for food and intra-venous heroin15,16,36, as well as to adapt their responding to changesin the dose of cocaine or heroin15 (and present results). These find-ings indicate intact learning of instrumental action-outcome contin-gencies and a largely intact efficacy of primary reinforcement,whether drug or food. The lack of a rightward shift in the dose-response curve is also strong evidence against the simple hypothesisthat acquisition impairments in cocaine-seeking are primarily due toan attenuation in drug reinforcement. Nevertheless, there was someevidence of elevated response rates on both the active and inactivelever under continuous reinforcement, which could be taken to indi-cate a mild attenuation of cocaine reinforcement. However, this gen-erally enhanced operant output could also be attributed to thelocomotor hyperactivity known to result from core lesions38,42. Thelesion-induced hyperactivity may also contribute to the persistence of

responding in extinction frequently seen in core-lesioned rats15,43,44.However, it is important to note that the hyperactive model of behav-ior produced by core lesions could not in itself account for the patternof effects seen in the drug-taking (where responding increased) anddrug-seeking (where responding decreased) paradigms, nor could itaccount for the differential responding on the active and inactivelevers in both situations.

NAc shell and cocaine-seekingCompared to NAc core lesions, shell lesions had no major effects onthe acquisition of cocaine-seeking behavior or self-administrationunder continuous reinforcement. Only at the most stringent stage ofthe second-order schedule was there an indication of significantlyreduced responding on the active lever. However, there were no differences in the rate of responding in the period prior to cocaineinfusion. Therefore, this effect can be entirely attributed to an atten-


Figure 5 Acquisition of cocaine self-administration under the second-orderschedule. (a) The proportion of sham, core and shell-lesioned rats attainingcriterion at each successive stage of acquisition of the second-orderschedule of cocaine reinforcement. Abbreviations: yS = FR10(FRy:S). Thecriterion was set as obtaining at least ten cocaine infusions within a 2-hself-administration session for three consecutive days. (b) Mean (±s.e.m.)responses on the active and inactive lever at each stage of the second-orderschedule of intravenous cocaine reinforcement in rats with lesions of theNAc core and shell.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

uation of the effect of cocaine itself to increase rates of respondingunder second-order schedules18,22. A similar, though quantitativelylarger, effect of similar NAc shell lesions on the potentiation ofresponding with conditioned reinforcement by amphetamine aswell as other psychomotor stimulant manifestations of the drug,such as locomotor hyperactivity, has previously been reported38.Furthermore, infusions of amphetamine into the NAc shell enhancethe ability of non-contingently presented Pavlovian cues to potenti-ate instrumental lever pressing39. Thus, these data, together withthose from the present study, show that the caudomedial NAc shellis not critical for the primary reinforcing effects of cocaine, but it isessential for the invigorating effect of stimulant drugs on condi-

tioned and unconditioned behavioral responses—that is, for theirpsychomotor stimulant actions.

Theoretical implicationsThe deficits observed in core-lesioned rats most likely reflect a loss ofthe impact of conditioned reinforcement on cocaine-seeking behav-ior. Highly relevant to such an interpretation is the observation thatlesions of the NAc core induce persistent impulsive choice behavior asevidenced by an inability to tolerate delays of food reinforcement45. Itshould be noted that a second-order schedule of reinforcement neces-sarily introduces a delay to primary reinforcement, but conditionedreinforcers normally help to mediate this delay by maintaining


Figure 6 Effects of NAc lesions on qualitative measures of cocaine self-administration under the second-order schedule. Representativeindividual records of (a) cumulative responses and (b) responses in sham,core and shell-lesioned rats under the second-order FR10(FR10:S)schedule of cocaine reinforcement. (a) The triangles represent the pointsat which the rat obtained a cocaine infusion (0.25 mg/infusion) and thecircles represent the presentation of a light CS that was contingent onevery tenth lever press. (b) The top panel shows individual records of thefirst 60 min of a 2-h session, with the arrows depicting the point ofcocaine infusion. This overall response record is then truncated intoresponse records for (i) the period up to the first cocaine infusion (middlepanel; a period of responding completely unaffected by any self-administered cocaine) and (ii) the period between the first and secondinfusion (bottom panel), for each lesion group.

Figure 7 Effects of NAc lesions on quantitative measures of cocaine self-administration under the second-order schedule. (a) Mean duration ofPRPs within a 2-h session, during each stage of second-order schedules ofcocaine reinforcement. **P < 0.01. Abbreviations: yS = FR10(FRy:S). (b) Mean duration of post-conditioned reinforcement pause (PCRP) duringthe first 100 responses in a drug-free state (pre-cocaine) and the next 100 responses following the first cocaine infusion (post-cocaine), under theFR10(FR10:S) schedule of reinforcement in sham, core and shell-lesionedrats. **P < 0.01, compared to pre-cocaine value, ++P < 0.01, compared to sham. (c) Mean rate of responding before (pre-cocaine) and after (post-cocaine) the first cocaine infusion of self-administration under theFR10(FR10:S) schedule of reinforcement in core, shell and sham-lesionedrats. **P < 0.01 compared to pre-cocaine rate of responding.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

instrumental responding34. As core-lesioned rats showed no deficitswhen responding for cocaine under a continuous reinforcementschedule, the impaired acquisition of cocaine-seeking behaviorreported here may depend on an inability of core-lesioned rats to tol-erate a delay in cocaine reinforcement. We infer that this reflects a lossof impact of cocaine-associated conditioned reinforcers. The similareffects of BLA lesions26 may implicate both of these structures in aneural system by which conditioned associations help to mediatedelays between responses and outcomes.

The integrity of the NAc core, unlike that of the caudomedial NAcshell, is not required for the response rate–increasing effects ofpsychostimulant drugs38. These data thus re-open the issue of therelationship between the stimulant and reinforcing effects ofpsychomotor stimulants such as cocaine, which have been suggestedto be isomorphic46. However, the doubly dissociable effects reportedhere indicate that the NAc core and shell mediate distinct processesassociated with cocaine self-administration behavior. The core seemsto mediate control by conditioned reinforcers, whereas the shellseems to mediate the potentiation of that control by cocaine, perhapsreflecting stimulant or motivational effects of the drug.

The additional importance of this double dissociation is in demon-strating a selective functional effect produced by the relatively discreteshell lesion (restricted to its caudomedial domain), that contrastswith the qualitatively distinct nature of the deficits produced by themore complete lesion of the nucleus accumbens core. It is significantthat this region of the NAc shell is most often implicated in mediatingboth the reinforcing and response-invigorating effects of psycho-motor stimulant drugs30–33,38.

How the NAc core and shell interact remains unclear, but recentanatomical evidence suggests that the striatum is organized in a hierar-chical fashion, with the shell and its limbic connections capable ofinfluencing the behavioral output of the core, via ‘spiralling’ connec-tions with the midbrain dopamine neurons47. The NAc shell can thusserve to amplify the expression in behavior of information flowingthrough the NAc core17. Such anatomical relationships may also under-lie different factors affecting intravenous drug self-administrationbehavior. These data emphasize the complex nature of cocaine rein-forcement mechanisms, while specifying a particular role for the NAccore subregion that has hitherto not been apparent.

METHODSAnimals. Male Lister Hooded rats (Charles River Ltd., UK) were housed in

pairs and then individually after surgery, in a room held at a temperature of

21 °C under a reversed 12-h light/dark cycle (lights off at 09:00). Food (labora-

tory chow, Purina) and water were available ad libitum but, after recovery from

surgery, food was restricted to 20 g of lab chow per day, sufficient to maintain

pre-operative body weight and growth. All experiments were carried out dur-

ing the dark phase, between 09:00 and 18:00 and in accordance with the

United Kingdom 1986 Animals (Scientific Procedures) Act Project License No.

80/1324.

Surgery. In all surgical procedures, animals were anesthetized with Avertin (10 g of 99% 2,2,2-tribromoethanol (Sigma-Aldrich) in 5 mg tertiary amylalcohol and 4.5 ml phosphate buffered saline (Dulbecco “A”, Unipath Ltd.) in40 ml absolute alcohol; 1 ml/100 g body weight, intraperitoneally (i.p.)).

Excitotoxic lesions. We used different excitotoxins to damage the core or shellsub-regions of the NAc (quinolinic acid for core; ibotenic acid for shell). Thisproduced selective lesions of these structures with little, if any, overlapbetween them38.

A 1-µl SGE syringe (SGE) was lowered stereotaxically into either the NAccore or shell, and the neurotoxin was infused bilaterally. For NAc core lesions,0.3 µl of 0.09 M quinolinic acid (Sigma-Aldrich) buffered to pH 7.3–7.4 in 0.1 M sterile phosphate buffer (sterile PB), was infused for 1 min in each hemi-sphere, using the following coordinates (in mm from bregma); AP: +1.2,L: ±1.8, DV: –7.1 from the skull surface (SS). For NAc shell lesions, three sepa-rate infusions of 0.06 M ibotenic acid (Sigma-Aldrich) buffered to pH 7.4using 0.1 M sterile PB were made at different points along the DV axis in eachhemisphere: (i) 0.2 µl at AP = +1.6, L = ±1.1, DV = –7.9 (SS); (ii) 0.1 µl at AP =+1.6, L = ±1.1, DV = –6,9; (iii) 0.1 µl at AP = +1.6, L = ±1.1, DV = –6.4. Shamand lesion groups were treated identically, except that sham controls receivedinjections of sterile PB instead of the toxin.

Intravenous catheterization. After a recovery period of at least 5 d with foodavailable ad libitum, rats were then implanted with chronic intravenous jugu-lar catheters as previously described48. Antibiotic treatment (daily subcuta-neous administration of 0.1 ml Baytrill; Bayer) was given for 5 d after surgery.Thereafter, before each self-administration session, the animals were flushedwith 0.1 ml sterile 0.9% saline and at the end of the session with 0.1 mlheparinised saline (CP Pharmaceuticals Ltd.; 30 units/ml 0.9% sterile saline)to maintain catheter patency.

Apparatus. Twelve operant chambers (24 cm wide × 20 cm high × 22 cm deep;Med Associates) contained within a sound-attenuating box with a ventilatingfan were used in the experiment. Each chamber was equipped with tworetractable levers, a stimulus light above each lever, a house light and threeinfrared beams (See Supplementary Methods online). Intravenous infusionsof cocaine were delivered by a software-operated infusion pump (SematTechnical Ltd.) placed outside the sound-attenuating box, through a counter-balanced single-channel liquid swivel. Animals were tethered to the counter-balanced arm by a metal spring and a skull-mounted plastic-post. Theapparatus was controlled by an Acorn Archimedes microcomputer (AcornComputers Ltd.) running a program written in the BASIC control language,Arachnid (Paul Fray Ltd.).

Drugs. Cocaine hydrochloride (McFarlan-Smith) was dissolved in sterile 0.9%saline. The dose of cocaine was calculated as the salt.

Behavioral procedures. Cocaine self-administration under continuousreinforcement: Animals were trained to acquire cocaine self-administrationunder a continuous reinforcement schedule (fixed ratio1) during daily 2-hsessions until stable baseline responding was achieved (defined as 2–3 con-secutive d of stable responding with as much as ±10% variation). For eachrat, one of the two levers was designated the active or drug lever, and theother was the inactive lever on which responding had no programmed con-sequence. No drug priming was given at any stage of training. The begin-ning of the session was marked by illumination of the house light.Subsequent depression of the active lever resulted in the retraction of bothlevers, extinction of the house light and simultaneous illumination of thedrug stimulus light for 20 s, as well as the activation of an infusion pump,


Figure 8 Effects of NAc lesions on cocaine-induced locomotor activityduring self-administration sessions. Vertical bars represent the mean ± s.e.m. photocell beam breaks for each group during a 2-h self-administration session on the first three and last days of acquisition. **P < 0.01, *P < 0.05 compared to the sham.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

delivering 0.1 ml intravenous infusion of cocaine solution (0.25 mg/infu-sion). On completion of the 20-s CS presentation per time out period, thelevers were re-extended, the house light illuminated and the stimulus lightextinguished. Throughout training, the maximum number of infusions persession was fixed at 50 to prevent overdosing. Once this number of cocaineinfusions had been reached, the session terminated.

Second-order schedule of cocaine reinforcement: Once self-administra-tion under a continuous reinforcement schedule had been attained, a sec-ond-order FRx(FRy:S) schedule of cocaine reinforcement was introduced.Under this schedule, rats were required to make y responses to obtain a sin-gle presentation of a 2 s light CS (or conditioned reinforcer) while comple-tion of x of these response units resulted in the delivery of cocaine, theillumination of the light CS for 20 s, the retraction of both levers and extinc-tion of the house light during a 20 s time out period. In the initial stage oftraining, x was set at 5 and y was 1. The value for x was then increased from 5to 10 and remained at this value throughout the training. The value for y wasprogressively increased from 1 to 10 until stable responding was establishedat FR10(FR10:S). Animals were allowed to move from one stage to the nextwhen at least ten cocaine infusions within a 2-h session were made over threeconsecutive days at each stage.

Cocaine dose-response function. A separate group of 18 rats (4 or 5 per treat-ment group) was subjected to a between-sessions cocaine dose-response func-tion once stable acquisition of cocaine self-administration under a CRFschedule at the training dose of 0.25 mg/infusion had been attained. The train-ing dose was substituted by 0.083, 0.125 or 0.50 mg per infusion of cocaine orsaline in a 2-h self-administration session on five consecutive days, and theorder in which the rats received each dose was counterbalanced.

Histological assessment of lesions. Within a week after completion of the test-ing, all rats were anesthetized with sodium pentobarbitone (1.5 ml/animal,200 mg/ml Euthatal, Rhone Merieux) and perfused intracardially via theascending aorta with 0.01 M phosphate-buffered saline (PBS) for 4 min, fol-lowed by 4% paraformaldehyde (PFA) in PBS for 6 min. Brains were thenremoved, stored in PFA and transferred to a 20% sucrose cryoprotectant solu-tion on the day before sectioning (See Supplementary Methods online). Forthe verification of lesions, coronal sections (60 µm) of the brain were cut usinga freezing microtome.

Statistical analysis. All behavioral data were analyzed using SPSS version 9.Responses during the acquisition of self-administration under CRF and sec-ond-order schedules were square-root transformed to preserve homogeneityof variance and analyzed using a three-factor analysis of variance (ANOVA)with group (core, shell, sham) as the between-group factor and training dayand lever (active vs. inactive) as repeated, within-subjects factors. Data col-lected from the seven core-lesioned animals that failed to complete the second-order schedule training were included in the statistical analyses, astheir exclusion did not alter statistical significance of the overall data.

Two-way ANOVAs (lesion group as between-subjects factor and trainingschedule or cocaine state as repeated measures) were conducted for all quan-titative data extracted from the response patterns for the second-orderschedule contingencies.

For all analyses, upon confirmation of significant main effects, differencesamong individual means were analyzed using the Newman-Keuls post-hoc test.Significant interactions were further analyzed as appropriate using two-way orone-way ANOVAs, with appropriate α adjustments using Sidak’s method (α′ = 1 – (1 – α)1/C, where C is the number of within-experiment analyses).


ACKNOWLEDGMENTSThis work was supported by an MRC Programme Grant (G9537855) andconducted within the MRC Centre for Behavioural and Clinical Neuroscience.R.I. was supported by an MRC research studentship. We thank D. Eagle, R.Cardinal and M. Aitken for helpful discussions on statistics.


Received 16 January; accepted 3 March 2004Published online at http://www.nature.com/natureneuroscience/

1. Roberts, D.C.S., Corcoran, M.E. & Fibiger, H.C. On the role of ascending cate-cholaminergic systems in intravenous self-administration of cocaine. Pharmacol.Biochem. Behav. 6, 615–620 (1977).

2. Pettit, H.O., Ettenberg, A., Bloom, F.E. & Koob, G.F. Destruction of dopamine in thenucleus accumbens selectively attenuates cocaine but not heroin self-administra-tion in rats. Psychopharmacology 84, 167–173 (1984).

3. Caine, S.B. & Koob, G.F. Effects of mesolimbic dopamine depletion on respondingmaintained by cocaine and food. J. Exp. Anal. Behav. 61, 213–221 (1994).

4. Robledo, P., Maldonado-Lopez, R. & Koob, G.F. Role of dopamine receptors in thenucleus accumbens in the rewarding properties of cocaine. Ann. NY Acad. Sci. 654,509–512 (1992).

5. Caine, S.B., Heinrichs, S.C., Coffin, V.L. & Koob, G.F. Effects of the dopamine D1antagonist SCH23390 microinjected into the accumbens, amygdala or striatum oncocaine self-administration in the rat. Brain Res. 692, 47–56 (1995).

6. Di Chiara, G.D. & Imperato, A. Drugs abused by humans preferentially increasesynaptic dopamine concentrations in the mesolimbic system of freely moving rats.Proc. Natl. Acad. Sci. USA 85, 5274–5278 (1988).

7. Hurd, Y., Weiss, F., Koob, G.F. & Ungerstedt, U. Cocaine reinforcement and extracel-lular DA overflow in rat nucleus accumbens: an in vivo microdialysis study. BrainRes. 498, 199–204 (1989).

8. Pettit, H.O. & Justice Jr, J.B. Dopamine in the nucleus accumbens during cocaineself-administration as studies by in vivo microdialysis. Pharmacol. Biochem. Behav.34, 899–904 (1989).

9. Bradberry, C.W., Barrett-Larimore, R.L., Jatlow, P. & Rubino, S.R. Impact of self-administered cocaine and cocaine cues on extracellular dopamine in mesolimbicand sensorimotor striatum in rhesus monkeys. J. Neurosci. 20, 3874–3883(2000).

10. Ito, R., Dalley, J.W., Howes, S.R., Robbins, T.W. & Everitt, B.J. Dissociation in con-ditioned dopamine release in the nucleus accumbens core and shell in response tococaine cues and during cocaine-seeking behavior in rats. J. Neurosci. 20,7489–7495 (2000).

11. Pulvirenti, L., Maldonado-Lopez, R. & Koob, G.F. NMDA receptors in the nucleusaccumbens modulate intravenous cocaine but not heroin self-administration in therat. Brain Res. 594, 327–330 (1992)

12. Cornish, J.L., Duffy, P. & Kalivas, P.W. A role for glutamate transmission in therelapse to cocaine-seeking behavior. Neuroscience 93, 1359–1367 (1999)

13. Zito, K.A., Vickers, G. & Roberts, D.C.S. Disruption of cocaine and heroin self-administration following kainic acid lesions of the nucleus accumbens. Pharmacol.Biochem. Behav. 23, 1029–1036 (1985).

14. Dworkin, S.I., Guerin, G.F., Goeders, N.E. & Smith, J.E. Kainic acid lesions of thenucleus accumbens selectively attenuate morphine self-administration. Pharmacol.Biochem. Behav. 29, 175–181 (1988).

15. Alderson, H.L., Parkinson, J.A., Robbins, T.W. & Everitt, B.J. The effects of excito-toxic lesions of the nucleus accumbens core or shell regions on intravenous heroinself-administration in rats. Psychopharmacology 153, 455–463 (2001).

16. Hutcheson, D.M., Parkinson, J.A., Robbins, T.W. & Everitt, B.J. The effects ofnucleus accumbens core and shell lesions on intravenous heroin self-administrationand the acquisition of drug-seeking behaviour under a second-order schedule ofheroin reinforcement. Psychopharmacology 153, 464–472 (2001).

17. Everitt, B.J., Dickinson, A. & Robbins, T.W. The neuropsychological basis of addic-tive behaviour. Brain Res. Rev. 36, 129–138 (2001).

18. Robbins, T.W. & Everitt, B.J. Drug addiction: bad habits add up. Nature 398,567–570 (1999).

19. Grant, S. et al. Activation of memory circuits during cue-elicited cocaine craving.Proc. Nat. Acad. Sci. USA 93, 12040–12045 (1996).

20. Childress, A.R. et al. Limbic activation during cue-induced cocaine craving. Am. J. Psychiatry 156, 11–18 (1999).

21. Stewart, J., de Wit, H. & Eikelboom, R. Role of unconditioned and conditioned drugeffects in the self-administration of opiates and stimulants. Psychol. Rev. 91,256–268 (1984).

22. Arroyo, M., Markou, A., Robbins, T.W. & Everitt, B.J. Acquisition, maintenance, andreinstatement of intravenous cocaine self-administration under a second-orderschedule of reinforcement in rats: effects of conditioned cues and continuousaccess to cocaine. Psychopharmacology 140, 331–344 (1998).

23. Everitt, B.J. & Robbins, T.W. Second-order schedules of drug reinforcement in ratsand monkeys: measurement of reinforcing efficacy and drug-seeking behaviour.Psychopharmacology 153, 17–30 (2000).

24. Weiss, F. et al. Control of cocaine-seeking behavior by drug-associated stimuli inrats: Effects on recovery of extinguished operant-responding and extracellulardopamine levels in amygdala and nucleus accumbens. Proc. Natl. Acad. Sci. USA97, 4321–4326 (2000).

25. See, R.E. Neural substrates of conditioned-cued relapse to drug-seeking behavior.Pharmacol. Biochem. Behav. 71, 517–529 (2002).

26. Whitelaw, R.B., Markou, A., Robbins, T.W. & Everitt, B.J. Excitotoxic lesions of thebasolateral amygdala impair the acquisition of cocaine-seeking behaviour under asecond order schedule of reinforcement. Psychopharmacology 127, 213–224(1996).

27. Meil, W.M. & See, R.E. Lesions of the basolateral amygdala abolish the ability ofdrug associated cues to reinstate responding during withdrawal from self-adminis-tered cocaine. Behav. Brain Res. 87, 139–148 (1997).


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

28. Voorn, P., Gerfen, C.R. & Groenewegen, H.J. Compartmental organization of the ven-tral striatum of the rat: Immunohistochemical distribution of enkephalin, substanceP, dopamine and calcium-binding protein. Comp. Neurol. 289, 189–201 (1989).

29. Heimer, L., Zahm, D.S., Churchill, L., Kalivas, P.W. & Wohltmann, C. Specificity inthe projection patterns of accumbal core and shell in the rat. Neuroscience 41,89–125 (1991).

30. Di Chiara G, Tanda G, Frau, R. & Carboni E. On the preferential release ofdopamine in the nucleus accumbens by amphetamine: further evidence obtainedby vertically implanted concentric dialysis probes. Psychopharmacology 112,398–402 (1993).

31. Pontieri, F.E., Tanda, G. & Di Chiara, G. Intravenous cocaine, morphine, andamphetamine preferentially increase extracellular dopamine in the “shell” as com-pared with the “core” of the rat nucleus accumbens. Proc. Natl. Acad. Sci. USA 92,12304–12308 (1995).

32. Carlezon, W.A. & Wise, R.A. Rewarding actions of phencyclidine and related drugsin nucleus accumbens shell and frontal cortex. J. Neurosci. 16, 3112–3122(1996).

33. Rodd-Hendricks, Z.A., McKinzie, D.L., Li, T-K., Murphy, J.M. & McBride, W.J.Cocaine is self-administered into the shell but not the core of the nucleus accum-bens of Wistar rats. J. Pharmacol. Exp. Ther. 303, 1216–1226 (2002).

34. Cardinal, R.N., Parkinson, J.A., Hall, J. & Everitt, B.J. Emotion and motivation: therole of the amygdala, ventral striatum, and prefrontal cortex. Neurosci. Biobehav.Rev. 26, 321–352 (2002).

35. Parkinson, J.A., Willoughby, P.J., Robbins, T.W. & Everitt, B.J. Disconnection of theanterior cingulate cortex and nucleus accumbens core impairs Pavlovian approachbehaviour: Further evidence for limbic cortical-ventral striatopallidal systems.Behav. Neurosci. 114, 42–63 (2000).

36. Hall, J., Parkinson, J.A., Connor, T.M., Dickinson, A. & Everitt, B.J. Involvement ofthe central nucleus of the amygdala and nucleus accumbens core in mediatingPavlovian influences on instrumental behaviour. Eur. J. Neurosci. 13, 1984–1992(2001).

37. Dalley, J.W. et al. Nucleus accumbens dopamine and discriminated approach learn-ing: interactive effects of 6-hydroxydopamine lesions and systemic apomorphineadministration. Psychopharmacology 161, 425–433 (2002).

38. Parkinson, J.A., Olmstead, M.C., Burns, L.H., Robbins, T.W. & Everitt, B.J.Dissociation in effects of lesions of the nucleus accumbens core and shell on appet-

itive Pavlovian approach behavior and the potentiation of conditioned reinforcementand locomotor activity by d-amphetamine. J. Neurosci. 19, 2401–2411 (1999).

39. Wyvell, C.L. & Berridge, K.C. Intra-accumbens amphetamine increases the condi-tioned incentive salience of sucrose reward: enhancement of reward ‘Wanting’ with-out enhanced ‘Liking’ or response reinforcement. J. Neurosci. 20, 8122–8130(2000).

40. Cador, M., Robbins, T.W. & Everitt, B.J. Involvement of the amygdala in stimulusreward associations- interaction with the ventral striatum. Neuroscience 30, 77–86(1989).

41. Burns, L.H., Robbins, T.W. & Everitt, B.J. Differential effects of excitotoxic lesionsof the basolateral amygdala, ventral subiculum and medial prefrontal cortex onresponding with conditioned reinforcement and locomotor activity potentiated byintra-accumbens infusions of d-amphetamine. Behav. Brain. Res. 55, 167–183(1993).

42. Maldonado-Irizarry, C.S. & Kelley, A.E. Excitotoxic lesions of the core and shell sub-regions of the nucleus accumbens differentially disrupt body weight regulation andmotor activity in rat. Brain Res. Bull. 38, 551–559 (1995).

43. Annett, L.E., McGregor, A. & Robbins, T.W. The effects of ibotenic acid lesions ofthe nucleus accumbens on spatial learning and extinction in the rat. Behav. Brain.Res. 31, 231–242 (1989).

44. Reading, P.J. & Dunnett, S.B. The effects of excitotoxic lesions of the nucleusaccumbens on a matching to position task. Behav. Brain Res. 46, 17–29 (1991).

45. Cardinal, R.N., Pennicott, D.R., Sugathapala, C.L., Robbins, T.W. & Everitt, B.J.Impulsive choice induced in rats by lesions of the nucleus accumbens core. Science292, 2499–2501 (2001).

46. Wise, R.A. & Bozarth, M.A. A psychomotor stimulant theory of addiction. Psychol.Rev. 94, 469–492 (1987).

47. Haber, S.N., Fudge, J.L. & McFarland, N.R. Striatonigrostriatal pathways in pri-mates from an ascending spiral from the shell to the dorsolateral striatum. J. Neurosci. 20, 2369–2382 (2000).

48. Caine, S.B., Lintz, R. & Koob, G.F. Intravenous self-administration techniques inanimals. in Behavioral Neuroscience: a Practical Approach Vol. 2 (ed. Sahgal, A.)117–143 (IRL Press, Oxford, UK, 1992).

49. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates (Academic Press,San Diego, 1998).


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Pairing a stimulus with intrinsic motivational power (unconditionedstimulus, US; e.g., pain, food) with a neutral sensory stimulus pro-duces changes in neural circuitry such that the previously neutralstimulus becomes capable of generating behavioral responses (condi-tioned responses, CRs). Thus the previously neutral stimulusbecomes a conditioned stimulus (CS). We use the term ‘teaching sig-nal’ to refer to a neural signal that is necessary and sufficient to pro-duce a conditioned response (CR). Accordingly, temporallycoincident activation of the pathways transmitting the aversive teach-ing signal and the initially neutral stimulus produces aversive associa-tive learning by strengthening the connections between the neuronsmediating the CS and those whose activity results in the CR.Associative learning using noxious stimuli as the US has been docu-mented behaviorally, and the mediating synaptic change has been elu-cidated in some systems1–4. However, the neural pathways thatmediate nociceptor-driven aversive teaching signals in mammals arenot well understood.

One candidate pathway for such signals includes projectionsfrom the spinal cord dorsal horn to the medial thalamus, and fromthere to the ACC5–8. This pathway was established by functionalimaging studies in humans9 and anatomical and electrophysiologi-cal studies in animals10–15. In human imaging studies, the degree ofACC activation is positively correlated with the magnitude ofunpleasantness in response to a noxious stimulus16. In addition, thehuman, primate, rodent and rabbit ACCs contain neurons thatrespond to noxious stimuli13,17–19. In chronic pain patients, lesionsof the ACC or cingulum bundle (an afferent and efferent ACC fibertract) reduce pain unpleasantness20,21. The ACC has extensive directinterconnections with limbic nuclei including the amygdala, hip-pocampus, posterior cingulate and ventral striatum22–25, each ofwhich has been implicated in CS-driven aversive behaviors2,26,27.

The observations that ACC neurons respond to noxious stimulationand that ACC activity is correlated with perceived unpleasantness inhumans are consistent with the hypothesis that ACC neuronsencode and transmit information related to the aversiveness of nox-ious stimuli and provide the teaching signal required for the acquisi-tion of conditioned aversion.

We recently found that excitotoxic lesions of the rostral ACC (r-ACC) selectively prevents avoidance learning elicited by tonicnoxious stimuli28. This is consistent with reports that lesions offrontal cortex or of both anterior and posterior cingulate corticesprior to conditioning reduce avoidance learning26,29. Although thesestudies show that the ACC is required for aversive learning, they donot distinguish between a role for ACC neurons in its acquisition(i.e., in providing an aversive teaching signal) versus expression (i.e.,in retrieval). Furthermore, the electrophysiological data are ambigu-ous on this question. In addition to responding to noxious stimuli(essential if they are to provide an aversive teaching signal and con-tribute to acquisition of the CR), some ACC neurons respond topain-predictive sensory stimuli. For example, human imaging androdent, rabbit and primate electrophysiology studies show activationof ACC neurons in response to a pain-predictive visual CS18,26,30–32.This activation supports the idea that ACC neurons encode andtransmit information that generates the motivational properties ofthe CS after conditioning, rather than generating an aversive teachingsignal during learning. In other words, this pattern of activity is moreconsistent with a role for ACC neurons in the expression rather thanthe acquisition of learned aversive behaviors. Finally, the facts thatneurons responding to both nociceptive (US) and aversive CS arefound in the ACC and that, after learning, some ACC neuronsrespond to both types of stimuli18 raise the intriguing possibility thatthe ACC is a critical site of plasticity for avoidance learning.

1Departments of Neurology and Physiology and 2The W.M. Keck Foundation Center for Integrative Neuroscience, University of California, San Francisco, 513Parnassus Avenue, S-784, San Francisco, California 94143-0453, USA. 3Present address: Interdepartmental Ph.D. Program for Neuroscience, UCLA, Los Angeles,California 90095, USA. Correspondence should be addressed to H.L.F. ([email protected]).

Published online 7 March 2004; corrected 12 March 2004 (details online); doi:10.1038/nn1207

Glutamatergic activation of anterior cingulate cortexproduces an aversive teaching signalJoshua P Johansen1–3 & Howard L Fields1,2

Noxious stimuli have motivational power and can support associative learning, but the neural circuitry mediating such avoidancelearning is poorly understood. The anterior cingulate cortex (ACC) is implicated in the affective response to noxious stimuli and themotivational properties of conditioned stimuli that predict noxious stimulation. Using conditioned place aversion (CPA) in rats, wefound that excitatory amino acid microinjection into the ACC during conditioning produces avoidance learning in the absence of aperipheral noxious stimulus. Furthermore, microinjection of an excitatory amino acid antagonist into the ACC during conditioningblocked learning elicited by a noxious stimulus. ACC lesions made after conditioning did not impair expression of CPA. Thus, ACCneuronal activity is necessary and sufficient for noxious stimuli to produce an aversive teaching signal. Our results support the ideathat a shared ACC pathway mediates both pain-induced negative affect and a nociceptor-driven aversive teaching signal.

A R T I C L E S


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

While not mutually exclusive, these hypotheses lead to clearlydifferent predictions of the effect of ACC manipulations on theacquisition and expression of avoidance learning. Thus, if ACC neu-rons are required to mediate the motivational effect of aversive con-ditioned stimuli, then lesions after conditioning should block theexpression of avoidance learning. In contrast, if ACC neurons arenecessary to provide a nociceptive aversive teaching signal, thenACC lesions before conditioning (or reversible inactivation duringconditioning) should block acquisition, but lesions made after con-ditioning should not affect expression of CPA after it has beenlearned. Furthermore, if ACC neuronal activity is sufficient to pro-vide an aversive teaching signal, direct activation of these ACC neu-rons during conditioning, in the absence of a peripheral noxiousstimulus, should produce an aversive teaching signal. Finally, if theaversive learning is associated with requisite synaptic plasticitywithin the ACC, lesions before conditioning should block acquisi-tion, and lesions after conditioning should block expression ofavoidance learning.

We previously examined the functional significance of the ACCusing a nociceptor-driven, associative avoidance-learning assay:formalin-induced conditioned place aversion (F-CPA)28. However,because the lesions in our earlier study were irreversible and madebefore conditioning, one could not distinguish an effect on acquisi-tion from one on expression. In the current study, to address thisquestion, we inactivated or lesioned the ACC in a temporally spe-cific manner. In addition, by activating ACC neurons directly, in theabsence of a peripheral nociceptive input, we explored whetheractivity of ACC neurons is sufficient to provide an aversive teachingsignal. Our results provide direct evidence that ACC neuronal activ-ity is sufficient to produce avoidance learning and necessary fornoxious stimuli to elicit an aversive teaching signal.

RESULTSr-ACC lesions do not affect the expression of avoidance learningExcitotoxin-induced r-ACC lesions were made after acquisition of theconditioned response to test whether the r-ACC is necessary for theexpression of previously learned avoidance behavior.

Bilateral infusions of the excitotoxinibotenic acid (IBO) made into r-ACC pro-duced neuronal cell loss and proliferation ofsmall glial cells (data not shown; see ref. 28).All animals included in our analyses metlesion inclusion criterion as described inMethods (Fig. 1a). Mean percent damagecalculations for each hemisphere and anoverall bilateral mean are as follows: lefthemisphere, 66 ± 11%; right hemisphere,58 ± 11%; mean, 62 ± 9%. Importantly, thelesion extents in this experiment were notdifferent from those in our previous study28.

When hindpaw formalin injections werepaired with a particular compartment in theplace-conditioning apparatus, rats with post-training r-ACC sham lesions spent less timein the formalin-paired room (i.e., CPA wasproduced; 389.8 ± 54.8 s pre-conditioning vs.211.6 ± 90.2 s post-conditioning; Student’s t-test, P < 0.05). Hindpaw formalin also pro-duced CPA in post-training r-ACC lesionedrats (392 ± 131.6 s pre-conditioning vs.184 ± 94.9 s post-conditioning; Student’s

t-test, P < 0.05). Group comparisons revealed no significant differ-ence between sham and lesion groups (Fig. 1b; Student’s t-test,P > 0.05). Thus, r-ACC lesions made after training have no effect onthe expression of F-CPA. Two critical conclusions can be drawn fromthis result. First, the r-ACC is not a significant site of plasticity for F-CPA learning and, second, it is not required for retrieval of infor-mation related to the prediction of aversive stimuli by contextual cues.

r-ACC glutamate receptor blockade prevents F-CPA acquisitionThe fact that lesions made before28 but not after conditioning blockF-CPA learning strongly supports the hypothesis that the r-ACC isnecessary specifically during the acquisition of F-CPA. The existenceof a significant spino-thalamo-cingulate nociceptive projection path-way10–15 is also consistent with a major role for the r-ACC in afferentnociceptive processing. Assuming that the thalamo-cingulate projec-tion is glutamatergic33, it is likely that glutamatergic activation ofr-ACC neurons by a prolonged noxious stimulus is necessary for theacquisition of CPA. To address this question, we made glutamatereceptor antagonist microinjections into the r-ACC during formalinconditioning sessions.

Microinjections of kynurenic acid (KyA) into the r-ACC beforeF-CPA conditioning blocked the acquisition of F-CPA learning (Fig. 2). There was no difference in the amount of time the r-ACCKyA animals spent in the formalin-paired context before versus afterconditioning (357.5 ± 50.6 s before, 320.6 ± 83.2 s after; Student’s t-test, P > 0.05). In contrast, microinjections of vehicle into the r-ACC during conditioning had no effect on F-CPA acquisition(388.3 ± 79.4 s before, 234.8 ± 106.5 s after conditioning; Student’s t-test, P < 0.01). For group comparisons, see Figure 2a. Notably, in aseparate group of animals, KyA alone did not produce motivationaleffects. KyA microinjected into the r-ACC in the absence of hindpawformalin had no effect on room preference (336.6 ± 86.3 s before vs.321.3 ± 137.1 s after conditioning; Student’s t-test, P > 0.05; Fig. 2b).Furthermore, the reduction of F-CPA by KyA injected into the r-ACC is unlikely to be due to a sedating effect since it did not altermotor activity (data not shown). In further support of this conclu-sion, our previous study showed that similarly located IBO-induced

A R T I C L E S


Figure 1 ACC lesions after training do not affect expression of place aversion. (a) Examples of the largest(gray) and smallest (black) lesions among animals in the group. Sections are in the coronal plane,numbers in mm anterior to Bregma in this and subsequent figures. (b) Rats with post-training lesions (n = 7) did not differ from those with sham lesions (n = 10). F-CPA scores are shown as mean ± s.e.m.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

r-ACC lesions had no effect on the place aversion elicited by sys-temic injection of the kappa opioid agonist U69,593 (ref. 28). As inour earlier lesion study, we found no significant main effect ofintracerebral treatment (vehicle vs. KyA) on acute formalin ratingscale scores (F1,12 = 0.97; P > 0.05) and no significant interactionbetween intracerebral treatment and time (F9,108 = 0.72, P > 0.05;Fig. 2c), indicating that KyA reduction of F-CPA is not due to a gen-eral decrease in nociceptive processing.

Glutamatergic r-ACC stimulation produces avoidance learningThe results of experiments 1 and 2 indicate that activation ofr-ACC neurons is necessary for acquisition, but not expression, ofF-CPA. However, they do not rule out the possibility that nocicep-tive activation of r-ACC neurons serves a permissive role duringconditioning and that activation of r-ACC neurons alone is not suf-ficient to produce F-CPA learning. To test this possibility, wedirectly stimulated the r-ACC by microinjecting an ionotropic glu-tamate receptor agonist into the r-ACC in the absence of a periph-eral nociceptive stimulus.

Homocysteic acid microinjected into the r-ACC produced sig-nificant, dose-dependent CPA learning. Rats spent significantly lesstime in the treatment-paired context (366.8 ± 44.8 before vs. 251.2 ±59.4 after conditioning; Student’s t-test, P < 0.01). Neither intra-

r-ACC vehicle nor low-dose homocysteic acid produced CPA (vehi-cle, 336.1 ± 58.3 s before vs. 308.4 ± 92.2 s after conditioning;Student’s t-test, P > 0.05; low-dose, 352.1 ± 34.6 s before vs. 365.75 ±69.8 s after conditioning; Student’s t-test, P > 0.05). Group compar-isons of magnitude of CPA scores analyzed using a one-way ANOVArevealed a significant effect of treatment (F2,27 = 6.46; P < 0.01).Further analysis revealed significantly higher CPA scores for the 100 mM HCA treatment group compared to the vehicle group, butno significant difference between vehicle and 5 mM HCA(Newman-Keuls test; P < 0.05 and P > 0.05, respectively; Fig. 3a). Toestablish the anatomical specificity of our r-ACC microinjections,we used off-site controls (Fig. 3b). High-dose HCA had no motiva-tional effects when injected into a cortical control site lateral to ourtarget r-ACC injections (n = 8; 347.4 ± 43.8 s before vs. 356.5 ± 157.6s after conditioning; Student’s t-test, P > 0.05).

DISCUSSIONPreviously we demonstrated that excitotoxic lesions of the r-ACC beforeconditioning abolish nociceptor-driven learned avoidance behavior (F-CPA) without affecting acute nociceptive behaviors or non-nociceptiveavoidance behavior28. Our current study extends those findings byshowing that activation of r-ACC neurons is required specifically for theacquisition of F-CPA, as lesions made after conditioning have no effect

on the expression of F-CPA. In addition, ourdata implicate r-ACC excitatory neurotrans-mission specifically in the acquisition ofF-CPA, as r-ACC microinjection of a gluta-mate receptor antagonist during acquisitionblocks F-CPA conditioning.

Importantly, our data provide critical evidence supporting the hypothesis that r-ACC neuronal activity is sufficient to gener-ate an aversive teaching signal. Thus, microin-jection of a glutamate receptor agonist intothe r-ACC, but not into an adjacent corticalsite, during conditioning produces robustCPA in the absence of a concomitant periph-eral noxious stimulus. That selective activa-tion of r-ACC neurons is sufficient to produceavoidance learning in the absence of inputfrom primary afferent nociceptors is directevidence that ACC neuronal activity is causalrather than permissive for avoidance learning.

A R T I C L E S


Figure 2 Intra-r-ACC microinjection of the ionotropic glutamate receptor antagonist kynurenic acid (KyA) blocks F-CPA. (a–c) Data are represented asmean ± s.e.m. (a) The effect of bilateral intra-r-ACC vehicle (n = 10) or KyA (n = 8) on the magnitude of F-CPA scores. (b) The magnitude of CPA scores forintra-r-ACC KyA in the absence of hindpaw formalin (n = 8). (c) Acute formalin-induced nociceptive scores (rating scale). (d) Injection sites for 50 mMKyA-treated rats. *P < 0.05, Student’s t-test as compared with vehicle-injected rats.

Figure 3 CPA is produced by glutamatergic stimulation of the r-ACC. (a) Magnitude of CPA scores inanimals given intra-r-ACC microinjection of vehicle (n = 9), 5 mM (n = 8) or 100 mM HCA (n = 11).Data are represented as mean ± s.e.m. *P < 0.05, as compared with vehicle injected rats. (b) Injection sites for 100 mM HCA r-ACC (circles) and off-site injection sites (triangles).

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Although the evidence is not conclusive, a parsimonious explanationof these results is that formalin injection produces an aversive teachingsignal through activation of r-ACC neurons during CPA conditioning.

A model: r-ACC pathway encodes an aversive teaching signalBecause dilute intradermal formalin selectively activates nociceptiveAδ and C-fiber primary afferent nociceptors34 and is painful inhumans35, F-CPA is, by definition, a nociceptor-driven learnedbehavior. Nociceptive stimuli reliably activate neurons in theACC13,17–19. Furthermore, together with the fact that inactivation ofthe medial thalamus13—an area that receives direct and indirectspinal cord projections and projects to the r-ACC10–12,14,15—reducesthis activation, the current results strongly support the idea that theACC is a major terminus or relay site for a nociceptive afferent path-way. Consistent with the idea that this region of the r-ACC con-tributes selectively to nociceptor-driven aversive processing, wepreviously showed that CPA produced by the systemic administrationof the kappa opioid agonist U69,593 is unaffected by lesions of the r-ACC28. Importantly, this result implies that r-ACC lesions do notproduce a general disruption of associative learning. In addition, weand others have shown that lesions of the rostral28 or caudal ACC36,37

(but see also ref. 38) spare other unconditioned behavioral responses(UR) elicited by noxious stimuli. On the other hand, caudal ACClesions appear to reduce acute escape responses to noxious heat36.

The results of the current experiment and previous work thusdemonstrate that a nociceptive pathway through the r-ACC is neces-sary and sufficient for peripheral noxious stimuli to produce aver-sive teaching signals. The r-ACC is not necessary for other URs tonociceptive stimuli (e.g., acute formalin behaviors). Thus at somepoint afferent to the r-ACC, the afferent pathway mediating theaversive teaching signal diverges from that mediating many of theacute behavioral responses elicited by noxious stimuli. Our data alsoindicate that the neural plasticity underlying the development ofavoidance learning occurs in areas of the brain that receive conver-gent input from the r-ACC neurons encoding the aversive teachingsignal and from other sensory pathways whose neurons encodeinformation about initially neutral conditioned stimuli. Workingwithin this model, glutamatergic activation of r-ACC neurons bynoxious stimuli is necessary to produce F-CPA, and direct activationof r-ACC neurons is sufficient to serve as a teaching signal for thistype of avoidance learning.

A teaching signal, in this model, serves to strengthen the CSinputs onto neurons that receive convergent input from both noci-ceptive (teaching input) and other sensory CS pathways. Thestrengthening of the CS input such that it becomes capable of elicit-ing the CR is manifest in the current study as the acquisition of CPA.This type of plasticity elicited by a noxious US has been reported atthe synaptic level in other neural systems. For example, using in-vivointracellular recordings, one study demonstrated enhanced synapticstrength of an olfactory input to an amygdala neuron by temporallycoincident activation of a noxious stimulus input onto the sameneuron4. Interestingly, a recent report suggests that ACC stimulationis also necessary and sufficient to produce amygdala-dependentaversive conditioning (Tang, T. & Zhuo, M. Soc. Neurosci. Abstr.293.4, 2003), suggesting that an aversive teaching signal generatedby ACC neuronal activity is involved in other forms of aversivelearning. Although we have shown that activation of r-ACC neuronsis necessary and sufficient to produce an aversive teaching signal,future studies in regions that receive ACC input and convergentcontextual sensory inputs are necessary to determine the site andmechanisms of the synaptic plasticity that underlies such learning.

CS-responsive neurons in ACCAlthough there are r-ACC neurons that respond to stimuli (CS) thatpredict a nociceptive stimulus (in the current experiments, contextualsensory cues in the chamber where the rats received either formalin orintra-ACC HCA), our results do not bear on the function of such CSresponses. Although our work does not preclude a role for r-ACCneurons with pain-predictive responses in F-CPA conditioning, it isclear that they are not required for the expression of F-CPA under theconditions of our experiment. One possibility is that different formsof aversive learning recruit the ACC differentially39–41. Another possi-bility is that r-ACC CS-responsive neurons are involved in a processother than aversive learning. Some studies have implicated the ACC innociceptive modulation42–45 and also in learned hormonal responsesto pain-predictive cues46. Further experiments are necessary toexplore these questions and to define other functional roles for CS-responsive neurons in the r-ACC.

Implications for chronic pain syndromesBecause psychological and emotional dysfunctions are characteristicof chronic pain syndromes, it may be of great clinical importance tounderstand how the nociceptive pathway through the r-ACC con-tributes to the long-term behavioral and subjective effects of chronicconditions associated with recurrent and/or prolonged nociceptoractivation. Indeed, animal studies report persistent activity47 andplastic changes within the ACC after nerve injury48,49, suggesting thatpersistent noxious input can lead to local ACC plasticity (sensitiza-tion). If the ACC nociceptive system is tonically sensitized underchronic pain conditions, an understanding of the processes that leadto this change and its consequences in downstream projection targetswould be of significant clinical importance.

In summary, the ACC is part of a nociceptor-activated circuitthat, when paired with a contextual CS, can produce a teaching sig-nal resulting in avoidance learning. Since lesions or glutamate recep-tor blockade of r-ACC neurons reduce acquisition, but post-conditioning lesions do not affect expression of such learning, theaversive teaching signal must act on other areas of the brain wherenociceptive US and contextual information (CS) converge to pro-duce the synaptic changes underlying the learned avoidanceresponse (CR). Consistent with this idea, excitatory amino acidstimulation of the r-ACC without peripheral noxious stimulation issufficient to produce CPA learning. Whereas human studies suggestthat the ACC processes information relating to the unpleasantnessof the stimulus, our data indicate that this signal is necessary to pro-duce avoidance learning. Together, the human and animal studiessupport the hypothesis that a circuit through the ACC encodes thenegative affective quality elicited by noxious stimuli and concomi-tantly provides an aversive teaching signal.

METHODSSubjects. Subjects were male Long Evans rats (Simonsen Laboratories)weighing 300–350 g at the start of the experiments. Rats were group-housedon a 12-h light-dark schedule with food and water available ad libitum. Allexperiments were carried out with the approval of the Institutional AnimalCare and Use Committee at the University of California, San Francisco. Allefforts were made to minimize animal suffering and reduce the numbers of animals used.

Drugs. Ibotenic acid (IBO, 1.9 M) was dissolved in 0.1 M PBS and adjusted topH 7.2–7.4 using 1.0 M NaOH. Stock formaldehyde solution (37% formalde-hyde or 100% formalin) was diluted to 2.5% formalin in isotonic saline. The glu-tamate agonist, homocysteic acid (HCA, 5 or 100 mM) was dissolved in isotonicsaline and adjusted to pH 7.2–7.4 using 1.0 M NaOH. The glutamate antagonist

A R T I C L E S


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

kynurenic acid (KyA, 50 mM) was dissolved in a vehicle solution (60% isotonicsaline/40% 0.1 M NaOH) and adjusted to pH 7.2–7.4 using 1.0 M HCl.

Surgery. Animals were anesthetized with an intraperitoneal (i.p.) injection ofsodium pentobarbital (50 mg/kg). Surgery was performed using a Kopf stereo-taxic apparatus. For lesion experiments, an injection cannula (30-gauge stainless steel tubing) filled with IBO or 0.1 M PBS was connected to amicroinfusion pump (Razel Scientific Instruments) via PE 10 tubing. Surgerydetails and coordinates for lesion procedures are as previously reported28.

For microinjection studies, chronic guide cannulae (33-gauge, Small Parts)were implanted using the stereotaxic procedure described above. Double (1.2 mm spacing between barrels) stainless steel guide cannulae wereimplanted 1 mm above the ACC injection site (coordinates from Bregma:anterior/posterior (AP), +2.6; dorsal/ventral (DV), –1.6; medial/lateral (ML),0.6 mm on each side). Single barreled stainless steel guide cannulae wereimplanted lateral to the ACC injection site for off-site control experiments(coordinates from Bregma: AP, +2.6; DV, –1.5; ML, 2.5 on each side). For bothon and off-site experiments, injectors were inserted into the guide cannulaeand extended 1 mm beyond the guide tip (see below for microinjectiondetails). Stainless steel dummy cannulae extending to the tip of the guide can-nulae were inserted and kept the guide free of debris during the recoveryperiod. All animals (lesion, sham and microinjection) recovered normallyfrom surgery as evidenced by a weight gain on the first test day.

Behavioral training and microinjections. All experiments were done asdescribed previously28 using a counterbalanced, unbiased CPA design. Theapparatus was exactly as described28: a box with three distinct compartments(a neutral room and two conditioning rooms with distinct olfactory and visualcues) with a removable door to allow room isolation when necessary andphoto beams along the floor to record the animal’s position and motor activ-ity. All animals were handled for 3 d prior to testing and habituated to theinjection chamber (for microinjection studies). The amount of time the ani-mal spent in the treatment-paired room before vs. after testing was recordedand used for analysis (see below). No initial preferences for any of the com-partments in the place-conditioning apparatus were detected before condi-tioning, indicating that the rats did not prefer any one compartment to theothers before conditioning.

Lesion study of F-CPA expression. Briefly, experiments began with a pre-testday, during which the animal was allowed to roam freely around all the rooms,and we recorded the amount of time spent in each. This was followed by fourconditioning days where the animals were confined to one of the conditioningrooms and received, on alternating days, either nothing in one context or a for-malin injection (alternating hindpaws) in the other context (2 UCS pairingstotal). Conditioning was followed on day 6 by a first post-test day on which theanimals were again given free access to all three rooms, and again we recordedthe amount of time spent in each room. Surgeries were performed the dayafter the first post-test, and testing began at least 6 d after surgery. After recov-ery, the animals were given a second post-test that was identical to the first.

Microinjection experiments. For all experiments, injectors were inserted intothe guide cannulae after removal of the dummy cannulae, and animals wereplaced in an injection chamber (injectors protruded 1 mm beyond the guidetip for on- and off-site experiments, so add 1 mm to coordinates given abovein “surgery” section for correct DV coordinates). The injectors were attachedto a microinfusion pump (Razel Scientific Instruments) via PE 10 tubing.Microinjections of drug or vehicle were made at a rate of 0.5 µl/1.5 min (0.5 µltotal volume/side), and the microinjection cannula was left in place for 2 minbefore and after microinjection.

For the glutamate antagonist experiments, microinjection of KyA was made5 min before the animal received a hindpaw formalin injection and was placedin the box for 50 min. Conditioning was accomplished in 2 d, not 4 d as in thelesion experiment, and included a pre-test and a post-test (4 d total). Thus, allanimals received treatment (drug/formalin or just drug) and vehicle contextpairings on the same day (counterbalanced by morning or afternoon) and notseparated by 1 d as in the lesion experiments. They still received the samenumber of formalin pairings (2) as in the lesion experiments and no differencein the magnitude of F-CPA was detected between the 4-d and 2-d conditioning

regimens (data not shown). Formalin behaviors were also scored using the rat-ing scale method50 on the first or second pairing day (counterbalanced).

For experiment 3, intra-ACC or off-site microinjections of a glutamate recep-tor agonist (HCA) were given without hindpaw formalin injections, and theanimals were placed in the conditioning context 5 min after microinjection for30 min. Experiment 3 was done using the same conditioning regimen as inexperiment 2, but three pairings of treatment (drug) and context were madeinstead of two (5 d total). Pre- and post-tests were identical to the first twoexperiments.

Histology. After completion of the experiments, animals were given a lethal doseof sodium pentobarbital and perfused transcardially with isotonic saline followedby 10% formalin. For microinjection experiments, microinjections of dilutemethylene blue were made into the r-ACC just before perfusion. The brains werethen removed and fixed first in formalin for 24 h, then in 30% sucrose 24–72 hbefore slicing. The brains were cut on a sledge microtome at a thickness of 50 µm,stained with cresyl violet and analyzed to assess the extent of the lesion (or injec-tion site) using a light microscope. Using a camera lucida (Nikon), lesions weretraced and analyzed using an unbiased stereological method28. Intra-ACCmicroinjection of IBO produced lesions with clearly definable borders of neu-ronal cell loss and gliosis as compared with intra-r-ACC microinjection of PBS.Based on past studies, areas of the rodent ACC rich in nociceptive input were tar-geted (see ref. 28 for detailed region-of-interest). Lesions meeting inclusion crite-ria had a minimum ‘percentage bilateral damage’ of 50% and at least 30%damage in the least damaged hemisphere within the region of interest.

Statistical analyses. For the CPA data, the amount of time spent in the condi-tioning compartment (i.e., compartment paired with formalin, drug/formalinor drug) on the post-conditioning day (i.e., final test day) was subtracted fromthe amount of time spent in the same compartment on the pre-conditioningday. This resulted in a ‘magnitude of CPA score’ for each rat. Magnitude ofCPA scores between groups were compared using a Student’s t-test when com-paring two groups (experiments 1 and 2) or a one-factor ANOVA (intracere-bral treatment) followed by a Newman-Keuls post-hoc test when comparingmore than two groups (experiment 3). In addition, the absolute amount oftime spent in the conditioning compartment on the pre-conditioning day ver-sus the post-conditioning day was compared in sham lesion, lesion, vehicle ordrug treated animals using correlated Student’s t-tests.

For analysis of formalin behaviors in experiment 2, rating scale nociceptivescores were collected either on day 1 or day 2 (counterbalanced) from forma-lin-treated rats during each 5-min time bin. The data then were analyzed inseparate two-factor ANOVAs (intracerebral treatment × time), with time ana-lyzed as a repeated measure. Post-hoc analyses were performed using theNewman-Keuls test. The accepted level of statistical significance for all experi-ments was P < 0.05.

ACKNOWLEDGMENTSThe authors thank I. Meng for valuable discussions throughout the course of thiswork. We also thank G. Hjelmstad, J. Levine, J. Mitchell and S. Nicola for readingthis manuscript, and C. Evans and C. Bryant for assistance in the completion ofthis study. Supported by a United States Public Health Service grant NS 21445.


Received 2 January; accepted 9 February 2004Published online at http://www.nature.com/natureneuroscience/

1. Kandel, E., Schwartz, J. & Jessel, T. Principles of Neural Science (McGraw-Hill, NewYork, 2000).

2. LeDoux, J.E. Emotion circuits in the brain. Annu. Rev. Neurosci. 23, 155–184(2000).

3. Quirk, G.J., Repa, C. & LeDoux, J.E. Fear conditioning enhances short-latency audi-tory responses of lateral amygdala neurons: parallel recordings in the freely behavingrat. Neuron 15, 1029–1039 (1995).

4. Rosenkranz, J.A. & Grace, A.A. Dopamine-mediated modulation of odour-evokedamygdala potentials during pavlovian conditioning. Nature 417, 282–287 (2002).

5. Fields, H.L. Pain: an unpleasant topic. Pain 6 (Suppl.), S61–S69 (1999).6. Vogt, B.A. & Sikes, R.W. The medial pain system, cingulate cortex, and parallel pro-

cessing of nociceptive information. Prog. Brain Res. 122, 223–235 (2000).

A R T I C L E S


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


7. Price, D.D. Psychological and neural mechanisms of the affective dimension ofpain. Science 288, 1769–1772 (2000).

8. Treede, R.D., Kenshalo, D.R., Gracely, R.H. & Jones, A.K. The cortical representa-tion of pain. Pain 79, 105–111 (1999).

9. Casey, K.L. Forebrain mechanisms of nociception and pain: analysis through imag-ing. Proc. Natl. Acad. Sci. USA 96, 7668–7674 (1999).

10. Cliffer, K.D., Burstein, R. & Giesler, G.J. Jr. Distributions of spinothalamic, spinohy-pothalamic, and spinotelencephalic fibers revealed by anterograde transport ofPHA-L in rats. J. Neurosci. 11, 852–868 (1991).

11. Peschanski, M. & Ralston, H.J. 3rd. Light and electron microscopic evidence oftransneuronal labeling with WGA-HRP to trace somatosensory pathways to the thal-amus. J. Comp. Neurol. 236, 29–41 (1985).

12. Berendse, H.W. & Groenewegen, H.J. Restricted cortical termination fields of themidline and intralaminar thalamic nuclei in the rat. Neuroscience 42, 73–102(1991).

13. Sikes, R.W. & Vogt, B.A. Nociceptive neurons in area 24 of rabbit cingulate cortex.J. Neurophysiol. 68, 1720–1732 (1992).

14. Hsu, M.M. & Shyu, B.C. Electrophysiological study of the connection betweenmedial thalamus and anterior cingulate cortex in the rat. Neuroreport 8,2701–2707 (1997).

15. Craig, A.D. Pain mechanisms: labeled lines versus convergence in central process-ing. Annu. Rev. Neurosci. 26, 1–30 (2003).

16. Rainville, P., Duncan, G.H., Price, D.D., Carrier, B. & Bushnell, M.C. Pain affectencoded in human anterior cingulate but not somatosensory cortex. Science 277,968–971 (1997).

17. Hutchison, W.D., Davis, K.D., Lozano, A.M., Tasker, R.R. & Dostrovsky, J.O. Pain-related neurons in the human cingulate cortex. Nat. Neurosci. 2, 403–405 (1999).

18. Koyama, T., Tanaka, Y.Z. & Mikami, A. Nociceptive neurons in the macaque anteriorcingulate activate during anticipation of pain. Neuroreport 9, 2663–2667 (1998).

19. Yamamura, H. et al. Morphological and electrophysiological properties of ACCx noci-ceptive neurons in rats. Brain Res. 735, 83–92 (1996).

20. Foltz, E.L. & White, L.E. Pain relief by frontal cingulumotomy. J. Neurosurg. 19,89–100 (1962).

21. Hurt, R.W. & Ballantine, H.T. Jr. Stereotactic anterior cingulate lesions for persistentpain: a report on 68 cases. Clin. Neurosurg. 21, 334–351 (1974).

22. Allen, G.V. & Hopkins, D.A. Mamillary body in the rat: topography and synaptology ofprojections from the subicular complex, prefrontal cortex, and midbrain tegmentum.J. Comp. Neurol. 286, 311–336 (1989).

23. Cassell, M.D. & Wright, D.J. Topography of projections from the medial prefrontalcortex to the amygdala in the rat. Brain Res. Bull. 17, 321–333 (1986).

24. Christie, M.J., Summers, R.J., Stephenson, J.A., Cook, C.J. & Beart, P.M. Excitatoryamino acid projections to the nucleus accumbens septi in the rat: a retrograde trans-port study utilizing D[3H]aspartate and [3H]GABA. Neuroscience 22, 425–439(1987).

25. Reep, R.L. & Corwin, J.V. Topographic organization of the striatal and thalamic con-nections of rat medical agranular cortex. Brain Res. 841, 43–52 (1999).

26. Gabriel, M., Kubota, Y., Sparenborg, S., Straube, K. & Vogt, B.A. Effects of cingulatecortical lesions on avoidance learning and training-induced unit activity in rabbits.Exp. Brain Res. 86, 585–600 (1991).

27. Salamone, J.D., Cousins, M.S. & Snyder, B.J. Behavioral functions of nucleusaccumbens dopamine: empirical and conceptual problems with the anhedoniahypothesis. Neurosci. Biobehav. Rev. 21, 341–359 (1997).

28. Johansen, J.P., Fields, H.L. & Manning, B.H. The affective component of pain inrodents: direct evidence for a contribution of the anterior cingulate cortex. Proc.Natl. Acad. Sci. USA 98, 8077–8082 (2001).

29. Peretz, E. The effects of lesions of the anterior cingulate cortex on the behavior ofthe rat. J. Comp. Physiol. Psychol. 53, 540–548 (1960).

30. Ploghaus, A. et al. Dissociating pain from its anticipation in the human brain.Science 284, 1979–1981 (1999).

31. Takenouchi, K. et al. Emotional and behavioral correlates of the anterior cingu-late cortex during associative learning in rats. Neuroscience 93, 1271–1287(1999).

32. Nishijo, H. et al. Single neuron responses in the monkey anterior cingulate cortexduring visual discrimination. Neurosci. Lett. 227, 79–82 (1997).

33. Gigg, J., Tan, A.M. & Finch, D.M. Glutamatergic excitatory responses of anterior cin-gulate neurons to stimulation of the mediodorsal thalamus and their regulation byGABA: an in vivo iontophoretic study. Cereb. Cortex 2, 477–484 (1992).

34. Puig, S. & Sorkin, L.S. Formalin-evoked activity in identified primary afferent fibers:systemic lidocaine suppresses phase-2 activity. Pain 64, 345–355 (1996).

35. Dubuisson, D. & Dennis, S.G. The formalin test: a quantitative study of the anal-gesic effects of morphine, meperidine, and brain stem stimulation in rats and cats.Pain 4, 161–174 (1977).

36. Pastoriza, L.N., Morrow, T.J. & Casey, K.L. Medial frontal cortex lesions selectivelyattenuate the hot plate response: possible nocifensive apraxia in the rat. Pain 64,11–17 (1996).

37. Vaccarino, A.L. & Melzack, R. Analgesia produced by injection of lidocaine into theanterior cingulum bundle of the rat. Pain 39, 213–219 (1989).

38. Donahue, R.R., LaGraize, S.C. & Fuchs, P.N. Electrolytic lesion of the anterior cin-gulate cortex decreases inflammatory, but not neuropathic nociceptive behavior inrats. Brain Res. 897, 131–138 (2001).

39. Morgan, M.A. & LeDoux, J.E. Differential contribution of dorsal and ventral medialprefrontal cortex to the acquisition and extinction of conditioned fear in rats. Behav.Neurosci. 109, 681–688 (1995).

40. Barros, D.M. et al. Molecular signaling pathways in the cerebral cortex are requiredfor retrieval of one-trial avoidance learning in rats. Behav. Brain Res. 114, 183–192(2000).

41. Kung, J.C., Su, N.M., Fan, R.J., Chai, S.C. & Shyu, B.C. Contribution of the ante-rior cingulate cortex to laser-pain conditioning in rats. Brain Res. 970, 58–72(2003).

42. Hardy, S.G. Analgesia elicited by prefrontal stimulation. Brain Res. 339, 281–284(1985).

43. Hardy, S.G. & Haigler, H.J. Prefrontal influences upon the midbrain: a possible routefor pain modulation. Brain Res. 339, 285–293 (1985).

44. Lee, D.E., Kim, S.J. & Zhuo, M. Comparison of behavioral responses to noxious coldand heat in mice. Brain Res. 845, 117–121 (1999).

45. Calejesan, A.A., Kim, S.J. & Zhuo, M. Descending facilitatory modulation of abehavioral nociceptive response by stimulation in the adult rat anterior cingulatecortex. Eur. J. Pain 4, 83–96 (2000).

46. Buchanan, S.L. & Powell, D.A. Cingulate cortex: its role in Pavlovian conditioning. J. Comp. Physiol. Psychol. 96, 755–774 (1982).

47. Mao, J., Mayer, D.J. & Price, D.D. Patterns of increased brain activity indicative ofpain in a rat model of peripheral mononeuropathy. J. Neurosci. 13, 2689–2702(1993).

48. Wei, F., Li, P. & Zhuo, M. Loss of synaptic depression in mammalian anterior cingu-late cortex after amputation. J. Neurosci. 19, 9346–9354 (1999).

49. Wei, F. & Zhuo, M. Potentiation of sensory responses in the anterior cingulate cortexfollowing digit amputation in the anaesthetised rat. J. Physiol. 532, 823–833(2001).

50. Manning, B.H. A lateralized deficit in morphine antinociception after unilateralinactivation of the central amygdala. J. Neurosci. 18, 9453–9470 (1998).

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Decision making refers to an evaluative process of selecting a particu-lar action from a set of alternatives. When the mapping between aparticular action and its outcome or utility is fixed, the decision toselect the action with maximum utility can be considered optimal orrational. However, animals face more difficult problems in a multi-agent environment, in which the outcome of one’s decision can beinfluenced by the decisions of other animals. Game theory provides amathematical framework to analyze decision making in a group ofagents1–4. A game is defined by a set of actions available to each player,and a payoff matrix that specifies the reward or penalty for eachplayer as a function of decisions made by all players. A solution orequilibrium in game theory refers to a set of strategies selected by agroup of rational players1,5,6. Nash has proved that any n-player com-petitive game has at least one equilibrium in which no players canbenefit by changing their strategies individually5. These equilibriumstrategies often take the form of a mixed strategy, which is defined as aprobability density function over the alternative actions available toeach player. This requires players to choose randomly among alterna-tive choices, as in the game of rock-paper-scissors during whichchoosing one of the alternatives (e.g., paper) exclusively allows theopponent to exploit such a biased choice (with scissors).

Many studies have shown that people frequently deviate from thepredictions of game theory7–21. Although the magnitudes of suchdeviations are often small, they have important implications regard-ing the validity of assumptions in game theory, such as the rational-ity of human decision-makers22–27. In addition, strategies of humandecision-makers change with their experience17–21. These adaptiveprocesses might be based on reinforcement learning algorithms28,which can be used to approximate optimal decision-making strate-gies in a dynamic environment. In the present study, we analyzed the

performance of monkeys playing a zero-sum game against a com-puter opponent to determine how closely their behaviors match thepredictions of game theory and whether reinforcement learningalgorithms can account for any deviations from such predictions. Inaddition, neural activity was recorded from the DLPFC to investigateits role during strategic decision making in a multi-agent environ-ment. The results showed that the animal’s choice behavior during acompetitive game could be accounted for by a reinforcement learn-ing algorithm. Individual prefrontal neurons often modulated theiractivity according to the choice of the animal in the previous trial,the outcome of that choice, and the conjunction between the choiceand its outcome. This suggests that the PFC may be involved inupdating the animal’s decision-making strategy based on a rein-forcement learning algorithm.

RESULTSBehavioral performanceTwo rhesus monkeys played a game analogous to matching penniesagainst a computer in an oculomotor free-choice task (Fig. 1a;Methods). The animal was rewarded when it selected the same tar-get as the computer that was programmed to minimize the animal’sreward by exploiting the statistical bias in the animal’s choice behav-ior. Accordingly, the optimal strategy for the animal was to choosethe targets randomly with equal probabilities, which corresponds tothe Nash equilibrium in the matching pennies game. To determinehow the animal’s decisions were influenced by the strategy of theopponent, we manipulated the amount of information that wasused by the computer opponent (see Methods). In algorithm 0, thecomputer selected its targets randomly with equal probabilities,regardless of the animal’s choice patterns. In algorithm 1, the com-

Department of Brain and Cognitive Sciences, Center for Visual Science, University of Rochester, Rochester, New York 14627, USA. Correspondence should beaddressed to D.L. ([email protected]).


Prefrontal cortex and decision making in a mixed-strategy gameDominic J Barraclough, Michelle L Conroy & Daeyeol Lee

In a multi-agent environment, where the outcomes of one’s actions change dynamically because they are related to the behaviorof other beings, it becomes difficult to make an optimal decision about how to act. Although game theory provides normativesolutions for decision making in groups, how such decision-making strategies are altered by experience is poorly understood.These adaptive processes might resemble reinforcement learning algorithms, which provide a general framework for findingoptimal strategies in a dynamic environment. Here we investigated the role of prefrontal cortex (PFC) in dynamic decision makingin monkeys. As in reinforcement learning, the animal’s choice during a competitive game was biased by its choice and rewardhistory, as well as by the strategies of its opponent. Furthermore, neurons in the dorsolateral prefrontal cortex (DLPFC) encodedthe animal’s past decisions and payoffs, as well as the conjunction between the two, providing signals necessary to update theestimates of expected reward. Thus, PFC might have a key role in optimizing decision-making strategies.

A R T I C L E S


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


puter analyzed only the animal’s choice history, but not its rewardhistory. In algorithm 2, both choice and reward histories were ana-lyzed. In both algorithms 1 and 2, the computer chose its target ran-domly if it did not find any systematic bias in the animal’s choicebehavior. Therefore, a reward rate near 0.5 indicates that the ani-mal’s performance was optimal.

Indeed, the animal’s reward rate was close to 0.5 for all algorithms,indicating that the animal’s performance was nearly optimal. Inalgorithm 0, the reward rate was fixed at 0.5 regardless of the ani-mal’s behavior, and therefore there was no incentive for the animal tochoose the targets with equal probabilities. In fact, both animalschose the right-hand target more frequently (P = 0.70 and 0.90 and n = 5,327 and 1,669 trials, for the two animals, respectively) than the

left-hand target. For the remaining two algo-rithms, the probability of choosing the right-hand target was much closer to 0.5 (Fig. 1c),which corresponds to the Nash equilibriumof the matching pennies game. In addition,the probability of choosing a given targetwas relatively unaffected by the animal’schoice in the previous trial. For example, theprobability that the animal would select thesame target as in the previous trial was alsoclose to 0.5 (P = 0.51 ± 0.06 and 0.50 ± 0.04and n = 120,254 and 74,113 trials, for algo-rithms 1 and 2, respectively). In contrast, the

animal’s choice was strongly influenced by the computer’s choice inthe previous trial, especially in algorithm 1. In the game of matchingpennies, the strategy to choose the same target selected by the oppo-nent in the previous trial can be referred to as a win-stay-lose-switch(WSLS) strategy, as this is equivalent to choosing the same target asin the previous trial if that choice was rewarded and choosing theopposite target otherwise. The probability of the WSLS strategy inalgorithm 1 (0.73 ± 0.14) was significantly higher than that in algo-rithm 2 (0.53 ± 0.06; P < 10−16; Fig. 1d). Although the tendency forthe WSLS strategy in algorithm 2 was only slightly above chance, thisbias was still statistically significant (P < 10−5). Similarly, averagemutual information between the sequence of animal’s choice andreward in three successive trials and the animal’s choice in the fol-lowing trial decreased from 0.245 (± 0.205) bits for algorithm 1 to0.043 (± 0.035) bits for algorithm 2.

Reinforcement learning modelUsing a reinforcement learning model19–21,28,29, we tested whetherthe animal’s decision was systematically influenced by the cumulativeeffects of reward history. In this model, a decision was based on thedifference between the value functions (that is, expected reward) forthe two targets. Denoting the value functions of the two targets (L andR) at trial t as Vt(L) and Vt(R), the probability of choosing each targetis given by the logit transformation of the difference between thevalue functions30. In other words,

logit P(R) ≡ log P(R)/(1 − P(R)) = Vt(R) − Vt(L).

Figure 1 Task and behavioral performance. (a) Free-choice task and payoff matrix for theanimal during the competitive game (1, reward;0, no reward). (b) Recording sites in PFC. Frontaleye field (gray area in the inset) was defined byelectrical stimulation50. (c) Frequencyhistograms for the probability of choosing theright-hand target in algorithms 1 and 2. (d) Probability of the win-stay-lose-switch(WSLS) strategy (abscissa) versus probability ofreward (ordinate). (e) Difference in the valuefunctions for the two targets estimated from areinforcement learning model (abscissa) versusthe probability of choosing the target at the right(ordinate). Error bars indicate standard error ofthe mean (s.e.m.). Histograms show thefrequency of trials versus the difference in thevalue functions. Solid line, prediction of thereinforcement learning model. In all panels, darkand light symbols indicate the results from thetwo animals, respectively.

Table 1 Parameters for the reinforcement learning model.

Algorithm Monkey α ∆1 ∆2

1 C 0.176 0.661 −0.554

(0.130, 0.220) (0.619, 0.704) (−0.597, −0.512)

E 0.170 0.941 −1.064

(0.143, 0.198) (0.903, 0.979) (−1.104, −1.024)

2 C 0.986 0.033 0.016

(0.983, 0.988) (0.028, 0.039) (0.012, 0.021)

E 0.828 0.195 −0.143

(0.801, 0.851) (0.171, 0.218) (−0.169, −0.118)

α, discount factor; ∆1 and ∆2, changes in the value function associated with rewardedand unrewarded targets selected by the animal, respectively. The numbers inparentheses indicate 99.9% confidence intervals.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

The value function, Vt(x), for target x, wasupdated after each trial according to the fol-lowing:

Vt+1(x) = αVt(x) + ∆t(x),

where α is a discount factor, and ∆t(x)denotes the change in the value functiondetermined by the animal’s decision and itsoutcome. In the current model, ∆t(x) = ∆1 ifthe animal selects the target x and isrewarded, ∆t(x) = ∆2 if the animal selects thetarget x and is not rewarded, and ∆t(x) = 0 ifthe animal does not select the target x. Weintroduced a separate parameter for theunrewarded target (∆2) because the proba-bility of choosing the same target after losinga reward was significantly different from theprobability of switching to the other targetfor all animals and for both algorithms 1 and2. Maximum likelihood estimates31 of themodel parameters (Table 1) showed that a frequent use of the WSLSstrategy during algorithm 1 was reflected in a relatively small dis-count factor (α < 0.2), a large positive ∆1 (> 0.6) and a large negative∆2 (< −0.5) in both animals. For algorithm 1, this led to a largelybimodal distribution for the difference in the value functions (Fig. 1e). In contrast, the magnitude of changes in value functionduring algorithm 2 was smaller, indicating that the outcome of pre-vious choices only weakly influenced the subsequent choice of theanimal. In addition, the discount factor for algorithm 2 was relativelylarge (α > 0.8). This suggests that the animal’s choice was systemati-cally influenced by the combined effects of previous reward historyeven in algorithm 2. The combination of model parameters for algo-rithm 2 produced an approximately normal distribution for the dif-

ference in value functions (Fig. 1e). This implies that for most trials,the difference in the value functions of the two targets was relativelysmall, making it difficult to predict the animal’s choice reliably. Theseresults suggest that during a competitive game, the monkeys mighthave approximated the optimal decision-making strategy using areinforcement learning algorithm.

Prefrontal activity during a competitive gameThe value functions in the above reinforcement learning model wereupdated according to the animal’s decisions and the outcomes ofthose decisions. To determine whether such signals are encoded in theactivity of individual neurons in PFC, we recorded single-neuronactivity in the DLPFC while the animal played the same free-choice

task. During the neurophysiological record-ing, the computer selected its target accord-ing to algorithm 2. A total of 132 neuronswere examined during a minimum of 128free-choice trials (mean = 583 trials; Fig. 1b).As a control, each neuron was also examinedduring 128 trials of a visual search task in

A R T I C L E S


Figure 2 Effects of relative expected reward (i.e., difference in value functions) and its trial-to-trialchanges on the activity of prefrontal neurons. (a) Percentage of neurons with significant correlation (t-test, P < 0.05) between their activity and the difference in the value functions for the two targets(VD = Vτ(R) − Vτ(L)) estimated for the current (τ = t) and previous (τ = t − 1) trials, or between theactivity and the changes in the value functions between the two successive trials (VD change = ∆t–1(R) − ∆t–1(L)). (b) Correlation coefficient between the VD change and the activity in a given neuron(ordinate), plotted against correlation coefficient between the VD in the previous trial (i.e., Vt−1(R) − Vt−1(L)) and the activity of the same neuron (abscissa). Black (gray) symbols indicate theneurons in which both (either) correlation coefficients were significantly different from 0 (t-test, P < 0.05). The numbers in each panel correspond to Spearman’s rank correlation coefficient (r) andits level of significance (P).

Figure 3 Percentages of neurons encodingsignals related to the animal’s decision. Whitebars show the percentage of neurons (n = 132)with significant main and interaction effects in athree-way ANOVA (P × R × C). Light gray barsshow the same information for the neurons with>256 free-choice trials, which was tested forstationarity in free-choice trials (n = 112). Darkgray bars show the percentage of neurons withsignificant effects in the three-way ANOVA thatalso varied with the task (search vs. choice) in afour-way ANOVA (Task × P × R × C). This analysiswas performed only for the neurons with >256free-choice trials for comparison with the controlanalysis to test stationarity. Black histogramsshow the percentage of neurons with significanteffects in the three-way ANOVA that also havesignificant non-stationarity in a control 4-wayANOVA across the two successive blocks of 128free-choice trials.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

the previous trial exerted a significant influ-ence on the activity before and during thefore-period, as well as during the delayperiod (3-way ANOVA, P < 0.001). In addi-tion, activity during the movement periodwas still influenced by the animal’s choice inthe previous trial and its outcome, as well asby their interactions with the animal’s choicein the current trial. To determine whetherany of these effects could be attributed to sys-tematic variability in eye movements, theabove analysis was repeated using the residu-als from a regression model in which the neu-ral activity related to a set of eye movementparameters was factored out (Methods). Theresults were nearly identical, with the onlydifference found in the loss of significancefor the effect of the current choice. Duringthe fore-period, 35% of neurons showed asignificant effect of the animal’s choice in theprevious trial on the residuals from the sameregression model.

It is possible that the animal’s choice in theprevious trial influenced the activity of thisneuron during the next trial through system-atic changes in unidentified sensorimotor

events, such as licking or eye movements during the inter-trial inter-val, that were not experimentally controlled. This was tested by com-paring the activity of the same neuron in the search and free-choicetrials. For the neuron shown in Figure 4, activity during search trialswas significantly affected by the position of the target in the previoustrial only during the fore-period, and this effect was opposite to andsignificantly different from that found in the free-choice trials (4-wayANOVA, P < 10−5). The raster plots show that this change occurredwithin a few trials after the animal switched from search to free-choicetrials (Fig. 4). These results suggest that the effect of the animal’schoice in the previous trial on the activity of this neuron did notmerely reflect nonspecific sensorimotor events. In 17% of the neuronsthat showed a significant effect of the animal’s previous choice duringthe fore-period, there was also a significant interaction between thetask type (search vs. free-choice) and the animal’s choice in the previ-ous trial (Fig. 3). This indicates that signals related to the animal’s pastchoice were actively maintained in the PFC according to the type ofdecision. It is unlikely that this was entirely due to an ongoing drift inthe background activity (i.e., non-stationarity), as the control analysisperformed on two successive blocks of free-choice trials did not pro-duce a single case with the same effect during the fore-period (Fig. 3).

During the fore-period, 39% of neurons showed a significant effectof the reward in the previous trial. For example, the activity of the neu-

which the animal’s decision was guided by sensory stimuli (Methods).During the free-choice trials, activity in some prefrontal neurons

was influenced by the difference in the value functions for the two tar-gets (that is, V(R) − V(L)), although the effects in individual neuronswere relatively small (Fig. 2). This was not entirely due to the animal’schoice and its outcome in the previous trial, as the value functionsestimated for the previous trial produced similar results (Fig. 2a). Ifindividual PFC neurons are involved in the temporal integration ofvalue functions according to the reinforcement learning modeldescribed above, differences in the value functions (i.e., V(x)) andtheir changes (i.e., ∆(x)) would similarly influence the activity of PFCneurons. Interestingly, such patterns were found for the delay andmovement periods, but not for the fore-period (Methods; Fig. 2b).These results suggest that some prefrontal neurons might be involvedin temporally integrating the signals related to previous choice and itsoutcome to update value functions.

To examine how the activity of individual PFC neurons is influ-enced by the animal’s choices and their outcomes, we analyzed neuralactivity by three-way ANOVA with the animal’s choice and reward inthe previous trial and its choice in the current trial as main factors.For 39% of PFC neurons, the activity during the fore-period wasinfluenced by the animal’s choice in the previous trial (Fig. 3). Forexample, in the neuron illustrated in Figure 4, the animal’s choice in

A R T I C L E S


Figure 4 Example neuron showing a significanteffect of the animal’s choice in the previous trial.Top, spike density functions averaged accordingto the animal’s choice (L and R) and reward (+, reward; −, no reward) in the previous trial andthe choice in the current trial. A dotted verticalline indicates the onset of the fore-period, and thetwo solid lines the beginning and end of the delayperiod. Bottom, raster plots showing the activity ofthe same neuron sorted according to the samethree factors during the search (gray background)and free-choice (white background) trials.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

ron in Figure 5 was higher throughout theentire trial after the animal was not rewardedin the previous trial, compared to when theanimal was rewarded. This effect was nearlyunchanged when we removed the eye-move-ment related activity in a regression analysis,both in this single-neuron example and forthe population as a whole. Overall, 37% ofneurons showed the effect of the previousreward when the analysis was performed onthe residuals from the regression model. Thepossibility that this effect was entirely due touncontrolled sensorimotor events is alsounlikely, as a substantial proportion of theseneurons (21%) also showed a significant interaction between the tasktype and the previous reward during the fore-period (Fig. 3).

To update the value functions in a reinforcement learning model,signals related to the animal’s choice and its outcome must be com-bined, because each variable alone does not specify how the valuefunction of a particular target should be changed. Similarly, activity ofthe neurons in the PFC was often influenced by the conjunction ofthese two variables. In the neuron in Figure 6, for example, there was agradual buildup of activity during the fore-period, but this occurred

only when the animal had selected the right-hand target in the previ-ous trial, and this choice was not rewarded. During the delay period,the activity of this neuron diverged to reflect the animal’s choice in thecurrent trial (Fig. 6, arrow). The same neuron showed markedlyweaker activity during the search trials, suggesting that informationcoded in the activity of this neuron regarding the outcome of choosinga particular target was actively maintained in free-choice trials (Fig. 6).For the fore-period, 20% of the neurons showed significant interac-tion between the animal’s choice and its outcome in the previous trial

(P < 0.05; Fig. 3). Activity related to eye move-ments was not an important factor: 90% ofthese neurons showed the same effect in theresiduals from the regression analysis that fac-tored out the effects of eye movements.Furthermore, during the fore-period, 27% ofthe same neurons showed significant three-way interactions among task type, animal’schoice in the previous trial and outcome ofthe previous trial. In contrast, the controlanalysis during the first two blocks of the free-choice task revealed such an effect only in 5%of the neurons (Fig. 3). These results indicatethat signals related to the conjunction of theanimal’s previous decision and its outcomeare processed differently in the PFC accordingto the type of decisions made by the animal.

DISCUSSIONInteraction with other intelligent beings isfundamentally different from—and morecomplex than—dealing with inanimateobjects32,33. Interactions with other animalsare complicated by the fact that their behav-ioral strategies often change as a result ofone’s own behavior. Therefore, the analysis

A R T I C L E S


Figure 5 Example neuron showing a significanteffect of the reward in the previous trial. Sameformat as in Figure 4.

Figure 6 Example neuron with a significantinteraction between the animal’s choice and itsoutcome in the previous trial. Same format as inFigure 4. Arrows indicate the time when theanimal’s choice in the current trial was firstreflected in the neural activity.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

of decision making in a group requires a more sophisticated analy-tical framework, which is provided by game theory. Matching pen-nies is a relatively simple zero-sum game that involves two playersand two alternative choices. The present study examined the behav-ior of monkeys playing a competitive game similar to matchingpennies against a computer opponent. It is not known whethermonkeys treated this game as a competitive situation with anotherintentional being. Nevertheless, the same formal framework ofgame theory is applicable to the task used in this study, and as pre-dicted, the animal’s behavior was influenced by the opponent’sstrategy. When the computer blindly played the equilibrium strat-egy regardless of the animal’s behavior, the animals selected one ofthe targets more frequently. In contrast, when the computer oppo-nent began exploiting biases in the animal’s choice sequence,the animal’s behavior approached the equilibrium strategy.Furthermore, when the computer did not examine the animal’sreward history (algorithm 1), the animals achieved a nearly optimalreward rate by adopting the win-stay-lose-switch (WSLS) strategy.This was possible because this strategy was not detected by thecomputer. Finally, the frequency of the WSLS strategy was reducedwhen the computer began exploiting biases in the animal’s choiceand reward sequences (algorithm 2).

These results also suggest that the animals approximated the opti-mal strategy using a reinforcement learning algorithm. This modelassumes that the animals base their decisions, in part, on the estimatesof expected rewards for the two targets and tend to select the targetwith larger expected reward. During zero-sum games such as match-ing pennies, strategies of the players behaving according to some rein-forcement learning algorithms would gradually converge on a set ofequilibrium strategies5–7. However, it is important to update the valuefunctions of different targets by a small amount after each play whenplaying against a fully informed rational player (such as algorithm 2in the present study). This is because large, predictable changes in thevalue functions would reveal one’s next choice to the opponent. In thepresent study, the magnitude of changes in the value function variedaccording to the strategy of the opponent and was adjusted throughthe animal’s experience.

Finally, neurophysiological recordings in the PFC revealed apotential neural basis for updating the value functions adaptivelywhile interacting with a rational opponent. Reward-related activity iswidespread in the brain34–38. In particular, signals related to expectedreward (i.e., value functions) are present in various brain areas39–43,including the DLPFC44–48. Our results showed that neurons in theDLPFC also code signals related to the animal’s choice in the previ-ous trial. Such signals might be actively maintained and processeddifferently in the DLPFC according to the type of informationrequired for the animal’s upcoming decisions. Furthermore, signalsrelated to the animal’s past choices and their outcomes are combinedat the level of individual PFC neurons. These signals might then betemporally integrated according to a reinforcement learning algo-rithm to update the value functions for alternative actions. Manyneurons in the PFC show persistent activity during a working mem-ory task, and the same circuitry might be ideally suited for temporalintegration of signals related to the animal’s previous choice and itsoutcome49. Although the present study examined the animal’s choicebehavior in a competitive game, reinforcement learning algorithmscan converge on optimal solutions for a wide range of decision-mak-ing problems in dynamic environments. Therefore, the results fromthe present study suggest that the PFC has an important role in opti-mizing decision-making strategies in a dynamic environment thatmay include multiple agents.

METHODSAnimal preparations. Two male rhesus monkeys were used. Their eye move-ments were monitored at a sampling rate of 250 Hz with either a scleral eyecoil or a high-speed video-based eye tracker (ET49, Thomas Recording). Allthe procedures used in this study conformed to National Institutes of Healthguidelines and were approved by the University of Rochester Committee onAnimal Research.

Behavioral task. Monkeys were trained to play a competitive game analogous tomatching pennies against a computer in an oculomotor free-choice task (Fig. 1a). During a 0.5-s fore-period, they fixated a small yellow square (0.9 ×0.9°; CIE x = 0.432, y = 0.494, Y = 62.9 cd/m2) in the center of a computerscreen, and then two identical green disks (radius = 0.6°; CIE x = 0.286,y = 0.606, Y = 43.2 cd/m2) were presented 5° away in diametrically opposedlocations. The central target disappeared after a 0.5-s delay period, and the ani-mal was required to shift its gaze to one of the targets. At the end of a 0.5-s holdperiod, a red ring (radius = 1°; CIE x = 0.632, y = 0.341, Y = 17.6 cd/m2)appeared around the target selected by the computer, and the animal main-tained its fixation for another 0.2 s. The animal was rewarded at the end of thissecond hold period, but only if it selected the same target as the computer. Thecomputer had been programmed to exploit certain biases displayed by the ani-mal in making its choices. Each neuron was also tested in a visual search task.This task was identical to the free-choice task, except that one of the targets inthe free-choice task was replaced by a distractor (red disk). The animal wasrequired to shift its gaze toward the remaining target (green disk), and this wasrewarded randomly with 50% probability. This made it possible to examine theeffect of reward on the neural activity. The location of the target was selectedfrom the two alternative locations pseudo-randomly for each search trial.

Algorithms for computer opponent. During the free-choice task, the computerselected its target according to one of three different algorithms. In algorithm 0,the computer selected the two targets randomly with equal probabilities, whichcorresponds to the Nash equilibrium in the matching pennies game. In algo-rithm 1, the computer exploited any systematic bias in the animal’s choicesequence to minimize the animal’s reward rate. The computer saved the entirehistory of the animal’s choices in a given session, and used this information topredict the animal’s next choice by testing a set of hypotheses. First, the condi-tional probabilities of choosing each target given the animal’s choices in thepreceding n trials (n = 0 to 4) were estimated. Next, each of these conditionalprobabilities was tested against the hypothesis that the animal had chosen bothtargets with equal probabilities. When none of these hypotheses was rejected,the computer selected each target randomly with 50% probability, as in algo-rithm 0. Otherwise, the computer biased its selection according to the probabil-ity with the largest deviation from 0.5 that was statistically significant (binomialtest, P < 0.05). For example, if the animal chose the right-hand target with 80%probability, the computer selected the left-hand target with the same probabil-ity. Therefore, to maximize reward, animals needed to choose both targets withequal frequency and select a target on each trial independently from previouschoices. In algorithm 2, the computer exploited any systematic bias in the ani-mal’s choice and reward sequences. In addition to the hypotheses tested in algo-rithm 1, algorithm 2 also tested the hypothesis that the animal’s decisions wereindependent of prior choices and their payoffs in the preceding n trials (n = 1 to4). Thus, to maximize total reward in algorithm 2, it was necessary for the ani-mal to choose both targets with equal frequency and to make choices independ-ently from previous choices and payoffs.

Neurophysiological recording. Single-unit activity was recorded from theneurons in the DLPFC of two monkeys using a five-channel multi-electroderecording system (Thomas Recording). The placement of the recording cham-ber was guided by magnetic resonance images, and this was confirmed in oneanimal by metal pins inserted in known anatomical locations. In addition, thefrontal eye field (FEF) was defined in both animals as the area in which sac-cades were evoked by electrical stimulations with currents <50 µA (ref. 50). Allthe neurons described in the present study were anterior to the FEF.

Analysis of behavioral data. The frequency of a behavioral event (e.g., reward)was examined with the corresponding probability averaged across recording

A R T I C L E S


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


sessions and its standard deviation. The values of mutual information werecorrected for the finite sample size. Null hypotheses in the analysis of behav-ioral data were tested using a binomial test or a t-test (P < 0.05). Parameters inthe reinforcement learning model were estimated by a maximum likelihoodprocedure, using a function minimization algorithm in Matlab (MathworksInc.), and confidence intervals were estimated by profile likelihood intervals31.

Analysis of neural data. Spikes during a series of 500-ms bins were countedseparately for each trial. The effects of the animal’s choice (P) and reward (R)in the previous trial and the choice in the current trial (C) were analyzed in a3-way ANOVA (P × R × C). The effect of the task (search versus free-choice)was analyzed in a four-way ANOVA (Task × P × R × C). As a control analysis todetermine whether the task effect was due to non-stationarity in neural activ-ity, the same four-way ANOVA was performed for the first two successiveblocks of 128 trials in the free-choice task (Fig. 3). To determine whether eyemovements were confounding factors, the above analysis was repeated usingthe residuals from the following regression model:

S = a0 + a1 Xpre80 + a2 Ypre80 + a3 XFP + a4 YFP + a5 XSV + a6 YSV

+ a7 SRT + a8 PV + ε

where S indicates the spike count, Xpre80 (Ypre80) the horizontal (vertical) eyeposition 80 ms before the onset of central fixation target, XFP (YFP) the averagehorizontal (vertical) eye position during the fore-period, XSV (YSV) the hori-zontal (vertical) component of the saccade directed to the target, SRT and PVthe saccadic reaction time and the peak velocity of the saccade, and ε the errorterm.

ACKNOWLEDGMENTSWe thank L. Carr, R. Farrell, B. McGreevy and T. Twietmeyer for their technicalassistance, J. Swan-Stone for programming, X.-J. Wang for discussions, and B. Averbeck and J. Malpeli for critically reading the manuscript. This work wassupported by the James S. McDonnell Foundation and the National Institutes ofHealth (NS44270 and EY01319).



1. Von Neumann J. & Morgenstern O. Theory of Games and Economic Behavior(Princeton Univ. Press, New Jersey, 1944).

2. Fudenberg, D. & Tirole, J. Game Theory (MIT Press, Cambridge, Massachusetts,1991).

3. Dixit, A. & Skeath, S. Games of Strategy (Norton, New York, 1999).4. Glimcher, P.W. Decisions, Uncertainty, and the Brain (MIT Press, Cambridge,

Massachusetts, 2003).5. Nash, J.F. Equilibrium points in n-person games. Proc. Natl. Acad. Sci. USA 36,

48–49 (1950).6. Robinson, J. An iterative method of solving a game. Ann. Math. 54, 296–301

(1951).7. Binmore, K., Swierzbinski, J. & Proulx, C. Does minimax work? An experimental

study. Econ. J. 111, 445–464 (2001).8. Touhey, J.C. Decision processes, expectations, and adoption of strategies in zero-

sum games. Hum. Relat. 27, 813–824 (1974).9. O’Neill, B. Nonmetric test of the minimax theory of two-person zerosum games.

Proc. Natl. Acad. Sci. USA 84, 2106–2109 (1987).10. Brown, J.N. & Rosenthal, R.W. Testing the minimax hypothesis: a re-examination of

O’Neill’s game experiment. Econometrica 58, 1065–1085 (1990).11. Rapoport, A. & Boebel, R.B. Mixed strategies in strictly competitive games: a further

test of the minimax hypothesis. Games Econ. Behav. 4, 261–283 (1992).12. Rapoport, A. & Budescu, D.V. Generation of random series in two-person strictly

competitive games. J. Exp. Psychol. Gen. 121, 352–363 (1992).13. Budescu, D.V. & Rapoport, A. Subjective randomization in one- and two-person

games. J. Behav. Dec. Making 7, 261–278 (1994).14. Ochs, J. Games with unique, mixed strategy equilibria: an experimental study.

Games Econ. Behav. 10, 202–217 (1995).

15. Sarin, R. & Vahid, F. Predicting how people play games: a simple dynamic model ofchoice. Games Econ. Behav. 34, 104–122 (2001).

16. Shachat, J.M. Mixed strategy play and the minimax hypothesis. J. Econ. Theory104, 189–226 (2002).

17. Walker, M. & Wooders, J. Minimax play at Wimbledon. Am. Econ. Rev. 91,1521–1538 (2001).

18. Fudenberg, D. & Levine, D.K. Theory of Learning in Games (MIT Press, Cambridge,Massachusetts, 1998).

19. Mookherjee, D. & Sopher, B. Learning behavior in an experimental matching pen-nies game. Games Econ. Behav. 7, 62–91 (1994).

20. Erev, I. & Roth, A.E. Predicting how people play games: reinforcement learning inexperimental games with unique, mixed strategy equilibria. Am. Econ. Rev. 88,848–881 (1998).

21. Camerer, C.F. Behavioral Game Theory (Princeton Univ. Press, Princeton, NewJersey, 2003).

22. Simon, H.A. Models of Man (Wiley, New York, 1957).23. Kahneman, D., Slovic, P. & Tversky, A. Judgement Under Uncertainty: Heuristics

and Biases (Cambridge Univ. Press, Cambridge, UK, 1982).24. O’Neill, B. Comments on Brown and Rosenthal’s reexamination. Econometrica 59,

503–507 (1991).25. Rapoport, A. & Budescu, D.V. Randomization in individual choice behavior. Psychol.

Rev. 104, 603–617 (1997).26. Chen, H-C., Friedman, J.W. & Thisse, J-F. Boundedly rational Nash equilibrium: a

probabilistic choice approach. Games Econ. Behav. 18, 32–54 (1997).27. Rubinstein, A. Modeling Bounded Rationality (MIT Press, Cambridge,

Massachusetts, 1998).28. Sutton, R.S. & Barto, A.G. Reinforcement Learning: an Introduction (MIT Press,

Cambridge, Massachusetts, 1998).29. Littman, M.L. Markov games as a framework for multi-agent reinforcement learning.

Machine Learning: Proc. 11th Int. Conf. pp. 157–163 (Morgan Kaufmann, SanFrancisco, California, 1994)

30. Christensen, R. Log-linear Models and Logistic Regression edn. 2 (Springer-Verlag,New York, 1997).

31. Burnham, K.P. & Anderson, D.R. Model Selection and Multimodel Inference 2ndedn. (Springer-Verlag, New York, 2002).

32. Byrne, R.W. & Whiten, A. Machiavellian Intelligence: Social Expertise and the Evolutionof Intellect in Monkeys, Apes and Humans (Oxford Univ. Press, Oxford, UK, 1988).

33. Whiten, A. & Byrne, R.W. Machiavellian Intelligence II: Extensions and Evaluations(Cambridge Univ. Press, Cambridge, UK, 1997).

34. Schultz, W. Predictive reward signal of dopamine neurons. J. Neurophysiol. 80,1–27 (1998).

35. Rolls, E.T. Brain and Emotion (Oxford Univ. Press, Oxford, UK, 1999).36. Amador, N., Schlag-Rey, M. & Schlag, J. Reward-predicting and reward detecting

neuronal activity in the primate supplementary eye field. J. Neurophysiol. 84,2166–2170 (2000).

37. Stuphorn, V., Taylor, T.L. & Schall, J.D. Performance monitoring by the supplemen-tary eye field. Nature 408, 857–860 (2000).

38. Ito, S., Stuphorn, V., Brown, J.W. & Schall, J.D. Performance monitoring by the ante-rior cingulate cortex during saccade countermanding. Science 302, 120–122 (2003).

39. Kawagoe, R., Takikawa, Y. & Hikosaka, O. Expectation of reward modulates cognitivesignals in the basal ganglia. Nat. Neurosci. 1, 411–416 (1998).

40. Platt, M.L. & Glimcher, P.W. Neural correlates of decision variables in parietal cor-tex. Nature 400, 233–238 (1999).

41. Shidara, M. & Richmond, B.J. Anterior cingulate: single neuronal signals related todegree of reward expectancy. Science 296, 1709–1711 (2002).

42. Ikeda, T. & Hikosaka, O. Reward-dependent gain and bias of visual responses in pri-mate superior colliculus. Neuron 39, 693–700 (2003).

43. McCoy, A.N., Crowley, J.C., Haghighian, G., Dean, H.L. & Platt, M.L. Saccadereward signals in posterior cingulate cortex. Neuron 40, 1031–1040 (2003).

44. Watanabe, M. Reward expectancy in primate prefrontal cortex. Nature 382,629–632 (1996).

45. Leon, M.I. & Shadlen, M.N. Effect of expected reward magnitude on the response ofneurons in the dorsolateral prefrontal cortex of the macaque. Neuron 24, 415–425(1999).

46. Kobayashi, S., Lauwereyns, J., Koizumi, M., Sakagami, M. & Hikosaka, O. Influenceof reward expectation on visuospatial processing in macaque lateral prefrontal cor-tex. J. Neurophysiol. 87, 1488–1498 (2002).

47. Roesch, M.R. & Olson, C.R. Impact of expected reward on neuronal activity in pre-frontal cortex, frontal and supplementary eye fields and premotor cortex. J. Neurophysiol. 90, 1766–1789 (2003).

48. Tsujimoto, S. & Sawaguchi, T. Neuronal representation of response outcome in theprimate prefrontal cortex. Cereb. Cortex 14, 47–55 (2004).

49. Wang, X.-J. Probabilistic decision making by slow reverberation in cortical circuits.Neuron 36, 955–968 (2002).

50. Bruce, C.J., Goldberg, M.E., Bushnell, M.C. & Stanton, G.B. Primate frontal eyefields. II. Physiological and anatomical correlates of electrically evoked eye move-ments. J. Neurophysiol. 54, 714–734 (1985).

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

Functional neuroimaging studies have identified a growing numberof sex differences in human brain function. In addition to cognitivedifferences1–3, men and women also differ markedly in aspects of sex-ual behavior, such as the reportedly greater male interest in andresponse to sexually arousing visual stimuli4–6. Animal studies haveidentified several sex differences in limbic brain regions that mediatereproductive behavior, which may provide clues to brain regionsunderlying sex differences in human sexual response. In rats, forexample, male but not female appetitive responses to distal olfactoryand visual sexual signals are critically mediated by the medial amyg-dala7. Lesions to the medial amygdala in male but not female rats dis-rupt appetitive, sexual behaviors involved in gaining access to areceptive mate, but these lesions leave consummatory, copulation-related behaviors intact7. In addition, male and female rats’ reproduc-tive functions are controlled by different hypothalamic regions7.Although the amygdala and hypothalamus have also been linked tomale responses to sexually arousing stimuli in neuroimaging stud-ies8,9 and these structures are considerably influenced by sex hor-mones10,11, it remains unclear whether sex differences in the functionof these regions also exist in humans.

Here we examined human sex differences in reactions to visualsexual stimuli using fMRI, contrasting neural responses of healthymen and women to sexually arousing photographs and controlstimuli. To permit sex differences in neural responses to be charac-terized while controlling for possible differences in brain responserelated to typically higher arousal levels for males, we selected stim-uli through prior testing that yielded equivalent sexual attractive-ness and physical arousal ratings from both sexes. On the basis ofconverging evidence from earlier studies7–11, we were particularlyinterested in whether males would show greater activation in theamygdala and hypothalamus.

Twenty-eight young adults (14 female) passively viewed alternatingshort blocks of four types of photographic stimuli via video goggles:two types of sexually arousing stimuli, including heterosexual couplesengaged in sexual activity (‘couples’ stimuli) and sexually attractiveopposite-sex nudes (‘opposite-sex’ stimuli), and two types of controlstimuli, including pleasant scenes depicting non-sexual male-femaleinteraction, such as therapeutic massage (neutral stimuli), or a fixa-tion cross (fixation). Subjects were screened to verify that they wereheterosexual and found visual erotica sexually arousing. Each blockcontained five stimuli of the corresponding type. Two runs were pre-sented, each containing four blocks of each type, and ratings of sexualattractiveness and physical arousal were assessed after scanning. Wefound that the amygdala and hypothalamus were more activated inmen than in women when viewing identical sexual stimuli, even whenfemales reported greater arousal.

RESULTSFemales and males rated the sexual stimuli as equivalently sexuallyattractive and physically arousing, and both sexes reported the cou-ples stimuli as more attractive and arousing than the opposite-sexstimuli (Fig. 1). A two-factor analysis of variance (ANOVA; sex ×stimulus type) conducted separately for attractiveness ratings andphysical arousal ratings showed a main effect of stimulus type, withcouples stimuli rated higher in attractiveness (F1,26 = 8.08, P < 0.01)and physical arousal (F1,26 = 17.88, P < 0.001) than opposite-sexstimuli. However, there was no difference between females andmales in overall ratings for either attractiveness (F1,26 = 1.21, P >0.28) or physical arousal (F1,26 = 1.01, P > 0.32), and there was nointeraction between sex and stimulus type for either attractiveness(F1,26 = 1.06, P > 0.31) or physical arousal (F1,26 = 2.04, P > 0.17).Thus, although the females seemed to show a somewhat larger dif-

1Department of Psychology, 532 North Kilgo Circle, Emory University, Atlanta, Georgia 30322, USA. 2Yerkes National Primate Research Center, 954 Gatewood Road,Emory University, Atlanta, Georgia 30322, USA. Correspondence should be addressed to S.H. ([email protected]).


Men and women differ in amygdala response to visualsexual stimuliStephan Hamann1, Rebecca A Herman1, Carla L Nolan1 & Kim Wallen1,2

Men are generally more interested in and responsive to visual sexually arousing stimuli than are women. Here we used functionalmagnetic resonance imaging (fMRI) to show that the amygdala and hypothalamus are more strongly activated in men than inwomen when viewing identical sexual stimuli. This was true even when women reported greater arousal. Sex differences werespecific to the sexual nature of the stimuli, were restricted primarily to limbic regions, and were larger in the left amygdala thanthe right amygdala. Men and women showed similar activation patterns across multiple brain regions, including ventral striatalregions involved in reward. Our findings indicate that the amygdala mediates sex differences in responsiveness to appetitive andbiologically salient stimuli; the human amygdala may also mediate the reportedly greater role of visual stimuli in male sexualbehavior, paralleling prior animal findings.


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

ference between their ratings for opposite-sex and couples stimuli(Fig. 1), this difference was not significantly greater than the differ-ence shown by the males.

Following spatial pre-processing of the functional images, activa-tion contrasts between conditions were estimated for each subject ateach voxel using linear regression, producing statistical parametric t-statistic maps12. Sex differences in activation were assessed usingsecond-level, mixed-effects t-tests. We focused on responses to thecouples stimuli because these stimuli elicited the highest arousal andallowed us to directly compare female and male responses to identi-cal stimuli. The entire brain was examined for regions where differential activity surpassed a statistical threshold of P < 0.001(uncorrected for multiple comparisons) and spanned a minimum offive contiguous 64-mm3 voxels.

Men had greater neural responses in the bilateral amygdala andhypothalamus than did women to the couples stimuli (P < 0.001;Fig. 2a). Sex differences were restricted to these regions, with theexception of the right cerebellum (Fig. 2a) and right posterior thala-mus (data not shown; P < 0.001). Within the a priori regions of inter-est (ROIs), the differential activations also survived a more stringentstatistical correction for multiple spatial comparisons: left amygdala,P < 0.001, corrected; maxima at –20, –4, –20 (x, y, z in MNI space, seeMethods; Z = 3.95) and –16, 0, –16 (Z = 3.77); right amygdala, twoclusters, P < 0.05, corrected; maxima at 24, –4, –24 (Z = 3.43) and 16,0, –16 (Z = 3.32); hypothalamus, P < 0.001, corrected; maximum at 4,0, –16 (Z = 3.58). Notably, in no region did females show significantlygreater activation than males at this statistical threshold. These sexdifferences were also evident when the couples vs. fixation contrastwas examined separately in each group in these same regions (Fig. 2b,c). For males, the left and right amygdala and the hypothala-mus were significantly activated (left amygdala, P < 0.001, corrected;right amygdala, P < 0.001, corrected; hypothalamus, P < 0.01, cor-rected; one group t-test), whereas females showed no significant acti-vations in these regions.

To establish that the observed sex differences were related specifi-cally to the sexual aspects of the couples stimuli, we contrasted theresponses to the couples stimuli with responses to the more closelymatched neutral, non-sexual stimuli that depicted male-female inter-action. This contrast controlled for non-sexual attributes of the cou-

ples stimuli, including pleasant, non-sexual physical interactionbetween males and females. We further restricted this contrast tothose regions where sex differences were previously identified in thesex-differences analysis for the couples stimuli vs. fixation contrast(masked at P < 0.01). This served to isolate group differences thatresulted from greater increases for the sexual stimuli relative to theneutral stimuli as well as those that resulted from increases relative tothe fixation baseline. This contrast revealed more focal differentialactivations (men > women) at a lower statistical threshold (P < 0.005,at least five contiguous voxels) in left amygdala and right amygdala(Fig. 2d; the same contrast without masking identified similar butmore extensive regions of differential activation, see SupplementaryFig. 1 online), as well as in the hypothalamus, bilateral posterior thal-amus and left hippocampus (data not shown). As before, no regionswere observed in which females showed significantly greater activa-tion than males at the same threshold. The left amygdala differentialactivation (men > women) survived a more stringent correction formultiple spatial comparisons (left amygdala, P < 0.05, corrected;maximum at –16, 0, –20; Z = 3.23), the right amygdala activation wasmarginally significant (P = 0.06; maximum at 16, 0, –16; Z = 2.89),and the hypothalamic activation did not reach significance. Theabsence of differential hypothalamic activation in this latter analysisstemmed largely from low-level activation (at P < 0.05, uncorrected)detected in the hypothalamus for non-sexual stimuli, in men but notwomen, possibly related to a greater propensity for the males toappraise the nominally non-sexual scenes as weakly sexually appeti-tive. Sex differences were also evident when the couples stimuli versusneutral stimuli contrast was examined separately in each group inthese same regions (Fig. 2e,f). For males, the left and right amygdalaand the hypothalamus were significantly activated: left amygdala,P < 0.01, corrected; maximum at –16, 0, –24 (Z = 3.64); right amygdala,P < 0.05, corrected, maximum at –20, 0, –20 (Z = 3.31); hypothalamus,P < 0.05, corrected; maximum at –4, 0, –12 (Z = 3.66); in contrast,females showed no significant activations in these same regions.

To compare differences in fMRI signal change across all stimulusconditions and between brain regions, we calculated the averagefMRI signal change relative to the fixation baseline for each subjectfor ROIs centered on the left and right amygdala and the hypothala-mus. Males showed significantly greater activations in the left amyg-dala than did females for the couples stimuli (P < 0.001) andmarginally greater activations in the right amygdala (P = 0.11) andthe hypothalamus (P = 0.11; Fig. 3). The two sexes did not differ sig-nificantly in any ROI for the opposite-sex or neutral stimuli (seeSupplementary Table 1 online for a complete listing of all ROI con-trast statistics). The sex difference in responses to the couples stimuliwas marginally greater in magnitude (P = 0.11, F1,26 = 2.71) in theleft amygdala than the right amygdala, and was larger in spatialextent in the left amygdala (Fig. 2d).

Men showed marginally greater activation for the couples stimuliversus opposite-sex stimuli in the left amygdala (P = 0.10) but not inthe right amygdala or hypothalamus. Men also showed greater acti-vation for the couples stimuli relative to neutral stimuli in the leftamygdala (P < 0.01), right amygdala (P < 0.05) and hypothalamus(P < 0.01), but activations for opposite-sex stimuli relative to neu-


0

0.5

1

1.5

2

2.5

3

0

0.5

1

1.5

2

2.5

3Opposite-sex

Couples

Attr

activ

enes

s ra

ting

Phy

sica

l aro

usal

rat

ing

Women Men

Figure 1 Women’s (n = 14) and men’s (n = 14) ratings of visual stimuliaccording to attractiveness and physical arousal. Each subject rated 40couples stimuli and 40 opposite-sex stimuli. A Likert-type rating scale wasused: 0 (lowest) to 3 (highest). Top, immediate post-scan ratings of sexualattractiveness. Bottom, post-experiment ratings of experienced physicalarousal. Error bars indicate the standard error of the mean (s.e.m.).

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

tral stimuli did not reach significance for any ROI. Opposite-sexstimuli depicted isolated nudes, whereas couples stimuli depictedvaried, explicit sexual activity. Because cross-cultural studies13

report that males prefer sexual variety more than females do, greatermale habituation14,15 and lower arousal may have attenuated poten-tial sex differences for the opposite-sex stimuli. Women showed thereverse pattern from males in the left amygdala, with greater activa-tion for the less-arousing opposite-sex stimuli than for couplesstimuli (P < 0.05); no corresponding differences were observed inthe right amygdala or hypothalamus. Across all activation contrastswith the neutral stimulus condition, women showed greater activa-tion for the couples stimuli (marginally, at P = 0.07) only in thehypothalamus ROI, whereas men showed greater activation for thecouples stimuli vs. neutral stimuli in all ROIs. In summary, the pat-tern of results from the ROI analysis was consistent with the whole-brain analysis and revealed a left-sided lateralization of amygdalaresponse to sexual stimuli for men.

Because reported sexual attractiveness and experienced physicalarousal was equivalent in females and males, the greater activationsfor males were unlikely to be attributable to greater subjective arousal.Moreover, when one female subject in the current study who reportedlow arousal ratings for the couples stimuli was excluded from analy-sis, reported arousal was greater for females than males (P < 0.005),yet the activation differences favoring males remained unchanged.

In addition to characterizing sex differences, we also examined acti-vations that women and men shared in common by computing thestatistical conjunction between activation maps for the two groupsfor the couples stimuli versus neutral stimuli contrast (Fig. 4). Threeregions of overlap were observed: (i) a large, bilateral parieto-tempo-ral-occipital activation spanning regions associated with visual pro-cessing, attention, and motor and somatosensory function (Fig. 4a,b;P < 0.0001; corrected maxima at 36, –84, 12; 28, –52, 60; –28, –56, 52;Z = 9.21), (ii) the anterior cingulate, which is linked to emotion,

attention and sexual motivation9 (Fig. 4a; P < 0.001, corrected; maxi-mum at 0, 40, 8; Z = 5.91); and (iii) the nucleus accumbens/ventralstriatum (Fig. 4a; P < 0.01 corrected, maxima at 0, 16, –4 (Z = 5.37);–8, 24, 0 (Z = 5.57); 8, 20, –8 (Z = 5.32)). Because of the close associa-tion of the ventral striatum with reward processes16–18, coactivationin this region suggests that the sexual stimuli were rewarding to a sim-ilar degree for both groups, corroborating the subjective reports fromboth groups that they found sexual stimuli significantly rewarding.The coactivation and lack of sex differences in these broadly distrib-uted regions contrasts with the marked and regionally localized sexdifferences observed in the amygdala and hypothalamus. Thus, thesex differences we observed in the processing of visual sexual stimuli


–0.1

0

0.1

0.2

0.3

0.4

Couples O.S. Neutral Couples O.S. Neutral Couples O.S. Neutral

Females

Males

Left amygdala Right amygdala Hypothalamus

Per

cent

sig

nal c

hang

e

Figure 2 Regional activation maps. Activationcontrast for the couples stimuli versus fixation(a–c) and the couples stimuli versus neutralcontrast (d–f) (P < 0.005, minimum fivecontiguous voxels). White circles indicate theapproximate location of the ROIs; the left andright circles and upper and lower circles showthe left and right amygdala ROIs on the coronaland axial views, respectively. The medial circlesshow the hypothalamic ROI, which is not visibleon the axial views at z = –20. Color bar indicatesmaximal Z values. Note that color scale bars varyfrom image to image. The right hemisphere is onthe right of the coronal images and bottom of theaxial images. (a) Left, coronal image (y = 0)showing greater bilateral amygdala andhypothalamic activations for males versusfemales for the couples versus fixation contrast.Right, axial view (z = –16) of the same contrast,showing additional right cerebellar activation. (b) Couples versus fixation contrast for males, atthe same coronal and axial views. (c) The samecontrast and views for females. (d) Left, coronalimage (y = 0) showing greater bilateral amygdalaactivations for males versus females for thecouples versus neutral stimuli contrast, withinthose regions showing greater activity for malesversus females for the couples versus fixation contrast (at P < 0.10). The region of greater hypothalamic activation for males is not visible at this coronallevel. Right, axial view (z = –20) of the same contrast, showing primarily left-sided amygdala activation. (e) Couples versus neutral stimuli contrast formales, at the same coronal and axial views. (f) The same contrast and views for females, showing an absence of differential activity in the ROIs.

Figure 3 Average fMRI signal change for males and females for couples,opposite-sex and neutral stimuli (vs. fixation baseline), for ROIs in theleft amygdala, right amygdala and hypothalamus. Couples = couplesstimuli; O.S. = opposite-sex stimuli; Neutral = neutral stimuli. Error barsindicate s.e.m.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

occurred against a background of considerable similarity in the pro-cessing of such stimuli by men and women.

DISCUSSIONSex differences in activations to sexual stimuli could arise from differ-ences in processing mode between men and women (e.g., differentcognitive styles or neural pathways), from activations related tohigher arousal, irrespective of biological sex, or from a combinationof these factors1. By the arousal hypothesis, when men and women arematched on levels of elicited arousal, sex differences in brain activa-tion should be eliminated. In contrast, the processing-mode hypothe-sis predicts that men should still show greater brain activation thanwomen in specific regions after controlling for arousal. Our presentresults support the second hypothesis. In addition, the highly local-ized nature of the sex differences is more consistent with the process-ing-mode hypothesis. Previous studies contrasting brain responses toaffectively positive visual stimuli with those to less arousing stimuliconsistently report arousal-related activations distributed across mul-tiple regions19,20. This stands in marked contrast to the localized dif-ferences found here.

A previous neuroimaging study examined sex differences inresponses to sexual stimuli21, but did not observe sex differences inthe amygdala, possibly because its design rendered it less sensitive tofMRI signal changes in this structure (see Supplementary Noteonline). Arousal was also substantially higher for males than femalesin this earlier study. After controlling for arousal, the only sex differ-ence observed (in the hypothalamus) was eliminated21 (seeSupplementary Note online). In our present study, reported arousalwas equated for females and males. Moreover, when one female sub-

ject in the current study who reported low arousal ratings for the cou-ples stimuli was excluded from analysis, reported arousal was greaterfor females than males, yet the activation differences favoring malesremained unchanged. We note, however, that because sexual arousalhas multiple psychological and physiological aspects, further studywill be required to determine to what extent these other aspects maycontribute to observed sex differences22.

Strong positive correlations between emotional arousal and amyg-dala activity have been reported for both appetitive and aversive stim-uli23,24, and arousal has been suggested as the primary factorinfluencing amygdala activity in response to olfactory and visualstimulation. Here, however, amygdala responses to appetitive visualsexual stimuli were not solely determined by arousal, but instead werestrongly influenced by the sex of the viewer. The amygdala has multi-ple functions, however, and although processes related to emotionalarousal are clearly of prime importance, in specific contexts theseother roles may take precedence in determining amygdala activity. Forexample, considerable evidence from humans and other animalspoints to a critical role for the amygdala in appetitive incentive moti-vation, whereby the amygdala mediates the acquisition of high moti-vational value by stimuli, which in turn drives instrumentalbehavior17,18,25–27. In this context, the greater amygdala activation inmales observed here may in part reflect a greater appetitive incentivevalue of visual sexual stimuli, either intrinsic or learned, rather thangreater emotional arousal. This would be consistent with the greatermale motivation to seek out and interact with such stimuli. Notably, ithas recently been reported that larger amygdala size is related tohigher sexual drive in humans, further supporting a role of thehuman amygdala in sexual motivation28. It is also possible that sexualstimuli could represent a specific type of biologically salient stimulusthat is processed differently from other types of appetitive visualstimuli, and for which the relation between arousal and amygdalaactivation is more complex.

The amygdala has an established role in processing biologicallysalient appetitive and aversive stimuli, and initiating rapid adaptiveresponses via activation of other brain regions including the hypo-thalamus1,15,29–34. The current results extend understanding ofamygdala function by showing that the amygdala acts to mediate sexdifferences in responses to appetitive, emotionally positive stimuli.This accords with two reports that the amygdala mediates sex differ-ences in memory for emotional visual stimuli, each of which foundgreater left-sided activation related to subsequent emotional mem-ory in women but greater right-sided activation in males1,2 and sug-gests that the amygdala may be implicated in a variety of sexdifferences in emotion processing. Here, sex differences betweenmen and women were greater for the left amygdala than the rightamygdala, consistent with the predominantly left-sided amygdalaactivations elicited by pleasant and unpleasant visual stimuli in pre-vious reports1,19. The possible differential roles of the left and rightamygdala in emotion processing have been discussed extensively inthe context of aversive stimuli34–36. However, the differential roles ofthe left and right amygdala in processing appetitive emotional stim-uli, and in mediating sex differences, remain unclear, in part becausefew studies have examined these issues to date. A parallel with thedifferential roles of the amygdala in male appetitive versus consum-matory sexual responses highlighted in previous animal studies issuggested by a recent positron emission tomography (PET) study ofbrain activity in men during consummatory sexual behavior elicitedby tactile stimulation by a female partner37. Relative to a restingbaseline, consummatory male sexual behavior (erection and orgasm)elicited decreased activity in only one brain region, the amygdala,


Figure 4 Regions of significant overlap (conjunction) between groupactivations for males and females. The statistical conjunction between theactivation maps for males and females (P < 0.05 corrected, ≥10contiguous voxels) for the couples stimuli versus neutral, non-sexualstimuli. (a) Axial view (z = –4) showing common activation in ventralstriatum and occipital cortex. Color bar indicates maximal Z values. (b) Brain-surface rendered view of the same map, showing parieto-temporal-occipital and frontal activations spanning regions associated withvisual processing, attention, and motor and somatosensory function.Regions in red surpassed a P < 0.05 corrected threshold. The righthemisphere is on the bottom of the image.

©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

bilaterally during erection and in the left amygdala during orgasm.Thus, whereas viewing appetitive sexual stimuli by males in the cur-rent study elicited highly localized increases in amygdala activation,consummatory sexual behavior elicited correspondingly focal deac-tivations in the amygdala. Further investigation will be required todetermine whether such parallels indeed reflect a conservation ofamygdala function across species.

For males, more than for females, visual sexual stimuli seem topreferentially recruit an amygdalo-hypothalamic pathway. Thisaccords with previous speculations that the amygdala is a critical ini-tial structure in a processing pathway recruited in human males dur-ing the processing of sexually arousing visual stimuli8. It is also in linewith findings from animal studies7,18,25 and with the efferent connec-tivity from the amygdala to the hypothalamus, which controls physio-logical reactions associated with sexual arousal7. In summary, thecurrent findings suggest a possible neural basis for the greater role ofvisual stimuli in human male sexual behavior3–6. Whether the sex dif-ferences observed here reflect inherent differences in neural functionor stem from differential experience is a matter for further study.

METHODSSubjects and task. Twenty-eight healthy subjects took part in the study, 14female (mean age 25.0 years) and 14 male (mean age 25.9 years). All subjectsgave informed consent to participate, and the study was approved by theEmory University human investigations committee. Subjects were pre-screened to verify that they were heterosexual (self-reported as having onlyopposite-sex sexual desire and sexual experiences), had experience viewingstimuli similar to those used in the study, and found such materials signifi-cantly sexually arousing. Thirty-four males were pre-screened: four (12%)were excluded because they reported same-sex desire or experience; no maleswere excluded because of insufficient response to erotica. Forty-five femaleswere pre-screened: 16 (36%) were excluded because they reported same-sexdesire or experience and 7 (16%) were excluded because of insufficientresponse to erotica. The remaining subjects who were not included in theanalysis were excluded either because of technical difficulties with the scanneror video goggle system, fMRI signal drop-out, or because a sufficient numberof subjects had already been tested.

Subjects viewed alternating 20.125-s blocks of four types of stimuli: hetero-sexual couples engaged in explicit sexual activity (couples stimuli), attractiveopposite-sex nudes in modeling poses (opposite-sex stimuli), pleasant socialinteraction between partially or fully clothed males and females with minimalor no overt sexual content (neutral stimuli; therapeutic massage, dancing,weddings) or a visual fixation cross. Sexual stimuli were pre-selected so thatthey would be maximally attractive to females, in an effort to match femalesand males on elicited arousal. Selection was conducted via computerizedanonymous ratings with a separate group of female subjects. Only sexual stim-uli rated as highly sexually attractive and physically arousing were selected foruse in the primary experiment; stimuli that elicited weak arousal or were ratedas aversive or humorous were eliminated. Stimuli were presented via MRI-compatible goggles (Resonance Technology, Inc.). We presented stimuli rap-idly, in alternating blocks of five stimuli of each type. Each block contained fivestimuli of the corresponding type, with each stimulus presented for 3,750 msfollowed by a fixation cross for 275 ms. Two runs were presented, each con-taining four blocks of each type presented in a pseudorandom order. Ratingsof sexual attractiveness were assessed immediately after each scan; physicalsexual arousal was assessed retrospectively immediately after all scanning hadconcluded. A Likert-type rating scale was used: 0 (lowest) to 3 (highest). Forthe re-analysis that omitted one female subject who reported very low arousalratings, physical arousal ratings for females and males were 2.85 ± 0.10 and2.31 ± 0.12 (P < 0.005), respectively. Physiological measurement of sexualresponse was not conducted because of incompatibility of female genitalplethysmography with MRI scanning and the lack of commonly acceptedmethods for comparing magnitudes of male and female genital responses.Subjects were instructed to view each stimulus attentively and to experiencewhatever reactions the stimuli might elicit. Overt responses were not required,

to avoid possible interference with emotional processes elicited by the stimuli.No stimuli were repeated during the experiment.

Imaging and data analysis. MRI scanning was performed on a 1.5-tesla PhilipsIntera scanner. After acquisition of a high-resolution T1-weighted anatomicalscan, subjects underwent whole-brain functional runs (echo-planar imaging,gradient recalled echo; TR = 3,000 ms; TE = 40 ms; flip angle, 90°; 64 × 64matrix, 25 5-mm axial slices) for measurement of blood oxygen level–depend-ent (BOLD) effects. The first four volumes were discarded to allow for T1equilibration effects. Data were analyzed using SPM99 software (http://www.fil.ion.ucl.ac.uk/spm). Functional EPI volumes were realigned to the firstvolume and normalized to a standard EPI template volume using 4 mm × 4 mm × 4 mm voxels. Images were subsequently smoothed with an 8-mm isotropic Gaussian kernel and band-pass filtered in the temporaldomain. Images were carefully inspected for regions of magnetic susceptibilityinduced signal dropout in the amygdala and hypothalamus. Three males andone female were scanned but had significant signal dropout in the amygdala orhypothalamus and were replaced by newly tested subjects. Thus, all 28 subjectsreported here had minimal signal dropout in the regions of interest. Becauseonly those voxels with sufficient signal in all subjects were included for analy-sis, any signal dropout would have tended to decrease the spatial extent ofobserved sex differences in activation.

Condition effects for the stimulus conditions were estimated using box-carregressors convolved with a canonical hemodynamic response function, sepa-rately for each subject at each voxel according to the general linear model(GLM), and regionally specific effects were compared using linear contrasts12.Contrasts between conditions produced statistical parametric maps for eachsubject of the t-statistic at each voxel. Sex differences in activation wereassessed with a second-level, mixed-effects analysis with subjects as the ran-dom-effects factor, using a two-group unpaired t-test on the individual sub-ject-specific contrast images, yielding statistical parametric maps. Thesecond-level mixed-effects conjunction analysis was conducted with the indi-vidual subject-specific contrast images contrasting the couples condition withthe neutral condition, using linear regression (P < 0.05, corrected for spatialcomparisons across the whole brain, extent threshold ≥10 voxels).

For the whole-brain analysis, we thresholded these summary statisticalmaps at a voxel-wise intensity threshold of P < 0.001 (uncorrected for multiplecomparisons) with a spatial extent threshold of ≥5 contiguous voxels. For thecomparison between males and females on the couples versus neutral stimulicontrast (Fig. 2d), a P < 0.005, ≥5 contiguous voxels threshold was used,masked inclusively by the group comparison for the couples versus fixationcontrast at a lenient P < 0.01 threshold (Fig. 2a). Based on previous studies, wedefined the amygdala and hypothalamus as ROIs. We did a correction for mul-tiple spatial comparisons within each region, as a more stringent test of our apriori hypotheses. The amygdala region was defined as an 8-mm sphere cen-tered on the following coordinates: left amygdala, –20, –4, –20; right amygdala,20, –4, –20. The hypothalamic ROI was an 8-mm sphere centered on the coor-dinates 0, –4, –8. For visualization of activation extent, the group activationmaps were thresholded at P < 0.005 uncorrected, with a five-voxel extentthreshold, and they were overlaid on a representative high-resolution struc-tural T1-weighted image from a single subject from the SPM99 canonicalimage set, coregistered to Montreal Neurological Institute (MNI) space—awidely used approximation of canonical Talairach space38. All coordinates arereported in MNI space, and may be converted to Talairach space using thefreely available MNI2TAL program (http://www.mrc-cbu.cam.ac.uk/Imaging/Common/mnispace.shtml). Anatomical localization of group activations wasassisted by reference to the atlas of Duvernoy39.

For the ROI-averaged analysis, to examine differences in the magnitude offMRI signal change across all stimulus conditions and between different brainregions, we calculated the average fMRI signal change relative to the fixationbaseline for each subject. We did this for each stimulus condition for 8-mm-radius spherical ROIs centered on the left amygdala, right amygdala and hypo-thalamus. Specifically, for each subject, hemodynamic response functions foreach condition type were estimated across each ROI using a finite impulseresponse formulation of the GLM, partialling out the modeled effects of theother conditions, as implemented in R. Poldrack’s SPM ROI Toolbox(http://spm-toolbox.sourceforge.net). Parameter estimates for this model are


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

estimates of the temporally evolving response magnitude across each point inperistimulus time, averaged across all occurrences of that peristimulus timeinterval. Response estimates were averaged across the five peristimulus timepoints occurring from 6 to 18 s after the onset of each 20.125-s block (at 3-sintervals). This is the temporal window where the average signal was predictedto be maximal based on the canonical hemodynamic lag. Statistical compar-isons between conditions, regions and groups were conducted using theseaveraged values from each subject using t-tests and ANOVAs using a criterionof P < 0.05 (two-tailed).


ACKNOWLEDGMENTSThis research was supported by the Center for Behavioral Neuroscience, a Scienceand Technology Center Program of the National Science Foundation, underagreement IBN-9876754.



1. Canli, T., Desmond, J.E., Zhao, Z. & Gabrieli, J.D.E. Sex differences in the neuralbasis of emotional memories. Proc. Natl Acad. Sci. USA 99, 10789–10794 (2002).

2. Cahill, L. et al. Sex-related difference in amygdala activity during emotionally influ-enced memory storage. Neurobiol. Learn. Mem. 75, 1–9 (2001).

3. Gur, R.C. et al. An fMRI study of sex differences in regional activation to a verbaland spatial task. Brain Lang. 74, 157–170 (2000).

4. Symons, D. The Evolution of Human Sexuality (Oxford Univ. Press, Oxford, UK,1979).

5. Laumann, E.O., Gagnon, J.H., Michael, R.T. & Michaels, S. The Social Organizationof Sexuality (University of Chicago Press, Chicago,1994).

6. Herz, R.S. & Cahill, E.D. Differential use of sensory information in sexual behavioras a function of gender. Hum. Nature 8, 275–286 (1997).

7. Newman, S.W. The medial extended amygdala in male reproductive behavior: a node inthe mammalian social behavior network. Ann. NY Acad. Sci. 877, 242–257 (1999).

8. Beauregard, M., Levesque, J. & Bourgouin, P.J. Neural correlates of conscious self-regulation of emotion. J. Neurosci. 21, 1–6 (2001).

9. Redoute, J. et al. Brain processing of visual sexual stimuli in human males. Hum.Brain Mapp. 11, 162–177 (2000).

10. Arnold, A.P. & Gorski, R.A. Gonadal steroid induction of structural sex differences inthe central nervous system. Annu. Rev. Neurosci. 7, 413–442 (1984).

11. Roselli, C.E., Klosterman, S. & Resko, J.A. Anatomic relationships between aro-matase and androgen receptor mRNA expression in the hypothalamus and amygdalaof adult male cynomolgus monkeys J. Comp. Neurol. 439, 208–223 (2001).

12. Friston, K.J. et al. Statistical parametric maps in functional imaging: a general lin-ear approach. Hum. Brain Mapp. 2, 189–210 (1995).

13. Schmitt, D.P. et al. Universal sex differences in the desire for sexual variety: tests from52 nations, 6 continents, and 13 islands J. Pers. Soc. Psychol. 85, 85–104 (2003).

14. Rauch, S.L. et al. Neural activation during sexual and competitive arousal in healthymen. Psychol. Res. Neuroimag. 91, 1–10 (1999).

15. Stoleru, S. et al. Neuroanatomical correlates of visually evoked sexual arousal inhuman males. Arch. Sex. Behav. 28, 1–21 (1999).

16. Kampe, K.K., Frith, C.D., Dolan, R.J. & Frith, U. Reward value of attractiveness andgaze. Nature 413, 589 (2001).

17. Gottfried, J.A., O’Doherty, J. & Dolan, R.J. Encoding predictive reward value inhuman amygdala and orbitofrontal cortex. Science 301, 1104–1107 (2003).

18. Everitt, B.J. Sexual motivation: a neural and behavioral analysis of the mechanismsunderlying appetitive and copulatory responses of male rats. Neurosci. Biobehav.Rev. 14, 217–232 (1990).

19. Hamann, S.B., Ely, T.D., Hoffman, J.M. & Kilts, C.D. Ecstasy and agony: activationof the human amygdala in positive and negative emotion. Psychol. Sci. 13,135–141 (2002).

20. Garavan, H., Pendergrass, J.C., Ross, T.J., Stein, E.A. & Risinger, R.C. Amygdalaresponse to both positively and negatively valenced stimuli. Neuroreport 12,2779–2783 (2001).

21. Karama, S.M. et al. Areas of brain activation in males and females during viewing oferotic film excerpts. Hum. Brain Mapp. 16, 1–13 (2002).

22. Bradley, M.M., Codispoti, M., Sabatinelli, D. & Lang, P. Emotion and motivation. II:Sex differences in picture processing. Emotion 1, 300–319 (2001).

23. Anderson, A.K. et al. Dissociated neural representations of intensity and valence inhuman olfaction. Nat. Neurosci. 6, 196–202 (2003).

24. Canli, T., Zhao, Z., Brewer, J., Gabrieli, J.D. & Cahill, L. Event-related activation inthe human amygdala associates with later memory for individual emotional experi-ence. J. Neurosci. 20, 1–5 (2000).

25. Everitt, B.J., Cardinal, R.N., Parkinson, J.A. & Robbins, T.W. Appetitive behavior:impact of amygdala-dependent mechanisms of emotional learning. Ann. NY Acad.Sci. 985, 233–250 (2003).

26. Arana, F.S. et al. Dissociable contributions of the human amygdala and orbitofrontalcortex to incentive motivation and goal selection. J. Neurosci. 23, 9632–9638(2003).

27. Gallagher, M. The amygdala and associative learning. in The Amygdala: a FunctionalAnalysis edn. 2 (ed. Aggleton, J.P.) 311–330 (Oxford Univ. Press, Oxford, UK,2001).

28. Baird, A.D. et al. The amygdala and sexual drive: insights from temporal lobeepilepsy surgery. Ann. Neurol. 55, 87–96 (2004).

29. Hamann, S.B., Ely, T.D., Grafton, S.T. & Kilts, C.D. Amygdala activity related toenhanced memory for pleasant and aversive stimuli. Nat. Neurosci. 2, 289–293(1999).

30. Anderson, A.K. & Phelps, E.A. Lesions of the human amygdala impair enhanced per-ception of emotionally salient events. Nature 411, 305–309 (2001).

31. Adolphs, R., Tranel, D. & Damasio, A.R. The human amygdala in social judgment.Nature 383, 470–473 (1998).

32. Davis, M. & Whalen, P.J. The amygdala: vigilance and emotion. Mol. Psychiatry 6,12–34 (2001).

33. Calder, A.J., Lawrence, A.J. & Young, A.W. Neuropsychology of fear and loathing.Nat. Rev. Neurosci. 2, 352–363 (2001).

34. Dolan, R.J. & Morris, J.S. The functional anatomy of innate and acquired fear: per-spectives from neuroimaging. in Cognitive Neuroscience of Emotion (eds. Lane,R.D. & Nadel, L.) 225–241 (Oxford Univ. Press, New York, 2000).

35. Funayama, E.S., Grillon, C., Davis, M. & Phelps E.A. A double dissociation in theaffective modulation of startle in humans: effects of unilateral temporal lobectomy.J. Cogn. Neurosci. 13, 721–729 (2001).

36. Morris, J.S., Ohman, A. & Dolan, R.J. Conscious and unconscious learning in theamygdala. Nature 393, 467–470 (1998).

37. Holstege, G. et al. Brain activation during human male ejaculation. J. Neurosci. 23,9185–9193 (2003).

38. Talairach, J. & Tournoux, P. Co-planar Stereotaxic Atlas of the Human Brain (ThiemeMedical, New York, 1988).

39. Duvernoy, H.M. The Human Brain: Surface, Three-dimensional Sectional Anatomyand MRI (Springer, New York, 1991).


©20

04 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Documents

Nature Neuroscience April 2004