nature neuroscience september 2005

www.nature.com/natureneuroscience

EDITORIAL OFFICE [email protected] Park Avenue South, New York, NY 10010-1707Tel: (212) 726 9319, Fax: (212) 696 0978Editor: Sandra Aamodt Senior Editor: Kalyani Narasimhan Associate Editors: Cara Allen, I-han Chou, Annette Markus Assistant Editor: Charvy NarainCopy Editor: Dorothy MooreSenior Production Editor: Nicole D. FournierProduction Editor: Ivelisse RoblesAssistant Production Editor: Lana ChengCover Design: Erin BoyleEditorial Assistant: Jessica Chen

MANAGEMENT OFFICESNPG New York345 Park Avenue South, New York, NY 10010-1707Tel: (212) 726 9200, Fax: (212) 696 9006Publisher: Beatrice RenaultExecutive Editor: Linda MillerChief Technology Officer: Howard RatnerGlobal Head of Advertising: Fabien SavenayHead of Nature Research & Reviews Marketing: Sara GirardMarketing Manager: Naomi MulgraveProduction Coordinator: Diane TempranoAssociate Director, Web Publishing: Timo HannayNew Technology Director: Greg SuprockHead of Web Services: Anthony BarreraWeb Production Editor: Edwina Hay

NPG LondonThe Macmillan Building, 4 Crinan Street, London N1 9XWTel: 44 207 833 4000, Fax: 44 207 843 4996Managing Director: Annette ThomasPublishing Director: Peter CollinsEditor-in-Chief, Nature Publications: Philip CampbellMarketing Director: Della Sar

NPG TokyoChiyoda Building 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843Tel: 81 3 3267 8751, Fax: 81 3 3267 8746Asia-Pacific Publisher: Antoine E. BocquetManager: Koichi NakamuraSenior Marketing Manager: Peter YoshiharaAsia-Pacific Sales Director: Kate YoneyamaAsia-Pacific Sales Manager: Rinoko Asami

DISPLAY ADVERTISING [email protected] (US/Canada) [email protected] (Europe) [email protected] (Japan)US Head of Display Advertising: Stephen Schwartz, Tel: (212) 726 9256, Fax: (212) 696 9481Global Head of Display Advertising Sales: John Michael, Tel: 44 207 843 4960, Fax: 44 207 843 4996Head of Display Advertising—Europe: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4996Business Development Manager: Claire Hines, Tel: 44 207 843 4960, Fax: 44 207 843 4996Asia-Pacific Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746Western Region Sales Manager: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805Sales Executives:New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717New York, Mid-Atlantic, Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481Northwest: Mathieu DesJardins, Tel: (415) 781 6422, Fax: (415) 781 3805Eastern England/Scotland/Italy/Spain/Israel: Matthew Wilkinson, Tel: 44 207 843 4960, Fax: 44 207 843 4749South/West United Kingdom/Scandinavia/Holland: Marianne Boulakas, Tel: 44 207 843 4969, Fax: 44 207 843 4749Northern Germany: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4749Southern Germany/Austria/Switzerland/France/Belgium: Sabine Hugi-Fürst, Tel: 41 52761 3386, Fax: 41 52761 3419

NATUREJOBS [email protected] (US/Canada) [email protected] (Europe) [email protected] (Japan)Publisher: Ben Crowe, Tel: (212) 726 9245, Fax: (212) 696 9482US Sales Manager: Peter Bless, Tel: (212) 726 9248, Fax: (212) 696 9482Japan Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746naturejobs Sales Director: Nevin Bayoumi, Tel: 44 207 843 4961, Fax: 44 207 843 4996

CUSTOMER SERVICE www.nature.com/helpSenior Global Customer Service Manager: Gerald CoppinFor all print and online assistance, please visit www.nature.com/helpPurchase subscriptions:Americas: Nature Neuroscience, Subscription Dept., 303 Park Avenue South #1280, New York, NY 10010-3601. Tel: (866) 363 7860, Fax: (212) 689 9108Europe/ROW: Nature Neuroscience, Subscription Dept., Macmillan Magazines Ltd., Brunel Road, Houndmills, Basingstoke RG21 6XS, United Kingdom. Tel: 44 1256 329 242, Fax: 44 1256 812 358Japan: Nature Neuroscience, Nature Japan K.K., MG Ichigaya Bldg. 5F, 19-1 Haraikatamachi, Shinjuku-ku, Tokyo 162-0841. Tel: 81 3 3267 8751, Fax: 81 3 3267 8746India: Harpal Singh Gill, Macmillan Magazines Ltd, 5A/12 Ansari Road, Darya Ganj, New Delhi, 110 002 India. Tel: 00 91 11 324 4186, Tel/Fax: 00 91 11 327 2010

REPRINTS [email protected] Neuroscience Reprint Department, Nature Publishing Group, 345 Park Avenue South, New York, NY 10010-1707, USA.For commercial reprint orders of 600 or more, please contact:UK Reprints Sales Executive: Christine Fothergill, Tel: 44 207 843 4967, Fax: 44 207 843 4749US Reprints Sales Executive: Sharda Tulsie, Tel: (212) 726 9631, Fax: (212) 679 0843

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

i

VOLUME 8 NUMBER 9 SEPTEMBER 2005

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature Publishing Group, a trading name of Nature America Inc. located at 345 Park Avenue South, New York, NY 10010-1707. Periodicals postage paid at New York, NY and additional mailing post offices. Editorial Office: 345 Park Avenue South, New York, NY 10010-1707. Tel: (212) 726 9319, Fax: (212) 696 0978. Annual subscription rates: USA/Canada: US$199 (personal), US$1,809 (institution). Canada add 7% GST #104911595RT001; Euro-zone: €271 (personal), €1,558 (institution); Rest of world (excluding China, Japan, Korea): £175 (personal), £1,005 (institution); 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. POSTMASTER: Send address changes to Nature Neuroscience, Subscriptions Department, 303 Park Avenue South #1280, New York, NY 10010-3601. Authorization to photocopy material for internal or personal use, or internal or personal use of specific clients, is granted by Nature Publishing Group to libraries and others registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided the relevant 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. CPC PUB AGREEMENT #40032744. Printed by Publishers Press, Inc., Lebanon Junction, KY, USA. Copyright © 2005 Nature Publishing Group. Printed in USA.

E D I TO R I A L

1123 A mercurial debate over autism

CO R R E S P O N D E N C E

1125 Motion processing in macaque V4

B O O K R E V I E W

1127 The Ethical Brainby Michael S GazzanigaReviewed by Charles Jennings

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

1129 Neuroeconomics: making risky choices in the brainDaeyeol Lee � see also p 1220

1130 Wnts send axons up and down the spinal cordBarry J Dickson � see also p 1151

1132 GABA puts the brake on stem cellsArnold R Kriegstein � see also p 1179

1134 Making the causal link: frontal cortex activity and repetition primingAlex Martin & Stephen J Gotts � see also p 1228

1136 Pathological tau: a loss of normal function or a gain in toxicity?John Q Trojanowski & Virginia M-Y Lee

1137 Imaging heterogeneity in synaptic transmissionKalyani Narasimhan � see also p 1188

Ryk-mediated repulsionin axonal growth

(pp 1130 and 1151)

Choosing when to seek or avoid risk can be critical for survival. McCoy and Platt report that monkeys consistently

took the riskier option when asked to choose between a sure bet and a

more uncertain reward, even when the risky choice led to a smaller reward

on average. Neurons in the posterior cingulate cortex responded to the

riskiness of these decisions. Cover illustration by Ann Thomson.

(pp 1129 and 1220)

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

iii


NATURE NEUROSCIENCE

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

1139 Selective inhibition of 2-AG hydrolysis enhances endocannabinoid signaling in hippocampusJudit K Makara, Marco Mor, Darren Fegley, Szilárd I Szabó, Satish Kathuria, Giuseppe Astarita, Andrea Duranti, Andrea Tontini, Giorgio Tarzia, Silvia Rivara, Tamás F Freund & Daniele Piomelli

1142 Sleep-disordered breathing after targeted ablation of preBötzinger complex neuronsLeanne C McKay, Wiktor A Janczewski & Jack L Feldman

1145 ‘Breaking’ position-invariant object recognitionDavid D Cox, Philip Meier, Nadja Oertelt & James J DiCarlo

1148 Extensive piano practicing has regionally specific effects on white matter developmentSara L Bengtsson, Zoltán Nagy, Stefan Skare, Lea Forsman, Hans Forssberg & Fredrik Ullén

A R T I C L E S

1151 Ryk-mediated Wnt repulsion regulates posterior-directed growth of corticospinal tractYaobo Liu, Jun Shi, Chin-Chun Lu, Zheng-Bei Wang, Anna I Lyuksyutova, Xuejun Song & Yimin Zou � see also p 1130

1160 Activation of GPCRs modulates quantal size in chromaffin cells through Gβγ and PKCXiao-Ke Chen, Lie-Cheng Wang, Yang Zhou, Qian Cai, Murali Prakriya, Kai-Lai Duan, Zu-Hang Sheng, Christopher Lingle & Zhuan Zhou

1169 Bag1 is essential for differentiation and survival of hematopoietic and neuronal cellsRudolf Götz, Stefan Wiese, Shinichi Takayama, Guadalupe C Camarero, Wilfried Rossoll, Ulrich Schweizer, Jakob Troppmair, Sibylle Jablonka, Bettina Holtmann, John C Reed, Ulf R Rapp & Michael Sendtner

1179 Nonsynaptic GABA signaling in postnatal subventricular zone controls proliferation of GFAP-expressing progenitorsXiuxin Liu, Qin Wang, Tarik F Haydar & Angélique Bordey � see also p 1132

1188 Heterogeneity in synaptic transmission along a Drosophila larval motor axonGiovanna Guerrero, Dierk F Rieff, Gautam Agarwal, Robin W Ball, Alexander Borst, Corey S Goodman & Ehud Y Isacoff � see also p 1137

1197 Neural basis of auditory-induced shifts in visual time-order perceptionJohn J McDonald, Wolfgang A Teder-Sälejärvi, Francesco Di Russo & Steven A Hillyard

1203 Auditory thalamus integrates visual inputs into behavioral gainsYutaka Komura, Ryoi Tamura, Teruko Uwano, Hisao Nishijo & Taketoshi Ono

1210 Neural codes for perceptual discrimination in primary somatosensory cortexRogelio Luna, Adrián Hernández, Carlos D Brody & Ranulfo Romo

1220 Risk-sensitive neurons in macaque posterior cingulate cortexAllison N McCoy & Michael L Platt � see also p 1129

Bag1 and neuronal differentiation (p 1169)

Nonsynaptic GABA signaling in postnatal subventricular zone

(pp 1132 & 1179)

Learning to predict pain relief involves reward and punishment

(p 1234)

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

v


NATURE NEUROSCIENCE

Filling-in of visual phantoms (p 1248)

Genetically targeted optical control of neural activity

(p 1263)

1228 Reductions in neural activity underlie behavioral components of repetition primingGagan S Wig, Scott T Grafton, Kathryn E Demos & William M Kelley� see also p 1134

1234 Opponent appetitive-aversive neural processes underlie predictive learning of pain reliefBen Seymour, John P O’Doherty, Martin Koltzenburg, Katja Wiech, Richard Frackowiak, Karl Friston & Raymond Dolan

1241 Structural and functional asymmetry of lateral Heschl’s gyrus reflects pitch perception preferencePeter Schneider, Vanessa Sluming, Neil Roberts, Michael Scherg, Rainer Goebel, Hans J Specht, H Günter Dosch, Stefan Bleeck, Christoph Stippich & André Rupp

1248 Filling-in of visual phantoms in the human brainMing Meng, David A Remus & Frank Tong

1255 Regret and its avoidance: a neuroimaging study of choice behaviorGiorgio Coricelli, Hugo D Critchley, Mateus Joffily, John P O’Doherty, Angela Sirigu & Raymond J Dolan

T E C H N I C A L R E P O R T

1263 Millisecond-timescale, genetically targeted optical control of neural activityEdward S Boyden, Feng Zhang, Ernst Bamberg, Georg Nagel & Karl Deisseroth

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.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 9 | SEPTEMBER 2005 1123

E D I TO R I A L

A mercurial debate over autism

In June, environmental lawyer and activist Robert F. Kennedy, Jr., accused the scientific community of covering up evidence that mer-cury in childhood vaccines causes autism1. Before 2001, US childhood

vaccines contained thimerosal, a preservative that includes ethyl mercury. As more vaccinations were recommended beginning in the 1980s, cumu-lative exposure increased, eventually exceeding the safety limit set by the US Environmental Protection Agency for mercury, a known neurotoxin. During this period, the number of autism cases increased, and thimerosal was suggested as a possible cause2. The hypothesis seemed plausible when first proposed, but recent epidemiological data do not support a causal relationship3. Some supporters of the hypothesis are not convinced by these data, and the scientific community has not helped its case with the public by appearing defensive or dismissive of this position.

Autism is the most severe of the autism spectrum disorders, pervasive developmental disorders characterized by impaired language, non-ver-bal communication and social interaction, and repetitive or stereotypi-cally restricted behaviors. Autism is strongly heritable, and epigenetic and environmental factors are likely to interact with a predisposing genetic background involving multiple risk genes. According to the US Centers for Disease Control and Prevention, the prevalence of autism spectrum disorders ranges from 2–6 per 1,000, and the number of cases has risen about tenfold over the last 20 years in the US and other western countries, with some reports claiming more dramatic increases. Some or all of this apparent increase, however, may be due to changes in diagnostic definitions and recognition of the disorder by parents and doctors. Yet, with so many affected children and so few answers, parents of autistic children are understandably frustrated. This feeling may enhance the attractiveness of the thimerosal hypothesis, which allows parents to identify a discrete cause and suggests avoiding future exposure as a reassuring preventative action.

Unfortunately, epidemiological studies do not support this link. In May 2004, a review by the Institute of Medicine (IOM)3 of over 200 studies (avail-able online at http://www.nap.edu/catalog/10997.html) concluded “that the evidence favors rejection of a causal relationship between thimerosal-containing vaccines and autism,” a view held by most of the international scientific community. The World Health Organization also maintains that

“there is no evidence supporting a causal association between neurobehav-ioural disorders and thiomersal-containing vaccines.”

Nonetheless, supporters of the thimerosal hypothesis continue to campaign aggressively for removal of any remaining trace of the preser-vative from medical products and for research that would confirm their hypothesis or develop therapies based on it. Largely through one vocal parent organization, Safe Minds, the idea has attracted media and politi-cal attention. However, according to Marie McCormick of the Harvard School of Public Health, chair of the IOM panel, other parent groups fear that the research process has been hijacked by the ongoing controversy. Many scientists see it as a distraction from other avenues of research that are more likely to yield insights into the disorders’ causes, prevention and treatment.

One promising avenue is to use the variability in clinical character-istics of autism to build more focused hypotheses. For example, most children with autism show abnormalities within the first year of life. In contrast, about a quarter of autistic children seem to develop normally until about 15–24 months of age, when their development appears to regress. Geraldine Dawson at the University of Washington, who studies regressive autism, notes that distinguishing between early-and late-onset Alzheimer disease led to important breakthroughs and hopes that the same may be true for autism. A planned intramural program at the US National Institutes of Health will look at psychological, immune and other measures across regressive and non-regressive cases to exam-ine what might precipitate the disease.

Improved understanding of phenotypic diversity is also guiding the search for autism risk genes. A subset of the symptoms of autism are often variably expressed in unaffected family members. The use of these endophenotypes increases the power of genetic analyses and may lead to identification of genetic homogeneities underlying individual traits. The hope, says Daniel Geschwind at UCLA, is to “take these genetic homogeneities and then work backward” to generate hypotheses about disease etiology, or about what types of gene-environment interactions may contribute to autism spectrum disorders.

In addition to interfering with such promising lines of research, the thimerosal controversy threatens to undermine the public’s trust that scientists are committed to studying the problem. Overcoming this mis-trust will require continued efforts from scientists to collaborate with the public. A good example is the US Department of Health and Human Services’ Interagency Autism Coordinating Committee, which includes parents or legal guardians of autistic patients. These members repre-sent the autism community and help guide policy decisions on autism research. Through this committee, the National Institutes of Health and other governmental agencies have developed partnerships for research and public education with national autism associations (http://www.nimh.nih.gov/autismiacc). Another successful collaborative effort is the Autism Genetic Resource Exchange (AGRE) program4 founded by the Cure Autism Now organization, which has provided researchers with biomaterials from hundreds of families with autistic children.

In the end, McCormick may be correct that some supporters of the thimerosal hypothesis are unlikely to be swayed in their beliefs by anything short of finding the “silver bullet that causes autism.” In the meantime, the absence of a clear mechanistic explanation should not be used to direct resources toward a single weak hypothesis. Parent groups should instead seek reassurance in continued collaboration with the scien-tific community as it moves forward in more promising directions.

1. Kennedy, R.F, Jr. Rolling Stone, published online 20 June 2005 <http://www.rollingstone.com/politics/story~/id/7395411>.

2. Bernard, S., Enayati, A., Redwood, L., Roger, H. & Binstock, T. Med. Hypotheses 56, 462–471 (2001).

3. Immunization Safety Review: Vaccines and Autism (National Academies Press, Washington, DC, 2004).

4. Geschwind, D.H. et al. Am. J. Hum. Genet. 69, 463–466 (2001).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


CO R R E S P O N D E N C E

Motion processing in macaque V4To the Editor:A recent study by Tolias et al.1 demonstrated convincingly that neurons in visual area V4 can acquire direction selectivity after adaptation to coherent motion. This is a valuable result because it bears on the role of V4 in motion processing and may also reconcile an appar-ent discrepancy between neurophysiology and functional imaging of V4 (ref. 2). Moreover, it underscores how dynamic aspects of all types of selectivity can be missed without specific tests. However, the premise of the study, that V4 neu-rons “are generally not selective for direction of motion”, is not entirely accurate. At least three reports have assessed direction selectivity in V4 quantitatively3–5 and have found that roughly one-third of V4 neurons are direction selective (preferred:null response ratio ≥ 2.0 or d-prime ≥ 1.0). A direct comparison5 between V4, MT and area 7a in the same animals and using coherent random dot stimuli showed that MT neurons had a mean preferred:null ratio of 6.4, compared with 1.8 for V4 and 1.7 for area 7a. Thus, although direction selectivity in V4 is not nearly as pronounced as in MT, it is compa-rable to other areas in the dorsal pathway. One must also consider that in the macaque, V4 is many times larger than MT6 and may there-fore contain a comparable number of highly direction-selective neurons even if the density of such neurons is much lower. Although the existence of conventional direction selectivity in V4 does not alter the authors’ conclusions about the role of adaptation in stimulus selec-tivity, we feel that V4 should not be overlooked as a potentially reliable source of conventional

motion signals outside of areas traditionally associated with motion processing.

Vincent P Ferrera1 & John H R Maunsell2

1Columbia University, Center for Neurobiology and Behavior, 1051 Riverside Drive, New York, New York 10032, USA. 2Howard Hughes Medical Institue and Baylor College of Medicine, Department of Neuroscience, One Baylor Plaza, Houston, Texas 77030 USA.e-mail: [email protected]

1. Tolias, A.S., Keliris, G.A., Smirnakis, S.M. & Logothetis, N.K. Nat. Neurosci. 8, 591–593 (2005).

2. Tolias, A.S., Smirnakis, S.M., Augath, M.A., Trinath, T. & Logothetis, N.K. J. Neurosci. 21, 8594–8601 (2001).

3. Desimone, R. & Schein, S.J. J. Neurophysiol. 57, 835–868 (1987).

4. Mountcastle, V.B., Motter, B.C., Steinmetz, M.A. & Sestokas, A.K. J. Neurosci. 7, 2239–2255 (1987).

5. Ferrera, V.P., Rudolph, K.K. & Maunsell, J.H.R. J. Neurosci. 14, 6171–6186 (1994).

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

the 95–99.8% coherent stimulus used by Ferrera et al.2. Second, as we simultaneously recorded from many neurons, we did not optimize the stimulus parameters (aperture size, dot size and density and speed) individually for each cell. Third, we used different tests and thresholds for statistical significance. In addition, Desimone and Schein3, who used moving gratings and a stricter tuning index threshold (<0.3, response to null/preferred direction) than did Ferrera et al., reported 13% of V4 neurons to be direc-tion-selective, in close agreement with our result. These reasons are likely to account for the difference in estimates of direction selec-tive neurons. It is clear, however, that there is a significant number of V4 neurons (13–33%, depending on estimates) that are tuned to direction of motion even before adaptation. We therefore believe that Ferrera and Maunsell are correct in suggesting that area V4 can be a significant source of conventional motion sig-nals before adaptation. Our paper shows that this role seems to be significantly enhanced after adaptation1.

Andreas S Tolias, Georgios A Keliris, Stelios M Smirnakis & Nikos K Logothetis

Max Planck Institute for Biological Cybernetics, Spemannstrasse 38, Tuebingen, 72076, Germany.e-mail: [email protected]

1. Tolias, A.S., Keliris, G.A., Smirnakis, S.M. & Logothetis, N.K. Nat. Neurosci. 8, 591–593 (2005).

2. Ferrera, V.P., Rudolph, K.K. & Maunsell, J.H.R. J. Neurosci. 14, 6171–6186 (1994).

3. Desimone, R. & Schein, S.J. J. Neurophysiol. 57, 835–868 (1987).

Tolias et al. reply:We agree with Ferrera and Maunsell that V4 could potentially serve as a source of conven-tional motion signals. This is also supported by our own study1, in which we found that 15% of V4 neurons were directionally selective before adaptation (Rayleigh test for circular statistics). The difference between our estimate of the per-centage of direction-selective neurons and the estimate of Ferrera et al.2 (they report around 33% direction-selective neurons) probably has a methodological origin. First, the random dot stimulus we used to test direction of motion selectivity had 60% coherence, as opposed to

We welcome short letters on matters arising from previous papers in Nature Neuroscience or

on other topics of widespread interest to the neuroscience community. Because space in this

section of the journal is limited, priority is given to short (fewer than 500 words), well-written

letters addressing the most topical issues. Typically, new data are not presented in this section,

although they may occasionally be allowed at the discretion of the editors. Letters concerning

material previously published in Nature Neuroscience are usually sent to the authors of the orig-

inal piece for their comments and/or a formal reply. Letters may be edited for brevity and clarity.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Neuroeconomics: making risky choices in the brainDaeyeol Lee

Choosing to accept enough risk, but not too much, is an important survival skill, and depending on the circumstances, animals may either seek or avoid risk. Given the choice between a sure bet and a larger but uncertain reward, a paper in this issue reports macaques consistently take the riskier option, and posterior cingulate cortex neurons represent the riskiness of those choices.

You are diagnosed with a disease for which the only available treatment is surgery. Without the surgery, you may survive for six months. If successful, the surgery will cure the disease, but there is a 50% chance that you may die of surgi-cal complications. This is a tragic and extreme example of decision-making involving risk, but our daily lives are full of such choices with uncertain outcomes. Economists and psycholo-gists have long studied how people and animals deal with uncertainty in making decisions. More recently, neurobiologists have begun to study the brain processes involved in decision-making1, but the cellular mechanisms underlying risky choices have not been systematically explored. Now McCoy and Platt2 report in this issue that individual neurons in posterior cingulate cortex respond according to the riskiness of the ani-mal’s choice. This is an excellent example of how neurobiological studies can enrich the formal economic theories of decision-making.

In economics, the numerical measure of an individual’s preference or subjective value for an object is referred to as utility. Although it may be difficult to compare apples and oranges, utilities of any items should incorporate the preferences of an individual. This way, his or her choices can be summarized by a parsimo-nious principle, such as utility maximization. In situations where the outcome of a decision is uncertain, risk can be formally defined as a spread from the mean in the objective values of possible outcomes (variance). For example, a lottery that pays either $10 or $20 is riskier than the one paying $14 or $16, although they have the same mean payoffs. By definition, risk is zero for a choice that has a fixed outcome. Expected utility theory provides a solution to the problem

of decision-making involving risk. In this theory, the utility of a choice is determined by summing the utility of each possible outcome weighted by its probability3. In other words, the utility of a choice with uncertain outcomes is its expected utility. For example, imagine that you are trying to choose between a piece of chocolate that is known to have a cherry filling and another one with a filling that is equally likely to be whis-key or pineapple. Let us also assume that you enjoy the one with a whiskey filling four times as much, and the one with cherry twice as much as the one with pineapple (4 units of utility, or ‘utils’), and the one with cherry twice as much (2 utils) than the one with pineapple (1 util). According to the expected utility theory, then, the utility of the chocolate with an unknown fill-ing would be the average of utility for whiskey-filled chocolate and utility for pineapple-filled chocolate (2.5 utils), and thus you should prefer this risky chocolate to the cherry-filled one.

The amount of risk involved in choosing a piece of chocolate might be difficult to quantify, but this can be done when different options are identified according to an objective measure (such as money). For example, imagine that the utility of money is proportional to its amount (Fig. 1a). In this case, the expected utility is

equivalent to the utility of the average outcome. For example, the utility of a lottery that is equally likely to pay 10 or 20 dollars would correspond to the utility of the average payoff of $15. In other words, uncertainty in the outcome (risk) does not affect utility when utility of a choice is pro-portional to its objective value. In this situation, the decision-maker is said to be ‘risk-neutral’, and is indifferent to any combination of possible out-comes as long as their mean remains the same.

However, such linear utility functions often do not fit actual behavior, because people and animals can and do seek or avoid risk. In his celebrated paper3 published in 1738, Daniel Bernoulli recognized the link between the risk preference and the shape of the utility function. He proposed that the utility function should be logarithmic rather than linear, consistent with the common intuition that a small amount of money should mean more (constitute a larger increase in utility) to a beggar than to a millionaire. An individual with such a concave utility function would tend to avoid a risky choice, because risk decreases expected utility (Fig. 1b). In contrast, an indi-vidual with a convex utility function would tend to prefer a risky choice (Fig. 1c). In real-ity, people show both of these tendencies. For

Daeyeol Lee is in the Department of Brain and

Cognitive Sciences, Center for Visual Science,

University of Rochester, Rochester, New York

14627, USA.

e-mail: [email protected]

a b c

Figure 1 Utility theory’s account of risk preference based on the comparison between the utility of the average outcome or reward, U(⎯x), and the expected utility, E{U(x)} (black disk). (a) When a small (x1) and large (x2) reward are expected with the same probability (P = 0.5), the expected utility and the utility of the average reward are equal for a risk-neutral individual. (b,c) The expected utility is smaller and larger for a risk-averse (b) and risk-prone (c) individual, respectively, than the utility of the average reward.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

1130 VOLUME 8 | NUMBER 9 | SEPTEMBER 2005 NATURE NEUROSCIENCE


The developing spinal cord is a major highway for growing axons. Axons enter and exit the highway at specific points, and, when appro-

priate, cross over to the other side. But the main traffic flow, as on any highway, is in both directions along the longitudinal axis—up to or down from the brain. What are the guid-ance cues that send these axons up or down the spinal cord? Over the past decade, researchers have uncovered many of the molecular sign-posts that regulate axon entry, exit and cross-

ing over. Yet, until recently, the signals that direct axons up and down the spinal cord had been elusive. The first breakthrough came a couple of years ago, when the Zou group dem-onstrated that Wnt proteins are important in directing axon growth toward the brain1. Now, further work from the same group, by Liu et al. in this issue2, suggests that other Wnt

example, an individual might insure a car used to drive to the casino. Such observations have led to more complex shapes of utility func-tions4 and other elaborations of utility theory5. People and animals tend to be risk-prone when the choice involves potential losses5,6, but risk-averse when they face potential gains or when the animal’s energy intake is sufficient for its daily metabolic requirement7.

Economic and psychological theories of deci-sion-making can successfully account for a broad range of human and animal choice behaviors, but the neural basis for this fundamental aspect of life is only beginning to be understood. Given the central role of utility in formal theories of decision-making, it is not surprising that much effort has been devoted to identifying neural sig-nals related to the utility of choices made by the animal. Indeed, signals resembling utility have been found in many brain areas, including the posterior cingulate cortex8 targeted in McCoy and Platt’s new study2. However, it has not been possible to determine whether such signals are actually related to utility (subjective value) or to the objective value of reward (such as its size). This can be accomplished by presenting a decision-maker with the task of choosing between two alternatives with the same mean outcome, one of which has a fixed outcome and the other of which has an uncertain outcome. This is precisely the approach used in McCoy and Platt’s study.

Monkeys were trained to choose between two targets, indicating their choice with an eye move-ment. Choosing one of the targets delivered a fixed amount of juice reward, but the amount of juice available from the other target was uncer-tain. By choosing the risky target, the animal had a 50:50 chance of receiving a larger or smaller reward than the mean, although the average reward was always the same for both targets. There were no other differences between the tar-

gets, so only the riskiness of the animal’s choice differentiated the two. Risk was systematically manipulated by changing the difference between the smaller reward and the larger reward avail-able from the risky target. The monkeys tested in McCoy and Platt’s experiment systematically preferred the risky target, and the riskier the tar-get, the more likely the animals were to choose it. Remarkably, the animals continued to show a bias for risky choices even when the probability of obtaining a larger reward from the risky tar-get was reduced so that the risky choice led to a smaller average reward.

McCoy and Platt also recorded the activity of individual neurons in the posterior cingu-late cortex while the animals were making their choices, and found that more than half of the neurons signaled not only the animal’s choice but also the riskiness of that choice. Because the animals were risk-prone in this experiment, the utility of the risky target must be larger than the utility of the average reward. Therefore, neu-rons responding more strongly to a risky target might have been signaling its utility, rather than merely the size of the expected reward. This is indeed what McCoy and Platt found. They rea-soned that such quantities as utility or expected reward size must be estimated from the animal’s recent experience9–12. However, they found that the activity in the posterior parietal cortex did not encode the size of reward in the previous trial. They then estimated the utility of each tar-get on a trial-by-trial basis according to the sum of reward size and risk, and found that this was more reliably reflected in the neural activity.

The study of McCoy and Platt raises several exciting questions for future studies. First, by providing quantitative data regarding the risk preference of monkeys, it lays the foundation for further neurobiological studies of risk prefer-ence in primate brains. It would be interesting, for example, to determine whether monkeys are

intrinsically risk-prone, or whether their risk preference can be manipulated by any environ-mental or cognitive factors6,7. Second, this study will stimulate similar future studies in other brain regions, as an animal’s ability to make adaptive decisions depends on cooperation among multiple cortical and subcortical areas11–

15. For example, are risk-related signals found in the present study first generated in the posterior cingulate cortex? If not, what is the function of risk-related signals represented in this particular brain area? How do signals related to risk or util-ity ultimately influence the choice of the animal? As McCoy and Platt demonstrated, answers to many of these questions may be within reach now. As with any other choice in our stochastic environment, the decision to study the neural basis of risky choices might be risky, but such studies will be surely rewarding.

1. Glimcher, P.W. & Rustichini, A. Science 306, 447–452 (2004).

2. McCoy, A.N. & Platt, M.L. Nat. Neurosci. 8, 1220–1229 (2005).

3. Bernoulli, D. (trans. by L. Sommer). Econometrica 22, 23–36 (1954).

4. Friedman, M. & Savage, L.J. J. Polit. Econ. 56, 279–304 (1948).

5. Kahneman, D. & Tversky, A. Econometrica 47, 263–291 (1979).

6. Marsh, B. & Kacelink, A. Proc. Natl. Acad. Sci. USA 99, 3352–3355 (2002).

7. Caraco, T. et al. Anim. Behav. 39, 338–345 (1990).

8. McCoy, A.N, Crowley, J.C., Haghighian, G., Dean, H.L. & Platt, M.L. Neuron 40, 1031–1040 (2003).

9. Sutton, R.S. & Barto, A.G. Reinforcement Learning: An Introduction (MIT Press, Cambridge, Massachusetts, USA, 1998).

10. Lee, D, Conroy, M.L., McGreevy, B.P. & Barraclough, D.J. Cogn. Brain Res. 22, 45–58 (2004).

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

12. Sugrue, L.P., Corrado, G.S. & Newsome, W.T. Science 304, 1782–1787 (2004).

13. Fiorillo, C.D., Tobler, P.N. & Schultz, W. Science 299, 1898–1902 (2003).

14. Rolls, E.T. Cereb. Cortex 10, 284–294 (2000).15. Roesch, M.R. & Olson, C.R. J. Neurophysiol. 90,

1766–1789 (2003).

Wnts send axons up and down the spinal cordBarry J Dickson

Certain Wnts attract ascending somatosensory axons up the spinal cord toward the brain. A study in this issue shows that other Wnts guide corticospinal axons down the spinal cord, not by an attractive mechanism but by repulsion through the receptor Ryk.

Barry J. Dickson is at the Institute of Molecular

Biotechnology of the Austrian Academy of Sciences,

Dr. Bohr-Gasse 3-5, A-1030 Vienna, Austria.


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



proteins direct axons in the opposite direc-tion, down the spinal cord.

Wnt proteins seem unlikely candidates for axon guidance cues. They are far better known for their roles in cell-fate specification and cell proliferation, which have been the focus of intense investigation for more than two decades. It thus came as a major surprise when clever genetic studies identified a Wnt protein, Wnt5, as a key factor in axon guidance in Drosophila melanogaster3. Specifically, Wnt5 was found to determine whether commissural axons—those that extend across the midline of the CNS—take the anterior or the poster-ior commissure of each segment. A further remarkable finding was that Wnt5 does not seem to act through a Frizzled (Fz) protein, the typical seven-transmembrane-domain receptors for Wnts. Instead, Wnt5 signals through a completely unrelated single-pass transmembrane receptor of the Ryk family, called Derailed (Drl).

While this Drosophila work was going on, members of the Zou group were looking for factors in vertebrates that instruct commis-sural axons to turn anteriorly after crossing. Remarkably, they too stumbled on Wnt pro-teins, including Wnt4, which are expressed in the floor plate at the ventral midline in an anterior-high to posterior-low gradient1 (Fig. 1a). Their report provided compelling evidence that Wnt proteins can also act as axon-guidance factors in vertebrates, just as they do in Drosophila3. But there were also

some intriguing differences. First, Drosophila Wnt5 clearly acts as a repellent cue for com-missural axons as they cross the midline, whereas mammalian Wnt4 instead seems to guide commissural axons by attraction, and only after they have crossed. Second, whereas Drosophila Wnt5 acts through a Ryk fam-ily receptor, mammalian Wnt4 seems to act through Fz receptors, including Fz3. Thus, a model emerged in which various Wnt proteins might act either as axonal attractants or repel-lents, depending upon whether a Fz or a Ryk receptor was involved, respectively4.

Guided by such a model, Liu et al.2 hypoth-esized that if Fz receptors could attract axons up a Wnt gradient in the spinal cord, then Ryk receptors might repel axons down a Wnt gra-dient. To test this idea, they focused on axons of the corticospinal tract, a major descending pathway from the brain that courses through the dorsal region of the spinal cord (Fig. 1b). As it turned out, their prediction was spot on. They first found that several other Wnts are indeed expressed in an anterior-high to poste-rior-low gradient in the neonatal dorsal spinal cord, and that two of these Wnts—Wnt1 and Wnt5a—are potent repellents for corticospinal tract axons in vitro. Dorsal spinal cord explants themselves also repel these axons in vitro, and just like Wnt1 and Wnt5a expression, this effect tapers off in more caudal regions of the spinal cord (Fig. 1b).

The authors then went on to show that Ryk, the vertebrate homolog of Drosophila

Drl, is indeed expressed on corticospinal tract axons. Moreover, by using anti-Ryk antibodies to block Ryk function, they show that Ryk is indeed required for the repulsion of cortico-spinal tract axons in vitro, in response both to Wnts and to dorsal spinal cord explants. Finally, Liu et al. show that injection of anti-bodies to Ryk directly into the spinal cord interferes with the posterior growth of cor-ticospinal tract axons. Overall, this series of experiments make a compelling case that endogenous Wnts, most likely including Wnt1 and Wnt5a, act through Ryk to help guide cor-ticospinal tract axons down the spinal cord. In the future, genetic manipulations of the Wnt1, Wnt5a and Ryk genes will be required to con-firm and extend these findings.

One intriguing aspect of this study is that corticospinal tract axons grow for some dis-tance in the brain before they enter the spinal cord, and so actually have to grow up a very steep Wnt gradient as they first enter the cord. This is not the response one would expect these axons to show when confronted with such a potent repellent! The explanation, it seems, is that young corticospinal tract axons are evidently insensitive to the repellent action of Wnts, and only acquire this sensitivity at about the time they enter the spinal cord2. Presumably, by this time corticospinal tract axons have passed the point of no return, and now have no choice but to continue on down the spinal cord. Such tight temporal regulation of guidance responses is a recurring theme in axon guidance, and has been particularly well documented for commissural5,6 axons. These axons switch their responses to several guid-ance cues as they cross the CNS midline, pos-sibly even becoming attracted by Wnt4 only after crossing1. In this regard, it is interesting to note that corticospinal tract axons also cross the midline just before they enter the spinal cord, and so it is tempting to speculate that midline crossing of corticospinal tract axons might similarly trigger the onset of Ryk expres-sion and aversion to Wnts.

This new work also adds to the growing evidence that Ryk proteins, like Fz proteins, are receptors for at least some of the Wnts3,7,8. An open issue, and a matter of some debate, is whether Ryk and Fz proteins are part of the same multifunctional receptor complex or are independent receptors with distinct functions. A recent report7 has argued that Ryk and Fz are part of the same receptor complex, though it should be noted that the case rests largely on evidence from overexpression studies in cell culture. In contrast, the studies discussed above clearly indicate that the axon guidance functions of Ryk and Fz receptors are mutu-ally independent1–3. A similar conclusion has

a b

DorsalVentral

E11.5

P0

Commissuralneuron

Fz3

Wnt4

CSTneuron

RykWnt1 and

Wnt5a

Wnt1andWnt5a

Corticalexplants

Dorsal spinalcord explants

CST

Figure 1 Wnts guide axons up and down the spinal cord. (a) Commissural axons turn anteriorly after crossing the floor plate. This turn seems to be mediated in part by Wnt4, and possibly by other Wnts, which are expressed in an anterior-posterior gradient in the floor plate and signal attraction through the Fz3 receptor1. (b) After passing through the mid- and hindbrain, corticospinal tract axons cross the midline and grow down the spinal cord in the dorsal funiculus. Growth down the spinal cord seems to be mediated in part by Wnt1 and Wnt5a, which are expressed in an anterior-posterior gradient in the dorsal spinal cord and signal repulsion through the Ryk receptor2.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



GABA puts the brake on stem cellsArnold R Kriegstein

In the adult brain, new neurons are generated from neural stem cells residing in the subventricular zone. Newborn neuroblasts release the transmitter GABA, which reduces the proliferation of stem cells—and thereby neurogenesis—by a nonsynaptic mechanism.

Arnold Kriegstein is at the Department of Neurology

and the Program in Developmental and Stem Cell

Biology, University of California, San Francisco,

California 94143, USA.


Uncontrolled proliferation of immature cell types can have devastating consequences, including cancer. Tight regulation of neuro-genesis in the adult brain is therefore essential. Neurons destined for the olfactory bulb are produced in the subventricular zone (SVZ) and added continuously through adulthood. However, the regulatory mechanisms that control neurogenesis are poorly understood. A variety of signaling molecules, including EGF, Shh, BMPs and the Eph/ephrin family, promote neurogenesis in SVZ cells, and at least one pathway, involving Notch signaling, seems to suppress SVZ neurogenesis (for review, see ref. 1). In this issue, Liu and colleagues2 use a wide variety of electrophysiological approaches to show that nonsynaptic, nonvesicular release of the neurotransmitter GABA provides nega-tive feedback to neural stem cells, inhibiting their production of new neuroblasts through inhibition of cell-cycle reentry.

These results build upon prior observations on the role of GABA in both embryonic and

adult neurogenesis. During embryonic stages of cortical development, GABA influences the proliferation of neural progenitor cells. Its net effect is to inhibit the number of cells entering the cell cycle3,4. GABA affects embry-onic neurogenesis at ages after GABAergic neurons are produced, but before the devel-opment of synaptic GABA transmission. The new report is consistent with these studies of embryonic neurogenesis, showing a GABA-mediated feedback regulation of adult neu-rogenesis controlled by nonsynaptic GABA release. This is also in line with observations that elimination of neuroblasts stimulates the proliferation of GFAP-positive neural stem cells, suggesting that a feedback signal pro-duced by neuroblasts may normally inhibit stem cell proliferation5. SVZ neuroblasts can synthesize and release GABA6,7, and the current report extends these observations by providing evidence that GABA may act as an inhibitory feedback signal to suppress neural stem cell proliferation. It is now clear that this neurotransmitter, once thought only to medi-ate signaling at synapses, is also important for regulating neurogenesis at both adult and embryonic ages.

Liu and colleagues very nicely characterize the electrophysiological properties of neu-

roblasts and GFAP-positive astrocyte-like neural stem cells. They find that the GFAP-positive cells are gap junction–coupled into small clusters and express functional GABAA receptors. They further show that nonsynaptic, nonvesicular GABA release by neuroblasts (type A cells) inhibits prolifera-tion of GFAP-positive progenitors (B cells; Fig. 1). The new findings, as with many new observations, raise more questions than they answer. One question concerns the role of C cells in the GABA regulatory pathway. These GFAP-negative cells are generated by B cells and serve as intermediates in neuroblast gen-eration (Fig. 1). The C cells are thought to act as transit-amplifying cells and are known to undergo rapid cell cycling. There are few reliable characteristics that can help identify C cells at present, perhaps the best being the highly invaginated nucleus with a reticulated nucleolus8, but this feature is best visualized by electron microscopy. Do the C cells syn-thesize and release GABA? Do they respond to GABA? Does GABA regulate C cell pro-liferation? Answering these questions will be difficult until better cell-specific markers are available to identify these cells in vivo. Another question involves the importance of GABA-mediated feedback regulation in vivo.

also come from recent studies of Ryk and Fz function in cell fate specification in the Caenorhabditis elegans vulva8. Thus, on bal-ance, the evidence seems to favor a model in which the two receptors can signal indepen-dently. Nevertheless, comparisons to another well-known family of bifunctional guidance cues, the netrins, may be helpful. Like Wnt proteins, netrins also signal attraction and repulsion through different receptors—DCC and Unc5, respectively. Although these two receptors can clearly act on their own9–11, they may also sometimes function together as part of a single receptor complex11–13.

Another important area for further research will be to define the signaling pathways that act downstream of Fz and Ryk receptors in axon guidance. Fz proteins activate at least three dis-tinct pathways, of which the so-called ‘planar cell polarity’ and ‘calcium’ pathways provide the more plausible links to the cytoskeleton14.

Signaling pathways for Ryk receptors are still unknown. These proteins have a cytoplasmic domain that resembles tyrosine kinases, but evidently lacks kinase activity. This cytoplas-mic domain is required in Drl for axon guid-ance in Drosophila15, but not in Lin-18/Ryk for cell fate specification in C. elegans8. Clearly, there is still a lot of work to do, and it would certainly help to develop in vitro growth cone turning assays for Wnts—a method that has proven particularly useful for other guidance cues such as netrins.

Finally, might these new results be of any help in efforts to develop therapies for the treat-ment of spinal cord injuries? It remains to be seen how, or whether, Wnt proteins and their Fz and Ryk receptors might act in the adult spinal cord. But, surely, learning how axons grow up and down the spinal cord during development can only increase the prospects for encouraging severed axons to do the same in the adult.

1. Lyuksyutova, A.I. et al. Science 302, 1984–1988 (2003).

2. Liu, Y. et al. Nature Neuroscience 8, 1151–1159 (2005).

3. Yoshikawa, S., McKinnon, R.D., Kokel, M. & Thomas, J.B. Nature 422, 583–588 (2003).

4. Imondi, R. & Thomas, J.B. Science 302, 1903–1904 (2003).

5. Shirasaki, R., Katsumata, R. & Murakami, F. Science 279, 105–107 (1998).

6. Zou, Y., Stoeckli, E., Chen, H. & Tessier-Lavigne, M. Cell 102, 363–375 (2000).

7. Lu, W., Yamamoto, V., Ortega, B. & Baltimore, D. Cell 119, 97–108 (2004).

8. Inoue, T. et al. Cell 118, 795–806 (2004).9. Hedgecock, E.M., Culotti, J.G. & Hall, D.H. Neuron 4,

61–85 (1990).10. Stein, E., Zou, Y., Poo, M. & Tessier-Lavigne, M.

Science 291, 1976–1982 (2001).11. Keleman, K. & Dickson, B.J. Neuron 32, 605–617

(2001).12. Colavita, A. & Culotti, J.G. Dev. Biol. 194, 72–85

(1998).13. Hong, K. et al. Cell 97, 927–941 (1999).14. Zou, Y. Trends Neurosci. 27, 528–532 (2004).15. Yoshikawa, S., Bonkowsky, J.L., Kokel, M., Shyn, S. &

Thomas, J.B. J. Neurosci. 21, RC119 (2001).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



The proposed model is not entirely consis-tent with a previous observation of adult neurogenesis after bulbectomy. Surgical removal of the olfactory bulb results in the accumulation of neuroblasts in the rostral migratory stream (RMS) and SVZ9. Despite the buildup of neuroblasts, proliferation is maintained for weeks, resulting in a dramatic increase in the size of the RMS and SVZ9. These observations seem to argue that feed-back inhibition may not exert a major regu-latory influence on neuroblast production in vivo. However, the role of GABA in such scenarios remains unclear. It will be impor-tant in future studies to examine GABA regu-lation of neurogenesis in vivo.

How does one reconcile the inhibitory effect of GABA on cell proliferation with the observation (reported by the same authors) that GABA inhibits neuroblast migration in the adult SVZ10? These results would argue for a model where GABA released by newly generated neuroblasts inhibits the further

generation of new neuroblasts and prevents the old ones from migrating away. This would result in stasis of neurogenesis. It is possible, however, that GABA levels could fine-tune the balance between proliferation and migration, but this would necessitate complicated spatial regulation of GABA levels. Interestingly, the neural stem cells themselves (B cells) express a high-affinity GABA transporter, GAT4, on the processes that ensheathe the migrating neuroblasts, and it has been suggested that activity of this transporter may regulate GABA concentra-tion in the microenvironment adjacent to migrating neuroblasts10. This makes for an interesting scheme in which GABA effects on proliferation and migration are mediated by receptors on stem cells and neuroblasts, respectively, and stem cells and neuroblasts jointly regulate the level of activation.

It will be interesting to investigate the downstream activators that translate GABAA receptor activation into an effect on cell pro-liferation. Unlike adult neurons that are inhib-ited by GABA activation, embryonic neurons are generally excited by GABA11,12. In cells at all ages, GABAA receptor activation gates a chloride channel, but whereas embryonic cells have a high intracellular chloride concentra-tion, mature cells have a lower chlor-ide con-centration owing to the action of a chloride transporter not found in immature cells13. As a result of the change in the intra- to extracellular chloride gradient, GABAA receptor activation depolarizes immature cells but hyperpolarizes mature cells. Consistent with this maturational change, GABA depolarizes immature neuro-blasts in the postnatal SVZ and rostral migra-tory stream7, suggesting that GABA-induced effects on neuroblast migration are likely to be governed by a depolarization-mediated mech-anism. Given this developmental sequence, it would be particularly interesting to assess the chloride gradient in the GFAP-positive adult stem cells and, consequently, whether GABA excites (depolarizes) or inhibits (hyperpolar-izes) them. Or, stated another way, is GABA’s effect on proliferation actually excitation-induced inhibition of cell cycle progression? If adult neural stem cells resemble embryonic neural stem cells, then the effect of GABA on GFAP-positive stem cells will involve GABA-induced depolarization and calcium entry through voltage-gated channels3.

During embryonic stages of develop-ment, neurotransmitters are released by neurons before synapses form. The mecha-nism responsible for nonsynaptic transmit-ter release in embryonic neurons is unknown, but it seems to be the same unconventional release mechanism as that described by Liu

et al. in adult-generated neuroblasts2. For example, nonsynaptic transmitter release in embryonic cortical neurons also does not involve calcium influx or vesicular machin-ery, and the kinetics of release are much lon-ger than at synaptic contacts, producing slow tonic currents in responding cells3,14,15. The release mechanism remains mysterious. As shown by Liu et al., one can be more confi-dent of what it is not, including that it is not vesicular, not dependent on extracellular cal-cium, not dependent on hemiconnexin chan-nel opening and not transporter mediated, than of what it may be. Evidence presented here indicates that neuroblast depolarization might induce GABA release, but the ques-tion is then, what triggers the spontaneous depolarizations in neuroblasts? The release mechanism and related machinery, therefore, remain an intriguing puzzle.

The neurotransmitter GABA turns out to be a versatile molecule with diverse roles in both the developing and adult nervous system. In early development, nonsynap-tic GABA seems to act as a paracrine signal mediating growth processes, including pro-liferation, migration and differentiation11,12. As demonstrated here, many of the functions of GABA in the embryonic brain also apply to the germinal centers in the adult brain. In parallel to its effect during development, GABA released spontaneously by adult-gen-erated neuroblasts activates adult neural stem cells and inhibits their proliferation. These observations suggest a developmentally con-served role of GABA as a paracrine feedback signal regulating neurogenesis.

1. Alvarez-Buylla, A. & Lim, D.A. Neuron 41, 683–686 (2004).

2. Liu, X., Wang, Q., Haydar, T.F. & Bordey, A. Nat. Neurosci. 8, 1179–1187 (2005).

3. LoTurco, J.J., Owens, D.F., Heath, M.J.S., Davis, M.B.E. & Kriegstein, A.R. Neuron 15, 1287–1298 (1995).

4. Haydar, T.F., Wang, F., Schwartz, M.L. & Rakic, P. J. Neurosci. 20, 5764–5774 (2000).

5. Doetsch, F., Caille, I., Lim, D.A., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Cell 97, 703–716 (1999).

6. Stewart, R.R., Hoge, G.J., Zigova, T. & Luskin, M.B. J. Neurobiol. 50, 305–322 (2002).

7. Wang, D.D., Krueger, D.D. & Bordey, A. J. Physiol. (Lond.) 550, 785–800 (2003).

8. Doetsch, F., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. J. Neurosci. 17, 5046–5061 (1997).

9. Kirschenbaum, B., Doetsch, F., Lois, C. & Alvarez-Buylla, A. J. Neurosci. 19, 2171–2180 (1999).

10. Bolteus, A.J. & Bordey, A. J. Neurosci. 24, 7623–7631 (2004).

11. Represa, A. & Ben-Ari, Y. Trends Neurosci. 28, 278–283 (2005).

12. Owens, D.F. & Kriegstein, A.R. Nat. Rev. Neurosci. 3, 715–727 (2002).

13. Rivera, C. et al. Nature 397, 251–255 (1999).14. Demarque, M. et al. Neuron 36, 1051–1061

(2002).15. Flint, A.C., Liu, X. & Kriegstein, A.R. Neuron 20,

43–53 (1998).

Figure 1 The production of new neurons in the adult SVZ begins when a neural stem cell (B cell) divides to generate a transit-amplifying cell (C cell). The stem cells are astrocyte-like cells that express glial fibrillary acid protein (GFAP)-containing intermediate filaments. The transit-amplifying cells divide to produce neuroblasts (A cells) that then migrate along the RMS to reach the olfactory bulb, where they contribute to the population of granule cells. Liu et al. provide evidence that neuroblasts inhibit their own production by releasing GABA, which inhibits B cell division and thereby applies a brake on neurogenesis. The proposed model depends upon the migration of neuroblasts away from the region of neurogenesis in order to release the proliferation brake. However, a previously characterized antimigration effect of GABA would seem to complicate this otherwise attractive scheme.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Making the causal link: frontal cortex activity and repetition primingAlex Martin & Stephen J Gotts

Object identification improves with repeated presentation, but neural activity decreases. In a new study, disrupting inferior frontal activity with transcranial magnetic stimulation during initial exposure to an object blocks later behavioral and neural changes.

As we all know, many things are easier the second time around. That includes object identification, which is faster and more effi-cient when we see an object for the second time. This behavioral phenomenon, termed repetition priming, is one of the most inter-esting and actively investigated forms of memory1. Priming is preserved in patients who, as a result of medial temporal lobe damage, are unable to consciously retrieve objects they have seen before. However, although repetition improves performance, on the neural level, priming is typically asso-ciated with decreased activity2. This paradox-ical coupling of reduced neural activity with improved behavior has attracted consider-able interest, as investigators have attempted to forge direct links between changes at the behavioral and neural levels. Imaging stud-ies show significant correlations between neural decreases in specific brain regions and priming3,4. However, as with all brain imaging studies, this relationship is neces-sarily indirect because it is based solely on correlation. Thus, whether decreased neu-ral activity causes priming has remained very much an open question. Now, by using transcranial magnetic stimulation (TMS) to transiently perturb neural processing, Wig and colleagues in this issue5 provide evidence for a causal link between decreased activity and priming.

The Wig et al. study involved three phases. In the first phase, the neural system associ-ated with priming was identified using event-related fMRI (functional magnetic resonance imaging). To accomplish this goal, subjects were shown a series of object pictures before scanning and asked to classify each concep-tually as living or non-living. They were then scanned while performing the same concep-tual task on already seen (repeated) objects, as well as on new objects. As expected, both behavioral priming (faster classification

of repeated objects), and ‘neural priming’ (less neural activity for repeated objects) were observed. In agreement with previous reports6, neural priming was seen through-out nearly all brain regions associated with the processing of common visual objects, including bilateral portions of the occipital lobe associated with perceptual processing and left posterior temporal and inferior frontal regions associated with conceptual processing (Fig. 1).

The second phase of the experiment took place approximately one week later. TMS was administered while subjects made liv-ing/non-living judgments on a new set of objects. The left inferior frontal cortex was chosen as the primary site of interest on the basis of its involvement in conceptual processing, and TMS was applied here dur-

ing the presentation of some objects. TMS was also applied to a control site—the hand region of left motor cortex—during presen-tation of the other objects. Importantly, TMS administration was tailored for each subject in two ways. A frameless stereotaxic system was used to allow the investigators to place the TMS probe directly over the point in the subject’s inferior frontal cortex that showed the largest repetition effect in the first, scan-ning phase. The investigators also used the subject’s behavioral performance from the first phase to begin administering TMS approximately 250 ms before each subject’s typical response time.

TMS did not prevent subjects from per-forming the task accurately, regardless of the site of administration. The investiga-tors then evaluated the impact of TMS on

Alex Martin and Stephen Gotts are in the Laboratory

of Brain and Cognition, National Institute of Mental

Health, Building 10 Room 4C-104, Bethesda,

Maryland 20892-1366, USA.


Figure 1 Summary of the neuroimaging findings. After control TMS was applied to left motor cortex (TMS-C), reduced activity in response to repeated (R) compared to novel (N) objects (‘neural priming’) was observed in regions associated with visual (occipital cortex, yellow) and conceptual (temporal, orange) object processing, and in the left inferior frontal cortex (red), associated with conceptual and lexical processing and selection. After TMS was applied to left inferior frontal cortex (TMS-F), behavioral priming was eliminated, as was neural priming in left inferior frontal and temporal cortices, but not in occipital cortex. The common pattern of neural priming in frontal and temporal cortices suggests a functional interaction between these two regions.

TMS-C TMS-F

N NR R

TMS-C TMS-F

N NR R

TMS-C TMS-F

N NR R

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



both behavioral and neural priming in a final scanning phase a few minutes later. The logic here is that because priming is presumed to result from events triggered during the initial processing (encoding) of an object, the consequences of TMS during encoding should be observable at a later time. During scanning after TMS, subjects again performed the same living/non-living task either on objects previously encoded during TMS or on new objects.

The findings were clear and striking. Behavioral priming was eliminated for objects encoded during TMS of left inferior frontal cortex, but not for objects encoded during stimulation of left motor cortex. Moreover, priming-related neural activity decreases were eliminated, but only in certain brain regions. Activity decreases were main-tained in regions of occipital cortex known to be associated with visual processing of object features, whereas decreases were not observed in either left inferior frontal cor-tex (the TMS site) or left posterior temporal cortex (Fig. 1). These results demonstrate a causal role for left inferior frontal cortex in at least one form of repetition priming. They also demonstrate a link between decreased neural activity and priming. Nevertheless, these two phenomena could be dissociated. Repetition-related decreases continued to occur in occipital cortex even when behav-ioral priming was absent.

The Wig et al. study5 is a tour de force with regard to the logic of its design and execution. To accomplish the authors’ goal required multiple behavioral and functional imaging sessions and the individual tailoring of TMS administration with regard to both site and timing of stimulation. As a result of their efforts, we now have evidence from the normal human brain of a causal relationship between the neural and behavioral aspects of priming. It is noteworthy, as well, that their finding con-verges nicely with those of fMRI correlational studies showing a stronger coupling between behavioral and neural priming in left inferior frontal than in posterior cortices3,4, thus sug-gesting that we are on the right track.

Nevertheless, many important issues remain to be resolved. First, converging evi-dence is needed from patients with focal left inferior frontal lesions (Broca’s aphasics). Broca’s aphasics show intact priming on some procedures, but not on others7,8. However, in contrast to the study by Wig et al., the prim-

ing procedure used in these studies have been markedly different with regard to stimuli (words rather than pictures) and task. Based on the Wig et al. finding, we would expect that the ability of these patients to classify object pictures as living or non-living would not improve with repetition. This possibility needs to be confirmed.

Several other predictions can be made. The procedure used by Wig et al. is a version of conceptual repetition priming9. The main features of this procedure are that the same object is presented at study and test and that the orienting task used at study and at test is conceptual—requiring a decision based on the meaning of the object—rather than per-ceptual. From a processing point of view, the task requires engagement of visual processing mechanisms largely located in occipital cor-tices10, access to conceptual object informa-tion largely stored in temporal cortices11 and engagement of left inferior frontal mecha-nisms for selecting and retrieving this infor-mation12. Viewed within this framework, one of the more intriguing aspects of the Wig et al. results is the lack of neural priming in left posterior temporal cortex induced by left inferior frontal TMS. Considerable neuro-psychological and neuroimaging evidence suggests left posterior temporal cortex is crit-ically involved in conceptual processing of objects, perhaps because information about object category–related properties is stored there11. For subjects to accurately classify the objects as living or nonliving requires access to the conceptual information represented in this region of the brain. The Wig et al. find-ing suggests that this information is accessed through the top-down influence of left infe-rior frontal cortex and that behavioral prim-ing is mediated by the interaction of these two regions. As a result, TMS to left posterior temporal region should show effects similar to those demonstrated for inferior frontal TMS: behavioral priming and neural prim-ing in left posterior temporal and frontal cortex, but not occipital cortex, should be eliminated.

But what about occipital cortex? In the Wig et al. study, neural priming was maintained in this brain region. One possible explanation for this finding is that perceptual priming is independent of the modulatory influence and control of frontal lobe mechanisms. A different possibility is that TMS applied to frontal cortex could eliminate neural priming

in occipital cortex if the orienting task were focused more exclusively on visual dimensions (for instance, a task requiring a difficult visual form discrimination).

Another important issue that remains to be explored is the effect of inferior frontal TMS on other forms of object memory. As the authors state, after TMS, subjects responded to the repeated presentations of the objects as if they were seeing them for the first time. However, would explicit memory be similarly impaired? Although there is abundant evidence showing that priming and explicit memory can be dis-sociated, TMS to inferior frontal cortex during encoding may disrupt both forms of memory, especially because the amount of left inferior frontal cortex activity during encoding is a strong predictor of whether an item will be later remembered13. The procedure developed by Wig et al. could be used to directly test the effects of inferior frontal TMS at encoding for both priming and remembering.

A final major puzzle remains. Wig. et al. have shown that disrupting one part of the circuitry that typically shows neural priming can disrupt behavioral priming. Nevertheless, how repetition-related decreases in neural activation cause behavioral facilitation is still unknown. To resolve this vexing ques-tion will require a clearer understanding of mechanism(s) producing repetition-related decreases in neural activity in different regions of the brain2,14,15.

1. Tulving, E. & Schacter, D.L. Science 247, 301–306 (1990).

2. Henson, R.N.A. Prog. Neurobiol. 70, 53–81 (2003).3. Dobbins, I.G., Schnyer, D.M., Verfaellie, M. & Schacter,

D.L. Nature 428, 316–319 (2004).4. Maccotta, L. & Buckner, R.L. J. Cognit. Neurosci. 16,

1625–1632 (2004).5. Wig, G.S., Grafton, S.T., Demos, K.E. & Kelley, W.M.

Nat. Neurosci. 8, 1228–1233 (2005). 6. van Turennout, M., Ellmore, T. & Martin, A. Nat.

Neurosci. 3, 1329–1334 (2000).7. Hagoort, P. Brain Lang. 56, 287–300 (1997).8. Blumstein, S.E. et al. Brain Lang. 72, 75–99

(2000).9. Wagner, A.D., Desmond, J.E., Demb, J.B., Glover, G.H.

& Gabrieli, J.D.E. J. Cognit. Neurosci. 9, 714–726 (1997).

10. Grill-Spector, K., Kushnir, T., Edelman, S., Itzchak, Y. & Malach, R. Neuron 21, 191–202 (1998).

11. Martin, A. & Chao, L.L. Curr. Opin. Neurobiol. 11, 194–201 (2001).

12. Thompson-Schill, S.L., D’Esposito, M., Aguirre, G.K. & Farah, M.J. Proc. Natl. Acad. Sci. USA 94, 14792–14797 (1997).

13. Wagner, A.D. et al. Science 281, 1188–1191 (1998).

14. Desimone, R. Proc. Natl. Acad. Sci. USA 93, 13494–13499 (1996).

15. Wiggs, C.L. & Martin, A. Curr. Opin. Neurobiol. 8, 227–233 (1998).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Pathological tau: a loss of normal function or a gain in toxicity?John Q Trojanowski & Virginia M-Y Lee

Neurofibrillary tangles, composed of tau protein, are a central feature of Alzheimer disease. A new paper challenges the idea that these tau inclusions alone cause disease by showing that they can be dissociated from memory impairment and neuronal loss.

Neurofibrillary tangles and senile plaques, the two diagnostic hallmarks of Alzheimer disease, are formed by intra-neuronal accu-mulation of abnormal tau filaments and extracellular deposits of Aβ fibrils, respec-tively1–4, which compromise the function and viability of neurons. Although a prevalent hypothesis is that misfolding and fibrillization of normal brain proteins causes their accumu-lation as disease-specific inclusions, the exact mechanisms by which brain degeneration in Alzheimer disease results from neurofibril-lary tangles and senile plaques remains poorly understood. Now, a new paper by SantaCruz et al.5 challenges the notion that tangles in neuronal perikarya alone cause disease, and at face value, seems to refute the notion that tau abnormalities cause the disease.

Mutations in the genes encoding tau, Aβ precursor proteins (APP) and presenilins are pathogenic for familial neurodegen-erative disorders. Drug discovery efforts for Alzheimer disease have focused mainly on targets defined by the Aβ hypothesis, though the normal function of Aβ remains an enigma4,6–8. In contrast, the tau hypothesis of Alzheimer disease neurodegeneration is based not only on studies of neurofibrillary tangles but also on the normal biology of tau2,4,9. Normal tau binds to and stabilizes microtubules, and phosphorylation of tau negatively regulates its binding to microtu-bules (Fig. 1). The paired helical filaments (PHFs) that form the neurofibrillary tangles seen in Alzheimer disease are abnormally phosphorylated tau (known as PHFtau), which loses the ability to bind to and stabilize microtubules2,4,9. However, although the bur-den of neurofibrillary tangles correlates well with the progression of the disease, over 90% of PHFtau is deposited in neuronal processes as dystrophic neurites and less than 10% is in

neurofibrillary tangles10. Thus, the deleteri-ous effects of tau pathology may well be due to a toxic gain of function by neurofibrillary tangles, but tau pathology could also reflect a loss-of-function defect.

Critical support for the tau hypothesis of neurodegeneration came from a series of dis-coveries showing that mutations in the gene encoding tau were pathogenic for hereditary frontotemporal dementia with parkinsonism linked to chromosome 17 (FTDP-17), a het-erogeneous group of disorders characterized by prominent tau pathologies, but no senile plaques or other disease-specific inclusions2–4. A number of tau gene mutations pathogenic for FTDP-17 cause losses of tau function (that is, a loss of the ability to bind to and promote the assembly of microtubules) and/or gains of potentially toxic properties by mutant tau isoforms (an increased amyloidogenic propensity)2,4. Neurofibrillary tangles and other tau pathologies also induce neurode-generation in worm, fly and mouse models of neurodegenerative tauopathies with persua-sive verisimilitude to their authentic human counterparts2,4.

The new study by SantaCruz et al.5 chal-lenges the idea that the neurodegenerative sequellae of pathological tau are due to the formation of neurofibrillary tangles alone5. The authors generated an inducible tau trans-genic mouse model, in which the expression of human tau with the P301L FTDP-17 mutation was regulated by tetracycline. These trans-genic mice developed memory impairments (as assessed in the Morris water maze) at 2.5 months, overt pathological tau deposits by 4 months and gross brain atrophy with abun-dant neurofibrillary tangles by 10 months of age. However, when the authors turned off P301L mutant tau expression by repressing the tetracycline promoter with doxycycline for varying lengths of time in multiple separate cohorts of transgenic mice at different ages beginning at 2.5 months, neurofibrillary tan-gle accumulation could be dissociated from the memory impairments and neuron loss. The mice then showed improved memory and a rescue of neuronal loss, but the accu-mulation of neurofibrillary tangles continued.

Thus, these data and other biochemical data on the tau pathology in these transgenic mice suggest that if the aggregation of abnormal tau filaments into neurofibrillary tangles is asso-ciated with a toxic gain of function, this alone may not be sufficient to cause the degenera-tion of neurons and the associated memory impairments seen in Alzheimer disease and related tauopathies, at least in transgenic mice. Indeed, SantaCruz et al. conclude that because memory function improved but neu-rofibrillary tangles continued to accumulate in transgenic mice treated with doxycycline at 4 months of age and later, the presence of the neurofibrillary tangles that remain after suppressing human mutant tau expression is insufficient to disrupt cognitive function. Further, these data also imply that some of the toxic consequences of tau pathology may be reversible, thereby opening up new avenues of tau-focused drug discovery for Alzheimer disease and other tauopathies.

How do these results fit with the prevailing tau hypothesis of Alzheimer disease? Because SantaCruz et al. did not test whether turning off mutant tau expression had any effect on neurofibrillary dystrophic tau neurites in the transgenic mice, it is plausible that these effects resulted from the elimination of tau pathologies in neuronal processes while the neurofibrillary tangles persisted in neuronal perikarya. These uncertainties notwithstand-ing, a significant prediction of the tau hypoth-esis of Alzheimer disease neurodegeneration is that by converting normal tau into function-ally impaired PHFtau, microtubules would be depolymerized, thereby disrupting axonal transport and compromising the function as well as the viability of affected neurons4,9. Thus, rather than being a consequence of a toxic gain of function by PHFtau in neurofi-brillary tangles, pathological tau may induce neurodegeneration through loss-of-function defects that would have several deleterious consequences—for example, impairing intra-neuronal transport or the function of organelles, such as the Golgi apparatus, that depend on the integrity of the microtubule network. Thus, the work of SantaCruz et al. is likely to focus attention on the compelling

John Trojanowski and Virginia Lee are at the

Center for Neurodegenerative Disease Research,

Department of Pathology and Laboratory

Medicine, and the Institute on Aging, University

of Pennsylvania School of Medicine, Philadelphia,

Pennsylvania 19104, USA.


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



possibility that pathological tau may induce neurodegeneration owing to loss-of-function defects by PHFtau rather than as a result of toxic gains of functions by accumulating neurofibrillary tangles.

Significantly, these and other insights into mechanisms of neurodegeneration in Alzheimer disease and related tauopathies suggest that tau pathologies may be tract-able or ‘druggable’ targets for the discovery of new therapies for these conditions4,6. For example, microtubule-stabilizing drugs could provide therapeutic benefits in these disor-

ders by offsetting the loss of tau function by causing its sequestration in neurofibril-lary tangles and dystrophic neurites, or its excessive phosphorylation11, whereas micro-tubule-stabilizing drugs such as paclitaxel also may ameliorate the neurotoxic effects of Aβ in Alzheimer disease12. Moreover, addi-tional strategies to develop therapies that tar-get tau abnormalities are being investigated, including high-throughput screening of large compound libraries to identify drugs that not only block tau fibrillization and aggregation but also reduce tau abundance or tau phos-

phorylation13–15. Notably, like microtubule-stabilizing drugs, inhibitors of tau kinases, such as glycogen synthase kinase-3, may be able to block pathways that lead to the for-mation of both Aβ and tau amyloid lesions in Alzheimer disease4.

Perhaps most importantly, the work of SantaCruz et al. suggests that rectifying losses of tau function may be as valuable as, or more valuable than, ameliorating gains of toxic properties by neurofibrillary tangles in drug-discovery programs for Alzheimer dis-ease and related tauopathies. Indeed, if this lesson can be extrapolated to other neuro-degenerative brain amyloidoses, researchers may now need to pay more attention to the normal functions of disease proteins such as tau, Aβ, α-synuclein, prions, and so on to fully understand the role of pathologically altered forms of these proteins in mechanisms of neurodegeneration.

1. Dobson, C.M. Nature 426, 884–890 (2003).2. Forman, M.S., Trojanowski, J.Q. & Lee, V.M.-Y. Nat.

Med. 10, 1055–1063 (2004).3. Selkoe, D.J. Nat. Cell Biol. 6, 1054–1061 (2004).4. Skovronsky, D.M., Lee, V.M.-Y. & Trojanowski, J.Q.

Annu. Rev. Pathol. (in the press).5. SantaCruz, K. et al. Science 309, 476–481 (2005).6. Fillit, H.M. & Refolo, L.M. Curr. Alzheimer Res. 2,

105–109 (2005).7. Lansbury, P.T. Jr. Nat. Med. 10 (Suppl.), S51–S57

(2004).8. Xiong, G. & Doraiswamy, P.M. Geriatrics 60, 22–26

(2005).9. Lee, V.M.-Y. & Trojanowski, J.Q. Curr. Opin. Neurobiol.

2, 653–656 (1992).10. Mitchell, T.W. et al. J. Histochem. Cytochem. 48,

1627–1638 (2000).11. Zhang, B. et al. Proc. Natl. Acad. Sci. USA 102,

227–231 (2005).12. Michaelis, M.L. et al. J. Pharmacol. Exp. Ther. 312,

659–668 (2005).13. Dickey, C.A. et al. Curr. Alzheimer Res. 2, 231–239

(2005).14. Pickhardt, M. et al. Curr. Alzheimer Res. 2, 219–226

(2005).15. Noble, W. et al. Proc. Natl. Acad. Sci. USA 102,

6990–6995 (2005).

Tau aggregation/hyperphosphorylationDecreased MT-bound tauMT depolymerizationImpaired axonal transportStasis and aggregation of axonal trafficSynaptic dysfunctionAxonal degeneration and disconnection

TauVesiclesPathological aggregates ofPHFtau as amyloid fibrils

Ann Thomson

Figure 1 A schematic of the misfolding, fibrillization and sequestration of tau into neurofibrillary tangles and neuritic tau pathology. Top, a normal neuron, in which normal tau maintains the network of microtubules (MTs) that support intraneuronal transport. Bottom, a diseased neuron, in which misfolded tau has formed amyloid fibrils that deposit as neurofibrillary tangles and neuritic tau pathology. The deleterious effects of tau pathology could result from toxic gains of functions by neurofibrillary tangles and neuritic tau pathology or from a loss of tau functions.

Imaging heterogeneity in synaptic transmissionNeurons can form synapses with just one postsynaptic target or with hundreds, and an individual neuron can make single or multiple connections with each postsynaptic cell. To add to this diversity, connections also vary in strength. These multiple variables are difficult to measure electrophysiologically. On page 1188 of this issue, Guerrero et al. report a new optical approach to compare transmission from different connections of an individual neuron in the neuromuscular junction of the fruitfly Drosophila.

The authors created an innovative genetically encoded calcium reporter called SynapCam to selectively study calcium influx through glutamate receptors (which in Drosophila are permeable to Ca2+). They were able to target SynapCam postsynaptically, and image single boutons. They report that calcium signals through glutamate receptors were uniform within a given connection but varied considerably among different connections made by a single axon at the same neuromuscular junction. Moreover, they observed a gradient of transmission strength along axonal branches, with the strongest synapses at the terminal end of an axonal branch (picture).

Kalyani Narasimhan

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Sleep-disordered breathing aftertargeted ablation of preBotzingercomplex neuronsLeanne C McKay1,2, Wiktor A Janczewski1,2 & Jack L Feldman1

Ablation of preBotzinger complex (preBotC) neurons, critical

for respiratory rhythm generation, resulted in a progressive,

increasingly severe disruption of respiratory pattern, initially

during sleep and then also during wakefulness in adult rats.

Sleep-disordered breathing is highly prevalent in elderly

humans and in some patients with neurodegenerative disease.

We propose that sleep-disordered breathing results from loss of

preBotC neurons and could underlie death during sleep in

these populations.

Breathing is an essential regulatory behavior, continuously activethroughout wakefulness and sleep from birth until death. Comparedto breathing during wakefulness, however, breathing during sleep isfragile, often disordered and frequently interrupted by apneic episodes.Sleep-disordered breathing (SDB) can result in intermittent hypoxia,the cumulative effect of which includes loss of brain gray matter,impaired cognitive function and increased mortality. Most instances ofSDB in children and adults are obstructive apneas, characterized bynarrowing or obstruction of the airways. Less common in young adultsbut increasingly prevalent with age are central sleep apneas (duringboth rapid eye movement (REM) and non-REM (NREM) sleep)1,which are characterized by loss of respiratory rhythm.

An essential component of the respiratory circuitry in rodents is thepreBotC, a small region of the ventrolateral medulla2,3. In the rat,approximately 300 preBotC neurons per side express high levels of theneurokinin 1 receptor (NK1R)4. Ablation of 480% of preBotC NK1Rneurons in adult rats leads to an ataxic breathing pattern during

Control (pre-injection)a c

b d

lamp

∫DIAEMG

NeckEMG

Day 6 post-injection



EEG

lamp

∫DIAEMG

NeckEMG

EEG

lamp

∫DIAEMG

NeckEMG

EEG

lamp

∫DIAEMG

NeckEMG

EEG

W NREM

NREM

NREM

NREMREM

REM W

W

Apnea

Apnea

W

REM W

Figure 1 Breathing pattern is progressively disrupted, initially during sleep and then also during wakefulness. (a) Pre-injection, breathing is regular during

sleep and wakefulness (W). (b) Post-injection, hypopnea and short central apneas (note the absence ofR

DIAEMG) occur initially during REM sleep. At day 6,

at the end of the REM-associated apnea, a number of small breaths precede awakening (arrow). (c) Hypopnea and short central apneas subsequently increase

in frequency and duration during NREM and REM sleep (day 8; Supplementary Video 1) and also during wakefulness. (d) Ultimately, an ataxic breathing

pattern develops (day 10). Iamp: inspiratory amplitude.R

DIAEMG: integrated diaphragm EMG. NeckEMG: dorsal neck EMG. EEG: electroencephalogram.

Scale bar, 2 s. See Supplementary Figures 2–6 for continuous 5-min data recordings.

Published online 7 August 2005; doi:10.1038/nn1517

1Department of Neurobiology, David Geffen School of Medicine, University of California Los Angeles, Los Angeles, California, 90095-1763, USA. 2These authorscontributed equally to this work. Correspondence should be addressed to J.L.F. ([email protected]).

1142 VOLUME 8 [ NUMBER 9 [ SEPTEMBER 2005 NATURE NEUROSCIENCE

BR I E F COMMUNICAT IONS©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

wakefulness5 that, in contrast to SDB, is rarely seen in humans. Wehypothesized that a less extensive disruption of preBotC function couldunderlie SDB.

Rats were implanted with electromyographic (EMG) electrodes(diaphragmatic, abdominal and neck) and electroencephalographic(EEG) electrodes (see Supplementary Methods online). All experi-mental procedures were approved by the UCLA Chancellor’s AnimalResearch Committee. Inspiratory amplitude, respiratory frequency anddisturbances on day 10 after electrode implantation surgery did notdiffer from control data before electrode implantation (sleep-wakecycle: B3–4 min; inspiratory amplitude: 1.02 ± 0.03 versus 1.00 ± 0.01arbitrary units (a.u.) for after implantation and before implantation,respectively; frequency: 105 ± 5.4 versus 109 ± 7.3 breaths/min;respiratory disturbances: 3.6 ± 0.9 versus 3.8 ± 1.2 episodes/h;n¼ 10; P4 0.05; Fig. 1a). Fourteen days after electrode implantation,either the toxin saporin conjugated to substance P (SP-SAP), whichselectively ablates neurons expressing NK1R5 (n ¼ 8), or substance Pmixed with saporin (unconjugated), which does not ablate neurons(n ¼ 4), was injected bilaterally into the preBotC. In rats injected withsubstance P mixed with saporin, breathing returned to pre-injectionlevels by day 3 post-injection (inspiratory amplitude: 1.02 ± 0.02 a.u;frequency: 110 ± 6.5 breaths/min; n ¼ 4; P 4 0.05) and respiratorydisturbances at day 10 did not significantly differ from data collectedpre-injection (wakefulness: 1.2 ± 0.12, NREM: 3.6 ± 0.12, REM:2.9 ± 2.0 episodes/h; P 4 0.05).

In contrast, in rats injected with SP-SAP at day 4 post-injection,respiratory disturbances significantly increased in number and dura-tion only during REM sleep (66.0 ± 13.8 episodes/h, 5.9 ± 1.1 s; n ¼ 5;P o 0.004; Fig. 2). Disturbances were characterized by markedlyreduced ventilation (that is, hypopnea), occasionally leading to ashort period with neither airflow nor diaphragmatic EMG inspiratoryactivity (that is, central apnea). However, during NREM and wakeful-ness, breathing was similar to that seen pre-injection (NREM 4.2 ± 0.6episodes/h, 4.0 ± 1.4 s, wakefulness: 5.4 ± 3.6 episodes/h, 3.2 ± 0.5 s;n¼ 5; P4 0.05; Fig. 2). Blood gases during wakefulness did not differfrom pre-injection levels (PCO2: 35.2 ± 1.5 versus 37.6 ± 2.3 mm Hg

for post-injection and pre-injection, respectively; PO2: 94.4 ± 2.2 versus96.0 ± 1.3 mm Hg; n ¼ 3; P 4 0.05).

On days 5–6 after SP-SAP injection, respiratory disturbances duringREM further increased in frequency and duration (132.0 ± 50.4episodes/h, 6.7 ± 0.8 s; n ¼ 5; P o 0.002; Figs. 1b and 2), andduring NREM there was a smaller but significant increase (19.2 ±8.4 episodes/h, 8.3 ± 0.7 s; n ¼ 5; P o 0.05; Figs. 1b and 2). A typicalrespiratory disturbance began with hypopnea during NREM thatcontinued into REM, when apnea(s) occurred (Fig. 1b), and endedwith several small breaths that preceded awakening (arrow in Fig. 1b).During wakefulness, short periods of increased frequency (194 ± 21breaths/min; P o 0.05) and brief apneas (13.8 ± 6.0 episodes/h, 5.8 ±1.3 s; n ¼ 5; P o 0.05) occurred; inspiratory amplitude did notsignificantly change (1.00 ± 0.09 a.u; n ¼ 5; P4 0.05; Figs. 1b and 2).Blood gases were slightly altered during wakefulness (PCO2: 38.7 ±1.6 mm Hg; PO2: 89.7 ± 2.4 mm Hg; n ¼ 3; P 4 0.05).

From day 7 after SP-SAP injection, rats were rarely able to complete asleep period without a marked respiratory disturbance (Figs. 1c and 2;Supplementary Video 1). During NREM, respiratory disturbances,characterized by hypopnea, noticeably increased in frequency but notduration (117.6 ± 100.8 episodes/h, 7.6 ± 1.7 s; n ¼ 5), with apneadeveloping upon transition to REM (199.8 ± 94.2 episodes/h, 7.4 ±1.4 s; n ¼ 5) and breathing resuming only upon waking (Fig. 1c). Thisresulted in a highly fragmented sleep pattern and decreased total sleeptime compared with that seen pre-injection (Fig. 3). Before day 7,arousals were a gentle transition to wakefulness, with slight headmovement and very little body movement; post day 7, the rat abruptlyawakened from each apnea and moved around the plethysmographseveral times before settling down. During wakefulness, breathingbecame increasingly irregular, with high-frequency breathing(frequency: 176 ± 16.9 breaths/min; n ¼ 5; P o 0.05; inspiratoryamplitude: 1.05 ± 0.07 a.u; n ¼ 5; P 4 0.05) interspersed withhypopnea and short apneas (15.0 ± 2.4 episodes/h, 5.1 ± 1.1 s;

300

240

180

*

* *

*

**

**

* * *

**

*

120

60

10

5

0

Num

ber

of r

espi

rato

ry d

istu

rban

ces

norm

aliz

ed to

1 h

Dur

atio

n of

res

pira

tory

dist

urba

nces

(s)

Control Day 4

Post–SP-SAP injection

Post–SP-SAP injection

Day 6 Day 8

Control Day 4 Day 6 Day 8

WakefulnessNREMREM

a

b

Figure 2 Number and duration of respiratory disturbances (apneas and

hypopneas). (a) Respiratory disturbances significantly increase during

REM sleep on day 4 post-injection, with no significant change during

NREM and wakefulness (mean ± s.e.m.; n ¼ 5). From day 6, respiratory

disturbances increase in all three states. (b) Duration of respiratory

disturbances increases in all three states up to day 6 post-injection.

* P o 0.05 compared with control.

Eupnea

Eupnea

Hypopnea

Hypopnea

Apnea

Apnea

NREM

NREM

Awake

Awake

0 30Time (min)

60

REM

REM


Control

Figure 3 Respiration and sleep become highly fragmented. In the control

(pre-injection) period (top), short respiratory disturbances occur infrequently

during sleep. On day 7 post-injection, respiratory disturbances occur during

REM and NREM, accounting for the highly fragmented sleep pattern and

decreased total sleep time. Respiratory disturbances are also more frequent

during wakefulness. Data analyzed and plotted in 5-s epochs for one

continuous hour of recording.

NATURE NEUROSCIENCE VOLUME 8 [ NUMBER 9 [ SEPTEMBER 2005 1143


2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

n ¼ 5; P o 0.05). Rats were hypoxic and hypercapnic duringwakefulness (PCO2: 40.2 ± 1.7 mm Hg; PO2: 76.7 ± 3.2 mm Hg;n ¼ 3; P o 0.05).

Beginning on day 9–10 after SP-SAP injection, a severely ataxicbreathing pattern developed during wakefulness5, consisting of anirregular sequence of high-frequency, large breaths (inspiratory ampli-tude: 1.32 ± 0.24 a.u; frequency: 206 ± 19 breaths/min; P o 0.05compared with pre-injection) interspersed with numerous hypopneasand apneas (18.0 ± 6.0 episodes/h, 4.9 ± 0.1 s; n¼ 5; Fig. 1d). Breathingterminated immediately upon sleep onset and did not restart until therat (abruptly) awakened. In two cases, a periodic Cheyne-Stokes–likepattern was intermingled with the ataxic breathing pattern.

The accumulating loss of preBotC neurons led to progressivedisturbances in breathing that proceeded in clearly demarcated stages,leading to an ataxic breathing pattern only when 480% of theseneurons were ultimately ablated5 (as determined by histological ana-lysis after day 10 post-injection). Lesions were confined to the pre-BotC5. NK1R staining was significantly reduced or completely absent inhistological sections representing B15% of the preBotC (2 ±1.3 NK1R+/side; n ¼ 5; P o 0.05; Supplementary Fig. 1) comparedwith non-injected controls (39 ± 7 NK1R+/side; SupplementaryFig. 1) or controls injected with substance P mixed with saporin(32 ± 5 NK1R+/side). In the surrounding rectangular area, NK1Rstaining was reduced (35 ± 15 NK1R+/side) compared with non-injected controls (76 ± 11 NK1R+/side; Supplementary Fig. 1) orcontrols injected with substance P mixed with saporin (70 ± 6 NK1R+/side; Supplementary Fig. 1).In vivo, neuronal death after SAP-induced lesions occurs over a

period of days; here, preBotC lesions caused repeated episodes ofhypoxia that would likely have resulted in further neuronal loss. Thiseffect was likely minor during the early stages of SDB (days 3–6)because the disturbances were limited, and blood gases taken duringwakefulness were normal. Beyond day 6, however, the cumulative effectof intermittent hypoxia resulting from prolonged apneas could haveinduced neuronal death (Supplementary Note) not necessarilyrestricted to the preBotC, further disrupting respiratory pattern.

In both rats6 and humans (particularly the elderly)1, breathing ismore irregular and spontaneous central apneas are most frequentduring REM sleep. Notably, neurons that release serotonin or norepi-nephrine are relatively inactive during REM7. As serotonin andnorepinephrine excite preBotC neurons, any decrease in their releasewould disfacilitate preBotC neurons, rendering breathing during REMvulnerable to preBotC insults (for instance, SP-SAP–induced lesions orneurodegeneration). Thus, as preBotC dysfunction progresses, distur-bances in breathing would be expected initially in REM, next in NREM,and, if sufficiently severe, even during wakefulness.

SDB is common and can go unnoticed in patients in later stages ofamyotrophic lateral sclerosis (ALS)8, multiple systems atrophy (MSA)9

or Parkinson disease10, whereas breathing during wakefulness is con-sidered otherwise normal (Supplementary Note). Many of thesepatients die during sleep8–10. In the ventrolateral medulla (includingthe presumptive preBotC), NK1R neurons are depleted by ~60%in individuals with Parkinson disease and by ~89% in individualswith MSA11, suggesting substantial damage to the preBotC. InALS, neurons with low levels of Ca2+ buffers, in particularmotoneurons, are particularly vulnerable to degeneration12. In the

ventrolateral respiratory column3, the only neurons lacking significantlevels of Ca2+ buffers are preBotC neurons13, suggesting that preBotCneurons are also vulnerable. If preBotC neurons were to degenerate,then significant disturbances of breathing during sleep would beexpected. The slow cumulative loss (compared with loss induced bySP-SAP) of preBotC neurons in these diseases could lead to extended,recurring sleep apneas with consequent intermittent hypoxias, withseemingly normal breathing during wakefulness. This SDB will likelycause an increase in the threshold for arousal from apnea14, leading to avicious cycle in which central apneas become more frequent and longer.Additionally, SP-SAP–induced preBotC NK1R neuron ablation in ratssignificantly reduces the frequency of sighs (data not shown) and bluntsregulatory responses to hypoxia and hypercapnea5; in humans thiscould further increase the arousal threshold14. Ultimately, a centralsleep apnea not terminated by arousal could result in anoxia and death.We suggest that this sequence of events may also occur in the elderly, inwhom the prevalence of central sleep apnea rises precipitously withage1, from the cumulative loss of the small number of preBotC neuronsover a lifetime. If this occurs, death due to preBotC damage wouldhappen before neuronal loss becomes sufficiently extensive to substan-tially alter breathing during wakefulness, as ataxic breathing in humansis rare. Furthermore, during sleep apneas, the induced hypoxia oftenresults in exceptionally high systolic blood pressures that can inducefatal cardiovascular or cerebrovascular incidents15. We suggest that inindividuals with preBotC neuron loss and with significantly compro-mised health, respiratory failure during sleep is the precipitating causeof death.

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

ACKNOWLEDGMENTSWe are grateful to J.M. Siegel for his insightful advice; we thank G.S. Mitchell,C.A. Del Negro, L. Kruger, N. Dale and P.A. Gray for comments on themanuscript and G. Li for histological assistance. This work was supported byUS National Institutes of Health grant HL70029.

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

Received 28 April; accepted 12 July 2005

Published online at http://www.nature.com/natureneuroscience/

1. Krieger, J., Turlot, J.C., Mangin, P. & Kurtz, D. Sleep 6, 108–120 (1983).2. Smith, J.C., Ellenberger, H.H., Ballanyi, K., Richter, D.W. & Feldman, J.L. Science 254,

726–729 (1991).3. Feldman, J.L., Mitchell, G.S. & Nattie, E.E. Annu. Rev. Neurosci. 26, 239–266

(2003).4. Gray, P.A., Rekling, J.C., Bocchiaro, C.M. & Feldman, J.L. Science 286, 1566–1568

(1999).5. Gray, P.A., Janczewski, W.A., Mellen, N., McCrimmon, D.R. & Feldman, J.L. Nat.

Neurosci. 4, 927–930 (2001).6. Mendelson, W.B. et al. Physiol. Behav. 43, 229–234 (1988).7. Siegel, J.M. J. Clin. Psychiatry 65 (suppl.), 4–7 (2004).8. Ferguson, K.A., Strong, M.J., Ahmad, D. & George, C.F. Chest 110, 664–669 (1996).9. Munschauer, F.E., Loh, L., Bannister, R. & Newsom-Davis, J. Neurology 40, 677–679

(1990).10. Maria, B. et al. Respir. Med. 97, 1151–1157 (2003).11. Benarroch, E.E., Schmeichel, A.M., Low, P.A. & Parisi, J.E. Brain 126, 2183–2190

(2003).12. Alexianu, M.E. et al. Ann. Neurol. 36, 846–858 (1994).13. Alheid, G.F., Gray, P.A., Jiang, M.C., Feldman, J.L. & McCrimmon, D.R. J. Neurocytol.

31, 693–717 (2002).14. Berry, R.B. & Gleeson, K. Sleep 20, 654–675 (1997).15. Leung, R.S. & Bradley, T.D. Am. J. Respir. Crit. Care Med. 164, 2147–2165 (2001).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

134 VOLUME 10 | NUMBER 1 | JANUARY 2007 NATURE NEUROSCIENCE

E R R ATA A N D CO R R I G E N DA

Erratum: Reduced sodium current in GABAergic interneurons in a mouse model of severe myoclonic epilepsy in infancyFrank H Yu, Massimo Mantegazza, Ruth E Westenbroek, Carol A Robbins, Franck Kalume, Kimberly A Burton, William J Spain, G Stanley McKnight, Todd Scheuer & William A CatterallNature Neuroscience 9, 1142–1149 (2006); published online 20 August; corrected after print 13 December 2006

In the version of this article initially published, the acceptance date was incorrect. The paper was accepted on 2 August 2006. This error has been corrected in the PDF versions of the article.

Erratum: The many roots of aggressionJordan Grafman, Maren Strezniak & Frank KruegerNature Neuroscience 9, 1347 (2006); published online 26 October 2006; corrected after print 21 November 2006

In the version of this article initially published, the second author’s name was spelled incorrectly. The correct name should be Maren Strenziok. The error has been corrected in the HTML and PDF versions of the article.

Corrigendum: Selective inhibition of 2-AG hydrolysis enhances endocannabinoid signaling in hippocampusJudit K Makara, Marco Mor, Darren Fegley, Szilárd I Szabó, Satish Kathuria, Giuseppe Astarita, Andrea Duranti, Andrea Tontini, Giorgio Tarzia, Silvia Rivara, Tamás F Freund & Daniele PiomelliNature Neuroscience 8, 1139–1141 (2005); Published online 7 August 2005

Our paper identified 6-methyl-2-p-tolylaminobenzo[d]oxazin-4-one (URB754; Specs) as a monoacylglycerol lipase (MGL) inhibitor that enhances hippocampal depolarization-induced suppression of inhibition (DSI). However, in subsequent tests of non-commercial URB754, we failed to replicate these results, suggesting that a bioactive impurity was present in the commercial material. We have identified this impurity as bis(methylthio)mercurane (Supplementary Results online). Because this compound interacts with multiple targets, we tested another MGL inhibitor, methylarachidonylfluo-rophosphonate (MAFP), which prolonged DSI (Fig. 1), confirming that monoacylglycerol lipase contributes to the termination of DSI, as others have reported1. Another generation of endocannabinoid metabolism inhibitors is needed to test this hypothesis further.


1. Szabo B et al. J. Physiol. (Lond.) 577, 263–280 (2006).

Figure 1 Effects of MAFP on DSI in hippocampal CA1 pyramidal cells. Top, traces from a representative experiment showing the effects of vehicle (ethanol, 0.00003%) or MAFP (Tocris, 45 nM) on the transient reduction of spontaneous inhibitory postsynaptic potentials (IPSCs) elicited by a depolarizing stimulus (arrow). Scale bars, 100 pA, 5 s. Bottom left, averaged time-course of DSI after administration of vehicle (solid squares) or MAFP (open squares). Bottom right, DSI area in the first 30 s after stimulus application was significantly larger in MAFP-treated than in control slices.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Selective inhibition of 2-AGhydrolysis enhancesendocannabinoid signalingin hippocampusJudit K Makara1, Marco Mor2, Darren Fegley3,5, Szilard I Szabo1,Satish Kathuria3,5, Giuseppe Astarita3,5, Andrea Duranti4,Andrea Tontini4, Giorgio Tarzia4, Silvia Rivara2,Tamas F Freund1 & Daniele Piomelli3,5

The functions of 2-arachidonoylglycerol (2-AG), the most

abundant endocannabinoid found in the brain, remain largely

unknown. Here we show that two previously unknown inhibitors

of monoacylglycerol lipase, a presynaptic enzyme that

hydrolyzes 2-AG, increase 2-AG levels and enhance retrograde

signaling from pyramidal neurons to GABAergic terminals in

the hippocampus. These results establish a role for 2-AG in

synaptic plasticity and point to monoacylglycerol lipase as a

possible drug target.

The endocannabinoids modulate brain neurotransmission by activat-ing CB1 cannabinoid receptors mostly localized on axon terminals ofGABAergic interneurons1. In the hippocampus, an endocannabinoidreleased from depolarized pyramidal cells transiently depresses GABArelease from basket cell terminals in a retrograde signaling processcalled depolarization-induced suppression of inhibition (DSI)2–4.Although the role of CB1 receptors in DSI is well documented, theidentity of its endocannabinoid mediator remains elusive. Two mole-cules, anandamide and 2-AG, meet the defining criteria of anendocannabinoid. They are produced by neurons in an activity-dependent manner, they engage CB1 receptors with high affinity andthey are eliminated through regulated transport and intracellularhydrolysis5. In neurons, the hydrolysis of anandamide and 2-AG iscatalyzed by two distinct serine hydrolases: fatty-acid amide hydrolase(FAAH), which cleaves anandamide and other lipid amides6, andmonoacylglycerol lipase (MGL), which hydrolyzes 2-AG and other2-monoacylglycerols7. This catabolic segregation offers the opportunityto investigate the functions of each endocannabinoid by blockingits deactivation and thereby amplifying its actions at CB1 receptors.Using this approach, it has been shown that inhibition of FAAH activitydoes not affect hippocampal DSI8, which suggests that neither ananda-mide nor other FAAH substrates with cannabinoid-like activity (forexample, virodhamine and N-arachidonoyl-dopamine)9,10 contributeto this process.

To test the alternative hypothesis that 2-AG mediates DSI, we firstused the compound URB602, a non-competitive MGL inhibitorthat blocks 2-AG hydrolysis in rat brain slices without affectingFAAH-catalyzed anandamide degradation11 (Fig. 1). Depolarization

–10 –9 –8 –7 –6 –5 –4 –30

25

50

75

100

Agent (log [M])

FA

AH

act

ivity

(%

con

trol

)

–9 –8 –7 –6 –5 –40

25

50

75

100

Agent (log [M])M

GL

activ

ity (

% c

ontr

ol)

150

0

300

AE

A (

% c

ontr

ol)

175

0

350

2-A

G (

% c

ontr

ol)

URB602URB597URB754

**

URB602URB597URB754

– – –– – –– 0.5 1

––3

– –– 100– –– – – –

– 1

*

– – –– – –– 0.5 1

––3

– –– 100– –– – – –

– 1

***

–3

OH

O

N

O NH2

O

O

HN

URB602 URB597 URB754

O

N

O

HN

a

b c

d e

Figure 1 URB754 is a potent MGL inhibitor. (a) Chemical structures of MGL

inhibitors URB602 (left), URB754 (right) and FAAH inhibitor URB597

(center). (b) Effects of URB754 (circles), URB602 (triangles) and URB597

(squares) on activity of recombinant rat brain MGL expressed in HeLa cells.

(c) Effects of URB597, URB754 and URB602 on native rat brain FAAH

activity. (d) URB754 (0.5–3 mM; 10 min) and URB602 (100 mM) elevated2-AG levels in rat forebrain slice cultures, whereas URB597 (1 mM) did not.

(e) URB597 elevated anandamide (AEA) levels in the same cultures, whereas

URB754 and URB602 did not. Changes in endocannabinoid levels are

expressed as percentage of control values, which were (in pmol mg–1

protein) 221.5 ± 67.6 (d) and 3.4 ± 1.9 (e). *, P o 0.05; **, P o 0.01;

***, P o 0.001, one-way ANOVA followed by Dunnett’s test (n ¼ 6–8).

Enzyme assays, culture conditions and endocannabinoid measurements

were conducted as outlined in Supplementary Methods. All experimental

procedures were in accordance with Society for Neuroscience and

European Union guidelines and were approved by the institutional

animal care and use committees.


1Institute of Experimental Medicine, Hungarian Academy of Sciences, Budapest, 8. Szigony u. 43., Budapest, H-1083 Hungary. 2Pharmaceutical Department, University ofParma, Parco Area delle Scienze, 27/A, Parma, I-43100 Italy. 3Department of Pharmacology, University of California Irvine, Irvine, California 92697-4625, USA. 4Instituteof Medicinal Chemistry, University of Urbino ‘Carlo Bo’, Piazza del Rinascimento, 6, Urbino, I-61029, Italy. 5Center for Drug Discovery, University of California Irvine, Irvine,California 92697-4625, USA. Correspondence should be addressed to D.P. (email: [email protected]) or T.F.F. ([email protected]).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

of CA1 pyramidal neurons elicited DSI in B50% of the cells tested.Peak DSI was reduced by 55.0 ± 2.6% compared with control (n ¼ 37)and was followed by a complete recovery of inhibitory activity(Fig. 2a–c). As previously reported, superfusion with the CB1 antago-nist AM251 (2 mM, 8–12 min) reduced peak DSI2–4 from 52.5 ± 10.2%to 13.4 ± 3.4% (P o 0.05; n ¼ 4), whereas 0.1 mM of FAAH inhibitorURB597 (refs. 8,12; Fig. 1a) had no effect on either DSI (charge transferreduction, measured as DSI area in the first 30 s after the depolarizingstimulus, was 3.1 ± 0.5 for control and 3.2 ± 0.5 for URB597; Fig. 2a,d;P 4 0.79, n ¼ 7) or basal charge transfer (84.0 ± 10.9% of control,Fig. 2e; P 4 0.19; n ¼ 7). In contrast, the MGL inhibitor URB602(100 mM) delayed the recovery from DSI and increased charge transferreduction (control: 3.1 ± 0.2, URB602: 5.2 ± 0.7; Fig. 2b,d; P o 0.05;n ¼ 8). This effect was blocked by the CB1 antagonist AM251(Supplementary Fig. 1) and was not accompanied by changes inamplitude and decay time constant of depolarization-evokedCa2+ signals, as measured by two-photon microscopy (SupplementaryFig. 2). Furthermore, though prolonged in the presence of URB602,DSI returned to control levels by the time of the next stimulation,indicating that occlusion did not occur at the drug concentra-tion used in these experiments (Fig. 2e; basal charge transfer: 98.6± 11.9%, P 4 0.91).

URB602 inhibits 2-AG hydrolysis selectively, but with low potency11.To discover more potent and selective agents, we screened a focusedlibrary of serine hydrolase inhibitors with electrophilic carbonyl groupsin different chemical environments, whose shape, size and lipophilicitywere comparable to those of URB602. This screening led to theidentification of URB754 (Fig. 1a), a benzoxazin-4-one related toknown elastase and intestinal lipase inhibitors13,14. URB754 inhibitedMGL activity with a half-maximal concentration (IC50) of 200 ± 16 nM(n ¼ 3), about two orders of magnitude more potently than didURB602 (IC50 ¼ 75 ± 7 mM in rat brain MGL expressed in HeLa cells(Fig. 1b) and 28 ± 4 mM in native rat brain MGL11). Kinetic analysesindicated that inhibition occurred through a noncompetitive mechan-ism (Supplementary Table 1). Moreover, overnight dialysis did notrestore activity in MGL preparations incubated with URB754 (1 mM),

suggesting that inhibition was irreversible (data not shown). Incontrast, much higher concentrations of URB754 were needed to affectrat brain FAAH activity (IC50 ¼ 31.8 ± 3.8 mM; n¼ 4; Fig. 1c), bindingof [3H]-Win-55212-2 to rat cerebellar CB1 receptors (IC50 ¼ 10 ±3.8 mM, n¼ 4) and cyclooxygenase (Cox)-1 or Cox-2 activities (IC50 4100 mM). Consistent with these results, incubation with URB754(0.1–3 mM) reduced MGL activity in intact rat forebrain slice cultures(IC50 ¼ 450 ± 7 nM, n¼ 6) and increased 2-AG accumulation (Fig. 1d)but did not change the levels of anandamide (Fig. 1e) or other FAAHsubstrates (Supplementary Fig. 3). URB602 (100 mM) produced asimilar, albeit weaker, effect (Fig. 1d), whereas the FAAH inhibitorURB597 (1 mM) acted in the opposite manner, preventing anandamidedegradation without influencing 2-AG levels (Fig. 1d,e). Finally,incubation of acutely prepared hippocampal slices with URB754(0.5 mM) significantly prolonged DSI (Fig. 2c), increasing chargetransfer reduction from 3.1 ± 0.8 to 4.3 ± 0.7 (Fig. 2d, P o 0.05;n¼ 7) without occlusion (Fig. 2e, basal charge transfer: 98.5 ± 9.3% ofcontrol, P 4 0.87). However, at 3 mM, URB754 elicited a gradualocclusion of DSI that was blocked by AM251 (Supplementary Fig. 4),suggesting that at this concentration the inhibitor caused a markedaccumulation of non-metabolized 2-AG (Fig. 1d).

The functions of 2-AG in synaptic plasticity have been previouslyinferred from experiments with non-specific biosynthesis inhibitors5 orCox-2 inhibitors8, whose effects on brain endocannabinoid metabo-lism are unknown. Our results, showing that blockade of intracellularMGL activity selectively increases brain 2-AG levels and prolongshippocampal DSI, provide unambiguous evidence that 2-AG mediatesthis form of retrograde signaling. Potent and selective MGL inhibitorsdesigned on the scaffolds of URB602 and URB754 will provide avaluable tool to explore the physiological roles of 2-AG and validateMGL as a drug target.


ACKNOWLEDGMENTSWe thank N. Hajos for valuable comments and J. Kim for help with cultures.This research was supported by grants from the US National Institute on Drug

URB URB URB597 602 754

0

50

100

150

Bas

al c

harg

e tr

ansf

er(1

00*d

rug/

cont

rol)

**

Ctr URB Ctr URB Ctr URB754602597

7

6

5

4

3

2

1

0DS

I are

a (a

rbitr

ary

units

)

t (s)403020100–10

t (s)403020100–10

t (s)403020100–10

0.00.2

0.4

0.60.8

1.0

1.21.4

Nor

mal

ized

cha

rge

0.00.20.4

0.60.81.0

1.21.4

Nor

mal

ized

cha

rge

0.00.2

0.4

0.60.8

1.0

1.21.4

Nor

mal

ized

cha

rge Ctr URB597

Ctr URB602

Ctr URB754a

b

c

d e

Figure 2 Inhibition of MGL, but not FAAH, activity prolongs DSI in CA1 pyramidal cells. (a–c) Time course of DSI measured in the same cell before (filled

squares) and after (open squares) treatment with (a) URB597 (0.1 mM; n ¼ 7); (b) URB602 (100 mM; n ¼ 8) or (c) URB754 (0.5 mM; n ¼ 7). Panels on theright show representative DSI traces from individual experiments before and after drug application. Arrows indicate the depolarizing stimuli used to elicit DSI

(from –60 mV to 0 mV, 1 s). Scale bars, 100 pA, 10 s. (d) Application of URB754 (0.5 mM) and URB602 (100 mM), but not URB597 (0.1 mM), led to greater

DSI area than in control (Ctr). (e) The drugs did not influence basal charge transfer. *, P o 0.05, Student’s t-test.



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Abuse and the University of California Discovery Program (D.P.); from MinisteroIstruzione Universita e Ricerca and University of Urbino ‘Carlo Bo’ andUniversity of Parma (G.T. and M.M.); and from the Howard Hughes MedicalInstitute, US National Institutes of Health, European Union FrameworkProgramme 6 and Orszagos Tudomanyos Kutatasi Alapprogramok (T.F.F).

COMPETING INTERESTS STATEMENTThe authors declare competing financial interests (see the Nature Neurosciencewebsite for details).



1. Freund, T.F., Katona, I. & Piomelli, D. Physiol. Rev. 83, 1017–1066 (2003).

2. Wilson, R.I. & Nicoll, R.A. Nature 410, 588–592 (2001).3. Ohno-Shosaku, T., Maejima, T. & Kano, M. Neuron 29, 729–738 (2001).4. Wilson, R.I., Kunos, G. & Nicoll, R.A. Neuron 31, 453–462 (2001).5. Piomelli, D. Nat. Rev. Neurosci. 4, 873–884 (2003).6. Cravatt, B.F. et al. Nature 384, 83–87 (1996).7. Dinh, T.P. et al. Proc. Natl. Acad. Sci. USA 99, 10819–10824 (2002).8. Kim, J. & Alger, B.E. Nat. Neurosci. 7, 697–698 (2004).9. Porter, A.C. et al. J. Pharmacol. Exp. Ther. 301, 1020–1024 (2002).10. Huang, C.C., Chen, Y.L., Lo, S.W. & Hsu, K.S. Mol. Pharmacol. 61, 578–585

(2002).11. Hohmann, A.G. et al. Nature 435, 1108–1112 (2005).12. Kathuria, S. et al. Nat. Med. 9, 76–81 (2003).13. Krantz, A. et al. J. Med. Chem. 33, 464–479 (1990).14. Hodson, H.F. et al. International patent application PCT WO 00/40247

(2000).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

‘Breaking’ position-invariantobject recognitionDavid D Cox, Philip Meier, Nadja Oertelt & James J DiCarlo

While it is often assumed that objects can be recognized

irrespective of where they fall on the retina, little is known

about the mechanisms underlying this ability. By exposing

human subjects to an altered world where some objects

systematically changed identity during the transient blindness

that accompanies eye movements, we induced predictable

object confusions across retinal positions, effectively ‘breaking’

position invariance. Thus, position invariance is not a rigid

property of vision but is constantly adapting to the statistics

of the environment.

Any given object can cast an essentially infinite number of differentimages on the retina, owing to variations in position, scale, view,lighting and a host of other factors. Nonetheless, humans effortlesslyrecognize familiar objects in a manner that is largely invariant to thesetransformations. The ability to identify objects in spite of these trans-forms is central to human visual object recognition, yet the neuralmechanisms that achieve this feat are poorly understood, andtransform-tolerant recognition remains a major stumbling block inthe development of artificial vision systems. Even for variations in theposition of an image on the retina, arguably the simplest transform thatthe visual system must discount, little is known about how invarianceis achieved.

Several authors have proposed that one solution to the invarianceproblem is to learn representations through experience with thespatiotemporal statistics of the natural visual world1–4. Visual featuresthat covary across short time intervals are, on average, more likely to

Normal exposure

'Swapped' exposure

Object A

Object A

Nor

mal

Left Center Center Right

Sw

appe

dN

orm

al

Nor

mal

Sw

appe

dN

orm

al

A

B

C

A′

B′

C′

A

A′

A

A′

B′

B

B

B′

C′ C′

C C

A

A′

A′

A

B

B′

B

B′

C′ C′

C C

Object A′Object A′

Object AObject A

Saccading

Saccading

a

b c

Figure 1 Experiment 1 and 2 design. Twelve naive subjects participatedin each experiment and provided informed consent in accordance with

the Massachusetts Institute of Technology Committee on the Use of

Humans as Experimental Subjects. (a) During the ‘exposure’ phase of

each experiment, subjects received two different types of exposure trials

randomly interleaved. In all trials, subjects started a trial by fixating on a

point, and then an object appeared in the periphery (61 to the left or right,

randomly). Subjects spontaneously saccaded to the object and were

required to decide if this object was the same object as in the preceding

trial. In ‘normal exposure’ trials, the object identity did not change, so

the same object was presented to both the peripheral retina (pre-saccade)

and the central retina (post-saccade). In ‘swapped’ exposure trials,

unknown to subjects, one object was swapped for a different object in

mid-saccade, such that one object was presented to the peripheral

retina pre-saccade, and a different object was presented to the central

retina post-saccade. (b) The objects used in this experiment were modified

versions of the publicly available ‘greeble’ stimuli (Supplementary Methods)

and were arranged in three pairs, with the differences within pair (for

example, A and A¢) being qualitatively smaller than the differencesbetween pairs (for example, A and B). Objects were chosen to be relatively

natural but unfamiliar to the subject. (c) A schematic representation of the

twelve exposure trial types for one subject. All such exposure trials occurred

equally often (pseudorandomly selected). Thus, each subject received an

equal number of presentations of all objects in each retinal location. The

letter on one side of the arrow indicates the peripherally presented object

(either on the right or left), with the arrow indicating the object identity

before (arrow tail) and after (arrowhead) the saccade. For all subjects, one

object pair was swapped on the right but was normal on the left (top row),

one pair was normal on the right but swapped on the left (middle row) and

one pair was not swapped on either side (bottom row). Subjects were tested

in two sets of six, with each set of six counterbalancing across all possible

assignments of the three object pairs to each of these three roles. Blue

panels indicate trials in which objects were not swapped; orange indicates

‘swapped’ trials and green indicates trials with object pairs that were not

swapped on either side.


McGovern Institute for Brain Research and Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA.Correspondence should be addressed to J.D. ([email protected]).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

correspond to different images of the same object than to differentobjects, and thus one might gradually build up invariant representa-tions by associating patterns of neural activity produced by successiveretinal images of an object. While some transformations of an object’sretinal image are played out smoothly across time (for example, scaleand pose), changes of an object’s retinal position often occur discon-tinuously as a result of rapid eye movements that sample the visualscene (saccades). A possible strategy, then, for building position-invariant object representations is to associate neural activity patternsacross saccades, preferably taking into account the direction andmagnitude of the saccade.

If correct position invariance is created through experience with thestatistical properties of the visual world, it might be possible to createunnatural or ‘incorrect’ invariances by manipulating those statistics. Inparticular, if objects consistently changed their identity as a function ofretinal position, then the visual system might incorrectly associate theneural representations of different objects at different positions into asingle object representation. The resulting representation would beactivated by one object at one retinal position and another object atanother position, and thus the two objects would be perceived as beingthe same object at different positions.

In the present study, we engineered such a situation, taking advan-tage of the fact that humans are effectively blind during the short time ittakes to complete a saccade5,6. By monitoring eye position in real time,we were able to present one object to a subject’s peripheral retina thatwas replaced by a particular different object in mid-saccade when thesubject attempted to foveate it. None of the subjects reported beingaware that objects were being swapped, despite being asked in a post-session debriefing whether they had seen objects change or appearotherwise unusual. After a brief period of exposure to these altered

spatiotemporal statistics (240–400 altered exposures in Experiment 1,and 120–180 altered exposures in Experiment 2), we used a same-different task to probe the subject’s representations of these objectsacross changes in position. The layout of Experiments 1 and 2 isdescribed in Figure 1 and in Supplementary Methods.

In both experiments, subjects significantly more often confusedobject pairs when they were tested across the retinal positions wherethose particular objects had been swapped during the exposure phasethan in tests across positions where the same objects had not beenswapped (P ¼ 0.0082 in experiment 1, P ¼ 0.022 in experiment 2;P ¼ 0.0007, both experiments pooled; one-tailed paired t-test; Fig. 2).That is, for previously swapped objects, subjects were more likely toperceive different objects at two retinal positions as the same object andperceive the same object at two positions as different objects.

These results show that confusions in invariant visual object proces-sing occur after relatively brief exposure (o1 h, total) to alteredspatiotemporal statistics across saccades, even though subjects wereunaware of this change. Moreover, the confusions are predictable in thatthey are what is expected if the visual system assumes that objectidentity is stable across the short time interval of a saccade. Although themagnitude of the observed effect is not large, and we have shown it onlyfor relatively similar objects, it should be borne in mind that theanomalous exposure provided represents a tiny fraction of each subject’slifetime experience with an unaltered, real-world visual environment.The ability to significantly shift object representations at all suggests thatposition-invariant visual object recognition is modifiable in adults, andit points to possible mechanisms by which sets of invariant featuresmight be acquired, especially during early visual learning.

To test whether the observed effect depends critically on the execu-tion of active eye movements, as opposed to spatiotemporal experiencealone, we ran a third set of experiments (experiment 3) with twelvesubjects with retinal experience matched to the subjects in experiment1, but without saccades. These subjects maintained fixation throughouteach trial during the exposure phase, and the retinal positions andtiming of object exposure was ‘replayed’, trial-by-trial, from thespatiotemporal retinal experience generated by their counterpart sub-ject in experiment 1. The testing phase was identical to experiments 1and 2. Subjects in experiment 3 showed no effect of anomalousspatiotemporal experience (P 4 0.6; one-tailed paired t-test, Fig. 2),suggesting that anomalous experience across saccades may be necessaryto produce later confusions in invariant object processing.

Although these results show that specific alterations in objectspatiotemporal experience can alter position-invariant recognitionwith test objects in the direction predicted by theory, we wondered ifsuch anomalous experience might also produce more general deficits inrecognition performance with those test objects. To examine this, wecompared recognition performance of test objects across positionswhere those objects had behaved normally (‘unswapped’ conditions)with recognition of control objects (which were never swapped in eitherposition). Although both experiments showed a trend toward reducedperformance with objects whose spatiotemporal statistics had beenaltered (Supplementary Fig. 1), no significant difference was found ineither experiment (experiment 1: P ¼ 0.48; experiment 2: P ¼ 0.094,two-tailed paired t-tests).

Like some recent perceptual learning studies, this study shows thatvisual processing can be altered by visual statistics that do not reachawareness7. However, in contrast to standard perceptual learningprocedures in which subjects improve on some sensory task over thecourse of many training sessions8, here, performance is impaired in apredictable way by brief exposure that runs counter to the subject’s pastvisual experience. This resembles other long-term perceptual adapta-

–4

–2

0

2

4

Swapped – unswapped

Exp. 1

P = 0.008

Exp. 2

P = 0.022

Exp. 1 & 2

P = 0.0007

Exp. 3

P > 0.6

Per

form

ance

diff

eren

ce (

% c

orre

ct)

** ***

Figure 2 Results. In testing after exposure, subjects in experiment 1 (2 d of

exposure; n ¼ 12) and experiment 2 (1 d of exposure; n ¼ 12) confused objects

significantly more often across retinal positions where they had been swapped

during the exposure phase (orange panels in Fig. 1c) than across positions where

the same objects behaved normally during exposure (‘unswapped’; blue panels in

Fig. 1c). These effects were not significantly different for trials where the correct

answer was the ‘same’ and trials where the correct answer was ‘different’ in either

experiment 1 or 2 (P 4 0.4 two-tailed paired t-tests). Subjects in experiment 3

(replay experiment; n ¼ 12) who received retinal exposure matched to subjects in

experiment 1 did not show a significant effect. Bars show effect magnitudes and

standard errors for experiment 1 (2 d of exposure), experiment 2 (1 d ofexposure), data from experiments 1 and 2 pooled together, and experiment 3

(replay). Mean performance with the control objects (green panels in Fig. 1c)

was 74%, 72% and 78% in experiments 1, 2 and 3, respectively, and was not

significantly different across the three experiments (P 4 0.1, one-way ANOVA).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

tion effects, such as the McCollough effect and prism adaptation and,like these effects, might represent an ongoing process to adapt to theenvironment and keep perception veridical9.

While adult transform-invariant object recognition is, for the mostpart, automatic and robust10, this finding adds to a growing body ofresearch suggesting that such invariance may ultimately depend uponexperience11–14. More broadly, this finding supports the developingbelief that visual representations in the brain are plastic and largely aproduct of the visual environment15. Within this context, invariantobject representations are not rigid and finalized, but are continuallyevolving entities, ready to adapt to changes in the environment.


ACKNOWLEDGMENTSWe would like to thank B. Balas, N. Kanwisher and P. Sinha for their helpfulcomments on earlier versions of this work and J. Deutsch for technical support.This work was supported by the US National Eye Institute (NIH-R01-EY014970)and the Pew Charitable Trusts (PEW UCSF 2893sc). D.D.C. is supported by aNational Defense Science and Engineering Graduate Fellowship. N.O. wassupported by the Paul E. Gray Memorial Undergraduate fund.


Received 17 May; accepted 15 July 2005


1. Foldiak, P. Neural Comput. 3, 194–200 (1991).2. Wallis, G. & Rolls, E.T. Prog. Neurobiol. 51, 167–194 (1997).3. Wiskott, L. & Sejnowski, T.J. Neural Comput. 14, 715–770 (2002).4. Edelman, S. & Intrator, N. Cogn. Sci. 27, 73–109 (2003).5. Ross, J., Morrone, M.C., Goldberg, M.E. & Burr, D.C. Trends Neurosci. 24, 113–121

(2001).6. McConkie, G.W. & Currie, C.B. J. Exp. Psychol. Hum. Percept. Perform. 22, 563–581

(1996).7. Watanabe, T., Nanez, J.E. & Sasaki, Y. Nature 413, 844–848 (2001).8. Karni, A. & Sagi, D. Nature 365, 250–252 (1993).9. Bedford, F. Trends Cogn. Sci. 3, 4–12 (1999).10. Biederman, I. & Bar, M. Vision Res. 39, 2885–2899 (1999).11. Dill, M. & Fahle, M. Percept. Psychophys. 60, 65–81 (1998).12. Nazir, T.A. & O’Regan, J.K. Spat. Vis. 5, 81–100 (1990).13. Dill, M. & Edelman, S. Perception 30, 707–724 (2001).14. Wallis, G. & Bulthoff, H.H. Proc. Natl. Acad. Sci. USA 98, 4800–4804 (2001).15. Simoncelli, E.P. & Olshausen, B.A. Annu. Rev. Neurosci. 24, 1193–1216

(2001).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Extensive piano practicing hasregionally specific effects on whitematter developmentSara L Bengtsson1, Zoltan Nagy1,2, Stefan Skare2, Lea Forsman1,Hans Forssberg1 & Fredrik Ullen1

Using diffusion tensor imaging, we investigated effects of

piano practicing in childhood, adolescence and adulthood

on white matter, and found positive correlations between

practicing and fiber tract organization in different regions for

each age period. For childhood, practicing correlations were

extensive and included the pyramidal tract, which was more

structured in pianists than in non-musicians. Long-term

training within critical developmental periods may thus

induce regionally specific plasticity in myelinating tracts.

Musicians are a useful group for the study of neural correlates ofextensive long-term training1. These correlates include structuraladaptations in gray matter regions extensive enough to be seen on amacroanatomical level with the use of morphometric techniques2.However, white matter in musicians has not been studied as extensively.In humans, the maturation of central fiber tracts continues at least untilthe age of 30 years, with regional differences in onset time and rate ofmyelination3 that have been related to the development of correspond-ing functions: maturation of frontal and left temporoparietal fibertracts coincides with the development of working memory capacity andreading ability, respectively4. Similarly, the maturation of corticospinalfibers parallels the development of fine finger movements5. In thepresent study, we tested if a fiber tract was susceptible to training-induced plasticity during the period when it was still under maturation.This hypothesis was inspired both by the aforementioned correlationsbetween structural and functional development and by the fact thatmyelination in the CNS can be stimulated by electrical activity inpremyelinated axons6. We investigated white matter structure in eightmale, right-handed professional concert pianists with a mean age of

32.6 ± 5.7 (s.d.) years, using the magnetic resonance techniquediffusion tensor imaging (DTI)7. A group of eight male, age-matchednon-musicians served as controls. Fractional anisotropy (FA)7 in eachvoxel was used as a measure of the degree of water diffusion anisotropy.FA can be used for inferences about the microstructural propertiesof white matter, as diffusion is faster along axons than in theperpendicular direction (see Supplementary Methods online).

We regressed FA on the estimated total number of hours practicedby each pianist during childhood (from the start of practicing until11 years), adolescence (12–16 years) and adulthood (17 years until timeof the magnetic resonance scan). These values were calculated frombiographical data collected from all participants on the self-estimatednumber of hours of practicing from the commencement of pianotraining until the present. The test-retest reliability of the practicingdata was assessed one year later. Significant differences in FA betweenthe pianists and the control group were evaluated with a two-samplet-test. The participating pianists started playing at a mean age of 5.8 ±1.4 years. The mean total number of hours practiced in childhood,adolescence and adulthood for all participants were 1,618 ± 662 h;3,195 ± 1,515 h and 22,971 ± 9,413 h, respectively. For all three ageperiods, the test-retest reliability was high. The reliabilities of themeasures of childhood, adolescent and adult practicing were r ¼ 0.81(P ¼ 0.015), r ¼ 0.86 (P ¼ 0.007) and r ¼ 0.95 (P ¼ 0.0004),respectively. Childhood practicing time correlated with practicing timein adolescence (r ¼ 0.78, P ¼ 0.02; Pearson product-moment correla-tion). This implies that in the case where a significant regression wasfound between FA and practicing in both childhood and adolescence inthe same brain region, we could not tell whether this reflectedpracticing in both or in only one of these age periods. Adult practicingtime did not correlate significantly with practicing time in childhood(P ¼ 0.19) or in adolescence (P ¼ 0.10).

Controls

Age ≤11

FA

0.6

0.4

0.2

0.00

FA

0.6

0.4

0.2

0.00.5 1.0

Practice (× 103 h)

Corpus callosum(isthmus)

1.5 2.52.0 3.0

0.5 1.0Practice (× 103 h)

1.5 2.52.0 3.0

Internal capsule

L

L

L

Pianists versus controls

z = 24

y = –16 y = –19

Age ≤11

a

c

b

Figure 1 Childhood practicing and white matter structure. (a) Clusters in the

internal capsule with significant FA correlations, overlaid on the mean FA

image from all participants. In the graph at right, each point shows the mean

FA value for all voxels in the cluster in one individual. Vertical lines represent

s.d. within that participant. Dashed lines are regression lines. (b) Cluster of

voxels in the right internal capsule with significantly higher FA values in the

pianists than in the control group. The mean FA values for each control

participant in this cluster are illustrated with gray points in the graph in a,

and the gray dashed line represents the total mean FA value of the wholecontrol group. (c) Same as a, for the isthmus of the corpus callosum.


1Department of Woman and Child Health, Karolinska Institutet, SE-171 77, Stockholm, Sweden. 2Karolinska MR Research Centre, Karolinska Hospital, SE-171 77,Stockholm, Sweden. Correspondence and requests for materials should be addressed to F.U. ([email protected]).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

A significant correlation between practicing time and FA was foundfor all age periods, but in different sets of brain regions (Figs. 1 and 2,Table 1). No correlations were found between FA and the age of theparticipant. All clusters were located within the white matter in allindividual subjects (Supplementary Figs. 1 and 2), and no potentiallyconfounding correlations between white matter density and practicingwere found in any regions (Supplementary Methods).

Childhood practicing correlated with FA in the bilateral posteriorlimbs of the internal capsule (Fig. 1a). The right posterior internalcapsule was the only region where significantly higher FA values (P o0.000, t ¼ 4.34) were found in the pianist group than in the non-musician group (Fig. 1a,b). This cluster had a volume of 1,322 mm3

with the peak coordinate in 27, –20, –36 (x, y, z) and partly overlappedwith the cluster in the same region found in the regression analysis.In addition, childhood practicing correlated with FA in two clusters inthe corpus callosum—one in the isthmus extending into the upper

splenium (Fig. 1c) and the other in the callosal body—and in fibertracts in the frontal lobe. The posterior limbs of the internal capsulecarry the corticospinal tracts5, which contain descending fibers fromthe primary sensorimotor and premotor cortices and which are ofcritical importance for independent finger movements in humans andother primates8. The isthmus and adjacent splenium of the corpuscallosum contain fibers connecting auditory regions in the superiortemporal gyri and the parietotemporal junctions9. The body of corpuscallosum connects superior frontal regions9, including the dorsalpremotor cortices and the mesial premotor areas, which play keyroles for bimanual coordination as well as learning and performanceof movement sequences10,11. Both the corticospinal tracts and thecorpus callosum continue their maturation throughout childhood.Recent work using magnetic resonance imaging has demonstrated age-related changes in corticospinal white matter density continuing at leastuntil an age of 17 years5. The maturation of these tracts thus continueswell into the period when our participants started practicing the piano.

Practicing during adolescence correlated with FA in the spleniumand body of the corpus callosum. The splenium cluster extended intothe white matter of the occipital lobe (Fig. 2a). As FA in the body andupper splenium of corpus callosum correlated also with childhoodpracticing, we cannot discuss the relative importance of these two ageperiods for training-related plasticity in these subregions. The spleniumcontains interhemispheric fibers from superior temporal and occipitalcortical areas9, which include auditory and visual processing regions,respectively. The cross-sectional area of the corpus callosum grows atleast until early adulthood12. These age-related changes are mainly dueto growth in the posterior and middle regions13, which is also where FAcorrelated with practicing in the present study, and could reflect bothmyelination, which continues at least until age ten3, and increases in thediameter of callosal axons, as indicated by a decrease in the signalintensity in anatomical magnetic resonance images of the corpuscallosum with age12.

Adult practicing correlated with FA in the left anterior limb of theinternal capsule and in a fiber bundle in the right temporoparietaljunction (Fig. 2b), most likely part of the arcuate fasciculus, which

Age 12–16

Age ≥17

x = 43

FA

0.6

0.4

0.2

0.0

FA

0.6

0.4

0.2

0.0

1 2

Corpus callosum(splenium)

Arcuate fasciculus

3 54 6

10 15

Practice (× 103 h)

Practice (× 103 h)

20 3025 35 40 45

L

z = 20

b

a

Figure 2 Correlations between adolescent and adult practicing and whitematter structure. Cluster images and graphs as in Figure 1 for (a) the

splenium of the corpus callosum and (b) the right arcuate fasciculus.

Table 1 Brain regions with significant correlations between FA values and mean hours of practicing in an age period

Cluster Side

Corrected

P-value

Cluster

size (mm3) r2Slope

(FA/h � 10–4)

Peak coordinate

(x,y,z)

Mean FA,

pianists (± s.d.)

Mean FA,

non-musicians (± s.d.)

Age r11

Corpus callosum (isthmus/splenium) L 0.02 350 0.88 1.0 –9, –49, 24 0.34 ± 0.10 0.33 ± 0.091

R o0.001 648 0.87 0.9 11, –49, 24 0.34 ± 0.11 0.30 ± 0.071

Corpus callosum (body) R 0.02 350 0.76 0.7 18, –7, 36 0.26 ± 0.063 0.31 ± 0.067

Capsula int., posterior limb L o0.001 635 0.83 0.8 –18, –16, 12 0.42 ± 0.10 0.44 ± 0.096

R 0.025 337 0.84 0.4 14, –13, 8 0.31 ± 0.055 0.33 ± 0.087

Superior frontal L 0.004 441 0.84 0.6 –25, –2, 36 0.30 ± 0.051 0.32 ± 0.053

R 0.012 376 0.90 0.6 18, 34, 40 0.27 ± 0.062 0.30 ± 0.051

Inferior frontal L 0.043 311 0.78 0.6 –18, 31, 0 0.32 ± 0.062 0.30 ± 0.065

Age 12–16

Corpus callosum (splenium) R o0.001 1,257 0.90 0.5 31, –67, 24 0.33 ± 0.083 0.31 ± 0.081

Corpus callosum (body) R o0.001 1,024 0.89 0.2 13, –7, 32 0.27 ± 0.059 0.30 ± 0.062

Superior frontal L o0.001 648 0.92 0.3 –22, 16, 36 0.27 ± 0.054 0.31 ± 0.064

Age Z17

Capsula int., anterior limb L 0.004 415 0.84 0.08 –16, 7, 12 0.37 ± 0.078 0.40 ± 0.066

Arcuate fasciculus R 0.004 415 0.90 0.07 47, –29, 0 0.29 ± 0.073 0.28 ± 0.067

The x, y and z coordinates of the peak voxel of each cluster in Talaraich space are given in millimeters.



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

connects temporal and frontal regions. The most extended myelinationcycle is found in corticocortical fibers. The long association fibersystems of the forebrain, which correlated with adult practicing,continue their maturation at least into the third decade of adult life3.

In summary, these results suggest that training can induce whitematter plasticity if it occurs in a period when the involved fiber tractsare still under maturation. We propose that increased myelination,caused by neural activity in fiber tracts during training, is onemechanism underlying the observed FA increases. In the mouse,induction of myelination by neural activity has been demonstratedboth in vivo and in vitro6, and myelination of the optic nerve isdecelerated by rearing in darkness14. A key finding here is that thelargest number of brain regions correlated with childhood practicing,although the total number of hours practiced in childhood wasconsiderably lower than in later life. In addition, the slope of theregression (that is, change in FA per practice hour) was steeper for earlypracticing than for adult practicing (Table 1). This illustrates both theimportance of early practicing for white matter plasticity and thelimited malleability of the system in adulthood. The one region where agroup difference between pianists and non-musicians was found wasthe posterior limb of the internal capsule, which also correlated onlywith childhood practicing. For the corpus callosum and fiber bundlesconnecting cortical association areas, no group difference was found,although a large proportion of the variance within the musician groupwas explained by practicing. This may reflect that these regions areinvolved also in a multitude of non-musical tasks that were practiced bythe non-musicians. Professional pianists on average both start theirtraining at an earlier age and practice much more intensively during thefirst years of training than do amateur players15. Our findings suggestone plausible explanation for this: training-induced white matteradaptations are likely to be important for the high-level performanceof the mature pianist, and the overall susceptibility to such plasticity ishigh in childhood, because at that time, a large number of fiber systems

used in piano performance have still not completed their maturation. Itappears likely that extensive training before the age when involvedwhite matter fiber tracts have fully matured, could be an importantfactor behind the development of high-level abilities in other domainsas well.


ACKNOWLEDGMENTSWe thank I. Agartz (Human Brain Informatics at Karolinska Institutet) andP. Lindberg for providing part of the control data and J. Andersson, S. Grillnerand P.E. Roland for valuable discussions and comments on the manuscript. Thiswork was supported by the Swedish Research Council, Karolinska Institutet’sResearch Funds, the Jeansson Foundations, Sallskapet Barnavard and theFreemasons in Stockholm Foundation for Children’s Welfare.


Received 1 June; accepted 13 July 2005


1. Schlaug, G. & Chen, C. in The Biological Foundations of Music (eds. Zatorre, R.J. &Peretz, I.) 281–299 (New York Academy of Sciences, New York, 2001).

2. Gaser, C. & Schlaug, G. J. Neurosci. 23, 9240–9245 (2003).3. Yakovlev, P.I. & Lecours, A.-R. in Regional Development of the Brain in Early Life

(ed. Minkowski, A.) 3–65 (Blackwell, Oxford, 1967).4. Nagy, Z., Westerberg, H. & Klingberg, T. J. Cogn. Neurosci. 16, 1227–1233 (2004).5. Paus, T. et al. Science 283, 1908–1911 (1999).6. Demerens, C. et al. Proc. Natl. Acad. Sci. USA 93, 9887–9892 (1996).7. Le Bihan, D. Nat. Rev. Neurosci. 4, 469–480 (2003).8. Armand, J., Olivier, E., Edgley, S.A. & Lemon, R.N. in Hand and Brain (eds. Wing, A.M.,

Haggard, P. & Flanagan, J.R.) 125–146 (Academic, San Diego, 1996).9. de Lacoste, M.C., Kirkpatrick, J.B. & Ross, E.D. J. Neuropathol. Exp. Neurol. 44,

578–591 (1985).10. Swinnen, S. & Wenderoth, N. Trends Cogn Sci. 8, 18–25 (2004).11. Tanji, J. Annu. Rev. Neurosci. 24, 631–651 (2001).12. Keshavan, M.S. et al. Life Sci. 70, 1909–1922 (2002).13. Giedd, J.N. et al. Dev. Brain. Res. 91, 274–280 (1996).14. Gyllensten, L. & Malmfors, T. J. Embryol. Exp. Morphol. 11, 255–266 (1963).15. Krampe, R.T. & Ericsson, K.A. J. Exp. Psychol. Gen. 125, 331–359 (1996).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Ryk-mediated Wnt repulsion regulates posterior-directed growth of corticospinal tract

Yaobo Liu1, Jun Shi1, Chin-Chun Lu1, Zheng-Bei Wang4, Anna I Lyuksyutova2, Xue-Jun Song4 & Yimin Zou1,2,3

Guidance cues along the longitudinal axis of the CNS are poorly understood. Wnt proteins attract ascending somatosensory axons

to project from the spinal cord to the brain. Here we show that Wnt proteins repel corticospinal tract (CST) axons in the opposite

direction. Several Wnt genes were found to be expressed in the mouse spinal cord gray matter, cupping the dorsal funiculus, in an

anterior-to-posterior decreasing gradient along the cervical and thoracic cord. Wnts repelled CST axons in collagen gel assays

through a conserved high-affinity receptor, Ryk, which is expressed in CST axons. Neonatal spinal cord secretes diffusible

repellent(s) in an anterior-posterior graded fashion, with anterior cord being stronger, and the repulsive activity was blocked by

antibodies to Ryk (anti-Ryk). Intrathecal injection of anti-Ryk blocked the posterior growth of CST axons. Therefore, Wnt proteins

may have a general role in anterior-posterior guidance of multiple classes of axons.

In the CNS, sensory pathways form ascending fibers that relay sensorystimuli to higher brain centers. Descending fibers send motor commandand regulatory signals down from the brain. In the spinal cord,somatosensory fibers project anteriorly, whereas motor pathways pro-ject posteriorly, making the spinal cord an excellent model system forstudying how ascending and descending axons are guided. The mechan-isms by which axons recognize the long anterior-posterior directionduring initial wiring of the spinal cord have been elusive. A signalingpathway that requires Wnt and Frizzled proteins is necessary for theanterior-directed growth of commissural axons after midline crossing1.

To address the molecular mechanisms that regulate posterior-directed axon growth in the spinal cord, we studied the CST axons.CSTaxons originate from the frontal and sensorimotor cortices, exit thecortex, traverse the internal capsule and project posteriorly through thebasal midbrain (Fig. 1a). When they reach the hindbrain, they projectdorsally at the caudal medulla and cross the midline just before enteringthe spinal cord at postnatal day (P) 0. Most CSTaxons cross the midline,forming the pyramidal decussation, and enter the spinal cord, project-ing posteriorly until they reach their proper anterior-posterior position.In rodents, most CST axons project posteriorly at the dorsal midline inthe ventral-most portion of the dorsal funiculus (Fig. 1b)2. LeadingCST axons in mice reach the caudal thoracic level at P7 (ref. 3).Although guidance and regeneration of CST axons has received muchattention, the molecular guidance cues that direct the posterior growthof the CST axons in the spinal cord have not been identified.

On the basis of our knowledge of the anterior-posterior guidance ofcommissural axons, we studied whether Wnt proteins also regulateanterior-posterior pathfinding of CST axons. We found that severalWnt genes are expressed in a high-to-low gradient in the gray matterthat cups the dorsal funiculus from cervical to thoracic spinal cord and

that Wnt proteins repel CST axons. The vertebrate homolog of therepulsive Wnt receptor Derailed, Ryk, is a high-affinity receptor and isexpressed on CST axons4,5. Polyclonal antibodies against the ecto-domain of Ryk blocked the repulsive effect of Wnt1 and Wnt5a. Theneonatal spinal cord secretes diffusible repellent(s) in a decreasingfashion along the anterior-posterior axis, and anti-Ryk blocked therepulsion. Intrathecal injection of anti-Ryk into neonatal cervical spinalcord caused a significant reduction in CST axons posterior to theinjection sites and stalling of CST axons anterior to the injection sites.These results indicate that Wnt proteins may control not only thepathfinding of ascending sensory axons but also that of descendingmotor pathways. Therefore, Wnts may convey general directionalinformation for axonal connections along the anterior-posterior axisin the spinal cord.

RESULTS

Anterior-posterior gradient of Wnts in neonatal spinal cord

Because CST axons project posteriorly along the dorsal funiculus of thespinal cord, we examined the expression patterns of Wnt genes aroundthe dorsal funiculus. We cloned in situ hybridization probes for theentire family of mouse Wnt genes (including 19 members) and carriedout in situ hybridization at P0 and P3 along the anterior-posterior axis.A transverse section was taken every 1.6 mm along the anterior-posterior axis of individual spinal cords and tested for expression ofWnt genes. We found that four Wnt genes are expressed in the graymatter that cups the dorsal midline and dorsal funiculus at these stages,with Wnt1 and Wnt5a being expressed at a higher level (Fig. 1c,d). Wenoticed that the expression patterns of Wnt1 and Wnt5a were notidentical, demonstrating the specificity of these probes. Wnt1 wasexpressed in wider areas in the dorsal half of the spinal cord

Received 18 June; accepted 15 July; published online 14 August 2005; corrected after print 14 December 2005; doi:10.1038/nn1520

1Department of Neurobiology, Pharmacology and Physiology, 2Committee on Developmental Biology and 3Committee on Neurobiology, University of Chicago, Chicago,Illinois 60637, USA. 4Parker College Research Institute, Dallas, Texas 75229, USA. Correspondence should be addressed to Y.Z. ([email protected]).


ART IC L E S©

2006

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

(Fig. 1e), whereas Wnt5a was expressed in anarrower area in the dorsal spinal cord but athigher levels in the ventral-medial domains ofthe spinal cord (Fig. 1f). Both Wnt1 andWnt5a were expressed in the gray matter immediately surroundingthe dorsal funiculus. Although the identities of the cells expressing highamounts of Wnt1 and Wnt5a are not known, Wnt1 was expressed inboth larger and smaller cells (Fig. 1g), whereas Wnt5a was expressedprimarily in larger cells (Fig. 1h). Wnt8a and Wnt9a were expressed atmuch lower amounts (data not shown). Along the anterior-posterioraxis, these four Wnt genes showed a general high-to-low gradient atcervical and thoracic spinal cord levels, although expression of the geneencoding ephrin B3 (Efnb3) did not show an anterior-posteriorgradient (Fig. 1d). The signal intensity of Wnts was quantified fromfour sets of in situ experiments (Fig. 1c). CST axons arrive at themedulla–spinal cord junction at P0 and during the first few days ofpostnatal life. Pioneering CST axons enter the cervical spinal cordand reach the posterior thoracic levels only at P5 (ref. 3). Therefore,CST axons pathfind within the Wnt1 and Wnt5a gradients inside thespinal cord.

Wnt proteins repel motor cortical axons

To test whether Wnts can guide CST axons, we carried out explantcultures to evaluate the function of Wnt proteins on the motor corticalaxons in collagen gel assays (Fig. 2). P0 brains were dissected out andsliced with a tissue chopper (see Supplementary Fig. 1 online). Layer 5cortical explants were dissected from the frontal and sensorimotorcortical regions and were cultured in collagen for 60 h (SupplementaryFig. 1). To ensure that the cultures were consistent and that only motorcortical explants were cultured, only three explants were dissected fromeach side of the cortical slices and only two slices (each was 250 mmthick) from the correct areas were used from each brain.

To test the function of Wnts, COS7 cells were transfected with Wntexpression constructs, aggregated into cell clumps and positioned nextto the cortical explants from P0 motor cortex (Fig. 2a). We found thatWnt1 and Wnt5a potently inhibited the outgrowth of axons from the

motor cortex in these assays, indicating that CST axons might respondto Wnt proteins as they pathfind along the spinal cord in vivo. Very fewaxons grew out in the collagen gel, and the axon length was markedlyreduced as well. To address the possibility that the cell aggregatessecreted excessive amounts of Wnt1 and Wnt5a proteins, preventingaxons from growing out of the explants, we diluted the transfected COScells with untransfected COS cells and began to observe strongrepulsion by cells expressing Wnt1 and Wnt5a (Fig. 2b).

Results were quantified from eight explants for each set of experi-ments (Fig. 2c,d and Supplementary Fig. 2), and the means (± s.d.) offive sets of experiments were calculated. We tested the function of Wnt1and Wnt5a on E18.5 cortical axons and found that in contrast to P0,Wnt1 and Wnt5a did not repel motor cortical axons from E18.5 (datanot shown). CST axons reach the spinal cord at P0. At E18.5, the CSTaxons are still in the midbrain and the hindbrain. The timing of Wnt1responsiveness is consistent with its potential role in pathfinding forCST axons as they enter the spinal cord.

These long axons that were growing out in the collagen gel and wererepelled by Wnts stained positively for two CST markers, a monoclonalantibody to neural cell adhesion molecule (N-CAM), 5A5 (ref. 6;Fig. 2e), and polyclonal antibodies to L1 (anti-L1; Fig. 2f). Therefore,these axons are CST axons, and the markers did not change aftercoculturing with Wnt1-expressing COS cell aggregates.

Ryk is expressed in CST axons

Axon guidance molecules are often bifunctional, attracting some axonsbut repelling others, depending on the responding neurons. Vertebratespinal cord commissural axons are attracted by Wnts1, whereas motorcortical axons are repelled by Wnts. In Drosophila melanogaster, Wnt5has a repulsive role in pathway selection before midline crossing4. Thisrepulsion is mediated by a Wnt receptor named Derailed throughdirect binding and is independent of Frizzled4. Therefore, the vertebrate

b Corticospinal tract

0

Wnt5a Wnt5aWnt1 Wnt1e f g h

180160140120100

806040

Rel

ativ

e de

nsity

1 2 3 4 5 6 7A-P position

Efnb3Wnt1Wnt5a

c

d Cervical Thoracic

Anterior-posterior positionAnterior Posterior

Efnb3

Wnt5a

Wnt1

a

CST trajectory

Decussation

Figure 1 Anterior-posterior gradient of Wnt

expression around the dorsal funiculus.

(a) Schematic showing lateral view of CST

trajectory from the frontal cortex to the spinal

cord. (b) Transverse view of neonatal spinal cord

showing the location of CST in the ventral-most

area of the dorsal funiculus surrounded by gray

matter. (c) Quantification of in situ signals. Digitalimages of in situ hybridizations were analyzed by

NIH Image J, and the density of the positive

grayscale signal was measured. The signal density

of Wnt1 and Wnt5a were averaged from four sets

of in situ hybridizations from four different mice.

A-P, anterior-posterior. (d) In situ hybridization of

transverse sections along the anterior-to-posterior

axis probed with Wnt1 and Wnt5a, showing

decreasing expression gradients of both genes

along the anterior-posterior axis. Efnb3 is

expressed at the dorsal midline (arrows) at the

same stages but does not show an anterior-

posterior gradient. (e,f) In situ hybridization of

transverse sections of P0 mouse spinal cord

for (e) Wnt1 and (f) Wnt5a. (g,h) Higher

magnification of (g) Wnt1 and (h) Wnt5a in situ

hybridization, showing expression in larger (large

arrowheads) and smaller (small arrowhead) cells.

Scale bars, 100 mm.


ART ICLES©

2006

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

homolog of Derailed, Ryk, is a candidatereceptor for Wnt-mediated repulsion ofCST axons.

We first generated an in situ probe for Rykand carried out in situ hybridization. Ryk wasexpressed in layers 5 and 6 of the frontal andsensorimotor cortices at P0 as well as in layers2 and 3 (Fig. 3a,b). At E18.5, Ryk wasexpressed in layers 2 and 3 but was barely detectable in layers 5 and6, and the signal was weaker than that of P0 (Fig. 3a,b). This correlateswith the observation that E18.5 cortical axons are not repelled by Wnts,whereas P0 axons are robustly repelled. We then generated polyclonalantibodies to the extracellular domain of Ryk and further confirmedthat Ryk protein is present in layer 5/6 cells in P0 motor cortex but notin layers 5/6 in E18.5 motor cortex (Fig. 3c). At E18.5, however, onlylayer 2/3 cells expressed Ryk protein (Fig. 3c). In sagittal sections, thepyramidal tracts showed highly specific Ryk staining. At P0, CSTaxons have just arrived at the medulla–spinal cord junction, and Rykexpression was stronger at the distal end of the tracts (Fig. 3d). At P5,when the CST axons have already entered the spinal cord and havestarted pathfinding in the dorsal funiculus, Ryk expression was higherin general (Fig. 3d). Anti-Ryk stained the CST axons specifically in thedorsal funiculus (Fig. 3d). Anti-L1 staining was carried out to show thetrajectory of the CST. The L1 signal was weaker in the distal end of theCST than in the proximal end (Fig. 3d). Additional L1 immunoreac-tivity was seen in areas anterior to the pons (Fig. 3d), suggesting that L1is expressed in broader areas in the brain than Ryk. The CST axons thatform the pyramidal decussation (Fig. 3d) and the pyramidal tracts inthe dorsal funiculus of the spinal cord (Fig. 3e) expressed high amountsof Ryk. Therefore, Ryk is expressed in CST axons at the appropriatetime to mediate Wnt repulsion and may be involved in the descendinggrowth of CST axons within the spinal cord.

To test whether Ryk is a vertebrate receptor for Wnts, we generatedWnt1–alkaline phosphatase (AP) fusion proteins and carried outbinding assays with COS cells transfected with Ryk. We found thatWnt1 can bind to Ryk, as well as to Frizzled (Fzd3) as a control

(Fig. 3f). To assess the affinity of the Wnt-Ryk interaction, wequantified binding and found that the Kd for Wnt1-AP to Ryk was7.89 nM. In parallel experiments, the affinity of Wnt1-AP to Fzd3 was53.91 nM (Fig. 3f). Therefore, Ryk is a high-affinity Wnt receptor.

To test whether Ryk is expressed in the motor cortical explantsshown in Figure 2 and whether Wnts affect the expression of thisreceptor by changing cell fate, we stained explants exposed to Wnt1-expressing cell aggregates with anti-Ryk (Fig. 3g). The cortical explantsthat were exposed to Wnt1 still expressed similar amounts of Rykprotein, and the Ryk-positive axons were actively repelled by Wnt1cell aggregates. Many axons were deflected away from the Wnt1 cellclumps (arrows in Fig. 3g).

Anti-Ryk blocks the repulsion by Wnts

To determine whether Ryk is involved in mediating Wnt repulsion invertebrate axons and whether it functions as a Wnt receptor, we usedpolyclonal antibodies generated to the ectodomain of Ryk (anti-Ryk)and tested whether anti-Ryk could block the repulsion by Wnts incollagen gel assays. We found that addition of purified anti-Ryk incollagen gel assays blocked the repulsive effects of Wnt proteins,whereas the preimmune control did not (Fig. 4a,b). In the presenceof preimmune serum, motor cortical axons tended to grow away fromthe point source of Wnt1 and Wnt5a. When anti-Ryk was included,motor cortical axons were no longer repelled. Therefore, Ryk isrequired for mediating Wnt repulsion of CST axons in collagen gelassays. Results (Fig. 4a,b) were quantified and averaged from five sets ofexperiments (Fig. 4g). In each set, eight explants were included for eachof the eight conditions (vector + preimmune, Wnt1 + preimmune,

1.2

1.0

0.8

0.6

0.4

0.2

0

Rel

ativ

e to

tal a

xon

bund

le le

ngth

Vector Wnt1 Vector Wnt5a

Undiluted cell aggregates

1.2

1.0

0.8

0.6

0.4

0.2

0

P/D

ratio

Vector Wnt1 Vector Wnt5a

Diluted cell aggregates

a b

c d

e f

β−Tubulin β−Tubulin β−Tubulin β−Tubulin

β−Tubulin

Vector

Vector

L1 L1

Vector

Vector (1:10)

Wnt5a(1:10)

Vector (1:20)

Wnt1 (1:20)

Wnt1 (1:20)

Wnt1 (1:20)

N-CAM(5A5)

N-CAM(5A5)

Vector Wnt5a

Wnt1

β−Tubulin β−Tubulin β−Tubulin

Figure 2 Wnts repel motor cortical axons.

(a) Wnt1- or Wnt5a-expressing COS cell aggre-

gates (undiluted) showed strong inhibition of the

growth of P0 layer 5 frontal cortical axons,

stained with b-tubulin antibodies. (b) Axons from

layer 5 explants of P0 frontal cortex were repelled

by Wnt1- or Wnt5a-transfected COS cells (diluted

to show repulsion: Wnt1, 1:20; Wnt5a, 1:10).(c) Quantification of the inhibitory effects of Wnts.

The relative total axon bundle lengths were

obtained by the ratio of total axons from explants

exposed to Wnt-transfected COS cell aggregates

over explants exposed to vector-transfected COS

cell aggregates. (d) Quantification of Wnt

repulsion of motor cortical axons. The repulsion

was measured by the proximal/distal (P/D) ratio:

ratio of axon length in the proximal quadrant

versus that in the distal quadrant. (e) Axons from

layer 5 frontal cortical explants stained positively

with a CST axon marker, 5A5 (N-CAM6) and were

repelled by Wnt1-expressing COS cell aggregates

(diluted 1:20 with untransfected cells). (f) Axons

from layer 5 frontal cortical explants stained

positively with another CST marker, L1, and were

repelled by Wnt-1-expressing COS cell aggregates

(diluted 1:20). Scale bars, 100 mm.


ART ICLES©

2006

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

vector + anti-Ryk, Wnt1 + anti-Ryk, vector + preimmune, Wnt5a +preimmune, vector + anti-Ryk, Wnt5a + anti-Ryk).

To test whether blocking of Wnt repulsion by anti-Ryk is specific, wetested whether anti-Ryk can block the repulsion of other repellents,such as the slit proteins. In the presence of preimmune serum, Slit2 canrepel postcrossing commissural axons in a collagen gel assay7. Inthe presence of anti-Ryk, Slit2 was still able to repel postcrossingcommissural axons (Fig. 4c,e; two sets of experiments were averaged).

Multiple explants were included in each of the four conditions (vector+ preimmune, n ¼ 11; Slit2 + preimmune, n ¼ 8; vector + anti-Ryk,n ¼ 8; Slit2 + anti-Ryk, n ¼ 17).

To test further whether anti-Ryk specifically blocks Wnt-mediatedrepulsion of CST axons, we tested whether anti-Ryk can block Wnt-mediated attraction. Wnt4 can attract postcrossing commissuralaxons1. When anti-Ryk was added to the culture, commissuralaxons were still attracted by Wnt4-expressing COS cell aggregates

E18.5 E18.5 E18.5

E18.5E18.5E18.5

P0 P0P5 P5

P5

L1

*

P5P5

P0P0 P0

P0P0P0

Ryk

Sense Sense

Ryk Ryk Ryk Ryk

Ryk

Wnt1-AP+Ryk

Wnt1-AP+Vector

Wnt1-AP+ Fzd3

Ryk Ryk

N-CAM(5A5)Ryk Ryk

Ryk Ryk

Ryk

Wnt1(1:20)

Ryk

Wnt1(1:20)

Vector

Ryk

RykRykRykRyk

II/III

V/VI

II/III II/III II/IIIII/III II/III

V/VI V/VI

V/VI

V/VI

V/VI V/VI

V/VIV/VI

V/VI

1.0

0.8

0.6

0.4

0.2

0.00 250

OD

500Conc. of Wnt1-AP

750

Wnt1 binding assay

Wnt1 binds RYK, Kd = 7.890 nM

Wnt1 binds Fzd3, Kd = 53.91 nM

1,000

a

d

f g

e

b c

Figure 3 A repulsive Wnt receptor, Ryk, is expressed on CST axons. (a,b) Temporal pattern of Ryk expression at E18.5 and P0 shown at (a) lower and (b) higher

magnification. Sense, sense control. (c) Temporal pattern of Ryk expression by immunostaining. (d) In sagittal and transverse sections, Ryk protein was highlyand specifically expressed along the CST trajectory from P0 to P5. Ryk immunostaining became highly enriched toward the distal end of the CST axons as they

approached the pyramidal decussation (arrow) as seen at P0. At P5, Ryk antibodies stained specifically the ventral-most regions of the dorsal funiculus where

CST axons project, similar to L1 antibody staining (big arrowhead). The L1 antibodies stained broader regions than did Ryk immunostaining (areas anterior to

the pons, indicated by asterisk). As shown in transverse sections, Ryk is highly enriched and specifically expressed on the crossing CST axons at the pyramidal

decussation at P0 (horizontal arrow). (e) Immunostaining of transverse sections of P5 cervical spinal cord. Anti-Ryk stained the ventral-most areas of the dorsal

funiculus, similar to the CST marker, 5A5 (arrows). (f) Quantification of Wnt1 binding to its receptors. Ryk is a higher-affinity Wnt1 receptor than Fzd3. AP,

alkaline phosphatase. (g) Axons from P0 layer 5 frontal cortical explants stained positively with the Ryk antibodies and were repelled by Wnt1-transfected COS

cell aggregates. Higher-magnification images (right) show Ryk-positive layer 5 cortical axons being deflected away from the Wnt1-expressing COS cells

(arrows). Scale bars: 200 mm in a, 100 mm (top) and 200 mm (bottom) in b, 100 mm in c and d, 50 mm in e and f, 200 mm (left) and 100 mm (right) in g.


ART ICLES©

2006

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

(Fig. 4d,f). Multiple explants were tested in each of the four conditions(vector + preimmune, n ¼ 7; Wnt4 + preimmune, n ¼ 6; vector+ anti-Ryk, n ¼ 7; Wnt4 + anti-Ryk, n ¼ 7). Therefore, anti-Rykdoes not block all Wnt-mediated effects on axons, documentingthe specificity of anti-Ryk in blocking Wnt-mediated repulsion ofCST axons.

Anterior-posterior graded chemorepulsion by neonatal cord

As Wnts are diffusible guidance cues and are expressed in an anterior-posterior graded fashion, it is possible that such an anterior-posteriorgradient provides directional information for CST axons along thelongitudinal axis of the spinal cord. We carried out cocultures of motorcortical axons with neonatal spinal cord tissue in collagen gel and foundthat neonatal spinal cord secretes diffusible repellents that repel CSTaxons. We first collected vibratome sections (slices) along the anterior-posterior axis of the spinal cord (Fig. 5a) and then cut the dorsal spinalcord areas where the CST axons pathfind and placed dorsal spinal cordtissue pieces of equal size next to the cortical explants (Fig. 5b).Anterior spinal cord tissue (from the cervical regions) showed strongerrepulsion than did posterior spinal cord tissue (from the thoracicregions). Along the anterior-posterior axis (positions 1, 2 and 3),there was a decreasing gradient of repulsive effect (Fig. 5b). Toinvestigate further the identity of the repellents, we tested whetheranti-Ryk can block the spinal cord–derived repellent(s) and foundthat in the presence of anti-Ryk, CST axons showed radial growth,whereas in the preimmune control, CST axons were repelledby neonatal spinal cord in an anterior-posterior graded fashion(Fig. 5b,c). The graded repulsive effect of neonatal spinal cord islikely due to the graded distribution of the Wnt proteins, which may

provide directional information for cortical motor axons to growposteriorly along the spinal cord.

Anti-Ryk blocked posterior growth of CST axons

To test whether a Wnt-Ryk interaction is required for the anterior-posterior guidance of CST axons in vivo, we injected purified anti-Rykinto cervical spinal cord. Mice were injected with 50 mg/ml antibodiesor preimmune control on P1 and P3 and were killed on P5 and fixed bycardiac perfusion. Along the anterior-posterior axis, sections weretaken every 800 mm and were analyzed by immunostaining withanti–N-CAM monoclonal antibody 5A5, a CST marker. Mice injectedwith anti-Ryk showed a marked reduction in CST fibers posterior tothe injection site but an increase in CST areas anterior to the injectionsites, whereas mice injected with the vehicle control, artificial cere-brospinal fluid (ACSF) or with preimmune serum showed normal CSTareas (Fig. 6a). Similar results were obtained in six different sets ofmice. Each set included ACSF-, preimmune- and anti-Ryk–injectedmice. CST areas were quantified from four spinal cords (Fig. 6b). CSTaxon areas were larger anterior to the injection sites, suggesting that theCST fibers may be stalling anterior to the injection site.

To characterize further the CST axon projections after anti-Rykinjection, we did DiI tracing and viewed the CST axons along theanterior-posterior longitudinal axis. Small DiI crystals were plantedinto the medulla–spinal cord junction at the pyramidal decussation,and DiI was allowed to diffuse for 6 weeks. Sagittal sections were takenat a thickness of 10 mm. The entire CST axon bundle was usuallycontained in three consecutive sagittal sections. Therefore, we evaluatedthe entire CST axon bundle by looking at these three sections(Fig. 6c,d). CSTaxons in mice injected with anti-Ryk were substantially

reduced posterior to the injection area ascompared with those injected with vehicle(ACSF) or preimmune serum. CST axon bun-dles anterior to the injection area appearedwider and less compact as compared withvehicle and preimmune controls (Fig. 6d). Atotal of five sets of mice were analyzed for eachtreatment (ACSF, preimmune or anti-Rykinjections) and yielded similar results. TheDiI tracing results are consistent with theobservation of CSTstaining in serial transversesections (Fig. 6a,b). A small amount of pur-ified antibodies (1 ml) was injected intrathe-cally (between the dura and spinal cordtissue), which should not damage the spinalcord tissue. In fact, the control vehicle and

Preimmune Vector

Preimmune Vector

αRyk αRykVector

αRyk Vector αRyk Wnt4

Preimmune Slit2

Preimmune Wnt4

Slit2

VectorPreimmune

Wnt1 Preimmune

VectorPreimmune

Wnt5aPreimmune

VectorαRyk

Wnt1 αRyk

VectorαRyk

Wnt5aαRyk

Rel

ativ

e to

tal a

xon

bund

le le

ngth

P/D

rat

io

Tota

l axo

n bu

ndle

leng

th (

mm

)1.2

0.8

0.4

0

Vecto

r + p

re

Vecto

r + p

re

Vecto

r + αRyk

Wnt

4 +

pre

Wnt

1 +

pre

Vecto

r + αRyk

Vecto

r + p

re

Wnt

1 +

αRyk

Wnt

5a +

pre

Vecto

r + αRyk

Wnt

5a +

αRyk

Wnt

4 +

αRyk

1.2

0.8

1.0

0.6

0.4

0.2

0

3,0002,5002,0001,5001,000

5000

a

b

e f

g

d

c

Figure 4 Ryk is required for Wnt1 and Wnt5a

repulsion. (a,b) Anti-Ryk (aRyk; 50 mg/ml) blocked

repulsion of P0 layer 5 motor cortical axons by

(a) Wnt1 and (b) Wnt5a. (c) Anti-Ryk (50 mg/ml)

did not block repulsion of postcrossing spinal

cord commissural axons by Slit2. (d) Anti-Ryk

(50 mg/ml) did not block the attraction of crossing

commissural axons by Wnt4. (e) Quantification

of Slit2 repulsion of postcrossing commissuralaxons in the presence and absence of anti-Ryk

antibodies. (f) Quantification of Wnt4 attraction of

postcrossing commissural axons in the presence

and absence of anti-Ryk. (g) Quantification of

Wnt1 and Wnt5a effects on P0 layer 5 motor

cortical axons in the presence and absence of

anti-Ryk. Scale bars, 100 mm.


ART ICLES©

2006

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

preimmune serum injections did not cause any defects or damage.Taken together, our results indicate that Wnt-Ryk signaling is requiredfor posterior growth of CST axons within the dorsal funiculus of theneonatal rodent spinal cord in vivo.

DISCUSSION

CST axons are one of the major descending motor pathways andcontrol voluntary body movement and fine motor functions of thelimbs. They are some of the longest axons in the human or animal bodyand connect the motor cortex to motor circuits at various anterior-posterior positions in the spinal cord. The development and regenera-tion of the CST have been studied intensely, but very little is knownabout the guidance cues that are involved. Until now, only a couple ofmolecules have been found to be important in regulating the pathfind-ing of CSTaxons from the brain to the spinal cord. Neural cell adhesionmolecules of the immunoglobulin superfamily (IgCAMs) have beenimplicated in both the fasciculation and guidance of axons. L1, amember of the IgCAM family, has been implicated in many neuralprocesses and is expressed widely in the embryonic and adult nervous

systems8. L1 function is necessary for the guidance of corticospinalaxons across the pyramidal decussation in mice. Although the pathwayto the caudal medulla seems normal, a substantial proportion of axonsfail to cross the midline to the opposite dorsal column in L1-deficientmice8. L1 is a component of the Sema3A receptor complex, and L1mutations may disrupt Sema3A signaling, leading to guidance errors9.The receptor tyrosine kinase EphA4 mediates midline pathfinding ofCST axons. In mice with null mutations in the gene encoding EphA4,anatomical studies and anterograde tracing experiments show majordisruptions of the CST within the medulla and spinal cord in the nullmutant mice, including aberrant recrossing of the dorsal midline,although CST pathfinding in the forebrain and midbrain is normal10.A recent study identified a gene, Bcl11b (also known as Ctip2), whichencodes a transcription factor, that is critical for the development ofaxonal projections of corticospinal motor neurons to the spinal cordin vivo11. In Bcl11b–/– mice, CST axons lack fasciculation in the sub-cerebral projections in the internal capsule after the postnatal stage.

A major question of CST guidance is how the corticospinal motoraxons pathfind along the anterior-posterior direction in the postnatalspinal cord. The directional cue has remained unknown. In thisstudy, we found that several Wnt proteins form a high-to-low ante-rior-posterior gradient in the cervical and thoracic spinal cord duringthe first week of postnatal life of mice, when CST axons first enter thespinal cord and project posteriorly. These Wnt proteins are potentrepellents of CST axons, and this repulsion is mediated by a conservedreceptor, Ryk, a protein tyrosine kinase–like transmembrane protein.Neonatal spinal cord secreted diffusible repellents in an anterior-posterior graded fashion, and this repulsion was blocked by anti-Ryk.Blocking Ryk in developing spinal cord led to a marked reductionin CST fiber formation posterior to the injection sites and to theaccumulation of CST axons anterior to the injection sites in vivo,suggesting that the Wnt proteins are essential for the proper develop-ment of CST axons.

Axon connections along the anterior-posterior (rostral-caudal) axisare an important aspect of nervous system wiring. Much attention hasbeen devoted to understanding axon connections along the dorsal-ventral axis, such as growing toward and away from the midline12. Verylittle is known about the molecular guidance cues along the anterior-posterior axis in any animal system. Wnts control the anterior turningof postcrossing commissural axons at the ventral midline of midgesta-tion embryos1. Our current study documents the function of Wntgradients in guiding the posterior growth of CST axons at the neonatalstage while they are pathfinding at the cervical and thoracic levels. Wethus propose that Wnts may form gradients to regulate axon pathfind-ing of both ascending sensory axons, such as the spinal cord commis-sural axons1, and descending motor axons, such as the CST axons, andthat Wnt family proteins are potentially general anterior-posteriorguidance signals for spinal cord axons.

c

b

a

1.2

1

A P

2 3

1.0

0.8

0.6

0.4

0.2

0Control Preimmune

Pre

imm

une

P/D

rat

io

αRyk

Con

trol

Cervical ThoracicαR

ykFigure 5 Neonatal spinal cord secretes diffusible repellent(s) in an anterior-

posterior graded fashion, which can be blocked by anti-Ryk. (a) Diagram

showing the three anterior-posterior positions where neonatal (P0) mouse

spinal cord slices were cut by tissue chopper. Equal-sized dorsal spinal cord

areas surrounding the dorsal funiculus were dissected out and positioned next

to the neonatal (P0) cortical explants from the same mouse. (b) The anterior-

posterior graded repulsive effects of neonatal spinal cord (anterior-posterior

positions 1, 2, 3) can be blocked by anti-Ryk. (c) Quantification of proximal/distal ratios. The results were averaged from three sets of assays. In each set,

three anterior-posterior positions (1, 2, 3) were tested in three conditions:

control, preimmune serum and anti-Ryk. For each condition and each

anterior-posterior position, two cortical explants were tested and averaged

in each set. Scale bar, 100 mm.


ART ICLES©

2006

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

The trajectory of CSTaxons is quite complex, particularly before theyreach the spinal cord (Fig. 1a). Therefore, it is conceivable that multipleguidance cues may be involved in defining their pathways. At P0, CSTaxons were strongly repelled by Wnts, but E18.5 CST axons were notrepelled. The onset of Wnt responsiveness coincides with the upregula-tion of Ryk immunoreactivity in the deep layers of frontal motor cortex(Fig. 3c) and the enrichment of Ryk protein at the pyramidal decussa-tion (Fig. 3d). Therefore, Wnts probably begin to repel CSTaxons oncethey have reached the medulla–spinal cord junction, and the Wntexpression gradient along the anterior-posterior axis provides direc-tionality for CST axons to grow posteriorly. Wnt1 and Wnt5a tran-scripts were expressed at even higher amounts in the medulla anteriorto the decussation (data not shown). It is possible that Wnt repulsion isinvolved in the posterior turning at the decussation and in thesubsequent descending growth of CST axons in the spinal cord. It istechnically challenging to deliver antibodies reliably to the pyramidaldecussation in neonatal mice to test the role of Ryk in the initialposterior turning. Intrathecal injection could only deliver anti-Rykbetween the joint of C4 and C5 because of the small size of neonatalmice. Injection of anti-Ryk into the upper neonatal cervical spinal corddid, however, result in stalling of CST axons and in the reduction of

CST fibers in the dorsal funiculus, supporting the role of Wnts in CSTguidance along the anterior-posterior axis inside the spinal cord in vivo.Anti-Ryk injection resulted primarily in the stalling of CST axons(Fig. 6). Posterior to the injection site, CST areas were greatly reduced,whereas CST axons appeared widened and less compact anterior to theinjection site. No other defects, such as random overshooting orabnormal branching were observed. It is possible that other guidancemolecules are involved in restricting the CST fibers within the ventralareas of the dorsal funiculus and that the primary function of Wnts is toguide the CST axons to grow posteriorly. The conventional Rykknockout mice typically die at the day of birth and display severepatterning defects5. The Ryk small interfering RNA transgenic mice alsotypically die at birth13. Therefore, phenotypic analyses of postnatal CSTaxon guidance in these mice would be complicated by issues of earlydevelopmental defects and potentially by other functions of Ryk.

Wnts are recently identified axon guidance molecules and seem tofunction as both attractants and repellents. Wnts attract postcrossingcommissural axons in a Fzd3-dependent manner1. Ryk may be requiredfor Wnt1- and Wnt3a-stimulated neurite outgrowth from dorsal rootganglion13. In the D. melanogaster midline, a subset of derailed-expressing commissural axons was found to be repelled by Wnt5

1a

b c

d

2 3 4 5 6 7 8 9

Anterior-posterior position

Anterior-posterior position

10 11 12 13 14

Section 1

VehicleVehicle (ACSF)

200180160140120100

CS

T a

rea

(10–

4 m

m2 )

806040200

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Preimmune

Preimmune

αRyk

Vehicle

Preimmune

αRyk

Vehicle

Preimmune

αRyk

αRyk

Section 2 Section 3

Section 1 Section 2 Section 3

15 16 17 18

Figure 6 Anti-Ryk blocked posterior growth of CST axons in vivo. (a) Immunostaining with 5A5 showing that the CST areas are reduced in size and intensity

posterior to the injection sites but are enlarged anterior to the injection sites (arrow) in mice injected with anti-Ryk. (b) Quantification of CST areas. CST areas

in antibody-injected mice anterior to the injection site are larger than in controls, particularly in anterior-posterior position 8, suggesting that CST axons were

stalling anterior to the injection site, anterior-posterior position 9 (arrow). (c) Serial sagittal sections (sections 1–3) of P5 mouse showing the entire CST

bundles in mice injected with vehicle, preimmune serum or anti-Ryk. Green arrows indicate the same anterior position that is consistent in all mice along the

anterior-posterior axis as a reference point. Red arrows indicate the posterior end of the DiI labeling that is clearly visible. The antibody-injected mice showed

markedly reduced growth of CST axons, similar to the results shown in a in serial transverse sections. (d) Higher-magnification pictures of c, showing details ofDiI tracing of the CST axons. CST axons failed to grow efficiently posteriorly when injected with anti-Ryk. CST bundles with anti-Ryk appear wider and less

compact (white arrows). Red arrows, approximate injection sites. Scale bars: 100 mm in a, 1,400 mm in c,d.


ART ICLES©

2006

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

(ref. 4). We show here direct evidence that vertebrate CST axons arerepelled by Wnts (Wnt1 and Wnt5a) and that the repulsion is mediatedby the vertebrate homolog of Derailed, Ryk5. In characterizing theWnt-Frizzled and Wnt-Ryk biochemical interaction, we found thatSfrp2 (secreted frizzle-related sequence protein 2) can block Wnt-Frizzled binding but cannot block Wnt-Ryk binding and, conversely,Ryk antibodies (to the WIF domain) can block Wnt-Ryk binding butcannot block Wnt-Frizzled binding (data not shown). We found thatSfrp2 did not block the repulsion of CSTaxons by Wnts, suggesting thatFrizzleds are not required to mediate Wnt repulsion of CSTaxons (datanot shown). Therefore, Frizzleds have so far been found to mediateonly attraction, and Ryk/Derailed is involved in repulsion and perhapsalso attraction13. We also tested whether other Wnts have similar effectson CST axon growth in collagen gel assays and found that Wnt4, Wnt6and Wnt7b did not have any effect (data not shown). These Wnts thathad no effect on CST axons were found to attract the postcrossingcommissural axons1. Therefore, there seems to be specificity of Wnts inregulating different classes of axons, which may be determined by thebinding specificity of Wnts to different classes of receptors, Ryk orFrizzleds. More and more studies document the evidence of involve-ment of Wnts in axon guidance, which will eventually lead to a betterunderstanding of the specificity of Wnts and their effects by means oftheir cognate receptors.

METHODSIn situ hybridization and immunohistochemistry. Specific in situ probes for

the entire mouse Wnt gene family, Wnt1, Wnt2, Wnt2b, Wnt3, Wnt3a, Wnt4,

Wnt5a, Wnt5b, Wnt6, Wnt7a, Wnt7b, Wnt8a, Wnt8b, Wnt9a, Wnt9b, Wnt10a,

Wnt10b, Wnt11 and Wnt16, were cloned by PCR with reverse transcription (RT-

PCR) from various stages of mouse embryos and adult mouse brains into the

pCRII vector (Invitrogen). Sections of neonatal spinal cords were collected every

1.6 mm along the anterior-posterior axis for evaluation of the expression of all

Wnt genes (19 members) along the anterior-posterior axis. The Ryk in situ probe

was cloned by RT-PCR from mouse E13.5 cDNA. The 1-kb probe included 500

nucleotides of 3¢ untranslated region and 500 nucleotides of the coding region at

the carboxy terminus. Fetal or neonatal mouse brains were sectioned for analysis

of Ryk expression. In situ hybridizations were carried out as described14.

Polyclonal anti-Ryk antibodies were generated against the ectodomain of

Ryk, amino acids 90–183, fused with maltose binding protein, which was

purified and injected into rabbits15. Fetal and neonatal brains were sectioned

for analyses of Ryk distribution in CST axons by use of purified anti-Ryk. The

N-CAM antibody 5A5 was purchased from the Developmental Studies Hybri-

doma Bank. Spinal cord sections were taken every 800 mm along the anterior-

posterior axis to evaluate CST axon projections using 5A5 as a CST marker6.

Immunohistochemistry was carried out as described16.

Wnt-receptor binding assays. Wnt-receptor binding assays were carried out as

described17, with slight modification. HEK293 cells were transfected with a

Wnt1-AP fusion construct using FuGENE6. Cells were switched to Opti-MEM

when they grew to 80–90% confluence after transfection. The medium was

conditioned for an additional 48 h and concentrated 20-fold using Centriprep

YM-30 (Millipore Corp) at 4 1C. COS cells transfected with a Ryk full-length

expression construct, cloned by RT-PCR from adult mouse cDNA, or a Fzd3

full-length expression construct, cloned by RT-PCR from E11.5 mouse cDNA,

were grown on glass coverslips, rinsed once with binding buffer (Hanks’

balanced salt solution with 0.5 mg/ml BSA, 0.1% NaN3 and 20 mM HEPES,

pH 7.0), incubated with Wnt1-AP-containing conditioned medium for 75 min

at room temperature and washed six times in binding buffer at room

temperature for a total of 45 min. Cells were then fixed for 1 min in 60%

acetone and 3% formaldehyde in 20 mM HEPES (pH 7.0), washed three times

in 150 mM NaCl and 20 mM HEPES (pH 7.0) and heated at 65 1C for 2 h.

After being washed once with alkaline phophatase buffer (0.1 M Tris-HCl,

pH 9.5; 0.1 M NaCl; 5 mM MgCl), cells were stained at room temperature

overnight with 5-bromo-4-chloro-3-indolyl-phosphate (165 mg/ml)/nitroblue

tetrazolium (330 mg/ml) in alkaline phosphatase buffer for color detection.

Explant assays. Whole brains from P0 mice were dissected out and kept

immersed in L15 medium. Unwanted portions (brain stem and hindbrain)

were trimmed with fine forceps and razor blade knives. The forebrain was

sectioned in the coronal plane with a McILWAIN tissue chopper into cortical

slices (250 mm thick), which were transferred to fresh L15 medium.

Layer 5 cortical explants were dissected out from cortical slices containing

frontal and sensorimotor cortices with sharp tungsten needles.

COS7 cells transfected with Wnt1 or Wnt5a expression constructs or with

vector only were made into hanging drops and were cut into aggregates.

Different COS cell aggregates and motor cortical explants were embedded in

collagen matrix and were cultured in 10% FBS MEM at 37 1C in 5% CO2

for 60 h. For antibody blocking assays, purified postimmune Ryk antibody or

preimmune control serum was added in culture medium with a final

concentration 50 mg/ml. After culture, motor cortical explants were

fixed, and whole-mount immunofluorescence was done with the monoclonal

b-tubulin antibody E7 or with 5A5. Using NIH Image J, the total length of

axons emerging from explants or the length of axons emerging from the distal

and proximate sides of explants toward COS cell aggregates was measured,

which was then converted into proximal/distal ratios. In every experiment,

the average relative outgrowth or proximal/distal ratio under each condition

(exposed to Wnts or vector control or with preimmune or postimmune serum)

was measured from eight explants. The means of five experiments

were calculated.

For spinal cord tissue and cortical explant coculture experiments, cervical

and thoracic spinal cords were cut into three parts of equal length along the

anterior-posterior axis, using a McILWAIN tissue chopper. Each part was

sectioned in the coronal plane into slices of spinal cord (250 mm thick). These

slices were transferred to fresh L15 medium and the dorsal part of the spinal

cord was dissected out with sharp needles.

Postcrossing commissural axon assays were carried out as previously

described7.

Intrathecal injection. Preimmune serum and postimmune anti-Ryk were

purified with protein A-G beads and were dialyzed into ACSF. Newborn ICR

mice were briefly anesthetized with halothane (1–3%). An incision was

made in the midline of the spine between the joint of the C4-C5

vertebrae for injection into the spinal cord at the cervical C1-C2 level. The

thin muscles were separated slightly so that the joints of the vertebrae could

be easily identified. Then a fine, blunt stainless steel needle

with a syringe filled with purified serum (50 mg/ml) in ACSF was inserted

approximately 3 mm into the vertebrae in a rostral direction. The reagent (1 ml)

was injected into each mouse, and the muscles and skin were sutured.

Intrathecal administrations were made on P1 and P3 with the same

protocol. On P5, mice were killed and fixed with cardiac perfusion of 4%

paraformaldehyde before collecting tissues. Four sets of experiments (ACSF,

preimmune and postimmune) were quantified (Fig. 6b). The intrathecal

injection experiments were carried out in agreement with the regulations of

the ethics committee of the International Association for the Study of Pain and

were approved by the Parker Research Institute Animal Care and Use

Committee. Serial sections were obtained along the anterior-posterior axis,

and the CST axons were examined by immunohistochemistry with 5A5

monoclonal antibody.

CST tracing. Anterograde DiI tracing of CST axons was done in fixed P5 tissue

including whole brain and spinal cord. Small DiI crystals were put in the

junction area between the hindbrain and cervical spinal cord, where CST axons

form the pyramidal decussation. DiI-labeled tissue in 4% paraformaldehyde

was left at 4 1C for 6 weeks. After embedding, DiI-labeled spinal cord tissues

were immediately sectioned sagittally at a thickness of 10 mm and were

analyzed. All sections containing CST bundles were collected and compared

as shown (Fig. 6c,d). Five sets of intrathecal injection and DiI tracing

experiments were carried out, yielding similar results.

All animal experiments were approved by the University of Chicago

Institutional Animal Care and Use Committee.

Accession codes. BIND identifiers (http://bind.ca): 315474 and 315475.



ART ICLES©

2006

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

ACKNOWLEDGMENTSThis work was supported by a Career Development Award from the SchweppeFoundation, Robert Packard ALS Center at Johns Hopkins University, the BrainResearch Foundation and the Jack Miller Peripheral Neuropathy Center at theUniversity of Chicago. We thank laboratory members L. King, A. Wolf,A. Schmitt, R. Sherman, J. Chen and A. Fenstermaker for comments andN. Milanesio for help with probes and reagents. We thank F.G. Rathjenfor the anti-L1.




1. Lyuksyutova, A.I. et al. Anterior-posterior guidance of commissural axons by Wnt-frizzledsignaling. Science 302, 1984–1988 (2003).

2. Paxinos, G. The Rat Nervous System (Academic, New York, 1995).3. Gianino, S. et al. Postnatal growth of corticospinal axons in the spinal cord of developing

mice. Brain Res. Dev. Brain Res. 112, 189–204 (1999).4. Yoshikawa, S., McKinnon, R.D., Kokel, M. & Thomas, J.B. Wnt-mediated axon guidance

via the Drosophila Derailed receptor. Nature 422, 583–588 (2003).5. Halford, M.M. et al. Ryk-deficient mice exhibit craniofacial defects associated with

perturbed Eph receptor crosstalk. Nat. Genet. 25, 414–418 (2000).6. Joosten, E.A., Reshilov, L.N., Gispen, W.H. & Bar, P.R. Embryonic form of N-CAM

and development of the rat corticospinal tract; immuno-electron microscopical

localization during spinal white matter ingrowth. Brain Res. Dev. Brain Res. 94,99–105 (1996).

7. Zou, Y., Stoeckli, E., Chen, H. & Tessier-Lavigne, M. Squeezing axons out of the graymatter: a role for slit and semaphorin proteins from midline and ventral spinal cord.Cell 102, 363–375 (2000).

8. Cohen, N.R. et al. Errors in corticospinal axon guidance in mice lacking the neural celladhesion molecule L1. Curr. Biol. 8, 26–33 (1998).

9. Castellani, V., Chedotal, A., Schachner, M., Faivre-Sarrailh, C. & Rougon, G. Analysis ofthe L1-deficient mouse phenotype reveals cross-talk between Sema3A and L1 signalingpathways in axonal guidance. Neuron 27, 237–249 (2000).

10. Dottori, M. et al. EphA4 (Sek1) receptor tyrosine kinase is required for the developmentof the corticospinal tract. Proc. Natl. Acad. Sci. USA 95, 13248–13253 (1998).

11. Arlotta, P. et al.Neuronal subtype-specific genes that control corticospinal motor neurondevelopment in vivo. Neuron 45, 207–221 (2005).

12. Dickson, B.J. Molecular mechanisms of axon guidance. Science 298, 1959–1964(2002).

13. Lu, W., Yamamoto, V., Ortega, B. & Baltimore, D. Mammalian Ryk is a Wnt coreceptorrequired for stimulation of neurite outgrowth. Cell 119, 97–108 (2004).

14. Frohman, M.A., Boyle, M. & Martin, G.R. Isolation of the mouse Hox-2.9 gene; analysisof embryonic expression suggests that positional information along the anterior-posterioraxis is specified by mesoderm. Development 110, 589–607 (1990).

15. Hovens, C.M. et al.RYK, a receptor tyrosine kinase-related molecule with unusual kinasedomain motifs. Proc. Natl. Acad. Sci. USA 89, 11818–11822 (1992).

16. Serafini, T. et al. Netrin-1 is required for commissural axon guidance in the developingvertebrate nervous system. Cell 87, 1001–1014 (1996).

17. Cheng, H.J. & Flanagan, J.G. Identification and cloning of ELF-1, a developmentallyexpressed ligand for the Mek4 and Sek receptor tyrosine kinases. Cell 79, 157–168(1994).


ART ICLES©

2006

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 1 | JANUARY 2006 147

CO R R I G E N DA

Corrigendum: Protocadherin Celsr3 is crucial in axonal tract developmentFadel Tissir, Isabelle Bar, Yves Jossin, & André M GoffinetNat. Neurosci. 8, 451–457 (2005)

In the print version of this article and the version initially published online, an author name was omitted. The fourth author should have been listed as Olivier De Backer of the Molecular Physiology Research Unit, University of Namur Medical School, 61, rue de Bruxelles, B5000 Namur, Belgium. The error has been corrected in the HTML and PDF versions of the article. This correction has been appended to the PDF version. The authors regret the error.

Corrigendum: Ryk-mediated Wnt repulsion regulates posterior-directed growth of corticospinal tractYaobo Liu, Jun Shi, Chin-Chun Lu, Zheng-Bei Wang, Anna I Lyuksyutova, Xuejun Song & Yimin ZouNat. Neurosci. 8, 1151–1159 (2005)

In the print version of this article and the version initially published online, one author’s name was spelled incorrectly. The correct spelling is Xue-Jun Song. The error has been corrected in the HTML and PDF versions of the article. This correction has been appended to the PDF version. The authors regret the error.

Corrigendum: Activity-dependent decrease of excitability in rat hippocampal neurons through increases in IhYuan Fan, Desdemona Fricker, Darrin H Brager, Xixi Chen, Hui-Chen Lu, Raymond A Chitwood & Daniel JohnstonNat. Neurosci. 8, 1542–1551 (2005)

In the print version of this article and the version initially published online, the units for anisomycin concentration in the figure labels for Fig. 8d and f were incorrect. The correct concentration is 20 µM. The error has been corrected in the HTML and PDF versions of the article. This correction has been appended to the PDF version. The authors regret the error.

©20

06 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Activation of GPCRs modulates quantal size inchromaffin cells through Gbc and PKC

Xiao-Ke Chen1,6, Lie-Cheng Wang1,6, Yang Zhou1,2, Qian Cai3, Murali Prakriya4, Kai-Lai Duan1,2,Zu-Hang Sheng3, Christopher Lingle4 & Zhuan Zhou1,2,5

Exocytosis proceeds by either full fusion or ‘kiss-and-run’ between vesicle and plasma membrane. Switching between these two

modes permits the cell to regulate the kinetics and amount of secretion. Here we show that ATP receptor activation reduces

secretion downstream from cytosolic Ca21 elevation in rat adrenal chromaffin cells. This reduction is mediated by activation of a

pertussis toxin–sensitive Gi/o protein, leading to activation of Gbc subunits, which promote the ‘kiss-and-run’ mode by reducing the

total open time of the fusion pore during a vesicle fusion event. Furthermore, parallel activation of the muscarinic acetylcholine

receptor removes the inhibitory effects of ATP on secretion. This is mediated by a Gq pathway through protein kinase C activation.

The inhibitory effects of ATP and its reversal by protein kinase C activation are also shared by opioids and somatostatin. Thus, a

variety of G protein pathways exist to modulate Ca21-evoked secretion at specific steps in fusion pore formation.

In neuroendocrine cells, the large dense-core vesicles can fuse to plasmamembrane by two alternative modes: full fusion and ‘kiss-and-run’. Fullfusion occurs when the vesicular and plasma membranes merge and allthe contents are released. Kiss-and-run releases vesicle contentsthrough a transient fusion pore1–4. The kiss-and-run mechanism allowspartial release by limiting the open time of the fusion pore. Both vesiclerelease probability and the switch between full fusion and kiss-and-runare subject to presynaptic modulation in synaptic transmission or tohormone secretion in neuroendocrine cells. Modulation of vesiclerelease probability has been intensively investigated. However, little isknown about what determines the switch between full fusion andkiss-and-run3–5.

A major modulatory mechanism of evoked secretion involvesmodulation of Ca2+ channels by G protein–coupled receptors(GPCRs), thereby altering the extent of Ca2+ influx available to initiateexocytosis6,7. The molecular steps by which G protein activation8 leadsto increases or decreases in Ca2+ current are fairly well understood.However, there are also suggestions that steps in the exocytotic processafter Ca2+ elevation may also be targets for regulation by G proteins.For example, exocytosis in chromaffin cells and hippocampal neuronsseems to be favored by activation of protein kinase C (PKC)9,10, whileGABAB receptor activation inhibits vesicle priming in the synapse calyxof Held11. However, at present, our understanding of the mechanismsof G protein–mediated regulation of secretion remains limited.

The adrenal chromaffin cell provides an important model forinvestigating neurosecretion3,12,13. Chromaffin granules, in additionto the principal molecules of catecholamines and ATP, contain severalpeptides including somatostatin and opioids14. Chromaffin cells also

express a variety of GPCRs for endogenous transmitters and modula-tors, including ACh15, various peptides16 and ATP17. Our results showthat ATP (as well as opioids and somatostatin) causes a Gi/o-mediatedinhibition of secretion by a mechanism involving Gbg subunits thatreduce the lifetime of the fusion pore. Furthermore, simultaneousactivation of PKC removes the ATP-mediated inhibition of secretion inrat adrenal chromaffin cells (RACCs).

RESULTS

Inhibition of secretion in RACCs by ATP

We examined the hypothesis that activation of a G protein pathway inRACCs inhibits secretion at steps subsequent to the regulation of Ca2+

influx. Direct application of 100 mM ATP to RACCs did not induce anycurrent or increase in [Ca2+]i (data not shown)18.

We monitored depolarization-induced secretion by a micro–carbonfiber electrode (CFE) (Fig. 1a). In the presence of 100 mM ATP, thedepolarization-induced secretion was inhibited by 80 ± 12% (10.8 ±2.2 pC versus 2.3 ± 0.4 pC; n ¼ 21, mean ± s.e.m.). In contrast, ATPinhibited the depolarization-induced Ca2+ current (ICa) by only25 ± 3% (pulse, n ¼ 9) and 21 ± 6% (action potential waveform(APW), n ¼ 11) (Fig. 1a, right; the inhibition of ICa reversed within10 s after washout of ATP). Because the relation between secretion and[Ca2+]i is not linear19,20, it is not clear whether the roughly threefoldlarger ATP inhibition of secretion compared with current is exclusivelydue to the reduced ICa, or whether other mechanisms downstream ofCa2+ contribute as well.

In RACCs, caffeine and muscarine induce Ca2+ release from caffeine-sensitive (ryanodine) and IP3-sensitive Ca2+ stores, respectively. If ATP


1Institute of Neuroscience, Shanghai Institutes for Biological Sciences and Graduate School, Chinese Academy of Sciences, Shanghai 200031, China. 2Institute ofMolecular Medicine, Peking University, Beijing 100871, China. 3Synaptic Function Unit, National Institute of Neurological Disorders and Stroke, Bethesda, Maryland20892, USA. 4Department of Anesthesiology, Washington University, St. Louis, Missouri 63110, USA. 5State Key Laboratory of Biomembrane Engineering, College of LifeSciences, Peking University, Beijing 100871, China. 6These authors contributed equally to this work. Correspondence should be addressed to Z.Z. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

directly inhibits some step in the secretory process subsequent to Ca2+

influx, it would be expected to reduce secretion evoked by Ca2+ storemobilization. To test this possibility, caffeine (20 mM) was applieddirectly to RACCs for 20 s, resulting in a burst of amperometric spikesafter a delay of 1–2 s (Fig. 1b, left panels). ATP (100 mM, co-puffed withcaffeine) inhibited the caffeine-induced amperometric spikes, resultingin a decrease in the integrated amperometric charge or secretion by52 ± 9% (mean ± s.e.m.) in 2 mM extracellular Ca2+ (12.8 ± 2.1 pC incontrol versus 6.6 ± 0.7 pC in ATP; n4100) and by 50 ± 10% in 0 mMextracellular Ca2+ (10.5 ± 2.6 pC in control versus 5.3 ± 1.3 pC in ATP;n¼ 7) (Fig. 1b, right). The inhibitory effect of ATP occurred less than5 s after onset of the application (data not shown). In contrast, ATP hadlittle effect on caffeine-induced intracellular Ca2+ transients (103 ± 4%of control, n ¼ 5, Fig. 1c). This result shows that in addition toinhibiting voltage-dependent Ca2+ influx (Fig. 1a), ATP inhibitssecretion downstream from the elevation of cytosolic Ca2+. Confirmingthis interpretation, 100 mM ATP also produced a 54 ± 11% reduction(16.5 ± 3.8 pC versus 7.6 ± 2.4 pC; n ¼ 8) of the secretion evoked by10 mM ionomycin (Fig. 1b, right), another method of elevatingcytosolic Ca2+ that bypasses Ca2+ channels. ATP had no effect on theionomycin-induced Ca2+ transient (data not shown).

ATP reduces quantal size

The inhibition of caffeine-induced secretion by ATP might reflect eithera decreased vesicle release probability or a reduction in total ampero-metric spike charge (or quantal size). To address this issue, wecompared isolated amperometric spikes elicited either by caffeinealone or co-puffed with ATP. Statistically, ATP reduced quantal size,as demonstrated by averaged amperometric spikes and quantal sizedistribution in the presence or absence of ATP (Fig. 2a). On average,ATP had little effect on the number of amperometric spikes induced bycaffeine (20 mM for 20 s; 19 ± 3 versus 17 ± 3 amperometric spikes/cell). This suggests that neither the size of the readily releasable pool northe release probabilities of individual vesicles were changed by ATP.

We also undertook a quantitative analysis of amperometric spikeproperties. Several features of amperometric spikes reflect importantsteps in the exocytotic process. Large, rapid amperometric spikes areoften preceded by a ‘foot’, thought to correspond to the initial openingof the fusion pore2,3,12. Compared with the control, ATP significantlyreduced quantal size by 44 ± 12% (0.54 ± 0.04 pC versus 0.31 ±0.03 pC) and foot charge by 73 ± 6% (44 ± 4 fC versus 12 ± 2 fC) (Po0.01, Fig. 2b). In addition, the half-height duration (HHD) and thefoot duration were reduced by 23 ± 7% (6.9 ± 0.6 ms versus 5.2 ±0.4 ms, Po 0.05) and 52 ± 4% (5.4 ± 0.2 ms versus 2.6 ± 0.2 ms, Po0.01), respectively, but ATP had no significant effect on the foot

150 pA30 ms

ControlATP

ATP

20 pA10 s

30 pA15 s

100

80

60

40

20

0

Inhi

bitio

n (%

)60

40

20

0

100

50

0

Inhi

bitio

n (%

)C

hang

e in

F34

0 /F

340

(%)

F34

0 /F

380

Secretion

Caffeine + ATP (2Ca)Caffeine + ATP (0Ca)Ionomycin + ATP

ICa (pulse)

ICa (APW)

****

Ctrl ATPCaffeineCaffeine Caffeine + ATP

CaffeineCaffeine Caffeine + ATP

20 s

900

800

3 min 3 min

Iampdt

Iampdt∫

∫

Iamp

Iamp

3-minbreak–70 mV

0 mV

a

b

c

Figure 1 ATP inhibits Ca2+-dependent secretion through multiple signaling

pathways. (a) ATP inhibits secretion more than ICa. Left panel shows typical

amperometric current traces (Iamp, upper trace) and the integrated current

signal (RIampdt) evoked by a 500-ms depolarization in an RACC with or

without 100 mM ATP. The right panel summarizes the ATP inhibition of

secretion elicited by a 500-ms depolarization, peak ICa elicited by a step

depolarization or by voltage-clamp action potential waveforms (APW). In this

and following histograms, the amounts of secretion were evaluated from theintegral of the amperometry. Inset at upper left shows the inhibitory effect of

ATP on ICa induced by a 50-ms depolarizing pulse. (b) ATP inhibits caffeine-

induced amperometric spikes. Left panel shows typical amperometric current

traces evoked by caffeine (20 mM for 10 s) co-puffed with or without ATP

(n 4100). (c) ATP has no effect on caffeine-induced [Ca2+]i elevation.

a

b

40

20

0

6

4

2

0

Num

ber

ofam

pero

met

ric s

pike

s

Num

ber

ofam

pero

met

ric s

pike

s

Foo

t dur

atio

n (m

s)

Foo

t fre

quen

cy (

%)

Qua

ntal

siz

e (p

C)

HH

D (

ms)

Foo

t cha

rge

(fC

)

****

***

Foot duration

Foot Halfheight

HHD

10 ms

25 pA

20 ms

20 pA

0

40

0

5

Control ATP

(pC)1.00.50 1.5 2.0 2.5

Control

Control

ATP

ATP

0

20

40

0

10

20

0.0

0.2

0.4

0.6

Figure 2 ATP modulates quantal size. (a) The distribution of quantal size

of caffeine-induced amperometric spikes before and after ATP treatment.

Histogram shows the number of amperometric spikes of different quantal size

evoked by caffeine (20 mM, 10 s) with or without ATP. Left inset shows the

numbers of amperometric spikes elicited by caffeine with ATP (n ¼ 195, ten

cells) or without ATP (n ¼ 177, ten cells). Right inset shows averaged traces

(each from 20 amperometric spikes, which were among the fastest 10%;

ref. 3) induced by caffeine with or without ATP. (b) Quantitative analysis of

amperometric spikes induced by caffeine with or without ATP. Data from 12

cells and 329 amperometric spikes that met the 5 s.d. threshold criterion.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

frequency (36 ± 6% versus 31 ± 8%, Fig. 2b). These results demonstratethat ATP does not reduce the vesicle release probability but does reducequantal size, presumably by reducing the open time of the fusion pore.The dose-response curve for ATP inhibition of quantal size gave an IC50

of 115 mM, and a Hill coefficient of 1 (Supplementary Fig. 1).The reduction of both foot duration and HHD suggests that ATP

reduces the stability of the fusion pore. If this is the case, manipulationsthat influence fusion pore stability might alter the ability of ATP toinhibit secretion. For example, dynamin is involved in the pinch-off ofendocytosed vesicles4,21,22 and thus may affect the termination of arelease event. Intracellular dialysis of a peptide derived from theproline-rich domain of dynamin (dynPRD, 250 mg/ml), which com-petitively inhibits endogenous dynamin function23, increased quantalsize, HHD and foot duration dramatically (Fig. 3). In the presence ofdynPRD, although quantal size, HHD and rise time were increased by204 ± 18%, 300 ± 36% and 327 ± 24%, the amperometric spikeamplitude was reduced by 36 ± 8% (Fig. 3c,d and data not shown). Inthe presence of intracellular dynPRD, quantal sizes were 0.56 ± 0.06 pC(control), 1.14 ± 0.11 pC (dynPRD) and 1.12 ± 0.09 pC (dynPRD +ATP). HHDs were 7.1 ± 0.7 ms (control), 21.3 ± 2.6 ms (dynPRD) and20.6 ± 1.9 ms (dynPRD + ATP). Foot durations were 5.4 ± 0.2 ms(control), 8.2 ± 0.3 ms (dynPRD) and 8.0 ± 0.3 ms (dynPRD + ATP).Foot charges were 44 ± 4 fC (control), 261 ± 20 fC (dynPRD) and 256 ±22 fC (dynPRD + ATP). This is consistent with a previous report usinganti-dynamin-IgG4. Importantly, dynPRD eliminated the ATP inhibi-tion of quantal size, HHD and foot duration (Fig. 3c,d). In contrast,intracellular dialysis of a scrambled control peptide (dynControl) waswithout effect (Fig. 3a,b). In the presence of intracellular dynControl,quantal sizes were 0.56 ± 0.06 pC (control), 0.54 ± 0.05 pC (dynCon-

trol, 8 min) and 0.27 ± 0.02 pC (dynControl + 100 mM ATP, 10 min).HHDs were 7.1 ± 0.7 ms (control), 6.9 ± 0.5 ms (dynControl) and 4.8 ±0.3 ms (dynControl + ATP). Foot durations were 5.4 ± 0.2 ms(control), 5.2 ± 0.2 ms (dynControl) and 2.6 ± 0.2 ms (dynControl+ ATP). Foot charges were 44 ± 4 fC (control), 42 ± 3 fC (dynControl)and 12 ± 2 fC (dynControl + ATP). Finally, neither dynPRD nordynControl affected the ATP inhibition of ICa (data not shown). Theseresults strongly suggest that ATP reduces quantal size by reducing theopen time of the fusion pore.

PKC reverses ATP-induced inhibition of secretion

Acetylcholine (ACh) is the endogenous transmitter for chromaffincells14. Activating the muscarinic ACh receptor (mAChR) elevatescytosolic Ca2+ (ref. 15) and evokes secretion in RACCs13. A 10-sapplication of 100 mM MCh, a selective mAChR agonist, resulted ina burst of amperometric spikes (Fig. 4a, left). Notably, in contrast to theeffect of ATP on caffeine- and ionomycin-induced secretion, ATP hadlittle effect on MCh-induced amperometric spikes (Fig. 4a, right).

Unlike the steps involved in caffeine- and ionomycin-evoked Ca2+

elevations, the mAChR activates a Gq protein that is coupled to aphospholipase C-inositol 1,4,5-trisphosphate (PLC-IP3) pathway8.This results in the parallel elevation of cytosolic IP3, leading to releaseof Ca2+ from cytosolic stores and an increase in diacylglycerol, whichactivates Ca2+-dependent PKC8. It therefore seemed possible that thesimultaneous activation of PKC by MCh could explain the lack ofinhibitory effects of ATP that we observed.

To test this possibility, we examined the ability of bisindolylmalei-mide (BIS, 500 nM), a specific membrane-permeant PKC inhibitor, toinfluence the ATP effects on MCh-induced secretion. In the presence ofBIS (Fig. 4b), MCh still induced bursts of amperometric spikes, but theresponse was less, suggesting that the control response to MCh mayreflect both the elevation of cytosolic Ca2+ and an effect of PKCactivation on secretion. When the effect of ATP on MCh-evokedsecretion was examined in the presence of BIS, the integrated ampero-metric spikes were inhibited by 55 ± 11% (Fig. 4b,d). In contrast, ATPhad no effect on the MCh-induced elevation of cytosolic Ca2+ in thepresence or absence of BIS (Fig. 4a, right, and data not shown). Theseresults suggest that the parallel activation of PKC can reverse the ATPinhibition. Consistent with this idea, in the presence of 5 mM stauro-sporine (a relatively nonspecific PKC inhibitor), ATP also inhibitedMCh-activated secretion (data not shown). Finally, 10 min pretreat-ment with 200 nM phorbol 12-myristate 13-acetate (PMA), a mem-brane-permeable agonist for PKC and presynaptic protein Munc13(refs. 24,25), abolished the ability of ATP to inhibit caffeine-evokedsecretion (Fig. 4c,d). ATP inhibited 52 ± 9% (12.8 ± 2.1 pC versus 6.6 ±0.7 pC) of caffeine-induced secretion (n4100), 9 ± 4% (22.1 ± 5.1 pCversus 20.4 ± 5.5 pC) of MCh-induced secretion (n¼ 22) and 55 ± 11%(12.2 ± 2.0 pC versus 5.4 ± 1.3 pC) of MCh-induced secretion when

dynControl

+ ATP

+ ATP

30 pA

100 pA

40 pA

5 s

0.4 s

2 s

Iamp

Iamp

Control

Control

Control

dynPRD

dynPRD

dynPRD

dynControl dynControl + ATP

dynPRD + ATP

Qua

ntal

siz

e (p

C)

Qua

ntal

siz

e (p

C)

HH

D (

ms)

HH

D (

ms)

Foo

t dur

atio

n (m

s)F

oot d

urat

ion

(ms)

Foo

t cha

rge

(fc)

Foo

t cha

rge

(fc)

0.0

0.5

0 0 0

00

5

0

10

20

0.0

0.5

1.0200

20

40

2

4

6

5

a

b

c

d

Figure 3 ATP inhibition was eliminated by blocking dynamin function.

(a) Intracellular dialysis of a scrambled control peptide (dynControl,

0.5 mg/ml, 8–10 min) had no effect on the reduction of quantal size by ATP.

(b) Statistically, intracellular dialysis of dynControl did not change the ATP

effects on quantal size, HHD, foot duration and foot charge. Data from

seven cells. (c) The mutant dynamin peptide dynPRD eliminated the

reduction of quantal size by ATP. A typical Iamp trace showed that intracellular

dialysis of dynPRD increased quantal size and HHD, and eliminated thereduction of quantal size by ATP. Inset shows the effect of dynPRD on

amperometric spikes at expanded time scale. (d) Statistically, dynPRD

eliminated the ATP effects on quantal size, HHD, foot duration and foot

charge. Data from 9 cells.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

cells were pretreated with 500 nM BIS (n ¼ 8). ATP inhibited 5 ± 9%(17.1 ± 3.2 pC versus 15.6 ± 2.8 pC) of caffeine-induced secretion whencells were pretreated with PMA (n ¼ 8). Taken together, these resultssuggest that activation of PKC removes the inhibitory effects of ATP onCa2+-dependent secretion.

Activation of PKC in chromaffin cells directly increases the size of thereadily releasable pool9, essentially enhancing the rate of recruitment ofvesicles for secretion26. In our experiments, the possibility that PKCeliminated the ATP inhibition by this mechanism could be ruled out, asPKC reversed the ATP effect on both quantal size and amperometricspike kinetics. In bovine chromaffin cells, PMA treatment also reducesthe size of digitonin (a non-physiological secretagogou)-inducedamperometric spikes27. In our experiments, however, PMA treatmenthad no obvious effect on quantal size of caffeine-induced secretion(Supplementary Fig. 2). We found that, like in bovine cells, PMA itselfreduced quantal size of digitonin-induced amperometric spikes in ratcells (Supplementary Fig. 3). Thus, PMA has no effect on quantal sizeif the release is induced by physiological stimuli. Analysis of the quantalsize of amperometric spikes evoked by MCh showed no difference inthe presence or absence of ATP. However, pretreatment with BIS toinhibit PKC restored the ability of ATP to reduce quantal size duringMCh-evoked secretion. Consistent with this, pretreatment with PMAeliminated the ATP-induced reduction of quantal size during caffeine-evoked secretion (Fig. 4e). ATP inhibited 44 ± 12% (0.54 ± 0.04 pCversus 0.31 ± 0.02 pC) of caffeine-induced quantal size (n ¼ 12), 2 ±2% (0.53 ± 0.03 pC versus 0.53 ± 0.05 pC) of MCh-induced quantalsize (n ¼ 22), and 45 ± 8% (0.53 ± 0.03 pC versus 0.29 ± 0.03 pC) ofMCh-induced quantal size when cells were pretreated with 500 nM BIS(n ¼ 8). PMA treatment eliminated the ATP inhibition in quantal size(0.53 ± 0.04 pC versus 0.54 ± 0.04 pC, n¼ 8). PMA also eliminated theATP-induced changes in amperometric spike kinetics (HHD and footduration; Fig. 4f). Comparing amperometric spikes evoked by caffeine+ ATP and caffeine + ATP + PMA, the HHDs were 4.8 ± 0.3 ms versus6.2 ± 0.4 ms, and the foot durations were 2.6 ± 0.2 ms versus 5.2 ±0.3 ms (n ¼ 8). In addition, like PMA, MCh eliminated the ATP-induced changes in amperometric spike kinetics (data not shown).Although PMA activates both PKC and Munc13 (refs. 24,25), the datafrom BIS and staurosporine are consistent with the hypothesis thatPKC is responsible for the MCh effect on removing the ATP inhibition.We conclude that PKC activation removes the inhibitory effect of ATPon the fusion pore.

ATP mediated effects by P2Y receptor coupled to Gi/o

Caffeine-evoked secretion was inhibited by only�5 ± 5% by ATP in thepresence of reactive blue-2 (RB-2, 30 mM), an antagonist of P2Ypurinoceptors (Fig. 5a, left, and Fig. 5b). The inhibition of caffeine-induced secretion by ATP was sensitive to pertussis toxin (PTX). In cellsfrom the same culture recorded on the same day, the extent of ATP-mediated inhibition was 46 ± 9% (10.4 ± 1.6 pC versus 5.0 ± 2.1 pC,

n¼ 8) in controls, but 5 ± 5% (12.4 ± 2.8 pC versus 13.1 ± 1.9 pC, n¼8) in RB-2, and �6 ± 4% (11.6 ± 3.3 pC versus 12.1 ± 3.5 pC, n¼ 14) inPTX-treated cells (Fig. 5a, right, and Fig. 5b). PTX also reversed theATP-induced inhibition of ICa in the same batch of cells (data notshown). Thus, the ATP-induced inhibition of secretion is mediated by aPTX-sensitive Gi/o signaling pathway.

Although the ATP concentration (100 mM) used in our study is closeto the physiological level28, the relatively high concentration might havehad some nonspecific effect on quantal size. To exclude this possibility,we used 2-methylsulfate ATP (2-MeS-ATP; 200 nM), a specific andhigh-affinity P2Y agonist18, to confirm that P2Y is responsible for theATP inhibition of quantal size in RACCs. Indeed, 2-MeS-ATP(200 nM) inhibited quantal size and amperometric spike kinetics tothe same extent as did ATP (100 mM) and had no effect on the caffeine-induced [Ca2+]i transient (Supplementary Fig. 4).

We next evaluated whether the pathway underlying the inhibitoryeffects of ATP on secretion is mediated by a or bg subunits of the Gi/o

protein. A primary effector pathway of the activated Gi/o a subunit issuppression of the activation of adenylate cyclase, thereby reducingcAMP and minimizing PKA activity8. To test the possible involvementof adenylate cyclase in the ATP-mediated effects, cells were preincu-bated with the cell-permeable cAMP/PKA reagents 8-bromine cAMP,forskolin or H89 for 10–15 min. None of these compounds altered theATP inhibition of total secretion and quantal size, excluding theinvolvement of a Gi/o a pathway (Fig. 5c). After 30 min preincubation

a

b

c

20 pA

10 s

10 pA

15 s

10 pA

15 s

3 min

3 min3 min

3 min 3 min

50 s

1,200

600MCh

MCh MCh

MCh MCh + ATP

MCh + ATP

Caffeine + ATPCaffeine Caffeine

MCh + ATP

F34

0 /F

380

Iamp

Iamp∫

∫

∫

Iamp

Iamp

Iamp

dt

Iampdt

MCh

MCh

MCh

Caf + ATP

Caf Caf

Caf + ATP + PMA

Tim

e (m

s)

Qua

ntal

siz

e (p

C)

d e f

Inhi

bitio

n (%

)

100

0

MCh

+ BIS

MCh

+ BIS

Caf +

PM

A

Caf +

PM

A HHD FD0

+ ATP + ATP

0.6

0

6 ***

Figure 4 Activation of PKC reverses the ATP-induced inhibition of secretion.

(a) 100 mM ATP had little inhibitory effect on secretion (left) and [Ca2+]ielevation (right) elicited by 0.1 mM MCh (n ¼ 6). (b) In presence of 500 mM

BIS, ATP inhibited MCh-induced secretion in a representative cell. (c) After

pretreatment with 200 nM PMA for 10 min, the inhibitory effects of ATP

on caffeine-evoked secretion were eliminated in a representative cell.

(d) Summary of effects of ATP on secretion evoked by MCh in the absence or

presence of BIS. (e) Summary of effects of ATP on quantal size evoked byMCh in the absence or presence of BIS and the effects of PMA on reduction

of quantal size by ATP. (f) PMA eliminated the effects of ATP on the kinetics

of amperometric spikes.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

with forskolin (50 mM, n ¼ 11), 8-bromine cAMP (8-Br cAMP,100 mM, n ¼ 9) or H89 (10 mM, n ¼ 8), ATP inhibited 45 ± 10%(17.9 ± 4.5 pC versus 9.9 ± 1.8 pC), 48 ± 12% (16.4 ± 3.6 pC versus 8.5± 2.2 pC), and 51 ± 10% (11.5 ± 3.6 pC versus 5.4 ± 2.1 pC) of caffeine-induced total secretion, respectively. The corresponding values forquantal size were 0.33 ± 0.03 pC (forskolin), 0.30 ± 0.02 pC (8-BrcAMP) and 0.32 ± 0.02 pC (H89). In Fig. 5c,d, ATP inhibited 52 ± 6%of total secretion. Quantal size was 0.31 ± 0.02 pC in the presence ofATP (control, n ¼ 100).bg subunits of the trimeric Gi/o protein are known to activate second

messenger signaling pathways of adenylate cyclase II, phosphatidylino-sitol 3-kinase (PI3K), phospholipase C-b2 (PLC-b2), serine/threoninekinases and tyrosine kinases8,29,30. We therefore tested the effects ofpretreatment with LY294002 for PI3K, U73122 for PLC-b2, stauro-sporine for serine/threonine kinase and K252A for tyrosine kinase.None of these compounds influenced the ATP inhibition of total sec-retion and quantal size, excluding the effector pathways listed abovefrom the ATP inhibition (Fig. 5d). After preincubation with U73122 (2mM, puff for 3 min, n¼ 8), LY294002 (40 mM, 30 min, n¼ 11), K252A(20 mM, 30 min, n¼ 9) or staurosporine (5 mM, 15 min, n¼ 13), ATPinhibited 48 ± 12% (9.6 ± 3.5 pC versus 5.0 ± 1.8 pC), 50 ± 13% (10.3 ±2.4 pC versus 5.1 ± 1.6 pC), 53 ± 14% (10.4 ± 2.8 pC versus 4.7 ± 1.4pC) and 52 ± 8% (10.2 ± 2.5 pC versus 4.8 ± 0.8 pC) of caffeine-induced secretion, respectively. The corresponding quantal size valueswere 0.32 ± 0.03 pC (U73122), 0.29 ± 0.02 pC (LY294002), 0.27 ± 0.02pC (K252A), and 0.31 ± 0.02 pC (staurosporine).

Direct role of Gbc subunits in inhibition of secretion

We next examined whether Gbg subunits directly participate in theATP-mediated inhibition of secretion. We used mSRIK, a membrane-permeable peptide that binds to the Gbg subunit, to shield theendogenous and activated Gbg subunits in chromaffin cells31,32. Afterpretreatment with mSRIK (30 mM) for 30 min, ATP was unable toreduce caffeine-induced secretion (Fig. 6a,b). It was possible thatmSRIK occluded the inhibitory effect of ATP by altering the activationof PKC in some way, thereby mimicking the effect of MCh. To excludethis possibility, we tested the effect of 500 nM BIS on the blockadeproduced by mSRIK, but it was without effect (Fig. 6b). Afterpretreatment with mSRIK, ATP inhibited only 9 ± 3% ofthe caffeine-induced secretion (12.2 ± 2.6 pC versus 11.2 ± 1.8 pC,n ¼ 13), but 500 nM BIS had no effect on the action of mSRIK on theATP inhibition (11 ± 4%; that is, 12.2 ± 2.6 pC versus 10.4 ± 2.3 pC) ofthe caffeine-induced secretion (n ¼ 8).

To obtain additional evidence that Gbg subunits directly inhibitsecretion in RACCs, we dialyzed purified Gb1g2 subunits33 through thewhole-cell patch pipette. In control cells, there was no significantreduction in secretion evoked by 1 s depolarization at 4-min intervals(Fig. 6c, left traces). Whole-cell dialysis of Gb1g2 subunits (0.5 ng/ml) for4 min reduced depolarization-induced secretion (Fig. 6c, right) by

reducing quantal size (Fig. 6d) by 69 ± 2% (from 0.61 ± 0.04 pC to 0.19± 0.02 pC ) of control and by reducing ICa by 22 ± 4% of control (n¼ 5,data not shown). Furthermore, Gb1g2 markedly reduced foot durationby 44 ± 9% (from 5.7 ± 0.7 ms to 3.2 ± 0.2 ms), foot charge by 82%(from 51 ± 6 fC to 9 ± 2 fC) and HHD by 26 ± 8% (from 6.3 ± 0.5 ms to4.7 ± 0.6 ms) compared with the control, indicating that the fusionpore is directly regulated by Gbg (Fig. 6d, n ¼ 5). The effects of Gbg onquantal size, foot duration and HHD are not due to the reduction ofCa2+ currents by Gb1g2, because these effects were also obtained whencaffeine or ionomycin, which increase [Ca2+]i without voltage-gatedCa2+ channels, were used to trigger secretion (Figs. 6e,f). Gb1g2

subunits significantly reduced the evoked secretion, with quantal sizedropping from 0.61 ± 0.08 pC to 0.29 ± 0.04 pC, HHD from 6.3 ± 0.8ms to 4.6 ± 0.6 ms, foot duration from 6.1 ± 1.3 ms to 1.5 ± 0.2 ms andfoot charge from 51 ± 9 fC to 7 ± 1 fC (n ¼ 3, P o 0.01 for all exceptHHD, for which P o 0.05).

ATP and Gb1g2 subunits inhibited both depolarization-activated ICa

and quantal size (Figs. 1a and Fig. 6c). On average, ATP reduced thenumber of depolarization-induced amperometric spikes and quantalsize by 53 ± 8% and 45 ± 3%, respectively (Supplementary Fig. 5). Toexamine whether the small (22–25%) inhibition of ICa by ATP or Gbgwas responsible for the reduction of quantal size, cells were depolarizedfrom �70 mV to 0 mV or 30 mV for 0.5 s (Supplementary Fig. 5). Inthese experiments, ICa and the number of amperometric spikes evokedat 30 mV were 83 ± 5% and 55 ± 12%, respectively, of that at 0 mV,while quantal size was similar at 0 mV and 30 mV. Thus, the reducedquantal size was not due to the inhibition (by 22–25%) of thedepolarization-induced Ca2+ current by ATP or Gbg (Supplementary

40 pA15 s

10 s

10 pA

3 min 3 min

Caffeine + ATP Caffeine Caffeine + ATP

Caf + ATP Caf + ATP + RB-2 Caf + ATP + PTX

Caffeine + ATP + RB-2

50

80Gi/oα signal pathway Gβγ signal pathway

60

40

20

0

0.2

0.0

CONCON

U7312

2 LYK25

2ASSP

FOS8-

BrH 8

9

0

Inhi

bitio

n (%

)

Inhi

bitio

n (%

)Q

uant

al s

ize

(pC

)

0.2

0.0

Qua

ntal

siz

e (p

C)

80

60

40

20

0

Inhi

bitio

n (%

)

Iamp Iamp

∫Iampdt ∫Iampdt

a

b

c d

Figure 5 ATP inhibits secretion through activation of a Gi/o protein. (a) ATP

inhibited caffeine-induced secretion through the P2Y receptor by means of

a PTX-sensitive Gi/o protein. Traces show representative amperometric spike

traces induced by caffeine with or without ATP following pretreatment of the

cell with 30 mM RB-2 (left) or 250 ng/ml PTX (right) for 24 h. (b) Summary

of statistical data in a. (c) The effects of reagents that interfere with cAMP

metabolism on the ATP inhibition of caffeine-induced secretion and quantal

size. FOS, forskolin. CON, control (applying 0.1 mM ATP only). 8-Br,8-bromine cAMP. (d) The effects of reagents that interfere with Gbg

metabolism on the ATP inhibition of caffeine-induced secretion and quantal

size. LY, LY294002. SSP, staurosporine. CON, control (as in c).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Fig. 5 and Fig. 6). Taken together, these results support the idea thatATP inhibits secretion via Gbg subunits that directly interact with thefusion machinery in RACCs.

Opioids and somatostatin share the ATP inhibitory effect

In addition to ATP and its purinoceptors, chromaffin cells containm-opioid and somatostatin and their receptors, which inhibit ICa orcatecholamine secretion14,16. Similar to ATP activation of P2Y recep-tors, both m-opioid and somatostatin receptors are known to selectivelyactivate Gi/o signaling pathways16,28,29,34. We therefore asked whetheractivation of these receptors shares features of action with ATP. Ourresults indicated that activation of both receptors inhibited secretion bya mechanism that is shared with ATP receptor activation and is likely toinvolve similar G protein signaling systems (Fig. 7). Somatostatinreduced quantal size from 0.51 ± 0.04 pC to 0.25 ± 0.04 pC (n ¼ 14)of that induced by caffeine but had no effect on quantal size of MCh-induced amperometric spikes (0.52 ± 0.05 pC to 0.50 ± 0.05 pC, n¼ 8,Fig. 7a). In addition, somatostatin inhibited 23 ± 2% (n¼ 7) of the ICa

and 74 ± 9% (12.1 ± 2.2 pC versus 3.1 ± 1.1 pC, n ¼ 7) of thedepolarization-induced secretion. Somatostatin inhibited 48 ± 5%(11.4 ± 1.7 pC versus 5.9 ± 0.6 pC, n ¼ 12) of the caffeine -inducedsecretion, 6 ± 4% (19.1± 4.2 pC versus 18.0 ± 0.7 pC, n ¼ 14)of the MCh-induced secretion and 46 ± 7% (12.6 ± 2.2 pC versus6.8 ± 0.8 pC, n ¼ 8) of the MCh-induced secretion after pretreatmentwith 500 nM BIS (Fig. 7c).

DAMGO reduced quantal size from 0.52 ± 0.05 pC to 0.28 ± 0.04 pC(n ¼ 13) of that induced by caffeine, but had no effect on quantalsize of MCh-induced amperometric spikes (0.52 ± 0.05 pC versus0.53 ± 0.04 pC, n ¼ 9, Fig. 7b). In addition, DAMGO inhibited 25 ±2% (n¼ 8) of ICa and 74 ± 8% (10.6 ± 2.1 pC versus 2.8 ± 0.8 pC, n¼8) of the depolarization-induced secretion. DAMGO inhibited 54 ± 6%(10.8 ± 2.0 pC versus 5.0 ± 0.6 pC, n ¼ 11) of the caffeine-inducedsecretion, 3 ± 2% (21.2 ± 4.6 pC versus 20.6 ± 0.4 pC, n ¼ 13) of theMCh-induced secretion and 44 ± 9% (11.4 ± 1.6 pC versus 6.4 ± 1.0

pC, n ¼ 9) of the MCh-induced secretion after pretreatment with500 nM BIS (Fig. 7d).

Endogenous PTX-sensitive Gi/o inhibits quantal size

Having established that exogenous application of ATP, opioids orsomatostatin reduces quantal size by means of activation of a PTX-sensitive Gi/o pathway, we next tested whether endogenous transmittersreleased from a RACC would have a similar effect on quantal size. Inorder to avoid the effect of gap junctions between adjacent cells, wepatch clamped an isolated cell (cell #1), and then lifted and placed it incontact with cell #2. A CFE was then placed on cell #1 for combinedpatch-clamp and amperometric recordings (Fig. 8a, right). In controlrecordings, cell #1 was depolarized for 1 s, triggering a burst ofamperometric spikes (Fig. 8a, left). To test the possible effects of releaseof endogenous transmitters on cell #1, immediately before applying asecond depolarizing pulse to cell #1, cell #2 was stimulated by localapplication of 70 mM KCl. Subsequent to stimulation of cell #2, theevoked transmitter release from cell #1 was smaller than the control,and, in particular, quantal size was reduced by 48 ± 1% (from 0.54 ±0.04 pC to 0.26 ± 0.03 pC, n¼ 12). Finally, after recovery, the secretionevoked by the depolarizing step recovered to the control level.

It is likely that ATP, opioids or other endogenous ligands releasedfrom cell #2 activated Gi/o pathways to cause inhibition in cell #1. Thishypothesis was confirmed by the observation that incubation ofchromaffin cell cultures with 250 ng/ml PTX for 24 h virtuallyabolished the reduction of quantal size (Fig. 8b). When pretreatedwith PTX, activating cell #2 resulted in a reduction of 5 ± 2% (from0.54 ± 0.04 pC to 0.57 ± 0.04 pC, n¼ 8) of quantal size in cell #1. Thesedata indicate that endogenous transmitters activate Gi/o and reducequantal size in RACCs (Fig. 8c).

DISCUSSION

ATP, opioids and somatostatin can each selectively activate a Gi/o

pathway in RACCs and inhibit secretion by two separate mechanisms:

5 pA 60

40

20

0

Inhi

bitio

n (%

)

0Ctrl mSRIK mSRIK

+ BIS

5 s

Caffeine Caffeine + ATP

Iamp

Iamp

∫Iampdt

–70 mV

0.5

Qua

ntal

siz

e (p

C)

0.0

5

HH

D (

ms)

0

5

Foo

t dur

atio

n (m

s)

0

20

40

60

Foo

t cha

rge

(fC

)

Qua

ntal

siz

e (p

C)

Foo

t dur

atio

n (m

s)

Foo

t cha

rge

(fC

)

0

0 mV

4 min

50 pA

5 s

Iamp

–70 mV0 mV

4 min

Gβ1γ2 (0.5 ng/µl)

Gβ1γ2 (0.5 ng/µl)

Gβ1γ2 dialysisControl

20 pA

5 s

40 pA

3 min 3 min

10 s50 pA

10 s

Caffeine

0.5

HH

D (

ms)

0.0

5

0

5

0 0

20

40

60

Caffeine Caffeine Caffeine

Gβ1γ2 dialysisControl

** **

**

***

** **

*

**

**

a

c

d

e

f

b

Figure 6 Gbg subunits mediate the ATP-induced inhibition of secretion. (a) In the presence of mSRIK (30 mM, 30 min, 37 1C), the inhibitory effect of ATP on

caffeine-induced secretion was reversed in a representative cell. (b) Statistics of panel a, and the effect of 500 nM BIS on the blockade produced by mSRIK.

(c) Secretion evoked by 1-s depolarizing pulses to 0 mV was compared for cells stimulated 4 min after initiation of whole-cell recording either without (left

panels) or with (right panels) 0.5 ng/ml Gb1g2 subunits in the recording pipette. (d) Statistical analysis of c. (e) Secretion evoked by caffeine (20 mM, 10 s)

was compared for cells stimulated 3 min after initiation of whole-cell recording either without (left panels) or with (right panels) 0.5 ng/ml Gb1g2 subunits in

the recording pipette. The same result was observed for ionomycin-induced secretion. (f) Statistical analysis of e.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

inhibition of ICa and inhibition of the exocytotic machinery. Whereasstimulation of the receptor(s) inhibits about only 25% of the total ICa,depolarization-elicited secretion is inhibited by 80% (Fig. 1a; see alsoref. 35). In bovine chromaffin cells, inhibition of ICa seems to accountalmost completely for inhibition of secretion by ATP18,28,34. Inhibitionof ICa is thought to be the major mechanism for presynaptic inhibi-tion7. However, the present results show that direct inhibition of thesecretory machinery by a physiological Gi/o pathway (B45%) may beof comparable importance to the inhibition of voltage-dependent Ca2+

channels (B55%) in RACCs (Supplementary Fig. 5).

Gbc subunits directly inhibit the secretory machinery

Scavenging of endogenous Gbg subunits by mSRIK, a membrane-permeable Gbg-binding peptide31,32, eliminated the ability of ATP toinhibit caffeine-evoked secretion. Although scavenging Gbg subunitsmay have other effects on cell function, the result implicates Gbgsubunits in the ATP-induced effects. Furthermore, direct dialysis of

Gbg subunits into RACCs mimicked the effect of ATP on secretion andamperometric spike kinetics.

Gbg subunits can modulate many cellular functions by a number ofdistinct effector pathways, including AC II, PI 3-kinase, PLC-b2 andseveral serine/threonine kinases and tyrosine kinases8. The failure of allof the tested compounds to block the ATP-mediated inhibition furthersupports the contention that a large number of signaling pathways arenot involved in this effect.

In lamprey spinal cord synapses, injection of Gbg into presynapticneurons inhibits neurotransmission downstream from presynapticvoltage-gated Ca2+ channels independently of Gbg -regulated cytoplas-mic messengers30. Although this is consistent with a direct effect of Gbgon presynaptic vesicle-release proteins, it remains to be establishedwhether this is also a general feature of secretory processes in mam-mals36. Furthermore, the mechanisms of the inhibitory effect (releaseprobability or fusion pore kinetics) on the lamprey synapses by Gbgwere unknown30. In the present work, we have demonstrated themechanism that Gbg inhibits secretion by reducing the open time ofthe fusion pore in RACCs. Although we favor the idea that free Gbgdirectly regulates fusion pore open time, we cannot exclude thepossibility that the apparent effect of Gbg could be caused by thewhole G protein complex (Ga + Gbg).

Gbc reduces the open time of the fusion pore

The reduction of quantal size resulting from the ATP inhibition andGbg action might arise from either inhibiting the refilling of vesicles; orshortening the lifetime of the fusion pore. It seems unlikely that theacute application of ATP (co-puffed with caffeine) should lead to arapid decrement in the average vesicle content. CFE measurements

10 pA

5 s

10 pA

10 s

30 pA

5 s

50 pA

10 s

Iamp

Iamp

Iamp

Iamp

3 min

0.6

0.4

0.2

0.0

0.6

0.4

0.2

0.0

0.6

0.4

0.2

0.0

0.6

100

80

60

Inhi

bitio

n (%

)

40

20

0

0.4

0.2

0.0

3 min

3 min

3 min

Caffeine

MCh MCh + SOM

MCh

MCh MCh + Bis

Somatostatin DAMGO

DAMGOCtrl

DAMGOCtrl

SOMCtrl

SOMCtrl

CafDepolICa

100

80

60

Inhi

bitio

n (%

)

40

20

0MCh MCh + BisCafDepolICa

MCh + DAMGO

Caffeine + SOM

Qua

ntal

siz

e (p

C)

Qua

ntal

siz

e (p

C)

Caffeine Caffeine + DAMGO

a

b

c d

**

**

Figure 7 Somatostatin and the m-receptor opioid, DAMGO, influence

secretion in a manner similar to ATP. (a) Representative amperometric spikes

induced by 20 mM caffeine (upper panels) or 100 mM MCh (lower panels)

with or without 500 nM somatostatin (left panels). Histograms show the

effect of somatostatin on quantal size evoked by caffeine (upper right) or

MCh (lower right). (b) DAMGO (3 mM) inhibited secretion evoked by 20 mM

caffeine (upper panels) but had no effect on secretion evoked by 100 mM

MCh (lower panels). (c) Additional effects of somatostatin. Depol,depolarization. Caf, caffeine. (d) Additional effects of DAMGO.

Iamp

∫Iampdt

Iamp

∫Iampdt

–70 mV 2 min

0 mV

Puff KCI

Puff KCI

Patch

Patch CFEKCI

Cell #2

Cell #2

Cell #1

Cell #1

20 pA

5 s

30 pA

5 s

2 min

2 min 2 min

0.6

0.4

0.2

0.0Ctrl

Qua

ntal

siz

e (p

C)

PTXEndoinhib.

a

b cFigure 8 Endogenous reduction of quantal size by Gi/o activation.

(a) Activation of cell #2 by local application of 70 mM KCl reduced

depolarization-induced secretion in cell #1 (left). Cell #1 was first patch

clamped and then placed in contact with cell #2 (right). The depolarizationpulse was applied to cell #1 during application of 70 mM KCl to cell #2.

(b) After pretreatment of cells with 250 ng/ml PTX (24 h), depolarization-

induced secretion in cell #1 was not inhibited by endogenous transmitter

release from cell #2. (c) Effect of cell #2 activation on quantal size without

and with PTX pretreatment.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

showed that ATP reduces foot duration and foot charge (Fig. 2).Intracellular dialysis of Gbg subunits directly inhibited both quantal sizeand foot duration (Fig. 6). The foot in an amperometric spike isthought to reflect slow efflux of catecholamine through a fusionpore3,12. Therefore, the reduction in foot duration argues that fusionpore open time is reduced.

To further confirm that the ATP inhibition is by means of the fusionpore, we tested the role of dynamin in the ATP inhibition. Dynamin is aprotein that is responsible for fission of endocytotic vesicles4,21,22.Intracellular dialysis of mutant dynamin (dynPRD), which had lostthe GTPase activity necessary to close the fusion pore21,23, increasedfusion pore duration and eliminated the ATP effect on quantal size,HHD and foot duration (Fig. 3). Note, although dynamin affects theregeneration of endocytosed vesicles, it does not influence the para-meters of amperometric spikes shown in Figure 3, because vesicleregeneration should affect only the total number of evoked ampero-metric spikes, but not the kinetics of single amperometric spikes(quantal size, HHD, foot duration, foot charge). Thus, these dynaminexperiments provide additional evidence for the ATP effect on fusionpore opening.

Reversal of the Gi/o-mediated inhibition by PKC

We found that despite the robust inhibitory effects of ATP on bothdepolarization- and caffeine-evoked secretion, it had no effect on thatinduced by MCh. However, both bisindolylmaleimide and staurospor-ine treatment enabled ATP to inhibit the secretion evoked by MCh.Furthermore, treatment with PMA reversed the ATP-mediated inhibi-tion of caffeine-induced secretion. These results argue that activation ofPKC removes the ATP-induced inhibition of secretion in RACCs,although the mechanisms by which it does this are unknown.

In chromaffin cells, intensive stimulation facilitates secretion bymeans of activation of Ca2+-dependent PKC37. PKC activationincreases secretion strength by increasing the readily releasable pooland accumulating vesicle recruitment9,26. In the present work, wedemonstrate another novel function of PKC activation; that PKC canremove the Gi-mediated inhibition of fusion pore opening. Futurework should address the mechanisms by which PKC removes theATP inhibition.

METHODSCulture of chromaffin cells. Rat adrenal medulla chromaffin cells (RACCs)

were prepared as described previously15,38. The use and care of animals in this

study complied with the guidelines of the Animal Advisory Committee at the

Shanghai Institutes for Biological Sciences.

Gb1g2 peptide was a gift from C. He (Second Military Medical University,

Shanghai)33. All other chemicals were from Sigma, except mSRIK, which was

from Calbiochem.

Electrophysiological methods. Voltage-gated membrane currents were

recorded using the nystatin perforated patch-clamp technique38. The holding

potential was �70 mV and the cell was depolarized to 0 mV for 0.5–2 s to evoke

secretion and current. For ICa recording, tetrodotoxin (100 nM) and

TEA (20 mM) were used to inhibit the Na+ and K+ currents, respectively. In

Figure 8a,b, as cell #1 was voltage-clamped at �70 mV, high KCl could not

depolarize it, but it induced an inward (KCl) current of about �20 pA. Thus,

there was no direct effect on secretion in cell #1 (data not shown).

A perfusion system (RCP-2B, INBIO) with a fast exchange time (o100 ms)

for electronic switching between seven channels was used to change the external

medium39. All experiments were carried out at room temperature (22–25 1C).

Data are given as mean ± s.e.m. The significance of differences was determined

using Student’s t-test (*P o 0.05, **P o 0.01).

Electrochemical amperometry. Highly sensitive, low-noise, 5-mm carbon fiber

electrodes (ProCFE, Dagan) were used for electrochemical monitoring of

quantal release of catecholamines from single RACCs as described previously3.

‘Foot’ analysis was as described previously3. The onset of the foot was

determined by a threshold of 5 s.d. above baseline, and the end of the foot

was determined by the onset of the major spike (Fig. 2b). In all ATP application

experiments, without preincubation, ATP was co-puffed with caffeine. For

analysis of the kinetic properties of amperometric spikes, only events 45 s.d.

were included. All the data were analyzed with Igor software (WaveMetrix) with

a custom-made macro program3.

[Ca2+]i measurements. To estimate changes in intracellular Ca2+, isolated

RACCs were incubated for 15 min in a bath solution containing 2 mM Fura-2/

AM (Molecular Probes) at 37 1C. Intracellular Ca2+ concentration [Ca2+]i was

measured by dual-wavelength ratiometric fluorometry. The Fura-2 was excited

with light alternating between 340 and 380 nm using a monochromator-based

system (TILL Photonics), and the resulting fluorescence signals were measured

using a cooled CCD. Relative changes in [Ca2+]i were calculated from the ratio

of F340 to F380, which were sampled at 1 Hz by fluorescence CCD imaging of a

single cell39. The image data were transferred and analyzed by Igor software

(WaveMetrix).


ACKNOWLEDGMENTSWe thank C. He for the Gb1g2 peptide, Y.T. Wang for the dynamin peptides andI. Bruce for reading the manuscript. This work was supported by grants from theNational Basic Research Program of China (G2000077800 and 2006CB500800),the National Natural Science Foundation of China (30330210, 303328013 andC010505 to Z.Z.) and the US National Institutes of Health (DK46564 to C.L.).




1. Ales, E. et al. High calcium concentrations shift the mode of exocytosis to the kiss-and-run mechanism. Nat. Cell Biol. 1, 40–44 (1999).

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

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

4. Elhamdani, A., Palfrey, H.C. & Artalejo, C.R. Quantal size is dependent on stimulationfrequency and calcium entry in calf chromaffin cells. Neuron 31, 819–830 (2001).

5. Wang, C.T. et al. Synaptotagmin modulation of fusion pore kinetics in regulatedexocytosis of dense-core vesicles. Science 294, 1111–1115 (2001).

6. Miller, R.J. Presynaptic receptors.Annu. Rev. Pharmacol. Toxicol.38, 201–227 (1998).7. Wu, L.G. & Saggau, P. Presynaptic inhibition of elicited neurotransmitter release. Trends

Neurosci. 20, 204–212 (1997).8. Hamm, H.E. The many faces of G protein signaling. J. Biol. Chem. 273, 669–672

(1998).9. Gillis, K.D., Mossner, R. & Neher, E. Protein kinase C enhances exocytosis from

chromaffin cells by increasing the size of the readily releasable pool of secretorygranules. Neuron 16, 1209–1220 (1996).

10. Stevens, C.F. & Sullivan, J.M. Regulation of the readily releasable vesicle pool by proteinkinase C. Neuron 21, 885–893 (1998).

11. Sakaba, T. & Neher, E. Direct modulation of synaptic vesicle priming by GABA(B)receptor activation at a glutamatergic synapse. Nature 424, 775–778 (2003).

12. Chow, R.H., von Ruden, L. & Neher, E. Delay in vesicle fusion revealed by electro-chemical monitoring of single secretory events in adrenal chromaffin cells. Nature 356,60–63 (1992).

13. Zhou, Z. & Misler, S. Action potential-induced quantal secretion of catecholamines fromrat adrenal chromaffin cells. J. Biol. Chem. 270, 3498–3505 (1995).

14. Artalejo, A.R. Electrical properties of adrenal chromaffin cells. in Electrophysiology ofNeuroendocrine Cells (eds. Scherubl, H. & Hescheler, J.) 259–294 (CRC, Boca Raton,Florida, 1995).

15. Neely, A. & Lingle, C.J. Effects of muscarine on single rat adrenal chromaffin cells.J. Physiol. (Lond.) 453, 133–166 (1992).

16. Livett, B.G. & Boksa, P. Receptors and receptor modulation in cultured chromaffin cells.Can. J. Physiol. Pharmacol. 62, 467–476 (1984).

17. Ennion, S.J., Powell, A.D. & Seward, E.P. Identification of the P2Y(12) receptor innucleotide inhibition of exocytosis from bovine chromaffin cells. Mol. Pharmacol. 66,601–611 (2004).

18. Powell, A.D., Teschemacher, A.G. & Seward, E.P. P2Y purinoceptors inhibit exocytosis inadrenal chromaffin cells via modulation of voltage-operated calcium channels.J. Neurosci. 20, 606–616 (2000).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

19. Augustine, G.J. & Neher, E. Calcium requirements for secretion in bovine chromaffincells. J. Physiol. (Lond.) 450, 247–271 (1992).

20. Dodge, F.A., Jr. & Rahamimoff, R. Co-operative action a calcium ions in trans-mitter release at the neuromuscular junction. J. Physiol. (Lond.) 193, 419–432(1967).

21. Holroyd, P., Lang, T., Wenzel, D., De Camilli, P. & Jahn, R. Imaging direct, dynamin-dependent recapture of fusing secretory granules on plasma membrane lawns fromPC12 cells. Proc. Natl. Acad. Sci. USA 99, 16806–16811 (2002).

22. Graham, M.E., O’Callaghan, D.W., McMahon, H.T. & Burgoyne, R.D. Dynamin-dependent and dynamin-independent processes contribute to the regulation of singlevesicle release kinetics and quantal size. Proc. Natl. Acad. Sci. USA 99, 7124–7129(2002).

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

24. Betz, A. et al. Munc13–1 is a presynaptic phorbol ester receptor that enhancesneurotransmitter release. Neuron 21, 123–136 (1998).

25. 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).

26. Nagy, G. et al. Protein kinase C-dependent phosphorylation of synaptosome-associatedprotein of 25 kDa at Ser187 potentiates vesicle recruitment. J. Neurosci. 22, 9278–9286 (2002).

27. Graham, M.E., Fisher, R.J. & Burgoyne, R.D. Measurement of exocytosis byamperometry in adrenal chromaffin cells: effects of clostridial neurotoxins andactivation of protein kinase C on fusion pore kinetics. Biochimie 82, 469–479(2000).

28. Currie, K.P. & Fox, A.P. ATP serves as a negative feedback inhibitor of voltage-gatedCa2+ channel currents in cultured bovine adrenal chromaffin cells. Neuron 16, 1027–1036 (1996).

29. Neves, S.R., Ram, P.T. & Iyengar, R. G protein pathways. Science 296, 1636–1639(2002).

30. Blackmer, T. et al. G protein betagamma subunit-mediated presynaptic inhibition:regulation of exocytotic fusion downstream of Ca2+ entry. Science 292, 293–297(2001).

31. Goubaeva, F. et al.Stimulation of cellular signaling and G protein subunit dissociation byG protein betagamma subunit-binding peptides. J. Biol. Chem. 278, 19634–19641(2003).

32. Scott, J.K. et al. Evidence that a protein-protein interaction ‘hot spot’ on heterotrimericG protein betagamma subunits is used for recognition of a subclass of effectors.EMBO J.20, 767–776 (2001).

33. He, C. et al. Identification of critical residues controlling G protein-gated inwardlyrectifying K(+) channel activity through interactions with the beta gamma subunits of Gproteins. J. Biol. Chem. 277, 6088–6096 (2002).

34. Carabelli, V., Carra, I. & Carbone, E. Localized secretion of ATP and opioids revealedthrough single Ca2+ channel modulation in bovine chromaffin cells. Neuron 20, 1255–1268 (1998).

35. Lim, W., Kim, S.J., Yan, H.D. & Kim, J. Ca2+-channel-dependent and -independentinhibition of exocytosis by extracellular ATP in voltage-clamped rat adrenal chromaffincells. Pflugers Arch. 435, 34–42 (1997).

36. Jarvis, S.E. & Zamponi, G.W. Interactions between presynaptic Ca2+ channels, cyto-plasmic messengers and proteins of the synaptic vesicle release complex. TrendsPharmacol. Sci. 22, 519–525 (2001).

37. Smith, C. A persistent activity-dependent facilitation in chromaffin cells is caused byCa2+ activation of protein kinase C. J. Neurosci. 19, 589–598 (1999).

38. Duan, K., Yu, X., Zhang, C. & Zhou, Z. Control of secretion by temporal patterns ofaction potentials in adrenal chromaffin cells. J. Neurosci. 23, 11235–11243 (2003).

39. Zhang, C. & Zhou, Z. Ca(2+)-independent but voltage-dependent secretion in mamma-lian dorsal root ganglion neurons. Nat. Neurosci. 5, 425–430 (2002).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Bag1 is essential for differentiation and survivalof hematopoietic and neuronal cells

Rudolf Gotz1,5, Stefan Wiese1,5, Shinichi Takayama3, Guadalupe C Camarero2, Wilfried Rossoll1,Ulrich Schweizer1, Jakob Troppmair2,4, Sibylle Jablonka1, Bettina Holtmann1, John C Reed3, Ulf R Rapp2 &Michael Sendtner1

Bag1 is a cochaperone for the heat-shock protein Hsp70 that interacts with C-Raf, B-Raf, Akt, Bcl-2, steroid hormone receptors

and other proteins. Here we use targeted gene disruption in mice to show that Bag1 has an essential role in the survival of

differentiating neurons and hematopoietic cells. Cells of the fetal liver and developing nervous system in Bag12/2 mice underwent

massive apoptosis. Lack of Bag1 did not disturb the primary function of Akt or Raf, as phosphorylation of the forkhead

transcription factor FKHR and activation of extracellular signal–regulated kinase (Erk)-1/2 were not affected. However, the defect

was associated with the disturbance of a tripartite complex formed by Akt, B-Raf and Bag1, in addition to the absence of Bad

phosphorylation at Ser136. We also observed reduced expression of members of the inhibitor of apoptosis (IAP) family. Our data

show that Bag1 is a physiological mediator of extracellular survival signals linked to the cellular mechanisms that prevent

apoptosis in hematopoietic and neuronal progenitor cells.

Proteins of the Bag family contain an evolutionarily conserved BAGdomain that allows them to bind and modulate the activity ofHsp70-family molecular chaperones1–4. Humans and mice containsix genes encoding the Bag proteins Bag1 (Rap46), Bag2, Bag3 (Bis),Bag4 (Sodd), Bag5 and Bag6 (Scythe)5.

Several Bag proteins have been implicated in the control of apop-tosis. For example, Bag1 interacts with the antiapoptotic protein Bcl-2and protects a variety of cell types from apoptosis in vitro6,7. Over-expression of Bag1 in neurons reduces stroke injury8 in mice andprovides enhanced protection of mouse retinas from apoptosis in vivo,but only if coexpressed with Bcl-2 (ref. 9). Bag3 (Bis) also interacts withBcl-2 and reduces apoptosis when coexpressed with that protein(ref. 10). Bag4 (Sodd) associates with certain receptors of thetumor necrosis factor family and blocks their signaling11,12. Bag6(Scythe) regulates a nuclear pathway that communicates with mito-chondria and regulates their release of cytochrome c, thereby control-ling apoptosis13,14.

The mechanisms by which Bag1 inhibits apoptosis are poorlyunderstood, and the physiological role of this protein during develop-ment is undefined. In addition to associating with Bcl-2, Bag1 may actas a scaffold protein that binds C-Raf, and possibly B-Raf, at themitochondrial surface15,16. In this context, C-Raf could act as aneffector kinase that phosphorylates Bad and other kinases, includingAkt, that have been shown to act as specific kinases for Bad17,18. Theconsequent dissociation of Bad from Bcl-XL seems to be essential for

the survival of neurons and other types of cells that depend onexogenous survival factors. Mice in which Ser112, Ser136 and Ser155of Bad have been mutated are viable but show enhanced cell death afterexposure to proapoptotic stimuli and have a reduced threshold forcytochrome c release, thus underlining the importance of Bad forcellular survival19. Nevertheless, the extent of developmental cell deathin those mice is much less than in mice in which the B-Raf kinase hasbeen inactivated20.

In addition to their role in apoptosis, Bag proteins are involved inother cellular functions. A longer isoform of Bag1 containing nucleartargeting sequences associates with several steroid hormone receptorsand regulates their transcriptional activity21. Both the shorter cytosolicBag1 and the longer nuclear Bag1L are cochaperones for Hsp70 (refs.1,22,23). In this regard, Hsp70 has a protective role in vivo24, but therelative importance of Bag1 in regulating apoptosis in vivo is unknown.

Here, we examined for the first time the in vivo function of a Bag-family gene using targeted gene ablation in mice. Inactivation of theBag1 gene has notable consequences for the survival of hematopoieticand neural stem cells, as massive cell death occurred in association withloss of Bad phosphorylation at Ser136. Notably, phosphorylation ofBad at Ser112 and Ser155 was not altered. Akt kinase activationremained unchanged, and phosphorylation of the forkhead-relatedtranscription factor FKHR, a specific substrate of Akt kinase, wasnormal, indicating that Bag1 is not necessary for the activation of Akt.Nevertheless, Bag1 was present as a complex with both B-Raf and Akt


1Institut fur Klinische Neurobiologie, University of Wurzburg, Josef Schneider Str. 11, D-97080 Wurzburg, Germany. 2Institut fur Medizinische Strahlenkunde undZellforschung (MSZ), University of Wurzburg, Versbacher Str. 5, D-97078 Wurzburg, Germany. 3The Burnham Institute, La Jolla, California 92037, USA. 4Present address:Daniel-Swarowski Research Lab, Department of General and Transplant Surgery, Innsbruck Medical University, Innsbruck, Austria. 5These authors contributed equally to thiswork. Correspondence should be addressed to M.S. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

in vivo, and formation of this complex was impaired in Bag1�/� mice.Our findings show that Bag1 is essential for survival of stem cells in thedeveloping brain and hematopoietic system and is thus a physiologicalmediator of the extracellular survival signals that prevent apoptosis.

RESULTS

Targeted inactivation of the mouse Bag1 locus

A targeting vector was designed and constructed to delete the first twocoding exons of Bag1 (Fig. 1a). This strategy was chosen to disrupt theexpression of all known isoforms of Bag1 (p29, Bag1S and Bag1L)25

that are generated by alternative use of start codons (Fig. 1a). G418-resistant embryonic stem cell clones were isolated and screened by PCRfor homologous recombination. Out of 480 clones, three showed theexpected band (data not shown) and were subjected to diagnosticSouthern blot analysis with a probe corresponding to the 5¢ flankingregion of the recombination construct (Fig. 1b). All three embryonicstem cell clones showed a 2-kb band corresponding to a mutant Bag1allele (Fig. 1b). After injection of these recombinant embryonic stemcell lines into C57Bl/6 blastocysts, germline-transmitting chimaeras werebred with C57Bl/6 females to generate F1 hybrid (129/SJ � C57Bl/6)heterozygotes. Bag1+/� mice were viable and fertile. The Bag1 mutationwas transferred to a C57Bl/6 background by further crossbreeding withC57Bl/6 wild-type mice for more than three generations.

Intercrossing of heterozygous animals did not yield live offspringwith a Bag1�/� genotype (n ¼ 28 matings), and Bag1+/� animals wereindistinguishable from wild-type littermates. Thus, the mutant allelebehaved as a recessive trait that is essential for mouse development.Genotyping of mice from heterozygous intercrosses at embryonic day12.5 (E12.5) demonstrated that Bag1�/� embryos were still present andviable (Fig. 1c). In contrast, all Bag1�/� embryos were highly growth-retarded at E13.5, and death of these animals was observed betweenE12.5 and E13.5 (Fig. 1d). Whereas the percentages of Bag1�/� mice atE10.5 and E11.5 corresponded approximately to the expected 25%, thepercentage was reduced at E12.5, and only 1% of Bag1�/� mice werealive at E13.5. Western blot analysis of E12.5 embryos showed thatdisruption of the Bag1 gene led to a complete loss of all known Bag1isoforms (p29, p32 and p50; Fig. 1e).

Bag1�/� embryos have liver and nervous system defects

At E12.5, Bag1�/� mice had normal body size compared with theirheterozygous and wild-type littermates (Fig. 2). Histological analysis

showed abnormalities in the fetal liver and forebrain corresponding tothe onset of endogenous expression, which was observed at E11.5 withthe highest levels in the nervous system and liver26. The fetal livers ofBag1�/� embryos were smaller than those of their Bag1+/� and Bag1+/+

littermates, suggesting a defect in hematopoiesis. Bag1�/� embryoswere also anemic (Fig. 2a) as a result of macroscopically detectabledefects in erythropoiesis in the fetal livers. In the nervous system,formation of the telencephalic vesicles, which form a major part of theforebrain, was severely disturbed (Fig. 2a), and the neuroepitheliumforming the telencephalic vesicles was substantially reduced in size(Fig. 2b). The spinal cord and dorsal root ganglia were present, butthe mantle zone of the spinal cord was much thinner than theinner epithelial zone consisting of the ependyma and periependymalcells (Fig. 3a).In situ TUNEL and propidium iodide staining showed massive

apoptosis in the fetal livers of Bag1�/� mice (Fig. 2d) and signi-ficantly (P o 0.001) greater apoptosis in the fetal brains compared toBag1+/+ controls (Fig. 2e,f). Quantification of embryonic liver cellswith apoptotic nuclei after propidium iodide staining showed that94.2 ± 6.1% of nuclei were apoptotic in Bag1�/� embryos at E12.5,compared with 0.2 ± 0.2% in wild-type embryos. Similar results wereobtained by TUNEL staining.

We also quantified cell death in the forebrains of Bag1+/+, Bag1�/�,Braf+/+ and Braf�/� embryos. In control Bag1+/+ and Braf+/+ fore-brains, 1.8 ± 0.5% and 1.7 ± 0.5% of nuclei, respectively, werepropidium iodide–positive and were shown by DAPI staining to becondensed. In contrast, 15.1 ± 1.3% and 6.5 ± 0.8% of cells in Bag1�/�

and Braf�/� forebrains, respectively, had propidium iodide–positive,condensed nuclei (Fig. 2e,f). The first indications of cell death inthe neuroepithelium were detected at E10.5 by caspase-3 staining(Supplementary Fig. 1 online).

Altered survival response of cultured Bag1�/� motoneurons

In mice, motoneurons become postmitotic between E10 and E13,and during the subsequent 5–10 d, about half of the newly generatedneurons undergo apoptosis27. This pattern of physiological celldeath allowed us to investigate the role of Bag1 in endogenousregulatory pathways for cellular survival. We isolated motoneuronsfrom Bag1�/� mice and tested their response to neurotrophicfactors. The number of motoneurons that could be isolated fromthe lumbar spinal cords of E12.5 Bag1�/� embryos was significantly

Figure 1 Targeted disruption of Bag1 by

homologous recombination. (a) Schematic

representation of the Bag1 genomic locus (WT),

the targeting vector and the mutant allele. The

neomycin-selectable marker replaced exons 1

and 2 of Bag1 to generate a null allele. The

region recognized by the hybridization probe

for Southern analysis and the primer pairs todiagnose the targeted allele are shown. X,

XbaI; EV, EcoRV. (b) Southern blot analysis of

embryonic stem cell clones. Genomic DNA was

digested with EcoRV and subjected to Southern

blot analysis with the probe shown in a. The wild-

type (WT) and targeted (KO) alleles produced a

3.5-kb and a 2-kb fragment, respectively. (c) PCR

analysis of genomic DNA from one litter of a

Bag1+/� intercross. DNA from E12.5 embryos

was tested with primer pairs P1/P2 to detect the

mutant allele and P1/P3 (P3 was deleted in the mutated allele) for the WT allele. (d) Percentage of viable Bag1�/� mice at various developmental stages.

***, significant at P o 0.001. (e) Western blot analysis (20 mg protein/lane) showed that all Bag1 protein isoforms (Bag1L, Bag1 and p29) were absent from

brains and livers of WT and Bag1-mutant E12.5 mouse embryos. The antibody used recognized an additional unspecific band at 35 kDa (*).

3.5 kb

+/–

+/–

+/–

+/–

+/+

+/+

+/+

+/+

+/–

–/–

+/–

+/–

+/–

+/–

–/–

–/–

+/+

+/+

+/+

Con

trol

Mar

ker –/–

–/–

–/–

Brain Liver

p50

p32p29

*

+/–

+/+

+/+

2 kb

1.5 kb Mutant

E10.5 E11.5

******

E12.5 E13.5

25.5+/–3.0

23.8+/–2.8

19.4+/–3.4

1.0+/–0.8

20

15

10

5

0

Per

cent

age

viab

leBag

1-m

utan

t em

bryo

s

WT1.5 kb

25

a b

c ed

WT

X EV

TK

Mutant

Neo

Neo

ProbeP3

3.5 kb

2 kb

1 2 3

3

4

4

3 4 5 6 7

Targeting vector

5 6 7

X

X

X X

EV

EV

EV EV

P1 P2PCR


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

(P o 0.05) lower compared with lumbar spinal cords of wild-typeand Bag1+/� littermates (Fig. 3b,c). The isolated neurons were platedonto laminin-coated culture dishes and incubated with the neuro-trophic factors glial-derived neurotrophic factor (GDNF), brain-derived neurotrophic factor (BDNF) or ciliary neurotrophic factor(CNTF) (1 ng/ml each). More than 45% of the originally platedwild-type motoneurons could be supported by these neurotrophicfactors over a period of at least 7 d in culture, whereas survivalwas severely reduced in Bag1�/� motoneurons (Fig. 3d,e). Motoneur-ons from Bag1+/� mice showed similar survival responses as didwild-type neurons.

Enhanced death of neuronal precursor cells in Bag1�/� mice

Bag1 promotes neuronal differentiation in the immortalized neuronalCSM14.1 cell line28. We therefore investigated whether Bag1 influencesdifferentiation of motoneurons in developing embryos by analyzingspinal cord sections with neuronal markers, including nestin, islet-1/2and p75NTR. The number of spinal motoneurons with islet-1/2-stainednuclei was significantly (P o 0.05) lower in Bag1�/� embryos than inwild-type embryos (Fig. 3a). Motoneurons that expressed p75NTR werestill present in Bag1�/� mice, although the number of these cells wasalso reduced in spinal cord sections of the Bag1-deficient mice(Fig. 3a). These p75NTR-positive neurons also stained positively forneurofilament-M (data not shown). Staining of E11.5 and E12.5 spinalcord sections showed a reduction of nestin-positive neural precursorcells in Bag1�/� mice, particularly in the population of motoneuron-like cells that had already migrated from the ventrolateral part of thespinal cord (Fig. 3a). This phenotype was discrete at E11.5 but becameprominent at E12.5.

To investigate the role of Bag1 in neuronal differentiation inmore detail, we isolated neural stem cells from the forebrains ofE11.5 Bag1�/� embryos. Initial growth and survival of neurospheres

from E11.5 Bag1�/� and Bag1+/+ littermates were not significantlydifferent when the cells were cultured with basic fibroblast growthfactor and epidermal growth factor on uncoated cell culture dishesto prevent differentiation. After six passages, the cells were trans-ferred to laminin-coated culture dishes to allow differentiation.The attached cells were fixed 24 h after plating and analyzed withnestin-specific antibodies (Fig. 4a) for the presence of neural precursorcells and neurofilament-M-specific antibodies for neuronal differentia-tion (Fig. 4b). The cells were also labeled with propidium iodide todetermine the number of apoptotic cells (Fig. 4c,d. Neurospheresfrom Bag1+/� embryos contained 10.4 ± 1.0% neurofilament-M–positive cells, whereas only 2.6 ± 0.9% of Bag1�/� neurosphereswere positive for neurofilament-M (Po 0.001; Fig. 4b,e). In contrast,the numbers of nestin-positive cells were similar in control andBag1�/� neurospheres (Fig. 4a). The proportion of apoptoticnuclei, as determined by propidium iodide staining, was significantlyhigher in Bag1�/� cells than in Bag1+/� cells (4.5 ± 1.3% versus0.5 ± 0.3%, P o 0.001; Fig. 4f).

Pax6 and doublecortin are markers of undifferentiated and earlydifferentiating neurons, respectively. To more clearly define the stage ofdifferentiation at which the enhanced cell death in the Bag1�/� neuronsoccurs, we determined the number of Pax6-positive and doublecortin-positive cells in the forebrain (Supplementary Fig. 2). There was amore than 100-fold reduction in the number of Pax6-positive cells indifferentiating Bag1�/� neural stem cell cultures compared to Bag+/+

cultures after 48 h. The number of early postmitotic doublecortin-positive cells was also reduced from 5,062 ± 570 in control cells to2,058 ± 617 in Bag1�/� cultures. At that time, the total number of cellswas reduced more than threefold in Bag1�/� cultures (SupplementaryFig. 2), whereas the number of cells with condensed nuclei wassignificantly (P o 0.01) enhanced (Supplementary Fig. 2). Thus, thelow percentage of neurofilament-M–positive neurons seems to result

Bag1+/+ Bag1+/+

Bag1+/+

Bag1–/–

Bag1+/+

Bag1–/–

Bag1+/+

Bag1+/+ Bag1–/– Braf –/–Braf +/+

Braf +/+

Bag1–/– Braf –/–

Bag1–/–

Bag1–/–

PI PITUNEL

PI DAPI Overlay PI DAPI Overlay

***

***17.5

15.0

12.5

10.0

7.5

5.0

2.5

0Per

cent

age

Pl-p

ositi

ve c

ells

a db c

e f

Figure 2 Bag1 mutant embryos show severe defects in the forebrain and liver. (a) Embryos from heterozygous Bag1 intercrosses were dissected at E12.5.

The forebrains of Bag1�/� embryos were massively reduced in thickness (open arrows), and the livers did not show any signs of erythropoiesis (closed arrows).Scale bar: 1 mm. (b,c) Paraffin sections of Bag1+/+ and Bag1�/� embryos stained with H&E. b, Open arrows indicate forebrains of Bag1+/+ and Bag1�/�

embryos; filled arrows indicate livers. c, Transverse section through neural tubes of wild-type and Bag1�/� embryos at E12.5. Scale bars: 1 mm in b and

100 mm in c. (d) Staining for apoptotic cells in the liver with Tunel and propidium iodide (PI) revealed massive cell death in E12.5 Bag1�/� embryos

compared to Bag1+/+ littermates. Higher magnification revealed fragmenting nuclei (arrows), which are typical of apoptosis. Scale bars in d: 100 mm in left

panels, 50 mm in middle panels and 10 mm in right panels. (e) Quantification of PI-positive cells revealed a significant increase in the number of pyknotic and

condensed nuclei in forebrains of Bag1�/� and Braf�/� compared with wild-type embryos.***, significant at P o 0.001. (f) Representative high-magnification

areas of forebrains from embryos in e. Red, PI; blue, DAPI. Scale bar: 20 mm.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

from the apoptosis of neurons as soon as they differentiate from Pax6-positive neural stem cells (Fig. 4e,f and Supplementary Fig. 2), ratherthan by a block to differentiation.

Similar results were obtained using an RNA interference (RNAi)approach to reduce the level of Bag1 in neural stem cells derived fromBag1+/+ embryos (Fig. 4g–j). This experiment was done to exclude the

possibility that Bag1 deficiency leads to defects at earlier stages ofdevelopment before neural stem cells differentiate, so that the observedphenotype would be caused by indirect effects or selection of differentcell populations. The transfection efficiency for the neural stem cellswith an eGFP-C1 reporter plasmid was 54.6 ± 4.6%. Quantification ofthe western blots from neural stem cells transfected with the Bag1 RNAi

Figure 3 Differentiation and survival of spinal

motoneurons is impaired in Bag1�/� embryos.

(a) Vibratome sections through lumbar spinal

cords of E11.5 and E12.5 embryos from Bag1+/�

intercrosses were stained with antibodies to the

neuronal markers islet-1/2 (E11.5), p75NTR

(E12.5) and nestin (E12.5). Lines mark borders

of the neural tubes. Staining of motoneuronnuclei for all markers was lower in Bag1�/� than

in wild-type embryos. All scale bars: 100 mm.

(b) The number of p75-positive motoneurons was

significantly lower in serial vibratome sections

from Bag1�/� mice than in sections from wild-

type mice. ***, significant at P o 0.001.

(c) After a 30-min panning period with p75NTR-

specific antibodies, the number of motoneurons

isolated from lumbar spinal cords of E12.5

embryos was determined. Spinal cords of Bag1�/�

embryos had significantly fewer motoneurons

than did wild-type spinal cords (P o 0.05).

(d) Motoneurons were plated at a density of

2,000 cells/well, and survival was determined

after 7 d in culture. Survival rates of Bag1�/�

motoneurons were significantly lower than those

of wild-type motoneurons, irrespective of the neu-

rotrophic factor that was added (BDNF, GDNF or CNTF). **, P o 0.01; ***, significant at P o 0.001. (e) Morphology of motoneurons isolated from Bag1+/+

and Bag1�/� embryos after 7 d in the presence of BDNF in vitro. The surviving Bag1�/� neurons displayed motoneuron-specific characteristics, such as a longaxon-like process and shorter dendritic processes. Scale bar: 100 mm.

Islet-1/2 p75NTR Nestin

***

50

40

30

20

10

0

Num

ber

of p

75+

cells

/sec

tion

Bag1

–/–

Bag1–/–

Bag1–/–

Bag1+/+

Bag1–/–Bag1+/–Bag1+/+

Bag1

+/+

Bag1

–/–

Bag1

+/–

Bag1

+/+

Bag1+/+

GDNF

GDNF

GDNF

BDNF

BDNF

BDNF

CNTF

No fac

tor

No fac

tor

No fac

tor

CNTF

CNTF

**

*********

0

10

20

30

40

50

Per

cent

age

surv

ival

of

orig

inal

ly p

late

d ce

lls

02468

101214

Num

ber

of is

olat

edce

lls ×

1,0

00

*

a

c d

e

b

Bag1 RNAiRNAi

Scrambled

100

80

60

40

20

0

Ban

d in

tens

ity (

%)

Actin

Bag1S p32

Bag1L p50

RNAi

Scram

bled

eGFP-C

1

RNAiBag1

Bag1–/–Bag1+/–Bag1–/–Bag1+/–

ScrambledRNAi

eGFP-C1RNAiBag1Scrambled

RNAieGFP-C1

0123456789

Per

cent

age

PI+

cel

ls

Per

cent

age

NF

-M+ c

ells

Per

cent

age

NF

-M+ c

ells

Per

cent

age

PI+

cel

ls

**

***

***

***

**

6

5

4

3

2

1

0

0

2

6

8

10

12

14

4

12

10

8

6

4

2

0

Pl

Nestin

Bag1+/–

Bag1–/–

Bag1+/–

Bag1–/–

PC

NF-M

45.3%± 6.8

a b e f

g h

i j

c d

Figure 4 Enhanced apoptosis of differentiating neurons from Bag1-mutant

neural stem cells. (a–d) Neural stem cells were isolated from forebrainsof E11.5 embryos from Bag1+/� intercrosses. Cells from passage 6

were transferred for 24 h to laminin-coated dishes, fixed and stained

for nestin (a) or neurofilament-M (NF-M; b) or with propidium iodide

(PI; c). Phase-contrast (PC) pictures for c are shown in d. Scale bar: 50 mm.

(e,f) Quantitative analysis of NF-M-positive (e) and PI-positive (f) cells in

cultures of differentiating Bag1+/+ and Bag1�/� neural stem cells revealed a

significant loss of neurons by apoptosis in Bag1�/� cultures. ***, P o 0.001 by

Student t-test. (g,h) Quantitative analysis of NF-M-positive (g) and PI-positive

(h) cells in differentiating Bag1+/+ neural stem cell cultures transfected with

Bag1 RNAi, scrambled RNAi or eGFP-C1 vector. **, significant at P o 0.01.

(i) Western blot analysis of Bag1 (10 mg protein/lane) from neural stem cells

transfected with Bag1 RNAi (Bag1), scrambled RNAi or eGFP-C1 vector.

Equal loading was controlled by probing the blot with an actin-specific antibody. Shown is a representative blot from four independent experiments.

(j) Quantification of Bag1 band intensity from blots as in i shows reduction in Bag1 expression neural stem cells after transfection of Bag1 RNAi compared to

scrambled RNAi.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

oligonucleotide showed a reduction of Bag1 protein levels to 45.3 ±6.8% of the Bag1 levels in cells transfected with the respective scrambledRNAi. Given the transfection efficiency, this indicates a nearly completeloss of Bag1 within successfully transfected cells (Fig. 4i,k). Immuno-histochemical detection of Bag1 in differentiating neural stem cellstransfected with the scrambled RNAi demonstrated that Bag1 waspresent in all cells. When the Bag1 RNAi was transfected, Bag1 proteinbecame undetectable in a number of cells. These cells had condensednuclei, indicating that cell death occurred when Bag1 was down-regulated (Supplementary Fig. 3). When these cells were allowed todifferentiate for 2 d on laminin in the absence of epidermal growthfactor or basic fibroblast growth factor, the number of propidiumiodide–positive cells was significantly (Po 0.05) higher in neural stemcells transfected with Bag1 RNAi (5.7 ± 2.7%) than in cells transfectedwith scrambled RNAi (1.0 ± 1.1%) or pEGFP-C1 vector alone (0.9 ±0.9%). Although the number of propidium iodide–positive cellsincreased, the number of neurofilament-M-positive cells droppedsignificantly (P o 0.01) in neural stem cells transfected with Bag1RNAi (6.4 ± 2.2%) compared with cells transfected with scrambledRNAi (14.2 ± 0.9%) or eGFP-C1 (13.2 ± 1.0%). We concluded thatBag1 is required for survival of differentiating neural stem cells and thatgene ablation and acute interference with gene expression leadto similar phenotypes in cultured stem cells derived from E11.5mouse forebrain.

Immunohistochemistry of E11.5 spinal cord and brain sections withantibodies for Pax6 and doublecortin showed that the number of Pax6-positive cells with condensed nuclei was significantly (Po 0.01) higherin Bag1�/� embryos compared to Bag1+/+ embryos (Fig. 5,Supplementary Fig. 4 and Supplementary Table 1), whereas thenumber of more differentiated doublecortin-positive cells with con-densed nuclei was not (Fig. 5 and Supplementary Fig. 4).

The Raf-Erk signaling pathway is not

disturbed in Bag1�/� cells

To study the molecular mechanisms ofaltered neural and hematopoietic stem cellsurvival in more detail, we investigatedthe Raf/mitogen-activated protein kinase(MAPK) signaling pathway because Bag1 isknown to interact with C-Raf. Phosphoryla-tion of Erk-1/2 was similar in livers fromBag1+/+ and Bag1�/� mice (Fig. 6a,b) andin primary embryonic fibroblasts derivedfrom Bag1+/+ and Bag1�/� embryos. Wethen tested the activation of the MAPK path-way by stimulating primary fibroblasts fromBag1+/+ and Bag1�/� embryos with insulin-like growth factor-1. Erk-1/2 activation wasnormal under these conditions (Fig. 6b).

Absence of Bag1 disturbs Bad

phosphorylation at Ser136

Phosphatidylinositol 3-kinase (PI3K) isimportant for the upregulation of IAPs29 andthe activation of Akt. We therefore investigatedthe phosphorylation of Bad (Fig. 6), a target ofAkt and a potential substrate of a complexbetween C-Raf and Bag1 (ref. 15). Bad phos-phorylation at Ser136 was not detectable bywestern blot analysis of Bag1�/� liver extractsusing phosphorylation-specific antibodies(Fig. 6c). Similar results were obtained with

E11.5 brain extracts (Fig. 6d). In contrast, phosphorylation at Ser112and Ser155 was not affected by Bag1 ablation (Fig. 6c,d), suggesting thatBag1 is essential for phosphorylation of Bad at Ser136 in vivo.

We have shown that B-Raf is essential for the survival of developingneurons29. To investigate the role of B-Raf in this context, we comparedBad phosphorylation in protein extracts from E11.5 Bag1+/+, Bag1�/�,Braf+/+ and Braf�/� brains. Pull-down assays for Bad demonstrated anabsence of Ser136 phosphorylation in Bag1�/� and Braf�/� brains,whereas the phosphorylation of Ser112 and Ser155 remained unaf-fected (Fig. 6e).

Bag1 deficiency does not disturb Akt activity

Akt and other kinases phosphorylate Bad at Ser112, and Akt specificallyphosphorylates Bad at Ser136 in response to activation of the PI3Kpathway17. We therefore tested whether the absence of Bag1 proteinleads directly or indirectly to abnormalities in Akt kinase activation.Analysis of Bag1+/+ and Bag1�/� fetal liver and brain extracts showedthat Akt phosphorylation at Thr308 and Ser473 was normal (Fig. 6e).These sites are known to activate Akt in response to PI3K, a mechanismthat seems important in coupling tyrosine kinase signaling to Badphosphorylation and thereby preventing cellular apoptosis. Thus, thepathway that leads to activation of Akt kinase seems to be intact inBag1�/� animals.

We then investigated whether Akt kinase activity is impaired bytesting phosphorylation of the forkhead homologous transcriptionfactor FKHR, a specific target of Akt kinase30. FKHR phosphorylationwas indistinguishable in E12.5 Bag1�/� and Bag1+/+ brains and livers(Fig. 6e), indicating that Bag1 deficiency does not reduce Akt kinaseactivity. Thus, the lack of Bad phosphorylation at Ser136 is caused bymechanisms other than a general defect in activation and enzymaticactivity of Akt.

Bag1–/–

Bag1+/+

cc

cc cc

cccc

cc

Pax6 DAPI Dc/Pax6/DAPI Dc/DAPIDc DAPI

Bag1–/–Bag1+/+Bag1–/–Bag1+/+0123456789

10

0123456789

10

Apo

ptot

ic P

ax6+

cells

/sec

tion

Apo

ptot

ic D

c+

cells

/sec

tion

a

c

b

Figure 5 Enhanced apoptosis of differentiating neurons from spinal cords of Bag1 mutants.

(a) Quantitative analysis of apoptotic cells in E11.5 spinal cords revealed increased apoptosis of

Pax6-positive cells in Bag1�/� compared to wild-type mice. (b) Apoptosis was not enhanced in

doublecortin (Dc)-positive cells in the spinal cords of E11.5 Bag1�/� mice. Error bars in a and b

represent s.d. (c) Immunohistochemical detection of Pax6 (red) and Dc (green) in spinal cords of

E11.5 Bag1+/+ and Bag1�/� embryos. DAPI (blue) was used as a nuclear marker to identify

condensed (arrowheads) or fragmented nuclei. Condensed nuclei were detected in Pax6-positive cells,

whereas Dc-positive neural precursor cells only rarely had condensed nuclei. Third panel from left

shows triple staining for Pax6, Dc and DAPI at low magnification to show the distribution of Pax6-

positive cells in neural tubes of Bag1�/� embryos. A reduction of Pax6-positive cells was particularlyevident in more peripheral parts of the neural tube. cc, central canal. Scale bars from left to right:

10 mm, 20 mm and 50 mm.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Altered expression of IAPs in Bag1�/� embryos

Previous studies have shown that neurotrophic factor–mediated survi-val of developing neurons depends on B-Raf and induction of IAP-family gene expression20. Similar observations were made in studiesinvestigating the survival of endothelial cells in response to vascularendothelial growth factor29. We therefore studied the expression ofIAPs in developing brains and fetal liver hematopoietic cells of E12.5Bag1�/� embryos. Expression of IAP-2 (Fig. 6g) and X-IAP (Fig. 6h)was much lower in Bag1�/� embryos than in wild-type embryos. Incontrast, IAP-1 (Fig. 6f) expression was reduced only in the livers, notthe brains, of Bag1�/� embryos.

Disturbed B-Raf, Akt and Bad association in Bag1�/� cells

C-Raf specifically interacts with Bag1, and it has been suggested thatC-Raf phosphorylates Bad at as-yet-unidentified serine residues15,31.Notably, B-Raf seems to be more important than C-Raf for survival ofdeveloping neurons20, and Braf�/� mice resemble Bag1�/� mice in thatthe survival response of Braf�/� mice to neurotrophic factors inembryonic motoneurons is strongly reduced and expression of membersof the IAP gene family is disturbed20. We therefore analyzed whetherB-Raf also interacts with Bag1. In PC12 pheochromocytoma cells,endogenous B-Raf and Bag1 both interacted with Hsp70 (Fig. 7a),and B-Raf and Bag1 mutually coimmunoprecipitated (Fig. 7b,c). Thesame effect was observed for C-Raf but not for A-Raf. We also

investigated the coimmunoprecipitation of B-Raf and Akt, on the basisof a prior report suggesting that these two kinases associate within aprotein complex20,32. When endogenous B-Raf or Bag1 were precipi-tated from PC12 cells, we detected Akt (Fig. 7d), thus confirming dataon the interaction of Akt and B-Raf after overexpression in HEK293cell extracts33.

To further analyze the interactions between Bad, Akt, B-Raf andHsp70 in wild-type and Bag1�/� cells, we prepared protein lysatesfrom E11.5 Bag1 heterozygous intercross embryos. In wild-typebrain extracts, we observed a strong interaction between B-Rafand Akt (Fig. 7e). This interaction was absent in Bag1�/� embryos(Fig. 7e), indicating that either Bag1 acts as a scaffold to bringthese two kinases into one complex, or the cochaperone activityof Bag1 is required for formation of this complex. The inter-action between Akt and Bad was weaker in brain extracts fromBag1�/� mice (Fig. 7), and the interaction between Bad and Hsp70was absent in extracts from Bag1�/� embryos (Fig. 7f), suggestingthat Hsp70 and Bag1 form a complex that brings Akt close to its target,Bad. This was confirmed by our observation that the interactionbetween B-Raf and Hsp70 was disrupted in extracts from Bag1�/�

embryos (Fig. 7f).We did additional immunoprecipitation experiments to determine

whether B-Raf also interacts with Bad. Using antibodies to B-Raf, theBad protein could be coprecipitated in extracts only from wild-type

LiverBrain

LiverBrain

LiverBrain

***

***

***

***

0

40

80

120

160

Arb

itrar

y un

its

0

40

80

120

160

Arb

itrar

y un

itsA

rbitr

ary

units

Bag1–/–Bag1+/+

Bag1–/–Bag1+/+

Bag1–/–Bag1+/+

80

60

40

20

0

Bad

30 kDa30 kDa

20 kDa20 kDa

20 kDa

30 kDa

P-BadP-Bad

P-Bad

20 kDa

30 kDaP-Bad

Ser155

+/+ –/– +/+ –/–

+/+ –/– +/+ –/–

Ser112 Ser136

EF-1

X-IAP

Brain Liver+/+ –/– +/+ –/–

Brain Liver+/+ –/– +/+ –/–

Brain Liver+/+ –/– +/+ –/–

IAP-2

IAP-1

Erk-1/2

P-Erk-1/2

IGF-1

Bag1–/–Bag1+/+

120′

60′

30′

30′′

0 10′

5′2′1′120′

60′

30′

30′′

0 10′

5′2′1′

43 kDa

43 kDa

45 kDa

45 kDa

Bag1 Braf+/+ +/+ –/––/–

P-BadSer112

P-BadSer136

P-Bad

Bad

Ser155

Hsp70

Akt

P-AktThr473

P-Akt

P-FKHR

Ser308

Ser256

66 kDa

97 kDa

45 kDa

66 kDa

45 kDa

66 kDa

45 kDa

66 kDa

66 kDa

97 kDa

Brain Liver+/+ –/– +/+ –/–

ErkP-Erk-1/2

p42p44

p42p44

43 kDa43 kDa

Liver Liver+/+ –/– +/+ –/–

a d

e

f

g

h

b

c

Figure 6 Specific lack of Bad phosphorylation at Ser136 is not caused by inhibition of MAPK and Akt in Bag1�/� embryos. (a) Western blotting showed that

phosphorylation of p44/p42Erk-1/2 was not absent from Bag1-mutant livers. (b) Protein extracts from serum-starved Bag1+/+ or Bag1�/� primary mouse

embryonic fibroblasts were stimulated with insulin-like growth factor (IGF)-1 for 30 s (") to 120 min ('). Western blots were probed with antibodies to P-Erk-1/2

and pan-Erk-1/2. Peak phosphorylation of Erk-1/2 was similar at 5 and 10 min for Bag1�/� and wild-type control cells. (c) Western blots of liver extracts were

probed with antibodies to pan-Bad and Bad phosphorylated (P-Bad) on Ser112, Ser136 or Ser155, revealing a specific absence of phosphorylation at Ser136

in Bag1�/� mice. (d) Pull-down assays of Bad in brain extracts from E11.5 mice were probed with the Bad-specific antibodies in c, revealing the absence of

Bad phosphorylation at Ser136in Bag1�/� and Braf�/� mice. (e) Western blots of single-embryo protein extracts were probed with antibodies to Ser256-phosphorylated FKHR, Ser473-phosphorylated Akt, Thr308-phosphorylated Akt, Akt and Hsp70. Each lane in a–c and e was loaded with 20 mg of protein

extract. Each experiment was done at least three times with protein extracts from different mice. (f–h) IAP expression in E12.5 brains and livers of Bag1+/+

and Bag1�/� embryos. Semiquantitative analysis of the RT-PCR results is shown on the right, and representative RT-PCR data are shown on the left. Each set

of experiments was performed three times with at least three individual genotypes. ***, P o 0.001 by Student t test. PCR for Elongation factor (EF) was used

as control for the RT reaction.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

embryos, not from Bag1�/� embryos (Fig. 7h). The expression of Baddid not differ between controls and Bag1�/� embryos.

Altered distribution of B-Raf and Akt in Bag1�/� neurons

To investigate the expression and subcellular distribution of Akt andB-Raf in Bag1+/+ and Bag1�/� cells, we focused on motoneurons, asthese cells depend on survival signaling at E11.5 and can easily beidentified in the developing spinal cord. In wild-type animals, mostphosphorylated Akt (P-Akt) immunoreactivity colocalized with cyto-chrome c oxidase at mitochondria-like structures in the cytoplasm(Fig. 8). In contrast, in Bag1�/� motoneurons, P-Akt was diffuselydistributed in the cytoplasm and was much less enriched at mitochon-dria (Fig. 8). Similar results were obtained for subcellular distributionof B-Raf (Fig. 8d,f).

We quantified the pixels in the P-Akt and B-Raf immunohistochem-ical stains that overlapped with cytochrome c oxidase immunoreactiv-ity in control and Bag1�/� cells. This analysis showed a reduction in thelocalization of P-Akt or B-Raf at mitochondria in the Bag1�/�

motoneurons (Fig. 8g,h). Thus, Bag1 seems to be important forconcentrating B-Raf and Akt at mitochondria33, where phosphoryla-tion of Bad and probably the activation of pathways leading to IAPupregulation are expected to occur.

DISCUSSION

Here we report that knocking out the Bag1 gene in mice causes massivecell death in the embryonic liver and severe defects in the differentiationand survival of neuronal cells. In the nervous system, enhanced celldeath occurs in Pax6-positive neural stem cells. The abnormal celldeath of neural stem cells and hematopoietic precursor cells correlateswith the absence of phosphorylation of the proapoptotic protein Bad atSer136. This site has been recognized as a specific target for Akt after

activation by neurotrophic and other prosurvival signals involvingtyrosine kinase receptors and PI3K stimulation. In addition, theexpression of X-IAP and IAP-2 is reduced in Bag1-deficient brains, aspreviously observed in Braf�/� mice20. The function of Raf kinases inthe activation of MEK and ERK is not blocked by the absence of Bag1,and phosphorylation of the forkhead transcription factor FKHR, aspecific target of the Akt kinase, is unchanged. However, the absence ofBag1 alters the subcellular distribution of B-Raf and Akt, which mayhave disturbed the function of these kinases in phosphorylating Bad atSer136, a process that takes place at mitochondria19.

Bag1 was originally identified as a Bcl-2–interacting proteinthat potentiates the survival-promoting activity of Bcl-2 (ref. 15).Bag1 also specifically interacts with C-Raf and, as shown here, withB-Raf but not with A-Raf. Binding of C-Raf to Bag1 stimulates itskinase activity and thus potentiates signals for cellular growth anddifferentiation. It has therefore been hypothesized that mitochondrialRaf, activated by a Bag1–dependent mechanism, acts as an effectorkinase for Bcl-2 by phosphorylating targets such as Bad15. However,knocking out the Bcl2 gene does not enhance apoptosis of developingmotoneurons34, which are severely affected in Bag1�/� embryos,suggesting either that other members of the Bcl-2 family also interactwith Bag1 and thus compensate for Bcl-2 deficiency or that Bag1mediates neurotrophic survival effects independently of Bcl-2 indeveloping neurons. Phosphorylation of Bad leads to its dissociationfrom a complex with Bcl-XL at the mitochondrial surface, and Bad isthen sequestered to the cytosol by the phosphoserine-binding protein14-3-3 (ref. 35). In another experimental procedure, deficiency ofmerosin led to decreased levels of Bag1 and reduced levels of Badphosphorylation36. Thus, this process seems to be important forcellular survival, particularly that of neurons and hematopoietic cells.The phenotype of Bag1�/� mice observed in our study is therefore

Figure 7 Bag1 is necessary to form a complex of

Bag1, Hsp70, Raf and Akt. (a) Immuno-

precipitation of endogenous B-Raf and Bag1

(using antibodies listed below blots for detection)

demonstrated their interaction with Hsp70.

Addition of the corresponding immunizing B-Raf

peptide abolished the signal for Hsp70.

Precipitating antibodies shown above blot.(b) C-Raf and B-Raf coprecipitated with Bag1.

A-Raf did not pull down Bag1. Addition of the

corresponding peptide for C-Raf antiserum

abolished the signal for Bag1. Pull-down assays

and antibody detection demonstrated that the

three Raf kinases were present. Precipitating

antibodies shown above blot. (c) Bag1

coprecipitated with B-Raf from native PC12 cells.

B-Raf antiserum cross-reacted with the Ig heavy

chain. Precipitating antibodies shown above blot.

(d) Akt coprecipitated with Bag1 and B-Raf in

native PC12 cells. Precipitating antibodies shown

above blot. Addition of the corresponding peptide

for B-Raf antiserum abolished the signal for Akt.

(e) Immunoprecipitation of Bad and B-Raf

revealed highly reduced interaction between Akt

and Bad and lack of interaction between B-Raf

and Akt in Bag1�/� in mice. (f) Top, interaction of

Hsp70 with Bad and B-Raf was abolished inBag1�/� embryo extracts. Bottom, loading

controls for Bad and B-Raf. (g) Top, immunoprecipitation of Akt revealed that Bag1 is necessary for interaction of Akt with Bad. Bad was also precipitated and

detected by western blotting to control for Bad protein levels. Bottom, loading control for Akt. (h) Top, immunoprecipitation of B-Raf coprecipitated Bad in

Bag1+/+ but not in Bag1�/� embryo extracts (shown are duplicate samples from independent extracts). Bad protein levels were not reduced in extracts from

Bag1�/� mice. Bottom, loading control for B-Raf.

B-Raf

B-Raf Bad

B-Raf

B-Raf

B-Raf

B-Raf

Bag1+/+ Bag1–/–

Detection: α BadIgG

+/+ –/– +/+ –/–

BadBad

BadB-Raf

A-RafC-RafB-Raf

B-Raf

B-Raf

Bad

BadBad AktAkt

Bad

Bad

–/–+/+–/–+/+–/–+/+–/–+/+

–/–+/+–/–+/+–/–+/+–/–+/+

Brain Trunk

Hsp70

Detection: α Akt

Detection: α Bag1Detection: α Bag1Detection: α Hsp70 α B-Raf

B-Raf

45 kDa

60 kDa

60 kDa

66 kDa

97 kDa 70 kDaBag1

Bag1

+ Pep

tide

B-Raf

Bag1+

Peptid

e

+ Pep

tide

B-Raf

B-Raf C-R

af

A-Raf Bag

1Bag

1

Bag1

Detection:

Bag1–/–Bag1+/+

Bag1–/–Bag1+/+

Akt

Akt Akt

IgG

IP:

IP:

Detection:

Detection: Bad

a

e

g

f

h

c db


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

consistent with a model in which Bag1 cooperates with Raf kinases tophosphorylate Bad, thus inhibiting apoptosis.

It has been hypothesized that Raf kinases also phosphorylateBad in parallel or together with Akt. However, attempts to identifyspecific sites in Bad that are phosphorylated by C-Raf in vivo havenot been successful. In vitro phosphorylation of recombinantBad at Ser136 by C-Raf occurs with low efficiency, in contrast to themore robust phosphorylation of Bad by activated Akt31. Nevertheless,B-Raf also seems to be essential for neuronal survival, as isolatedneurons from Braf�/� mice cannot survive in the presence of neuro-trophic factors20. Braf�/� brain extracts also showed a strong reductionin Bad phosphorylation at Ser136, similar to what is observed in theBag1�/� brain and liver extracts. It remains to be determined whetherAkt or B-Raf is responsible for Bad Ser136 phosphorylation.The observation that Ser136 phosphorylation is also reduced inBraf�/� brains suggests that B-Raf in complex with Bag1 is necessary,either directly or indirectly, to phosphorylate Bad at Ser136 at thesurface of mitochondria.

In the absence of Bag1, a complex between Akt, B-Raf and Hsp70 isdisrupted, and B-Raf and Akt kinase lose their contact with mitochon-dria. These findings suggest that Bag1 has a role in assembly of theB-Raf/Akt/Hsp70 complex, which is important for Bad phosphoryla-tion and other mechanisms that regulate induction and executionof cell death. We do not know whether the functions of Hsp70 and achaperone complex involving these molecules are deficient in Bag1�/�

cells. Expression of Hsp70 is not decreased in brains and livers ofBag1�/� mice at E12.5 (Fig. 4a), when cell death is prominent. Also ourfinding that activation of the MAPK pathway and FKHR phosphoryla-tion was normal suggests that Bag1 is not necessary for the activity ofRaf and Akt. On the other hand, we cannot exclude the possibility thatthe presence of Bag1 and Hsp70 is necessary for the activity of B-Rafand Akt, specifically in the context of Bad phosphorylation at Ser136.Alternatively, Bag1 and Hsp70 may be necessary for changing theconformation of Bad so that Ser136 is accessible for phosphorylation byAkt or B-Raf. We therefore propose a model in which Bag1 and Hsp70

must be present in a complex at the surface of mitochondria to allowphosphorylation of Bad at Ser136 by Akt and/or B-Raf (see model inSupplementary Fig. 5). If such a complex exists, it will be interesting tostudy the activity of the kinases in this complex.

Although Bag1 is required for proper targeting of Raf kinases andAkt to the mitochondria and for Bad Ser136 phosphorylation, theabsence of Bag1 probably has other effects on cell survival. In thisregard, mice in which Ser112, Ser136 and Ser155 of Bad are mutated toalanine residues show enhanced cell death under various conditions,although the phenotype of these mice is much less severe than thephenotype of Bag1�/� or Braf�/� mice19. This indicates that Badphosphorylation at Ser136 is not the only event regulating the survivalof neuronal and hematopoietic cells. In contrast to Bad-mutant mice,Braf�/� mice show excessive cell death, particularly in the nervoussystem, and thus resemble Bag1�/� mice.

Neurotrophin-induced expression of IAPs is essential for neuronalsurvival, and this survival pathway is also dependent on B-Raf 37. Weobserved here that IAP gene expression is disturbed in Bag1�/�

embryos in a manner similar to that seen in Braf�/� mice20. IAPgene expression is regulated by NFkB38, and the massive cell death inthe livers of Bag1�/� embryos resembles that of mouse mutants withdisturbed NFkB signaling. RelA-deficient mice die at E15 to E16 bymassive apoptosis in liver cells39. However, in contrast to Bag1�/� mice,no major defects are observed in the developing brains of RelA-mutantmice, and the erythropoietic cells in the livers of RelA-deficient mice areunaffected compared to Bag1�/� mice. It remains to be determinedwhether this finding is related to the observation that additionalsignaling pathways are affected in Bag1�/� mice, such as the pathwayleading to Bad phosphorylation at Ser136.

Notably, the initial development and generation of neuronal pre-cursor cells is much less affected in Bag1�/� mice before neuronal cellsdifferentiate and become postmitotic. This observation suggests thatBag1 becomes important at a crucial point of development at whichneurons and other cells become competent to die40. The observationthat phosphorylation of ERKs is normal in Bag1�/� mice is consistentwith the observation that general growth of the embryos is notsignificantly altered up to E12. We previously observed that most ofthe B-Raf and C-Raf kinase immunoreactivity in postmitotic moto-neurons is localized at mitochondria20. Our finding here that C-Rafand B-Raf coprecipitate with Bag1 suggests that Bag1 becomes highlyimportant for these postmitotic cells after they become competent todie. The observations that the subcellular distribution of Akt kinase alsochanged and that much less Akt kinase is located at mitochondria of

P-Akt

P-Akt

Cyto c oxidase

Cyto c oxidase

Cyto c oxidaseP-Akt

Bag

1+

/+

Bag

1+

/+

Bag

1–/

–

Bag

1–/

–

Bag1

–/–

Bag1

+/+

P-Akt

P-Akt

Bag1

–/–

Bag1

+/+

B-Raf

B-Raf

Bag

1+

/+Bag

1–/

–

Overlay B-RafCyto c oxidase Overlay

100

80

60

40 ***

*

20

Per

cent

age

over

lapp

ing

pixe

ls

Per

cent

age

over

lapp

ing

pixe

ls

Cyto c oxidase/P-Akt Cyto c oxidase/B-Raf

0

100

80

60

40

20

0

a b

dc

f

hg

e

Figure 8 Loss of Bag1 leads to changes in mitochondrial localization of Akt

and Raf in isolated motoneurons. (a–h) Lumbar spinal cord sections and

isolated motoneurons from E12.5 Bag1+/+ (a,c,e) and Bag1�/� mice (b,d,f)

were fixed with 4% pradormaldehyde and stained with Cy3-conjugated

antibodies to phosphorylated Akt or B-Raf and Cy-2 conjugated antibody to

cytochrome c oxidase. Akt immunoreactivity was reduced in the ventrolateral

part of the spinal cord in Bag1�/� embryos (a), thus reflecting the loss of

neurons at E12.5. When motoneurons were isolated and cultured for 2 d withCNTF and BDNF (c–f), the accumulation of P-Akt immunoreactivity at

mitochondria was reduced (d) compared with wild-type controls (c). Similarly,

B-Raf immunoreactivity, which colocalizes with cytochrome c oxidase at

mitochondria in control cells (e), became more dispersed in Bag1�/�

motoneurons (f). (g,h) Overlapping pixels of the signal for cytochrome c

oxidase and P-Akt (g) or B-Raf (h) were quantified using a channel plot

overlap probability function. There was a reduction in overlay of pixels for

P-Akt and cytochrome c oxidase (g) and for B-Raf and cytochrome c oxidase

(h). Scale bars: 100 mm for a,b; 5 mm for c–f. ***, significant at P o 0.001;

*, significant at P o 0.05.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Bag1�/� neurons could provide an explanation for the massiveapoptosis that occurs at this developmental stage in Bag1�/� mice.

In conclusion, the phenotype of Bag1�/� mice suggests that thismolecule has a central role in the survival of neurons and other celltypes once they become competent to die. The interaction of Bag1 withsteroid receptors, Hsp70 and Raf kinases provides a structural counter-part to the observation that cellular signaling pathways involving thesemolecules work together to regulate cellular survival. The absence ofBag1 disrupts such coordinated functions and eventually results inmassive apoptosis in the nervous system and other organs duringdevelopment.

METHODSGeneration of Bag1�/� mice. The procedure for generating Bag1�/� mice is

described in detail in the Supplementary Methods. All animal procedures were

approved by the Bavarian State authorities for animal experimentation.

Cell culture techniques. Cultures of spinal motoneurons from E12.5 mice were

prepared by a panning technique using a monoclonal rat antibody to p75NTR

(Chemicon) according to a previously published method20.

PC12 cells were propagated under standard medium conditions (DMEM

with 10% horse serum and 5% fetal calf serum). For the immunoprecipitation

experiments, the cells were synchronized at low serum and treated for 24 h with

nerve growth factor. The cells were then lysed (5 � 106 cells for each

immunoprecipitation experiment initially), and the extracts were treated with

the indicated antibodies.

Cultures of neurospheres from E12.5 and 11.5 mice were prepared by

dissecting the forebrain as described in detail in the Supplementary Methods.

Primary embryonic fibroblasts were isolated from E12.5 Bag1+/+ and

Bag1�/� single embryos using standard techniques as described in the

Supplementary Methods. The embryos were decapitated and the intestinal

organs removed. The embryo bodies were then washed three times in PBS and

treated with 0.1% trypsin/EDTA (Invitrogen) for 10 min at 37 1C. Each single

embryo body was passed through a 21-gauge needle, and the cells were

transferred to individual culture dishes. The plates were filled with DMEM

containing 10% fetal calf serum and 1% nonessential amino acids (Invitrogen).

Cells were passaged three times and then grown on individual 10-cm cell

culture plates under low-serum conditions (0.5% fetal calf serum) for 24 h

before stimulation with insulin-like growth factor-1 for the indicated times.

Protein isolation, western blotting and immunoprecipitation. Embryonic

tissue was obtained from litters derived from intercrosses of Bag1+/�

mice. After lysis in radioimmunoprecipitation assay buffer, electro-

phoresis and blotting, the membranes were stained with Ponceau S (Sigma)

to mark molecular mass standards. For immunoprecipitation, tissue or

cell extracts were precleared with protein A-Sepharose and then incubated

with 1 mg/ml antibodies to Akt, B-Raf or Bad (New England Biolabs/Cell

Signaling) or antibody to Bag1 (Santa Cruz Biotechnology) overnight. Pre-

cipitates were processed according to standard protocols32 for further western

blot analysis.

Immunofluorescence analysis of cell cultures and tissue sections of embry-

onic mice. Vibratome sections and cell cultures derived from E11.5 and E12.5

embryos from Bag1+/� intercrosses were processed for immunostaining and

confocal microscope analysis (described in detail in Supplementary Methods).

RT-PCR for IAP-1, IAP-2 and X-IAP. RNA from brains and livers of E12.5

mice was isolated by TRIzol reagent (Invitrogen). For each RT-PCR reaction,

10 ng of RNA was used. RT-PCR was done according to the manufacturer’s

instructions (Invitrogen) with random hexamer primers. The primer sequences

used to amplify IAP-1, IAP-2, X-IAP and elongation factor-1 were designed

according to earlier published sequences20.

Statistical analyses. Neuronal counts were expressed as means ± s.e.m. or s.d.,

as indicated. Statistical significance was assessed by the Student t-test or

ANOVA followed by the Bonferroni test, using GraphPad Prism software.

The channel plot overlap probability function from Leica was used to quantify

the overlapping immunohistochemical signals.


ACKNOWLEDGMENTSWe thank M. Pfister, J. Marcano and K. Kalus for excellent technical assistance;R. McKay (US National Institutes of Health) for providing the nestin-specificantibody; T. Jessell for providing the islet-1/2 hybridoma (39.4D5) through theDevelopmental Studies Hybridoma Bank; and R. Timpl for the kind donation oflaminin. This work was supported by the Deutsche Forschungsgemeinschaft (SE697/3-3, SFB487, TP C4 and SFB581 TP B4 and B17), the US National Institutesof Health (CA67385) and the Hermann and Lilly Schilling Foundation.




1. Takayama, S. et al. BAG-1 modulates the chaperone activity of Hsp70/Hsc70. EMBO J.16, 4887–4896 (1997).

2. Bimston, D. et al. BAG-1, a negative regulator of hsp70 chaperone activity, uncouplesnucleotide hydrolysis from substrate release. EMBO J. 17, 6871–6878 (1998).

3. Stuart, J.K. et al. Characterization of interactions between the anti-apoptoticprotein BAG- 1 and Hsc70 molecular chaperones. J. Biol. Chem. 273, 22506–22514(1998).

4. Brehmer, D. et al. Tuning of chaperone activity of Hsp70 proteins by modulation ofnucleotide exchange. Nat. Struct. Biol. 8, 427–432 (2001).

5. Takayama, S. & Reed, J.C. Molecular chaperone targeting and regulation by BAG familyproteins. Nat. Cell Biol. 3, E237–E241 (2001).

6. Takayama, S. et al. Cloning and functional analysis of BAG-1: a novel Bcl-2-bindingprotein with anti-cell death activity. Cell 80, 279–284 (1995).

7. Wang, H.G. & Reed, J.C. Bc1–2, Raf-1 and mitochondrial regulation of apoptosis.Biofactors 8, 13–16 (1998).

8. Kermer, P. et al.BAG1 over-expression in brain protects against stroke.Brain Pathol. 13,495–506 (2003).

9. Eversole-Cire, P. et al. Synergistic effect of Bcl-2 and BAG-1 on the prevention ofphotoreceptor cell death. Invest. Ophthalmol. Vis. Sci. 41, 1953–1961 (2000).

10. Lee, M.Y. et al. Reactive astrocytes express bis, a bcl-2-binding protein, after transientforebrain ischemia. Exp. Neurol. 175, 338–346 (2002).

11. Jiang, Y., Woronicz, J.D., Liu, W. & Goeddel, D.V. Prevention of constitutive TNF receptor1 signaling by silencer of death domains. Science 283, 543–546 (1999).

12. Miki, K. & Eddy, E.M. Tumor necrosis factor receptor 1 is an ATPase regulated by silencerof death domain. Mol. Cell. Biol. 22, 2536–2543 (2002).

13. Thress, K., Song, J., Morimoto, R.I. & Kornbluth, S. Reversible inhibition of Hsp70chaperone function by Scythe and Reaper. EMBO J. 20, 1033–1041 (2001).

14. Thress, K., Henzel, W., Shillinglaw, W. & Kornbluth, S. Scythe: a novel reaper-bindingapoptotic regulator. EMBO J. 17, 6135–6143 (1998).

15. Wang, H.G., Takayama, S., Rapp, U.R. & Reed, J.C. Bcl-2 interacting protein, BAG-1,binds to and activates the kinase Raf-1. Proc. Natl. Acad. Sci. USA 93, 7063–7068(1996).

16. Wang, H.G., Rapp, U.R. & Reed, J.C. Bcl-2 targets the protein kinase Raf-1 tomitochondria. Cell 87, 629–638 (1996).

17. Datta, S.R. et al. Akt phosphorylation of BAD couples survival signals to the cell-intrinsicdeath machinery. Cell 91, 231–241 (1997).

18. Datta, S.R. et al. 14–3-3 proteins and survival kinases cooperate to inactivate BAD byBH3 domain phosphorylation. Mol. Cell 6, 41–51 (2000).

19. Datta, S.R. et al. Survival factor-mediated BAD phosphorylation raises the mitochondrialthreshold for apoptosis. Dev. Cell 3, 631–643 (2002).

20. Wiese, S. et al. Specific function of B-Raf in mediating survival of embryonic motoneur-ons and sensory neurons. Nat. Neurosci. 4, 137–142 (2001).

21. Froesch, B.A., Takayama, S. & Reed, J.C. BAG-1L protein enhances androgen receptorfunction. J. Biol. Chem. 273, 11660–11666 (1998).

22. Hohfeld, J. & Jentsch, S. GrpE-like regulation of the hsc70 chaperone by the anti-apoptotic protein BAG-1. EMBO J. 16, 6209–6216 (1997).

23. Schmidt, U. et al. Essential role of the unusual DNA-binding motif of BAG-1for inhibition of the glucocorticoid receptor. J. Biol. Chem. 278, 4926–4931(2003).

24. Yenari, M.A., Giffard, R.G., Sapolsky, R.M. & Steinberg, G.K. The neuroprotectivepotential of heat shock protein 70 (HSP70). Mol. Med. Today 5, 525–531 (1999).

25. Coldwell, M.J. et al. The p36 isoform of BAG-1 is translated by internal ribosome entryfollowing heat shock. Oncogene 20, 4095–4100 (2001).

26. Crocoll, A., Blum, M. & Cato, A.C. Isoform-specific expression of BAG-1 in mousedevelopment. Mech. Dev. 91, 355–359 (2000).

27. Lo, A.C., Houenou, L.J. & Oppenheim, R.W. Apoptosis in the nervous system: morpho-logical features, methods, pathology, and prevention. Arch. Histol. Cytol. 58, 139–149(1995).

28. Kermer, P. et al. Bag1 is a regulator and marker of neuronal differentiation. Cell DeathDiffer. 9, 405–413 (2002).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

29. Tran, J. et al. Marked induction of the IAP family antiapoptotic proteins survivin andXIAP by VEGF in vascular endothelial cells. Biochem. Biophys. Res. Commun. 264,781–788 (1999).

30. Brunet, A. et al. Akt promotes cell survival by phosphorylating and inhibiting a Forkheadtranscription factor. Cell 96, 857–868 (1999).

31. von Gise, A. et al. Apoptosis suppression by Raf-1 and MEK1 requires MEK- andphosphatidylinositol 3-kinase-dependent signals. Mol. Cell. Biol. 21, 2324–2336(2001).

32. Jablonka, S., Wiese, S. & Sendtner, M. Axonal defects in mouse models of motoneurondisease. J. Neurobiol. 58, 272–286 (2004).

33. Guan, K.L. et al. Negative regulation of the serine/threonine kinase B-Raf by Akt. J. Biol.Chem. 275, 27354–27359 (2000).

34. Michaelidis, T.M. et al. Inactivation of the bcl-2 gene results in progressive degenerationof motoneurons, sensory and sypathetic neurons during early postnatal development.Neuron 17, 75–89 (1996).

35. Gross, A., McDonnell, J.M. & Korsmeyer, S.J. BCL-2 family members and the mitochon-dria in apoptosis. Genes Dev. 13, 1899–1911 (1999).

36. Laprise, P. et al. Merosin (laminin-2/4)-driven survival signaling: complex modulationsof Bcl-2 homologs. J. Cell. Biochem. 89, 1115–1125 (2003).

37. Wiese, S. et al. The anti-apoptotic protein ITA is essential for NGF-mediated survival ofembryonic chick neurons. Nat. Neurosci. 2, 978–983 (1999).

38. Wang, C.Y., Mayo, M.W., Korneluk, R.G., Goeddel, D.V. & Baldwin, A.S.J. NF-kappaBantiapoptosis: induction of TRAF1 and TRAF2 and c-IAP1 and c-IAP2 to suppresscaspase-8 activation. Science 281, 1680–1683 (1998).

39. Beg, A.A., Sha, W.C., Bronson, R.T., Ghosh, S. & Baltimore, D. Embryonic lethality andliver degeneration in mice lacking the RelA component of NF-kappa B. Nature 376,167–170 (1995).

40. Deshmukh, M. & Johnson, E.M., Jr. Programmed cell death in neurons: focus on thepathway of nerve growth factor deprivation-induced death of sympathetic neurons. Mol.Pharmacol. 51, 897–906 (1997).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Nonsynaptic GABA signaling in postnatalsubventricular zone controls proliferation ofGFAP-expressing progenitors

Xiuxin Liu1, Qin Wang1, Tarik F Haydar2 & Angelique Bordey1

In the postnatal subventricular zone (SVZ), local cues or signaling molecules released from neuroblasts limit the proliferation of

glial fibrillary acidic protein (GFAP)-expressing progenitors thought to be stem cells. However, signals between SVZ cells have not

been identified. We show that depolarization of neuroblasts induces nonsynaptic SNARE-independent GABAA receptor currents in

GFAP-expressing cells, the time course of which depends on GABA uptake in acute mouse slices. We found that GABAA receptors

are tonically activated in GFAP-expressing cells, consistent with the presence of spontaneous depolarizations in neuroblasts that

are sufficient to induce GABA release. These data demonstrate the existence of nonsynaptic GABAergic signaling between

neuroblasts and GFAP-expressing cells. Furthermore, we show that GABAA receptor activation in GFAP-expressing cells limits their

progression through the cell cycle. Thus, as GFAP-expressing cells generate neuroblasts, GABA released from neuroblasts provides

a feedback mechanism to control the proliferation of GFAP-expressing progenitors by activating GABAA receptors.

The postnatal SVZ contains the largest pool of dividing neuralprogenitors in the adult brain. The SVZ consists of interconnectedchannels comprised of three major cell types1. The channel lumen ispacked with neuroblasts. Many of these proliferate and migrate towardsthe olfactory bulb where they become interneurons2–5. GFAP-immunopositive progenitors, also called astrocytes1,6, line the channelwalls and thus closely encapsulate neuroblasts. These GFAP-expressingcells constitute a pool of stem cells6–8. A small number of highlyproliferative progenitors are present in clusters along the channels.Overall, the anterior SVZ is mostly composed of neuroblasts (B60–70%) ensheathed by GFAP-expressing cells (B20%) that are thus in aprime location to send and receive signals from neuroblasts. Further-more, elimination of neuroblasts stimulates the proliferation of GFAP-expressing cells that regenerate the entire SVZ9, indicating that localcues or signaling molecules released from neuroblasts regulate theproliferation of GFAP-expressing cells. However, the signals betweenneuroblasts and GFAP-expressing cells have not been identified. Iden-tifying these signals and their receptors is of considerable interestbecause they provide therapeutic targets for controlling the prolifera-tion of GFAP-expressing progenitors and ultimately neurogenesis.

Among the possible signaling molecules, the neurotransmitterGABA is a prominent candidate for providing intercellular signalsbetween neuroblasts and GFAP-expressing progenitors, because SVZprogenitors synthesize and release GABA, and GFAP-expressing cellsexpress GABA transporters (GATs)10,11. GABA plays an importantsignaling role in developmental processes such as embryonic cell

proliferation and migration12–17 as well as in regulating the migratoryspeed and production of neuroblasts in postnatal SVZ11,18. Theexpression of GABA receptors in GFAP-expressing cells, which wouldprovide evidence in support of GABAergic signaling from neuroblaststo GFAP-expressing cells, has yet to be demonstrated. Furthermore, it isunknown whether neuroblasts release GABA that activates GABAreceptors on GFAP-expressing cells in a synaptic or nonsynaptic fashionand whether this release depends on the classical SNARE complex. TheSNARE complex, formed by the vesicle protein synaptobrevin alsoknown as VAMP (R-SNARE) and by the plasma membrane proteinsSNAP25 and syntaxin (Q-SNARE), is the minimal machinery requiredfor vesicle exocytosis. GABA release onto synaptically silent hippocam-pal neurons19 and from CA1 neurons20 is Q-SNARE–independent.

To determine whether neuroblasts communicate with GFAP-expressing cells by means of GABA signaling, we recorded fromGFAP-expressing cells in the postnatal SVZ using transgenic miceexpressing green fluorescent protein (GFP) driven by the humanGFAP promoter (GFAPP-GFP). Our data demonstrate the existenceof nonsynaptic, SNARE-independent GABAergic communicationbetween neuroblasts and presumed stem cells. Furthermore, wefound that GABAA receptor (GABAAR) inhibition increases the num-ber of proliferative GFAP-expressing cells. Thus, GABA released fromneuroblasts acts as a stop signal to limit GFAP-expressing cell prolif-eration, and this GABAergic communication between SVZ progenitorsmay contribute to maintaining a balance between amplification andmobilization of progenitors.


1Departments of Neurosurgery, and Cellular and Molecular Physiology, Yale University School of Medicine, New Haven, Connecticut 06520-8082, USA. 2Departments ofPediatrics and Pharmacology, George Washington University School of Medicine, Washington, D.C. 20010, USA. Correspondence should be addressed to A.B.([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

RESULTS

GFAP-expressing cells of the SVZ express functional GABAARs

Whole-cell patch-clamp recordings were made in acute slices frompostnatal transgenic GFAPP-GFP mice21. Fluorescent (GFP+) cells werevisualized before recording and then were filled with lucifer yellowduring recording to better examine their morphology. GFP+ cellsrecorded in the SVZ had cell bodies of irregular shape (B10-mm-long) with one or two main processes (B30- to 60-mm-long, Fig. 1a).Although striatal astrocytes adjacent to the SVZ were also GFP+, theywere easily distinguished from GFP+ cells within the SVZ owing to theirmore complex morphology (Supplementary Fig. 1 online). Striatalastrocytes near the SVZ were characterized by several main processesconfined to a sphere or ellipse of 80–150 mm in diameter. GFP+ cells inthe SVZ had a low mean input resistance (RIN, 50.9 ± 4.7 MO, mean ±s.e.m., n¼ 48) and a hyperpolarized mean resting potential (VR, �80.7± 0.6 mV, n ¼ 48) similar to those of astrocytes22 and radial glia23 butdifferent from those of GFP-negative progenitors (mean RIN of B3 GOand zero-current VR of �28 mV), which are presumably neuro-blasts11,24. We also performed immunostaining for nestin, a class VIintermediate filament protein expressed in SVZ cells but not inastrocytes1,25. Lucifer yellow–filled GFP+ cells recovered after fixationof slices stained positive for nestin (n ¼ 5/5, Fig. 1b), confirming that

GFP+ cells with an immature morphology recorded in the SVZ are theGFAP-expressing progenitors thought to be stem cells6–8.

All GFAP-expressing cells examined responded to focal applicationsof GABA (100 mM), which induced inward currents that rangedfrom �20 to �300 pA in amplitude at a holding potential of�80 mV (n ¼ 42, Fig. 1c). The variability in current amplitude islikely due to variability in the GABA concentration that reaches cellsrecorded at variable depths (50–150 mm) in the slice. GABA-inducedcurrents slowly desensitized and were accompanied by an increase innoise, which was mostly visible in the presence of 100 mM meclofen-amic acid (MFA), a gap junction blocker26. This increase in noise withMFA was accompanied by a decrease in GABA current amplitudes anda significant increase in RIN (to 305.4 ± 14.8 MO, n ¼ 178, Po 0.001)without affecting the series (access) resistance (Fig. 1d). These effectswere likely due to gap junction blockade by MFA26. Indeed, no dyecoupling was observed in GFP+ cells recorded in the presence of MFA(Fig. 1a,b), whereas GFP+ cells recorded without MFA showed dyecoupling between four to eight surrounding GFP+ cells (data notshown). Although MFA affects responses at certain GABAARs27,MFA did not alter the amplitude of gabazine-sensitive currents inoutside-out patches of GFAP-expressing cells (data not shown). MFA(100 mM) was then routinely bath applied after obtaining whole-cellrecordings of GFP+ cells to improve the space clamp of the recording.GABA responses in GFAP-expressing cells were reversibly blocked bythe GABAAR antagonists gabazine (20–50 mM, n¼ 5, Fig. 1e), bicucul-line (100 mM; n¼ 6) or picrotoxin (30–50 mM, n¼ 5, data not shown),and were mimicked by pressure application of a GABAAR agonistisoguvacine (10 mM, n¼ 3/3 and 100 mM, n¼ 3/3, data not shown). Asexpected for GABAAR-mediated (GABAA) currents, the mean reversalpotential for GABA-induced currents (mean of –52.8 ± 6.8 mV, n¼ 7)was near the Nernst equilibrium potential for chloride (ECl ¼ �54 mVwith an intracellular [Cl�] of 18.5 mM, Fig. 1f). The mean EC50 ofGABA was 15.1 mM (n ¼ 5, Fig. 2a,b), obtained in cells pulled out andraised above the slice. Furthermore, single-channel activity couldbe induced with 3 and 10 mM GABA in outside-out patches ofGFAP-expressing progenitors (data not shown). The dose-dependentinhibition of 100 mM GABA responses by picrotoxin gave an IC50 of11.9 mM (n ¼ 4, Fig. 2c).

Stimulus-evoked nonsynaptic GABAA currents in GFAP cells

SVZ neuroblasts synthesize and contain GABA10,11. We thus set out todetermine whether GABAARs could be activated by GABA releasedfrom SVZ progenitors after focal electrical stimulation in the SVZ.Single-pulse electrical stimulation in the SVZ did not induce currentsin GFAP-expressing cells (Fig. 3a). However, increasing the number ofpulses (2–5 pulses at 50 Hz) induced slow, long-lasting inward currentsof increasing amplitude (Fig. 3a). These evoked inward currents were

10 µm

100 µM MFA

100 µM GABA

Gabazine

10 µM MFA

MFA

100 pA

5 s

100 pA

5 ms

I (pA) 400

200

0

–200–100 –80 –60 –40 –20

–400

–600

V (mV)

100 pA

Control 3 s

Control

Control

10 µm

a

c

d

e f

b

Figure 1 GFAP-expressing cells in the SVZ express functional GABAARs.

(a) Photograph of a GFP+ cell recorded at the edge of the SVZ (upper panel).

Photograph of the lucifer yellow fill of the cell shown above (lower panel).

(b) Photograph of a lucifer yellow–filled GFP+ cell (green) immunostained for

nestin (red), a marker of immature cells. (c) GABA-induced currents before

and during application of MFA (100 mM). (d) Currents in response to a 10-mVhyperpolarizing pulse before and during MFA application from the same cell

as in c. (e) GABA-induced currents before and during application of the

GABAAR antagonist, gabazine (50 mM). (f) Current-voltage curve of GABA

responses obtained by applying a ramp protocol (from �120 to 0 mV

in 200 ms) near the peak of the current. GABA response reverses at

�55 mV as expected for a Cl–-carried current in our recording conditions

(ECl ¼ �54 mV). GFP+ cells were recorded at a holding potential of �80 mV.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

usually accompanied with an increase in noise and were reversiblyreduced by bicuculline (100 mM, n ¼ 3, Fig. 3b) and picrotoxin(50 mM, n ¼ 3, data not shown), suggesting that they were mediatedby GABAAR activation in GFAP-expressing cells. In addition, evokedcurrents persisted in the presence of AMPA/kainate and NMDAreceptor antagonists (20 mM DNQX and 50 mM D-AP5, respectively,n ¼ 5). Evoked GABAA currents ranged from �20 to �60 pA,suggesting that GABA release was small. Evoked currents showedslower kinetics than did synaptic currents recorded in other brainregions. Evoked currents developed slowly with a rise time between 200and 700 ms and were long-lasting, with a mean monoexponential decaytime constant (t) of 1.16 ± 0.10 s (n¼ 24, Fig. 3c). Fast inward currentsresembling synaptic currents were never observed after either a single ortetanic stimulation (Fig. 3c inset) or during bath application of 1 nMa-latrotoxin, which stimulates vesicular exocytosis from presynapticterminals28 (Fig. 3d, n ¼ 4). Although a-latrotoxin increased thefrequency of spontaneous synaptic currents in striatal neurons(Fig. 3d, n ¼ 6), the lack of an effect of a-latrotoxin could bedue to the lack of receptors necessary for the toxin internalizationto the cytoplasm28.

To further ensure that GABA released by SVZ progenitors diffusesand activates receptors in GFAP-expressing cells in a nonsynaptic

manner, we tested whether the amplitude and kinetics of evokedcurrents changed with the distance between the stimulating electrodesand recorded cells. The rise-time of evoked GABAA currents decreasedand amplitude increased when the stimulating electrode was movedcloser to the recorded cell (Fig. 3e,f). Stimulating electrodes wereusually placed 40 to 60 mm from the recorded cell (between positions band c in Fig. 3e), close to the lateral ventricle. GFP+ cells were recordedmidway in the SVZ except for experiments shown in Figure 3e.Variability in the distance between the stimulating electrode andrecorded cell and the presence of GATs are likely to contribute tothe variability in the evoked current amplitude and kinetics. Wenext examined whether evoked currents were independent of actionpotentials and extracellular or intracellular Ca2+. Evoked GABAA

currents in GFAP-expressing cells persisted in the presence of thevoltage-dependent sodium channel blocker tetrodotoxin (TTX,1 mM, n ¼ 4), in a Ca2+-free extracellular solution applied for 30 min(0 mM Ca2+ plus 1 or 2 mM EGTA, n ¼ 4) or in the presence of thevoltage-gated Ca2+ channel blockers Ni2+ and Cd2+ (each at 100 mM,n ¼ 4, Fig. 3g). These data suggested a nonsynaptic release of GABA.TTX, Ni2+ and Cd2+ or a Ca2+-free solution abolished or significantlyreduced GABAA synaptic currents in Purkinje neurons induced by one-pulse stimulation in the molecular layer (data not shown). When sliceswere incubated for 30 min in 30 mM BAPTA-AM in a Ca2+-freesolution to buffer intracellular Ca2+, no evoked currents were induced

I/Imax

EC50 = 15.1 µM

[GABA] (µM)

0.0 0.5 1.0 1.5 2.0 2.50

20

40

60

80

100

IC50 = 11.9 µM

Imax–I/Imax

Log[PTX] (µM)1 10 100 1,000

0

20

40

60

80

100

10 pA5 s

GABA3

10

30

b ca

100

500

Figure 2 GABA and picrotoxin affinities for GABAARs in GFAP-expressing

cells of the SVZ. (a) GABA responses induced by increasing GABA

concentrations. (b,c) Mean dose-response curve for GABA (b) and picrotoxin

(PTX, c) in GFAP-expressing cells gave an EC50 of 15.1 mM (n ¼ 5) and an

IC50 of 11.9 mM (n ¼ 4), respectively. Four to six GABA concentrations weresuccessively bath applied to each cell. The resulting averaged dose-response

curve was fitted with a classical Hill equation. Errors bars represent s.e.m.

1 pulsea b

c d

e f

g

i

h

3 pulses 5 pulses

20 pA

5 s

20 pA

20 pA

20 pA

20 pA

30 µm

Normalized amplitudeRise-time

Distance (µm)20

0

20

40

60

80

100%

30 40 50 60 70 802 s

2 s20 pA

3 s20 pA

2 s

20 pA

2 s

10 pA

4 s

TTX 0 Ca2+/1 mM EGTA Cd2+ + Ni2+ BAPTA-AM

Stimulation in the SVZ In the striatum Back in the SVZ

200 ms

a

ab

b

cc

4 s

30 pA

5 s Control

Control (striatal neuron)

α-latrotoxin

α-latrotoxin (GFP+ cell)

100 µM BIC

Figure 3 Electrical stimulation of SVZ progenitors evoked nonsynaptic

GABAA currents in GFAP-expressing cells. (a) Repetitive focal electrical

stimulation in the SVZ (one, three and five pulses of 200 ms, 50 Hz)

induces inward currents in GFAP-expressing cells. (b) Currents evoked by a

five-pulse stimulus are reversibly blocked with bicuculline (BIC, 100 mM).

(c) Evoked currents develop and decay slowly (rise time: 680 ms;

monoexponential decay time constant: 1.7 s). Inset: trace at higher time

scale. (d) Spontaneous synaptic events are observed in striatal neurons (top

two traces) but not in GFAP-expressing cells (bottom trace) during 1-nM

a-latrotoxin applications. Transient currents represent responses to 10-mV

depolarizing pulses. Scale bars: 50 pA and 20 s (neuron), and 20 pA and

20 s (GFAP+ cell). (e) GABAA currents evoked by electrical stimulation at

positions a, b and c (arrows) in photograph at right. The arrowhead points to

the recorded cell. (f) Plots of the normalized amplitudes and rise times of

evoked GABAA currents against the distance between the stimulating

electrodes and recorded cells. (g) Evoked currents recorded with 1 mM TTX

(left), in a zero-Ca2+ solution (30 min, middle) and in Cd2+ and Ni2+ for30 min (each at 100 mM, right). (h) Currents were not evoked in slices

incubated in a Ca2+-free solution containing BAPTA-AM for 30 min before

recording. (i) Focal electrical stimulation in the SVZ induces inward currents

in GFAP-expressing cells recorded midway in the SVZ, whereas no current

was induced with an identical stimulation in the striatum. Stimulation of the

SVZ after stimulation of the striatum still induced evoked currents.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

in GFAP-expressing progenitors (n¼ 9 cells, 72 stimulations, five slices,Fig. 3h). Nevertheless, GABA responses could still be induced in SVZprogenitors by exogenous GABA applications (n¼ 4, data not shown).These data suggest that increases in intracellular Ca2+ levels arenecessary for inducing GABA release.

Consistent with a nonsynaptic release, we found no immunostainingfor synapsin I near GFAP-immunopositive cells in the SVZ (Supple-mentary Fig. 2 online). Synapsin I is a specific marker of synapses29.Because the striatum contains GABAergic terminals, we next examinedwhether tetanic stimulation in the striatum adjacent to the SVZinduced GABA responses in GFAP-expressing cells. Although tetanicstimulation in the SVZ induced inward currents in GFAP-expressingcells recorded midway in the SVZ, the same stimulation applied in thestriatum the same distance from the recorded cells did not induce suchcurrents (n ¼ 4, Fig. 3i). Although GABA is released from synapticterminals in the striatum, dilution by diffusion and uptake into bothastrocytes and GFAP-expressing cells is likely to prevent significantlevels of synaptically released GABA from reaching recorded cells in the

SVZ. These data indicate that GABA or a GABAAR agonist is non-synaptically released from SVZ progenitors and activates GABAARs inGFAP-expressing cells.

SNARE-independent GABA release from SVZ progenitors

Because evoked GABA currents were dependent on intracellular Ca2+,we examined whether GABA release was abolished after inhibitionof the SNARE complex. We incubated slices with botulinum neuro-toxin A (BoNT/A) or B (BoNT/B) to block the Q- or R-SNAREcomplex, respectively. Bath applications of BoNT/A or BoNT/B (eachat 200 ng/ml) for 30 min significantly reduced (P o 0.001) thefrequency of spontaneous gabazine (20 mM)-sensitive synaptic currentsby 80% in striatal neurons without affecting cell capacitance and RIN

(n ¼ 4, Fig. 4a). Synaptic activity was studied in the presence ofAMPA/kainate and NMDA receptor antagonists (20 mM DNQX and50 mM D-AP5). However, similar treatment with BoNT/A or BoNT/Bfor 30 min did not affect evoked GABAA currents in GFAP-expressingcells (n¼ 4 for each toxin, Fig. 4b). Treated and control slices were thenincubated for 18 h in a solution containing 100 ng/ml of BoNT/A or

50 pA

20 s

20 pA

3 s

a

BoNT/B (30 min)

c

50 pA

1 min

b

d

Striatal neuron GFAP-expressing cell

30 pA

1 min

No BoNT/B (18 h)

Striatal neuronBoNT/B (18 h)

GFAP-expressing cell

Control

BoNT/B (18 h)

10 pA

1 s

10 pA

2 s

BoNT/B (30 min)

100 pA

10 ms

Figure 4 GABA release from SVZ progenitors is independent of SNARE.

(a) Spontaneous synaptic currents in a striatal neuron under control

conditions and after 30 min application of 200 ng/ml BoNT/B. BoNT/B

did not affect series resistance or cell capacitance measured by applying

a 10-mV hyperpolarizing pulse before and after BoNT/B application

(traces at right). (b) Evoked GABAA currents in GFAP-expressing cells after a

30-min application of BoNT/B. (c) Records of synaptic activity in striatal

neurons in slices incubated for 18 h in a solution with or without 100 ng/mlBoNT/B. (d) Evoked GABAA currents in GFAP-expressing cells in slices

treated with BoNT/B for 18 h.

a Control(with DNQX + AP5)

NO-711 + SNAP5114 PTX

0

50

100

150

200

250

300

Per

cent

age

incr

ease

with

NO

-711

+ S

NA

P51

14

*

b

10 pA

2 s

Amplitude τ

c*

NO-711 + SNAP5114

10 pA

20 s

d

0 2 4 6 8 100.0

0.5

1.0

1.5

NO-711 SNAP5114

3

1

2

4

20 pA

2 s

NO-711

SNAP5114

12

34

123

Dec

ay (

s)

Time (min)

1 s

GABA

80 pA

40 se f g

Figure 5 GABA uptake controls GABAAR activation in GFAP-expressing

cells of the SVZ. (a) Evoked GABAA currents in control, in the presence of

N0-711 and SNAP5114 and in the presence of PTX. Bath application of

GABA transporter antagonists (50 mM NO-711 for GAT1 and 50 mM

SNAP5114 for GAT3/GAT4) enhances the amplitude and prolongs the decay

time of picrotoxin-sensitive currents evoked by tetanic stimulation in the SVZ.

(b) Percentage increases of the evoked GABAA current amplitude and decay

time constant in the presence of NO-711 and SNAP5114. Errors bars

represent s.e.m. (c) NO-711 and SNAP5114 (50 mM each) reversibly induce

an inward shift of the holding current in an GFAP-expressing cell. (d) Inward

currents evoked by pressure applications of GABA (100 mM, 100 ms, every

20 s; marked by arrowheads) before and during bath application of NO-711

(50 mM) and SNAP5114 (50 mM). Note that GABA responses were preceded

by short 10-mV depolarizations (transients on the recordings) to monitor the

series and input resistance. In addition, NO-711 and SNAP5114 induced

inward currents, suggesting that GABA transporter inhibition induced an

increase in ambient GABA levels. (e) Average trace of three successive GABA

responses at a higher time scale under control (1), in the presence of NO-711 (2) or SNAP5114 (3) and after washout (4). (f) Superimposed GABA

responses (averaged) under control (top trace) and with NO-711 (middle

trace) or SNAP5114 (bottom trace). (g) Monoexponential decay time

constant of GABA response against the recording time.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

BoNT/B (without MFA) and no toxin, respectively. SpontaneousGABAergic synaptic currents were absent in neurons recorded intreated slices but were present in neurons from control slices (Fig. 4cfor BoNT/B; data not shown for BoNT/A). Furthermore, a-latrotoxin(1 nM) did not induce spontaneous synaptic currents in neurons fromtreated slices (n ¼ 5, data not shown). Evoked slow GABAA currentsin GFAP-expressing cells persisted in slices treated with either toxin(n ¼ 5, BoNT/A; n ¼ 6, BoNT/B (Fig. 4d)). These data suggest thatGABA release from SVZ progenitors is SNARE-independent.

GABA uptake controls GABAAR activation in GFAP cells

Four subtypes of GATs (in mice, GAT1 to GAT4) have been reported30.GAT2 is less widespread in the brain and has the lowest affinity forGABA31. Bath application of non-transportable antagonists of the threemajor GATs, 50 mM NO-711 for GAT1 and 50 mM SNAP5114 forGAT3/GAT4, significantly enhanced the amplitude and prolonged thedecay of picrotoxin- and bicuculline-sensitive inward currents evokedby tetanic stimulation (n ¼ 5, Po 0.05, Fig. 5a,b). Furthermore, NO-711 and SNAP5114 induced inward shifts of the holding current ofabout �5 to �20 pA (n ¼ 6/10, Fig. 5c). These results suggest thatGATs decrease ambient GABA levels and limit GABAAR activation inGFAP-expressing cells. To determine whether both GAT1 and GAT3/GAT4 contributed to the clearance of GABA, we tested the effects ofeach blocker on GABA responses induced by pressure applications ofGABA (100 mM, 100 ms, Fig. 5d). SNAP5114 significantly (P o 0.05)increased the amplitude of GABA-induced responses by 61 ± 39% in3/5 cells, whereas NO711 had no effect on the amplitude (n ¼ 5,Fig. 5e). SNAP5114 and NO-711 significantly increased t of GABAresponses by 91 ± 26% (n¼ 7) and 33 ± 15% (n¼ 5), respectively (Po0.05, Fig. 5f,g). These data suggest that GAT3/GAT4 transporters aremore efficient than the GAT1 transporter at regulating ambient andnonsynaptically released GABA levels in the SVZ.

Depolarization-induced GABAA currents in GFAP cells

We examined whether subtle and near-physiological depolarization ofSVZ progenitors induced GABAAR activation in GFAP-expressing cells.Bath applications of 7.5 and 12.5 mM K+ or focal (pressure) applica-tions of 12.5 mM K+ depolarized neuroblasts by 10 and 15 mV,respectively (n ¼ 9, without MFA in the bath, data not shown). Bathand pressure applications of K+, which depolarizes every progenitor ora local group of SVZ progenitors, respectively, induced inward currentsin GFAP-expressing cells that were reversibly reduced by bicuculline(n ¼ 8, Fig. 6a) or picrotoxin (n ¼ 6, Fig. 6b). Neither picrotoxinnor bicuculline affected the series (access) resistance or capacitancecalculated in response to a 10-mV depolarizing pulse (Fig. 6b for

picrotoxin). As a control for high K+ application, application of acontrol solution into which 5 mM sucrose was added induced nocurrents in GFAP-expressing cells (n ¼ 3). As expected for GABAA

currents, the mean reversal potentials for high K+–induced GABAA

currents were �58.7 ± 2.3 mV and +1.9 ± 1.4 mV (n ¼ 6, data notshown) when using a low (ECl ¼ �54 mV) and high internal Cl–

solution (ECl ¼ 1.3 mV), respectively. High K+–induced GABAA

currents persisted in the presence of TTX (data not shown) or in aCa2+-free solution (n ¼ 4, Fig. 6a). Furthermore, no fast, transientcurrents resembling synaptic currents were observed during high K+

applications. Picrotoxin-sensitive currents induced by 12.5 mM K+

(bath) had significantly larger amplitudes (�34.0 ± 5.8 pA) thanthose induced by 7.5 mM K+ (�22.0 ± 4.0 pA, n ¼ 6, P o 0.05).Focal (pressure) application of 12.5 mM K+ induced GABAA currentsof �10 to �20 pA (n ¼ 5). High K+ applications also inducedbicuculline- and picrotoxin-insensitive currents (Fig. 6a,b). BecauseGFAP-expressing cells have a resting K+ conductance like astrocytes(data not shown), bicuculline- and picrotoxin-insensitive inwardcurrents induced by K+ are likely to represent a change in the restingK+ current due to a shift in the Nernst equilibrium potential for K+.

We next examined whether 10- to 15-mV depolarizations sponta-neously occurred in neuroblasts. Neuroblasts were recorded in currentclamp mode from �60 mV, which is near their estimated restingpotential10, and without MFA. Nevertheless, MFA had no effect onthe resting potential, RIN, or on the spontaneous depolarizationsobserved in neuroblasts (data not shown). Seventy-one percent of theneuroblasts displayed spontaneous depolarizations of 15–20 mV(Fig. 6c), which occurred at a frequency of 0.3 to 12/min and were in-sensitive to GABAAR blockers. These large spontaneous depolarizations

0 Ca2+ + 1 mM EGTA

100 µM BIC7.5 mM K+ (bath)

30 pA

5 s

a

40 pA

50 ms

Control

Control

10 pA

2 s

12.5 mM K+ (focal)

50 µM PTX50 µM PTX

b

c Neuroblast recordings

10 mV

5 s

d e PTX with NO-711 + SNAP5114

10 pA

30 s

BIC

1 2

31 2

3

5 pA

30 s

Figure 6 Near-physiological depolarization of SVZ progenitors induces

activation of GABAARs in GFAP-expressing cells that are also tonically

activated. (a) Bath application of 7.5 mM K+ induces bicuculline-sensitive

inward currents in a GFAP-expressing cell recorded in a Ca2+-free solution.

(b) Focal pressure application of 12.5 mM K+ to a cluster of putative

neuroblasts induced picrotoxin-sensitive currents in GFAP-expressing cells.

High K+ applications are not accompanied by changes in series (access)

resistance or capacitance calculated in response to a 10-mV depolarizingpulse applied before and during picrotoxin application. (c) Spontaneous

depolarizations of 15–20 mV in neuroblasts at �60 mV. (d) Bicuculline

(100 mM) induces a 5-pA outward shift of the holding current in a GFAP-

expressing cell without affecting series resistance or cell capacitance

measured by applying a 10-mV depolarizing pulse before, during and after

bicuculline application. The extracellular solution contained 2.5 mM K+.

Scale bar: 100 pA, 20 ms. (e) Picrotoxin (50 mM) induces a 20-pA outward

shift of the holding current in a GFAP-expressing cell in the presence of

NO-711 and SNAP5114 (50 mM each).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

were observed during the whole length (45–75 min) of stable whole-celland perforated recordings, suggesting that they were not due to celldamage. Spontaneous single channel currents of �5 pA amplitude thatcorrespond to a 15–20 mV depolarization due to the large RIN ofneuroblasts (3–4 GO) were also observed in voltage clamped neuro-blasts. Forty percent of the neuroblasts also showed small gabazine-sensitive depolarizations ofB5–8 mV (data not shown), consistent withtonic GABAAR activation in neuroblasts11,18. These results show thatK+-induced depolarizations of SVZ progenitors similar in amplitude tothe spontaneous depolarizations found in SVZ neuroblasts can inducenonsynaptic GABAAR activation in surrounding GFAP-expressing cells.

Neuroblasts mediate tonic GABAA current in GFAP cells

The presence of spontaneous depolarizations in neuroblasts suggeststhat GABAARs are tonically activated in some GFAP-expressing cellsowing to spontaneous GABA release. Bath application of bicuculline(100 mM, Fig. 6d) or picrotoxin (50 mM, Fig. 6e) induced a reversible+5 to +30 pA outward shifts of the holding currents when recorded in a2.5 or 5 mM K+ extracellular solution with GAT inhibitors (50 mMNO-711 and 50 mM SNAP5114, n ¼ 12/17) or without GAT inhibitors(n ¼ 7/12). This outward shift was not accompanied by a change inseries resistance or cell capacitance (Fig. 6d). These results indicate thatthere is sufficient ambient GABA in the SVZ to tonically activateGABAARs in GFAP-expressing cells.

Although electrical and high-K+ stimulations indiscriminately depo-larize all SVZ cells, B70% of the SVZ cells that stain positive for GABAare neuroblasts11, suggesting that neuroblasts are the major source ofGABA in the SVZ. Furthermore, GFP+ cells did not stain positive forGABA (Fig. 7a). To conclusively show that neuroblasts release GABAspontaneously and upon depolarization, we used outside-out patchesof striatal neurons or GFAP-expressing cells held at –80 mV to monitorGABAA single channels (without MFA) and detect GABA release fromsimultaneously recorded neuroblasts (Fig. 7b–e). Neuroblasts wererecorded in the cell-attached mode to keep the intracellular milieuintact. Cell-attached patches were held at a pipette potential (Vpip) of0 mV (patch potential Vpatch ¼ VR, using Vpatch ¼ VR � Vpip).Experiments were performed with AMPA/kainate and NMDA receptor

blockers (DNQX and D-AP5). Because gabazine (20 mM)-sensitivesingle channel activity was detected when outside-out patches wereB5 mm above the slices, outside-out patches were raised at B50 mmabove the slice where no single channel activity was observed (Fig. 7c).The cell-attached neuroblasts were first positioned at 50 mm from thepatches and from the slice. A total of 21 paired recordings (fourneuron-neuroblast and 17 GFAP-expressing cell–neuroblast pairs)were obtained. Fifty-two percent of the outside-out patches (sevenfrom GFAP-expressing cells and four from striatal neurons) displayedsingle-channel activity when placed o5 mm from the recorded neuro-blast (Fig. 7b–d). This activity was abolished by 20 mM gabazine inoutside-out patches in neurons (n ¼ 3, Fig. 7c) and GFAP-expressingcells (n ¼ 3, Fig. 7d). Furthermore, single channels had a meanamplitude of �2.0 ± 0.1 pA in GFAP-expressing cells (n ¼ 13) and�2.2 ± 0.1 pA in neurons (n¼ 4) at�80 mV corresponding to a typicalGABAAR conductance of 24.5 ± 0.7 pS and 27.7 ± 0.9 pS, respectively(assuming a reversal potential of B0 mV in our recording condition).Six of ten of the remaining silent patches from GFAP-expressing cellsplaced at o5 mm from a recorded neuroblast displayed single-channelactivity upon a 5-s-long depolarization of the progenitor from a Vpatch

of �60 to +60 mV (using VR of �60 mV; refs. 10,24). The single-channel activity lasted 30–60 s and was blocked by gabazine (n¼ 3; seeinsets in Fig. 7c–e showing single channels at a higher time scale andcorresponding amplitude histograms). These data demonstrate thatneuroblasts release GABA both spontaneously and upon depolarizationand that released GABA can directly activate GABAARs on neighboringGFAP-expressing cells.

GABAAR activation reduces the proliferation of GFAP cells

We examined whether activation of GABAARs in GFAP-expressing cellsinfluence the proliferation of these cells, as previously shown for

SVZ

Striatum

10 µm

40 µm

Patch (neuron) >50 µm from neuroblast

Before neuroblast depolarization (GFP+ cells <5 µm from NP)

After neuroblast depolarization (5 s)

Patch <5 µm from NP, control

Patch (GFP+ cell) <5 µm from neuroblast

Gabazine

Gabazine

3 pA

5 min

10 min

3 pA

3 s

3 s

Wash

3 pA

3 s200

200

100

100

0

0

–3 –2Current (pA)

–1 0

–3 –2Current (pA)

–1 0

Cou

nt/0

.08

pA

Cou

nt/0

.08

pA

a b

c

e

dFigure 7 Neuroblasts release GABA spontaneously and upon depolarization.

(a) GABA immunostaining (red) in the SVZ of GFAPP-GFP mice.

(b) Photograph of a neuroblast recorded in cell-attached mode (lower arrow)above the slice. The second patch electrode contains an outside-out patch

from a GFP+ cell (or a striatal neuron, as in c, or a GFP+ cell, as in d,e)

positioned o5 mm from the simultaneously recorded neuroblast (upper

arrow). (c) Simultaneous recordings from a cell-attached neuroblast (Vpip of

0 mV) and an outside-out patch from a striatal neuron held at –80 mV. No

single-channel activity is detected when the patch (NP) is 50 mm away from

the neuroblast, but channel activity is observed when the patch is o5 mm

from the neuroblast shown in b and is blocked by 20 mM gabazine. Trace at

bottom shows trace between the arrows at a higher time scale (scale bar,

200 ms). Amplitude histogram reflects middle records. Horizontal dashed

lines in all panels represent channel in a closed state. (d,e) Simultaneous

recordings from a cell-attached neuroblast and an outside-out patch from a

GFP+ cell positioned o5 mm from the neuroblast. No channel activity is

observed when recording with 20 mM gabazine (d; top), but single-channel

activity develops upon gabazine removal. Times at right represent time after

wash start. Trace at bottom shows trace between the arrows at a higher time

scale (scale bar, 500 ms). Spontaneous channel activity develops (e) after

a 5-s-long depolarization of the cell-attached neuroblast (Vpip from 0 to�120 mV). Traces at bottom represents area between arrows at higher time

scale (scale bar, 500 ms). Amplitude histogram of depolarization-induced

channels is shown at left. These experiments were performed in the presence

of 20 mM DNQX and 50 mM D-AP5, but without MFA.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

embryonic cells14, including SVZ cells13. Pairs of hemisected slices wereincubated in the presence of the mitotic marker 5-bromo-2¢-deoxy-uridine (0.05% BrdU) for 18 h either under control conditions (with-out MFA) or in the presence of bicuculline (50 mM). After 18 h,hemisected slices were fixed and processed for GFP and BrdU immu-nostaining, and confocal z-stacks (ten sections spaced by 1 mm) of twoto three fields of view were taken in control and treated (opposite)hemispheres. Figure 8a represents a reconstruction of a 10-mm z-stackin control sections and in bicuculline-treated sections. Inhibitionof GABAARs with bicuculline alone significantly (Po 0.001, unpairedt-test with unequal variance) increased the percentage of GFP-immu-nopositive cells staining positive for BrdU from 8.8 ± 1.5% (665 GFP+

cells counted, 95 ± 41 cells, mean ± s.e.m.) to 17.9 ± 1.5% (1,064 cells,mean of 142 ± 47 cells, Fig. 8b, six pairs of hemisected slices from fivemice). To prevent an increase in the slice excitability and release ofvarious factors from the slice, these experiments were repeated inthe presence of the glutamate receptor blockers DNQX (20 mM)to block AMPA/kainate receptors and D-AP5 (50 mM) to blockNMDA receptors. In this condition, inhibition of GABAARs alsosignificantly (P o 0.001) increased the percentage of GFP-immuno-positive cells that stained positive for BrdU from 8.8 ± 1.1% (1,643GFP+ cells, mean of 182 ± 7 cells) to 19.7 ± 1.7% (1,238 cells, mean of137 ± 15 cells, Fig. 8c; four pairs of slices, nine resected sections, threemice). We next tested whether the blocker of GAT3/GAT4 transportersSNAP5114 (100 mM) would limit the number of proliferative GFAP-expressing cells, as we have shown previously that SNAP5114 increasedambient GABA levels in the SVZ11 (see also Fig. 5). Inhibition of GAT3/GAT4 transporters with SNAP5114 significantly (P o 0.05) reducedthe percentage of GFP-immunopositive cells that stained positive forBrdU from 10.9 ± 1.9% (1,909 GFP+ cells, mean of 127 ± 15) to 6.7 ±1.2% (1,979 cells, mean of 116 ± 13, Fig. 8c; six pairs of slices,15 resected sections, four mice). These data suggest that GABAARactivation in GFAP-expressing cells by ambient GABA reduces theproliferation of these cells.

DISCUSSION

We demonstrate here that nonsynaptically, SNARE-independentrelease of GABA from neuroblasts activates GABAARs in GFAP-

expressing cells of the SVZ, thought to be stem cells6–8, andthat tonic GABAAR activation in GFAP-expressing cells reducestheir proliferation.

Nonsynaptic GABAergic signaling between SVZ progenitors

Recorded cells were identified as GFAP-expressing cells before record-ing by direct visualization of fluorescence in GFAPP-GFP mice21.Recorded GFP+ cells had passive membrane properties similar tothose of astrocytes but had a less elaborated morphology22. In addition,recorded GFP+ cells stained positive for nestin, a marker of immaturecells but not astrocytes1,25. GABA currents in GFAP-expressing cellswere blocked by several GABAAR antagonists and were carried by Cl–

ions, identifying these currents as GABAAR-mediated responses. Elec-trical or high-K+ stimulations of SVZ progenitors induced nonsynapticGABAA currents in GFAP-expressing cells, suggesting that SVZ pro-genitors released GABA. Evoked currents developed slowly and decayedover several seconds. This time course is significantly slower than thatof synaptic currents. The four following results support nonsynapticGABA release: (i) synapsin I staining was absent in the SVZ, (ii) thekinetic course and amplitude of evoked GABAA currents were depen-dent on the distance between stimulating electrodes and recorded cells,(iii) blockade of action potentials or voltage-gated Ca2+ channels orremoval of extracellular Ca2+ did not prevent evoked GABAA currents,and (iv) a-latrotoxin, which stimulates vesicular exocytosis frompresynaptic terminals28, did not induce any synaptic currents inGFAP-expressing cells. The persistence of evoked GABAA currents inGFAP-expressing cells after BoNT treatments suggests that GABArelease is independent of a classical SNARE complex. Taken together,these results demonstrate that GABA is nonsynaptically, and indepen-dently of SNARE, released in the postnatal mammalian SVZ. Suchnonsynaptic, Q-SNARE–independent GABAergic communication hasbeen observed in embryonic and neonatal hippocampal neurons19.However, in contrast with GABAergic signaling in the neonatalhippocampus, we found that GATs control ambient GABA levels andlimit the diffusion of released GABA. GAT3/GAT4, and to a lesserextent GAT1 transporters, have a major role in regulating ambientGABA levels in the SVZ.

GABA is present in neuroblasts11 but not in GFAP-expressing cells.We cannot rule out that GABA is also contained in glioblasts, but thispopulation is a minority at the ages studied11,32,33. It has beenpreviously reported by us and others that SVZ neuroblasts expressglutamic acid decarboxylase10,34,35. Neuroblasts demonstrated sponta-neous depolarizations, and our paired recordings indicate that neuro-blasts release GABA spontaneously and upon depolarization.Collectively, these data strongly suggest that GABA is spontaneouslyreleased from neuroblasts, diffuses and activates GABAARs on neigh-boring GFAP-expressing cells known to ensheath neuroblasts1.

BICBIC Control SNAPControl0

10

20

30

0

10

20

30

Per

cent

age

ofG

FP

+-B

rdU

+/G

FP

+ c

ells

Per

cent

age

ofG

FP

+-B

rdU

+/G

FP

+ c

ells**

***

+D-AP5 + DNQX

20 µm20 µm

Control (+D-AP5 + DNQX)a

b c

Bicuculline Figure 8 GABAAR activation limits the proliferation of GFAP-expressing cells.

(a) Reconstructions of 10-mm z-stacks of BrdU (red) and GFP (green)

immunostaining in control and bicuculline-treated slices. White arrows

indicate some GFP+ cells that are BrdU+. The reconstructed stack in the

bicuculline-treated slice included a region outside the SVZ. BrdU+ cells were

within the SVZ. (b) Percentage of GFP+ cells that are BrdU+ in control slices

and in 50 mM bicuculline–treated slices. (c) Percentage of GFP+ cells that

are BrdU+ in control slices in the presence of AMPA/kainate and NMDAreceptor blockers (20 mM DNQX and 50 mM D-AP5, respectively), and in

50 mM bicuculline– or 100 mM SNAP5114–treated slices. *: P o 0.05,

**: P o 0.001 (unpaired t-test with unequal variance). Box: 25th and

75th percentiles. Short horizontal lines at top and bottom: 5th and 95th

percentiles. Filled circle: minimum and maximum. Open square: mean.

Horizontal line inside box: median.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Although further studies are required to define the mechanisms ofGABA release from neuroblasts, the following two mechanisms can beruled out: (i) GABA transport reversal (although release via GAT2transport reversal cannot be ruled out) and (ii) release throughhemichannels as observed for ATP and glutamate36,37, as MFA is ablocker of hemichannels26. One possible mechanism of GABA release isunconventional exocytosis triggered by Ca2+ release from intracellularstores, as buffering intracellular Ca2+ prevents the induction of evokedGABAA currents. Furthermore, neuroblasts display spontaneous Ca2+

transients (F. Huang and A.B., unpublished observations) and containsynapsin III, a protein associated with synaptic vesicles38,39.

Function of GABAergic signaling between SVZ progenitors

Our data indicate that in the SVZ, neuroblasts communicate withGFAP-expressing cells by releasing GABA that activates GABAARs inGFAP-expressing cells (Supplementary Fig. 3 online). Furthermore, weshow that GABAAR activation in GFAP-expressing cells limits theirprogression through the cell cycle. As more neuroblasts are generated, itis possible that more GABA is released in the extracellular space,resulting in increased ambient GABA levels and GABAAR activationin GFAP-expressing cells. Because a subset of GFAP-expressing cells arethought to generate neuroblasts6–8, an increase in the number ofneuroblasts serves as negative feedback, decreasing GFAP-expressingcell proliferation and neuroblast production by activating GABAARs.This is consistent with earlier studies showing that GABA altersproliferation14,15,18 and, in particular, reduces the proliferation ofembryonic SVZ cells13, which are thought to be lineally related toGFAP-expressing cells in the postnatal SVZ40. Thus, with respect toambient GABA’s proliferative action, there is a parallel between resultsfrom embryonic and postnatal SVZ cells. Such a feedback mechanismmight also apply for glioblast production from GFAP-expressing cells orradial glia in the neonatal SVZ, assuming that glioblasts containGABA32,33,41. Furthermore, the feedback mechanism provided byneuroblasts on GFAP-expressing cell proliferation fits well with theconstant migration of neuroblasts away from the SVZ to the olfactorybulb5,42,43, which would limit ambient GABA accumulation. Thisfeedback also fits well with the increased proliferation of GFAP-expressing cells after elimination of neuroblasts9. In conclusion, changesin GABA levels due to neuroblast migration or production provide amechanism to control the proliferation of GFAP-expressing cells andmaintain a balance between neuroblast production and mobilization.

METHODSSlice preparation and electrophysiology. Coronal (250-mm-thick) brain slices

from 16- to 41-day-old transgenic mice (Jackson Laboratory)21 were prepared

as previously described10,24. Experimental procedures were in accordance with

the animal welfare guidelines of Yale University. The standard external solution

contained (in mM) NaCl 125, KCl 2.5, CaCl2 2, MgCl2 1, NaH2PO4 1.25,

NaHCO3 25, glucose 10, pH 7.4 when equilibrated with 95% O2/5% CO2.

Whole-cell and gramicidin-perforated patch-clamp recordings were obtained as

previously described10,22,44 and as detailed in Supplementary Methods online.

GFAP-expressing cells had a mean capacitance of 17.7 ± 0.5 pF (n ¼ 111) with

100 mM MFA. Electrical stimulation was performed using a 6–8 MO patch

pipette (tip diameter o1 mm) filled with ACSF and surrounded with a silver

wire. For BoNT treatment, acutely prepared slices were transferred to tissue

culture inserts (0.4 mm membrane, Becton Dickinson) in 35 mm six-well plates,

which contained external solution (1 ml/well) with antibiotics with or without

100 ng/ml BoNT/A or BoNT/B, and were incubated for 18 h at 371C in a 95%

O2/5% CO2-saturated atmosphere.

Immunohistochemistry and proliferation assay. Slices were fixed for 3 h

in 4% paraformaldehyde (PAF) in PBS containing 4% sucrose (PBSS).

Immunostaining was performed as previously described10,11. The primary

antibodies included mouse anti-nestin IgG1 antibody (1:1,000, Rat-401,

Developmental Studies Hybridoma Bank, developed under the auspices of

the National Institute for Child Health and Human Development and main-

tained by the University of Iowa), rabbit anti-GABA (1:500, Sigma), mouse

Alexa 594-conjugated anti-BrdU IgG1 antibody (1:1,000, Molecular Probes)

and rabbit anti-GFP (1:200, Abcam). Secondary antibodies were the appro-

priate AlexaFluor (1:1,000, Molecular Probes). Stained sections were viewed on

a confocal scanning laser microscope (BioRad MRC600) or on an epifluores-

cence microscope (Olympus BX51) using standard procedures. Proliferation

experiments were exactly performed as previously described45 and as detailed in

Supplementary Methods.

Data analysis. Data acquisition and analysis were performed using PClamp 8

(Axon Instruments). For analysis of cell proliferation, confocal z-stacks (ten

sections spaced by 1 mm) were obtained in two to three fields of the proximal

RMS or anterior SVZ per slice using a LSM 510 META NLO confocal

microscope (Carl Zeiss). The number of GFP+ cells per field and then the

number of cells that were both GFP+ and BrdU+ were counted using Adobe

Photoshop. None of the control proliferation data (for SNAP5114 and the

bicuculline experiments) were significantly different. We combined the two

control data for SNAP5114 and bicuculline that were performed with gluta-

mate receptor blockers. Data are given as mean ± s.e.m., n being the number of

cells or slices when noted. Levels of significance were determined by Student’s

t-test (Statview).


ACKNOWLEDGMENTSWe thank C. Broberger, J. Fitzpatrick, U. Misgeld, M. Sarkisian, S. Titz andA. Williamson for valuable comments on the manuscript. We thank C.A. Greerfor providing us with a scanning confocal microscope. This work was supportedby a grant from the National Multiple Sclerosis Society (A.B.) and the USNational Institutes of Health (NS042189 and NS048256; A.B.).




1. Doetsch, F., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Cellular composition and three-dimensional organization of the subventricular germinal zone in the adult mammalianbrain. J. Neurosci. 17, 5046–5061 (1997).

2. Lois, C., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Chain migration of neuronalprecursors. Science 271, 978–981 (1996).

3. Luskin, M.B., Zigova, T., Soteres, B.J. & Stewart, R.R. Neuronal progenitor cells derivedfrom the anterior subventricular zone of the neonatal rat forebrain continue to proliferatein vitro and express a neuronal phenotype. Mol. Cell. Neurosci. 8, 351–366 (1997).

4. Belluzzi, O., Benedusi, M., Ackman, J. & LoTurco, J.J. Electrophysiological differentia-tion of new neurons in the olfactory bulb. J. Neurosci. 23, 10411–10418 (2003).

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

6. Doetsch, F., Caille, I., Lim, D.A., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Subven-tricular zone astrocytes are neural stem cells in the adult mammalian brain. Cell 97,703–716 (1999).

7. Capela, A. & Temple, S. LeX/ssea-1 is expressed by adult mouse CNS stem cells,identifying them as nonependymal. Neuron 35, 865–875 (2002).

8. Garcia, A.D., Doan, N.B., Imura, T., Bush, T.G. & Sofroniew, M.V. GFAP-expressingprogenitors are the principal source of constitutive neurogenesis in adult mouseforebrain. Nat. Neurosci. 7, 1233–1241 (2004).

9. Doetsch, F., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Regeneration of a germinallayer in the adult mammalian brain. Proc. Natl. Acad. Sci. USA 96, 11619–11624(1999).

10. Wang, D.D., Krueger, D.D. & Bordey, A. GABA depolarizes neuronal progenitors of thepostnatal subventricular zone via GABAA receptor activation. J. Physiol. (Lond.) 550,785–800 (2003).

11. Bolteus, A.J. & Bordey, A. GABA release and uptake regulate neuronal precursormigration in the postnatal subventricular zone. J. Neurosci. 24, 7623–7631 (2004).

12. Fueshko, S.M., Key, S. & Wray, S. GABA inhibits migration of luteinizing hormone-releasing hormone neurons in embryonic olfactory explants. J. Neurosci. 18, 2560–2569 (1998).

13. Haydar, T.F., Wang, F., Schwartz, M.L. & Rakic, P. Differential modulation of proliferationin the neocortical ventricular and subventricular zones. J. Neurosci. 20, 5764–5774(2000).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

14. LoTurco, J.J., Owens, D.F., Heath, M.J., Davis, M.B. & Kriegstein, A.R. GABA andglutamate depolarize cortical progenitor cells and inhibit DNA synthesis. Neuron 15,1287–1298 (1995).

15. Barker, J.L. et al. GABAergic cells and signals in CNS development. Perspect. Dev.Neurobiol. 5, 305–322 (1998).

16. Owens, D.F. & Kriegstein, A.R. Is there more to GABA than synaptic inhibition?Nat. Rev.Neurosci. 3, 715–727 (2002).

17. Nguyen, L. et al. Neurotransmitters as early signals for central nervous system devel-opment. Cell Tissue Res. 305, 187–202 (2001).

18. Nguyen, L. et al. Autocrine/paracrine activation of the GABA(A) receptor inhibits theproliferation of neurogenic polysialylated neural cell adhesion molecule-positive (PSA-NCAM+) precursor cells from postnatal striatum. J. Neurosci. 23, 3278–3294 (2003).

19. Demarque, M. et al. Paracrine intercellular communication by a Ca2+- and SNARE-independent release of GABA and glutamate prior to synapse formation. Neuron 36,1051–1061 (2002).

20. Verderio, C. et al. SNAP-25 modulation of calcium dynamics underlies differences inGABAergic and glutamatergic responsiveness to depolarization. Neuron 41, 599–610(2004).

21. Zhuo, L. et al. Live astrocytes visualized by green fluorescent protein in transgenic mice.Dev. Biol. 187, 36–42 (1997).

22. Bordey, A. & Sontheimer, H. Ion channel expression by astrocytes in situ: comparison ofdifferent CNS regions. Glia 30, 27–38 (2000).

23. Noctor, S.C. et al. Dividing precursor cells of the embryonic cortical ventricularzone have morphological and molecular characteristics of radial glia. J. Neurosci. 22,3161–3173 (2002).

24. Wang, D.D., Krueger, D.D. & Bordey, A. Biophysical properties and ionic signature ofneuronal progenitors of the postnatal subventricular zone in situ. J. Neurophysiol. 90,2291–2302 (2003).

25. Hockfield, S. & McKay, R.D. Identification of major cell classes in the developingmammalian nervous system. J. Neurosci. 5, 3310–3328 (1985).

26. Harks, E.G. et al. Fenamates: a novel class of reversible gap junction blockers.J. Pharmacol. Exp. Ther. 298, 1033–1041 (2001).

27. Smith, A.J., Oxley, B., Malpas, S., Pillai, G.V. & Simpson, P.B. Compounds exhibitingselective efficacy for different beta subunits of human recombinant gamma-amino-butyric acid A receptors. J. Pharmacol. Exp. Ther. 311, 601–609 (2004).

28. Sudhof, T.C. alpha-Latrotoxin and its receptors: neurexins and CIRL/latrophilins. Annu.Rev. Neurosci. 24, 933–962 (2001).

29. Hilfiker, S. et al. Synapsins as regulators of neurotransmitter release. Phil. Trans. R. Soc.Lond. B 354, 269–279 (1999).

30. Borden, L.A. GABA transporter heterogeneity: pharmacology and cellular localization.Neurochem. Int. 29, 335–356 (1996).

31. Lopez-Corcuera, B., Liu, Q.R., Mandiyan, S., Nelson, H. & Nelson, N. Expression of amouse brain cDNA encoding novel gamma- aminobutyric acid transporter. J. Biol. Chem.267, 17491–17493 (1992).

32. Levison, S.W. & Goldman, J.E. Multipotential and lineage restricted precursorscoexist in the mammalian perinatal subventricular zone. J. Neurosci. Res. 48, 83–94(1997).

33. Levison, S.W., Chuang, C., Abramson, B.J. & Goldman, J.E. The migrational patternsand developmental fates of glial precursors in the rat subventricular zone are temporallyregulated. Development 119, 611–622 (1993).

34. De Marchis, S. et al. GABAergic phenotypic differentiation of a subpopulationof subventricular derived migrating progenitors. Eur. J. Neurosci. 20, 1307–1317(2004).

35. Stewart, R.R., Hoge, G.J., Zigova, T. & Luskin, M.B. Neural progenitor cells of theneonatal rat anterior subventricular zone express functional GABA(A) receptors.J. Neurobiol. 50, 305–322 (2002).

36. Cotrina, M.L. et al. Connexins regulate calcium signaling by controlling ATP release.Proc. Natl. Acad. Sci. USA 95, 15735–15740 (1998).

37. Ye, Z.C., Wyeth, M.S., Baltan-Tekkok, S. & Ransom, B.R. Functional hemichannels inastrocytes: a novel mechanism of glutamate release. J. Neurosci. 23, 3588–3596(2003).

38. Pieribone, V.A. et al. Expression of synapsin III in nerve terminals and neurogenic regionsof the adult brain. J. Comp. Neurol. 454, 105–114 (2002).

39. Kao, H.T. et al. A third member of the synapsin gene family. Proc. Natl. Acad. Sci. USA95, 4667–4672 (1998).

40. Tramontin, A.D., Garcia-Verdugo, J.M., Lim, D.A. & Alvarez-Buylla, A. Postnatal devel-opment of radial glia and the ventricular zone (VZ): a continuum of the neural stem cellcompartment. Cereb. Cortex 13, 580–587 (2003).

41. Levison, S.W. & Goldman, J.E. Both oligodendrocytes and astrocytes develop fromprogenitors in the subventricular zone of postnatal rat forebrain. Neuron 10, 201–212(1993).

42. Lois, C. & Alvarez-Buylla, A. Long-distance neuronal migration in the adult mammalianbrain. Science 264, 1145–1148 (1994).

43. Luskin, M.B. & Boone, M.S. Rate and pattern of migration of lineally-related olfactorybulb interneurons generated postnatally in the subventricular zone of the rat. Chem.Senses 19, 695–714 (1994).

44. Edwards, F.A., Konnerth, A., Sakmann, B. & Takahashi, T. A thin slice preparation forpatch clamp recordings from neurones of the mammalian central nervous system.Pflugers Arch. 414, 600–612 (1989).

45. Haydar, T.F., Bambrick, L.L., Krueger, B.K. & Rakic, P. Organotypic slice cultures foranalysis of proliferation, cell death, and migration in the embryonic neocortex. BrainRes. Brain Res. Protoc. 4, 425–437 (1999).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Heterogeneity in synaptic transmission along aDrosophila larval motor axon

Giovanna Guerrero1, Dierk F Rieff 2, Gautam Agarwal3, Robin W Ball3, Alexander Borst2,Corey S Goodman1,3,5 & Ehud Y Isacoff1,3,4

At the Drosophila melanogaster larval neuromuscular junction (NMJ), a motor neuron releases glutamate from 30–100 boutons

onto the muscle it innervates. How transmission strength is distributed among the boutons of the NMJ is unknown. To address

this, we created synapcam, a version of the Ca21 reporter Cameleon. Synapcam localizes to the postsynaptic terminal and

selectively reports Ca21 influx through glutamate receptors (GluRs) with single-impulse and single-bouton resolution. GluR-based

Ca21 signals were uniform within a given connection (that is, a given bouton/postsynaptic terminal pair) but differed considerably

among connections of an NMJ. A steep gradient of transmission strength was observed along axonal branches, from weak proximal

connections to strong distal ones. Presynaptic imaging showed a matching axonal gradient, with higher Ca21 influx and exocytosis

at distal boutons. The results suggest that transmission strength is mainly determined presynaptically at the level of individual

boutons, possibly by one or more factors existing in a gradient.

Neurons form synaptic connections with one, several or hundreds ofpostsynaptic cells, and an individual neuron may make single ormultiple connections with a postsynaptic partner. These connectionscan change in number, strength and properties of short and long-termplasticity, both during development and as a consequence of experiencein the mature nervous system1–4. Classically, transmission has beenmeasured electrophysiologically, enabling an assessment of the overallstrength of transmission between a pair of cells, but without knowledgeof the number of connections between them. For simplicity, it hasusually been assumed that all connections between two cells sharesimilar properties. With advances in imaging, it has become feasible toselectively measure transmission at individual connections, either byfollowing the presynaptic release of FM dyes, amphipathic moleculeswhose fluorescence increases on membrane binding, during transmit-ter exocytosis or by monitoring the rise in postsynaptic Ca2+ due toinflux through transmitter-gated receptors, voltage-gated Ca2+ chan-nels or transmitter-triggered release from intracellular stores5–8. It hasalso become possible to test whether or not synaptic connectionsbetween a presynaptic cell and a postsynaptic partner have equalstrength. This question has not yet been extensively addressed, butelegant evidence has been obtained that Ca2+ dynamics, release prob-ability, and short-term plasticity may vary across the different connec-tions from one presynaptic cell8–10.

Here we investigate how transmission is distributed among themultiple connections between a presynaptic neuron and its musclepartner in the developing NMJ of the Drosophila larva, a synapse that

can be readily accessed for electrophysiology and easily imaged in asemi-intact preparation. The Drosophila larval NMJ shares importantstructural and molecular properties with mammalian CNS synapses. Itis glutamatergic, with non-NMDA-type ionotropic glutamate recep-tors, and both the presynaptic active zones and postsynaptic specializa-tions are organized through PDZ interactions in a manner similar tomammalian synapses11,12. In addition, as in certain CNS synapses suchas hippocampal CA1 inputs13, the NMJ involves multiple neuronalconnections onto the postsynaptic muscle. The NMJ also exhibitsstructural and functional plasticity. During larval development, musclesize increases more than 100-fold, causing a decrease in input resis-tance. To effectively depolarize and contract the muscle, synapticcurrents must also increase as the larva grows. Two general mechanismsupregulate synaptic currents during development: one that regulatespresynaptic structure and another that regulates transmission strength.During larval development, the degree of axonal branching as well asthe number of boutons and active zones increase14,15. Nascent boutonsemerge either de novo or by budding from pre-existing boutons andcome equipped with vesicles and active zones16. A muscle-secretedbone morphogenetic protein (BMP) is essential for this developmentalgrowth17, and its presynaptic receptor, Wishful thinking (Wit)18,19, hasbeen implicated both in the structural growth of the NMJ as well as inthe retrograde signaling that strengthens synaptic transmission20. It isnot known how synaptic strength is distributed at this multisynapticconnection or how the mechanisms that regulate its growth andtransmission may help to establish and maintain this distribution.


1Department of Molecular and Cell Biology, 279 Life Sciences Addition, University of California Berkeley, Berkeley, California 94720, USA. 2Department of Systems andComputational Neurobiology, Max-Planck-Institute of Neurobiology, Am Klopfersptiz 18 A, 82152 Martinsried, Germany. 3Helen Wills Neuroscience Institute, 279 LifeSciences Addition, University of California Berkeley, Berkeley, California 94720, USA. 4Physical Bioscience and Material Science Divisions, Lawrence Berkeley NationalLaboratory, Berkeley, California 94720, USA. 5Present address: Renovis, Inc., Two Corporate Drive, South San Francisco, California 94080, USA. Correspondence should beaddressed to E.Y.I. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

To clarify this, we compared transmissionfrom different connections of an individualneuron in the Drosophila NMJ using a newoptical approach. A genetically encoded fluor-escent Ca2+ reporter was used to detect post-synaptic Ca2+ flux through GluRs, which inDrosophila are permeable to Ca2+ (ref. 21).Versions of the fluorescence resonance energytransfer (FRET)-based reporter Cameleon22,23

were targeted to postsynaptic sites by means of fusion to the single-passtransmembrane domain of CD8 as well as to the C-terminal PDZinteraction domain of the Shaker K+ channel16. These genetic chimeras(synapcams) were localized to the muscle’s subsynaptic reticulum (SSR;that is, postsynaptic membrane folds that envelop presynaptic bou-tons), through the interaction of the Shaker C terminus with the PDZprotein Discs large (Dlg)11,12. This targeting, combined with the lowerCa2+ affinity of a mutated version of the fluorescent reporter, allowedsimultaneous monitoring of synaptic transmission across a populationof connections in response to single action potentials. We found thattransmission strength varies among connections in a gradient along thelength of axonal branches, with distal connections making strongerfunctional associations. We consider mechanisms that may be respon-sible for generating this gradient.

RESULTS

Synapcam design: a postsynaptically targeted cameleon

We measured synaptic activity at the Drosophila larval NMJ electricallyusing voltage-clamp and optically with genetically encoded Ca2+

sensors whose expression was driven in muscles by the myosin heavychain promoter (MHC)24. Various versions of the FRET-based Ca2+

sensor Cameleon22,23 were used to construct reporters, which we calledsynapcams, that were localized to the postsynaptic cell membrane. Thereporters were targeted to the muscle plasma membrane by theheterologous transmembrane protein CD8 and to the SSR by thePDZ interaction domain of the Shaker C terminus (Fig. 1a), as donepreviously16 for the targeting of a green fluorescent protein (GFP).The attachment of CD8 was necessary for efficient targeting. Fusionto the Shaker C terminus without CD8 produced weaker accumula-tion at postsynaptic sites (data not shown). Synapcams, however,were mostly localized at postsynaptic sites that surround presynapticboutons (that is, postsynapse), with lower levels in the nonsynapticmuscle membrane (Fig. 1b). These reporters thus permittedvital imaging of the subsynaptic structure. Muscle 6, the muscleused in these experiments, is innervated by type Ib boutons (3–8mm) from motor neuron RP3, and type Is boutons from motorneuron 6/7b (1–3 mm)15,25. Synapcam fluorescence was greaterat type Ib postsynapses, presumably because these are enveloped with

3.53.02.52.01.51.00.50.0mEJC amplitude (nA)

0

100

200

300

400

Num

ber

of e

vent

s

100 ms10 ms

2 nA

synapcam3.1

Wild type

Wild

type

Wild

type

50 n

Asc3.1

sc3.1

sc3.1sc3.1sc3.1 Wild typeWild type

Sh C′

Cameleon

CD8

Cytosol

Extracell.

Synapcam 2.1 Synapcam 3.1 Synapcam nulla

b

g

e

h

f

c d

Figure 1 Synapcam expression has no affect on

NMJ development or physiology. (a) Synapcams

are cameleons with CD8 at the N terminus and

the PDZ interaction domain of the Shaker K+

channel at the C terminus. Three versions of

cameleon, differing in their sensitivity to Ca2+

ions (red circles), were used: synapcam2.1, with

all four Ca2+ binding sites intact, synapcam3.1,with one site mutated to reduce Ca2+ affinity and

synapcamNull, with all four sites mutated to serve

as a Ca2+-insensitive control. (b) When expressed

under the control of the MHC promoter,

synapcam3.1 (shown in red, indicating YFP

fluorescence) localized to muscle sites underlying

presynaptic terminals of type Ib boutons (shown

in green, stained with anti-Nc82, an active zone

marker). (c–e) Flies expressing synapcam3.1

(sc3.1) showed no observable differences in the

levels or localization of pre- or postsynaptic

markers when compared to control larvae

(wild type). Glutamate receptor subunits DGluRIIA

and DGluRIIB (c,d), Dlg (e) and Syt (f) were not

perturbed by expression of synapcam3.1.

All images shows muscles 6/7, except for

f, which shows muscle 8. Scale bar: 10 mm.

(g,h) Physiological parameters were also

unaffected. Neither EJCs (g, n ¼ 10 NMJs) northe amplitude distribution of spontaneous

miniature quantal events (mEJCs, h) were

affected by expression of synapcam3.1 or the

experimental conditions (2 mM thapsigargin and

500 mM ryanodine) that prevented muscle con-

traction. The histogram in h depicts six NMJs and

1,953 events for w1118 (black), and nine NMJs

and 2,191 events from synapcam3.1 (gray). The

holding potential for g and h was �80 mV.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

more extensive SSR14,16,26. We thus focused our analysis on typeIb postsynapses.

Expression of synapcams did not affect muscle viability, nor did itaffect NMJ morphology. In addition, the localization of synapticproteins such as GluRs, Dlg or synaptotagmin (Syt) (Fig. 1c–f) aswell as physiological parameters such as the resting membrane poten-tial and the amplitudes of spontaneous and evoked junctional poten-tials (EJPs) and currents (EJCs) in synapcam+/+ larvae wereindistinguishable from wild-type larvae (Fig. 1g,h and SupplementaryTable 1). Thus, expression of synapcams did not interfere with synaptictransmission or perturb the development of the NMJ.

Synapcam reports on glutamate neurotransmission

Cameleon is a ratiometric reporter that makes use of fluorescenceresonance energy transfer (FRET)22. When the calmodulin domain of

cameleon binds Ca2+, the protein undergoes a conformational changethat increases the efficiency of FRET between the cyan fluorescentprotein (CFP) and yellow fluorescent protein (YFP) fluorophores,producing a decrease in CFP emission and an increase in YFP emission.Single stimulation of the motor axon resulted in reciprocal changes inthe CFP and YFP intensities of postsynaptic areas expressing synapcam(Fig. 2a,b) indicating an increase in FRET efficiency. Nonsynapticmuscle regions of synapcam+/+ larvae (Fig. 2a,b) and muscle fibersexpressing cytosolic Cam2.1 or CD8-Cam2.1 showed few or nofluorescence changes in response to axon stimulation (data not shown).

We created two different synapcam constructs and compared theiroptical responses. synapcam2.1 retains all four Ca2+ binding sites ofcalmodulin, and synapcam3.1 encodes a mutation (E104Q) thatabolishes one binding site in the N-terminal lobe22. We found thatboth reporters showed significant changes in FRET (DFRET) at

Figure 3 Synapcam reveals transmission

heterogeneity at the Drosophila NMJ. (a) Muscle

sites postsynaptic to individual boutons were

analyzed for Ca2+-dependent FRET changes in

response to a single motor axon stimulus. Each

color represents a different postsynaptic region

highlighted in the CCD image. The colors in the

FRET trace at right correspond to the colors in the

image at left. DFRET values for each region are

shown. A 4.6-fold difference of DFRET

magnitudes was observed for the NMJ displayed,

with a mean difference of 5.3-fold for all

experiments. Scale bar: 10 mm. (b) In addition to

a small decrease in DFRET as a result of

depression, small fluctuations (arrows) in the

magnitude of DFRET were observed within singlepostsynapses upon repeated stimulation

(0.25 Hz). These fluctuations however were not

observed in the average response for the entire

postsynapse (mean FRET trace, black).

Fluctuations in DFRET were independent of the

performance of other postsynapses despite

physical proximity. For example, at two pairs of

postsynapses (green and blue, or pink and yellow), FRET changes showed different fluctuation behavior, regardless of proximity. Therefore, the DFRET for a

single postsynapse is not influenced by FRET changes at other postsynapse. (c) Comparison of FRET response for postsynapse pairs against the distance

between pairs shows no correlation (156 trials, 61 boutons, six NMJs, r ¼ 0.009). Distance is the pythagorean distance between postsynapse centers.

3.532.521.510.50–0.1

00.10.20.30.40.50.60.70.80.9

Cor

rela

tion

in F

RE

T tr

ace

Distance (a.u.)

200 ms

0.02

YF

P/C

FP

100

nA

150

nA

Mean FRET

Individual boutons

1000 ms

0.05

Y/C

0.1250.1180.0990.0610.0600.0580.0470.0380.0310.027

a

c

b

100 ms200 ms 0.05

YF

P/C

FP

0.05

YF

P/C

FP

CFP

YFP

100 ms

2 a.

u.2

a.u.

100

nA

100

nA

synapcam3.1

synapcam2.1

CFP YFP

a c db

Figure 2 Synapcams report Ca2+ flux through GluRs as an increase in FRET. (a) An NMJ at muscle 6 during nerve stimulation. Synaptic and non-synaptic

(boxes) areas expressing synapcam3.1 were selected for analysis (in blue for CFP and yellow for YFP). Scale bar: 10 mm. (b) The total fluorescence of synaptic

regions showed reciprocal CFP (black traces at bottom) and YFP (gray) intensity changes following stimulation of the motor axon (EJC, black trace at top).Nonsynaptic areas (dotted lines) did not show fluorescent changes. Fluorescence levels for nonsynaptic areas were adjusted to synaptic levels for display

purposes. a.u., arbitrary units. (c) Fluorescence changes were indicative of an increase in FRET between CFP and YFP upon synaptic transmission. Single

stimuli to the motor axon evoked large FRET (YFP/CFP) changes measured from the entire synaptic area for both synapcam3.1 (lower black trace, 16 NMJs)

and synapcam2.1 (gray, 13 NMJs). The synapcam2.1 DFRET was on average 18% greater than the synapcam3.1 response. FRET changes had a rapid onset

(peak intensity reached after o200 ms) and a gradual offset, which was fit with a first-order exponential, with synapcam2.1 showing a slower decay lasting up

to 2 s. (d) FRET increases (black, single response for the synapse in a) were induced by Ca2+ transients, dependent on GluR activity, and synapcams did not

report voltage-dependent Ca2+ influx. The Ca2+-insensitive synapcamNull showed no change in FRET (green, six NMJs). FRET changes were abolished upon

application of desensitizing concentrations of glutamate (1.5mM; red, three NMJs). A voltage step from –80 to 0 mV (gray bar) did not elicit a change in FRET

(gray trace, three NMJs).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

postsynaptic regions (Fig. 2c). DFRET beganincreasing B20 ms after the postsynapticcurrent and reached a peak within 120 msfor synapcam3.1 and within 200 ms for synap-cam2.1. The mean peak response for synap-cam2.1 was 18.18% greater than that forsynapcam3.1 (0.385 ± 0.138 and 0.315 ±0.120, respectively), and the decay of theresponse was slower (t ¼ 192.63 ms ± 5.95and t ¼ 599.65 ± 11.04, respectively). Theseresults were consistent with the greater Ca2+

affinity of Cameleon2.1 (refs. 22,23). Becausethe rise phases overlapped for synapcam2.1and 3.1, the earlier peaking of synapcam3.1 islikely attributable to a faster recovery rate.Rapid kinetics and a lower Ca2+affinity madesynapcam3.1 a more attractive reporter forsubsequent experiments.

Although results strongly suggested thatsynapcams functioned in Drosophila muscleas synaptic Ca2+ sensors, it remained possiblethat the GFP variants were in part respondingto changes in pH or halides. Synapses expres-sing synapcamNull, a reporter with all fourCa2+ binding sites abolished by mutation(Fig. 1a), showed no fluorescence change inresponse to neurotransmission (Fig. 2d, greentrace). This result demonstrated that DFRETwas indeed reporting an increase in postsyn-aptic Ca2+ resulting from neurotransmission.

To focus exclusively on Ca2+ influx throughGluRs, Ca2+ from internal stores was depletedwith thapsigargin27 while residual Ca2+ effluxfrom the stores was blocked with ryanodine28.In addition, the interrogated muscle was vol-tage clamped to prevent activation of voltage-gated Ca2+ channels. Under these conditions,desensitization of GluRs by bath-applied glutamate abolished both theFRET response and the EJC (current 94 ± 5%, fluorescence 88 ± 9%decrease, three NMJs, three traces each; Fig. 2d, red trace), demon-strating that GluR-derived currents were necessary for the opticalresponse of synapcam. Depolarizing voltage jumps designed to openvoltage-gated Ca2+ channels did not induce DFRET (Fig. 2d, graytrace), indicating that voltage-gated Ca2+ channels did not contributedetectable synaptic Ca2+ under our recording conditions. Thus, synap-cams provided a selective optical report of postsynaptic currentsgenerated entirely from activated GluRs.

Heterogeneous transmission among connections of the NMJ

To assess the contribution of individual connections to overall trans-mission at the NMJ, discrete postsynaptic areas under presynapticboutons were individually analyzed (Fig. 3a). Each postsynapsemeasured (801 in 65 NMJs from 62 larvae) showed a DFRET inresponse to axon stimulation (that is, no silent connections wereobserved, consistent with previous results from focal electrophysiolog-ical recordings29). However, postsynaptic DFRETs sometimes fluctu-ated in amplitude during repetitive presynaptic stimulation (Fig. 3b).These fluctuations were independent of fluctuations occurring at otherpostsynapses, including close neighbors (Fig. 3b, compare greenpostsynapse versus blue postsynapse). Similarly, there was no pairwisecorrelation between the DFRET response of a postsynapse and the

distance separating it from another postsynapse (Fig. 3c). Theseobservations indicated that Ca2+ influx at one postsynapse did notactivate sensor readout at a nearby postsynapse, thus providing singlepostsynapse (that is, single-bouton) resolution.

The magnitude of DFRET differed considerably among postsynapsesof a single axon (Fig. 3a,b). Of 65 NMJs tested, the amplitudes differedtwo- to 14-fold among postsynapses of the same NMJ. This observedheterogeneity in DFRET might stem from biological variation, such aspresynaptic differences in transmitter release or differences in post-synaptic sensitivity. Alternatively, it could arise from differencesbetween postsynapses in synapcam numbers or readout. However,there was no correlation between relative reporter quantity and DFRET(Fig. 4c) and moreover, DFRET responses peaked sharply, arguingagainst saturation of synapcam (Fig. 3). This possibility was testeddirectly by comparing the DFRET response after a single stimulusagainst the response to a pair of stimuli given in rapid succession. Ifsmall FRET changes were due to saturated postsynapses, stimulus pairswould not be expected to have an increased FRET response.

Stimuli were separated by 10 ms and produced two separate EJCs.Although the current amplitude of the second pulse was depressed byapproximately 28% (first pulse, –363.8 ± 18.5 nA; second pulse, �261.9± 8.4 nA), as expected for the Ca2+ concentration used, the stimuluspair produced DFRET amplitudes larger than a single stimulus, owingto summation of the synapcam responses (Fig. 4a). Taken together, 41

5004003002001000220

0.00

0.02

0.04

0.06

0.08

0.10

Area

∆FR

ET

∆FR

ET

∆FR

ET

two

puls

es (

YF

P/C

FP

)

2001801601401201008060

0.00

rYFP

0.02

0.04

0.06

0.08

0.10

0.160.120.080.040.000.00

0.04

0.08

0.12

0.16

∆FRET one pulse (YFP/CFP)

100 ms0.02

5 Y

FP

/CF

P

0.05

YF

P/C

FP

1 s

6 7

50 n

A

54

1 2 3

Average

6

5

3

4

1

2

7a

b c d

Figure 4 Synapcam3.1 is not saturated by single stimuli to the motor axon. (a) Example NMJ that was

subjected to trials of single stimulation (FRET responses are black traces) and trials where two stimuli

were separated by 10 ms (gray traces). Although currents were depressed after the second pulse, the

magnitude of DFRET was higher for dual stimulation recordings in all boutons imaged as a result of Ca2+

summation in the synapcam response. Scale bar: 10 mm. Numbers identify single postsynapses.

Numbers in image correspond with numbers above traces. (b) This increase was observed in all boutons

of five NMJs tested. When compared with the response after a single pulse, the pooled responses for two

stimuli showed on average an 83.1 ± 5.3% increase, and the responses were fit by linear regression

(r ¼ 0.897, P o 0.0001), indicating a lack of saturation, even for boutons that responded strongly to a

single stimulus. DFRET numbers in b are the mean ± s.e.m. of three single stimulation and three doublestimulation trials for each NMJ. (c) No correlation was found between the level of reporter expression, as

assayed by the average resting levels of sensitized YFP (rYFP) and the FRET response of the postsynaptic

terminal (black line: linear fit, r ¼ �10.047, P ¼ 0.25). (d) No correlation was observed between the

size of a bouton and the FRET response (black line: linear fit, r ¼ 0.032, P ¼ 0.33). For c and d,

n ¼ 625 boutons and 45 NMJs.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

boutons from five NMJs showed linear increases in the FRET responseto the stimulus pair as compared with the single stimulus (Fig. 4b),indicating that for single pulses at low frequency, as used throughoutour analysis, the reporter is not saturated at any boutons along theaxon, whether they are proximal or distal or their responses large orsmall. The ratio of DFRET from a pulse pair relative to the response forsingle pulse (mean of 1.84 ± 0.05) showed a slight negative dependenceon the DFRET after a single stimulus (Supplementary Fig. 1). Thisdependence could be due to presynaptic depression, as the EJC showeda similar trend (Supplementary Fig. 1). Taken together, the resultsindicate that heterogeneity in DFRET stems from biological variationin neurotransmission.

Active zone or GluR numbers do not determine heterogeneity

Because synapcam responds to Ca2+ influx through GluRs, the in-homogeneity in synaptic transmission might arise from differencesamong the postsynaptic sites in the number of GluRs. Alternatively,Ca2+ influx might also be affected by GluR subunit composition. FiveGluRs are expressed in the muscles of Drosophila larvae30–32. DGluR-IIA and DGluRIIB compete with each other for assembly with the othersubunits and for membrane expression31,32. In addition, DGluRIIBdesensitizes more rapidly than DGluRIIA, leading to reduced macro-scopic currents in muscles overexpressing DGluRIIB33. Differencesbetween postsynapses in the numbers of DGluRIIB subunits andDGluRIIA subunits might thus affect Ca2+ influx. To test thesepossibilities, NMJs were fixed immediately after imaging synapcamand stained with antibodies against DGluRIIA and DGluRIIB. Theamplitude of the DFRET signals showed very weak correlation or no

correlation to DGluRIIA or DGluRIIB staining or to the ratioof the number of DGluRIIA to DGluRIIB subunits (SupplementaryFig. 2). On average, postsynapses that differed by no more than 50% innumber of GluRs could differ in DFRET amplitude by as much as1000% (Supplementary Fig. 2). Thus, inhomogeneity in synapcamresponses among postsynapses of an NMJ could not be attributed todifferences in the number or composition of GluRs.

Transmission heterogeneity might also stem from presynaptic dif-ferences among boutons. Type Ib boutons contain multiple activezones14, and larger boutons tend to contain more active zones. Wetested if differences in the number of active zones accounted for theobserved transmission inhomogeneity. There was no correlationbetween the amplitude of DFRET response and the size of thepresynaptic bouton (Fig. 4d). Moreover, immunostaining with theantibody Nc82, which stains presynaptic active zones opposite post-synaptic patches of GluRs (ref. 34; Fig. 1b), showed that neither theimmunofluorescence intensity nor the density of Nc82 punctaper bouton correlated with the DFRET from individual boutons(Supplementary Fig. 3).

A gradient of transmission strength along axonal branches

To describe the spatial distribution of synaptic transmission across theconnections of the motor axon, continuous imaging (Figs. 2–4) wasreplaced by episodic imaging (Fig. 5a; see Methods) to minimizephotobleaching and allow the acquisition of data for hundreds ofevents. FRET images of the synaptic region (Fig. 5b), averaged fromresponses to 50–300 stimuli, provided high signal-to-noise DFRETmeasurements from type Ib boutons and also enabled optical detection

rFRETYFP

1.4

1.5

1.6

1.7

0.01

0.02

0.03

0.04

0.05

1.3

1.4

1.5

100908070605040302010Stimulus number

0

0.02

0.04

0.06

0.08

0.1

MeanEJC

8

6

4

2

∆FRET Partitioned ∆FRET

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.1

PostPre

Stimulus

1

2

3

4

5

6

7

8

a

b

c

d

e

∆FRET0

0.02

0.04

0.06Figure 5 Prolonged imaging shows the distribution of transmission strength of

an NMJ. (a) Imaging conditions were optimized to allow extended imaging of the

NMJ. One frame was acquired 200 ms before, and another 100 ms after nerve

stimulation (0.125 Hz). Each frame represents 50 ms of exposure. The protocol

was repeated for at least 30 trials, but more typically for 100–200 trials.

(b) Image analysis produced mean DFRET scans of the entire synaptic region(left) and scans where the NMJ was partitioned into postsynaptic regions of

interest with mean DFRET values for each postsynapse (right). (c) Array of 100

DFRET responses for the eight postsynapses numbered in b. The y-axis reflects

postsynapse identity. The last two rows (separated by a blue bar) are mean DFRET for all postsynapses and EJC responses for each stimulus. (d,e) YFP, mean

rFRET and mean DFRET images of two different NMJs. (d) FRET changes were sometimes observed for type Is postsynapses (short arrows), even though

reporter localization at these sites was low (see YFP image). (e) Adjacent postsynapses of similar YFP and rFRET values frequently produced different FRET

changes (two examples, white arrowheads and arrows) indicating that the mean DFRET of a postsynapse was not determined by reporter expression or rFRET

values. The scale bar for all images is 8 mm.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

of transmission from type Is boutons (Fig. 5e). However, because thetype Is boutons were more difficult to identify in the YFP images, wedid not pursue them at this stage.

Pronounced differences were observed in the potency of post-synapses. Although there was some variation between individualstimuli, strong postsynapses tended to stay strong throughout thetrial (Fig. 5c). We never observed a drastic change in the transmissiondistribution across postsynapses. FRET images from type Ib boutonsshowed that adjacent postsynapses of similar size, reporter expression,and resting FRET (rFRET) could differ markedly in transmissionstrength (Fig. 5e), confirming results from single stimulationthat suggest Ca2+ entry is confined to the physical limits of a post-synapse and that reporter expression or rFRET did not determine apostsynapse’s DFRET.

Our analysis of 440 postsynapses from 91 axonal branches showedthat strong and weak synaptic connections were not randomly dis-tributed throughout the motor axon, but rather were organized in agradient along the length of axonal branches (Fig. 6). Horseradishperoxidase (HRP)-specific staining after imaging traced the axonbranches from the first point of muscle contact, to secondary branchpoints, to terminal boutons (Fig. 6a). Postsynapses at the ends ofbranches consistently showed the greatest DFRET responses(Fig. 6c,d,f–i), whereas postsynapses at the origins of branches werealways among the weakest. The DFRET gradient was not due todifferences in the amount of synapcam, because rFRET (Fig. 6b) didnot show consistent trends in intensity along a branch (Fig. 6e,j).DFRET decreased linearly with greater distance from the end of a

branch and with the connection position along an axonal branch(Fig. 6j,k). This distal-to-proximal gradient of transmission strength,from strong distal connections to weak proximal ones, was observed inboth simple and complex branching patterns (Fig. 6h). Terminalpostsynapses from different branches within the same NMJ alsodiffered significantly in terms of transmission potency (Fig. 6i). Thegradient in transmission strength was not dependent upon the relativelocation of the axon branch along the muscle surface. Transmission atterminal connections was always stronger regardless of location alongthe length of the muscle fiber. Moreover, the gradient was alsoindependent on branch orientation, whether the branch ran parallel,perpendicular to, or obliquely across the muscle fiber (Fig. 6h,i). Theseobservations suggested that the gradient in transmission strength wasnot a function of postsynaptic polarity but instead depended solely onthe location of a bouton relative to the axon branch.

Two exceptions were found which disrupted the commonlyseen gradient of transmission strength. One of these was at boutonsthat were budding-off from established boutons. Budding boutonsat the terminus of a branch had lower transmission strengththan mature boutons from which they budded (Fig. 6d, lowergreen bouton budding from yellow bouton). A second exceptionoccurred when an axon branch had a sharp kink. Boutons locatedon the proximal side of the bend had higher DFRET valuesthan boutons following it, as if they were terminal boutons(Fig. 6g). The dependence of transmission strength on fineaxon morphology supported the idea that the gradient is estab-lished presynaptically.

876543210140Bouton order from end

0.0

0.2

0.4

0.6

0.8

1.0

Nor

mal

ized

DF

RE

T

Partitioned ∆FRET

∆FR

ET

rFR

ET

rFRET ∆FRET

120100806040200–20Distance from end

0.0

0.2

0.4

0.6

0.8

1.0

Nor

mal

ized

FR

ET

152129

416076

90

90

* ** *

* **

0

0.01

0.02

0.03

0.04

0.05

6040200Distance from end (pixels)

6040200Distance from end (pixels)

0 *

***

*

*

*

0.01

0.02

0.03

0.04

0

0.02

0.04

0.06

0

0.02

0.04

0.06

0

5

10

15

0

5

10

15

× 10–3

× 10–3

1.1

1.2

1.3

1.4

1.5

1.5

1.4

1.3

1.2

1.60.03

0.02

0.01

0

a b ec

f g h i

j k

d

Figure 6 A proximal-distal gradient in transmission strength.

(a) HRP staining (green), overlaid on a synapcam3.1 YFP image

(red), confirms the origin and termini of two branches (black and gray

asterisks). (b–d) Mean rFRET (b), mean DFRET (c) and DFRET

partitioned and averaged within each synapse (d) for the NMJ in a.

(e) Scatter plots of the mean rFRET (top) or DFRET (bottom) of each

postsynapse against its distance from the end of a branch for the

NMJ depicted in a–d. Black and gray corresponds to color of

asterisks in a, and lines are fits to depict trend. (f–i) Additionalexamples of the transmission strength gradient along the length of

axonal branches. Branch ends are marked with asterisks, branch origins are marked with arrows, and muscle border is depicted by a dashed white line. Color

bars represent rFRET for b and DFRET for all other images. Scale bar for all images is 8 mm. (j) Pooled data of all postsynapses DFRET (gray) and rFRET

(black) values against the distance from the branch’s end show a stronger correlation for DFRET values. Linear fits, DFRET: r ¼ �0.649, P o 0.0001; rFRET:

r ¼ �0.461, P o 0.0001. (k) Postsynapses at the branch end (first column) give on average greater DFRET than the postsynapses that follow them (to the

right). Numbers within bars are number of postsynapses averaged, mean ± s.e.m., ** P o 0.001, * P o 0.005 independent t-test. For j and k, FRET values

were normalized to the highest value within each branch for n ¼ 440 postsynapses, 90 branches, 34 NMJs.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Presynaptic release is stronger at end boutons

We have documented a gradient in synaptic transmission that makesthe connections of distal boutons stronger than proximal ones alongthe length of the motor axon. These measurements were madepostsynaptically, but the lack of correlation to GluR levels and thedependence on axon morphology suggested that the locus of thegradient was presynaptic. To test this idea, presynaptic fluorescentreporters were used to measure Ca2+ influx and synaptic vesicle fusionat each bouton. Presynaptic Ca2+ was monitored with a version ofcameleon, Cam2.3 (ref. 35), expressed pan-neuronally. Motor axonswere subjected to 40-Hz trains of stimuli, a frequency that, in thispreparation, does not saturate the reporter36. Larger rises in Ca2+ wereobserved at distal boutons than at boutons 2–3 positions from the end(Fig. 7a,b), indicating greater Ca2+ influx and suggesting greatertransmitter release at the ends of axons. This finding led us to assayvesicle release directly. This was done in flies expressing supereclipticSynapto-pHluorin (SpH), a GFP-based sensor that reports vesiclefusion by exploiting differences in pH between the intravesicular andextracellular space37. A greater activity-dependent increase in fluores-cence, indicating greater exocytosis, was observed at distal boutonsthan at boutons 2–3 positions from the end (Fig. 7c). This provideddirect evidence for greater release from distal boutons. Thus, transmis-sion at the Drosophila NMJ seemed to be greater at the ends of axonalbranches, and this gradient seems to be due in part to greater Ca2+

influx and vesicle fusion presynaptically.

DISCUSSION

By targeting a genetically encoded Ca2+ sensor to postsynaptic sites,near Ca2+-permeant GluRs, we have monitored glutamatergic trans-mission optically with single-bouton resolution at the Drosophilalarval NMJ. While highly responsive to Ca2+ influx through GluRs,the sensor is relatively blind to influx through voltage-gated Ca2+

channels (Fig. 2), and in response to a presynaptic action potential,the DFRET at one postsynapse is independent of DFRET at an adjacentpostsynapse (Figs. 3 and 5). As a result, synapcam enables thesimultaneous examination of transmission at multiple synaptic con-nections in response to single presynaptic action potentials. Thisapproach is promising in that similar targeting of low-affinity opticalCa2+ sensors to specific Ca2+ channels such as NMDA and IP3receptors or to voltage-gated Ca2+ channels in mammalian neuronscould enable visualization in dendrites of the spatial spread of the Ca2+

fluctuations that trigger long-lasting changes in synaptic strength.Although all connections at the Drosophila NMJ participated in

neurotransmission, the strength of transmission was found to varygreatly from bouton to bouton. The position of a particular synaptic

connection with respect to the origin of a branch or sub-branchdetermines its transmission potency. Distal boutons are the strongest,with a progressive decline in transmission strength in boutons nearer tothe origin of the axon branch (Fig. 6). Although this type of organizedtransmission heterogeneity has never before been documented for theDrosophila NMJ, transmission gradients have been observed before atcrayfish, frog and mouse NMJs38–41. These studies used serial single-bouton or single-muscle fiber recordings to compare proximal anddistal connections. The targeting of a Ca2+ fluorescent reporter topostsynaptic sites made it possible for us to capture the same kind ofinformation in parallel in single snapshots. In those earlier NMJstudies, differences in release probability, the size of SSR folds, orfailure of action potential propagation were proposed to explain thegradients42,43. Action potential propagation failure cannot explain ourobservation of stronger transmission at distal connections in Droso-phila larvae. In addition, we do not observe a gradient in eitherpostsynapse size or quantity of synapcam along the synaptic connec-tions made by an axon (Fig. 4), arguing against systematic differencesin the SSR. Finally, the quantity of GluRs and the molecular identity ofthe GluR subunits at a postsynapse are not organized in a gradient(Supplementary Fig. 2). Notably, in addition to receptor quantity andsubunit composition, the phosphorylation state of GluRs has beenshown to affect quantal size at the NMJ44. Although we did notinvestigate the phosphorylation state of the GluRs at each bouton,the effect of phosphorylation on GluRs cannot quantitatively accountfor the gradient observed. At most phosphorylation decreases quantalsize by 60% (ref. 44). This cannot account for the 1,000% differencesthat we observe. In summary, we have not found a strong postsynapticcorrelate for the establishment of a transmission gradient. However, wedid find increased presynaptic Ca2+ and vesicle fusion at distal boutons

Proxim

al

Distal

Proxim

al

Distal

5

10

15

20

(67)

(32)

(19)

(52)

*

*

1086420

Time (s)

1086420

Time (s)

–5

0

5

10

15

20

–5

0

5

10

15

20

∆F/F

(%

)∆F

RE

T/F

RE

T (

%)

∆FRET/FRET

Proximal

Distal

Proximal

Distal

20

15

10

rYFP rFRET120100806040201201008060402012010080604020

180

160

140

120

100

80

60

40

20a

b

c

Figure 7 Presynaptic contribution to the gradient of transmission strength.

(a) Axons expressing cytoplasmic Cam2.3 were subjected to 2.2 s of 40-Hz

stimuli (dashed lines in b and c). Before stimulation, no axonal gradient was

detected either in the amount of Cam2.3 at the synapse (rYFP) or the resting

FRET (rFRET). However, during stimulation, boutons at the ends of axonal

branches showed higher DFRET responses than more proximal boutons.

(b) Average DFRET traces show higher presynaptic Ca2+ increase for end

boutons than for those 2–3 boutons away (DFRET/FRET ¼ 18.95 ± 0.88distal, 15.69 ± 0.73 proximal, P o 0.005, independent t-test). Data from

19 distal and 32 proximal boutons from19 axonal branches of seven NMJs.

(c) Vesicle fusion was examined in animals expressing SpH and mDsRed (to

aid in visualization of the axonal arbor) presynaptically. Distal boutons

showed bigger fluorescent changes upon 40-Hz stimulation indicative of

higher exocytosis (DF/F ¼ 17.97 ± 1.49 distal, 13.20 ± 0.91 proximal;

P o 0.005, independent t-test). Data from 52 axonal branches in 14 NMJs,

including 52 distal and 67 proximal boutons.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

(Fig. 7) indicating that a presynaptic mechanism that controls trans-mitter release has a major role in the establishment of the transmissiongradient. Although release probability was not measured directly in thisstudy, the number of active zones was not organized in a spatial gradientalong the boutons of an NMJ (Supplementary Fig. 3). Still, from earlierwork at the Drosophila and Crayfish NMJs, it seems that more efficienttransmitter release occurs at active zones that are associated withelectron dense specializations known as ‘T-bars’20,42,45,46. Hence, it isconceivable that more distal boutons have larger numbers of T-bars.Alternatively, differences in several other parameters that regulaterelease probability or the mode of release could have a role.

We find that the gradient in transmission strength does not dependon the relative position of a bouton on the muscle surface, suggesting itis determined presynaptically by the polarity of the axon. This idea issupported by our finding of greater Ca2+ influx and vesicle fusion atdistal boutons. In principle, this finding is consistent with either apositive regulator of transmission that is preferentially concentrated atthe ends of axon branches, or a negative regulator of transmission thatis delivered in limiting amounts and accumulates more in the firstboutons it reaches during anterograde transport (SupplementaryFig. 4). Because bouton addition occurs throughout the length of theaxon, not only at the ends of branches but also between previouslyestablished boutons16, a proximal-distal gradient in a bouton’s devel-opmental age is not likely to fully explain the gradient in transmissionstrength. Notably, a gradient has been observed in the arrangement ofthe microtubule associated protein 1B (MAP1B) along microtubules inthe presynaptic terminal. Immunofluorescent mapping of MAP1Bshows tight bundles of microtubules along axon branches with a higherincidence of microtubule ‘loops’ at branch points and branch ends, aswell as disarrayed microtubule patterns at the most distal boutons47,48.It is possible that as well as promoting the growth of the presynapticterminal, microtubules may also have a role transporting componentsneeded for synaptic transmission down the length of axon branches.Perhaps microtubule loops in axon kinks and at the end of branches, aswell as microtubule disarray at the most distal boutons, favor thedeposition of positive regulatory factors that increase transmissionstrength. It will be interesting in the future to examine the role ofmicrotubules in determining transmission strength and to identify anyregulatory factors existing in an axonal gradient.

METHODSDNA constructs and flies. Yellow cameleon (cam) 3.1 (gift from R.Y. Tsien,

University of California San Diego) was converted to cam2.1 by reverting the

Q104E mutation through site-directed mutagenesis (Quickchange, Stratagene).

cam-null was derived from cam3.1 by mutations E31Q, E67Q and E140Q,

which abolish the Ca2+ binding sites. To target cameleon variants to the synapse

(in other words, to create the synapcam construct), cam2.1, cam3.1 and cam-

null were amplified with SpeI primers (5¢-GG ACT AGT GCC GCC ACC ATG

GTG AGC-3¢ and 5¢-GG ACT AGT GCA GAA TTC CTT GTA CAG-3¢) and

inserted into the SpeI sites of the CD8-GFP-Sh construct16, exchanging GFP for

each cam variant. The genetic chimeras were placed under control of the MHC

promoter24. Transgenic flies were made using standard germline transforma-

tion by embryo injection. Single insertions on the X chromosome were

identified by orange eye color and confirmed by segregation analysis. Female

larvae were used in all experiments.

For presynaptic exocytosis and calcium imaging, respectively, UAS-SpH37

and UAS-Cam2.3 (ref. 36) were driven pan-neuronally with elavC155-Gal4.

UAS-monomeric-DsRed flies were a gift from G. Tavosanis, European Mole-

cular Biology Laboratory. The genotype of animals expressing SpH and

mDsRed was elavc155-Gal4+/�, UAS-SpH+/�, UAS-mDsRed+/�.

Immunohistochemistry and analysis. Larvae were fixed immediately after

physiological recordings, either with Bouin’s fixative for 5 min (DGluRIIA and

DGluRIIB) or with 4% formaldehyde for 30 min (Nc82, Syt, HRP). The

following primary antibodies were used: mouse anti-DGluRIIA (1:100)31, mouse

anti-Syt (1:5), mouse anti-Dlg (1:100), mouse anti-HRP (1:100; Sigma), rabbit

anti-DGluRIIB (1:2,500)31 and NC-82 (1:100, gift from E. Buchner, University

of Wurzburg). Alexa 647–conjugated goat anti-mouse and Cy3-conjugated goat

anti-rabbit secondary antibodies were used for double-labeling experiments

(Molecular Probes).

Metamorph (Universal Imaging) was used to calculate the intensities of YFP,

GluRs and Nc82 staining from confocal images. Because GluRs and Nc82

antibodies label in a punctal pattern, images were thresholded in Metamorph to

discard grey levels below minimal staining. Total intensity was calculated for

values above threshold within each postsynaptic terminal. The density of staining

was taken as the percentage of area above threshold for a postsynaptic terminal.

Electrophysiology. Two-electrode voltage clamp recordings were done with an

AxoClamp 2B amplifier (Axon Instruments) on Drosophila muscle 6 at

segments A2 or A3 of third instar larvae. Unless otherwise stated, recording

solution consisted of physiological saline HL3 (ref. 49) containing 1.5 mM

Ca2+, 20 mM Mg+2, 2 mM thapsigargin and 500 mM ryanodine. For presynaptic

imaging, 7 mM glutamate was added to prevent muscle contraction during

high-frequency stimulation. Recording electrodes contained 3 M KCl and had

resistances between 10–20 MO. Only muscles with a resting membrane

potential below �60 mV were chosen for study. For EJC studies, the nerve

was stimulated at 0.1 Hz and muscles clamped at �80 mV. During image

acquisition, muscles were held at �100 mV. The more negative holding

potential (�100 mV) improved cameleon signaling, presumably by increasing

the Ca2+ driving force. Data were filtered at 1 KHz and recorded using a

Digidata 1200A/B board and Clampex 8.0 software (Axon Instruments). mEJCs

were analyzed with MiniAnalysis software (Synaptosoft), and other electro-

physiological data were analyzed with Clampfit 8.0 (Axon Instruments).

Optical recording of FRET. Larvae were imaged with an ORCA-ER charge-

coupled device (CCD) camera (Hamamatsu) and an Olympus BX-50WI

microscope (Olympus) with a 75-W Xenon lamp and a 60� 0.9NA objective

(Olympus). Excitation was at 434 ± 10 nm using a 460-nm dichroic filter.

Emission wavelengths were separated with a dual-emission beam splitter

(Optical Insights) with a 510 nm dichroic filter and 480 ± 20 nm and 535 ±

20 nm emission filters. Images were collected with SimplePCI software

(Compix) at either 53 or 35 frames/s (8 � 8 or 4 � 4 binning, respectively).

For presynaptic vesicle fusion experiments, mDsRed was excited at 550 nm

with Q585LP and HQ620/60 filters, and SpH was excited at 470 nm with a

Q480LP and a HQ535/50 band pass. All filters and dichroics were from

Chroma Technology.

To obtain an accurate representation of amplitude and spatial distributions

of transmission across synapses, the initial continuous imaging (Figs. 2–4) was

substituted with episodic imaging, thereby minimizing photobleaching and

allowing more responses to be measured. A pair of CCD images was acquired

per stimulus, one before stimulation (baseline) and one 100 ms after stimula-

tion (around the peak of the synapcam response; Fig. 5a). Under these

conditions, the synapse could be imaged for up to 30 min while recording

hundreds of responses. Responses were sub-saturating at all boutons (Fig. 4a),

with a similar doubling of DFRET at distal (1.98 ± 0.15) and proximal (1.96 ±

0.8) postsynaptic terminals in response to a pulse pair.

Image analysis. Image analysis (Figs. 2–4) was performed with Bouton Project,

software that enables overlaying CFP and YFP data, defining regions of interest

and quantifying fluorescence change over time (D. Raymond, University of

California Berkeley, personal communication). For extended experiments

where two frames were acquired before and after stimulation (Figs. 5–7),

images were analyzed in Matlab 7.0 (Mathworks). Details about image analysis

are described in Supplementary Methods.


ACKNOWLEDGMENTSWe thank D. Raymond for developing the Bouton Project software; K. Zito forinitial cloning and transfection of synapcam3.1 and M.-M. Poo, R. Zucker andM. Neff for comments on the manuscript. This work was funded by a US


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

National Institutes of Health grant (E.Y.I. and C.S.G.), the Max-Planck-Society(D.F.R. and A.B.) and the Howard Hughes Medical Institute (G.G. and C.S.G.).




1. Hua, J.Y. & Smith, S.J. Neural activity and the dynamics of central nervous systemdevelopment. Nat. Neurosci. 7, 327–332 (2004).

2. Constantine-Paton, M. & Cline, H.T. LTP and activity-dependent synaptogenesis: themore alike they are, the more different they become. Curr. Opin. Neurobiol. 8, 139–148(1998).

3. Turrigiano, G.G. & Nelson, S.B. Hebb and homeostasis in neuronal plasticity. Curr. Opin.Neurobiol. 10, 358–364 (2000).

4. Liu, G. Presynaptic control of quantal size: kinetic mechanisms and implicationsfor synaptic transmission and plasticity. Curr. Opin. Neurobiol. 13, 324–331(2003).

5. Oertner, T.G. Functional imaging of single synapses in brain slices. Exp. Physiol. 87,733–736 (2002).

6. Fetcho, J.R. & O’Malley, D.M. Imaging neuronal networks in behaving animals. Curr.Opin. Neurobiol. 7, 832–838 (1997).

7. Murthy, V.N. Optical detection of synaptic vesicle exocytosis and endocytosis.Curr. Opin.Neurobiol. 9, 314–320 (1999).

8. Koester, H.J. & Johnston, D. Target cell-dependent normalization of transmitter releaseat neocortical synapses. Science 308, 863–866 (2005).

9. Murthy, V.N., Sejnowski, T.J. & Stevens, C.F. Heterogeneous release properties ofvisualized individual hippocampal synapses. Neuron 18, 599–612 (1997).

10. Koester, H.J. & Sakmann, B. Calcium dynamics associated with action potentials insingle nerve terminals of pyramidal cells in layer 2/3 of the young rat neocortex.J. Physiol. (Lond.) 529, 625–646 (2000).

11. Zito, K., Fetter, R.D., Goodman, C.S. & Isacoff, E.Y. Synaptic clustering of Fasciclin IIand Shaker: Essential targeting sequences and role of Dlg. Neuron 19, 1007–1016(1997).

12. Tejedor, F.J. et al. Essential role for dlg in synaptic clustering of shaker K+ channels invivo. J. Neurosci. 17, 152–159 (1997).

13. Sorra, K.E. & Harris, K.M. Occurrence and three-dimensional structure of multiplesynapses between individual radiatum axons and their target pyramidal cells inhippocampal area CA1. J. Neurosci. 13, 3736–3748 (1993).

14. Atwood, H.L., Govind, C.K. & Wu, C.-F. Differential ultrastructure of synaptic terminalson ventral longitudinal abdominal muscles inDrosophila larvae. J. Neurobiol. 24, 1008–1024 (1993).

15. Schuster, C.M., Davis, G.W., Fetter, R.D. & Goodman, C.S. Genetic dissection ofstructural and functional components of synaptic plasticity. I. Fasciclin II controlssynaptic stabilization and growth. Neuron 17, 641–654 (1996).

16. Zito, K., Parnas, D., Fetter, R.D., Isacoff, E.Y. & Goodman, C.S. Watching a synapsegrow: Noninvasive confocal imaging of synaptic growth in Drosophila. Neuron 22, 719–729 (1999).

17. McCabe, B.D. et al. The BMP homolog Gbb provides a retrograde signal that regulatessynaptic growth at the Drosophila neuromuscular junction. Neuron 39, 241–254(2003).

18. Aberle, H. et al. wishful thinking encodes a BMP type II receptor that regulates synapticgrowth in Drosophila. Neuron 33, 545–558 (2002).

19. Marques, G. et al. The Drosophila BMP type II receptor Wishful Thinkingregulates neuromuscular synapse morphology and function. Neuron 33, 529–543(2002).

20. Haghighi, A.P. et al.Retrograde control of synaptic transmission by postsynaptic CaMKIIat the Drosophila neuromuscular junction. Neuron 39, 255–267 (2003).

21. Chang, H., Ciani, S. & Kidokoro, Y. Ion permeation properties of the glutamate receptorchannel in cultured embryonic Drosophila myotubes. J. Physiol. (Lond.) 476, 1–16(1994).

22. Miyawaki, A. et al. Fluorescent indicators for Ca2+ based on green fluorescent proteinsand calmodulin. Nature 388, 882–887 (1997).

23. Miyawaki, A., Griesbeck, O., Heim, R. & Tsien, R.Y. Dynamic and quantitative Ca2+measurements using improved cameleons. Proc. Natl. Acad. Sci. USA 96, 2135–2140(1999).

24. Chiba, A., Snow, P., Keshishian, H. & Hotta, Y. Fasciclin III as a synaptic targetrecognition molecule in Drosophila. Nature 374, 166–168 (1995).

25. Johansen, J., Halpern, M.E., Johansen, K.M. & Keshishian, H. Stereotypic morphologyof glutamatergic synapses on identified muscle cells ofDrosophila larvae. J. Neurosci. 9,710–725 (1989).

26. Jia, X.X., Gorczyca, M. & Budnik, V. Ultrastructure of neuromuscular junctions inDrosophila: comparison of wild type and mutants with increased excitability.J. Neurobiol. 24, 1025–1044 (1993).

27. Treiman, M., Caspersen, C. & Christensen, S.B. A tool coming of age: thapsigargin as aninhibitor of sarcoendoplasmic reticulum Ca2+-ATPases. Trends Pharmacol. Sci. 19,131–135 (1998).

28. Sullivan, K.M.C., Scott, K., Zuker, C.S. & Rubin, G.M. The ryanodine receptor isessential for larval development in Drosophila melanogaster. Proc. Natl. Acad. Sci.USA 97, 5942–5947 (2000).

29. Davis, G.W. & Goodman, C.S. Synapse-specific control of synaptic efficacy at theterminals of a single neuron. Nature 392, 82–86 (1998).

30. Petersen, S.A., Fetter, R.D., Noordemeer, J.N., Goodman, C.S. & DiAntonio, A. Geneticanalysis of glutamate receptors in Drosophila reveals a retrograde signal regulatingpresynaptic transmitter release. Neuron 19, 1237–1248 (1997).

31. Marrus, S.B., Portman, S.L., Allen, M.J., Moffat, K.G. & DiAntonio, A. Differentiallocalization of glutamate receptor subunits at the Drosophila neuromuscular junction.J. Neurosci. 24, 1406–1415 (2004).

32. Qin, G. et al. Four different subunits are essential for expressing the synaptic glutamatereceptor at neuromuscular junctions of Drosophila. J. Neurosci. 25, 3209–3218(2005).

33. DiAntonio, A., Petersen, S.A., Heckmann, M. & Goodman, C.S. Glutamate receptorexpression regulates quantal size and quantal content at the Drosophila neuromuscularjunction. J. Neurosci. 19, 3023–3032 (1999).

34. Wucherpfennig, T., Wilsch-Brauninger, M. & Gonzalez-Gaitan, M. Role of DrosophilaRab5 during endosomal trafficking at the synapse and evoked neurotransmitter release.J. Cell Biol. 161, 609–624 (2003).

35. Griesbeck, O., Baird, G.S., Campbell, R.E., Zacharias, D.A. & Tsien, R.Y. Reducing theenvironmental sensitivity of yellow fluorescent protein. Mechanism and applications.J. Biol. Chem. 276, 29188–29194 (2001).

36. Reiff, D.F. et al. In vivo performance of genetically encoded indicators of neural activityin flies. J. Neurosci. 25, 4766–4778 (2005).

37. Miesenbock, G., De Angelis, D.A. & Rothman, J.E. Visualizing secretion and synaptictransmission with pH-sensitive green fluorescent proteins. Nature 394, 192–195(1998).

38. Atwood, H.L. Variation in physiological properties of crustacean motor synapses. Nature215, 57–58 (1967).

39. Lavidis, N.A. & Bennett, M.R. Probabilistic secretion of quanta from visualizedsympathetic nerve varicosities in mouse vas deferens. J. Physiol. (Lond.) 454, 9–26(1992).

40. Bennett, M.R., Jones, P. & Lavidis, N.A. The probability of quantal secretion alongvisualized terminal branches at amphibian (Bufo marinus) neuromuscular synapses.J. Physiol. (Lond.) 379, 257–274 (1986).

41. Bittner, G.D. Differentiation of nerve terminals in the crayfish opener muscle and itsfunctional significance. J. Gen. Physiol. 51, 731–758 (1968).

42. Cooper, R.L., Harrington, C.C., Marin, L. & Atwood, H.L. Quantal release at visualizedterminals of a crayfish motor axon: intraterminal and regional differences. J. Comp.Neurol. 375, 583–600 (1996).

43. Robitaille, R. & Tremblay, J.P. Non-uniform responses to Ca2+ along the frog neuromus-cular junction: effects on the probability of spontaneous and evoked transmitter release.Neuroscience 40, 571–585 (1991).

44. Davis, G.W., DiAntonio, A., Petersen, S.A. & Goodman, C.S. Postsynaptic PKA controlsquantal size and reveals a retrograde signal that regulates presynaptic transmitterrelease in Drosophila. Neuron 20, 305–315 (1998).

45. Reiff, D.F., Thiel, P.R. & Schuster, C.M. Differential regulation of active zone densityduring long-term strengthening ofDrosophila neuromuscular junctions. J. Neurosci. 22,9399–9409 (2002).

46. Quigley, P.A., Msghina, M., Govind, C.K. & Atwood, H.L. Visible evidence for differencesin synaptic effectiveness with activity-dependent vesicular uptake and release of FM1–43. J. Neurophysiol. 81, 356–370 (1999).

47. Roos, J., Hummel, T., Ng, N., Klambt, C. & Davis, G.W. Drosophila Futsch regulatessynaptic microtubule organization and is necessary for synaptic growth. Neuron 26,371–382 (2000).

48. Ruiz-Canada, C. et al. New synaptic bouton formation is disrupted by misregulation ofmicrotubule stability in aPKC mutants. Neuron 42, 567–580 (2004).

49. Stewart, B.A., Atwood, H.L., Renger, J.J., Wang, J. & Wu, C.-F. Improved stability ofDrosophila larval neuromuscular preparations in haemolymph-like physiological solu-tions. J. Comp. Physiol. A 175, 179–191 (1994).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

E R R ATA

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 10 | OCTOBER 2005 1411

Erratum: Heterogeneity in synaptic transmission along a Drosophila larval motor axonGiovanna Guerrero, Dierk F Rieff, Gautam Agarwal, Robin W Ball, Alexander Borst, Corey S Goodman & Ehud Y IsacoffNat. Neurosci., 8, 1188–1196 (2005)

In the version of this article initially published online, the second author’s name was misspelled. The correct spelling is Dierk F Reiff.

Erratum: Notch signaling in the mammalian central nervous system: insights from mouse mutantsKeejung Yoon & Nicholas GaianoNat. Neurosci., 8, 709 – 715 (2005)

The version of this article that was published contained typographical errors in some gene names. On page 710, in the right column, third paragraph, the fourth sentence should have read as follows: “That study examined the expression both of pathway components such as Hes1, Hes5 and Delta-like 1 (Dll1) and of early differentiation markers such as Math4A (also known as Neurog2), NeuroD and NSCL-1 (also known as Nhlh1).” The last sentence of that paragraph should have read as follows: “This notion is supported by previous findings that Hes1 can be upregulated in PC12 cells cultured in the growth factors NGF, FGF2 or EGF64 and in postnatal cerebellar granule cells cultured in Sonic hedge-hog65.” The fourth and fifth sentences in the second paragraph, right column, on page 713 should have read as follows: “These studies found that Notch activation reduced neurite extension, but presumed signaling blockade (via expression of Numb, Numbl or Dx) could promote neurite extension. Subsequent studies have found that Numb deletion disrupts neuronal maturation in the developing cerebellum31, whereas deletion of Numb and Numbl disrupts axonal arborization in sensory ganglia in vivo32.” In addition, on page 712, in the right column, top line, the authors would like to revise the sentence to read as follows: “Third, several reports have identified ErbB2 as a Notch target that has a role during mammalian radial glial maintenance91,92.”

CO R R I G E N D U M

Corrigendum: Visual field maps and stimulus selectivity in human ventral occipital cortexAlyssa A Brewer, Junjie Liu, Alex R Wade & Brian A WandellNat. Neurosci., 8, 1102-1109 (2005)

The discussion section contains an incorrect citation. In the 3rd paragraph on page 1107, “Tootell et al. 16 (subsequent to Halgren et al.)” should read: “Tootell et al. 16 (subsequent to Hadjikhani et al.)”. The authors regret the error

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Neural basis of auditory-induced shifts in visualtime-order perception

John J McDonald1, Wolfgang A Teder-Salejarvi2, Francesco Di Russo3,4 & Steven A Hillyard2

Attended objects are perceived to occur before unattended objects even when the two objects are presented simultaneously.

This finding has led to the widespread view that attention modulates the speed of neural transmission in the various perceptual

pathways. We recorded event-related potentials during a time-order judgment task to determine whether a reflexive shift of

attention to a sudden sound modulates the speed of sensory processing in the human visual system. Attentional cueing influenced

the perceived order of lateralized visual events but not the timing of event-related potentials in visual cortex. Attentional cueing

did, however, enhance the amplitude of neural activity in visual cortex, which shows that attention-induced shifts in visual time-

order perception can arise from modulations of signal strength rather than processing speed in the early visual-cortical pathways.

The effect of attention on the perceived timing of simultaneous andnearly simultaneous events has been of interest to psychologists for overa century. According to Titchener’s ‘law of prior entry’1, objects towhich we attend enter our consciousness before objects to which we donot attend. In more recent years, prior entry has been conceived notonly in terms of the order in which attended and unattended stimulienter consciousness, but also in terms of the speed with whichinformation arising from attended and unattended stimuli is trans-mitted through sensory pathways2–7. According to this view, sensorysignals arising from an attended stimulus arrive at some critical braincenter earlier than sensory signals arising from unattended stimuli.Thus, directing attention to a stimulus was proposed to accelerate therate of information processing in the cortical sensory pathways.

Critical evidence that attention affects the perceived timing of visualstimuli has come from the temporal order judgment (TOJ) task. Intypical TOJ experiments, two target stimuli are presented simulta-neously or in rapid succession, and observers are required to judge theirtemporal sequence, for example, by indicating which of the two stimulicame first3–8. Attention is directed to the location of one of theimpending targets in advance, either by means of a symbolic cue oran attention-capturing peripheral stimulus. Two related findings haveprovided support for the attention-induced prior entry hypothesis.First, participants are maximally uncertain as to which stimulus arrivedfirst when the unattended stimulus is presented before the attendedstimulus. The unattended-stimulus lead time that results in maximaluncertainty about temporal order has been labeled the point ofsubjective simultaneity (PSS). Second, when the attended and unat-tended stimuli are presented simultaneously, participants often reportthat the attended stimulus appeared first. We will refer to these effectsjointly as ‘TOJ effects’.

Despite the evidence obtained from TOJ experiments, someinvestigators have questioned whether paying attention really resultsin prior entry to perceptual stages of processing3,4,9–12. The mainconcern is that attention-induced TOJ effects could arise from changesat post-perceptual stages of processing rather than perceptual stagesof processing. In particular, participants might report that theyperceived the attended stimulus first simply because they are biasedto respond in the direction of the cue. To reduce response bias,participants have been given additional response options (such as‘simultaneous’3,7), different tasks (such as ‘which came first’ versus‘which came last’5) or instructions to respond to target features that areindependent of the direction of the attentional manipulation5,6.Unfortunately, the response-bias problem may not be eliminated bythese strategies alone, as demonstrated by the observation of largerTOJ effects in ‘which came first’ tasks than in ‘which came last’ tasks,even when the response dimension is orthogonal to the attentionalcueing dimension5.

In the present study, we used recordings of event-related brainpotentials (ERPs) to investigate the neural basis of attention-inducedTOJ effects. We used a cross-modal procedure in which an attention-capturing sound was presented to the left or right of fixation before thepresentation of a pair of simultaneous or nearly simultaneous visualtargets (Fig. 1). To minimize response bias, we used an orthogonalresponse procedure wherein observers indicated whether the targetperceived to occur first was green or red. Our neurophysiological inves-tigation was premised on the lateralized organization of the visual path-ways from retina to cortex and the known lateralized attention effectson the ERPs to bilateral stimulus arrays. Specifically, the early compo-nents of the visual ERP arise predominantly from the hemispherecontralateral to the side of stimulation owing to the organization of

Published online 31 July 2005; doi:10.1038/nn1512

1Department of Psychology, Simon Fraser University, 8888 University Drive, Burnaby, British Columbia, V5A 1S6, Canada. 2Department of Neurosciences, Universityof California San Diego, 9500 Gilman Drive, La Jolla, California, 92093-0608, USA. 3Department of Education in Sport and Human Movement, University for HumanMovement (IUSM), Piazza Lauro De Bosis, 15, 00194, Rome, Italy. 4Santa Lucia Foundation IRCCS, Via Ardeatina, 306, 00179, Rome, Italy. Correspondence shouldbe addressed to J.J.M. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

the visual pathways13,14. Consequently, directing attention to one sideof a bilaterally symmetrical visual display results in an enlarged earlypositive ERP (at 80–140 ms) over the contralateral relative to theipsilateral occipital cortex15–18.

From these findings, we hypothesized that if a reflexive shift ofattention to a sudden sound affected the speed of information proces-sing through the visual system, then early ERP activity in the 80–140 msrange elicited by simultaneous visual targets would occur at a shorterlatency over the contralateral occipital scalp than over the ipsilateraloccipital scalp. Such a latency shift would provide compelling evidencefor the accelerated-processing view of prior entry (see ref. 6, p. 823). If,however, a reflexive shift of attention to a sudden sound influenced theprocessing of subsequent visual targets only at post-perceptual stages,then the timing and morphology of early ERPs would be highly similarover the contralateral and ipsilateral occipital cortices. A third possibleoutcome consistent with the aforementioned ERP studies would bethat early ERP components would be larger, but not earlier, over theoccipital cortex contralateral to the preceding sound cue. Indeed, thegeneral finding has been that spatial attention modulates the amplitudeof early ERP components without substantially affecting their laten-cies13,19. Such a result would indicate that the attention-inducedTOJ effects are caused by signal enhancements at early levels of sensoryprocessing that are interpreted as timing differences at a subsequentstage of processing. In the present study we analyzed the amplitudesand latencies of early ERP components in order to distinguish betweenthese alternative mechanisms that may underlie attention-inducedTOJ effects.

RESULTS

The nonpredictive auditory cue had a substantial effect on TOJperformance (Fig. 2). When the two targets were presented simulta-neously, participants reported seeing the target on the cued side firstsignificantly more often than chance (78.9% versus 50%; t27 ¼ 12.7,Po 0.000001). The PSS was estimated to be �68.5 ms, which indicatesthat the target on the uncued side needed to occur 68.5 ms before thetarget on the cued side in order for participants to have no lateralizedpreference in their temporal order judgments.

Amplitude of neural activity is related to perceived order

To investigate the effects of the auditory cue on visual processing, wefirst examined the ERPs elicited by the simultaneously presented visualtargets that followed left and right auditory cues. Figure 3 shows thesimultaneous-target ERPs averaged over the 79% of trials in whichparticipants judged that the target on the cued side appeared first. TheERP waveforms recorded over the posterior lateral scalp consisted of aseries of positive and negative peaks, including prominent C1 (meanlatency of 72 ms), P1 (110 ms), N1 (160 ms), P2 (210 ms) and N2(260 ms) components (Fig. 3a). Starting at approximately 80 ms post-stimulus, the visual ERPs recorded over the left posterior scalp becamemore positive after the right auditory cue than after the left auditorycue. The converse was true for the right posterior scalp. This enhancedcontralateral positivity lasted for approximately 140 ms and overlappedthe P1, N1 and P2 components. Statistical tests confirmed that theenhanced contralateral positivity was significant between 80 ms and220 ms post-stimulus (Fig. 3b).

To estimate the neural sources of these cue-induced, lateralizedmodulations of the target ERP, we first created spline-interpolatedvoltage maps of the ERPs to the simultaneous visual targets after leftand right cues and the difference waveforms formed by subtracting theright-cued from the left-cued ERPs (Fig. 4). An enlarged positivityoverlapped the time ranges of the P1 (90–120 ms post-stimulus) andsubsequent N1 (135–175 ms post-stimulus) components: the P1 waslarger over the hemisphere contralateral to the cued side, whereas the

LEDs

Speaker

a

b

Figure 1 Experimental setup. (a) Schematic illustration of audiovisual

apparatus. (b) Example of stimulus sequences on simultaneous-target and

nonsimultaneous-target trials. The top and bottom rows illustrate

nonsimultaneous-target trials (50% of all trials; target on cued side presented

first in top row; target on uncued side presented first in bottom row), and the

middle row illustrates a simultaneous-target trial (50% of all trials). The

stimulus onset asynchrony (SOA) between the auditory cue and the first

visual target event was 100–300 ms, and on nonsimultaneous-target trials,

the SOA between targets was either 35 or 70 ms. Targets appearedsimultaneously on 50% of the trials.

–70 –35 0 35 700

25

50

75

100

Cued side onset advantage (ms)

Cue

d si

de r

epor

ted

first

(%

)

Cued target first

Point of subjectivesimultaneity

Actual simultaneity

Uncued target first

Figure 2 Mean percentage of trials in which participants reported seeing

the target on the cued side before the target on the uncued side as a

function of the cued side onset advantage (CSOA). The 0-ms CSOA

indicates that the two targets appeared simultaneously. Negative CSOAs

indicate that the target on the uncued side was presented before the

target on the cued side, whereas positive CSOAs indicate that the target

on the cued side was presented before the target on the uncued side.

The PSS was estimated by fitting the data points with a third-order

polynomial and interpolating the CSOA at which participants would

have reported seeing the target on the cued side first 50% of thetime. Data were collapsed across cued side (left, right) and target color

(red, green).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

N1 was smaller (that is, more positive, owing to the overlappingpositivity) over the hemisphere contralateral to the cued side. Eachdifference map in Figure 4 shows a focal positivity over the rightoccipito-temporal scalp and a focal negativity over the left occipito-temporal scalp; the negativity appears because the subtraction invertedthe sign of the contralateral positivity that was elicited when the cue wason the right.

The neural generators of the enlarged contralateral ERP positivitywere modeled as dipolar current sources fit to the difference topogra-phies shown in Figure 4. A pair of dipoles in ventral occipito-temporalcortex accounted for the contralateral positivity over a 100-ms latencyrange that spanned the P1 and N1 components (90–190 ms, Fig. 5).These dipoles were projected onto a standardized brain and were foundto be situated in the fusiform gyrus of the occipito-temporal cortex(Talairach coordinates: x ¼ ±35, y ¼ �46, z ¼ �20). A second pair ofdipoles fit over 220–250 ms accounted for a brief phase of greaternegativity over both hemispheres after the left cue. These dipoles weresituated in the superior temporal region. The residual variance of thisdipole model over the 96–240 ms latency range was 3.5%.

Timing of neural activity is unrelated to

perceived order

In contrast to the significant effects of audi-tory cueing on the early positive amplitude ofthe simultaneous-target ERPs, auditory cue-ing had little or no effect on the peak latenciesof the principal ERP components. The C1, P1,N1 and N2 components elicited by the simul-taneous targets did not differ in latencybetween the hemispheres contralateral andipsilateral to the preceding cue (all P values4 0.05), whereas the P2 peak latency was 5ms earlier over the contralateral hemisphere(P ¼ 0.022). To further examine the possibi-lity of any cue-induced effects on the timing oftarget-elicited neural activity, we calculatedbest-fitting dipoles for each of the earlyERP components elicited by the simultaneoustargets in the following latency ranges:80–120 ms (P1), 150–175 ms (N1) and180–230 ms (P2). The resulting three dipolepairs (P1: x¼ ±28, y¼�62, z¼ �5; N1: x¼±21, y ¼ �77, z ¼ 12; P2: x ¼ ±26, y ¼ �41,z ¼ 17) provided a good fit to these early

components (residual variance ¼ 1.9% within the 80- to 230-msrange). Notably, the source waveforms of dipoles located contralateraland ipsilateral to the auditory cue showed nearly identical time courses,with peaks in the relevant latency ranges occurring at approximatelythe same times (P1: 113 and 112 ms, N1: 158 and 161 ms, and P2: 204and 204 ms, at contralateral and ipsilateral sites, respectively).

Finally, we examined the ERPs to nonsimultaneous visual targets todetermine whether our procedure was sufficiently sensitive to latencyshifts in the early ERP components. Of particular interest were theERPs elicited when the target on the uncued side preceded the target onthe cued side by 70 ms, because this particular target asynchrony wasvery close to the PSS (68.5 ms). If the timing of the activity in the visualsystem is related to the actual temporal asynchrony between events, theearly ERP components elicited by the successive targets should be offsetby approximately 70 ms (that is, the physical asynchrony). If, however,the timing of the activity in the visual system is related to the perceivedtemporal asynchrony between events, then the corresponding earlyERP components elicited by the successive targets should occur atapproximately the same time. The waveform contralateral to T1consisted of T1-elicited P1 (at 116 ms), N1 (at 165 ms), P2 (at

ERPs to simultaneous visual targetspreceded by left cue and right cue

Preceded by left cuePreceded by right cue

N1N2

C1

–100 0 100

0 100

200 300

Time post-stimulus (ms)

–2 µV

–2 µV

P1

P2

ERPs from hemispheres contralateraland ipsilateral to side of cue

Time post-stimulus (ms)–100 200 300

Ipsilateral to side of cueContralateral to side of cue

p < 0.05n.s.

p < 0.01p < 0.001

a b Figure 3 Grand-averaged ERPs to simultaneous

visual targets, averaged over the 79% of trials in

which participants indicated that the target on the

cued side appeared first. (a) Target ERPs recorded

over parietal (P1, P2) and occipital (PO7, PO8)

brain areas, shown separately for left-cue and

right-cue trials. (b) Target ERPs at the occipital

sites collapsed over left- and right-cue conditionsand left and right hemispheres so as to yield

waveforms recorded contralateral and ipsilateral

to the side of the cue. Statistical tests on the

differences between the mean ERP amplitudes at

contralateral versus ipsilateral recording sites were

performed in successive 20-ms intervals from

40 ms to 300 ms post-stimulus. The P values

from these tests are illustrated by shading above

the time axis.

90–120ms

Preceded byleft cue

Preceded byright cue

More positivecontralateral to cued side

More positive (less negative)contralateral to cued side

Left-rightdifference

135–175ms

0.8

–0.8

1.0

–1.0

1.70

–1.70

2.33

–2.33

Figure 4 Scalp topographies of the simultaneous-target ERP waveforms after a left or right cue (left and

center columns) and the difference wave created by subtracting the right-cued target ERP from the left-

cued target ERP (right column).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

200 ms) and N2 (at 240 ms) components (Fig. 6a), whereas thewaveform contralateral to T2 consisted of T2-elicited P1 (at 185 ms),N1 (at 234 ms) and N2 (at 280 ms) components. The positive peak at120 ms in the contralateral-to-T2 waveform was the delayed P1 elicitedby T1 that is transferred from the contralateral to the ipsilateral hemi-sphere across the corpus callosum13,14. Notably, the P1 and N1 compo-nents that were elicited by the successive targets were each offset by69 ms, which closely approximates the actual time asynchrony betweentargets. A further analysis showed that the latencies of the P1, N1, P2and N2 components did not differ appreciably between trials when thecued target was reported to be first and when the uncued target wasreported to be first (Supplementary Fig. 1). The T1- and T2-elicited P1peaks were maximal over the same ventral occipital areas (Fig. 6b).

DISCUSSION

The behavioral results presented here fit well with recent reports thatinvoluntary shifts of attention to sound can influence the perception ofsubsequent visual stimuli20–24. As in the present study, a previousstudy24 reported that a sound appearing to the left or right of fixationcan affect the temporal perception of subsequent visual events when thecue-target stimulus-onset asynchrony (SOA) was between 0 and300 ms. Similar results were obtained in that study when nonpredictivevisual, auditory and tactile cues were used to orient attention andwhen participants were asked to judge the perceived motion of a

stationary line rather than the perceived tem-poral order of two visual stimuli. Takentogether, these behavioral results indicatethat modulations of visual perception bycross-modal attention are robust over a vari-ety of psychophysical techniques.

Recent behavioral work on prior entry hasshown that attention-induced shifts in thePSS may arise in part from post-perceptualeffects such as response bias3–6. In light of this,we used an orthogonal-response procedure,in which reflexive shifts of attention areknown to produce substantial perceptualeffects above and beyond any response-biaseffects5. Thus, although response bias mightnot have been completely eliminated, the TOJeffects observed here were largely due to actualchanges in the perceived timing of the stimuli.The exact contributions of perceptual andpost-perceptual effects, however, remain tobe determined.

The electrophysiological results presentedhere argue against the hypothesis that direct-ing attention to one side accelerates the sen-

sory processing of contralateral visual stimuli. A difference in visualcomponent latencies between the hemispheres would be expected iftheir underlying neural generators were activated more rapidly in thehemisphere contralateral to the nonpredictive auditory cue. No inter-hemispheric differences in the latencies of the early ERP components orof their occipital dipole source waveforms were observed, however,when the targets were presented simultaneously, even though partici-pants judged the target on the cued side to appear first on the majorityof trials. The small difference in P2 latency (5 ms) that was measured

180115–10

0

10

–10

0

10

–10

0

10

–10

–100 100 200Time post-stimulus (ms)

Sou

rce

ampl

itude

(µV

)

300 400

A P

L R

0

0

10

110176 235

235

Figure 5 Dipole model of the neural sources of the left-cued minus right-cued difference wave. Dipoles

1 and 2 (red; x ¼ ±35, y ¼ �46, z ¼ �20) were fit over the 90–190 ms interval, and dipoles 3 and 4

(blue; x ¼ ±35, y ¼ �26, z ¼ 18) were fit over the 220–250 ms interval. The source waveforms on the

left show the time courses of activity for the computed dipoles, and the images on the right show the

locations of the dipoles with respect to brain anatomy.

400Time post-stimulus (ms)

300

N2

N1

N2

P2

200

Contralateral to T2Contralateral to T1

P1

N1

P1

100

T2T1

0

–1 µV

P1 to T2P1 to T1

2.2–2.20.78–1.36

a

bFigure 6 Target ERPs obtained at the point of subjective simultaneity.

(a) Grand-averaged ERPs obtained when the uncued target was presented

70 ms before the cued target. The ERPs shown were recorded from occipital

electrodes (PO7 and PO8) and were collapsed over cue side (left, right) and

perceived temporal order (target on cued side first, target on uncued side

first). The onsets of the uncued target (T1) and the cued target (T2) aredenoted on the time axis, and the major contralateral ERP components

elicited by T1 and T2 are labeled in red and blue, respectively. (b) Scalp

topographies of the P1 components elicited by T1 (left) and T2 (right),

plotted at 124 ms and 185 ms, respectively. The maps were constructed

as if T1 were presented on the left and T2 were presented on the right.

The arrows point to the contralateral P1 foci.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

between the hemispheres may be attributed to the overlap of P2 withthe terminal phase of the enhanced positivity, which was larger atcontralateral hemisphere sites. In contrast, substantial interhemisphericdifferences in the latencies of the P1 and N1 were observed when thetargets were presented successively with a time separation correspond-ing to that of the PSS, even though participants showed no lateralpreference in their timing judgments because the reflexive attentionalcueing counteracted the actual timing difference. In this case theinterhemispheric latency difference corresponded to the actual timedelay (70 ms) between the successive targets. This pattern of resultsindicates that the timing of early, stimulus-evoked neural activity in thevisual cortex is more closely related to the actual timing of stimuli thanto the perceived timing of stimuli. It should be cautioned, however, thatthe perceptual contribution to the shifted PSS (r68.5 ms) may havebeen appreciably smaller than the actual target asynchrony.

Although auditory cueing did not affect the latencies of theearly visual ERP activity, it did affect the amplitude of the ERPwaveform. The earliest such effect was an enlarged positivity over thehemisphere contralateral to the cued side of the simultaneous-targetdisplay, which started 80 ms after target onset and lasted for about140 ms. Topographical mapping and source analysis indicated thatthis enlarged contralateral positivity originated in the ventral occipito-temporal cortex. Very similar enhancements of early contra-lateral positivity have been reported in studies where attention wasdirected by endogenous cues to one side of a bilaterally symmetricalvisual display15–17.

The present study advances our understanding of how the timing ofbrain activity relates to the timing of our visual perceptions. In recentyears, investigators have argued that sensory signals arising fromattended objects are transmitted through the visual system morerapidly than sensory signals arising from unattended objects2,5–7. Wefound no evidence for this widely accepted view. Instead, our studyprovides clear evidence that the timing of visual perception is notinevitably based on the timing of neural events in the visual-corticalpathways. On the basis of our ERP findings, we propose the alternativehypothesis that attention-induced effects on time-order perceptionmay arise from changes in the strength of neural signals in ventraloccipital areas that underlie visual object perception. It follows fromthis hypothesis that attention-induced enhancements in signal strengththat occur at early stages of visual processing are interpreted as a timingdifference by a later comparator mechanism. The challenge for futurework is to define the subsequent neural events that convert the initialsignal enhancement into a perceived time-order difference.

METHODSParticipants. Twenty-eight healthy adults (mean age 26.0 ± 6.3 s.d.; 16 female;

3 right-handed) participated after giving informed written consent. Each

participant had normal or corrected-to-normal visual acuity and reported

having normal color vision and hearing.

Apparatus. The experiment was conducted in a dimly lit sound-attenuated

chamber with an ambient sound level of 32 dB (A). Auditory and visual stimuli

were delivered from a custom apparatus consisting of two audiovisual displays

and a fixation light arranged on a semicircular arc (Fig. 1). Participants sat 1 m

from the arc and faced the fixation light. The audiovisual displays were

positioned 201 to the left and right of fixation, and each consisted of a

loudspeaker, four red light-emitting diodes (LEDs) and four green LEDs.

The red and green LEDs were arranged in interleaved squares and were

mounted directly above the speaker cone (Fig. 1).

Stimuli and procedure. At the beginning of each trial, a spatially nonpredictive

auditory cue (500- to 15,000-Hz ‘pink’ noise burst, 76 dB SPL, 83 ms duration)

was delivered randomly from either the left or right speaker. After a

randomized delay of 17–217 ms (100- to 300-ms SOA, rectangular distribu-

tion), a pair of visual targets was presented, one to each visual field. The target

on one side was a green flash produced by turning on all green LEDs on that

side for 10 ms; the target on the other side was a red flash produced by turning

on all red LEDs on that side for 10 ms. The locations of the red and green

targets were determined randomly, with equal numbers of left-red/right-green

and left-green/right-red trials. The two targets were presented simultaneously

on 50% of the trials; on the remaining trials, the left target appeared 35- or 70-

ms before the right target, or the right target appeared 35- or 70-ms before the

left target. These trials were delivered in random order. Participants were told to

ignore the auditory stimulus and to indicate the color of the target that was

perceived to occur earlier by pressing one of two buttons held in separate

hands. Participants were instructed to respond as accurately as possible and

were given 2 s to respond before the onset of the next trial. The color-button

mapping was counterbalanced across participants. The experiment comprised

20 blocks of 32 trials.

Electrophysiological recording and analysis. Continuous EEG was acquired

from tin electrodes mounted in an elastic cap (Electro-cap International). Fifty-

six electrodes were positioned according to the 10-10 system, and four

additional electrodes were positioned inferior to the occipital sites to ensure

adequate spatial sampling of the posterior scalp23. The horizontal electro-

oculogram (EOG) was acquired using a bipolar pair of electrodes positioned at

the external ocular canthi. All EEG electrodes were referenced to an electrode

on the right mastoid (A2). The EEG and EOG channels were amplified with a

gain of 20,000, filtered (2-pole Butterworth) with a bandpass of 0.1–100 Hz

(�3 dB point; �12 dB per octave) and digitized at 250 Hz.

After the recording session, the EEG and EOG were averaged over 3-s epochs

that started 1.5 s before the onset of each cue and target. Automated artifact

rejection was performed before averaging to discard trials during which an

eye movement, blink or amplifier blocking occurred. The averaged waveforms

were digitally low-pass filtered (�3 dB point at 25 Hz; zero-phase) and re-

referenced to the average of the left and right mastoids. Deflections in the

averaged EOG waveforms were small (o1.5 mV), which indicated good

maintenance of fixation.

Cue and target ERPs were averaged separately for all combinations of cue

side (left, right), target asynchrony (�70 ms, �35 ms, 0 ms, 35 ms, 70 ms,

where negative values denote trials with the first target on the uncued side) and

perceived target order (green first, red first). As the cue-target SOA was varied

over a 200-ms range, the cue ERPs did not consistently overlap the target ERPs.

The adjacent-response (Adjar) filter procedure25 was used to remove any

residual overlap originating from the cue ERPs. The target ERPs were collapsed

across the two color combinations (left-red/right-green and left-green/right-

red) to eliminate possible ERP lateralizations related to target color. The

analysis of ERP amplitudes focused on the simultaneously occurring targets.

Mean amplitudes of the bilateral-target ERP waveforms were measured with

respect to a 100-ms prestimulus period in successive 20-ms intervals starting at

60 ms post-stimulus. These measurements were taken from a pair of lateral-

occipital electrodes (PO7 and PO8) at which cueing effects were maximal. The

mean amplitudes in each interval were analyzed in a repeated-measures analysis

of variance (ANOVA) with cued side (left, right) and electrode lateralization

(relative to the cued side; contralateral, ipsilateral) serving as within-subject

factors. The peak latencies of the C1, P1, N1 and P2 components of the

bilateral-target ERP waveform were also determined at posterior electrodes (C1:

electrodes P1 and P2; P1, N1 and P2: electrodes PO7 and PO8) and were

subjected to a similar ANOVA. Finally, the peak latencies of the P1 and N1

components of the nonsimultaneous target ERP waveforms were measured at

posterior electrodes (PO7 and PO8).

Topographical mapping and source localization. Difference waveforms were

calculated by subtracting the right-cued from the left-cued simultaneous target

ERPs. Topographical maps of the simultaneous-target ERPs and correspond-

ing difference waveforms were constructed by spherical spline interpolation26.

The cortical generators of the difference-wave potentials were estimated

using Brain Electrical Source Analysis (BESA 2000 version 5.0). The BESA

algorithm estimates the location and the orientation of multiple equivalent

dipolar sources by calculating the scalp distribution that would be obtained for


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

a given dipole model (forward solution) and comparing it with the actual

scalp-recorded voltage distribution. The algorithm interactively adjusts (fits)

the location and orientation of the dipole sources in order to minimize the

residual variance (RV) between the model and the observed spatiotemporal

ERP distribution.

To investigate the anatomical sources of the attention-induced changes

in the amplitude of visual cortical activity, we sequentially fit two pairs of

dipoles to distinctive components in the difference waveforms within specified

intervals. Each dipole pair was constrained to be mirror-image in location

only. The first dipole pair was fit to the left-cued minus right-cued target

difference wave topography in the 90–190 ms post-stimulus interval. The

second pair was fit over the 220–250 ms interval. The coordinates of each

dipole were registered on the standardized finite element model (FEM) of BESA

2000, which was created from an averaged head using 24 individual MRIs in

Talairach space.

To investigate the attention-induced changes in the timing of visual cortical

activity, we fit separate pairs of dipoles to the occipital P1, N1 and P2

components in the original ERPs (not difference waves) elicited by the

simultaneous targets. Dipole pairs were fit within intervals that encompassed

these early peaks in the grand-averaged waveforms (P1: 80–120 ms, N1:

150–175 ms and P2: 180–230 ms). Each dipole pair was constrained to be

mirror-image in location only. The coordinates of each dipole were registered

in Talairach space. The latencies of major peaks in the dipole source waveforms

were taken as indices of neural response timing.


ACKNOWLEDGMENTSThe authors thank D.R. Heraldez and M.M. Marlow for technical assistanceand L.M. Ward for providing access to research equipment. This research wassupported by grants from the National Institute for Mental Health (MH 25594),Natural Sciences and Engineering Research Council of Canada (227959) and theCanadian Foundation for Innovation.


Received 23 February; accepted 5 July 2005


1. Titchener, E.N. Lectures on the Elementary Psychology of Feeling and Attention(Macmillan, New York, 1908).

2. Carrasco, M. & McElree, B. Covert attention accelerates the rate of visual informationprocessing. Proc. Natl. Acad. Sci. USA 98, 5363–5367 (2001).

3. Jaskowski, P. Selective attention and temporal-order judgment. Perception 22,681–689 (1993).

4. Schneider, K.A. & Bavelier, D. Components of visual prior entry. Cognit. Psychol. 47,333–366 (2003).

5. Shore, D.I., Spence, C. & Klein, R.M. Visual prior entry. Psychol. Sci. 12, 205–212(2001).

6. Spence, C., Shore, D.I. & Klein, R.M. Multisensory prior entry. J. Exp. Psychol. Gen.130,799–832 (2001).

7. Stelmach, L.B. & Herdman, C.M. Directed attention and perception of temporal order.J. Exp. Psychol. Hum. Percept. Perform. 17, 539–550 (1991).

8. Stone, S.A. Prior entry in the auditory-tactile complication. Am. J. Psychol. 37,284–287 (1926).

9. Downing, P.E. & Treisman, A.M. The line-motion illusion: Attention or impletion? J. Exp.Psychol. Hum. Percept. Perform. 23, 768–779 (1997).

10. Frey, R.D. Selective attention, event perception and the criterion of acceptabilityprinciple: evidence supporting and rejecting the doctrine of prior entry. Hum. Mov.Sci. 9, 481–530 (1990).

11. Pashler, H.E. The Psychology of Attention (MIT Press, Cambridge, Massachusetts,1998).

12. Schneider, K.A. & Bavelier, D. Components of visual prior entry. Cognit. Psychol. 47,333–366 (2003).

13. Di Russo, F., Martinez, A. & Hillyard, S.A. Source analysis of event-related corticalactivity during visuo-spatial attention. Cereb. Cortex 13, 486–499 (2003).

14. Di Russo, F., Martinez, A., Sereno, M.I., Pitzalis, S. & Hillyard, S.A. Cortical sources ofthe early components of the visual evoked potential. Hum. Brain Mapp. 15, 95–111(2001).

15. Heinze, H.J., Luck, S.J., Mangun, G.R. & Hillyard, S.A. Visual event-related potentialsindex focused attention within bilateral stimulus arrays. I. Evidence for early selection.Electroencephalogr. Clin. Neurophysiol. 75, 511–527 (1990).

16. Heinze, H.J. et al. Combined spatial and temporal imaging of brain activity during visualselective attention in humans. Nature 372, 543–546 (1994).

17. Luck, S.J., Heinze, H.J., Mangun, G.R. & Hillyard, S.A. Visual event-related potentialsindex focused attention within bilateral stimulus arrays. II. Functional dissociationof P1 and N1 components. Electroencephalogr. Clin. Neurophysiol. 75, 528–542(1990).

18. Mangun, G.R., Hopfinger, J.B., Kussmaul, C., Fletcher, E. & Heinze, H.J. Covariations inERP and PET measures of spatial selective attention in human extrastriate visual cortex.Hum. Brain Mapp. 5, 273–279 (1997).

19. Hillyard, S.A. & Anllo-Vento, L. Event-related brain potentials in the study of visualselective attention. Proc. Natl. Acad. Sci. USA 95, 781–787 (1998).

20. Dufour, A. Importance of attentional mechanisms in audiovisual links. Exp. Brain Res.126, 215–222 (1999).

21. Frassinetti, F., Bolognini, N. & Ladavas, E. Enhancement of visual perception bycrossmodal visuo-auditory interaction. Exp. Brain Res. 147, 332–343 (2002).

22. McDonald, J.J., Teder-Salejarvi, W.A. & Hillyard, S.A. Involuntary orienting to soundimproves visual perception. Nature 407, 906–908 (2000).

23. McDonald, J.J., Teder-Salejarvi, W.A., Di Russo, F. & Hillyard, S.A. Neural substrates ofperceptual enhancement by crossmodal spatial attention. J. Cogn. Neurosci. 15, 10–19(2003).

24. Shimojo, S., Miyauchi, S. & Hikosaka, O. Visual motion sensation yielded by non-visuallydriven attention. Vision Res. 37, 1575–1580 (1997).

25. Woldorff, M.G. Distortion of ERP averages due to overlap from temporally adjacentERPs: analysis and correction. Psychophysiology 30, 98–119 (1993).

26. Perrin, F., Pernier, J., Bertrand, O. & Echallier, J.F. Spherical splines for scalp potentialand current density mapping. Electroencephalogr. Clin. Neurophysiol. 72, 184–187(1989).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Auditory thalamus integrates visual inputs intobehavioral gains

Yutaka Komura1, Ryoi Tamura2,3, Teruko Uwano2,3, Hisao Nishijo3,4 & Taketoshi Ono3,5

By binding multisensory signals, we get robust percepts and respond to our surroundings more correctly and quickly. How and

where does the brain link cross-modal sensory information to produce such behavioral advantages? The classical role of sensory

thalamus is to relay modality-specific information to the cortex. Here we find that, in the rat thalamus, visual cues influence audi-

tory responses, which have two distinct components: an early phasic one followed by a late gradual buildup that peaks before re-

ward. Although both bimodal presentation and reward value had similar effects on behavioral performance, the cross-modal effect

on neural activity showed unique temporal dynamics: it affected the amplitude of the early component and starting level of the

late component, whereas reward value affected only the slope of the late component. These results demonstrate that cross-modal

cueing modulates gain in the sensory thalamus, potentially providing a priming influence on the choice of an optimal behavior.

How does the brain integrate signals from separate modality-specificpathways? The classical role of the sensory cortex is to process modality-specific sensory information. Recent animal and human studies haveshown, however, that sensory stimuli of different modalities affectresponses in the early stage of the sensory cortex1–5. For example, visualstimuli such as lip movements activate the auditory cortex6. Little isknown about the mechanisms that underlie such cross-modal effects inthe auditory system. Some studies indicate that top-down informationfrom the multisensory cortical areas may affect auditory cortical activityby means of polysynaptic back-projections7,8. Other studies thatare concerned with perceptual illusions, such as McGurk andventriloquist effects, indicate that vision may influence auditoryperception preattentively1,9,10. These latter observations raise thepossibility that bottom-up subcortical information also stronglyinfluences audiovisual integration in the auditory cortex. To addressthe question of whether or not the auditory thalamic regionparticipates in cross-modal binding, we recorded the activity ofsingle neurons in the medial geniculate complex of a rat while it carriedout an auditory spatial discrimination task with or without visual cues.As a result, we found that cross-modal interaction emerges in theauditory thalamus.

In principle, multisensory cues enhance an animal’s ability todiscriminate among sensory stimuli and to respond to the environ-ment11–13. Such enhancement is often measured on the basis ofbehavioral markers, such as reaction times and error rates. Thesemarkers depend on both sensory and decision processes for theselection of an appropriate response. To clarify whether the auditory

thalamic modulation that was observed in the present study reflects astimulus-driven multisensory effect or a subject-related decision pro-cess, it was necessary to set a condition that was independent of themultisensory conditions in the task. Many experiments have shownthat rewards exert a profound effect on the decision process (but not onthe sensory process) and therefore on behavioral markers14–16. For thisreason (to bias only the decision process), we altered reward values thatwere associated with an identical sensory cue. This manipulationsucceeded in dissociating the multisensory effect from the rewardeffect, leading us to conclude that multisensory presentation affectsearly and later parts of thalamic activity, whereas reward valueinfluences a later part only.

RESULTS

Task paradigm

In the auditory spatial discrimination task, the auditory and/or visualcues were presented for 2 s on the rat’s left or right side. After a 1-s delayincluding no explicit auditory and/or visual cue stimuli, a spoutautomatically protruded close to the rat’s mouth. If it licked the spoutwithin 2 s after the delay period in a ‘go’ trial, the rat could obtain areward. In a ‘no-go’ trial, reward was not given to the rat (asymmetricalreward contingency). This task consisted of three kinds of sessions:unimodal, bimodal and variable-reward sessions (Fig. 1). In the uni-modal sessions (Fig. 1a), the rat was required to perform a go or a no-goresponse when the auditory cue was presented at the left or rightside, respectively; the visual cues were not associated with rewardwhether presented from the left or right side. In the bimodal


1Neuroscience Research Institute, National Institute of Advanced Industrial Science and Technology, 1-1-1 Umezono, Tsukuba 305-8568, Japan. 2Department ofPhysiology, Faculty of Medicine, Toyama Medical and Pharmaceutical University, 2630 Sugitani, Toyama 930-0194, Japan. 3Core Research for Evolutionary Science andTechnology, Japan Science and Technology Agency, 4-1-8 Honcho, Kawaguchi 332-0012, Japan. 4System Emotional Science and 5Molecular and Integrative EmotionalNeuroscience, Graduate School of Medicine, Toyama Medical and Pharmaceutical University, 2630 Sugitani, Toyama 930-0194, Japan. Correspondence should beaddressed to T.O. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

sessions (Fig. 1b), the auditory and visual cues were simultaneouslypresented on the same or opposite side (matched- or conflicting-bimodal condition, respectively). As in the unimodal sessions, therat was required to perform a go or no-go response, based onlyon the position of the auditory stimulus, irrespective of theposition of the visual stimulus. In both the unimodal and bimodalsessions, the reward was constant (50 ml of 0.3 M sucrose solution).In variable-reward sessions (Fig. 1c), auditory cues were used that werethe same as those in the unimodal sessions, but the reward value wasvaried (100 ml of 0.3 M sucrose solution in high-reward sessions and50 ml of water in low-reward sessions). Each session proceeded in a blockdesign, and there were no cues for altering reward values. As in the

unimodal sessions, the rat was required to perform a go or no-goresponse when the auditory cue was presented at the left or the rightside, respectively.

Neural basis for cross-modal signals

We isolated 627 single neurons from the right auditory thalamus. Ofthese, 435 (69.4%) had significant selectivity for the position (left orright, Po 0.01) of the auditory stimulus and were subjected to furtheranalysis (see Supplementary Methods online). Among these position-selective neurons, 384 (88.3%) responded more vigorously when theauditory cue was contralateral to the recording site than when it wasipsilateral (Supplementary Fig. 1). We tested whether or not visualinputs modulated the auditory responses of these cells. Among the totalsample tested (n ¼ 384), 56 (14.6%), 38 (9.9%) and 72 (18.8%)neurons showed bimodal modulation, reward modulation and bothtypes of modulation, respectively.

A typical example of neurons that had phasic responses and showedbimodal modulation was found (Fig. 2a,b). With unimodal stimula-tion, the neuron showed greater responses to the auditory cue on theleft than on the right, but was unresponsive to the visual cue either onthe left or right (Fig. 2a). With bimodal stimulation, such phasicresponses to the left tone were enhanced when the visual cue waspresented simultaneously on the same side as the auditory cue but weredepressed when it was presented simultaneously on the opposite side(Fig. 2b). The neural response to the right tone showed neitherenhancement nor depression. Fifty-six neurons had similar phasicresponses and properties (Fig. 2c). When the tone was presented onthe left (contralateral side to the recording sites), the bimodalresponses were significantly higher and lower in the matched- andconflicting-bimodal conditions, respectively, than was predicted bythe sum of the unimodal responses (Wilcoxon signed-rank test;P o 0.001). When the tone was presented on the right (ipsilateralside), however, the bimodal responses were not significantly differentfrom the sum of the unimodal responses.

Next, we tested the bimodal enhancement of these neurons duringthe extinction of cue-reward association (Fig. 3a). After the pre-extinction trials, the rat continued to lick the spout during the firstseveral extinction trials (extinction 1 block) and then ceased licking

vs.

vs.

ToneTone

+light

Tone+

light

Light

Tone

Tone

Left

Left

Left

Left

Left

Matched

Conflictingvs.

Right

Right

Right

Right

Right

Right

High reward

Low reward

Unimodal session Bimodal session

Variable-reward session

Left

vs.

vs.

vs.

a b

c

Figure 1 Task design. (a–c) The squares represent the types of cue

stimulus in the (a) unimodal, (b) bimodal and (c) variable-reward sessions.

Each session consisted of go or no-go trials with asymmetrical reward

contingency. Filled and open areas indicate go trials with reward and no-go

trials without reward, respectively. Line colors (magenta, light blue, black,

gray, red and blue) and kinds of squares (solid and dotted) correspond to

those of the graphs shown in Figure 2. Note that the position

of the auditory cue is relevant to the task, but the visual cue

is irrelevant.

60

50

40

30

20

0–200 0 200

2.5

2.0

1.5

1.0

0.5

0.0

2.5

2.0

1.5

1.0

0.5

0.0400

Time (ms)

Contra Ipsi

Spi

kes

s–1

Nor

mal

ized

firin

g ra

te

600

Tone (L)

Tone

(L)

Tone

(L) +

light

(L)

Tone

(L) +

light

(R)

Sum (p

redic

ted)

Sum (p

redic

ted)

Tone

(R)

Tone

(R) +

light

(R)

Tone

(R) +

light

(L)

Sum (p

redic

ted)

Sum (p

redic

ted)

Tone (R)

Tone (L) + Light (L)

Tone (R) + Light (R)

Tone (L) + Light (R)

Tone (R) + Light (L)Light (L)

Light (R)

800 1,000 –200 0 200 400Time (ms)

600 800 1,000

10

60

50

40

30

20

0

Spi

kes

s–1

10

*

*

a cb

Figure 2 The cross-modal effect on neural activity in the auditory thalamus. (a,b) A representative neuron in the dorsal part of the right medial geniculate body.

Rastergrams and SDFs of the neuronal activity to each (a) unimodal or (b) bimodal cue stimulus. Time 0 is the onset of cue stimulus. (c) The histograms

summarize the neural responses (mean ± s.d.) to each sensory cue, normalized and averaged across 56 neurons with phasic responses, similar to the examplesshown in a and b. The colors correspond to those of the graphs in Figure 1. White bars indicate the sum of responses to the visual and auditory cues. Black

and gray bars in the white bars indicate the responses to the visual cues on the left and right, respectively. *P o 0.001.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

the spout in the following trials (extinction 2 block). During theseextinction trials, the magnitude of the early responses did not changesignificantly. We found the same tendency in all seven neurons tested.The normalized activities in the pre-extinction trials as well as in theextinction 1 and 2 blocks were higher than those predicted by thesum of the unimodal responses (Fig. 3b; Wilcoxon signed-rank test;P o 0.05). These results indicate that bimodal modulation may beindependent of reward contingency or of a behavioral response.

Neural basis for cross-modal and reward signals

Seventy-two neurons showed, in addition to the early phasic response,a late gradual increase in the discharge rate, which peaked beforereward delivery (Fig. 4a). In trials with a left auditory cue (go trials),both components were larger in the matched-bimodal condition and

smaller in the conflicting-bimodal condition than in the unimodalcondition (Fig. 4b). In addition, the late component was selectivelyenhanced or suppressed in the unimodal condition when the rewardvalue increased or decreased, respectively (Fig. 4c). The net effect ofeach condition on the normalized population activity after subtractionof the unimodal effect is shown (Fig. 4d). The bimodal and rewardeffects on the auditory thalamic responses had different time courses.To facilitate the comparison of these effects, we used two indices of thelate components, the starting level (the normalized differential activityat 700 ms after the cue onset) and slope (connecting the two points at700 ms and 3,000 ms after the cue onset). The starting level (mean ±s.d.) differed between the matched- and conflicting-bimodalconditions (0.52 ± 0.29 versus �0.38 ± 0.23; Wilcoxon signed-ranktest; P o 0.001), whereas the slopes did not differ significantly(�0.0016 ± 0.26 versus �0.095 ± 0.27 s�1). The starting levels didnot differ significantly between the low- and high-reward conditions(�0.14 ± 0.27 versus �0.0054 ± 0.32), although the slopes differed(�0.26 ± 0.16 versus 0.22 ± 0.20 s�1; Wilcoxon signed-rank test;P o 0.001). These results indicate that the bimodal effect was presentfrom the beginning of a task trial and was maintained up to the timepoint before the reward delivery, whereas the effect of varying thereward value developed gradually in the later phase.

Relations between neural responses and behavioral accuracy

Next, we examined the relationship between behavioral accuracyand these neural responses (Supplementary Methods). Behavioralresponses were subdivided into four groups: correct go responses(hit), wrong go responses (false alarm), correct no-go responses (reject)and wrong no-go responses (miss). In the auditory unimodal sessions,15.4% of the go trials resulted in a miss, and 17.0% of the no-go trialsresulted in a false alarm. The conflicting-bimodal condition increasedthe incidence of both false alarms (25.4%; one-way ANOVA: F2,381 ¼128.212, P o 0.001; with post hoc Bonferroni test, P o 0.001) andmisses (23.7%; one-way ANOVA: F2,381 ¼ 182.335, P o 0.001; withpost hoc Bonferroni test, P o 0.001), whereas the effects of the high-and low-reward values increased the incidence of false alarms (27.8%;one-way ANOVA: F2,381 ¼ 96.346, Po 0.001; with post hoc Bonferronitest, P o 0.001) and misses (24.5%; one-way ANOVA: F2,381 ¼253.541, P o 0.001; with post hoc Bonferroni test, P o 0.001),respectively. We carried out a nonparametric signal detection analysisand calculated A¢ and B¢¢ for each condition17,18 (Fig. 5). A¢ inthe matched- and conflicting-bimodal conditions was higher and

Tone (L)

Time (ms)

Pre-extinction

Pre-e

xtinc

tion

Extinction 1

Extinc

tion

1

Extinc

tion

2

Extinction 2

20 s

pike

s s–1

Tone (L) + light (L)

Light (L)

–200 200 400 600 800 1,0000

Trial block

2.5

* **

2.0

1.5

1.0

0.5

0.0

Nor

mal

ized

firin

g ra

te

Tone

(L)

Sum (p

redic

ted)

a b

Figure 3 Bimodal modulation in extinction trials. (a) Response examples of

a neuron. Red, dotted magenta and dotted black lines indicate SDFs during

pre-extinction trials under the matched-bimodal, auditory-unimodal and

visual-unimodal conditions, respectively. Light blue and dark blue lines

indicate SDFs under the matched-bimodal condition during extinction

1 trials (without reward contingency and with behavioral responses) and

during extinction 2 trials (with neither reward contingency nor behavioral

responses), respectively. (b) The histograms summarize neural responses

(mean ± s.d.) under each condition, normalized and averaged across seven

neurons that were tested in extinction. The colors correspond to those

in a. *P o 0.05.

60

50

40

30

20

Time (s)

High reward

Low reward

00 1 2 3 4

Time (s)0 1 2 3 4

Time (s)0 1 2 3 4

Time (s)

Matched-bimodal

Uni

mod

al s

ubtr

acte

d di

ffere

nce

in n

orm

aliz

ed a

ctiv

ities

Conflicting-bimodal

High reward

Low reward

0 1 2 3 4

Spi

kes

s–1

10

60

50

40

30

20

0

Spi

kes

s–1

10

60

50

40

30

20

0

Spi

kes

s–1

10

1.5

1.0

0.5

0.0

–0.5

–1.0

–1.5

a c db

Figure 4 The effects of bimodal cues and reward values on the activity of neurons in the auditory thalamus. (a–c) A response example of a neuron in the

posterior intralaminar nucleus. Rastergrams and SDFs were generated for (a) unimodal and (b) bimodal cue stimuli and for (c) different reward values. The

cue stimuli were presented for 2 s (black bars), and a 1-s delay period was imposed before reward delivery (arrowheads). The colors, kinds of lines and panel

positions in a and b are the same as those in Figure 2. In c, data on the trials of auditory stimulus from the left are shown. (d) Time course of response

enhancement (solid line) or suppression (dotted line) by different bimodal presentations (black) or reward values (gray). The ordinate shows the differencebetween the normalized population activity of each condition and the activity recorded in the unimodal condition with a moderate reward (n ¼ 72).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

lower, respectively, than that in the unimodal condition (one-wayANOVA: F4,635 ¼ 109.164, P o 0.001; with post hoc Bonferroni test,Po 0.001), whereas A¢ in the high- and low-reward conditions was notsignificantly different from that in the unimodal condition (Fig. 5a). B¢¢in the high- and low-reward conditions was lower and higher, respec-tively, than that in the unimodal condition (one-way ANOVA: F4,608 ¼91.876, P o 0.001; with post hoc Bonferroni test, P o 0.001), whereasB¢¢ in the matched- and conflicting-bimodal conditions was notsignificantly different from that in the unimodal condition (Fig. 5b).These results indicate that the bimodal condition affects the discrimin-ability of the stimulus, whereas the reward value condition affects therat’s response criteria. Typical neural responses for one cell in the errortrials are shown for the conflicting-bimodal condition (Fig. 6a) and forthe condition with variable reward values (Fig. 6b). In both cases, thelater responses declined in the miss trials and increased in the false-alarm trials, whereas the early components did not change. To evaluatethe relationship between the neural responses and behavioral accuracyacross the whole set, we also plotted the normalized neural activities inthe error trials against those of the same neurons in the correct trials(Fig. 6c). The firing rates in the late phase, but not in the early phase,were found to correlate well with the behavioral performance.

Relations between neural responses and reaction time

Next, we examined the relationship between these neural responses andreaction time (Supplementary Methods). The reaction time (thelatency to the first lick) was shorter in the matched-bimodal condition(mean ± s.d.: 108.0 ± 39.0 ms) and longer in the conflicting-bimodalcondition (198.4 ± 45.5 ms) when compared with that in the unimodalcondition (142.7 ± 42.8 ms; one-way ANOVA: F2,3,771 ¼ 1,417.685,P o 0.001; with post hoc Bonferroni test, P o 0.001). Also, it wasshorter in the high-reward condition (97.8 ± 38.2 ms) and longer in thelow-reward condition (214.0 ± 48.1 ms) when compared with that inthe medium-reward unimodal condition (142.7 ± 42.8 ms; one-wayANOVA: F2,3,747 ¼ 2,306.754, Po 0.001; with post hoc Bonferroni test,P o 0.001). These results support the view that the matched-bimodalpresentation and larger reward value should allow an animal to assessthe situation and respond more quickly. The early response of the sameneuron (Fig. 4a–c), based on the spike count during the first 200 msafter the cue onset, showed a clear inverse relationship with the reactiontime in the matched- and conflicting-bimodal conditions (Fig. 7a;y ¼ 56.6 � 235.4x, r2 ¼ 0.626). This early neural response showed norelationship, however, to the reaction time in either the low- or high-reward condition (Fig. 7a; y ¼ 17.9 � 3.04x, r2 ¼ 0.015). In contrast,the late neural response, based on the spike count over the period of2,500 to 3,000 ms after the cue onset (Fig. 7b), showed a clear inverserelationship with reaction time both in the bimodal conditions(y ¼ 43.0 � 130.3x, r2 ¼ 0.766) and in the reward value conditions(y ¼ 38.4 � 107.5x, r2 ¼ 0.753).

To evaluate the relationship between neural responses and reactiontime across the whole population of these cells (n ¼ 72), we calculatedthe linear regression of the normalized neural activities in the earlyand late phases for each neuron on the reaction time of the rat(similar to the manner used for Fig. 7a,b). Then we plotted slopes ofthe early phase against those of the late phase for the bimodal andreward value conditions (Fig. 7c). This scatter plot showed that, inmost neurons, the effect of the bimodal presentation was apparentin the early and late phases, whereas the effect of changes in rewardvalue appeared only in the late phase. Furthermore, to track the timecourse of the regression slopes in fine time resolution at the populationlevel, we calculated the slope of the linear regression of the normalizedspike density function (SDF) value at successive times (1-ms bins) onthe reaction time and plotted the average of the instantaneous regres-sion slopes across the 72 neurons as a function of time in a trial(Fig. 7d). The time course clearly indicated that the bimodal effectwas apparent from the beginning of the early responses, whereas the

1.0 * *

*

0.9

0.8

0.4

0.0

–0.4

–0.8

–1.2

0.8

A′

B′′

0.7

Low re

ward

Unimod

al

Mat

ched

-bim

odal

Conflic

ting-

bimod

al

High re

ward

Low re

ward

Unimod

al

Mat

ched

-bim

odal

Conflic

ting-

bimod

al

High re

ward

*

a b

Figure 5 Signal detection analysis on behavioral performance. (a,b) The bars

(mean ± s.d.) indicate (a) A¢ and (b) B¢¢ measured under unimodal (white),

matched- or conflicting-bimodal (black) and high- or low-reward (gray bars)conditions. *P o 0.001.

30

20

10Spi

kes

s–1

0

30

20

10

Spi

kes

s–1

00 1 2

Time (s)3 4 0 1 2

Time (s)3 4

Activities in correct trials0.5 1.0 2.0 2.51.5

a b

Act

iviti

es in

err

or tr

ials

2.5

2.0

1.5

1.0

0.5

0.0

c

Miss

Hit/miss (early)Reject/false alarm (early)Hit/miss (late)Reject/false alarm (late)

HitFalse alarmReject

Figure 6 Correlation between neural responses and accuracy. (a,b) SDF and rastergrams of a neuron (Fig. 4) in miss (black solid line) and false-alarm (gray

solid line) trials under the conditions with (a) bimodal presentation or (b) altered reward value. The dotted lines are SDFs in the correct trials requiring go (hit;

black) and no-go (reject; gray) responses. (c) Analysis of error trials from the population data (n ¼ 151). Scatter plots in each error type and each phase (earlyand late phases were defined as 0–200 ms and 2,500–3,000 ms after cue onset, respectively). Only the data recorded in the sessions in which the rat

produced four or more errors were used for this analysis.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

reward value effect appeared later and increased gradually, peakingbefore the reward was delivered.

Functional topography

Finally, we investigated the functional anatomy of these responses. Interms of the thalamocortical connections, the auditory thalamus iscategorized into two parallel core and belt regions19–23 (Supplemen-tary Fig. 2). Almost all neurons with bimodal or reward modulation(such as those in Figs. 2 and 4) were located in the belt region of theauditory thalamic nuclei: the dorsal and medial part of the medialgeniculate nucleus, the suprageniculate nucleus and the posterior intra-laminar nucleus. In contrast, neurons in the ventral medial geniculatenucleus seldom showed bimodal or reward modulation (Supplemen-tary Fig. 2).

DISCUSSION

Our data provide new insights on cross-modal information processingin the brain. Rewiring the retinal projection in neonatal ferrets inducesvisual signals into the auditory thalamus, which functions in theauditory cortex to mediate visual behavior24. Using an auditory spatialdiscrimination task that embedded visual cues, we obtained for the firsttime physiological evidence that visual signals drive the early phase ofauditory thalamic responses, even in intact adult rats. Some imagingstudies indicate that the posterior thalamic region may be associatedwith cross-modal binding1,4,25. Because of the limited spatial resolutionof imaging studies, it remains unclear which parts of the posteriorthalamic region participate in cross-modal binding. Epicortical map-ping of auditory- or visual-evoked potentials combined with retrogradetracing indicates that polysensory interactions may occur at localizedareas of the sensory thalamus5. The polymodal zone, in which jointauditory and visual stimulation evokes supra-additive responses (thatis, responses greater than the sum of the unimodal responses),primarily covered the secondary sensory and association cortices.These areas received inputs exclusively from the belt region of theauditory thalamus (suprageniculate, medial part of the medial genicu-late, dorsal part of the medial geniculate and posterior intralaminarnuclei) but not from the core region (ventral medial geniculatenucleus). In addition, the evoked responses in the polymodal zonehad a unique positive and negative waveform that is characteristic ofthalamocortical activation26. Electrical stimulation of the posteriorintralaminar nuclei increases cortical gamma band oscillations27,which appear in cross-modal interactions28. The functional distribu-tion (Supplementary Fig. 2), together with these previous findings,

is in accord with the possibility that the early components in the beltregions of the auditory thalamic responses instruct the auditorycortices with cross-modal information.

How cross-modal activities are generated in the auditory thalamusremains unclear. At least three possibilities should be considered. Thefirst is the influence of the superior colliculus. Only the belt region ofthe auditory thalamus receives inputs from the superior colliculus, avision-dominant site21,29. The neurons in the deep layer of the superiorcolliculus also show supra-additive responses to multisensory stimula-tion30. The manner of cross-modal interaction in the auditory thala-mus differs, however, from that in the superior colliculus: in the latter,neurons respond predominantly to visual cues31, whereas most neu-rons in the former seldom respond to visual cues alone. The visualimpact on the auditory thalamic responses observed in the presentstudy was apparent solely when the visual cue was presented togetherwith the auditory cue. Therefore, the cross-modal interactions in theauditory thalamus may not simply reflect the activity in the superiorcolliculus, although activity in the auditory thalamus might bemodulated by visual input from the superior colliculus. The secondpossibility is the involvement of corticofugal modulation on thalamicactivities. Previous studies indicate that unimodal sensory cortices maybe affected by the back-projections from multisensory cortices7,8,32.Thus the thalamic activities may be modulated by multisensory corticalactivities through the feedback projections from the sensory cortices.The third possibility is the influence of the inferior colliculus. The beltregion of the auditory thalamus receives inputs from the pericentralnucleus (belt region) of the inferior colliculus21. Only the pericentralnucleus of the inferior colliculus receives direct retinal inputs, whereasthe central nucleus (core region) of the inferior colliculus does not33.Further studies will be necessary to elucidate how the visual signals sentto the auditory thalamus are gated or accessed.

The present report also relates the neural basis of cross-modal signalsexplicitly to behavioral performance. A number of animal experimentshave focused on the superior colliculus and cortices, which have supra-additive neural activities evoked by multisensory stimulation3,31,34,35

and mediate multisensory behavior12,13,36,37. Therefore, we also exam-ined the relationship between thalamic responses and behavioralresponses. We found that the auditory thalamus had early responsesas well as subsequent late responses, which were found to be goodpredictors of the behavioral results. The late components were modu-lated by bimodal presentation and reward value. Bimodal presentationimproves the intelligibility of the stimulus, whereas changes in rewardvalue bias the subject’s behavioral responses14,18,38,39 (Fig. 5). Such

a b c d40

30

20

10

00

BimodalReward

–30 –20

–20

–10

–10

10

LS (s–1)

ES (s–1)

100 200Reaction time (ms)

300 0

0

–5

–10 Bimodal effectReward effect

0 1 2Time (s)

3 4–15

100 200Reaction time (ms)

300

Spi

kes

s–1

40

30

20Unimodal /medium rewardMatched-bimodalConflicting-bimodalHigh rewardLow reward

10

0

Spi

kes

s–1

Reg

ress

ion

slop

e (s

–1)

Figure 7 Correlation between neural responses and reaction time. (a,b) Scatter plots (n ¼ 36 trials) of the reaction time and the firing rates of a neuron (Fig. 4)

in the (a) early phase and (b) late phase (defined as in Fig. 6c). The fitted lines show the regression slopes of the bimodal (black line; for filled triangles and

open circles) and rewarding (gray line; for open triangles and circles) effects. (c) Scatter plots (n ¼ 72 neurons) showing the relations between early slope (ES;

abscissa) and late slope (LS; ordinate) under the bimodal (black) and variable-reward (gray) conditions. ES and LS are the regression slopes for normalized

neural firing rates versus reaction time in the early and late response phases, respectively. (d) Time course of the averaged regression slope for normalized SDF

value versus reaction time, calculated from the population data (n ¼ 72 neurons).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

effects probably caused differences between the two conditions in thedevelopment of the late gradual component (Fig. 4d). Both bimodalpresentation and reward value are powerful influences on the decisionprocess, however, which determines the speed and accuracy of ananimal’s behavioral response14,18,25. Notably, only the belt region of theauditory thalamus bypasses the sensory cortices and communicatesdirectly with other brain areas, such as the basal ganglia andfrontal cortex40–44, that contribute to the decision process14,15,39,45

(Supplementary Fig. 2). These areas, including the posterior thalamicregion, are important in anticipatory behaviors that link sensory andmotor events46,47. Patients with posterior thalamic lesions respond lesscorrectly or more slowly to a contralesional target than to an ipsile-sional target48,49. These findings, when taken together, indicate that theauditory thalamus may provide a shortcut to the neural network thatunderlies the selection of an optimal behavior, thereby allowingbimodal exogenous factors to advance the endogenous decision pro-cesses and in turn to produce a quick response. Also, the schemeproposed here revises the traditional view of the sensory thalamicfunction, which is restricted to feeding modality-specific sensorysignals to the sensory cortex.

METHODSThe experimental procedures were similar to those used previously16, and the

experiments were carried out in accordance with both the US National

Institutes of Health Guide for the Care and Use of Laboratory Animals and

the Guiding Principles for the Care and Use of Animals in the Field of

Physiological Sciences recommended by the Physiological Society of Japan.

Rat preparation. Twelve male Wistar rats, weighing 270–330 g, were anesthe-

tized (40 mg per kg sodium pentobarbital, i.p.) and had a cranioplastic cap

attached to the skull. This made it possible to fix the rat’s head painlessly in the

stereotaxic device. A hole for chronic recording was drilled through the

cranioplastic cap and the underlying skull.

Behavioral tasks. The sessions proceeded in a block design. Each block consisted

of more than 18 trials. When a block was altered, the initial four trials were used

to acclimatize the rat to the new situation, and the data obtained from these four

initial trials were discarded from the analysis. Within each block, a left or right

sensory cue was presented in a pseudorandom order. In the extinct-

ion trials, the reward, which was previously contingent on the cue, was omitted.

Electrophysiology. Single units were recorded with glass-insulated tungsten

microelectrodes (Z ¼ 1.0–1.5 MO at 1,000 Hz). At the start of each recording

session, we first searched for auditory neural responses by clapping and thereby

localized the margins of the medial geniculate body. Based on the coordinates

derived from this localizing procedure, the microelectrode was then inserted

stereotaxically stepwise with a micromanipulator into various parts of the

posterior thalamus. Extracellular discharges of single neurons were recorded

using conventional recording procedures.

Data analysis. Spike trains were smoothed by convolution with a Gaussian

kernel (s ¼ 40 ms) to obtain SDFs. Spontaneous activity was defined as the

mean discharge rate during the 500 ms just preceding cue onset. The early

response was defined as the number of spike counts in the first 200 ms after cue

onset. The late response was defined as the number of spike counts in the

500 ms between 2,500 and 3,000 ms after cue onset. A neuron was classified as

responsive to any cue stimuli if the difference in neural activity in the three

periods (spontaneous activity, early response and late response) was significant

(one-way ANOVA; P o 0.01). A bimodal modulation (enhancement or

depression) in neural responses was defined as a significant increase or decrease

in firing rates in the bimodal condition when compared with those in the

auditory unimodal condition (one-way ANOVA; P o 0.01)50. A neuron was

classified as modulated by reward value if the difference in the firing rates in

three conditions (low, medium and high reward) was significant (one-way

ANOVA; P o 0.01). To compare the normalized data in the different

conditions, we carried out a nonparametric estimation using the Wilcoxon

signed-rank test (P o 0.05). To normalize the firing rates across neurons, we

first averaged the early response to the auditory cue on the left and then used

this mean as a reference value for each neuron. Next, we divided the averaged

response of each neuron in the different sensory-cue or reward value conditions

with respect to the reference value, and this quotient was defined as a

normalized activity in each condition.


ACKNOWLEDGMENTSWe thank F. Miles and M. Shidara for comments on the manuscript andS. Kitazawa and T. Kitamura for discussions. This work was partly supported bySpecial Coordination Funds for Promoting Science & Technology (Y.K.) and byGrants-in-Aid for Scientific Research nos. 17021015 and 17500273 (R.T.) fromthe Japanese Ministry of Education, Culture, Sports, Science and Technology.


Received 23 February; accepted 28 July 2005


1. Bushara, K.O. et al. Neural correlates of cross-modal binding. Nat. Neurosci. 6,190–195 (2003).

2. Shams, L., Kamitani, Y., Thompson, S. & Shimojo, S. Sound alters visual evokedpotentials in humans. Neuroreport 12, 3849–3852 (2001).

3. Wallace, M.T., Ramachandran, R. & Stein, B.E. A revised view of sensory corticalparcellation. Proc. Natl. Acad. Sci. USA 101, 2167–2172 (2004).

4. Sekiyama, K., Kanno, I., Miura, S. & Sugita, Y. Auditory-visual speech perceptionexamined by fMRI and PET. Neurosci. Res. 47, 277–287 (2003).

5. Barth, D.S., Goldberg, N., Brett, B. & Di, S. The spatiotemporal organization of auditory,visual, and auditory-visual evoked potentials in rat cortex. Brain Res. 678, 177–190(1995).

6. Calvert, G.A. et al. Activation of auditory cortex during silent lipreading. Science 276,593–596 (1997).

7. Macaluso, E., Frith, C.D. & Driver, J. Modulation of human visual cortex by crossmodalspatial attention. Science 289, 1206–1208 (2000).

8. Schroeder, C.E. et al. Anatomical mechanisms and functional implications of multi-sensory convergence in early cortical processing. Int. J. Psychophysiol. 50, 5–17(2003).

9. McGurk, H. & MacDonald, J. Hearing lips and seeing voices. Nature 264, 746–748(1976).

10. Driver, J. Enhancement of selective listening by illusory mislocation of speech soundsdue to lip-reading. Nature 381, 66–68 (1996).

11. Stein, B.E., Meredith, M.A., Huneycutt, W.S. & McDade, L. Behavioral indicesof multisensory integration: Orientation to visual cues is affected by auditory stimuli.J. Cogn. Neurosci. 1, 12–24 (1989).

12. Wilkinson, L.K., Meredith, M.A. & Stein, B.E. The role of anterior ectosylvian cortexin cross-modality orientation and approach behavior. Exp. Brain Res. 112, 1–10(1996).

13. Jiang, W., Jiang, H. & Stein, B.E. Two corticotectal areas facilitate multisensoryorientation behavior. J. Cogn. Neurosci. 14, 1240–1255 (2002).

14. Glimcher, P.W. The neurobiology of visual-saccadic decision making. Annu. Rev.Neurosci. 26, 133–179 (2003).

15. Schultz, W. Multiple reward signals in the brain. Nat. Rev. Neurosci. 1, 199–207(2000).

16. Komura, Y. et al. Retrospective and prospective coding for predicted reward in thesensory thalamus. Nature 412, 546–549 (2001).

17. Grier, J.B. Nonparametric indexes for sensitivity and bias: computing formulas. Psychol.Bull. 75, 424–429 (1971).

18. Lovelace, C.T., Stein, B.E. & Wallace, M.T. An irrelevant light enhances auditorydetection in humans: a psychophysical analysis of multisensory integration in stimulusdetection. Brain Res. Cogn. Brain Res. 17, 447–453 (2003).

19. Winer, J.A., Kelly, J.B. & Larue, D.T. Neural architecture of the rat medial geniculatebody. Hear. Res. 130, 19–41 (1999).

20. Rauschecker, J.P., Tian, B., Pons, T. & Mishkin, M. Serial and parallel processing inrhesus monkey auditory cortex. J. Comp. Neurol. 382, 89–103 (1997).

21. Kaas, J.H., Hackett, T.A. & Tramo, M.J. Auditory processing in primate cerebral cortex.Curr. Opin. Neurobiol. 9, 164–170 (1999).

22. Hu, B., Senatorov, V. & Mooney, D. Lemniscal and non-lemniscal synaptic transmissionin rat auditory thalamus. J. Physiol. (Lond.) 479, 217–231 (1994).

23. Rauschecker, J.P. & Tian, B. Mechanisms and streams for processing of ‘‘what’’and ‘‘where’’ in auditory cortex. Proc. Natl. Acad. Sci. USA 97, 11800–11806(2000).

24. von Melchner, L., Pallas, S.L. & Sur, M. Visual behaviour mediated by retinal projectionsdirected to the auditory pathway. Nature 404, 871–876 (2000).

25. Bushara, K.O., Grafman, J. & Hallett, M. Neural correlates of auditory-visual stimulusonset asynchrony detection. J. Neurosci. 21, 300–304 (2001).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

26. Brett-Green, B., Fifkova, E., Larue, D.T., Winer, J.A. & Barth, D.S. A multisensory zone inrat parietotemporal cortex: intra- and extracellular physiology and thalamocorticalconnections. J. Comp. Neurol. 460, 223–237 (2003).

27. Barth, D.S. & MacDonald, K.D. Thalamic modulation of high-frequency oscillatingpotentials in auditory cortex. Nature 383, 78–81 (1996).

28. Bhattacharya, J., Shams, L. & Shimojo, S. Sound-induced illusory flash perception: roleof gamma band responses. Neuroreport 13, 1727–1730 (2002).

29. Linke, R., De Lima, A.D., Schwegler, H. & Pape, H.C. Direct synaptic connections ofaxons from superior colliculus with identified thalamo-amygdaloid projection neurons inthe rat: possible substrates of a subcortical visual pathway to the amygdala. J. Comp.Neurol. 403, 158–170 (1999).

30. Stein, B.E. Neural mechanisms for synthesizing sensory information and producingadaptive behaviors. Exp. Brain Res. 123, 124–135 (1998).

31. Stein, B.E., Jiang, W., Wallace, M.T. & Stanford, T.R. Nonvisual influences on visual-information processing in the superior colliculus. Prog. Brain Res. 134, 143–156(2001).

32. Weeks, R. et al. A positron emission tomographic study of auditory localization in thecongenitally blind. J. Neurosci. 20, 2664–2672 (2000).

33. Itaya, S.K. & Van Hoesen, G.W. Retinal innervation of the inferior colliculus in rat andmonkey. Brain Res. 233, 45–52 (1982).

34. Wallace, M.T., Meredith, M.A. & Stein, B.E. Multisensory integration in the superiorcolliculus of the alert cat. J. Neurophysiol. 80, 1006–1010 (1998).

35. Hikosaka, K., Iwai, E., Saito, H. & Tanaka, K. Polysensory properties of neurons in theanterior bank of the caudal superior temporal sulcus of the macaque monkey.J. Neurophysiol. 60, 1615–1637 (1988).

36. Bell, A.H., Corneil, B.D., Meredith, M.A. & Munoz, D.P. The influence of stimulusproperties on multisensory processing in the awake primate superior colliculus.Can. J. Exp. Psychol. 55, 123–132 (2001).

37. Frens, M.A. & Van Opstal, A.J. Visual-auditory interactions modulate saccade-related activity in monkey superior colliculus. Brain Res. Bull. 46, 211–224(1998).

38. Ernst, M.O. & Bulthoff, H.H. Merging the senses into a robust percept. Trends Cogn. Sci.8, 162–169 (2004).

39. Kawagoe, R., Takikawa, Y. & Hikosaka, O. Expectation of reward modulates cognitivesignals in the basal ganglia. Nat. Neurosci. 1, 411–416 (1998).

40. Romanski, L.M. et al. Dual streams of auditory afferents target multiple domains in theprimate prefrontal cortex. Nat. Neurosci. 2, 1131–1136 (1999).

41. Kobler, J.B., Isbey, S.F. & Casseday, J.H. Auditory pathways to the frontal cortex of themustache bat, Pteronotus parnellii. Science 236, 824–826 (1987).

42. Takada, M., Itoh, K., Yasui, Y., Sugimoto, T. & Mizuno, N. Topographical projections fromthe posterior thalamic regions to the striatum in the cat, with reference to possible tecto-thalamo-striatal connections. Exp. Brain Res. 60, 385–396 (1985).

43. LeDoux, J.E., Farb, C.R. & Romanski, L.M. Overlapping projections to the amygdala andstriatum from auditory processing areas of the thalamus and cortex.Neurosci. Lett. 134,139–144 (1991).

44. Shammah-Lagnado, S.J., Alheid, G.F. & Heimer, L. Efferent connections of the caudalpart of the globus pallidus in the rat. J. Comp. Neurol. 376, 489–507 (1996).

45. Dalley, J.W., Cardinal, R.N. & Robbins, T.W. Prefrontal executive and cognitive functionsin rodents: neural and neurochemical substrates. Neurosci. Biobehav. Rev. 28,771–784 (2004).

46. Brunia, C.H. & van Boxtel, G.J. Wait and see. Int. J. Psychophysiol. 43, 59–75 (2001).47. Rektor, I., Kanovsky, P., Bares, M., Louvel, J. & Lamarche, M. Event-related potentials,

CNV, readiness potential, and movement accompanying potential recorded from poster-ior thalamus in human subjects. A SEEG study. Neurophysiol. Clin. 31, 253–261(2001).

48. Rafal, R.D. & Posner, M.I. Deficits in human visual spatial attention following thalamiclesions. Proc. Natl. Acad. Sci. USA 84, 7349–7353 (1987).

49. Harvey, M., Olk, B., Muir, K. & Gilchrist, I.D. Manual responses and saccades in chronicand recovered hemispatial neglect: a study using visual search. Neuropsychologia 40,705–717 (2002).

50. Meredith, M.A. & Stein, B.E. Interactions among converging sensory inputs in thesuperior colliculus. Science 221, 389–391 (1983).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Neural codes for perceptual discrimination in primarysomatosensory cortex

Rogelio Luna1, Adrian Hernandez1, Carlos D Brody2 & Ranulfo Romo1

We sought to determine the neural code(s) for frequency discrimination of vibrotactile stimuli. We tested five possible candidate

codes by analyzing the responses of single neurons recorded in primary somatosensory cortex of trained monkeys while they

discriminated between two consecutive vibrotactile stimuli. Differences in the frequency of two stimuli could be discriminated

using information from (i) time intervals between spikes, (ii) average spiking rate during each stimulus, (iii) absolute number of

spikes elicited by each stimulus, (iv) average rate of bursts of spikes or (v) absolute number of spike bursts elicited by each

stimulus. However, only a spike count code, in which spikes are integrated over a time window that has most of its mass in the

first 250 ms of each stimulus period, covaried with behavior on a trial-by-trial basis, was consistent with psychophysical biases

induced by manipulation of stimulus duration, and produced neurometric discrimination thresholds similar to behavioral

psychophysical thresholds.

Investigations in several sensory systems have shown how neuralactivity represents the physical parameters of sensory stimuli in boththe periphery and central areas of the brain. This knowledge has pavedthe way for new questions that are more closely related to cognitiveprocessing. For example, are the neural representations of sensorystimuli related to perception? In this respect, it has been shown thatquickly adapting neurons of the primary somatosensory cortex (S1) aredirectly involved in frequency discrimination of vibrotactile stimuli1,2.But exactly which components of these neurons’ stimulus-evokedactivity are associated with the discrimination process is not known.Most of the quickly adapting neurons of S1 show phase-lockedresponses to the periodic mechanical sinusoids, in the form of singlespikes or bursts of spikes3–6. This suggests that discrimination could bebased on observing the temporal intervals between responses to eachstimulus period3. However, about one-third of the quickly adaptingneurons in S1 also have a firing rate, averaged over the duration of astimulus, that is a function of the periodic stimulus frequency, withhigher firing rates in response to higher stimulus frequencies5,6. Thus,an observer of the stimulus-evoked activity in the quickly adaptingneuronal population of S1 could discriminate between two periodicvibrotactile stimuli either by comparing the precise temporal intervalsbetween spikes or by comparing the overall spike rates elicited by thetwo stimuli7,8.

Previously, we have computed neurometric thresholds6 for bothspike periodicity–based codes and overall firing rate–based codes; wefound that periodicity-based neurometric thresholds were far smallerthan psychometric thresholds. In contrast, firing rate–based neuro-metric thresholds were similar to psychometric thresholds. This resultfavored firing rate over spike timing as the neural code for perception of

vibrotactile stimuli5,6. Notably, monkeys are also able to discriminatethe mean frequency of aperiodic stimuli, which lack any temporalregularity1,5,6. It is assumed that under aperiodic stimulation, discri-mination would be based on a comparison of overall spike rates5,6.Monkeys could use different coding strategies for periodic versusaperiodic stimuli, but a parsimonious account covering boththe periodic and aperiodic cases once again favored firing rate asthe neural code.

There are, however, some further unexplored possibilities. Forexample, quickly adapting neurons of S1 typically respond to eachstimulus pulse with a discrete burst of spikes. Encoding of vibrotactilestimuli could therefore be based on the number or rate of events, whereeach event is defined as a burst instead of being defined as a single spike.An observer counting bursts would obtain a good estimate of the countof stimulus pulses, and this estimate would be independent of varia-bility in the number of spikes fired in response to each pulse. Indeed,there is experimental evidence suggesting that bursting activity couldefficiently encode the stimulus features9–12. But whether burstingactually contributes directly to psychophysical behavior is notknown. Finally, the temporal window on which vibrotactile discrimi-nation is based has not been determined. In our previous experiments,stimulus periods were always 500 ms long. Under those conditions, theuse of a code based on counting events and the use of a code based onthe rate of events could not be distinguished.

To distinguish between all these alternatives, we conducted newcombined psychophysical and neurophysiological experiments in thevibrotactile discrimination task. We reasoned that if an observer usesfiring rate, bursting rate or a measure of periodicity, then increases ordecreases in the duration of either of the two stimuli used in each trial


1Instituto de Fisiologıa Celular, Universidad Nacional Autonoma de Mexico, 04510 Mexico, D.F., Mexico. 2Cold Spring Harbor Laboratory, Cold Spring Harbor, New York11724, USA. Correspondence should be addressed to R.R. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

of the task should not lead to a systematic bias in discrimination ineither of the two possible directions. (However, under this hypothesis,stimulus duration could affect the sensory signal-to-noise ratio andtherefore the psychometric threshold.) Alternatively, if the observeruses a strategy based on the total number of spikes or bursts fired inresponse to each stimulus, then manipulation of the stimulus durationshould systematically bias performance, with longer stimuli beingperceived as having been of higher frequency than they actually were.We found that when the duration of one of the two stimuli was changedby 50% relative to the other stimulus, monkeys indeed biased theirdiscrimination performance. Monkeys treated shortened stimuli as ifthe applied stimulus frequency had been slightly but significantly lowerthan it actually was; conversely, monkeys treated lengthened stimuli asif the applied frequency had been slightly butsignificantly higher than it actually was. Theseeffects were observed with both periodic andaperiodic stimuli. We sought an explanationfor these psychophysical biases by recordingquickly adapting neurons of S1 while themonkeys performed in variable–stimuluslength conditions. We found that the effectscan be qualitatively explained if one assumesthat the neural signal used by the observer tosolve the task is a weighted sum of eitherspikes or bursts over a time window thatconcentrates most of its weight within thefirst 250 ms of the stimulus but also has asmall tail in later parts of the stimulus. Finally,examining trial-by-trial covariations ofweighted counts of spikes and weightedcounts of bursts, we found that only theweighted count of spikes covaried with per-formance on a trial-by-trial basis.

RESULTS

Stimulus duration biases performance

Two monkeys (Macaca mulatta) were trainedin the vibrotactile discrimination task (Fig. 1).Initially, the monkeys were trained to discri-minate between pairs of periodic stimulusfrequencies of equal duration (500 ms) upto their psychophysical thresholds3,13. Wethen asked whether they could discriminatebetween pairs of aperiodic stimuli1,6. In each

of these two task conditions, and in separate trial blocks, monkeyscompared a second stimulus frequency that varied from trial to trial(range, 14–30 Hz) against a fixed first stimulus frequency (22 Hz), orthey compared a fixed second stimulus frequency (22 Hz) against avarying first stimulus frequency (range, 14–30 Hz). The differencesbetween the psychometric thresholds for the four cases (data notshown) were not significant (permutation test, n ¼ 1,000, P ¼ 0.32)14.

Experiments using fixed-duration stimuli cannot distinguishbetween a code based on the total number of events produced inresponse to each stimulus and a code based on the rate at which theevents are produced. However, if discrimination is based on a totalnumber of events over the stimulus periods, changing this number bychanging the stimulus period durations should affect discriminationperformance. In contrast, if we assume that the periodicity measureand the spike and burst rate measures are time invariant during astimulus, and one of these is the code used, then changing stimulusduration should not affect discrimination performance. We used blocksof trials in which, on a pseudorandom trial-by-trial basis, monkeyswere required to discriminate either between two vibrotactile stimuli ofequal duration (500 ms, control condition) or between two stimuliwhere one of the two stimuli was modified in length. In separate blocks,the modified stimulus either shortened by 50% to 250 ms, orlengthened by 50% to 750 ms. The other stimulus was kept at500 ms. As before, in some blocks of trials we kept f1 fixed at 22 Hzand varied f2 (Fig. 2a,b); in other blocks of trials, we varied f1 andkept f2 fixed (Fig. 2c,d). We compiled psychometric curves for thedifferent stimulus conditions and fit a logistic function to eachpsychometric curve.

Quantitative changes in the psychometric curves can be assessedthrough two parameters of the logistic fits (Fig. 2): (i) the psychometric

f1 f2PD KD KU PB

500 ms

Figure 1 Discrimination task. Sequence of events during each trial. The

mechanical probe is lowered (PD), indenting the glabrous skin of one digit

of the hand, and the monkey places its free hand (KD) on an immovable key.

The probe oscillates vertically at base frequency f1, and after a delay (3 s),

a second mechanical vibration is delivered at the comparison frequency (f2).The monkey releases the key (KU) after a delay (3 s) between f2 and KU

and presses one of two push buttons (PB) to indicate whether the second

stimulus was higher or lower. In separate stimulus sets, monkeys

discriminated between periodic (black line) and aperiodic (gray line)

stimulus frequencies.

f1 f2

Periodic Aperiodic

500 ms250 ms

750 ms

500 ms250 ms

750 ms500 ms250 ms

750 ms

500 ms250 ms

750 ms

f1 f2

f1 f2 f1 f2

14 18 22 26 300

1

p (f

2 hi

gher

than

f1)

p (f

1 hi

gher

than

f2)

f2 (Hz)

n = 22n = 12

n = 10

n = 77n = 42

n = 35n = 90n = 52

n = 38

n = 24n = 15

n = 9

PT = 4.41 ± 0.32PT = 3.79 ± 0.46

PT = 5.24 ± 0.61

X0 = 20.3 ± 0.54X0 = 19.7 ± 0.72

X0 = 22.6 ± 0.78

PT = 4.73 ± 0.16PT = 5.44 ± 0.34

PT = 5.56 ± 0.42

X0 = 21.6 ± 0.2X0 = 19.8 ± 0.42

X0 = 24.9 ± 0.39

PT = 5.74 ± 0.49PT = 6.14 ± 0.34

PT = 6.56 ± 0.52

X0 = 22.0 ± 0.18X0 = 19.3 ± 0.32

X0 = 26.3 ± 0.32

PT = 4.53 ± 0.31PT = 5.01 ± 0.29

PT = 6.78 ± 0.62

X0 = 21.3 ± 0.49X0 = 19.7 ± 0.64

X0 = 24.4 ± 0.93

14 18 22 26 300

1

p (f

2 hi

gher

than

f1)

p (f

1 hi

gher

than

f2)

f2 (Hz)

14 18 22 26 300

1

f1 (Hz)14 18 22 26 30

0

1

f1 (Hz)

a b

c d

Figure 2 Psychophysical performance during the manipulation of the duration of one of the two stimuli.

(a) Psychophysical performance when the duration of the second (f2) periodic stimulus increased (red) or

decreased (cyan) with respect to the first (f1) periodic stimulus. (b) The same as in a, but with aperiodic

stimuli. (c) Psychophysical performance when the duration of the periodic f1 stimulus increased (red) or

decreased (cyan) with respect to f2. (d) The same protocol as in c, but with aperiodic stimuli. n, number

of sessions for each stimulus condition; PT, psychometric thresholds for each stimulus condition (mean

± s.d.); X0, the frequency (mean ± s.d.) that corresponds to a y-axis value of 0.5 value in the logistic fit

for each stimulus condition.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

threshold, which measures the steepness of the logistic curve andrepresents the minimal change in stimulus frequency that produces areliable change in behavior, and (ii) the X0 value, which is a measure ofthe displacement of the curve along the horizontal axis and whichrepresents the frequency at which the stimulus being varied (f1 or f2) isindistinguishable from the stimulus kept fixed at 22 Hz. Values of X0greater than 22 Hz (rightward displacement of the psychometric curve)indicate that the subject judges the varying stimulus to have a lowerfrequency than its actual value. Values of X0 lower than 22 Hz (leftwarddisplacement of the psychometric curve) indicate that the subjectjudges the varying stimulus to have a higher frequency than its actualvalue. In extreme cases, large displacements of the logistic curvepreclude an accurate estimation of X0.

In general, we found that psychometric thresholds were only mini-mally affected by changes in stimulus duration (Fig. 2). The singleexception was the condition in which f2 was decreased in length by 50%with respect to f1 for aperiodic stimuli (Fig. 2b, cyan; permutation test,n¼ 1,000, Po 0.001)14. In contrast, X0 values were systematically andsignificantly affected by stimulus duration. Monkeys treated shortenedstimuli as if they had a frequency that was 2.3–4.3 Hz lower than theactual applied value (compare X0 values and psychometric curve shiftsin Fig. 2a–d for shortened stimuli (cyan) versus control-length stimuli(black); permutation test, n ¼ 1,000, P o 0.04)14. This bias effect wasobserved for both periodic (Fig. 2a,c) and aperiodic (Fig. 2b,d) stimuli.The opposite effect was observed when lengthened stimuli were used:monkeys treated lengthened stimuli as if they had a frequency that was0.6–2.7 Hz higher than the actual applied value (compare X0 values andpsychometric curve shifts in Fig. 2a–d for lengthened stimuli (red)versus control-length stimuli (black); permutation test, n ¼ 1,000,P o 0.04)14. Although the sign of the lengthening effect was the samein all conditions, the effect was strongest and statistically significant

only when the first stimulus, f1, was lengthened (red in Fig. 2c,d;Permutation test, n ¼ 1,000, P o 0.001)14. Again, the effect wasobserved for both periodic (Fig. 2c) and aperiodic (Fig. 2d)stimuli. The bias effects persisted over many trials despite the factthat monkeys were rewarded only for correct discrimination of theactual applied frequencies.

These results show that manipulations of the stimulus durationbiased psychophysical performance in a direction consistent with anaccumulative–event number code, such as integrating the number ofspikes or bursts over each stimulus. However, the magnitude of theeffect suggests that the accumulation of spikes or bursts does not occurequally over the entire stimulus period. For example, if firing rates wereconstant over the stimulus periods, and spikes were accumulated withequal weight over the entire period, then halving the stimulus lengthwould have halved X0 values with respect to the control (that is,shortened stimuli would have had X0 ¼ 11 Hz), and increasingstimulus lengths by 50% should have led to an increase of 50% in X0values (lengthened stimuli would have had X0¼ 33 Hz). Although thesign of the observed effect was in all cases consistent with the sign of thisprediction, the observed magnitude was much lower. In addition, theeffect was consistently stronger for shortening of stimuli than forlengthening of stimuli, which suggests that the initial part of thestimulus may have greater weight than the later part of the stimulusin determining discrimination performance. However, as there was adiscernible effect when stimuli were lengthened from 500 to 750 ms, thelater part of the stimulus must also have some influence on theperceptual process, though perhaps less influence than the earlierpart of the stimulus.

There are two distinct alternatives that could contribute towards agreater weighting for the initial part of this stimulus than for the laterpart. First, the response of S1 neurons, which are known to be causally

0 320

32

Res

pons

es(2

50 to

750

ms)

Periodicity

0 3000

300Firing rate

0 3000

300

Responses (0 to 250 ms)

No. of spikes

0 250

25No. of bursts

0 500

50Bursting rate

0 250 500 7500

1

Ave

rage

of s

lope

s

0 250 500 7500

4

0 250 500 7500

2.5

Time (ms)0 250 500 750

0

0.8

0 250 500 7500

1

a

b

Figure 3 Responses of S1 neurons as a function of time during the stimulus period. (a) Comparison of responses, using 750-ms-long stimuli, from the first

250 ms versus the last 500 ms for each of the five measures. Each point represents data for one stimulus condition from one neuron. In the first panel, only

the periodic stimulus trials are used, as no information can be calculated from the periodicity for aperiodic stimuli. Measures calculated during both periodic

and aperiodic stimuli are plotted in all other panels. Diagonal gray line is the expected value when the response is time invariant; black line is the value

obtained using the regression analysis in the data. (b) Response measure sensitivity, expressed as the linear regression slope (in units of response measure

change per 1 Hz change in stimulus frequency), averaged over all neurons, as a function of time for each of the five measures in panel a. Aperiodic stimulus

trials are not used in the first panel; both periodic and aperiodic trials are used in all other panels.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

related to perception of vibrotactile stimuli1,2, could adapt duringthe stimulus, being more sensitive to stimulus parameters duringthe earlier versus the later parts of the stimulus15–18. Alternatively, aprocess subsequent to S1 involved in perception could preferentiallyweight the S1 responses from the earlier, versus the later, parts ofthe stimulus. We first explore S1 neuron adaptation and then turn tothe second alternative.

Response adaptation in S1 does not explain performance

To investigate whether S1 neuron responses were differentially stimulus-sensitive during different portions of the stimulus, we recorded single,quickly adapting neurons in S1 (Table 1) while the monkeys discrimi-nated between the periodic or aperiodic stimulus pairs shown inFigure 2. We examined the sensitivity over time of five candidateneural codes: spike timing periodicity, overall spike rate, overall spikecount, overall burst rate and overall burst count. We took responses to750-ms-long stimuli and compared, for each measure, the first 250 msof the stimulus to the last 500 ms of the stimulus (Fig. 3a). We foundthat spike periodicity was time invariant (Fig. 3a) but that for all othermeasures, responses were significantly attenuated from the initial250 ms to the final 500 ms of the stimulus (black slope below grayline; permutation test on distribution of responses for initial versuslater part of the stimulus period, n ¼ 1,000; P o 0.01)14. We alsocompiled trials in which both the first (f1) and the second (f2) stimuluswere 500 ms long, and compared the response attenuation in the firststimulus with the response attenuation in the second stimulus (datanot shown). No significant differences were found (permutation test,

n ¼ 1,000; P ¼ 0.41)14. The initially stronger response to the stimuluscould carry information about the stimulus value itself, or it could be aresponse common to all stimulus values and therefore uninformative.For each measure, we calculated a linear regression of the measure as afunction of stimulus frequency (as in Fig. 4 below) and used the slopeof this linear regression to quantify the sensitivity of each measure tochanges in stimulus frequency. We did this for a range of time windowsall beginning at the start of the stimulus and extending into thestimulus in steps of 50 ms. We averaged these sensitivity slopes overneurons (Fig. 3b, lower panels) and found that the sensitivity of theperiodicity and burst rate measures remained roughly constant overtime during the stimulus. However, the sensitivity of the spike ratemeasure peaked approximately 200 ms after stimulus onset, suggestingthat spikes from this time period would be particularly informativewhen used for stimulus discrimination.

These results suggest that for some codes, differential stimulussensitivity in S1 neurons to different times during the stimulus couldcontribute to psychophysical biases induced by using varying stimuluslengths. But for each of the five candidate codes (periodicity, spikenumber, spike rate, burst number and burst rate), we must address theissue in a manner that allows quantitative comparison between theneuronal response measure and psychophysical results. We thereforeused our five candidate measures to compute neurometric thresholdsfrom S1 neuron responses6. These can be directly compared with themonkeys’ psychometric threshold6,19. In our initial neurometric calcu-lations, we weighted all parts of each stimulus equally, corresponding toan observer central to S1 that weights all parts of the stimulus equally.

14 300

36

Per

iodi

city

(H

z)

b

14 300

1

NT = 0.22X0 = 22NT = 0.22X0 = 22NT = 1X0 = 22

p (f

1 >

f2)

c

14 300

135F

iring

rat

e

14 300

1

NT = 3.5X0 = 21NT = 1.9X0 = 18

14 300

85

Num

ber

of s

pike

s

Frequency pairs

14 300

1

f1 (Hz)

14 300

20

Num

ber

of b

urst

s

14 300

1

NT = 3.5X0 = 21

14 300

30

Bur

stin

g ra

te

14 300

1

NT = 3.5X0 = 21NT = 3.9X0 = 21NT = 3.5X0 = 20

30:22

28:22

26:22

24:22

20:22

18:22

16:22

14:22

f1a f2 f1 f2 f1 f2

22 22 22 22 22 22 22 22 22 22

NT = 3.4X0 = 22

NT = 3.4X0 = 22

Figure 4 Responses of an area 1 neuron during the discrimination of the periodic stimulus set in Figure 2c. (a) Raster plots. Here, the duration of the first

stimulus (f1) decreased (50%, cyan) or increased (50%, red) with respect to the comparison frequency (500 ms, gray). Middle panel: responses duringdiscrimination of frequencies of equal duration. (b) Five different response measures, plotted as a function of stimulus frequency, during the manipulation of

the stimulus duration. (c) Neurometric curves using the different measures during each stimulus condition. NT, neurometric threshold; X0, the frequency that

gives a y-axis value of 0.5 value from the logistic fit.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

The results for one sample neuron are shown in Figure 4. As can beseen in the spike rasters of Figure 4a, this neuron was stronglyentrained by the periodic stimuli. Consistent with Figure 3a, theperiodicity measure for this neuron was similar for all stimulusdurations (leftmost panel, Fig. 4b). Similarly, the burst rate measurewas similar for all stimulus durations (rightmost panel, Fig. 4b).However, other measures were significantly affected by stimulusduration. Spike rates, calculated over each stimulus period length,were slightly but significantly higher for the short, 250-ms stimulusdurations (Fig. 4b, cyan), compared with spike rates for 500 and750 ms durations (Fig. 4b, black and red lines, respectively).Compared with control-stimulus period lengths, total spike orburst numbers were markedly higher for long, 750-ms stimulusdurations (Fig. 4b, red versus black) and markedly lower for 250-msstimulus durations, (Fig. 4b, cyan versus black). Similar trends wereobserved during discrimination of aperiodic stimuli (with the excep-tion of periodicity, which cannot be calculated when aperiodic stimuliare used). Thus, except for periodicity and burst rate, differentmeasures were considerably affected by the manipulation of thestimulus duration.

The neuronal response distributions elicited by the stimuli are thebasis for constructing neurometric functions, which can then becompared directly to the psychometric functions. Figure 4c showsneurometric curves, based on response distributions using differentmeasures, for the neuron of Figure 4a. As expected from Figure 4b, X0values based on the periodicity measure were not affected by the

stimulus duration. In contrast, neurometricX0 values calculated from a spike-rate codewere affected by stimulus duration but in adirection opposite to the effect found withpsychometric X0 values (compare Fig. 4c withFig. 2c). Neurometric X0 values for eitherspike number or burst number calculatedfrom a total-event-number code were affectedin the same direction as the psychometric X0values, but the magnitude of the neurometriceffect was far stronger than the psychometriceffect. Finally, neurometric X0 values calcu-lated from a burst-rate code were not signifi-cantly affected by stimulus duration. Thetrends shown for the example neuron ofFigure 4 were similar across the populationof recorded neurons (Fig. 5).

This result shows that when all parts ofthe stimulus response of S1 neurons areweighted equally, none of the five codes con-

sidered so far produces stimulus duration effects on neurometric curvesthat mimic those seen in the psychometric curves. Thus, none of thefive codes can explain the psychophysical biases produced by themanipulation of the stimulus duration.

Weighted integration of S1 responses explains performance

We therefore considered the alternative option: namely, that a processinvolved in perception but subsequent to S1 could differentially weightdifferent portions of the stimulus. This corresponds to carrying outneurometric calculations that assign different weights to differentportions of the stimulus response. Because the periodicity code didnot depend on the portion of time used to measure it, we did notinclude periodicity in this analysis, restricting ourselves to spike- andburst-based codes.

Having measured psychometric curves at three different stimuluslengths, we used a weighting window composed of three fixed-durationintervals corresponding to the three stimulus lengths (Fig. 6a). We nowassume that an observer central to S1 would use the same weightingwindow for all stimulus lengths. This makes event-rate and event-countcodes equivalent to each other: the relationship between event-rate andevent-count codes is defined by the weighting window in that rate canbe defined as the weighted event count divided by the area of theweighting window. In this sense, Figures 4 and 5 assume a rectangularweighting window whose width varies with the stimulus length and isalways as long as the stimulus. But here we turn to the assumption thatthe weighting window is constant over all stimulus lengths. The rate

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

14 22 300

1

Periodic stimuli

Aperiodic stimuli

Periodicity Firing rate No. of spikes No. of bursts Bursting rate

Periodicity Firing rate No. of spikes No. of bursts Bursting rate

p (c

ompa

rison

freq

uenc

y ju

dged

hig

her)

Comparison frequency [Hz]

p (c

ompa

rison

freq

uenc

y ju

dged

hig

her)

Comparison frequency [Hz]

Figure 5 Individual neurons’ neurometric

functions, using different measures, while

monkeys performed the discrimination task using

the stimulus sets in Figure 2. Color codes and

display format as in Figure 4c. Note the large

changes produced by manipulation of the stimulus

duration compared to the control neurometric

functions (black). These changes occurredfor both periodic and aperiodic stimulus

discriminations. Black, equal duration of the two

stimuli: 500 ms. Cyan, the duration of one of the

two stimuli decreased (50%) as compared with

the other. Red, the duration of one of the two

stimuli increased (50%) as compared with

the other.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

codes and weighted-event-count codes therefore differ only in a fixednormalization constant (the area of weighting window) and thusbecome equivalent to each other.

What should be the shape of this weighting window? Only therelative weights for the different stimulus portions are of importance indetermining shape. We therefore kept the weight of the initial 250-msportion fixed at an arbitrary value of 1 and systematically varied theweights for each of the other two intervals in the range [0, 1.9] in stepsof 0.1 (20 different values for each interval, leading to 400 different totalwindow shapes). Each of the possible window shapes was used tointegrate spikes or bursts, and the resulting values were considered asspike- or burst-response measures. Distributions of these responsemeasures were then used to compute neurometric functions as before,and the result was averaged over neurons. We computed the sum of

squared differences between the averaged neurometric curves and themonkeys’ psychometric curves. The window shapes that gave theminimal squared difference are shown in Figure 6a. Figures 6c and eshow the corresponding neurometric curves, averaged over neurons,for spike integration and for burst integration, respectively. Resultsbased on these integration measures show that such windows canindeed lead to psychophysical biases of a sign and magnitude compar-able to those found experimentally, with shortening (cyan) having agreater effect than lengthening (red). These results are also consistentwith the idea that the earlier components of the neuronal responseshave a greater impact than later components on the perceptual signalused to perform the task.

The weighting window in Figure 6a was based on three rectangularportions, but this was determined by data from three specific stimuluslengths. In general, our data are roughly consistent with window shapesthat have a gradual fall-off as a function of time, and any actual windowused by the subjects is unlikely to have a strictly stepwise shape. Wetherefore constructed a time window with a square shape for 230 msfollowed by an exponential fall-off, with a time constant of 60 ms(Fig. 6b). We used this window, placed beginning at stimulus onset, as aweighting window to integrate spikes or bursts. The results of using thiswindow (Fig. 6d and f, for spike integration and for burst integration,respectively, with the same window used for both) are essentiallysimilar to those obtained using the stepwise window of Figure 6a,with shortening (cyan) having a greater effect than lengthening(red). The window is placed at the peak of spike sensitivity tostimulus frequency.

0 250 500 750

0

1W

eigh

t

Time (ms)

p (c

ompa

rison

freq

uenc

y ju

dged

hig

her)

14 18 22 26 30

0

1

Comparison frequency (Hz)

Spikes

Bursts

Wd = 0.1 + e–(t – 223)/60

(230 < t < 750 ms)

0 250 500 750

0

1

Time (ms)

14 18 22 26 30

0

1

14 18 22 26 30

0

1

Spikes

Bursts

Comparison frequency (Hz)

n = 42n = 28n = 39

n = 87n = 83n = 76

14 18 22 26 30

0

1

a

c

e

b

d

f

Figure 6 An integration time window, for event-number codes, that produces

neurometric biases consistent with the psychophysical biases induced by

stimulus lengthening and shortening. (a) Weighting windows, W(t), composed

of three rectangular portions, each 250 ms wide. The response measure for

events at time ti using this kernel is defined as the sum over i of W(ti). Time

t ¼ 0 corresponds to the start of the stimulus. The windows for spike events

and for burst events shown here are those that produced the neurometric

curves in c and e below most similar to the psychometric curves. (b) Asmoother weighting window with properties similar to those shown in a.

(c) Neurometric curves, averaged over neurons, that follow from using

the spike weighting window in a when each event is an individual spike.

(d) Neurometric curve as in c but using the weighting window of b. (e) The

neurometric curves that follow from using the burst weighting window in a

when each event is a burst of spikes. (f) Neurometric curve as in e but using

the weighting window of b.

–3 –2 –1 0 1 2 3

–3 –2 –1 0 1 2 3

0

1

Cum

ulat

ive

resp

onse

–3 –2 –1 0 1 2 3Normalized spikes

–3 –2 –1 0 1 2 30

1

0

1

0

1

–3 –2 –1 0 1 2 3Normalized bursts

–3 –2 –1 0 1 2 3

Cum

ulat

ive

resp

onse

0

1

0

1

250:500 ms 500:500 ms 750:500 ms

250:500 ms 500:500 ms 750:500 ms

ROC = 0.55P = 0.01n = 1,00028 neurons

ROC = 0.49P = 0.32n = 1,00083 neurons

ROC = 0.51P = 0.23n = 1,00087 neurons

ROC = 0.51P = 0.39n = 1,00076 neurons

ROC = 0.54P = 0.001n = 1,00042 neurons

ROC = 0.52P = 0.26n = 1,00039 neurons

Spikes

Bursts

Figure 7 Weighted counts of spikes covary with

behavior on a trial-by-trial basis, but weighted

counts of bursts do not. Each panel shows the

cumulative distribution of normalized responses

for correct (solid lines) versus incorrect (dotted

lines) trials for trials with the stimulus lengths

shown. Events in each stimulus are integrated

using the weighting window of Figure 6a. In the

top row, events are defined as single spikes,

whereas in the bottom row, events are defined

as bursts of spikes. Only neurons for which the

weighted measure (spikes or bursts) was

significantly stimulus-dependent were used ineach panel, leading to different numbers of

neurons in upper versus lower panels. ROC,

receiver operating characteristic measure

comparing the two distributions. The probability

of observing this ROC value or greater by chance

is 0.5, estimated using a permutation test.

n, number of permutations.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Weighted sum of spikes covaries with performance

Our results using the weighting windows suggest that either spikesor bursts, when integrated using the windows of the top panels ofFigure 6, could form the basis for an S1 neuronal code that is consistentwith the psychophysical results of Figure 2. To further test the relation-ship between these two neuronal response measures and behavioralperformance, we carried out an analysis that distinguished betweencorrect and error trials. We assumed that on each trial, the animal’sperformance was based on comparing the activity of S1 neurons duringthe first stimulus with the activity of S1 neurons during the secondstimulus. Trial-to-trial variations in the activity of these neurons is thenexpected to affect the animal’s performance5,19. For each neuron, foreach trial and for each of the two response measures (weighted spikesand weighted bursts, using the window of Fig. 6a), we calculated thedifference between the measure applied to the spikes fired in responseto the second stimulus in the trial, and the measure applied to theresponse to the first stimulus in the trial. We then grouped trials intoclasses defined by the f1, f2 stimulus frequency pair used in each trial.When f2 4 f1, we expected correct trials to be associated with morepositive measure differences than error trials; conversely, whenf2 o f1, we expected correct trials to be associated with more negativemeasure differences. We multiplied each trial’s measure difference bythe sign of (f2 – f1) so that all trial classes would have the same expectedtendencies. We then normalized each trial’s measure difference into aZ-score by (i) subtracting the mean measure difference over those trialsthat shared the same (f1,f2) stimulus values and were recorded from thesame neuron and then (ii) dividing by the standard deviation of thatgroup of trials. We could now collapse together all trials from all classes,allowing us to search for what could be very weak correlations betweensingle S1 neurons and behavior. We asked whether the distribution ofnormalized measure differences for correct trials had a significantlymore positive mean than the distribution of normalized measuredifferences for error trials. We found that the two distributions werevery slightly but quite significantly different only for the weightedspiking rate measure (Fig. 7; permutation test, n¼ 1,000, Po 0.01)14,but not for the weighted bursting rate measures (Fig. 7; permutationtest, n ¼ 1,000, P 4 0.2)14. The effect was of a magnitude comparableto similar correct-versus-error trial tendencies found in the middletemporal area during a perceptual discrimination task20. The effect wasfound for weighted spike measures during comparison of 250-ms-longstimuli with 500-ms-long stimuli, as well as during comparison of500-ms-long stimuli with 500-ms-long stimuli. However, no significanteffect was found when one stimulus was 750 ms long and the other was500 ms long (upper panels of Fig. 7); we have no explanation for thelack of effect in this case.

DISCUSSION

The frequency of the vibrotactile stimulus can be defined as (i) thenumber of pulses per unit of time, or as (ii) the inverse of the period of

time between two consecutive mechanical sinusoid periods. To identifythe stimulus frequency, a subject could count the number of mechan-ical sinusoid periods per unit of time or measure the interval betweentwo consecutive sinusoid periods. Here, we have addressed the follow-ing questions: first, whether we can identify in the neuronal activity ofS1 which strategy an observer might use to discriminate between twovibrotactile stimuli, and second, which of the potential strategies areactually used by the subjects. Quickly adapting neurons of S1 typicallyproduce a brief burst of spikes in response to each mechanical sinusoidperiod. We found that depending on the stimulus sets and conditions,an observer of the evoked-neuronal responses in S1 could extractinformation from either (i) the overall spiking rates during the stimuli,(ii) the rates of bursting, (iii) a count of the number of spikes or (iv) acount of the number of bursts elicited by the vibrotactile stimuli. Countand rate codes are different only when the stimulus can vary induration. We note that by ‘count codes’ we do not necessarily meanthat the observer consciously counts the exact number of pulses duringthe stimulus—an observer could simply judge whether there are morepulses in one stimulus period that there are in the other. However, as weexplain further below, when we further consider which of thesepotential codes might be actually used by subjects performing thetask, we find reasons to reject burst-based codes, suggesting that themost likely neural code for vibrotactile discrimination is one based onspike rate or count.

To distinguish whether a code based on spike rate or a code based onspike count was more to be used by subjects when discriminatingbetween two stimuli, we manipulated stimulus duration. We assumedthat if the observer used a rate-based code, or even a periodicity-basedcode, manipulation of stimulus duration should have no effect onpsychometric curves. But if the observer used a simple accumulativecounting strategy, stimulus duration manipulation should result inconsistent and predictable biases in behavior. Monkeys biased theirpsychophysical performance in a manner consistent with an accumu-lative counting strategy. In other words, when the duration of one ofthe two stimuli increased, monkeys behaved as if the frequency of thatstimulus were higher than it actually was, and when the duration of astimulus was shortened, monkeys behaved as if the frequency of thatstimulus were lower than it actually was. However, although the sign ofthe bias was consistent with an accumulative-counting code, themagnitude of the bias was much smaller than that predicted byaccumulation over the entirety of the stimulus periods. Consequently,we considered a weighting window, defining a kernel over which spikesmight be accumulated (that is, integrated), with most of the windowconcentrated over a time period significantly shorter than the standard500-ms-long stimulus. We found that a spike integration time windowwith a 230-ms width, followed by an exponential fall-off with a timeconstant of 60 ms, can qualitatively account for both sign andmagnitude of the psychophysical biases observed experimentally,can be used for both periodic and aperiodic stimuli, covaries on a

Table 1 Neurons recorded in primary somatosensory cortex (S1) during vibrotactile discrimination with variable stimulus length

f1:f2 duration (ms) Periodic (f1: 14–30 Hz, f2: 22 Hz) Aperiodic (f1: 14–30 Hz, f2: 22 Hz) Periodic (f1: 22 Hz, f2: 14–30 Hz) Aperiodic (f1: 22 Hz, f2: 14–30 Hz)

250:500 83 (p ¼ 61, fr ¼ 22, b ¼ 63) 86 (p ¼ 5, fr ¼ 27, b ¼ 64) – –

750:500 101 (p ¼ 75, fr ¼ 31, b ¼ 78) 131 (p ¼ 6, fr ¼ 38, b ¼ 95) – –

500:250 – – 23 (p ¼ 16, fr ¼ 8, b ¼ 17) 15 (p ¼ 0, fr ¼ 6, b ¼ 11)

500:750 – – 29 (p ¼ 20, fr ¼ 9, b ¼ 21) 40 (p ¼ 2, fr ¼ 7, b ¼ 28)

f1, first stimulus. f2, second stimulus. First number in each column under ‘periodic’ or ‘aperiodic’ represents number of those neurons tested with a modification in the stimulus duration of a total of146 neurons. Each stimulus set of variable stimulus length had an equal number of trials in which the two stimulus periods were always 500 ms long. Numbers in parentheses after ‘¼’ correspond tothe numbers of neurons that had significant slopes for the measures of periodicity (p), firing rate (fr) and bursts (b). The slopes were calculated using the stimulus periods of 500:500 ms.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

trial-by-trial basis with behavior and produces neurometric discrimi-nation thresholds that are similar to psychometric thresholds. We donot propose this time window as the precisely shaped unique windowwith these properties; other windows of approximately the same shapeand size, but differing in the details of their shape (for instance, agamma function instead of flat followed by exponential fall-off) wouldshare the same properties. In sum, spike integration over a window ofthis form is the single candidate neural code for vibrotactile discrimi-nation that is consistent with all the data examined to date. The outputof such spike integration from quickly adapting neurons of S1 couldform the input to more central areas, in which a spike-rate codeencodes the stimulus frequency during the stimulus presentation,working memory, comparison and decision-making processes of thistask6,21–26. The specific mechanisms by which integration over such atime window might be carried out remain to be determined.

Our findings are closely reminiscent of psychophysical evidencefound for integration time windows during detection of vibrotactilestimuli in the vibration frequency range (250 Hz; ref. 27). Some of theseauthors found no evidence for integration in sensations transduced bynon-Pacinian receptors, which are thought to underlie perception ofstimuli in the flutter frequency range (6–40 Hz; ref. 27). But their use ofa detection task at high frequencies, as opposed to the discriminationtask in the flutter frequency range we used here, precludes a directcontrast between the two studies.

We have previously found that modifying the amplitude of themechanical stimuli has no detectable effect on discrimination bias orperformance, as long as the stimuli remain well above threshold fordetection1, as they are here. This is consistent with our present results:well above detection threshold, modest changes in stimulus amplitudedo not change the firing rate of mechanosensory afferents withreceptive fields centered at the stimulation site. Instead, amplitudechanges affect the number of afferents with receptive field centers awayfrom the stimulation site that are recruited into responding to thestimulus28. Thus, we might expect S1 neurons with receptive fieldscentered at the stimulation site to be similarly insensitive to modestchanges in stimulus amplitudes29. If discrimination is based on theweighted integration of spikes from these neurons, then discriminationwould not be affected by amplitude changes that keep the stimulusabove detection threshold.

An important criterion for accepting any of the neural measures wehave considered as candidates for encoding the stimuli is that the codeshould covary, on a trial-by-trial basis, with behavior19. In S1, we foundthat weighted spike counts, but not weighted burst counts or ourperiodicity measure, covaried with discrimination performance (inref. 6 and the current study), supporting weighted spike count (that is,spike rate when computed with a kernel window with a shape similarto those of Fig. 6) as the most likely neural code for frequencydiscrimination (Fig. 7). In all areas central to S1 studied duringthe vibrotactile task, spike rate not only carries information aboutstimulus frequency during the different phases of the vibrotactilediscrimination task, but also covaries, on a trial-by-trial basis, withthe behavioral performance6,21–26.

In conclusion, over the past several years there has been a debate overhow information of sensory stimuli is encoded by cortical neurons.Depending on the stimulus characteristics, tasks and brain areaschosen, some investigators have proposed the firing rate30,31, detailedprecision of the neuronal responses3,29,32–35, bursting rate9–12,36 andsparse temporal codes37 as codes for the sensory stimuli. In our case, wehave shown that the frequency of the vibrotactile stimulus is encoded inseveral different measures of the stimulus-evoked responses of S1neurons. However, the acid test is whether any of these neural codes

accounts for behavior during task performance. Our results show thatfiring rate of S1 neurons, computed as a weighted spike count with aweighted window that has most of its mass in the first 250 ms and yethas a small tail extending beyond 500 ms, best covaries with theanimal’s psychophysical performance and therefore is the most likelyneural code for vibrotactile discrimination. Although we have rejectedall but one of the corresponding codes as the basis of vibrotactilediscrimination, there is nevertheless the possibility that each of therejected codes might be useful for different purposes during thevibrotactile task. Given that our analysis is based on single units, it ispossible that a temporal code based on interactions between multipleneurons (for example, one based on spike synchrony), for either orboth periodic or aperiodic stimuli, has escaped our scrutiny.

METHODSGeneral. Two monkeys (Macaca mulatta) were trained to discriminate the

difference in frequency between two mechanical vibrations delivered sequen-

tially to their fingertips (Fig. 1). Neurophysiological recordings were made in

S1 (areas 3b and 1) contralateral to the mechanical stimulation while the

monkeys performed the discrimination task1,3,5,6,13. The neurons selected for

study had small, cutaneous receptive fields confined to the smooth, glabrous

skin of one fingertip of digits 2, 3 or 4. All neurons had quickly adapting

properties. The neuronal responses from S1 were collected while the monkeys

discriminated frequencies at psychophysical thresholds (Figs. 1 and 2). Animals

were handled according to institutional standards of the US National Institutes

of Health and the Society for Neuroscience.

Discrimination task. The discrimination task used here has been described

before3,13. Briefly, stimuli were delivered to the skin of the distal segments of

one digit of the right, restrained hand by means of a computer-controlled

motor stimulator (BME Systems; 2-mm round tip). The initial indentation was

500 mm. Vibrotactile stimuli were trains of short mechanical pulses. Each of

these pulses consisted of single-cycle sinusoid lasting 20 ms. Stimulus ampli-

tudes were adjusted to equal subjective intensities3,13: for example, 71 mm at

12 Hz and 51 mm at 34 Hz (1.4% per Hz). During trials, two vibrotactile stimuli

were delivered consecutively to the glabrous (hairless) skin, separated by an

inter-stimulus delay period of 3 s, and the animal was rewarded with a drop of

liquid. Discrimination was indicated by pressing one of two push-buttons.

Performance was measured through psychometric techniques1,3,6,13. Initially,

the two monkeys were trained to discriminate between pairs of periodic

stimulus frequencies up to their psychophysical thresholds3,13. We then tested

whether the two animals could discriminate between pairs of aperiodic stimuli.

Aperiodic stimuli were also composed of pulses that were each 20 ms wide. To

generate an aperiodic stimulus with N pulses, the 500-ms-long stimulus period

was first divided into 20-ms bins, the first and last of the bins were then

assigned to contain a pulse, and (N – 2) of the remaining bins were randomly

chosen to also contain a pulse. Fixing the initial and final pulse ensured that

aperiodic stimuli could not be discriminated based on total stimulus length.

Periodic and aperiodic stimuli were used in different blocks of trials. For

both periodic and aperiodic conditions, and again in separate blocks of

trials, monkeys either compared a variable second stimulus frequency (range

14–30 Hz) against a fixed first stimulus frequency (22 Hz; Fig. 2a,b), or they

compared a fixed second stimulus frequency (22 Hz) against a variable first

stimulus frequency (range 14–30 Hz; Fig. 2c,d). In these stimulus sets, monkeys

discriminated between stimulus frequencies of equal duration (500 ms) or

unequal duration (Fig. 2), with one stimulus 50% longer (750 ms) or shorter

(250 ms) than the other.

Recording sessions and sites. Neuronal recordings were obtained with an array

of seven independent, moveable microelectrodes (2–3 MO, inserted into S1;

areas 3b and 1; two monkeys)1,3. Recording sites changed from session to

session, and standard histological procedures were used to construct surface

maps of all of the penetrations in S1. This was done first by marking the edges

of the small chamber (7 mm in diameter) placed above S1. Additionally, in

the last recording sessions, we made small lesions at different depths in

the recording area. Neurons recorded from the top of the cortex to


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

2,500 mm below the surface fell into area 1; neurons recorded 2,500 mm from

the insertion site and below fell into area 3b. All of these neurons had small

cutaneous receptive fields confined to the distal segments of fingertips 2, 3 or 4

and had quickly adapting properties.

Data analysis. For each neuron studied during the discrimination task, offline

analysis and statistical tests were done with custom and Matlab software

(Mathworks). The analysis was restricted to the stimulus periods according

to three criteria. First, we devised a measure that quantified the capacity of the

neurons to represent the periodicity of the stimulus. For each trial, the power

spectrum of the spike train evoked during the stimulus period was computed

(fast Fourier transform, n ¼ 216; sampling frequency, 10 kHz; resolution, 0.15;

range, 6–100 Hz)38,39. As an estimate of the periodicity, we calculated the

median frequency around the peak power spectrum frequency. The frequencies

used for this measure were limited to those within a factor of 1.8 of the peak (to

avoid contamination by harmonics) and to frequencies with a power greater

than 0.15 of the power at the peak (to avoid noise). The median frequency

calculated in this way was considered a quantitative measure of periodicity

evoked in S1 neurons by the periodic or aperiodic mechanical stimuli. Second,

neurons were classified as responding with bursts to the mechanical stimuli

according to two criteria. First, an elevated firing rate criterion: we required the

number of spikes recorded during the interval that separates two consecutive

mechanical sinusoids to be higher than the activity of 950 of 1,000 intervals of

the same duration, randomly selected within the period of 1.5 s that preceded

the stimulus presentation (permutation test, n ¼ 1,000, P o 0.05)14. Second, a

mechanical event detection criterion: we required that in 95% of the intervals

between two mechanical stimuli, the number of spikes in the first half of the

interval was higher than the number of spikes in the second half of the same

interval (Supplementary Fig. 1). Individual cycles for which the mechanical

event detection criterion was satisfied were considered as containing a burst.

This definition of bursts was intended to capture how an observer of an S1

neuron might try to detect the application of individual mechanical pulse

events. The definition was not intended to capture detection of spike bursts

caused by intrinsic biophysical properties of the neuron. Third, for each trial,

we calculated the mean firing rate over the stimulus periods. For each stimulus

frequency, we computed the mean ± s.d. of periodicity, bursting rate and firing

rate over all trials with that stimulus frequency. For further analysis, we selected

those neurons that had the best linear fit (w2, Q 4 0.05) of the periodicity,

bursting rate, number of bursts, number of spike or firing rate values as a

function of the stimulus frequency38,39. We also required the slope of this linear

fit to be significantly different from zero (permutation test, n ¼ 1,000,

P o 0.05)14. Under this combined test, not all neurons carried information

in all of the codes tested. For each of the codes tested, we selected for further

analysis only the subset of neurons that carried statistically significant stimulus

information using that code. The discrimination task requires the comparison

of the second stimulus frequency against the first. The quickly adapting

neurons of S1 provide a reliable representation of the two stimulus frequen-

cies3,5,6. We determined the probability that an observer (for example, a cortical

region central to S1) could distinguish the difference between the two stimuli.

This could be based on a comparison of the neuronal response distributions of

the second stimulus frequency (f2) made against the neuronal response

distributions of the first stimulus frequency (f1). According to this, the observer

could use a simple rule: if the number of spikes or bursts during f2 is higher

than during f1, then f2 is higher than f1. The same rule can be used when

considering the periodicity values: if the periodicity values during f2 are higher

than during f1, then f2 is higher than f1 (refs. 6,40). This rule can be tested by

determining the area under the curve receiver operating characteristic (ROC)

generated by the neuronal response distributions for each pair of stimulus

frequencies, using periodicity, bursting rate and firing rate values6,40. In pairs of

stimulus frequencies in which the neuronal response distributions of f2 are

much higher than the neuronal distributions of f1, ROC values are close to 1. If

the neuronal response distributions of f2 are much lower than the neuronal

response distributions of f1, ROC values are close to 0. For overlapping

distributions, intermediate ROC values are found (0.5). The ROC values were

then used to compute neurometric functions. Psychometric and neurometric

discrimination thresholds were calculated as half of the difference between the

stimulus frequency identified as higher than the base in 75% of the trials and

that frequency identified as higher in 25% of the trials3,6,13. These were directly

read from the logistic functions (Boltzmann’s equation) expressed in Hz.

Because the manipulation of the stimulus duration altered both psychometric

and neurometric curves, these changes can be quantified by calculating two

parameters in the logistic function: (i) the psychometric and neurometric

threshold is the minimal difference (in Hz) between f1 and f2 that the subject

and the neuron can discriminate, and (ii) the X0 value is the frequency with a

0.5 probability in the logistic function. The X0 value measures the displacement

of the logistic function along the x-axis. Rightward displacement of the

psychometric function (compared with the control psychometric function

calculated in the same run) indicates that the observer judges the comparison

stimulus frequency lower than the first, whereas leftward displacement indicates

the opposite.


ACKNOWLEDGMENTSThe research of R.R. was supported by an International Research Scholars Awardfrom the Howard Hughes Medical Institute and grants from Consejo Nacional deCiencia y Tecnologıa and Direccion del Personal Academico of the UniversidadNacional Autonoma de Mexico. C.D.B. is supported in part by the US NationalInstitutes of Health (grant R01-MH067991).


Received 29 March; accepted 8 July 2005


1. Romo, R., Hernandez, A., Zainos, A. & Salinas, E. Somatosensory discrimination basedon cortical microstimulation. Nature 392, 387–390 (1998).

2. Romo, R., Hernandez, A., Zainos, A., Brody, C. & Lemus, L. Sensing without touching:psychophysical performance based on cortical microstimulation. Neuron 26, 273–278(2000).

3. Mountcastle, V.B., Steinmetz, M.A. & Romo, R. Frequency discrimination in the sense offlutter: psychophysical measurements correlated with postcentral events in behavingmonkeys. J. Neurosci. 10, 3032–3044 (1990).

4. Recanzone, G.H., Merzenich, M.M. & Schreiner, C.E. Changes in the distributed temporalresponse properties of S1 cortical neurons reflect improvements in performance on atemporally based tactile discrimination task. J. Neurophysiol. 67, 1071–1091 (1992).

5. Salinas, E., Hernandez, A., Zainos, A. & Romo, R. Periodicity and firing rate as candidateneural codes for the frequency of vibrotactile stimuli. J. Neurosci. 20, 5503–5515(2000).

6. Hernandez, A., Zainos, A. & Romo, R. Neuronal correlates of sensory discrimination inthe somatosensory cortex. Proc. Natl. Acad. Sci. USA 97, 6191–6196 (2000).

7. Romo, R. & Salinas, E. Touch and go: decision-making mechanisms in somatosensation.Annu. Rev. Neurosci. 24, 107–137 (2001).

8. Romo, R. & Salinas, E. Flutter discrimination: Neural codes, perception, memory anddecision making. Nat. Rev. Neurosci. 4, 203–218 (2003).

9. Reinagel, P., Godwin, D., Sherman, M. & Koch, C. Encoding of visual information by LGNbursts. J. Neurophysiol. 81, 2558–2569 (1999).

10. Martınez-Conde, S., Macknik, S.L. & Hubel, D.H. The function of bursts of spikes duringvisual fixation in the awake primate lateral geniculate nucleus and primary visual cortex.Proc. Natl. Acad. Sci. USA 99, 13920–13925 (2000).

11. Kepecs, A., Wang, X.J. & Lisman, J. Bursting neurons signal input slope. J. Neurosci.22,9053–9062 (2002).

12. Krahe, R. & Gabbiani, F. Burst firing in sensory systems. Nat. Rev. Neurosci. 5, 13–24(2004).

13. Hernandez, A., Salinas, E., Garcıa, R. & Romo, R. Discrimination in the sense of flutter:new psychophysical measurements in monkeys. J. Neurosci. 17, 6391–6400 (1997).

14. Siegel, S. & Castellan, N.J. Nonparametric Statistics for the Behavioral Science(McGraw-Hill, New York, 1988).

15. Connors, B.W. & Gutnik, M.J. Intrinsic firing patterns of diverse neocortical neurons.Trends Neurosci. 13, 99–104 (1990).

16. Carandini, M. Visual cortex: fatigue and adaptation. Curr. Biol. 10, R605–R607 (2000).17. Nowak, L.G., Azouz, R., Sanchez-Vives, M., Gray, C.M. & McCormick, D.A. Electro-

physiological classes of cat primary visual cortical neurons in vivo as revealed byquantitative analyses. J. Neurophysiol. 89, 1541–1566 (2003).

18. Kohn, A. & Movshon, J.A. Neuronal adaptation to visual motion in area MT of themacaque. Neuron 39, 681–691 (2003).

19. Parker, A.J. & Newsome, W.T. Sense and the single neuron: probing the physiology ofperception. Annu. Rev. Neurosci. 21, 227–277 (1998).

20. Britten, K.H., Newsome, W.T., Shadlen, M.N., Celebrini, S. & Movshon, J.A. A relation-ship between behavioral choice and the visual responses of neurons in macaque MT.Vis. Neurosci. 13, 87–100 (1996).

21. Romo, R., Hernandez, A., Zainos, A., Lemus, L. & Brody, C.D. Neuronal correlates ofdecision-making in secondary somatosensory cortex. Nat. Neurosci. 5, 1217–1225(2002).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

22. Romo, R., Hernandez, A., Zainos, A. & Salinas, E. Correlated neuronal discharges thatincrease coding efficiency during perceptual discrimination. Neuron 38, 649–657(2003).

23. Romo, R., Hernandez, A. & Zainos, A. Neuronal correlates of a perceptual decision inventral premotor cortex. Neuron 41, 165–173 (2004).

24. Hernandez, A., Zainos, A. & Romo, R. Temporal evolution of a decision-making processin medial premotor cortex. Neuron 33, 959–972 (2002).

25. Romo, R., Brody, C.D., Hernandez, A. & Lemus, L. Neuronal correlates of parametricworking memory in the prefrontal cortex. Nature 399, 470–473 (1999).

26. Brody, C.D., Hernandez, A., Zainos, A. & Romo, R. Timing and neural encoding ofsomatosensory parametric working memory in macaque prefrontal cortex. Cereb. Cortex13, 1196–1207 (2003).

27. Gescheider, G.A., Berryhill, M.E., Verrillo, R.T. & Bolanowski, S.J. Vibrotactile temporalsummation: probability summation or neural intergration? Somatosens. Mot. Res. 16,229–242 (1999).

28. Talbot, W.H., Darian-Smith, I., Kornhuber, H.H. & Mountcastle, W.T. The sense of flutter-vibration: comparison of human capacity with response patterns of mechanoreceptorsafferents from the monkey hand. J. Neurophysiol. 31, 301–334 (1968).

29. Mountcastle, V.B., Talbot, W.H., Sakata, H. & Hyvarinen, J. Cortical neuronal mechan-isms in flutter vibration studied in unanesthetized monkeys. J. Neurophysiol. 32, 452–484 (1969).

30. Shadlen, M.N. & Newsome, W.T. Noise, neural codes and cortical organization. Curr.Opin. Neurobiol. 4, 569–579 (1994).

31. Shadlen, M.N. & Newsome, W.T. The variable discharges of cortical neurons: Implica-tions for connectivity, computation, and information coding. J. Neurosci. 18, 3870–3896 (1998).

32. Ahissar, E., Sosnik, R. & Haidarliu, S. Transformation from temporal to rate coding in asomatosensory thalamocortical pathway. Nature 406, 302–306 (2000).

33. Poggio, G.F. & Viernstein, L.J. Time series analysis of impulse sequences of thalamicsomatic sensory neurons. J. Neurophysiol. 27, 517–545 (1964).

34. Abeles, M. Corticonics (Cambridge Univ. Press, Cambridge, 1990).35. Bialek, W., Rieke, F., Vansteveninck, R.R.D. & Warland, D. Reading a neural code.

Science 252, 1854–1857 (1991).36. Bair, W., Koch, C., Newsome, W.T. & Britten, K. Power spectrum analysis of

bursting cells in area MT in the behaving monkey. J. Neurosci. 14, 2870–2892 (1994).37. deCharms, R.C. & Zador, A. Neural representation and the cortical code. Annu. Rev.

Neurosci. 23, 613–647 (2000).38. Draper, N. & Smith, H. Applied Regression Analysis 2nd edn.(Wiley, New York, 1966).39. Press, W.H., Flannery, B.P., Teukolsky, S.A. & Vetterling, W.T. Numerical Recipes in C

2nd edn. (Cambridge Univ. Press, Cambridge, 1992).40. Green, D.M. & Swets, J.A. Signal Detection Theory and Psychophysics (Wiley, New York,

1966).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Risk-sensitive neurons in macaque posteriorcingulate cortex

Allison N McCoy1 & Michael L Platt1,2,3

People and animals often demonstrate strong attraction or aversion to options with uncertain or risky rewards, yet the neural

substrate of subjective risk preferences has rarely been investigated. Here we show that monkeys systematically preferred the risky

target in a visual gambling task in which they chose between two targets offering the same mean reward but differing in reward

uncertainty. Neuronal activity in posterior cingulate cortex (CGp), a brain area linked to visual orienting and reward processing,

increased when monkeys made risky choices and scaled with the degree of risk. CGp activation was better predicted by the

subjective salience of a chosen target than by its actual value. These data suggest that CGp signals the subjective preferences

that guide visual orienting.

To survive and thrive, animals must make choices that relate internalstates to the current environment. For example, choosing to pursuefood or water depends not only upon available resources but also onwhether hunger or thirst is greater. The decision to make a particularaction thus depends on subjective needs and desires as well as anyobjectively measurable gains1–3.

In addition to state-dependent variables such as hunger, thirst andeven wealth, subjective biases also contribute to decision making. Sincethe 18th century, it has been known that people’s choices reflect rewarduncertainty as well as reward value4. When confronted with twooptions of the same mean value but differing in uncertainty, bothpeople and animals typically avoid choosing the uncertain, or risky,option5,6. The idea that subjective preferences guide decision makinghas since become a core concept in the decision sciences2,7. However,the impact of subjective preferences on neural mechanisms of decisionmaking remains largely unexplored (but see ref. 8).

The simplest economic models of decision making posit thatrational choosers select the alternative with the highest expectedvalue9,10. Recent neurophysiological studies of visual orienting deci-sions have demonstrated that neurons in several brain areas linkingvisual perception with eye movements also track target value11–15.These observations suggest that orienting decisions are computed, inpart, by scaling neuronal responses by target value11,14,16. One questionthese observations raise, however, is whether reward modulation ofneuronal activity in these brain areas reflects scaling by subjectivevalue17, predicted reinforcement14,18 or motivation19.

As people and animals often demonstrate strong attractions oraversions to options with uncertain rewards2,6,20, risk preferenceprovides a promising behavioral framework for exploring neuralmechanisms underlying decision making and offers a potential wayto dissociate subjective value from objective rewards. Specifically,

neurons participating in the decision process should be sensitive tosubjective risk preferences, even when available options have the sameobjective value.

To test this prediction, we recorded from single neurons in posteriorcingulate cortex (CGp), a limbic area linking reward with spatialattention21,22 and orienting23,24. Two adult male rhesus macaquesperformed a visual gambling task in which they chose between twovisual targets offering the same mean reward but differing in rewarduncertainty (Fig. 1a). We found that monkeys preferred orienting totargets offering uncertain rewards, and neuronal activity in CGpreflected these risk preferences. Our data suggest that neuronalresponses in CGp signal subjective spatial biases that guide orienting.

RESULTS

Behavioral risk preferences in monkeys

Although numerous studies have sought to understand risk preferencesin humans, birds and insects (reviewed in refs. 3,6), risk preferences inmonkeys remain largely unstudied. Therefore, we first probed mon-keys’ behavioral sensitivity to reward uncertainty in a visual gamblingtask. Shifting gaze to the ‘certain’ target resulted in 150-ms access tofruit juice; shifting gaze to the ‘risky’ target resulted in the randomreceipt of less than 150 ms on one-half of the trials and more than150 ms on the other half of trials (mean ¼ 150 ms). The locations of thecertain and risky targets, as well as the degree to which risky rewardvalues deviated from the mean, were varied across blocks of 50 trials(Fig. 1a, lower panel). Here we define risk as the coefficient of varia-tion (CV) of rewards associated with the risky target, a dimen-sionless measure of relative risk permitting direct comparisons withother studies3.

Reward CV ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðððx1 � meanÞ2+ðx2 � meanÞ2Þ=nÞ=mean

q


1Department of Neurobiology, 2Center for Cognitive Neuroscience and 3Department of Biological Anthropology and Anatomy, Duke University Medical Center, Box 3209,Durham, North Carolina 27710, USA. Correspondence should be addressed to M.L.P. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

where x1 and x2 are values for risky target andn is the number of risk values. Overall, bothmonkeys preferred the risky target and thefrequency of choosing it increased system-atically with the degree of risk (Fig. 2a).

Sensitivity to reward uncertainty is notpredicted by standard models of reinforce-ment learning. Specifically, reinforcementlearning models predict that the associativestrength of any stimulus for controlling beha-vior is determined by rewards delivered inassociation with that stimulus25,26. Becauseaverage reward size was the same for riskyand certain targets, such models would notpredict a preference for one option over theother. To explore this surprising behavioralpattern further, we therefore examined theimpact of prior reward outcomes on subse-quent choices. We found that receipt of smal-ler-than-average rewards on the previous trialblunted the likelihood of choosing the riskytarget on the next trial (Fig. 2b) but not nearlyas much as would be expected by standardreinforcement learning25. Moreover, monkeysbehaved as if they overvalued relatively largerewards. Our behavioral dataset was largeenough to examine sequences of up to sevenprior reward outcomes with statisticalconfidence. Analysis showed that monkeyssignificantly preferred the risky target evenafter a sequence of six smaller rewards thatfollowed a large reward (F ¼ 5.709, P o 0.00001). These data makeplain that the choices monkeys make depend not only on expectedreward value, as shown previously16, but also on reward uncertainty.

We next examined the impact of both received rewards and riskexperienced for prior choices on the probability of choosing the riskytarget using logistic regression27. We found that both the degree of riskand normalized reward value associated with the target chosen on theprevious trial biased the probability of choosing the risky target on thefollowing trial (logistic regression coefficients: target risk ¼ 2.768;target reward value ¼ 4.16 (risky), 3.378 (certain); all P-values |0.001); moreover, including both risk and rewards received for priorchoices significantly improved the explanatory power of the model overany other single variable or combination of variables (Akaike’s Infor-mation Criterion (AIC)combined ¼ 14,629.66; AICtarget risk ¼ 15,476.45;AICrisky target reward ¼ 14,847.19; AICcertain target reward ¼ 18,079.97; allother combinations, AICs 4 (AIC)combined).

Next, we developed a simple model of monkeys’ subjective prefer-ence for the risky target—which we refer to here as subjective targetutility—on the basis of the difference in the experienced value ofthe risky and certain targets. Since the logistic regression analysisshowed that both experienced risk and rewards received influencedthe probability of choosing the risky target on subsequent trials, wedenoted the experienced value of the risky target Vrisky on each trial asthe sum of the risk and reward associated with that target when themonkey chose it:

Vrisky ¼ Reward receivedrisky target + Riskrisky target ð1Þ

Similarly, we denoted the experienced value of the certain target Vcertain

on each trial as the sum of the reward and risk associated with that

target when the monkey chose it:

Vcertain ¼ Reward receivedcertain target + Riskcertain target ð2Þ

where the risk of the certain target was always 0.We next estimated the subjective utility of the risky target Urisky on a

given trial (t) according to the following algorithm:

UriskyðtÞ ¼ Sn¼ 1 to i½Vriskyðt � nÞ � Vcertainðt � nÞ� � an ð3Þ

where an is the logistic coefficient for the difference in the experiencedvalue of the risky and certain targets lagged n trials. Multiple logisticregression analysis of the probability of a risky choice as a function ofthe difference in experienced value for the two targets (Vrisky � Vcertain)on each of up to ten prior trials was used to derive the weighting factoran. This analysis showed that the value function difference (Vrisky �Vcertain) significantly influenced the probability of choosing the riskytarget at all lags up to five trials (AIClags1–5 o AICs for all othercombinations) but declined rapidly thereafter (Fig. 2c). As the additionof further lags did not significantly improve the model, i was set to 5trials. Using the weighting factor an, our estimate of subjective targetutility, derived from prior choices and their associated rewards and risk,provided a good prediction of the probability of choosing the riskytarget (logistic regression coefficient ¼ 5.2222, Wald statistic ¼2,560.93, P o 0.00001).

In animals and humans, preference for risky options has beenassociated with impoverished physiological6,28 or financial2,29 status.We therefore asked how the risk sensitivity of our monkeys was affectedby their hydration status. In our experiments, access to fluids waslimited during the week but freely available on weekends. Risksensitivity was therefore examined as a function of the day of the

T1T1

T2T2

Fixation on

Fixationoff

FixTarget T1Target T2

EyeReward

Sample reward schedulesReward CV 0.0667 0.1 0.167 0.333 0.5

20

20Horizontal amplitude

Ver

tical

am

plitu

de Response (H

z)

–20–20

0

0

0.667Certain target 150

160140

50 trials

135 125 100 75 50165 175 200 225 250150 150 150 150 150

Risky target

Target onDelay

Time

MoveReward

CSMR

CC

RF

Fix

T2

T1

a b

c

Figure 1 Method for investigating risk sensitivity in macaque posterior cingulate cortex. (a) Visual

gambling trials were used to investigate risk sensitivity. On each trial, subjects initially fixated (± 1–21) a

central yellow LED (200–800 ms). Two peripheral yellow LEDs were then illuminated diametrically

opposite the fixation LED (200–800 ms). The fixation LED was extinguished, cueing the monkey to shift

gaze to either target (± 3–51) within 350 ms. Correct trials were rewarded with a 300-ms noise burst and

juice. Lower panel: example of reward schedule. Mean reward size for each target was 150 ms; the range

of reward differences for the risky target was 20–250 ms across blocks of trials. (b) Recording sites in

posterior cingulate cortex (CGp), estimated by digital ultrasound imaging. Diagonal hatches indicate

approximate neuron locations within areas 31 and 23. Recording chamber projection (red box) and major

landmarks in the ultrasounds are indicated (CS: cingulate sulcus, horizontal limb; MR: marginal ramus;

CC: corpus callosum). (c) Example of target geometry. One target was inside the response field (RF) while

the other was diametrically opposite the fixation point. Neuronal response is plotted as a function oftarget location using an arbitrary color scale from blue to red (low to high firing rate).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

week, as well as time during an experimentalsession, under the assumption that fluid bal-ance would decrease over the course of theweek and increase as juice was consumedduring each daily session. We found no effectof day of the week (logistic regression coeffi-cients, day of the week ¼ 0.005, P 4 0.97;subject ¼ 0.23, P o 0.000001; risk ¼ 0.281,Po 0.000001) or number of trials performedin each session (logistic regression coeffi-cients: trial ¼ –0.00004, P 4 0.42; subject¼ 0.22, P o 0.000001; risk ¼ 2.44, P o0.000001). Thus, monkeys showed consistentpreferences for uncertain fluid rewards, andthese preferences were apparently indepen-dent of fluid balance.

One important question is whether anynonlinearity in the computer-driven solenoidcontrolling fluid delivery might explain whymonkeys preferred the risky option. In fact,fluid reward size was a linear function ofsolenoid open time (volume ¼ 0.0026 +0.001 � (open time in ms); Fig. 2d). Thus,the risk preferences of our monkeys could notbe explained by an asymmetry in the size ofrewards delivered across the range of valuestested. We also performed a control experi-ment in which choosing the risky target wasfollowed by delivery of larger-than-averagerewards on one-third of trials and smaller-than-average rewards on two-thirds of trials(as compared with one-half larger-than-average and one-half smaller-than-averagerewards in standard visual gambling trials).The certain target, as before, offered 150-msaccess to juice on all trials, and the specificvalues of high and low rewards wereunchanged from standard gambling trials. Choosing the risky targetthus resulted in a net loss of juice compared with choosing the certainoption, and this loss increased with increasing risk. Despite the fact thatthe expected value of the risky target now declined with increasing risk,monkeys continued to prefer it (Fig. 2e).

Another question these data raise is whether the observed riskpreferences might simply reflect a preference for the novelty, orvariability, of rewards associated with the risky target. In otherwords, perhaps a changing target or reward is simply more interestingfor monkeys. This is an important potential confound which wesought to address in a control experiment. Reward size was heldconstant at 150 ms access to juice for both targets while novelty wasintroduced by systematically changing the color of one of the targetsduring reward delivery. For the ‘monotonous’ target, the color of thetarget remained yellow throughout the trial. For the ‘novel’ target,target color changed from yellow to green on half of the trials and fromyellow to red on the other half of the trials. Notably, both targets wereyellow until the delivery of reward such that any difference inluminance by color would not affect the monkeys’ choices. Thelocations of the novel and monotonous targets were reversed acrossblocks of trials to mitigate any spatial bias in the monkeys’ choices.Neither monkey showed any preference for the novel colored target(Fig. 2f). Thus, monkeys’ preference for risky targets is unlikely to beexplained by novelty alone.

Neural correlates of reward uncertainty in CGp

Having established the risk preferences of two monkeys performing avisual gambling task, we next examined the activity of single neurons inposterior cingulate cortex (Fig. 1b) for evidence of similar risksensitivity. Neurons in this area respond to the illumination of con-tralateral visual stimuli30, after contraversive gaze shifts16,24,30 and afterreward delivery16, and the strength of these responses is modulated byreward size and expectancy16. Because most CGp neurons respondselectively for a broad range of contralateral saccades, experiments wereconducted such that one target was placed inside the response field ofthe neuron under study while the other target was diametricallyopposite the fixation point (Fig. 1c). The response field was determinedduring 100–400 standard mapping trials in which monkeys were askedto shift gaze to targets throughout the visual field.

The activity of a single CGp neuron recorded during visualgambling trials is shown in Figure 3. Neuronal activity increasedafter movement onset, and this activity was modulated both by whetherthe movement was into or out of the response field as well as whetherthose choices were for risky or certain rewards (Fig. 3a). During theepoch 200–400 ms after movement onset, firing rate was modulated byboth movement direction and risk (Fig. 3b). Neuronal activityincreased systematically with increasing risk, and this risk-relatedmodulation was stronger for movements in the neuron’s preferreddirection (Fig. 3c).

0.9a

0.8

0.7

0.6

0.5

0.0 0.2 0.4 0.6 0.8

Risk (reward CV)

Pro

babi

lity

of r

isky

cho

ice

1.0

Monkey NikoMonkey Broome

b

0.9

0.8

0.7

0.6

0.5

1.0

1.2

Reward on previous trial (ms)

Pro

babi

lity

of r

isky

cho

ice

0 50 100 150 200 250 300

c

Lag (trials)

Coe

ffici

ent f

or r

isky

val

ue

10–1.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

9 8 7 6 5

n = 2032 trialsn = 2008 trials

4 3 2 1

d

0

0.01

0.02

0.03

0.04

0.05

100 200 300Solenoid open time (ms)

Juic

e vo

lum

e (m

l)

e

0.0 0.2 0.4 0.6 0.8 1.0Risk (reward CV)

0.9

0.8

0.7

0.6

0.5

Pro

babi

lity

of r

isky

cho

ice

f

0.0

0.2

0.4

0.6

0.8

1.0

Broome NikoMonkey

Pro

babi

lity

of n

ovel

cho

ice

Figure 2 Monkeys prefer targets offering uncertain rewards. (a) Probability of choosing the risky target as

a function of risk for each monkey. Risk preference increased with increasing risk (logistic regression

coefficients: Broome, 2.442, P o 0.0000001; Niko, 2.426, P o 0.0000001). (b) Monkeys discount

low payoffs at the risky target. Probability of choosing the risky target is plotted as a function of reward

received on the previous trial for both monkeys. Both monkeys were indifferent to the average reward size(150 ms) but systematically preferred the risky target after either small or large payoffs. (c) Influence of

rewards received and risk on current choice declines with time. Logistic regression coefficient for the

experienced value of the risky target, estimated as the sum of the reward and risk received for choosing

that target, plotted as a function of trial lag. (d) Juice volume varies linearly with solenoid open-time.

(e) Monkeys preferred the risky target despite receiving a net loss of juice. Probability of choosing the

risky target as a function of risk for control experiment in which choosing the risky target resulted in a

two-thirds chance of a lower-than-average reward and a one-third chance of a higher-than-average reward.

(f) Monkeys were indifferent to targets that changed color. Probability of choosing novel colored target

when reward size was equal is plotted for both monkeys. Broome, n = 2,032 trials; Niko, n = 2,008 trials.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Neuronal activity was modulated both by which target was chosenand whether that target was associated with uncertain rewards. Overall,the activity of 22/41 (53.7%) studied neurons was significantly modu-lated by whether the risky target was chosen (14/25 neurons in monkeyBroome and 8/16 neurons in monkey Niko). Across the population,average neuronal activity was greater for risky target choices than forcertain target choices (ANCOVA: fixation epoch, F ¼ 3.957, Po 0.05;pre-movement, F ¼ 11.321, P o 0.001; post-movement, F ¼ 2.346, P4 0.10), even when the effects of movement amplitude, latency, peakvelocity and direction were removed statistically.

We collected data from 39/41 CGp neurons in the same fourconditions of risk. Population neuronal activity increased system-atically with increases in risk throughout trials (Fig. 4; multipleregression: fixation epoch, rrisk ¼ 0.030, n ¼ 9,153, P o 0.005; pre-movement epoch, rrisk ¼ 0.048, n ¼ 9153, P o 0.00001; post-move-ment epoch, rrisk ¼ 0.047, n¼ 9153, Po 0.00001), roughly parallelingthe frequency of risky choices made by monkeys in these experiments(compare to Fig. 2a). Notably, the effect of risk on neuronal firing ratespersisted as a tonic change throughout trials but was maximal whensuperimposed upon phasic, movement-related responses (Fig. 4c).Consistent with previous reports, the CGp population was alsosensitive to whether the chosen target was in the neuronal responsefield, even while monkeys maintained central fixation before targetonset (Fig. 4; fixation epoch: rdirection ¼ 0.056, n ¼ 9,153, P o0.000001; pre-movement epoch: rdirection ¼ 0.067, n ¼ 9,153, P o0.000001; post-movement epoch: rdirection ¼ 0.125, n ¼ 9,153,P o 0.000001). These data indicate that CGp neuronal activity issensitive to reward uncertainty as well as to target choice. For anyparticular choice, either into or out of the response field, CGp neuronalactivity varied with risk.

Our data also indicate that the spatial selectivity of CGp neurons wasenhanced by increasing risk: in high-risk blocks, the neuronal popula-tion more accurately discriminated movement direction, visualized

here as a larger separation between black and gray lines (Fig. 4; post-hoc Tukey’s honestly significant difference tests for saccade direction;fixation epoch: low-risk, not significant (n.s.); high-risk, Po 0.000001;pre-movement epoch: low-risk, n.s.; high-risk, P o 0.000001; post-movement: low-risk, P o 0.01; high-risk, P o 0.00000001). Thus,increasing risk did not seem to be associated with global or uniformchanges in neuronal activity, but rather with selective enhancement oftask-related neuronal activity.

The enhancement of spatial sensitivity by risk is difficult to explainby global changes in arousal. Nevertheless, we sought to examinewhether changes in neuronal activity associated with risky choicesreflected changes in autonomic arousal. Correlates of autonomicarousal such as heart rate, galvanic skin response, and cortisol levelsare elevated during gambling in humans31–33 and are attenuated inmedial prefrontal lesion patients with poor impulse control in gam-bling tasks34. We therefore recorded the heart rates of both monkeyswhile they performed visual gambling trials in a separate set ofexperiments (monkey Niko: 4,471 trials; monkey Broome: 3,780 trials).Although there were fluctuations in heart rate over the course ofbehavioral sessions as well as upon block changes, we found nosystematic effect of risk on heart rate (Fig. 5).

We examined other potential behavioral correlates of arousal ormotivation that might vary with reward uncertainty35. Reactiontimes were not significantly faster when monkeys made risky choices(F ¼ 2.7, P 4 0.1) but were significantly slower on trials after deliveryof either unusually large or small rewards (F ¼ 13.8, P o 0.000001).Similarly, risky choices had no influence on peak saccade velocity scaledby saccade amplitude (ramplitude ¼ 0.519, P o 0.000001; rrisky choice ¼�0.00850, P 4 0.322), but peak saccade velocity was significantlyhigher after delivery of both the smallest and largest rewards

Firi

ng r

ate

(Hz)

30

15

0

500 msMovement onset

BR030429

a

b c

Risky choices out of RFCertain choices out of RF

Risky choices into RFCertain choices into RF

Firi

ng r

ate

(Hz)

Choice

30

0

15

Out of RF

Into RF

CertainRisk (reward CV)

0.80.40.0 0.2 0.6Risky

4

8

6

10

0.80.60.40.20.0

Firi

ng r

ate

(Hz)

Risk (reward CV)

Fixation Pre-movement Post-movement

Into RFOut of RF

cba

Figure 4 Target risk enhances neuronal activity in CGp as well as sensitivity

to movement direction. Plots of average (± s.e.m.) neuronal firing rate as a

function of risk (reward CV) for the epochs after fixation (a), before movement

onset (b) and after movement onset (c). Activity was greater for movements

made into the RF and scaled with the degree of risk.

Figure 3 Posterior cingulate neurons are risk-sensitive. (a) Post-stimulus

time histogram for a single CGp neuron aligned on movement onset.

Points indicate average (± s.e.m.) firing rate measured in 100-ms bins.

Firing rate was greater for choices into RF than out of RF, as well as for

risky choices than for certain choices. Gray shaded box indicates 200-ms

epoch analyzed in b,c. (b,c) Average firing rate of the same CGp neuron

plotted as a function of choice (risky or certain) and reward CV for choices

in the neuron’s preferred (black) and non-preferred (gray) directions.

Firing rate systematically increased for risky choices as well as with

increasing reward CV for movements in the preferred direction (b: t-test,

t ¼ 5.55, df ¼ 375, P o 0.000001; c: multiple regression, r ¼ 0.230,

df ¼ 375, P o 0.00003).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

(F ¼ 23.858, P o 0.000001). Thus, while the saccade metrics andchoices of monkeys suggested that monkeys were, in fact, sensitive torisk, the effects of risk on firing rate are not readily explained by globalchanges in arousal.

CGp neurons track subjective target preferences

Despite the fact that the amount of reward received from choosing eachtarget was equivalent over time, monkeys systematically preferred therisky target. Moreover, neuronal activity in CGp increased with theselection of risky targets. These data suggest that, under these condi-tions, the activation of CGp neurons reflects subjective biases for targetsassociated with uncertain rewards. If this is true, then firing rate on anyparticular trial should be more closely associated with subjectivepreferences for a particular target than with the actual rewardsharvested by choosing it. We tested this hypothesis by first examiningneuronal activity as a function of prior rewards received and, second, bycomputing an estimate of subjective target utility on the basis of theinfluence of both risk and reward received for previous choices27 andasking whether neuronal activity was related to this measure.

First, we examined neuronal activity in the CGp population on eachtrial as a function of the size of the rewards delivered on the previoustrial to ask whether neuronal activity reflected in any simple way theactual rewards received (Fig. 6). In all three measured epochs, firingrate discriminated between target choices into and out of the responsefield (fixation: F ¼ 29.951, P o 0.000001; pre-movement: F ¼ 42.226,P o 0.0000001; post-movement: F ¼ 148.152, P o 0.0000001).Moreover, firing rate was elevated when monkeys received rewardsthat deviated from the average, certain value (fixation: F ¼ 5.358,P o 0.0000001; pre-movement: F ¼ 7.022, P o 0.0000001; post-movement: F ¼ 5.649, P o 0.0000001). However, CGp neuronalactivity did not distinguish between the lowest and highest rewardsreceived on previous trials (post-hoc Tukey honestly significant differ-ence tests, P 4 0.15 in all epochs). Thus, neuronal responses in CGpdid not monotonically reflect the actual value of rewards received. Wesuspected that, instead, CGp responses for a particular target choicereflected monkeys’ subjective valuation of the target on the basis oftheir own internal preferences (compare with Fig. 2b).

As a second test of this hypothesis, we computed a local estimate ofsubjective target preference, which we refer to as subjective targetutility. Analysis of behavioral data, described above, showed that boththe risk and reward value associated with prior choices roughly equallybiased the probability of choosing the risky target on subsequent trials.

We therefore first computed the experienced values of the target in theresponse field (RF) and the target outside the response field (ORF)using the risk and reward outcomes associated with prior choices, asabove. The experienced value of the RF target VRF and the experiencedvalue of the ORF target VORF on a trial were computed as

VRF ¼ Reward receivedRF + RiskRF ð4Þ

VORF ¼ Reward receivedORF + RiskORF ð5Þ

The subjective utility of the target in the response field URF on a giventrial t was then estimated according to the following algorithm:

URFðtÞ ¼ Sn¼ 1 to i½VRFðt � nÞ � VORFðt � nÞ� � an ð6Þ

where an is the logistic coefficient for the difference in the experiencedvalue of the RF and ORF targets lagged n trials. Multiple logisticregression analysis of the probability of an RF target choice as afunction of the difference in experienced value for the two targets(VRF � VORF) on each of up to ten prior trials was used to derive theweighting factor an. We found that differences in the experienced valueof the two targets significantly influenced the probability of choosingthe RF target at all lags up to ten trials (AIClags1–10 o AICs for all othercombinations), so the weighting term an was therefore weighted on thebasis of the logistic regression coefficients for each trial lag with i set at10 trials. The utility of the target outside the response field wascomputed as the sign-reversed utility of the response field target.High target utility implied that the monkey frequently chose, andtherefore preferred, a particular target in the past ten trials. Thesecomputations were performed across the entire dataset.

We examined the percentage of cells in the population that weresignificantly correlated with the subjective utility of the response fieldtarget and whether the response field target was chosen, using amultiple linear regression analysis with movement latency, amplitudeand peak velocity as co-regressors. The firing rates of 64% of cells weresignificantly modulated by target utility as well as target choice in any oftwelve epochs examined (see Methods; Fig. 7). Modulations of firingrate by target utility and target choice, however, varied over time aswell. Over 20% of the studied population of neurons showed asignificant correlation between firing rate and subjective target utilityduring initial fixation before target onset (Fig. 7a), and the percentageof neurons with a significant correlation between firing rate and targetutility gradually increased and peaked at around one-third of the

Risk (reward CV)0.80.60.40.20.0 1.0

Hea

rt r

ate

(bea

ts/m

in)

120

110

105

100

130

125

115

Monkey BroomeMonkey Niko

Figure 5 Average heart rate does not increase with increasing risk. Average

heart rate (± s.e.m.) measured at 12 Hz by means of pulse oximetry is

plotted as a function of risk (reward CV) for both monkeys.

4

10

8

12

25020015010050

6

Firi

ng r

ate

(Hz)

Reward on previous trial (ms)

Fixation Pre-movement Post-movement

Into RFOut of RF

a b c

Figure 6 Population neuronal activity reflects monkeys’ preference for riskytargets but not prior reward outcomes. Firing rate is plotted as a function of

reward on previous trial for epochs after fixation (a), before movement onset

(b) and after movement onset (c). Activity was greater for movements made

into the RF as well as when relatively small or large rewards were delivered on

previous trial.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

population in the epochs after movement onset. In contrast, fewer than5% of neurons showed a correlation between eventual target choice andfiring rate during fixation before target onset, and the percentage ofcells with a significant correlation peaked just after target choice andpersisted through the end of the trial. These trends suggest a gradualtemporal shift in the information carried by CGp neurons, initiallyfavoring the subjective utility of available targets and later reflectingtarget choice and, to a lesser degree, target utility.

Across the studied population of neurons, firing rate was positivelycorrelated with target utility for all trials in which monkeys chose thetarget in the neuronal response field (Fig. 7b, black lines; post-move-ment epoch: r ¼ 0.146, Po 0.0000001) and negatively correlated withtarget utility when monkeys chose the target out of the response field(Fig. 7b, gray lines; post-movement epoch: r ¼ �0.0893, P o 0.005).Activity in the CGp population was therefore well correlated with ameasure of target utility, estimated from the monkeys’ history ofchoices, rewards received and associated risk. Note that black andgray lines overlap at low measures of target utility and diverge withincreasing values, indicating that the ability of CGp neurons todiscriminate target choice depended on the utility of those targets.

If, as these data suggest, CGp neurons carry information about boththe direction and subjective utility of movements monkeys make, animportant question remaining is whether such activity causes—orreflects—the animals’ choices. To address this issue, we examined thetime course of neuronal activity and the pattern of behavioral choicesafter a switch in the location of risky and certain targets. We found thatthe proportion of risky choices rose steeply from indifference (0.5) andreached a plateau within 15–20 trials (averaged across multiple valuesof risk; Fig. 8a). Similarly, neuronal activity in the 200–400 ms epochafter movement onset increased by approximately 50% and began toplateau within 15–20 trials from a block change (Fig. 8b). Piecewiselinear regression analysis for the probability of a risky choice andneuronal activity as a function of time after a switch in the location ofthe risky target showed break points, respectively, of 19.4 (r ¼ 0.889,r2 ¼ 79.1%) and 24.6 (r ¼ 0.868, r2 ¼ 75.4%) trials, suggesting thatneuronal activity in CGp closely followed changes in monkeys’ choices.This interpretation should be viewed with caution, however, as piece-wise linear regression break points can be sensitive to the range of dataincluded in the model.

Taken together, our results indicate that a subset of neurons inposterior cingulate cortex carries information about the subjectiveutility of targets in the visual world. Indeed, neuronal activity inCGp mirrors the behavioral sensitivity of monkeys to risk. Thesedata are consistent with a role for CGp in signaling the subjectivesalience of locations in the visual scene.

DISCUSSION

Economists, experimental psychologists and behavioral ecologistshave long argued that decision making depends on the conversionof external variables into a common internal currency of value. With-out such a common currency, neither animals nor people would beable to choose adaptively between apples and oranges, much lessactivities as disparate as eating and mating. While it is readilyaccepted that internal representations of value lie at the very core ofdecision making, most neurobiological studies of the decisionprocess have manipulated objective value, typically by changing rewardsize or probability, because these factors are easily controlled andquantified. On the other hand, subjective value can be measuredonly indirectly, and attempts to correlate it with neuronal activityare, not surprisingly, rare14,17.

The visual gambling task used in the current study affords a uniqueopportunity to examine neural activity under conditions in whichsubjective value, but not objective value, varied, as indicated bysubjects’ choices. Although there was no apparent reason for monkeysto prefer one option or the other, they showed systematic preferencesfor targets offering uncertain rewards, much like hungry birds6,20,28,36,typical adolescents37–39 and people addicted to drugs40 or pathologicalgambling32. Moreover, preference for the uncertain reward increasedparametrically with the coefficient of variation of reward or risk, inaccord with some recent findings in humans3.

Similarly, neurons in posterior cingulate cortex were also risk-sensitive, carrying information about both the direction of impendingmovement, as shown previously24,30, and the uncertainty of rewardsassociated with this movement. These findings both corroborate andextend the findings of a previous study16, in which posterior cingulateneurons were reported to be sensitive to reward size and predictability,manipulations that also presumably influenced subjective value. How-ever, the previous study could not discern whether such modulations inneuronal activity reflected subjective preferences for larger or moresurprising rewards. In the present study, the average reward valueof each target was held constant, yet monkeys demonstrated clear

4

10

8

12

1.00.50.0–0.5 1.5

6

2.0 2.5

Target utilityRF choice

Cel

ls s

igni

fican

tly c

orre

late

d (%

)

Movementonset

Time during trials

Rewardoffset

Targetonset

0

30

20

40

10

50a b

Pos

t-m

ovem

ent r

espo

nse

(Hz)

Target utility

Into RF

Out of RF

Figure 7 CGp neurons carry information about both target choice and

subjective target utility. Multiple linear regression analysis was performed

with utility, RF choice and saccade kinematic parameters as co-factors.

(a) Percentage of cells with a significant correlation are plotted as a function

of time during trials. (b) Population activity is correlated with subjective

target utility. Firing rate measured after saccade onset is plotted as a function

of target utility, estimated from the influence of risk and rewards received on

target choices over the previous ten trials.

Pro

babi

lity

of r

isky

cho

ice

Trial within block Trial within block403020100

0.8

0.7

0.6

0.5

0.9

50 403020100 50

Firi

ng r

ate

(Hz) 10

8

6

4

12

a b Post-movement activityBehavior

Figure 8 Both the frequency of risky target choices and neuronal activity

gradually increase after block changes. Probability of risky choice (a) and

neuronal activity (b) are plotted as a function of trial after a change in the

location of the risky target. Behavioral preference for risky target and

neuronal firing rate increased at a similar rate and reached asymptote

within 15–20 trials.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

preferences for one option over the other, thus permitting dissociationof subjective and objective value. Under these conditions, CGp neuronswere sensitive to subjective target utility, consistent with recent findingsin posterior parietal cortex, a premotor area that has been implicated inoculomotor decision making14,17.

Recent studies have suggested that orienting decisions are computed,in part, by scaling neuronal responses by target value and thencomparing them with a threshold41,42. Our data indicate that CGpneurons signal subjective biases for uncertain rewards (or, perhaps, thepotential to receive a large reward) rather than objective target value.Thus, CGp appears to carry spatial information that is scaled bysubjective preferences for particular patterns of reward outcome.Other investigators have shown recently that the responses of apopulation of midbrain dopamine neurons are selectively enhancedby reward uncertainty43. Such dopamine responses may indirectlyfacilitate neuronal activity in CGp through projections to anteriorcingulate cortex, a major input to posterior cingulate cortex44. Severalneuroimaging studies have also shown hemodynamic responses tooutcome uncertainty in midline cortical areas such as anterior cingulatecortex, orbitofrontal cortex and precuneus45,46. Our data extend thosefindings, demonstrating a direct relationship between subjective pre-ferences for uncertain rewards, or the opportunity to harvest arelatively large reward, and the activity of neurons thought to partici-pate in the allocation of attention21,22.

Neuronal activity in CGp reflects subjective preferences for riskytarget locations during fixation as well as before and after saccade onset,suggesting that this area contributes to visuospatial biases guidingorienting22,30,47. Our data are in sharp contrast with findings inposterior parietal cortex, where modulations by local fractionalincome14 or subjective desirability17 are found to emerge around thetime of target onset and end after the eye movement. Therefore,posterior parietal cortex has been proposed to signal the relativevalue of potential eye movements17 but is unlikely to actually computethis value, as the signals seem to be ‘reset’ at the start of each trial14. Incontrast, modulations in neuronal activity in CGp seem to persistacross trials, and the timing of these modulations is consistent with arole in predicting and/or evaluating subjective value16,41. Thus, CGpmay convey information about subjective target value that scalesneuronal signals in parietal cortex.

Our data also indicate that the spatial sensitivity of neurons in CGpis enhanced under conditions of risk or uncertainty. This result echoes arecent finding that the spatial selectivity of parietal neurons is greatestwhen target value is high14. Because both parietal and cingulate corticeshave been implicated in the allocation of spatial attention, suchenhancement may reflect heightened attention to regions of spacewith high subjective value48. We speculate that enhanced neuronalactivity associated with risky rewards biases attention spatially, markinglarge payoffs as salient for guiding behavior48 and thereby favoringbehavioral responses to risky targets. Such a link between risk pre-ference, salience, attention and action has profound implications notonly for oculomotor decision making, but also for why people andanimals sometimes demonstrate irrational and even harmful prefer-ences for risky behaviors.

In conclusion, two monkeys were systematically risk prone whenoffered choices of targets associated with certain and uncertain fluidrewards. Posterior cingulate neurons were similarly risk sensitive, andtheir firing rates conveyed information about the direction of impend-ing movements as well as the subjective utility of those movements.Moreover, the spatial sensitivity of CGp neurons was enhanced underconditions of high risk. Neurophysiological studies of risk preferences,as reported here, may serve as an important model for probing the

neural processes that underlie pathological risk taking in individualswith addictions to drugs, sex, food or gambling.

METHODSSurgical and training procedures. All procedures were approved by the Duke

University Institutional Animal Care and Use Committee and were designed

and conducted in compliance with the Public Health Service’s Guide for the

Care and Use of Animals. Initially, a head restraint prosthesis and scleral search

coil49 were implanted using standard surgical techniques30. Six weeks later,

animals were habituated to head restraint and trained to perform oculomotor

tasks for liquid rewards. A second surgical procedure was then performed to

implant a stainless steel recording chamber (Crist Instruments) over posterior

cingulate cortex at the intersection of the interaural and midsagittal planes. The

chamber was kept sterile with regular antibiotic washes and sealed with sterile

caps. Animals received analgesics and antibiotics after all surgeries.

Behavioral techniques. Horizontal and vertical eye positions were sampled at

500 Hz (Riverbend Instruments) and recorded by computer (ryklinsoftware.

com). Visual stimuli were LEDs (LEDtronics), which were illuminated to

appear yellow, red or green to normal human observers, fixed on a tangent

screen 144.78 cm (57 inches) from the animals’ eyes and forming a grid of

points separated by 11, spanning 491 horizontally and 411 vertically.

Behavioral datasets were collected for 12 sessions from both monkeys before

physiological recording. A 300-ms broadband noise before juice delivery served

as a secondary reinforcer on all correct trials. During visual gambling trials, one

target was placed in the response field of the neuron under study, while the

other target was placed diametrically opposite the fixation point. One target

was associated with a ‘certain’ reward outcome of 150 ms access to juice on

every trial, while the other ‘risky’ target was randomly rewarded with less than

150 ms on half of trials and greater than 150 ms on the other half of trials

(mean ¼ 150 ms across trials). The locations of the certain and risky targets, as

well as the coefficient of variation in reward for the risky target, were varied

every 50 trials. Heart rate was measured at 12 Hz by pulse-oximetry (SurgiVet)

on eight sessions for both monkeys.

Two control experiments were performed for risk and novelty. First, for the

risk control, visual gambling trials were as above with the following important

difference: the risky target was associated with larger-than-average rewards on

one-third of trials and smaller-than-average rewards on two-thirds of trials (as

compared with one-half larger-than-average and one-half smaller-than-average

rewards in standard visual gambling trials). Second, for the novelty control,

reward size was held constant at 150-ms access to juice for both targets while

novelty was introduced by systematically changing the color of one of the

targets during reward delivery. The color of the ‘monotonous’ target remained

yellow throughout trials, while the color of the ‘novel’ target randomly changed

color from yellow to green on half of trials, and from yellow to red on the other

half of trials.

Microelectrode recording techniques. Single electrodes (Frederick Haer) were

lowered under physiological guidance until the waveform of a single neuron

was isolated. Individual action potentials were identified in hardware by time

and amplitude criteria (BAK Electronics) and recorded by computer at 25 KHz.

Neurons were selected for recording experiments on the basis of the quality of

isolation and apparent task-sensitivity. Neuronal activity was first monitored

during 100–400 single-target trials to identify the neuron’s response field and

select appropriate target locations for subsequent visual gambling trials. Data

were collected for 4 to 14 blocks of gambling trials for each neuron, depending

on the duration and quality of isolation.

Following some recording sessions, we confirmed the location of the

electrode using a hand-held digital ultrasound device (Sonosite 180) placed

against the recording chamber50. Ultrasound images taken in the sagittal plane

showed that recordings were made in areas 23 and 31 in the cingulate gyrus and

ventral bank of the cingulate sulcus, anterior to the intersection of the marginal

and horizontal rami30.

Analysis. Data were analyzed off-line using custom software (Eyemove,

supported by K. Pearson, D. Sparks Laboratory, Baylor College of Medicine),

which computed saccade direction, amplitude, latency, peak velocity and times


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

of spike occurrence. For behavioral data, logistic regression was used to

estimate the effects of experienced rewards and risk associated with prior

choices on the probability of choosing the risky target. Statistics were computed

using Statistica 6 or Matlab.

Firing rates were measured for each trial during 12 200-ms intervals

aligned with the time of an event during the trial: (i) 0–200 ms after the

onset of fixation; (ii) 0–200 ms after the illumination of the eccentric target;

(iii) 200–0 ms before target offset; (iv) 200–0 ms before movement onset;

(v) 0–200 ms after movement onset; (vi) 200–400 ms after movement onset;

(vii) 0–200 ms after the reinforcing noise burst; (viii) 200–0 ms before juice

onset and (ix–xii) 0–200 ms, 200–400 ms, 400–600 ms and 600–800 ms after

juice delivery. Analysis of firing rates focused on three 200 ms intervals in

particular: 0–200 ms after the onset of fixation (fixation epoch), 200–0 ms

before target offset (pre-movement epoch) and 200–400 ms after movement

onset (post-movement epoch). Multiple regression was used to quantify the

relationship between neuronal firing rate and target risk, independent of the

effects of latency, amplitude and peak velocity of eye movements.

In addition, neuronal activity was also analyzed as a function of subjective

target utility, which was estimated by assuming that the experienced value of

each target was the sum of both the received rewards and risk associated with

prior choices. Experienced value for each target was incorporated into the

model of subjective target utility by first multiplying this value by a weighting

factor and then summing across up to ten previous trials. Weighting factors for

the influence of experienced value at each trial lag were estimated by logistic

regression of the effects of experienced target value on the probability of

choosing the response field target. Aikake’s Information Criterion (AIC) was

used to evaluate the inclusion of experienced target value for trials at different

lags in the model.

ACKNOWLEDGMENTSWe thank G. Haghighian and J. Crowley for their contribution to the earlystages of this work. We also thank S. Roberts for assistance in animal careand S. Huettel, J. Stowe, P. Glimcher, R. Deaner, J. Roitman, M. Bendiksby,S. Shepherd and A. Khera for valuable comments on the manuscript.Supported by the Klingenstein Foundation, the Duke Provost’s CommonFund and the National Eye Institute.




1. Bateson, M. & Kacelnik, A. Starlings’ preferences for predictable and unpredictabledelays to food. Anim. Behav. 53, 1129–1142 (1997).

2. Kahneman, D. & Tversky, A. Prospect theory: an analysis of decision under risk.Econometrica 47, 263–291 (1979).

3. Weber, E.U., Shafir, S. & Blais, A.R. Predicting risk sensitivity in humans and loweranimals: risk as variance or coefficient of variation. Psychol. Rev. 111, 430–445(2004).

4. Bernouilli, D. The Works (Birkhauser, Boston, 1758).5. Rode, C., Cosmides, L., Hell, W. & Tooby, J. When and why do people avoid unknown

probabilities in decisions under uncertainty? Testing some predictions from optimalforaging theory. Cognition 72, 269–304 (1999).

6. Bateson, M. Recent advances in our understanding of risk-sensitive foraging prefer-ences. Proc. Nutr. Soc. 61, 509–516 (2002).

7. Von Neumann, J.V. & Morgenstern, O. Theory of Games and Economic Behavior(Princeton University Press, Princeton, New Jersey, 1944).

8. McClure, S.M. et al. Neural correlates of behavioral preference for culturally familiardrinks. Neuron 44, 379–387 (2004).

9. Arnauld, A. & Nichole, P. The Art of Thinking: Port-Royal Logic (Bobbs-Merrill,Indianapolis, 1662).

10. Stephens, D.W. & Krebs, J.R. Foraging Theory (Princeton University Press, Princeton,New Jersey, 1986).

11. Platt, M.L. & Glimcher, P.W. Neural correlates of decision variables in parietal cortex.Nature 400, 233–238 (1999).

12. 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).

13. Kawagoe, R., Takikawa, Y. & Hikosaka, O. Expectation of reward modulates cognitivesignals in the basal ganglia. Nat. Neurosci. 1, 411–416 (1998).

14. Sugrue, L.P., Corrado, G.S. & Newsome, W.T. Matching behavior and the representationof value in the parietal cortex. Science 304, 1782–1787 (2004).

15. Lauwereyns, J., Watanabe, K., Coe, B. & Hikosaka, O. A neural correlate of response biasin monkey caudate nucleus. Nature 418, 413–417 (2002).

16. McCoy, A.N., Crowley, J.C., Haghighian, G., Dean, H.L. & Platt, M.L. Saccade rewardsignals in posterior cingulate cortex. Neuron 40, 1031–1040 (2003).

17. Dorris, M.C. & Glimcher, P.W. Activity in posterior parietal cortex is correlated with therelative subjective desirability of action. Neuron 44, 365–378 (2004).

18. Schultz, W. Neural coding of basic reward terms of animal learning theory, game theory,microeconomics and behavioural ecology. Curr. Opin. Neurobiol. 14, 139–147 (2004).

19. Roesch, M.R. & Olson, C.R. Neuronal activity related to reward value and motivation inprimate frontal cortex. Science 304, 307–310 (2004).

20. Kacelnik, A. Normative and descriptive models of decision making: time discountingand risk sensitivity. Ciba Found. Symp. 208, 51–67 (1997).

21. Hopfinger, J.B., Buonocore, M.H. & Mangun, G.R. The neural mechanisms of top-downattentional control. Nat. Neurosci. 3, 284–291 (2000).

22. Small, D.M. et al. The posterior cingulate and medial prefrontal cortex mediate theanticipatory allocation of spatial attention. Neuroimage 18, 633–641 (2003).

23. Berman, R.A. et al. Cortical networks subserving pursuit and saccadic eye movements inhumans: an FMRI study. Hum. Brain Mapp. 8, 209–225 (1999).

24. Olson, C.R., Musil, S.Y. & Goldberg, M.E. Single neurons in posterior cingulate cortexof behaving macaque: eye movement signals. J. Neurophysiol. 76, 3285–3300(1996).

25. Rescorla, R.A. & Wagner, A.R. A theory of Pavlovian conditioning. Variations in theeffectiveness of reinforcement and nonreinforcement. in Classical Conditioning II:Current Research and Theory (eds. Black, A.H. & Prokasy, W.F.) (Appleton-Century-Crofts, New York, 1972).

26. Sutton, R.S. & Barto, A.G. Toward a modern theory of adaptive networks: expectation andprediction. Psychol. Rev. 88, 135–170 (1981).

27. Barraclough, D.J., Conroy, M.L. & Lee, D. Prefrontal cortex and decision making in amixed-strategy game. Nat. Neurosci. 7, 404–410 (2004).

28. Caraco, T. White crowned sparrows (Zonotricha leucophrys): foraging preferences in arisky environment. Behav. Ecol. Sociobiol. 8, 820–830 (1983).

29. McGlothlin, W.H. Stability of choices among uncertain alternatives. Am. J. Psychol. 69,604–615 (1956).

30. Dean, H.L., Crowley, J.C. & Platt, M.L. Visual and saccade-related activity in macaqueposterior cingulate cortex. J. Neurophysiol. 92, 3056–3068 (2004).

31. Ladouceur, R., Sevigny, S., Blaszczynski, A., O’Connor, K. & Lavoie, M.E. Video lottery:winning expectancies and arousal. Addiction 98, 733–738 (2003).

32. Sharpe, L. Patterns of autonomic arousal in imaginal situations of winning and losing inproblem gambling. J. Gambl. Stud. 20, 95–104 (2004).

33. Meyer, G. et al. Casino gambling increases heart rate and salivary cortisol in regulargamblers. Biol. Psychiatry 48, 948–953 (2000).

34. Bechara, A. & Damasio, H. Decision-making and addiction (part I): impaired activationof somatic states in substance dependent individuals when pondering decisions withnegative future consequences. Neuropsychologia 40, 1675–1689 (2002).

35. Hikosaka, O., Sakamoto, M. & Usui, S. Functional properties of monkey caudateneurons. III. Activities related to expectation of target and reward. J. Neurophysiol.61, 814–832 (1989).

36. Marsh, B. & Kacelnik, A. Framing effects and risky decisions in starlings. Proc. Natl.Acad. Sci. USA 99, 3352–3355 (2002).

37. Kelley, A.E., Schochet, T. & Landry, C.F. Risk taking and novelty seeking in adolescence:introduction to part I. Ann. NY Acad. Sci. 1021, 27–32 (2004).

38. Doremus, T.L., Varlinskaya, E.I. & Spear, L.P. Age-related differences in elevated plusmaze behavior between adolescent and adult rats. Ann. NY Acad. Sci. 1021, 427–430(2004).

39. Chambers, R.A., Taylor, J.R. & Potenza, M.N. Developmental neurocircuitry of motiva-tion in adolescence: a critical period of addiction vulnerability. Am. J. Psychiatry 160,1041–1052 (2003).

40. Bickel, W.K., Giordano, L.A. & Badger, G.J. Risk-sensitive foraging theory elucidatesrisky choices made by heroin addicts. Addiction 99, 855–861 (2004).

41. McCoy, A.N. & Platt, M.L. Expectations and outcomes: decision-making in the primatebrain. J. Comp. Physiol. A Neuroethol. Sens. Neural Behav. Physiol. 191, 201–211(2005).

42. Gold, J.I. & Shadlen, M.N. Banburismus and the brain: decoding the relationshipbetween sensory stimuli, decisions, and reward. Neuron 36, 299–308 (2002).

43. Fiorillo, C.D., Tobler, P.N. & Schultz, W. Discrete coding of reward probability anduncertainty by dopamine neurons. Science 299, 1898–1902 (2003).

44. Vogt, B.A. & Pandya, D.N. Cingulate cortex of the rhesus monkey: II. Cortical afferents.J. Comp. Neurol. 262, 271–289 (1987).

45. Critchley, H.D., Mathias, C.J. & Dolan, R.J. Neural activity in the human brain relating touncertainty and arousal during anticipation. Neuron 29, 537–545 (2001).

46. Dickhaut, J. et al. The impact of the certainty context on the process of choice. Proc.Natl. Acad. Sci. USA 100, 3536–3541 (2003).

47. Mesulam, M.M. Spatial attention and neglect: parietal, frontal and cingulate contribu-tions to the mental representation and attentional targeting of salient extrapersonalevents. Phil. Trans. R. Soc. Lond. B 354, 1325–1346 (1999).

48. Damasio, A. Descartes’ Error: Emotion, Reason, and the Human Brain (Avon, New York,1995).

49. Judge, S.J., Richmond, B.J. & Chu, F.C. Implantation of magnetic search coils formeasurement of eye position: an improved method. Vision Res. 20, 535–538(1980).

50. Glimcher, P.W. et al. Application of neurosonography to experimental physiology.J. Neurosci. Methods 108, 131–144 (2001).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Reductions in neural activity underlie behavioralcomponents of repetition priming

Gagan S Wig, Scott T Grafton, Kathryn E Demos & William M Kelley

Repetition priming is a nonconscious form of memory that is accompanied by reductions in neural activity when an experience

is repeated. To date, however, there is no direct evidence that these neural reductions underlie the behavioral advantage afforded

to repeated material. Here we demonstrate a causal linkage between neural and behavioral priming in humans. fMRI (functional

magnetic resonance imaging) was used in combination with transcranial magnetic stimulation (TMS) to target and disrupt activity

in the left frontal cortex during repeated classification of objects. Left-frontal TMS disrupted both the neural and behavioral

markers of priming. Neural priming in early sensory regions was unaffected by left-frontal TMS—a finding that provides evidence

for separable conceptual and perceptual components of priming.

Recent experience with an item leads to quicker recognition andclassification of that item upon subsequent encounters. This implicitform of memory is commonly referred to as ‘‘behavioral priming’’ andoccurs even in the absence of conscious remembering1,2. Neuroscien-tific investigations consistently reveal reductions in neural activity thataccompany this repetition-based learning facilitation. These activityreductions are seen both at the level of single cells in nonhumanprimates3–5 and across a host of brain areas extending from posteriorsensory to frontal cortices in humans6–10. Although this neural phe-nomenon generalizes across a range of behavioral priming situations,the loci of such ‘‘neural priming’’ effects vary and are restricted to asubset of the brain regions engaged during task performance with novelmaterial. Repeated semantic classification of visually presented objects,for example, consistently yields reduced activity in extrastriate visualregions and in the left inferior frontal gyrus (LIFG)11,12.

One speculation is that neural priming reflects fine-tuning of theneuronal response, or a suppression of neurons within the neuronalpopulation, perhaps because those neurons that are no longer neededdrop out of the responsive pool3,13,14. Such effects could occur eitherearly in the processing stream at the level of object recognition insensory cortices (perceptual priming) or at later stages during semanticclassification in frontal and temporal cortices (conceptual priming)15.The precise neurophysiological mechanism that reduces neural activityis unclear, and the relationship between neural priming and behavioralpriming remains indirect—evidenced only by the co-occurrence ofthese two phenomena. A fundamental question remains: does neuralpriming in a given brain region contribute to the behavioral facilitationafforded to repeated items, or is it epiphenomenal to behavior?

One possibility is that neural priming in brain regions thought to beinvolved in conceptual priming (e.g., LIFG) is necessary for behavioralpriming. Alternatively, neural priming in sensory cortices may subserve

behavioral priming and the neural reductions observed in frontalregions may simply reflect a feed-forward propagation of the changesin neural activity arising earlier in perceptual regions. To test thesepossibilities, we combined fMRI and TMS to target and transientlydisrupt left-frontal activity during an object classification task (Fig. 1).

In an initial fMRI session, subjects performed a semantic classifica-tion task (living/nonliving) for a series of objects that were eitherrepeated or novel. In a second session, subjects received TMS whileperforming the classification task on a new set of objects. The use ofTMS allowed for noninvasive and transient disruption of corticalactivity in a circumscribed region of cortex. We used activation mapsfrom the initial fMRI session, which compared trials with novel objectsto those with repeated objects, to identify subject-specific neuralpriming foci within the LIFG. These single-subject activation mapswere then superimposed on the subject’s anatomical brain image andused to guide the positioning of the TMS coil on the subject’s head.This approach permitted real-time, continuous monitoring of the coilposition with respect to the site of interest.

For each presented object, TMS was delivered to either the LIFGregion identified during fMRI scanning (left-frontal TMS) or the handregion of left motor cortex (control-site TMS). The motor region wasincluded as a control site to ensure that TMS effects were specific toleft-frontal stimulation and not a property of global cortical disruption.During the TMS session, each object was presented twice andwas accompanied by a 10-Hz train of stimulation lasting500 ms. Onset of the TMS was tailored to each subject’s individualresponse profile from the initial fMRI session (Fig. 2; see Methods).

To assess both the behavioral and neurophysiological consequences ofTMS, we performed a second fMRI scan on each subject immediatelyafter the TMS session. Critically, this post-TMS scanning sessionallowed behavioral responses to be recorded in the absence of potentially


Department of Psychological and Brain Sciences, Center for Cognitive Neuroscience, Dartmouth College, Hanover, New Hampshire 03755, USA. Correspondence should beaddressed to G.S.W. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

confounding TMS effects (such as peripheral nerve stimulation) andalso permitted simultaneous recording of the neural correlates of thebehavior. During this final session, subjects performed the living/nonliving classification task for three types of objects: novel objects,repeated objects that were previously presented with left-frontal TMSduring the first two presentations, and repeated objects that werepreviously presented with control-site TMS during the first twopresentations. Of interest were the neural activity and response latenciesthat accompanied this third presentation of each repeated object.

RESULTS

Session 1: fMRI

Behaviorally, repeated objects were classified more quickly (560 ±27 ms (mean ± s.e.m. throughout)) than novel objects (621 ±35 ms) in the first fMRI session (t10 ¼ 5.99, P o 0.001). Consistentwith prior neuroimaging work, this behavioral facilitation for repeatedobjects was accompanied by reduced neural activity in a network ofbrain regions (Fig. 3).

Session 2: TMS

Accuracy during the semantic classification task while undergoing TMSwas near perfect and was unaffected by TMS site (left-frontal mean,99.5% correct; control-site mean, 98.9% correct; P 4 0.55). Similarly,response latencies during TMS administration (left-frontal TMS: firstpresentation, 549 ± 40 ms; second presentation, 539 ± 35 ms; control-site TMS: first presentation, 601 ± 56 ms; second presentation, 546 ±38 ms) did not differ as a function of TMS site or repetition (maineffect of TMS site, F o 1; main effect of repetition, F1,10 ¼ 2.41,P ¼ 0.15; interaction, F o 1).

Response latencies acquired during TMS administration were notconsidered to be a pure measure of behavioral priming, as thesemeasures were contaminated by peripheral effects resulting fromTMS. Specifically, TMS at the intensity and frequency used hereproduced discomfort due to contraction of facial muscles. Also, asexpected, left-motor (control) TMS produced right-hand movement

while subjects responded. Instead, the behavioral and neural conse-quences of functionally targeted TMS were examined simultaneously ina subsequent fMRI session.

Session 3: fMRI

Subjects viewed three classes of objects during the final fMRI session:repeated objects that had been paired with left-frontal TMS (repeated-frontal), repeated objects that had been paired with left-motor TMS(repeated-control) and novel objects (novel). To assess the effects ofTMS on neural priming, we conducted hypothesis-driven region-of-interest (ROI) analyses on six brain regions: left posterior-inferiorfrontal cortex (Brodmann’s area 44 (LIFG; BA 44); x y z peak location:–42 9 24), left anterior-inferior frontal cortex (BA 45/47; �54 30 0), leftposterior temporal cortex (BA 21/37; �57 �66 �6), left fusiform gyrus(BA 37; �33 �75 �21), left middle occipital gyrus (left MOG; BA 19;�36 �84 9) and the left inferior occipital gyrus (left IOG; BA 18; �39,�83, �1). The six brain regions were identified in an unbiased manner

StudyTest R

N

R-C

R-F

N

Livingor

nonliving?

Novel > repeatedSession 1: fMRI

LIFG Primary motor cortex

Session 2: TMS

Session 3: fMRI

Figure 1 Experimental design. In Session 1, subjects made living/nonliving judgments for a series of colored objects. Each object was repeated six times

prior to fMRI scanning. During fMRI, subjects judged repeated objects and novel objects that were presented for the first time. Single-subject activation maps

comparing novel (N) to repeated (R) objects were used to identify neural priming effects in the LIFG. In Session 2, subjects received TMS to either the LIFG or

a control site (left motor cortex) while performing the same task on a new set of objects. Each object was presented twice and was accompanied by TMS for

each presentation. In Session 3, subjects were again imaged with fMRI while judging repeated objects that were previously paired with left-frontal TMS (R-F),

repeated objects that were previously paired with control-site TMS (R-C) and novel objects (N).

500 ms

10 Hz

500 ms

4,500 ms

800

600

Cla

ssifi

catio

n tim

e (m

s)

400

Median RTfor subject 1

TMS onsetfor subject 1200

0

a b

Figure 2 TMS timing parameters. (a) Objects were presented for 500 ms,

with a 4,500-ms inter-trial interval. Each stimulation consisted of a 10-Hztrain lasting for 500 ms. (b) TMS onset was catered to each subject’s

individual response times, with TMS stimulation time-locked to occur

250–310 ms before each subject’s median response time based on their

response latencies from the first MRI session.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

based on the group statistical map comparing novel to repeated objectsin Session 1.

Results revealed a functional dissociation in neural priming betweensensory and frontal cortices. An analysis of variance (ANOVA) exam-ining two factors, object type (repeated-frontal, repeated-control andnovel) and brain region (LIFG [BA45/47] and MOG [BA 18/19]),showed a significant main effect of object type (F2,20 ¼ 15.44,P o 0.001), no significant main effect of region (F1,10 ¼ 3.93,P ¼ 0.10) and a significant interaction between object type and region(F2,20 ¼ 9.80, P o 0.005). Planned comparisons showed that left-frontal TMS had no impact on the subsequent neural priming effectsobserved in the left MOG (novel 4 repeated, t10 ¼ 2.21, P o 0.05;Fig. 4a). By contrast, left-frontal TMS disrupted the ensuing neuralpriming reductions in the LIFG that normally accompany repeatedsemantic classification of objects (novel ¼ repeated, P 4 0.56;Fig. 4a). Importantly, control-site TMS did not disrupt neural primingin either region (MOG: novel 4 repeated, t10 ¼ 2.37, Po 0.05; LIFG:novel 4 repeated, t10 ¼ 2.05, P o 0.05; Fig. 4b). Additionally,when signal change for the two classes of repeated objects werecompared directly, a significant difference was noted in the LIFG(repeated-frontal 4 repeated control; t10 ¼ 2.25, P o 0.05) but notin the MOG (P 4 0.5).

A region of the left posterior temporal cortex (middle temporalgyrus; BA 21/37) showed a qualitatively similar pattern to the leftinferior frontal region. Left-frontal TMS also diminished the neuralreductions in the left posterior temporal cortex that are typicallyassociated with repeated semantic classifications (novel versusrepeated, P 4 0.55), whereas objects previously presented duringcontrol-site TMS showed a trend toward neural priming when com-pared to novel objects (novel 4 repeated, t10 ¼ 1.47, P ¼ 0.086;Supplementary Fig. 1 online). When compared directly, thedifference in neural activity between repeated-frontal objects andrepeated-control objects was significant (repeated-frontal 4 repeated-control, t10 ¼ 1.84, P o 0.05).

Three of the neural priming regions identified in Session 1 (left BA44, left fusiform gyrus and left IOG) did not show significant neural

priming effects (novel 4 repeated) in Session 3 after left-frontal TMSand control-site TMS. This is not surprising, given that Session 3 wasnecessarily less sensitive to neural priming effects than Session 1because of TMS constraints. In Session 1, repeated objects werepresented for the seventh, eight and ninth times. In Session 3, repeatedobjects were presented for only the third time.

The ROI analysis was based on activations observed in Session 1. Todetermine whether neural priming effects were present in any addi-tional brain regions during Session 3, we used a conjunction analysis.The conjunction analysis identified brain regions that were jointlyactive when classification of novel objects was contrasted with bothtypes of repeated objects (that is, novel 4 repeated-frontal AND novel4 repeated-control; P o 0.025). Results of this analysis revealedsignificant neural priming effects in the same region of the left MOG(as identified in Session 1) and a region of the left IOG (BA 18;�41 �87 �4) that overlapped with the IOG region identified inSession 1. Further analysis of the IOG region defined in this wayrevealed intact neural priming for both frontal-stimulated and control-site-stimulated objects (novel 4 repeated-frontal, t10 ¼ 2.52, Po 0.05;novel 4 repeated-control, t10 ¼ 3.86, P o 0.005). Neural activitydid not differ between classification of repeated-frontal and repeated-control objects (P ¼ 0.47).

Importantly, the behavioral effects mirrored the neural effectsobserved in the LIFG (Fig. 5). An ANOVA examining object type(repeated-frontal, repeated-control and novel) revealed a significantmain effect (F2,20 ¼ 4.62, P o 0.05). Left-frontal TMS significantlyreduced the subsequent behavioral facilitation that normally accom-panies repeated classification of objects. Specifically, responses torepeated objects that had been presented twice during left-frontalTMS (repeated-frontal) were no faster during their third repetitionthan responses to novel objects (novel [602 ms] ¼ repeated-frontal[594 ms], t10 ¼ 1.07, P¼ 0.15). Put simply, subjects responded to theserepeated objects as if they were doing so for the first time. Moreover, the

P < 0.01

Novel > repeated

Max

Figure 3 Neural priming before TMS. Whole-brain group statistical activation

map comparing novel to repeated items (P o 0.01) overlaid on an inflated

cortical rendering of the left hemisphere. Regions that survived a more

stringent statistical threshold (P o 0.001) were investigated further using

an ROI analysis in the post-TMS fMRI scan (Session 3). At this threshold,

repetition-related reductions were observed in the posterior (A) and anterior

(B) portions of the inferior frontal gyrus, middle temporal gyrus (C), fusiform

gyrus (D), middle-occipital gyrus extending into the superior occipital

gyrus (E) and the IOG (F). Weaker activations (P o 0.01) were noted

along the middle frontal gyrus in BA 9 and BA 10, and in the inferior

temporal gyrus in BA 21/37.

0.10

0.00

–0.10

–0.20

–0.30

Frontal stimulation

a

b

Control stimulation

–0.40

LIFG MOG

LIFG MOG

Neu

ral p

rimin

g

0.10

0.00

–0.10

–0.20

–0.30

–0.40

Neu

ral p

rimin

g

Figure 4 Neural priming after TMS. Graphs depict hemodynamic reductions

in activity in the LIFG and MOG following left-frontal TMS (a) and control-site

TMS (b). Signal change is reported as the difference in activity between

semantic classification of repeated objects and that of novel objects. Gray

bars denote a significant neural priming effect. Error bars indicate s.e.m.

Left-frontal TMS eliminated neural priming in the LIFG but did not disrupt

neural priming in the MOG. Control-site TMS had no effect on neural priming

in either region.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

disruption of behavioral priming after TMS was not a result of globalcortical disruption. Normal behavioral priming was observed forrepeated objects that had been presented during control-site TMS(repeated-control); these repeated objects were classified faster thannovel objects (novel [602 ms] 4 repeated-control [581 ms], t10 ¼ 3.34,Po 0.005). When compared directly, the difference in response latencybetween repeated-frontal objects and repeated-control objects was alsosignificant (t10 ¼ 1.82, P ¼ 0.05).

DISCUSSION

These findings provide direct evidence that neural priming in the left-frontal cortex is the basis for conceptual priming. Numerous studiesreport instances of neuronal3,4 and hemodynamic6–9,16–19 reductionsassociated with implicit memory and repetition-accompanied learning,but the correlative nature of these investigations has, until now,precluded a direct linkage between these two phenomena (for review,see ref. 11).

Decreases in neural activity can accompany other forms of learningas well. For example, practice-related hemodynamic decreases arecommonly observed as tasks become automated9,20–25, andsimilar decreases are reported in patients during the recovery ofmotor function following stroke26. Such practice-related reductionsoften occur in conjunction with increases in other corticalregions9,21–25,27–29, a finding that has prompted debate over whetherthe neural reductions reflect circuit efficiency (analogous to priming)or whether the neural reductions are simply a byproduct of the fact thatthe task is being performed in a different way (via recruitment ofdifferent brain regions). The present findings suggest that the neuro-physiological reductions that accompany procedural learning in theseother domains may well serve a causal purpose.

Neural priming may reflect a mechanism of pruning irrelevantconnections between or within regions and tuning the representationsof those neurons that still remain in the responsive pool3,5,13. Whenconsidered within the framework of a semantic classification task, thesechanges may serve to facilitate more fluent access to pertinent objectinformation. Such an account is consistent with recently demonstratedcorrelations between the magnitude of neural priming and reductionsin response latencies during semantic classification of objects10,30.

Here we show that when initial experience with an object isaccompanied by cortical disruption of LIFG activity, neural regionsinvolved in the retrieval, selection or representation of object

knowledge must be re-engaged at full capacity the next time thatobject is encountered, and the behavioral advantage normally asso-ciated with repetition is not realized. In line with this idea, the patternof activity in a region of left posterior temporal cortex following TMSwas similar to that of the LIFG. Activity in these two regions is oftensimilar in conceptual priming tasks, regardless of the sensory modalityused9,31–33. One speculation is that posterior temporal cortex activity ismediated via top-down modulation from the frontal cortex, andbehavioral priming is supported via the interaction of these two regionsduring semantic search and retrieval.

Although we have reported clear differences in neural and behavioralpriming following left-frontal but not control-site TMS, it should benoted that TMS to left-inferior frontal regions can produce greaterlevels of discomfort than does TMS to motor cortex. This leaves openthe possibility that differences in discomfort between the two stimula-tion sites influenced the reported effects and highlights one of the manychallenges researchers face when selecting appropriate control condi-tions in TMS studies. Several factors influenced our selection of motorcortex as the control condition. An ideal control site should (i) showcomparable levels of activation for both trial types of interest (in thiscase novel and repeated objects), (ii) be homologous to the TMS site ofinterest (e.g., right inferior frontal gyrus, RIFG) and (iii) be accessible toTMS (on the midline or lateral surface of the cortex).

The RIFG would normally fulfill these criteria; presumably, theuse of a right-frontal control site would also control for the effects ofdiscomfort and distraction. Several brain imaging studies of priming,however, have shown repetition-related reductions in the right-hemisphere homolog of the LIFG10,34–37. Indeed, repetition-relatedreductions were observed at a more lenient threshold (P o 0.01) inRIFG in the present study during Session 1. For this reason, indesigning our study, we explicitly avoided the RIFG as a control site.Instead, we selected the left motor cortex as the control site because thisregion is active (and hence easy to identify by fMRI) during objectclassifications that require a button press, but it does not showrepetition-related reductions in imaging studies of priming. Althoughthe use of a motor cortex control site does not necessarily control forTMS-induced discomfort levels associated with the LIFG site, webelieve that the behavioral data acquired during TMS administration(Session 2) minimize this concern considerably, as response latenciesduring left-frontal and control-site TMS did not differ.

The observation of spared neural priming effects in the middle andinferior occipital gyri following left-frontal TMS indicates that con-ceptual and perceptual components of priming can operate indepen-dently. Indeed, patient studies reveal dissociable impairments inconceptual and perceptual priming following damage to frontal-temporal38 and occipital39 cortices, respectively. Our results provide acontext in which to consider these findings in the intact brain byshowing that neural priming effects in the visual cortex are not linkedto upstream effects in the frontal cortex. Although neural priming insensory regions was unaffected by left-frontal or control-site stimula-tion, behavioral priming in the intact brain may be realized through theaggregate contributions of neural priming in frontal and sensorycortices. That is, neural priming in sensory regions may contribute tobehavioral priming during semantic classification tasks, but it is notsufficient to produce behavioral priming in the absence of neuralpriming in frontal cortex.

METHODSSubjects. Twelve right-handed subjects between the ages of 22 and 30 were

recruited from the Dartmouth community. All subjects were native speakers of

English, were strongly right-handed as measured by the Edinburgh handedness

5

0

–5

–10

–15

–20

–25

–30Frontal Control

TMS location

Beh

avio

ral p

rimin

g (m

s)

Figure 5 Behavioral priming after left-frontal and control-site TMS. Graph

depicts the behavioral facilitation for repeated objects (measured as the

difference in response latencies between repeated and novel objects). Gray

bar denotes a significant behavioral priming effect. Error bars indicate s.e.m.

Left-frontal TMS significantly reduced the behavioral facilitation afforded to

repeated objects.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

inventory40 and gave informed consent in accordance with the guidelines set by

the Committee for the Protection of Human Subjects at Dartmouth College. Of

the 12 subjects, one subject failed to show a behavioral measure of priming in

the initial fMRI session and therefore did not participate in the TMS session.

Results reported here reflect data analyzed from the remaining 11 participants

(5 male; mean age 26).

Study procedure. The study was carried out in three sessions. In Session 1,

subjects were scanned using event-related fMRI while they made semantic

classifications (living/nonliving) of novel and repeated line-drawn colored

objects (Supplementary Methods). Functional data were acquired in two runs.

Repeated objects (presented for the seventh, eighth and ninth times) and novel

objects were presented individually in the center of the screen for 500 ms, at a

rate of one every 2,000 ms, and subjects were instructed to make their decisions

as quickly as possible (by button press), without sacrificing accuracy. Func-

tional data from this fMRI session were analyzed to identify regions showing

neural priming effects by contrasting classifications of novel objects to

classifications of repeated objects.

In Session 2, subjects underwent TMS while performing the object classi-

fication task on a new item set. Regions of the left-inferior frontal gyrus and left

motor cortex were identified for TMS on a subject-by-subject basis. Each

subject’s high-resolution anatomical image, overlaid by his or her functional

data from Session 1, was displayed as a three-dimensional representation. The

focus of frontal TMS was defined functionally by locating subject-specific

regions that demonstrated repetition-related reductions (novel 4 repeated)

within the pars triangularis or pars opercularis of the left hemisphere

(Brodmann’s areas 44/45/47). The hand area of the primary motor cortex

was used as a TMS control site to ensure that the effects of TMS were not due

to global cortical disruption. The control site (left motor cortex) was identified

from functional activation maps in Session 1 by comparing all object classifica-

tion judgments to the baseline control task (visually fixating a cross-hair). The

TMS coil position was then adjusted within the motor activation area until a

visible motor-evoked potential (MEP) was elicited in the right hand muscles.

TMS intensity was set at 110% MEP.

For each TMS site, a set of 30 objects was presented. Objects were presented

for 500 ms, with a 4,500-ms intertrial interval. Each object was presented twice,

and TMS was administered to the same site for each presentation of the object.

Each stimulation consisted of a 10-Hz train lasting for 500 ms. TMS onset was

catered to each subject’s individual response times. Median response times were

calculated from each subject’s performance during fMRI (median response

times ranged from 506 to 810 ms). To isolate the conceptual components of

priming, TMS stimulation was time-locked to occur 250–310 ms before a

subject’s median response time. The mean stimulation onset across the group

was 334 ms after stimulus onset (range 254–500 ms).

Subjects underwent two runs of TMS, one in which objects were accom-

panied by left-frontal TMS and one in which objects were accompanied by

control-site TMS. Stimulation and timing protocols were identical for both left-

frontal and control-site TMS.

Session 3 immediately followed TMS. Subjects were rescanned using

event-related fMRI. The average time between TMS and this final fMRI scan

was 15 min. Functional data were acquired in three runs. Once again, subjects

made living/nonliving judgments about colored, line-drawn objects. Subjects

were presented with novel objects, repeated objects that were previously

presented with left-frontal TMS, and repeated objects that were previously

presented with control-site TMS. Trial types were pseudorandomly intermixed

with trials of fixation, such that each trial type followed every other trial type

equally often.

fMRI data analysis. Functional data were analyzed using the general linear

model for event-related designs in SPM99 (Wellcome Department of Cognitive

Neurology; Supplementary Methods). Data from Session 3 were analyzed

using hypothesis-driven ROI analyses. Frontal and visual ROIs were identified

in an unbiased manner based on the group statistical map comparing novel to

repeated objects in Session 1 (threshold, P ¼ 0.001, uncorrected). To calculate

signal intensities for each of these regions, spherical ROIs of 8-mm radius were

created. For each subject, signal intensities from each ROI were calculated

separately for each condition in Session 3, and then examined statistically. This

ROI method was done in normalized MNI atlas space to permit a random

effects analysis that would generalize across the human population.


ACKNOWLEDGMENTSWe thank R. Henson and A. Martin for their helpful comments on an earlierversion of this manuscript, and R. Magge and T. Laroche for their technicalassistance. This work was supported by a US National Institutes of Healthgrant (MH64667) to W.M.K. and the Dartmouth Brain Imaging Center.G.S.W. is a graduate fellow of the Natural Sciences and EngineeringResearch Council of Canada.




1. Schacter, D.L., Chiu, C.Y. & Ochsner, K.N. Implicit memory: a selective review. Annu.Rev. Neurosci. 16, 159–182 (1993).

2. Tulving, E. & Schacter, D.L. Priming and human memory systems. Science 247,301–306 (1990).

3. Desimone, R. Neural mechanisms for visual memory and their role in attention. Proc.Natl. Acad. Sci. USA 93, 13494–13499 (1996).

4. Miller, E.K., Li, L. & Desimone, R. A neural mechanism for working and recognitionmemory in inferior temporal cortex. Science 254, 1377–1379 (1991).

5. Rainer, G. & Miller, E.K. Effects of visual experience on the representation of objects inthe prefrontal cortex. Neuron 27, 179–189 (2000).

6. Buckner, R.L. et al. Functional-anatomic correlates of object priming in humans revealedby rapid presentation event-related fMRI. Neuron 20, 285–296 (1998).

7. Buckner, R.L. et al. Functional anatomical studies of explicit and implicit memoryretrieval tasks. J. Neurosci. 15, 12–29 (1995).

8. Demb, J.B. et al. Semantic encoding and retrieval in the left inferior prefrontal cortex:a functional MRI study of task difficulty and process specificity. J. Neurosci. 15,5870–5878 (1995).

9. Raichle, M.E. et al. Practice-related changes in human brain functional anatomy duringnonmotor learning. Cereb. Cortex 4, 8–26 (1994).

10. Maccotta, L. & Buckner, R.L. Evidence for neural effects of repetition thatdirectly correlate with behavioral priming. J. Cogn. Neurosci. 16, 1625–1632(2004).

11. Henson, R.N. Neuroimaging studies of priming. Prog. Neurobiol. 70, 53–81 (2003).12. Schacter, D.L. & Buckner, R.L. Priming and the brain. Neuron 20, 185–195

(1998).13. Wiggs, C.L. & Martin, A. Properties and mechanisms of perceptual priming. Curr. Opin.

Neurobiol. 8, 227–233 (1998).14. Henson, R.N. & Rugg, M.D. Neural response suppression, haemodynamic repetition

effects, and behavioural priming. Neuropsychologia 41, 263–270 (2003).15. Schacter, D.L., Dobbins, I.G. & Schnyer, D.M. Specificity of priming: a cognitive

neuroscience perspective. Nat. Rev. Neurosci. 5, 853–862 (2004).16. Henson, R., Shallice, T. & Dolan, R. Neuroimaging evidence for dissociable forms of

repetition priming. Science 287, 1269–1272 (2000).17. van Turennout, M., Ellmore, T. & Martin, A. Long-lasting cortical plasticity in the object

naming system. Nat. Neurosci. 3, 1329–1334 (2000).18. Wagner, A.D., Desmond, J.E., Demb, J.B., Glover, G.H. & Gabrieli, J.D.E. Semantic

repetition priming for verbal and pictorial knowledge: A functional MRI study of leftinferior prefrontal cortex. J. Cogn. Neurosci. 9, 714–726 (1997).

19. Dobbins, I.G., Schnyer, D.M., Verfaellie, M. & Schacter, D.L. Cortical activity reductionsduring repetition priming can result from rapid response learning. Nature 428,316–319 (2004).

20. Friston, K.J., Frith, C.D., Passingham, R.E., Liddle, P.F. & Frackowiak, R.S. Motorpractice and neurophysiological adaptation in the cerebellum: a positron tomographystudy. Proc. Biol. Sci. 248, 223–228 (1992).

21. Grafton, S.T., Hazeltine, E. & Ivry, R. Functional mapping of sequence learning in normalhumans. J. Cogn. Neurosci. 7, 497–510 (1995).

22. Grafton, S.T., Hazeltine, E. & Ivry, R.B. Abstract and effector-specific representations ofmotor sequences identified with PET. J. Neurosci. 18, 9420–9428 (1998).

23. Grafton, S.T. et al. Functional anatomy of human procedural learning determined withregional cerebral blood flow and PET. J. Neurosci. 12, 2542–2548 (1992).

24. Hazeltine, E., Grafton, S.T. & Ivry, R. Attention and stimulus characteristicsdetermine the locus of motor-sequence encoding. A PET study. Brain 120, 123–140(1997).

25. van Mier, H., Tempel, L.W., Perlmutter, J.S., Raichle, M.E. & Petersen, S.E. Changes inbrain activity during motor learning measured with PET: effects of hand of performanceand practice. J. Neurophysiol. 80, 2177–2199 (1998).

26. Ward, N.S., Brown, M.M., Thompson, A.J. & Frackowiak, R.S. Neural correlates ofmotor recovery after stroke: a longitudinal fMRI study. Brain 126, 2476–2496(2003).

27. Karni, A. et al. Functional MRI evidence for adult motor cortex plasticity during motorskill learning. Nature 377, 155–158 (1995).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

28. Petersen, S.E., van Mier, H., Fiez, J.A. & Raichle, M.E. The effects of practice on the func-tional anatomy of task performance. Proc. Natl. Acad. Sci. USA 95, 853–860 (1998).

29. Shadmehr, R. & Holcomb, H.H. Neural correlates of motor memory consolidation.Science 277, 821–825 (1997).

30. Maccotta, L. & Buckner, R.L. Evidence for neural effects of repetition that directlycorrelate with behavioral priming. J. Cogn. Neurosci. 16, 1625–1632 (2004).

31. Buckner, R.L., Koutstaal, W., Schacter, D.L. & Rosen, B.R. Functional MRI evidence fora role of frontal and inferior temporal cortex in amodal components of priming. Brain123, 620–640 (2000).

32. Dale, A.M. et al. Dynamic statistical parametric mapping: combining fMRI and MEG forhigh-resolution imaging of cortical activity. Neuron 26, 55–67 (2000).

33. Donaldson, D.I., Petersen, S.E. & Buckner, R.L. Dissociating memory retrieval processesusing fMRI: evidence that priming does not support recognition memory. Neuron 31,1047–1059 (2001).

34. Bergerbest, D., Ghahremani, D.G. & Gabrieli, J.D.E. Neural correlates of auditoryrepetition priming: Reduced fMRI activation in the auditory cortex. J. Cogn. Neurosci.16, 966–977 (2004).

35. Donaldson, D.I., Petersen, S.E. & Buckner, R.L. Dissociating memory retrieval processesusing fMRI: Evidence that priming does not support recognition memory. Neuron 31,1047–1059 (2001).

36. Koutstaal, W. et al. Perceptual specificity in visual object priming: functional magneticresonance imaging evidence for a laterality difference in fusiform cortex. Neuropsycho-logia 39, 184–199 (2001).

37. Wagner, A.D., Desmond, J.E., Demb, J.B., Glover, G.H. & Gabrieli, J.D.E. Semanticrepetition priming for verbal and pictorial knowledge: A functional MRI study of leftinferior prefrontal cortex. J. Cogn. Neurosci. 9, 714–726 (1997).

38. Keane, M.M., Gabrieli, J.D., Fennema, A.C., Growdon, J.H. & Corkin, S. Evidence for adissociation between perceptual and conceptual priming in Alzheimer’s disease. Behav.Neurosci. 105, 326–342 (1991).

39. Gabrieli, J.D.E., Fleischman, D.A., Keane, M.M.I., R.S. & Morell, F. Double dissociationbetween memory systems underlying explicit and implicit memory in the human brain.Psychol. Sci. 6, 76–82 (1995).

40. Raczkowski, D., Kalat, J.W. & Nebes, R. Reliability and validity of some handednessquestionnaire items. Neuropsychologia 12, 43–47 (1974).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Opponent appetitive-aversive neural processes underliepredictive learning of pain relief

Ben Seymour1, John P O’Doherty1,2, Martin Koltzenburg3, Katja Wiech1, Richard Frackowiak1,4, Karl Friston1 &Raymond Dolan1

Termination of a painful or unpleasant event can be rewarding. However, whether the brain treats relief in a similar way as it

treats natural reward is unclear, and the neural processes that underlie its representation as a motivational goal remain poorly

understood. We used fMRI (functional magnetic resonance imaging) to investigate how humans learn to generate expectations of

pain relief. Using a pavlovian conditioning procedure, we show that subjects experiencing prolonged experimentally induced pain

can be conditioned to predict pain relief. This proceeds in a manner consistent with contemporary reward-learning theory (average

reward/loss reinforcement learning), reflected by neural activity in the amygdala and midbrain. Furthermore, these reward-like

learning signals are mirrored by opposite aversion-like signals in lateral orbitofrontal cortex and anterior cingulate cortex. This dual

coding has parallels to ‘opponent process’ theories in psychology and promotes a formal account of prediction and expectation

during pain.

Self-preservation and evolution ordain that animals act optimally ornear-optimally to minimize harm. One of the principal mechanismsfor detecting harm is the pain system, and early prediction is essentialto direct appropriate pre-emptive behavior. However, any simplecorrespondence between predicted sensory input and behavioral out-put is challenged by considering the nature of relief: for example, mildpain will be rewarding if it directly follows severe pain. This illustrates acritical issue in our understanding of pain relief as an affective andmotivational state1–3 and poses a broader question in emotion research:how do the neural processes that underlie motivation adapt to thecontext provided by the ongoing affective state?

According to psychological theories4–7, tonic aversive states recruitreward processes to help direct behavior toward homeostatic equili-brium (which becomes the motivational goal). This may offer insightinto why relief is often pleasurable: for example, the experience ofcooling oneself in a swimming pool on a hot day. Indeed, the euphoriaof relief has been used to help explain a number of seeminglyparadoxical behaviors, from sky diving to sauna bathing8, in whichrelief is thought to become the dominant motivational drive. Despitesupportive psychological evidence9–12, direct observations of neuralactivity consistent with such appetitive processes are lacking.

Conceptually related issues arise in diverse areas such as engineer-ing, economics and computer science and offer potential insightinto the underlying neural processes involved in relief in animals.Notably, computational reinforcement learning models have provedparticularly useful in formalizing how the brain learns to predictrewards and punishments13–19. These models learn to make

predictions by assessing previous contingencies between environ-mental cues and motivationally salient outcomes. In theory, thesemodels can be extended to deal with tonic reinforcement andrelief, by computing predictions relative to an average rate of reinforce-ment, rather than according to absolute values20,21. However, the extentto which average reward/loss reinforcement learning strategies areimplemented in the brain is still unclear. With respect to pain, thismay have added importance, as motivational predictions (of pain orrelief) are thought to exert substantial influence on the subsequentperception of pain22,23. Understanding the neural mechanisms bywhich predictions are learned is therefore key to our understandingof how the brain intrinsically modulates pain in physiological andclinical situations.

We used fMRI to investigate the pattern of brain responses innineteen healthy subjects as they learned to predict the occurrence ofphasic relief from or exacerbation of tonic pain (see Methods). Weemployed a first-order pavlovian conditioning procedure with a partial(50%) reinforcement schedule (Fig. 1a). Tonic pain was induced usingthe capsaicin-heat model. Capsaicin is the pain-inducing component ofchili pepper; it induces sensitization to heat by activation of tempera-ture-dependent TRPV1 ion channels expressed on peripheral nocicep-tive neurons. This temperature sensitivity allowed us to deliverconstant but easily modifiable levels of pain for long durations, adaptedfor each individual subject, at temperatures which do not cause skindamage. This provides a unique experimental tool to study pain, as itspecifically permits investigation of the neural processes underlying theoffset of pain: that is, relief. The model has the further advantage that it


1Wellcome Department of Imaging Neuroscience, 12 Queen Square, London WC1N 3BG, UK. 2Division of the Humanities and Social Sciences 228-77, California Instituteof Technology, Pasadena, California 91125, USA. 3Institute of Child Health, University College London, 30 Guildford Street, London WC1N 1EH, UK. 4NeuroimagingLaboratory, Fondazione Santa Lucia, Rome 00179, Italy. Correspondence should be addressed to B.S. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

induces the characteristic molecular and cellular changes that mimicphysiological injury, and so presents a biologically realistic model ofrelief in natural and clinical environments.

We applied capsaicin topically to an area (12.5 cm2) of skin on theleft leg, which caused a localized area of burning pain (which feelssimilar to sunburn), and manipulated the intensity of this pain with anoverlying temperature thermode that matched the capsaicin-treatedarea. Temperature was adjusted for individual subjects to aim forevoking an average baseline magnitude of pain rated as 6 on a 0–10categorical scale. Phasic decreases in the baseline temperature to 20 1Ccaused complete relief of pain, and temperature increases causedexacerbation. We used visual cues (which were abstract colored images)as pavlovian conditioned predictors of these changes. Thus, in the fMRIscanner, subjects learned that certain images tended to predict immi-nent relief or exacerbation of pain.

We used a computational reinforcement learning (temporal differ-ence) model to identify neural activity consistent with reward-likeprocessing. The characteristic teaching signal of these models is the

prediction error, which is used to direct acqui-sition and refinement of predictions relatingto individual cues. The prediction errorrecords any change in expected affective out-come, and it thus occurs whenever predictionsare generated, updated or refuted. By treatingrelief of pain as reward, and exacerbation asnegative reward, we sought to identify activitythat correlated with this prediction error sig-nal. We calculated the value of the predictionerror for each subject according to thesequence of stimuli they received in order toprovide a statistical predictor of fMRI data(as has been done previously17,18,24). The useof a partial (probabilistic) reinforcement strat-egy, in which the cues are only 50% predictiveof their outcomes, ensures constant learningand updating of predictions and generatesboth positive and negative prediction errorsthroughout the course of the experiment(Fig. 1b,c). Thus, inference is based onidentification of this dynamic and highlycharacteristic signal.

In support of the model, our data show thatbrain activity (that is, blood oxygen level–dependent, or BOLD, activity) in the amyg-dala and midbrain correlates with the rewardprediction error signal predicted by averagereward temporal difference learning. In addi-

tion, we show an opponent, aversive representation of the predictionerror in lateral orbitofrontal and genual anterior cingulate cortex.Furthermore, these two signals appear to be coexpressed in theventral striatum.

RESULTS

Behavioral and autonomic results

Subjects rated the baseline thermal stimulation as painful and thedecreases and increases in temperature as pleasant or more painful,respectively (Fig. 2a). In addition, pleasantness and pain ratings weresignificantly greater than equivalent temperature changes on adjacentskin not treated with capsaicin (P o 0.05, all pair-wise comparisons;see Methods).

In a behavioral version of the task outside of the fMRI scanner,we demonstrated conditioning to the relief and exacerbations of pain byengaging the subjects in a supplementary cue-preference task, after thelearning task. In this, subjects (n¼ 14) made a forced choice preferencejudgement of pairs of cues, presented side by side. This demonstrated a

Time4 s

Time4 s

–10

1

–10

1

Aversive prediction error:

Appetitive prediction error:

Low (relief)

Pain Medium

High

Cue

Trial type:(Random order)

A A B B C

Reinforced Unreinforced Reinforced Unreinforced

Relief Pain

Controlcue

a

b

c

Figure 1 Experimental design and computational model. (a) Experimental design. There were five

trial types: cue A was followed by a temperature/pain decrease on 50% of occasions (reinforced and

unreinforced relief cue), cue B was followed by a temperature/pain increase on 50% of occasions

(reinforced and unreinforced pain cue) and cue C was followed by no change in temperature/pain (control

cue). (b) Appetitive computational model: predicted neuronal response. Schematic showing the mean

representation of the temporal difference prediction error according to the different cue types, where

relief is represented as reward. (c) Aversive computational model: predicted neuronal response.

Schematic showing the aversive temporal difference prediction error, which treats pain exacerbation as

punishment. b and c represent the average predicted neuronal response; the corresponding predicted

BOLD response is shown in Figs. 3c and 4c, respectively, following convolution with a canonical

hemodynamic response function.

ReliefBaselineExacerbation–8

–6

–4

–2

0

2

4

6

8

10

Pai

nful

Ple

asan

t

Sub

ject

ive

ratin

gs

20

18

16

14

12

10

8

6

4

2

0Exacerbation Control Relief

Cue type

Pre

fere

nce

scor

e

a bCapsaicin-treated skinUntreated skin

Figure 2 Behavioral measures. (a) Pain ratings.

Pain and pleasantness ratings for the baseline

level of thermal stimulation, and the phasicincreases and decreases in temperature. Scores

are on a 0–10 magnitude rating, with error bars

representing the s.e.m. The graph shows results

for the capsaicin-treated skin and an adjacent

area of unaffected skin. (b) Preference scores.

After the learning experiment, subjects made

forced choices between randomized pairs of

cues. The scores are out of a maximum of

20 pairings for each cue (with higher scores

indicating more preferred).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

significant preference ordering, with the relief cue preferred to theneutral cue (P o 0.05, Wilcoxon sign rank test), which was, in turn,preferred to the exacerbation cue (P o 0.01, Wilcoxon sign rank test;Fig. 2b). On post-experimental debriefing (see Methods), only four outof the 14 subjects could report any contingent relationship between thecues and the outcomes.

During the fMRI version of the task, we used physiological measuresto assess the acquisition of cue expectations. Heart rate changesinduced by the cues correlated with the magnitude of expectations(that is, cue-specific temporal difference values) both of pain relief(P o 0.01) and pain exacerbation (P o 0.01), calculated from themodel (see Methods). This supports the hypothesis that cue expecta-tions are acquired in a manner consistent with the (temporal differ-ence) learning model, albeit in a valence-insensitive manner. That is, weobserved increased heart rate with higher valued cues, whether positiveor negative, consistent with a learned arousal-like response associatedwith the expectations.

fMRI results

We used the model to identify a representation of the appetitiveprediction error in the brain (Fig. 1b, appetitive model). Activity inleft amygdala and left midbrain (in a region consistent with thesubstantia nigra) correlated with this signal (Fig. 3a,b). Time-courseanalysis illustrates the average pattern of response associated with thedifferent trial types in the amygdala, illustrating a strong correspon-dence with the predictions of the model (Fig. 3c). These data supportthe hypothesis that relief learning involves a reward-like learning signal.

Recent evidence indicates that temporal difference models alsoprovide an accurate description of aversive learning, suggesting theexistence of a separate reinforcement learning mechanism encodingaversive events18. We therefore sought to identify whether an aversive

representation of the prediction error was expressed, in whichexacerbation of pain was treated as positive punishment, and relief asnegative punishment (Fig. 1c, aversive model). Activity in bilaterallateral orbitofrontal cortex and genual anterior cingulate cortex corre-lated with this signal (Fig. 4a,b). The time-course of this activity(Fig. 4c) illustrates the opposite pattern of response to the appetitiveprediction error. These data indicate the existence of an aversivereinforcement signal, distinct from the reward-like signal.

Psychological studies of appetitive-aversive interactions predict thatopposing, learning-related activities should converge in some areas10.This might occur in areas such as the ventral striatum (and insulacortex), where predictive activity has been observed in both reward andpain learning tasks, albeit in separate studies17,18,25–28. This raises aquestion about how coexpressed aversive and appetitive predictionerrors would be represented by the BOLD signal, particularly if theyinteract. We therefore created a new statistical model that included tworegressors, modelling prediction error for relief and exacerbationseparately. This model revealed coexpression in the ventral putamen,anterior insula and rostral anterior cingulate cortex (Fig. 5a–c). Theresponses in these regions showed an appetitive prediction error for therelief-related cue, and an aversive prediction error for the exacerbation-related cue (Fig. 5d). This pattern of activity is notable, as it cannotresult simply from the linear superposition of appetitive and aversivesignals, but implies either an interaction between prediction errorand cue-valence, or the expression of a single valence-independentprediction error.

DISCUSSION

Drawing on theoretical considerations provided by computationalreinforcement learning11, our data provide evidence in support of anopponent motivational model of tonic pain. We observed two distinct

151050151050

151050151050

–0.5

0.5

0

–0.5

0.5

0

–0.5

0.5

0

–0.5

0.5

0

Sig

nal c

hang

e (%

)

Time (s) Time (s)

UnreinforcedReinforcedRelief

Pain

RR

a

c

b

Figure 3 Appetitive temporal difference prediction error. (a,b) Statistical

parametric maps (P o 0.001) showing (a) left substantia nigra (axial plane)

and (b) left amygdala (coronal plane). (c) Time course of inferred mean

neuronal activity for the four principal trial types in left amygdala. Black line

represents the data (error bars represent 1 s.e.m.), and thin gray line is the

model appetitive temporal difference prediction error (from Fig. 1b) after

convolution with a canonical hemodynamic response function.

151050–0.5

0.5

0

Time (s)

151050–0.5

0.5

0

Time (s)

151050–0.5

0.5

0

151050–0.5

0.5

0

Sig

nal c

hang

e (%

)

Pain

UnreinforcedReinforced

Relief

R

a

c

b

Figure 4 Aversive temporal difference prediction error. Statistical parametric

maps (P o 0.001) showing (a) lateral orbitofrontal cortex (axial plane) and

(b) genual anterior cingulate cortex, highlighted (sagittal plane). (c) Time

course of inferred mean neuronal activity for the four principal trial types in

left orbitofrontal cortex. Black line shows data (error bars represent 1 s.e.m.),

and thin black line is the model aversive temporal difference prediction error

(Fig. 1c) after convolution with a canonical hemodynamic response function.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

patterns of neural activity, distinguishable by their expression inseparate brain areas, that correlated with the prediction error signalsof an opponent temporal difference model. This extends our under-standing of human predictive learning beyond the occurrence of phasicevents arising from a neutral baseline. Thus, during tonic pain, aversiveand appetitive systems seem to be simultaneously involved to encodeappropriate goal-directed predictions across the spectrum of positiveand negative outcomes. Our observations suggest a formal frameworkfor understanding the homeostatic and motivational processes engagedby pain and may offer a paradigmatic account of motivation duringtonic affective states.

The use of the temporal difference algorithm to represent positiveand negative deviations of pain intensity from a tonic backgroundlevel approximates the class of reinforcement learning model termedaverage-reward models20,21,29. Accordingly, predictions are judgedrelative to the average level of pain, rather than according to anabsolute measure. This comparative treatment of motivationally salientpredictions is consistent with both neurobiological and economicaccounts of homeostatic motivation, which rely critically on changein affective state2,30,31.

Implicit in any such model is a representation of the average rate ofreinforcement, although the short time window of fMRI precludesinvestigation of this directly. From an implementational perspective,one argument for opponency relates to consideration of how a long-run average affective state might be represented. Given our demonstra-tion that positive and negative prediction errors are both encoded byone system and are fully mirrored by opposite signals in an opponent

system, the requirement for one system to fully represent both the toniclevels of reinforcement (that is, by sustained elevated activity) withpositive and negative phasic predictions simply superimposed, wouldseem to be obviated. If this is the case, the tonic level of pain would befree to have a distinct representation, a signal that has been suggested tobe conveyed by tonic dopamine release11.

Mirror opponency has many similarities to the appetitive-aversivereciprocity characteristic of early psychological ‘opponent process’theories4–7. In their various forms, these theories grew out of arequirement both to explain the adaptive changes that occur duringand after tonic reinforcement, and to understand the interactionsbetween appetitive and aversive processes that arise in certain specificlearning procedures such as conditioned inhibition and trans-reinforcerblocking. Notably, recent electrophysiological recordings of neuronalactivity in mice directly indicate the involvement of opponent processesin (context-related) conditioned inhibition, specifically implicating theventral striatum and amygdala32. Thus it seems possible (and fullyconsistent with a computational account) that, at least in the ventralstriatum, a ‘safety signal’ that predicts the absence of future pain mightshare the same neural substrate as the relief-prediction error seen here.However, we show an appetitive representation in the amygdala, ratherthan an opponent aversive representation (which we observe in lateralorbitofrontal and genual anterior cingulate cortex). This points to theexpression of multiple learning-related neural signals in the amygdala,consistent with the complex, integrative role of this structure (and thevarious nuclei within) in associative learning and pain33,34.

The finding that lateral orbitofrontal cortex demonstrates an aversiveprediction error signal is consistent with previous reports of a role forthis region in aversive learning35. In particular, this area has been shownto be involved in evaluation of aversive stimuli in the context of differentmotivational states36 as well as in short-time-scale pain predictionrelative to a changing (learned) baseline rate of phasic pain37. Takenwith the present results, this suggests that learning of aversive valuepredictions in this region may be mediated by an aversion–specificprediction error signal, particularly in circumstances that requireadaptive representations following changing motivational state or con-text. However, it should also be noted that lateral orbitofrontal cortexmay not be exclusively involved in aversive processing, as reward-relatedresponses have also been reported in this region in some circumstances.

In relation to pain, other cortical areas, specifically insula and anteriorcingulate cortex, have clear motivational roles and have previously beenimplicated in the processing of relief-related information3. For example,recent neuroimaging studies investigating the expectation and receipt ofplacebo analgesia implicate these areas in endogenously mediatedanalgesia38,39. Our findings provide further support that these areashave a key role in homeostatic functions relating to pain2.

The BOLD signal is thought to correspond to changes (increases ordecreases) in synaptic activity, and thus the activity we describe mayreflect specific afferent neuromodulatory influences that originateelsewhere40,41. Substantial evidence indicates that mesolimbic dopa-mine neurons both encode reward-related prediction error16,19 andhave a key role in analgesia42, suggesting that dopamine could conveyan appetitive relief-related prediction error. This draws attention toactivity in the ventral striatum, a region that receives strong mesolimbicdopaminergic projections. Comparison with previous data in this areahighlights the observation that cues signaling lower-than-predictedpain cause deactivation in the context of a neutral baseline, as opposedto activation in the context of a tonic pain baseline18,26. This implicatesadaptive changes occurring during tonic pain, influencing ventralstriatal activity and consistent with the representation of an appetitivesignal for relief-related cues. However, taken alone, it is possible that

151050–0.5

0.5

0

Time (s)

151050–0.5

0.5

0

Time (s)

151050–0.5

0.5

0

151050–0.5

0.5

0

Sig

nal c

hang

e (%

)

Pain

UnreinforcedReinforced

Relief

R R

a

d

b c

Figure 5 Appetitive relief-related plus aversive exacerbation-related

prediction error. Statistical parametric maps showing activity that correlateswith the appetitive prediction error for the relief cue (P o 0.001), masked

with the aversive prediction error for the exacerbation cue (P o 0.001).

(a) Bilateral ventral putamen. (b) Bilateral ventral putamen and right anterior

insula. (c) Rostral anterior cingulate cortex. (d) Time course of inferred mean

neuronal activity for the four principal trial types in left ventral putamen.

Thick black line shows the data (error bars represent 1 s.e.m.), and the

thin gray and black lines are the model appetitive and aversive temporal

difference prediction error, respectively (from Fig. 1b,c) after convolution

with a canonical hemodynamic response function.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

this ventral striatal activity is modulated by a single prediction-errorsignal for both relief and exacerbation cues43,44, although recentelectrophysiological evidence demonstrating suppression of midbraindopaminergic neurons to aversive stimuli would seem to require adistinct aversive opponent45. Either way, this signal must interact withvalence-specific information by some additional mechanism, possiblythrough the involvement of different intrinsic sub-populations ofappetitive and aversive neurons within the ventral striatum46.

That pain relief and reward might share a common neural substrateis also suggested by the fact that many drugs that have rewarding effectshave analgesic properties. Aside from dopamine, there are manyneurotransmitters with clear combined roles in appetitive and aversivemotivation, for example opioid peptides, serotonin, substance P andglutamate3,47,48. Of particular interest are serotonin-releasing neuronsprojecting from the dorsal raphe nucleus to the ventral striatum,which have emerged as a plausible candidate to mediate an aversiveprediction error11.

In addition to a role in pavlovian motivation, it is also clear that painand relief-related expectations exert a strong influence on the actualsubsequent experience of pain, in that perception (of intensity) isweighted by the prior expectancies acquired through conditioning.How predictive motivational values influence perceptual inferencessuch as pain intensity is not yet clear, although probabilistic perceptualmodels that incorporate economic cost functions, such as decisiontheory, may offer insight at a theoretical level49. From an implementa-tional perspective, one putative mechanism exploits an influence of‘higher’ brain areas on ascending pain pathways via descendingmodulatory control centers. A possible target is the ‘on-’ and ‘off-’cells of the periaqueductal grey and rostral ventromedial medulla,which show opponent anticipatory pain-related activity under appar-ent higher control3. Whatever the mechanisms, these influences arethought to be clinically important both in endogenous pain modula-tion (including placebo analgesia) and in the pathogenesis of somechronic pain syndromes3,23,38,39, and we suggest that integrated psy-chological, neurophysiological and computational approaches offersome promise in furthering their understanding.

METHODSSubjects. Thirty-three healthy right handed subjects (14 in a behavioral version

of the task, and 19 in the fMRI version of the task), free of pain or medication,

gave informed consent and participated in the study, approved by the Joint

National Hospital for Neurology and Neurosurgery (University College

London, National Health Service Trust) and Institute of Neurology (University

College London) Ethics Committee. Subjects were remunerated for their

inconvenience (40 GBP).

Stimuli: capsaicin model. We applied topical 1% capsaicin (8-methyl-N-

vanillyl-6-nonenamide, 98%, Sigma, diluted in 5% ethanol-KY jelly) to the

lateral aspect of the left leg over an area of 2.5 � 5 cm, under an occlusive

dressing, and left it for 40 min, after which all subjects reported feeling

persistent (though bearable) pain, at which time the capsaicin and dressing

was removed and the skin cleaned. A thermode matching the size of the

capsaicin application area was applied with a loose tourniquet (easily removable

in case of unbearable pain) to the treated skin. Temperature was then

manipulated using an fMRI-compatible Peltier thermode (MSA thermotest,

Somedic). Phasic variations in temperature were made at a rate of 5 1C/s to the

predetermined upper and lower levels and were controlled by in-house software.

Stimuli and pre-experimental set-up. Before the experiment, required tem-

perature levels for each individual subject were set by slowly increasing the

cutaneous temperature overlying the capsaicin treatment site from 20 1C in

steps of 0.5 1C, with continual monitoring of pain ratings (on a 0–10 rating

scale) to achieve a baseline level of 6/10. Subsequently, subjects received

progressively higher phasic increases to determine a satisfactory temperature

for the pain exacerbations, to at least 8/10 (‘just tolerable’). Pain relief was

induced by phasic cooling to 20 1C, which abolished pain in all subjects.

We obtained subjective ratings of pain for the increase, baseline and

decreases in pain. We asked the subjects, ‘‘Can you give a score, on a scale of

0 to 10, as to how painful the pain is, where 0 is no pain at all, and 10 is the

worst imaginable pain?’’ We also took subjective ratings of pleasantness for the

phasic relief. We first asked the subjects, ‘‘Did you find the change in

temperature unpleasant or pleasant?’’ to check that no subjects found the

cooling as unpleasant, and then, ‘‘Can you give a score, on a scale of 0 to 10, as

to how pleasant you found it, where 0 is not at all, and 10 is highest imaginable

pleasure?’’ Phasic changes were repeated with pain and pleasantness ratings on

capsaicin-treated skin and on a distant area of non–capsaicin treated skin on

the same limb well beyond the area of secondary hyperalgesia, and repeated at

the end of the experiment. We achieved mean ratings (s.e.m. in parentheses) for

the baseline tonic pain of 5.5/10 (1.1) on capsaicin treated skin and 0.9/10 (1.5)

on untreated skin. Phasic increases were rated at 9.3/10 (0.9) for capsaicin-

treated skin and 3.3/10 (3.6) on untreated skin. Phasic decreases (relief;

measured on the pleasantness scale) were rated at 7.0/10 (2.4) and 4.6/10

(2.3) on untreated skin. All comparisons (treated versus untreated) were

significant at P o 0.01 with corresponding t-tests. After transfer into the

scanner or behavioral testing room (with the thermode attached) subjects were

in pain for approximately 40 min to 1 h by the time the experiment started. The

visual cues were abstract colored pictures.

Task. The task was a classical pavlovian delay-conditioning procedure of

temperature increases (exacerbations of pain) or decreases (relief of pain).

Visual cues were presented for 4 s, at the end of which the phasic pain

perturbation was applied for 5 s. The precise timing was determined in

psychophysical pilot testing (to accommodate thermode and C-fiber latencies).

There were three different visual cues, each presented 30 times. Cue A (relief-

related cue) was followed by decreased temperature on 15/30 occasions (50%),

cue B (pain exacerbation related cue) was followed by increased temperature on

15/30 occasions (50%), and cue C was followed by no change in temperature

on 30/30 occasions. The control condition provides additional control in our

parametric design, although it was initially included to permit a more

conventional analysis (data not shown). The five different trial types were

presented in random order.

Behavioral measures. Subjects performed a reaction-time task which consisted

of judging whether the visual cue appeared to the left or right of center on the

display monitor, as quickly as possible. The resulting reaction times were taken

as a behavioral index of conditioning. Performance on this task was not

contingent on the stimuli presented, and subjects were told before imaging that

their success or failure at quickly judging the position would not affect the

amount of pain or relief received. The task was performed with a two-button

key press using the right hand. Heart rate was recorded using a pulse oximeter

in conjunction with Spike 2 software (CED).

A behavioral version of the task was performed that was identical to

that performed in the fMRI scanner, except that it was performed in a testing

room with the subject seated in front of a computer monitor. After this task,

we performed a supplementary cue-preference task designed to investigate

whether the subjects had acquired appetitive and aversive preferences for the

cues as a result of the conditioning procedure. In this task, we presented two

cues side-by-side and asked the subject to judge which cue they preferred,

indicated by a left or right key-press. Each cue-pairing was repeated ten times

and was randomized as to which side the cue appeared on. We calculated

the preference scores by summing the total number of preference choices made

for each cue (as in an all-play-all games table, with a maximum score of 20).

Mean scores for each cue were compared across subjects using Wilcoxon sign

rank tests.

We did not attempt to formally address the issue of conscious versus non-

conscious acquisition of conditioned expectancies. However, to gain some

insight into the level of explicit expectancy learning, we asked the question,

‘‘Did you recognize any relationship between the pictures and subsequent

change in pain level?’’ at the end of the experiment (for the behavioral version

of the task only). Subjects were not told the experiment was a learning and

conditioning study beforehand but rather were simply told that it was a study


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

of pain and temperature processing. Ten of fourteen subjects were unable to

report any association between cues and outcomes.

Computational model. We used a temporal difference model to generate a

parametric regressor corresponding to the appetitive prediction error, which

was applied to the imaging data, as previously described17,18. Here, we used a

two–time point temporal difference model with a learning rate (a ¼ 0.3)

determined from behavioral results (see below). In this model, the value v of a

particular cue (referred to as a state s) is updated according to the learning rule:

v(s) v(s) + ad, where d is the prediction error. This is defined as d ¼ r – a

+ v(s)t+1 – v(s)t, where r is the return (that is, the amount of pain) and a is the

average amount of reinforcement (tonic pain) that was assumed to be constant.

We assigned relief and exacerbations of pain as returns of 1 and –1, respectively

(that is, a linear scale of pain from relief to exacerbation). This is an arbitrary

specification, given that it is difficult to precisely scale the relative oppositely

valenced utilities of relief and exacerbations of pain. Thus, the model treats

predictions relating to relief of pain on equal par with unexpected omission of

exacerbation of pain, and, similarly, it treats exacerbation-related predictions

equivalently to unexpected omissions of relief.

Data acquisition and analysis: behavioral and autonomic measures. These

were taken as measures of cue reinforcement and correlated with the temporal

difference value (that is, the cue expectancy). Reaction time data were

individually (that is, on a subject-by-subject basis) fit to a gamma cumulative

distribution function (using a maximum likelihood function), to allow analysis

across subjects, and correlated with the temporal difference value. This yielded

a best fit with a learning rate of 0.3, and a significant correlation for both the

relief-related and exacerbation-related trials, independently, and in the same

direction. That is, reaction times were shorter for both high reward values and

high aversive values. To remove any possible confounding effects of early trials,

during which reaction time data habituate substantially, we repeated this

procedure after removing the first ten trials. This yielded a correlation which

just failed to reach significance (P ¼ 0.056), across both cue types. We also

looked at sensitivity to the initial temporal difference value by setting this to the

average value of 0.5, which yielded a non-significant correlation.

The heart rate was found to be approximately normally distributed and was

normalized to permit analysis across subjects. We found significant heart rate

correlations with both relief and pain cue types (independently, as for the

reaction time). For both exacerbation and relief trial types, this yielded a best fit

with a learning rate of 0.3. Across both cue types, this remained significant

(Po 0.05, r¼ 0.19) after removal of the first ten trials and with use of different

initial temporal difference values. This is a robust correlation and is reported in

the main text. Consequently, we used a learning rate of 0.3 for the temporal

difference model used in the fMRI analysis.

fMRI. Functional brain images were acquired on a 3-T Allegra Siemens scanner.

Subjects lay in the scanner with foam head restraint pads to minimize any

movement associated with the painful stimulation. Images were realigned with

the first volume, normalized to a standard EPI template and smoothed using a

6-mm FWHM Gaussian kernel. Realignment parameters were inspected

visually to identify any potential subjects with excessive head movement; none

was found. Images were analyzed in an event-related manner using the general

linear model, with the onsets of each stimulus represented as a delta function to

provide a stimulus function. We used a parametric design, in which the

temporal difference prediction errors modulated the stimulus functions on a

stimulus-by-stimulus basis. The statistical basis of this approach has been

described previously50. Regressors were then generated by convolving the

stimulus function with a hemodynamic response function (HRF). Effects of

no interest included the onsets of visual cues, the pain relief and exacerbations

themselves and realignment parameters from the image preprocessing to

provide additional correction for residual subject motion. Linear contrasts of

appetitive prediction errors were taken to a group level (random effects)

analysis by way of a one-sample t-test, and the aversive prediction error was

taken as the inverse. MNI coordinates and statistical z-scores are found in

Table 1. This analysis determines areas which correlate to univalent appetitive

or aversive prediction error and does not identify areas in which these signals

overlap. To explore the possible representation of distinct prediction error

signals for the pain relief and exacerbation trials, we generated two independent

regressors for the prediction error occurring at each. Then, we took the

appetitive relief and aversive exacerbation components of the prediction error

to a second level analysis of variance and exclusively masked the two individual

contrasts (that is, we looked for areas of overlap of the independent appetitive-

relief and aversive-exacerbation prediction errors, both at Po 0.001; Fig. 5a–c).

Group level activations were localized according to the group-averaged

structural scan. Activations were checked on a subject-by-subject basis using

their individual normalized structural scans to ensure correct localization, as

some of the reported activations are in small nuclei (for example, substantia

nigra). We report activity in areas in which we had prior hypotheses on the

basis of previous data, though without specification of laterality. These regions

have established roles in both aversive and appetitive predictive learning, and

included ventral putamen, head of caudate, midbrain (substantia nigra),

anterior insula cortex, cerebellum, anterior cingulate cortex, amygdala, lateral

orbitofrontal cortex, medial orbitofrontal cortex, dorsal raphe and ventral

tegmental area. We report activations at a threshold of P o 0.001, with a

minimum size of five contiguous voxels. We also report brain activations

outside our areas of interest that survive whole-brain correction for multiple

comparisons (Table 1) using family-wise error correction at P o 0.05.

We performed a supplementary fixed-effects analysis on a trial basis to

determine impulse responses, as previously described18. Note that this analysis

refers to the average impulse response across each trial throughout the

experiment and does not embody the time-dependent nature of learning

incorporated within the main parametric analysis.

ACKNOWLEDGMENTSWe wish to thank P. Dayan and N. Daw for many helpful discussions andO. Josephs, B. Johanssen and C. Rickard for technical assistance. This researchwas funded by The Wellcome Trust.


Received 15 June; accepted 2 August 2005


1. Cabanac, M. Physiological role of pleasure. Science 173, 1103–1107 (1971).2. Craig, A.D. A new view of pain as a homeostatic emotion. Trends Neurosci. 26, 303–307

(2003).

Table 1 MNI coordinates and statistical z-scores for the appetitive,

aversive and joint coexpressed appetitive-aversive temporal difference

prediction error

Region Laterality x y z z-score

Appetitive prediction error

Midbrain (substantia nigra) L –18 –12 –8 3.99

Amygdala L –20 2 –26 3.33

Aversive prediction error

Lateral orbitofrontal cortex R 40 34 –20 3.72

L –34 34 –20 3.71

Genual anterior cingulate cortex R 10 42 –6 4.24

Motor cortex R 14 0 60 5.35a

Combined appetitive-aversive prediction error

Ventral putamen R 18 8 0 4.08

22 10 –10 3.32

L –18 8 –12 3.62

Anterior insula R 30 22 6 3.87

36 2 16 4.78

L –34 12 12 4.55

Rostral anterior cingulate cortex R 2 34 20 3.61

aSignificant after whole brain correction.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

3. Fields, H. State-dependent opioid control of pain. Nat. Rev. Neurosci. 5, 565–575(2004).

4. Solomon, R.L. & Corbit, J.D. An opponent-process theory of motivation. I. Temporaldynamics of affect. Psychol. Rev. 81, 119–145 (1974).

5. Konorski, J. Integrative Activity of the Brain: an Interdisciplinary Approach (Chicago,University of Chicago Press, 1967).

6. Schull, J. A conditioned opponent theory of Pavlovian conditioning and habituation. inThe Psychology of Learning andMotivation (ed. Bower, G.) 57–90 (Academic, New York,1979).

7. Grossberg, S. Some normal and abnormal behavioral syndromes due to transmittergating of opponent processes. Biol. Psychiatry 19, 1075–1118 (1984).

8. Solomon, R.L. The opponent-process theory of acquired motivation: the costs of pleasureand the benefits of pain. Am. Psychol. 35, 691–712 (1980).

9. Solomon, R.L. Recent experiments testing an opponent-process theory of acquiredmotivation. Acta Neurobiol. Exp. (Wars.) 40, 271–289 (1980).

10. Dickenson & Dearing, MF. Appetitive-aversive interactions and inhibitory processes. inMechanisms of Learning and Motivation. (eds. Dickinson, A. & Boakes, R.A.) 203–231(Erlbaum, Hillsdale, New Jersey, 1979).

11. Daw, N.D., Kakade, S. & Dayan, P. Opponent interactions between serotonin anddopamine. Neural Netw. 15, 603–616 (2002).

12. Tanimoto, H., Heisenberg, M. & Gerber, B. Experimental psychology: event timing turnspunishment to reward. Nature 430, 983 (2004).

13. Barto, A.G. Adaptive critics and the basal ganglia. in Models of Information Processingin the Basal Ganglia (eds. Houk, J.C., Davis, J.L. & Beiser, D.G.) 215–232 (MIT Press,Cambridge, Massachusetts, 1995).

14. Sutton, R.S. & Barto, A.G. Reinforcement Learning: an Introduction. (MIT Press,Cambridge, Massachusetts, 1998).

15. Montague, P.R., Dayan, P. & Sejnowski, T.J. A framework for mesencephalic dopaminesystems based on predictive Hebbian learning. J. Neurosci. 16, 1936–1947 (1996).

16. Schultz, W., Dayan, P. & Montague, P.R. A neural substrate of prediction and reward.Science 275, 1593–1599 (1997).

17. O’Doherty, J.P., Dayan, P., Friston, K., Critchley, H. & Dolan, R.J. Temporal differencemodels and reward-related learning in the human brain. Neuron 38, 329–337 (2003).

18. Seymour, B. et al. Temporal difference models describe higher-order learning in humans.Nature 429, 664–667 (2004).

19. Dayan, P. & Balleine, B.W. Reward, motivation, and reinforcement learning. Neuron 36,285–298 (2002).

20. Schwartz, A. A reinforcement learning method for maximizing undiscounted rewards. inProceedings of the Tenth International Conference on Machine Learning. 298–305(Morgan Kaufmann, San Mateo, California, 1993).

21. Mahadevan, S. Average reward reinforcement learning: Foundations, algorithms andempirical results. Mach. Learn. 22, 1–38 (1996).

22. Fields, H.L. Pain modulation: expectation, opioid analgesia and virtual pain. Prog. BrainRes. 122, 245–253 (2000).

23. Price, D.D. Psychological Mechanisms of Pain and Analgesia (IASP, Seattle, 1999).24. Tanaka, S.C. et al. Prediction of immediate and future rewards differentially recruits

cortico-basal ganglia loops. Nat. Neurosci. 7, 887–893 (2004).25. Ploghaus, A. et al. Dissociating pain from its anticipation in the human brain. Science

284, 1979–1981 (1999).26. Jensen, J. et al. Direct activation of the ventral striatum in anticipation of aversive

stimuli. Neuron 40, 1251–1257 (2003).27. McClure, S.M., Berns, G.S. & Montague, P.R. Temporal prediction errors in a passive

learning task activate human striatum. Neuron 38, 339–346 (2003).

28. Setlow, B., Schoenbaum, G. & Gallagher, M. Neural encoding in ventral striatum duringolfactory discrimination learning. Neuron 38, 625–636 (2003).

29. Daw, N.D. & Touretzky, D.S. Long-term reward prediction in TD models of the dopaminesystem. Neural Comput. 14, 2567–2583 (2002).

30. Markowitz, H. The utility of wealth. J. Polit. Econ. 60, 151–158 (1952).31. Camerer, C., Loewenstein, G. & Prelec, D. Neuroeconomics: how neuroscience can

inform economics. J. Econ. Lit. (in the press).32. Rogan, M.T., Leon, K.S., Perez, D.L. & Kandel, E.R. Distinct neural signatures for

safety and danger in the amygdala and striatum of the mouse. Neuron 46, 309–320(2005).

33. Watkins, L.R. et al. Neurocircuitry of conditioned inhibition of analgesia: effects ofamygdala, dorsal raphe, ventral medullary, and spinal cord lesions on antianalgesia inthe rat. Behav. Neurosci. 112, 360–378 (1998).

34. Holland, P.C. & Gallagher, M. Amygdala-frontal interactions and reward expectancy.Curr. Opin. Neurobiol. 14, 148–155 (2004).

35. O’Doherty, J., Kringelbach, M.L., Rolls, E.T., Hornak, J. & Andrews, C. Abstract rewardand punishment representations in the human orbitofrontal cortex. Nat. Neurosci. 4,95–102 (2001).

36. Small, D.M., Zatorre, R.J., Dagher, A., Evans, A.C. & Jones-Gotman, M. Changesin brain activity related to eating chocolate: from pleasure to aversion. Brain 124,1720–1733 (2001).

37. Glascher, J. & Buchel, C. Formal learning theory dissociates brain regions with differenttemporal integration. Neuron 47, 295–306 (2005).

38. Petrovic, P., Kalso, E., Petersson, K.M. & Ingvar, M. Placebo and opioid analgesia–imaging a shared neuronal network. Science 295, 1737–1740 (2002).

39. Wager, T.D. et al. Placebo-induced changes in FMRI in the anticipation and experienceof pain. Science 303, 1162–1167 (2004).

40. Logothetis, N.K., Pauls, J., Augath, M., Trinath, T. & Oeltermann, A. Neurophysiologicalinvestigation of the basis of the fMRI signal. Nature 412, 150–157 (2001).

41. Stefanovic, B., Warnking, J.M. & Pike, G.B. Hemodynamic and metabolic responses toneuronal inhibition. Neuroimage 22, 771–778 (2004).

42. Altier, N. & Stewart, J. The role of dopamine in the nucleus accumbens in analgesia. LifeSci. 65, 2269–2287 (1999).

43. Horvitz, J.C. Mesolimbocortical and nigrostriatal dopamine responses to salient non-reward events. Neuroscience 96, 651–656 (2000).

44. Smith, A.J., Becker, S. & Kapur, S. A computational model of the functional role of theventral-striatal D2 receptor in the expression of previously acquired behaviors. NeuralComput. 17, 361–395 (2005).

45. Ungless, M.A., Magill, P.J. & Bolam, J.P. Uniform inhibition of dopamine neurons in theventral tegmental area by aversive stimuli. Science 303, 2040–2042 (2004).

46. Roitman, M.F., Wheeler, R.A. & Carelli, R.M. Nucleus accumbens neurons are innatelytuned for rewarding and aversive taste stimuli, encode their predictors, and are linked tomotor output. Neuron 45, 587–597 (2005).

47. Johansen, J.P. & Fields, H.L. Glutamatergic activation of anterior cingulate cortexproduces an aversive teaching signal. Nat. Neurosci. 7, 398–403 (2004).

48. Gadd, C.A., Murtra, P., De Felipe, C. & Hunt, S.P. Neurokinin-1 receptor-expressingneurons in the amygdala modulate morphine reward and anxiety behaviors in the mouse.J. Neurosci. 23, 8271–8280 (2003).

49. Dayan, P. & Abbott, L.F. Theoretical Neuroscience: Computational and MathematicalModeling of Neural Systems (MIT Press, Cambridge, Massachusetts, 2001).

50. Buchel, C., Holmes, A.P., Rees, G. & Friston, K.J. Characterizing stimulus-responsefunctions using nonlinear regressors in parametric fMRI experiments. Neuroimage 8,140–148 (1998).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Structural and functional asymmetry of lateral Heschl’sgyrus reflects pitch perception preference

Peter Schneider1, Vanessa Sluming2,3, Neil Roberts3, Michael Scherg1, Rainer Goebel4, Hans J Specht5,H Gunter Dosch5, Stefan Bleeck6, Christoph Stippich1 & Andre Rupp1

The relative pitch of harmonic complex sounds, such as instrumental sounds, may be perceived by decoding either the

fundamental pitch (f0) or the spectral pitch (fSP) of the stimuli. We classified a large cohort of 420 subjects including symphony

orchestra musicians to be either f0 or fSP listeners, depending on the dominant perceptual mode. In a subgroup of 87 subjects,

MRI (magnetic resonance imaging) and magnetoencephalography studies demonstrated a strong neural basis for both types of

pitch perception irrespective of musical aptitude. Compared with f0 listeners, fSP listeners possessed a pronounced rightward,

rather than leftward, asymmetry of gray matter volume and P50m activity within the pitch-sensitive lateral Heschl’s gyrus. Our

data link relative hemispheric lateralization with perceptual stimulus properties, whereas the absolute size of the Heschl’s gyrus

depends on musical aptitude.

Pitch perception is an essential prerequisite for understanding musicand speech intonation. Although there is common agreement that theperceived pitch of harmonic complex tones like instrumental soundsor vowels in the singing voice is closely related to the fundamentalfrequency (f0) of the sound spectrum, large individual differences inpitch and timbre perception challenge this one-to-one relationship1,2.In particular, if f0 is not physically present, pitch is perceived either asthe missing f0 or as spectral pitch (fSP), corresponding to the dominantperceptual mode (historically referred to as ‘synthetic’ versus ‘analytic’pitch perception3,4).

At the cortical level, both frequency processing and pitch perceptionhave been found to correlate with neural activity changes in theauditory cortex, related to different processing stages of hierarchicallyorganized auditory subareas5. Physical stimulus properties such asperiodicity6, temporal regularity7,8 and frequency spectrum9 areencoded in both subcortical and cortical structures of the auditoryascending pathway. In primary auditory cortex, sound frequencyis represented in mirror-symmetric tonotopic frequency maps9,10

by spatiotemporal integration11. The more complex the stimuli andthe processing tasks, ranging from pitch perception12 to melody7,timbre or tonality13 processing, the more lateral and anterior arethe main peaks of activation in Heschl’s gyrus (HG) and anteriorsupratemporal gyrus (aSTG)13–16. Furthermore, the representationof pitch as a perceptual, rather than physical, stimulus property wasfound to correlate with neural activity changes in the non-primary

auditory cortex15. Consistent with this finding, numerous functionalimaging studies point to the existence of a ‘pitch processing center’12

immediately anterolateral to primary auditory cortex within the lateralHeschl’s gyrus (lHG), subserving the processing of fixed pitch16, pitchchroma17, pitch salience15, pitch direction18, pitch sequences19 andlively pitch20.

Both relative hemispheric asymmetries21–26 and the absolute mag-nitude of the neural auditory cortex substrate27 are important inenabling understanding of how brain structure maps with the observedfunctional specialization. In particular, recent functional imagingstudies show a relative left-hemispheric specialization for rapid tem-poral processing23,25, whereas right auditory cortex shows a strongersensitivity for spectral processing23 and a slower temporal processingmode25. Motivated by these findings of asymmetry and strong mutualcorrelations between early auditory evoked activity, anatomical size andbehavioral predisposition in the anteromedial portion of HG27, wehypothesize here that f0 versus fSP perception may serve as a predictorreflecting both functional and structural aspects of the pitch-sensitiveareas in lHG. The main purpose of this study was to investigate(i) individual psychometric differences in f0 and fSP perception inrelation to musical aptitude, (ii) the neural basis for type of pitchperception by using MRI and magnetoencephalography (MEG) and(iii) the influence of relative hemispheric lateralization versus absolutemagnitude of both gray matter volume structurally and auditoryevoked activity functionally.


1Department of Neurology, University Hospital Heidelberg, INF 400, D-69120 Heidelberg, Germany. 2School of Health Sciences, Division of Medical Imaging, University ofLiverpool, Johnston Building, The Quadrangle Brownlow Hill, Liverpool L69 3GB, UK. 3Magnetic Resonance and Image Analysis Research Centre (MARIARC), University ofLiverpool, Pembroke Place, PO Box 147, Liverpool, L69 3BX, UK. 4Department of Cognitive Neuroscience, Faculty of Psychology, Universiteit Maastricht, Postbus 616,6200MD Maastricht, The Netherlands. 5Department of Physics, University of Heidelberg, Philosophenweg 12, D-69120 Heidelberg, Germany. 6Institute of Sound andVibration Research, University of Southampton, University Road Highfield, Southampton S017 1BJ, UK. Correspondence should be addressed to P.S.([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

RESULTS

Psychometric testing and grouping

A psychometric pitch test was designed on the basis of standardmethodology1,2 for a parametric range that was extended with respectto the typical pitch range of musical instruments. The test requiredparticipants to state the dominant direction of pitch shift betweenitems in each of a total of 144 pairs of complex tones with systematicvariation of frequency (f), order of harmonics (n) (that is, a multipleinteger of the fundamental frequency (fSP ¼ n � f0)) and number (N)of present harmonics. The stimuli were such that the perceiveddirection of the shift in pitch between the two tones was dependentupon whether they are decoded in f0 or fSP pitch (Fig. 1a). Accordingly,for each individual a psychometric asymmetry coefficient was derivedby recording the number of occasions of f0 listening versus fSP listeningand computing an ‘index of pitch perception preference’ dp ¼ (f0 – fSP)/ (f0 + fSP). Analysis of the psychometric data of a large sample of 373musicians including symphony orchestra musicians28 and 48 non-musicians showed that dp demonstrated a broad bimodal distribution(Fig. 1b), which allowed for a dichotomous classification of all subjectsas belonging to one of two behavioral groups, either ‘f0 listeners’ (dp o0) or ‘fSP listeners’ (dp 4 0). The overall strong separation was mostpronounced in the lower (Fig. 1c, spectral frequencies o1,500 Hz,F1,419 ¼ 731.4, P o 0.0001) as compared with the higher spectralfrequency range. Furthermore, the tendency to base direction of pitchjudgments on the implied f0 increased systematically with increasingnumber (N) of components (Fig. 1c, F2,838 ¼ 498.5, P o 0.0001) andwith decreasing order (n) of harmonic number (Fig. 1d, F2,838¼ 352.8,Po 0.0001). The separation of f0 and fSP listeners was such that a two-component stimulus (N ¼ 2) had even stronger fundamental pitchcharacter for f0 listeners as a four-component stimulus (N¼ 4) had forfSP listeners.

Neural basis of pitch perception

MRI of brain structure and functional MEG of neural activity inresponse to harmonic complex tones were performed in a subgroup of34 f0 and 53 fSP listeners and demonstrated a strong neural basis fortype of pitch perception. The individual surfaces of all 87 left and rightauditory cortices were segmented and reconstructed three dimension-ally from the T1-weighted MRI slices (Fig. 2a). The pronouncedoblique crescent-shaped gyral entity in anterior auditory cortex,including HG in its mediolateral extent and aSTG anterolaterally,was always identified by detection of the first complete Heschl’s sulcus(cHS) as posterior boundary of HG and the first transverse sulcus(FTS) as the anterior boundary (Fig. 2b–d, red). HG may includeincomplete duplications by a sulcus intermedius (SI)29–32 indentinglocally its crown or a medial Heschl’s sulcus (mHS), not reaching thelateral end. We next calculated the gray matter volume along themedial-lateral-anterior progression of this anterior gyrus (Fig. 2b,dashed lines), by marking the corresponding gray values successively

in cross-sectional slices perpendicular to the orientation of HG andaSTG. Overall, the gray matter volume increased successively frommHG to aSTG (Fig. 2e,f). When comparing the volumes of the left andthe right hemispheres, we found a characteristic asymmetry exclusivelywithin the lateral aspect of HG irrespective of musical aptitude. The f0listeners demonstrated a pronounced leftward asymmetry (Fig. 2e),whereas fSP listeners demonstrated a pronounced rightward asymmetry(Fig. 2f). The asymmetry started at the lateral border of mHG, peakedwithin the lateral edge of HG and was absent in aSTG.

This structural asymmetry was paralleled by a correspondingfunctional asymmetry. In particular, we performed an MEG studyin which subjects were instructed to listen passively to harmoniccomplex tones covering the large parametric range of the pitch test.Auditory evoked fields were recorded continuously over both hemi-spheres. The source activity was calculated from the sensor distributionby modeling one equivalent dipole in each hemisphere. When fittedto the secondary P50m response peaking 50 ms after tone onset(P50m), the dipoles localized in the lateral portion of HG in mostcases. Figure 2g,h shows the group-averaged source waveformsfor professionals and non-musicians. P50m magnitude was larger inthe left hemisphere for f0 listeners (professionals: factor 1.3 ± 0.2,F1,33 ¼ 6.2, P o 0.01; non-musicians: factor 1.5 ± 0.2, F1,9 ¼ 14.5,P o 0.01) and in the right hemisphere for fSP listeners (professionals:factor 1.3 ± 0.2, F1,29 ¼ 22.4, P o 0.0001; non-musicians: factor1.6 ± 0.3, F1,9 ¼ 7.7, Po 0.01). No significant hemispheric asymmetrywas observed for the early P30m (ref. 27) and the pitch-sensitiveN100m response10,33.

Morphometry of auditory cortex subregions

Based on this specific connection of structural and functional asym-metry in relation to pitch perception, we identified objective criteria todefine lHG. After normalization34, a grand-average auditory cortex

Lowest harmonic number nAverage spectral frequency (kHz)7432

f0 listenersf0 listeners

fSP listeners

f0

fSP

fSP listeners

f0 listeners fSP listeners

0.25 0.44 0.79 1.40 2.49 4.43

–1

–0.5

0

0.5

1

0

δp

δp < 0 δp > 0

–1

–0.5

0

0.5

1

δp

N

43

2

4

32

N

2

3

4

2

34

Complex tones–1 –0.5 0 0.5 1

Index of pitch perception preference δp

25

50

500 ms

n

11

3

44

5

5

6

Tone 1 Tone 2

Pitch test

a

c d

bFigure 1 Psychometric testing and grouping. (a) Experimental design:

participants were required to state the dominant direction of pitch shift

between tone pairs. Solid lines represent the harmonics of the test tones and

dashed lines the harmonics which are not physically present, such as the

missing fundamental, indicated as number 1. (b) Bimodal distribution of

fundamental (f0) and spectral pitch (fSP) listeners. (c,d) Perceptual changes

as a function of the acoustic variables. The dependence on frequency (f )

and lowest order of harmonics are depicted in separate curves for stimulicomposed of N ¼ 2, 3 and 4 adjacent harmonics. n indicates the order of

harmonics in a complex tone; that is, an integer multiple of the fundamental

frequency (fSP ¼ n � f0). N is the number of physically present harmonics

(on the left of each curve).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

map of all 87 brains was calculated from the individual landmarks ofFTS, cHS and posterior border of planum temporale (PT)35. The topview of this map is depicted in Figure 3. Despite large individualdifferences in shape and progression of HG, the grand average overall 87 right and left auditory cortices demonstrated a completelysymmetric organization with respect toangulation, extent and transition from HGto aSTG. However, the PT clearly showedthe expected leftward asymmetry. Secondly,functional activation peaks in relation topitch perception8,15–17,19,20,23,36 were plottedon this averaged auditory cortex map (Fig. 3,magnified HG). The main activation peaksconcerning pitch perception are all confinedto the lateral portion of HG (x-range: ± 45to ± 65; y-range: –20 to 0). Melody-specificactivation7,13,16,23 originates more anterior(y 4 0) in aSTG. Our own findingson lateralization (Fig. 2e,f), the currentknowledge of primary auditory cortexextent16,29–31,37–40 and the functional sep-aration of adjacent pitch-sensitive areas14–17

were used as criteria to define objectiveboundaries of lHG: (i) a line perpendicularto the progression of HG at the mediolateraltwo-thirds to separate the region of mHG andlHG; (ii) a line at y ¼ 0 to separate the region

of lHG (y o 0, highlighted) and aSTG (y 4 0, colored in Fig. 3).By strict application of these boundary definitions, the gray mattervolumes of mHG, lHG, aSTG and PT were calculated. Only lHGdemonstrated a strong leftward asymmetry in f0 listeners and rightwardasymmetry in fSP listeners (Table 1). The individual auditory cortexmorphology of 32 professional musicians, eight amateur musicians andeight non-musicians (Fig. 4) illustrates (i) the large individual differ-ences with respect to angulation and progression of HG, (ii) thedifferences in structural left-right asymmetry of lHG in relation topitch perception and (iii) the structural enlargement of the entireanterior crescent-shaped gyrus in musicians as compared with non-musicians. The frequency of duplications or sulci depends onhemisphere and on perceptual preference (Table 2).

Structural, functional and perceptual asymmetry

Corresponding to the definition of perceptual asymmetry, the struc-tural and functional asymmetry of lHG was measured in terms of a‘structural asymmetry index’ ds ¼ (RlHG – LlHG)/(RlHG + LlHG) and a‘functional P50m asymmetry index’ df ¼ (RP50m – LP50m)/(RP50m +LP50m). The correlation of ds versus dp was calculated separately bysystematically including or excluding HG duplications: (i) all duplica-tions including complete posterior duplications (PDs) were included

100500–50Time (ms)

100500–50–30

–20

–10

Time (ms)

0

10

20

30

Dip

ole

ampl

itude

(nA

m)

–30

–20

–10

0

10

20

30

Dip

ole

ampl

itude

(nA

m)

Prof

Non

LR

LR

Prof

Non

LR

LR

Prof

Non

LR

LR

Prof

Non

LR

LR

P50m P50m

fSP listeners

fSP listeners

f0 listeners

f0 listeners

7043221Direction of HG/aSTG (mm)

7043221Direction of HG/aSTG (mm)

0

0.2

0.4

0.6

0.8

Vol

ume

of 3

-mm

slic

es (

cm3 )

0

0.2

0.4

0.6

0.8

Vol

ume

of 3

-mm

slic

es (

cm3 )

mHG mHG

IHG IHG

aSTG aSTG

cHSmHS

mHG

Posterior

IHGFTS SI

aSTG

AnteriorACa b

e f

g h

c d

30L R

Tonality13

Melody condition7

Melody vs. fixed pitch16

Pitch chroma17

Spectral variation23

Fixed pitch vs. noise16

Pitch change19

Pitch salience15

P50m harmonic complex tonesLively vs. fixed pitch20

Temporal regularity8

Pitch onset response36

Spectral vs. temporal variation23

a

pRL

20

10

–10

–20

–30

–40

–50

–60 –50 –40 60

x (mm)

y (m

m)

50

PT

PD

FTSSI, mHS

cHS mHG

PT

aSTG

IHG IHG

aSTG

40

0

Figure 3 Averaged landmarks of 87 auditory cortices (top view, standard stereotaxic coordinates34;

line width of the dashed landmarks corresponds to averaged s.e.m.). The crescent-shaped anterior

gyrus including HG and anterior supratemporal gyrus (aSTG) showed a completely symmetric shape

and progression with respect to angulation, extent, duplications and curvature. A complete posterior

duplication (PD) was considered to be part of the planum temporale (PT). The main activation peaks of

functional imaging studies are plotted on the magnified map. Black open circles indicate the averaged

localization of the auditory evoked P50m response in lHG measured by MEG. Pitch-specific activation

localized in lHG, melody-specific activation more anterior in aSTG. In key at right, superscript numbers

refer to references.

Figure 2 Neural basis of fundamental and spectral pitch perception. (a) 3D

reconstruction of an individual auditory cortex and auditory evoked activity

(blue and red dipoles). (b–d) The top view of three individual 3D surface

reconstructions of right auditory cortex shows the pronounced gyral entity

including HG and aSTG (colored red), bordered anteriorly by the first

transverse sulcus (FTS) and posteriorly by the first complete Heschl’s sulcus

(cHS). HG may include a sulcus intermedius (SI), indenting locally the crown

of HG, or a medial Heschl’s sulcus (mHS), not reaching the lateral end. Graymatter volume was successively calculated in cross-sectional slices along

the medial-lateral-anterior progression. (e,f) Fundamental pitch listeners

demonstrated a pronounced leftward asymmetry of gray matter volume and

spectral pitch listeners a rightward asymmetry, peaking within the lateral

one-third of HG. (g,h) Functional asymmetry of the auditory evoked P50m

source activity of lHG in response to harmonic complex tones. P50m

magnitude was relatively larger in the left hemisphere for f0 listeners and vice

versa for fSP listeners, irrespective of musical aptitude. Prof, professional

musician; non, non-musician, in all figures.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

(r ¼ 0.47, P o 0.0001); (ii) right and left PD’s were excluded; thatis, cHS was always the posterior boundary (r ¼ 0.77, P o 0.0001);(iii) only left PDs were excluded, (r ¼ 0.81, P o 0.0001); (iv) allduplications posterior to cHS and mHS were excluded (r ¼ 0.70,Po 0.0001); or (v) all duplications posterior to cHS, mHS and SI wereexcluded; that is, the anterior HG27 was considered (r ¼ 0.38,Po 0.001). Overall, the correlation remained strongly robust irrespec-tive of inclusion or exclusion of duplications. However, with respect to asymmetric definition of boundaries, the effect was most pronounced ifcHS was always the posterior boundary (case 2, Fig. 5a). The scatterplot of P50m asymmetry (df) versus pitch perception asymmetry (dp)demonstrated again a robust effect (r¼ 0.63, Po 0.0001, Fig. 5b). As aconsequence, structural and functional asymmetry of lHG was stronglylinked (r ¼ 0.55, P o 0.001). However, ifabsolute magnitude of the neural substratewas considered instead of the relative hemi-spheric asymmetries, the correlation droppedto insignificance (P50m dipole amplitude ver-sus gray matter volume of lHG: r¼ 0.04, n.s.).

Influence of musical ability

With respect to the absolute magnitude of theneural lHG substrate, large group-specificdifferences were found, corroborating andextending earlier findings27. The crescent-shaped anterior convolution of auditory cor-tex including mHG, lHG and aSTG, coloredin red (right hemisphere) and in blue (lefthemisphere, Figs. 2b–d and 4), demonstratedstrongly enlarged gray matter volume in pro-fessional musicians as compared with non-musicians (Table 1). The gray matter volumecorrelated significantly with musical aptitudeas measured by the Advanced Measure ofMusic Audiation (AMMA) test27,41 (lHG:r ¼ 0.71, P o 0.0001). Furthermore, theauditory evoked P50m response showed afivefold larger magnitude in professionals ascompared with non-musicians (Fig. 2e,f; pro-fessional: 25.1 ± 1.9 nAm, non-musician:5.3 ± 1.2 nAm, factor 4.7 ± 0.8, F1,85 ¼ 51.2,Po 0.0001) and correlated with the intensityof musical practice during the last ten years(r ¼ 0.80, P o 0.0001). No significant

correlation was found between any musical ability parametersand neural (df, ds) or perceptual (dp) asymmetries. As a consequence,the correlations shown in Figure 5a,b remained strong for thenon-musicians (ds versus dp: r ¼ 0.75, P o 0.0001; df versus dp:r ¼ 0.67, P o 0.0001).

DISCUSSION

Application of an auditory judgment task that is known to producelarge perceptual differences across individuals1,2 in the investigation ofa large sample of musicians and non-musicians enabled a systematiccategorization in f0 and fSP listeners. The f0 listeners tended to basedirection of pitch change judgments on the implied fundamentalfrequency, whereas the fSP listeners performed the pitch change

Table 1 Gray matter volume of auditory cortex subregions

f0 prof (21) fSP prof (30) f0 non (10) fSP non (10) Prof vs. non

Left hemisphere

mHGL 1.42 ± 0.07 1.26 ± 0.06 0.70 ± 0.05 0.73 ± 0.06 ***

lHGL (highlighted) 2.09 ± 0.14*** 1.67 ± 0.09 1.20 ± 0.06** 0.87 ± 0.11 ***

aSTGL 5.14 ± 0.32 5.02 ± 0.21 3.48 ± 0.26 3.47 ± 0.23 ***

PTL 4.03 ± 0.32* 4.38 ± 0.28** 3.63 ± 0.41 3.59 ± 0.34* n.s.

Right hemisphere

mHGR 1.25 ± 0.07 1.37 ± 0.06 0.61 ± 0.11 0.79 ± 0.07 ***

lHGR (highlighted) 1.61 ± 0.12 2.21 ± 0.12** 0.73 ± 0.10 1.22 ± 0.09** ***

aSTGR 5.09 ± 0.42 4.55 ± 0.20 3.49 ± 0.26 3.05 ± 0.14 ***

PTR 3.23 ± 0.31 2.65 ± 0.16 3.26 ± 0.29 2.58 ± 0.25 n.s.

*P o 0.05, **P o 0.01, ***P o 0.001 ANOVA, left versus right hemisphere (designating the dominant hemisphere), Prof vs. non, overall significance of professionals versus non-musicians).mHG, medial Heschl’s gyrus; lHG, lateral Heschl’s gyrus; aSTG, anterior supratemporal gyrus; PT, planum temporale. Values: mean (cm3) ± s.e.m., standard stereotactic voxel size29. Prof, professionalmusician. Non, non-musician; n.s., not significant.

f0 listeners, musicians

L R

fSP listeners, musicians

f0 listeners, non-musicians fSP listeners, non-musicians

Figure 4 Individual HG morphology. f0 listeners demonstrate a larger left lHG and fSP listeners a larger

right lHG in most cases (lHG highlighted, red and blue arrows). The occurrence of sulci and duplications

(SI, asterisks; mHS, ‘+’ symbols; PD, black open circles) depends on hemisphere and pitch perception

preference. Professional musicians and amateurs (A) showed greater gray matter volume of the entire

anterior convolution including HG and aSTG (colored structure) than non-musicians (bottom).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

judgment on the basis of the spectral envelope rather than fundamentalfrequency. By using MRI and MEG we found a strong neural basis ofboth types of pitch perception, which corroborated the functionalspecialization of left auditory cortex for rapid temporal23,25 and theright hemisphere for spectral23 processing.

The pitch test allowed systematic analysis of the influence of acousticstimulus properties on pitch perception. First, the separation of f0 andfSP listeners was much stronger within a spectral frequency range below1,500 Hz and decreased towards the higher frequencies. This may bereflected in the psychometric ‘dominance principle’42, stating that thecenter of harmonic order which is dominant for pitch perceptiondecreases from n ¼ 5 (f0 ¼ 200 Hz) to n ¼ 1 (f0 ¼ 2,000 Hz). Second,the general increase of f0 dominance with decreasing n and increasingN(Fig. 1c,d) corroborates the current knowledge of fundamental pitchsaliency2,10,42. However, our data emphasize that pitch perceptiondepends on both spectral envelope and fundamental frequency infor-mation with different weighting and cannot be explained by a simpleone-to-one relationship between perceived pitch and fundamentalfrequency6,10. A functional separation of periodicity coding and spec-tral integration at the level of auditory cortex may account for thedifferent pitch percepts, varying between individuals by up to three orfour octaves, when the same sound was presented.

The large perceptual dissimilarity in pitch perception was paralleledby a strong inter-individual structural and functional variability inauditory cortex. To visualize this pronounced variation, which wasanticipated a long time ago from myelogenetic studies40, we depictedfor the first time the full shape and progression of HG and aSTG,forming together a crescent-shaped gyral entity of the anterolateralstream in human auditory cortex14. The huge morphological differ-ences with respect to angulation, extent of HG and its obliquetransition towards aSTG were superposed by a conspicuous increaseof gray matter volume in musicans27,43 (Fig. 4). Averaged over 87subjects, the shape was completely symmetric, whereas some morpho-metric studies show asymmetric average maps29,32. However, these andour studies are not sufficiently comparable owing to differences insample size, definition of boundaries and extent of the region ofinterest. Furthermore, detailed morphometric analysis showed thatthe correlation between preference of f0 versus fSP perception andneural asymmetry was confined to lHG and was not present for theentire body of HG or aSTG. In particular, f0 listeners demonstrated anasymmetry favoring the left lHG in terms of both cortical gray mattervolume and auditory evoked P50m activity, whereas fSP listeners

showed the opposite asymmetry. This corroborated the functionalspecialization of lHG as a pitch processing center12. Our results implya left-hemispheric specialization for (missing) fundamental pitchperception and a right hemispheric specialization for spectral pitchperception, consistent with a recent functional imaging studycomparing the neural processing of spectral and temporal variation23.Furthermore, left auditory cortex is sensitive to short time scales(25–50 ms)25,44 and right auditory cortex to slower time scales(200–300 ms)25. The fundamental pitch (f0) of an instrumentalsound (f0 4 25 Hz) reflects its periodicity6 T ¼ 1/ f0, correspondingto time segments shorter than 40 ms. Thus, the existence of two pitchcenters may facilitate the extraction of fundamental pitch in leftauditory cortex and spectral pitch in right auditory cortex. Indeed,most professional musicians perceive simultaneously both fundamen-tal and spectral pitch from an ambiguous tone, and the subjectivedifferences are rather relative than absolute24. Here, these relativeperceptual differences were found to correlate strongly to neuralasymmetries, as anticipated by earlier studies on cerebral dominance45.Thus, a greater volume on the left may predispose one to hear the f0in an ambiguous tone, and vice versa, a greater volume on the rightmay lead to a dominant perception of spectral pitch or singleharmonics. A psychophysical study on patients with temporal lobelesions demonstrates fourfold higher thresholds for determining thedirection of pitch changes in patients with right hemisphere lesions thatencroached on HG. This study concludes that detecting the direction ofpitch changes may depend largely on the right HG18. However, there isno conflict with our results, because the magnitudes of pitch changesused in our study were largely above the f0 discrimination thresholdsinvestigated in the lesion study. Thus, the latter may reflect in particularthe accuracy of pitch direction judgment relative to the magnitude off0 discrimination, irrespective of general left-right hemispheric later-alization effects in pitch perception.

The direct link between structural and functional asymmetriesreported here seems to be confined to a local mechanism within thepitch center of lateral HG, present only in a small time range of 30–40ms around P50m activity of the MEG recordings. The main activationpeaks from comparable fMRI15,16 studies depict the activity within amuch larger time frame of several seconds, including other activitieswhich originate partially from lateral HG, in particular the pitch-sensitive N100 activity10,33 and the pitch onset response36 occurringabout 130 ms after tone onset. Overall, functional asymmetries are notalways observed in every task, subject, stimulus condition or specified

Table 2 Frequency of HG duplications and sulci

SI mHS PD Single HG

f0 listener (33)

Only right HG 6 1 3 5

Only left HG 4 2 6 6

Left and right HG 3 0 2 19

Without 20 30 22 3

fSP listener (54)

Only right HG 29 9 6 0

Only left HG 0 1 10 36

Left and right HG 6 1 1 11

Without 19 43 37 9

SI, sulcus intermedius; mHS medial Heschl’s sulcus; PD, complete posterior duplication. Thefrequency of HG duplications or sulci (given by the number of observed cases from all 87subjects) depends on hemisphere and on perceptual preference. In particular, for all 39 caseswhere either in the left or in the right hemisphere a SI was present, the frequency differedsignificantly for fSP listeners as compared to f0 listeners (w2(1) ¼ 12.9, P o 0.001).

f0 listeners

δs δf

δp δp

0.3

0

–0.3

–0.5

0.3

0

–0.3

–0.5

–1 –0.5 0 0.5 –1 –0.5 0 0.5

fSP listeners f0 listeners fSP listeners

MusiciansNon-musicians

MusiciansNon-musicians

a b

Figure 5 Pitch perception preference and neural asymmetries. (a) Correlation

of pitch asymmetry (dp) versus structural asymmetry (ds) of gray matter

in lHG. (b) Correlation of pitch asymmetry (dp) versus functional asymmetry

(df) of P50m magnitude. The correlations are strong irrespective of

musical aptitude.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

processing level. In some cases or individuals the activation maps reallyare symmetric14,16,20. However, with respect to pitch- or melody-related tasks, the majority of functional imaging studies14,16,23,24

showed right-sided asymmetries. Multimodal functional imagingwith professional musicians may help to clarify under which conditionsthe observed pitch asymmetries arise.

Our findings demonstrate a strong correlation between the relativehemispheric lateralization of structure and function and perceptualpreference, as confirmed by morphometric31,45,46 and functional ima-ging studies25. In contrast, absolute magnitudes of the neural HGsubstrate depend on musical expertise consistent with previous stu-dies27,33,47,48. At a more fundamental level, a recent post-mortem studyobserved characteristic asymmetries in auditory belt areas at the level ofthe underlying microanatomical architecture26. However, the exactrelation between absolute changes at the microanatomical and macro-anatomical level still remains unclear and needs further clarification.Use-dependent subcortical changes such as dendritic arborizationchanges or interdigitation of neuronal clusters26 may enhance themagnitude of synchronized postsynaptic potentials as measured byMEG in professional musicians without changing, however, the corticalthickness or volume of the underlying gray matter tissue. Likewise, alarger HG gray matter volume implies a larger neural network per se,independent of neural connectivity, and may reflect a greater potentialof musical aptitude. A strong relation between absolute magnitudes ofstructure and function was observed at the early automatic processinglevel in anteromedial HG27 and obviously disappears in secondaryauditory cortex. Post-mortem studies29,30 showed considerable varia-tion between structure and function when comparing the boundariesof individual microanatomical structure and macroanatomical visiblemagnetic resonance landmarks. These results suggest that the func-tional areas do not correspond to areas defined on the basis ofmacroscopic boundaries. Nevertheless, when calculating micro- andmacroanatomical probability maps, the corresponding centroids of thelocation were found to be almost identical14,16,29. As a consequence,despite large individual variability29, the medial two-thirds of anteriorHG were considered a reliably good approximation of the anterolateralextent of primary auditory cortex16,24,37, as confirmed by cytoarchi-tectonic38, histochemical39 and myelogenetic40 studies. Overall, weconclude that the relative hemispheric lateralization of functionaland structural size reflects the type of pitch processing irrespective ofmusical aptitude, whereas the absolute magnitudes of the neural HGsubstrate depends on musical expertise. Further studies may clarifywhether the observed lateralization is linked to a preference forcharacteristic physical sound properties (in particular, the fastertemporal structure of percussive sounds versus the slower time scaleof sustained sounds25,49,50) therefore influencing musical instrumentpreference and musical performance.

METHODSSubjects. A large sample of 420 right handed healthy subjects (125 professional

musicians including members of the Royal Liverpool Philharmonic Orches-

tra28, 181 graduate students in music, 66 amateur musicians and 48 non-

musicians) were recruited for the psychometric evaluation of pitch perception

and evaluation of musical aptitude. A subgroup of 87 subjects participated in

the MRI and MEG measurements (out of 51 professionals, including 21

members of the Royal Liverpool Philharmonic Orchestra, 16 amateurs and

20 non-musicians, 34 were f0 and 53 fSP listeners). Averaged over the groups, no

significant differences in age, sex and head size were observed. Experimental

procedures were approved by the relevant local research ethics committee.

Pitch test. The pitch test included 144 different pairs of harmonic complex

tones. Each tone pair consisted of two consecutive harmonic complex tones

(duration 500 ms, 10-ms rise-fall time, interstimulus interval 250 ms). Each test

tone comprised two, three or four adjacent harmonics, leaving out the

fundamental frequency. Overall, the tone pairs were designed with six different

upper component frequencies (293, 523, 932, 1,661, 2,960 and 5,274 Hz)

chosen to be equidistant on a logarithmical frequency scale corresponding to

the musical interval of a major ninth, beginning with D3 (293 Hz) up to C8

(5,274 Hz). The upper component frequency of both tones in each tone pair

was identical to minimize the perception of edge pitch. Furthermore, the lowest

presented harmonic number transitions (n1,min - n2,min) within a tone pair

was one of the following four conditions: 2 - 3, 3 - 4, 4 - 6 or 7 - 9.

Thus, the spectral components ranged between 146 and 5,274 Hz and f0between 29 to 1,318 Hz. The magnitude of f0 pitch changes ranged between

a factor of 1.1 (major second) to 1.8 (minor seventh); the fSP pitch changes of

the lowest harmonic number ranged from a factor of 1.1 to 3.1. Thus, the

magnitude of these pitch changes was considerably larger than pitch discrimi-

nation threshold18. By using additionally complete harmonic complex tones

(nmin ¼ 1) as reference tones (conditions 2 - 1, 3 - 1, 4 - 1 and 5 - 1), the

pitch test allowed detection of octave-shifted fundamental pitches (for instance,

one octave above f0). This case occurred only significantly for three-component

stimuli within the higher spectral range (41,000 Hz) and was not considered

to be part of f0 perception. However, if fundamental pitch perception would

mean both f0 and octave-shifted f0 perception, our results would not qualita-

tively change. All stimuli were presented binaurally in pseudorandomized order

using a Hammerfall DSP Multiface System with a stimulus level of 50 dB nSL to

avoid the interfering superposition of combination tones. Each tone pair was

repeated once and the next new tone pair presented after a pause of 2 s.

Subjects were instructed to select the dominant pitch direction or to answer

according to the first, spontaneous impression, if either both directions were

perceived at the same time or if tones lacked a clear pitch. Test duration was

22 min. All subjects were tested on an identical set of stimuli. A subgroup of 37

subjects repeated the pitch test about 6 months later and demonstrated strong

individual re-test reliability (r ¼ 0.96, P o 0.0001).

Morphometry. The three-dimensional (3D) gray matter surface reconstruc-

tions of all individual auditory cortices were calculated from T1-weighted

structural MRI data (Siemens, Symphony, 1.5-T) after segmentation using

BrainVoyager software (Brain Innovation). All brains were rotated in direction

of the antero-posterior commissural line and normalized by unfiltered trans-

formation in Talairach space34. Using standard definitions of the anatomical

auditory cortex landmarks29–32,35,37, the sagittal MRI slices of the individual

auditory cortices were segmented along the Sylvian fissure to obtain PT, HG

and aSTG. The inclusion range of image gray values was calculated in a

normalized box around left and right auditory cortex. For gray matter surface

reconstruction and morphometry, the ‘gray value inclusion range’ was defined

individually from the intensity histogram of gray values for each left and right

auditory cortex, by identifying (i) the half-amplitude side-lobe of the gray

matter peak distribution towards cerebral spinal fluid and (ii) the saddle point

between the gray and white matter peak. All gray value voxels inside this

inclusion range were marked and used for 3D reconstruction and morpho-

metry. The non-automated parts of this structural analysis (in particular, the

identification of individual landmarks from the individual 3D surface recon-

structions of auditory cortex) were obtained by observers who were blind to

subject group and hemisphere.

HG subregions. Here we used the most obvious and well-accepted definition

of HG29,30 by identifying the first complete Heschl’s sulcus (cHS) as its

posterior and the crescent-shaped first transverse sulcus (FTS)29–32 as its

anterior boundary. cHS was identified by virtue of having a clear lateral

indentation (Fig. 2d), large mediolateral extent and pronounced depth, and

divided auditory cortex in two parts: (i) an anterior auditory stream including

HG and aSTG and (ii) a posterior stream including the PT. Based on normal-

ization, the pronounced crescent-shaped gyrus anterior to cHS was subdivided

systematically in mHG, lHG and aSTG. By using functional and structural

criteria (Fig. 3), the y ¼ 0 line was found to be an appropriate borderline to

separate HG and aSTG. Individual extrapolation of FTS towards the lateral end

of HG, as proposed by some morphometric studies31,32,37, was impossible in

our sample owing to large individual differences with respect to angulation and


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

asymmetric progression (Fig. 4). The extent of mHG was defined by the medial

two-thirds of HG along the mediolateral direction of HG, similar to the

estimated extent of primary auditory cortex16,29,30. lHG was the remaining part

of the gyrus between aSTG and mHG. The gray matter volume of each

specified subregion was calculated by marking and counting all included gray

values of the individual gray matter peak distribution (see above). Finally, the

volume of lHG was the difference between the volumes of HG and mHG.

Magnetoencephalography. Using a Neuromag-122 whole-head MEG system,

we recorded auditory evoked fields to twelve harmonic complex sounds

covering the parametric range as used for the pitch test (f0: 100 and 500 Hz,

lowest number of harmonics: 1, 4 and 10, complete spectrum and three

adjacent harmonics). Subjects were instructed to listen passively to the sounds,

each of which was presented 240 times in pseudorandomized order. Cortical

responses were averaged for each frequency using the BESA program (MEGIS

Software) and collapsed into an individual grand average for source analysis

(3,600 averages). The source activity of the auditory evoked P50m response was

separated from the earlier P30m and later N100m response by spatiotemporal

source modeling27, using one equivalent dipole in each hemisphere. The fitting

intervals were adjusted to the individual source waveforms in time intervals

around the peaks defined by their half–side lobes. Signal strength was calculated

for each peak relative to a 100 ms baseline.

ACKNOWLEDGMENTSWe thank K. Sartor for providing the 3D-MRI in Heidelberg, the radiographicstaff at MARIARC for assistance with MRI data acquisition in Liverpool andE. Hofmann (Music Academy, Basel); D. Geller, R. Schmitt and T. van der Geld(University of Music and Performing Arts, Mannheim); C. Klein (Institute ofMusic Pedagogy, Halle) and D. Schmidt (Conservatory of Music and PerformingArts, Stuttgart) for assistance with collecting the psychometric data.




1. Smoorenburg, G.F. Pitch perception of two-frequency stimuli. J. Acoust. Soc. Am. 48,924–942 (1970).

2. Laguitton, V., Demany, L., Semal, C. & Liegeois-Chauvel, C. Pitch perception: adifference between right-and left-handed listeners. Neuropsychologia 36, 201–207(1998).

3. von Helmholtz, H.L.F. On the Sensations of Tone (Longmans, London, 1885).4. Terhardt, E. Pitch, consonance and harmony. J. Acoust. Soc. Am. 55, 1061–1069

(1974).5. Wessinger, C.M. et al.Hierarchical organization of the human auditory cortex revealed by

functional magnetic resonance imaging. J. Cogn. Neurosci. 13, 1–7 (2001).6. Langner, G. Periodicity coding in the auditory system. Hear. Res. 60, 115–142 (1992).7. Griffiths, T.D., Buchel, C., Frackowiak, R.S.J. & Patterson, R.D. Analysis of temporal

structure in sound by the human brain. Nat. Neurosci. 1, 422–427 (1998).8. Gutschalk, A., Patterson, R.D., Rupp, A., Uppenkamp, S. & Scherg, M. Sustained

magnetic fields reveal separate sites for sound level and temporal regularity in humanauditory cortex. Neuroimage 15, 207–216 (2002).

9. Formisano, E. et al. Mirror-symmetric tonotopic maps in human primary auditory cortex.Neuron 40, 859–869 (2003).

10. Pantev, C., Hoke, M., Lutkenhoner, B. & Lehnerz, K. Tonotopic organization of theauditory cortex: pitch versus frequency representation. Science 246, 486–488 (1989).

11. Seifritz, E. et al. Spatiotemporal pattern of neural processing in the human auditorycortex. Science 297, 1706–1708 (2002).

12. Griffiths, T.D. Functional imaging of pitch analysis. Ann. NY Acad. Sci. 999, 40–49(2003).

13. Janata, P. et al. The cortical topography of tonal structures underlying western music.Science 298, 2167–2170 (2002).

14. Hall, D. et al. Spectral and temporal processing in human auditory cortex. Cereb. Cortex12, 140–149 (2002).

15. Penagos, H., Melcher, J.R. & Oxenham, A.J. A neural representation of pitch salience innonprimary human auditory cortex revealed with functional magnetic resonance ima-ging. J. Neurosci. 24, 6810–6815 (2004).

16. Patterson, R.D., Uppenkamp, S., Johnsrude, I.S. & Griffiths, T.D. The processing oftemporal pitch and melody information in auditory cortex.Neuron36, 767–776 (2002).

17. Warren, J.D., Uppenkamp, S., Patterson, R. & Griffiths, T.D. Separating pitch chromaand pitch height in the human brain. Proc. Natl. Acad. Sci. USA 100, 10038–10042(2003).

18. Johnsrude, I.S., Penhune, V.B. & Zatorre, R.J. Functional specificity in the right humanauditory cortex for perceiving pitch direction. Brain 123, 155–163 (2000).

19. Warren, J.D. & Griffiths, T.D. Distinct mechanisms for processing spatial sequences andpitch sequences in the human auditory brain. J. Neurosci. 23, 5799–5804 (2003).

20. Griffiths, T.D., Uppenkamp, S., Johnsrude, I., Josephs, O. & Patterson, R.D. Encoding ofthe temporal regularity of sound in the human brainstem. Nat. Neurosci. 4, 633–637(2001).

21. Tervaniemi, M. Lateralization of auditory-cortex functions. Brain Res. Brain Res. Rev.43, 231–246 (2003).

22. Devlin, J.T. et al. Functional asymmetry for auditory processing in human primaryauditory cortex. J. Neurosci. 23, 11516–11522 (2003).

23. Zatorre, R. & Belin, P. Spectral and temporal processing in human auditory cortex.Cereb. Cortex 11, 946–953 (2001).

24. Zatorre, R.J., Belin, P. & Penhune, V. Structure and function of auditory cortex: musicand speech. Trends Cogn. Sci. 6, 37–46 (2002).

25. Boemio, A., Fromm, S. & Poeppel, D. Hierarchical and asymmetric temporal sensitivityin human auditory cortices. Nat. Neurosci. 8, 389–395 (2005).

26. Galuske, R.A., Schlote, W., Bratzke, H. & Singer, W. Interhemispheric asymmetries ofthe modular structure in human temporal cortex. Science 289, 1946–1949 (2000).

27. Schneider, P. et al. Morphology of Heschl’s gyrus reflects enhanced activation in theauditory cortex of musicians. Nat. Neurosci. 5, 688–694 (2002).

28. Sluming, V. et al. Voxel-based morphometry reveals increased gray matter density inBroca’s area in male symphony orchestra musicians. Neuroimage 17, 1613–1622(2002).

29. Rademacher, J. et al. Probabilistic mapping and volume measurement of human primaryauditory cortex. Neuroimage 13, 669–683 (2001).

30. Morosan, P. et al. Human primary auditory cortex: cytoachitectonic subdivisions andmapping into a spatial reference system. Neuroimage 13, 684–701 (2001).

31. Penhune, V.B., Zatorre, R.J., MacDonald, J.D. & Evans, A.C. Interhemispheric ana-tomical differences in human primary auditory cortex: probabilistic mapping and volumemeasurement from magnetic resonance scans. Cereb. Cortex 6, 661–672 (1996).

32. Leonard, C.M., Puranik, C., Kuldau, J.M. & Lombardino, L.J. 1998. Normal variation inthe frequency and location of human auditory cortex. Heschl’s gyrus: where is it? Cereb.Cortex 8, 397–406 (1998).

33. Pantev, C. et al. Increased auditory cortical representation in musicians. Nature 392,811–813 (1998).

34. Talairach, J. & Tournoux, P. Co-planar Stereotaxic Atlas of the Human Brain (Thieme,New York, 1988).

35. Westbury, C.F., Zatorre, R.J. & Evans, A.C. Quantifying variability in the planumtemporale: a probability map. Cereb. Cortex 9, 392–405 (1999).

36. Ritter, S., Dosch, H.G., Specht, H.J. & Rupp, A. Neuromagnetic responses reflect thetemporal pitch change of regular interval sounds. Neuroimage (in the press).

37. Penhune, V.B., Cismaru, R., Dorsaint-Pierre, R., Petitto, L. & Zatorre, R. The morpho-metry of auditory cortex in the congenitally deaf measured using MRI. Neuroimage 20,1215–1225 (2003).

38. Galaburda, A. & Sanides, F. Cytoarchitectonic organization of the human auditory cortex.J. Comp. Neurol. 190, 597–610 (1980).

39. Wallace, M.N., Johnston, P.W. & Palmer, A.R. Histochemical identification of corticalareas in the auditory region of the human brain. Exp. Brain Res. 143, 499–508(2002).

40. Pfeifer, R.A. Pathologie der Horstrahlung und der corticalen Horsphare. inHandbuch derNeurologie Vol. 6 (eds. Bumke, O. & Forster, O.) 533–626 (Springer, Berlin, 1936).

41. Gordon, E.E. Introduction to Research and the Psychology of Music (GIA, Chicago,1998).

42. Plomp, R. Aspects of Tone Sensation (Academic, London, 1976).43. Meyer, A. The search for a morphological substrate in the brains of eminent persons

including musicians: a historical review. in Music and the Brain (eds. Critchley, M. &Henson, R.A.) 255–281. (Heinemann, London, 1977).

44. Rupp, A., Gutschalk, A., Uppenkamp, S. & Scherg, M. Middle latency auditory-evokedfields reflect psychoacoustic gap detection thresholds in human listeners.J. Neurophysiol. 92, 2239–2247 (2004).

45. Galaburda, A.M., Le May, M. & Kemper, T.L. Right-left asymmetries in the brain.Structural differences between the hemispheres may underlie cerebral dominance.Science 199, 852–856 (1978).

46. Schlaug, G., Jancke, L., Huang, Y. & Steinmetz, H. In vivo evidence of structural brainasymmetry in musicians. Science 267, 699–701 (1995).

47. Munte, T.F., Kohlmetz, C., Nager, W. & Altenmuller, E. Superior auditory spatial tuning inconductors. Nature 409, 580 (2001).

48. Gaser, C. & Schlaug, G. Brain structures differ between musicians and non-musicians.J. Neurosci. 23, 9240–9245 (2003).

49. Joliveau, E., Smith, J. & Wolfe, J. Tuning of vocal tract resonance by sopranos. Nature427, 116 (2004).

50. Belin, P., Zatorre, R.J. & Lafaille, P. Voice-selective areas in human auditory cortex.Nature 403, 309–312 (2000).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Filling-in of visual phantoms in the human brain

Ming Meng1, David A Remus1 & Frank Tong2

The constructive nature of perception can be demonstrated under viewing conditions that lead to vivid subjective impressions

in the absence of direct input. When a low-contrast moving grating is divided by a large gap, observers report seeing a ‘visual

phantom’ of the real grating extending through the blank gap region. Here, we report fMRI evidence showing that visual phantoms

lead to enhanced activity in early visual areas that specifically represent the blank gap region. We found that neural filling-in

effects occurred automatically in areas V1 and V2, regardless of where the subject attended. Moreover, when phantom-inducing

gratings were paired with competing stimuli in a binocular rivalry display, subjects reported spontaneous fluctuations in conscious

perception of the phantom accompanied by tightly coupled changes in early visual activity. Our results indicate that phantom

visual experiences are closely linked to automatic filling-in of activity at the earliest stages of cortical processing.

A particularly vivid and powerful form of perceptual completioninvolves the formation of moving visual phantoms. When a low-contrast moving grating is divided by an orthogonal gap, subjectstypically perceive a dimmer version of the surrounding dynamicpattern continuing across the blank gap region (Fig. 1a; Supplemen-tary Movie 1). Visual phantoms are greatly enhanced by motion of thesurrounding inducers, and can occur anywhere in the normal visualfield across gaps as large as 101 (ref. 1). These illusory phantoms seemto match the pattern, motion, color and texture of the physicallysurrounding inducers, and, notably, they can mimic the perceptualeffects of real stimuli. For example, moving phantoms can induce localmotion aftereffects, suggesting that phantom impressions are activelyrepresented in the brain2. However, the neural basis of visual phantomshas not been studied previously. Such knowledge is important forunderstanding how the brain fills in gaps in sensory information andforms representations of subjective perceptual content in the absence ofdirect input.

We used functional magnetic resonance imaging (fMRI) to measureneural responses to visual phantoms in corresponding regions of thehuman visual cortex. Subjects maintained central fixation while low-contrast oscillating gratings were presented in the upper- and lower-leftvisual field, separated by a large 71 � 71 gap. The gratings werepresented either vertically (Fig. 1a), leading to the perception of aphantom, or horizontally (Fig. 1b), to serve as a ‘no-phantom’ controlcondition. A sequence of letters was presented concurrently at fixation,and before each fMRI trial, subjects were cued to perform an atten-tionally demanding task involving either the central letters or theperipheral gratings. This manipulation of spatial attention served asan important control, as it is known that focal attention can activatecorresponding regions of visual cortex even when no stimulus ispresent3,4. Moreover, this manipulation allowed us to test if perceptualfilling-in of visual phantoms occurs automatically or requires focused

attention. Psychophysical studies of other types of filling-in suggest thatsome forms of perceptual completion can occur preattentively5,6. Wepredicted that brain areas involved in visual phantom formation shouldshow greater activity to the vertical phantom condition than to thehorizontal no-phantom condition, that enhanced activity to visualphantoms should be specific to retinotopic regions corresponding tothe location of the phantom and that phantom filling-in should involvean automatic process that operates independent of spatial attention. Ina second experiment, we further tested if activity corresponding to theblank gap region is tightly coupled to moment-to-moment changes inconscious perception of the visual phantom.

RESULTS

Behavioral results

For the grating task, subjects were instructed to press one of twocorresponding buttons to report whenever the top or bottom gratingbriefly decreased in contrast at random intervals. The letter identifica-tion task required subjects to report whenever a ‘J’ or ‘K’ appeared in arapid sequence of letters presented at fixation. The two tasks werematched for difficulty by independently varying magnitude of contrastchange and letter presentation rate. Mean performance was nearlyidentical for the two tasks and did not significantly differ (grating task:83% correct, letter task: 82% correct, T ¼ 0.34, P ¼ 0.743).

In a separate psychophysical experiment performed after the fMRIstudy, we confirmed that the subjects perceived reliable visual phan-toms in the appropriate experimental conditions. Subjects were askedto adjust the contrast of a real grating to match the strength of anyperceptual impression they might have in the blank gap between thetwo gratings. All eight subjects reported stronger visual phantomswhen presented with vertically aligned gratings than when presentedwith horizontal gratings (Fig. 1a,b; mean perceived contrast, 1.52%and 0.27%, respectively, T ¼ 5.7, P o 0.001).


1Psychology Department, Green Hall, Princeton University, Princeton, New Jersey 08544, USA. 2Psychology Department, 301 Wilson Hall, Vanderbilt University, 111 21st

Avenue South, Nashville, Tennessee 37203, USA. Correspondence should be addressed to M.M. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Regions of interest

The critical regions of interest (ROIs) consisted of the retinotopicregions in visual areas V1 through V4 that corresponded to the visual-field location of the blank gap (Fig. 1c). Retinotopic regions corre-sponding to the surrounding gratings served as control ROIs toevaluate the spatial specificity of phantom filling-in effects. All ROIswere localized in separate reference scans by presenting high-contrastflickering checkerboards in the same locations as the surroundinggratings and blank gap region (see Methods). ROIs correspondingto the center of the blank gap were identified by selecting asmall, restricted region in each visual area consisting of the most highlyactive voxels found during the reference scan. Visual areas weredelineated on flattened cortical surfaces using standard retinotopicmapping techniques7–9.

Cortical responses to visual phantoms

Average fMRI time courses in regions of V1 and V2 corresponding tothe blank gap showed enhanced activity for visual phantoms (Fig. 2).Peripheral oscillating gratings and central letters were presented duringthe 12-s stimulus period, which was preceded and followed by 16-sfixation-baseline periods. Even though there was no physical stimulusin the gap, corresponding regions of V1 and V2 (Fig. 2a,b) showedsubstantially greater fMRI responses to the vertical phantom conditionthan to the horizontal no-phantom condition, irrespective of whethersubjects were attending to the peripheral gratings or to the centralletters. These enhanced neural responses to illusory visual phantomswere highly reliable, with all eight subjects showing effects in thepredicted direction.

Increased responses to the vertical phantom condition were foundeven under conditions of strong attentional suppression. When subjectswere required to attend to the central letters and to ignore theperipheral gratings, negative BOLD responses were found in V1 andV2 regions corresponding to the gap. These results are consistent withprevious reports of attentional suppression in early visual areas10 andthe fact that stimulation of neighboring cortical regions (by thesurrounding gratings) can lead to local suppression of BOLD activityand neuronal firing rates11 (A. Shmuel et al., Soc. Neuro. Abstr. 125.12003). Despite these powerful effects of attentional suppression, V1 and

V2 showed greater responses to the visual phantom condition than tothe control condition, even when attention was directed away fromthe peripheral gratings. These results suggest that neural filling-in ofthe phantom occurred automatically, independent of the locus ofattention, in these early visual areas.

Cortical locus of phantom-enhanced activity

To determine the retinotopic and cortical foci of these responsesto visual phantoms, we compared fMRI response amplitudes tothe vertical and horizontal grating conditions for all regions of interest.If the enhanced fMRI responses found in V1 and V2 truly correspondto an internal representation of the visual phantom, then theseenhancement effects should be specific to the blank gap regionand absent from the stimulus surround. This prediction can bedistinguished from the well-documented effects of collinear facilitationobserved in primary visual cortex. These studies show that V1 neuronsrespond more strongly to an oriented target in their receptivefield if collinear flanking stimuli are presented outside of thereceptive field, even though the flanking stimuli alone evoke negligibleresponse in absence of the target12,13. Therefore, collinear facilitationleads to the opposite prediction of phantom filling-in: namely,that response enhancement should occur primarily in retinotopicregions that receive direct stimulation from the collinear verticalgratings but not in regions corresponding to the blank gap where thephantom is perceived.

We measured cortical responses to the visual phantom by calculatingthe difference between fMRI activity in the vertical phantom condition

V1 V2 V3 V4v

V3aV3V2V1

* * *

a

c

b Figure 1 Experimental design and stimuli. Subjects viewed letters at central

fixation and oscillating sine-wave gratings in the upper and lower left visual

field, separated by a 71 gap. Subjects attended to the identity of the letters

or to brief decrements in the contrast of the peripheral gratings. (a) Phantom

condition: vertically aligned gratings led to strong impressions of a phantom

grating extending through the blank gap region. A weaker impression of a

phantom may be observed here with static gratings. (b) Control condition:

horizontal gratings appeared to move as a perceptual group but did not leadto an impression of a phantom. (c) Retinotopic regions of interest in areas

V1–V4 corresponding to the location of the blank gap (red) and surrounding

gratings (blue), shown on the cortical flatmap of a representative subject.

Regions were identified in separate scans using flickering checkerboards and

were aligned to retinotopic maps collected from the same subject.

1612840–4Time (s)

1612840–4–0.6

–0.4

–0.2

Time (s)

0

0.2

0.4

0.6

–0.6

–0.4

–0.2

0

0.2

0.4

0.6

Sig

nal c

hang

e (%

)

Attendto letters

Attend togratings

Area V1 Area V2

Attend togratings

Attendto letters

a b

Figure 2 Average fMRI time courses for retinotopic regions corresponding to

the blank gap in areas V1 and V2, plotted as a function of grating orientation

and attentional task (n ¼ 8). Both V1 (a) and V2 (b) showed significantlygreater responses to the vertical phantom condition (solid lines) than to the

horizontal no-phantom condition (dashed lines) during the 12-s stimulus

period (gray shading), regardless of whether subjects had to attend to the peri-

pheral gratings (black lines) or to the central letters (gray lines). Error bars at

right indicate the average standard error (±1 s.e.m.) for all data points shown.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

and activity in the horizontal no-phantom condition (phantom minusno-phantom) for each visual area, retinotopic location and attentionalcondition, separately (Fig. 3). A significant positive difference indicateenhanced responses to the visual phantom. Early visual areas corre-sponding to the blank region showed strong evidence of phantom-enhanced activity (Fig. 3a), whereas no such enhancement was foundin retinotopic regions corresponding to the physically surroundinggratings (Fig. 3b). These results provide strong support for the notionthat visual phantom formation involves filling-in of neural activity inearly visual areas that specifically represent the blank gap.

Additional analyses showed that V1 and V2 were the only visualareas corresponding to the blank gap that demonstrated consistentlygreater responses to the visual phantom (Fig. 3a), independent ofattention. Area V3 showed enhanced responses to visual phantomswhen attention was directed to the peripheral gratings but showedunreliable effects when the gratings were ignored. Activity in highervisual areas V3a and V4v did not reliably differ for the vertical phantomcondition and the horizontal no-phantom condition (F o 1). Onepossible interpretation is that higher visual areas are less involved in

phantom filling-in. Alternatively, it is possible that enhanced activity tovisual phantoms was more difficult to detect in higher areas because ofthe larger point-spread function of visual projections to these areas. Inany case, the positive results indicate robust effects in areas V1 and V2,suggesting that activity in these early visual areas may be important forperceptual filling-in of visual phantoms.

As a further test of the retinotopic specificity of these responses tovisual phantoms, we compared effect sizes for small and large ROIs.If enhanced V1 responses to the vertical gratings truly reflect filling-inof the illusory phantom, then one would predict stronger enhance-ment effects for small ROIs that closely correspond to the center ofthe gap where only the phantom is perceived. However, if enhancedactivity instead reflects a stimulus-driven response to the neighboringborder of the surrounding gratings, then one would predict strongerenhancement effects for large ROIs that extend toward the boundarybetween gap and stimulus. Whereas all previous analyses focused onsmall ROIs (B200 mm3) identified in separate reference scans at highthreshold, here we generated enlarged ROIs (B630 mm3) by loweringstatistical thresholds until ROI sizes matched V1 cortical magnificationestimates of the blank gap region9. Statistical comparisons showedsignificantly greater response enhancement for visual phantoms insmall ROIs than in large ROIs (0.23% versus 0.16% signal changerespectively, F1,7 ¼ 5.9; P o 0.05). These results suggest that theenhanced responses in V1 reflect phantom impressions around the

V4vV3aV3V2V1V4vV3aV3V2V1

**

–0.1–0.1

0.1

0.2

0.3

00

0.1

0.2

0.3 Attend to letters

Attend to gratings**

*****

**

*

ROI: blank gap region ROI: surrounding grating region

Sig

nal d

iffer

ence

(%

)(p

hant

om –

no-

phan

tom

)

a b Figure 3 Magnitude of responses to visual phantoms. (a) Retinotopic regions

corresponding to the blank gap. (b) Retinotopic regions corresponding to

the surrounding gratings. Ordinate axis shows the difference between fMRI

activity for the vertical phantom condition and activity for the horizontal no-

phantom condition (vertical phantom minus horizontal no-phantom). Gray

bars indicate attention to peripheral gratings; white bars indicate attention

to central letters. Error bars denote s.e.m. across subjects. Significant

differences in fMRI activity between the vertical phantom and horizontalcontrol conditions are indicated (* P o 0.05; ** P o 0.01; *** P o 0.001).

V4vV3aV3V2V1

V4vV3aV3V2V1

0

0.2

0.4

0.6

0.8

1.0

1.2

0

0.2

0.4

0.6

0.8

1.0

1.2

Vertical gratingsHorizontal gratings

ROI: surroundinggrating region

ROI: blank gap region

Sig

nal d

iffer

ence

(%

)

(atte

nd g

ratin

gs –

atte

nd le

tters

)

Sig

nal d

iffer

ence

(%

)

(atte

nd g

ratin

gs –

atte

nd le

tters

)

–4.0

–8.0

t

t

4.0

8.0

V1

V1V2 V3 V4v

V2V3 V3a

* * *

a b

c

Figure 4 Attentional modulation effects

across visual areas. (a) Activation map of a

representative subject, displayed on the flattened

cortical surface with visual areas delineated.

Attention to peripheral gratings activated

peripheral retinotopic regions (yellow-red),

whereas attention to the central letters activated

foveal regions (green-blue) more strongly.

Asterisks indicate the location of the foveal

cortical representation. (b,c) Magnitude of

attentional modulation averaged across subjects

in retinotopic regions corresponding to the

blank gap (b) and the surrounding gratings (c).

Ordinate axis shows the difference between fMRI

activity for the ‘attend gratings’ condition and

activity for the ‘attend letters’ condition (‘attendgratings’ minus ‘attend letters’). Error bars, ±1

s.e.m. Retinotopic regions in each visual area

showed highly significant attentional modulation

effects (P o 0.0001), and modulations were

significantly stronger in higher visual areas

(F4,34 ¼ 6.9, P o 0.0005).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

center of the gap rather than a stimulus-drivenresponse to the border of the surroundinggratings.

Attentional modulation effects

The effects of spatial attention were wide-spread and independent of the effects ofphantom filling-in. Activation maps plottedon flattened cortical surfaces showed thatextensive regions of visual cortex were modu-lated by spatial attention (Fig. 4a). Corticalrepresentations of the visual periphery weremuch more active when subjects attended tothe peripheral gratings, whereas foveal regionswere more active when subjects attended to the central letters. Theperipheral regions showing strong attentional enhancement overlappedconsiderably with ROIs corresponding to the blank gap and surround-ing gratings (Fig. 1c).

We further measured the magnitude of attentional enhancement foreach grating condition and ROI by calculating the difference betweenactivity for attention to gratings and activity for attention to letters(attention to gratings minus attention to letters). Unlike the effects ofphantom filling-in, which were spatially specific to V1 and V2 regionscorresponding to the blank gap, attentional modulation effects wereequally powerful in regions corresponding to the gap (Fig. 4b) andregions corresponding to the surrounding gratings (Fig. 4c). Thesemodulation effects were highly significant for every visual area (P o0.0001) but were significantly stronger in higher visual areas than inearly visual areas (F4,34 ¼ 6.9, P o 0.0005). In contrast, phantomfilling-in effects were found only in early visual areas. Taken together,these results indicate that the effects of visual attention, across bothretinotopic space and visual hierarchy, were independent of the effectsof phantom filling-in.

Neural correlates of conscious perception

We conducted an additional experiment in three subjects to evaluate ifthe activity in early visual areas is tightly coupled to the consciousperception of visual phantoms, independent of possible low-level visualdifferences between stimulus conditions. To address this issue, wedevised a novel binocular rivalry display to manipulate perception ofthe visual phantom while the physical stimuli were held constant.Binocular rivalry occurs when different stimuli are presented to the twoeyes, leading to spontaneous alternations in conscious perceptionbetween the two monocular images14. In our rivalry display, verticallyaligned gratings, which would normally elicit stable perception of avisual phantom when presented alone to one eye, were paired withcompeting horizontal gratings presented to corresponding locations ofthe other eye (Fig. 1). When subjects viewed this display while

maintaining steady fixation, they reported experiencing spontaneousalternations between perceptual dominance and suppression of thevisual phantom every few seconds. The visual phantom was perceivedonly when both vertical gratings were dominant; no phantom wasperceived when either or both of the horizontal gratings predominated.Subjects were instructed to press one of three keys to indicate whetherthey perceived the visual phantom, both horizontal gratings and nophantom or mixed dominance and no phantom. To perform the task,subjects had to attend steadily to the spatial regions corresponding tothe surrounding gratings and intervening gap. Therefore, unlike thefirst experiment, spatial attention was held constant in this experimentwhile perception fluctuated spontaneously. Because the physical stimuli

128401284012840128401284012840Time (s) Time (s) Time (s)



Stimulusalternation

Binocularrivalry

Phantom

No phantom

Stimulusalternation

Phantom

No phantomNo phantom

Binocularrivalry

Stimulusalternation

Phantom

Binocularrivalry

V3V2V1

–0.2

–0.1

0

0.1

0.2

–0.2

–0.1

0

0.1

0.2

–0.2

–0.1

0

0.1

0.2

Sig

nal c

hang

e (%

)S

igna

l cha

nge

(%)

Sig

nal c

hang

e (%

)S1

S2

S3

Figure 5 Event-related activity for reported

percepts of phantom (black solid lines) or no

phantom (gray dashed lines) during rivalry and

stimulus alternation. Average time courses show

activity corresponding to the blank gap in areas

V1–V3 for all three subjects (error bars, ±1 s.e.m.).

Activity corresponding specifically to the blank

gap was measured as the difference betweenfMRI activity in blank gap region and activity

in the surrounding region (blank gap region minus

surrounding region). Vertical lines at time 0 s

indicate when the subject reported seeing a

phantom or no phantom.

0.4

0.2

0.2

V1

S1

S2

S3

V2 V3Subject

Stimulus alternation (% signal change)

Riv

alry

(%

sig

nal c

hang

e)

0–0.2

–0.2

–0.4

–0.4

0.4

Figure 6 Comparison of response amplitudes for rivalry (ordinate) and

stimulus alternation (abscissa) for reported percepts of phantom (positive

points) or no phantom (negative points). Amplitudes reflect the peak-to-

trough difference in fMRI response (time window 0–7 s) for each subject,

visual area and condition. All data points fall close to line of unity (line ofbest fit: slope 0.98, intercept 0.002), indicating equivalent response

amplitudes for rivalry and stimulus alternation.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

remained constant throughout the experiment, any activity that variedin accordance with perceptual dominance of the phantom can beattributed only to changes in conscious perception rather than changesin the physical stimuli.

We used event-related fMRI analyses to isolate awareness-relatedactivity15 corresponding to when subjects perceived a visual phantomor no phantom. Awareness-related activity obtained under conditionsof rivalry viewing was further compared to the activity evoked byphysical stimulus alternation. In separate fMRI runs, subjects viewedphysical alternations between the two monocular displays (Fig. 1a,b)consisting of the same temporal sequence of alternations reported in aprevious rivalry run.

Event-related analyses of retinotopic regions corresponding specifi-cally to the blank gap region demonstrated a tight coupling betweenearly visual activity and conscious perception of the visual phantom(Fig. 5). When the visual phantom was dominant during rivalryviewing, a concomitant rise in fMRI activity was observed in areasV1 through V3, with peak activity occurring about 5–7 s after thereported emergence of the phantom. In contrast, perceptual suppres-sion of the phantom was accompanied by suppressed activity in thesesame cortical areas, with similar time courses for negatively peakingactivity. Linear trend analyses showed that activity changes in V1 andV2 were statistically reliable for all three subjects (P o 0.05) and werereliable for two out of three subjects in V3. Awareness-related activitychanges during rivalry were very similar to activity changes evoked byphysical stimulus alternation, both in amplitude and time course. Weconstructed a scatter plot to compare the amplitude of activity changesfor rivalry and stimulus alternation on the basis of the peak-to-troughdifference in fMRI signal intensity for each subject, percept type andvisual area (Fig. 6). Positive points indicate positive fMRI responsescorresponding to visual phantom perception; negative points indicatenegative fMRI responses corresponding to no-phantom perception.All points clustered closely to the line of unity (slope 0.98, intercept0.002, R2 ¼ 0.93, T¼ 14.27, Po 10�9) and did not differ reliably froma predicted slope of 1 (T ¼ 0.26, P ¼ 0.80) or an intercept of 0(T ¼ 0.12, P ¼ 0.91), corresponding to identical response amplitudesfor rivalry and stimulus alternation. Therefore, changes in consciousperception of the visual phantom, in the absence of any physicalstimulus change, led to cortical responses that were as strong asthose evoked by stimulus-driven changes. These data demonstrate atight coupling between activity in early visual areas that represent theblank gap region and moment-to-moment fluctuations in consciousperception of the visual phantom.

DISCUSSION

Our study suggests that the formation of visual phantoms results fromautomatic filling-in of activity in early visual areas. Illusory visualphantoms that spanned a large blank gap led to spatially specificenhancement of activity in corresponding regions of V1 and V2. Thesephantom filling-in effects seemed to occur automatically, independentof the locus of spatial attention, and remained robust even whensubjects had to attend away from the peripheral gratings. Unlike thespatially restricted effects of phantom filling-in in V1 and V2, manip-ulations of spatial attention led to global modulations in activity acrossall visual areas in regions corresponding to both the blank gap and thestimulus surround. In a second experiment, subjects were presentedwith a novel binocular rivalry display to evaluate if moment-to-moment changes in conscious perception of the visual phantom, inthe absence of any physical stimulus change, would lead to concomitantchanges in early visual activity. We found a tight coupling betweenawareness of the visual phantom and activity in early visual areas. These

awareness-related activity changes are unlikely to be explained in termsof shifts in spatial attention, judging from the results of the firstexperiment, and instead seem to reflect the neural representation ofthe phenomenal visibility of the visual phantom.

The combined results suggest that phantom filling-in involves afairly automatic, bottom-up cortical mechanism that does not requirefocal attention, yet this mechanism cannot be triggered by the simplepresentation of phantom-inducing stimuli to the retina. Previous fMRIstudies have shown that rivalry suppression can strongly modulateactivity in human V1, despite the presence of steady retinal input16–18.These studies led us to predict that binocular rivalry might be capableof suppressing the neural formation of visual phantoms at this sameearly site. Consistent with this prediction, we found that rivalrysuppression of the phantom-inducing gratings led to the suppressionof both the phenomenal visibility of the phantom and correspondingactivity in visual areas as early as V1. Therefore, intact corticalprocessing of the surrounding inducer gratings seems to be necessaryfor phantom filling-in to occur. Our results provide support for theview that attentional and perceptual mechanisms offer distinct con-tributions to visual awareness; both seem to be necessary for reporting avisual experience19,20. Attention is commonly thought to reflect aflexible, domain-general system that is needed to access and to reportthe contents of perception (presumably by enhancing and ‘broad-casting’21 a subset of sensory signals), whereas the representation ofperceptual content itself is thought to depend on domain-specificsensory areas of the brain. Our results suggest that the neural repre-sentation of visual phantoms occurs in early visual areas and that theperceptual contents of these representations are readily available toawareness once they are accessed by attentional mechanisms.

These findings suggest that early visual areas have an importantrole in representing subjective visual content in the absence of directsensory input. The results pertain to an ongoing debate regarding therelative roles of early and high-level visual areas in conscious percep-tion20,22 and to philosophical and psychological discussions on whetherthe brain needs to fill in information that is absent23–26. Our datasuggest that the brain does recreate visual representations in ‘zones ofabsence’ and that these representations of subjective content can berealized at the earliest stages of cortical processing, including primaryvisual cortex20.

The present findings also contribute to current understanding of theneural interactions that mediate various forms of perceptual filling-in.Filling-in is known to occur in many sensory modalities, acrossmultiple scales of space and time, and likely involves cooperativemechanisms that operate at multiple spatial-temporal scales and stagesof processing26–31. In the visual domain, single-unit studies have foundevidence of receptive field enlargement in V1 due to small retinalscotomas32–35, enhanced responses to illusory contours in V2 (ref. 36)and dynamic filling-in of artificial scotomas in V3 (ref. 37). Here, usingfMRI, we were able to monitor activity across multiple visual areas andshow that enhanced activity to moving visual phantoms occurredindependent of the locus of attention in areas V1 and V2. Our resultsprovide evidence that missing information about visual orientation ormoving patterns can be automatically filled in at the earliest stages ofcortical processing, even when spatial attention is directed elsewhere.These findings agree with a recent fMRI study showing color filling-ineffects in human V1 (ref. 38) and indicate that primary visual cortex isinvolved in the perceptual completion of multiple types of visualfeatures, including color, pattern and motion.

What types of neural interactions might account for the long-rangeperceptual completion of visual phantoms? As discussed, top-downeffects of attentional feedback seem to be unlikely to account for our


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

results. Instead, the prominent filling-in effects found in V1 and V2indicate that activity corresponding to the surrounding gratings canautomatically propagate to retinotopic regions representing the blankgap. This propagating activity might arise from long-range horizontalconnections between neighboring columns with similar orientationpreferences39, dynamic interactions between V1 and V2 (ref. 40) orautomatic feedback interactions between higher visual areas with largereceptive fields and early areas41. The present study found positiveevidence in favor of the first two possibilities; areas V3a and V4v didnot show reliable effects of phantom filling-in. However, the largerpoint-spread function in these higher visual areas may have led to morespatially diffuse effects of filling-in. Consistent with this possibility,intermediate areas such as V3 showed more mixed evidence of filling-in. Regardless of the underlying mechanisms for propagating activity,the results indicate that phantom filling-in effects emerge at the earlieststage of cortical processing in the primary visual cortex.

This study also provides evidence linking neural filling-in activitywith moment-to-moment changes in conscious perception, underconditions of rivalry viewing. Our results indicate that the neuralmechanisms underlying binocular rivalry can strongly influence andeven suppress the mechanisms underlying visual phantom formation.Suppression effects were robust; rivalry responses to visual phantomswere as strong as the responses evoked by stimulus alternation. Currenttheories suggest that rivalry can result from competition at multiplelevels of visual processing42. Nonetheless, several functional imagingstudies have demonstrated highly reliable effects of rivalry suppressionin human V1 (refs. 16–18). According to the present study, binocularrivalry not only suppresses the neural representation of real stimuli butcan also suppress the neural representation of illusory stimuli in regionsof visual cortex that do not receive direct stimulation.

Our study provides concurrent evidence suggesting that earlyvisual areas, including V1, may be important in both selective andconstructive aspects of conscious perception. Selective perception isevident when observers are aware of only one of two rivaling imagespresent on the retinae; constructive perception is evident when obser-vers experience phantom visual impressions in regions that lack directretinal input. Future studies may further demonstrate the contributionof early visual areas to the selective and constructive nature of humanvisual experience.

METHODSSubjects. Eight right-handed healthy adults (two females) with normal or

corrected-to-normal visual acuity participated in the first experiment. Three

male subjects participated in the second experiment. All subjects gave informed

written consent. The study was approved by the Institutional Review Panel for

Human Subjects at Princeton University.

Experimental design and stimuli. Visual stimuli were rear projected onto a

screen in the scanner bore using a luminance-calibrated LCD projector driven

by a Macintosh G3 computer. The top grating, bottom grating and the middle

blank area were each 71� 71 in size, with the center of the blank gap positioned

81 to the left of fixation. Stimulus parameters were set to optimize the

perceptual salience of the moving visual phantom in the vertical grating

condition (contrast 15%, spatial frequency 0.286 cyc/deg, temporal frequency

1.13 Hz, direction reversal every 1 s) with the mean luminance of the back-

ground set to match the lowest luminance portion of the gratings43. Horizontal

peripheral gratings consisted of the same stimuli rotated by 901. During

fixation-rest periods, gratings were replaced by uniform gray squares that

matched the mean luminance of the gratings.

In the first experiment, a randomized fMRI block design was used to

measure cortical responses to vertical and horizontal peripheral gratings

(Fig. 1) while subjects maintained fixation on a sequence of centrally presented

letters (letter size B0.41, fixation size 0.61). Stimuli were presented for 12-s

periods, interleaved between 16-s fixation-rest periods. A letter cue was

presented above the fixation point for 1 s before each stimulus period to

indicate whether subjects had to monitor for the appearance of a ‘J’ or ‘K’ in

the letter sequence or monitor for brief contrast decrements (180 ms duration)

of the top or the bottom peripheral grating. To balance the difficulty level of the

two tasks, letter presentation rates and the magnitude of the contrast decrement

were adjusted on each run to achieve a performance level of about 80% correct

detection (letter presentation rates 160–220 ms/item, contrast decrements

8–10%). Both types of targets appeared on six occasions in every stimulus

block at randomly selected intervals. The four experimental conditions (two

grating orientations � two attentional tasks) appeared in a randomized order

twice in each run, and each subject performed 8–12 runs in a single

experimental session.

In the second experiment, red-green anaglyph filters were used to present

vertical gratings to one eye and horizontal gratings of equal luminance and

contrast to the other eye (Fig. 1a,b). Stimulus parameters were otherwise

identical to those used in the first experiment. During rivalry fMRI runs, both

vertical and horizontal gratings were continuously presented over the entire

2-min period while subjects reported the perceptual dominance of the visual

phantom. On stimulus alternation runs, the two monocular displays were

presented in alternation using the same temporal sequence of reported

alternations from a previous rivalry run. Stimulus alternation was achieved

by adjusting the relative contrast of each display over a 250-ms time window.

Each subject performed a total of 20–26 rivalry runs and an equal number of

stimulus alternation runs over the period of two 2.5-h fMRI sessions. To

control for possible effects of eye dominance, eye assignment of red-green filters

was counterbalanced across sessions, and subjects received an equal number of

fMRI runs in which vertical/horizontal stimuli were assigned to either eye.

MRI acquisition. Scanning was performed on a 3.0-T Siemens MAGNETOM

Allegra scanner using a standard head coil at the Center for the Study of Brain,

Mind and Behavior, Princeton University. A high-resolution T1-weighted three-

dimensional MPRAGE anatomical scan was acquired for each participant (FOV

256 � 256; resolution: 1 mm3). To measure BOLD contrast, standard T2*-

weighted gradient-echo echoplanar imaging was used to acquire 25 slices

perpendicular to the calcarine sulcus to cover the entire occipital lobe (first

experiment: TR 2000 ms, TE 30 ms, flip angle 901, slice thickness 3 mm, slice

gap 0.75 mm, in-plane resolution 3 � 3 mm; second experiment: TR 1000 ms,

TE 35 ms, flip angle 651). Visual areas were delineated in a separate session by

using rotating wedges and expanding rings to map the boundaries between

visual areas on flattened cortical representations7–9. ROIs corresponding to the

locations of the blank gap and the surrounding gratings (Fig. 1c) were identified

in separate reference scans using high-contrast flickering checkerboard stimuli.

MRI analysis. All fMRI data underwent three-dimensional (3D) motion

correction and were analyzed using Brain Voyager software and custom

routines in Matlab. Slow drifts in signal intensity were removed by linear

detrending; no spatial or temporal smoothing was applied. The general linear

model was used to identify voxels that significantly correlated with a predicted

hemodynamic time course derived by convolving a gamma function with the

relevant stimulus periods in the flickering checkerboard reference scans. ROIs

corresponding to the blank gap were selected at a high statistical threshold and

adjusted to yield ROI sizes of 200 mm3 for each visual area (minimum

threshold T ¼ 3.3, P o 0.001, max T ¼ 16.8). ROIs for V1, V2 and V3 were

200 mm3 in volume when pooled across the most active voxels in both dorsal

and ventral regions. Our rationale for choosing this fairly small ROI size

(equivalent to 7.4 voxels of 3 � 3 � 3 mm size) was to focus on activity

corresponding to the center of the blank gap region without compromising

signal-to-noise in our fMRI measurements. Matching of ROI sizes further

ensured that statistical comparisons across visual areas and regions of interest

were as comparable as possible. ROIs corresponding to the locations of both

surrounding gratings were selected in a similar manner by adjusting thresholds

to yield a total volume of 200 mm3 in each visual area. Average event-related

fMRI time courses were calculated for each ROI, experimental condition and

subject. For the first experiment, average fMRI amplitudes were measured

based on the peak of the hemodynamic response (averaged across time points

6–12 seconds post–stimulus onset) relative to prestimulus activity levels (from


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

time points �4 to 0 s). Within-subjects analysis of variance and planned

comparisons were used to test for statistically reliable differences (P o 0.05) in

fMRI response amplitudes for each experimental condition and retinotopic

region of interest.

In the second experiment, fMRI intensities were normalized relative to the

mean intensity of each run, after discarding the first 12 s to avoid transient

activity corresponding to the onset of the visual display. Activity corresponding

specifically to the blank gap was isolated by calculating the difference between

fMRI activity in the blank gap region and activity in the surrounding region

(blank gap region minus surrounding region) for every fMRI run, before event-

related averaging. fMRI data were sorted and binned in an event-related fashion

according to the time of reported percepts of phantom or no-phantom, using

previously described methods15,17. Because fMRI responses are dependent

on the duration of neural activity, brief percepts of o2 s were excluded from

the analysis. Event-related fMRI amplitudes were measured based on the peak-

to-trough difference in activity by identifying points of maximal deflection

for the initial peak/trough (time window 0–2 s) and final peak/trough (time

window 5–7 s).


ACKNOWLEDGMENTSWe thank Y. Kamitani, Y. Sasaki and A. Seiffert for comments on earlier versionsof this manuscript, and the Center for the Study of Brain, Mind and Behavior,Princeton University, for MRI support. This work was supported by grant R01EY14202 from the US National Institutes of Health to F.T.




1. Tynan, P. & Sekuler, R. Moving visual phantoms: a new contour completion effect.Science 188, 951–952 (1975).

2. Weisstein, N., Maguire, W. & Berbaum, K. Phantom-motion aftereffect. Science 198,955–958 (1977).

3. Kastner, S., Pinsk, M.A., De Weerd, P., Desimone, R. & Ungerleider, L.G. Increasedactivity in human visual cortex during directed attention in the absence of visualstimulation. Neuron 22, 751–761 (1999).

4. Ress, D., Backus, B.T. & Heeger, D.J. Activity in primary visual cortex predictsperformance in a visual detection task. Nat. Neurosci. 3, 940–945 (2000).

5. Ramachandran, V.S. Filling in the blind spot. Nature 356, 115 (1992).6. Rauschenberger, R. & Yantis, S. Masking unveils pre-amodal completion representation

in visual search. Nature 410, 369–372 (2001).7. Sereno, M.I. et al. Borders of multiple visual areas in humans revealed by functional

magnetic resonance imaging. Science 268, 889–893 (1995).8. DeYoe, E.A. et al. Mapping striate and extrastriate visual areas in human cerebral cortex.

Proc. Natl. Acad. Sci. USA 93, 2382–2386 (1996).9. Engel, S.A., Glover, G.H. & Wandell, B.A. Retinotopic organization in human visual

cortex and the spatial precision of functional MRI. Cereb. Cortex 7, 181–192 (1997).10. Somers, D.C., Dale, A.M., Seiffert, A.E. & Tootell, R.B. Functional MRI reveals spatially

specific attentional modulation in human primary visual cortex. Proc. Natl. Acad. Sci.USA 96, 1663–1668 (1999).

11. Shmuel, A. et al. Sustained negative BOLD, blood flow and oxygen consumptionresponse and its coupling to the positive response in the human brain. Neuron 36,1195–1210 (2002).

12. Polat, U., Mizobe, K., Pettet, M.W., Kasamatsu, T. & Norcia, A.M. Collinear stimuliregulate visual responses depending on cell’s contrast threshold.Nature 391, 580–584(1998).

13. Kapadia, M.K., Ito, M., Gilbert, C.D. & Westheimer, G. Improvement in visual sensitivityby changes in local context: parallel studies in human observers and in V1 of alertmonkeys. Neuron 15, 843–856 (1995).

14. Levelt, W.J. On Binocular Rivalry (Institute of Perception Rvo-Tno, Soesterberg,The Netherlands, 1965).

15. Tong, F., Nakayama, K., Vaughan, J.T. & Kanwisher, N. Binocular rivalry and visualawareness in human extrastriate cortex. Neuron 21, 753–759 (1998).

16. Polonsky, A., Blake, R., Braun, J. & Heeger, D.J. Neuronal activity in human primaryvisual cortex correlates with perception during binocular rivalry.Nat. Neurosci.3, 1153–1159 (2000).

17. Tong, F. & Engel, S.A. Interocular rivalry revealed in the human cortical blind-spotrepresentation. Nature 411, 195–199 (2001).

18. Lee, S.H., Blake, R. & Heeger, D.J. Traveling waves of activity in primary visual cortexduring binocular rivalry. Nat. Neurosci. 8, 22–23 (2005).

19. Lamme, V.A. Why visual attention and awareness are different. Trends Cogn. Sci. 7,12–18 (2003).

20. Tong, F. Primary visual cortex and visual awareness. Nat. Rev. Neurosci. 4, 219–229(2003).

21. Baars, B.J. The conscious access hypothesis: origins and recent evidence. Trends Cogn.Sci. 6, 47–52 (2002).

22. Rees, G., Kreiman, G. & Koch, C. Neural correlates of consciousness in humans. Nat.Rev. Neurosci. 3, 261–270 (2002).

23. Dennett, D.C. Consciousness Explained (Little Brown, Boston, 1991).24. Ramachandran, V.S. Filling in gaps in logic: reply to Durgin et al. Perception 24,

841–845 (1995).25. Durgin, F.H., Tripathy, S.P. & Levi, D.M. On the filling in of the visual blind spot: some

rules of thumb. Perception 24, 827–840 (1995).26. Pessoa, L., Thompson, E. & Noe, A. Finding out about filling-in: a guide to perceptual

completion for visual science and the philosophy of perception. Behav. Brain Sci. 21,723–748 (1998).

27. Ramachandran, V.S. & Gregory, R.L. Perceptual filling in of artificially induced scotomasin human vision. Nature 350, 699–702 (1991).

28. Ramachandran, V.S., Rogers-Ramachandran, D. & Stewart, M. Perceptual correlates ofmassive cortical reorganization. Science 258, 1159–1160 (1992).

29. Shams, L., Kamitani, Y. & Shimojo, S. Illusions. What you see is what you hear. Nature408, 788 (2000).

30. Chen, L.M., Friedman, R.M. & Roe, A.W. Optical imaging of a tactile illusion in area 3bof the primary somatosensory cortex. Science 302, 881–885 (2003).

31. Paradiso, M.A. & Nakayama, K. Brightness perception and filling-in. Vision Res. 31,1221–1236 (1991).

32. Kaas, J.H. et al. Reorganization of retinotopic cortical maps in adult mammals afterlesions of the retina. Science 248, 229–231 (1990).

33. Gilbert, C.D. & Wiesel, T.N. Receptive field dynamics in adult primary visual cortex.Nature 356, 150–152 (1992).

34. Fiorani, M. Jr., Rosa, M.G.P., Gattass, R. & Rocha-Miranda, C.E. Dynamic surrounds ofreceptive fields in primate striate cortex: A physiological basis for perceptual comple-tion? Proc. Natl. Acad. Sci. USA 89, 8547–8551 (1992).

35. Komatsu, H., Kinoshita, M. & Murakami, I. Neural responses in the retinotopicrepresentation of the blind spot in the macaque V1 to stimuli for perceptual filling-in.J. Neurosci. 20, 9310–9319 (2000).

36. von der Heydt, R., Peterhans, E. & Baumgartner, G. Illusory contours and cortical neuronresponses. Science 224, 1260–1262 (1984).

37. De Weerd, P., Gattass, R., Desimone, R. & Ungerleider, L.G. Responses of cells inmonkey visual cortex during perceptual filling-in of an artificial scotoma. Nature 377,731–734 (1995).

38. Sasaki, Y. & Watanabe, T. The primary visual cortex fills in color. Proc. Natl. Acad. Sci.USA 101, 18251–18256 (2004).

39. Stettler, D.D., Das, A., Bennett, J. & Gilbert, C.D. Lateral connectivity and contextualinteractions in macaque primary visual cortex. Neuron 36, 739–750 (2002).

40. Shmuel, A. et al. Retinotopic axis specificity and selective clustering of feedbackprojections from V2 to V1 in the owl monkey. J. Neurosci. 25, 2117–2131 (2005).

41. Angelucci, A. et al. Circuits for local and global signal integration in primary visualcortex. J. Neurosci. 22, 8633–8646 (2002).

42. Blake, R. & Logothetis, N.K. Visual competition. Nat. Rev. Neurosci. 3, 13–21(2002).

43. Gyoba, J. Disappearance of stationary visual phantoms under high luminant or equi-luminant inducing gratings. Vision Res. 34, 1001–1005 (1994).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Regret and its avoidance: a neuroimaging study ofchoice behavior

Giorgio Coricelli1, Hugo D Critchley2, Mateus Joffily1, John P O’Doherty2, Angela Sirigu1 & Raymond J Dolan2

Human decisions can be shaped by predictions of emotions that ensue after choosing advantageously or disadvantageously.

Indeed, anticipating regret is a powerful predictor of future choices. We measured brain activity using functional magnetic

resonance imaging (fMRI) while subjects selected between two gambles wherein regret was induced by providing information

about the outcome of the unchosen gamble. Increasing regret enhanced activity in the medial orbitofrontal region, the anterior

cingulate cortex and the hippocampus. Notably, across the experiment, subjects became increasingly regret-aversive, a cumulative

effect reflected in enhanced activity within medial orbitofrontal cortex and amygdala. This pattern of activity reoccurred just

before making a choice, suggesting that the same neural circuitry mediates direct experience of regret and its anticipation. These

results demonstrate that medial orbitofrontal cortex modulates the gain of adaptive emotions in a manner that may provide a

substrate for the influence of high-level emotions on decision making.

Cost-benefit analyses in everyday decision making are often difficultbecause our evidence about future outcomes is incomplete or, atbest, probabilistic. According to standard economic theory, rationaldecision makers should optimize their choice strategies throughreliance on expected utility. However, it is known that human decisionsdeviate from this ideal and are influenced by other, less rational,considerations. For instance, the Dutch postal code lottery is popular,although playing the lottery can be considered an irrational behavior.Its success has been explained by the possibility that people anticipatehow bad they would feel if, not having bought a ticket, their postalcode is drawn.

A sense of responsibility in human decision making operatesthrough a process of counterfactual reasoning1,2 that enables us torelate the outcome of a previous decision with what we would haveobtained had we opted for a rejected alternative. Regrets are what weexperience when this comparison is to our disadvantage3–6, and it canbe defined as a cognitively enriched emotion embodying a feeling ofresponsibility for negative outcomes of choices7,8. This contrasts withdisappointment, which is an emotion related to an unexpected negativeoutcome, without an obligatory sense of personal responsibility9,10. Asregret is an unpleasant feeling that encapsulates a sense of responsi-bility, we learn from past experience to minimize its likely reoccurrencewhen considering a new choice decision11.

The interplay between decision making and emotional processingcan be assumed to involve a contribution from several brain structures,including areas associated with executive and emotional processing.Several lines of evidence indicate that orbitofrontal cortex (OFC)assigns relative value to stimuli and updates the salience of primaryand secondary reinforcers12–14. Both simple and abstract complex

instrumental reinforcers such as monetary gain and loss evokeemotions that serve to guide behavior, and the OFC is acandidate substrate for the generation of such emotions15–18. Wewould also suggest that the OFC modulates the gain of emotionsusing a top-down process in which a paradigmatic cognitive process,specifically counterfactual thinking, contributes to an emotionalresponse and ensuing choice behavior. Evidence that high-level emo-tions such as regret depend on a specific neuroanatomical substratecomes from a study showing that the normal expression of thiscognitively based emotion depends on the integrity of OFC19. Speci-fically, patients with selective lesions to anterior medial OFC do notexperience regret and, unlike healthy controls, are unable to adjust theirbehavior to avoid regret-inducing situations. In order to define pre-cisely the conditions under which the OFC and related areas areengaged by the experience and anticipation of regret and to determinehow the latter experience influences learning, we used functionalmagnetic resonance imaging (fMRI) while subjects performed agambling task. We demonstrated that activity in the OFC mediatesthe experience of adaptive emotions such as regret. Moreover, learningto anticipate this emotion during choice reactivated the OFC inconjunction with the amygdala.

On each trial, the subject viewed two gambles where differentprobabilities involving financial gain or loss were represented by therelative size of colored sectors of a circle (Fig. 1). There were two kindsof trials, each with a different type of feedback information indicatingfinancial gain or loss for the subject. In the ‘partial feedback’ condition,only the outcome of the selected gamble was shown, whereas in the‘complete feedback’ condition, the outcome of both the selected andunselected gambles were available to the subject. Complete feedback


1Neuropsychology Group, Institut des Sciences Cognitives, Centre National de la Recherche Scientifique, 67 Boulevard Pinel 69675, Bron, France. 2Wellcome Departmentof Imaging Neuroscience, 12 Queen Square, London, WC1N 3BG, UK. Correspondence should be addressed to R.J.D. ([email protected]) or A.S. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

trials enabled subjects to judge not only the financial consequence oftheir decision but also the outcome they would have achieved had theyselected the unselected option.

As a sense of responsibility is critical to the experience of regret, wecompared subjects’ responses when they had a choice (the ‘choose’condition) with their responses when they had no choice, but ratherfollowed a computer-selected choice (the ‘follow’ condition), thusremoving any feeling of responsibility.

RESULTS

Ventral striatum response to wins and losses

Agency is reflected in cognitive and physiological engagement ofsubjects. During task performance, subjects’ physiological responses(heart rate) were significantly higher in ‘choose’ trials than in ‘follow’trials (P ¼ 0.001). During fMRI, the processing of outcome wasmodulated as a function of whether outcome (wins or losses) wasevaluated in the context of a ‘choose’ or ‘follow’ trial (that is, whethersubjects were agents). We found activation of anterior ventral striatumduring wins and a relative deactivation during losses solely in ‘choose’trials (Fig. 2), highlighting the dependency of reward-related signalingin this region on instrumentality20. This pattern of activity in ventralstriatum for ‘choose’ trials is consistent with a ‘reward prediction error’response21 insofar as in ‘follow’ trials (where choice was computer-selected, meaning the subject had no agency) there is no need for

prediction22. In other words, this area processes mismatches betweenpredicted and actual outcome and is activated when an outcome isbetter than expected and relatively deactivated in the alternative case. Inlight of this agency effect, we restricted our subsequent analyses ofoutcome-related activity to ‘choose’ trials alone.

Disappointment and regret

The psychological and behavioral impact of outcome (wins and losses)was influenced by the amount of feedback information provided tosubjects. Disappointment arises when, on a selected gamble, thealternative outcome is more positive than an experienced outcome.The magnitude of disappointment (that is, the discrepancy between the‘unobtained outcome’ and actual outcome of the selected gamblecorrelated with enhanced activity in middle temporal gyrus and dorsalbrainstem (including periaqueductal gray matter), a region implicatedin processing aversive signals such as pain23 (Table 1a).

Regret represents an emotion based on counterfactual processing,but it differs from disappointment in its abstract point of reference.

At outcome

2.0 s

2.0 s

4.0 s

4.0 s1.0 s

At choiceAnalysis of imaging data

CC

CF

'Choose'

'Choose'

'Follow'

'Follow'

Completefeedback

feedback

PC

PF

Partial

1.0 s 4.0 s

4.0 s

2.0 s

2.0 s

12 + 1 (dummy)trials per run

16 runs divided intwo sections: A & B

200 –200 –5050 200 –200 –5050 200 –200 –5050

200 –200 –5050 200 –200 –5050 200 –200 –5050

Trial 12Trial 11Trial 2Trial 1 Dummy

PC CC PF CF CC PF PC CF PF CC CF PC CF PF PC CC

Ventral striatum0.80.60.40.2

0–0.2–0.4–0.6–0.8–1.0 Win Loss Win Loss

Choose Follow

Rel

ativ

e %

cha

nge

inB

OLD

sig

nal

VStr VStr

Figure 1 Experimental design. On each trial, the

subject viewed two gambles where different

probabilities of financial gain or loss were

represented by the relative size of colored sectors

of a circle. The preferred gamble was indicated by

the subject by means of a left or right button

press. Once selected, the chosen gamble was

highlighted on the screen by a green square. Arotating arrow then appeared in the center of the

gamble circle, stopping after 4 s. The outcome of

the selected gamble, indicated by the resting

position of the arrow, resulted in financial gain or

loss for the subject. Half of the trials were ‘choose’

trials; in half of those, only the outcome of the

selected gamble was given to the subject (‘partial

feedback choose’, PC). In the other half, the

outcome of both selected and unselected gambles

were available (‘complete feedback choose’, CC).

An equal number of trials were ‘follow’ trials, in

which the subject was informed that the computer

would randomly choose one of the two gambles. A

green square appeared behind one of the two

gambles, and the subject had to press a button on

the corresponding side. Follow trials were likewise

divided into complete feedback (‘complete

feedback follow’, CF) and partial feedback (‘partial

feedback follow’, PF) trials.

Figure 2 Activity at outcome is related to win and loss. Activity within the

striatum, encompassing regions of ventral striatum, discriminated between

financial gain and loss at trial outcome. This effect, however, was significant

only for ‘choose’ trials, in which the subject was responsible for the choice

(that is, when the subject rather than the computer selected between two

gambles). Group data (thresholded at P o 0.001, uncorrected) is plotted onsagittal and coronal sections of a normalized canonical template brain.

Striatal activations (VStr) were centered on MNI coordinates (8, 18, 0),

(6, 18, 2) and (12, 24, 8). The bottom panel plots the average parameter

estimates (± s.e.m.) for relative difference in BOLD activity at outcome for

wins and losses in ‘choose’ and ‘follow’ trials.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Regret arises from a discrepancy between the actual outcome and anoutcome that would have pertained had an alternative choice beentaken. In our experiment, regret is represented as the difference betweenthe outcome of the unselected gamble and the outcome of the selectedgamble; hence, it occurs only on complete feedback trials. Across bothwin and loss trials, the magnitude of the difference in unselected andselected outcomes correlated with enhanced activity within anteriorcingulate, putamen, inferior parietal lobule (which is implicated inprocessing number comparison24) and lateral OFC (Table 1b).

Effect of the unobtained outcome

Activity in medial OFC clearly discriminated between trials resulting ina win or a loss, but only in the complete feedback condition (Fig. 3). Wemodeled the corresponding outcome of the nonselected gamble as aparametric regressor for each outcome (win or loss); in partial feedbacktrials, we used the unobtained outcome of the chosen gamble as aparametric regressor. By adopting such a procedure, we could dis-sociate comparative assessments underlying counterfactual thinking.The counterfactual process in the complete feedback condition inlosing trials resulted in greater activation of bilateral regions of medialOFC, and greater deactivation for winning trials. Notably, in thisanalysis, losses in the complete feedback condition resulted in theemotion of regret (with the exception of the case in which the subjectlost �50 and felt relief at forgoing an outcome of �200), whereas winsresulted in relief (with the exception of the case in which the subjectwon 50 and felt regret for a forgone outcome of +200; Fig. 3).

Activity in medial OFC, extending from subgenual cingulate, corre-lated with the degree of regret (that is, the difference between theoutcome of the unchosen gamble and the obtained outcome) in the‘choose’/complete feedback condition (referred to as the ‘completechoose’ condition). These data are plotted in Figure 4a, showing thechange in magnitude of the fitted response in OFC relative to the degreeof regret and relief experienced. In addition, anterior hippocampusactivity correlated with regret. From this analysis, we extracted data onbrain activity in response to the obtained outcomes of �50 and +50 asa function of the outcome of the unselected gamble (�200 and +200)in the complete feedback condition (Fig. 4b). Notably, activity in theOFC, dorsal anterior cingulate cortex (ACC) and anterior hippo-campus discriminated between the two unobtained outcomes.More specifically, these areas were activated when the actual outcomewas compared with a more advantageous forgone outcome (+200),

leading to regret, and are relatively deactivated when the sameactual outcome is compared with a less advantageous alternative(�200), leading to relief. Self-reported emotional ratings (in thepractice session before scanning) were consistent with this result inthat the ratings did not simply reflect wins or losses on the selectedgambles but rather were strongly influenced by the provision ofinformation regarding outcome of the alternative nonselected gamble(Fig. 4c). These results suggest that there are three main regions thatcontribute to the experience of regret: dorsal anterior cingulate cortex,medial OFC and anterior hippocampus.

Choice behavior and brain activity

During the fMRI experiment, we predetermined the presentation ofindividual gambles and the outcome of each gamble such that subjectswere exposed, in both ‘choose’ and ‘follow’ trials, to a range of positiveand negative outcomes and, for complete feedback trials, were exposedto alternative outcomes that provided either positive and negativevalues. However, for ‘choose’ trials, the selection of individual gambleswas determined by the subject, and the individual behavior in thispattern of selection provided data on subject sensitivity to wins andlosses and counterfactual information.

It is indeed counterfactual thinking between the obtained and theunobtained outcomes that modulates the experience of regret anddisappointment. When the subject obtains an outcome that is lowerthan expected, he or she might feel disappointment. The greater thedifference between the expected and the obtained outcome, the moreintense is this negative feeling. Thus, the subject can avoid futuredisappointment by choosing a gamble that minimizes a differencebetween lowest and highest outcome, weighted by the probability of theworst possible outcome.

We tested a model of choice (see Methods) using the data from thepartial-feedback/‘choose’ condition (the ‘partial choose’ condition)that incorporated the effects of anticipating disappointment in addi-tion to the maximization of the expected values. The result from aregression analysis (Table 2a) shows that subjects chose maximizing

Table 1 Activity at outcome

(a) Activity related to the comparison between the unobtained and the obtained

outcome of the selected gamble (‘partial choose’ condition)

Location Side Coordinates Z-score

Midbrain (periaqueductal gray region) – 2, �34, �20 3.84

Precentral gyrus S2 L �60, 4, 8 3.74

Subcallosal gyrus L �4, 0, �8 3.61

Middle temporal gyrus L �42, 0, �20 3.54

(b) Activity related to the comparison between the outcome of the unselected gamble

and the obtained outcome of the selected gamble (‘complete choose’ condition)

Anterior cingulate cortex – 10, 24, 34 4.68

Putamen L –14, 0, 6 3.92

Lateral OFC R 42, 42, –18 3.77

Inferior parietal lobule R 54, –50, 36 3.67

Win Loss Win LossPartialComplete

–0.6

–0.4

–0.2

0

0.2

0.4

0.6

Rel

ativ

e %

cha

nge

inB

OLD

sig

nal

Orbitofrontal cortex

OFC

Figure 3 The effect of the unobtained outcome: counterfactual processing of

value. Orbitofrontal cortex (OFC) activity for the comparison between the

outcome of the selected gamble and the alternative outcome in the ‘complete

choose’ versus ‘partial choose’ conditions (right). For the ‘complete choose’

condition we modeled the corresponding outcome of the nonselected gamble

as a parametric regressor for each actual outcome (win or loss), whereas in

partial feedback trials, we used the unobtained outcome of the chosen

gamble as a parametric regressor. The counterfactual process between losses

(or wins) and any forgone outcome in the complete condition resulted in

much greater activation (or deactivation) of bilateral regions of medial OFC.

Error bars indicate s.e.m. In the partial feedback condition, we observed only

a relative deactivation of the OFC when actual losses were compared withunobtained outcome in the chosen gamble. Group data (thresholded at P o0.001, uncorrected) is plotted on an axial section of a normalized canonical

template brain (left panel). Peak orbitofrontal activity occurred at MNI

coordinates (�8, 32, �14).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

expected values (P o 0.001) and minimizing future disappointment(P ¼ 0.001). Notably, brain activity in motor and premotor cortex,anterior cingulate cortex and superior parietal cortex related to a pre-ceding choice of maximal expected value (Table 3a). Choice behaviorthat deviates from the maximization of expected values, motivated byavoidance of future disappointment, produced activity in the substan-tia nigra, a dopaminergic midbrain area involved in anticipatoryreward (reward prediction, refs. 25,26; Supplementary Figure 1).This activity is related to the activity at outcome observed whensubjects evaluated wins and loss in terms of prediction error (Fig. 2).

The feeling of regret depends on the comparison between theobtained and the unobtained outcome across the two gambles, acomparison possible only in the complete feedback condition. Thegreater the difference between the outcome of the unchosen gamble

and the obtained outcome, the more intense the feeling of regret. Weparameterized regret as the absolute value of the difference between thelowest and the highest outcome across gambles. Thus, the subject canavoid future regret by choosing a gamble that minimizes this difference(see equation 3 in Methods). In our analysis of subjects’ behavior in thecomplete choose condition, we considered anticipated disappointment,anticipated regret and maximization of the expected values as the maincomponents of subjects’ choices. Results from a regression analysis

–50 +50–50 +50–50 +50–50 +50

Obtained outcomeObtained outcomeObtained outcomeObtained outcome

1

3

5

7

9

Very bad

Very good

Em

otio

nal e

valu

atio

n

–200

+200

–200

+200

Unobtainedoutcome

Unobtainedoutcome

–0.5

0.0

0.5

Rel

ativ

e %

cha

nge

inB

OLD

sig

nal

Rel

ativ

e %

cha

nge

inB

OLD

sig

nal

OFC ACC Hippocampus

11 22 33 44Regret →← Relief

–0.4

–0.2

0.0

0.2

0.4

0.6

0.8Data across subjects

OFC

a b c

Figure 4 Regret and relief. (a) Change in magnitude of fitted response in the medial OFC relative to the degree of regret or relief experienced in the CC

condition. Medial OFC activity (�10, 30, �12) correlates with the level of regret experienced. Regret and relief are measured as the positive and negative

differences, respectively, between the outcome of the unselected gamble and the actual outcome. The relative change in BOLD signal between unobtained

and actual outcomes is plotted for each of the four levels of regret (numbered 1–4, from lowest to highest regret). The first value in each pair represents the

unobtained outcome and the second value represents the actual outcome: 1: [50, �50]; 2: [200, 50], [�50, �200]; 3: [200, �50], [50, �200]; and 4:

[200, �200]. The four levels of relief are the result of the following comparisons (numbered 1–4, as for regret): 1: [�50, 50]; 2: [�200, �50], [50, 200]; 3:

[�200, 50], [�50, 200]; and 4: [�200, 200]. (b) Left plot: medial OFC activity; center: anterior cingulate cortex activity (coordinates 8, 32, 24); and right:

hippocampal activity (30, �10, �12) in processing the comparison between �50 and +50 obtained with a forgone outcome of �200 or + 200 in the CC

condition. Error bars, s.e.m. (c) Mean emotional evaluation (± s.e.m.) measured in the practice session before scanning for two obtained outcomes (�50 or 50)

as a function of the outcome of the unchosen gamble (�200 or 200) in the complete feedback condition.

Table 2 Regression analysis (panel logit procedure with individual

random effect)

(a) Subjects’ choice behavior as a function of anticipated disappointment (d) and

maximization of the expected value (e) in the ‘partial choose’ condition.

Variable name Coefficient Standard error Z P

Constant 0.22014 0.13002 1.69 0.09

d 0.00259 0.0008 3.23 0.001

e 0.01464 0.00106 13.79 0.000

Number of subjects ¼ 15; number of observations ¼ 720. Log likelihood ¼ �331.4631,Wald w2(2) ¼ 193.71, Prob 4 w2 ¼ 0.000. The dependent variable ‘choice’ is equal to 1 ifsubject chose gamble 1 and equal to 0 if subject chose gamble 2.

(b) Subjects’ choice behavior as a function of anticipated disappointment (d),

maximization of the expected value (e) and anticipated regret (r) in the ‘complete

choose’ condition.

Variable name Coefficient Standard error Z P

Constant 0.06958 0.09706 0.72 0.473

d 0.00211 0.00137 1.62 0.106

r 0.00463 0.00114 4.03 0.000

e 0.00726 0.00184 3.94 0.000

Number of subjects ¼ 15; number of observations ¼ 720. Log likelihood ¼ �331.1425;Wald w2(3) ¼ 205.82, Prob 4 w2 ¼ 0.000. The dependent variable ‘choice’ is equal to 1if subject chose gamble 1 and is equal to 0 if subject chose gamble 2.

Table 3 Activity at choice

(a) Activity preceding choice of maximal expected values in the ‘partial choose’

condition

Location Side Coordinates Z-score

Motor cingulate cortex L �12, �10, 46 4.85

Premotor cortex L �34, �20, 62 4.08

Motor/posterior cingulate cortex R 14, �28, 54 4.43

Genual anterior cingulate cortex R 18, 36, 8 4.04

Superior parietal cortex R 42, �52, 52 3.73

(b) Activity preceding choice of maximal expected values in the ‘complete choose’

condition

Lateral occipital cortex L �38, �76, �6 4.43

Lateral geniculate nucleus L �20, �31, 6 3.89

Somatomotor cortex L �28, �28, �52 3.98

Superior parietal cortex R 24, �56, 50 3.96

Parahippocampus L 32, 2, �26 3.29

Mid cingulate cortex – �6, �4, 32 4.36

(c) Influence of cumulative experience of regret (CR) on choice-related activity

Somatomotor cortex R 68, �12, 20 5.19

Inferior parietal lobule L �38, �42, 40 4.8

Medial OFC L �10, 40, �24 4.24

Amygdala L �8, �4, �24 4.21

(d) Influence of immediately prior experience of regret (t – 1) on choice

Dorsolateral prefrontal cortex R 46, 28, 38 4.99

Lateral OFC R 42, 26, �16 4.72

Inferior parietal lobule R 54, �58, 48 4.62


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

(panel logit procedure with individual random effect, Table 2b) showthat subjects chose maximizing expected values (P o 0.001) andminimizing future regret (P o 0.001).

Anticipating disappointment (as defined in equation 2 in Methods)would correspond to risk avoidance. The absence of this behavior (asshown in Table 2b for the variable d, P¼ 0.11) in the complete choosecondition indicates a hierarchical relationship between risk and regret.Indeed, the subjects chose minimizing regret independently of the riskcomponent in their choice responses.

How the experience of regret affects decision making

The experience of regret has a powerful influence on subsequentbehavioral choice, leading to a pattern of behavior that can becharacterized as regret-aversive. In our experiment, this was manifestin two distinct ways. First, we observed a behavioral bias in subjects’choices over the course of the gambling task away from choices thatmight engender an emotion of regret (Table 2b). Second, regret aversionwas also evident in a bias away from choices that had previously led tonegative outcomes, on the basis of subjects’ own cumulative experienceof regret. The proportion of regret-avoiding choices increased over timewith the cumulative effect of the experience of regret (Fig. 5a).

On the basis of the above data, we determined how these behavioralbiases were expressed in patterns of neural activity at the time ofchoice behavior. We found enhanced activity (during the epochbetween trial onset and subject response in the complete choosecondition) in the dorsal anterior cingulate cortex and in the substantianigra, when subjects chose minimizing regret over maximizingexpected values (Supplementary Figure 2). Both anterior cingulateand midbrain activities are related to a reward anticipation process (interms of error prediction signal)25–28. Activity related to a precedingchoice of maximal expected values in the ‘complete choose’ condition isshown in Table 3b.

We next assessed the effects of cumulative history using a reinforce-ment-learning model based on past emotional experience (see Meth-ods). For cumulative regret experience, we observed modulation of

choice-related activity in the medial OFC,right somatomotor, inferior parietal lobuleand left amygdala (Fig. 5b and Table 3c).Notably, this expression of cumulative regretat the time of decision making involved simi-lar anatomical regions (medial OFC and ante-rior medial temporal lobe) as that elicited byregret at the time of outcome feedback. Thissuggests that the experience-dependent influ-ence of regret on decision making may besupported by reactivation of processes med-iating regret as a reactive emotion.

The more immediate experience of regretin the preceding trial also influenced choice-related activity, enhancing responses in rightdorsolateral prefrontal cortex, particularlyaround the border between middle and infer-ior frontal gyri, perhaps representing an influ-ence of immediate regret on self-monitoringat decision making. We also found thatenhanced activity during choice selection inright inferior parietal lobule and right caudo-lateral OFC correlated similarly with the mag-nitude of regret experienced in the precedingtrial (Table 3d and Fig. 5c).

DISCUSSION

Regret is a complex emotion based on a counterfactual process thatjuxtaposes the outcome of choices we make with a better outcome for arejected alternative. We show that activity in response to this negativeemotion is distinct from activity seen for mere outcome evaluation. Inour brain imaging data, the influence of personal responsibility on theprocessing of outcome was evident in contrasting outcome-relatedactivity for ‘choose’ trials (where the subject selected which gamble to‘play’) with ‘follow’ trials (where the ‘choice’ was computer-selected).In accordance with psychological theory5,11, we also find a neuroana-tomical dissociation of regret versus disappointment. Thus, outcomeevaluation is influenced by the level of responsibility in the process ofchoice (agency) and by the available information regarding alternativeoutcomes (complete or partial feedback). The level of regret, calculatedin terms of the magnitude of the difference between the forgoneoutcome and the obtained outcome, was strongly correlated withactivity in the medial OFC.

In a number of studies (including tasks in which outcome is notdependent on operant action) medial OFC activity reflects rewardattainment17,18,29,30. This has been interpreted as suggesting thatmedial OFC may support positive emotions (and that lateral OFCmay support emotions with negative valence). Nevertheless, otherneuroimaging studies highlight a more complex role in reinforcementrepresentations that is also suggested by lesion data. Thus, enhance-ment of medial OFC activity reflects devaluation in extinctionof conditioned aversive stimuli and inflation of aversive stimuli12.Significant activity with monetary gain and loss has been reported inboth medial and lateral OFC31, whereas monetary gain in a probabil-istic reversal task has been associated with activity in both medial andlateral OFC. Similarly, lesions of medial OFC do not impair processingof primary rewards but seem to interfere with relative reward dis-crimination that includes conditions involving prospective and coun-terfactual appraisal19,32. These findings point to a more complexrelationship in OFC than a simple medial-lateral specialization forreward or punishment. Our data would suggest that cognitive context,

DLPFCDLPFCIPL

OFC

OFCAmg

FinalMiddleInitial0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Pro

port

ion

Anticipated regret

Blocks

a b

c

Figure 5 Activity at choice: learning from the experience of regret. (a) Proportion of choice (± s.e.m.)

related to anticipated regret in ‘complete choose’ trials. Anticipated regret increased over time as the

experiment proceeded. (b) Activity at choice reflecting cumulative regret. We found activity in the medial

left amygdala (Amg; coordinates �8 to �16, �4, �25) and medial OFC (�10, 40, �24). Group data is

plotted on coronal and axial slices of a template brain in normal space at a threshold of P o 0.005,

uncorrected (for illustrative purposes in this figure only). (c) Activity reflecting prior regret at choice.

Individual subject analytic designs modeled the parametric modulation of activity during the epoch

between trial onset and subject response. Experience of regret in the preceding trial profoundly

influenced choice-related activity, enhancing responses in right dorsolateral prefrontal (DLPFC),

right lateral OFC, and inferior parietal lobule (IPL; Table 3c). Group activity is plotted at

P o 0.001, uncorrected.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

exemplified by counterfactual thinking in relation to states of the world,exerts a modulatory influence on OFC activation to reward andpunishment. On this basis, our neuroimaging data complement andconfirm results from lesion studies19 that assign a critical role to themedial OFC in experiencing and anticipating the emotion of regret.

To our knowledge, this is the first demonstration of involvement ofmedial OFC and amygdala in choice behavior that reflects regretavoidance. We suggest that, as a result of cumulative experience ofregret, inputs from these regions to decision-making processes providean updated representation of value that incorporates a weighting of therelative emotional value of different options for choice12,33. In addition,there is enhanced lateral OFC activity in the choice condition thatimmediately follows the experience of regret, perhaps reflecting cogni-tive processes associated with a need (across gamble comparisons) toavoid future regrettable outcomes. Such processes are analogous tothose normally subsumed under procedures involving reversal learn-ing, where subjects need to change behavioral strategies that are nolonger advantageous34. We also found enhanced activity in the parietalcortex during choice selection, in accordance with findings in animalstudies related to action desirability35. Thus, the experience of regrethas a major impact on the process of choice that is expressed at twolevels, with a net result of biasing subjects to forgo choices that mightlead to future experience of this highly negative emotion.

Regret is also an emotion based on a declarative cognitive processthat requires an ability to assess the consequences of our actions withinconscious awareness. It is not surprising that there is involvement ofbrain structures such as the hippocampus and dorsal anterior cingulateregions that are critical to declarative memory (indicating what weneed to remember) and cognitive-induced arousal responses, respec-tively36,37. Dorsal anterior cingulate activity is related to the experienceof regret at outcome and to regret-aversive decision making at choice.In the context of decision making, regret involves an appraisal ofpotential outcomes and is likely to evoke behavioral adjustments onsubsequent trials. Indeed, choice behavior in our complete task involveseither maximizing expected value in the face of potential regret, orminimizing regret at a cost to expected value. Thus, the decisionprocess necessarily involves a degree of conflict. There is substantialneuroimaging literature that highlights a role for dorsal anteriorcingulate and pre-supplementary motor area in response monitoringand error detection38–41. Notably, dorsal anterior cingulate cortexactivity may also mediate an attentional focus on subjective emotionalstates42,43 and the cognitive and emotional processing engendered instates of autonomic arousal44,45. Our findings would conform to aunified model of error detection, response conflict and attendantemotional arousal whereby anterior cingulate activation arises duringappraisal of conflicts (between maximizing expected value and mini-mizing regret) in decision making that would include a signal forpotential behavioral adjustment.

However, the critical finding in our study concerns the role of OFC,which we suggest integrates cognitive and emotional mechanisms aftera declarative process in which distinct counterfactual processes engen-der a high-level emotion of regret. Our data suggest a mechanism bywhich comparing a choice with its alternative outcome, along with theassociated feeling of responsibility, promotes behavioral flexibility andexploratory strategies in dynamic environments so as to minimize thelikelihood of emotionally negative outcomes.

METHODSSubjects. Fifteen healthy right-handed subjects were recruited to take part in a

study at the Wellcome Department of Imaging Neuroscience. These volunteers

gave fully informed consent for the project which was approved by the Joint

Ethics Committee of the National Hospitals and Institute of Neurology,

London. Each participant was screened to exclude medication and conditions

including psychological or physical illness or history of head injury. Mean age

of participants was 23.3 years.

Experimental design and task. Each participant underwent fMRI scanning

while performing a total of 192 trials of the experimental task illustrated in

Figure 1. During scanning, the subject viewed a projection of a computer

screen and made a two-choice button-press response with the right index or

middle finger. The task was adapted from refs. 5 and 19 and involved stimuli

resembling ‘wheels of fortune’. On each trial, the subject viewed two gambles

where different probabilities of financial gain or loss were represented by the

relative size of colored sectors of a circle.

Half the trials were designated ‘choose’ trials in which the preferred gamble

was indicated by the subject by means of a left or right button press. Once

selected, the chosen gamble was highlighted on the screen by a green square,

and a rotating arrow then appeared in the center of the gamble circle, stopping

after 4 s. The outcome of the selected gamble, indicated by the resting position

of the arrow, resulted in financial gain or loss. There were two kinds of choose

trials: in 48 of the 96 ‘choose’ trials, the outcome (and the spinning arrow)

appeared for the selected gamble alone (‘partial feedback’). In the other 48

‘choose’ trials, the spinning arrow and outcome of both the selected and

unselected gambles were visible to the subject (‘complete feedback’). Complete

feedback trials enabled the subject to judge not only the financial consequence

of their decision, but also the outcome had they selected the other option.

An equal number of trials were designated ‘follow’ trials, which were also

divided into complete and partial feedback trials. In ‘follow’ trials, the subject

was informed that the computer would randomly choose one of the two

gambles. A green square appeared behind one of the two gambles, and the

subject was required to press the corresponding button. The rotating arrow

then appeared, as in the ‘choose’ condition, in the selected gamble circles (or in

both circles, for the complete feedback condition), stopping to indicate the

outcome. The ‘follow’ condition was a control condition, as it eliminated the

responsibility element of decision making but still produced outcomes repre-

senting financial gains and losses.

Partial and complete feedback and ‘choose’ and ‘follow’ conditions were

presented in pseudo-randomized blocks of 12 stimuli. Null events were also in-

cluded in each block to allow estimation of low-level baseline brain activity and

to desynchronize timings of event types. Behavioral responses were logged by

means of a desktop computer running Cogent software on a Matlab platform.

Incentive procedure. Each subject was told that the outcome of both ‘choose’

and ‘follow’ trials would result in financial gain or loss. The subjects were

informed before starting the experiment that they would be paid a show-up fee

of 20 GBP and an additional amount that would depend on their performance

in terms of cumulative points earned during the experiment, in both ‘choose’

and ‘follow’ conditions: 0 GBP, 5 GBP and 10 GBP corresponded to low,

medium and high cumulative earnings, respectively. No quantitative references

about the ranges of the cumulative earnings were provided. The subjects were

informed about their earnings and paid in cash outside the fMRI scanner. No

information about their cumulative outcomes was provided during the

experiment. For ethical reasons, we were not able to pay different amounts

to subjects; thus, in practice we programmed the computer to always assign 5

extra GBP. Every subject ended up with final earnings of 25 GBP.

Parameter structure. Each individual gamble presented paired combinations of

200, 50, �50 and �200 points and represented one of three levels of outcome

probability (0.2, 0.5 and 0.8). Displayed and actual probabilities were identical.

The two gambles always differed in their expected values (that is, probability

times the outcome of each alternative choice) and in the value of their actual

outcomes. There were six possible outcome pairs (outcome obtained in the

chosen gamble and outcome of the unselected gamble, or vice versa): (i) �50

and 200, (ii) 50 and 200, (iii) �50 and �200, (iv) 50 and �200, (v) 50 and �50

and (vi) �200 and 200. The set of pairs of gambles was the same for each

condition. The order of presentation was pseudo-random and differed for each

condition. In the ‘follow’ condition, the predetermined pattern of choices did

not resemble any particular choice strategy. The set of trials and the computer’s

pattern of choice in ‘follow’ conditions can be found in Supplementary Table 1.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Emotional rating. Before scanning, the subject read a standard description of

the task and was familiarized with the computerized task events. The subject

then performed a practice session covering the four trial types (‘complete

choose’, ‘partial choose’, ‘complete follow’ and ‘partial follow’) and were asked

to rate how they felt about each outcome on a nine-point scale (where 9

signaled a very good outcome and 1 very bad).

Physiological monitoring. Blood volume pressure (BVP) was recorded with a

Nonin 8600 Pulse Oximeter (Nonin Medical) and sampled at 300 Hz at the

same time as task performance and fMRI data acquisition. BVP signal and

fMRI scanning pulses were coregistered by means of an analog-to-digital

converter (CED1901) and Spike 3.3 software (CED). Heart rate was estimated

from BVP inter-beat intervals and resampled at 2 Hz for later analysis. We

considered physiological responses in each trial of two time windows of 3 s

each. The first time window (anticipatory) started when the stimulus was

presented at the beginning of each trial and included the choice (or follow)

event and the period of waiting for the outcome; the second time window

(feedback) corresponded to the feedback presentation.

Analysis of behavioral data. We tested (by regression analysis and the panel

logit procedure with individual random effect; Table 2) a model of choice that

incorporates the effects of anticipating disappointment and regret in addition

to maximization of expected values. x1 and y1 represent the highest and the

lowest outcome of gamble 1 (g1), and x2 and y2 represent the highest and the

lowest outcome of gamble 2 (g2). The probability of x1 is p and the probability

of y1 is 1 – p; the probability of x2 is q, and the probability of y2 is 1 – q. The

probability of choosing gamble 1 is

Prðg1itÞ ¼ 1 � Prðg2itÞ ¼ F½dit ; rit ; eit � ð1Þ

where i ¼ individual and t ¼ time. The function F[y] denotes the function

ey/(1 + ey). The variables d and r, as described in equations 2 and 3 indicate the

process of minimizing future disappointment and future regret, respectively;

e indicates the result of maximizing expected values.

d ¼ ðjy2 � x2jð1 � qÞÞ � ðjy1 � x1jð1 � pÞÞ ð2Þ

r ¼ jy2 � x1j � jy1 � x2j ð3Þ

e ¼ EVðg1Þ � EVðg2Þ ¼ ðpx1+ð1 � pÞy1Þ � ðqx2+ð1 � qÞy2Þ ð4Þ

Subjects would choose g1, minimizing disappointment (equation 2), if the

difference in absolute value between the lowest and the highest possible

outcome in g2, weighted by the probability of the lowest outcome, were larger

than the same weighted difference in g1. The process of anticipating regret is

described by the minimization of the difference between the lowest and the

highest outcome across gambles (equation 3). Finally, subjects would choose g1

if its expected value (EV) is higher than that of g2 (equation 4). A restricted

version of the model, where we considered only the effect of anticipated

disappointment (d) and maximization of expected values (e), was tested with

the data from the ‘partial choose’ condition (Table 2a).

Functional imaging data: acquisition, pre-processing and analysis. Subjects

were scanned at 3 T (Siemens Allegra) performing the experimental task over

two counterbalanced sessions. T2-weighted echoplanar images, optimized for

blood oxygenation level–dependent (BOLD) contrast, were acquired (TE:

30 ms, TRvol: 2.86 s, 44 slices angled at 301 in anterior-posterior axis). A

preparation pulse (duration 1 ms, amplitude �2 mT/m) was used in the slice

selection direction to compensate for through-plane susceptibility gradients for

enhancement of imaging of orbitofrontal and medial temporal lobe regions46.

The efficient imaging of these regions (cross-hairs) is illustrated in Supple-

mentary Figure 3, depicting smoothed normalized EPI images and corre-

sponding locations in a normalized structural template image.

Image pre-processing and subsequent analyses were done using statistical

parametric mapping (SPM2; http://www.fil.ion.ucl.uk/~spm/SPM2.html) on a

Matlab platform. Images were initially realigned and unwarped, correcting for

motion artifacts. Differences in the timing of image slices across each individual

volume were corrected, and each volume was transformed into standard

stereotaxic space and smoothed with a Gaussian filter (full-width half-max-

imum 8 mm). Voxel-wide differences in BOLD contrast within the smoothed

normalized images resulting from the different task conditions and trial types

were examined using SPM. Standard neuroimaging methods using the

general linear model were used with the first level (individual subject

analyses) providing contrasts for group effects analyzed at the second level.

All individual analyses modeled the period of stimulus presentation

up to choice selection as a mini-epoch. Activity at outcome was likewise

modeled as a mini-epoch, segregating complete and partial feedback trials

for ‘choose’ and ‘follow’ conditions. In the fMRI data analysis, the outcomes

of each trial were modeled as 3-s epochs beginning at 1 s before the

outcome display stopped (that is, when the spinning arrow began to slow,

enabling prediction of its end point) and continuing for a further 2 s

while the trial outcome was displayed (and processed). Choice-related neural

activity at the time of choice was studied during the epoch between trial onset

and subject response. To model the hemodynamic lag of the BOLD response

relative to underlying evoked neural activity, regressors for task effects were

obtained by convolving these mini-epochs with a canonical hemodynamic

response function.

We did not jitter timings within trials to discriminate overlapping ‘raw’

evoked hemodynamic BOLD responses related to the choice and outcome

phases (although some temporal variability arose from the time it took the

subject to select between gambles on ‘choose’ conditions). Thus, the temporal

proximity (and lack of jittering) between the decision-making and outcome

components of each trial resulted in overlapping of BOLD responses between

these different phases. However, our principal analyses focused on modulation

of outcome-related activity by parameters not embodied within the decision or

anticipation phases (regret, disappointment, win/loss). Furthermore, contrasts

(Student’s t-tests) within our regression analyses identify activity that is not

otherwise accountable for in the analytic design. Thus, the impact of shared

variance in activity between choice and parametric modulators of outcome is

minimized or excluded in results from first- and second-level analyses by virtue

of this orthogonalization with analyses modeling decision and outcome epochs

independently of their parametric modulation.

Three sets of analyses were performed. In the first, outcome trials were

partitioned according to whether the subject experienced a loss or a win and

additional regressors parametrically modeling the degree of disappointment

and, for complete-feedback trials, regret experienced at the time of outcome. In

two further analyses, activity for win and loss trials were not modeled

separately, and parametrical regressors of activity at choice modeled experience

of regret on the preceding trial and mean regret experienced over the course of

the task prior to the choice, respectively. For group analysis of outcome-related

activity, second-level analyses of contrast for regret and disappointment for

different trial types were computed as ANOVAs with sphericity correction for

repeated measures. Post-hoc exploration of individual data is also reported to

illustrate specific effects as a function of different trial types. For choice-related

activity, contrasts relating to prior experience of regret were modeled for

complete and partial trials in a second-level ANOVA, taking into account group

effects of prior regret experienced. Adjusted activity represents BOLD signal

changes proportionally adjusted for the analytic model. Although general

threshold significance was set at P o 0.05, corrected, we tabulate group effects

at P o 0.001, uncorrected, to highlight regions of interest. For illustrative

purposes, we show Figure 5b at P o 0.005, uncorrected.

Modeling cumulative regret. The regressor to test activity at choice that

reflected cumulative regret (Fig. 5b and Table 3c) represented the difference

between missed payoff and payoff realized from past choice in the complete

feedback condition (‘complete choose’ condition (CC)) over time, that is,

CRt ¼ (Aunobtained, t – 1 – Aobtained, t – 1), where CR is cumulative regret, t is trial,

Aobtained is the average payoff realized and Aunobtained is the average payoff of the

unselected gambles.


ACKNOWLEDGMENTSThis work was supported by grants from the Human Frontier Science Program(RGP 56/2005), the Action Concertee Incitative, Systemes Complexes from theCentre National de la Recherche Scientifique to A.S. and G.C., the Coordenacao


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

de Aperfeicoamento de Pessoal de Nıvel Superior to M.J., a Wellcome TrustProgramme Grant to R.J.D. and a Wellcome Senior Fellowship in Clinical Scienceto H.D.C.




1. Roese, N.J. & Olson, J.M. What Might Have Been: The Social Psychology of Counter-factual Thinking (Erlbaum, Mahwah, New Jersey, USA, 1995).

2. Byrne, R.M.J. Mental models and counterfactual thinking. Trends Cogn. Sci. 6, 426–445 (2002).

3. Kahneman, D. & Tversky, A. The psychology of preferences. Sci. Am. 246, 136–142(1982).

4. Kahneman, D. & Miller, D. Norm theory: comparing reality to its alternatives. Psychol.Rev. 93, 136–153 (1986).

5. Mellers, B., Schwartz, A. & Ritov, I. Emotion-based choice. J. Exp. Psychol. Gen. 128,332–345 (1999).

6. Zeelenberg, M. & van Dijk, E. On the comparative nature of regret. in The Psychologyof Counterfactual Thinking (eds. Mandel, D., Hilton D. & Catelani, P.) 147–161(Routledge, London, 2005).

7. Bell, D.E. Regret in decision-making under uncertainty. Oper. Res. 30, 961–981(1982).

8. Loomes, G. & Sugden, R. Regret theory: an alternative theory of rational choice underuncertainty. Econ. J. 92, 805–824 (1982).

9. Bell, D.E. Disappointment in decision making under uncertainty. Oper. Res. 33, 1–27(1985).

10. Loomes, G. & Sugden, R. Disappointment and dynamic inconsistency in choice underuncertainty. Rev. Econ. Stud. 53, 271–282 (1986).

11. Zeelenberg, M. et al. Consequences of regret aversion: effects of expected feedback onrisky decision making. Organ. Behav. Hum. Decis. Process. 65, 148–158 (1996).

12. Gottfried, J.A. & Dolan, R.J. Human orbitofrontal cortex mediates extinction learningwhile accessing conditioned representations of value. Nat. Neurosci. 7, 1144–1152(2004).

13. Rolls, E.T. The orbitofrontal cortex and reward. Cereb. Cortex 10, 284–294 (2000).14. Tremblay, L. & Schultz, W. Relative reward preference in primate orbitofrontal cortex.

Nature 398, 704–708 (1999).15. Elliott, R. et al. Dissociable neural responses in human reward systems. J. Neurosci. 20,

6159–6165 (2000).16. Kringelbach, M. & Rolls, E. The functional neuroanatomy of the human orbitofrontal

cortex: evidence from neuroimaging and neuropsychology. Prog. Neurobiol. 72, 341–372 (2004).

17. Breiter, H.C., Ahron, I., Kahneman, D., Dale, A. & Shizgal, P. Functional imaging ofneural responses to expectancy and experience of monetary gains and losses. Neuron30, 619–639 (2001).

18. O’Doherty, J. et al. Abstract reward and punishment representations in the humanorbitofrontal cortex. Nat. Neurosci. 4, 95–102 (2001).

19. Camille, N. et al. The involvement of the orbitofrontal cortex in the experience of regret.Science 304, 1167–1170 (2004).

20. Corlett, P.R. et al. Prediction error during retrospective revaluation of causal associationsin humans: fMRI evidence in favor of an associative model of learning. Neuron 44,877–888 (2004).

21. Schultz, W., Dayan, P. & Montague, P.R. A neural substrate of prediction and reward.Science 275, 1593–1599 (1997).

22. Miceli, M. & Castelfranchi, C. The mind and the future: The (negative) power ofexpectations. Theory Psychol. 12, 335–366 (2002).

23. Peyron, R., Laurent, B. & Garcia-Larrea, L. Functional imaging of brain responses topain. A review and meta-analysis. Neurophysiol. Clin. 30, 263–288 (2000).

24. Gobel, S.M., Johansen-Berg, H., Behrens, T. & Rushworth, M.F.S. Response-selection-related parietal activation during number comparison. J. Cogn. Neurosci. 16, 1536–1551 (2004).

25. Tobler, P.N., Fiorillo, C.D. & Schultz, W. Adaptive coding of reward value by dopamineneurons. Science 307, 1642–1645 (2005).

26. Dreher, J.C., Kohn, P. & Berman, K.F. Neural coding of distinct statistical properties ofreward information in humans. Cereb. Cortex (in the press).

27. Holroyd, C.B. et al. Dorsal anterior cingulated cortex shows fMRIresponse to internal andexternal error signal. Nat. Neurosci. 7, 497–498 (2004).

28. Shidara, M. & Richmond, B.J. Anterior cingulated: single neuronal signal related todegree of reward expectancy. Science 296, 1709–1711 (2002).

29. Berns, G.S., McClure, S.M., Pagnoni, G. & Montague, P.R. Predictability modulateshuman brain response to reward. J. Neurosci. 21, 2793–2798 (2001).

30. Gottfried, J.A., O’Doherty, J. & Dolan, R.J. Encoding predictive reward value in humanamygdala and orbitofrontal cortex. Science 301, 1104–1107 (2003).

31. Elliott, R., Newman, J.L., Longe, O.A. & Deakin, J.F. Differential response patterns in thestriatum and orbitofrontal cortex to financial reward in humans: a parametric functionalmagnetic resonance imaging study. J. Neurosci. 23, 303–307 (2003).

32. Bechara, A., Tranel, D. & Damasio, H. Characterization of the decision-making deficit ofpatients with ventromedial prefrontal cortex lesions. Brain 123, 2189–2202 (2000).

33. Schoenbaum, G., Chiba, A.A. & Gallagher, M. Orbitofrontal cortex and basolateralamygdala encode experience outcomes during learning. Nat. Neurosci. 1, 155–159(1998).

34. Rolls, E.T., Hornak, J., Wade, D. & McGrath, J. Emotion-related learning in patients withsocial and emotional changes associated with frontal lobe damage. J. Neurol. Neurosurg.Psychiatry 57, 1518–1524 (1994).

35. Dorris, M.C. & Glimcher, P.W. Activity in posterior parietal cortex is correlated with therelative subjective desirability of action. Neuron 44, 365–378 (2004).

36. Bush, G., Phan, L. & Posner, M.I. Cognitive and emotional influences in anteriorcingulate cortex. Trends Cogn. Sci. 4, 215–222 (2000).

37. Critchley, H.D. The human cortex responds to an interoceptive challenge. Proc. Natl.Acad. Sci. USA 101, 6333–6334 (2004).

38. Carter, C.S., Botvinick, M.M. & Cohen, J.D. The contribution of the anterior cingulatecortex to executive processes in cognition. Rev. Neurosci. 10, 49–57 (1999).

39. Kiehl, K.A., Liddle, P.F. & Hopfinger, J.B. Error processing and the rostral anteriorcingulate: an event-related fMRI study. Psychophysiology 37, 216–223 (2000).

40. Garavan, H., Ross, T.J., Murphy, K., Roche, R.A. & Stein, E.A. Dissociable executivefunctions in the dynamic control of behavior: inhibition, error detection, and correction.Neuroimage 17, 1820–1829 (2002).

41. Kerns, J.G. et al. Anterior cingulate conflict monitoring and adjustments in control.Science 303, 1023–1026 (2004).

42. Lane, R.D. et al. Neural correlates of levels of emotional awareness. Evidence of aninteraction between emotion and attention in the anterior cingulate cortex. J. Cogn.Neurosci. 10, 525–535 (1998).

43. Phan, K.L., Liberzon, I., Welsh, R.C., Britton, J.C. & Taylor, S.F. Habituation of rostralanterior cingulate cortex to repeated emotionally salient pictures. Neuropsychopharma-cology 28, 1344–1350 (2003).

44. Critchley, H.D. et al. Activity in the human brain predicting differential heart rateresponses to emotional facial expressions. Neuroimage 24, 751–762 (2005).

45. Critchley, H.D. et al. Human cingulate cortex and autonomic control: convergingneuroimaging and clinical evidence. Brain 126, 2139–2152 (2003).

46. Deichmann, R., Gottfried, J.A., Hutton, C. & Turner, R. Optimized EPI for fMRI studies ofthe orbitofrontal cortex. Neuroimage 19, 430–441 (2003).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Millisecond-timescale, genetically targeted optical controlof neural activityEdward S Boyden1, Feng Zhang1, Ernst Bamberg2,3, Georg Nagel2,5 & Karl Deisseroth1,4

Temporally precise, noninvasive control of activity in well-

defined neuronal populations is a long-sought goal of systems

neuroscience. We adapted for this purpose the naturally

occurring algal protein Channelrhodopsin-2, a rapidly gated

light-sensitive cation channel, by using lentiviral gene delivery in

combination with high-speed optical switching to photostimulate

mammalian neurons. We demonstrate reliable, millisecond-

timescale control of neuronal spiking, as well as control of

excitatory and inhibitory synaptic transmission. This technology

allows the use of light to alter neural processing at the level of

single spikes and synaptic events, yielding a widely applicable

tool for neuroscientists and biomedical engineers.

Neural computation depends on the temporally diverse spiking pat-terns of different classes of neurons that express unique genetic markersand demonstrate heterogeneous wiring properties within neural net-works. Although direct electrical stimulation and recording of neurons

in intact brain tissue have provided many insights into the function ofcircuit subfields (for example, see refs. 1–3), neurons belonging to aspecific class are often sparsely embedded within tissue, posing funda-mental challenges for resolving the role of particular neuron types ininformation processing. A high–temporal resolution, noninvasive,genetically based method to control neural activity would enableelucidation of the temporal activity patterns in specific neurons thatdrive circuit dynamics, plasticity and behavior.

Despite substantial progress made in the analysis of neural networkgeometry by means of non–cell-type-specific techniques like glutamateuncaging (for example, see refs. 4–7), no tool has yet been inventedwith the requisite spatiotemporal resolution to probe neural coding atthe resolution of single spikes. Furthermore, previous geneticallyencoded optical methods, although elegant8–10,11, have allowed controlof neuronal activity over timescales of seconds to minutes, perhapsowing to their mechanisms for effecting depolarization. Kineticsroughly a thousand times faster would enable remote control of

100 mV

500 ms

10

5

0

Tim

e (m

s)

Time

to

thre

shold

Time

to

spike

pea

k

Spike

jitter

15 ms

10 ms

5 ms

400 pA

400 pA

100 ms

500 ms

100 ms

40 mV

100 pA2 s

Peak

Stead

y sta

te0

500

1,000

curr

ent (

pA)

a

d e

b cFigure 1 ChR2 enables light-driven neuron

spiking. (a) Hippocampal neurons expressing

ChR2-YFP (scale bar 30 mm). (b) Left, inward

current in voltage-clamped neuron evoked by 1 s

of GFP-wavelength light (indicated by black bar);

right, population data (right; mean ± s.d. plotted

throughout; n ¼ 18). Inset, expanded initial

phase of the current transient. (c) Ten overlaid

current traces recorded from a hippocampal

neuron illuminated with pairs of 0.5-s light pulses

(indicated by gray bars), separated by intervalsvarying from 1 to 10 s. (d) Voltage traces showing

membrane depolarization and spikes in a current-

clamped hippocampal neuron (left) evoked by

1-s periods of light (gray bar). Right, properties

of the first spike elicited (n ¼ 10): latency to

spike threshold, latency to spike peak, and

jitter of spike time. (e) Voltage traces in

response to brief light pulse series, with light

pulses (gray bars) lasting 5 ms (top), 10 ms

(middle) or 15 ms (bottom).


1Department of Bioengineering, Stanford University, 318 Campus Drive West, Stanford, California 94305, USA. 2Max-Planck-Institute of Biophysics, Department ofBiophysical Chemistry, Max-von-Laue-Str. 3, D-60438 Frankfurt am Main, Germany. 3Department of Biochemistry, Chemistry and Pharmaceutics, University of Frankfurt,Marie-Curie-Str. 9, 60439 Frankfurt am Main, Germany. 4Department of Psychiatry and Behavioral Sciences, Stanford School of Medicine, 401 Quarry Road, Stanford,California 94305, USA. 5Present address: Julius-von-Sachs-Institut, University of Wurzburg, Julius-von-Sachs-Platz 2–4, D-97082 Wurzburg, Germany. Correspondenceshould be addressed to K.D. ([email protected]).


TECHNICAL REPORT©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

individual spikes or synaptic events. We have therefore devised a newstrategy using a single-component ion channel with submillisecondopening kinetics, to enable genetically targeted photostimulation withfine temporal resolution.

Two rhodopsins in the unicellular green alga Chlamydomonas rein-hardtii were recently identified independently by three groups12–15.One of them is a light-gated proton channel (Channelrhodopsin-1;ref. 13), whereas the other, Channelrhododopsin-2 (ChR2), is a light-gated cation channel12. The N-terminal 315 amino acids of ChR2 arehomologous to the seven-transmembrane structure of many microbial-type rhodopsins; they compose a channel with light-gated conductance(as proposed earlier16). Inward currents in ChR2-expressing cells couldbe evoked within 50 ms after a flash of blue light in the presence ofall-trans retinal12, suggesting the possibility that ultrafast neuronalstimulation might be possible with equipment commonly used forvisualizing green fluorescent protein (GFP). ChR2 therefore combinessome of the best features of previous photostimulation methods,including the speed of a monolithic ion channel9, and the efficacy ofnatural light-transduction machinery11.

We found that ChR2 could be expressed stably and safely inmammalian neurons and could drive neuronal depolarization. Whenactivated with a series of brief pulses of light, ChR2 could reliablymediate defined trains of spikes or synaptic events with millisecond-timescale temporal resolution. This technology thus brings opticalcontrol to the temporal regime occupied by the fundamental buildingblocks of neural computation.

RESULTS

Rapid kinetics of ChR2 enables driving of single spikes

To obtain stable and reliable ChR2 expression for coupling light toneuronal depolarization, we constructed lentiviruses containing aChR2-yellow fluorescent protein (YFP) fusion protein for geneticmodification of neurons. Infection of cultured rat CA3/CA1 neuronsled to membrane-localized expression of ChR2 for weeks after infection(Fig. 1a). Illumination of ChR2-positive neurons with blue light(bandwidth 450–490 nm via Chroma excitation filter HQ470/40�;300-W xenon lamp) induced rapid depolarizing currents, whichreached a maximal rise rate of 160 ± 111 pA/ms within 2.3 ± 1.1 ms

after light pulse onset (mean ± s.d. reported throughout paper, n¼ 18;Fig. 1b, left). Mean whole-cell inward currents were large: 496 pA ±336 pA at peak and 193 pA ± 177 pA at steady-state (Fig. 1b, right).Light-evoked responses were never seen in cells expressing YFP alone(data not shown). Consistent with the known excitation spectrumof ChR2 (ref. 12), illumination of ChR2-expressing neurons withYFP-spectrum light in the bandwidth 490–510 nm (300-W xenonlamp filtered with Chroma excitation filter HQ500/20�) resulted incurrents that were smaller (by 42% ± 20%) than those evoked with theGFP filters. Despite the inactivation of ChR2 with sustained lightexposure (Fig. 1b and ref. 12), we observed rapid recovery of peakChR2 photocurrents in neurons (Fig. 1c; recovery t ¼ 5.1 ± 1.4 s;recovery trajectory fit with Levenberg-Marquardt algorithm; n ¼ 9).This rapid recovery is consistent with the well-known stability of theSchiff base (the lysine in transmembrane helix seven, which bindsretinal) in microbial-type rhodopsins, and the ability of retinal tore-isomerize to the all-trans ground state in a dark reaction, without theneed for other enzymes. Light-evoked current amplitudes remainedunchanged in patch-clamped neurons during 1 h of pulsed lightexposure (data not shown). Thus ChR2 was able to sustainably mediatelarge-amplitude photocurrents with rapid activation kinetics.

We next examined whether ChR2 could drive spiking of neuronsheld in current-clamp mode, with the same steady illuminationprotocol we used for eliciting ChR2-induced currents (Fig. 1d, left).Early in an epoch of steady illumination, single neuronal spikes wererapidly and reliably elicited (8.0 ± 1.9 ms latency to spike peak, n¼ 10;Fig. 1d, right), consistent with the fast rise times of ChR2 currentsdescribed above. However, any subsequent spikes elicited during steadyillumination were poorly timed (Fig. 1d, left). Thus, steady illumina-tion is not adequate for controlling the timing of ongoing spikes withChR2, despite the reliability of the first spike. Earlier patch-clampstudies using somatic current injection showed that spike times weremore reliable during periods of rapidly rising membrane potential thanduring periods of steady high-magnitude current injection17. This isconsistent with our finding that steady illumination evoked a singlereliably timed spike, followed by irregular spiking.

In searching for a strategy to elicit precisely timed series of spikeswith ChR2, we noted that the single spike reliably elicited by steady

λ = 1

00

λ = 2

00

10

5

0

Jitte

r, ne

uron

-to-

neur

on(m

s)

0 2 4 6 0 2 4 6

# of neurons firing a given spike

46 spikes59 spikesλ = 200,λ = 100,

λ = 1

00

λ = 2

00

λ = 1

00

λ = 2

00

λ = 1

00

λ = 2

00

λ =

200

λ =

100

λ =

200

λ =

100

30

20

10

0

30

20

10

0# of

spi

kes

Three different neurons:same pulse series

1 s

60 mV

Timeto

spike

Jitterthroughout

train

20

10

0

i ii

ms

100%

50%

0%

100%

50%

0%

Per

cent

age

fidel

ity o

fsp

ike

tran

smis

sion

10

5

0Jitte

r, tr

ial-t

o-tr

ial

(ms)

Rep

eata

bilit

y,tr

ial-t

o-tr

ial

One neuron: repeated pulse seriesa

f

g h

b c

d e

Figure 2 Realistic spike trains driven by series of

light pulses. (a) Voltage traces showing spikes in

a single current-clamped hippocampal neuron,

in response to three deliveries of a Poisson train

(with mean interval l ¼ 100 ms) of light pulses

(gray dashes). (b) Trial-to-trial repeatability of

light-evoked spike trains, as measured by

comparing the presence or absence of a spike

in two repeated trials of a Poisson train (either

l ¼ 100 ms or l ¼ 200 ms) delivered to the

same neuron (n ¼ 7 neurons). (c) Trial-to-trial

jitter of spikes, across repeated light-evoked

spike trains. (d) Percent fidelity of spike

transmission throughout entire 8-s light pulseseries. (e) Latency of spikes throughout each

light pulse series (i) and jitter of spike times

throughout train (ii). (f) Voltage traces showing

spikes in three different hippocampal neurons,

in response to the same temporally patterned light

stimulus (gray dashes) used in a. (g) Histogram

showing how many of the seven neurons spiked in

response to each light pulse in the Poisson train.

(h) Neuron-to-neuron jitter of spikes evoked by

light stimulation.


TECHNICAL REPORT©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

illumination had extremely low temporal jitter from trial to trial, asreflected by the small standard deviation of the spike times across trials(Fig. 1d, right; 0.5 ± 0.3 ms, average of n ¼ 10 neurons). Thisobservation led us to devise a pulsed-light strategy that would takeadvantage of the low jitter of the single reliable spike evoked at lightpulse onset. In order for this to work, the conductance and kinetics ofChR2 would have to permit peak currents of sufficient amplitude toreach spike threshold, during a light pulse of duration shorter than thedesired interspike interval. We found that multiple pulses of lightwith interspersed periods of darkness could elicit trains of multiplespikes (Fig. 1e; shown for a 25-Hz series of four pulses). Longer lightpulses evoked single spikes with greater probability than shortlight pulses (Fig. 1e). In the experiments described here, we usedlight pulse durations of 5, 10 or 15 ms. Thirteen high-expressingneurons fired reliable spikes, and five low-expressing neurons couldreliably be depolarized to subthreshold levels. The ability to easilyalter light pulse duration suggests that a straightforward methodfor eliciting spikes, even in multiple neurons having differentChR2 current densities, would involve titrating the light pulse durationuntil single spikes were reliably obtained in all the neurons beingilluminated. Modulation of light intensity would also allow for thiskind of control.

Precise spike trains elicited by series of light pulses

The precise control described above raised the prospect of generatingarbitrary spike patterns, even mimicking natural neural activity. To testthis possibility, we generated series of light pulses, the timings of whichwere selected according to a Poisson distribution, commonly used tomodel natural spiking. A single hippocampal neuron could firerepeatable spike trains in response to multiple deliveries of the samePoisson-distributed series of light pulses (Fig. 2a; response to arepeated 59-pulse-long series of light pulses, with each pulse of10-ms duration, and with mean interpulse interval of l ¼ 100 ms).These optically driven spike trains were very consistent across repeateddeliveries of the same series of light pulses: on average, 495% of thelight pulses in a series elicited spikes during one trial if and only if theyelicited spikes on a second trial, for both the l¼ 100 ms series (Fig. 2a)and a second series (mean interval l ¼ 200 ms) comprising 46 spikes(Fig. 2b; n ¼ 7 neurons). We increased light pulse duration untilreliable spiking was obtained: we used trains of 10-ms light pulses forfour of the seven neurons and trains of 15-ms light pulses for the otherthree (for the analyses of Fig. 2, all data were pooled). The trial-to-trialjitter for each individual spike was very small across repeated deliveriesof the same Poisson series of light pulses (on average, 2.3 ± 1.4 ms and1.0 ± 0.5 ms for l ¼ 100 ms and l ¼ 200 ms, respectively; Fig. 2c).Throughout a series of pulses, the efficacy of eliciting spikes throughoutthe train was maintained (76% and 85% percent of light pulses

successfully evoked spikes, respectively; Fig. 2d), with small latencies(Fig. 2e). As another indication of how precisely spikes can be elicitedthroughout an entire series of pulses, we measured the standarddeviation of the latencies of each spike across all the spikes in thetrain. This ‘throughout-train’ spike jitter was quite small (3.9 ± 1.4 msand 3.3 ± 1.2 ms; Fig. 2e), despite presumptive channel inactivationduring the delivery of a series of pulses. Hence, pulsed optical activationof ChR2 elicits precise, repeatable spike trains in a single neuron,over time.

Even in different neurons, the same precise spike train couldbe elicited by a particular series of light pulses (shown for threehippocampal neurons in Fig. 2f). Although the large hetero-geneity of different neurons—for example, in their membrane capaci-tance (68.8 ± 22.6 pF) and resistance (178.8 ± 94.8 MO)—mightbe expected to introduce significant variability in their electricalresponses to photostimulation, the strong nonlinearity of light-spikecoupling dominated this variability. Indeed, different neuronsresponded in similar ways to a given light pulse series, with 80–90%of the light pulses in a series eliciting spikes in at least half the neuronsexamined (Fig. 2g). To quantitatively compare the reliability ofspike elicitation in different neurons, for each pulse, we calculatedthe standard deviation of spike latencies (jitter) across all theneurons. Remarkably, this across-neuron jitter (3.4 ± 1.0 ms and 3.4± 1.2 ms for the pulses in the l ¼ 100 ms and l ¼ 200 ms trains,respectively; Fig. 2h) was similar to the within-neuron jitter measuredthroughout the light pulse series (Fig. 2e). Thus, heterogeneouspopulations of neurons can be controlled in concert, withpractically the same precision observed for the control of singleneurons over time.

Having established the ability of ChR2 to drive sustained naturalistictrains of spikes, we next probed the frequency response of light-spikecoupling. ChR2 enabled driving of spike trains from 5 to 30 Hz (Fig. 3a;here tested with series of twenty 10-ms light pulses). It was easier toevoke more spikes at lower frequencies than at higher frequencies(Fig. 3b; n ¼ 13 neurons). Light pulses delivered at 5 or 10 Hz couldelicit long spike trains (Fig. 3b), with spike probability dropping off athigher frequencies of stimulation. For these experiments, the lightpulses used were 5 ms (n ¼ 1), 10 ms (n ¼ 9) or 15 ms (n ¼ 3)in duration (data from all 13 cells were pooled for the populationanalyses of Fig. 3). As expected from the observation that light pulsesgenerally elicited single spikes (Figs. 1d and 2), almost no extraneousspikes occurred during the delivery of trains of light pulses (Fig. 3c).Even at higher frequencies, the throughout-train temporal jitter ofspike timing remained very low (typically o5 ms; Fig. 3d) and thelatency to spike remained constant across frequencies (B10 msthroughout; Fig. 3e). Thus ChR2 can mediate spiking across aphysiologically relevant range of firing frequencies.

5 Hz

10 Hz

20 Hz

30 Hz

60 mV1 s 0 10 20 30

Stimulus frequency(Hz)

0 10 20 300 10 20 30Stimulus frequency

(Hz)Stimulus frequency

(Hz)

0 10 20 30Stimulus frequency

(Hz)

20

10

0

20

10

0

20

10

0# of

exc

ess

spik

es

Tim

e to

spi

ke(m

s)

10

5

0

# of

suc

cess

ful s

pike

s(o

ut o

f 20

poss

ible

)

Jitte

r th

roug

hout

trai

n (m

s)

a b

c e

d Figure 3 Frequency dependence of coupling

between light input and spike output. (a) Voltage

traces showing spikes in a current-clamped

hippocampal neuron evoked by 5-, 10-, 20-

or 30-Hz trains of light pulses (gray dashes).

(b) Population data showing the number of spikes

(out of 20 possible) evoked in current-clamped

hippocampal neurons. (c) Number of extraneousspikes evoked by the trains of light pulses, for the

experiment described in b. (d) Jitter of spike times

throughout the train of light pulses for the

experiment described in b. (e) Latency to spike

peak throughout the light pulse train for the

experiment described in b.


TECHNICAL REPORT©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Remote activation of subthreshold and synaptic responses

For many cellular and systems neuroscience processes (and fornonspiking neurons in species like Caenorhabditis elegans) subthres-hold depolarizations convey physiologically significant information.For example, subthreshold depolarizations are potent for acti-vating synapse-to-nucleus signaling18, and the relative timing ofsubthreshold and suprathreshold depolarizations can determine thedirection of synaptic plasticity19. But compared with driving spiking, itis in principle more difficult to drive reliable and precisely sizedsubthreshold depolarizations. The sharp threshold for action potentialproduction facilitates reliable ChR2-induced spiking, and the all-or-none dynamics of spiking produces virtually identical waveforms fromspike to spike (as seen throughout Figs. 1–3), even in the presence ofsignificant neuron-to-neuron variability in electrical properties. Incontrast, subthreshold depolarizations, which operate in a more linearregime of membrane voltage, will lack these intrinsic normalizingmechanisms. Nevertheless, subthreshold depolarizations evoked byrepeated light pulses were reliably evoked over a range of frequencies(Fig. 4a), with spaced repeated depolarizations resulting in a coefficientof variation of 0.06 ± 0.03 (Fig. 4b; n ¼ 5). Thus, ChR2 can be used todrive subthreshold depolarizations of reliable amplitude.

Finally, synaptic transmission was also easily controlled with ChR2.Indeed, both excitatory (Fig. 4c) and inhibitory (Fig. 4d) synapticevents could be evoked in ChR2-negative neurons receiving synapticinput from ChR2-expressing neurons.

Expression of ChR2 has minimal side effects

We conducted extensive controls to test whether simply expressingChR2 would alter the electrical properties or survival of neurons.Lentiviral expression of ChR2 for at least 1 week did not alter neuronalmembrane resistance (212 ± 115 MO for ChR2+ cells versus

239.3 ± 113 MO for ChR2� cells; Fig. 5a;P 4 0.45; n ¼ 18 each) or resting potential(�60.6 ± 9.0 mV for ChR2+ cells versus �59.4± 6.0 mV for ChR2� cells; Fig. 5b; P4 0.60),when measured in the absence of light. Thissuggests that in neurons, ChR2 has little basalelectrical activity or passive current-shuntingability. It also suggests that expression ofChR2 did not compromise cell health, asindicated by electrical measurement of mem-brane integrity. As an independent measure ofcell health, we stained live neurons with themembrane-impermeant DNA-binding dyepropidium iodide. ChR2 expression did notaffect the percentage of neurons that took uppropidium iodide (1/56 ChR2+ neurons ver-sus 1/49 ChR2� neurons; P4 0.9 by w2 test).Neither did we see pyknotic nuclei, indicativeof apoptotic degeneration, in cells expressingChR2 (data not shown). We also checked foralterations in the dynamic electrical propertiesof neurons. In darkness, there was no differ-ence in the voltage change resulting from100 pA of injected current, in either thehyperpolarizing (�22.6 ± 8.9 mV for ChR2+

neurons versus �24.5 ± 8.7 mV for ChR2�

neurons; P 4 0.50) or depolarizing (+18.9 ±4.4 mV for ChR2+ versus 18.7 ± 5.2 mV forChR2� neurons; P 4 0.90) directions. Norwas there any difference in the number of

spikes evoked by a 0.5-s current injection of +300 pA (6.6 ± 4.8 forChR2+ neurons versus 5.8 ± 3.5 for ChR2– neurons; Fig. 5c; P4 0.55).Thus, ChR2 does not significantly jeopardize cell health or basalelectrical properties of the expressing neuron.

We also measured the electrical properties described above, 24 h afterexposing ChR2+ neurons to a typical light pulse protocol (1 s of 20-Hz15-ms light flashes, once per minute, for 10 min). Neurons expressingChR2 and exposed to light had basal electrical properties similar tonon-flashed ChR2+ neurons: cells had normal membrane resistance(178 ± 81 MO; Fig. 5a; P4 0.35; n¼ 12 and resting potential (�59.7 ±7.0 mV; Fig. 5b; P 4 0.75). Exposure to light also did not predisposeneurons to cell death, as measured by live-cell propidium iodide uptake(2/75 ChR2+ neurons versus 3/70 ChR2- neurons; P4 0.55 by w2 test).Finally, neurons expressing ChR2 and exposed to light also had normalspike counts elicited from somatic current injection (6.1 ± 3.9; Fig. 5c;P4 0.75). Thus, membrane integrity, cell health and electrical proper-ties were normal in neurons expressing ChR2 and exposed to light.

20 Hz

10 Hz

10 Hz10 Hz

500 ms

40 mV

5 Hz

1 s

500 pA

+GabazineInhibitory synaptic transmission

20 Hz20 Hz

1 s

20 mV

5 Hz5 Hz

Sub-threshold responses Excitatory synaptic transmission

100 pA

1 s

+NBQXa c

db

Figure 4 Simulated and natural synaptic transmission evoked via ChR2. (a) Voltage traces showing trains

of subthreshold depolarizations in a current-clamped hippocampal neuron in response to trains of light

pulses (gray dashes). (b) Repeated light pulses induced reliable depolarizations. (c) Excitatory synaptic

transmission driven by light pulses. The selective glutamatergic transmission blocker NBQX abolished

these synaptic responses (right). (d) Inhibitory synaptic transmission driven by light pulses. The

selective GABAergic transmission blocker gabazine abolished these synaptic responses (right).

400

300

200

100

0Mem

bran

e re

sist

ance

(MO

hm)

ChR2+

ChR2+

flash

edChR

2–

ChR2+

ChR2+

flash

edChR

2–

ChR2+

ChR2+

flash

edChR

2–

–70

0Res

ting

pote

ntia

l (m

V) 15

10

5

0

Spi

kes

evok

ed b

y cu

rren

tin

ject

ion

(0.5

s, +

300

pA)a b c

Figure 5 Basal and dynamic electrical properties of neurons expressing

ChR2. (a) Membrane resistance of neurons expressing ChR2 (black; n ¼ 18),

not expressing ChR2 (white; n ¼ 18) or expressing ChR2 and measured 24 h

after exposure to a typical light-pulse protocol (gray; n ¼ 12). (b) Membrane

resting potential of the same neurons described in a. (c) Number of spikes

evoked by a 300-pA depolarization of the same neurons described in a.


TECHNICAL REPORT©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

DISCUSSION

Combining the best aspects of earlier approaches that use light to driveneural circuitry, the technology described here demonstrates voltagecontrol significantly faster than previous genetically encoded photo-stimulation methods8–11. Notably, the ChR2 method does not rely onsynthetic chemical substrates or genetic orthogonality of the transgeneand the host organism. Although the ChR2 molecule does require thecofactor all-trans retinal for light transduction12, no all-trans retinalwas added either to the culture medium or recording solution for anyof the experiments described here. Background levels of retinal may besufficient in many cases; moreover, the commonly used culturemedium supplement B-27 used here (see Methods) includes retinylacetate, and additional supplementation with all-trans retinal or itsprecursors may assist in the application of ChR2 to the study of neuralcircuits in various tissue environments.

We used pulsed light delivery to take full advantage of the fast kineticsand high conductance of ChR2. This strategy was made possible withfast optical switches, but other increasingly common equipment, suchas pulsed lasers, would also suffice. Unlike electrical stimulation,glutamate uncaging20–22 and high-powered laser excitation methods23,ChR2 can be genetically targeted to allow probing of specific neuronsubclasses within a heterogeneous neural circuit, avoiding fibers ofpassage and the simultaneous stimulation of multiple cell types.

Because ChR2 is encoded by a single open reading frame of only 315amino acids, it is feasible to express ChR2 in specific subpopulations ofneurons in the nervous system through genetic methods includinglentiviral vectors (as we have done) and in transgenic mice, thuspermitting the study of the function of individual types of neuronsin intact neural circuits and even in vivo. Cell-specific promoters willallow targeting of ChR2 to various well-defined neuronal subtypes,which will permit future exploration of their causal function in drivingdownstream neural activity (measured using electrophysiological andoptical techniques) and animal behavior. ChR2 also could be used toresolve functional connectivity of particular neurons or neuron classesin intact circuits in response to naturalistic spike trains (Fig. 2) orrhythmic activity (Fig. 3), for example, by using acute slice prepara-tions after intracranial viral injections.

Recent papers have explored the topics of static and dynamicmicrocircuit connectivity using calcium imaging of spontaneousactivity24, multi-neuron patch-clamping25,26 and glutamate uncaging6.These studies have reported surprisingly refined and precise connec-tions between neurons. However, finer-scale dissection of micro-circuits, at the level of molecularly defined neuron classes (suchas cannabinoid receptor–expressing cortical neurons, parvalbumin-positive interneurons or cholinergic modulatory neurons) would begreatly facilitated through use of a genetically targeted, temporallyprecise tool like ChR2. This holds true also for recent microstimulationexperiments that have demonstrated profound influences of a cluster ofneurons in controlling attention, decision making or action2,3,27,28.Understanding precisely which cell types contribute to these functionscould provide great insight into how they are computed at the circuitlevel. Because the light power required for ChR2 activation (8–12 mW/mm2) is fairly low, it is possible that ChR2 will be an effective tool forin vivo studies of circuit maps and behavior, even in mammals. Finally,the efficacious and safe transduction of light with a single naturalbiological component also could serve biotechnological needs, in high-throughput studies of activity-dependent signal transduction and geneexpression programs, for example, in guiding stem cell differentiation29

and screening for drugs that modulate neuronal responses to depolar-ization. Thus, the technology described here may fulfill the long-soughtgoal of a method for noninvasive, genetically targeted, temporally

precise control of neuronal activity, with potential applications rangingfrom neuroscience to biomedical engineering.

METHODSPlasmid constructs. The ChR2-YFP gene was constructed by in-frame fusing

EYFP (Clontech) to the C terminus of the first 315 amino acid residues of

ChR2 (GenBank accession number AF461397) via a NotI site. The lentiviral

vector pLECYT was generated by PCR amplification of ChR2-YFP with pri-

mers 5¢-GGCAGCGCTGCCACCATGGATTATGGAGGCGCCCTGAGT-3¢ and

5¢-GGCACTAGTCTATTACTTGTACAGCTCGTC-3¢ and ligation into pLET

(gift from E. Wexler and T. Palmer, Stanford University) via the AfeI and SpeI

restriction sites. The plasmid was amplified and then purified using MaxiPrep

kits (Qiagen).

Viral production. VSVg pseudotyped lentiviruses were produced by triple

transfection of 293FT cells (Invitrogen) with pLECYT, pMD.G and

pCMVDR8.7 (gifts from E. Wexler and T. Palmer) using Lipofectamine 2000.

The lentiviral production protocol is the same as previously described30 except

for the use of Lipofectamine 2000 instead of calcium phosphate precipitation.

After harvest, viruses were concentrated by centrifuging in a SW28 rotor

(Beckman Coulter) at 20,000 rpm for 2 h at 4 1C. The concentrated viral titer

was determined by FACS to be between 5 � 108 and 1 � 109 infectious units

(IU) per ml.

Hippocampal cell culture. Hippocampi of postnatal day 0 (P0) Sprague-

Dawley rats (Charles River) were removed and treated with papain (20 U/ml)

for 45 min at 37 1C. The digestion was stopped with 10 ml of MEM/Earle salts

without L-glutamine along with 20 mM glucose, Serum Extender (1:1000), and

10% heat-inactivated fetal bovine serum containing 25 mg of bovine serum

albumin (BSA) and 25 mg of trypsin inhibitor. The tissue was triturated in a

small volume of this solution with a fire-polished Pasteur pipette, and

B100,000 cells in 1 ml plated per coverslip in 24-well plates. Glass coverslips

(prewashed overnight in HCl followed by several 100% ethanol washes and

flame sterilization) were coated overnight at 37 1C with 1:50 Matrigel

(Collaborative Biomedical). Cells were plated in culture medium: Neurobasal

containing 2� B-27 (Life Technologies) and 2 mM Glutamax-I (Life Technol-

ogies). The culture medium supplement B-27 contains retinyl acetate, but no

B-27 was present during recording and no all-trans retinal was added to the

culture medium or recording medium for any of the experiments described.

One-half of the medium was replaced with culture medium the next day, giving

a final serum concentration of 1.75%.

Viral infection. Hippocampal cultures were infected on day 7 in vitro (DIV 7)

using fivefold serial dilutions of lentivirus (B1 � 106 IU/ml). Viral dilutions

were added to hippocampal cultures seeded on coverslips in 24-well plates and

then incubated at 37 1C for 7 d before experimentation.

Confocal imaging. Images were acquired on a Leica TCS-SP2 LSM confocal

microscope using a 63� water-immersion lens. Cells expressing ChR2-YFP

were imaged live using YFP microscope settings, in Tyrode solution containing

(in mM) NaCl 125, KCl 2, CaCl2 3, MgCl2 1, glucose 30 and HEPES 25 (pH 7.3

with NaOH).

Propidium iodide (Molecular Probes) staining was carried out on live cells

by adding 5 mg/ml propidium iodide to the culture medium for 5 min at 37 1C,

washing twice with Tyrode solution and then immediately counting the

number of ChR2+ and ChR2� cells that took up propidium iodide. Coverslips

were then fixed for 5 min in PBS + 4% paraformaldehyde, permeabilized for

2 min with 0.1% Triton X-100 and then immersed for 5 min in PBS containing

5 mg/ml propidium iodide for detection of pyknotic nuclei. At least eight fields

of view were examined per coverslip.

Electrophysiology and optical methods. Cultured hippocampal neurons were

recorded at approximately DIV 14 (7 d post-infection). Neurons were recorded

by means of whole-cell patch clamp, using Axon Multiclamp 700B (Axon

Instruments) amplifiers on an Olympus IX71 inverted scope equipped with

a 20� objective lens. Borosilicate glass (Warner) pipette resistances were

B4 MO, range 3–8 MO. Access resistance was 10–30 MO and was monitored

throughout the recording. Intracellular solution consisted of (in mM) 97


TECHNICAL REPORT©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

potassium gluconate, 38 KCl, 0.35 EGTA, 20 HEPES, 4 magnesium ATP, 0.35

sodium GTP, 6 NaCl, and 7 phosphocreatine (pH 7.25 with KOH). Neurons

were recorded in Tyrode solution (above). All experiments were performed at

room temperature (22–24 1C). For all experiments except for the synaptic

transmission data shown in Figure 4b,c, we patched fluorescent cells immersed

in Tyrode solution containing 5 mM NBQX and 20 mM gabazine to block

synaptic transmission.

Photocurrents were measured while holding neurons in voltage clamp at

�65 mV. Recovery from inactivation was measured by measuring photo-

currents while illuminating neurons with pairs of 500-ms duration light pulses,

separated by periods of darkness lasting 1–10 s.

Spiking was measured while injecting current to keep the voltage of the cell

at approximately �65 mV (holding current ranging from 0 pA to �200 pA).

For synaptic transmission experiments, we patched nonfluorescent neurons

near ChR2-expressing neurons, immersed in Tyrode solution containing either

20 mM gabazine to isolate the excitatory postsynaptic response or in 5 mM

NBQX to isolate the inhibitory postsynaptic response. To confirm whether the

evoked potentials were indeed synaptically driven, after photostimulation, we

blocked all postsynaptic receptors with solution containing both 20 mM

gabazine and 5 mM NBQX and carried out photostimulation again.

pClamp 9 software (Axon Instruments) was used to record all data, and a

DG-4 high-speed optical switch with 300-W xenon lamp (Sutter Instruments)

was used to deliver the light pulses for ChR2 activation. An Endow GFP filter

set (excitation filter HQ470/40�, dichroic Q495LP; Chroma) was used for

delivering blue light for ChR2 activation. YFP was visualized with a standard

YFP filter set (excitation HQ500/20�, dichroic Q515LP, emission HQ535/30 m;

Chroma). Through a 20� objective lens, power density of the blue light was

8–12 mW/mm2, measured with a power meter (Newport).

Pulse series were synthesized by custom software written in MATLAB

(MathWorks) and then exported through pClamp 9 via a Digidata board

(Axon) attached to a PC. Poisson trains were generated in MATLAB by creating

series of pulses with inter-pulse intervals independently picked from a Poisson

distribution with mean l. Poisson trains were 8 s long, with mean interval

l ¼ 100 or 200 ms. For biophysical realism, a 10-ms minimum refractory

period was enforced between consecutive light pulses.

Membrane resistance was measured in voltage clamp mode with 20-mV

depolarizing steps lasting 75 ms. Spike rates due to direct current injection were

measured with 300-pA current steps lasting 0.5 s.

Data analysis. Data was analyzed using Clampfit (Axon) and custom software

written in MATLAB. Spikes were extracted by looking for voltage crossings of a

threshold (typically 60 mV above resting potential), and latencies were

measured from the onset of the light pulse to the spike peak. Extraneous

spikes were measured as the number of extra spikes after a single light pulse,

plus any spikes occurring later than 30 ms after the onset of a light pulse.

Jitter was calculated as the standard deviation of spike latencies, measured

either across all the spikes throughout a spike train (‘throughout-train’ jitter),

or for a particular spike across different trains (when gauging trial-to-trial

reliability, or across different neurons). For display of population data,

throughout-train jitter and trial-to-trial jitter were averaged across all neurons,

and neuron-to-neuron jitter was averaged across all spikes. For all jitter

analyses, light pulses which failed to elicit a spike were ignored. For the

across-neuron jitter analysis shown in Figure 2h, light pulses that did not elicit

spikes in all seven neurons were ignored (leaving 31/59 light pulses for the

l ¼ 100 ms stimulus, and 30/46 light pulses for the l ¼ 200 ms stimulus).

ACKNOWLEDGMENTSWe would like to thank L. Meltzer and N. Adeishvili for experimental assistance;C. Niell, C. Chan and J.P. Levy for helpful discussions and D. Ollig for technicalhelp. E.B. and G.N. are supported by the Max-Planck-Society and acknowledgea grant from the German Research Foundation (DFG) in the research unit 472(Molekulare Bioenergetik). E.S.B. is supported by the Helen Hay WhitneyFoundation, the Dan David Prize Foundation, and National Institute onDeafness and Other Communication Disorders, and F.Z. is supported by a USNational Institutes of Health predoctoral fellowship. K.D. is supported by theNational Institute of Mental Health, the Stanford Department of Bioengineering,the Stanford Department of Psychiatry and Behavioral Sciences, the Neuroscience

Institute at Stanford, the National Alliance for Research On Schizophrenia andDepression and the Culpeper, Klingenstein, Whitehall, McKnight, and Albert Yuand Mary Bechmann Foundations.




1. Kandel, E.R., Spencer, W.A. & Brinley, F.J., Jr. Electrophysiology of hippocampalneurons. I. Sequential invasion and synaptic organization. J. Neurophysiol. 24,225–242 (1961).

2. Ditterich, J., Mazurek, M.E. & Shadlen, M.N. Microstimulation of visual cortex affectsthe speed of perceptual decisions. Nat. Neurosci. 6, 891–898 (2003).

3. Salzman, C.D., Britten, K.H. & Newsome, W.T. Cortical microstimulationinfluences perceptual judgements of motion direction. Nature 346, 174–177(1990).

4. Shepherd, G.M., Pologruto, T.A. & Svoboda, K. Circuit analysis of experience-dependentplasticity in the developing rat barrel cortex. Neuron 38, 277–289 (2003).

5. Pettit, D.L., Helms, M.C., Lee, P., Augustine, G.J. & Hall, W.C. Local excitatory circuits inthe intermediate gray layer of the superior colliculus. J. Neurophysiol. 81, 1424–1427(1999).

6. Yoshimura, Y., Dantzker, J.L. & Callaway, E.M. Excitatory cortical neurons form fine-scalefunctional networks. Nature 433, 868–873 (2005).

7. Dalva, M.B. & Katz, L.C. Rearrangements of synaptic connections in visual cortexrevealed by laser photostimulation. Science 265, 255–258 (1994).

8. Lima, S.Q. & Miesenbock, G. Remote control of behavior through genetically targetedphotostimulation of neurons. Cell 121, 141–152 (2005).

9. Banghart, M., Borges, K., Isacoff, E., Trauner, D. & Kramer, R.H. Light-activatedion channels for remote control of neuronal firing. Nat. Neurosci. 7, 1381–1386(2004).

10. Zemelman, B.V., Nesnas, N., Lee, G.A. & Miesenbock, G. Photochemical gating ofheterologous ion channels: remote control over genetically designated populations ofneurons. Proc. Natl. Acad. Sci. USA 100, 1352–1357 (2003).

11. Zemelman, B.V., Lee, G.A., Ng, M. & Miesenbock, G. Selective photostimulation ofgenetically chARGed neurons. Neuron 33, 15–22 (2002).

12. Nagel, G. et al. Channelrhodopsin-2, a directly light-gated cation-selective membranechannel. Proc. Natl. Acad. Sci. USA 100, 13940–13945 (2003).

13. Nagel, G. et al. Channelrhodopsin-1: a light-gated proton channel in green algae.Science 296, 2395–2398 (2002).

14. Sineshchekov, O.A., Jung, K.H. & Spudich, J.L. Two rhodopsins mediate phototaxis tolow- and high-intensity light in Chlamydomonas reinhardtii. Proc. Natl. Acad. Sci. USA99, 8689–8694 (2002).

15. Suzuki, T. et al. Archaeal-type rhodopsins in Chlamydomonas: model structure andintracellular localization. Biochem. Biophys. Res. Commun. 301, 711–717 (2003).

16. Harz, H. & Hegemann, P. Rhodopsin-regulated calcium currents in Chlamydomonas.Nature 351, 489–491 (1991).

17. Mainen, Z.F. & Sejnowski, T.J. Reliability of spike timing in neocortical neurons. Science268, 1503–1506 (1995).

18. Mermelstein, P.G., Bito, H., Deisseroth, K. & Tsien, R.W. Critical dependence of cAMPresponse element-binding protein phosphorylation on L-type calcium channels supportsa selective response to EPSPs in preference to action potentials. J. Neurosci. 20, 266–273 (2000).

19. Bi, G.Q. & Poo, M.M. Synaptic modifications in cultured hippocampal neurons:dependence on spike timing, synaptic strength, and postsynaptic cell type. J. Neurosci.18, 10464–10472 (1998).

20. Katz, L.C. & Dalva, M.B. Scanning laser photostimulation: a new approach for analyzingbrain circuits. J. Neurosci. Methods 54, 205–218 (1994).

21. Dantzker, J.L. & Callaway, E.M. Laminar sources of synaptic input to cortical inhibitoryinterneurons and pyramidal neurons. Nat. Neurosci. 3, 701–707 (2000).

22. Schubert, D. et al. Layer-specific intracolumnar and transcolumnar functional connec-tivity of layer V pyramidal cells in rat barrel cortex. J. Neurosci. 21, 3580–3592(2001).

23. Hirase, H., Nikolenko, V., Goldberg, J.H. & Yuste, R. Multiphoton stimulation of neurons.J. Neurobiol. 51, 237–247 (2002).

24. Ikegaya, Y. et al. Synfire chains and cortical songs: temporal modules of cortical activity.Science 304, 559–564 (2004).

25. Kozloski, J., Hamzei-Sichani, F. & Yuste, R. Stereotyped position of local synaptictargets in neocortex. Science 293, 868–872 (2001).

26. Pouille, F. & Scanziani, M. Routing of spike series by dynamic circuits in thehippocampus. Nature 429, 717–723 (2004).

27. Graziano, M.S., Taylor, C.S. & Moore, T. Complex movements evoked by microstimula-tion of precentral cortex. Neuron 34, 841–851 (2002).

28. Moore, T. & Armstrong, K.M. Selective gating of visual signals by microstimulation offrontal cortex. Nature 421, 370–373 (2003).

29. Deisseroth, K. et al. Excitation-neurogenesis coupling in adult neural stem/progenitorcells. Neuron 42, 535–552 (2004).

30. Dull, T. et al. A third-generation lentivirus vector with a conditional packaging system.J. Virol. 72, 8463–8471 (1998).


TECHNICAL REPORT©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

nature neuroscience september 2005

Documents