Nature Neuroscience November 2005.pdf

www.nature.com/natureneuroscience

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

EDITORIAL OFFICE [email protected] 345 Park Avenue South, New York, NY 10010-1707 Tel: (212) 726 9319, Fax: (212) 696 0978 Editor: Sandra Aamodt Senior Editor: Kalyani Narasimhan Associate Editors: Cara Allen, I-han Chou, Annette Markus, Charvy Narain Copy Editor: Dorothy Moore, Lavanya Reddy Senior Production Editor: Nicole D. Fournier Production Editor: Jessica Iannuzzi Cover Design: Kimberly Caesar Editorial Assistant: Jessica Chen MANAGEMENT OFFICES NPG New York 345 Park Avenue South, New York, NY 10010-1707 Tel: (212) 726 9200, Fax: (212) 696 9006 Publisher: Beatrice Renault Executive Editor: Linda Miller Chief Technology Officer: Howard Ratner Global Head of Advertising: Fabien Savenay Head of Nature Research & Reviews Marketing: Sara Girard Assistant Marketing Manager: Amy Maurer Production Coordinator: Diane Temprano Associate Director, Web Publishing: Timo Hannay New Technology Director: Greg Suprock Head of Web Services: Anthony Barrera Web Production Editor: Edwina Hay NPG London The Macmillan Building, 4 Crinan Street, London N1 9XW Tel: 44 207 833 4000, Fax: 44 207 843 4996 Managing Director: Annette Thomas Publishing Director: Peter Collins Editor-in-Chief, Nature Publications: Philip Campbell Marketing Director: Della Sar NPG Tokyo Chiyoda Building 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843 Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 Asia-Pacific Publisher: Antoine E. Bocquet Manager: Koichi Nakamura Senior Marketing Manager: Peter Yoshihara Asia-Pacific Sales Director: Kate Yoneyama Asia-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 9481 Global Head of Display Advertising Sales: John Michael, Tel: 44 207 843 4960, Fax: 44 207 843 4996 Head of Display Advertising—Europe: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4996 Business Development Manager: Claire Hines, Tel: 44 207 843 4960, Fax: 44 207 843 4996 Asia-Pacific Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 Western Region Sales Manager: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805 Sales Executives: New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717 New York, Mid-Atlantic, Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481 Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481 Northwest: Mathieu DesJardins, Tel: (415) 781 6422, Fax: (415) 781 3805 Eastern England/Scotland/Italy/Spain/Israel: Matthew Wilkinson, Tel: 44 207 843 4960, Fax: 44 207 843 4749 South/West United Kingdom/Scandinavia/Holland: Marianne Boulakas, Tel: 44 207 843 4969, Fax: 44 207 843 4749 Northern Germany: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4749 Southern 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 9482 US Sales Manager: Peter Bless, Tel: (212) 726 9248, Fax: (212) 696 9482 Japan Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 naturejobs Sales Director: Nevin Bayoumi, Tel: 44 207 843 4961, Fax: 44 207 843 4996 SITE LICENSE BUSINESS UNIT Americas: Tel: (888) 331 6288 Asia/Pacific: Tel: 81 3 3267 8751 Australia/New Zealand: Tel: 61 3 9825 1160 India: Tel: 00 91 11 324 4186 ROW: Tel: 44 207 843 4609

[email protected] [email protected] [email protected] [email protected] [email protected]

CUSTOMER SERVICE www.nature.com/help Senior Global Customer Service Manager: Gerald Coppin For all print and online assistance, please visit www.nature.com/help Purchase subscriptions: Americas: Nature Neuroscience, Subscription Dept., 303 Park Avenue South #1280, New York, NY 10010-3601. Tel: (866) 363 7860, Fax: (212) 689 9108 Europe/ROW: Nature Neuroscience, Subscription Dept., Macmillan Magazines Ltd., Brunel Road, Houndmills, Basingstoke RG21 6XS, United Kingdom. Tel: 44 1256 329 242, Fax: 44 1256 812 358 Japan: 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 8746 India: 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] Nature 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 4749 US Reprints Sales Executive: Sharda Tulsie, Tel: (212) 726 9631, Fax: (212) 679 0843


VOLUME 8 NUMBER 11 NOVEMBER 2005

E D I TO R I A L 1413 Taking addiction research into the clinic

BOOK REVIEW Addiction is pervasive, affecting millions of people around the world. The progression from recreational drug use to drug dependence and addiction is influenced by many factors, including the nature of the drug, the personality of the user, and environmental stressors. In this issue, we present reviews and opinion pieces on the neurobiology of drug abuse, decision making and habit formation, as well as a commentary on how the neuroscience of addiction should guide public policy and treatment. This special focus is sponsored by the National Institute on Drug Abuse and the National Institute on Alcohol Abuse and Alcoholism. (pp 1427–1489)

1415 The War of the Soups and the Sparks By Elliot S Valenstein Reviewed by Nicholas C Spitzer

NEWS AND VIEWS 1417 How neurons keep in touch Fekrije Selimi & Nathaniel Heintz see also p 1534 1418 Synaptic plasticity and self-organization in the hippocampus György Buzsáki & James J Chrobak see also p 1560 1420 Glial cells under remote control Klaus-Armin Nave & Markus H Schwab 1422 Senseless makes sense for spinocerebellar ataxia-1 Vikram Khurana, Tudor A Fulga & Mel B Feany 1424 How the brain recovers following damage Yalçin Abdullaev & Michael I Posner see also p 1603 1425 Time to smell the roses Cara Allen see also p 1568

I N T RO D U C T I O N : N E U RO B I O LO G Y O F A D D I C T I O N 1427 Neurobiology of addiction I-han Chou & Kalyani Narasimhan

S P O N S O R S ’ F O R E WO R D : N E U RO B I O LO G Y O F ADDICTION 1429 The neuroscience of addiction Nora Volkow & Ting-Kai Li

Semaphorin controls cerebellar granule cell migration (p 1516)

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.

i



C O M M E N TA R I E S : N E U R O B I O L O G Y O F A D D I C T I O N 1431 Neurobiology of addiction: treatment and public policy ramifications Charles Dackis & Charles O’Brien 1437 The role of neuroadaptations in relapse to drug seeking Yavin Shaham and Bruce T Hope 1440 How do we determine which drug-induced neuroplastic changes are important? Peter W Kalivas 1442 Plasticity of reward neurocircuitry and the ‘dark side’ of drug addiction George F Koob & Michel Le Moal

P E R S P E C T I V E S : N E U RO B I O LO G Y O F A D D I C T I O N 1445 Is there a common molecular pathway for addiction? Eric J Nestler 1450 Genetic influences on impulsivity, risk taking, stress responsivity and vulnerability to drug abuse and addiction Mary Jeanne Kreek, David A Nielsen, Eduardo R Butelman & K Steven LaForge 1458 Decision making, impulse control and loss of willpower to resist drugs: a neurocognitive perspective Antoine Bechara

R E V I E W S : N E U RO B I O LO G Y O F A D D I C T I O N 1465 Nicotine addiction and comorbidity with alcohol abuse and mental illness John A Dani and R Adron Harris 1471 Laboratory models of alcoholism: treatment target identification and insight into mechanisms David M Lovinger & John C Crabbe 1481 Neural systems of reinforcement for drug addiction: from actions to habits to compulsion Barry J Everitt & Trevor W Robbins

B R I E F C O M M U N I C AT I O N S 1491 The cerebellum communicates with the basal ganglia E Hoshi, L Tremblay, J Féger, P L Carras & P L Strick 1494 Shift of activity from attention to motor-related brain areas during visual learning S Pollmann & M Maertens 1497 The essential role of stimulus temporal patterning in enabling perceptual learning S-G Kuai, J-Y Zhang, S A Klein, D M Levi & C Yu 1500 COMT genotype predicts longitudinal cognitive decline and psychosis in 22q11.2 deletion syndrome D Gothelf, S Eliez, T Thompson, C Hinard, L Penniman, C Feinstein, H Kwon, S Jin, B Jo, S E Antonarakis, M A Morris & A L Reiss

ARTICLES 1503 MPS-1 is a K+ channel β-subunit and a serine/threonine kinase S-Q Cai, L Hernandez, Y Wang, K H Park & F Sesti

NATURE NEUROSCIENCE

iii



1510 Lbx1 and Tlx3 are opposing switches in determining GABAergic versus glutamatergic transmitter phenotypes L Cheng, O Abdel Samad, Y Xu, R Mizuguchi, P Luo, S Shirasawa, M Goulding & Q Ma Activity-dependent decrease in global excitability (p 1542)

1516 The transmembrane semaphorin Sema6A controls cerebellar granule cell migration G Kerjan, J Dolan, C Haumaitre, S Schneider-Maunoury, H Fujisawa, K J Mitchell & A Chédotal 1525 TARP γ-8 controls hippocampal AMPA receptor number, distribution and synaptic plasticity N Rouach, K Byrd, R S Petralia, G M Elias, H Adesnik, S Tomita, S Karimzadegan, C Kealey, D S Bredt & R A Nicoll 1534 Cbln1 is essential for synaptic integrity and plasticity in the cerebellum H Hirai, Z Pang, D Bao, T Miyazaki, L Li, E Miura, J Parris, Y Rong, M Watanabe, M Yuzaki & J I Morgan see also p 1417 1542 Activity-dependent decrease of excitability in rat hippocampal neurons through increases in Ih Y Fan, D Fricker, D H Brager, X Chen, H-C Lu, R A Chitwood & D Johnston 1552 Fine-scale specificity of cortical networks depends on inhibitory cell type and connectivity Y Yoshimura & E M Callaway 1560 Induction of sharp wave–ripple complexes in vitro and reorganization of hippocampal networks C J Behrens, L P van den Boom, L de Hoz, A Friedman & U Heinemann see also p 1418

Binocular rivalry in the lateral geniculate nucleus (p 1595)

1568 Encoding a temporally structured stimulus with a temporally structured neural representation S L Brown, J Joseph & M Stopfer see also p 1425 1577 Drosophila melanogaster homolog of Down syndrome critical region 1 is critical for mitochondrial function K T Chang & K-T Min 1586 Transcriptional and behavioral interaction between 22q11.2 orthologs modulates schizophrenia-related phenotypes in mice M Paterlini, S S Zakharenko, W-S Lai, J Qin, H Zhang, J Mukai, K G C Westphal, B Olivier, D Sulzer, P Pavlidis, S A Siegelbaum, M Karayiorgou & J A Gogos 1595 Neural correlates of binocular rivalry in the human lateral geniculate nucleus K Wunderlich, K A Schneider & S Kastner 1603 Neural basis and recovery of spatial attention deficits in spatial neglect M Corbetta, M J Kincade, C Lewis, A Z Snyder & A Sapir see also p 1424 1611 Perceptions of moral character modulate the neural systems of reward during the trust game M R Delgado, R H Frank & E A Phelps

Down Syndrome Critical Region 1 and mitochondrial function (p 1577)

TECHNICAL REPORT 1619 A hybrid approach to measuring electrical activity in genetically specified neurons B Chanda, R Blunck, L C Faria, F E Schweizer, I Mody & F Bezanilla

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

NATURE NEUROSCIENCE

v


E D I TO R I A L

Taking addiction research into the clinic

A

s the focus articles in this issue demonstrate, our understanding of the neurobiology of addiction has progressed substantially over the last decade, largely through studies in rodents. However, despite having some of the best animal models in neuroscience, researchers have been less successful in translating this knowledge into effective therapies. To solve this problem, we need to remove the roadblocks to development and testing of new treatments. Some medications are available. For alcoholism, the opioid receptor antagonist naltrexone and a newer drug, acamprosate, are approved in the US and Europe. (Acamprosate affects GABA and glutamate receptors, but its mechanism of action is not fully understood.) Each drug moderately reduces the amount and frequency of drinking in clinical trials, and combination therapy with both drugs seems to be more effective than either alone. For cocaine addiction, modafinil (an atypical stimulant that is approved for other uses) seems to increase abstinence during treatment1. For heroin addicts, the µ-opioid receptor partial agonist buprenorphine is available in Europe and the US. Buprenorphine is as effective as methadone (a full agonist that is tightly regulated), and has no abuse potential or overdose liability. However, in the US, federal regulations prohibit physicians from treating more than 30 patients. Such restrictions are common for treatments that target addiction to illegal drugs, and they unfairly reduce the availability of help for patients who need it most. Other therapies affect not only drug addiction, but also the response to natural rewards, such as food. This is a potentially serious problem for drug design, as the brain circuits that respond to natural rewards partially overlap with those that respond to addictive drugs. For example, topiramate, an anticonvulsant that acts at AMPA/kainate glutamate receptors, is used for alcoholism and cocaine addiction, but it is also effective against binge eating, suggesting that it may affect responses to food. Rimonabant is a cannabinoid receptor 1 antagonist that reduces the expression of several addictions in animal models, and is

now in clinical trials in humans. However, the compound is also in clinical trials as a treatment for obesity, and seems to promote modest weight loss, so it too is likely to affect eating behavior. Preliminary results from clinical trials suggest that other approaches may be promising. Vaccines against cocaine2 or nicotine seem to reduce drug intake in patients who produce enough antibodies, which then intercept the drug before it reaches receptors in the brain. N-acetylcysteine (a food supplement available over the counter) reduces cocaine addiction without affecting the response to food in rats, and is now being tested in people. Some animal models of addiction mimic the diagnostic characteristics of the human disorder very closely, but unfortunately they are too complicated for use in large-scale drug screening. For example, in one model of drug seeking and relapse, rats are trained first to press a lever for access to a drug, and then to work for an intermediate cue that predicts the drug’s future availability (as money might do for human addicts). If this behavior fails to produce the drug, the animal eventually will stop pressing the lever, but later exposure to stress, the addictive drug, or cues associated with the drug can cause the drug-seeking behavior to start again, mimicking relapse in human addicts. This entire assay takes 10 to 12 weeks. Such models are time-consuming and expensive, so investigators typically begin screening for drug-induced neural changes with simpler tests. Promising hypotheses can then be studied with more complex behaviors. The disadvantage of this approach is that it does not allow the identification of neural adaptations that are specific to compulsive drug seeking and relapse—and such neural changes might be the most directly relevant to addiction treatment. For example, drug self-administration by rodents produces distinct effects in the brain that are not seen when the same amount of drug is simply injected by researchers. Beyond the practical problems involved in identifying targets, pharmaceutical companies lack incentives to develop medicines for addiction. Treating addiction to illegal drugs

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 11 | NOVEMBER 2005

raises legal issues, and the common view of addiction as a character defect rather than a neurobiological disorder3 creates a public relations problem. In addition, the addicts who most need treatment are least likely to have jobs or medical insurance. Insurance companies often restrict the availability of treatment, even for the many alcoholics and nicotine addicts who have coverage. For these reasons, pharmaceutical companies are often reluctant to undertake fullscale development of new drugs, or even to release the compounds that they already have for testing in addiction. Instead clinical trials for addiction typically involve drugs that are already approved for other uses, which reduces the costs of bringing them to market. In contrast, most drug targets proposed by basic researchers have not been tested properly in the clinic, either because drugs to target these mechanisms do not exist or because the companies that own the compounds have not made them available to researchers. What can be done to improve the situation? Governments and charitable foundations could provide better incentives for drug development, such as promising to purchase a certain amount of any effective drug that is developed, or they could purchase candidate compounds from pharmaceutical companies for testing in their own trials. Doctors and insurance companies should begin to think of drug addiction as a chronic disease that must be treated over the long term, despite difficulties with patient compliance, like schizophrenia or hypertension. Behavioral choices contribute to many health problems—diet and exercise to heart disease, for example, or smoking to lung cancer—but we do not refuse these patients medical attention because they are culpable for their illness. A similar attitude toward addicted people would go a long way toward improving their care. 1. Dackis, C.A., Kampman, K.M., Lynch, K.G., Pettinati, H.M. & O’Brien, C.P. Neuropsychopharmacology 30, 205–211 (2005). 2. Martell, B.A., Mitchell, E., Poling, J., Gonsai, K. & Kosten, T.R. Biol. Psychiatry 58, 158–164 (2005). 3. Dackis, C. & O’Brien, C. Nat. Neurosci. 8, 1431–1436 (2005).

1413

How neurons keep in touch Fekrije Selimi & Nathaniel Heintz Cerebellin 1 is abundant in the cerebellum, but its function remains a mystery. Hirai et al. now show that this gene is required to maintain parallel fiber-Purkinje cell synapses, via the orphan glutamate receptor subunit Grid2. These findings provide further evidence that there is a molecular pathway devoted to maintenance of synapses. Neurons face a classic issue: it is not enough to find the right partner; they also need to work hard to keep it. Forming and maintaining the proper synaptic contacts between specific types of neurons is the last step in establishing functional neuronal networks in the brain. Any dysfunction during this process—which includes formation, maturation and elimination of synapses—can lead to cognitive disorders1. Moreover, learning and memory formation requires continuous synaptic plasticity at both the functional and structural level2. Thus, regulating the stability of synapses must be a dynamic process that responds to bidirectional signaling between pre- and postsynaptic elements, altering synaptic properties and stability in response to specific cues. Whereas the role of activity in the maintenance of synapses seems clear3, very little is known about the identity of the molecules that mediate this particular transsynaptic action. In this issue, Hirai et al.4 demonstrate that the transsynaptic function of the secreted glycoprotein Cbln1 is essential for the stability and plasticity of a specific synapse: the parallel fiber–Purkinje cell (PF-PC) synapse. Cerebellin, discovered two decades ago, is a 16-mer peptide associated with Purkinje cells5. Subsequent studies have focused on the relationship of the full-length cerebellin protein Cbln1 to the C1Q/TNFα family of secreted proteins and on its biochemical properties6 but have not yielded significant insight into its role in the cerebellum. In this new study, Hirai

Fekrije Selimi and Nathaniel Heintz are at the Laboratory for Molecular Biology, and the Howard Hughes Medical Institute, The Rockefeller University, New York, New York 10021, USA. Fekrije Selimi is also at the Centre National de la Recherche Scientifique, UMR7102, Paris, France. e-mail: [email protected]

Figure 1 A model for Cbln1 action in Purkinje cell–parallel fiber synapses. Cbln1 is secreted from the parallel fiber bouton, where it acts in combination with Grid2 and other unknown Glutamate spillover? molecules (‘X’) to stabilize the PF-PC interaction. + Cbln1 This process involves as-yet unidentified intracellular signaling cascades and possibly X X Grid2 Grid2 retrograde signals to the PF bouton (dashed arrow). The actions of Cbln1 are restricted to the Destabilization PSD immediate vicinity of the presynaptic terminal of PSD owing to its properties as a large, poorly diffusible Stabilization glycoprotein complex. This counteracts the destabilizing effect of Grid2 activation by a small, diffusible ligand (presumably glutamate (blue arrows)). At sites distal to the actions of Cbln1, activation of Grid2 has a destabilizing effect on postsynaptic structures, perhaps through its ability to control autophagy locally. In this way, the actions of Cbln1 and Grid2 in Purkinje cell spines can control both synapse stabilization and refinement.

and colleagues describe in detail the phenotype of a mouse lacking Cbln1, providing new and important insight into Cbln1 function. The gross phenotype of Cbln1 knockout mice is unremarkable. They are ataxic, although no obvious abnormality of cerebellar histology is evident in the light microscope. However, when the authors examined the Cbln1 knockout cerebellum at the electron microscope level, a striking phenotype emerged: 78% of the distal spines that are the site of contact for parallel fibers on Purkinje cells lacked a presynaptic partner. Moreover, detailed analysis of serial sections by electron microscopy showed that a significant fraction of the remaining synapses had mismatched pre- and postsynaptic specializations. Electrophysiological studies confirmed this result, demonstrating a deficit of transmission between parallel fibers and Purkinje cells. This phenotype is a pleasant surprise, because it is similar to that of the glutamate receptor δ2 (Grid2) knockout mouse. The function of the Grid2 gene, which is specifically expressed in cerebellar Purkinje cells, is not yet fully understood. However, strong evi-


Rebecca Henretta


NEWS AND VIEWS

dence that Grid2 is involved in stabilization of PF-PC synapses has accumulated from detailed analysis of the Grid2-null phenotype during development7 and from the observation that inactivation of the Grid2 gene in adult animals also leads to detachment of the parallel fibers from Purkinje cell spines and mismatching of the pre- and postsynaptic specializations8. These results provided strong evidence for the existence of a specific pathway dedicated to governing the structural stability of synapses in the brain, rather than their formation. Given the phenotypic similarities between Cbln1 and Grid2 knockout mice, the authors produced and analyzed Cbln1/Grid2 doubleknockout animals to test the involvement of these two genes in a common pathway. The phenotype of the double mutant animals was identical to that of the Grid2 knockout, demonstrating that these two genes do indeed function as part of the same signaling pathway. Taken together, these data strongly implicate Cbln1 in synapse stabilization in the cerebellum, although an additional role for Cbln1 in the development of synapses seems probable.

1417


NEWS AND VIEWS Resolution of this important issue must await further experimentation. The extensive similarities between the Cbln1 and Grid2 knockout mice, including deficits in cerebellar long-term depression4,9 and genetic evidence that these two genes act in the same pathway, suggested that Cbln1 acts with Grid2 in the postsynaptic density of PC spines. However, Hirai and colleagues show—using a combination of in vitro studies, hybridization in situ and a transgenic approach—that Cbln1 is a glycoprotein secreted by granule cells, and not by Purkinje cells. Thus, Cbln1 and Grid2 are part of a transsynaptic pathway controlling the stability and plasticity of the PF-PC synapse. The transsynaptic action of Cbln1 is particularly interesting given the biochemical properties of other C1q/TNFα family proteins10. These proteins form large complexes, which interact with a variety of molecules through their globular C1q domains. They often are important in tissue remodeling. In some cases, this action requires associated protease activities—certainly a handy feature if one hopes to remodel extracellular structures that promote synapse stability. Furthermore, the demonstration that Grid2 is tethered to a molecular complex regulating autophagy11 and that activation of Grid2 receptors in Lurcher mice results in stimulation of this important catabolic pathway11,12 suggests an intracellular mechanism through which the Cbln1/Grid2 pathway could effect changes in PF-PC synaptic structure. A very interesting parallel can be made with the neuromuscular junction. Agrin is a glycoprotein secreted by motoneurons at the neuro-

muscular junction13. It was thought to promote synapse formation by clustering acetylcholine receptors. Indeed, agrin knockout mice have very few acetylcholine receptor clusters and lack neuromuscular junctions. However, this phenotype can be rescued by inactivating the gene encoding choline acetyltransferase in those same mice, thus preventing acetylcholine receptor activation. These results, together with in vitro studies, have indicated that the role of agrin is to inhibit the destabilizing effect of the activation of acetylcholine receptors14,15. Given the results of Hirai et al., and previous studies of Grid2, a similar model (Fig. 1) can be proposed for the roles of the Cbln1/Grid2 signaling pathway at the PF-PC synapse. Grid2 seems to have a dual function. Its regulation of autophagy suggests a destabilizing role for this receptor in PF-PC synapses, whereas the phenotype of the Grid2 knockout suggests that it is necessary for synapse stabilization and maintenance. One role for Cbln1 secretion from PF boutons may be to locally inhibit the destabilizing action of Grid2 receptors, as agrin inhibits the destabilizing effects of nAChR activity at the neuromuscular junction. This would promote stabilization of PC-PF contacts immediately adjacent to the active zone. Outside of the active zone, which is presumably free of Cbln1 owing to its limited ability to diffuse out of the cleft, Grid2 would retain its destabilizing actions. This would explain the phenotypes observed by Hirai et al. as well as the role of this new pathway in promoting both stabilization of the synaptic contact and matching of the pre- and postsynaptic specializations.

Although a great deal of additional information will be required to understand in detail the mechanisms regulating the structural integrity of central synapses, Hirai et al. have made an important step in identifying a mechanism for synapse stabilization that operates transsynaptically in the brain. Their studies reveal a common logic for this important process at the neuromuscular junction and at central synapses, and suggest that the transsynaptic actions of large, secreted glycoproteins on neurotransmitter receptors may provide a key function for structural remodeling of these critical CNS structures. 1. Zoghbi, H.Y. Science 302, 826–830 (2003). 2. Chklovskii, D.B., Mel, B.W. & Svoboda, K. Nature 431, 782–788 (2004). 3. Hua, J.Y. & Smith, S.J. Nat. Neurosci. 7, 327–332 (2004). 4. Hirai, H. et al. Nat. Neurosci. 8, 1534–1541 (2005). 5. Slemmon, J.R., Danho, W., Hempstead, J.L. & Morgan, J.I. Proc. Natl. Acad. Sci. USA 82, 7145–7148 (1985). 6. Bao, D., Pang, Z. & Morgan, J.I. J. Neurochem. 95, 618–629 (2005). 7. Kurihara, H. et al. J. Neurosci. 17, 9613–9623 (1997). 8. Takeuchi, T. et al. J. Neurosci. 25, 2146–2156 (2005). 9. Kashiwabuchi, N. et al. Cell 81, 245–252 (1995). 10. Kishore, U. et al. Trends Immunol. 25, 551–561 (2004). 11. Yue, Z. et al. Neuron 35, 921–933 (2002). 12. Selimi, F. et al. Neuron 37, 813–829 (2003). 13. McMahan, U.J. et al. Curr. Opin. Cell Biol. 4, 869–874 (1992). 14. Misgeld, T., Kummer, T.T., Lichtman, J.W. & Sanes, J.R. Proc. Natl. Acad. Sci. USA 102, 11088–11093 (2005). 15. Lin, W. et al. Neuron 46, 569–579 (2005).

Synaptic plasticity and self-organization in the hippocampus György Buzsáki & James J Chrobak A new paper reports that long-term potentiation in the hippocampus, a model of learning and memory, can induce sharp wave-ripple complexes, which are thought to be critical for the stabilization of memory traces in cortex. After buying a new cell phone, we quickly transfer our phone book to the new gadget György Buzsáki is at the Center for Molecular and Behavioral Neuroscience, Rutgers, the State University of New Jersey, 197 University Avenue, Newark, New Jersey 07102, USA. James Chrobak is at the Department of Psychology, University of Connecticut, 406 Babbidge Road, Storrs, Connecticut 06269, USA. e-mail: [email protected]

1418

and relegate the old instrument to the recycling bin. Likewise, sometime after an event, memories that initially depend on the activity of hippocampal neuronal assemblies are transferred to and consolidated in the neocortex and no longer depend on the hippocampus. How does this change take place, and how do patterns of activity within hippocampal cell assemblies transfer information to the neocortex and consolidate it there? And how can cell assemblies produce the patterns of neuronal

discharge required to induce synaptic change? Self-organized population discharges in the hippocampus such as hippocampal sharp wave-ripple (SPW-R) complexes, mainly seen in vivo, are thought to represent stored information that is then transferred to the neocortex. Nonetheless, the mechanisms responsible for the induction of these SPW-R complexes are unclear. Now Behrens and colleagues1 report that they can induce in vitro SPW-R complexes,

VOLUME 8 | NUMBER 11 | NOVEMBER 2005 NATURE NEUROSCIENCE

similar to the fast-frequency ensemble patterns commonly seen in vivo. Moreover, SPW-R can be induced with stimulation protocols known to induce LTP, a popular neurophysiological model of learning and memory. The authors also demonstrate that these events induce synaptic change among CA3 neurons. Thus, they have established a link between the induction of LTP and the emergence of a physiological network pattern believed to be involved in shaping memories. Their report takes one big step in bridging the chasm separating synaptic mechanisms studied in vitro and the consolidation of memory traces in the intact brain. The hippocampal SPW-R complex has features that make it a candidate pattern for the consolidation of synaptic plasticity and the transfer of neuronal patterns2. Importantly, it also has a widespread effect. In the approximately 100-ms time window of a hippocampal SPW, between 50,000 and 100,000 neurons—10–20% of the total neuronal population of the rat hippocampus—discharge simultaneously in the CA3-CA1–subicular complex–entorhinal axis, qualifying it as the most synchronous network pattern in the brain3 (Fig. 1). The SPW-R complex arises in the recurrent collateral system of the CA3 region and spreads downstream. Although its recruitment dynamics are delicately controlled by various classes of interneurons4, a three- to fivefold gain of network excitability is achieved transiently5. The observation that the SPW-R is shaped by previous experience6 gives further support for the fundamental role of these population patterns. However, there are several missing links in the story, including how SPW-R complexes emerge and whether this pattern is actually accompanied by changes in synaptic connectivity. Behrens and colleagues1 demonstrate that stimuli that induce LTP lead to the generation of SPW-R complexes in slices of the dorsal hippocampus of the rat. Further, the induction, but not the expression, of SPW-R complexes is NMDA receptor– dependent. They also provide evidence that the induction of SPW-R complexes is paralleled by changes in both excitation and inhibition in the CA3 region. SPW-R complexes in vitro have been observed only in the mouse hippocampus and from the ventral hippocampus of the rat7,8 and the mouse9,10. One possible explanation for the absence of spontaneous SPW-R in slices of the rat dorsal hippocampus is that the density of recurrent axon collaterals and the mutual excitation are simply not sufficient to bring about a self-organized population burst. If so, strengthening the surviving synapses could rescue the compromised circuit. This is precisely what Behrens and colleagues1

Figure 1 Selforganized burst of activity in the hippocampal CA3 region produces a CA1 field potential in the dendritic layer of Sub CA1 and a short-lived fast-frequency field EC oscillation (200-Hz ripple) within stratum CA3 pyramidale, as well as a phase-related discharge of the neurons. Hippocampal Para output, in turn, produces similar sharp wave-ripple complexes in the subiculum (‘Sub’), parasubiculum (‘Para’) and deep layers of the entorhinal cortex (‘EC’). Behrens and colleagues1 show that the rules of synaptic plasticity govern the emergence and the recruitment of particular cell groups in these hippocampal output events.

Ann Thomson


NEWS AND VIEWS

have found. During repeated stimulus trains (400 ms at 100 Hz), repeated every 40 s, SPW-R complexes gradually increased in incidence and amplitude after the third to fifth stimulus. Trains that failed to induce LTP failed to induce SPW-R. After approximately the fifth LTP train, the incidence of SPW-Rs reached a plateau, perhaps because the synapses involved in the generation of the spontaneous patterns were ‘saturated’—that is, they reached their maximum possible strength. Both MK-801 (a noncompetitive antagonist of the NMDA subtype of glutamate receptor) and D-AP5 (a competitive NMDA receptor antagonist) could prevent the induction of SPW-R complexes. However, once these complexes were established, these drugs did not prevent SPW-Rs. Indeed, the incidence of established SPW-R incidence increased in the presence of these drugs, probably via mechanisms that involve decreased calcium influx through NMDA receptors and a subsequent reduction in the activation of SK2 calcium-activated potassium channels11. Importantly, the established SPW-R events were reduced or abolished by low-frequency stimulation, a protocol that induces long-term depression of synapses. The blockade of gap junctions with carbenoxolone also attenuated ripple occurrence, in accordance with earlier observations of SPW-R complexes in vitro12. In short, the spontaneously emerging SPW-R complexes show a strong parallel with synaptic changes observed in vivo13. Using combined intracellular and extracellular recordings, Behrens and colleagues also examined intracellular responses in pyramidal cells during the development of SPW-R complexes. They observed extracellular post-


synaptic potential–intracellular postsynaptic potential (EPSP-IPSP) sequences, IPSP-EPSP sequences and prominent IPSPs, but they never found isolated EPSPs. Thus, inhibitory inputs are prominent in the development of ripple complexes and may contribute to the temporal precision of EPSP-spike coupling14,15. Importantly, the pattern of inhibition-excitation in any particular neuron remained stable during the development of SPW-Rs in individual neurons; this indicated that the discharge sequence of the participating neurons during the SPW-R complexes is determined largely by a unique distribution of synaptic strengths at both excitatory and inhibitory connections2,13. Although these findings show substantial homology between a hippocampus-generated event in vivo and population events in vitro, some differences are worth pointing out. First, SPW-Rs are associated with large population events and ripple oscillations in both the CA1 and CA3 pyramidal layer in vitro. In contrast, the in-vivo ripple event in CA1 reflects a convergence of small-amplitude excitatory inputs from a large area of the CA3 region, without synchronous ripple oscillations. This difference may be interpreted as an artificial augmentation of excitation of the truncated CA3 collateral circuitry, reminiscent of epileptiform activity. Second, in contrast to the regularly occurring and uniformly sized SPW-Rs in the slice, their invivo counterparts are very irregular in both their temporal distribution and magnitude. Nevertheless, the replication of an endogenous brain pattern in vitro allows for the investigation of a number of important

1419


NEWS AND VIEWS mechanisms that are difficult to explore in vivo. For example, using simultaneous intracellular recordings from two or more neurons, future experiments will be able to reveal if the induced but otherwise self-generated patterns involve stable recruitment mechanisms. If so, such findings would provide evidence for the hypothesis that endogenous patterns preserve the information about the perturbations that gave rise to the patterns. Monitoring large numbers of neurons and modifying targeted parts of the circuit may identify the elementary mechanisms involved in the consolidation of network patterns. The finding that such modi-

fications can be brought about by stimulation protocols that induce LTP and can be altered by those that produce long-term depression is even more exciting. Memories may really be made in the hippocampus, and SPW-R complexes may contribute to the process after all. 1. Behrens, J., van den Boom, L P., de Hoz, L., Friedman, A. & Heinemann, U. Nat. Neurosci. 8, 1560–1567 (2005). 2. Buzsaki, G. Neuroscience 31, 551–570 (1989). 3. Chrobak, J.J. & Buzsaki, G. J. Neurosci. 14, 6160– 6170 (1994). 4. Klausberger, T. et al. Nature 421, 844–848 (2003). 5. Csicsvari, J., Hirase, H., Czurko, A., Mamiya, A. & Buzsaki, G. J. Neurosci. 19, 274–287 (1999). 6. Wilson, M.A. & McNaughton, B.L. Science 265, 676–

679 (1994). 7. Papatheodoropoulos, C. & Kostopoulos, G. Brain Res. Bull. 57, 187–193 (2002). 8. Kubota, D., Colgin, L.L., Casale, M., Brucher, F.A. & Lynch, G. J. Neurophysiol. 89, 81–89 (2003). 9. Yanovsky, Y., Brankack, J. & Haas, H.L. Neuroscience 64, 319–325 (1995). 10. Maier, N. Nimmrich, & Draguhn, A. J. Physiol. (Lond.) 550, 873–887 (2003). 11. Colgin, L.L., Jia, Y., Sabatier, J.M. & Lynch, G. Neurosci. Lett. 385, 46–51 (2005). 12. LeBeau, F.E., Traub, R.D., Monyer, H., Whittington, M.A. & Buhl, E.H. Brain Res. Bull. 62, 3–13 (2003). 13. King, C., Henze, D.A., Leinekugel, X. & Buzsaki, G. J. Physiol. (Lond.) 521, 159–167 (1999). 14. Axmacher, N. & Miles, R. J. Physiol. (Lond.) 555, 713–725 (2004). 15. Pouille, F. & Scanziani, M. Science 293, 1159–1163 (2001).

Glial cells under remote control Klaus-Armin Nave & Markus H Schwab Not all axons in a peripheral nerve are myelinated. A recent study shows that the expression of neuregulin-1 on an axon membrane determines whether immature Schwann cells will differentiate into myelinating Schwann cells. Like the insulation on electrical wires in your house, myelin sheaths are essential for rapid impulse propagation throughout the vertebrate nervous system. The multilayered myelin membranes are synthesized by highly specialized glial cells, termed oligodendrocytes, in the CNS and Schwann cells in the PNS. In the PNS, the decision of Schwann cells to myelinate resembles a cell lineage decision that is triggered by axonal signals, but the nature of these signals has remained unclear. Recently in Neuron1, Taveggia et al. demonstrated that axonal neuregulin-1 type III, a regulator of myelin growth, is required to induce the differentiation of myelinating Schwann cells in dorsal root ganglion (DRG) and superior cervical ganglion (SCG) explant cultures. Ever since electron microscopists demonstrated the complexity of compact myelin sheaths, cellular neurobiologists have been hooked on the myelin-forming glia, their complex interaction with axons and their spectacular membrane growth. Numerous proteins of myelin and the axon-glia junction have been identified; nevertheless, major questions remain unanswered. We do not know the driving force of glial ensheathment. Crosssections suggest that myelin assembly is a spiral membrane growth process, but this has not been verified. It is unclear why axons larger than Klaus-Armin Nave and Markus Schwab are at the Department of Neurogenetics, Max Planck Institute of Experimental Medicine, Hermann-Rein-Str. 3, 37075 Göttingen, Germany. e-mail: [email protected]

1420

1 µm are myelinated, but small-caliber axons are only ensheathed, and dendrites seem not to interact at all with oligodendroglia. We do not know what signals provide specificity of axonglia interaction and instruct glia to myelinate, or whether these mechanisms are the same for oligodendrocytes and Schwann cells. One would expect neurons and myelinating glia to communicate through a complex assembly of developmentally regulated signaling proteins. It is therefore surprising that, at least in the PNS, one signaling system, comprised of axonal neuregulin-1 and glial ErbB receptors, seems to operate at all stages of Schwann cell development and myelination. Neuregulin-1 (Nrg1) comprises a family of more than 15 membrane-associated and secreted growth factors that are derived from a single gene by alternative splicing and promotor use. Major subgroups are Nrg1 type I (including proteins named NDF, heregulin, and ARIA) and type II (‘glial growth factor’ or GGF) isoforms, both of which are potentially secreted or shed upon proteolytic cleavage. In contrast, Nrg1 type III has a second transmembrane domain and remains a membrane-associated ligand2. Common to all isoforms is an EGF-like domain that activates ErbB receptor tyrosine kinases. In the developing nerve, ErbB2 and ErbB3 are expressed at the cell surface of Schwann cell progenitors and are essential for their survival and subsequent differentiation3. Nrg1 signaling also contributes to synaptogenesis (at least in vitro), the migration of cortical interneurons and cardiac development in the embryo. The latter has greatly hampered the

conventional genetic analysis of Nrg1 function in the postnatal nervous system. Taveggia et al.1 now provide experimental evidence that neuregulin-1 type III is necessary for myelination in the PNS and can instruct immature ensheathing cells to become true myelinforming Schwann cells. These in vitro findings close a gap between related in vivo findings by other groups, including the requirement of neuregulin-1 for the survival of precursor cells and immature Schwann cells4, the requirement of glial ErbB2 receptors for normal myelination in conditional mouse mutants5, and the identification of neuregulin-1 type III as an axonal signal that regulates myelin sheath thickness in mice with altered neuregulin gene dosage6. Mice that selectively lack Nrg1 type III die perinatally and have a marked decrease in Schwann cells and degenerating motor and sensory nerves7. To study the competence of these mutant axons to be myelinated, Taveggia et al. used a coculture system in which in vitro myelination is initiated by the addition of ascorbate8. Sensory DRG neurons from wild-type and Nrg1 type III–deficient mice were cocultured with Schwann cells from normal rats. The Nrg1 type III mutant axons never became myelinated, even in the presence of a fivefold excess of Schwann cells. Lentiviral expression of Nrg1 type III was sufficient to restore myelination competence. In agreement with a report of hypermyelination in mice that overexpress Nrg1 type III, but not type I (ref. 6), multiple examples of unusually thick myelin profiles were seen in ‘rescued’ cocultures.


Because Nrg1/ErbB2 signaling is required for myelination in vivo5, the authors then asked whether normally unmyelinated axons would become myelinated if they ectopically expressed Nrg1 type III. By transducing cultured SCG neurons (with axons that are normally ensheathed by Schwann cells, but not myelinated), they could show that indeed low levels of Nrg1 type III are sufficient to induce ectopic myelination. This suggests that increasing Nrg1 expression in neurons above a certain level may confer myelination competence. An important next question is whether insufficient Nrg1 expression can also explain why axons smaller than 1 µm in diameter (such as C-fibers devoted to pain perception) are typically unmyelinated, or whether axon size and membrane curvature pose neuregulin-independent thresholds for myelination. What are the effector molecules in neuregulin-induced Schwann cells? The inhibition of PI3 kinase blocks myelin formation in cocultures9. PI3 kinase and MAP kinase in Schwann cells are activated by GGF, a soluble type II neuregulin, or by contact with neurite membranes. Now Taveggia et al. extend these observations by showing that neurite membranes from wild-type (but not from Nrg1 type III–deficient) neurons activate the PI3 kinase pathway. In contrast, MAPK activation was not impaired in mutant cultures. Thus, Nrg1 type III emerges as a critical activator of glial PI3 kinase, whereas other signaling molecules underlie the activation of MAP kinases. These other molecules may or may not include the secreted neuregulin isoforms. Assuming that the paracrine neuregulins (type I and II) are functionally not equivalent to the juxtacrine type III (localized to the axon surface), the authors asked whether a type III protein promotes myelination when offered as a paracrine signal. To address that issue, they added a recombinant type III ectodomain to the coculture system. The soluble protein retained mitogenic activity, but did not support myelination. The authors also tested an ectopic juxtacrine source of Nrg1 type III. Schwann cells were cultured under myelinating conditions on a monolayer of CHO cells that stably expressed Nrg1 type III on the surface. Again, the factor retained some mitogenic potential, but Schwann cell differentiation (as determined by myelin gene expression) could not be induced. These observations suggest that neurons produce multiple neuregulin-1 isoforms, but only type III—in concert with other axonal signals—promotes ensheathment and myelination as a contact-dependent signal. Taveggia et al. also analyzed heterozygous Nrg1 type III mutant mice, which are phenotypically normal7. In agreement with previous work6, they

Neuronal cell bodies

Axon

1

Mitotic Schwann cell progenitors

Immature Schwann cell 2 3

Remak bundle

4 5

Debbie Marzels


NEWS AND VIEWS

Neuregulin Axon surface Myelinating Schwann cell

Figure 1 Neuregulin-1 directs Schwann cell differentiation. In peripheral nerves, neural crest–derived Schwann cell progenitors (1) proliferate and populate axon bundles. Later, as immature Schwann cells (2), they face a binary choice: they either stay tightly associated with several axons to form a Remak bundle (3), or alternatively they single out larger axons and differentiate into myelinating Schwann cells (4). Work in several laboratories indicates that the entire path of Schwann cell development and myelination is remote controlled by the neurons through expression of the neuregulin-1 (5) on the axonal surface.

found that the sciatic nerves of adult mice are hypomyelinated and have reduced nerve conduction velocity. Additionally, the authors observed a higher number of small caliber axons (C-fibers) grouped into so-called Remak bundles. The latter suggests that Nrg1 type III also signals between C-fibers and ‘non-myelinating’ Schwann cells. Indeed, the importance of this interaction has been demonstrated before by dominant-negative perturbation of ErbB receptor function, which causes disorganization of Remak bundles and peripheral neuropathy in transgenic mice10. Most likely, the conclusions presented here will be rederived, and possibly modified, in vivo using conditional mutants and transgenic mice. But when the new data1 are combined with those from other studies3,5,6,10 an unexpected picture emerges already: evolution of the nervous system has recruited a neuregulin-based signaling system for virtually all steps of Schwann cell differentiation (Fig. 1), including the fine-tuning of myelin


growth. It will be most interesting to extend this line of research into the CNS. Like Schwann cells, oligodendrocytes respond to Nrg1 (refs. 11,12). Myelination control in the brain and spinal cord differs with respect to other growth factors13, and preliminary evidence suggests a more complex regulation by different Nrg1 isoforms (B. Brinkmann, M.H.S. and K.A.N., unpublished data). This research may also have clinical implications, if axonal Nrg1 contributes to remyelination (or the lack thereof) in multiple sclerosis. From a neuronal point of view, glial cell differentiation emerges as largely remote controlled. If quantitative differences of neuronal Nrg1 gene expression determine Schwann cell fate, even myelin thickness, then how does a neuron know its own size, or how much neuregulin to make? It should matter whether an axon measures 1 or 10 µm in diameter, and whether it is 1 mm or 1 meter long. This issue of whether axon length and caliber control neuronal gene expression

1421

NEWS AND VIEWS may be relevant for many more proteins, but it has never been recognized as such. Once again, glial cells have evoked important questions.


1. Taveggia, C. et al. Neuron 47, 681–694 (2005). 2. Falls, D.L. Exp. Cell Res. 284, 14–30 (2003). 3. Jessen, K.R. & Mirsky, R. Nat. Rev. Neurosci. 6, 671–

682 (2005). 4. Meyer, D. & Birchmeier, C. Nature 378, 386–390 (1995). 5. Garratt, A.N., Voiculescu, O., Topilko, P., Charnay, P. & Birchmeier, C. J. Cell Biol. 148, 1035–1046 (2000). 6. Michailov, G.V. et al. Science 304, 700–703 (2004). 7. Wolpowitz, D. et al. Neuron 25, 79–91 (2000). 8. Eldridge, C.F., Bunge, M.B., Bunge, R.P. & Wood, P.M.

J. Cell Biol. 105, 1023–1034 (1987). 9. Maurel, P. & Salzer, J.L. J. Neurosci. 20, 4635–4645 (2000). 10. Chen, S. et al. Nat. Neurosci. 6, 1186–1193 (2003). 11. Park, S.K., Miller, R., Krane, I. & Vartanian, T. J. Cell Biol. 154, 1245–1258 (2001). 12. Fernandez, P.A. et al. Neuron 28, 81–90 (2000). 13. Chan, J.R. et al. Neuron 43, 183–191 (2004).

Senseless makes sense for spinocerebellar ataxia-1 Vikram Khurana, Tudor A Fulga & Mel B Feany Why are some neurons selectively targeted for death in neurodegenerative diseases? A recent paper combines genetics in the fruit fly and mouse to uncover mechanisms underlying the vulnerability of Purkinje cells in spinocerebellar ataxia-1. Neurodegenerative diseases share many features, including a progressive loss of neurons and the formation of proteinaceous aggregates. These similarities have motivated research into common underlying pathogenic processes, including dysfunction of the ubiquitin-proteasome system, impaired axonal transport and oxidative stress. A recent paper in Cell by Hiroshi Tsuda and colleagues1 reminds us that neurodegenerative diseases also have important features that distinguish them from one another, including the selective vulnerability of particular groups of neurons. The authors focus on a type of spinocerebellar ataxia (SCA). SCAs are debilitating neurodegenerative diseases characterized by progressive gait incoordination and cerebellar atrophy. Tsuda et al. delineate a physical and functional interaction between the AXH domain of ataxin-1, a protein of unknown function, and the transcription factor known as Senseless in Drosophila melanogaster and Gfi-1 in vertebrates. The authors provide compelling evidence in animal models that this interaction contributes to the progressive demise of Purkinje neurons in SCA-1. Autosomal-dominant polyglutamine (polyQ) expansion disorders, including Huntington disease and a number of SCAs, are all caused by the expansion of unstable CAG repeat sequences within the coding region of the causative gene2. Neurodegeneration accompanies the intraneuronal aggregation of the polyQ-expanded proteins in each disease. The dominant mode of inheritance, together with the recapitulation of disease phenotypes in overexpression but not knock-

The authors are at the Department of Pathology at Brigham and Women’s Hospital and Harvard Medical School, 77 Louis Pasteur Ave., Boston, Massachusetts 02115, USA. e-mail: [email protected]

1422

out animal models, suggests a toxic gain-offunction mechanism whereby the expanded polyQ tract confers new molecular functions upon the causative protein. Notably, however, despite ubiquitous expression in the nervous system, only certain neuronal groups are targeted for death in these diseases. Furthermore, differing polyQ repeat lengths are required to initiate neurodegeneration in the different diseases. These differences strongly implicate sequences outside the CAG repeat region in disease pathogenesis. Tsuda et al.1 have now taken us a significant step closer to understanding the unique features of ataxin-1 that mediate degeneration of Purkinje cells in SCA-1. This disease is caused by a polyglutamine expansion in ataxin-1 and accompanied by nuclear aggregation of this protein in neurons. The authors previously showed that overexpressing human ataxin-1 (hAtx-1) with an expanded polyQ tract in Drosophila resulted in neurodegeneration3; further, they showed that phosphorylation of Ser776 by the kinase Akt is critical to toxicity by enabling an interaction with 14-3-3 and increasing hAtx-1 stability4. Tsuda et al.1 now report that expressing the fly homolog of ataxin-1 (dAtx-1), but not a polyQ repeat alone, recapitulates hAtx-1–induced phenotypes in different fly tissues, albeit with reduced severity. Intriguingly, dAtx-1 does not contain a polyQ domain but shares an AXH (ataxin-1/HBP1) domain with hAtx-1, a domain recently implicated in RNA binding and self-association5. The authors further demonstrate that dAtx-1 physically interacts with the transcription factor Senseless (Sens) by means of this domain. An in vitro transcriptional assay and a functional analysis in the fly reveal that both Sens activity and protein abundance are downregulated by dAtx-1. Furthermore, expressing hAtx-1 with an expanded polyQ tract reduces Sens levels more potently than dAtx-1, whereas over-

expressing the polyQ tract alone, or polyQexpanded hAtx-1 with the AXH domain deleted, has no effect on Sens levels. The study proceeds with a logical series of experiments relating the findings in Drosophila to a mouse model system, thus strengthening the relevance of AXH domain interactions to the human disease. The authors show that hAtx-1 binds to the vertebrate homolog of Sens, Gfi-1, also through its AXH domain. Importantly, Gfi-1 is maximally expressed in the nervous system within the Purkinje neurons of the cerebellum, one group of neurons that selectively degenerate in SCA-1. In mammalian cells, as in flies, Gfi-1 levels are downregulated by hAtx-1, an effect that is post-translational and depends on the ubiquitin-proteasome system. Significantly, these findings are recapitulated in a mouse model of SCA- 1, where expression of polyQexpanded hAtx-1 in Purkinje cells leads to an early decrease in the abundance of Gfi-1, preceding Purkinje cell loss and ataxia. Furthermore, removing a single copy of Gfi1 enhances the neurodegeneration phenotype in this model. In a final elegant proof-of-principle experiment, the authors show that a progressive loss of Purkinje neurons accompanies ataxia in Gfi1 knockout mice, thus demonstrating that decreased Gfi-1 is sufficient to cause Purkinje cell loss. These findings are important at multiple levels. Most directly, they suggest that the interaction between the AXH domain of hAtx-1 and Gfi-1 is important for mediating neurodegeneration and that this is potentially a therapeutic target. However, although the authors show that reducing Gfi-1 levels results in an enhancement of hAtx-1–induced neurodegeneration, establishing whether ectopic expression of Sens or Gfi-1 rescues neurodegeneration in flies or mice, respectively, would further strengthen the case for Gfi-1 stabilization as a potential treatment. Beyond


the direct therapeutic implications, however, this study provides critical evidence implicating sequences outside the polyQ region in the selectivity of neurodegeneration in polyQ disorders. Future studies might explore whether the hAtx-1–Gfi1-1 interaction mediates neurodegeneration in other cell populations vulnerable in SCA1, including the inferior olivary nucleus or spinocerebellar tracts, or whether other interactions are involved. In this regard it would be interesting to determine if, in addition to Purkinje neurons, these populations are also vulnerable in Gfi-1–null mice. The methodology used by Tsuda et al.1 also deserves attention. Whereas several models of autosomal-dominant neurodegenerative diseases have been made by transgenic overexpression of causative human genes in Drosophila6, the present study adopts the normal fly protein as a starting point. The observation that expression of an expanded polyQ tract alone does not phenocopy certain phenotypes shared by dAtx-1 and hAtx-1–polyQ leads the authors to infer the existence of functionally important sequences outside the polyQ region. Flies certainly provide an ideal system to make such comparisons. Further, by recapitulating the biochemical and functional interactions in the mouse model system, the study supports the utility of Drosophila in modeling human diseases. Indeed, there are fly homologs for proteins, such as tau, that are involved in other neurodegenerative diseases, raising the possibility that such an approach might be fruitful for these diseases also. How do we place the present findings in the context of what is known about the pathogenesis of SCA-1 and related disorders? PolyQ expansion clearly initiates the disease process. The model proposed in this study would implicate this expansion in the stabilization of hAtx-1, abnormally potentiating the AXH domain–Gfi-1 interaction. Neurotoxicity would follow from proteasomal degradation of Gfi-1 and transcriptional dysregulation (Fig. 1). Here, toxic gain-of-function is caused, not by the protein attaining an entirely aberrant function, but rather from the abnormal activation of a physiological pathway. In contrast, many previous studies have concentrated on abnormal interactions mediated by the polyQ tracts themselves. Because the nuclear localization of causative proteins is essential for neurotoxicity7, these studies have focused on abnormal effects on transcription, either directly or through sequestration of transcription factors8. For example, the polyQ tract of hAtx-1 binds to the nuclear protein PQBP1 in a manner dependent on polyQ tract length9. A resultant complex forming between hAtx-1, PQBP-1 and RNA polymerase II led to

Mouse Purkinje neurons AXH

PolyQ

PolyQ aggregation

Nucleus

hAtx1 Gfi-1 AXH

Gfi-1

Neurotoxicity Proteasome

dAtx1

Sens

AXH Nucleus

Drosophila

Ann Thompson


NEWS AND VIEWS

Figure 1 Tsuda et al. demonstrate homologous pathways that mediate ataxin-1–induced neurodegeneration in mouse and Drosophila models of SCA-1. In Drosophila, dAtx-1 (which lacks polyQ repeats) binds, through its AXH domain, to the transcription factor Senseless and targets it for proteasomal degradation. A homologous interaction, potentiated by abnormal polyQ expansion and aggregation, occurs between hAtx-1 and Gfi-1 in mouse Purkinje neurons. The resultant transcriptional dysregulation mediates neurotoxicity in both model systems. Direct dysregulation of transcription by polyQ-expanded repeats may also make an important contribution to neurotoxicity9.

a decrease in basal transcription. Intriguingly, PQBP-1 is enriched in the cerebellum, and a polyQ-dependent interaction could therefore contribute to selective neuronal vulnerability in SCA-1 (ref. 10). Other groups have provided data supporting different toxic gain-of-function mechanisms in polyQ-associated diseases, including disruption of axonal transport and downregulation of survival pathways11. Wild-type ataxin-3 suppresses degeneration in multiple polyQ models in flies by proteasomal activation, implying not only that common mechanisms might operate in different polyQ-associated diseases, but that loss-of-function mechanisms might also be involved12. Taking these studies together, a picture emerges of both the common mechanisms in polyQ expansion disorders that could be mediated by the polyQ tracts themselves and the distinct effects that could be attributable to non–polyQ-encoding sequences. In keeping with this idea, transcriptional profiling and microarray studies reveal both common and distinct changes in different polyQ animal models13. A clear challenge for future studies is to try and integrate the multiple pathways downstream of polyQ expansion into a cohesive picture. For example, in SCA-1, do the implicated transcription factors function as


a network influencing common downstream processes? Gfi-1, for example, downregulates proapoptotic genes14. Is it possible that downregulation of Gfi-1 by hAtx-1 leads to neuronal apoptosis? Could dysregulation of other transcription factors converge on apoptosis also? To guide investigations into events downstream of transcriptional dysregulation, it would clearly be useful to define the molecular pathways mediating cell death in these diseases, whether apoptotic or non-apoptotic. At present, the mechanisms of cell death in SCA-1 remain unclear, with neurodegeneration in the SCA-1 mouse seeming to be p53 dependent but not classically apoptotic15. In summary, Tsuda et al. present us with an important and thought-provoking study that provides hope for targeted SCA-1 therapies in the future. Establishing the direct relevance of a physiological protein interaction to disease pathogenesis has implications extending beyond SCA-1 and polyQ disorders to other neurodegenerative diseases for which toxic gain-of-function mechanisms have been proposed, including familial Alzheimer and Parkinson diseases. For mice and fruit flies, the message is loud and clear: resolving the similarities and differences among related neurodegenerative diseases should guarantee employment for many generations to come.

1423

NEWS AND VIEWS


1. Tsuda, H. et al. Cell 122, 633–644 (2005). 2. Taroni, F. & DiDonato, S. Nat. Rev. Neurosci. 5, 641– 655 (2004). 3. Fernandez-Funez, P. et al. Nature 408, 101–106 (2000). 4. Chen, H.K. et al. Cell 113, 457–468 (2003). 5. de Chiara, C. et al. FEBS Lett. 551, 107–112 (2003). 6. Muqit, M.M. & Feany, M.B. Nat. Rev. Neurosci. 3,

237–243 (2002). 7. Klement, I.A. et al. Cell 95, 41–53 (1998). 8. Margolis, R.L. & Ross, C.A. Trends Mol. Med. 7, 479– 482 (2001). 9. Okazawa, H. et al. Neuron 34, 701–713 (2002). 10. Humbert, S. & Saudou, F. Neuron 34, 669–670 (2002). 11. Lipinski, M.M. & Yuan, J. Curr. Opin. Pharmacol. 4,

85–90 (2004). 12. Warrick, J.M. et al. Mol. Cell 18, 37–48 (2005). 13. Sugars, K.L. & Rubinsztein, D.C. Trends Genet. 19, 233–238 (2003). 14. Jafar-Nejad, H. & Bellen, H.J. Mol. Cell. Biol. 24, 8803–8812 (2004). 15. Shahbazian, M.D., Orr, H.T. & Zoghbi, H.Y. Neurobiol. Dis. 8, 974–981 (2001).

How the brain recovers following damage Yalçin Abdullaev & Michael I Posner Individuals with neglect fail to process stimuli on the left. A new paper uses functional imaging to show that a restricted lesion, usually caused by a stroke, may influence the network of areas associated with attention shifts. After right-hemisphere stroke, some people see objects in their world as having no left side (Fig. 1). In the acute stage immediately following the stroke, such individuals with ‘spatial neglect’ may fail to orient to people approaching from their left, to recognize their left arm as their own and to eat from the left side of their plate. How can a stroke to a local area of the brain be associated with such a mysterious array of symptoms? In this issue, Corbetta et al.1 report that the activity of an interconnected network of dorsal, ventral parietal and frontal areas may be influenced by comparatively restricted lesions of the ventral right hemisphere. The dorsal network includes the superior parietal lobe and the frontal eye fields, and the ventral network involves the temporal parietal junction and the ventral lateral prefrontal cortex1. In normal people, imaging studies show that this interconnected network is important for shifts of spatial attention2. The authors tracked the recovery of this network in individuals with neglect through successive functional magnetic resonance imaging (fMRI) scans taken while the patients performed a task. In an fMRI scanner, participants were instructed to direct their gaze to the center of the screen in front of them. An arrow appeared at this central point, directing the subjects to attend to either the left or the right side of the screen, without moving their eyes. Participants were then asked to press a button when they detected a target. The target appeared on the side indicated by the arrow most of the time, but occasionally it appeared on the other, unattended side. Normal participants have longer reaction times when the

Yalçin Abdullaev and Michael I. Posner are in the Department of Psychology, University of Oregon, Eugene, Oregon 97403, USA. e-mail: [email protected]

1424

Figure 1 Sample drawings by individuals with spatial neglect. (a,b) The pictures demonstrate how they ignore their left visual field in trying to make simple line drawings of a clock (a) or a house (b).

target appears on the side opposite to where they are directing their attention3. People with neglect show particularly long reaction times in response to targets on the left side when they are first cued to attend to the right4. In the acute stage immediately after the stroke, they may miss such targets completely, and even after many years they have a large deficit in reaction time3. The Corbetta et al. study1 found a dramatic alteration in the pattern of activation in the parietal-frontal network four weeks after the stroke (the acute stage), even though the individual nodes showed no evidence of structural damage. Seven months later (in the chronic stage), the participants had considerably improved in their ability to orient and detect stimuli on the left. Functionally, the most striking change was that the dorsal right parietal lobe, which was not activated at all during the acute phase, was now strongly activated in the chronic phase. This was in contrast with an actual reduction of

activity in similar areas in the left hemisphere. Therefore, the dorsal parietal area, which is critical for voluntary shifts of attention in normal people2, was the only brain area that showed increased activity on the lesioned side from the acute to the chronic stage. This was accompanied by reduced activity in the non-lesioned hemisphere. This effect is likely to be responsible for the observed reduction in rightward bias from the acute to the chronic stage. How do these parietal findings relate to what is found in visually specific areas of the cortex? Several studies5 indicate that the source of attention effects (manifested as increased activity in visual cortex during stimulus detection) lies in parietal areas. The results of Corbetta et al.1 show that activity in the left visual cortex is reduced in the chronic phase as compared to the acute phase, whereas right hemisphere visual cortex activation is increased. These findings mirror those in the parietal lobe. Individuals recovering from neglect also


improve their ability to spot targets on the unattended side while they focus their attention on the side indicated by the arrow. This recovery seems to depend mostly on the right temporal parietal junction, a brain area thought to be responsible for interrupting attention and allowing a shift of attention toward the target location. This area is active only in the right hemisphere during shifts of attention to targets in either direction, whereas most other areas are symmetric6. The right lateralization of this area seems to be a major reason for the strong neglect of the left side apparent in the participants’ performance (Fig. 1). The Corbetta et al. study1 reveals that damage to the ventral areas of the spatial attention system produces neglect because it causes dysfunction of the dorsal system as well. Although this was not a specific focus of the current study, the results raise the question of why more dorsal lesions do not cause malfunctions of the ventral parietal area and thus also produce neglect. Apparently, there is an asymmetry between the ventral and dorsal parietal areas: lesions of the dorsal

parietal area alone do not cause neglect7. The mechanisms underlying this strong asymmetry in the remote effects of these two critical areas of the parietal lobe remain unclear. Evidence for greater than normal left hemisphere activity during the acute stage of neglect provides some basis for rehabilitation methods as well. The authors suggest that competition between hemispheres may provide the basis for helping recovery, either by increasing activity in the ipsilesional cortex or by reducing it in the contralesional cortex. Increased right hemisphere activity through warning signals, already known to reduce neglect, could serve as one method for improving hemispheric balance8. The need for hemispheric balance could be an important reason why forcing the increased use of a paralyzed limb can foster recovery9. Among patients with neglect, those with right hemisphere lesions, who show a chronic rightward bias, may also be helped by inhibiting left superior parietal lobe activity1. There are more general lessons that might be gleaned from this work. Remote effects of lesions provide some explanation of how local-

ized computations, as revealed in many functional imaging studies, may be consistent with the occurrence of a syndrome with many complex sensory and motor features. This finding may be important for all of neuropsychology. The authors also argue that the general principles of recovery that they have found might apply to aphasia or sensory motor deficits. In any case, the paper provides a strong argument for using imaging during recovery to determine the effects of various forms of therapy. 1. Corbetta, M., Kincade, M.J., Lewis, C. & Sapir, A. Nat. Neurosci. 8, 1603–1610 (2005). 2. Corbetta, M. & Shulman, G.L. Nat. Rev. Neurosci. 3, 201–215 (2002). 3. Posner, M.I. Q. J. Exp. Psychol. 32, 3–25 (1980). 4. Posner, M.I., Walker, J.A., Friedrich, F.J. & Rafal, R.D. J. Neurosci. 4, 1863–1874 (1984). 5. Hillyard, S.A., DiRusso, F. & Martinez, A. in Functional Neuroimaging of Visual Cognition (eds. Kanwisher, N. & Duncan, J.) 381–388 (Oxford Univ. Press, Oxford, 2004). 6. Perry, R.J. & Zeki, S. Brain 123, 2273–2288 (2000). 7. Friedrich, F.J., Egly, R., Rafal, R.D. & Beck, D. Neuropsychology 3, 193–207 (1998). 8. Robertson, I.H., Mattingley, J.B., Rorden, C. & Driver, J. Nature 395, 169–172 (1998). 9. Taub, E., Uswatte, G. & Elbert, T. Nat. Rev. Neurosci. 3, 228–236 (2002).

Time to smell the roses Timing is a thorny issue for the chemical senses. Principal neurons in the vertebrate olfactory bulb and insect antennal lobe have dynamic odor-evoked responses that can long outlast odorant exposure. The temporal pattern of these responses is thought to be important for distinguishing different odorants, but in a natural environment, odor stimuli have their own temporal structure. With an animal’s movements and the intermittent arrival of odors in the air, new odors are likely to interrupt responses to previous ones. Experience tells us that a second sniff of a rose still smells like a rose, but how does the olfactory system sort out these potentially conflicting time courses? On page 1568 of this issue, Mark Stopfer and colleagues address this question in the olfactory system of the locust. They exposed adult locusts to trains of brief odorant pulses with natural interpulse intervals, and made intracellular and extracellular recordings from projection neurons (PNs) in the antennal lobe. In most cases, responses varied over successive pulses, showing that the temporal structure of odorant presentation did influence the temporal pattern of PN responses. To determine the downstream consequences of this interference, the authors considered known features of the locust olfactory system. Each antennal lobe contains 830 PNs, and more than 100 PNs converge onto each of about 50,000 Kenyon cells in the mushroom body, a brain area important for olfactory memory. Odorant stimulation evokes oscillations in the antennal lobe, and Kenyon cells seem to integrate convergent PN input over 50 ms, approximately one oscillation cycle. The authors pooled 117 PN recordings from multiple experiments and constructed activity vectors over 50-ms windows as well as ensemble activity trajectories over the entire PN responses to different patterns of odor pulses. In contrast to the varied responses in individual PNs, ensemble PN activity showed highly similar responses to multiple pulses. Response trajectories for repeated odor pulses overlapped, with successive pulses appearing to partially reset the circuitry such that each response began similarly and followed a similar trajectory, irrespective of the pulse timing. A template chosen from any 50-ms window of a response could accurately classify odorant identity for responses in different presentation patterns. These results suggest that the problem of temporal interference may be adequately solved by converging PN input onto Kenyon cells, which also responded reliably to repeated odorant pulses. However, this solution depends on the ability of newly arriving odors to reset ensemble activity in the antennal lobe. What restricts PNs to repeatedly return to the same response trajectory despite ongoing dynamic activity? Can an ongoing response to one odorant also be reset by the arrival of a different smell? Researchers undoubtedly will continue to sniff around for these answers.

Stacey Brown


NEWS AND VIEWS

Cara Allen


1425


N E U R O B I O LO G Y O F A D D I C T I O N

INTRODUCTION

Neurobiology of addiction

T

he pleasant sensation of sipping a drink after a hard day’s work is familiar to many people, but for some, recreational use easily slips into dependence and tolerance. Some users then progress to addiction. Even in the face of harmful consequences to self and others, addicts cannot resist the urge to engage in the addictive activity. Moreover, if they do stop taking drugs, even after years of abstinence, addicts may relapse into drug use under stress or when faced with otherwise benign cues that remind them of the addicting drug. Drug use and addiction are pervasive. The 2005 World Drug Report from the United Nations estimates that 200 million people, or 5% of the global population, consumed illicit drugs at least once in the last 12 months. The US Department of Health estimates that in 2004, 22.5 million Americans aged 12 or older (9.4% of the population) experienced substance dependence or abuse. During this period, about 21.1 million people needed but did not get treatment for their addiction in the US alone. Although drug abuse cuts across all societal strata and age groups, the young and poor are affected most. Addiction has very high overall health costs, once related factors such as heart disease, cancer and accidents are considered. The National Institute on Drug Abuse estimated the cost of drug and alcohol abuse at about $246 billion in 1992 (without considering nicotine addiction). This figure includes health consequences from drug abuse and their effects on the health care system, criminal behavior, negligent driving, job loss and the effects of impaired productivity on these individuals and their employers. The progression from initial drug use to addiction is influenced by the drug, the user’s personality, peer influences and environmental stressors. These complex interactions determine why some individuals are more easily addicted than others. In this focus, we highlight the biology of the most commonly abused substances, explore the genetics of predisposition to addiction, and examine the components of addictive behavior itself. Drug addiction can clearly vary with the drug. Cocaine, marijuana, LSD or amphet-

amine can create psychological dependence, in which the individual feels satisfaction and euphoria and is driven by a need to repeat the experience. Heroin or alcohol can produce physical dependence. Drugs also act on specific receptors and brain areas. Given this complexity, can addiction be treated as a unitary disorder? Are there common brain targets for all addictive substances that could be exploited to provide a ‘magic bullet’ for addiction treatment? A perspective by Eric Nestler addresses this issue. Exposure to drugs causes plasticity in neural circuits related to reward and motivation, supporting the idea that addiction is a biological disorder. Plasticity (of synapses and circuits) results from drug use and drug abuse. How do we make sense of the multitude of observations in so many different areas under different circumstances? What animal models are likely to have the most validity for studying addiction, and what specific changes should we examine? In three separate commentaries, George Koob and Michel LeMoal, Peter Kalivas, and Yavin Shaham and Bruce Hope discuss which changes are likely to be critical to addiction. Taking drugs may begin as a voluntary choice to seek a pleasant stimulus, but for addicts, that choice is no longer volitional, even in the face of terrible personal consequences. Barry Everitt and Trevor Robbins review the cortical and subcortical circuits that mediate reinforcing effects of drugs, presenting a framework for how occasional behaviors become habits and then compulsions through pavlovian and instrumental learning. Antoine Bechara proposes that volitional decisions involve a balance between neural systems signaling the immediate and delayed consequences of actions. He discusses how drugs may tip this balance, leading to an inability to weigh future consequences and the urge to make impulsive decisions. What makes certain individuals more vulnerable to drug use and abuse? Mary Jeanne Kreek and colleagues discuss the genetic influences on complex personality traits such as impulsivity, risk taking and stress responsiveness, and their relationship to addiction vulnerability. They also discuss the difficulty in teasing out genetic vulnerability factors, in


light of the strong comorbidity between addiction and other mental disorders. Alcohol and nicotine are legal drugs that are prone to abuse. Nicotine is one of the most widely abused substances, and tobacco addiction kills more than 430,000 Americans each year. John Dani and Adron Harris review progress in understanding nicotine addiction and its comorbidity with alcoholism. John Crabbe and David Lovinger discuss the neurobiology of alcohol abuse and genetic influences that may predispose animals (and humans) to alcoholism. Despite the enormous social and economic cost of addiction, long-term treatments are few and far between. Only a handful of pharmaceutical therapies exist. In a commentary, Charles Dackis and Charles O’Brien discuss social issues that may be hampering development and access to treatment, pointing out that loss of control, the hallmark of addiction, is the source of its societal stigma. A naive public is likely to conceptualize addiction as a character flaw rather than a bona fide brain disorder. Dackis and O’Brien argue that to effectively develop treatments for addiction, we must change this perception. We are grateful to the National Institute of Drug Abuse and the National Institute of Alcoholism and Alcohol Abuse for their generous financial support for this focus issue. With their help, we are making the content of this focus freely available on the web for three months at http://www.nature.com/neuro/ focus/addiction/index.html. Apart from the sponsors’ foreword, the editorial team of Nature Neuroscience is entirely responsible for the content of the focus issue. We hope that our readers will find this collection of articles useful and enlightening and that it may contribute toward understanding and eventually solutions to this critical medical and societal problem.

I-han Chou Associate Editor Kalyani Narasimhan Senior Editor View background material on Connotea at http://www.connotea.org/user/NatNeurosci/tag/ addictionfocus.

1427


S P ON S O R ’ S F O R E WO R D

The neuroscience of addiction The burden of substance abuse and addiction to society is enormous, with an estimated annual economic impact in the United States of approximately half a trillion dollars arising from medical consequences, loss of productivity, accidents and crime1. The impact of drugs and alcohol on children is particularly problematic, as adolescents are significantly more vulnerable than adults to substance abuse and to addiction2. Also, because many of the molecular targets affected by drugs are involved with brain development, substance abuse during childhood and adolescence has the potential to be particularly deleterious. Indeed, it has been shown that children who begin using alcohol early in childhood (ages 14 or younger) are four times more vulnerable to becoming addicted to alcohol later in life than are those who begin drinking at 20 years of age or older3. Scientists are now able to portray addiction as a medical disease with physiological and molecular changes thanks to the scientific and technological advances that have occurred over the past decade. The articles in this issue highlight some of the remarkable progress that has revolutionized our understanding of the neurobiology of addiction and the way we treat it. Here we highlight some of the compelling neuroscientific questions about substance abuse, the answers for which will further improve prevention and treatment of addiction. In this editorial we use the term addiction rather than drug dependence, which is the clinical term favored by the Diagnostic and Statistical Manual of Mental Disorders (fourth edition; DSM-IV), to avoid confusion with physical dependence. Physical dependence refers to the adaptations that result in withdrawal symptoms when drugs such as alcohol and heroin are discontinued. Those are distinct from the adaptations that result in addiction, which refers to the loss of control over the intense urges to take the drug even at the expense of adverse consequences. Why do some people become addicted and others do not? Addiction has a significant genetic component. In fact, it is estimated that 40–60% of the vulnerability to addiction can be attributed to genetic factors4,5. These estimates of heredity include the percentage of the variance attributed to genetic factors by themselves as well as the percentage of the variance that is attributed to gene-environment interactions. Genotypic vulnerability for addiction is often suggested to reflect both variability in metabolism of the drug and variability in the sensitivity to the reinforcing effects of the abused substance6. However, addiction-prone and addiction-resistant phenotypes may also reflect sensitivity to the various stressors and alternative reinforcers in an individual’s environment7,8. As


we gain knowledge of the individual differences in genes and the geneenvironment interactions that make a person more vulnerable to addiction, we will be able to tailor interventions for those at high risk. Why does addiction begin most frequently during adolescence? Experimentation with drugs and alcohol often starts in adolescence, and so does the process of addiction1. This could reflect normal adolescent-specific behaviors (risk-taking, novelty-seeking, response to peer pressure) that increase the probability of someone experimenting with drugs and alcohol, and perhaps could also reflect the incomplete development of brain regions involved in the processes of executive control and motivation (for example, myelination of frontal lobe regions)9. Furthermore, preclinical studies indicate that the neuroadaptations that occur in adolescents exposed to certain drugs such as nicotine or cannabinoids are different from those that occur during adulthood10–12. Much research is currently focused on finding out whether the sensitivity to neuroadaptations during adolescence generalizes to other drugs and to alcohol, and whether this phenomenon could underlie the greater vulnerability to addiction in individuals who start using alcohol, nicotine and marijuana early in life13,14. Better knowledge of the adolescent brain, its normal functioning and how it responds to social stressors and reinforcers will allow us to develop strategies to engage adolescents in productive and creative ways that will minimize their chances of experimenting with drugs. Why do addicted people often have other mental illnesses? Individuals suffering from a variety of different disorders (such as depression, anxiety disorder, ADHD and schizophrenia) are at a much higher risk of abusing drugs and alcohol. Similarly, substance abusers and addicted individuals have a higher prevalence of mental disorders than the rest of the population. These robust comorbidities are likely to reflect overlapping environmental, genetic and neurobiological factors that influence substance abuse and mental illness. Comorbidities may emerge, in certain instances, when individuals afflicted by a mental disorder attempt to self-medicate (for example, when individuals with depression or schizophrenia use nicotine and alcohol). A more controversial interpretation, for which there is still not sufficient evidence, is the possibility that early exposure to certain drugs of abuse might increase the vulnerability to other mental disorders, particularly in those genotypes that confer increased susceptibility. What are the neural consequences of environmental risks? Drug availability is the most obvious environmental factor that influences addiction. Indeed, increased availability of cocaine and methamphetamine has contributed to the recent epidemics of addiction to these drugs. Low socioeconomic class and poor parental support are two other factors that are consistently associated with a propensity to self-administer drugs, and stress might be a common feature of these environmental factors. The mechanisms responsible for stress-induced increases in vulnerability to

1429


S P ON S O R ’ S F O R E WO R D drug use and to relapse in those addicted are not yet well understood. However, there is evidence that corticotropin-releasing factor (CRF) might play a linking role through its effects on the mesocorticolimbic dopamine system and the hypothalamic-pituitary–adrenal axis15,16. Additional preclinical studies have provided tantalizing insights on how environmental factors affect the brain and how these, in turn, affect the behavioral responses to drugs of abuse. For example, in nonhuman primates, social status affects dopamine (DA) D2 receptor expression in the brain; low status decreases expression and increases the propensity for cocaine selfadministration17. Also, animal studies have shown that an increase in DA D2 receptors in the nucleus accumbens markedly decreases drug consumption, and this could provide a mechanism by which a social stressor modifies the propensity to self-administer drugs. If we understand the neurobiological consequences underlying the adverse environmental factors that increase the risks for drug use and for addiction, we will be able to develop interventions to counteract these changes How can we repair the brain circuits disrupted by drugs? The adaptations in the brain from chronic drug exposure seem to be long-lasting and implicate multiple brain circuits (reward, motivation, learning, inhibitory control, executive function). This suggests that new interventions for drug addiction should include strategies that enhance the saliency value of natural reinforcers (including social support), strengthen inhibitory control and executive function, decrease conditioned responses and improve mood if disrupted. An interesting approach is the development of medications that act synergistically with an effective behavioral intervention. Although not yet evaluated for addiction, a proof of principle for such a concept has been recently established in a report showing that D-cycloserine administration facilitates the extinction of fear in phobic individuals through the pharmacological strengthening of the relearning events triggered during a desensitization session18.

and alcohol can disrupt volitional mechanisms by hijacking the brain mechanisms involved in seeking natural reinforcement and weakening brain mechanisms that inhibit these processes19. This new knowledge has started to provide explanations of why the addicted person relapses even in the face of dire consequences such as loss of a child’s custody or incarceration. However, despite these advances in understanding the neuroplastic changes to drugs and alcohol, addicted individuals continue to be stigmatized by the pernicious yet enduring popular belief that their affliction stems from voluntary behavior. The loss of behavioral control in the addicted individual should spur a renewed discussion of what constitutes volition, challenge us to identify the neurobiological substrates that go haywire, and influence our evolving strategies to direct our efforts to prevent and treat substance abuse and addiction more effectively. 1. 2. 3. 4. 5. 6. 7. 8.

Volkow, N. & Li, T.K. Pharmacol. Ther. (in the press). Kelley, A.E., Schochet, T. & Landry, C.F. Ann. NY Acad. Sci. 1021, 27–32 (2004). Grant, B.F. & Dawson, D.A. J. Subst. Abuse 9, 103–110 (1997). Goldman, D., Oroszi, G. & Ducci, F. Nat. Rev. Genet. 6, 521–532 (2005). Hiroi, N. & Agatsuma, S. Mol. Psychiatry 10, 336–344 (2005). Crabbe, J.C. Annu. Rev. Psychol. 53, 435–462 (2002). Kosten, T.A. et al. Brain Res. 778, 418–429 (1997). Ranaldi, R., Bauco, P., McCormick, S., Cools, A.R. & Wise, R.A. Behav. Pharmacol. 12, 527–534 (2001). 9. Sowell, E.R., Thompson, P.M., & Toga, A.W. Neuroscientist 10, 372–392 (2004). 10. Pistis, M. et al. Biol. Psychiatry 56, 86–94 (2004). 11. Adriani, W. & Laviola, G. Behav. Pharmacol. 15, 341–352 (2004). 12. Belluzzi, J.D., Lee, A.G., Oliff, H.S. & Leslie, F.M. Psychopharmacology (Berl.) 174, 389–395 (2004). 13. Baumeister, S.E. & Tossmann, P. Eur. Addict. Res. 11, 92–98 (2005). 14. Brook, D.W., Brook, J.S., Zhang, C., Cohen, P. & Whiteman, M. Arch. Gen. Psychiatry 59, 1039–1044 (2002). 15. Wang, B. et al. J. Neurosci. 25, 5389–5396 (2005). 16. Breese, G.R. et al. Alcohol Clin. Exp. Res. 29, 185–195. 17. Morgan, D. et al. Nat. Neurosci. 5, 169–174 (2002). 18. Ressler, K.J. et al. Arch. Gen. Psychiatry 61, 1136–1144 (2004). 19. Volkow, N.D. & Fowler, J.S. Cereb. Cortex 10, 318–325 (2000).

What is volition and how do drugs disrupt it? Remarkable scientific advances have emerged in the neuroscience of addiction that offer new insights into how chronic drug use affects the inner workings of the brain and how this leads to the aberrant behavioral manifestations of addiction. We have learned how some drugs

Nora Volkow

1430


Director, National Institute on Drug Abuse

Ting-Kai Li Director, National Institute on Alcohol Abuse and Alcoholism



C O M M E N TA RY

Neurobiology of addiction: treatment and public policy ramifications Charles Dackis & Charles O’Brien In the United States, efforts to treat addiction are hampered by prejudice and a public view that treats it as a disorder of selfcontrol, not a disease. We highlight select advances in addiction research that, if disseminated to the public, could reverse these misconceptions and facilitate changes in policy to improve treatment access and care delivery for this highly prevalent disease.

The infrastructure to treat addictive illness, when compared with treatment for other traditional medical illnesses, is lacking in the United States. This situation is tolerated by a public that views addiction more as a social problem than an actual disease, despite scientific evidence supporting a disease concept of addiction based on neuronal mechanisms, heritability, treatment responses and a characteristic progressive clinical course1. Pejorative views toward addicted individuals also exist and contribute to policies that would be simply unacceptable if applied to ‘real’ medical disorders. These policies have created limited access, insufficient capacity and a dearth of trained providers in most geographical regions, especially for adolescent patients who might avoid progressive addiction with appropriate treatment. Even patients with access to treatment typically discover that its duration is severely limited by insurance company policies (managed care), even though addiction is a chronic illness requiring sustained aftercare. Imagine limiting treatment duration for diabetes, chronic heart failure or hypertension. Stigma and misconception create formidable obstacles to a more enlightened public policy toward addictive illness. Rather than being treated as patients, afflicted individuals are often blamed for their illness, discriminated against

Charles Dackis is at the Department of Psychiatry, 3900 Chestnut Street, University of Pennsylvania, Philadelphia, Pennsylvania, USA, and Charles O’Brien is at the Philadelphia Veterans Affairs Medical Center, Treatment Research Center, 3900 Chestnut Street, Philadelphia, Pennsylvania, USA. e-mail: [email protected]

and readily criminalized. Specialized treatment for addiction is even viewed as unnecessary (why not ‘just say no’ to drugs?) or misperceived as being ineffective. In contrast, treatment response does not dictate availability of care for other medical conditions like cancer, stroke and heart failure. Why should it be considered an appropriate standard for the availability of addiction treatment? This blatant discrepancy in access suggests that, despite therapeutic advances and improved clinical outcome, treatment parity will not be achieved until addiction is widely viewed as a disease. Much of our knowledge about addiction neurobiology is based on decades of animal studies that model the dynamic clinical components of the illness. Elegant study designs assessing self-administration, conditioned place preference, reinstatement (after cues, stress and drug priming) and intracranial self-stimulation have provided a tremendous amount of behavioral and neurochemical information. Although this research has identified neuronal mechanisms underlying drug reward, craving, relapse and hedonic dysregulation, the predictive value of animal models varies considerably. Naltrexone treatment for alcohol dependence stemmed directly from animal studies showing that opioid antagonists reduce alcohol selfadministration. On the other hand, the robust phenomenon of sensitization has received considerable emphasis even though its clinical significance is questionable, and it has not produced new treatments. Neuroimaging may ultimately circumvent the limitations of animal models and delineate brain mechanisms associated with clinical features of addictive illness. Scientific discoveries that


substantiate the biological basis of addiction and improve treatment outcome should ultimately erode entrenched societal attitudes that prevent addiction from being evaluated, treated and insured as a medical disorder. Addiction is best conceptualized as a disease of brain reward centers that ensure the survival of organisms and species2. Given their function, reward centers have evolved the ability to grip attention, dominate motivation and compel behavior directed toward survival goals, even in the presence of danger and despite our belief that we are generally rational beings. By activating and dysregulating endogenous reward centers, addictive drugs essentially hijack brain circuits that exert considerable dominance over rational thought, leading to progressive loss of control over drug intake in the face of medical, interpersonal, occupational and legal hazards. There is even evidence that denial, once thought to be purely ‘psychological’, may be associated with drug-induced dysfunction of the prefrontal cortex3. Loss of control is both the hallmark of addiction and the source of its societal stigma. An uneducated yet strongly opinionated public does not understand the technical field of addiction neurobiology and is more likely to conceptualize addiction as a character flaw (for example, addictive personality) than a brain disease. Therefore, the dissemination of understandable information about this brain disease could change public perceptions and hence public policy toward addictive illness. Considerable emphasis has been placed on the prevention of addiction through widespread educational initiatives targeting children, adolescents and parents, but much less emphasis has been placed on the disease of addiction. Addiction is a disease

1431

C O M M E N TA R Y

Addictive agent

+ Drug euphoria Positive reinforcement Activated reward pathways

Neuroadaptations Withdrawal and tolerance Protracted hedonic dysregulation


Drug administration Drug-seeking behavior Failed impulse suppression

Loss of control Denial/poor decision-making Hypofrontality/low D2 Reduced gray matter density

– Drug craving Negative reinforcement Dysregulated reward pathways

Stress

of brain regions that are intrinsically interesting to the general public because they subserve the human experiences of pleasure, craving and motivation. Fascination with this topic could be exploited by educational initiatives to gain ground against moralistic attitudes that stigmatize, ostracize and often criminalize patients with addictive illness. Access to treatment for millions of addicted patients is a costly proposition. However, there would be offset savings in the cost of medical care, lost productivity, neighborhood destruction, crime and prison capacity4. Even though the United States has a disproportionate number of prisoners, and most have been incarcerated for drug-related crimes, their addiction is seldom treated within the prison walls or, more importantly, after they are released to a druginfested environment. Similarly, although medical complications of addiction are commonly encountered in clinical practice, their cause is seldom addressed and treated5. By fully integrating addiction treatment into our medical care delivery and judicial systems, we could dramatically improve medical care and justice. Here we highlight select areas of addiction research that illustrate brain involvement and would probably stimulate public interest if conveyed in an understandable fashion. Perhaps the dissemination of these and other examples of current knowledge could begin to reverse popular misconceptions about addictive illness, increase compassion and tolerance and facilitate changes in public policy that improve treatment access and care delivery for this highly prevalent disease. The cycle of addiction Addiction neurobiology ties clinical phenomena of the illness to specific neuronal

1432

Drug-related cues Limbic activation

mechanisms, which provides insight into the pathological process and identifies new treatments. Discrete clinical phenomena of this pleasure-reinforced illness are integrated into a dynamic cycle of addiction (Fig. 1) that becomes progressively more entrenched and uncontrollable as the brain becomes addicted6. The biological basis of these clinical components has been increasingly delineated through an explosion in addiction research that initially involved animal models and has since expanded to neuroimaging studies of addicted patients. Addictive drugs produce euphoria by activating brain pleasure centers, and it is noteworthy that diverse agents (for example, opioids, stimulants, alcohol, nicotine, marijuana) all increase extracellular dopamine (DA) levels in the shell of the nucleus accumbens (NAc). Drug-induced euphoria in humans has also been closely linked to DA receptor (D2) binding by several elegant positron emission tomography (PET) studies6. Animal studies demonstrate that natural rewards (sex, food, water) also elevate DA levels in the NAc, although to a lesser extent, and human neuroimaging studies report that DA release in the dorsal striatum correlates with meal pleasantness7. These studies link drug euphoria to natural reward centers that have evolved to ensure survival. Once experienced, drug euphoria promotes the repeated use of an addictive drug, especially if genetic traits enhance the pleasurable experience. For instance, there is considerable evidence that individuals with a genetic predisposition toward alcoholism experience more pleasure from this drug because it produces an exaggerated β-endorphin response. Over time, addictive drugs disrupt reward circuits and produce dysphoric states such as withdrawal, craving and hedonic dysregulation that provide

Figure 1 The cycle of addiction is positively reinforced by drug euphoria and negatively reinforced by withdrawal, craving and hedonic dysregulation. Drug-related cues and stress increase craving, and loss of control may stem in part from prefrontal cortical dysfunction. Neuronal mechanisms for these cardinal components of addiction have been increasingly delineated with animal models and human neuroimaging studies.

negative reinforcement, and alternate with the positive reinforcement of euphoria to drive the cycle of addiction (Fig. 1). Chronic exposure to heroin, cocaine or alcohol produces a number of common neuroadaptations8, including DA hypoactivity, that contribute to a remarkably similar clinical course in severely addicted individuals. The cycle of addiction becomes etched in midbrain and frontal structures that reinforce the pursuit of survival-related behaviors by dominating attention and decision-making. Addictive illness reminds us that desire and pleasure can be impervious to rational thought, clashing with deeply engrained cultural values placed on stoicism and self-control. Craving is a complicated phenomenon that can be dramatically amplified by stimuli (cues) that have become associated with drugs through conditioned learning. Neuroimaging studies of addicted human patients demonstrate a fascinating link between brain function and cue-induced craving, which is arguably the most persistent and insidious clinical component of addictive illness. Cues associated with diverse substances (for example, cocaine, heroin, alcohol and nicotine) produce robust activation of limbic structures on PET and functional magnetic resonance imaging (fMRI). Images depicting limbic activation during cue-induced craving provide an interesting and graphic means of demonstrating the neuronal basis of cue-induced craving to the general public (Fig. 2). Another interesting neuroimaging finding associated with addictive illness is that of hypofrontality (reduced baseline metabolism in the prefrontal cortex)6. Baseline hypofrontality involves the same frontal regions that become hypermetabolic during cueinduced craving, and the exaggerated change (∆ metabolism; peak minus baseline) in frontal metabolism might contribute to the remarkable salience of drug-related cues3. In addition to hypofrontality, cocaine-addicted individuals show reductions in frontal gray matter density9 and poor performance on neuropsychological tests assessing prefrontal cortical function3. As the seat of executive function in the brain, the prefrontal cortex is involved in decisionmaking, risk/reward assessment, impulse


C O M M E N TA R Y


the ventral striatum that correlates with the severity of their reported craving for alcohol15. Thus, both genetic and alcohol-induced alterations in β-endorphin are important in the neurobiology of alcoholism. The involvement of endogenous opioids in alcoholism led directly to the development of naltrexone as an approved treatment for this condition.

Figure 2 Cue-induced craving (produced by a cocaine video compared to a nature video) is associated with significant limbic activation on PET, which graphically demonstrates the neuronal basis of this important clinical component of addictive illness. Reprinted with permission from the American Journal of Psychiatry, Copyright 1999. American Psychiatric Association.

control and perseverance. Functional and structural abnormalities in the prefrontal cortex might therefore contribute to clinical characteristics of addicted patients (such as poor impulse control, lack of resolve, faulty decision making) that are viewed prejudicially by the general public. Hypofrontality is associated with reduced D2 receptor availability on PET, which may be a marker for reduced DA function in addicted patients. The next sections review specific neurobiological findings that are likely to be of great interest to the general population and convey the biological basis of addiction. It is not widely known that the brain produces opioids and opioid receptors, that heroin binds to these receptors, and that alcohol pleasure involves opioid function. Also unappreciated is the importance of cue-induced craving, its basis in limbic activation, and evidence that addicted individuals have impairments in executive and hedonic function. These findings should be disseminated to the general public in understandable and interesting forums to promote the disease concept of addiction. Endogenous opioids Endogenous opioid pathways activated by addictive drugs are involved in pain, pleasure, appetite, sexual function and natural drive states, and it is noteworthy that separate and antagonistic enkephalin and dynorphin populations of medium spiny cells in the NAc are involved in addictive illness8. Although alcohol is ubiquitous in our society, few people know that alcohol reward is mediated by

endogenous opioids and influenced by genetic factors affecting opioid function. One of the earliest reports10 pertaining to this topic was published in 1980, showing that naltrexone pretreatment extinguishes alcohol self-administration in rhesus monkeys. Several lines of animal research subsequently demonstrated that alcohol acutely increases opioid activity, especially in animals bred to prefer alcohol, and that alcohol is not self-administered by µ-opioid receptor knockout mice11. Human studies also demonstrate involvement of endogenous opioid systems in alcohol reward. Compared with normal subjects, individuals with a genetic predisposition for alcoholism have low baseline blood β-endorphin levels and enhanced β-endorphin release and euphoria after alcohol administration12. Enhanced release of β-endorphin against low baseline levels constitutes a surge in the concentration of this rewarding endogenous opioid that may explain why these individuals experience more pleasure from alcohol. Other studies corroborate an interaction between β-endorphin levels and alcohol consumption. Cerebrospinal levels of β-endorphin are three times higher in normal subjects than in patients with chronic alcoholism, and there is evidence that β-endorphin levels might become depleted after chronic alcohol intake13. Alcoholics experiencing withdrawal symptoms have plasma β-endorphin levels only half as high as those in normal subjects, yet their levels normalized after several weeks of sobriety14. In addition, abstinent alcoholics show increased µ-opioid receptor binding in


Cue-induced craving and limbic activation We have known for decades that environmental stimuli (people, places and things) associated with drug use can trigger intense craving in addicted patients. Aside from perpetuating active drug use, cue-induced craving triggers relapse after protracted abstinence because it persists for months or years, and even perhaps indefinitely, as a direct avenue to recidivism. Seeing a syringe in the doctor's office, smelling a cigarette, or glancing at a vodka advertisement are innocuous experiences for most of us but can be painfully compelling for vulnerable individuals. Largely unknown to the general public, neuroimaging studies have demonstrated dramatic limbic responses to drug-related cues that correlate with the degree of reported craving. This phenomenon graphically demonstrates the biological nature of addictive illness and provides one of the most fascinating examples of the mind/brain interface. Neuroimaging studies of patients addicted to various substances demonstrate the activation of similar frontal regions as a common pathway of cue-induced craving. Cocainedependent patients have been studied extensively in PET and fMRI experiments, and they consistently show activation of the amygdala and anterior cingulate cortex that correlates closely with their reports of craving severity3. Alcoholic subjects also show activation of the anterior cingulate, medial prefrontal cortex and striatum in response to alcohol-related cues on fMRI16,17, and the intensity of the cue reactivity correlates with their likelihood of relapse18. PET studies of heroin-dependent subjects also demonstrate a strong correlation between cue-induced opioid craving and hypermetabolic responses in the inferior frontal and orbitofrontal cortex19, and patients with nicotine dependence show increased metabolism in the anterior cingulate on fMRI during exposure to cigarette-related cues20. These studies demonstrate a common neuronal response to cues associated with diverse substances, and they justify the commonly held notion that various drug dependencies should be conceptualized as a single disorder. Natural drive states are also associated with activation of glutamate-rich cortical regions. Remarkably, the same frontal regions that are activated by cocaine-related cues in

1433


C O M M E N TA R Y cocaine-dependent patients are also activated in normal subjects viewing sexually explicit videos21. Furthermore, neuroimaging studies demonstrate that the subjective report of hunger in response to food-related cues is temporally associated with marked activation of frontal regions22. These studies link drugrelated craving with natural drive states, and graphically support the idea that addictive drugs hijack endogenous reward circuits that have evolved to ensure survival. Prefrontal cortical regions that are activated during cue-induced craving receive DA projections from neurons originating in the ventral tegmentum. A series of elegant singlecell recording studies demonstrate that these midbrain DA neurons fire during unpredicted hedonic activity, but their firing habituates to predictable reward and shifts instead to cues that reliably predict impending reward23. These and other studies reviewed elsewhere3 suggest that DA firing is correlated with cue-induced limbic activation. Interestingly, studies demonstrate that DA firing in the ventral tegmentum plunges below baseline when anticipated reward is not delivered, linking DA hypoactivity to an animal model of acute deprivation (craving). Furthermore, animals chronically exposed to stimulants, alcohol or opioids show dramatic depletion of extracellular DA in the NAc24, and DA depletion might contribute to craving and hedonic dysregulation in addictive illness8. Because cue-induced craving is associated with DA firing and hypermetabolic responses in glutamate-rich cortical regions, medications that reduce DA neurotransmission have been widely proposed as potential treatments for this phenomenon. Glutamatergic neurotransmission is also implicated in cue-induced craving, and glutamate-releasing neurons in the orbitofrontal cortex (which receive reward-related sensory input from the thalamus) fire during cues related to natural rewards and send excitatory projections to the VTA and the NAc25. Cueinduced craving often leads directly to relapse, and an effective treatment for this phenomenon should dramatically improve outcome. DA and glutamate antagonists should be tested in the laboratory before concluding that they reduce limbic activation during cue presentation, especially as addicted patients may already be DA depleted. Indeed, limbic activation during drug-related cues provides a unique biological marker that should be exploited with further targeted research. The D2 story One of the most interesting findings in addiction research is the reduced availability

1434

of striatal D2 receptors in patients with addictive illness. PET studies using [11C]raclopride, a radioligand that competes with DA at D2 receptors, demonstrate persistently low striatal D2 availability (↓D2) in patients addicted to cocaine26, alcohol16, methamphetamine27 or opioids6. Individuals with morbid obesity also have ↓D2 that is inversely related to their body mass index28. It is not known whether ↓D2 in addicted patients precedes or results from their drug exposure, and there is evidence that both possibilities may occur. That ↓D2 persists beyond detoxification from alcohol and opiates suggests that it might be a predisposing factor or at least a persistent drug-induced finding6. The possibility that ↓D2 represents an inherited trait is compelling because D2 binding varies considerably across individuals, and nonaddicted individuals with ↓D2 report significantly more pleasure after receiving stimulant drugs6. Similarly, monkeys with ↓D2 are significantly more likely to self-administer cocaine than those with increased striatal D2 receptor availability on PET (↑D2; ref. 29). Indeed, ↑D2 may protect against addiction because alcohol intake is significantly reduced in rats after D2 receptor expression has been increased with an adenoviral vector30. The importance of genetic factors in addictive illness, especially those affecting the intensity of drug reward, reinforces the biological basis of this disorder. Although ↓D2 may represent a constitutional trait and addiction vulnerability, it can also result from cocaine exposure because chronic cocaine treatment produces ↓D2 in monkeys. In addition, D2 varies with social dominance rank in cynomolgus monkeys and is reduced with social demotion, leading to an increased propensity to self- administer cocaine 31. There is considerable evidence from animal studies supporting DA hypoactivity after chronic exposure to stimulants, opioids, and alcohol 24,32,33, and human studies also report DA hypoactivity in alcohol-34, heroin- 35 and cocaine-addicted patients 6, with the latter group showing evidence of DA hypoactivity on neuroimaging and a host of neuroendocrine and autopsy studies reviewed elsewhere 25 . DA hypoactivity after chronic cocaine administration is associated with the downregulation of D2 autoreceptors that are abundant in the striatum3. Consequently, ↓D2 may reflect autoreceptor downregulation and may serve as a marker for DA dysregulation in addictive illness.

Autoreceptor downregulation might also contribute to the controversial finding of sensitization, which has unclear relevance to addictive illness despite its considerable emphasis by many animal researchers36. Whereas tolerance is defined as a reduced dose response after repeated drug administration, sensitization involves accentuated responses, classically in the form of enhanced locomotion with a repeated fixed dose of a stimulant or opioid agent. Enhanced cocaineinduced elevations of DA37 and glutamate38 in the NAc of cocaine-pretreated animals are associated with sensitization and could be explained by persistently downregulated D2 (ref. 3) and mGluR2/3 (ref. 39) autoreceptors, especially as mice lacking D2 expression (having no autoreceptor function) show strikingly enhanced striatal DA levels after the administration of cocaine and morphine40. Thus, sensitized DA and glutamate responses to cocaine, often invoked as a rationale for testing DA- or glutamate-inhibiting agents, may merely reflect a homeostatic autoreceptor response to DA hypoactivity. Sensitization in animals has led some researchers to speculate that cocaine euphoria actually increases over time, even though patients typically report the opposite and escalate their daily consumption of cocaine. This area of research provides an excellent example of why animal models must be reconciled with clinical experience. In cocaine and methamphetamine abusers, ↓D2 is correlated with reduced metabolism in the orbitofrontal cortex6. As previously noted, hypofrontality in cocaine-dependent patients may contribute to poor impulse control, elements of denial and compulsive drug use2,6. Cocaine-dependent subjects with reduced anterior cingulate and right prefrontal cortical metabolism have concomitant difficulty controlling impulses during formal neuropsychological testing41. These findings suggest that agents that increase metabolic activity in frontal regions, such as modafinil, might improve impulse control in addicted patients3. Through animal models and human neuroimaging studies, researchers are elucidating neuronal mechanisms that underlie the dynamic clinical elements of addictive illness. First, this body of research strongly supports the disease concept by linking the activity of reward-related structures in the brain to clinical manifestations of this disease. Common neurobiological phenomena also justify categorizing addiction to diverse agents under a single general disorder. Diverse agents like cocaine, heroin and alcohol increase striatal DA levels during intoxication, whereas chronic exposure to these agents is associated with DA hypo-


C O M M E N TA R Y


activity, ↓D2 and limbic activation during cue-induced craving. Addiction also has genetic determinants and a similar progressive clinical course across various substances. These attributes are certainly consistent with the idea that addiction is a brain disorder, despite popular misconceptions. Pharmacological treatments for addiction Several approved and promising treatments for addiction have been identified through neurobiological research3,42. Pharmacological strategies are emerging that target specific clinical components of addiction, including drug-induced euphoria, hedonic dysregulation, cue-induced craving and even denial. The development of treatments that dramatically improve clinical outcome should reverse social stigma and justify an expanded care delivery system. However, the clinical impact of new treatments also depends on their translation into clinical practice. Treatments for nicotine dependence (such as bupropion and the nicotine patch) have been promoted by the pharmaceutical industry, are often prescribed by primary care physicians and are covered by some insurance plans. Naltrexone treatment for alcoholism, on the other hand, has not been sponsored by industry, is seldom prescribed by primary care physicians, and is greatly underused. Naltrexone (an opioid receptor antagonist) was originally developed to treat heroin dependence by blocking euphoria, an established pharmacological strategy that improves outcome by weakening the addiction cycle. However, this strategy has been limited in opioid dependence because naltrexone does not convincingly ameliorate opioid craving, and patients often stop the drug and resume heroin use. Still, reducing opioid reward with naltrexone provides benefits for some patients, and adherence has recently been addressed with the development of depot delivery systems that allow for monthly medication injections. The initial controlled study of naltrexone in alcoholics reported a reduction in clinically significant daily drinking and alcohol craving in active versus placebo groups43. After these findings were replicated at another site44, naltrexone gained FDA approval for the treatment of alcoholism. Since then, most controlled studies have reported significant reductions in daily drinking with naltrexone treatment. Furthermore, the efficacy of naltrexone might be more dramatic in a subgroup of genetically defined alcoholics. One of the polymorphisms (Asp40) for the gene encoding the µ-opiate receptor produces a receptor with high affinity for β-endorphin, and individuals with this variant have increased risk of alcoholism45 and heroin addiction46. Alcoholics with this variant are reported to be

significantly more likely to benefit from naltrexone than patients without the variant are47. If this finding is replicated, clinicians will have an available genotype to match alcoholic patients with effective treatment. Studies showing naltrexone efficacy in alcoholic outpatients prompted laboratory testing to assess how the beneficial effects are mediated. One of these studies suggests that naltrexone diminishes alcohol-induced craving, which fuels the common phenomenon in which the first drink leads to uncontrollable drinking. This controlled study evaluated the effect of naltrexone pretreatment on baseline and alcohol-induced craving48, finding that placebo- versus naltrexone-treated patients reported higher alcohol craving at baseline and after alcohol priming. This study also tested drinking behavior after the priming dose of alcohol by asking subjects to choose either alcohol or money, and the placebo group chose alcohol significantly more often. Conversely, the naltrexone group reported less craving, even after the priming dose when additional alcohol was available, consumed fewer drinks and drank more slowly. Therefore, the tendency for alcoholics to lose control once they begin to drink is an important clinical feature of alcoholism that may be specifically ameliorated by naltrexone. Although naltrexone represents a success story that stemmed directly from neurobiological research, this treatment for alcoholism is markedly underused in clinical practice. One problem has been patient nonadherence, sometimes in response to side effects but more often to recapture the experience of alcohol-induced euphoria. This issue is being addressed by the development of depot delivery systems that will eliminate the need for patients to make a daily decision about naltrexone. However, the greatest translational problem likely stems from a curious lack of awareness among primary care physicians regarding naltrexone. Despite FDA approval for alcoholism since 1994, naltrexone is not widely prescribed to the enormous population of active alcoholics, even though alcohol often produces the very illnesses for which these patients seek medical treatment5. Pharmaceutical companies have just begun to view alcoholic patients as an important population, and the recent approval of acamprosate might signal a change in industry attitudes toward addictive illness. Acamprosate modulates N-methyl-D-aspartate (NMDA) receptor subunit expression49. However, the pharmaceutical industry is still reluctant to develop treatments for illegal addictions, exemplified by the fact that cannabinoid receptor antagonists have not been made available for testing


in marijuana-dependent patients. Our government may also be reluctant to promote treatments for illegal addictions. Buprenorphine, an effective partial µ-opioid agonist that was recently approved for office-based treatment of opioid dependence, has many advantages over methadone, a full agonist with many legal restrictions. Nonetheless, its use is curtailed by FDA rules stipulating which physicians can prescribe the drug and how many patients they can treat. New pharmacological strategies that target specific elements of the addiction cycle are currently under intense investigation. Modafinil has been reported to attenuate cocaine euphoria in two controlled studies and is under investigation in three large clinical trials3. Cocaine euphoria is also being targeted with cocaine vaccines that prevent the drug from entering the brain50, and other promising medications for cocaine dependence (disulfiram, topiramate, propranolol and baclofen) are being tested3. Cue-induced craving in cocaine, opioid, heroin and nicotine dependence is a logical target for candidate medications that might be screened in the neuroimaging laboratory before being tested in large clinical trials. Potential pharmacological treatments for other clinical components of addiction, including stress-induced craving, hedonic dysregulation and hypofrontality, will likely be identified through expanding research3. Public policy implications Changes in public policy are needed to improve the access, capacity and quality of addiction services. These changes would be facilitated by public acceptance of the disease concept and through the development of more effective treatments. Both goals could be attained with advances in addiction neurobiology and continued funding in this essential area of research. Animal and human research should be closely coordinated and focused on developing practical treatment interventions. Animal models with demonstrated relevance to the clinical setting, especially those assessing self-administration and reinstatement, should be prioritized as a means of guiding treatment development. It is also imperative that research findings in this technical area be made available to the public in an understandable manner that conveys the biological nature of addiction. We have reviewed select findings from this research that illustrate brain involvement in alcohol euphoria, cue-induced craving and the genetic vulnerability to addiction. These interesting examples are largely unknown to the general public, and their dissemination is now indicated to promote the disease concept of addiction.

1435


C O M M E N TA R Y Even when effective treatments for addiction have been identified, as illustrated by naltrexone treatment for alcoholism, they have not always been adequately translated into clinical practice. The pharmaceutical industry has recently become interested in alcohol and nicotine dependence, and their considerable resources could potentially expand addiction treatment. Addiction treatment should also be integrated into mainstream medicine. Although medical complications of substance abuse are commonly encountered in clinical practice, addiction and medical treatments are seldom coordinated5. Primary-care physicians are essential in this regard and should have a much greater role in the assessment and treatment of addicted patients. The training of physicians to assess and treat addiction should be expanded in our medical schools and residency programs. We cannot allow the stigma of addiction to influence training policies, as it undoubtedly has, if we wish to provide comprehensive and effective medical care. Primary-care physicians should competently evaluate their patients for addictive illness, especially those with addiction-related medical conditions, and view pharmacological treatments as part of their clinical arsenal. Referrals to addiction specialists should be made with the same frequency as those to other medical specialties. The general quality of care delivery in this country will be improved to the extent that addiction treatment is placed in the mainstream of medicine. The judicial approach to addicted patients is another area in dire need of guidance and policy change. Active addiction often involves criminal behaviors related to drug use and procurement, and addicted patients typically engage in activities they would never consider during recovery. Such individuals should receive innovative judicial interventions that promote treatment over criminalization and recovery over incarceration. The judicial system should develop an integrated interface with specialized treatment teams to ensure that appropriate interventions are closely coordinated. Addicted patients already incarcerated should receive treatment within the prison walls, and drug testing of inmates and correctional officers should be applied to eliminate the widespread use of drugs within our prisons. Ending inappropriate criminalization of this disease would produce tremendous financial and humanitarian benefit to our society. Education designed to prevent addiction is already integrated into our schools and should be further developed with research-based inter-

1436

ventions. Because perceived risk is an important determinant of drug experimentation, the dangers of drugs should be conveyed accurately and credibly to our youth. Schools provide a natural setting for these interventions and a means of identifying students who may require treatment. Unfortunately, there are few practitioners qualified to evaluate and treat adolescent substance abusers in most geographical regions. Given the progressive nature of addiction, and the opportunity of early intervention, inadequate access to treatment for adolescent substance abusers is entirely unacceptable. Public-policy changes that improve access, capacity and quality in addiction treatment will require significant investment in our health delivery system. Treatment for this chronic disorder is labor intensive, requiring a comprehensive assessment by qualified practitioners, as well as ongoing individual, group and family interventions. Although pharmacological treatments for addiction will continue to improve and streamline treatment, a ‘magic bullet’ for this chronic debilitating disorder will probably not be found. If and when the public begins to view addiction as a medical disorder, the need for treatment parity could be added to arguments regarding the cost of untreated addiction in dollars and lives. It remains to be seen whether improvements in our care delivery system will occur in a climate that focuses more on cost savings than quality enhancement. ACKNOWLEDGMENTS We thank A.R. Childress for contributing Figure 2. COMPETING INTEREST STATEMENT The authors declare that they have no competing financial interests.

1. MpcLellan, A.T., Lewis, D.C., O’Brien, C.P. & Kleber, H.D. J. Am. Med. Assoc. 284, 1689–1695 (2000). 2. Dackis, C.A. & O’Brien, C.P. J. Subst. Abuse Treat. 21, 111–117 (2001). 3. Dackis, C.A. Drug Des. Discov. 2, 79–86 (2005). 4. Belenka, S., Patapis, N. & French, M. Economic Benefits of Drug Treatment (National Rural Alcohol and Drug Abuse Network, Menominie, Wisconsin, 2005). 5. Weisner, C., Mertens, J., Parthasarathy, S., Moore, C. & Lu, Y. J. Am. Med. Assoc. 286, 1715–1723 (2001). 6. Volkow, N.D., Fowler, J.S. & Wang, G.J. Neuropharmacology 47 (Suppl.) 3–13 (2004). 7. Small, D.M., Jones-Gotman, M. & Dagher, A. Neuroimage 19, 1709–1715 (2003). 8. Dackis, C.A. & O’Brien, C.P. in Diseases of the Nervous System (eds. Asbury, A., McKhann, G., McDonald, W., Goadsby, P. & McArthur, J.) 431–444 (Cambridge Univ. Press, Cambridge, 2002). 9. Franklin, T.R. et al. Biol. Psychiatry 51, 134–142 (2002). 10. Altshuler, H.L. et al. Pharmacol. Biochem. Behav. 13 Suppl 1, 233–240 (1980). 11. Roberts, A.J. et al. J. Pharmacol. Exp. Ther. 293,

1002–1008 (2000). 12. Gianoulakis, C., Krishnan, B. & Thavundayil, J. Arch. Gen. Psychiatry 53, 250–257 (1996). 13. Genazzani, A.R. et al. J. Clin. Endocrinol. Metab. 55, 583–586 (1982). 14. Vescovi, P.P., Coiro, V., Volpi, R., Giannini, A. & Passeri, M. Alcohol Alcohol. 27, 471–475 (1992). 15. Heinz, A. et al. Arch. Gen. Psychiatry 62, 57–64 (2005). 16. Heinz, A. et al. Am. J. Psychiatry 161, 1783–1789 (2004). 17. Myrick, H. et al. Neuropsychopharmacology 29, 393– 402 (2004). 18. Grusser, S.M. et al. Psychopharmacology (Berl.) 175, 296–302 (2004). 19. Sell, L.A. et al. Drug Alcohol Depend. 60, 207–216 (2000). 20. Brody, A.L. et al. Psychiatry Res. 130, 269–281 (2004). 21. Garavan, H. et al. Am. J. Psychiatry 157, 1789–1798 (2000). 22. Wang, G.J. et al. Neuroimage 21, 1790–1797 (2004). 23. Schultz, W. Neuroscientist 7, 293–302 (2001). 24. Rossetti, Z.L., Melis, F., Carboni, S. & Gessa, G.L. Ann. NY Acad. Sci. 654, 513–516 (1992). 25. Dackis, C. & O’Brien, C. Ann. NY Acad. Sci. 1003, 328–345 (2003). 26. Martinez, D. et al. Neuropsychopharmacology 29, 1190–1202 (2004). 27. Volkow, N.D. et al. Am. J. Psychiatry 158, 2015–2021 (2001). 28. Wang, G.J., Volkow, N.D., Thanos, P.K. & Fowler, J.S. J. Addict. Dis. 23, 39–53 (2004). 29. Morgan, D. et al. Nat. Neurosci. 5, 169–174 (2002). 30. Thanos, P.K. et al. J. Neurochem. 78, 1094–1103 (2001). 31. Czoty, P.W., Morgan, D., Shannon, E.E., Gage, H.D. & Nader, M.A. Psychopharmacology (Berl.) 174, 381– 388 (2004). 32. Georges, F. & Aston-Jones, G. Neuropsychopharmacology 28, 1140–1149 (2003). 33. Gerrits, M.A., Petromilli, P., Westenberg, H.G., Di Chiara, G. & van Ree, J.M. Brain Res. 924, 141–150 (2002). 34. Tupala, E. & Tiihonen, J. Prog. Neuropsychopharmacol. Biol. Psychiatry 28, 1221–1247 (2004). 35. Kish, S.J. et al. Neuropsychopharmacology 24, 561– 567 (2001). 36. Robinson, T.E. & Berridge, K.C. Brain Res. Brain Res. Rev. 18, 247–291 (1993). 37. Kalivas, P.W. & Duffy, P. Synapse 5, 48–58 (1990). 38. Reid, M.S. & Berger, S.P. Neuroreport 7, 1325–1329 (1996). 39. Xi, Z.X. et al. J. Pharmacol. Exp. Ther. 303, 608–615 (2002). 40. Rouge-Pont, F. et al. J. Neurosci. 22, 3293–3301 (2002). 41. Hester, R. & Garavan, H. J. Neurosci. 24, 11017– 11022 (2004). 42. O’Brien, C.P. Am. J. Psychiatry 162, 1423–1431 (2005). 43. Volpicelli, J.R., Alterman, A.I., Hayashida, M. & O’Brien, C.P. Arch. Gen. Psychiatry 49, 876–880 (1992). 44. O’Malley, S.S. et al. Arch. Gen. Psychiatry 49, 881–887 (1992). 45. Bart, G. et al. Neuropsychopharmacology 30, 417–422 (2005). 46. Bart, G. et al. Mol. Psychiatry 9, 547–549 (2004). 47. Oslin, D.W. et al. Neuropsychopharmacology 28, 1546– 1552 (2003). 48. O’Malley, S.S., Krishnan-Sarin, S., Farren, C., Sinha, R. & Kreek, M.J. Psychopharmacology (Berl.) 160, 19–29 (2002). 49. Rammes, G. et al. Neuropharmacology 40, 749–760 (2001). 50. Haney, M. & Kosten, T.R. Expert Rev. Vaccines 3, 11–18 (2004).




C O M M E N TA RY

The role of neuroadaptations in relapse to drug seeking Yavin Shaham and Bruce T Hope One of the most difficult problems in treating addiction is not withdrawing addicts from drugs, but preventing relapse. Persistent neuroadaptations are thought to underlie aspects of addiction, including relapse. This commentary assesses the degree to which these neuroadaptations, primarily identified in preclinical studies on cocaine, induce relapse.

Relapse to drug use is one of the core features of addiction and probably the most difficult clinical problem in addiction treatment1. After prolonged abstinence, drug relapse and craving is often precipitated by acute re-exposure to the drug itself, drug-associated cues or stress2,3. This clinical scenario can be modeled in a reinstatement procedure, as stimuli that trigger human relapse also reinstate drug seeking in drug-abstinent animals4. In this model, mice, rats or monkeys are trained to self-administer drugs by pressing a lever and then undergo ‘extinction’ training, during which lever-presses do not deliver the drug. Subsequently, the effect of ‘drug priming’ (acute noncontingent drug injections) or exposure to drug cues or stress on reinstatement of nonreinforced lever responding (the operational measure of drug seeking) is measured5. The effect of drug cues on drug seeking can also be measured in extinction tests administered after different durations of abstinence6. These extinction tests, during which rats are exposed to the drug-associated cues, permit the characterization of the time course of relapse vulnerability. The reinstatement and extinction procedures described above are regarded as valid animal models, and many investigators use these procedures to investigate the neuronal mechanisms underlying relapse to drug seeking4,7.

Yavin Shaham and Bruce T. Hope are in the Behavioral Neuroscience Branch, Intramural Research Program, National Institute on Drug Abuse, National Institutes of Health, Department of Health and Human Services, 5500 Nathan Shock Drive, Baltimore, Maryland 21224, USA. e-mail: [email protected]

According to a popular hypothesis, chronic drug exposure causes long-lasting molecular, cellular and neurochemical adaptations in the brain that underlie different facets of addiction, including prolonged relapse vulnerability after cessation of drug use7–9. Here we assess the degree to which specific drug-induced neuroadaptations contribute to reinstatement of drug seeking induced by drug priming, drug cues or stress. We also discuss the role of druginduced neuroadaptations in the progressive increase in cocaine seeking after withdrawal, a phenomenon called incubation of cocaine craving6 that was recently demonstrated in humans (T.R. Kosten, T.A. Kosten, J. Poling & A. Oliveto, College of Problems on Drug Dependence Annual Meeting Abstracts, p. 90, 2005). Our assessment suggests that although drug-induced neuroadaptations are involved in relapse to cocaine seeking, the available preclinical data do not allow us to conclude that these neuroadaptations are the main cause of long-term relapse vulnerability in humans. Neuroadaptations and reinstatement Over a decade ago, the authors of two influential reviews10,11 hypothesized that neuroadaptations induced by repeated drug exposure that produce enduring psychomotor sensitization also underlie drug-, cueand stress-induced relapse to drug seeking. Sensitization refers to the enhanced psychomotor response (quantified by measuring locomotor activity and stereotypy) that occurs after repeated exposure to psychostimulant or opiate drugs; this sensitized response to drugs can persist for many months after the last drug exposure10.


The neuronal systems involved in enduring psychostimulant and opiate sensitization are also involved in drug priming–induced reinstatement of drug seeking. The magnitude of drug-induced reinstatement of psychostimulant or opiate seeking is associated with sensitized drug-induced locomotor response, and both behaviors are associated with enhanced dopamine release in the nucleus accumbens12,13. Sensitizing regimens of psychostimulants diminish cysteineglutamate transporter activity, which leads to decreases in basal non-synaptic glutamate levels in the nucleus accumbens; this decrease in nonsynaptic glutamate is critical for acute cocaine-induced synaptic glutamate release7. Reversal of this neuroadaptation with systemic injections of cysteine prodrugs prevents both cocaine-induced reinstatement and cocaineinduced glutamate release14, demonstrating a link between a neuroadaptation involved in psychomotor sensitization and relapse to cocaine seeking. These and related findings led to the suggestion that pharmacotherapies for drug relapse prevention should aim at reversing cocaine-induced neuroadaptations7. A recent clinical trial based on this hypothesis, in which cocaine-dependent patients were treated with modafinil (a drug that increases glutamate transmission), has led to tentatively promising results15. Several issues should be considered in evaluating the implications of the above findings for general treatment strategies. These data were derived primarily from studies using cocaine-trained rats in which drug priming provoked relapse behavior. However, drug relapse in both humans and laboratory animals is often provoked by

1437


C O M M E N TA R Y conditions other than reexposure to the selfadministered drug, including exposure to drug-associated cues1 and stress2,5,16. There is evidence that the neuronal mechanisms underlying drug seeking induced by stress, cues or drug priming are not identical16, so it is important to verify whether manipulations that reverse drug-induced neuroadaptations also prevent cue- or stress-induced reinstatement. Preliminary data suggest that this might not be the case (Z.X. Xi, J. Gilbert, A. Campos, C.R. Ashby & E.L. Gardner, Soc. Neurosci. Abstr. 691.9, 2004). Cue- or footshock stress–induced reinstatement was not affected by systemic injections of an antagonist of the mGluR5 metabotropic glutamate receptor, which increases basal nucleus accumbens glutamate levels and blocks cocaine-induced glutamate release and cocaine-induced reinstatement. On the basis of studies of crosssensitization of locomotor activity between drug and stressors11, it has been suggested that neuroadaptations associated with psychomotor sensitization are also involved in stress-induced reinstatement10. However, unlike drug priming, the effect of footshock stress on reinstatement is not correlated with its effect on locomotor activity and nucleus accumbens dopamine release17. These findings suggest that the neuroadaptations that mediate enduring psychomotor sensitization, which also potentially contribute to persistent drug priming–induced reinstatement (see above), are not likely to be involved in stress-induced reinstatement of drug seeking. A recent study, however, may be the first demonstration of cocaine-induced neuroadaptation specific to stress-induced reinstatement18. Intermittent footshock stress increases the levels of the stress neurohormone corticotropin-releasing factor (CRF) in the ventral tegmental area (VTA, the cell body region of the mesolimbic dopamine reward system), and blockade of CRF receptors in this brain area attenuates stress-induced reinstatement of cocaine seeking. Furthermore, in cocaine-experienced but not in cocainenaive rats, exposure to stress or local infusion of CRF induces glutamate release in the VTA, which in turn activates VTA dopamine neurons; such activation is known to induce reinstatement of drug seeking4. These data suggest that long-lasting cocaine-induced neuroadaptations in VTA neurotransmission are important in stress-induced relapse to cocaine seeking. This drug-induced neuroadaptation in CRF effects is likely to be specific to stress; CRF receptor antagonists block stress-induced but not drug priming– induced reinstatement of drug seeking17.

1438

Neuroadaptations and incubation of craving To account for the persistent propensity for relapse to cocaine use, it has been suggested that craving induced by cocaine cues increases over the first several weeks of withdrawal and remains high over extended drug-free periods19. We and others identified an analogous phenomenon in laboratory animals: time-dependent increases in cocaine seeking induced by exposure to cocaine cues over the first months of withdrawal (incubation of craving)20,21. We subsequently explored whether cocaine-induced neuroadaptations underlie the incubation of cocaine craving. We found that the time-dependent increase in cocaine seeking after withdrawal is associated with increasing peptide levels of the plasticity-related growth factor BDNF in the VTA, nucleus accumbens and amygdala22, and that injections of BDNF directly into the VTA increase cocaine seeking during early withdrawal23. However, sucrose-trained rats demonstrated short-term (several weeks) time-dependent increases in sucrose seeking without alterations in BDNF expression23. On the basis of these findings and other considerations, we suggested that cocaineinduced alterations of mesolimbic BDNF augment, rather than directly mediate, an ongoing incubation process6. We also explored whether the timedependent increase in cocaine seeking involves activation of the extracellular signal-regulated kinases (ERK) signaling pathway in the amygdala24. Cocaine activates the ERK pathway in mesolimbic dopamine areas25. In the amygdala and other brain areas, this pathway is involved in learning and memory processes26 that were hypothesized to be involved in drug addiction and relapse27,28 and may be altered by chronic exposure to abused drugs8,29. We found that exposure to cocaine cues increases ERK phosphorylation in the central, but not basolateral, amygdala after 30 d but not after 1 d of withdrawal, and that after 30 d, inhibition of central amygdala ERK phosphorylation attenuates cocaine seeking. Our data also indicate that glutamate is involved in the activation of the ERK pathway by cocaine cues. We concluded that time-dependent increases in the responsiveness of the central amygdala ERK pathway to cocaine cues mediate the incubation of cocaine craving24. This time-dependent, sensitized response to cocaine cues may be due to cocaine-induced neuroadaptations of the central amygdala ERK pathway. Alternatively, the time-dependent increase in the amygdala ERK’s responsiveness to cues may be a general mechanism for the incubation of craving that also occurs with non-drug rewards6.

Conclusions The results from the studies described above support the notion that drug-induced neuroadaptations are involved in drug relapse, as measured in preclinical models. However, this conclusion is applicable only to cocaine; the role of neuroadaptations in relapse to other abused drugs has not been assessed. Furthermore, to determine whether specific neuroadaptations are causally involved in drug relapse, experimental manipulations should aim to reverse endogenous neuroadaptations and then determine the effect of this reversal on relapse behavior. This experimental approach so far has been applied only to neuroadaptations’ role in cocaine priming– induced reinstatement7,14. Thus, although there is evidence that specific cocaine-induced neuroadaptations are involved in specific forms of cocaine relapse, it is too early to conclude from the available preclinical data that drug-induced brain neuroadaptations underlie long-term relapse vulnerability in humans. In this regard, given that relapse rates in humans are similar for different drugs of abuse30, it seems important to determine the generality of the findings from studies on the role of cocaine-induced neuroadaptations in relapse to other drugs. Data from studies assessing this issue of generality across drugs would also influence medication development. Most drug addicts are polydrug users1, and thus treatment approaches that involve pharmacological reversal of cocaine-induced neuroadaptations7 are more likely to succeed if neuroadaptation-related findings from studies using cocaine do generalize to other drugs. On the basis of the aforementioned similarity in relapse rates across drugs and the clinical situation of polydrug use, in our view the critical issue for future preclinical studies is the identification of specific neuroadaptations that similarly contribute to relapse in rats with a history of stimulant, opiate, alcohol and nicotine self-administration. ACKNOWLEDGMENTS Supported by the National Institute on Drug Abuse, Intramural Research Program. We thank E. Wentzell for editorial assistance and K. Preston for helpful comments. We also would like to acknowledge L. Lu for his contribution to the work described in the section on the role of neuroadaptation in incubation of cocaine craving. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. 1. O’Brien, C.P. Am. J. Psychiatry 162, 1423–1431 (2005). 2. Sinha, R. Psychopharmacology (Berl.) 158, 343–359 (2001). 3. de Wit, H. Exp. Clin. Psychopharmacol. 4, 5–10 (1996).



4. Shaham, Y., Shalev, U., Lu, L., De Wit, H. & Stewart, J. Psychopharmacology (Berl.) 168, 3–20 (2003). 5. Stewart, J. Nebr. Symp. Motiv. 50, 197–234 (2004). 6. Lu, L., Grimm, J.W., Hope, B.T. & Shaham, Y. Neuropharmacology 47 (Suppl.) 1, 214–226 (2004). 7. Kalivas, P.W. Curr. Opin. Pharmacol. 4, 23–29 (2004). 8. Nestler, E.J. Nat. Rev. Neurosci. 2, 119–128 (2001). 9. Wolf, M.E., Sun, X., Mangiavacchi, S. & Chao, S.Z. Neuropharmacology 47 (Suppl.) 61–79 (2004). 10. Robinson, T.E. & Berridge, K.C. Brain Res. Brain Res. Rev. 18, 247–291 (1993). 11. Kalivas, P.W. & Stewart, J. Brain Res. Brain Res. Rev. 16, 223–244 (1991). 12. De Vries, T.J., Schoffelmeer, A.N., Binnekade, R., Mulder, A.H. & Vanderschuren, L.J. Eur. J. Neurosci.

10, 3565–3571 (1998). 13. Vezina, P., Lorrain, D.S., Arnold, G.M., Austin, J.D. & Suto, N. J. Neurosci. 22, 4654–4662 (2002). 14. Baker, D.A. et al. Nat. Neurosci. 6, 743–749 (2003). 15. Dackis, C.A., Kampman, K.M., Lynch, K.G., Pettinati, H.M. & O’Brien, C.P. Neuropsychopharmacology 30, 205–211 (2005). 16. Shalev, U., Grimm, J.W. & Shaham, Y. Pharmacol. Rev. 54, 1–42 (2002). 17. Shaham, Y., Erb, S. & Stewart, J. Brain Res. Brain Res. Rev. 33, 13–33 (2000). 18. Wang, B. et al. J. Neurosci. 25, 5389–5396 (2005). 19. Gawin, F.H. & Kleber, H.D. Arch. Gen. Psychiatry 43, 107–113 (1986). 20. Neisewander, J.L. et al. J. Neurosci. 20, 798–805 (2000).


21. Lu, L., Grimm, J.W., Dempsey, J. & Shaham, Y. Psychopharmacology (Berl.) 176, 101–108 (2004). 22. Grimm, J.W. et al. J. Neurosci. 23, 742–747 (2003). 23. Lu, L., Dempsey, J., Liu, S.Y., Bossert, J.M. & Shaham, Y. J. Neurosci. 24, 1604–1611 (2004). 24. Lu, L. et al. Nat. Neurosci. 8, 212–219 (2005). 25. Berhow, M.T., Hiroi, N. & Nestler, E.J. J. Neurosci. 16, 4707–4715 (1996). 26. Adams, J.P. & Sweatt, J.D. Annu. Rev. Pharmacol. Toxicol. 42, 135–163 (2002). 27. Wise, R.A. J. Abnorm. Psychol. 97, 118–132 (1988). 28. White, N.M. Addiction 91, 921–949 (1996). 29. Everitt, B.J., Dickinson, A. & Robbins, T.W. Brain Res. Brain Res. Rev. 36, 129–138 (2001). 30. Hunt, W.A., Barnett, L.W. & Branch, L.G. J. Clin. Psychol. 27, 455–456 (1971).

1439


C O M M E N TA RY


How do we determine which drug-induced neuroplastic changes are important? Peter W Kalivas Although many drug-induced neural changes are known, progress has been slow in identifying the ones that actually mediate addiction. Identifying changes that are specific to particular elements of the transition from initial to habitual to relapsing drug use may be a fruitful strategy for pinpointing which forms of drug-induced plasticity are critical for addiction.

Adaptive behavioral responses to important environmental events result from a combination of genetic disposition, memories of related events and neuroplasticity induced by the event1,2. A fundamental hypothesis in addiction research is that the pharmacological properties of drugs serve as important environmental cues for inducing neuroplasticity and creating drug-related memories3–6. These processes are thought to mediate the development of the pathological behaviors such as habitual drug use and the overwhelming drive to get drugs (as in relapse). This hypothesis has led to a deluge of experiments identifying cellular changes that may cause neuroplastic events underlying addiction. Although this research has been useful in cataloguing drug-induced neuroplasticity, it does not efficiently distinguish neuroadaptations that mediate addiction from those unrelated to addiction. This distinction is critical, not only for understanding the neurobiology of addiction but also for identifying new pharmacotherapeutic targets for treating addiction. In part, our inefficiency in recognizing the most important addiction-related neuroplastic events results from the sheer enormity of deciphering the biology of neuroplasticity. However, the absence of a procedural framework for integrating animal models of addiction with the discovery and interpretation of drug-induced neuroplasticity hampers investigative efficiency. Addiction develops as a transition from initial to habitual to relapsing drug use. Each stage of this transition can be modPeter W. Kalivas is in the Department of Neurosciences, Medical University of South Carolina, Charleston, South Carolina, USA. e-mail: [email protected]

1440

eled in animals and is paralleled by a transition in drug-induced neuroplasticity7,8. By designing experiments to investigate drug-induced neuroplastic events that are specific to a particular part of this temporal sequence, investigators can distinguish plasticity related to the acute effects of the drug from the changes required for habitual drug use, from the permanent adaptations mediating vulnerability to relapse. The following three phases of neuroplasticity can be inferred using drug self-administration protocols. First, the acquisition of drug selfadministration represents acute neuroplasticity. Second, patterns of daily self-administration represent neuroplasticity associated with the habitual drug use. Third, the reinstatement of drug seeking represents neuroplasticity underlying relapse. The involvement of specific neuroplastic changes in the three stages of addiction can be validated directly using mice with a deletion/induction of the relevant gene. However, because drug-induced neuroplasticity is often selective for a particular brain nucleus, it is preferable to locally stimulate or inhibit protein expression using in vivo DNA, RNA or protein transfer techniques such as viral transfection, small interfering RNA (siRNA) or Tat-fusion proteins. Examples are shown below of how temporal categorizing can be used to determine the relevance of druginduced neuroplasticity in addiction. The cellular responses to acute drug administration are generally short-lived, are closely tied to the molecular site of drug action and are similar to plasticity identified in in vitro models of synaptic plasticity, including the induction of immediate early gene (IEG) products such as c-Fos, Homer1a, NAC-1

and Narp. Unfortunately, it is unclear which of these changes are antecedents to habitual drug use because they have not been validated in animal models of the acquisition of drug self-administration. However, a few IEGs induced by psychostimulant administration have been linked to behavioral sensitization, a form of enduring behavioral plasticity in which repeated noncontingent psychostimulant administration produces a progressive increase in locomotor activity3. This lessthan-optimal model of addiction-related neuroplasticity has revealed that rather than promoting psychostimulant-induced behavioral plasticity, the induction of NAC-1 in the nucleus accumbens inhibits the development of cocaine-induced locomotor sensitization9, whereas the induction of Homer1a is without measurable consequence10. The second temporal category of druginduced neuroplasticity mediates habitual drug use and is characterized by changes in proteins that emerge with repeated drug exposure and then disappear during withdrawal. Included in this category are transcription factors such as ∆FosB and proteins regulating dopamine and glutamate transmission. The induction of ∆FosB is a well-characterized neuroplastic event in terms of relevance to addiction, and many of the gene deletion and transfection techniques outlined above show that ∆FosB is necessary for the development of behavioral sensitization and conditioned place preference8,11. Similarly, increases in GluR1 in the ventral tegmental area are necessary for the development of cocaineinduced behavioral plasticity12. However, the relevance of most cellular changes in this temporal category is not yet validated in animal models of drug-induced plasticity. Importantly, as most



C O M M E N TA R Y studies use investigator-administered drugs, it is also unknown if many drug-induced adaptations are elicited by self-administered drugs. This lack of information is a critical shortcoming for validating claims that extended periods of self-administration induce behavioral changes that more accurately model addiction than do more common limited access protocols13,14. Unfortunately, without objective measures of cellular plasticity, it remains unclear whether distinct behaviors are emerging or the behaviors are dose-dependent extensions resulting from higher daily self-administered drug intake. A systematic comparison of the cellular plasticity induced by the limited- versus extended-access protocols could clarify this issue. The form of drug-induced neuroplasticity that has been most successfully evaluated in animal models endures for weeks of drug abstinence and may underlie relapse. For example, plasticity in Erk signaling in the amygdala and proteins regulating glutamate transmission in the prefrontal cortex projection to the accumbens are critical in the reinstatement of drug-seeking3,15,16. In contrast, neuroadaptations in dopamine transmission in the nucleus accumbens that are critical for maintaining cocaine self-administration7 are not necessary for the reinstatement of cocaine seeking17. The relatively reduced importance of dopamine in the nucleus accumbens in the

C O M M E N TA R Y reinstatement of cocaine seeking contrasts with the critical role identified for enhanced dopamine transmission in the expression of behavioral sensitization3, highlighting the importance of validating neuroadaptations in the self-administration/reinstatement protocol. Moreover, the validation between druginduced neuroplasticity and the reinstatement model of relapse has led to identification of new pharmacotherapeutic targets such as the cysteine-glutamate exchanger and glutamate receptor subtypes18,19. Attempts to link molecular neuroplasticity with the physiology of neuronal networks have been another important outcome of integrating enduring drug-induced plasticity with the reinstatement model15,16,20. This linkage is a necessary intermediary toward an integrated understanding of the neurobiology of addiction. Over the next decade, many new addictionrelated neuroplastic changes will be discovered. These drug-induced neuroadaptations should be identified and catalogued according to the temporal characteristics of their appearance and disappearance, and animal models should be used to target specific temporal stages of addiction. This approach will delineate an integrated sequence of neuroplastic changes underlying addiction and, in this way, identify the neuroplastic events that can best serve as pharmacotherapeutic targets for treating addiction.


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

1. Keitz, M., Martin-Soelch, C. & Leenders, K.L. Neural Plast. 10, 121–128 (2003). 2. Wise, R.A. Nat Rev. Neurosci. 5, 483–494 (2004). 3. Robinson, T.E. & Berridge, K.C. Annu. Rev. Psychol. 54, 25–53 (2003). 4. Kauer, J.A. Annu. Rev. Physiol. 66, 447–475 (2004). 5. Kelley, A.E. Neuron 44, 161–179 (2004). 6. Jones, S. & Bonci, A. Curr. Opin. Pharmacol. 5, 20–25 (2005). 7. Koob, G.F. et al. Neurosci. Biobehav. Rev. 27, 739–749 (2004). 8. Nestler, E.J., Barrot, M. & Self, D.W. Proc. Natl. Acad. Sci. USA 98, 11042–11046 (2001). 9. Mackler, S.A. et al. J. Neurosci. 20, 6210–6217 (2000). 10. Szumlinski, K.K. et al. Neuropsychopharmacology published online 14 September 2005 (doi:10.1038/ sj.npp.1300890). 11. McClung, C.A. & Nestler, E.J. Nat. Neurosci. 6, 1208– 1215 (2003). 12. Carlezon, W.A., Jr. & Nestler, E.J. Trends Neurosci. 25, 610–615 (2002). 13. Ahmed, S.H., Kenny, P.J., Koob, G.F. & Markou, A. Nat. Neurosci. 5, 625–626 (2002). 14. Vanderschuren, L.J. & Everitt, B.J. Science 305, 1017– 1019 (2004). 15. Lu, L. et al. Nat. Neurosci. 8, 212–219 (2005). 16. Kalivas, P.W., Volkow, N. & Seamans, J. Neuron 45, 647–650 (2005). 17. Kalivas, P.W. & Volkow, N.D. Am. J. Psychiatry 162, 1403–1413 (2005). 18. Baker, D.A. et al. Nat. Neurosci. 6, 743–749 (2003). 19. Baptista, M.A., Martin-Fardon, R. & Weiss, F. J. Neurosci. 24, 4723–4727 (2004). 20. Canales, J.J. Neurobiol. Learn. Mem. 83, 93–103 (2005).

1441


C O M M E N TA RY


Plasticity of reward neurocircuitry and the ‘dark side’ of drug addiction George F Koob & Michel Le Moal Drug seeking is associated with activation of reward neural circuitry. Here we argue that drug addiction also involves a ‘dark side’—a decrease in the function of normal reward-related neurocircuitry and persistent recruitment of anti-reward systems. Understanding the neuroplasticity of the dark side of this circuitry is the key to understanding vulnerability to addiction.

Drug addiction has been conceptualized as a progression from impulsive to compulsive behavior, ending in chronic, relapsing drug taking. Patients with impulse control disorders experience an increasing sense of tension or arousal before committing an impulsive act; pleasure, gratification or relief at the time of committing the act; and then regret, selfreproach or guilt after the act1. In contrast, patients with compulsive disorders experience anxiety and stress before committing a compulsive repetitive behavior, then relief from the stress by performing the behavior1. In addiction, drug-taking behavior progresses from impulsivity to compulsivity in a three-stage cycle: binge/intoxication, withdrawal/negative affect and preoccupation/anticipation2. In the impulsive stage, the drive for the drug-taking behavior is positive reinforcement, in which stimuli increase the probability of the response. As individuals move to the compulsive stage, the drive transitions to negative reinforcement, in which removal of the aversive state increases the probability of the response. Different theoretical perspectives from experimental psychology (positive and negative reinforcement framework), social psychology (self-regulation failure framework) and neurobiology (counteradap-

George F. Koob is in the Molecular and Integrative Neurosciences Department, The Scripps Research Institute, La Jolla, California, 92037, USA, and Michel Le Moal is at the Laboratoire de Physiopathologie des Comportements, Institut National de la Santé et de la Recherche Médicale, Unite 588, Université Victor Segalen Bordeaux 2, Bordeaux, France. e-mail: [email protected]

1442

tive and sensitization framework) can be superimposed on the stages of the addiction cycle2. These stages are thought to feed into each other, becoming more intense and ultimately leading to the pathological state known as addiction. Our thesis is that addiction involves a longterm, persistent plasticity in the activity of neural circuits mediating two different motivational systems: decreased function of brain reward systems driven by natural rewards, and recruitment of anti-reward systems that drive aversive states. The concept of anti-reward is based on the hypothesis that there are brain systems in place to limit reward (see footnote in ref. 2), an ‘opponent process’ concept that is a general feature of biological systems3. From a neurobiological perspective, progression through the three stages of the addiction cycle induces plasticity in neural circuitry that drives compulsive drug taking, narrowing the behavioral repertoire to drug seeking. Animal models have been developed that have face validity (resembles the human condition) and some construct validity (possesses explanatory power) for all three stages of the addiction cycle and the transition to drug addiction. Acute selfadministration of drugs (intravenous and oral) has construct validity for drug intoxication and elements of drug binges in humans. Self-stimulation and place conditioning (learning to avoid a location previously paired with an aversive stimulus or state) are sensitive measures of ‘motivational’ withdrawal. Cue-induced or stress-induced reinstatement has face validity and is currently under test for construct validity. Although more construct validation relative to the human condition is needed, neural substrates for each of the stages have already been identified using these models4. Different

theoretical positions favor models from each of the three stages, although in our view, models of the transition to dependence have the most heuristic value for the human condition. For the binge-intoxication stage, studies of the acute reinforcing effects of drugs of abuse per se have identified key neurobiological substrates. Important anatomical circuits include the mesocorticolimbic dopamine system originating in the ventral tegmental area and projecting to the nucleus accumbens and the extended amygdala. The extended amygdala comprises the central nucleus of the amygdala, the bed nucleus of the stria terminalis and a transition area in the medial (shell) part of the nucleus accumbens and a major projection to the lateral hypothalamus. Neurotransmitter/ neuromodulator systems implicated in the acute reinforcing effects of drugs of abuse in these neuroanatomical sites include dopamine, opioid peptides, γ-aminobutyric acid (GABA), glutamate, neuropeptide Y and glucocorticoids of the hypothalamic-pituitary-adrenal (HPA) axis5. There is strong evidence for a role of dopamine in the acute reinforcing actions of psychostimulants, for opioid peptide receptors in the acute reinforcing effects of opioids, and for GABA and opioid peptides in the acute reinforcing actions of alcohol. Although acute drug use is not, in and of itself, addiction, the study of the neuropharmacological mechanisms for the acute reinforcing effects of drugs of abuse has had heuristic value in two major domains. Such studies provide a framework for examining neuroadaptive changes in the reward circuits with the development of addiction, and they also provide a valid model for development of medications to treat excessive drug intake (such as naltrexone for excessive drinking).



C O M M E N TA R Y For the purposes of this opinion piece, the withdrawal/negative affect stage can be defined as the presence of motivational signs of withdrawal in humans: chronic irritability, emotional pain, malaise, dysphoria, alexithymia and loss of motivation for natural rewards. It is characterized in animals by increases in reward thresholds during withdrawal from all major drugs of abuse. Significant plasticity occurs in the neurotransmitter circuits identified above as critical for the acute reinforcing effects of drugs of abuse. In animal models of the transition to addiction, similar changes in brain reward threshold occur that temporally precede and highly correlate with escalation in drug intake6. During such acute withdrawal, there is decreased activity of the mesocorticolimbic dopamine system as measured by electrophysiological recordings and in vivo microdialysis, and there is also decreased activity in opioid peptide, GABA, glutamate and neuropeptide Y in elements of the extended amygdala and/or nucleus accumbens. Human imaging studies of addicts during withdrawal or protracted abstinence give results that are consistent with the animal studies, including decreases in dopamine D2 receptors (hypothesized to reflect hypodopaminergic functioning) and hypoactivity of the orbitofrontal-infralimbic cortex system7. These neurotransmitter/neuromodulator system changes may persist during protracted abstinence and include hypofunctioning of the HPA axis8. More importantly for the present thesis, as dependence and withdrawal develop, brain anti-reward systems such as corticotropinreleasing factor (CRF), norepinephrine and dynorphin are recruited. For example, extracellular CRF in the extended amygdala is increased during acute withdrawal from drugs of abuse, and critically, CRF receptor antagonists block excessive drug taking during dependence8. These neurotransmitter systems are activated during the development of excessive drug taking, and this activation is manifest when the drug in removed (acute withdrawal and protracted abstinence). The observation that CRF receptor antagonists in the amygdala can block excessive drug intake associated with the development of dependence provides a compelling example of a key player in the plasticity of the extended amygdala in the development of addiction. We hypothesize that anti-reward circuits are recruited as between-system neuroadaptations9 during the development of addiction, producing aversive or stress-like states8,10,11. We further hypothesize that within the motivational circuits of the extended amygdala, the combination of decreases in reward neurotransmitter function and recruitment of anti-reward systems

provides a powerful source of negative reinforcement that defines compulsive drug-seeking behavior and addiction. The development of the aversive emotional state that drives the negative reinforcement of addiction is here termed the ‘dark side’ of addiction. We further hypothesize that this chronic aversive state manifested by motivational signs of withdrawal in humans is produced in part by recruitment of the brain anti-reward systems. We believe that research in this domain has been largely neglected by the field, mainly because of an excessive focus on psychostimulant drugs and reward pathways (largely misattributed to the mesolimbic dopamine system). A critical problem in drug addiction is chronic relapse, in which addicts return to compulsive drug taking long after acute withdrawal. This corresponds to the preoccupation/ anticipation stage of the addiction cycle, outlined above. Both animal and human neuroimaging studies show that the prefrontal cortex system (orbitofrontal, medial prefrontal, prelimbic/cingulate) and the basolateral amygdala are key mediators of drug- and cue-induced reinstatement in animal models and craving and relapse in humans. Neurotransmitter systems implicated in drug- and cue- or context-induced craving again include dopamine, opioid peptides, glutamate and GABA. Neurotransmitter/neuromodulator systems implicated in stress-induced relapse include CRF, glucocorticoids and norepinephrine, suggesting that there is reactivation of both reward and anti-reward systems during relapse12–14. Although the relapse models have face validity, there remain serious concerns about construct validity relative to the human condition. Most reinstatement studies to date have been done with nondependent animals and may be of little more relevance to the study of addiction than studies of reinstatement of responding for a nondrug, high-incentive stimulus such as a saccharin solution, a control rarely explored in reinstatement studies. In other words, do the neuropharmacological substrates for the reinstatement of responding for saccharin—or any other nondrug reinforcer of high incentive value—parallel those of the neural substrates for nondependent doses of cocaine or heroin? We also hypothesize that the dysregulations that constitute the dark side of drug addiction persist during protracted abstinence to set the tone for vulnerability to ‘craving’ by activation of the drug-, cue- and stress-induced reinstatement neurocircuits now driven by a reorganized and hypofunctioning prefrontal system15. A reward allostasis model is proposed to explain how dysregulation of the reward system associated with the development of motivational aspects of withdrawal is a major


source of potential allostatic changes that drive and maintain addiction2. In this context, ‘allostasis’ is defined as the process of achieving stability of the reward system through change. An allostatic state is a state of chronic deviation of the reward system from its normal (homeostatic) operating level, which ultimately leads to the pathological state of addiction. More specifically, in drug addiction, allostasis is the process of attempting to maintain apparent reward function stability by changes in reward and anti-reward system neurocircuitry5. Neuroplasticity in the natural reward system is highlighted by decreased dopaminergic activity and hypofrontality. Neuroplasticity in the anti-reward system is highlighted by increased CRF function and is hypothesized to be particularly slow to return to homeostasis. This makes the system that drives the dark side potentially more important for driving dependence than decreases in natural reward function. For example, there is evidence of residual dysregulation of the HPA axis16 and of the brain CRF system weeks after acute withdrawal from alcohol17. Similar observations are made in human addicts18. In contrast, withdrawal-induced decreases in dopaminergic function are relatively transient10. Thus, the drug addict, futilely in the short term, attempts to misregulate these drug-induced neuroplasticities by taking more drug, which only serves in the long-term to dysregulate the system further, leading to a worsening of the condition. The most prominent functional increase of the anti-reward system identified to date involves activation of the CRF-HPA axis and subsequent activation of the CRF– extended amygdala system, but other neuroadaptive processes associated with behavioral responses to stressors also may have potential roles such as neuropeptide Y, dynorphin and norepinephrine. The allostatic dysregulated reward state not only produces the motivational symptoms of acute withdrawal and protracted abstinence, but also provides the background by which drug priming, drug cues and acute stressors acquire even more power to elicit drug-seeking behavior. Clearly appropriate, construct-validated animal models for the stages of the addiction cycle, the motivational aspects of drug-seeking and genetic vulnerability to addiction are critical to test these hypotheses. Recently, animal models with excellent face validity for the transition to addiction have been developed, which include escalated drug intake driven by dependence6,15, responding for drug despite adverse consequences19,20 and a narrowing of the behavioral repertoire for drug19,20. These models have strong face validity for the Diagnostic and Statistical Manual of

1443


C O M M E N TA R Y Mental Disorders1 and International Statistical Classification of Diseases21 criteria for addiction, are currently under test for construct validity and show promise for measuring the genetic and environmental contributions to vulnerability to addiction. Thus, in our view, a perspective often overlooked in the drug abuse field is that there is a long-term persistent decrease in function of normal motivational systems driven by two sources: decreased function of brain reward systems (mediating natural rewards) and increased anti-reward systems (recruited as an opponent process to excessive activation of the brain reward system)22. It is the deficit state for normal reward, produced by excessive drug taking, that provides the core element of the motivation to seek drugs, not a hyperactive or sensitized reward state for drugs per se. In our view, understanding the neuroplasticity of the dark side of this circuitry will be the key to understanding individual vulnerability to addiction.

1444

ACKNOWLEDGMENTS We thank M. Arends for his assistance with the preparation of this manuscript. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. 1. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders 4th edn. (American Psychiatric Press, Washington, D.C., 1994). 2. Koob, G.F. & Le Moal, M. Science 278, 52–58 (1997). 3. Martin, W.R. in The Addictive States (ed. Wikler, A.) 206–225 (Williams and Wilkins, Baltimore, 1968). 4. Shippenberg, T.S. & Koob, G.F. in Neuropsychopharmacology: The Fifth Generation of Progress (eds. Davis, K.L., Charney, D., Coyle, J.T. & Nemeroff, C.) 1381–1397 (Lippincott Williams and Wilkins, Philadelphia, 2002). 5. Koob, G.F. & Le Moal, M. Neuropsychopharmacology 24, 97–129 (2001). 6. Ahmed, S.H., Kenny, P.J., Koob, G.F. & Markou, A. Nat. Neurosci. 5, 625–626 (2002). 7. Volkow, N.D., Fowler, J.S. & Wang, G.J. J. Clin. Invest. 111, 1444–1451 (2003). 8. Koob, G.F. Alcohol. Clin. Exp. Res. 27, 232–243 (2003).

9. Koob, G.F. & Bloom, F.E. Science 242, 715–723 (1988). 10. Nestler, E.J. Nat. Rev. Neurosci. 2, 119–128 (2001). 11. Aston-Jones, G., Delfs, J.M., Druhan, J. & Zhu, Y. Ann. NY Acad. Sci. 877, 486–498 (1999). 12. Piazza, P.V. & Le Moal, M. Annu. Rev. Pharmacol. Toxicol. 36, 359–378 (1996). 13. Shaham, Y., Erb, S. & Stewart, J. Brain Res. Brain Res. Rev. 33, 13–33 (2000). 14. See, R.E., Fuchs, R.A., Ledford, C.C. & McLaughlin, J. Ann. NY Acad. Sci. 985, 294–307 (2003). 15. Le Moal, M. in Psychopharmacology: The Fourth Generation of Progress (eds. Bloom, F.E. & Kupfer, D.J.) 283–294 (Raven, New York, 1995). 16. Rasmussen, D.D. et al. Alcohol. Clin. Exp. Res. 24, 1836–1849 (2000). 17. Valdez, G.R. et al. Alcohol. Clin. Exp. Res. 26, 1494– 1501 (2002). 18. Kreek, M.J. & Koob, G.F. Drug Alcohol Depend. 51, 23–47 (1998). 19. Deroche-Gamonet, V., Belin, D. & Piazza, P.V. Science 305, 1014–1017 (2004). 20. Vanderschuren, L.J. & Everitt, B.J. Science 305, 1017– 1019 (2004). 21. World Health Organization. International Statistical Classification of Diseases and Related Health Problems, 10th revision (World Health Organization, Geneva, 1992). 22. Solomon, R.L. & Corbit, J.D. Psychol. Rev. 81, 119– 145 (1974).




PERSPECTIVE

Is there a common molecular pathway for addiction? Eric J Nestler Drugs of abuse have very different acute mechanisms of action but converge on the brain’s reward pathways by producing a series of common functional effects after both acute and chronic administration. Some similar actions occur for natural rewards as well. Researchers are making progress in understanding the molecular and cellular basis of these common effects. A major goal for future research is to determine whether such common underpinnings of addiction can be exploited for the development of more effective treatments for a wide range of addictive disorders.

Drugs of abuse are highly diverse chemical substances. Accordingly, each drug binds to its distinct initial protein target in the brain and periphery and elicits a distinct combination of behavioral and physiological effects upon acute administration. However, despite these disparate mechanisms of action and pharmacological effects, all drugs of abuse cause certain common effects after both acute and chronic exposure. The drugs are all acutely rewarding, which promotes repeated drug intake and leads eventually, in vulnerable individuals, to addiction—a loss of control over drug use. All drugs also produce similar negative emotional symptoms upon drug withdrawal, a prolonged period of sensitization, and associative learning toward drug-related environmental cues. These adaptations are thought to contribute to the intense drug craving and relapse that can persist even after long periods of abstinence, although the relative contribution of each mechanism remains a subject of considerable controversy. The question addressed by this Perspective is whether there are common neural and molecular pathways underlying these shared rewarding and addicting actions of drugs of abuse. Common actions on brain reward circuits There is now considerable evidence, from animal models and more recently from humans, that all drugs of abuse converge on a common circuitry in the brain’s limbic system1–5. Most attention has been given to the mesolimbic dopamine pathway, which includes dopaminergic neurons in the ventral tegmental area (VTA) of the midbrain and their targets in the limbic forebrain, especially the nucleus accumbens (NAc). This VTA-NAc pathway is one of the most important substrates for the

Eric J. Nestler is in the Department of Psychiatry and Center for Basic Neuroscience, The University of Texas Southwestern Medical Center, 5323 Harry Hines Blvd., Dallas, Texas 75390-9070, USA. e-mail: [email protected] Published online 26 October 2005; doi:10.1038/nn1578


acute rewarding effects of all drug of abuse, and research over the past several decades has delineated how each drug, regardless of its distinct mechanism of action, converges on the VTA and NAc with common acute functional effects (Fig. 1). Each drug activates dopaminergic transmission in the NAc and many produce dopamine-like, yet dopamine-independent effects on the same NAc neurons, in many cases via indirect, circuit-level actions1–8. In addition, several drugs (see Fig. 1 legend) seem to activate the brain’s endogenous opioid and cannabinoid systems within the VTA-NAc pathway, as exemplified by reduced drug effects in cannabinoid and opioid receptor knockout mice, which further underscores shared acute mechanisms of drug action1,8. On the basis of these common acute actions, one would expect that chronic exposure to drugs of abuse would also cause common chronic functional changes in the VTA-NAc pathway. Indeed, numerous common chronic adaptations have been described, examples of which are discussed in the next sections. Consistent with common mechanisms of addiction are the observations that certain drugs of abuse, under particular experimental conditions, can induce crosstolerance and cross-sensitization to one another with respect to their locomotor activating and rewarding effects9,10. More recent work has established that several additional brain areas that interact with the VTA and NAc are also essential for acute drug reward and chronic changes in reward associated with addiction. These regions include the amygdala (and related structures of the socalled ‘extended amygdala’), hippocampus, hypothalamus and several regions of frontal cortex, among others1,2,4,10–13. Some of these areas are part of the brain’s traditional memory systems; this has led to the notion, now supported by increasing evidence, that important aspects of addiction involve powerful emotional memories2,4,5,11–13. Growing evidence indicates that the VTA-NAc pathway and the other limbic regions cited above similarly mediate, at least in part, the acute positive emotional effects of natural rewards, such as food, sex and social interactions14,15. These same regions have also been implicated in the so-called ‘natural addictions’ (that is, compulsive consumption of natural rewards) such as pathological overeating, pathological gambling and sexual addictions. Preliminary findings suggest that shared pathways may be involved: two examples are cross-sensitization that occurs between natural rewards and drugs of abuse16 and similar abnormalities found in brain imaging scans in drug and natural addictions4. However, it must be emphasized that the mechanisms underlying natural addictions are much less well understood than those underlying drug addictions because the animal models are far less straightforward, and the clinical syndromes of the natural addictions are likely to be much more heterogeneous. Nevertheless, early findings in the field raise the possibility that the similar behavioral pathology that characterizes drug addictions and certain natural addictions may be mediated, at least in part, by common neural and molecular mechanisms.

1445

PERSPECTIVE Nicotine Alcohol

the negative emotional symptoms as well as many of the somatic symptoms that occur Glutamate inputs upon drug withdrawal, and may contribute (e.g., from cortex) – to drug craving and relapse as well. CRF can therefore be viewed as an example of ‘oppoAlcohol Opiates VTA nent process’–like changes that drugs induce GABA ? interneuron PCP in the brain that serve to counteract drug – – effects and drive withdrawal symptoms when Alcohol ? Stimulants the drug is discontinued1. Current work is + focusing on the molecular basis of this hyperfunctional CRF system, which presumably + DA DA Nicotine involves adaptations within amygdala neurons or their inputs (see below). Cannabinoids – Glutamate + Another common adaptation to chronic drug inputs use is cortical ‘hypofrontality’: namely, reduced (e.g., from amygdala baseline activity of several regions of frontal VTA NAc PPT/LDT) cortex, as inferred from brain imaging studies4. These regions (for example, prefrontal cortex, Figure 1 Highly simplified scheme of converging acute actions of drugs of abuse on the VTA-NAc. anterior cingulate cortex and orbitofrontal corDrugs of abuse, despite diverse initial actions, produce some common effects on the VTA and NAc1–8. tex) control executive function, including workStimulants directly increase dopaminergic transmission in the NAc. Opiates do the same indirectly: ing memory, attention and behavioral inhibition they inhibit GABAergic interneurons in the VTA, which disinhibits VTA dopamine neurons. Opiates and are important in controlling an individual’s also directly act on opioid receptors on NAc neurons, and opioid receptors, like D2 dopamine (DA) receptors, signal via Gi; hence, the two mechanisms converge within some NAc neurons. The actions response to environmental stimuli, in part via of the other drugs remain more conjectural. Nicotine seems to activate VTA dopamine neurons directly glutamatergic projections from these regions via stimulation of nicotinic cholinergic receptors on those neurons and indirectly via stimulation of its to the NAc and VTA. Impressive evidence from receptors on glutamatergic nerve terminals that innervate the dopamine cells. Alcohol, by promoting rodent models and from human brain imaging GABAA receptor function, may inhibit GABAergic terminals in VTA and hence disinhibit VTA dopamine studies demonstrates that chronic exposure to neurons. It may similarly inhibit glutamatergic terminals that innervate NAc neurons. Many additional any of several drugs of abuse causes complex mechanisms (not shown) are proposed for alcohol. Cannabinoid mechanisms seem complex, and they involve activation of CB1 receptors (which, like D2 and opioid receptors, are Gi linked) on glutamatergic changes in these frontal cortical regions and and GABAergic nerve terminals in the NAc, and on NAc neurons themselves. Phencyclidine (PCP) may their glutamatergic outputs, which are impliact by inhibiting postsynaptic NMDA glutamate receptors in the NAc. Finally, there is some evidence cated in the profound impulsivity (acting on that nicotine and alcohol may activate endogenous opioid pathways and that these and other drugs sudden urges to take a drug) and compulsivof abuse (such as opiates) may activate endogenous cannabinoid pathways (not shown). PPT/LDT, ity (being driven by irresistible inner forces to peduncular pontine tegmentum/lateral dorsal tegmentum. take a drug) that characterizes a state of addiction4,9,11,18. The chronic drug-treated state is associated with reduced basal activity of cortical pyramidal neurons and Common circuit-level adaptations Just as all drugs of abuse increase dopaminergic transmission to the NAc a reduced sensitivity of the neurons to activation by natural rewards. This after acute administration, they also produce common adaptations in presumably underlies the hypofrontality noted in human brain scans. dopamine function after chronic exposure. These adaptations seem to be In contrast, these neurons are hypersensitive to activation by drugs of complex in that different effects have been reported by numerous labora- abuse as well as drug-associated stimuli. These drug-induced changes in tories even for the same drug, partly because of differing drug doses, routes glutamatergic transmission to the NAc parallel the changes reported in of administration and dosing regimens. Nevertheless, it is possible to piece dopaminergic transmission to the NAc discussed earlier (Fig. 2). together the following scheme1,3,5. Chronic exposure to any of several drugs of abuse causes an impaired dopamine system, which can be viewed as a Common cellular and molecular adaptations homeostatic response to repeated drug activation of the system (in other Chronic exposure to drugs of abuse causes numerous common adapwords, tolerance; Fig. 2). After chronic drug use, baseline levels of dopamine tations at the cellular and molecular level in the VTA-NAc and other function are reduced, and normal rewarding stimuli may be less effective brain reward regions. There are too many adaptations to describe here at eliciting typical increases in dopaminergic transmission. These changes comprehensively; only a few illustrative examples are included (Fig. 2). may contribute to the negative emotional symptoms observed between However, whereas the occurrence of such shared adaptations is indisputdrug exposures or upon drug withdrawal. At the same time, chronic drug able, the behavioral consequences of each of these adaptations remain exposure seems to sensitize the dopamine system, with greater increases in uncertain, along with the extent to which they mediate common behavdopaminergic transmission occurring in response to the drug in question ioral abnormalities associated with drug and natural addictions. Numerous types of drugs of abuse, including cocaine, amphetamine, and to drug-associated cues5,9,10,13.This sensitization can last long after opiates, alcohol or nicotine, induce a long-term potentiation (LTP)-like drug-taking ceases and may relate to drug craving and relapse. Chronic drug states are also associated with common changes in state in VTA dopamine neurons19–22. This sensitized state is mediated via central corticotropin releasing factor (CRF) systems. Abrupt with- increases in AMPA glutamate receptor responsiveness, which may occur drawal from virtually any drug of abuse leads to activation of CRF- via induction of the GluR1 AMPA glutamate receptor subunit and altered containing neurons in the amygdala17. These neurons, classically intracellular trafficking of AMPA receptors in these neurons21,23. These characterized for their involvement in fear and other aversive states, adaptations in glutamatergic transmission have been related directly to innervate many forebrain and brainstem regions. We now know that sensitized behavioral responses to drugs of abuse, although aspects of this activation of these neurons during drug withdrawal partly mediates model remain controversial23. Alterations in GABAergic regulation of VTA Opioid peptides

Ann Thomson

Opiates


+

1446


dopamine neurons have also been implicated for opiates24, but whether similar changes occur with other drugs of abuse is not yet known. Chronic administration of any of several drugs of abuse, including cocaine, amphetamine, opiates, alcohol and nicotine, also increases levels of tyrosine hydroxylase (TH), the rate-limiting enzyme in dopamine biosynthesis, in the VTA25,26. Concomitantly, decreased TH levels or activity in VTA nerve terminals in the NAc are reported under certain experimental conditions. This latter adaptation could mediate the reduction in dopaminergic signaling seen after chronic drug exposure (see above). Recent evidence implicates the transcription factor CREB (cAMP response element binding protein), which is also activated in VTA by several drugs of abuse after chronic administration, in mediating drug induction of GluR1 and TH in this region27 as well as in some of the behavioral plasticity associated with addiction27–29. Reduced amounts of neurofilament proteins seen within the VTA after chronic opiate, cocaine or alcohol exposure may be a biochemical marker of common morphological changes to VTA neurons induced by these drugs25,30. To date, such changes have been documented only for opiates, which, after chronic exposure, reduce the size of cell bodies and the caliber of proximal processes of VTA dopamine neurons. Reduced neurofilament levels could also account for the impaired axonal transport from the VTA to the NAc observed after chronic opiates. This latter finding could, in turn, explain the disparity between the higher levels of TH seen in VTA and the lower levels seen in NAc. Although the functional consequences of these changes are not known, one can speculate that they reflect a fundamental impairment of the dopamine cells. Interestingly, infusion of any of several neurotrophic factors into the VTA prevents these morphological changes and also produces sensitized behavioral responses to several drugs of abuse, whereas blockade of endogenous neurotrophic factors exerts the opposite effects30–33. Most evidence for common chronic effects of drugs of abuse in the NAc is biochemical. One of the most dramatic examples is induction

of the transcription factor ∆FosB, a Fos family protein, which accumulates in the NAc after chronic exposure to all drugs of abuse, including cocaine, amphetamine, opiates, alcohol, nicotine, cannabinoids and phencyclidine34,35. It is also induced in this same region by chronic consumption of natural rewards, such as high levels of wheel running and sucrose drinking. In contrast to ∆FosB, cFos and other Fos family members are induced in NAc after acute exposure to drugs or natural rewards, whereas ∆FosB accumulates in this region uniquely after chronic exposure, when induction of the other family members shows desensitization. ∆FosB accumulates during chronic exposure owing to its unique stability at the protein level. There is now considerable evidence that ∆FosB accumulation within NAc neurons contributes to a state of sensitization34,35. Overexpression of ∆FosB in NAc increases behavioral responses to cocaine and opiates as well as to sucrose and wheel-running, including increased incentive drive for these rewards. Conversely, blockade of ∆FosB function in the NAc by overexpression of a dominant negative antagonist causes the opposite effects. Our hypothesis is that induction of ∆FosB mediates many shared aspects of drug and natural addictions by regulating a set of common target genes34–36. Activation of CREB is another common adaptation in the NAc, although it is not as universal as induction of ∆FosB. Repeated exposure to cocaine, amphetamine or opiates induces CREB activity in this region37–39, whereas alcohol and nicotine reduce CREB phosphorylation (an indirect marker of CREB activity) in the NAc (although they induce CREB activity elsewhere)40–42. Despite this opposite regulation of CREB in the NAc by different drugs, activation of CREB seems to produce similar behavioral effects: in numerous experimental systems, increased CREB activity in the NAc decreases behavioral responses to cocaine, opiates and alcohol, whereas decreased CREB activity increases such responses28,36,39,41,43,44. CREB is also induced in the NAc by natural rewards (such as sucrose) and similarly reduces an animal’s sensitivity to sucrose’s rewarding effects39. The molecular mechanisms governing drug regulation of CREB in NAc remain

Figure 2 Highly simplified scheme of some Glutamate common, chronic actions of drugs of abuse on Control inputs the VTA-NAc. The top panel (Control) shows a VTA neuron innervating an NAc neuron and glutamatergic inputs to the VTA and NAc neurons, under normal conditions. After chronic drug administration (lower panel), several NAc adaptations occur. In VTA, drug exposure induces TH and increases AMPA glutamatergic VTA responses (Glut) via regulation of glutamate receptors. There is also evidence that VTA dopamine neurons decrease in size. Induction of CREB activity and alterations in neurotrophic factor (NTF) signaling may partly mediate these Basal glut effects. In NAc, all drugs of abuse induce the Stimulated glut transcription factor ∆FosB, which may then AGS3 mediate some of the shared aspects of addiction Glia Addicted by regulation of numerous target genes. Glut xc– Several, but not all, drugs of abuse induce Glut CREB activity in this region, which may be ∆FosB mediated by upregulation of the cAMP pathway. Basal DA cAMP-CREB TH Several additional changes have been found for Stimulated DA ∆ NTF stimulant exposure; it is not yet known whether CREB they generalize to other drugs. Stimulants decrease AMPA glutamatergic responses in NAc neurons, possibly by regulation of glutamate receptors or postsynaptic density proteins (such as PSD95 and Homer-1). These changes in postsynaptic glutamate responses are associated with complex changes in glutamatergic innervation of the NAc, effects mediated in part by upregulation of AGS3 in cortical neurons and downregulation of the cystine-glutamate transporter (system xc–) in glia. Stimulants and nicotine also induce dendritic outgrowth of NAc neurons, although opiates produce the opposite action. The net effect of these adaptations on glutamatergic transmission remains uncertain.


Ann Thomson


PERSPECTIVE

1447


PERSPECTIVE unknown but could involve upregulation of the cAMP pathway observed in this region after repeated exposure to any of several drugs of abuse, including cocaine, amphetamine, opiates and alcohol2,45,46. Consistent with this view is that in most behavioral assays, cAMP pathway activation also causes reduced behavioral responses to drugs of abuse, whereas inhibition of the pathway facilitates drug responses. One target gene in the NAc through which CREB may exert these effects on drug and natural rewards is the opioid peptide dynorphin, which decreases dopaminergic tone and opposes reward via activation of κ opioid receptors on VTA dopamine neurons and their nerve terminals44,46. A major gap in our knowledge is whether chronic exposure to drugs or natural rewards also elicits common changes in the electrophysiological properties of NAc neurons or of their synaptic inputs. This type of information is available for stimulants like cocaine and amphetamine but has not been examined sufficiently for the other drugs or natural rewards. Stimulants cause a long-term depression (LTD)-like state in NAc neurons and also reduce postsynaptic responses to glutamate21. This could occur in part via ∆FosB-mediated induction of GluR2 (which reduces AMPA receptor conductance and Ca2+ permeability) in the NAc34,35. Changes in postsynaptic responses could also be mediated by altered AMPA receptor trafficking or by adaptations in the neurons’ postsynaptic densities (PSDs), including reduced levels of PSD95 and Homer or increased levels of F-actin18,47. These findings, along with the evidence of abnormal glutamatergic innervation of the NAc from frontal cortical regions (discussed in the next section) would suggest a profound dysfunction in cortical control over the NAc18, which could in turn relate to the impulsive and compulsive features of stimulant addiction. A major need for future investigations is to determine whether similar changes occur with other drugs of abuse or natural rewards. Fewer studies have searched for common drug-induced molecular changes in other brain areas (outside the VTA and NAc) associated with addiction. This is unfortunate, as these other regions are also key to the addiction process. For example, one can presume that the hyperactivity of central CRF pathways upon precipitation of drug withdrawal17 reflects underlying molecular adaptations in amygdala neurons. An interesting candidate is CREB, because it is induced in amygdala by stimulants, opiates and alcohol, and the CRF gene is regulated by CREB37,38,41. As another example, one can presume that the hypofrontality demonstrated in rodents and humans after chronic exposure to several drugs of abuse is mediated via common molecular and cellular adaptations in these cortical regions (Fig. 2). To date, such adaptations have been characterized only for stimulants and suggest an interesting pathophysiology13,18. According to this scheme, which requires further analysis, dopamine has a profound effect on activity of prefrontal cortical neurons, with D1 and D2 dopamine receptors exerting opposite effects: D2 receptor activation tends to promote the neurons’ responsiveness to diverse environmental stimuli, whereas D1 receptor activation favors activation by the strongest stimuli only. Chronic stimulant administration seems to cause a shift toward the D1 state in part by induction of AGS3 (activator of G protein signaling-3), which is a negative regulator of Gi and hence of D2 signaling. This could explain the functional observations cited above that baseline activity of these neurons and their responses to natural rewards is dampened, whereas responses to cocaine and cocaine-associated stimuli are enhanced. Induction of AGS3 is accompanied by decreased levels of the cystine-glutamate transporter in glial cells in the NAc; this transporter promotes release of glutamate from prefrontal cortical glutamatergic nerve terminals, perhaps further exaggerating glutamatergic transmission to the NAc when the cell bodies fire in response to cocaine and associated cues13,18. Clearly, these changes are complex and interact with postsynaptic adaptations in glutamate receptor function in NAc neurons in ways that remain incompletely understood. Nevertheless, these impor-

1448

tant findings now highlight the need to characterize the effects of other drugs of abuse, as well as natural rewards, on these same endpoints. We know that several drugs of abuse, after chronic administration, converge to reduce neurogenesis—birth of new neurons—in the dentate gyrus of adult hippocampus, an effect reported to date for cocaine, opiates, alcohol, nicotine and cannibinoids48. The function of adult hippocampal neurogenesis is a subject of great debate: the new nerve cells may be critical for the formation of new memories, although this remains unproven. In any event, the finding that reduced neurogenesis is a common consequence of chronic drug administration raises interesting questions. For example, might this effect contribute to common abnormalities in memory or other cognitive functions seen in many addicts? Drug-specific actions in brain reward circuits The argument that acute and chronic drug effects are mediated by common mechanisms is inconsistent with the knowledge that drugs of abuse can readily be distinguished from one another. This is probably due in large part to the very different effects the drugs elicit outside the brain’s reward circuitry. For example, opiates are analgesics and sedatives owing to their actions on opioid receptors, which are located in brainstem and spinal cord, whereas cocaine activates the cardiovascular system because of its actions on monoamine transporters, which are located in heart and vascular tissue. Nevertheless, there are likely to be very important differences in the effects of each drug of abuse on reward circuitry as well, and such differences may help explain why most addicts have their preferred drug of abuse. The most striking example of different chronic cellular actions of drugs of abuse comes from morphological changes observed in NAc and prefrontal cortical pyramidal neurons. Chronic exposure to cocaine, amphetamine or nicotine causes long-lasting increases in dendritic arborization and in the density of dendritic spines of these neurons49. These morphological changes, which may be partly mediated via ∆FosB and some of its target genes35, could contribute to the state of sensitization seen after stimulant exposure, although this has not yet been demonstrated with certainty. In contrast, chronic opiate administration causes opposite changes in the dendritic arbor of NAc and prefrontal cortical neurons49. This opposite action of opiates is surprising, given that chronic opiate administration, like chronic stimulant administration, causes very similar sensitization to the behavioral effects of the drugs, and even cross-sensitization, as mentioned earlier. More work is needed to better understand this paradox. One possibility is that the different morphological effects may still lead to a common behavioral endpoint owing to the more specific changes in synaptic transmission associated with the altered morphology. Hence, much more detailed examination is needed of pre- and postsynaptic elements as a function of chronic drug administration. There are also large numbers of molecular adaptations reported for one drug of abuse that are not seen with the others2,13,46. One general principle is that such drug-specific adaptations may increase in likelihood with increasing proximity of a protein to the immediate drug target, whereas more common adaptations may be expected more distally. For example, chronic administration of cocaine, opiates, alcohol or nicotine is specifically associated with changes in dopamine transporters, opioid receptors, GABAA receptors and nicotinic cholinergic receptors, respectively, in numerous brain regions4,6,7,9,46, even though all of the drugs alter dopaminergic transmission3 and induce ∆FosB34,35 in the NAc. Implications for treatment of drug and natural addictions Because common mechanisms seem to contribute to at least some aspects of all drug addictions, and possibly to natural addictions as well, it might be possible to develop treatments that would be effective for a wide range of addictive disorders. Drugs that target the brain’s dopamine, glutamate,



PERSPECTIVE CRF, opioid or cannabinoid systems might well be expected to exert common palliative effects in individuals addicted to a wide range of drugs or natural rewards. Studies are underway in each of these areas. In the more distant future, it might be possible to exploit our increasing knowledge of common molecular adaptations to drugs and natural rewards, for example, in the cAMP-CREB pathway, ∆FosB, or any of their numerous target genes, or in the complex molecular constituents of the glutamatergic synapse, although this approach remains highly speculative. At the same time, however, our knowledge of shared mechanisms of action of drugs of abuse and natural rewards raises serious red flags about the ultimate safety and effectiveness of such treatments. Is it possible to dampen common mechanisms of impulsive and compulsive consumption of drug or natural rewards without affecting the normal functioning of the individual (that is, healthy responses to natural rewards)? Thus far, all established treatments for addiction are drug-specific and are aimed at the acute protein target of the drugs46. For example, methadone, buprenorphine and naltrexone, which are, respectively, an agonist, partial agonist and antagonist at the µ opioid receptor, are effective for opiate addiction, whereas nicotine patches and chewing gum are effective for nicotine addiction. Naltrexone can also be of some use in the treatment of alcohol and nicotine addictions, but the magnitude of its effect is modest in most individuals. Acamprosate, which may act by reducing NMDA glutamatergic receptor function (although this remains speculative), is reported to be effective for some alcohol addicts50. Thus, no treatment aimed at common drug mechanisms has yet been fully validated across of range of addictions to multiple drugs of abuse and to natural rewards. A high priority for current research should be to focus on bringing some of the most promising common anti-addiction mechanisms into the clinic for broad trials across several addictive disorders. Such trials, while clearly difficult, are critically important to help us understand whether addictions can be treated, at least in part, as a unitary disorder. ACKNOWLEDGMENTS Preparation of this review was supported by the National Institute on Drug Abuse. COMPETING INTERESTS STATEMENT The author declares that he has no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Koob, G.F. & Le Moal, M. Drug addiction, dysregulation of reward, and allostasis. Neuropsychopharmacology 24, 97–129 (2001). 2. Nestler, E.J. Molecular basis of long-term plasticity underlying addiction. Nat. Rev. Neurosci. 2, 119–128 (2001). 3. Di Chiara, G. et al. Dopamine and drug addiction: the nucleus accumbens shell connection. Neuropharmacology 47 (suppl.) 227–241 (2004). 4. Volkow, N.D., Fowler, J.S., Wang, G.J. & Swanson, J.M. Dopamine in drug abuse and addiction: results from imaging studies and treatment implications. Mol. Psychiatry 9, 557–569 (2004). 5. Wise, R.A. Dopamine, learning and motivation. Nat. Rev. Neurosci. 5, 483–494 (2004). 6. Boehm, S.L., II et al. gamma-Aminobutyric acid A receptor subunit mutant mice: new perspectives on alcohol actions. Biochem. Pharmacol. 68, 1581–1602 (2004). 7. Dani, J.A., Ji, D. & Zhou, F.M. Synaptic plasticity and nicotine addiction. Neuron 31, 349–352 (2001). 8. Howlett, A.C. et al. Cannabinoid physiology and pharmacology: 30 years of progress. Neuropharmacology 47 (suppl.) 345–358 (2004). 9. Everitt, B.J. & Wolf, M.E. Psychomotor stimulant addiction: a neural systems perspective. J. Neurosci. 22, 3312–3320 (2002). 10. Robinson, T.E. & Berridge, K.C. Addiction. Annu. Rev. Psychol. 54, 25–53 (2003). 11. Hyman, S.E. & Malenka, R.C. Addiction and the brain: the neurobiology of compulsion and its persistence. Nat. Rev. Neurosci. 2, 695–703 (2001). 12. Everitt, B.J., Cardinal, R.N., Parkinson, J.A. & Robbins, T.W. Appetitive behavior: impact of amygdala-dependent mechanisms of emotional learning. Ann. NY Acad. Sci. 985, 233–250 (2003). 13. Kalivas, P.W. Glutamate systems in cocaine addiction. Curr. Opin. Pharmacol. 4, 23–29 (2004). 14. Kelley, A.E. & Berridge, K.C. The neuroscience of natural rewards: relevance to addictive drugs. J. Neurosci. 22, 3306–3311 (2002). 15. Tobler, P.N., Fiorillo, C.D. & Schultz, W. Adaptive coding of reward value by dopamine neurons. Science 307, 1642–1645 (2005).


16. Avena, N.M. & Hoebel, B.G. A diet promoting sugar dependency causes behavioral crosssensitization to a low dose of amphetamine. Neuroscience 122, 17–20 (2003). 17. Heinrichs, S.C. & Koob, G.F. Corticotropin-releasing factor in brain: a role in activation, arousal and affect regulation. J. Pharmacol. Exp. Ther. 311, 427–440 (2004). 18. Kalivas, P.W., Volkow, N. & Seamans, J. Unmanageable motivation in addiction: a pathology in prefrontal-accumbens glutamate transmission. Neuron 45, 647–650 (2005). 19. Saal, D., Dong, Y., Bonci, A. & Malenka, R.C. Drugs of abuse and stress trigger a common synaptic adaptation in dopamine neurons. Neuron 37, 577–582 (2003). 20. Borgland, S.L., Malenka, R.C. & Bonci, A. Acute and chronic cocaine-induced potentiation of synaptic strength in the ventral tegmental area: electrophysiological and behavioral correlates in individual rats. J. Neurosci. 24, 7482–7490 (2004). 21. Thomas, M.J. & Malenka, R.C. Synaptic plasticity in the mesolimbic dopamine system. Phil. Trans. R. Soc. Lond. B Biol. Sci. 358, 815–819 (2003). 22. Kauer, J.A. Learning mechanisms in addiction: synaptic plasticity in the ventral tegmental area as a result of exposure to drugs of abuse. Annu. Rev. Physiol. 66, 447–475 (2004). 23. Carlezon, W.A., Jr. & Nestler, E.J. Elevated levels of GluR1 in the midbrain: a trigger for sensitization to drugs of abuse? Trends Neurosci. 25, 610–615 (2002). 24. Bonci, A. & Williams, J.T. Increased probability of GABA release during withdrawal from morphine. J. Neurosci. 17, 796–803 (1997). 25. Nestler, E.J. Molecular mechanisms of drug addiction. J. Neurosci. 12, 2439–2450 (1992). 26. Lu, L., Grimm, J.W., Shaham, Y. & Hope, B.T. Molecular neuroadaptations in the accumbens and ventral tegmental area during the first 90 days of forced abstinence from cocaine self-administration in rats. J. Neurochem. 85, 1604–1613 (2003). 27. Olson, V.G. et al. Regulation of drug reward by CREB: Evidence for two functionally distinct subregions of the ventral tegmental area. J. Neurosci. 25, 5553–5562 (2005). 28. Walters, C.L., Godfrey, M., Li, X. & Blendy, J.A. Alterations in morphine-induced reward, locomotor activity, and thermoregulation in CREB-deficient mice. Brain Res. 1032, 193– 199 (2005). 29. Walters, C.L., Cleck, J.N., Kuo, Y.C. & Blendy, J.A. mu-Opioid receptor and CREB activation are required for nicotine reward. Neuron 46, 933–943 (2005). 30. Bolanos, C.A. & Nestler, E.J. Neurotrophic mechanisms in drug addiction. Neuromol. Med. 5, 69–83 (2004). 31. Pierce, R.C. & Bari, A.A. The role of neurotrophic factors in psychostimulant-induced behavioral and neuronal plasticity. Rev. Neurosci. 12, 95–110 (2001). 32. Lu, L. et al. A single infusion of brain-derived neurotrophic factor into the ventral tegmental area induces long-lasting potentiation of cocaine seeking after withdrawal. J. Neurosci. 24, 1604–1611 (2004). 33. Hall, F.S., Drgonova, J., Goeb, M. & Uhl, G.R. Reduced behavioral effects of cocaine in heterozygous brain-derived neurotrophic factor (BDNF) knockout mice. Neuropsychopharmacology 28, 1485–1490 (2003). 34. Nestler, E.J., Barrot, M. & Self, D.W. ∆FosB: A molecular switch for addiction. Proc. Natl. Acad. Sci. USA 98, 11042–11046 (2001). 35. McClung, C.A. et al. ∆FosB: A molecular switch for long-term adaptation. Mol. Brain Res. 132,146–154 (2004). 36. McClung, C.A. & Nestler, E.J. Regulation of gene expression and cocaine reward by CREB and ∆FosB. Nat. Neurosci. 6, 1208–1215 (2003). 37. Shaw-Lutchman, T.Z. et al. Regional and cellular mapping of CRE-mediated transcription during naltrexone-precipitated morphine withdrawal. J. Neurosci. 22, 3663–3672 (2002). 38. Shaw-Lutchman, T.Z., Impey, S., Storm, D. & Nestler, E.J. Regulation of CRE-mediated transcription in mouse brain by amphetamine. Synapse 48, 10–17 (2003). 39. Barrot, M. et al. CREB activity in the nucleus accumbens shell controls gating of behavioral responses to emotional stimuli. Proc. Natl. Acad. Sci. USA 99, 11435–11440 (2002). 40. Brunzell, D.H., Russell, D.S. & Picciotto, M.R. In vivo nicotine treatment regulates mesocorticolimbic CREB and ERK signaling in C57Bl/6J mice. J. Neurochem. 84, 1431–1441 (2003). 41. Pandey, S.C., Roy, A., Zhang, H. & Xu, T. Partial deletion of the cAMP response elementbinding protein gene promotes alcohol-drinking behaviors. J. Neurosci. 24, 5022–5030 (2004). 42. Constantinescu, A., Wu, M., Asher, O. & Diamond, I. CAMP-dependent protein kinase type I regulates ethanol-induced cAMP response element-mediated gene expression via activation of CREB-binding protein and inhibition of MAPK. J. Biol. Chem. 279, 43321–43329 (2004). 43. Walters, C.L. & Blendy, J.A. Different requirements for cAMP response element binding protein in positive and negative reinforcing properties of drugs of abuse. J. Neurosci. 21, 9438–9444 (2001). 44. Carlezon, W.A., Jr., Duman, R.S. & Nestler, E.J. The many faces of CREB. Trends Neurosci. 28, 436–445 (2005). 45. Self, D.W. et al. Involvement of cAMP-dependent protein kinase in the nucleus accumbens in cocaine self-administration and relapse of cocaine-seeking behavior. J. Neurosci. 18, 1848–1859 (1998). 46. Kreek, M.J. Drug addictions. Molecular and cellular endpoints. Ann. NY Acad. Sci. 937, 27–49 (2001). 47. Yao, W.D. et al. Identification of PSD-95 as a regulator of dopamine-mediated synaptic and behavioral plasticity. Neuron 41, 625–638 (2004). 48. Eisch, A.J. Adult neurogenesis: implications for psychiatry. Prog. Brain Res. 138, 315–342 (2002). 49. Robinson, T.E. & Kolb, B. Structural plasticity associated with exposure to drugs of abuse. Neuropharmacology 47 (suppl.) 33–46 (2004). 50. Littleton, J. & Zieglgansberger, W. Pharmacological mechanisms of naltrexone and acamprosate in the prevention of relapse in alcohol dependence. Am. J. Addict. 12 (suppl.) S3–S11 (2003).

1449


PERSPECTIVE


Genetic influences on impulsivity, risk taking, stress responsivity and vulnerability to drug abuse and addiction Mary Jeanne Kreek, David A Nielsen, Eduardo R Butelman & K Steven LaForge

Genetic variation may partially underlie complex personality and physiological traits—such as impulsivity, risk taking and stress responsivity—as well as a substantial proportion of vulnerability to addictive diseases. Furthermore, personality and physiological traits themselves may differentially affect the various stages of addiction, defined chronologically as initiation of drug use, regular drug use, addiction/dependence and potentially relapse. Here we focus on recent approaches to the study of genetic variation in these personality and physiological traits, and their influence on and interaction with addictive diseases.

Published online 26 October 2005; doi:10.1038/nn1583

contributions and genetic variants have been identified and verified by multiple studies. However, the identified variants, in their entirety, comprise only a small proportion of the estimated genetic contribution. Studying the genetics of complex psychiatric or behavioral disorders such as addiction poses additional challenges. These include precise phenotypic characterization of individuals and the characterization of ethnic/cultural backgrounds (as different backgrounds yield differences in allelic frequencies). These challenges also must be faced in the study of other complex genetic disorders. Despite the complexity of the problem, the costs to society of drug and alcohol addiction are too enormous to ignore. Addiction has some of the highest overall medical health costs of any medical disorder, once comorbid disorders such as HIV/AIDS, hepatitis C and lung cancer are factored in. Loss of productivity, interdiction and the criminal justice system incur additional economic costs. It is therefore imperative that all components contributing to addiction be studied, including genetics, with the goal of improving primary prevention, early intervention and chronic treatment. Family and twin epidemiological studies show that genes contribute to the vulnerability to addictive disease, with estimates of heritability of 30–60%. Addiction heritability was first demonstrated with alcoholism, which is influenced by distinct genetic factors such as the aldehyde dehydrogenase 2 genotype. Predisposition to addiction may be due both to genetic variants that are common to all addictions and to those specific to a particular addiction. For example, a genetic variance shared by multiple classes of drugs of abuse is demonstrated in twin studies4,5. However, some genetic variance is specific to drug class, as is particularly well documented for opiate addiction4. Moreover, there are different influences of environment versus genetic factors on the transitions from initiation of drug use, to regular drug use, to drug addiction/dependence and then potentially to relapse6 (Fig. 1). The genetics of addiction encompasses heritable factors that influence the different stages in the trajectory of initiation and progression to drug addiction, including severity of dependence or withdrawal and risk of relapse. Variation in personality dimensions, such as impulsivity, risk taking and novelty seeking, may contribute to the initiation of drug use as well as the transitions from initial use to regular use to addiction (Fig. 1). Each of these personality dimensions may have, in part, its own genetic basis. A number of inventories have been developed for the description and classification of personality dimensions and to tease out the influence of genetics on personality. Four instruments often used in genetics

1450


Vulnerability to develop a drug addiction is influenced by a combination of genetic and environmental factors. Both factors couple with direct drug-induced effects to influence the progression from intermittent to regular drug use, the transition from abuse to addiction, and the propensity for repeated relapse after achievement of a drug-free state1,2. Chronic exposure to drugs of abuse causes persistent changes in the brain, including changes in expression of genes or their protein products, in protein-protein interactions, in neural networks, and in neurogenesis and synaptogenesis, all of which ultimately affect behavior. In rodents, there are inbred strains and selectively bred lines that readily self-administer drugs of abuse (implying genetic vulnerability) as well as strains that do not readily self-administer drugs (implying genetic resistance). Different strains show differences in the cellular and molecular response to drugs3. Genetic factors may also be involved in direct drug-induced effects, including alteration of pharmacodynamics (a drug’s effects at a receptor, including the physiological consequences of receptor activity) or pharmacokinetics (a drug’s absorption, distribution, metabolism and excretion) of a drug of abuse or of a treatment agent. Many medical disorders have some genetic component, but most, including cancer, obesity and heart disease, involve complex genetic contributions based on multiple variants of multiple genes and different combinations of these variants in different people. For some of the most studied diseases, such as certain cancers, the specific genetic

Mary Jeanne Kreek, David A. Nielsen, Eduardo R. Butelman & K. Steven LaForge are in the Laboratory of the Biology of Addictive Diseases, The Rockefeller University, New York, New York, USA. e-mail: [email protected]

PERSPECTIVE

Impulsivity* (genetics?) Risk taking* (genetics?) Comorbidity (genetics)

Initiation of drug use

Intermittent to regular use

Addiction and relapse

Environmental factors (~ 100% of cases) Genetic factors for addiction (30-60% of cases) Drug induced effects (with some genetic factors) (~ 100% of cases)

research to quantify personality dimensions are the Tridimensional Personality Questionnaire (TPQ) or the more complete version, the Temperament and Character Inventory (TCI; which measures novelty seeking, harm avoidance, reward dependence and persistence), the NEO Personality Inventory-Revised (NEO-PI-R; which measures neuroticism, extroversion, openness, agreeableness and conscientiousness) and the Barratt Impulsiveness Scale. Some of these questionnaires and variables are based on the concept of factor analysis, in which a large number of individual questions contribute to a smaller number of underlying traits. Of the traits measured by these tests, some addiction research has focused on impulsivity, with or without aggression or suicidality, and risk taking, which is often associated with novelty seeking. In addition, addictions can be defined with scales such as the KMSK, which measure duration and magnitude of drug use. The TPQ, TCI, NEO-PI-R provide a broad and more time-intensive characterization of personality traits. By contrast, the Barratt and KMSK scales provide a relatively rapid evaluation of a particular phenotype (impulsiveness and degree of exposure to a drug of abuse, respectively). Use of common questionnaires does facilitate the comparative interpretation of different studies. However, optimization of more focused instruments may also be advantageous for the study of more refined phenotypes relevant to a particular clinical situation. Identifying genetic factors in personality traits and addiction Until recently, family-based linkage studies have been most widely used. Linkage studies investigate the transmission of genetic markers on specific genomic regions of interest and phenotypes in pedigrees consisting of, preferably, two or more generations, including studies of affected sibling pairs (more powerful when both siblings and parents are included). The alternative is association studies, which ask whether a particular gene allele is more prevalent in patients, compared with control subjects, than would be expected by chance. This is increasingly the experimental approach of choice for identifying genes responsible for complex traits. Association studies are able to detect linked variants involved in a disease (i) if they are within 40,000–80,000 nucleotides of the genotyped variant, (ii) if linkage disequilibrium (which occurs when there is a non-random distribution of allele combinations; for example, in a haplotype) is relatively high, and (iii) if the effect sizes are moderate to high. This is a much smaller distance than is possible with family-based linkage studies. Although family-based studies may be feasible for personality traits such as impulsivity, risk taking and stress responsivity, family


Ann Thomson


Stress responsivity atypical (genetics)

Figure 1 Diverse contribution of genetic influences to initial drug use, abuse and addiction. We suggest that impulsivity and risk taking contribute most to the initiation of drug use and the progression to regular drug use. We expect that these personality factors contribute less to addiction and relapse after substantial changes to the brain, effected by chronic exposure to the drug of abuse. These two personality factors, comorbidity and stress responsivity (top) and the three domains (bottom) interact to influence the progression to addiction, as depicted. *Lifelong or identified in early childhood. ↓↓↓, greatest relative influence; ↓↓, medium relative influence; ↓, small relative influence; ↓, minor relative influence; 0, no influence. These ratings reflect our estimates and opinion based on current information.

studies in illicit drug addiction are difficult to conduct because of the enormous stigma of addiction, the disruption of involved families and the difficulty in ascertaining family members. However, outstanding family studies have been done in the field of alcoholism, notably the Collaborative Study on the Genetics of Alcoholism sponsored by United States National Institute on Alcoholism and Alcohol Abuse7. Strong evidence has been provided by these studies for the involvement of several genes, including the GABA receptor subunit A2 (GABRA2) and muscarinic acetylcholine receptor M2 (CHRM2), in alcohol dependence8,9. One general approach for identifying specific genes involved in a disease is hypothesis-oriented selection. It is frequently useful to investigate specific genes involved in diseases based on a prior understanding of the diseases and/or addiction and based on specific hypotheses about these factors1,2. In studying drug addiction, one can initially consider genes governing direct and downstream molecular events altered by chronic exposure to a drug of abuse. For example, cocaine produces a surge in extracellular dopamine by blocking the action of the dopamine transporter. Cocaine also increases gene expression and promotes release of the κ opioid ligand dynorphin in the striatum. Variants of the preprodynorphin gene (PDYN) have been associated with vulnerability to develop cocaine addiction2. Another strategy is to use positional approaches—conducting genome-wide scans to identify chromosomal positions that may be associated with a specific disorder or addiction. Further fine mapping in identified chromosomal regions is then required. Until recently, microsatellite marker panels were used to scan the whole genome. However, over the last few years, various approaches using single nucleotide polymorphism (SNP) arrays or other panels of single SNPs have allowed the identification of more defined regions for fine mapping in a far simpler manner, a more refined approach than microsatellite marker panels. As SNP panels become more inclusive of the common variants in the human genome, it should be possible to examine the variants associated with a phenotype more quickly. Variants in the coding region of genes may change the protein product, as in the A118G variant of the µ opioid receptor gene (OPRM1). Other variants may alter the amount of gene expression (for example, prodynorphin promoter region variants), and yet other variants may alter the rate of mRNA degradation (for example, a dopamine receptor D2 variant, DRD2), all of which can contribute to functionality2. Such variants can affect both normal physiology and specific aspects of addiction pathophysiology.

1451

PERSPECTIVE


Table 1 Genes having one or more variants that have been reported to be associated with one or more addictions Gene

Protein

System

Chromosomal locationa

I

R

E

S

A

Drug

Status

OPRM1

µ opioid receptor

Opioid

6q24-q25

–

–

–

+

+

H/O; Alc

D/Ab

OPRK1

κ opioid receptor

Opioid

8q11.2

–

–

–

–

+

H/O

D/A

PDYN

Preprodynorphin

Opioid

20pter-p12.2

–

–

–

–

+

C/S

D/A

TH

Tyrosine hydroxylase

Dopaminergic

11p15.5

–

–

+

–

+

Alc

D/A

DRD2

Dopamine receptor D2

Dopaminergic

11q23

–

–

–

–

+

Alc

D/Ab

DRD3


Dopaminergic

3q13.3

–

+

–

–

+

Alc; C/S

D/Ab

DRD4


Dopaminergic

11p15.5

+

+

–

–

+

H/O; C/S; Alc

D/Ab

DBH

Dopamine β-hydroxylase

Dopaminergic

9q34

–

–

–

–

+

C/S

D/A

DAT (SLC6A3)

Dopamine transporter

Dopaminergic

5p15.3

+

–

–

–

+

Alc

D/Ab

TPH1

Tryptophan hydroxylase 1

Serotonergic

11p15.3-p14

+

–

–

–

+

Alc

D/Ab

TPH2

Tryptophan hydroxylase 2

Serotonergic

12q21.1

–

–

–

–

+

H/O; Alc

CSA; D/Ab

HTR1B

Serotonin receptor 1B

Serotonergic

6q13

–

–

–

–

+

Alc; H/O

D/Ab

HTR2A

Serotonin receptor 2A

Serotonergic

13q14-q21

–

–

–

–

+

Alc

CSA; D/Ab

SERT (SLC6A4)

Serotonin transporter

Serotonergic

17q11.1-q12

+

–

+

–

+

H/O; Alc

D/Ab

MAOA

Monoamine oxidase A

Catecholaminergic, serotonergic

Xp11.23

+

–

+

–

+

Alc

D/A

COMT

Catechol-O-methyl transferase

Catecholaminergic

22q11.2

+

–

–

+

+

Alc; H/O

D/Ab

GABRA1

GABA receptor subunit α-1

GABAergic

5q34-q35

+

–

–

–

+

Alc

D/Ab

GABRA6

GABA receptor subunit α-6

GABAergic

5q31.1-q35

+

–

–

–

+

Alc

D/A

GABRB1

GABA receptor subunit β-1

GABAergic

4p13-p12

+

–

–

–

+

Alc

D/A

CHRM2

Muscarinic acetylcholine receptor M2

Cholinergic

7q35-q36

–

–

–

–

+

Alc

D/Ab

CNR1

Cannabinoid receptor 1

Cannabinoid

6q14-q15

–

–

–

–

+

Alc; C/S

CSA; D/Ab

FAAH

Fatty acid amide hydrolase

Cannabinoid

1p35-34

–

–

–

–

+

Alc

CSA

NPY

Neuropeptide Y

Neuromodulatory

7p15.1

–

–

–

–

+

Alc

CSA; D/Ab

ADH1B

Alcohol dehydrogenase 1B

Ethanol metabolism

4q22

–

–

–

–

+

Alc

D/Ab

ADH1C

Alcohol dehydrogenase 1C

Ethanol metabolism

4q22

–

–

–

–

+

Alc

D/Ab

ALDH2

Aldehyde dehydrogenase 2

Ethanol metabolism

12q24.2

–

–

–

–

+

Alc

D/Ab

CYP2D6

Cytochrome CYP450

Drug metabolism

22q13.1

–

–

–

–

+

H/O

D/A

ANKK1

Ankyrin repeat and kinase domain–containing 1

Signal transduction (predicted)

11q23.2

–

+

–

–

+

Alc

D/Ab

I: impulsivity, R: risk taking, E: environment, S: stress responsivity, A: addiction. H/O: heroin or opiate, Alc: alcohol, C/S: cocaine or stimulants, CSA: continued substance abuse, D/A: dependence/addiction. aGene

map locus: Online Mendelian Inheritance in Man, Johns Hopkins University, Baltimore (http://www.ncbi.nlm.nih.gov/omim) as of July 2005. with drug addiction in two or more studies

bAssociation

Ultimately, however, rigorous phenotypic assessment is essential for all studies of addiction genetics because poor or inadequate phenotypic assessments lead to incorrect results. Such assessment entails the use of a diverse battery of instruments to evaluate personality traits, comorbid disorders, detailed histories of initiation of drug use, and progression to addiction. Precise phenotyping takes time, and requires highly trained personnel. Moreover, because of the time and expense, it can lead to a decrease in numbers of subjects studied. There is therefore an inherent trade-off that most geneticists have to make—whether to study large numbers of subjects, which would in turn give the study and the statistics greater validity, or to do very careful phenotyping of the subjects, without which one runs the risks of generating more false positives or negatives. Population genetics can also be influenced by additional factors—for instance, there are significant ethnic/cultural differences in allelic frequencies of variants of many specific genes. These must be controlled for or analyzed using a variety of newly developing techniques involving primarily combinations of SNPs or other variants. There are innumerable further challenges to molecular genetics studies of any complex disorder, including other diseases that are present at the same time, the specificity in diagnosis of subjects, vigilance for error of any type in

the molecular work, and rigorous state-of-the-art statistical genetics analyses10. Statistical genetics methods involve techniques that are evolving, such as methods for statistically determining inferred haplotypes . Here we have included only a selection of studies that we consider of potential importance, primarily from established research teams using acceptable or optimal study designs, phenotypic assessments, molecular techniques and statistical genetics analyses. This perspective is not intended to be a comprehensive review. Moreover, we must emphasize that evidence of enhanced genetic vulnerability to addiction does not imply that addiction will occur. Many factors, such as environmental influences or availability of drugs, strongly influence the development of drug abuse or addiction. Conversely, a ‘genetically resistant’ individual (or strain of rat) may self-administer a drug of abuse under specific environmental conditions (Fig. 1)3.

1452


Personality traits and addiction: impulsivity Impulsivity is a personality trait characterized by behavioral disinhibition, defined as acting suddenly in an unplanned manner to satisfy a desire: for example, acting on the spur of the moment, not thinking through the potential impact before carrying out actions, or making


PERSPECTIVE statements without thinking in advance about what is to be said. Acts of impulsivity may include aggression, violence and suicide. However, impulsivity, as a trait, occurs on a continuum; thus, impulsivity per se is not an indicator of pathology. Early work implicated low serotonin levels and its metabolites in various forms of impulsivity. Low levels of cerebrospinal fluid 5-hydroxylindolacetic acid, a major metabolite of serotonin and an indicator of serotonin metabolism, is related to impulsivity, aggression and depression, as well as to early-onset alcoholism (reviewed in refs. 11,12). In addition, prolactin release after fenfluramine challenge, a biomarker of serotonin metabolism, demonstrates a relationship between low serotonin metabolism and impulsive behavior13, which is also associated with an increased risk for impulsive personality traits in first-degree relatives13. Serotonergic neurons project from the raphe throughout the brain to diverse regions, including the hippocampus, frontal cortex and amygdala. Loss of impulse control may be due to impaired inhibitory control resulting from drug-induced changes in the frontal cortex. Experimentation with addictive drugs and the onset of drug abuse generally occur in adolescence, with the rare exception of some reported cases of alcoholism and prescription opiate addiction, which may occur later, even in the elderly. Neurodevelopmental processes and reproductive hormone changes during adolescence and early adulthood may modulate impulse control, which may contribute to vulnerability to experimentation with drugs of abuse, with possible progression to addiction. Behaviors characterized by a deficit in impulse control have been studied for association and linkage with candidate genes (Table 1) in the serotonergic system (for example, tryptophan hydroxylase 1 and 2 [TPH1 and TPH2] and serotonin transporter [SERT]), the dopaminergic system (tyrosine hydroxylase [TH], dopamine receptors, and dopamine transporter [DAT]), the monoamine metabolism pathway (monoamine oxidase A [MAOA] and catechol-O-methyltransferase [COMT]), and the noradrenergic system (dopamine β-hydroxylase [DBH]), inhibitory system, GABAergic and nitric oxide systems, as well as other genes)11,14,15. Each of these genes is reportedly associated with alcoholism or some other addiction (Table 1). In addition, the neurotransmitter systems coded by these genes are interactively involved in the acute and chronic effects of most drugs of abuse and, thus, underlie addiction as well as the initiation of drug use. The earliest candidate gene studies on impulsivity were conducted on TPH1, which codes for the rate-limiting enzyme in the production of serotonin. In impulsive violent offenders, a TPH1 gene variant was associated with reduced CSF 5-HIAA and suicidal behavior12,14. TPH1 variants are also associated with impulsivity, aggression and various forms of suicidality. Other genes such as SERT, DRD3, MAOA, 5-HT2A, and dopamine receptors D3 and D4 (DRD3 and DRD4) are related to impulsivity (Table 1)11,14. Although a substantial body of knowledge is accumulating from these genetic studies, the results gleaned from these investigations may not extrapolate to all people. These studies vary in their assessment instruments used, the ethnic/cultural populations studied and the statistical methods applied. Hence, the results of apparently similar studies cannot be directly compared and, for this reason, meta-analyses of these studies may be fraught with pitfalls. Thus, several major candidate genes with variants associated with impulsivity have been reported. Most of these candidate genes code for proteins that control major neurotransmitter systems, for which a wealth of basic data is available. Furthermore, impulsivity itself is associated with specific addictive diseases. Future studies on the role of impulsivity, and its genetic variants, at specific stages of addiction could therefore shed light on neurobiological mechanisms underlying clinically defined stages in the process of addiction, relapse and recovery.


Personality traits and addiction: risk taking Risk taking is characterized by behaviors performed under uncertainty, with or without inherent negative consequences, with or without any possible or probable harm to oneself or others, and without robust contingency planning. Risk taking may be operationally measured in tasks that require an evaluation of relative risk versus reward (for instance, in the choice of career opportunities or selection of automobiles). Pathological gamblers and addicted patients may exhibit signs of risk taking, which can be assessed by specific clinical questionnaires such as the South Oaks Gambling Screen. Novelty seeking, often defined as one aspect of risk taking, with potentially high reactivity to novel stimuli, can alternatively be considered a personality trait, detected in certain psychometric instruments (such as the Temperament and Character Inventory). It is part of a constellation of traits observed in individuals with a propensity to experiment with novel stimuli, including those produced by drugs of abuse. Novelty seeking may be correlated with progression from abuse to addiction for several drugs. DRD4 receptors are members of the ‘D2like’ family of Gi-coupled dopamine receptors. Some studies report an association between novelty seeking and DRD4 receptor variants, for example, between high Tridimensional Personality Questionnaire novelty-seeking scores and a particular allelic variant16,17. In human brain tissue, DRD4 binding is found in brain regions that include the prefrontal and entorhinal cortex, hippocampus, dorsomedial thalamus, lateral septal nucleus and hypothalamus18. Notably, no apparent DRD4 binding is detected in the nucleus accumbens, caudate or putamen, which are major sites of D2 receptor binding and mediate the direct psychostimulant and reinforcing effects of drugs of abuse. In contrast, the DRD4 receptor distribution pattern suggests roles in attentional, motivational, emotional and mnemonic processing, on the basis of some major functions thought to be mediated by these brain areas. Notably, the prefrontal cortex is a site of cognitive and executive functions and decision making. Although several studies have identified associations of different DRD4 polymorphisms with novelty seeking, these findings have not been consistently replicated16,17. These heterogeneous findings may result from differences in age of subjects, phenotyping instruments used and ethnic composition of patient populations, among other issues, in different studies16. Other molecular targets involved in monoaminergic function have been related to novelty seeking and drug abuse (Table 1). The DRD2 Taq1A polymorphism is widely studied and reported in the scientific literature and popular press for its association with alcoholism and various psychiatric disorders. However, this association has yet to be solidly documented, with conflicting meta-analyses from different groups19,20. Interestingly, the Taq1A variant is located approximately 10,000 nucleotides downstream (3′) from the DRD2 gene and has recently been reported to reside in the neighboring ANKK1 gene, which codes for a serine/threonine kinase21. Hence, results reported for the Taq1A variant may be potentially ascribed to the action of the ANKK1 gene product. Comorbid disorders For many addicts, substance abuse does not occur as an isolated disorder. Four psychiatric conditions (depression, anxiety, antisocial personality disorder and attention deficit/hyperactivity disorder) are commonly present in and probably are involved in psychopathology or physiology of addiction to opiates and alcohol22. The most common comorbid conditions are depression and anxiety, with unipolar depression being the most common. In epidemiological studies, 20% to over 50% of people with alcoholism, cocaine and other stimulant addiction, or opiate addiction have depressive and/or anxiety disor-

1453

Stress responsivity The modern concept of stress and its importance for many human diseases was developed by the pioneering neuroendocrinologist Hans Selye, who discovered that various noxious stimuli caused what he called a ‘general adaptation syndrome,’ mediated in part by the pituitary and adrenal glands. An important component of the stress-responsive system is the hypothalamic-pituitary-adrenal (HPA) axis. Exposure to stress activates the HPA axis (Fig. 2). HPA axis activation or suppression influences addiction24–27. The question can therefore be posed: is there a genetic link

between HPA axis function and addiction? In addition to the classical feedback regulation of the HPA axis by corticosteroids, clinical studies with opioid antagonists demonstrate that the endogenous opioid system, via both µ and κ opioid receptors, also tonically inhibits the HPA axis. We have proposed that an atypical responsivity to stress and stressors, with a particular focus on the HPA axis, contributes to the continuation of specific addictions, as well as to relapse once the brain has undergone plasticity due to addiction1,24,25. We have found that active heroin addicts have a hypo-responsive HPA system and that patients with cocaine dependence, including former heroin addicts in methadone maintenance treatment with ongoing dependence on cocaine, show a hyper-responsive HPA axis1,26. In these clinical studies, it is not possible to distinguish whether this atypical stress responsivity preceded (because of underlying physiological and ultimately genetic conditions) or was caused by long-term self-administration of opiates or cocaine. In animal models of conditioned place preference and drug selfadministration, acute and chronic stress affect the HPA axis, as well as other components of stress responsivity in the brain, and may increase the reinforcing effects of drugs of abuse. Stressors can influence the rewarding properties of drugs at each of the stages in laboratory animal self-administration studies, including initiation, maintenance, extinction and reinstatement, which are thought to model human states of initiation and maintenance of addictions, withdrawal and relapse. Thus, in general, stress can enhance acquisition, increase resistance to extinction, and induce reinstatement of self-administration. Animal studies document physiological and corresponding molecular alterations in components of the HPA axis caused by acute or chronic administration of drugs of abuse. For example, administration of cocaine in a ‘binge’ protocol for 1 or 2 days to rats causes an increase in plasma corticosterone levels, which is significantly attenuated following chronic 14-day binge cocaine administration. Corticotropin releasing factor mRNA in the hypothalamus is also increased following 1 or 2 days of administration but is significantly reduced after 14 days28. In a series of clinical studies, recently abstinent cocaine-dependent subjects were read individually tailored scripts designed to provoke stressful, drug-cue related or neutral, relaxing experiences. Stressful and drug-cue related, but not neutral, scripts evoked increased craving, anxiety and cardiovascular measures, as well as increased plasma levels of ACTH, cortisol, prolactin and norepinephrine, not only indicating involvement of the HPA axis, but also suggesting that the sympatho-adreno-medullary system is involved in cocaine craving during abstinence29. The endogenous opioid system, specifically µ and κ opioid receptors, demonstrate inhibitory control over the HPA axis (Fig. 2). This is apparently tonic inhibition, rather than feedback and circadian inhibition, as is the case with glucocorticoid regulation of the axis25. The µ opioid receptor is the primary target of addictive opioid drugs. Mice lacking the µ opioid receptor gene (OPRM1) show dramatically reduced or absent analgesia, reward, physical dependence and respiratory depression in response to opiates, such as morphine (reviewed in ref. 2). Numerous polymorphisms of the OPRM1 gene have been identified. The most common coding region polymorphism is the A118G SNP with allelic frequencies that vary widely among populations (allele frequencies from 0.01 to 0.48) and results in an asparagine (Asn) to an aspartic acid (Asp) substitution at amino acid position 40, thereby abolishing a putative glycosylation site in the N terminus2,11,30. In in vitro studies, we found that the endogenous opioid peptide β-endorphin bound the 118G (Asp40) receptor variant with threefold greater affinity than the prototype 118A (Asn40) receptor30. Also, β-endorphin binding to the Asp40 receptors showed threefold greater potency in activation of G protein–coupled inwardly rectifying potassium (GIRK) channels, one of the important intracellular signaling systems of this

1454


Figure 2 Stress causes _ increased mRNA Hypothalamus synthesis and release of hypothalamic CRF _ corticotropin releasing CRF factor (CRF) into the portal circulation, + Anterior which acts on CRFR1 _ pituitary receptors in the anterior pituitary. This β-endorphin POMC _ induces synthesis of Endogenous proopiomelanocortin opioids ACTH (POMC) mRNA and (µ-,κ-) peptide in the anterior + Adrenal pituitary and release a Human, into the circulation guinea pig Glucocorticoids Cortisola b of β-endorphin and Mouse, rat Corticosteroneb adrenocorticotropic hormone (ACTH), which are derived from processing of POMC. ACTH acts on ACTH receptors in the adrenal cortex and induces release of the stress hormone cortisol (in humans and guinea pigs) or corticosterone (in rats and mice), which are primary mediators of the stress response. Cortisol or corticosterone exert negative feedback regulation at both the hypothalamus and the pituitary to inhibit the synthesis of POMC and release of ACTH and β-endorphin. In addition to this classical circadian negative feedback regulation by glucocorticoids, the endogenous opioid system, including both µ and κ opioid receptors, tonically inhibits this axis.

Ann Thomson


PERSPECTIVE

ders22. However, the prevalence of comorbidity in people when they first try a drug of abuse is not well defined. Certainly, many people who are already addicted to illicit drugs potentially could be classified as having antisocial personality disorder because of their criminal activity pertaining to the acquisition and use of illicit drugs. Attention deficit/hyperactivity disorder in the childhood or adult form is common, especially in people who are dependent on cocaine and other stimulants (reviewed in ref. 23). For each of the comorbid disorders, it is important to use more refined psychiatric diagnostic tools to determine if the psychiatric disorder preceded or followed the development of the addictive disorder. It has been established that genetics are somewhat involved in each of the psychiatric disorders just as in the addictive diseases discussed here. Thus, in the presence of comorbidity, it is difficult to determine which of the gene variants contribute to the psychiatric disease, to the addictive disease, or to both. The role of comorbidity in the genetics of addiction remains an area of controversy. For example, comorbidity may be mechanistically important in the vulnerability to or severity of addiction, requiring focused studies of its influence. Conversely, studies are needed in addictive disease populations, taking comorbidity into account as an independent variable, thus investigating genetic variation in addiction with and without comorbidities.


PERSPECTIVE receptor30. No other agonist tested showed differences in binding to, or GIRK activation of, the variant receptors30. The in vitro findings of changes in responses of the 118G variant µ opioid receptor led us to predict that HPA-mediated stress responsivity may be altered in people expressing the variant30,31. Although the molecular or cellular mechanisms have yet to be fully elucidated, these predictions have been borne out in clinical studies in which healthy individuals were administered a µ opioid receptor antagonist, naloxone or naltrexone, which causes immediate activation of the HPA axis by blocking the µ opioid receptor; that is, by disinhibition. Subjects heterozygous for the 118G allele showed a greater HPA response to opioid antagonist than did subjects with the only prototype receptor, as measured by serum ACTH and cortisol levels2,11. Additionally, people with the 118G variant receptors had a more favorable clinical response to treatment for alcoholism with the opioid antagonist naltrexone2,11. This difference in response to treatment may be mediated by differences in HPA axis activation owing to receptor genotype, as modest activation of this axis is desired by at least some alcoholics27. This difference in HPA axis responsivity may be a factor in the possible contribution of this variant to the risk for developing opiate addiction and alcoholism reported in some studies32,33. A second gene linking the HPA axis, stress response and addiction is COMT, which encodes an enzyme that catalyzes the degradative metabolism of the catecholamine neurotransmitters dopamine, norepinephrine and epinephrine, as well as hydroxylated estrogens. A common guanine-to-adenine transition34 in exon 4 causes the substitution of methionine for valine at residue 158. The methionine form has greater thermolability and a three- to fourfold lower enzymatic activity than the valine form11. Genetic linkage and association studies suggest that this polymorphism may be involved in several different psychiatric disorders. The low-activity methionine form is associated with increased risk for alcoholism in several studies. Genotype at this polymorphism may influence HPA axis function. After administration of naloxone, subjects with the homozygous Met/Met genotype have greater increases in plasma ACTH and cortisol than do people with one or more high-activity valine alleles (Val/Met or Val/Val)35. In this study, all subjects were A/A homozygous for the OPRM1 A118G SNP, as this polymorphism also affects HPA response to opiate antagonist challenge. Overall, the activity of the HPA axis seems to undergo extensive plasticity as a result of exposure to drugs of abuse. Furthermore, HPA responsivity is affected by genetic variants. Along with the finding that stress is a precipitating factor in relapse, these results point to the importance of more extensive studies of genetic variants in the HPA axis and drug addiction. Environmental factors The expression of a genetic predisposition may be, in part, conditional on exposure to environmental determinants. In twin studies, environmental factors, including family environment, influence the development of alcohol dependence in individuals with a relatively high genetic risk. The influence of family and non-family environmental factors also contribute to abuse of or dependence on other drugs of abuse4,5. Among maltreated children, those with the MAOA variant that directs high expression levels were not as likely to develop antisocial problems in adulthood as children with the low-expression variants36. MAOA metabolizes a variety of neurotransmitters, including serotonin, norepinephrine and dopamine; defects in the MAOA gene have been linked to aggression. Although the environment contributes to the development of antisocial traits, in these children, the resultant antisocial behavior was moderated by genetic factors.


Another association study investigated why stressful events may lead to depression in some individuals but not in others37. SERT has a repeat polymorphism in the promoter region, with the long form of the repeat polymorphism expressing higher levels of SERT mRNA. At 26 years of age, people with the long or short form of the SERT promoter polymorphism had similar depressive symptoms and episodes, as well as suicidal ideations, if they lacked ‘life events’ such as employment, relationship or health stressors from age 21 to 25. However, in people who had experienced stressful life events and had two copies of the short SERT alleles, depression and suicidal ideation increased at a much higher rate, whereas an intermediate increase was observed in heterozygous subjects. These results suggest that common genetic variants maintained at a high frequency in the population promote resistance to environmental stressors. Furthermore, the lack of replication in many genetic studies may be due to the specific gene-environment interaction that must occur for an effect to be observed. On the other hand, these studies may have to be revisited in light of the recent report of an A → G variant in the long repeat of the SERT promoter polymorphism that affects expression38. These two specific variants (the MAOA and SERT promoter polymorphisms) are each associated with alcoholism (Table 1). Childhood abuse also increases the risk of developing alcoholism or other drug addiction. These studies point to the critical interaction between specific genetic variants and the environment as leading to associations with addiction. Genetic factors directly associated with addiction As noted previously, genetic factors account for 30–60% of the overall variance in the risk for the development of drug addictions, but there may be different influences of environmental or genetic factors at different stages4–6. The potential influences of the personality traits of impulsivity and risk-taking, of stress responsivity, and comorbid psychiatric conditions, along with potential gene variants involved in each of these factors, have been discussed above. We will now highlight direct genetic studies of addiction to alcohol, opiates and cocaine and other stimulants. That is, these studies focus on genetic variants and addictive diseases without analyzing the personality traits mentioned above. Many of the genes for which there is evidence of association or linkage are those already discussed as potentially contributing to impulsivity, risk-taking, anxiety, depression and stress-responsivity (Table 1). Linkage studies have been conducted to identify genetic determinants of addictive diseases2,11,39–41. The Collaborative Study on the Genetics of Alcoholism (COGA), a multi-center effort to identify genes involved in alcoholism, was an early project7. New techniques allow association studies to be done on thousands of genes using microarray technology. A multiple pooling technique with a 1,494-SNP microarray identified 42 chromosomal regions that may be involved in vulnerability to drug abuse in African-Americans and European-Americans. All the affected subjects had polysubstance abuse, including nicotine and alcohol abuse or addiction, so the regions identified may contain genes that are involved in addictions to multiple substances41. This study of polysubstance abuse showed that at least 15 large chromosomal regions were shared with regions identified in one or more other linkage studies of alcoholism and nicotine addiction, suggesting that there may be general genetic factors for addiction40. Genetic variants may also contribute to opiate addiction. One promising candidate is the µ opioid receptor gene (OPRM1). Several individual variants and haplotypes at the OPRM1 locus are associated with opiate dependence2,11,31. The 118G allele of the common functional A118G SNP was associated with heroin addiction in two relatively non-admixed populations, one of Han Chinese and the other in central Sweden2. In the latter study, the population-attributable risk for the 118G allele was 21% for Swedish individuals with

1455


PERSPECTIVE two Swedish parents32. Some studies of the A118G SNP, as well as of other polymorphisms in this gene, have not identified association or linkage to an addiction with this locus, possibly owing to differences in the genetic makeup of the populations under study, differences in population substructure or the use of different assessment criteria. Another association of the OPRM1 118G allele with alcohol dependence has been reported in Swedish individuals from central Sweden, further indicating the importance of ethnic/cultural background33. We found an association between a single SNP and also a specific haplotype of variants of the κ opioid receptor gene (OPRK1) and opiate addiction42. Prodynorphin is the precursor to dynorphin peptides, the endogenous ligands of the κ opioid receptor that can attenuate cocaineinduced increases in perisynaptic dopamine levels in reward-related areas of the brain1. We found that a 68-base repeat polymorphism in the promoter of the dynorphin gene was associated with cocaine abuse or dependence and also with cocaine-alcohol dependence2. The specific associations of addictive status with µ and κ opioid receptor systems can be viewed in the context of the importance of these two systems in the neurobiology of reinforcement and reward by different drugs of abuse (including opiates and psychostimulants1). Alleles of the DRD2 gene are associated with alcoholism, cocaine dependence, psychostimulant abuse or polysubstance abuse2. Some studies report the DRD4 gene to be associated with opiate addiction or alcoholism, although these findings have not always been replicated12. The high-activity Val158 allele of the COMT gene V158M polymorphism is associated with polysubstance abuse43, with alcoholism11 and, in a family-based haplotype relative-risk study, with heroin addiction44. Functional magnetic resonance imaging shows that individuals with the high activity valine/valine genotype of the COMT gene have enhanced prefrontal cortex function when given amphetamine during a working memory task, whereas amphetamine caused deterioration of cortical efficiency in individuals with the methionine/methionine genotype (see ref. 11 for review). Alleles of the DRD4 and COMT genes also interact with methamphetamine abuse45. Cocaine-induced psychosis is associated with a potentially functional variable nucleotide tandem repeat in the 3′ untranslated region of DAT (reviewed in ref. 2). Variants of this gene have also been associated with amphetamine-induced psychosis2 and with alcoholism11. A functional polymorphism in the promoter region of DBH that causes lower plasma dopamine β-hydroxylase activity is associated with cocaine-induced paranoia46. Two studies reported an association of heroin dependence with polymorphisms in SERT, but this finding was not replicated in other studies (reviewed in ref. 2). Variants in SERT, TPH2 and MAOA and genes encoding serotonin receptors 5-HT1B and 5-HT2A have all been associated with alcoholism11. Alcohol dependence is associated with variants of the GABRA2 gene, which codes for the α2 subunit of GABAA; this gene is located in a region of chromosome 4p, which is linked and associated with alcoholism8. The endogenous cannabinoid system is also implicated in genetic studies of addictions. A trinucleotide repeat polymorphism in the 3′ flanking region of the cannabinoid receptor 1 (CNR1) gene is associated with intravenous drug abuse (cocaine, amphetamine or heroin)47. A synonymous coding region SNP in the 3′ untranslated region is associated with symptoms of delirium in alcohol withdrawal48. A study of 22 polymorphisms in CNR1 identified a haplotype in an intronic 5′ region of the gene that is associated with substance (cocaine, opiate, alcohol or other drug) abuse49. Fatty amide acid hydrolase, encoded by the FAAH gene, is an enzyme that metabolizes endogenous ligands of the cannabinoid receptors. A functional SNP that alters the sensitivity of the enzyme to protease in vitro is associated with drug and alcohol abuse50.

A pharmacokinetic gene product, the cytochrome CYP450 gene CYP2D6, has been studied in codeine dependence. This enzyme biotransforms codeine and several of its congeners into metabolites with greater opioid potency. The CYP2D6 gene is highly polymorphic, resulting in large differences in enzyme activity11. As detailed above, variants of genes involved in specific neurotransmitter systems are implicated in vulnerability to alcoholism; genes involved in biotransformation or degradation of alcohol are also implicated11. The alcohol-metabolizing enzymes alcohol dehydrogenase (ADH1B and ADH1C), and aldehyde dehydrogenase (ALDH) genes have variants that are protective against alcoholism11. The reports of associations of these alcohol-metabolizing gene variants with protection from alcoholism are diverse, robust and exhaustively reviewed elsewhere11. Some studies in Table 1 were primarily designed to test for associations between genetic variation and addictive diseases. Other studies focus on the association of genetic variants with personality traits such as impulsivity and risk taking. Many of these variants are associated in independent studies with both addictive diseases and personality traits (for example, SERT, TPH1, TPH2, COMT). One major focus for the future could be integrated studies on the role of personality trait variants in addictive diseases.

1456


Summary and conclusions Addiction is a complex disorder with interacting factors, including environmental factors, drug-induced neurobiological changes, comorbidity, personality traits and stress responsivity. Clearly, multiple genetic variants that affect these factors may work in concert to affect vulnerability and severity of addiction. As a concrete example, a functional SNP in the OPRM1 gene (A118G) influences the µ opioid receptor, as defined by molecular and cellular studies and in human studies, and results in clinically observable changes in stress responsivity, vulnerability to opiate addiction and alcoholism in defined populations, as well as in response to a specific addiction pharmacotherapy. The advent of more modern technologies, such as SNP microarrays, enhances our capacity to study genetic influence on the addictive diseases. Several important challenges remain for the near future; in particular, the refinement of phenotyping in the addictive diseases, which may focus on clinically relevant aspects of this disorder, such as age of initiation, speed of progression to regular drug use, severity of dependence or withdrawal, vulnerability to relapse, and response to specific pharmacotherapeutic treatments. Molecular resequencing of new and previously studied genes is of critical value in the discovery of genetic variants of potential interest. A relative standardization across laboratories in phenotyping and statistical approaches (and the sharing of these data) is desirable to assess more directly replicability and generalization across different populations. Without such relative standardization, meta-analyses of studies using highly disparate methodologies are difficult. Meta-analyses focus on particular questions (such as an association between a genetic variant and a personality trait or an addiction) and combine results from multiple studies into a coherent summary. Analyses are based on individual or aggregate patient data, with the former being the preferred type, although the use of the latter is more common. Phenotypic assessments, ethnic/cultural group studies and statistical methods used must be similar to decrease heterogeneity in the combined data. Hence, the results of meta-analyses of apparently similar studies may not be directly compared, and meta-analyses of disparate studies may be misleading. Additional information, including references for additional reading, is available in the Supplementary Note online.

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


ACKNOWLEDGMENTS This work was supported in part by US National Institutes of Health-National Institute on Drug Abuse (NIH-NIDA) Research Scientist Award Grant K05DA00049; NIH-NIDA Research Center Grant P60-DA05130; NIH-GCRC General Research Center Grant MOI-RR00102; and the New York State Office of Alcoholism and Substance Abuse (OASAS). Thanks to K. Lavoie for assistance in the preparation of this manuscript. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. Kreek, M.J., LaForge, K.S. & Butelman, E. Pharmacotherapy of addictions. Nat. Rev. Drug Discov. 1, 710–726 (2002). 2. Kreek, M.J., Bart, G., Lilly, C., LaForge, K.S. & Nielsen, D.A. Pharmacogenetics and human molecular genetics of opiate and cocaine addictions and their treatments. Pharmacol. Rev. 57, 1–26 (2005). 3. Kosten, T.A. et al. Miserendino, M.J. & Haile, C.N., DeCaprio, J.L., Jatlow, P.I., Nestler, E.J. Acquisition and maintenance of intravenous cocaine self-administration in Lewis and Fischer inbred rat strains. Brain Res. 778, 418–429 (1997). 4. Tsuang, M.T. et al. Co-occurrence of abuse of different drugs in men: the role of drugspecific and shared vulnerabilities. Arch. Gen. Psychiatry 55, 967–972 (1998). 5. Kendler, K.S., Jacobson, K.C., Prescott, C.A. & Neale, M.C. Specificity of genetic and environmental risk factors for use and abuse/dependence of cannabis, cocaine, hallucinogens, sedatives, stimulants, and opiates in male twins. Am. J. Psychiatry 160, 687–695 (2003). 6. Tsuang, M.T. et al. Genetic and environmental influences on transitions in drug use. Behav. Genet. 29, 473–479 (1999). 7. Foroud, T. et al. Alcoholism susceptibility loci: Confirmation studies in a replicate sample and further mapping. Alcohol. Clin. Exp. Res. 24, 933–945 (2000). 8. Edenberg, H.J. et al. Variations in GABRA2, encoding the α2 subunit of the GABAA receptor, are associated with alcohol dependence and with brain oscillations in the beta frequency. Am. J. Hum. Genet. 74, 705–714 (2004). 9. Wang, J.C. et al. Evidence of common and specific genetic effects: association of the muscarinic acetylcholine receptor M2 (CHRM2) gene with alcohol dependence and major depressive syndrome. Hum. Mol. Genet. 13, 1903–1911 (2004). 10. Ott, J. Issues in association analysis: error control in case-control association studies for disease gene discovery. Hum. Hered. 58, 171–174 (2004). 11. Kreek, M.J., Nielsen, D.A. & LaForge, K.S. Genes associated with addiction: alcoholism, opiate and cocaine addiction. Neuromolecular Med. 5, 85–108 (2004). 12. Nielsen, D.A. et al. A tryptophan hydroxylase gene marker for suicidality and alcoholism. Arch. Gen. Psychiatry 55, 593–602 (1998). 13. Coccaro, E.F. et al. Familial correlates of reduced central serotonergic system function in patients with personality disorders. Arch. Gen. Psychiatry 51, 318–324 (1994). 14. Nielsen, D.A. et al. Suicidality and 5-hydroxyindoleacetic acid concentration associated with a tryptophan hydroxylase polymorphism. Arch. Gen. Psychiatry 51, 34–38 (1994). 15. Limosin, F. et al. Association between dopamine receptor D3 gene BalI polymorphism and cognitive impulsiveness in alcohol-dependent men. Eur. Psychiatry 20, 304–306 (2005). 16. Lusher, J.M., Chandler, C. & Ball, D. Dopamine D4 receptor gene (DRD4) is associated with novelty seeking (NS) and substance abuse: the saga continues. Mol. Psychiatry 6, 497–499 (2001). 17. Schinka, J.A., Letsch, E.A. & Crawford, F.C. DRD4 and novelty seeking: results of metaanalyses. Am. J. Med. Genet. B Neuropsychiatr. For. Genet. 114, 643–648 (2002). 18. Primus, R.J. et al. II. Localization and characterization of dopamine D4 binding sites in rat and human brain by use of the novel, D4 receptor-selective ligand [3H]NGD 94–1. J. Pharmacol. Exp. Ther. 282, 1020–1027 (1997). 19. Uhl, G., Blum, K., Noble, E. & Smith, S. Substance abuse vulnerability and D2 receptor genes. Trends Neurosci. 16, 83–88 (1993). 20. Gelernter, J., Goldman, D. & Risch, N. The A1 allele at the D2 dopamine receptor gene and alcoholism: a reappraisal. J. Am. Med. Assoc. 269, 1673–1677 (1993). 21. Neville, M.J., Johnstone, E.C. & Walton, R.T. Identification and characterization of ANKK1: a novel kinase gene closely linked to DRD2 on chromosome band 11q23.1. Hum. Mutat. 23, 540–545 (2004). 22. Rounsaville, B.J., Weissman, M.M., Crits-Christoph, K., Wilber, C. & Kleber, H. Diagnosis and symptoms of depression in opiate addicts. Course and relationship to


treatment outcome. Arch. Gen. Psychiatry 39, 151–156 (1982). 23. Levin, F. & Kleber, H. Attention deficit hyperactivity disorder and substance abuse: relationships and implications for treatment. Harv. Rev. Psychiatry 2, 246–258 (1995). 24. Kreek, M.J. Medical safety, side effects and toxicity of methadone. Proceedings of the Fourth National Conference on Methadone Treatment, National Association for the Prevention of Addiction to Narcotics (NAPAN)-NIMH, 171–174 (1972). 25. Kreek, M.J. Opiates, opioids and addiction. Mol. Psychiatry 1, 232–254 (1996). 26. Schluger, J.H., Borg, L., Ho, A. & Kreek, M.J. Altered HPA axis responsivity to metyrapone testing in methadone maintained former heroin addicts with ongoing cocaine addiction. Neuropsychopharmacology 24, 568–575 (2001). 27. O’Malley, S.S., Krishnan-Sarin, S., Farren, C., Sinha, R. & Kreek, M.J. Naltrexone decreases craving and alcohol self-administration in alcohol dependent subjects and activates the hypothalamo-pituitary-adrenocortical axis. Psychopharmacology (Berl.) 160, 19–29 (2002). 28. Zhou, Y. et al. Corticotropin-releasing factor and type 1 corticotropin-releasing factor receptor messenger RNAs in rat brain and pituitary during ‘binge’-pattern cocaine administration and chronic withdrawal. J. Pharmacol. Exp. Ther. 279, 351–358 (1996). 29. Sinha, R. et al. Hypothalamic-pituitary-adrenal axis and sympatho-adreno-medullary responses during stress-induced and drug cue-induced cocaine craving states. Psychopharmacology (Berl.) 170, 62–72 (2003). 30. Bond, C. et al. Single nucleotide polymorphism in the human mu opioid receptor gene alters beta-endorphin binding and activity: Possible implications for opiate addiction. Proc. Natl. Acad. Sci. USA 95, 9608–9613 (1998). 31. LaForge, K.S., Yuferov, V. & Kreek, M.J. Opioid receptor and peptide gene polymorphisms: potential implications for addictions. Eur. J. Pharmacol. 410, 249–268 (2000). 32. Bart, G. et al. Substantial attributable risk related to a functional mu-opioid receptor gene polymorphism in association with heroin addiction in central Sweden. Mol. Psychiatry 9, 547–549 (2004). 33. Bart, G. et al. Increased attributable risk related to a functional mu-opioid receptor gene polymorphism in association with alcohol dependence in central Sweden. Neuropsychopharmacology 30, 417–422 (2005). 34. Lachman, H.M. et al. Human catechol-O-methyltransferase pharmacogenetics: description of a functional polymorphism and its potential application to neuropsychiatric disorders. Pharmacogenetics 6, 243–250 (1996). 35. Oswald, L.M., McCaul, M., Choi, L., Yang, X. & Wand, G.S. Catechol-O-methyltransferase polymorphism alters hypothalamic-pituitary-adrenal axis responses to naloxone: a preliminary report. Biol. Psychiatry 55, 102–105 (2004). 36. Caspi, A. et al. Role of genotype in the cycle of violence in maltreated children. Science 297, 851–854 (2002). 37. Caspi, A., Sugden, K. & Moffitt, T.E. Influence of life stress on depression: moderation by a polymorphism in the 5-HTT gene. Science 301, 386–389 (2003). 38. Hu, X. et al. An expanded evaluation of the relationship of four alleles to the level of response to alcohol and the alcoholism risk. Alcohol. Clin. Exp. Res. 29, 8–16 (2005). 39. Gelernter, J. et al. Genomewide linkage scan for cocaine dependence and related traits: significant linkages for a cocaine-related trait and cocaine-induced paranoia. Am. J. Med. Genet. B Neuropsychiatr. For. Genet. 136, 45–52 (2005). 40. Uhl, G.R., Liu, Q.R. & Naiman, D. Substance abuse vulnerability loci: converging genome scanning data. Trends Genet. 18, 420–425 (2002). 41. Uhl, G. Molecular genetics of substance abuse vulnerability: remarkable recent convergence of genome scan results. Ann. NY Acad. Sci. 1025, 1–13 (2004). 42. Yuferov, V. et al. Redefinition of the human kappa opioid receptor (OPRK1) structure and association of haplotypes with opiate addiction. Pharmacogenetics 14, 793–804 (2004). 43. Vandenbergh, D.J., Rodriguez, L.A., Miller, I.T., Uhl, G.R. & Lachman, H.M. High-activity catechol-O-methyltransferase allele is more prevalent in polysubstance abusers. Am. J. Med. Genet. 74, 439–442 (1997). 44. Horowitz, R. et al. Confirmation of an excess of the high enzyme activity COMT val allele in heroin addicts in a family-based haplotype relative risk study. Am. J. Med. Genet. 96, 599–603 (2000). 45. Li, T. et al. Association analysis of the DRD4 and COMT genes in methamphetamine abuse. Am. J. Med. Genet. B Neuropsychiatr. For. Genet. 129, 120–124 (2004). 46. Cubells, J.F. et al. A haplotype at DBH, associated with low plasma dopamine-βhydroxylase activity, also associates with cocaine-induced paranoia. Mol. Psychiatry 5, 56–63 (2000). 47. Comings, D.E. et al. Cannabinoid receptor gene (CNR1): Association with IV drug use. Mol. Psychiatry 2, 161–168 (1997). 48. Schmidt, L.G. et al. Association of a CB1 cannabinoid receptor gene (CNR1) polymorphism with severe alcohol dependence. Drug Alcohol Depend. 65, 221–224 (2002). 49. Zhang, P-W. et al. Human cannabinoid receptor 1: 5′ exons, candidate regulatory regions, polymorphisms, haplotypes and association with polysubstance abuse. Mol. Psychiatry 9, 916–931 (2004). 50. Sipe, J.C., Chiang, K., Gerber, A.L., Beutler, E. & Cravatt, B.F. A missense mutation in human fatty acid amide hydrolase associated with problem drug use. Proc. Natl. Acad. Sci. USA 99, 8394–8399 (2002).

1457


PERSPECTIVE


Decision making, impulse control and loss of willpower to resist drugs: a neurocognitive perspective Antoine Bechara Here I argue that addicted people become unable to make drug-use choices on the basis of long-term outcome, and I propose a neural framework that explains this myopia for future consequences. I suggest that addiction is the product of an imbalance between two separate, but interacting, neural systems that control decision making: an impulsive, amygdala system for signaling pain or pleasure of immediate prospects, and a reflective, prefrontal cortex system for signaling pain or pleasure of future prospects. After an individual learns social rules, the reflective system controls the impulsive system via several mechanisms. However, this control is not absolute; hyperactivity within the impulsive system can override the reflective system. I propose that drugs can trigger bottom-up, involuntary signals originating from the amygdala that modulate, bias or even hijack the goal-driven cognitive resources that are needed for the normal operation of the reflective system and for exercising the willpower to resist drugs. Imagine yourself at a party during the first year in college, your friends offering you alcoholic drinks and drugs. In the back of your mind, you hear the voice of your parents warning you against such activities. What would you do? This is a hard decision, but you are the one who will ultimately decide, with a clear sense of exercising free will. Willpower, as defined by the Encarta World English Dictionary, is a combination of determination and self-discipline that enables somebody to do something despite the difficulties involved. This mechanism enables one to endure sacrifices now in order to obtain benefits later, or vice versa. There are similarities in behavior between patients with ventromedial prefrontal cortex (VMPC) damage and drug addicts. Both often deny, or are not aware, that they have a problem. When faced with a choice that brings immediate reward, even at the risk of incurring future negative outcomes, including loss of reputation, job, and family, they appear oblivious to the consequences of their actions. (For the purposes of this piece, VMPC is defined as the ventral medial prefrontal cortex and the medial sector of the orbitofrontal cortex, thus encompassing Brodmann’s areas (BA) 25, lower 24, 32 and medial aspect of 11, 12 and 10.) After injury to this area, patients tend to recover normal intelligence, memory and other cognitive functions, but emotion, affect and social behavior change

completely. The patients begin to make choices that often lead to financial losses, loss in social standing, and even loss of family and friends. When this syndrome was initially described1, the decision making deficit seen in these patients was puzzling because their poor decision making and failure to learn from repeated mistakes was obvious in their everyday lives, but there was no laboratory probe to detect and measure their impairment. This challenge was overcome by the development of the Iowa Gambling Task2. In this task, subjects choose from four decks of cards, each with a different potential payoff, to maximize their monetary gain. After each choice, subjects receive feedback telling them how much money they won or lost. Through this feedback, normal decision-makers learn to avoid decks that yield high immediate gains but larger future losses down the line. In contrast, patients with VMPC damage and drug addicts persist in making disadvantageous choices despite the rising losses associated with their choices2. Early on, abnormalities in the VMPC region were observed in cocaine addicts3. These deficits were linked to the decision making impairments of VMPC patients when cocaine addicts were shown to make poor decisions on the Iowa Gambling Task4. This linkage energized a new line of research aimed at understanding the relationship between substance abuse and poor decision making (see refs. 2,5–8 for reviews). The aim of this perspective is to highlight the key role of choice in addiction, and to present a broad conceptual framework that brings together several disparate lines of research on addiction. The main purpose is to provide a gross picture of how multiple brain mechanisms come together in addiction, instead of focusing on one specific process of addiction, or one specific brain region. The view I present here is that addiction is a condition in which the neural mechanisms that enable one to choose according to long-term outcomes are weakened, thus leading to loss of willpower to resist drugs. This complements previous proposals that disruption of the VMPC leads to loss of self-directed behavior in favor of more automatic sensory-driven behavior3.

Published online 26 October 2005; doi:10.1038/nn1584

A neural system for willpower The somatic marker hypothesis is a systems-level neuroanatomical and cognitive framework for choosing according to long-term, rather than short-term, outcomes1. The key idea of this hypothesis is that the process of decision making depends in many important ways on neural substrates that regulate homeostasis, emotion and feeling8. The term ‘somatic’ refers to the collection of body- and brain-related responses that are hallmarks of affective and emotional responses. Both the amygdala and VMPC are critical for triggering somatic states, but as I will explain shortly, the amygdala responds to events that occur in the environment, whereas the VMPC triggers somatic states from memories, knowledge and cognition. In order for somatic signals to influence

1458


Antoine Bechara is at the Institute for the Neurological Study of Emotion and Creativity, Department of Psychology, University of Southern California, Los Angeles, California 90089-2520, USA. e-mail: [email protected]

cognition and behavior, they must act on appropriate neural systems. As I will explain, there are several target sites through which somatic (affective) signals modulate cognition and behavior, and I will propose that this modulation is in fact mediated by neurotransmitter systems (Fig. 1). Thus, during the process of pondering decisions, the immediate and future prospects of an option may trigger numerous affective (somatic) responses that conflict with each other; the end result is that an overall positive or negative signal emerges. We have proposed that the mechanisms that determine the valence of the dominant pattern of affective signaling are consistent with the principles of natural selection (that is, survival of the fittest)9. In other words, numerous and conflicting signals may be triggered simultaneously, but stronger ones gain selective advantage over weaker ones. Over the course of pondering a decision, positive and negative signals that are strong are reinforced, and weak ones are eliminated. This process can be very fast, and ultimately a winner takes all: in other words, an overall, more dominant, pattern of affective signaling emerges that then can act on appropriate neural systems to modulate cognition and behavior. On the basis of this neural framework, I propose that willpower emerges from the dynamic interaction of two separate, but interacting, neural systems: an impulsive system, in which the amygdala is a critical neural structure involved in triggering the affective/emotional signals of immediate outcomes, and a reflective system, in which the VMPC is a critical neural structure involved in triggering the affective/emotional signals of long-term outcomes (Fig. 1). This framework addresses one important question in drug addiction: of the millions of people who drink alcohol or experiment with drugs, why do only about 10% become addicted? The view I present here challenges the old thinking that people may be equally vulnerable to addiction once drugs are made available, as drug use can induce neuronal changes that lead to addiction. I argue that before one gets to the stage where a certain pattern of drug use can cause changes to the brain, there is a decision by the person to use, or not to use the drug. This mechanism protects most individuals who have used drugs from losing control and succumbing to addiction. For some individuals, however, this decision making mechanism is relatively weak. Such individuals are vulnerable to addiction because the process that enables one to inhibit actions elicited by the impulsive system is dysfunctional. The source of this dysfunction, I will suggest, can be genetic or environmentally induced. The impulsive system Physiological evidence suggests that responses triggered through the amygdala are short lived and habituate very quickly10. Therefore, we have suggested that pleasant or aversive stimuli, such as encountering an object that induces fear (a ‘fear object’, such as a snake) or a cue predictive of a fear object, trigger quick, automatic and obligatory affective/emotional responses through the amygdala system11. According to the somatic marker framework, the amygdala links the features of the stimulus to its affective/emotional attributes. The affective/emotional response is evoked through visceral motor structures such as the hypothalamus and autonomic brainstem nuclei that produce changes in internal milieu and visceral structures, as well as through behaviorrelated structures such as the striatum, periaqueductal gray (PAG) and other brainstem nuclei that produce changes in facial expression and specific approach or withdrawal behaviors1. Unlike food and water, money does not initially have affective properties, but acquires them with learning, such that exposure to monetary reward triggers affective signals through the amygdala system. We have shown that autonomic responses to large sums of monetary gains or losses depend on the integrity of the amygdala, as patients with bilateral amygdala damage fail to show such responses11. This is consistent with research


Figure 1 A schematic diagram illustrating key structures belonging AC DLPC to the impulsive Striatum system (red) and the Insula reflective system (blue). An emergent dominant pattern of affective signaling can DA modulate activity of VMPC Hip A several components 5-HT of the impulsive and reflective systems. These include regions involved in (i) representing patterns of affective states (e.g., the insula and somatosensory cortices); (ii) triggering of affective states (e.g., amygdala (A) and VMPC); (iii) memory, impulse and attention control (e.g., lateral orbitofrontal, inferior frontal gyrus and dorsolateral prefrontal (DLPC), hippocampus (Hip) and anterior cingulate (AC); and (iv) behavioral actions (e.g., striatum and supplementary motor area). 5-HT: serotonin; DA: dopamine.

Ann Thomson


PERSPECTIVE

showing that the brain can encode the value of various options on a common scale12, thus suggesting that there may be a common neural ‘currency’ that encodes the value of different options, thus allowing the reward value of money to be compared with that of food, sex or other rewards. Similarly, drugs may acquire powerful affective and emotional properties. In addicts, fast, automatic and exaggerated autonomic responses are triggered by cues related to the substance they abuse, similar to the effects of monetary gains2. Several lines of direct and indirect behavioral evidence have supported the view that conditioned approach behavior to drug cues relates to abnormal activity in the amygdala–ventral striatum system, thereby resulting in exaggerated processing of the incentive values of substance-related cues13. This ascribes a functional role to the striatum in the motivational and behavioral aspects of drug seeking, and it is consistent with the currently proposed framework of addiction. The reflective system Affective reactions can also be generated from recall of personal—or imagination of hypothetical—affective/emotional events. Affective state patterns develop in brainstem nuclei (such as the parabrachial nuclei) and in somatosensory cortices (for example, insula, somatosensory and posterior cingulate cortices) from prior experiences of reward and punishment1. After an affective state has been experienced at least once, a neural pattern for this state is formed. Subsequent evocation of memories of a previous experience reactivates the pattern of affective state belonging to an original experience. Provided that representations of these affective state patterns develop normally, the VMPC is a critical substrate in the neural system necessary for triggering affective states from recall or from imagination11. This hypothesis is based on evidence from patients with lesions in the VMPC11. However, it is also reasonable to suggest based on this evidence that recalling the experience of a drug reactivates the pattern of affective state belonging to the actual previous encounter with that drug. This mechanism should also bring up the negative consequences associated with drug use. These negative consequences are not simply aversive experiences resulting from the actual consumption of the drug. Rather, they relate to social (such as trouble with the law, family or finances) and psychological harms associated with drug use. The affective state patterns of these negative consequences become represented in the brain when individuals learn from parents or society about

1459


PERSPECTIVE the dangers of drug use. Therefore, one does not need to use drugs in order to fear their consequences; these negative consequences should be there, even before experimenting with drugs. However, having poor mechanisms of decision making renders individuals oblivious to these negative consequences, thus facilitating their escalation of drug use, and vulnerability to succumb to addiction. Normal functioning of the VMPC is contingent upon the integrity of other neural systems. One system involves the insula and other somatosensory cortices, especially on the right side, that are critical for representing patterns of emotional/affective states1. Patients with right parietal damage (encompassing insula and somatosensory cortex) show impairments in decision making11; addicts show functional abnormalities in these parietal regions when performing decision making tasks7. The other system involves the dorsolateral sector of the prefrontal cortex and the hippocampus, which are critical for memory11. Indeed, maintaining an active representation of memory over a delay period involves the dorsolateral sector of the prefrontal cortex, and patients with damage to this structure show compromised decision making14; addicts who have deficits in working memory also show compromised decision making15,16. Thus, decision making depends on systems for memory as well as for emotion and affect. Damage to any of these systems compromises the ability to make decisions that are advantageous in the long term. The VMPC region links these systems together, and therefore when it is damaged, there are many manifestations, including alterations of emotional/affective experience, poor decision making and abnormal social functioning11,14. Several voxel-brain-morphometry studies of brain scans of addicts found varying degrees of structural abnormalities in main components of the reflective system (Fig. 1), including the VMPC, anterior cingulate, insular cortex17, dorsolateral prefrontal cortex and lateral orbitofrontal/ inferior frontal gyrus18. Abnormalities have also been detected in white matter pathways connecting these structures19,20. Convergent results have also been obtained from functional neuroimaging studies (see refs. 3,7,8 for reviews). However, it is difficult to determine whether these abnormalities preceded or were the consequences of drug use. My view is that a degree of abnormality pre-existed the addiction state, by facilitating the progress from experimentation to addiction. However, any subsequent excessive and chronic use of drugs can exacerbate these abnormalities. Top-down control mechanisms of the reflective system Decision making reflects a process in which a choice is made after reflecting on the consequences of that choice. The choice between another drug use episode and the potential of losing a job, family breakdown and financial ruin down the line presents a dilemma to an addict, and a decision has to be made. Individuals with a weakness in this process (that is, those who do not reflect on the consequences of their decisions) may be similar to individuals with the personality trait of ‘nonplanning impulsivity’, a tendency to live for the moment with no regard for the future21, or individuals that lack the trait of ‘premeditation’, a tendency to think and reflect on the consequences of an act before engaging in that act22. Several tasks are now used to study this decision making processes, including the Iowa Gambling Task and the Cambridge Gamble and Risk Tasks14,23. A critical neural region for this mechanism is the VMPC region, but other neural components outlined earlier are also important11. Impairments in decision making are evident in addicts, regardless of the type of drug they abuse, which suggests that poor decision making may relate to addiction in general, rather than the effects of one specific type of drugs. Alcohol, cannabis, cocaine, opioid and methamphetamine abusers show impairments in decision making on a variety of tasks2,5,6,23. Although the differences in cognitive impairments brought

by the use of different drugs remains elusive, we have obtained preliminary evidence suggesting that chronic use of methamphetamine may be more harmful to decision making than use of other drugs24. Direct comparison of the decision making impairments in addicts on the Iowa Gambling Task versus patients with VMPC damage showed that a significantly high proportion of addicts (63%, versus 27% of normal controls) performed within the range of VMPC patients, whereas the rest performed within the range of the majority of normal controls25. Further characterization of these decision making deficits, using skin conductance response (SCR) measures as indices of affective states during performance of the task, showed that this small minority of addicts (the 37% of addicts who performed normally) matched normal controls in all respects. However, the remainder of the addicts (the 63% who performed abnormally) had two profiles: one subgroup matched the VMPC patients in all respects (that is, they had abnormal SCRs when they pondered risky decisions), but another subgroup did not match the VMPC patients. This pattern of abnormal physiological responses when making risky decisions in addicts was also obtained with the Cambridge Gamble Task26. A minority of normal controls performed like addicts and VMPC patients on the Iowa Gambling Task, and with additional SCR measures, some of them matched the profile of VMPC patients. The remainder of the controls were more like the addicts who did not match the VMPC patients2,25. These studies suggest that decision making deficits in addicts, and surprisingly, in some normal controls, are not uniform across all individuals. My view is that attention to individual, as opposed to group, differences in these decision making deficits is the key to understanding the nature of the addiction problem, its prognosis and possible treatment. There may be more than one mechanism by which the reflective system exerts control over the impulsive system. Besides decision making, there are other mechanisms of inhibitory control, one of which is the ability to deliberately suppress dominant, automatic or pre-potent responses27. For instance, acting quickly without an intention to act (as in the case of acting impulsively and using a drug without thinking) reflects an instance of weakness in this mechanism. Poor performance on several laboratory instruments requiring response inhibition reflects deficits in this mechanism of impulse control27. A critical neural region for this mechanism seems to be the more posterior area of the VMPC region, which includes the anterior cingulate and the basal forebrain, as patients with lesions in this area demonstrate signs of disinhibition and poor impulse control11. Disturbances in this mechanism may relate to the personality trait of motor impulsivity, the tendency to act without thinking21, or the trait of ‘urgency’, the tendency to experience strong impulses, frequently under conditions of negative affect22. Addicts show poor performance on tasks requiring the inhibition of pre-potent motor responses, and functional neuroimaging studies in addicts with inhibition deficits reveal diminished activity in neural systems involved in these inhibitory control mechanisms6,8. Another mechanism of impulse control is the ability to resist the intrusion of information that is unwanted or irrelevant27. Difficulties inhibiting particular thoughts or memories, such as thinking about drugs, and shifting attention to something else, reflect instances of weakness in this mechanism. Poor performance on tasks requiring internal inhibition of intrusive information reflects weakness in this mechanism27, and a critical neural region for this mechanism appears to be the lateral orbitofrontal and dorsolateral (inferior frontal gyrus) regions of the prefrontal cortex. Patients with damage in these areas make perseverative errors and have difficulties shifting attention28. Disturbances in this mechanism may relate to the personality trait of ‘cognitive impulsivity’, the tendency to make up one’s mind quickly or have problems concentrating21, or the trait

1460


PERSPECTIVE


of ‘perseverance’, the ability to remain focused on a task that may be boring or difficult22. Addicts show deficits in this mechanism of impulse control, as they demonstrate poor performance on tasks requiring the internal inhibition of an intention to act28 (E.A. Crone, C. Cutshall, E. Recknor, W.P.M. Van den Wildenberg & A.B., Soc. Neurosci. Abstr. 33,427, 2003). Bottom-up influence of the impulsive system The reflective system may generate affective states through top-down mechanisms, but then ascending signals from these affective states can exert bottom-up influence on cognition. Thus, when one is pondering a decision, numerous affective signals that conflict with each other may be triggered simultaneously through both the impulsive and reflective systems. The result is emergence of an overall positive or negative affective state. Ascending signals from this overall affective state can then modulate activity of several components of the impulsive and reflective systems (Fig. 1). We have previously proposed that the key mechanism by which these bottom-up signals modulate synaptic activity at telencephalic targets is pharmacological9. The cell bodies containing the neurotransmitter dopamine, serotonin, noradrenaline and acetylcholine are located in the brainstem; the axon terminals of these neurotransmitter neurons make synapses on cells and/or terminals throughout cortex. Anatomically, both the amygdala and VMPC have direct access to these neurotransmitter cell bodies in the brainstem. For affective states and homeostatic signals generated in the body, a number of channels can convey their signals to these neurotransmitter nuclei, but we have suggested that the vagus nerve is the most critical11. Changes in neurotransmitter release can modulate synaptic activity in several components of the impulsive and reflective systems. First, changes in representation of patterns of affective states (for example, in the insula and other somatosensory cortices) can lead to an increase in the reward utility of the drug. Second, changes in triggering of affective states (for example, in amygdala and VMPC) can lower the threshold for triggering subsequent affective signals related to drugs. Third, alterations in impulse control and the inhibition of unwanted memories or thoughts (for example, in lateral orbitofrontal, inferior frontal gyrus and dorsolateral prefrontal, hippocampus, and anterior cingulate) can strengthen thoughts about drugs and make shifting attention to other thoughts more difficult. Finally, changes in regions involved in behavior (striatum and supplementary motor area) can translate into drug use (Fig. 1). The outline of these pharmacological systems given here is very simplistic, mainly because there are many excellent reviews that describe the molecular mechanisms by which neurotransmitters affect synaptic activity in addictive states and that explain how these activities influence cognitive systems such as memory (see refs. 29,30 for reviews). Other excellent lines of research have attempted to differentiate the specific roles of dopaminergic, serotonergic, or noradrenergic systems in decision making, impulse control31,32 and time delay33. Therefore, the main purpose here is not to detail the processes and mechanisms of any one specific pharmacological system. Rather, the goal is to illustrate (i) how one can relate molecular and pharmacological studies on drug addiction to neural systems concerned with mechanisms of affect and emotion and (ii) the influence of drug addiction on cognition. The proposed arrangement provides a way for affective signals to exert a bottom-up influence on the reflective system. If, for instance, the signals triggered by the impulsive system were relatively strong, they would have the capacity to hijack the top-down goal-driven cognitive resources needed for the normal operation of the reflective system and exercising the willpower to resist drugs.


Hyperactive impulsive system Hyperactivity in bottom-up mechanisms of the impulsive system can weaken control of the reflective system. Evidence suggests that conditions leading to hyperactivity in this system include hypersensitivity, and attention bias, to reward. Addicts trigger exaggerated autonomic responses to cues related to the substances they abuse (see refs. 2,25 for reviews). Although addicts show blunted affective responses to affective stimuli that are not drug related34, we have shown that addicts trigger exaggerated autonomic responses when exposed to monetary reward in the Iowa Gambling Task2,25. Perhaps money represents a special case, in that it may be automatically linked to buying drugs. Using different versions of the Iowa Gambling Task, combined with SCR measures, we identified a subgroup of addicts that were different from both VMPC patients and the majority of normal controls; this subgroup of addicts was drawn to choices that yielded larger gains, irrespective of the losses that were encountered, and they generated exaggerated SCRs when they won money2,25. Direct autonomic responses to wins and losses are blocked in patients with bilateral amygdala damage. In contrast, in VMPC patients, the SCR defect is specific to the anticipatory phase when they are pondering which option to choose11. This suggests that addicts suffer from the opposite condition of amygdala lesion patients; that is, their amygdala is overresponsive to reward. This is supported by functional neuroimaging studies showing increased amygdala activity in response to drug-related cues35,36 and that this exaggerated brain response generalizes to monetary reward37. Other studies using tasks in which subjects were required to respond to targets (drug-related stimuli) but not respond to distracters (neutral stimuli) suggested that substance-related cues trigger bottom-up mechanisms in substance abusers, influencing top-down cognitive mechanisms such as motor impulse and attention control38. Another approach for studying these attention biases has been to use cognitive models6 that deconstruct complex behavioral decisions, such as those made in the Iowa Gambling Task, into simpler component processes of decision making. One of the component processes is the tendency of a subject to pay more attention to gains or losses encountered on previous trials in order to make future decisions. Addicts show patterns of high attention to monetary gains (which are more frequent in men than in women6) thus providing indirect evidence for the hypothesis that the amygdala system in addicts is hyperactive in response to monetary reward. Modulating factors The control function of the reflective system is complex, and even under normal circumstances, several factors can modify the strength of affective signals triggered by the reflective system, thus influencing its control over the impulsive system. Indeed, one of the fundamental questions in decision making research is how humans assign value to options. Several factors affect the value of a choice, and research has begun to explore the neural basis of these factors. We have proposed a neural framework for how factors that affect decision making—such as time delay, the probability of the outcome or the tangibility of the reward—could be implemented in the VMPC9. We have suggested that information conveying immediacy (the near future) engages more posterior VMPC (including anterior cingulate, basal forebrain and nucleus accumbens), whereas information conveying delay (distant future) engages more anterior VMPC (such as frontal pole)9. This is on the basis of the finding that major advancement in the size, complexity and connectivity of the frontal lobes in humans has occurred in relation to Brodmann area (BA) 10 (that is, the frontal pole)39. Furthermore, the more posterior areas of the VMPC (such as BA 25) are directly connected to brain structures involved in triggering (autonomic, neu-

1461


PERSPECTIVE rotransmitter nuclei) or representing (sensory nuclei in the brainstem, insular and somatosensory cortices) affective states, whereas access of more anterior areas is polysynaptic and indirect40. It follows that coupling of information to representations of affective states via posterior VMPC is associated with relatively fast, effortless, and strong affective signals, whereas the signaling via more anterior VMPC is relatively slowed, effortful and weak. This view is supported by recent functional imaging studies addressing how the perceived delay to receiving a reward modulates activity in reward-related brain areas33. This discounting mechanism of time is also relevant to addiction, as addicts tend to exhibit a higher temporal discounting rate than normal people; that is, they prefer smaller, sooner rewards over larger, later rewards23. Thus, events that are more immediate in time (such as having the drug now as opposed to the delayed consequences) have a stronger capability to influence decision making and hijack cognition in the direction of short-term outcomes. Similarly, we have suggested that information conveying higher certainty (or higher probability) engages posterior VMPC, whereas information conveying lower certainty engages anterior VMPC9. Functional imaging studies implicating the parietal cortex and anterior cingulate cortex in computing the probability of outcomes on the basis of available options (see ref. 7 for a review) are supportive of this view. This mechanism for processing probabilities is also relevant to addiction, as cocaine addicts show abnormalities in the activity of neural structures critical for decision making in proportion to the degree of certainty (or uncertainty) that they have about receiving their drug at the end of a brain scanning session3. Finally, reward values are processed by the VMPC region, and representations of these values are modulated by homeostatic factors such as hunger41. Given the view that neural systems supporting drug reward have evolved to subserve natural motivational functions, such as feeding42, drug withdrawal can be viewed like hunger43 in that once it is present, it increases the utility of drug reward, and, in doing so, it influences the decision to use drugs. This suggestion is consistent with the incentive motivational view of drug addiction proposing that although physical withdrawal signs are neither necessary nor sufficient for taking drugs, they exaggerate the incentive impact of drugs, thereby increasing the motivation to use drugs42. Thus in the presence of withdrawal, the capacity of bottom-up homeostatic signals to hijack control mechanisms of the reflective system is increased. Implications for treatment and directions for future research Most addicts show behavioral signs of poor decision making, but in the profiles of their physiological responses, some addicts match VMPC patients, and some do not (see above). We have suggested that addicts who match VMPC patients are characterized by insensitivity to future consequences; that is, they are oblivious to future positive or negative consequences, and instead they are guided by immediate prospects. Addicts who partially match VMPC patients are suggested to be hypersensitive to reward, so that the prospect of drugs outweighs the prospect of future consequences. These differences may have implications for prognosis, and they provide testable hypotheses that could be addressed in future research: addicts who match VMPC patients may have a harder time recovering from addiction and remaining abstinent in comparison with addicts who partially match the VMPC patients. One subgroup of addicts appeared normal and did not show behavioral or physiological signs of decision making deficits. This suggests that not every drug user has impaired decision making. We have described these addicts as ‘functional’ addicts, because a closer inspection of their everyday lives has shown that they have suffered minimal social and psychological harm as a consequence of their drug use: for

1462

example, they manage to keep their jobs2. Therefore, my view is that poor decision making in addiction is evident only when individuals persist in escalating their drug use in the face of rising adverse consequences. According to this view, people described as addicted to coffee, sweets, the internet and so on do not necessarily have impaired decision making, unless their choices bring increasing social, physical or psychological harms. However, an alternate possibility is that the lack of evidence for decision making deficits in this subgroup of addicts is a limitation of the proposed somatic marker framework, in that it does not capture all instances of addiction. Finally, one subgroup of normal controls shows behavioral and physiological profiles that matches VMPC patients. This raises the question of whether these individuals are predisposed, or at higher risk, for addiction than individuals with normal decision making capabilities. This suggestion is reasonable in light of the evidence that one predisposing factor to addiction is heredity, and genes can act in general fashion (such as the serotonin transporter gene) to predispose individuals to multiple, as opposed to specific, drug addictions44. Future research using functional imaging methods could focus on relationships between (i) genotypes related to specific neurotransmitter systems (for example, the serotonin transporter gene) (ii) the level of neural activity in specific neural circuits, and (iii) quality of choice, as shown by complex laboratory tasks of decision making. This will reveal whether genetic factors lead to suboptimal function in specific neural systems, which then leads to behaviors reflecting poor decision making. However, not all predisposing factors are necessarily genetic; other factors could be environmental (such as drug neurotoxicity), or the product of gene-environment interactions. Although the evidence for neurotoxicity resulting from drug use remains questionable45, the potential for harm remains relatively higher if drugs were abused during adolescence. Indeed, evidence suggests that the functions of the prefrontal cortex may not develop fully until the age of 21, and until such a time, the development of neural connections that underlie decision making, and the control over powerful temptations, is still taking place46–48. Therefore, exposing the prefrontal cortex to drugs before its maturity could be harmful to decision making, just like exposing the fetus to drugs during pregnancy. However, the fact remains that not every adolescent who tries drugs ends up addicted; it takes more than mere exposure to drugs to become addicted. Therefore, my hypothesis is that poor decision making in addiction is not the product of drug use; rather, poor decision making is what leads to addiction. Future systemic and longitudinal studies on decision making in young adolescents should test this hypothesis and determine whether neurocognitive development can serve as a marker predictive of addictive disorders. This research should also take into consideration models of addiction that describe a progressive dysregulation of reward brain circuitry concomitant with a spiraling path from controlled drug use to addiction49 and should examine whether drug users undergo a slow and gradual hijacking of their willpower as they move from controlled use to addiction. However, my proposal is that not every individual who tries drugs ends up on this down-spiraling path; those with poor decision making capabilities are more vulnerable, and those with normal decision making capabilities are more resistant. These are testable hypotheses with clear predictions that can be addressed in future research. ACKNOWLEDGMENTS The research described in this article was supported by the following grants from the US National Institute on Drug Abuse (NIDA): DA11779, DA12487, and DA16708. COMPETING INTERESTS STATEMENT The author declares that he has no competing financial interests.


PERSPECTIVE


Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Damasio, A.R. Descartes’ Error: Emotion, Reason, and the Human Brain (Grosset/Putnam, New York, 1994). 2. Bechara, A. Neurobiology of decision making: risk and reward. Semin. Clin. Neuropsychiatry 6, 205–216 (2001). 3. Volkow, N.D., Fowler, J.S. & Wang, G.J. The addicted human brain viewed in the light of imaging studies: brain circuits and treatment strategies. Neuropharmacology 47, 3–13 (2004). 4. Grant, S., Contoreggi, C. & London, E.D. Drug abusers show impaired performance in a laboratory test of decision making. Neuropsychologia 38, 1180–1187 (2000). 5. Rahman, S., Sahakia, B., Rudolph, N.C., Rogers, R.D. & Robbins, T.W. Decision making and neuropsychiatry. Trends Cogn. Sci. 6, 271–277 (2001). 6. Garavan, H. & Stout, J.C. Neurocognitive insights into substance abuse. Trends Cogn. Sci. 9, 195–201 (2005). 7. Ernst, M. & Paulus, M.P. Neurobiology of decision making: a selective review from a neurocognitive and clinical perspective. Biol. Psychiatry published online 10 August 2005 (doi:10.1016/j.biopsych.2005.06.004). 8. Goldstein, R.Z. & Volkow, N.D. Drug addiction and its underlying neurobiological basis: Neuroimaging evidence for the involvement of the frontal cortex. Am. J. Psychiatry 159, 1642–1652 (2002). 9. Bechara, A. & Damasio, A.R. The somatic marker hypothesis: a neural theory of economic decision. Games Econ. Behav. 52, 336–372 (2005). 10. Büchel, C., Morris, J., Dolan, R.J. & Friston, K.J. Brain systems mediating aversive conditioning: an event-related fMRI study. Neuron 20, 947–957 (1998). 11. Bechara, A. Disturbances of emotion regulation after focal brain lesions. Int. Rev. Neurobiol. 62, 159–193 (2004). 12. Montague, P.R. & Berns, G.S. Neural economics and the biological substrates of valuation. Neuron 36, 265–284 (2002). 13. Everitt, B.J. et al. Associative processes in addiction and reward: the role of amygdala and ventral striatal subsystems: the role of amygdala-ventral striatal subsystems. Ann. NY Acad. Sci. 877, 412–438 (1999). 14. Clark, L., Cools, R. & Robbins, T. The neuropsychology of ventral prefrontal cortex: decision making and reversal learning. Brain Cogn. 55, 41–53 (2004). 15. Bechara, A. & Martin, E. Impaired decision making related to working memory deficits in substance addicts. Neuropsychology 18, 152–162 (2004). 16. Martin, E.M. et al. Delayed nonmatch-to-sample performance in HIV-seropositive and HIV-seronegative polydrug abusers. Neuropsychology 17, 283–288 (2003). 17. Franklin, T.R. et al. Decreased gray matter concentration in the insular, orbitofrontal, cingulate, and temporal cortices of cocaine patients. Biol. Psychiatry 51, 134–142 (2002). 18. Matochik, J.A., London, E.D., Eldreth, D.A., Cadet, J.L. & Bolla, K.I. Frontal cortical tissue composition in abstinent cocaine abusers: a magnetic resonance imaging study. Neuroimage 19, 1095–1102 (2003). 19. Bartzokis, G. et al. The incidence of T2-weighted MR imaging signal abnormalities in the brain of cocaine-dependent patients is age-related and region-specific. AJNR Am. J. Neuroradiol. 20, 1628–1635 (1999). 20. Lim, K.O., Choi, S.J., Pomara, N., Wolkin, A. & Rotorsen, J.P. Reduced frontal white matter integrity in cocaine dependence: a controlled diffusion tensor imaging study. Biol. Psychiatry 51, 890–895 (2002). 21. Patton, J.H., Stanford, M.S. & Barratt, E.S. Factor structure of the Barratt impulsiveness scale. J. Clin. Psychol. 51, 768–774 (1995). 22. Whiteside, S. & Lynam, D. The five factor model and impulsivity: using a structural model of personality to understand impulsivity. Pers. Individ. Dif. 30, 669–689 (2001). 23. Monterosso, J., Ehrman, R., Napier, K., O’Brien, C.P. & Childress, A.R. Three decision making tasks in cocaine-dependent patients: Do they measure the same construct? Addiction 96, 1825–1837 (2001).


24. Martin, E.M. & Bechara, A. Decision making and drug of choice in substance-dependent individuals: a preliminary report. Biol. Psychiatry 53, 97S (2003). 25. Bechara, A. Risky business: Emotion, decision making and addiction. J. Gambl. Stud. 19, 23–51 (2003). 26. Fishbein, D. et al. Cognitive performance and autonomic reactivity in abstinent drug abusers and nonusers. Exp. Clin. Psychopharmacol. 13, 25–40 (2005). 27. Friedman, N.P. & Miyake, A. The relations among inhibition and interference control functions: A latent variable analysis. J. Exp. Psychol. Gen. 133, 101–135 (2004). 28. Aron, A.R., Robbins, T.W. & Poldrack, R.A. Inhibition and the right inferior frontal cortex. Trends Cogn. Sci. 8, 170–177 (2004). 29. Hyman, S.E. Addiction: A disease of learning and memory. Am. J. Psychiatry 162, 1414–1422 (2005). 30. Nestler, E.J. Molecular basis of long-term plasticity underlying addiction. Nat. Rev. Neurosci. 2, 119–128 (2001). 31. Clarke, H.F., Dalley, J.W., Crofts, H.S., Robbins, T.W. & Roberts, A.C. Cognitive inflexibility after prefrontal serotonin depletion. Science 304, 878–880 (2004). 32. Rogers, R.D., Lancaster, M., Wakeley, J. & Bhagwagar, Z. Effects of beta-adrenoreceptor blockade on components of human decision making. Psychopharmacology (Berl.) 172, 157–164 (2004). 33. McClure, S.M., Laibson, D.I., Loewenstein, G. & Cohen, J.D. Separate neural systems value immediate and delayed monetary rewards. Science 306, 503–507 (2004). 34. Aguilar de Arcos, F., Verdejo, A., Peralta, M.I., Sanchez-Barrera, M. & Perez-Garcia, M. Experience of emotions in substance abusers exposed to images containing neutral, positive, and negative affective stimuli. Drug Alcohol Depend. 78, 159–167 (2004). 35. London, E.D., Ernst, M., Grant, S., Bonson, K. & Weinstein, A. Orbitofrontal cortex and human drug abuse: functional imaging. Cereb. Cortex 10, 334–342 (2000). 36. Childress, A.R. et al. Limbic activation during cue-induced cocaine craving. Am. J. Psychiatry 156, 11–18 (1999). 37. Breiter, H.C., Aharon, I., Kahneman, D., Dale, A. & Shizgal, P. Functional imaging of neural responses to expectancy and experience of monetary gains and losses. Neuron 30, 619–639 (2001). 38. Noel, X., Van Der Linden, M., Verbanck, P., Pelc, I. & Bechara, A. Deficits of inhibitory control and of shifting associated with cognitive bias in polysubstance abusers with alcoholism. Addiction 100, 1302–1309 (2005). 39. Semendeferi, K., Armstrong, E., Schleicher, A., Zilles, K. & Van Hoesen, G.W. Prefrontal cortex in humans and apes: A comparative study of area 10. Am. J. Phys. Anthropol. 114, 224–241 (2001). 40. Ongur, D. & Price, J.L. The organization of networks within the orbital and medial prefrontal cortex of rats, monkeys and humans. Cereb. Cortex 10, 206–219 (2000). 41. Kringelbach, M.L. & Rolls, E.T. The functional neuroanatomy of the human orbitofrontal cortex: Evidence from neuroimaging and neuropsychology. Prog. Neurobiol. 72, 341–372 (2004). 42. Wise, R.A. & Bozarth, M.A. A psychomotor stimulant theory of addiction. Psychol. Rev. 94, 469–492 (1987). 43. Nader, K., Bechara, A. & van der Kooy, D. Neurobiological constraints on behavioral models of motivation. Annu. Rev. Psychol. 48, 85–114 (1997). 44. Goldman, D. & Bergen, A. General and specific inheritance of substance abuse and alcoholism. Arch. Gen. Psychiatry 55, 964–965 (1998). 45. Ricaurte, G.A., Yuan, J., Hatzidimotriou, G., Cord, B.J. & McCann, U.D. Retraction. Science 301, 1479 (2003). 46. Eslinger, P.J. Conceptualizing, describing, and measuring components of executive function. in Attention, Memory and Executive Function (eds. Lyon, G.R. & Krasnegor, N.A.) 420–441 (Paul H. Brooks, Baltimore, 1999). 47. Crone, E.A., Jennings, J.R. & Van der Molen, M.W. Developmental change in feedback processing as reflected by phasic heart rate changes. Dev. Psychol. 40, 1228–1238 (2004). 48. Overman, W.H. et al. Performance on the IOWA card task by adolescents and adults. Neuropsychologia 42, 1838–1851 (2004). 49. Koob, G.F. & Le Moal, M. Drug abuse: Hedonic homeostatic dysregulation. Science 278, 52–58 (1997).

1463



REVIEW

Nicotine addiction and comorbidity with alcohol abuse and mental illness John A Dani and R Adron Harris The World Health Organization estimates that one-third of the global adult population smokes. Because tobacco use is on the rise in developing countries, death resulting from tobacco use continues to rise. Nicotine, the main addictive component of tobacco, initiates synaptic and cellular changes that underlie the motivational and behavioral alterations that culminate in addiction. Nicotine addiction progresses rapidly in adolescents and is most highly expressed in vulnerable people who have psychiatric illness or other substance abuse problems.

Tobacco use is the leading cause of preventable death in developed countries1. Smoking commonly begins during adolescence, and about half of those who do not quit eventually die from smokingrelated diseases2. In the United States alone, smoking annually causes over 400,000 deaths and $50 billion in medical costs3. The addictive power of tobacco is exemplified by the difficulty in quitting4–6. Most smokers wish to quit and try repeatedly. About one-third of smokers attempt to quit each year, but fewer than 10% succeed. Despite imperative medical reasons, 50% of heart attack survivors and of those hospitalized for other serious smoking-related illness relapse to cigarettes within weeks of leaving the hospital. Of the roughly 3,000 ingredients in cigarette smoke, nicotine is the main addictive component that motivates continued tobacco use despite its harmful effects4,6–10. Nicotine is addictive in the absence of tobacco, and it supports self-administration, enhances reward from brain stimulation and reinforces preference for the place where nicotine is administered (place preference). It also produces a withdrawal syndrome that is relieved by nicotine replacement6–8,10,11. Tobacco use is most highly prevalent and is more intense in psychiatric patients and drug abusers12,13. The comorbidity with mental illness is particularly high for schizophrenia and depression. These individuals may be more susceptible to nicotine addiction because tobacco provides desired positive mood influences14. Furthermore, they often experience more severe withdrawal symptoms, making it more difficult to quit. A great majority of those who abuse other substances also smoke, and there is a particularly strong correlation between smoking and abuse of the other most commonly abused drug, alcohol. More severely dependent drinkers smoke more and

John A. Dani is in the Department of Neuroscience, Menninger Department of Psychiatry & Behavioral Sciences, Baylor College of Medicine, Houston, Texas 77030, USA, and R. Adron Harris is at the Waggoner Center for Alcohol and Addiction Research, University of Texas, Austin, Austin, Texas 78712, USA. e-mail: [email protected] Published online 26 October 2005; doi:10.1038/nn1580


are less likely to quit. Thus, particularly vulnerable groups within the overall population consume a disproportionately high fraction of all the cigarettes that are smoked12. Smoking begins in adolescence The vast majority of those who initiate tobacco use are young. In the US, more than 60% of young people try smoking, and about one-third to half of them become daily smokers15. In those who initiate smoking, cigarette consumption escalates over a couple of years, but the addiction process proceeds quickly in adolescents5,16. Nearly one-quarter of the adolescents report symptoms of addiction at about the time they establish a routine of smoking on a monthly basis. In young rats, synaptic changes and neuroadaptations to nicotine occur after only one exposure17,18. In addition, adolescent rats show hypersensitivity to the reinforcing actions of nicotine, as demonstrated by intravenous self-administration and conditioned place preference19,20. Tobacco can have positive effects on behavior and mood, but the first exposure to smoking often highlights the aversive impact21–23. Adolescents, however, report fewer aversive effects and more positive effects than adults after their first smoking episode16. Furthermore, cigarettes are an ideal drug delivery system. Smokers adjust their dose precisely to avoid discomfort while achieving the most desirable impact5,6. Once addicted, smokers report pleasure, arousal, relaxation, improved attention, reduced anxiety, relief from stress, relief from hunger and eventually relief from withdrawal symptoms5. Nicotine is a mood leveler in humans and other animals, causing arousal during fatigue and relaxation during anxiety. Smoking is a learned (conditioned) behavior reinforced by nicotine. Cigarettes are excellent vehicles for the conditioning because the dosing via puffs is precise and repeated very often5,6. Furthermore, the drug-taking behavior is associated with common events of the day, such as waking in the morning. The behavioral conditioning occurs more frequently, and is associated with more common everyday events, for cigarettes than for any other addictive drug. Therefore, the associations that become cues for smoking are almost unavoidable parts of smokers’ lives. In summary, adolescents experience aspects of dependence after only a few cigarettes, and nicotine exposure in adolescent rats increases selfadministration tested later in life24. The conditioned association of the

1465

REVIEW a NAc

β2*

α4β2* DA α7*

b NAc

–

α4β2* GABA

β2* α4β2*

VTA

and memory9,17,29. Thus, the functions of DA within the overall mesocorticolimbic system involve the learning and integration of salient environmental information. Subsequently, that information is used in the preparation, initiation and execution of behaviors that serve a beneficial goal8,9,29,30.

Nicotine later Some desensitization

DA

–

α4β2* GABA

α7*

Nicotine influences synapses of the VTA Nicotine obtained from tobacco reaches the brain in 10–60 seconds, and is initially at a LDT/PPT LDT/PPT concentration of roughly 100–500 nM in the Glu ACh ACh Glu arterial blood, lung and brain5–7,10,31. The distribution half-life of approximately 8 minutes dictates the initial actions of nicotine. The elimination half-life of about 2 hours allows Figure 1 A simplified illustration of several synaptic connections and nicotine-induced events that control nicotine to accumulate with ongoing smokDA release in the NAc. (a) Initially, nicotine causes some activation of most nAChR subtypes. ing and persist for hours. Thus, smokers often The active presynaptic α7* nAChRs enhance glutamatergic excitatory drive (upward arrow), whereas deliver a small pulse of nicotine each time they active α4β2* nAChRs directly excite DA neurons (upward arrow). This coincidence of presynaptic smoke, and nicotine accumulates and lingers glutamate release and postsynaptic firing increases the likelihood of synaptic potentiation (for example, LTP). The cholinergic input from the LDT/PPT (laterodorsal tegmentum and pedunculopontine in the body (and brain) as the day progresses. tegmentum) provides excitatory drive onto GABAergic interneurons. (b) The more prolonged presence Upon smoking, nicotine initially activates of nicotine causes some desensitization (indicated by nAChRs in white text), particularly of subtypes nAChRs throughout the brain, including containing the β2 subunit. As a consequence, direct nicotine excitation of VTA DA neurons ceases, and those on VTA DA neurons (Fig. 1a). NicotineGABA interneurons decrease their inhibition (downward arrow) onto VTA DA neurons. Although the nAChR induced activity of nAChRs produces a direct subtypes are not perfectly segregated as shown, they are the main subtypes mediating nicotinic influence depolarization of the DA neurons, causing at the given locations, as shown in rodent studies. an increase in burst firing and overall firing rate10,29,32–34. The nAChRs on the VTA DA neurons are mainly composed of α4 (ref. addictive drug with common daily events motivates progression along 35) and β2 (refs. 36,37) subunits, in combination with other nAChR subunits38. After a few minutes, the high-affinity α4β2-containing the path to daily cigarette use and spurs relapse during abstinence. (α4β2*) receptors, in particular, desensitize (Fig. 1b), which decreases Mesocorticolimbic dopamine system or terminates the direct stimulation of the DA neurons by nicotine32,39. Nicotine binds selectively to nicotinic acetylcholine receptors (nAChRs), Microdialysis studies in rats show, however, that a single injection of which are ligand-gated cationic channels that normally bind acetylcho- nicotine elevates DA in the NAc for hours8,40. Synaptic changes in the line25,26. Neuronal nAChRs are pentameric, containing combinations of circuitry that controls the firing of VTA DA neurons produce the proα and β subunits or exclusively α subunits, and in the mammalian brain longed DA signal in the NAc. The VTA receives massive convergent afferent inputs, including only α7 subunits commonly form homo-oligomeric nAChRs. Because nAChRs are widely distributed, nicotine influences cellular events and glutamatergic projections from the prefrontal cortex and GABAergic produces neuroadaptations in many brain areas that are directly or projections from the NAc and ventral pallidum10,41,42. Another major indirectly important during the addiction process23,27. source of innervation into the midbrain DA areas arises from the nearby Although many areas of the brain participate, the mesocorticolimbic pedunculopontine tegmentum (PPT) and the laterodorsal tegmentum dopamine (DA) system has a vital role in the acquisition of behaviors (LDT), which are a loose collection of cholinergic neurons interspersed that are inappropriately reinforced by psychostimulant drugs, includ- with GABAergic and glutamatergic neurons42. The PPT projects mainly ing nicotine4,6–8,10,28. An important dopaminergic pathway originates to the substantia nigra compacta, and the LDT projects mainly to the in the ventral tegmental area (VTA) of the midbrain and projects to VTA. The PPT and LDT contribute to events associated with drug the prefrontal cortex as well as limbic and striatal structures, including taking27, as shown by the observation that lesions in the PPT reduce the nucleus accumbens (NAc). A wide range of evidence supports the nicotine self-administration43. Although the VTA receives a strong role of the mesocorticolimbic DA system in nicotine addiction4,7,8. For excitatory glutamate input from the prefrontal cortex, that excitation example, blocking DA release in the NAc with antagonists or lesions is mainly onto DA neurons that project back to the cortex, not to the attenuates the rewarding effects of nicotine, as indicated by reduced NAc41. Rather, the PPT and LDT provide the main glutamatergic excitaself-administration11,28. That result is consistent with the general find- tion to the DA neurons projecting to the NAc (Fig. 1)42. ing that addictive drugs (such as cocaine, heroin and amphetamine) Glutamatergic afferents onto DA neurons commonly have presynaptic elevate DA in the NAc8,9. nAChRs composed of α7 subunits10,18,29,40. Because α7-containing Recent, more sophisticated theories explain the mounting data that (α7*) nAChRs have a relatively low affinity for nicotine, the low concencontradict the simplest notions about the rewarding properties of trations of nicotine achieved by smokers do not strongly desensitize the DA8,9,29,30. Clearly, DA concentrations in the NAc are not a direct indi- α7* nAChRs (Fig. 1)32,39. Thus, the activity of presynaptic α7* nAChRs cation of reward. Rather, DA may participate in the ongoing associa- enhances glutamatergic afferent excitation onto DA neurons while tive learning of adaptive behaviors as an animal continually updates nicotine concentrations are elevated10,18. The enhanced presynaptic a construct of environmental saliency. The hypothesis suggests that glutamate release is paired initially with the increased firing of the postaddictive drugs act upon mechanisms that normally underlie learning synaptic DA neurons caused by α4β2* nAChRs before they desensitize. Ann Thomson


VTA

Nicotine initially nAChR activation

1466



REVIEW The combination of enhanced presynaptic drive and strong postsynaptic response favors the production of long-term synaptic potentiation (LTP) of the glutamatergic afferents (Fig. 1a)44. Even though nicotine directly activates α4β2* nAChRs on the DA neurons for only a short time before those nAChRs desensitize, that brief enhanced depolarization of the postsynaptic DA neurons is sufficient to produce LTP when paired with the boosted glutamate release caused by presynaptic α7* nAChRs. While those nicotine-induced mechanisms enhance glutamatergic excitation of DA neurons, related mechanisms decrease the inhibition from GABAergic interneurons in the midbrain. Although a small subset of DA neurons receive cholinergic inputs from the PPT and LDT, there is greater cholinergic innervation and endogenous excitatory drive of GABAergic interneurons in the VTA (Fig. 1)45. That endogenous cholinergic activity has a significant excitatory influence over the firing of inhibitory VTA GABAergic interneurons through the α4β2* nAChRs10,40,46. Although other minority subtypes are present, α4β2* nAChRs make up the majority of the nAChR subtypes on midbrain GABAergic interneurons10,18,29,38–40,46. As is the case for the DA neurons, nicotine desensitizes the α4β2* nAChRs on the GABAergic interneurons in a matter of minutes (Fig. 1b). The inhibitory GABAergic activity declines rapidly because nAChR desensitization removes the direct excitation caused by nicotine and decreases the endogenous cholinergic drive onto the GABAergic interneurons arising from the PPT and LDT40. In summary, smokers deliver a small pulse of nicotine with each episode of smoking, and nicotine accumulates as the day progresses. That situation initially causes some activation of most nAChR subtypes, but then the prolonged low levels of nicotine favor significant desensitization of most non-α7 nAChR subtypes (such as α4β2*). As a result of these pharmacodynamics, nicotine initiates cellular and synaptic events in the VTA that enhance excitation and decrease inhibition to the DA neurons. As a consequence, DA neurons fire more frequently34, and the concentration of DA is elevated in the NAc for a prolonged time8,40. Although most research has focused on the midbrain DA centers, nicotinic mechanisms are also important in the target areas of the DA projections. The striatum is richly innervated throughout by cholinergic interneurons, and this cholinergic activity regulates DA release47–50, acting mainly through presynaptic non-α7 nAChR subtypes on DA terminals. When nicotine is applied in vivo, it desensitizes nAChRs on DA terminals (Fig 1b). By itself, this desensitization would decrease DA release—particularly release evoked by low-frequency action potentials (that is, tonic single-afferent pulses along the DA fibers)48–50. However, by acting on the midbrain source of DA, nicotine causes DA neurons to fire more bursts of action potentials32–34. Nicotine also acts at the fibers and terminals in the target neuron to alter DA signaling so as to favor DA release in response to phasic bursts while simultaneously depressing release in response to tonic, single action potentials. In that way, nicotine boosts DA concentrations in the NAc. Acting in the target region, nicotine alters the relationship between afferent activity along DA fibers and DA release, thereby altering DA signaling. Nicotine also acts in the target region to alter intrinsic GABAergic feedback mechanisms, thus modulating information processing along reward pathways51,52. As research progresses, it is likely that such nicotinic mechanisms in the target areas will be better appreciated as important contributors to nicotine addiction. Additional influences on reward, withdrawal and relapse Although there is strong support for a role of DA and the overall mesocorticolimbic system in reinforcing nicotine use, evidence also indicates roles for other neurotransmitters and peptides. A fundamental role of


nAChRs in the brain is the presynaptic enhancement of neurotransmitter release25,26,29. Enhanced release has been seen for GABA, glutamate and dopamine and also for other neurotransmitters that influence mood and emotional balance, such as serotonin, norepinephrine and endogenous opioids. Associated with opioid influences, chronic nicotine use upregulates µ-opioid receptors and alters the transcription factor CREB (cAMP response element binding protein)53,54. Addictive drugs commonly alter CREB activity, which in turn influences the longterm behaviors associated with addiction55 (also see the perspective by Nestler56 in this issue). Likewise, those events are important for the reinforcing influences of nicotine53,54. Environmental cues that are associated with the reinforcing properties of nicotine regulate CREB54, and thus cues linked to smoking become conditioned stimuli that initiate molecular events contributing to the craving and relapse of abstinent smokers. Environmental cues linked to withdrawal also may stimulate relapse. Withdrawal from nicotine decreases the sensitivity of reward systems, as detected by elevated thresholds for intracranial self-stimulation in rats57,58. Nicotine elevates glutamate levels, and group II metabotropic glutamate (mGluII) receptors serve in a negative feedback capacity for homeostatic regulation of glutamate. It is hypothesized that among the neuroadaptations induced by chronic nicotine, altered mGluII receptor function decreases glutamate levels and contributes to the discomfort of withdrawal57. Stimuli repeatedly paired with such withdrawal discomfort come to elevate reward thresholds on their own58. Thus, the deficits in reward pathways normally caused by nicotine withdrawal eventually arise from conditioned stimuli that then cue smoking to relieve the symptoms. The β4 nAChR subunit is likely to have a role in withdrawal because mice lacking β4 show much milder symptoms when nicotine withdrawal is induced59. Often the withdrawal symptoms and the ‘priming’ cues arising from internal states are more severe for smokers with psychiatric illness, making abstinence more difficult. Comorbidity with mental illness Nicotine dependence is much more prevalent among psychiatric patients than in the general population12,13. Most notable are schizophrenic patients, who have smoking rates of 70% to 90% compared to about 25% for the general population. In a US study, patients with mood, anxiety or personality disorders show nicotine dependence twice as commonly as the general population12. Remarkably, 7% of the overall population—those who have a psychiatric disorder and are nicotine dependent—consume 34% of all cigarettes. Adolescent smoking is particularly important because early tobacco use is associated with higher risk of later psychiatric problems and, conversely, early behavioral problems are linked to a greater risk of later tobacco use. For example, the prevalence of psychiatric disorders is about 70% in adolescents who are daily smokers60. Although attention-deficit/ hyperactivity disorder (ADHD) does not increase smoking prevalence in all studies, children with ADHD often initiate smoking earlier and have more trouble quitting13,61. In addition, depression and anxiety are associated with higher risk for smoking initiation and for transition to daily smoking62. An evaluation of female twins suggested that the relationship between lifetime smoking and major depression arises largely from familial factors (likely genetic) that predispose individuals to both smoking and depression63. It is commonly argued that psychiatric patients use tobacco for selfmedication. That hypothesis applies most readily to schizophrenia. Nicotine normalizes several deficits in sensory processing associated with schizophrenia, and nAChRs influence those sensory events13,64. Although nicotine seems to improve attention in schizophrenia, it does not improve most symptoms65. However, there is other intrigu-

1467


REVIEW ing evidence linking nAChRs and schizophrenia. The chromosomal region containing the nAChR α7 subunit is linked to genetic risk for schizophrenia, and α7 expression is reduced in schizophrenics13,64. The self-medication hypothesis also may have some validity among ADHD patients. Nicotine increases the release of DA and improves attention, and similar effects are produced by stimulant drugs used to treat ADHD. Cigarettes also may provide a medicating influence because an unknown ingredient inhibits brain monoamine oxidase, and monoamine oxidase inhibitors have antidepressant actions66,67. Another factor that may exacerbate nicotine addiction arises from psychiatric medications. For example, antipsychotic drugs block DA receptors, and nicotine can overcome this action by enhancing DA release and reducing unwanted side effects. Indeed, schizophrenics who smoke have a lower incidence of neuroleptic-induced parkinsonism, and haloperidol increases smoking in schizophrenics21. This influence is only part of the motivation for smoking, however, because even firstepisode patients who have not received antipsychotic drugs have a high incidence of smoking68. The links between stress, depression, anxiety and tobacco use also are contributing factors for comorbidity. Depression sensitizes smokers to the influences of stress, which increases the motivation and vulnerability for drug use67. During the acquisition phase, stress increases the sensitivity to addictive drugs, making the individual more susceptible to drug reward. During abstinence, stress can stimulate reinstatement of drug seeking and drug self-administration. Craving and relapse can be elicited by the drug itself or by environmental cues that become salient through their repeated association with previous use. Just as environmental events associated with nicotine withdrawal serve as priming cues for drug seeking58, the stress response mimics a motivating internal state69. Depression and anxiety often accompany nicotine withdrawal, particularly for abstinent smokers with psychiatric illness, and relief from specific aspects of those symptoms motivates relapse. Thus, smokers become conditioned to expect nicotine to provide partial relief from stress and depression as it does from the symptoms of withdrawal67. This hypothesis has mechanistic support because stress produces synaptic plasticity in the VTA similar to that produced by nicotine17,70. Furthermore, anhedonia often accompanies mental illnesses such as schizophrenia and depression. By boosting DA release, nicotine may ameliorate particularly the anhedonic aspects of the illness. It is reasonable to conclude that common underlying mechanisms influencing motivation and behavior contribute to the high comorbidity between forms of mental illness and nicotine addiction. Comorbidity with alcohol Nicotine and alcohol seem to share few pharmacological similarities. Nicotine has specific receptors, promotes alertness and is proconvulsant, whereas alcohol affects multiple receptor types, diminishes alertness and is anticonvulsant. Although both drugs produce tolerance and dependence, the characteristics of the withdrawal syndromes also differ markedly71. Despite these clear distinctions, there are remarkable commonalities between the two drugs, including their legal status and wide use. The prevalence of nicotine dependence is very high among alcohol abusers12. In addition, the amount of tobacco smoked is positively correlated with the amount of alcohol consumed and the severity of alcohol dependence. Although a past history of alcohol dependence does not influence the subjective effects of nicotine, it does render nicotine more reinforcing than for those who have never been alcohol dependent72. There also may be shared genetic influences over the development of nicotine and alcohol dependence (also see Crabbe and Lovinger73, in this issue). Twin studies indicate a substantial genetic

correlation between nicotine and alcohol dependence, with only a modest environmental contribution74. A potential mechanistic link arises from the observation that smokers report less intoxication from the same amounts of alcohol than either nonsmokers or former smokers and that result is not due to differences in alcohol metabolism75. Low responsiveness to alcohol is a risk factor for the development of alcohol dependence76, and chronic nicotine use may decrease the effects of alcohol, thereby increasing alcohol consumption and dependence. Nicotine and alcohol also both relieve pain, and that action requires the GIRK2 potassium channel77. Although nicotine does not act directly on GIRK channels, the overall action demonstrates a convergence of drug actions on a common effector protein. Further convergent action is indicated by animal studies of motor activity and body temperature that show cross-tolerance for nicotine and alcohol78. Emerging results from human brain imaging suggest that neuroadaptations and neurotoxicity produced by chronic alcohol abuse may also be altered by smoking. When alcoholics stop drinking, they show a time-dependent increase in the number of neuronal GABAA receptors. There also is evidence that increased GABAA receptors result in more severe withdrawal symptoms. Smoking reduces upregulation of GABAA receptors and may reduce the severity of alcohol withdrawal79. It is possible that nicotine boosts DA signaling that is diminished during alcohol withdrawal. If alcohol enhances GABA inhibition of the mesolimbic DA neurons, chronic nicotine may counter the upregulation of GABAA receptors as well as acting acutely to decrease GABAergic inhibition by desensitizing nAChR subtypes that help drive GABA interneurons (Fig. 1b). The link between nicotine and alcohol use is of particular importance for adolescents. Initiation of smoking at an early age is a risk factor for the development of alcohol dependence and other substance-abuse disorders80. A large Finnish study gathered data on 14-year-olds and followed them until age 32. Regular smoking at age 14 was the most powerful predictor of drunk driving offenses at age 32 (ref. 81). A key issue is whether exposure to nicotine during development increases alcohol use and dependence. There could be psychosocial influences as well as a pharmacological role for smoking in later substance abuse. Animal studies show that chronic nicotine administration does increase alcohol self-administration, supporting a possible causal link between smoking and alcohol reinforcement82. Although unproven, it is possible that preventing or delaying initiation of nicotine use in adolescence would reduce development of alcohol abuse later in life. Commonalities between nicotine and alcohol also occur at the molecular and genetic level. Genetically selected lines of mice and rats show different specific behavioral responses to ethanol, and some of those animals also have differing reactions to nicotine. Rat and mouse lines sensitive to the hypnotic actions of alcohol show a different sensitivity to the locomotor effects of nicotine83,84, suggesting common genetic determinants for those actions. One genetic factor influencing nicotine and alcohol modulation of acoustic startle in mice is a polymorphism in the α4 nAChR subunit85. The presence of alanine or threonine at position 529 of the α4 subunit influences the effects of nicotine and alcohol on the response to acoustic startle and to the severity of an alcohol withdrawal sign in mice86. Studies with mice lacking the β2 subunit have shown that those actions of nicotine and alcohol are mediated by α4β2* nAChRs85. These studies provide hints of possible molecular interactions between nicotine and alcohol. Indeed, ethanol alters the function of neuronal nAChRs, possibly by binding to a site analogous (but not identical) to the alcohol site proposed for GABAA and glycine receptors87. Even though the A529T polymorphism is in the intracellular loop of the α4 subunit and distant from the proposed alcohol binding site, it is possible that the polymorphism influences

1468



REVIEW the alcohol sensitivity of the α4* nAChR88. Although the mechanistic details are not yet conclusively known, there are interactions between nicotine and alcohol at the α4β2* nAChR. Other subtypes also are influenced, and additional complexities are likely to contribute to the commonalities between the two addictive drugs85,89. The substantial interactions between nicotine and alcohol raise the possibility of pharmacotherapies that could either simultaneously treat nicotine and alcohol dependence or treat alcohol dependence by blocking nAChRs that may contribute to alcohol consumption. The nonselective nAChR antagonist mecamylamine reduces the reinforcing actions of alcohol in humans90, but this treatment is not practical because mecamylamine produces autonomic side effects. Several drugs that do not directly affect nAChRs show some promise: cannabinoid CB1 receptor antagonists and the anticonvulsant drug topiramate may be useful in treating both nicotine and alcohol dependence91,92. The common comorbidity of tobacco use with mental illness or drug dependence suggests that a more complete understanding of nicotine addiction will have a broad impact on society. The mechanisms underlying nicotine addition may indicate common modes of treatment and prevention for particularly vulnerable members of the population. ACKNOWLEDGMENTS The authors are supported by the National Institute on Alcoholism and Alcohol Abuse, the National Institute on Drug Addiction and the National Institute of Neurological Disorders and Stroke. We thank D. Balfour, C. Borghese, A. Collins, M. De Biasi, L. O’Dell and the members of the Dani laboratory for comments. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Peto, R., Lopez, A.D., Boreham, J., Thun, M. & Heath, C., Jr. Mortality from tobacco in developed countries: indirect estimation from national vital statistics. Lancet 339, 1268–1278 (1992). 2. Tobacco or Health: A Global Status Report (World Health Organization, Geneva, 1997). 3. Epping-Jordan, M.P., Watkins, S.S., Koob, G.F. & Markou, A. Dramatic decreases in brain reward function during nicotine withdrawal. Nature 393, 76–79 (1998). 4. Balfour, D.J. The neurobiology of tobacco dependence: a preclinical perspective on the role of the dopamine projections to the nucleus. Nicotine Tob. Res. 6, 899–912 (2004). 5. Benowitz, N.L. Nicotine addiction. Prim. Care 26, 611–631 (1999). 6. Karan, L., Dani, J.D. & Benowitz, N. The pharmacology of nicotine and tobacco. in Principles of Addiction Medicine 3rd Ed., 225–248 (American Society of Addiction Medicine, Chevy Chase, Maryland, 2003). 7. Dani, J.A. & Heinemann, S. Molecular and cellular aspects of nicotine abuse. Neuron 16, 905–908 (1996). 8. Di Chiara, G. Role of dopamine in the behavioural actions of nicotine related to addiction. Eur. J. Pharmacol. 393, 295–314 (2000). 9. Di Chiara, G. et al. Dopamine and drug addiction: the nucleus accumbens shell connection. Neuropharmacology 47 (Suppl.) 227–241 (2004). 10. Mansvelder, H.D. & McGehee, D.S. Cellular and synaptic mechanisms of nicotine addiction. J. Neurobiol. 53, 606–617 (2002). 11. Stolerman, I.P. & Shoaib, M. The neurobiology of tobacco addiction. Trends Pharmacol. Sci. 12, 467–473 (1991). 12. Grant, B.F., Hasin, D.S., Chou, S.P., Stinson, F.S. & Dawson, D.A. Nicotine dependence and psychiatric disorders in the United States: results from the national epidemiologic survey on alcohol and related conditions. Arch. Gen. Psychiatry 61, 1107–1115 (2004). 13. Leonard, S. et al. Smoking and mental illness. Pharmacol. Biochem. Behav. 70, 561–570 (2001). 14. Quattrocki, E., Baird, A. & Yurgelun-Todd, D. Biological aspects of the link between smoking and depression. Harv. Rev. Psychiatry 8, 99–110 (2000). 15. Henningfield, J.E., Moolchan, E.T. & Zeller, M. Regulatory strategies to reduce tobacco addiction in youth. Tob. Control 12 (Suppl.) i14–i24 (2003). 16. DiFranza, J.R. et al. Initial symptoms of nicotine dependence in adolescents. Tob. Control 9, 313–319 (2000). 17. Jones, S. & Bonci, A. Synaptic plasticity and drug addiction. Curr. Opin. Pharmacol. 5, 20–25 (2005). 18. Mansvelder, H.D. & McGehee, D.S. Long-term potentiation of excitatory inputs to brain reward areas by nicotine. Neuron 27, 349–357 (2000).


19. Levin, E.D., Rezvani, A.H., Montoya, D., Rose, J.E. & Swartzwelder, H.S. Adolescentonset nicotine self-administration modeled in female rats. Psychopharmacology (Berl.) 169, 141–149 (2003). 20. Vastola, B.J., Douglas, L.A., Varlinskaya, E.I. & Spear, L.P. Nicotine-induced conditioned place preference in adolescent and adult rats. Physiol. Behav. 77, 107–114 (2002). 21. Laviolette, S.R. & van der Kooy, D. Blockade of mesolimbic dopamine transmission dramatically increases sensitivity to the rewarding effects of nicotine in the ventral tegmental area. Mol. Psychiatry 8, 50–59 (2003). 22. Laviolette, S.R. & van der Kooy, D. The neurobiology of nicotine addiction: bridging the gap from molecules to behaviour. Nat. Rev. Neurosci. 5, 55–65 (2004). 23. Wonnacott, S., Sidhpura, N. & Balfour, D.J. Nicotine: from molecular mechanisms to behaviour. Curr. Opin. Pharmacol. 5, 53–59 (2005). 24. Adriani, W. et al. Evidence for enhanced neurobehavioral vulnerability to nicotine during periadolescence in rats. J. Neurosci. 23, 4712–4716 (2003). 25. Jones, S., Sudweeks, S. & Yakel, J.L. Nicotinic receptors in the brain: correlating physiology with function. Trends Neurosci. 22, 555–561 (1999). 26. Role, L.W. & Berg, D.K. Nicotinic receptors in the development and modulation of CNS synapses. Neuron 16, 1077–1085 (1996). 27. Picciotto, M.R. & Corrigall, W.A. Neuronal systems underlying behaviors related to nicotine addiction: neural circuits and molecular genetics. J. Neurosci. 22, 3338– 3341 (2002). 28. Corrigall, W.A. Nicotine self-administration in animals as a dependence model. Nicotine Tob. Res. 1, 11–20 (1999). 29. Dani, J.A., Ji, D. & Zhou, F.M. Synaptic plasticity and nicotine addiction. Neuron 31, 349–352 (2001). 30. Schultz, W., Dayan, P. & Montague, P.R. A neural substrate of prediction and reward. Science 275, 1593–1599 (1997). 31. Rose, J.E., Behm, F.M., Westman, E.C. & Coleman, R.E. Arterial nicotine kinetics during cigarette smoking and intravenous nicotine administration: implications for addiction. Drug Alcohol Depend. 56, 99–107 (1999). 32. Pidoplichko, V.I., DeBiasi, M., Williams, J.T. & Dani, J.A. Nicotine activates and desensitizes midbrain dopamine neurons. Nature 390, 401–404 (1997). 33. Schilstrom, B., Nomikos, G.G., Nisell, M., Hertel, P. & Svensson, T.H. N-methyl-Daspartate receptor antagonism in the ventral tegmental area diminishes the systemic nicotine-induced dopamine release in the nucleus accumbens. Neuroscience 82, 781–789 (1998). 34. Schilstrom, B., Rawal, N., Mameli-Engvall, M., Nomikos, G.G. & Svensson, T.H. Dual effects of nicotine on dopamine neurons mediated by different nicotinic receptor subtypes. Int. J. Neuropsychopharmacol. 6, 1–11 (2003). 35. Tapper, A.R. et al. Nicotine activation of alpha4* receptors: sufficient for reward, tolerance, and sensitization. Science 306, 1029–1032 (2004). 36. Picciotto, M.R. et al. Acetylcholine receptors containing the beta2 subunit are involved in the reinforcing properties of nicotine. Nature 391, 173–177 (1998). 37. Maskos, U. et al. Nicotine reinforcement and cognition restored by targeted expression of nicotinic receptors. Nature 436, 103–107 (2005). 38. Klink, R. de Kerchove d’Exaerde, A., Zoli, M. & Changeux, J.P. Molecular and physiological diversity of nicotinic acetylcholine receptors in the midbrain dopaminergic nuclei. J. Neurosci. 21, 1452–1463 (2001). 39. Wooltorton, J.R., Pidoplichko, V.I., Broide, R.S. & Dani, J.A. Differential desensitization and distribution of nicotinic acetylcholine receptor subtypes in midbrain dopamine areas. J. Neurosci. 23, 3176–3185 (2003). 40. Pidoplichko, V.I. et al. Nicotinic cholinergic synaptic mechanisms in the ventral tegmental area contribute to nicotine addiction. Learn. Mem. 11, 60–69 (2004). 41. Carr, D.B. & Sesack, S.R. Projections from the rat prefrontal cortex to the ventral tegmental area: target specificity in the synaptic associations with mesoaccumbens and mesocortical neurons. J. Neurosci. 20, 3864–3873 (2000). 42. Omelchenko, N. & Sesack, S.R. Laterodorsal tegmental projections to identified cell populations in the rat ventral tegmental area. J. Comp. Neurol. 483, 217–235 (2005). 43. Lanca, A.J., Adamson, K.L., Coen, K.M., Chow, B.L. & Corrigall, W.A. The pedunculopontine tegmental nucleus and the role of cholinergic neurons in nicotine self-administration in the rat: a correlative neuroanatomical and behavioral study. Neuroscience 96, 735–742 (2000). 44. Malenka, R.C. & Nicoll, R.A. Long-term potentiation–a decade of progress? Science 285, 1870–1874 (1999). 45. Garzon, M., Vaughan, R.A., Uhl, G.R., Kuhar, M.J. & Pickel, V.M. Cholinergic axon terminals in the ventral tegmental area target a subpopulation of neurons expressing low levels of the dopamine transporter. J. Comp. Neurol. 410, 197–210 (1999). 46. Mansvelder, H.D., Keath, J.R. & McGehee, D.S. Synaptic mechanisms underlie nicotine-induced excitability of brain reward areas. Neuron 33, 905–919 (2002). 47. Salminen, O. et al. Subunit composition and pharmacology of two classes of striatal presynaptic nicotinic acetylcholine receptors mediating dopamine release in mice. Mol. Pharmacol. 65, 1526–1535 (2004). 48. Zhang, H. & Sulzer, D. Frequency-dependent modulation of dopamine release by nicotine. Nat. Neurosci. 7, 581–582 (2004). 49. Zhou, F.M., Liang, Y. & Dani, J.A. Endogenous nicotinic cholinergic activity regulates dopamine release in the striatum. Nat. Neurosci. 4, 1224–1229 (2001). 50. Rice, M.E. & Cragg, S.J. Nicotine amplifies reward-related dopamine signals in striatum. Nat. Neurosci. 7, 583–584 (2004). 51. de Rover, M., Lodder, J.C., Kits, K.S., Schoffelmeer, A.N. & Brussaard, A.B. Cholinergic modulation of nucleus accumbens medium spiny neurons. Eur. J. Neurosci. 16, 2279– 2290 (2002). 52. Mansvelder, H.D., De Rover, M., McGehee, D.S. & Brussaard, A.B. Cholinergic modu-

1469


REVIEW lation of dopaminergic reward areas: upstream and downstream targets of nicotine addiction. Eur. J. Pharmacol. 480, 117–123 (2003). 53. Berrendero, F., Kieffer, B.L. & Maldonado, R. Attenuation of nicotine-induced antinociception, rewarding effects, and dependence in mu-opioid receptor knock-out mice. J. Neurosci. 22, 10935–10940 (2002). 54. Walters, C.L., Cleck, J.N., Kuo, Y.C. & Blendy, J.A. mu-opioid receptor and CREB activation are required for nicotine reward. Neuron 46, 933–943 (2005). 55. Nestler, E.J. Molecular mechanisms of drug addiction. Neuropharmacology 47 (suppl.) 24–32 (2004). 56. Nestler, E.J. Is there a common molecular pathway for addiction? Nat. Neurosci. 8, 1445–1449 (2005). 57. Kenny, P.J., Gasparini, F. & Markou, A. Group II metabotropic and alpha-amino-3hydroxy-5-methyl-4-isoxazole propionate (AMPA)/kainate glutamate receptors regulate the deficit in brain reward function associated with nicotine withdrawal in rats. J. Pharmacol. Exp. Ther. 306, 1068–1076 (2003). 58. Kenny, P.J. & Markou, A. Conditioned nicotine withdrawal profoundly decreases the activity of brain reward systems. J. Neurosci. 25, 6208–6212 (2005). 59. Salas, R., Pieri, F. & De Biasi, M. Decreased signs of nicotine withdrawal in mice null for the beta4 nicotinic acetylcholine receptor subunit. J. Neurosci. 24, 10035–10039 (2004). 60. Kandel, D., Chen, K., Warner, L.A., Kessler, R.C. & Grant, B. Prevalence and demographic correlates of symptoms of last year dependence on alcohol, nicotine, marijuana and cocaine in the U.S. population. Drug Alcohol Depend. 44, 11–29 (1997). 61. Upadhyaya, H.P., Deas, D., Brady, K.T. & Kruesi, M. Cigarette smoking and psychiatric comorbidity in children and adolescents. J. Am. Acad. Child Adolesc. Psychiatry 41, 1294–1305 (2002). 62. Patton, G.C. et al. Depression, anxiety, and smoking initiation: a prospective study over 3 years. Am. J. Public Health 88, 1518–1522 (1998). 63. Kendler, K.S. et al. Smoking and major depression. A causal analysis. Arch. Gen. Psychiatry 50, 36–43 (1993). 64. Freedman, R. et al. Linkage disequilibrium for schizophrenia at the chromosome 15q13–14 locus of the alpha7-nicotinic acetylcholine receptor subunit gene (CHRNA7). Am. J. Med. Genet. 105, 20–22 (2001). 65. Harris, J.G. et al. Effects of nicotine on cognitive deficits in schizophrenia. Neuropsychopharmacology 29, 1378–1385 (2004). 66. Fowler, J.S. et al. Brain monoamine oxidase A inhibition in cigarette smokers. Proc. Natl. Acad. Sci. USA 93, 14065–14069 (1996). 67. Balfour, D.J. & Ridley, D.L. The effects of nicotine on neural pathways implicated in depression: a factor in nicotine addiction? Pharmacol. Biochem. Behav. 66, 79–85 (2000). 68. McEvoy, J.P. & Brown, S. Smoking in first-episode patients with schizophrenia. Am. J. Psychiatry 156, 1120–1121 (1999). 69. Goeders, N.E. Stress and cocaine addiction. J. Pharmacol. Exp. Ther. 301, 785–789 (2002). 70. Saal, D., Dong, Y., Bonci, A. & Malenka, R.C. Drugs of abuse and stress trigger a common synaptic adaptation in dopamine neurons. Neuron 37, 577–582 (2003). 71. Hughes, J.R., Higgins, S.T. & Bickel, W.K. Nicotine withdrawal versus other drug withdrawal syndromes: similarities and dissimilarities. Addiction 89, 1461–1470 (1994). 72. Hughes, J.R., Rose, G.L. & Callas, P.W. Nicotine is more reinforcing in smokers with a past history of alcoholism than in smokers without this history. Alcohol. Clin. Exp. Res. 24, 1633–1638 (2000). 73. Crabbe, J. & Lovinger, D.M. Animal models of alcoholism: treatment target identifica-

tion and insight into mechanisms. Nat. Neurosci. 8, 1471–1480 (2005). 74. True, W.R. et al. Common genetic vulnerability for nicotine and alcohol dependence in men. Arch. Gen. Psychiatry 56, 655–661 (1999). 75. Madden, P.A., Heath, A.C., Starmer, G.A., Whitfield, J.B. & Martin, N.G. Alcohol sensitivity and smoking history in men and women. Alcohol. Clin. Exp. Res. 19, 1111–1120 (1995). 76. Schuckit, M.A. & Smith, T.L. Changes over time in the self-reported level of response to alcohol. Alcohol Alcohol. 39, 433–438 (2004). 77. Blednov, Y.A., Stoffel, M., Alva, H. & Harris, R.A. A pervasive mechanism for analgesia: activation of GIRK2 channels. Proc. Natl. Acad. Sci. USA 100, 277–282 (2003). 78. Lopez, M.F., White, N.M. & Randall, C.L. Alcohol tolerance and nicotine cross-tolerance in adolescent mice. Addict. Biol. 6, 119–127 (2001). 79. Staley, J.K. et al. Cortical gamma-aminobutyric acid type A-benzodiazepine receptors in recovery from alcohol dependence: relationship to features of alcohol dependence and cigarette smoking. Arch. Gen. Psychiatry 62, 877–888 (2005). 80. John, U., Meyer, C., Rumpf, H.J. & Hapke, U. Probabilities of alcohol high-risk drinking, abuse or dependence estimated on grounds of tobacco smoking and nicotine dependence. Addiction 98, 805–814 (2003). 81. Riala, K., Hakko, H., Isohanni, M., Jarvelin, M.R. & Rasanen, P. Teenage smoking and substance use as predictors of severe alcohol problems in late adolescence and in young adulthood. J. Adolesc. Health 35, 245–254 (2004). 82. Clark, A., Lindgren, S., Brooks, S.P., Watson, W.P. & Little, H.J. Chronic infusion of nicotine can increase operant self-administration of alcohol. Neuropharmacology 41, 108–117 (2001). 83. De Fiebre, C.M., Medhurst, L.J. & Collins, A.C. Nicotine response and nicotinic receptors in long-sleep and short-sleep mice. Alcohol 4, 493–501 (1987). 84. de Fiebre, N.C., Dawson, R., Jr. & de Fiebre, C.M. The selectively bred high alcohol sensitivity (HAS) and low alcohol sensitivity (LAS) rats differ in sensitivity to nicotine. Alcohol. Clin. Exp. Res. 26, 765–772 (2002). 85. Owens, J.C. et al. Alpha 4 beta 2* nicotinic acetylcholine receptors modulate the effects of ethanol and nicotine on the acoustic startle response. Alcohol. Clin. Exp. Res. 27, 1867–1875 (2003). 86. Butt, C.M., King, N.M., Stitzel, J.A. & Collins, A.C. Interaction of the nicotinic cholinergic system with ethanol withdrawal. J. Pharmacol. Exp. Ther. 308, 591–599 (2004). 87. Borghese, C.M., Henderson, L.A., Bleck, V., Trudell, J.R. & Harris, R.A. Sites of excitatory and inhibitory actions of alcohols on neuronal alpha2beta4 nicotinic acetylcholine receptors. J. Pharmacol. Exp. Ther. 307, 42–52 (2003). 88. Butt, C.M. et al. A polymorphism in the alpha4 nicotinic receptor gene (Chrna4) modulates enhancement of nicotinic receptor function by ethanol. Alcohol. Clin. Exp. Res. 27, 733–742 (2003). 89. Cardoso, R.A. et al. Effects of ethanol on recombinant human neuronal nicotinic acetylcholine receptors expressed in Xenopus oocytes. J. Pharmacol. Exp. Ther. 289, 774–780 (1999). 90. Blomqvist, O., Hernandez-Avila, C.A., Van Kirk, J., Rose, J.E. & Kranzler, H.R. Mecamylamine modifies the pharmacokinetics and reinforcing effects of alcohol. Alcohol. Clin. Exp. Res. 26, 326–331 (2002). 91. Cohen, C., Kodas, E. & Griebel, G. CB(1) receptor antagonists for the treatment of nicotine addiction. Pharmacol. Biochem. Behav. 81, 387–395 (2005). 92. Johnson, B.A. Topiramate-induced neuromodulation of cortico-mesolimbic dopamine function: a new vista for the treatment of comorbid alcohol and nicotine dependence? Addict. Behav. 29, 1465–1479 (2004).

1470




REVIEW

Laboratory models of alcoholism: treatment target identification and insight into mechanisms David M Lovinger & John C Crabbe Laboratory models, including animal tissues and live animals, have proven useful for discovery of molecular targets of alcohol action as well as for characterization of genetic and environmental factors that influence alcohol’s neural actions. Here we consider strengths and weaknesses of laboratory models used in alcohol research and analyze the limitations of using animals to model a complex human disease. We describe targets for the neural actions of alcohol, and we review studies in which animal models were used to examine excessive alcohol drinking and to discover genes that may contribute to risk for alcoholism. Despite some limitations of the laboratory models used in alcohol research, these experimental approaches are likely to contribute to the development of new therapies for alcohol abuse and alcoholism.

The euphoria that follows the first drink at a party is familiar to many, as is the loss of judgment and control after continuing to imbibe alcohol. Unfortunately, the escalation that leads to the desperate craving for alcohol, destroying lives and families, is also experienced by far too many people. Alcohol abuse and alcoholism involve interactions among a number of neural mechanisms, including acute sensitivity to alcohol, development of tolerance to and dependence upon the drug, and development of an intense desire to consume the drug (sometimes called ‘craving’). Alcohol has direct actions on molecules that lead to intoxication and influence responses to chronic drinking1 (Fig. 1). Other molecules, most notably several neural proteins, interact with ethanol less directly by influencing expression or function of molecules that are direct ethanol targets or by altering the function of neural circuits that participate more generally in addictive processes (Fig. 1). Molecules in both groups contribute to the acute and chronic stages of alcohol use, influencing the likelihood of abuse and the expression of alcoholism. Inherited factors contribute a great deal to an individual’s susceptibility to alcohol abuse and alcoholism (Fig. 2). Genetic makeup and environmental experience interact to alter both direct alcohol actions and molecular mechanisms that indirectly affect ethanol-related behaviors. Such interactions influence the acute sensitivity to ethanol intoxication and the neuroadaptive changes that take place in response to chronic

David M. Lovinger is in the Laboratory for Integrative Neuroscience, National Institute on Alcohol Abuse and Alcoholism, National Institutes of Health, Rockville, Maryland 20852, USA, and John C. Crabbe is in the Department of Behavioral Neuroscience, Oregon Health & Science University, the Portland Alcohol Research Center and the Department of Veterans Affairs Medical Center, Portland, Oregon 97239, USA. e-mail: [email protected] Published online 26 October 2005; doi:10.1038/nn1581


alcohol abuse. Initial sensitivity to alcohol differs among different individuals, and there is some evidence that these differences contribute to susceptibility to later alcoholism2. Animal and human studies suggest that differences in alcohol sensitivity, seeking and drinking behaviors have a strong inherited component and may involve polymorphisms in genes encoding proteins involved in intoxication3,4. Chronic alcohol exposure causes neuroadaptations that foster continued alcohol abuse5,6. Tolerance to ethanol (that is, decreased sensitivity to intoxication) and dependence on it, evidenced by both physical withdrawal and increased desire for the drug, result from repeated exposure to alcohol. The neuroadaptations underlying these behavioral adaptations to alcohol involve molecular mechanisms that are affected both directly and indirectly by alcohol. The propensity for certain alcohol-related neuroadaptations is also influenced by genetic factors and gene-environment interactions (Fig. 2). As with other drugs of abuse, alcohol seeking in dependent individuals may reflect its increased value as a reward, its ability to reduce the undesirable effects of alcohol withdrawal, or both. The changes in reward value and the consequences of withdrawal result from the neuroadaptations brought about by chronic alcohol, and thus it is important to understand the molecular, cellular and genetic basis of these neuroadaptations. Addiction to alcohol shares common neural substrates with other addictions at the molecular, cellular and circuit levels. These commonalities include alcohol and drug actions at similar subsets of neural proteins and neural systems that are targets for other abused substances (such as the GABAA receptor7, brain glutamatergic systems8,9, nuclear signaling proteins such as CREB10, stimulation of VTA dopaminergic neurons11 and involvement of mesocorticolimbic and extended amygdala circuitry12). However, alcohol abuse and alcoholism also involve neural processes distinct from other addictions. Attempts to understand the common and unique aspects of alcohol addiction have spurred

1471

REVIEW

Chronic exposure (neuroadaptation)

Functional/ biophysical analysis

Transgenics, knockouts, QTL analysis, etc.

Pharmacological, biochemical, molecular biological analyses

Direct alcohol targets


Forward and reverse genetics

Alcohol-associated proteins

Preclinical drug screening

Pharmacotherapeutic target candidates

Ann Thomson

Acute pharmacology

Figure 1 Use of animals to identify direct and indirect alcohol targets can lead to development of pharmacotherapies for alcohol abuse and alcoholism. Pharmacological and genetic techniques are used alone and in combination to uncover molecular targets that are directly affected by alcohol (direct alcohol targets) and to identify proteins that are indirectly altered by alcohol or that influence neural responses to alcohol (alcoholassociated proteins). These molecules can then be targeted in preclinical screens designed to determine if pharmacotherapeutic agents or other target-specific treatments alter responses that are predictive of alcohol abuse or alcoholism (such as alcohol intake, alcohol reinforcement, craving for alcohol or relapse). Molecular targets that show promise in preclinical testing are then the focus of clinical testing for safety and efficacy.

Alcohol’s difficult pharmacology: how do we identify a ‘hit’? Animals, animal tissues and molecules arising from experimental animals are widely used in studies aimed at finding molecular targets of alcohol actions. Before examining some of the latest findings in this field, it is worthwhile to consider the limitations of experimental searches for alcohol targets. Investigators searching for direct targets of alcohol actions in the CNS often use pharmacological approaches involving direct application of ethanol to neuronal preparations16. Numerous complications arise when using these approaches. Because of the simple structure of the ethanol molecule, only two reactive sites are present: the OH group,

which is capable of hydrogen bonding, and the short carbon backbone, which contributes to weak hydrophobic interactions (although metal ion interactions do occur in alcohol dehydrogenase)17. The molecule does not form ionic or covalent bonds. These characteristics generate poor reactivity that results in low potency of the drug, such that acute neural effects ranging from intoxication to anesthesia18 are observed only at blood and brain concentrations from ∼5 mM to 100 mM. This low potency prevents the application of some pharmacological techniques, such as radioligand binding, that can detect sites of direct molecular interactions between proteins and compounds with high binding affinities. Furthermore, biophysical studies of specific ethanol interactions with molecules are often hampered by the distribution of ethanol into many cellular compartments and the weak interactions of the molecule with many potential target molecules. These experimental limitations make it difficult to measure actual affinity of alcohol for potential molecular targets, leaving only measures of potency and efficacy that are difficult to relate to occupancy of a particular molecular site. Without measures of direct molecular interactions, it is difficult to be certain that a particular moiety is truly an ‘alcohol binding site’. Contrast this with the pharmacology of opiate drugs, which have high affinity for a well-characterized G protein–coupled receptor19, and one begins to see the problems inherent in the molecular neuropharmacology of alcohol. Methodologies for examining direct alcohol interactions have been applied to some proteins (such as the Drosophila melanogaster LUSH protein; see below), but until these types of experiments can be performed on a wide variety of proteins, it is best to avoid inferring too much about direct alcohol targets and affinity of alcohol at these targets. The foregoing discussion provides some of the reasons why identification of direct alcohol target molecules has lagged behind the discovery of other drug targets. Ethanol, within the range of physiologically relevant concentrations, nearly always produces small effects. It is not unusual to observe only a 30% change in any given measure of cellular or molecular function even at concentrations that are near the higher limit of the sublethal range (for example, 100 mM). This may be a blessing for the person drinking alcohol as an intoxicant, as it ensures that the desired effects occur with a large margin of safety from outright toxicity. However, for the neuropharmacologist, the small effect sizes are problematic. Attempts to determine the molecular interactions that underlie relevant ethanol effects are often confounded by signal-to-noise issues inherent in looking for changes in a small response. Small effects are often overlooked, and changes in molecular or cellular function that result from timedependent changes in the experimental system might be mistaken for ethanol actions. Consequently, the experimenter must have confidence in the accuracy and range of variability of the measures that are made.

1472


investigators to adopt new animal models and research approaches that have not been as widely used in investigation of other addictions. Here we focus on two aspects of the use of animal models in alcoholism-related research that most distinguish this line of research from investigation of other addictions. First, the basic molecular mechanisms of alcohol action are not shared by most other drugs of abuse, and the hunt for alcohol targets has spurred investigators to use unique experimental approaches in animal tissues, cells and molecules. Second, the use of genetic animal models to explore mechanisms of ethanol action has traditionally been more widespread in alcohol research than in investigations of other addictions. Given the large role of vertebrate animal models, and more recently, invertebrate models, in alcohol research, we feel it is important to critically analyze the strengths and weaknesses of the animal models. We highlight findings from animal models that indicate molecular targets and genetic underpinnings of alcoholism and consider the limitations of using animal models as surrogates for understanding human alcoholism. The behavioral repertoire of laboratory animals differs from that of humans, and many animals—most notably invertebrates—have nervous systems that differ radically from that found in man. In typical studies, rats are offered a choice of drinking water versus ethanol solutions, or animals are tested in a protocol of operant self-administration, in which a specific response yields access to alcohol. Drug state can also be repeatedly paired with specific cues, and the animal can be tested on the choice it makes between a saline-paired or alcohol-paired cue13. Most behavioral assays of alcohol intoxication and tolerance target some aspect of motor coordination (which turns out to be a very complex domain of behavior genetically14) or dysregulation of body temperature, and these methods are also used for studying a broad spectrum of drugs with sedative-hypnotic effects. Anxiolytic effects are also studied. Laboratory animals cannot verbally report their subjective states, which only adds to this problem. Consequently, when using these organisms, it is difficult to infer affective and cognitive states that are likely to be important determinants of alcohol intoxication, abuse and alcoholism15.

REVIEW a

Genes Proteins Pathways

Behaviors

Figure 2 Complexity of gene-environment-behavioral interactions in the neural actions of alcohol. (a) Individual genes through their expression lead to synthesis of specific proteins. These proteins in turn participate in numerous cellular- and systems-level pathways that ultimately modulate specific addiction-related behaviors. A behavioral difference can be determined essentially by a single gene: some complex traits show purely mendelian inheritance. More commonly, a single gene can influence multiple behaviors (pleiotropy), or multiple genes can exert converging influence on a single behavior, which is then variously termed multigenic, oligogenic or polygenic. This describes what we know of addictive behaviors. The bottom of panel a depicts the range of interest of genomics, proteomics and metabolomics analyses. Their common feature is the concentration of the analysis on a single behavior (such as impulsive drug-taking). Here we also show that the relationships between proteins, pathways and behaviors are complex. The usual case under investigation must account for both multigenic and pleiotropic effects. (b) All of the above gene-behavior relationships take place in environments that themselves may differ. The only differences between the cases shown in environments 1 and 2 in the example are that the effects of the second and third genes on the first two behaviors differ depending on the environment—a gene-environment interaction. The final panel shows that behaviors can themselves interact, and that they can influence the expression of genes. The range of interest of behavioral genomics includes all relevant genes and behaviors, including assessment of their interactions with each other and the environments in which they are assessed. This level of analysis will be critical to understanding the addictions.

Gene effects Mendelian Pleiotropic

Genomics Proteomics Metabolomics

b

Genes Proteins Pathways

Behaviors

1

1

2

2

3

3

4

4

1

1

2

2

3

3

4

4

Environments

Environment 1 Gene ҂ environment interaction

number of cycles of exposure and withdrawal to be modeled. These reduced systems have contributed much to our understanding of cellular and molecular neuroadaptations to alcohol exposure.

Environment 2

Behavioral interactions Effects of behavior on gene expression

Behavioral genomics

Ann Thomson


Multigenic Oligogenic Polygenic Multigenic + Pleiotropic

Sample sizes used in experiments with ethanol often must be larger than those used with more efficacious drugs. In addition, it is especially important to demonstrate reversibility of the alcohol effect to ensure that small changes are not due to ‘drift’ in the measurement over time. The high concentrations of alcohol used in most experiments bring into play possible artifacts related to ethanol’s solvent properties (that is, the possibility that ethanol may redissolve a hydrophobic compound that had previously adhered to plastic tubing or may leach plasticizing agents out of tubing or storage vessels). Thus, great care must be taken in designing pharmacological studies with ethanol. An alternative approach to identifying molecules with important roles in the neural actions of alcohol is to examine molecular changes brought about by chronic alcohol exposure. Alcohol-induced neuroadaptations that lead to tolerance, dependence, withdrawal signs and increased alcohol intake often involve prolonged exposure to alcohol. Many techniques for chronic ethanol exposure in reduced neuronal preparations (such as dispersed primary neuronal cultures or organotypic slice cultures) allow investigators to examine molecular adaptations to ethanol in wellcontrolled experimental systems20,21. Important variables include the concentration of alcohol applied, ethanol evaporation at physiological temperatures, the pattern and duration of alcohol exposure, and the


Modeling alcohol genetics in animals Largely by historical accident, animal models have been used more often in the field of alcohol genetics than in any other area of psychiatric genetics. The wealth of alcohol response data on genetic animal models gives us a window through which to assess the strengths and limitations of animal models in biomedical research more generally. Some issues may be traced to the complex genetics of alcohol dependence disorders (Box 1; Fig. 2), and there are parallels with the pharmacological challenges just discussed. Taken together, this means that no single animal model will convincingly capture all the features of this complex disorder, and a useful model should target certain well-defined features. Most research with genetic animal models has therefore concentrated on a few relatively straightforward phenotypes. Although there are many problems in research on the cellular and molecular neuropharmacology of ethanol and many complexities that must be addressed by animal models, there is an abundance of candidate molecules that seem to be targets for direct and indirect actions of ethanol in the nervous system. However, the abundance of targets also makes it difficult to determine the most important contributors to the neural effects of ethanol. The direct approach: pharmacological methods Heterologous expression of molecules allows researchers to examine the effects of ethanol on the function of proteins expressed in a cellular context free of many neural proteins. Heterologous expression combined with examination of the function of the same protein in neurons has been used to characterize ethanol sensitivity of a variety of proteins, including neurotransmitter receptors, ion channels and neurotransmitter transporters1,22. The discussion of all these molecular targets is beyond the scope of this short review, but the following description of alcohol actions on GABAergic synaptic transmission illustrates the usefulness of this approach. Interactions between ethanol and GABAergic inhibitory synaptic

1473

REVIEW

transmission have been the subject of intensive study for over 30 years7,23. Evidence from in vivo and in vitro studies supports the idea that ethanol produces many of its intoxicating actions by enhancing GABAergic synaptic transmission23. The GABAA receptor is one potential molecular target for ethanol, and a variety of evidence sug-

gests that alcohol potentiates GABAA receptor function. However, there is ongoing controversy as to the prevalence and importance of ethanol actions on this receptor. Ethanol potentiation of GABAA receptor function has been observed in some isolated neuron preparations7,24,25. However, negative results have also been reported in numerous neuronal subtypes24. Studies in the Xenopus laevis oocyte expression system have generally found ethanol potentiation, but at relatively high ethanol concentrations26. However, very few studies in mammalian cell systems have produced evidence for this potentiating effect24. Clearly, understanding the molecular basis for these discrepant results will be a key step in resolving this controversy and determining the role of GABAA receptors in alcohol effects. Diversity among GABAA receptor subtypes could contribute to the differential effects of ethanol in different cells. GABAA receptors exist in the plasma membrane as heteropentamers of a variety of subunits that can coassemble in various combinations. The most recent studies carried out in oocytes, isolated neurons and brain slices indicate that ethanol can potentiate responses to GABA that are mediated by receptors containing the α4 or α6, β2 or β3 and δ subunits25,27,28 at ethanol concentrations as low as 1–3 mM, well within the range associated with in vivo intoxication, although there is unresolved disagreement about the effective concentrations and the shape of the ethanol concentration-response functions observed in different laboratories. These receptors are often found outside the synapse, where they are thought to mediate tonic GABAergic inhibition28,29. Studies of granule neurons in brain slices from the cerebellar cortex and dentate gyrus demonstrate ethanol potentiation of a tonic GABAA-mediated current that seems to involve this type of receptor28,29. What role, if any, does this ethanol potentiation have in acute intoxication and alcohol abuse? One clue comes from a study28 concluding that a naturally occurring polymorphism in the rat GABAA α6 subunit may account for differences in the alcohol sensitivity of GABAA receptors. This polymorphism is also associated with a change in acute alcohol motor impairment observed in the rats in vivo28, and the polymorphism segregates in rat lines bred for differential alcohol sensitivity30–32. Differences in alcohol sensitivity have been related to differences in alcohol intake, and thus this line of research may prove useful for understanding factors that contribute to this link. Indeed, studies of human α6 polymorphisms suggest some association between this subunit and alcoholism3. These findings indicate the potential importance of direct ethanol effects on α6-containing GABAA receptors. However, the role of the rat α6 polymorphism in differential alcohol sensitivity has been questioned. The specific polymorphism leading to reduced GABA function is not observed in human or even mouse α6, although there are interesting parallels with the human α6 P385S polymorphism. Other polymorphisms cosegregate strongly with the α6 difference31 and this may suggest that there is coordinated regulation of the cluster of GABAA subunit genes in this chromosomal region. Furthermore, several acute alcohol sensitivity phenotypes do not simply cosegregate with this polymorphism in the AT and ANT rat lines that were selectively bred to differ in motor impairment by ethanol32. Finally, gene-targeted mice lacking the α6 subunit do not differ in sensitivity to ethanol motor impairment33, although this may be due to compensations during development. Thus, it is premature to conclude that this polymorphism is the only factor contributing to alcohol sensitivity in rats28. Expression of the α6 subunit is restricted to cerebellar granule neurons30, and thus this subunit would seem less likely to participate in aspects of intoxication that do not involve motor impairment. For this reason, it is unlikely that the α6 polymorphism contributes to alcohol sensitivity for most aspects of intoxication. In addition, there is little information about the direct effect of ethanol

1474



BOX 1 GENETICS OF COMPLEX TRAITS The complexities involved in relating genes to addictive behavior are becoming more amenable to analysis, thanks to the development of new ways to measure and manipulate the function of genes as well as large-scale applications of classical breeding methods. Although the field is moving rapidly, addiction will not be explained by concentrating solely on functional genomics and its related reductions (proteomics, metabolomics; Fig. 2a). Equal attention should be paid to the other part of the gene-behavior nexus, the behavioral phenotypes. Most behaviors taken to index different aspects of drug or alcohol dependence vary continuously over a wide range in populations, and each relevant gene influencing susceptibility or severity usually has a small, graded (polygenic) effect on that trait (Fig. 2a). Hence, the involvement of many genes is required to explain the full range of genetic contribution to risk for or severity of alcohol dependence. A gene that influences one behavioral contributor to risk is likely to be relevant for other behaviors or even things that may have no obvious connection with behavior (pleiotropy). Furthermore, genes do not act independently but typically interact with each other (epistasis). Finally, in the same way that any statistic describes the specific population from which it is drawn, genetic influence reflects the environment in which a gene’s effects on behavior are studied, and if a different environment (such as set of families, rearing conditions, stressors) is studied, different genes may be important, they may interact differently, or both. (Fig. 2b). Alcoholism and drug dependence are also complex at the behavioral level. Although the severity of any of the various symptoms that in the aggregate lead to diagnosis is continuous, the diagnoses are categorical. Furthermore, unlike many genetically influenced diseases, diagnosis does not imply a particular pathophysiology. Thus, members of any diagnostic category (such as ‘alcohol dependence’) are heterogeneous. Individuals diagnosed with drug-related disorders are also highly likely to have other (comorbid) diagnoses as well, such as depression or anxiety disorders, each of which has its own set of definitional and genetic complexities. Addictive behaviors may interact as well. Anxiety may drive attempts to alleviate it through escalation of drug ingestion, but drug withdrawal may then engender more anxiety, not less. In addition, there are developmental shifts in the characteristics of the disorder across the life span. For example, any amount of drinking in an 8-year-old may be considered excessive, whereas high consumption levels in 20-year-olds are seen as less diagnostic than a similar level would be in a 65-year-old. Confronted with this daunting complexity, the field often progresses in small steps. A study may identify one or two relevant genes and assess their interactions with other factors. Gradually, genetic knowledge from many studies then can be assembled into a larger system of interactants that enables us to understand a set of related behaviors. We term this perspective behavioral genomics (Fig. 2b).


REVIEW in mammalian heterologous expression systems or on isolated neurons containing α6-δ receptors. It is always reassuring to observe drug effects in a single neuron under tightly controlled physiological and pharmacological conditions, because this rules out many indirect effects that could contribute to ethanol effects in more intact tissues. Clearly, further work is needed to determine the factors that contribute to ethanol effects on GABAA receptors and the scope of the involvement of this receptor in alcohol effects on the brain and behavior. The long-held belief that ethanol potentiation of GABAergic transmission arises solely from increased GABAA receptor function is being challenged by studies indicating that ethanol potentiates GABA release in several brain regions. This potentiation is observed at reasonable ethanol concentrations and thus could certainly contribute to intoxication and other alcohol-related behaviors. As yet, there is little information about the molecular mechanisms underlying this potentiation. A recent review34 nicely describes this growing area of investigation. Here we simply note that ethanol increases GABAergic synaptic input onto neurons, including those in the central amygdala, cerebellum and hippocampus35–37. At many of these synapses, ethanol alters several electrophysiological measurements indicating an increase in presynaptic GABA release. These include increased paired-pulse facilitation and increases in the frequency of miniature inhibitory postsynaptic currents (mIPSCs). In cerebellum, an increase in firing rate of GABAergic interneurons also contributes to the increased GABAergic input36. Because transmission at most GABAergic CNS synapses is mediated predominantly by GABAA receptors, presynaptic ethanol potentiation must be considered when formulating hypotheses about effects of GABAA receptor–targeted drugs on alcohol-related behaviors and changes in alcohol responsiveness or alcohol drinking in gene-targeted mice lacking GABAA receptor subunits. The spineless approach: invertebrate models Studies in invertebrate models have unearthed molecules with potential roles in alcohol intoxication and alcoholism, along with unexpected concordance with previous findings from studies of vertebrates. For example, several genes that influence acute alcohol sensitivity and tolerance have been identified in mutant Drosophila melanogaster fruit flies, and intracellular signaling pathways involving cAMP and protein kinase A (PKA), as well as transcription-associated proteins38,39, are beginning to be implicated. Powerful molecular genetic methods can be applied in D. melanogaster to alter protein expression in only a subset of neurons within the nervous system much more rapidly than is possible in mammals. Selective overexpression of a PKA inhibitor leads to localization of the cAMP-mediated signaling important for regulation of ethanol sensitivity in a subset of putative neurosecretory neurons in the CNS40. This finding rules out a host of more trivial interpretations of previous findings (such as alterations in muscle function or metabolism), and helps validate D. melanogaster as a model for study of ethanol effects on the CNS. Studies using cell lines derived from rodents had already implicated cAMP-associated signaling proteins in effects of acute and chronic ethanol41. Furthermore, alcohol intake can be regulated by treatments that affect the cAMP-PKA signaling system in the nucleus accumbens42. Thus, symmetry is emerging between the findings in D. melanogaster and rodent models, but it is not yet clear whether any of these proteins are direct targets of alcohol. Another D. melanogaster mutant may provide one of the best models for direct ethanol binding to a protein. The lush mutant lacks an olfactory protein involved in detecting ethanol (from rotting fruit)43. Flies lacking the LUSH protein do not avoid alcohol, in contrast to wild-type flies. The alcohol avoidance behavior typical of wild-type flies can be reinstated by reexpression of the LUSH protein in mutant flies43. The


lush gene product is putatively involved in olfactory detection of ethanol and other short carbon-chain alcohols such as butanol, and it has an alcohol-binding site identified by NMR and X-ray crystallography44. However, other researchers disagree that LUSH is an alcohol binding protein and have challenged the idea that flies avoid high ethanol concentrations45. These investigators attribute the avoidance behavior to plasticizing contaminants in ethanol solutions, again raising the specter of problems brought about by alcohol’s solvent properties. This issue has yet to be resolved, but the crystal structure of LUSH certainly indicates how an alcohol-binding pocket in a protein might look. The nematode Caenorhabditis elegans has also been used to probe the genetic and molecular basis of acute ethanol sensitivity. Mutant C. elegans lines with decreased sensitivity to ethanol-induced disruption of acute movement and egg-laying behavior have a loss-of-function mutation in the gene that codes for the pore-forming subunit of the large-conductance calcium-activated potassium channel (BK channel)46. Acute ethanol exposure also enhances BK channel function in C. elegans neurons46, which could alter motor and egg-laying function. This elegant work demonstrates the power of invertebrate models for analysis of susceptibility genes and molecular targets of ethanol action. Similarly, in rodents, ethanol potentiates BK channels in neurohypophyseal peptidergic terminals47, which likely contributes to alcohol-induced changes in vasopressin secretion and diuresis within this neuroendocrine system. The role of BK channels in mammalian intoxication remains to be determined. The use of invertebrate animal models has also revealed potential molecular targets of ethanol action that had not been previously examined. For example, ethanol sensitivity in D. melanogaster is regulated by insulin, the insulin receptor and insulin-producing neurons. Mutation of the insulin receptor or its substrate in a subset of neurons within the D. melanogaster CNS increases sensitivity to intoxication48. It will be interesting to determine to what extent the interactions between insulin and the cAMP signaling system are involved in fruit fly intoxication. Behavioral screening methods for zebra fish (a vertebrate model, of course) have also been developed to study effects of ethanol on behavior49–51. Thus, the rapidly emerging use of invertebrate models for alcohol research may be complemented by the use of simpler vertebrate models that are amenable to genetic analysis and genetic manipulation. The early promise of invertebrate and non-mammalian vertebrate model systems has spurred intense interest in the alcohol research field. The tractability of these organisms for genetic studies such as mutagenesis screens suggests that further interesting developments may be on the way. However, their utility for understanding human alcohol abuse and alcoholism is still in question. It remains to be seen if the nervous system circuitry of invertebrates can be generalized meaningfully to the mammalian brain, as it is possible that neural circuits contributing to behavioral effects of alcohol differ in invertebrates and mammals even if the behavioral outcomes are similar in D. melanogaster and C. elegans. For example, the effects of ethanol on basic motor control circuits seem to underlie incoordination in invertebrates, whereas more sophisticated circuits such as the basal ganglia and cerebellum contribute to motor impairment in rodents. Future studies must focus on determining the comparability of molecular and cellular effects of ethanol in the different circuits found in vertebrate and invertebrate organisms. Invertebrate and vertebrate model organisms, however, do share considerable commonalities in neurochemistry. Neurotransmitters such acetylcholine, dopamine and glutamate are present in the nervous systems of all these organisms, and many neuropeptides are also commonly expressed. However, there are notable differences in neurotransmitter types and receptors that could alter the effect of ethanol in the different

1475


REVIEW systems. For example, octopamine is a monoamine neurotransmitter implicated in ethanol actions in D. melanogaster38. This neurotransmitter is not found in mammals, but may have analogous cellular functions to norepinephrine in the mammalian CNS. Significant questions remain as to whether neuronal systems implicated in alcohol actions in mammals will have a role in invertebrate responses to the drug. Thus far, we have stressed similarities between molecular substrates of alcohol responsiveness in D. melanogaster, C. elegans and rodents. However, there are no findings to date implicating GABAergic or glutamatergic transmission in invertebrate alcohol effects, despite the known importance of these neurotransmitters in mammalian responses to alcohol7–9,28. It may well be that mammalian brains have evolved alcohol-sensitive circuitry and neurotransmitter systems that are not prominent in the invertebrate brain. It will also be important to determine which invertebrate proteins are direct targets of ethanol action (such as the BK channel) and which indirectly influence alcohol sensitivity. Examination of the effects of alcohol in invertebrate models is a relatively new research area, and thus it is premature to speculate too strongly on the similarities and differences in the systems. Ultimately, more complete characterization of the molecular actions of ethanol in multiple nervous systems will be needed to determine which mechanisms are most common. Comparison of the effects of alcohol on neural circuits and molecules across organisms should yield some consensus on the similarity of intoxication in the different models. Invertebrate models show great promise for alcohol research at present, and it is likely that information gained from such models will complement, but certainly not replace, that gained from the continued use of mammalian laboratory animals. Genetic animal models and alcohol responses Evaluation of the validity and utility of rodent animal models is also important for developing better tools to analyze the genetics of alcohol-related behaviors. A wealth of historical data has been reviewed elsewhere52,53. Selective breeding has been used since the late 1940s to develop lines of rats and mice that differ markedly in voluntary alcohol drinking. Inbred strains of mice that preferred to drink alcohol, or shunned it, were first identified in 1959. At least seven different pairs of lines of rats have been bred for high versus low alcohol preference. These genetically high and low drinkers have some fairly consistent differences in the neurobiology of alcohol responses53–55. High drinkers tend to have low synaptic levels of serotonin and dopamine, and activation of serotonin 5-HT1A or 5-HT2 receptors reduces alcohol intake, as does stimulation of dopamine D2 or D3 or GABAA receptors. Blockade of opioid or 5-HT3 receptors also reduces intake, and these classes of agents have some efficacy in the clinic with alcoholics56,57. Rather than review the older material, we concentrate on a few recent examples of genetic animal model research drawn from quantitative trait locus (QTL) gene mapping approaches and studies of null mutants, including some gene expression studies.

alcohol58. The primitive tools then available and the sparsity of the mouse map of genomic markers made progress initially slow, but as the technology has improved, the rate of progress has greatly accelerated in the past few years. To date in addiction research, a single QTL has been definitively traced to the underlying gene59. In multiple genetic animal models (standard and recombinant inbred mouse strains, segregating populations, short-term selectively bred lines, congenic strains), a QTL on mouse chromosome 4 has been linked to the severity of acute alcohol and pentobarbital withdrawal60. The QTL region initially contained several hundred genes, and polymorphisms and gene expression data allowed the strong inference that the gene Mpdz, which codes for a multiple PDZ domain protein, was the only remaining gene in the region that affected the withdrawal response59. MPDZ protein facilitates coupling of ligands and receptors and interacts with 5-HT2A, 5-HT2B and 5-HT2C receptors as well as cKIT (a membrane tyrosine kinase receptor) and p75 (a Trk-associated neurotrophin receptor). A more detailed description of this project, which is also pursuing withdrawal QTL on chromosomes 1 and 11, and of the strengths and weaknesses of QTL mapping approaches, is given elsewhere61. Because of the close similarity (>85%) between the mouse and human genomes owing to their shared ancestor, gene findings in mice and humans are increasingly reciprocally informative62. Not surprisingly, alcoholrelated QTL do not act in isolation. The influence on acute alcohol and pentobarbital withdrawal of a chromosome 11 QTL, whose interval contains multiple GABAA receptor subunit genes (α1, α6, β2, γ2), depends upon the presence of one or more genes within the chromosome 1 QTL63. Similarly, some QTL affecting chronic ethanol withdrawal severity are apparent only when considered epistatically with another genome region. The chromosome 4 QTL containing the Mpdz gene also affects chronic alcohol withdrawal severity in combination with a QTL on chromosome 8 (ref. 64). Numerous QTL affecting the tendency of mice to seek to drink or to avoid alcohol solutions are also being pursued55. One of the methods recently brought into play to accelerate the progress of gene mapping efforts is to synthesize information about both gene expression- and gene sequence–based sources of genetic influence. Until recently, QTL mapping efforts simply identified associations between the occurrence of the mapped trait and markers that varied in DNA base pair sequence. Positional cloning of a candidate gene near the markers was followed by functional studies showing that the gene’s different protein variants had different neurobiological functions affecting the trait. We now know that QTL ‘signals’ may also derive ultimately from differential expression of the underlying gene. For example, a specific genetic variation in a promoter region leading to greater expression of a gene could be found preferentially in high alcohol responders. Methods exploring both gene sequence and gene expression variation are proving to be powerful tools to identify genes for complex traits65 and are now being used on alcohol-related QTL66–70. For example, a locus very near the dopamine D2 receptor gene strongly affects its expression68. Levels of DRD2 protein are correlated with ethanol’s motor stimulant effects, as well as with its efficacy to produce a conditioned taste aversion. Both these traits, which are thought to model different aspects of ethanol’s reinforcing effects, show a strong QTL signal in this region as well.

Finding new genes through gene mapping Mapping approaches to identify genes affecting alcohol responses were initiated in the early 1990s. For reasons discussed earlier (Box 1), individual differences in responses to alcohol are not generally all-or-none but rather are quantitatively distributed in populations (that is, they are quantitative traits). In gene mapping, individuals’ behavioral responses are first compared with their genotype at many polymorphic markers scattered across the chromosomes. A pattern of association between a marker and degree of response provisionally identifies a quantitative trait locus (QTL). Of well over 100 QTL for many responses to seven different drugs of abuse cited in an early review, over half were for

Genetically engineered candidate genes Mouse stocks genetically engineered to have a null mutation for one of about 50 genes or to overexpress them have been studied for one or more alcohol responses. We know of no recent, systematic review of the mutant studies. The GABAA receptor has been fairly thoroughly studied using genetic engineering strategies, as this receptor system has long

1476



REVIEW been a target of mechanistic studies (see above). Null mutants and/or transgenic overexpression mutants have been generated for the α1, α2, α5, α6, β2, β3, γ2L, γ2S and δ receptor subunit genes, as well as for the GABA transporter gene. A recent review of this work compiled behavioral data for several ethanol responses, including high-dose sensitivity (loss of righting reflex), withdrawal severity and tendency toward self-administration71. This review compared the transgenic behavioral data with those for expression of the various subunits in brain. Because the chromosomal location of these genes is known, it was also possible to compare behavioral sensitivity in the mutants and gene expression patterns taken from a publicly available set of informatics tools and data sets, WebQTL (http://www.genenetwork.org) with the location of QTL that had been mapped for the same behavioral responses. For example, by QTL mapping efforts, the GABAA receptor subunit– rich chromosome 11 region was found to harbor a gene or genes affecting sensitivity to alcohol-induced loss of righting reflex in some mouse genotypes. The α1 subunit gene was differentially expressed in mouse strains that differed in sensitivity, and gene deletion of the α1 subunit gene affected sensitivity in one of two such null mutant models. This, and similar data for the α2 subunit gene on chromosome 5, strongly suggests a role for the α1 and α2 subunit genes in this response71. Because many different ethanol-related responses have been studied and because many different genotypes were used, the data matrix available for analysis is frustratingly incomplete. However, such syntheses of available data will serve to identify the lacunae in our knowledge, allowing the complexities of GABA-ethanol interactions to be reduced to a more tractable set of possibilities. Genetic models: challenges and future directions Studies of tolerance and dependence seem to have reasonable face validity across the biomedical science community. Dependent animals, including those from which alcohol is subsequently withdrawn, display apparent dysregulation of multiple neural systems that normally are held within bounds by homeostatic feedback loops. Alcohol dependence in humans obviously must disrupt in some way the overall calculation of an individual of the risks and benefits of drinking a lot, but how these miscalculations play out in the brain’s reward and stress axis circuitry is not yet clear. Functional and structural brain imaging studies suggest that chronic alcohol abuse has localized effects on brain circuits72, and functional MRI signals measured during a behavioral inhibition task may prove to be useful to predict individuals’ expectancies about the positive or negative effects of drinking alcohol73. Some rhesus monkeys given voluntary access to alcohol daily for long periods will develop patterns of repeated drinking that are obviously excessive74. However, all the rodent genetic animal models of alcohol drinking mentioned above, even those animals intensely selectively bred for preference, show patterns of drinking that do not look exactly like human alcoholism. Mice and rats will drink until they begin to approach brain alcohol levels that produce clear signs of intoxication, but they will rarely continue to drink thereafter until their blood levels subside somewhat. In other words, rodents seem to have some internal controls limiting intake that are not shared by susceptible humans. Several behavioral manipulations can overcome these controls in the laboratory75–79, but these generally require fairly long-duration exposure and are labor intensive. Thus, it would be useful to have more highthroughput rodent models in which animals voluntarily self-administer ethanol to excess. Attempts to create such models are supported by the Integrative Neuroscience Initiative on Alcoholism of the US National Institute on Alcohol Abuse and Alcoholism. Results have supported the idea that scheduling access to alcohol during the circadian dark78,80,


limiting access to fluids for a brief period80–82 or establishing chronic levels of alcohol before offering it for self-administration83,84 can convince some rodents to drink to intoxication on a daily basis. To date, most of these studies have focused on the C57BL/6J mouse, known to drink large quantities of alcohol. Although it is encouraging that we seem to be able to overcome the (unknown) factors that tend to selflimit ingestion in these creatures, the ultimate utility of these newer models will be achieved if it proves possible to selectively breed for very high intakes. Other challenges facing animal model research are more generic. Some of the principal defining characteristics of the alcohol dependence disorders are behavioral (for example, interference with work or social life). Because animals cannot self-report, we must infer their subjective state from their behavior, which is not as straightforward as it seems15. There are many rodent behavioral assays designed to reflect an anxious state, and anxiety is reported by alcoholics when they are withdrawing from a period of chronic drinking. However, a recent review of rat and mouse studies of anxiety during alcohol withdrawal found that nearly all such studies (at least with mice) showed a marked reduction in general activity of the animals during withdrawal. Because reduced activity in an apparatus such as the elevated plus maze makes it difficult to infer anxiety specifically, it is methodologically tricky to study withdrawal anxiety in rodents85, although some protocols seem to work reproducibly. Invertebrate models are not exempt from these challenges. To date, they have been able to model simple sensitivity to the sedative effects of alcohol, and the reduction in that sensitivity (tolerance) with chronic exposure. Given the behavioral repertoire of flies and worms, it will be difficult to devise assays tapping such internal states as anxiety, response inhibition or its opposite (impulsivity), craving and reduced or increased sensitivity to reward. We are fairly confident that the rodent models in place are a reasonable reflection of these psychological aspects of alcohol dependence. However, some features (for example, the intense social pressures loosely called ‘peer pressures’ experienced by adolescents) will never be modeled in non-human species. In general, the feelings experienced by alcoholics at different stages of their disease have been less convincingly modeled than the biological sequelae of chronic administration. Serotonin and alcohol: translating animal research to humans The neurotransmitter serotonin (5-HT) is important for normal and dysregulated emotional behavior, including depression and alcoholism. Compounds targeted at brain serotonergic systems, including SSRIs and 5-HT3 antagonists, were proposed as potential treatments for alcoholism on the basis of animal studies. Brain levels of 5-HT are regulated by several proteins, including the serotonin transporter (5-HTT) that removes serotonin from the synapse. The specific serotonin reuptake inhibitors (SSRIs), most frequently used to treat depression, act by inhibiting 5-HTT. Activity in the 5-HTT gene is affected by a polymorphism (5-HTTLPR) in its promoter region, with three common genotypes termed s/s, l/l, or s/l for short (s) or long (l) variants. One or more short alleles leads to reduced 5-HTT function, but attempts to relate the 5-HTTLPR polymorphism to depression or other psychiatric diagnoses have met with varied success86. However, low serotonin transporter activity is also seen in rhesus monkeys with an analogous short variant promoter polymorphism in the transporter gene87. The s/l heterozygotes show greater alcohol selfadministration than l/l homozygotes. Experimental studies seek to constrain complexity by using uniform environmental conditions, but this approach may obscure genetic effects. The effect was seen only in monkeys separated from their mothers at birth and reared with peers after

1477


REVIEW five weeks of nursery rearing. Peer rearing is demonstrably stressful, so there is an intriguing parallel with a study showing a similar gene-byenvironment interaction between the transporter polymorphism and life stress on human depression88. When subjected to isolation stress, monkeys of both sexes showed elevated adrenocorticotropic hormone (ACTH) if they were l/s heterozygotes, but l/s females showed an even greater increase in ACTH if they were peer reared and no increase if they were mother reared89. In a human genetics study, one variant of the l allele of 5-HTTLPR significantly predicted both alcohol use disorders and a low response to alcohol, which itself is predictive of later alcohol-related diagnoses90. This study awaits replication. The 5-HTTLPR polymorphism has pleiotropic effects. A different (missense) mutation in the 5-HTT gene itself has been identified in two generations of a family with a high incidence of obsessive-compulsive disorder. Individuals with the missense mutation who were also l/l homozygotes have extremely high levels of 5-HT transporter activity, possibly owing to their ‘double hit’ at the two mutant genes, and suffer from multiple disorders, including alcohol and drug abuse, Asperger’s syndrome and social phobia91. Basic-clinical translation: developing pharmacotherapies As the ultimate goal of much alcohol research is development of effective therapeutics, there is great interest in translating basic findings on the neural actions of alcohol into potential clinical applications. Two medications currently used in the treatment of alcoholism—acamprosate (a taurine analog with no well-established molecular target in CNS) and naltrexone (an opiate receptor antagonist)—target neural circuitry involved in alcohol ‘craving’ and relapse56. Acamprosate was not developed on the basis of data from animal models, whereas laboratory animal studies indicated that opiate receptor antagonists reduce alcohol intake before the first tests of naltrexone in alcoholics92. Researchers have focused on developing and testing new pharmacotherapies aimed at molecular targets identified by their sensitivity to alcohol or the apparent involvement of the target in control of alcohol drinking in rodent models. A parallel effort is examining the efficacy of potential therapeutics in the broader array of behavioral assays mentioned above. Analogous to the situation with null mutants, and with the exception of naltrexone, acamprosate and the SSRIs, the current matrix of compounds and behavioral tests is frustratingly patchy93. Some promising new targets are emerging. For example, noncompetitive antagonists of the NMDA receptor, a neurotransmitter that is directly inhibited by ethanol94, are being tested for efficacy in reducing alcoholic relapse95. Antagonists of the CB1 cannabinoid receptor and the mGluR5 metabotropic receptors reduce alcohol intake in several animal models, and gene-targeted animals lacking CB1 show reduced intake in some behavioral models96,97. Plans are underway to evaluate the therapeutic potential of CB1 and mGluR5 antagonists in human alcoholics. The roles of CB1 and mGlu receptors in brain reinforcement and control of addictive behaviors are not confined to interactions with ethanol. There is a substantial and growing literature implicating the brain cannabinoid system in opiate, nicotine and even food-related addictions98,99. Indeed, early clinical indicators suggest that drugs targeted at the CB1 receptor will be useful in the treatment of obesity and cessation of cigarette smoking100. These findings suggest that CB1targeted therapeutics are successful because they tap into generalized brain reward systems. The finding that alcohol abuse is likely to be closely related to other addictions that involve these systems comes as no surprise. However, the finding that the glutamate and cannabinoid systems may have generalized roles suggests that caution should be exercised in touting these neurochemicals as targets for treatment of any one addiction disorder. Long-term studies may yet reveal that general

1478

interference with brain reinforcement systems might have untoward consequences, such as anhedonia or disruption of normal neurophysiological drives. The CB1 and mGlu receptors are just two among many molecular targets being considered for alcoholism pharmacotherapy93. Two metabolic genes (alcohol dehydrogenase and acetaldehyde dehydrogenase) have specific polymorphic variants whose unpleasant effects in their bearers significantly protect against alcoholism3. The known therapeutic agent disulfiram (Antabuse) is directed at this metabolic susceptibility. The wide array of potential therapeutic targets raises the hope that effective treatments will be found in the not-too-distant future. Summary Much recent progress has been achieved on both fronts highlighted in this review. Several interesting targets for potential pharmacotherapies have emerged from continuing pharmacological and physiological studies. New genetic animal models are emerging in both vertebrate and invertebrate organisms that will lead to greater levels of behavioral dysregulation of alcohol self-administration, as well as identification of new genes and their proteins that are involved in alcohol’s neural actions. Significant barriers still must be surmounted before we can successfully use information gained from animal models for translation of basic findings into clinical treatments, but that end point seems closer than ever. ACKNOWLEDGMENTS The authors are supported by the US Department of Veterans Affairs (J.C.C.), and the US National Institute on Alcohol Abuse and Alcoholism (AA10760, AA12714 and AA13519 to J.C.C) and the Division of Intramural Clinical and Basic Research (D.M.L.). We thank M. Rutledge-Gorman for help in preparing the manuscript, and G. McClearn for many previous versions of Figure 2. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. 2.

3. 4. 5.

6. 7. 8. 9.

10. 11.

12. 13.

14. 15.

Diamond, I. & Gordon, A.S. Cellular and molecular neuroscience of alcoholism. Physiol. Rev. 77, 1–20 (1997). Schuckit, M.A., Smith, T.L. & Kalmijn, J. The search for genes contributing to the low level of response to alcohol: patterns of findings across studies. Alcohol. Clin. Exp. Res. 28, 1449–1458 (2004). Oroszi, G. & Goldman, D. Alcoholism: genes and mechanisms. Pharmacogenomics 5, 1037–1048 (2004). Crabbe, J.C. Genetic contributions to addiction. Annu. Rev. Psychol. 53, 435–462 (2002). Heinz, A., Schafer, M., Higley, J.D., Krystal, J.H. & Goldman, D. Neurobiological correlates of the disposition and maintenance of alcoholism. Pharmacopsychiatry 36 (Suppl.) S255–S258 (2003). Weiss, F. et al. Compulsive drug-seeking behavior and relapse. Neuroadaptation, stress, and conditioning factors. Ann. NY Acad. Sci. 937, 1–26 (2001). Ueno, S. et al. Alcohol actions on GABA(A) receptors: from protein structure to mouse behavior. Alcohol. Clin. Exp. Res. 25, 76S–81S (2001). Carlezon, W.A., Jr. & Nestler, E.J. Elevated levels of GluR1 in the midbrain: a trigger for sensitization to drugs of abuse? Trends Neurosci. 25, 610–615 (2002). Krystal, J.H., Petrakis, I.L., Mason, G., Trevisan, L. & D’Souza, D.C. N-methyl-Daspartate glutamate receptors and alcoholism: reward, dependence, treatment, and vulnerability. Pharmacol. Ther. 99, 79–94 (2003). Nestler, E.J. Molecular mechanisms of drug addiction. Neuropharmacology 47. Suppl 1, 24–32 (2004). Gonzales, R.A., Job, M.O. & Doyon, W.M. The role of mesolimbic dopamine in the development and maintenance of ethanol reinforcement. Pharmacol. Ther. 103, 121–146 (2004). Nestler, E.J. Is there a common molecular pathway for addiction? Nat. Neurosci. 8, 1445–1449 (2005). Cunningham, C.L. & Phillips, T.J. Genetic basis of ethanol reward. in Molecular Biology of Drug Addiction (ed. Maldonado, R.) 263–294 (Humana, Totowa, New Jersey, 2003). Crabbe, J.C., Metten, P., Cameron, A.J. & Wahlsten, D. An analysis of the genetics of alcohol intoxication in inbred mice. Neurosci. Biobehav. Rev. 28, 785–802 (2005). Crabbe, J.C. & Morris, R.G.M. Festina lente: Late night thoughts on high-throughput


REVIEW 16. 17. 18. 19. 20.


21.

22. 23.

24.

25. 26. 27.

28.

29.

30.

31.

32. 33.

34. 35.

36.

37. 38. 39. 40.

41. 42.

43.

44.

45.

46. 47.

48.

screening of mouse behavior. Nat. Neurosci. 7, 1175–1179 (2004). Liu, Y. & Lovinger, D.M. Methods in Alcohol-Related Neuroscience Research (CRC, Washington, D.C., 2000). Dwyer, D.S. & Bradley, R.J. Chemical properties of alcohols and their protein binding sites. Cell. Mol. Life Sci. 57, 265–275 (2000). Schuckit, M.A. Drug and Alcohol Abuse (Plenum, New York, 1979). Simantov, R. & Snyder, S.H. The opiate receptor. Biochem. Soc. Trans. 5, 62–65 (1977). Carpenter-Hyland, E.P., Woodward, J.J. & Chandler, L.J. Chronic ethanol induces synaptic but not extrasynaptic targeting of NMDA receptors. J. Neurosci. 24, 7859–7868 (2004). Thomas, M.P. & Morrisett, R.A. Dynamics of NMDAR-mediated neurotoxicity during chronic ethanol exposure and withdrawal. Neuropharmacology 39, 218–226 (2000). Lovinger, D.M. Alcohols and neurotransmitter gated ion channels: past, present and future. Naunyn Schmiedebergs Arch. Pharmacol. 356, 267–282 (1997). Grobin, A.C., Matthews, D.B., Devaud, L.L. & Morrow, A.L. The role of GABA(A) receptors in the acute and chronic effects of ethanol. Psychopharmacology (Berl.) 139, 2–19 (1998). Aguayo, L.G., Peoples, R.W., Yeh, H.H. & Yevenes, G.E. GABA(A) receptors as molecular sites of ethanol action. Direct or indirect actions? Curr. Top. Med. Chem. 2, 869–885 (2002). Mihic, S.J. et al. Sites of alcohol and volatile anaesthetic action on GABA(A) and glycine receptors. Nature 389, 385–389 (1997). Sundstrom-Poromaa, I. et al. Hormonally regulated alpha(4)beta(2)delta GABA(A) receptors are a target for alcohol. Nat. Neurosci. 5, 721–722 (2002). Wallner, M., Hanchar, H.J. & Olsen, R.W. Ethanol enhances alpha 4 beta 3 delta and alpha 6 beta 3 delta gamma-aminobutyric acid type A receptors at low concentrations known to affect humans. Proc. Natl. Acad. Sci. USA 100, 15218–15223 (2003). Hanchar, H.J., Dodson, P.D., Olsen, R.W., Otis, T.S. & Wallner, M. Alcohol-induced motor impairment caused by increased extrasynaptic GABA(A) receptor activity. Nat. Neurosci. 8, 339–345 (2005). Wei, W., Faria, L.C. & Mody, I. Low ethanol concentrations selectively augment the tonic inhibition mediated by delta subunit-containing GABAA receptors in hippocampal neurons. J. Neurosci. 24, 8379–8382 (2004). Korpi, E.R., Kleingoor, C., Kettenmann, H. & Seeburg, P.H. Benzodiazepine-induced motor impairment linked to point mutation in cerebellar GABAA receptor. Nature 361, 356–359 (1993). Congeddu, E. et al. Molecular characterization of new polymorphisms at the β2, α1, γ2 GABA(A) receptor subunit genes associated to a rat nonpreferring ethanol phenotype. Brain Res. Mol. Brain Res. 110, 289–297 (2003). Radcliffe, R.A. et al. Behavioral characterization of alcohol-tolerant and alcohol-nontolerant rat lines and an F(2) generation. Behav. Genet. 34, 453–463 (2004). Homanics, G.E. et al. Gene knockout of the alpha6 subunit of the gamma-aminobutyric acid type A receptor: lack of effect on responses to ethanol, pentobarbital, and general anesthetics. Mol. Pharmacol. 51, 588–596 (1997). Siggins, G.R., Roberto, M. & Nie, Z. The tipsy terminal: presynaptic effects of ethanol. Pharmacol. Ther. 107, 80–98 (2005). Ariwodola, O.J. & Weiner, J.L. Ethanol potentiation of GABAergic synaptic transmission may be self-limiting: role of presynaptic GABA(B) receptors. J. Neurosci. 24, 10679–10686 (2004). Carta, M., Mameli, M. & Valenzuela, C.F. Alcohol enhances GABAergic transmission to cerebellar granule cells via an increase in Golgi cell excitability. J. Neurosci. 24, 3746–3751 (2004). Nie, Z. et al. Ethanol augments GABAergic transmission in the central amygdala via CRF1 receptors. Science 303, 1512–1514 (2004). Wolf, F.W. & Heberlein, U. Invertebrate models of drug abuse. J. Neurobiol. 54, 161–178 (2003). Scholz, H., Franz, M. & Heberlein, U. The hangover gene defines a stress pathway required for ethanol tolerance development. Nature 436, 845–847 (2005). Rodan, A.R., Kiger, J.A., Jr. & Heberlein, U. Functional dissection of neuroanatomical loci regulating ethanol sensitivity in Drosophila. J. Neurosci. 22, 9490–9501 (2002). Thiele, T.E. et al. High ethanol consumption and low sensitivity to ethanol-induced sedation in protein kinase A-mutant mice. J. Neurosci. 20, RC75, 1–6 (2000). Yao, L. et al. βγ dimers mediate synergy of dopamine D2 and adenosine A2 receptor-stimulated PKA signaling and regulate ethanol consumption. Cell 109, 733–743 (2002). Kim, M.S., Repp, A. & Smith, D.P. LUSH odorant-binding protein mediates chemosensory responses to alcohols in Drosophila melanogaster. Genetics 150, 711–721 (1998). Kruse, S.W., Zhao, R., Smith, D.P. & Jones, D.N. Structure of a specific alcohol-binding site defined by the odorant binding protein LUSH from Drosophila melanogaster. Nat. Struct. Biol. 10, 694–700 (2003). Zhou, J.J. et al. Revisiting the odorant-binding protein LUSH of Drosophila melanogaster: evidence for odour recognition and discrimination. FEBS Lett. 558, 23–26 (2004). Davies, A.G. et al. A central role of the BK potassium channel in behavioral responses to ethanol in C. elegans. Cell 115, 655–666 (2003). Dopico, A.M., Lemos, J.R. & Treistman, S.N. Ethanol increases the activity of large conductance, Ca(2+)-activated K+ channels in isolated neurohypophysial terminals. Mol. Pharmacol. 49, 40–48 (1996). Corl, A.B., Rodan, A.R. & Heberlein, U. Insulin signaling in the nervous system regulates ethanol intoxication in Drosophila melanogaster. Nat. Neurosci. 8, 18–19


49. 50.

51.

52.

53.

54. 55. 56.

57.

58. 59. 60.

61. 62. 63.

64. 65. 66. 67.

68.

69. 70. 71.

72. 73.

74.

75.

76.

77. 78.

79.

80.

(2005). Gerlai, R. Zebra fish: an uncharted behavior genetic model. Behav. Genet. 33, 461–468 (2003). Lockwood, B., Bjerke, S., Kobayashi, K. & Guo, S. Acute effects of alcohol on larval zebrafish: a genetic system for large-scale screening. Pharmacol. Biochem. Behav. 77, 647–654 (2004). Carvan, M.J., III, Loucks, E., Weber, D.N. & Williams, F.E. Ethanol effects on the developing zebrafish: neurobehavior and skeletal morphogenesis. Neurotoxicol. Teratol. 26, 757–768 (2004). Browman, K.E., Crabbe, J.C. & Li, T.K. Genetic strategies in preclinical substance abuse research. in Psychopharmacology: The Fourth Generation of Progress (eds. Bloom, F.E. & Kupfer, D.J.) (Lippincott, Williams & Wilkins, Philadelphia, 2000). Murphy, J.M. et al. Phenotypic and genotypic characterization of the Indiana University rat lines selectively bred for high and low alcohol preference. Behav. Genet. 32, 363–388 (2002). McBride, W.J. & Li, T.K. Animal models of alcoholism: neurobiology of high alcoholdrinking behavior in rodents. Crit. Rev. Neurobiol. 12, 339–369 (1998). Crabbe, J.C. & Phillips, T.J. Pharmacogenetic studies of alcohol self-administration and withdrawal. Psychopharmacology (Berl.) 174, 539–560 (2004). Soyka, M. & Chick, J. Use of acamprosate and opioid antagonists in the treatment of alcohol dependence: a European perspective. Am. J. Addict. 12 Suppl 1, S69–S80 (2003). Johnson, B.A. & Ait-Daoud, N. Neuropharmacological treatments for alcoholism: scientific basis and clinical findings. Psychopharmacology (Berl.) 149, 327–344 (2000). Crabbe, J.C., Belknap, J.K. & Buck, K.J. Genetic animal models of alcohol and drug abuse. Science 264, 1715–1723 (1994). Shirley, R.L., Walter, N.A., Reilly, M.T., Fehr, C. & Buck, K.J. Mpdz is a quantitative trait gene for drug withdrawal seizures. Nat. Neurosci. 7, 699–700 (2004). Fehr, C., Shirley, R.L., Belknap, J.K., Crabbe, J.C. & Buck, K.J. Congenic mapping of alcohol and pentobarbital withdrawal liability loci to a <1 centimorgan interval of murine chromosome 4: identification of Mpdz as a candidate gene. J. Neurosci. 22, 3730–3738 (2002). Crabbe, J.C. Identifying genes affecting addiction risk in animal models. in The Cell Biology of Addiction (eds. Madras, B. et al.) (Cold Spring Harbor Press, in press). Uhl, G.R. Molecular genetics of substance abuse vulnerability: remarkable recent convergence of genome scan results. Ann. NY Acad. Sci. 1025, 1–13 (2004). Hood, H.M., Belknap, J.K., Crabbe, J.C. & Buck, K.J. Genomewide search for epistasis in a complex trait: Pentobarbital withdrawal convulsions in mice. Behav. Genet. 31, 93–100 (2001). Bergeson, S.E. et al. Chromosomal loci influencing chronic alcohol withdrawal severity. Mammal. Genome 14, 454–463 (2003). Jansen, R.C. & Nap, J.P. Genetical genomics: the added value from segregation. Trends Genet. 17, 388–391 (2001). Hitzemann, R. et al. On the integration of alcohol-related quantitative trait loci and gene expression analyses. Alcohol. Clin. Exp. Res. 28, 1437–1448 (2004). Tabakoff, B., Bhave, S.V. & Hoffman, P.L. Selective breeding, quantitative trait locus analysis, and gene arrays identify candidate genes for complex drug-related behaviors. J. Neurosci. 23, 4491–4498 (2003). Chesler, E.J. et al. Complex trait analysis of gene expression uncovers polygenic and pleiotropic networks that modulate nervous system function. Nat. Genet. 37, 233–242 (2005). Bystrykh, L. et al. Uncovering regulatory pathways that affect hematopoietic stem cell function using ‘genetical genomics’. Nat. Genet. 37, 225–232 (2005). Hubner, N. et al. Integrated transcriptional profiling and linkage analysis for identification of genes underlying disease. Nat. Genet. 37, 243–253 (2005). Boehm, S.L., II et al. gamma-Aminobutyric acid A receptor subunit mutant mice: new perspectives on alcohol actions. Biochem. Pharmacol. 68, 1581–1602 (2004). Sullivan, E.V. & Pfefferbaum, A. Neurocircuitry in alcoholism: a substrate of disruption and repair. Psychopharmacology (Berl) 180, 583–594 (2005). Anderson, K.G., Schweinsburg, A., Paulus, M.P., Brown, S.A. & Tapert, S. Examining personality and alcohol expectancies using functional magnetic resonance imaging (fMRI) with adolescents. J. Stud. Alcohol 66, 323–331 (2005). Vivian, J.A. et al. Induction and maintenance of ethanol self-administration in cynomolgus monkeys (Macaca fascicularis): long-term characterization of sex and individual differences. Alcohol. Clin. Exp. Res. 25, 1087–1097 (2001). Middaugh, L.D., Szumlinski, K.K., Van Patten, Y., Marlowe, A.L. & Kalivas, P.W. Chronic ethanol consumption by C57BL/6 mice promotes tolerance to its interoceptive cues and increases extracellular dopamine, an effect blocked by naltrexone. Alcohol. Clin. Exp. Res. 27, 1892–1900 (2003). Mittleman, G., Van Brunt, C.L. & Matthews, D.B. Schedule-induced ethanol selfadministration in DBA/2J and C57BL/6J mice. Alcohol. Clin. Exp. Res. 27, 918–925 (2003). Falk, J.L. Production of polydipsia in normal rats by an intermittent food schedule. Science 133, 195–196 (1961). Sharpe, A.L., Tsivkovskaia, N.O. & Ryabinin, A.E. Ataxia and c-Fos expression in mice drinking ethanol in a limited access session. Alcohol. Clin. Exp. Res. 29, 1419–1426 (2005). Rodd-Henricks, Z.A. et al. Effects of concurrent access to multiple ethanol concentrations and repeated deprivations on alcohol intake of alcohol-preferring rats. Alcohol. Clin. Exp. Res. 25, 1140–1150 (2001). Rhodes, J.S., Best, K., Belknap, J.K., Finn, D.A. & Crabbe, J.C. Evaluation of a simple model of ethanol drinking to intoxication in C57BL/6J mice. Physiol. Behav.

1479

REVIEW 81. 82.

83.


84.

85. 86.

87.

88. 89.

90.

84, 53–63 (2005). Finn, D.A. et al. A procedure to produce high alcohol intake in mice. Psychopharmacology (Berl.) 178, 471–480 (2005). Cronise, K., Finn, D.A., Metten, P. & Crabbe, J.C. Scheduled access to ethanol results in motor impairment and tolerance in female C57BL/6J mice. Pharmacol. Biochem. Behav. 81, 943–953 (2005). Fidler, T.L., Clews, T.W. & Cunningham, C.L. Reestablishing an intragastric ethanol self-infusion model in rats. Alcohol. Clin. Exp. Res. (in press). Lopez, M.F. & Becker, H.C. Effect of pattern and number of chronic ethanol exposures on subsequent voluntary ethanol intake in C57BL/6J mice. Psychopharmacology (Berl.) published online 7 July 2005 (doi:10.1007/s00213005-0026-3). Kliethermes, C.L. Anxiety-like behaviors following chronic ethanol exposure. Neurosci. Biobehav. Rev. 28, 837–850 (2005). Anguelova, M., Benkelfat, C. & Turecki, G. A systematic review of association studies investigating genes coding for serotonin receptors and the serotonin transporter: I. Affective disorders. Mol. Psychiatry 8, 574–591 (2003). Barr, C.S. et al. Interaction between serotonin transporter gene variation and rearing condition in alcohol preference and consumption in female primates. Arch. Gen. Psychiatry 61, 1146–1152 (2004). Caspi, A. et al. Influence of life stress on depression: moderation by a polymorphism in the 5-HTT gene. Science 301, 386–389 (2003). Barr, C.S. et al. Sexual dichotomy of an interaction between early adversity and the serotonin transporter gene promoter variant in rhesus macaques. Proc. Natl. Acad. Sci. USA 101, 12358–12363 (2004). Hu, X. et al. An expanded evaluation of the relationship of four alleles to the level

1480

of response to alcohol and the alcoholism risk. Alcohol. Clin. Exp. Res. 29, 8–16 (2005). 91. Ozaki, N. et al. Serotonin transporter missense mutation associated with a complex neuropsychiatric phenotype. Mol. Psychiat. 8, 895, 933–936 (2003). 92. Oswald, L.M. & Wand, G.S. Opioids and alcoholism. Physiol. Behav. 81, 339–358 (2004). 93. Egli, M. Can experimental paradigms and animal models be used to discover clinically effective medications for alcoholism? Addict. Biol. (in press). 94. Lovinger, D.M., White, G. & Weight, F.F. Ethanol inhibits NMDA-activated ion current in hippocampal neurons. Science 243, 1721–1724 (1989). 95. Nagy, J. Renaissance of NMDA receptor antagonists: do they have a role in the pharmacotherapy for alcoholism? IDrugs 7, 339–350 (2004). 96. Backstrom, P., Bachteler, D., Koch, S., Hyytia, P. & Spanagel, R. mGluR5 antagonist MPEP reduces ethanol-seeking and relapse behavior. Neuropsychopharmacology 29, 921–928 (2004). 97. Hungund, B.L. & Basavarajappa, B.S. Role of endocannabinoids and cannabinoid CB1 receptors in alcohol-related behaviors. Ann. NY Acad. Sci. 1025, 515–527 (2004). 98. Di, M.V. & Matias,I. Endocannabinoid control of food intake and energy balance. Nat. Neurosci. 8, 585–589 (2005). 99. Le Foll, B. & Goldberg, S.R. Cannabinoid CB1 receptor antagonists as promising new medications for drug dependence. J. Pharmacol. Exp. Ther. 312, 875–883 (2005). 100. Van Gaal, L.F., Rissanen, A.M., Scheen, A.J., Ziegler, O. & Rossner, S. Effects of the cannabinoid-1 receptor blocker rimonabant on weight reduction and cardiovascular risk factors in overweight patients: 1-year experience from the RIO-Europe study. Lancet 365, 1389–1397 (2005).




REVIEW

Neural systems of reinforcement for drug addiction: from actions to habits to compulsion Barry J Everitt & Trevor W Robbins Drug addiction is increasingly viewed as the endpoint of a series of transitions from initial drug use—when a drug is voluntarily taken because it has reinforcing, often hedonic, effects—through loss of control over this behavior, such that it becomes habitual and ultimately compulsive. Here we discuss evidence that these transitions depend on interactions between pavlovian and instrumental learning processes. We hypothesize that the change from voluntary drug use to more habitual and compulsive drug use represents a transition at the neural level from prefrontal cortical to striatal control over drug seeking and drug taking behavior as well as a progression from ventral to more dorsal domains of the striatum, involving its dopaminergic innervation. These neural transitions may themselves depend on the neuroplasticity in both cortical and striatal structures that is induced by chronic self-administration of drugs.

The nucleus accumbens is well known to mediate the reinforcing effects of drugs, but more recent research emphasizes the role of the striatum as a whole, including the shell and core components of the nucleus accumbens, in the processes leading first to drug abuse and then to addiction. This view has been stimulated by progress in understanding the dopamine-dependent, serial communication between the various domains of the striatum via a cascading loop interconnectivity1, and by an improved understanding of associative learning mechanisms that conceive of behavioral output as an interaction between pavlovian and instrumental learning processes2,3. In particular, the description of two processes that seem to function partly in parallel, but with the second eventually dominating behavioral output, has led to the concepts of action-outcome and stimulus-response (‘habit’) learning. Here we elaborate the hypothesis that these behavioral processes can be mapped onto the parallel and serial, dynamic functioning of corticostriatal circuitry (Fig. 1) to mediate the ‘switches’4,5 between drug reinforcement, drug abuse and drug addiction. Reinforcement, conditioning and the nucleus accumbens The reinforcing effects of addictive drugs are multidimensional (Box 1). Drugs act as ‘instrumental reinforcers’—that is, they increase the likelihood of responses that produce them, resulting in drug self-administration or ‘drug taking’ (defined in Box 2). Environmental

Barry J. Everitt and Trevor W. Robbins are in the Department of Experimental Psychology and the MRC-Wellcome Behavioural and Clinical Neuroscience Institute, University of Cambridge, Downing Street, Cambridge CB2 3EB, UK. e-mail: [email protected] Published online 26 October 2005; corrected after print 13 April 2006; doi:10.1038/nn1579


stimuli that are closely associated in time and space with the effects of self-administered drugs gain incentive salience through the process of pavlovian conditioning (Box 2). Drugs produce subjective or ‘discriminative’ effects, which include the sensing of autonomic activity (‘feelings’) or distortions in sensory processing. Stimulant drugs such as cocaine and amphetamine (along with others) also exaggerate the perceptual impact or incentive salience of environmental stimuli, especially those that already predict important environmental events, which are known as conditioned stimuli (CSs). We postulate that any combination of these effects may constitute the ‘rewarding’ effect of a drug—that is, the subjective effects produced by attributions made about the conditioned stimuli. In particular, we argue that it is the sense of expectancy, or perhaps even more importantly, the sense of ‘control’ over such interoceptive and exteroceptive states, including the overall level of arousal accompanying them, acquired through action-outcome learning (Box 1) that constitutes instrumental drug reinforcement. CSs that predict natural reinforcers, such as a light that predicts food, can have several effects on behavior, in addition to eliciting pavlovian (that is, automatic or reflexive) elements of approach and consummatory behavior. The locomotor stimulation produced by psychomotor stimulants such as amphetamine and cocaine may arise in this way. CSs can have motivational effects: for example, increasing rates of responding for food when the CS is presented unexpectedly (called pavlovian– instrumental transfer, PIT; Box 2)2. These motivational effects of CSs can be ascribed to a hypothetical process of pavlovian arousal, which serves to energize or activate responding, whether in terms of enhanced locomotor activity or increasing rates of instrumental (operant) behavior. Considerable evidence now shows that the midbrain dopamine neurons show fast phasic burst firing in response to such CSs6 but may also be active, at least in their tonic mode, under other circumstances in response to such factors as unpredictability7, novelty, stress and

1481

Figure 1 Representation of limbic circuitry, with tentative localization of functions involved in drug addiction. (a) Key connectivities in human brain (redrawn from ref. 90). (b) Limbic corticalventral striatopallidal circuitry. (i) Processing of conditioned reinforcement and delays by basolateral amygdala and of contextual information by hippocampus. (ii) Goal-directed actions involve interaction of prefrontal cortex with other structures, possibly including nucleus accumbens but also dorsomedial striatum. (iii) ‘Habits’ depend on interactions between prefrontal cortex and dorsolateral striatum. (iv) ‘Executive control’ depends on prefrontal cortex and includes representation of contingencies, representation of outcomes and their value and subjective states (craving and, presumably, feelings) associated with drugs. (v) Drug craving involves activation of orbital and anterior cingulate cortex, and temporal lobe including amygdala, in functional imaging studies. (vi) Connections between dopaminergic neurons and striatum reflect ‘spirals’—serial interactions organized in a ventral-to-dorsal cascade. (vii) Reinforcing effects of drugs may engage stimulant, pavlovian-instrumental transfer and conditioned reinforcement processes in the nucleus accumbens shell and core and then engage stimulus-response habits that depend on dorsal striatum. Green/blue arrows, glutamatergic projections; orange arrows, dopaminergic projections; pink arrows, GABAergic projections; Acb, nucleus accumbens; AMG, amygdala; BLA, basolateral amygdala; CeN, central nucleus of the amygdala; VTA, ventral tegmental area; SNc, substantia nigra pars compacta. GP, globus pallidus (D, dorsal; V, ventral); Hipp, hippocampus; mPFC, medial prefrontal cortex; AC, anterior cingulate cortex; OFC, orbitofrontal cortex; VS, ventral striatum; DS, dorsal striatum; Thal, thalamus. Modified from refs. 91,92.

Acb shell

Acb core

Dorsal striatum

Ann Thomson


REVIEW

food deprivation8. For this reason, we have previously used the term ‘activation’ to describe the important role of the mesolimbic and mesostriatal dopamine systems in behavioral output, to distinguish it from general changes (such as in EEG) associated with the term ‘arousal’8. Thus, in terms of drug abuse, in addition to obvious direct influences of drugs such as cocaine on extracellular dopamine, it might be feasible in certain circumstances to detect the effects of CSs themselves on striatal dopamine function. In testing this hypothesis, we have found that unexpected presentations of drug-paired CSs elicit dopamine release in the core but not in the shell region of the nucleus accumbens9. Consistent with these data, selective lesions of the nucleus accumbens core10 or infusions of NMDA or dopamine receptor antagonists into the nucleus accumbens core during training11 greatly retard the acquisition of pavlovian approach responses, whereas infusions of NMDA or dopamine D1 receptor antagonists into this region after a training trial disrupt the consolidation of this response into memory12. Lesions of the nucleus accumbens core also abolish PIT13, and increasing dopamine in the nucleus accumbens shell potentiates PIT14. Therefore, it might logically be thought that pavlovian approach is involved in maladaptively attracting humans toward sources of addictive drug reinforcers, and that drug-associated CSs that occur unexpectedly invigorate their efforts to seek and take or ‘want’ drugs as emphasized in the incentive salience theory of addiction15. However,

neither phenomenon (neither approach to a CS predictive of a drug, nor enhancement of drug seeking by the unexpected presentation of a drugassociated CS) has been clearly demonstrated in laboratory studies of drug seeking or relapse16–18, although both are readily seen in animals responding for natural rewards. It may be that the experimental conditions for demonstrating these phenomena in a drug seeking setting have not yet been optimized, but it may also be that the behavioral influence of CSs associated with drugs and natural reinforcers differ fundamentally in this regard19. The neural basis of pavlovian approach behavior and PIT has been reviewed extensively elsewhere19,20 and will not further be considered here, as it relates exclusively to studies with CSs associated with high-incentive natural reinforcers. In certain circumstances CSs can also function as conditioned reinforcers. Conditioned reinforcement occurs when stimuli that were initially motivationally neutral, such as a light, become reinforcing in their own right via association with primary reinforcers such as food or drugs. These stimuli help to maintain instrumental responding by bridging delays to the ultimate goal, such as food or cocaine, and affect the responses to D-amphetamine on a delayed gratification task21. It is well known that drugs such as amphetamine, nicotine and (under certain circumstances) opiates greatly increase responding with conditioned reinforcement. For example, infusion of amphetamine into the nucleus accumbens increases the acquisition of responding for a

1482


REVIEW


BOX 1 THE SIGNIFICANCE OF SUBJECTIVE RESPONSES IN REINFORCEMENT MECHANISMS Precisely what is a reinforcer, and why might a drug of abuse have reinforcing properties? These fundamental questions remain difficult to answer definitively. Early theories of motivation, as described in any textbook of psychology, suggested that reinforcers produce (i) drive (or need) reduction operating according to a homeostatic regulatory model, (ii) memory consolidation (for example, of the association of a conditioned stimulus predicting a reinforcer) or (iii) incentive-motivational effects by which the expectation of a reinforcer (mediated presumably by its representation in the brain) evokes appropriate preparatory (appetitive) responses, such as approach behavior or physiological adjustment that constitutes a sequence of motivated behavior terminated by consummatory behavior (such as eating food or sexually mounting a female in heat) usually elicited by the reinforcer. There is clearly some truth in each of these accounts, and it is no surprise that they are reflected in contemporary theories of drug addiction15,88,93,94. These theoretical schemes do not, however, refer directly to subjective responses or ‘feelings’ associated with drug effects, although an eventual explanation will have to accommodate these. The incentive-motivational class of theories has often been assumed to emphasize the hedonic properties of the reinforcer, especially when there is no obvious deficit or need state: for example, reinforcers such as intracranial electrical selfstimulation of the brain, cocaine, sucrose or novel objects. This conceptualization has led to the use of terms such as ‘reward’ and ‘liking’ that connote subjective responses associated with reinforcers (generally their post-presentational consequences, which can become associated with incentive-motivation via conditioning). Although these subjectively loaded terms refer to hypothetical processes of attribution that are associated with reinforcement, the processes themselves have never been identified or localized to particular brain regions or networks. This is in part because they have been confounded with the more implicit processes of reinforcement itself. For example, the use of the term ‘reward’ for ‘reinforcement’ might have led to the possibly mistaken view that those structures subserving reinforcement, such as the nucleus accumbens and its dopaminergic innervation, also mediate ‘reward’95, including its subjective, attributional aspects. It is clearly much harder to test the hypothesis that the nucleus accumbens is implicated in ‘reward’ than ‘reinforcement’. Similarly, the identification of ‘hedonic’ responses associated with

food-related conditioned reinforcer22. This behavior depends on two major influences; the nucleus accumbens core mediates the effects of the conditioned reinforcer10 via its afferent inputs from limbic cortical structures19 (see below). By contrast, the mesolimbic dopamine projection, especially to the nucleus accumbens shell10 mediates the response rate–increasing or psychomotor stimulant effects of the drug, hypothetically by simulating the behaviorally activating effects of pavlovian arousal and affecting the incentive salience of the conditioned reinforcer. This dopamine-dependent potentiation of conditioned reinforcement is a key component of the reinforcing effects of stimulant drugs such as cocaine, amphetamine and nicotine and likely other drugs as well. The effects of conditioned reinforcers, perhaps especially drug-related conditioned reinforcers, are pervasive and profound. For example, they support the learning of new drug seeking responses (Box 2), an effect that persists for at least two months without any further experience


acceptance and rejection reflexes96 may have led to confusion in separating those structures associated with attribution from those involved in controlling the reflex itself. Whereas it is clear that the latter reside in the brainstem, it seems less likely that subjective attribution does. This analysis, however, is consistent with the view that processes constituting reinforcement can be dissociated: for example, those mechanisms controlling appetitive behavior, such as instrumental lever pressing and locomotor approach behavior that generally occur remote from the reinforcer in space and time, and the consummatory responses associated with its proximal occurrence. Subjective responses associated with reinforcement are similarly dissociable. However, this is likely to be testable only in cases of motivated behavior in humans, and it is compatible with the often erratic capability of verbal expressions of craving for drugs. This should not at all be construed to mean that we discount emotional subjective responses and the brain mechanisms mediating them in the analysis of affective behavior. Such responses have to be translated into a (usually subvocal) verbal code, but might also involve nonverbal representations. These processes are presumably the product of interactive cortical mechanisms. Burgeoning evidence links the orbitofrontal cortex to the sensory representation of reinforcers as well as their value and the relative utility of different courses of action producing them. However, the neural mediation of those aspects of the reinforcer conveying its hedonic properties remain elusive because the use of functional neuroimaging procedures thus far has confounded the sensory properties of a reinforcer with hedonic subjective responses associated with it. We assume that these subjective responses arise in some way as a post hoc ‘commentary’ on the sensory representation itself. Defining sensory properties of reinforcers is more difficult than initially seems. For food, it must involve a combination of gustatory, olfactory, somatosensory (textural) and visual elements97. The visual aspects obviously gain their hedonic properties through conditioning. However, it is likely that we have to learn about virtually all of the hedonic properties of food and that even tastes and smells may not be as ‘unconditioned’ as hitherto believed. In addition, the hedonic effects of food may not arise simply from the exteroceptive stimuli themselves but from their capacity to evoke visceral changes, such as alterations in heart rate and other autonomic responses, ‘sensed’ as ‘feelings’ by mechanisms that depend on the insular cortex98.

of self-administered cocaine and that is resistant to extinction of the original CS-drug association23. Drug-associated conditioned reinforcers also help to maintain responding under second-order schedules of reinforcement (Box 2), where drugs are provided only after a prolonged time interval, thus more realistically modeling drug seeking behavior24. Responding in the interim is maintained by the presence of drug-associated CSs that are presented as a consequence of instrumental seeking responses (Box 2). The CSs must be presented as conditioned reinforcers (that is, their presentation must depend on the animal’s behavior); merely presenting them unexpectedly fails to increase drug seeking18. This seems to contradict the ‘incentive salience’ model of drug seeking behavior, which would predict enhancement from pavlovian, or unexpected, presentations of the CS. Drug self-administration behavior, including drug seeking under second-order schedules of reinforcement, initially involves action-

1483

REVIEW BOX 2 DEFINITION OF KEY TERMS


Positive reinforcer. An event that increases the probability of a response on which it is contingent. For example, drug infusions increase the probability of lever pressing for the drug; alcohol ingestion increases the probability of licking or drinking. Negative reinforcer. An event that when omitted or terminated increases the probability of the response on which it is contingent. For example, withdrawal symptoms precipitated by scheduled administration of naloxone in morphine-dependent animals, which can be avoided by lever pressing to postpone the naloxone. Incentive. A stimulus that elicits approach behavior (positive incentive) or withdrawal behavior (negative incentive). A conditioned incentive acquires such properties via pavlovian conditioning. Incentives and conditioned incentives may also function as reinforcers and conditioned reinforcers, respectively, depending on environmental contingencies. Pavlovian (or classical) conditioning. The process by which a conditioned stimulus (CS), such as a tone, after a number of pairings, comes to elicit conditioned responses like salivation that are normally elicited by an unconditioned stimulus, such as food. Such conditioned responses are normally considered to be involuntary reflexes. The pairings require the onset of the CS to occur before the unconditioned stimulus (temporal contiguity) and a positive temporal correlation (predictive contingency) between the two events. Conditioned reinforcer. A stimulus that acquires its reinforcing properties (positive or negative) by pairings with other, generally primary, reinforcers such as food, drugs, sex or electric shock. A stimulus can function as a conditioned reinforcer or discriminative stimulus in the same situation. Contingency. A consistent temporal relationship between two (or more) events that reduces the uncertainty of the subsequent event: for example, a situation in which a tone always occurs at the same time as a shock. Operant. A response on which the presentation of a reinforcer is contingent, such as lever pressing. Such behavior is either called ‘instrumental’ in obtaining a goal (or ‘outcome’ or ‘reinforcer’), or else it is a voluntary action. The learning of such behavior is termed instrumental conditioning. Action-outcome learning. When instrumental actions are goal directed, the actions (lever pressing) are made with the intention of obtaining the goal. The actions are sensitive to devaluation of the goal: for instance, an animal that has learned to lever-press for food will respond much less or not at all for that food if it is devalued either by making the animal ill after ingesting the food, or by pre-feeding it to satiety with the same food. This is called ‘reinforcer devaluation’. It is easy to devalue ingestive reinforcers, but it is much more difficult to devalue intravenously selfadministered drugs such as cocaine. Stimulus-response or ‘habit’ learning. In habit learning, instrumental performance is acquired through the association of responses with stimuli present during training. It therefore reflects the formation of stimulus-response associations, and reinforcers primarily serve the function of strengthening the stimulus-response association but do not become encoded as a goal. Therefore, devaluing the reinforcer does not affect instrumental responding acquired by habit learning.

1484

Pavlovian-instrumental transfer (PIT). Appetitive pavlovian stimuli (associated with positive reinforcers such as food) can greatly enhance instrumental responding for the same reinforcer when presented unexpectedly (independent of the instrumental response), and this defines the pavlovian-instrumental transfer effect. PIT has been interpreted as evidence that CSs exert a motivational influence over instrumental performance. Drug taking and drug seeking. ‘Drug taking’ is a term used to describe drug self-administration when the drug is readily available: for instance, following each lever press or the simple drinking of alcohol (‘continuous reinforcement’). The subject does not need to forage, or to work, for the drug, nor does it need to mediate delays in acquiring it; that is, it does not actively need to ‘seek’ the drug. ‘Drug seeking’ can be studied in a number of ways. A ‘second-order schedule of drug reinforcement’ (in contrast to continuous reinforcement) emphasizes the role of drugassociated conditioned reinforcers in maintaining drug seeking behavior over relatively prolonged periods24. Rats or monkeys (and also humans) are initially trained to self-administer cocaine or heroin under a simple, continuous reinforcement schedule, each drug infusion being paired with a light CS (simple drug taking). Subsequently, the animal responds for periods of time (usually 15 minutes, but occasionally up to an hour) for each infusion of drug, and responding during that period is maintained by response-dependent presentations of the CS, which act as conditioned reinforcers of the instrumental seeking responses; omission of the contingent CS results in a marked decrease in drug seeking. This behavior models aspects of drug seeking in the real world, in which drugs are not immediately available, and drug-associated stimuli reinforce and maintain drug seeking. The ‘reinstatement of drug seeking’ after extinction of the instrumental seeking response (i.e., the decrement in responding caused by non-delivery of the drug)99 or the maintenance of drug seeking responses in the absence of drug100 are widely used procedures because they model a critical aspect of drug addiction: namely, the propensity to relapse after sometimes prolonged periods of withdrawal (or abstinence). The ability of drug-associated conditioned reinforcers to maintain or reinstate drug seeking responses may actually increase with the duration of withdrawal100. Not only drug-associated stimuli but also injections of the drug itself and stressors can all reinstate drug seeking measured in this way. The subject has been reviewed extensively and is not considered in detail here59,60. Recently, another method of measuring drug seeking has adapted the ‘acquisition of a new response’ procedure, in which the ability of a drug-associated CS to support the learning of a new instrumental seeking response is measured23. This procedure models another feature of conditioned reinforcers: namely, their ability to act as goals themselves and thereby support the learning of new behavioral strategies directed toward obtaining the primary reinforcer—in this case, a drug. This behavior is remarkably persistent, as is drug seeking in drugaddicted individuals. Discriminative stimulus. A stimulus in the presence of which responding is reinforced according to some schedule of reinforcement. For example, drug cues can act as discriminative stimuli (i.e., can set the occasion) for behavioral responding that is maintained by drug reinforcement.


REVIEW


outcome learning (Box 2), before extended training additionally leads to the formation of stimulus-response (‘habit’) associations that help maintain responding20,25. Interacting roles of striatal subregions in reinforcement The experiments discussed above and further below show that the nucleus accumbens core, as distinct from the shell, is important in the maintenance of instrumental behavior involving delays in the provision of cocaine, in particular in the capacity of CSs to bridge that delay. This conclusion begs the question of where the drug exerts its primary reinforcing effects. One possibility is the nucleus accumbens shell, which is connected to the full network of descending neural influences over reflexive autonomic and motor responses26,27 (Fig. 1). This idea is consistent with evidence that this region is necessary for the direct psychomotor stimulant effects of the cocaine, including response rate–enhancing and locomotor activity effects10. Some of the unconditioned effects of food reinforcers may be similarly mediated via the nucleus accumbens shell, though involving mechanisms specifically associated with opiate, rather than dopamine, receptors27. Cocaine (and other drugs) have positive reinforcing effects when infused responsedependently directly into another region of the ventral striatum, the olfactory tubercle28. We do not wish necessarily to draw a sharp distinction between the unconditioned and conditioned aspects of reinforcement, as we have already pointed out that these may merge into one another. It is possible, for example, that conditioning to interoceptive aspects of reinforcers, such as taste or smell, does depend on the shell or other regions of the ventral striatum29, whereas the core is especially associated with exteroceptive (for instance, visual) conditioning19. Important issues to be resolved include how the contributory factors such as pavlovian arousal and instrumental reinforcement, including conditioned reinforcement, are integrated within the nucleus accumbens circuitry. Perhaps the most obvious mechanism could stem from the cascading loop circuitry by which output from the nucleus accumbens shell can influence the functioning of the ascending dopamine projections to the core, and similarly, from the output of the core via the substantia nigra to other domains of the dorsal striatum1 (Fig. 1). Thus, several phenomena, such as the potentiation of conditioned reinforcement by stimulant drugs and pavlovian-instrumental transfer during instrumental learning, could arise from the sequential operation of the drug’s impact in the nucleus accumbens shell, influencing processing of CSs in the core. By a similar token, such sequential operations may result in drug seeking (action-outcome learning) that seems to depend on the interaction of the dorsomedial striatum30 with its afferents from specific regions of the medial prefrontal cortex (mPFC)31. Extended training leads to the development of habits, where the emphasis is on slow stimulus-response learning mechanisms with less involvement of the goal itself25 (Box 2). Data for food-maintained habits suggest that yet another sector of the striatum, the dorsolateral striatum, is implicated in habit learning32 (see below). These sequential phases of pavlovian and instrumental learning may be especially relevant for the transition from initial drug use to drug abuse, and finally compulsive drug taking and drug seeking behavior. Compulsive drug seeking and drug taking are the hallmarks of the definitions of drug addiction (or ‘dependence’ in the Diagnostic and Statistical Manual IV), which is becoming increasingly acknowledged by neuroscientists modeling this behavior33–35. In theoretical terms, it seems reasonable to characterize such compulsive behavior as a maladaptive stimulus-response habit in which the ultimate goal of the behavior has been devalued so that the behavior is not directly under the control of the goal20,25. Rather, responding is governed by a succession of discriminative stimuli, which also function—when they are


presented as a consequence of instrumental responses—as conditioned reinforcers. Hypothetically, such stimulus-response associative (‘habit’) learning occurs in parallel with instrumental action-outcome learning but, with extended training, eventually dominates behavioral output. Crucial to drug addiction is the persisting quality of these habits, which has been likened to the subjective state of ‘wanting’15, but which we would suggest corresponds more obviously to the subjective state of ‘must do!’—although this subjective response could arise post hoc as a rationalization of the ‘out-of control’ habitual behavior rather than being the driving influence. These behavioral patterns are maintained by the enhanced significance of drug-associated conditioned reinforcers, which act as discriminative stimuli for continued drug seeking behavior, especially once the drug itself has been taken18,36. The obvious analogy is with obsessive-compulsive disorder. It is, of course, possible, as with obsessive-compulsive behavior, that the habitual behavior is maintained in part by negative reinforcement37; active avoidance behavior in monkeys can have a similarly persistent quality, especially after treatment with psychomotor stimulant drugs38. Two details of this hypothesis are important: it applies to instrumental behavior such as intravenous drug self-administration under a ‘drug seeking’ second-order schedule, and it is not an example of a procedural skill, such as playing the piano or tying one’s shoelace—although it is plausible that such skills result from even more extended training. The analogy with drug addiction would be a persistence or constant reinitiation of such activities. Evidence for this concept of drug addiction as a maladaptive and persistent habit comes from several sources, which also increasingly point to the dorsal striatum as a major contributor to this form of learning. An operational definition of a habit is that the behavior continues even after the controlling influence of the goal is reduced by devaluation procedures, such as satiation or even poisoning in the case of a food goal (Box 2)39. The extent to which instrumental behavior is maintained under these conditions reveals the degree of control by stimulus-response mechanisms. This approach has led to the definition of a role for the dorsolateral striatum32 and its dopaminergic innervation40 in instrumental habit learning in rats. However, devaluing drugs as reinforcers seems to be quite difficult and probably depends on understanding the precise nature of their reinforcing effects (see above and Box 1). This cannot readily be achieved by simple pharmacological antagonism, which does not devalue a reinforcer so much as remove it. In studies of oral cocaine and alcohol self-administration, however, habitual responding—evidenced by resistance to devaluation by gastric malaise—develops more rapidly for a drug than for a food reinforcer41,42. Dopamine release in the nucleus accumbens core and shell, as measured by microdialysis in vivo, is not generally coincident with the provision of drug-paired CSs in rats extensively trained under second-order schedules9, but such conditioned reinforcers do evoke dopamine release in the dorsal striatum36. Thus, although the acquisition of drug seeking under a second-order schedule depends on the nucleus accumbens core, which is part of the ventral striatum, control over performance may ultimately devolve to the dorsal striatum. Indeed, the mixed dopamine receptor antagonist α-flupenthixol infused into the dorsal striatum greatly reduces well-established cocaine seeking under a second-order schedule, yet it has no effect when infused into the nucleus accumbens core43,44. This is consistent with the habit hypothesis and also with the presence of ‘error prediction’ dopamine neurons innervating the entire striatum, including its dorsal as well as ventral regions6. Perhaps of even greater significance is that these findings provide further evidence of devolved control from the shell and core regions of the nucleus accumbens now to include the dorsal striatum, thereby supporting the capability of ventralto-dorsal unidirectional cascades of information processing mediated by corticostriatal ‘loop’ circuitry1. This proposed sequence of changes

1485


REVIEW caused by drug-self-administration is further supported by observations in rhesus monkeys self-administering cocaine over an extended period. The downregulation of striatal dopamine D2 receptors, as well as other cellular markers, known to occur in human chronic cocaine abusers45 can also be observed to occur first in ventral and then in dorsal territories of the striatum in cocaine-taking rhesus monkeys46–48. The findings of both parallel and serial cascading mechanisms of associative learning suggested by these studies require further investigation. They are broadly compatible with functional neuroimaging evidence in humans that the ventral striatum is implicated in pavlovian conditioning and the dorsal striatum with instrumental learning49. From the neurocomputational perspective, the ventral and dorsal striatum could conceivably correspond to the ‘critic’ and ‘actor’ components, respectively, of contemporary models of reinforcement learning49. The critic learns to predict future rewards, and the actor maintains information about the rewarding outcome of actions; in other words, the interaction of pavlovian and instrumental learning through the intermediary of conditioned reinforcement. However, it would be very misleading to imply that it is only the striatum that is implicated in these aspects of learning. First, of course, the striatum is also implicated in performance, involving the retrieval of appropriate stimulus-response rules and goal representations. Second, the striatum is only one part of a much more extended network defined by its intimate, roughly topographical inputs from limbic cortical structures, such as the basolateral amygdala, the hippocampus and the prefrontal cortex, that are primarily focused on the ventral striatum and discrete regions of the dorsal striatum, as well as from neocortical areas.

data indicate that associative information in the basolateral amygdala is translated into goal-directed, drug seeking behavior via its interactions with the nucleus accumbens core (Fig. 1). Selective lesions of the orbital prefrontal cortex (OFC) also impair the acquisition of cocaine seeking56 and responding with conditioned reinforcement57, but without affecting continuously reinforced cocaine self-administration56. The OFC and basolateral amygdala are richly interconnected, as well as projecting to the nucleus accumbens core and overlying anterior dorsal striatum (Fig. 1). The observation that basolateral amygdala and OFC are involved, along with the nucleus accumbens core, in the neural mechanisms underlying the ability to seek drugs over long delays bridged by conditioned reinforcers is consistent with a growing body of data that the basolateral amygdala and OFC cooperate to regulate goal-directed behavior58. Studies of the cued reinstatement of extinguished responding for cocaine (Box 2) have been reviewed extensively elsewhere59,60. These also emphasize the involvement in drug seeking of the basolateral amygdala61, lateral OFC62 and nucleus accumbens core63 as well as dopamine and glutamate transmission in the basolateral amygdala and nucleus accumbens core64,65—but in the context of relapse, which is a key aspect of drug addiction.

The basolateral amygdala–nucleus accumbens core system Selective lesions of the basolateral amygdala or the nucleus accumbens core impair the acquisition of cocaine or heroin seeking under a second-order schedule50–53. These studies also show that continuously instrumental responding for cocaine is completely unaffected by core lesions, consistent with other evidence that this region is not directly implicated in instrumental learning per se: lesion-induced deficits are found only when the drug infusions are delayed. Thus, the mechanisms underlying drug taking are dissociable from those underlying drug seeking. The effects of lesions in basolateral amygdala or nucleus accumbens are likely to reflect the interacting roles of these structures in conditioned reinforcement19 and also their roles in mediating delays to reinforcement. Basolateral amygdala lesions, like nucleus accumbens core lesions, increase the choice of small, immediate rewards over larger, delayed rewards—indicating greater impulsivity54. The core is also necessary for instrumental learning when there is a delay between the response and the reinforcer55. Presumably, it acts by allowing CSs occurring during the delay (either discrete, or forming part of the context) to act as conditioned reinforcers for instrumental responding, leading to the reward. Using a disconnection procedure (unilateral manipulation of structures within a putative neural system, but on opposite sides of the brain), we have shown the functional importance of serial interactions between the basolateral amygdala and nucleus accumbens core in drug seeking sustained by conditioned reinforcers43. Dopamine (but not AMPA) receptor blockade bilaterally in the basolateral amygdala impairs cocaine seeking under a second-order schedule, whereas AMPA (but not dopamine) receptor blockade bilaterally in the nucleus accumbens core has a similar effect. Most importantly, unilateral blockade of dopamine receptors in the basolateral amygdala combined with unilateral blockade of AMPA receptors in the core in the contralateral hemisphere (neither of which has any effect alone) reduces cue-controlled cocaine seeking as much as bilateral manipulations of either structure43. These

Hippocampus—nucleus accumbens system There is general consensus on the functions of the amygdala, nucleus accumbens core and OFC and their interactions in the control over goaldirected behavior by discrete CSs acting as conditioned reinforcers. In contrast, the hippocampal formation, which is also a major source of glutamatergic afferents to the nucleus accumbens, especially the nucleus accumbens shell26 (Fig. 1), has received somewhat less attention in studies of drug seeking. Inactivation of the dorsal hippocampus prevents the reinstatement of extinguished responding for cocaine by contextual stimuli, but not by discrete CSs66. Theta-burst stimulation of the hippocampus reinstates extinguished cocaine seeking, acting via glutamatergic transmission in the VTA, which was suggested to mimic the way that reinstatement occurs when animals are placed in a context associated with drug taking, rather than in response to discrete cocaine cues67. These data are generally consistent with the view that, whereas the amygdala mediates conditioning to discrete CSs, the hippocampal formation underlies conditioning to contextual or spatial stimuli68 and may therefore underlie the motivational impact of contextual stimuli on drug seeking. Hippocampal contextual information and amygdala-dependent discrete CSs may compete for control over goal-directed behavior3. Thus, amygdala lesions not only impair appetitive behavioral responses under the control of discrete CSs but also result in enhanced control by contextual cues; similarly, hippocampal lesions impair contextual conditioning but can also result in enhanced conditioning to discrete CSs3 (R. Ito, T.W.R., B.L. McNaughton & B.J.E., unpublished observations). The neural basis of such competition between associative influences on behavior is unclear, but may depend upon the projections of the basolateral amygdala and hippocampus to the nucleus accumbens core and shell69. Electrophysiological and in vivo neurochemical studies show that hippocampal, amygdala and PFC projections interact in the nucleus accumbens. This interaction is modulated by mesolimbic dopamine and, in turn, can modulate the release of dopamine70–73. Indeed, D1 and D2 dopamine receptors differentially regulate the influence of mPFC versus hippocampal afferents on the activity of nucleus accumbens neurons, and this modulation influences performance in appetitive behavioral tasks70. Dorsal subiculum–lesioned rats are hyperactive in tests of exploratory locomotion, whereas ventral subiculum–lesioned rats show an attenuated locomotor response to amphetamine and impaired acquisi-

1486



REVIEW tion of cocaine self-administration74. Lesions of the ventral subiculum also completely abolish the locomotor response to intra-accumbens infusions of D-amphetamine in addition to blocking the potentiative effect of the same treatment on responding with conditioned reinforcement75. These data suggest a key role for hippocampal projections to the nucleus accumbens, especially the shell, in regulating its dopaminergic tone and mediating the psychomotor stimulant effects of drugs such as amphetamine and cocaine10,53. We hypothesize that, in psychological terms, hippocampal mechanisms provide the contextual background that defines the motivational arousal upon which goal-directed responding occurs. Inactivating this mechanism at the hippocampus or nucleus accumbens shell level reduces exploration, activity and contextual conditioning and also the potentiation of these responses by psychomotor stimulants—providing, therefore, an additional basis for understanding the reinforcing effects of drugs acting on the dopamine and other systems in the nucleus accumbens (see above). The prefrontal executive system By far the most detailed investigation of the prefrontal cortex (PFC) in drug seeking measures the reinstatement of cocaine seeking after extinction. Inactivation of the dorsomedial part of the PFC prevents the reinstatement of responding elicited by drug cues, contexts, priming injections of drugs or stress59,60. Moreover, the involvement of the mPFC in the reinstatement of drug seeking depends on glutamate release in the nucleus accumbens core and also on the integrity of the ventral pallidum, providing clear evidence of the function of specific limbic cortical-ventral striatopallidal circuits59,64. Hippocampal, amygdala and mPFC mechanisms may therefore all influence drug seeking through their convergent projections to the nucleus accumbens (Fig. 1), perhaps competing for access to response strategies dependent on different limbic cortical–striatal circuitries70. Lesions of the mPFC (including the prelimbic and infralimbic cortex) result in increased responding for cocaine under a second-order schedule of reinforcement and also enhance the acquisition of cocaine selfadministration76. However, these effects of mPFC lesions seem unlikely to result from any change in conditioned reinforcement75 and may reflect instead an impairment in executive control over behavior (including behavioral inhibition processes)77. This is consistent with burgeoning neuroimaging and neuropsychological evidence from human studies suggesting that chronic drug abusers show deficits in tests of inhibitory control and decision making78–80. Studies using PET, especially, highlight changes in metabolism in the OFC in abstinent drug abusers45,81, but despite the involvement of this region in reinforcer processing, few experiments have examined its role in controlling drug seeking behavior in animals. Distinct changes in neuronal plasticity in this region, however, do result from chronic stimulant self-administration82. Overall, we hypothesize that the transition from voluntary actions (governed mainly by their consequences) to more habitual modes of responding in drug seeking behavior represents a transition from prefrontal cortical to striatal control over responding, and from ventral to more dorsal striatal subregions (Fig. 1). Some of that transition may reflect important changes in the balance of activity in those brain regions mediating the executive control over behavior: for example, in the acquisition of action-outcome learning itself, the detection of altered instrumental contingencies with associated changes in subjective attribution, and in related processes of goal revaluation by components of the prefrontal cortex, including prelimbic and infralimbic regions31,77,83,84. Impairments in such processes, perhaps arising in part as the direct consequence of toxic drug effects, may contribute to the shift in balance of behavioral control processes toward those promoting habitual behavior. This hypothesis is plausibly supported in


neural terms by neuroimaging data in humans showing reductions in the activity of the prefrontal cortex, including the orbitofrontal region of abstinent addicts45,81. Habitual responding by itself, however, does not capture the persistent, indeed, compulsive aspects of ‘out-of-control’ drug bingeing; some additional factor seems to be required. In the ‘incentive-sensitization’ model, the potentiated responding is postulated to depend on druginduced sensitization of behavior85. A related but distinct view is that the drug effect itself may produce the enhanced drive to responding, thus prolonging the duration of a drug taking binge. On this account, sensitization reflects the normal processes of tolerance and inverse tolerance that modify the effects of many drugs. Whether sensitization directly affects instrumental drug self-administration seems less clear, although sensitization can augment responding for conditioned reinforcers enhanced by intra-accumbens amphetamine86 and increase ‘break points’ (the highest number of responses rats will make) for cocaine assessed using a progressive ratio schedule87. According to the DSM-IV, another characteristic of drug addiction is that it persists despite adverse consequences. This, too, has been modeled in rats, which continue to seek cocaine after a prolonged, but not brief, drug taking history in the face of conditioned or unconditioned aversive stimuli34,35. Alternatively, as in the case of obsessive-compulsive disorder itself, which has similarly been associated with dysfunctional orbitofrontal-striatal circuitry, it may be necessary to postulate a source of negative reinforcement that maintains responding, for example, through opponent motivational systems also engaged by drug abuse88,89. How such systems interact at a neural level with those inducing the habitual appetitive behavior associated with drug addiction remains a central question for future research. ACKNOWLEDGMENTS The research is supported by the UK Medical Research Council. We acknowledge the major contributions of H. Alderson, M. Arroyo, R. Cardinal, J. Dalley, P. Di Ciano, A. Dickinson, L. Fattore, J. Hall, K. Hellemans, D. Hutcheson, R. Ito, J. Lee, F. Miles, C. Olmstead, J. Parkinson, M. Pilla, Y. Pelloux, K. Thomas, L. Vanderschuren, R. Weissenborn and R. Whitelaw. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1.

Haber, S.N., Fudge, J.L. & McFarland, N.R. Striatonigral pathways in primates form an ascending spiral from the shell to the dorsolateral striatum. J. Neurosci. 20, 2369–2382 (2000). 2. Dickinson, A. & Balleine, B. Motivational control of goal-directed action. Anim. Learn. Behav. 22, 1–18 (1994). 3. White, N.M. & McDonald, R.J. Multiple parallel memory systems in the brain of the rat. Neurobiol. Learn. Mem. 77, 125–184 (2002). 4. O’Brien, C.P. & McLellan, A.T. Myths about the treatment of addiction. Lancet 347, 237–240 (1996). 5. Leshner, A.I. Addiction is a brain disease, and it matters. Science 278, 45–47 (1997). 6. Schultz, W. & Dickinson, A. Neuronal coding of prediction errors. Annu. Rev. Neurosci. 23, 473–500 (2000). 7. Fiorillo, C.D., Tobler, P.N. & Schultz, W. Discrete coding of reward probability and uncertainty by dopamine neurons. Science 299, 1898–1902 (2003). 8. Robbins, T.W. & Everitt, B.J. Functions of dopamine in the dorsal and ventral striatum. Seminars in the Neurosciences 4, 119–128 (1992). 9. Ito, R., Dalley, J.W., Howes, S.R., Robbins, T.W. & Everitt, B.J. Dissociation in conditioned dopamine release in the nucleus accumbens core and shell in response to cocaine cues and during cocaine-seeking behavior in rats. J. Neurosci. 20, 7489– 7495 (2000). 10. Parkinson, J.A., Olmstead, M.C., Burns, L.H., Robbins, T.W. & Everitt, B.J. Dissociation in effects of lesions of the nucleus accumbens core and shell on appetitive Pavlovian approach behavior and the potentiation of conditioned reinforcement and locomotor activity by D-amphetamine. J. Neurosci. 19, 2401–2411 (1999). 11. Di Ciano, P., Cardinal, R.N., Cowell, R.A., Little, S.J. & Everitt, B.J. Differential involvement of NMDA, AMPA/kainate, and dopamine receptors in the nucleus

1487

REVIEW

12.

13.


14.

15. 16. 17.

18.

19. 20. 21.

22.

23.

24.

25. 26.

27. 28.

29.

30. 31.

32.

33. 34. 35. 36.

37. 38.

39.

40.

41.

accumbens core in the acquisition and performance of Pavlovian approach behavior. J. Neurosci. 21, 9471–9477 (2001). Dalley, J.W. et al. Time-limited modulation of appetitive Pavlovian memory by D1 and NMDA receptors in the nucleus accumbens. Proc. Natl. Acad. Sci. USA 102, 6189–6194 (2005). Hall, J., Parkinson, J.A., Connor, T.M., Dickinson, A. & Everitt, B.J. Involvement of the central nucleus of the amygdala and nucleus accumbens core in mediating Pavlovian influences on instrumental behaviour. Eur. J. Neurosci. 13, 1984–1992 (2001). Wyvell, C.L. & Berridge, K.C. Intra-accumbens amphetamine increases the conditioned incentive salience of sucrose reward: enhancement of reward “wanting” without enhanced “liking” or response reinforcement. J. Neurosci. 20, 8122–8130 (2000). Robinson, T.E. & Berridge, K.C. The neural basis of drug craving: an incentivesensitization theory of addiction. Brain Res. Brain Res. Rev. 18, 247–291 (1993). Kearns, D.N. & Weiss, S.J. Sign-tracking (autoshaping) in rats: A comparison of cocaine and food as unconditioned stimuli. Learn. Behav. 32, 463–476 (2004). Grimm, J.W., Kruzich, P.J. & See, R.E. Contingent access to stimuli associated with cocaine self- administration is required for reinstatement of drug-seeking behavior. Psychobiology 28, 383–386 (2000). Di Ciano, P. & Everitt, B.J. Differential control over drug-seeking behavior by drugassociated conditioned reinforcers and discriminative stimuli predictive of drug availability. Behav. Neurosci. 117, 952–960 (2003). Cardinal, R.N. & Everitt, B.J. Neural and psychological mechanisms underlying appetitive learning: links to drug addiction. Curr. Opin. Neurobiol. 14, 156–162 (2004). Everitt, B.J., Dickinson, A. & Robbins, T.W. The neuropsychological basis of addictive behaviour. Brain Res. Brain Res. Rev. 36, 129–138 (2001). Cardinal, R.N., Robbins, T.W. & Everitt, B.J. The effects of d-amphetamine, chlordiazepoxide, alpha- flupenthixol and behavioural manipulations on choice of signalled and unsignalled delayed reinforcement in rats. Psychopharmacology (Berl.) 152, 362–375 (2000). Taylor, J.R. & Robbins, T.W. Enhanced behavioural control by conditioned reinforcers following microinjections of d-amphetamine into the nucleus accumbens. Psychopharmacology (Berl.) 84, 405–412 (1984). Di Ciano, P. & Everitt, B.J. Conditioned reinforcing properties of stimuli paired with self-administered cocaine, heroin or sucrose: implications for the persistence of addictive behaviour. Neuropharmacology 47 (Suppl.) 202–213 (2004). Everitt, B.J. & Robbins, T.W. Second-order schedules of drug reinforcement in rats and monkeys: measurement of reinforcing efficacy and drug-seeking behaviour. Psychopharmacology (Berl.) 153, 17–30 (2000). Robbins, T.W. & Everitt, B.J. Drug addiction: bad habits add up. Nature 398, 567–570 (1999). Groenewegen, H.J., Wright, C.I. & Beijer, A.V.J. The nucleus accumbens: Gateway for limbic structures to reach the motor system? Prog. Brain Res. 107, 485–511 (1996). Kelley, A.E. & Berridge, K.C. The neuroscience of natural rewards: Relevance to addictive drugs. J. Neurosci. 22, 3306–3311 (2002). Ikemoto, S., Qin, M. & Liu, Z.H. The functional divide for primary reinforcement of D-amphetamine lies between the medial and lateral ventral striatum: Is the division of the accumbens core, shell, and olfactory tubercle valid? J. Neurosci. 25, 5061–5065 (2005). Fenu, S., Bassareo, V. & Di Chiara, G. A role for dopamine D1 receptors of the nucleus accumbens shell in conditioned taste aversion learning. J. Neurosci. 21, 6897–6904 (2001). Yin, H.H., Ostlund, S.B., Knowlton, B.J. & Balleine, B.W. The role of the dorsomedial striatum in instrumental conditioning. Eur. J. Neurosci. 22, 513–523 (2005). Ostlund, S.B. & Balleine, B.W. Lesions of medial prefrontal cortex disrupt the acquisition but not the expression of goal-directed learning. J. Neurosci. 25, 7763–7770 (2005). Yin, H.H., Knowlton, B.J. & Balleine, B.W. Lesions of dorsolateral striatum preserve outcome expectancy but disrupt habit formation in instrumental learning. Eur. J. Neurosci. 19, 181–189 (2004). Wolffgramm, J. & Heyne, A. From controlled drug intake to loss of control - the irreversible development of drug-addiction in the rat. Behav. Brain Res. 70, 77–94 (1995). Deroche-Gamonet, V., Belin, D. & Piazza, P.V. Evidence for addiction-like behavior in the rat. Science 305, 1014–1017 (2004). Vanderschuren, L.J. & Everitt, B.J. Drug seeking becomes compulsive after prolonged cocaine self-administration. Science 305, 1017–1019 (2004). Ito, R., Dalley, J.W., Robbins, T.W. & Everitt, B.J. Dopamine release in the dorsal striatum during cocaine-seeking behavior under the control of a drug-associated cue. J. Neurosci. 22, 6247–6253 (2002). Jakes, I. Theoretical Approaches to Obsessive-Compulsive Disorder (Cambridge Univ. Press, Cambridge, 1996). Barrett, J.E., Katz, J.L. & Glowa, J.R. Effects of D-amphetamine on responding of squirrel-monkeys maintained under 2nd-order schedules of food presentation, electric-shock presentation or stimulus-shock termination. J. Pharmacol. Exp. Ther. 218, 692–700 (1981). Dickinson, A., Nicholas, D.J. & Adams, C.D. The effect of instrumental training contingency on susceptibility to reinforcer devaluation. Q. J. Exp. Psychol. 35B, 35–51 (1983). Faure, A., Haberland, U., Conde, F. & El Massioui, N. Lesion to the nigrostriatal dopamine system disrupts stimulus-response habit formation. J. Neurosci. 25, 2771–2780 (2005). Dickinson, A., Wood, N. & Smith, J.W. Alcohol seeking by rats: Action or habit? Q. J. Exp. Psychol. B 55, 331–348 (2002).

1488

42. Miles, F.J., Everitt, B.J. & Dickinson, A. Oral cocaine seeking by rats: action or habit? Behav. Neurosci. 117, 927–938 (2003). 43. Di Ciano, P. & Everitt, B.J. Direct interactions between the basolateral amygdala and nucleus accumbens core underlie cocaine-seeking behavior by rats. J. Neurosci. 24, 7167–7173 (2004). 44. Vanderschuren, L.M.J., Di Ciano, P. & Everitt, B.J. Involvement of the dorsal striatum in cue-controlled cocaine seeking. J. Neurosci. 25, 8665-8770 (2005). 45. Goldstein, R.Z. & Volkow, N.D. Drug addiction and its underlying neurobiological basis: Neuroimaging evidence for the involvement of the frontal cortex. Am. J. Psychiatry 159, 1642–1652 (2002). 46. Letchworth, S.R., Nader, M.A., Smith, H.R., Friedman, D.P. & Porrino, L.J. Progression of changes in dopamine transporter binding site density as a result of cocaine self-administration in rhesus monkeys. J. Neurosci. 21, 2799–2807 (2001). 47. Porrino, L.J., Lyons, D., Smith, H.R., Daunais, J.B. & Nader, M.A. Cocaine selfadministration produces a progressive involvement of limbic, association, and sensorimotor striatal domains. J. Neurosci. 24, 3554–3562 (2004). 48. Nader, M.A. et al. Effects of cocaine self-administration on striatal dopamine systems in rhesus monkeys: Initial and chronic exposure. Neuropsychopharmacology 27, 35–46 (2002). 49. O’Doherty, J. et al. Dissociable roles of ventral and dorsal striatum in instrumental conditioning. Science 304, 452–454 (2004). 50. Whitelaw, R.B., Markou, A., Robbins, T.W. & Everitt, B.J. Excitotoxic lesions of the basolateral amygdala impair the acquisition of cocaine-seeking behaviour under a second-order schedule of reinforcement. Psychopharmacology (Berl.) 127, 213–224 (1996). 51. Alderson, H.L., Robbins, T.W. & Everitt, B.J. The effects of excitotoxic lesions of the basolateral amygdala on the acquisition of heroin-seeking behaviour in rats. Psychopharmacology (Berl.) 153, 111–119 (2000). 52. Hutcheson, D.M., Parkinson, J.A., Robbins, T.W. & Everitt, B.J. The effects of nucleus accumbens core and shell lesions on intravenous heroin self-administration and the acquisition of drug-seeking behaviour under a second-order schedule of heroin reinforcement. Psychopharmacology (Berl.) 153, 464–472 (2001). 53. Ito, R., Robbins, T.W. & Everitt, B.J. Differential control over cocaine-seeking behavior by nucleus accumbens core and shell. Nat. Neurosci. 7, 389–397 (2004). 54. Winstanley, C.A., Theobald, D.E.H., Cardinal, R.N. & Robbins, T.W. Contrasting roles of basolateral amygdala and orbitofrontal cortex in impulsive choice. J. Neurosci. 24, 4718–4722 (2004). 55. Cardinal, R.N. & Cheung, T.H. Nucleus accumbens core lesions retard instrumental learning and performance with delayed reinforcement in the rat. BMC Neurosci. 6, 9 (2005). 56. Hutcheson, D.M. & Everitt, B.J. The effects of selective orbitofrontal cortex lesions on the acquisition and performance of cue-controlled cocaine seeking in rats. Ann. NY Acad. Sci. 1003, 410–411 (2003). 57. Pears, A., Parkinson, J.A., Hopewell, L., Everitt, B.J. & Roberts, A.C. Lesions of the orbitofrontal but not medial prefrontal cortex disrupt conditioned reinforcement in primates. J. Neurosci. 23, 11189–11201 (2003). 58. Schoenbaum, G., Setlow, B., Saddoris, M.P. & Gallagher, M. Encoding predicted outcome and acquired value in orbitofrontal cortex during cue sampling depends upon input from basolateral amygdala. Neuron 39, 855–867 (2003). 59. Kalivas, P.W. & McFarland, K. Brain circuitry and the reinstatement of cocaine-seeking behavior. Psychopharmacology (Berl.) 168, 44–56 (2003). 60. Shaham, Y., Shalev, U., Lu, L., de Wit, H. & Stewart, J. The reinstatement model of drug relapse: history, methodology and major findings. Psychopharmacology (Berl.) 168, 3–20 (2003). 61. Meil, W.M. & See, R.E. Lesions of the basolateral amygdala abolish the ability of drug associated cues to reinstate responding during withdrawal from self-administered cocaine. Behav. Brain Res. 87, 139–148 (1997). 62. Fuchs, R.A., Evans, K.A., Parker, M.P. & See, R.E. Differential involvement of orbitofrontal cortex subregions in conditioned cue-induced and cocaine-primed reinstatement of cocaine seeking in rats. J. Neurosci. 24, 6600–6610 (2004). 63. McFarland, K. & Kalivas, P.W. The circuitry mediating cocaine-induced reinstatement of drug seeking behavior. J. Neurosci. 21, 8655–8663 (2001). 64. McFarland, K., Lapish, C.C. & Kalivas, P.W. Prefrontal glutamate release into the core of the nucleus accumbens mediates cocaine-induced reinstatement of drug seeking behavior. J. Neurosci. 23, 3531–3537 (2003). 65. See, R.E., Kruzich, P.J. & Grimm, J.W. Dopamine, but not glutamate, receptor blockade in the basolateral amygdala attenuates conditioned reward in a rat model of relapse to cocaine-seeking behavior. Psychopharmacology (Berl.) 154, 301–310 (2001). 66. Fuchs, R.A. et al. The role of the dorsomedial prefrontal cortex, basolateral amygdala, and dorsal hippocampus in contextual reinstatement of cocaine seeking in rats. Neuropsychopharmacology 30, 296–309 (2005). 67. Vorel, S.R., Liu, X., Hayes, R.J., Spector, J.A. & Gardner, E.L. Relapse to cocaine-seeking after hippocampal theta burst stimulation. Science 292, 1175–1178 (2001). 68. Selden, N.R.W., Everitt, B.J., Jarrard, L.E. & Robbins, T.W. Complementary roles for the amygdala and hippocampus in aversive conditioning to explicit and contextual cues. Neuroscience 42, 335–350 (1991). 69. Pennartz, C. M.A., Groenewegan, H. J. & Lopas da Silva, F. H. The nucleus accumbens as a complex of functionally distinct neuronal ensembles: an integration of behavioural, electrophysiological and anatomical data. Prog. Neurobiol. 42, 719–761 (1994). 70. Goto, Y. & Grace, A.A. Dopaminergic modulation of limbic and cortical drive of nucleus accumbens in goal-directed behavior. Nat. Neurosci. 8, 805–812 (2005).



REVIEW 71. Grace, A.A., Floresco, S.B., West, A.R. & Goto, Y. Dissociation of tonic and phasic dopamine neuron activity by afferent pathway activation: Relationship to patterns of dopamine release. Int. J. Neuropsychopharmacol. 7, S15–S15 (2004). 72. O’Donnell, P. Dopamine gating of forebrain neural ensembles. Eur. J. Neurosci. 17, 429–435 (2003). 73. Floresco, S.B., Blaha, C.D., Yang, C.R. & Phillips, A.G. Modulation of hippocampal and amygdalar-evoked activity of nucleus accumbens neurons by dopamine: cellular mechanisms of input selection. J. Neurosci. 21, 2851–2860 (2001). 74. Caine, S.B., Humby, T., Robbins, T.W. & Everitt, B.J. Behavioral effects of psychomotor stimulants in rats with dorsal or ventral subiculum lesions: Locomotion, cocaine self- administration, and prepulse inhibition of startle. Behav. Neurosci. 115, 880–894 (2001). 75. Burns, L.H., Robbins, T.W. & Everitt, B.J. Differential effects of excitotoxic lesions of the basolateral amygdala, ventral subiculum and medial prefrontal cortex on responding with conditioned reinforcement and locomotor activity potentiated by intra-accumbens infusions of D-amphetamine. Behav. Brain Res. 55, 167–183 (1993). 76. Weissenborn, R., Robbins, T.W. & Everitt, B.J. Effects of medial prefrontal or anterior cingulate cortex lesions on responding for cocaine under fixed-ratio and second-order schedules of reinforcement in rats. Psychopharmacology (Berl.) 134, 242–257 (1997). 77. Dalley, J.W., Cardinal, R.N. & Robbins, T.W. Prefrontal executive and cognitive functions in rodents: neural and neurochemical substrates. Neurosci. Biobehav. Rev. 28, 771–784 (2004). 78. Rogers, R.D. & Robbins, T.W. Investigating the neurocognitive deficits associated with chronic drug misuse. Curr. Opin. Neurobiol. 11, 250–257 (2001). 79. Bolla, K.I. et al. Orbitofrontal cortex dysfunction in abstinent cocaine abusers performing a decision-making task. Neuroimage 19, 1085–1094 (2003). 80. Hester, R. & Garavan, H. Executive dysfunction in cocaine addiction: evidence for discordant frontal, cingulate, and cerebellar activity. J. Neurosci. 24, 11017–11022 (2004). 81. Volkow, N.D., Fowler, J.S. & Wang, G.J. The addicted human brain viewed in the light of imaging studies: brain circuits and treatment strategies. Neuropharmacology 47, 3–13 (2004). 82. Crombag, H.S., Gorny, G., Li, Y.L., Kolb, B. & Robinson, T.E. Opposite effects of amphetamine self-administration experience on dendritic spines in the medial and orbital prefrontal cortex. Cereb. Cortex 15, 341–348 (2005). 83. Killcross, S. & Coutureau, E. Coordination of actions and habits in the medial prefrontal cortex of rats. Cereb. Cortex 13, 400–408 (2003).


84. Corbit, L.H. & Balleine, B.W. The role of prelimbic cortex in instrumental conditioning. Behav. Brain Res. 146, 145–157 (2003). 85. Robinson, T.E. & Berridge, K.C. Addiction. Annu. Rev. Psychol. 54, 25–53 (2003). 86. Taylor, J.R. & Horger, B.A. Enhanced responding for conditioned reward produced by intra-accumbens amphetamine is potentiated after cocaine sensitization. Psychopharmacology (Berl.) 142, 31–40 (1999). 87. Vezina, P. Sensitization of midbrain dopamine neuron reactivity and the selfadministration of psychomotor stimulant drugs. Neurosci. Biobehav. Rev. 27, 827– 839 (2004). 88. Koob, G.F. & Le Moal, M. Drug addiction, dysregulation of reward, and allostasis. Neuropsychopharmacology 24, 97–129 (2001). 89. Koob, G.F. et al. Neurobiological mechanisms in the transition from drug use to drug dependence. Neurosci. Biobehav. Rev. 27, 739–749 (2004). 90. The Frontal Lobes and Neuropsychiatric Illness (Salloway, S.P., Malloy, P.E. & Duffy, J.D., eds.) (American Psychiatric Press, Washington, D.C., 2001). 91. Altman, J. et al. The biological, social and clinical bases of drug addiction: Commentary and debate. Psychopharmacology (Berl.) 125, 285–345 (1996). 92. Everitt, B.J. & Wolf, M.E. Psychomotor stimulant addiction: A neural systems perspective. J. Neurosci. 22, 3312–3320 (2002). 93. Stewart, J., de Wit, H. & Eikelboom, R. The role of unconditioned and conditioned drug effects in the self administration of opiates and stimulants. Psychol. Rev. 91, 251–268 (1984). 94. Di Chiara, G. A motivational learning hypothesis of the role of mesolimbic dopamine in compulsive drug use. J. Psychopharmacol. 12, 54–67 (1998). 95. Wise, R.A. Dopamine, learning and motivation. Nat. Rev. Neurosci. 5, 483–494 (2004). 96. Berridge, K.C. Pleasures of the brain. Brain Cogn. 52, 106–128 (2003). 97. Kadohisa, M., Rolls, E.T. & Verhagen, J.V. Orbitofrontal cortex: Neuronal representation of oral temperature and capsaicin in addition to taste and texture. Neuroscience 127, 207–221 (2004). 98. Critchley, H.D., Wiens, S., Rotshtein, P., Ohman, A. & Dolan, R.J. Neural systems supporting interoceptive awareness. Nat. Neurosci. 7, 189–195 (2004). 99. Stewart, J. & de Wit, H. in Methods of Assessing the Reinforcing Properties of Abused Drugs (ed. Bozarth, M.A.) 211–227 (Springer-Verlag, New York, 1987). 100. Lu, L., Grimm, J.W., Hope, B.T. & Shaham, Y. Incubation of cocaine craving after withdrawal: a review of preclinical data. Neuropharmacology 47, 214–226 (2004).

1489


E R R AT U M

Erratum: Neural systems of reinforcement for drug addition: from actions to habits to compulsion Barry J Everitt & Trevor W Robbins Nature Neuroscience 8, 1481–1489 (2005); corrected after print, 31 March 2006 In the version of this article initially published, there was an error in Figure 1. The correct version of the figure is below. The error has been corrected in the HTML and PDF versions of the article.

Acb shell

NATURE NEUROSCIENCE

Acb core

Dorsal striatum


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

The cerebellum communicates with the basal ganglia Eiji Hoshi1–3, León Tremblay4, Jean Fe´ger4, Peter L Carras1,2 & Peter L Strick1,2,5 The cerebral cortex is interconnected with two major subcortical structures: the basal ganglia and the cerebellum. How and where cerebellar circuits interact with basal ganglia circuits has been a longstanding question. Using transneuronal transport of rabies virus in macaques, we found that a disynaptic pathway links an output stage of cerebellar processing, the dentate nucleus, with an input stage of basal ganglia processing, the striatum. The basal ganglia and cerebellum are two major subcortical structures that influence multiple aspects of motor, cognitive and affective behavior1–5. Both structures are densely interconnected with the cerebral cortex. For example, large numbers of cortical neurons project to the input stages of the basal ganglia (the caudate and putamen) and the cerebellum (the pontine nuclei). Similarly, the output stages of the basal ganglia (the internal segment of the globus pallidus (GPi) and the substantia nigra pars reticulata) and the cerebellum (the deep cerebellar nuclei) project to subdivisions of the ventroanterior and ventrolateral thalamus6,7. These regions of the thalamus then project back upon the cerebral cortex. Thus, a major architectural feature of these circuits is the formation of multiple ‘loops’ between cerebral cortex and basal ganglia and between cerebral cortex and cerebellum. Basal ganglia and cerebellar loops are believed to operate largely in isolation from one another because the outputs from the two circuits project to neighboring, but separate, thalamic nuclei6,7. The major site for interaction between these circuits was thought to be at the level of Figure 1 Tracer injection sites. Rabies virus (N2C strain) was injected into different locations within the basal ganglia: (a) putamen, animal GP7; (b,c) external segment of the globus pallidus (GPe), animals GP1 and GP4. The survival time was 40 h in a and c and 50 hours in b (Supplementary Fig. 1). The shaded ellipse in each panel indicates the injection site; the dashed line indicates the track of the injection cannula. Scale bars, 1 mm. AC, anterior commisure; C, caudate; GPe, external segment of the globus pallidus; GPi, internal segment of the globus pallidus; P, putamen. AC – 1.0: 1.0 mm caudal to AC. AC – 1.5 mm: 1.5 mm caudal to AC.

a

b

GP7 AC – 1.0

the cerebral cortex. We now provide evidence for a pathway that enables the output stage of cerebellar processing to have a direct influence over the input stage of basal ganglia processing. We injected the N2C strain of rabies virus into sites within the basal ganglia of six macaque monkeys (Fig. 1, Supplementary Methods and Supplementary Fig. 1). Rabies virus is transported transneuronally in a time-dependent fashion in the CNS of nonhuman primates8. We used this feature of the virus to determine if neurons in the deep cerebellar nuclei project to the basal ganglia and to define the links in this connection. All experimental procedures were approved by the institutional animal care and biosafety committees of the University of Pittsburgh and were in accordance with the regulations detailed in the US National Institutes of Health Guide for the Care and Use of Laboratory Animals. In the first experiments, we injected a small amount of the N2C strain of rabies into the putamen (Fig. 1a, n ¼ 2; Supplementary Figs. 1,2) and allowed the animals to survive for 40 h. With the N2C strain, 40 h is long enough for retrograde transport of the virus from the injection site to ‘first-order’ neurons and, subsequently, retrograde transneuronal transport to ‘second-order’ neurons that innervate the first-order neurons. After the putamen injections we observed retrograde transport of the virus to first-order neurons in the thalamus and then retrograde transneuronal transport from these first-order neurons to second-order neurons in the deep cerebellar nuclei. Although our putamen injections were relatively small and localized (Fig. 1a, Supplementary Fig. 2), we labeled an average of 149 neurons in the cerebellar nuclei. This number reflects counts from every other section through the cerebellum. Of these labeled neurons, 88% were located in the contralateral nuclei; 67% of the contralateral neurons were located in the dentate, 29% were in interpositus and 4% were in fastigial. The labeled neurons were most concentrated in dorsal and ventral portions of the rostral dentate. The morphology of these labeled neurons was typical of dentate neurons that project to cortex via the thalamus9.

C

c

GP1 AC – 1.0

C

AC – 1.5 C P

P

P

GP4

GPe GPe

GPe GPi

1Department of Neurobiology and 2Center for the Neural Basis of Cognition, University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania 15261, USA. 3Tamagawa University Research Institute, Machida, Tokyo 194-8610 Japan. 4Neurologie et The´rapeutique expe´rimentale (INSERM U679), Hoˆpital de la Salpeˆtrie`re, 47 boulevard de l’Hoˆpital, 75651 Paris CEDEX 13, France. 5Pittsburgh Veterans Affairs Medical Center, Pittsburgh, Pennsylvania 15261, USA. Correspondence should be addressed to P.L.S. ([email protected]).

Received 26 April; accepted 22 August; published online 2 October 2005; doi:10.1038/nn1544

NATURE NEUROSCIENCE VOLUME 8

[

NUMBER 11

[

NOVEMBER 2005

1491

B R I E F C O M M U N I C AT I O N S (Fig. 1b), labeled neurons were most numerous in ventral and caudal regions of the dentate (Fig. 3b). In a second animal, an injection of virus was placed B1 mm caudally in GPe, and the region with dense labeling shifted to more dorsal regions of the dentate. We have previously presented evidence that the dentate contains distinct ‘motor’ and ‘nonmotor’ domains10 (Fig. 3c). These regions of the nucleus contain neurons that project via the thalamus to primary motor, premotor, prefrontal and posterior parietal areas of the cerebral cortex9–11. The neurons labeled after virus injections into GPe (and into the putamen) were located in one or both Figure 2 Dentate neurons labeled by retrograde transneuronal transport of virus from GPe. of these domains (Fig. 3b). Furthermore, the (a) Labeled neurons on a coronal section through the dentate nucleus. The animal was allowed to survive for 50 h after an injection of rabies into GPe. Each arrow points to a labeled neuron. Scale bar, 200 mm. number and density of dentate neurons (b) An enlargement of the boxed area in a. Scale bar, 50 mm. labeled after GPe injections were comparable to the number and density of dentate neurons labeled after virus injections into some posterIn the second experiments, we injected a small amount of the N2C ior parietal and prefrontal areas of cortex9–11. This result implies that strain into the external segment of the globus pallidus (GPe) and the dentate influence on a stage of basal ganglia processing is as allowed the animals to survive for 50 h (Fig. 1b, n ¼ 2; Supplementary substantial as its influence on some areas of the cerebral cortex. Fig. 1). Fifty hours is long enough for transneuronal transport of the In the third experiments (Fig. 1c, n ¼ 2; Supplementary Fig. 1), we N2C strain to ‘third-order’ neurons. After the GPe injections we injected a small amount of the N2C strain into GPe and allowed the observed retrograde transport of the virus from the injection site to animals to survive for 40 h. As noted above, this survival time is long first-order neurons in the striatum, retrograde transneuronal transport enough for only one stage of retrograde transneuronal transport. After from these first-order neurons to second-order neurons in the thala- these injections into GPe we observed retrograde transport from the mus, and then another stage of retrograde transneuronal transport injection site to first-order neurons in the striatum and then retrograde from these second-order neurons to third-order neurons in the deep transneuronal transport from these first-order neurons to second-order cerebellar nuclei (Figs. 2,3). The small injections of virus into GPe neurons in the thalamus. The labeled neurons in the thalamus were labeled an average of nearly 1,400 neurons in the cerebellar nuclei, or found in subdivisions of ventroanterior/ventrolateral thalamus approximately ten times the number of neurons labeled after similar- and were particularly numerous in regions of several intralaminar sized injections into the putamen. As with the putamen injections, nuclei including the paracentral, central lateral and centromedian88% of the labeled neurons were located in the contralateral cere- parafascicular complex6 (Supplementary Fig. 2). In contrast, we found bellar nuclei; of these, 69% were located in the dentate, 14% were only one or two labeled neurons in the contralateral dentate of in interpositus and 17% were in fastigial. After the GPe injection each animal.


a

b

Figure 3 Location of dentate neurons that project to GPe. (a) Cross-sections of the dentate. Dots show the location of third-order neurons labeled by retrograde transneuronal transport of virus from GPe. (b) Distribution of labeled neurons on an unfolded map of the dentate (for details of map construction, see ref. 10). Arrows at the top of the map in b indicate locations of slices in a. Arrows in a indicate the level of the horizontal line through the middle of the map in b. The vertical dashed line marks the rostrocaudal center of the nucleus. Filled squares indicate the density of labeled neurons found in 200 mm 200 mm bins through the nucleus. (c) Motor and nonmotor domains of the dentate (modified from ref. 10). This map shows the origin of dentate projections to different cortical areas (‘M1 face’, ‘M1 arm’ and ‘M1 leg’: face, arm and leg representations in primary motor cortex.; PMv: ventral premotor area; 7b: area 7b in posterior parietal cortex; 9L and 46d: lateral area 9 and dorsal area 46 in prefrontal cortex). D, dorsal; C, caudal. The curved dotted line indicates the border between motor and nonmotor domains of the dentate (for details, see ref. 10).

1492

a

Sections a′

b

c

Dentate map a′

158 – 162

b′

c′

Summary map (cebus)

d′

M1 b′

e′

178 – 182

Face

M1 'Motor'

c′

Arm

PMv

M1

198 – 202

7b

Leg

9L 46d 'Nonmotor' d′

218 – 222

e′

238 – 242

D

1 mm

D

3-6 2 1 1 mm

VOLUME 8

[

C

NUMBER 11

1 mm

[

C

NOVEMBER 2005 NATURE NEUROSCIENCE


B R I E F C O M M U N I C AT I O N S These observations provide evidence that the output of the dentate is linked to the striatum via a disynaptic connection and to GPe via a trisynaptic connection (Supplementary Fig. 3, Supplementary Discussion). It is likely that these connections are mediated by intralaminar nuclei and/or ventroanterior/ventrolateral thalamus6. The cerebellar nuclei are known to project to these thalamic regions6, and there is evidence that these thalamic nuclei (especially intralaminar nuclei) project to the striatum12. Our findings support a previous study13 that demonstrated a disynaptic connection between the rat cerebellum and the striatum. The present results extend these observations to a nonhuman primate and demonstrate four new findings: (i) the disynaptic projection to the striatum originates from both the motor and nonmotor domains of the dentate, (ii) the striatum also receives less substantial inputs from fastigial and interpositus, (iii) the projection from the dentate to the striatum connects with medium spiny stellate cells that innervate GPe and thus influences the so-called ‘indirect’ pathway of basal ganglia processing14 and (iv) the number of dentate neurons that influence localized portions of GPe is comparable to the number of dentate neurons that influence some areas of posterior parietal and prefrontal cortex9–11. The demonstration of a pathway that links the output stage of cerebellar processing to the input stage of basal ganglia processing has broad functional implications. For example, it raises the possibility that the cerebellum adaptively adjusts basal ganglia activity on the basis of some internal model and error signal, in a manner similar to the cerebellar mechanisms for adjusting voluntary movement15. Our findings also lead to questions about cerebellar input associated with the motor and cognitive disorders that are characteristic of basal ganglia dysfunction. When basal ganglia activity is abnormal, is cerebellar input part of the problem or part of the solution?


[

NUMBER 11

[

NOVEMBER 2005

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

ACKNOWLEDGMENTS This work was supported by the Veterans Affairs Medical Research Service; the National Parkinson Foundation; US Department of Health and Human Services grants MH56661, NS047126, RR018604 (P.L.S.) and a long-term fellowship from the Human Frontier Science Program Organization (E.H.). We thank B. Dietzschold and M. Schnell (Thomas Jefferson University, Philadelphia) for supplying rabies virus and A. Wandeler (Animal Disease Research Institute, Nepean, Ontario, Canada) for supplying antibodies to rabies. We thank M. Page for developing the computer programs; C. Lovell, K. McDonald, M. O’Malley, M. Ratajeski and M. Watach for their technical assistance; and D. Akkal, F. Delis, R.P. Dum, D. Hoffman, and J.-A. Rathelot for scientific discussions. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. 2. 3. 4. 5. 6. 7.

Alexander, G.E., DeLong, M.R. & Strick, P.L. Annu. Rev. Neurosci. 9, 357–381 (1986). Graybiel, A.M. Neurobiol. Learn. Mem. 70, 119–136 (1998). Masterman, D.L. & Cummings, J.L.J. Psychopharmacology 11, 107–114 (1997). Middleton, F.A. & Strick, P.L. Brain Res. Brain Res. Rev. 31, 236–250 (2000). Schmahmann, J.D. The Cerebellum and Cognition (Academic, San Diego, 1997). Jones, E.G. The Thalamus (Plenum, New York, 1985). Percheron, G., Francois, C., Talbi, B., Yelnik, J. & Fenelon, G. Brain Res. Brain Res. Rev. 22, 93–181 (1996). 8. Kelly, R.M. & Strick, P.L. J. Neurosci. Methods 103, 63–71 (2000). 9. Middleton, F.A. & Strick, P.L. J. Neurosci. 21, 700–712 (2001). 10. Dum, R.P., Li, C. & Strick, P.L. Ann. NY Acad. Sci. 978, 289–301 (2002). 11. Clower, D.M., Dum, R.P. & Strick, P.L. Cereb. Cortex 15, 913–920 (2005). 12. Smith, Y., Raju, D.V., Pare, J.F. & Sidibe, M. Trends Neurosci. 27, 520–527 (2004). 13. Ichinohe, N., Mori, F. & Shoumura, K. Brain Res. 880, 191–197 (2000). 14. Wichmann, T. & DeLong, M.R. Curr. Opin. Neurobiol. 6, 751–758 (1996). 15. Doya, K. Curr. Opin. Neurobiol. 10, 732–739 (2000).

1493



Shift of activity from attention to motor-related brain areas during visual learning Stefan Pollmann1,2 & Marianne Maertens1–3 With practice, we become increasingly efficient at visual object comparisons. This may be due to the formation of a memory template that not only binds individual features together to create an object, but also links the object with an associated response. In a longitudinal fMRI study of object matching, evidence for this link between perception and action was observed as a shift of activation from visual-attentive processing areas along the posterior intraparietal sulcus to hand-sensory and motor-related areas. When we perceive an object, we often link a specific response to it. For instance, perception of a red traffic light primes a movement of the car driver’s foot onto the brake pedal. Thus, object perception does not only involve perceptual processes but may have direct effects on the selection of appropriate actions1. In this example, the perception of a red traffic light not only activates an object representation but also creates a link to the motor actions required for braking. Of course, we do not automatically brake whenever we see a red traffic light, but the link between object representation and response facilitates braking when we do decide to execute this action. In a longitudinal fMRI study, we tested this hypothesis of a link between perception and response by investigating learning-related

changes in brain activation during visual object-matching. In this task, the attentional demands for analyzing feature differences between two objects are initially high, but they should decrease with learning when a memory template of the object pair is created2. In turn, strengthening the memory template may strengthen the link to the associated response1,3. In terms of neural activation, we expect a decrease over the course of learning in brain areas that support attentive processing of stimulus features, whereas we expect an increase in areas that link visual input to manual response preparation. Attentive visual processing depends on posterior parietal cortex, as indicated, for example, by deficits in attentionally demanding visual search performance after temporary disruption of parietal function by transcranial magnetic stimulation (TMS)4. In particular, the cortex along the horizontal segment5 of the intraparietal sulcus (IPS) is involved in shifting attention in space6,7 or between feature dimensions8. Disruption of attentionally demanding visual search by parietal TMS disappears when search becomes automatic after learning4. Similarly, activation in the superior parietal lobule decreases after skill learning9. In contrast, activity related to the representation of conditional manual responses would be expected to occur in the hand motor area and in dorsal premotor cortex, which supports conditional responses associated with arbitrary stimuli10.

Figure 1 Stimuli and procedures. Top, the stimulus set consisted of five categories of geometric shapes: square, triangle, parallelogram, trapezoid and hexagon. Each category consisted of two exemplars that were identical in form but different in orientation. Four items were presented in a trial, two on each side of fixation. In the physical-identity (PI) matching task, a difference in orientation between otherwise identical shapes precluded a ‘match’ judgment. In the category-identity (CI) matching task, participants had to match category membership (for example, ‘triangle’, ‘hexagon’) irrespective of orientation. Bottom, a trial began with the presentation of a fixation cross for either 100 ms or 600 ms. Next, two red frames appeared for 100 ms and cued the locations of two of the four items to be matched. The cues were followed by the presentation of four geometric shapes appearing simultaneously at the four display locations for 80 ms. Participants had to indicate match or mismatch by pressing one of two buttons with their index finger or middle finger, respectively. Response time registration began with the onset of the target display and ended with the participant’s response or after a maximum of 2,220 ms or 2,720 ms, for a total trial duration of 3,000 ms. Participants received a 2,000-Hz feedback tone after every incorrect response. See Supplementary Methods for further details.

+ 100 or 600 ms

+ 100 ms

+ 80 ms

+ 2,220/2,720 ms 'same'

1Department of Experimental Psychology, Otto von Guericke University, Postbox 4120, 39016 Magdeburg, Germany. 2Day Clinic of Cognitive Neurology, University of Leipzig, Liebigstrasse 22a, 04103 Leipzig, Germany. 3Department of Cognitive Neurology, Max Planck Institute for Human Cognitive and Brain Sciences, Stephanstrasse 1a, 04103 Leipzig, Germany. Correspondence should be addressed to S.P. ([email protected]).

Received 27 May; accepted 24 August; published online 2 October 2005; doi:10.1038/nn1552

1494

VOLUME 8

[

NUMBER 11

[



a

b 0.5

0.5 First session

First session Third session Fifth session

0.3 0.2

z = 62

0.1

Third session Fifth session

0.4

0.0

Percent signal change



0.4

0.3 0.2 0.1 0.0 –0.1

–0.1

postCG

IPSh

z = 47

–0.2

–0.2 0

3.5

5

–3.5

–5

2

4

6 8 Time (s)

10

12

14

0

2

4

6

8 10 Time (s)

12

14

z = 57

Figure 2 Learning-related activation changes. The figure shows the activation changes between sessions 5 and 1. The red–yellow scale shows signal increase over sessions, whereas the blue scale shows signal decrease (Z-scores). Left hemisphere is on the left. (a) Activation in the horizontal segment of the intraparietal sulcus (IPSh, Talairach coordinates 25, –64, 47). (b) Activation in postcentral gyrus (postCG, 31, 36, 62) bordering the superior postcentral sulcus (arrow), and local activation maxima in precentral gyrus and superior frontal gyrus shown in two axial planes. z, z-coordinate in Talairach space. The graphs indicate average signal time courses (mean ± s.e.m.) in sessions 1, 3 and 5.

In our visual object matching task, we expected that, during the course of learning, activation would decrease along the horizontal segment of the IPS, as would the need to attentively process individual object features. By contrast, we expected activation to increase in the motor and sensory hand areas and in dorsal premotor cortex as objectresponse associations were strengthened. To test these predictions, we used a geometric figure matching task (Fig. 1). Geometric objects were chosen because they are familiar (but not as overlearned as letters); thus object matching was initially based on feature comparisons. Stimuli were presented tachistoscopically to prevent eye movements. Participants were asked to judge common geometric figures by either their physical (PI) or their categorical (CI) identity. They were required to indicate matches and non-matches by a forced choice response, and a feedback tone indicated an incorrect response; thus a consistent association between object pairs and correct responses could be learned in the course of the experiment. Participants took part in five sessions over a period of up to 21 d. fMRI data were collected in the first, third and fifth sessions (see Supplementary Methods). Participants considerably improved their performance on the task, as indicated by a decrease in response latency from the first session (757 ms) to the third (694 ms) (t6 ¼ 4.62; P o 0.05) and from the third session to the fifth (651 ms) (t6 ¼ 3.22; P o 0.05). Likewise, error rates declined from 8.3% in the first session to 4.5% in the third and 2.8% in the final session. In agreement with our hypothesis, we observed a reduction in activation between the first and last fMRI sessions (equivalent to the fifth training session) along the banks of the horizontal segment5 of the right intraparietal sulcus (Fig. 2a). Notably, the reverse (that is, an increase in activation between sessions 1 and 5) was observed in the right precentral gyrus, extending anteriorly into the superior frontal gyrus and posteriorly into the postcentral gyrus. Within these large activated areas, local activation maxima were observed in the middle genu of the precentral gyrus (the motor hand area; ref. 11), posteriorly adjacent in the somatosensory hand representation area in postcentral gyrus, further anterior at the banks of the superior precentral sulcus (dorsal premotor cortex; ref. 10) and in the posterior portion of the superior frontal gyrus (Fig. 2b, Supplementary Table 1).


[

NUMBER 11

[

NOVEMBER 2005

In the right postcentral gyrus, there was a strong increase in signal amplitude in session 5 as compared to sessions 1 and 3 (Fig. 2b); these last did not differ from each other in signal amplitude. This pattern was observed for trials requiring left hand and right hand responses alike (Supplementary Fig. 1); thus, the signal increase over sessions did not reflect processes related to contralateral motor execution. In contrast, we observed a decrease in signal amplitude between session 1 and sessions 3 and 5 (with no differences between the last two) along the horizontal segment of the right intraparietal sulcus (Fig. 2a). These learning-related activation changes were comparable for physical and category matching (Supplementary Fig. 2); this shows that the activations were not associated with task-specific operations such as mental rotation (a potentially useful strategy in the CI, but not the PI, task). We did not observe learning-related signal changes in visual object processing areas in the lateral occipital cortex or the fusiform gyrus (the lateral occipital complex12). This may be due to the fact that no novel object representation needed to be generated (as subjects were familiar with the geometric objects used in this study); rather, an association between existing object representations needed to be built. We have hypothesized that in the present study, training should lead to memory-based processing, thus reducing the demands on attentive processing. The decrease in activation along the horizontal segment of the right intraparietal sulcus, at a location consistently reported to subserve attentive processing, confirms the first part of our hypothesis. More importantly, we have posited that learning object matching includes associating object pairs with the appropriate response. The signal increase over sessions in the hand representation areas in precentral and postcentral gyri supports this prediction. It may seem puzzling that a simple two-alternative forced-choice response should require learning. However, in our view, it is not the response that needs to be learned but the link between a particular object pair and its associated response (which is strengthened when the same object pair is presented repeatedly and is followed by the same response). Postcentral gyrus activation, that is, activation in sensorimotor cortex (S1), is commonly observed in motor tasks, often with greater signal changes

1495


B R I E F C O M M U N I C AT I O N S than in precentral motor areas13. The rostral part of the dorsal premotor cortex, along the banks of the superior precentral sulcus, and the cortex along the horizontal segment of the IPS—both activated in the present study—have previously been reported to support conditional motor selection10. Our data agree well with these data; moreover, they show that the contributions of the posterior parietal and more anterior components can be further dissociated in the time course of establishing a link between object representation and action. The early decrease in activation along the IPS (between sessions 1 and 3) precedes the later increase in activation in the postcentral gyrus (between sessions 3 and 5). This shows that the latter is not immediately contingent on the former; it may indicate that learning in the object matching task occurred in two steps: an initial step in which a perceptual representation of the object pair was generated, reducing visual attentional demands, and a second step in which the perceptual object representation was associated with the appropriate response4. According to such a sequential model, fast learners may have begun responserelated learning even before session 3, whereas slow learners may reach this transition only after session 3. Indeed, the increase in postcentral signal between sessions 3 and 5 was significantly correlated with the ratio of late to early response-time reductions (r ¼ 0.784, P o 0.05; Supplementary Fig. 3); the signal increase was stronger for late learners. However, the relation between perceptual and response learning may depend on numerous factors, such as object complexity or salience on the perceptual learning side and the number of response alternatives on the response learning side. Furthermore, perceptual learning can proceed much faster than in the present experiment and the contribution of parietal cortex may decrease rapidly14. Are object-response associations created equally fast under these circumstances? Longitudinal fMRI studies may be a valuable tool to address these issues.

1. Hommel, B. Vis. Cogn. 5, 183–216 (1998). 2. Logan, G.D. Psychol. Rev. 109, 376–400 (2002). 3. Hommel, B., Musseler, J., Aschersleben, G. & Prinz, W. Behav. Brain Sci. 24, 849–878 (2001). 4. Walsh, V., Ashbridge, E. & Cowey, A. Neuropsychologia 36, 45–49 (1998). 5. Duvernoy, H.M. The Human Brain: Surface, Three-Dimensional Sectional Anatomy with MRI, and Blood Supply 2nd edn. (Springer-Verlag, New York, 1999). 6. Corbetta, M. et al. Neuron 21, 761–773 (1998). 7. Vandenberghe, R., Gitelman, D.R., Parrish, T.B. & Mesulam, M.M. Neuroimage 14, 661–673 (2001). 8. Weidner, R., Pollmann, S., Mu¨ller, H.J. & von Cramon, D.Y. Cereb. Cortex 12, 318–328 (2002). 9. Poldrack, R.A., Desmond, J.E., Glover, G.H. & Gabrieli, J.D.E. Cereb. Cortex 8, 1–10 (1998). 10. Grafton, S.T., Fagg, A.H. & Arbib, M.A. J. Neurophysiol. 79, 1092–1097 (1998). 11. Yousry, T.A. et al. Brain 120, 141–157 (1997). 12. Malach, R., Levy, I. & Hasson, U. Trends Cogn. Sci. 6, 176–184 (2002). 13. Maldjian, J.A., Gottschalk, A., Patel, R.S., Detre, J.A. & Alsop, D.C. Neuroimage 10, 55–62 (1999). 14. Walsh, V., Allison, A., Ashbridge, E. & Cowey, A. Neuropsychologia 37, 245–251 (1999).

1496

VOLUME 8

In summary, the decrease in activation in the posterior part of the IPS, along with an increase in activation in the postcentral gyrus and the frontal cortex, support the idea that learning object matching proceeds from an initial attentionally demanding feature comparisons stage to a later template-based processing stage in which object pairs are linked to the appropriate response. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS This work was supported by a grant from the Gertrud Reemtsma Stiftung to M.M. and by Deutsche Forschungsgemeinschaft grant Po 548/3–1 to S.P. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

[

NUMBER 11

[


The essential role of stimulus temporal patterning in enabling perceptual learning

can be readily learned when stimuli are practiced in a fixed temporal pattern. This temporal patterning does not facilitate learning by reducing stimulus uncertainty; further, learning enabled by temporal patterning can later generalize to randomly presented stimuli.

Shu-Guang Kuai1,3, Jun-Yun Zhang1,3, Stanley A Klein2, Dennis M Levi2 & Cong Yu1

Perceptual learning refers to improvement, through practice, in the ability to discriminate fine differences in visual and other sensory features such as contrast1,2, orientation3,4 and Vernier and other positional acuities5,6 (see refs. 7 and 8 for recent reviews). Many studies have investigated the effects of spatial factors—such as stimulus contrast, spatial frequency and orientation—on perceptual learning. However, the role of stimulus temporal factors in perceptual learning has been largely overlooked.

2

1 3 51 3 51 3 51 3 5 Session

10 8 6 5 4 3 2

n=6 2

3 4 5 6 78 10 Pre-training ∆D

20

0.2

n = 10 0.2

0.3

0.1 0.08 0.06 0.05 0.04 0.03

0.04 0.05 0.06 0.08 0.1

1 3 51 3 51 3 5 1 3 5 Session

i D = 22.5° D = 67.5° D = 112.5° D = 157.5°

20

10 8 6 5 4 3

0.1 0.08 0.06 0.05 0.04 0.03

0.3

Pre-training ∆C

j 20 10 8 6 5 4 3 2

20

YS Post-training ∆D

ZJ

e YH

0.2

Pre-training ∆C

h 20

0.3

0.03

n=4

Contrast threshold ∆C

0.1 0.08 0.06 0.05 0.04 0.03

0.2

1 3 51 3 51 3 5 1 3 5 Session

g

0.2

0.3

0.1 0.08 0.06 0.05 0.04 0.03

C = 0.20 C = 0.30 C = 0.47 C = 0.63

0.04 0.05 0.06 0.08 0.1

0.2

d

0.3

Direction threshold (deg)

f

SA

0.03

Time

c

0.3

Post-training ∆C

92 ms

b

Post-training ∆D

92 ms

Contrast threshold ∆C

a

Post-training ∆C

Little is known about how temporal stimulus factors influence perceptual learning. Here we demonstrate an essential role of stimulus temporal patterning in enabling perceptual learning by showing that ‘unlearnable’ contrast and motion-direction discrimination (resulting from random interleaving of stimuli)

Direction threshold (deg)



1 3 5 1 3 51 3 5 1 3 5 Session

10 8 6 5 4 3 2

n=5 2

3 4 5 6 78 10 Pre-training ∆D

20

Figure 1 Effects of stimulus roving and temporal patterning on perceptual learning of contrast and motion-direction discrimination. (a) Stimuli for contrast discrimination in a two-alternative forced-choice (2-AFC) trial. A fixation cross was followed by two Gabor stimuli (spatial frequency sf ¼ 6 cycles per degree, presented for 92 ms each separated by a 600-ms interstimulus interval; s.d. s ¼ 0.071.) The observers’ task was to judge which stimulus had higher contrast. Discrimination thresholds were measured with a three-down–one-up forced-choice staircase method. (b) Observer SA’s unchanged session-by-session contrast thresholds (DC) for each reference contrast with contrast roving. Throughout: error bars, s.e.m.; solid lines, linear fits. Training sessions were typically 2 h and included 1,000–1,200 trials. (c) Comparison of mean post- and pre-training contrast thresholds obtained with contrast roving showed no significant learning (F1,3 ¼ 3.26, P ¼ 0.169; data points significantly below dashed diagonal line indicate that learning has taken place). (d) Observer YH’s session-by-session reduction in contrast thresholds during practice with temporally patterned contrasts. (e) Comparison of mean post- and pre-training contrast thresholds obtained with temporal patterning showed significant learning (F1,9 ¼ 40.8, P ¼ 0.000). (f) Stimuli for motion direction discrimination. A circular window of diameter 81 held 1,000 random dots, all moving in the same direction at a speed of 101 per s. In a 2-AFC trial, two sets of dots (for clarity, fewer dots are shown) were presented for 500 ms each, with a 200-ms interval between presentations. The observers’ task was to judge in which interval the random dots moved more clockwise. (g) Observer ZJ’s session-by-session motion direction thresholds (DD) with direction roving. (h) Comparison of mean post- and pre-training motion direction thresholds showed no significant improvement after practice with direction roving (F1,5 ¼ 0.007, P ¼ 0.936). (i) Observer YS’s session-by-session lowering of direction thresholds with temporally patterned directions. (j) Comparison of mean post- and pre-training motion direction thresholds showed that significant learning took place under stimulus temporal patterning (F1,4 ¼ 17.7, P ¼ 0.014).

1Institute of Neuroscience, Chinese Academy of Sciences, Shanghai 200031, China. 2School of Optometry and Helen Wills Neuroscience Institute, University of California, Berkeley, California 94720, USA. 3These authors contributed equally to this work. Correspondence should be addressed to C.Y. ([email protected]).

Received 1 August; accepted 23 August; published online 16 October 2005; doi:10.1038/nn1546


[

NUMBER 11

[

NOVEMBER 2005

1497


Pre-training ∆C

0.03 1

2

3

4 5 6 7 8 10

Post-training ∆D (deg)

n=4

Pre-training ∆C

Pre-training ∆D (deg)

0.3

n=4

0.1 0.08 0.07 0.06 0.05 0.04

0.2

2

b

C = 0.20 C = 0.30 C = 0.47 C = 0.63

0.2

3

1

0.3

0.04 0.05 0.06 0.07 0.08 0.1

0.3

0.2

0.04 0.05 0.06 0.07 0.08 0.1

n=4

a

D = 22.5° D = 67.5° D = 112.5° D = 157.5°

8 7 6 5 4

0.03

0.1 0.08 0.07 0.06 0.05 0.04

Post-training ∆D (deg)

0.2

0.03

b 10

C = 0.20 C = 0.30 C = 0.47 C = 0.63

Post-training ∆C

0.3

0.03


Post-training ∆C

a

10 8 7 6 5 4

D = 22.5° D = 67.5° D = 112.5° D = 157.5°

3 2 n=3 1

1

2

3

4 5 6 7 8 10

Pre-training ∆D (deg)

Figure 2 The effects of pre-cueing on contrast and motion direction learning with stimulus roving. (a,b) Comparisons of the mean post- and pre-training thresholds for (a) contrast discrimination and (b) motion direction discrimination showed no significant learning (see text for F-test results).

Figure 3 Perceptual learning generalizes to unpracticed temporal conditions. (a,b) After learning with stimulus temporal patterning, discrimination thresholds with stimulus roving were significantly lower than the pre-training levels for (a) contrast discrimination and (b) motion direction discrimination (see text for F-test results).

Recently1 we found that contrast discrimination for Gabor stimuli (Gaussian-windowed sinusoidal gratings; Fig. 1a) is unlearnable if the several contrasts that are to be learned are randomly interleaved trial by trial (a condition we refer to as ‘contrast roving’; see also ref. 2). On average, the ratio of contrast discrimination thresholds after training to the same thresholds before training is 0.93 ± 0.09 (mean ± s.e.m.; data replotted in Fig. 1b,c), indicating no significant threshold reduction or perceptual learning after practice. Learning is possible only when contrast discrimination is practiced one contrast at a time in block trials1. The ‘knock-out’ of learning by contrast roving suggests that stimulus information may need to be organized in a certain temporal pattern if perceptual learning is to take place. In the current study, we presented the same four contrasts as in ref. 1 in two fixed temporal patterns: for seven observers, the contrast increased monotonically (0.2–0.3–0.47–0.63), and for three other observers, the contrast varied non-monotonically (0.2–0.47–0.3– 0.63). After five such ‘temporal patterning’ practice sessions, the observers’ contrast discrimination improved significantly (Fig. 1d,e). The ratio of the mean post-training to mean pre-training thresholds (Fig. 1e) was 0.61 ± 0.06, comparable to the 0.64 ± 0.04 we obtained in our original blocked-trials learning1. An analysis of variance (ANOVA) further confirmed that the effects of contrast roving and temporal patterning on contrast learning were significantly different (F1,12 ¼ 13.2, P ¼ 0.003). Thus, practice with contrast interleaving is just as effective as practice with blocked trials in facilitating perceptual learning, as long as the contrasts are temporally patterned. The role of stimulus temporal patterning in perceptual learning was also evident in a completely different task: motion-direction discrimination (Fig. 1f). As in the contrast learning task, observers who practiced with ‘direction roving’—with motion stimuli in four reference directions that varied randomly from trial to trial—showed postto pre-training threshold ratios either close to 1 (indicating that no learning had taken place) or even higher (indicating that observers’ performance was actually worse after practice) (Fig. 1g,h); on average, post- to pre-practice threshold ratios were 1.17 ± 0.23. However, observers who practiced an equivalent amount of trials with stimuli possessing a fixed temporal pattern (same four directions as in the roving condition, changing clockwise as follows: 22.51–67.51–112.51– 157.5) showed improved discrimination: the overall post- to pretraining threshold ratio was 0.64 ± 0.11 (Fig. 1i,j), comparable to the 0.61 ± 0.06 ratio in contrast learning with temporal patterning (Fig. 1e). Again, roving and temporal patterning had significantly

different effects on learning motion direction (F1,9 ¼ 6.1, P ¼ 0.036). Taken together, the results of our contrast and motion-direction learning experiments indicate that stimulus temporal patterning has an essential role in enabling at least low-level perceptual learning. Does temporal patterning facilitate perceptual learning by reducing stimulus uncertainty1,2,9? With stimulus roving, observers may be uncertain about the contrast or motion direction in the first interval of a 2-AFC trial, resulting in judgments within the uncertain range of the contrast or motion direction. This stimulus uncertainty may knock out learning2,9. To test this possibility, we had observers practice contrast and motion-direction discrimination with roving, but we presented a pre-cue to minimize stimulus uncertainty. The pre-cue, appearing 1 s before the first interval of each roving trial, was an identical Gabor or moving random-dot stimulus with the same reference contrast (Gabor) or motion direction (moving dot) and the same duration as the stimulus itself. The results demonstrated that, with stimulus roving, pre-cueing was insufficient to significantly improve contrast or motion-direction discrimination (Fig. 2). The ratio of mean post-training to mean pre-training thresholds was 0.88 ± 0.1 for both tasks (F1,3 ¼ 5.47, P ¼ 0.101 for contrast discrimination; F1,3 ¼ 3.57, P ¼ 0.155 for motion-direction discrimination). For both contrast and motiondirection discrimination, the effects of pre-cued roving were not significantly different from those of uncued roving (F1,6 ¼ 0.404, P ¼ 0.549 for contrast discrimination; F1,8 ¼ 0.487, P ¼ 0.505 for motion-direction discrimination) but were significantly different from the effects of temporal patterning (F1,12 ¼ 9.54, P ¼ 0.009 for contrast discrimination; F1,7 ¼ 7.01, P ¼ 0.003 for motion-direction discrimination). These results ruled out reduction in contrast uncertainty as a plausible explanation for the facilitation of perceptual learning by stimulus temporal patterning. Does the improved discrimination in the post-practice period remain specific to the practiced temporal pattern? This kind of specificity would be uneconomical because observers would have to re-learn the same stimuli whenever the stimulus temporal pattern changes. To examine this issue, we measured both the contrast and the motion-direction discrimination with stimulus roving, but we used observers who had demonstrated successful learning in the temporal patterning condition. These observers’ post-training discrimination thresholds were significantly lower than their pre-training thresholds (Fig. 3; F1,3 ¼ 29.1, P ¼ 0.012 for contrast discrimination; F1,2 ¼ 30.8, P ¼ 0.031 for motion-direction discrimination). The ratios of the mean post- and pre-training thresholds were 0.61 ± 0.06

1498

VOLUME 8

[

NUMBER 11

[



B R I E F C O M M U N I C AT I O N S and 0.55 ± 0.05 for contrast and motion-direction discrimination, respectively, suggesting that learning obtained with temporal patterning generalized to the stimulus roving condition. A similar learning generalization was also found with unpracticed fixed temporal patterns in contrast discrimination (data not shown). Our data therefore provide no evidence for post-learning temporal specificity and instead suggest that human visual learning is highly efficient. We hypothesize that the fixed temporal patterns may temporally chunk discrete stimuli together, thus facilitating the encoding of stimulus information (as memory traces) into visual long-term memory (LTM), but that interference by stimulus roving interrupts such encoding. For learning to occur, the top-down developing LTM traces must interact with the bottom-up sensory input and guide stimulus discrimination. This continuous interaction between the bottom-up stimulus inputs and top-down LTM traces enhances and refines the stored memory traces, which eventually improves discrimination. Further work will be needed to explain how temporally chunked LTM can apply to roving and other unpracticed temporal patterns after learning has taken place (Fig. 3).


[

NUMBER 11

[

NOVEMBER 2005

ACKNOWLEDGMENTS We thank S.-G. He for helpful comments. This research was supported by the Chinese Academy of Sciences, the Shanghai Municipal Government and the US National Institutes of Health (R01-04776 and R01-01728). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. Yu, C., Klein, S.A. & Levi, D.M. J. Vis. 4, 169–182 (2004). 2. Adini, Y., Wilkonsky, A., Haspel, R., Tsodyks, M. & Sagi, D. J. Vis. 4, 993–1005 (2004). 3. Lu, Z.L. & Dosher, B.A. J. Vis. 4, 44–56 (2004). 4. Fiorentini, A. & Berardi, N. Nature 287, 43–44 (1980). 5. Li, R.W., Levi, D.M. & Klein, S.A. Nat. Neurosci. 7, 178–183 (2004). 6. Saarinen, J. & Levi, D.M. Vision Res. 35, 519–527 (1995). 7. Fine, I. & Jacobs, R.A. J. Vis. 2, 190–203 (2002). 8. Fahle, M. Curr. Opin. Neurobiol. 15, 154–160 (2005). 9. Tsodyks, M., Adini, Y. & Sagi, D. Neural Netw. 17, 823–832 (2004).

1499

COMT genotype predicts longitudinal cognitive decline and psychosis in 22q11.2 deletion syndrome Doron Gothelf1,2, Stephan Eliez3, Tracy Thompson1, Christine Hinard4, Lauren Penniman1, Carl Feinstein1, Hower Kwon5, Shuting Jin1, Booil Jo1, Stylianos E Antonarakis6, Michael A Morris4 & Allan L Reiss1 Although schizophrenia is strongly hereditary, there are limited data regarding biological risk factors and pathophysiological processes. In this longitudinal study of adolescents with 22q11.2 deletion syndrome, we identified the catechol-Omethyltransferase low-activity allele (COMT L) as a risk factor for decline in prefrontal cortical volume and cognition, as well as for the consequent development of psychotic symptoms during adolescence. The 22q11.2 deletion syndrome is a promising model for identifying biomarkers related to the development of schizophrenia. Although there is robust genetic vulnerability to schizophrenia, efforts to identify its molecular pathophysiology have yielded limited results, probably owing to the etiological heterogeneity of this symptom-defined disorder1,2. One approach for contending with this heterogeneity is to study subjects with more homogeneous risk for psychosis. The 22q11.2 deletion syndrome (22q11.2DS) is caused by a microdeletion on chromosome 22 and occurs in about 1 of every 4,000 live births. One-third of the individuals with this condition develop schizophrenia or a related psychotic disorder, making 22q11.2DS the most commonly known risk factor for the development of psychosis and a unique model for elucidating neurodevelopmental pathways leading to psychotic disorders3–5. The syndrome is also associated with congenital malformations and cognitive deficits3–5. Among the genes in the deleted region6, variants of COMT and of proline dehydrogenase (PRODH) have been reported to increase susceptibility to schizophrenia7,8. The COMT gene seems especially relevant to schizophrenia, as it encodes the COMT enzyme that is critical for dopamine metabolism in the prefrontal cortex (PFC)2. COMT contains a polymorphism resulting in the amino acid polymorphism V108/158M and concomitant high- and low-activity variants of the enzyme9. Human tissues homozygous for the high-activity allele (COMTH) have 40% higher COMTactivity in the PFC than do tissues homozygous for the low-activity allele (COMTL)9.

COMT is particularly critical for dopamine metabolism in the PFC, a region thought to play an important role in the pathophysiology of schizophrenia2. As subjects with 22q11.2DS carry only one copy of the COMT gene, those with the COMTL are probably exposed to unusually large amounts of dopamine in the brain. We hypothesized that hemizygosity for COMT L would be a risk factor for greater reduction in PFC gray matter volume, cognitive decline and the evolution of psychotic symptoms during adolescence. We also investigated whether other genetic variations, including key polymorphisms in the COMT and PRODH genes, influence neuropsychiatric outcome in 22q11.2DS. Here we report the results of a prospective, longitudinal study of 24 subjects with 22q11.2DS and 23 subjects with idiopathic developmental disabilities, matched for age, gender, ethnicity and IQ (Supplementary Table 1 online). The two groups were first evaluated during childhood (T1) and re-evaluated during late adolescence or early adulthood (T2). Written informed consent was obtained from the subjects and their parents under protocols approved by the institutional review board of Stanford University. Using standard methods, subjects were genotyped for the following single-nucleotide polymorphisms: COMT rs165688 (COMTH/ COMTL), rs2097603, rs737865 and rs165599; and PRODH rs372055 and rs450046. The severity of subjects’ psychotic symptoms at T2 was assessed using the Brief Psychiatric Rating Scale (BPRS). Cognitive testing was conducted at both T1 and T2, using the age-appropriate Wechsler Intelligence Scale and the Clinical Evaluation of Language

15 Change in verbal IQ (T2− T1 scores)



10 5 0 −5 − 10 − 15 − 20 COMT L

COMT H

Figure 1 Box plots show the range of values for each group, as well as the 25th, 50th (dark bar) and 75th percentiles. Significant longitudinal effects (P o 0.05) of COMT genotype on change in VIQ in subjects with 22q11.2DS.

1Center for Interdisciplinary Brain Sciences Research, Stanford University School of Medicine, 401 Quarry Road, Stanford, California 94305–5795, USA. 2Sackler Faculty of Medicine, Tel Aviv University, Ramat Aviv, Israel. 3Department of Psychiatry, University of Geneva School of Medicine, 41 Ch. des Creˆts-de-Champel, CH-1206 Geneva, Switzerland. 4Medical Genetics Service, University Hospitals, 1Rue Michel Servet, 1211 Geneva, Switzerland. 5Department of Psychiatry and Behavioral Sciences, University of Washington, Box 359911, Seattle, Washington 98104, USA. 6Department of Genetic Medicine and Development, University Medical School, Centre Medicale Universitaire, 1 Rue Michel Servet, 1211 Geneva, Switzerland. Correspondence should be addressed to A.L.R. ([email protected]).

Received 29 June; accepted 23 September; published online 23 October 2005; doi:10.1038/nn1572

1500

VOLUME 8

[

NUMBER 11

[


B R I E F C O M M U N I C AT I O N S 70


BPRS score at T2

60 50 40 30 20 10 0 −20

− 15

−10 −5 0 5 Decrease in VIQ (T2−T1 scores)

10

15

Figure 2 Correlation between change in VIQ and BPRS scores in subjects with 22q11.2DS (r ¼ 0.71, P o 0.0001). m, subjects with psychosis; }, nonpsychotic subjects.

Fundamentals–III (CELF-III). Magnetic resonance images (MRI) were acquired with a GE Signa 1.5- scanner at both time points, for 18 subjects with 22q11.2DS. The PFC was defined as all portions of the frontal cortex anterior to the genu of the corpus callosum (see Supplementary Methods online). At T1, no subject had a psychotic disorder. At T2, seven subjects with 22q11.2DS (29.2%) had developed a psychotic disorder (five schizophrenia or schizoaffective, one schizophreniform and one psychotic depression) versus only one (4.2%) from the developmental disability group (psychotic depression; Fisher’s exact test, P o 0.05). In the 22q11.2DS group, we found a significant decline in verbal IQ (VIQ) and CELF-III expressive language (CELFE) scores from T1 to T2 (P o 0.05); further, a repeated-measures analysis of variance (ANOVA) showed a significant interaction between time and COMTH/COMTL allele for VIQ and CELFE scores and PFC gray matter volume. Specifically, the COMT L subgroup had a more robust decrease in VIQ and CELFE scores and PFC volume than did the COMT H subgroup (all P-values o 0.05; Fig. 1 and Supplementary Table 2 online). There were no marked time-related changes (or interactions) in the developmental disability group for either cognitive variable of interest (DVIQ ¼ +1.7 ± 9.4; DCELFE ¼ –2.8 ± 13.5). At T2, BPRS scores were significantly different between the COMT L, COMTH and developmental disability groups (F2,44 ¼ 9.4, P o 0.0001). On Scheffe´ pairwise comparison, the 22q11.2DS COMT L subgroup had significantly higher BPRS scores (40.5 ± 12.5) than did the 22q11.2DS COMT H subgroup (29.2 ± 9.8; P o 0.05) and the developmental disability control group (26.1 ± 7.8; P o 0.0001). There was a significant correlation (P o 0.0001) between DVIQ and BPRS scores in the 22q11.2DS group (Fig. 2) but not in the developmental disability controls. There was no marked association between COMTH/ COMTL genotype in the developmental disability group and BPRS scores (COMTL/L: 27.3 ± 10.0; COMTH/L: 26.8 ± 7.3; COMTH/H: 23.3 ± 7.5). Except for the COMTH/COMTL polymorphism, no marked associations were detected between COMT single-nucleotide polymorphisms and BPRS, DPFC or DVIQ in the 22q11.2DS or developmental disability groups. A significant (P o 0.05) time genotype interaction was observed for COMT rs165599 in the 22q11.2DS group for CELFE scores only (Supplementary Table 2). For the PRODH single-nucleotide polymorphism rs372055, a threegroup analysis of the 22q11.2DS ‘hemizygous T’ and ‘hemizygous C’ subgroups and the developmental disability controls showed a significant difference in BPRS scores (F2,44 ¼ 6.9, P ¼ 0.002). Post-hoc comparisons showed that the 22q11.2DS ‘T’ subgroup scored significantly higher (37.5 ± 12.2) than the developmental disability controls


[

NUMBER 11

[

NOVEMBER 2005

did (P o 0.005) and marginally higher than the ‘C’ subgroup did (27.2 ± 11.5; P ¼ 0.10). In the developmental disability group, no association was found between the PRODH genotype and any outcome measure. This longitudinal study indicates that the COMT genotype makes a significant contribution to brain development and neuropsychiatric outcome in adolescent subjects with 22q11.2DS. Because cognitive decline and the emergence of psychosis were observed in subjects with 22q11.2DS and not in the matched developmental disability control group, it seems likely that genetic factors specific to the 22q11.2DS syndrome are responsible for these phenomena. As hypothesized, 22q11.2DS subjects with COMTL demonstrated a more robust decrease in PFC volume and cognition, as well as more severe psychotic symptoms, than did those with COMTH. Being hemizygous for COMT, subjects with 22qDS11.2 and COMTL are presumably most deficient in COMT activity and thus most likely to have increased amounts of synaptic dopamine on a long-term developmental basis. At T1, the COMTH subgroup showed a trend for higher cognitive function; this finding is similar to that from a previous cross-sectional investigation of children with 22q11.2DS10. Results from other studies indicate that a narrow window of dopamine receptor D1 activation in the PFC is required for optimal cognitive functioning11–13. We propose that neurodevelopmental changes associated with adolescence further increase dopaminergic signaling in adolescents with 22q11.2DS COMTL, leading to reduced efficiency of cognitive processing. This could reflect a shift of cognition onto the ‘down slope’ of an inverted-U curve that represents the relationship between amounts of dopamine in the PFC and cognitive function11,12. Elevation in dopaminergic tone also increases the risk for emergence of psychotic symptoms12 and adverse effects on PFC development11,12. In support of this hypothesis, COMT knockout mice are known to have increased brain dopamine, especially in the frontal cortex, and to show aberrant behavior14. It is likely that decline in VIQ in 22q11.2DS subjects precedes the onset of psychotic symptoms: longitudinal studies of schizophrenia demonstrate a decline in IQ several years before psychotic symptoms emerge, and during adolescence, this decline is most robust in the language domain15. To detect the sequence in which neuropsychiatric abnormalities emerge in 22q11.2DS, future studies should use a larger sample size and make multiple assessments during adolescence, separated by shorter time intervals. Future studies should also include refined measures of PFC subregions and more specific assessment of prefrontal-related cognitive tasks such as working memory. Finally, although the COMTH/COMTL polymorphism seemed to contribute the most to outcome in our 22q11.2DS group, it is likely that other genes from the critical deletion region also play a role in the development of neuropsychiatric and cognitive symptoms. Thus, the effect, in individuals with 22q11.2DS COMTL, of severe deficiency in PFC COMT activity may be to worsen these symptoms. Subsequent studies should evaluate the potential contribution of the PRODH genotype and other genes from the 22q11.2 deletion region to neuropsychiatric outcome. In summary, this is the first study to evaluate the longitudinal effects of COMT alleles in a specific model disorder. Our findings suggest that extreme deficiency in COMT activity, as present in COMTL subjects with 22q11.2DS, is an important neurodevelopmental risk factor for decline in PFC volume and cognition and for the emergence of psychotic symptoms during adolescence. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank L. Xiaoyan and J.F. Hallmayer for DNA extraction, and J. Keller for cognitive assessments. This work was supported by US National Institutes

1501

B R I E F C O M M U N I C AT I O N S of Health grants MH50047, HD31715 and MH19908 (A.L.R.) and by the Swiss National Science Foundation, the European Union Federal Office of Education and the ‘Child Care’ Foundation (S.E.A.).

1. Sullivan, P.F., Kendler, K.S. & Neale, M.C. Arch. Gen. Psychiatry 60, 1187–1192 (2003).

2. Weinberger, D.R. Lancet 346, 552–557 (1995). 3. Murphy, K.C., Jones, L.A. & Owen, M.J. Arch. Gen. Psychiatry 56, 940–945 (1999). 4. Karayiorgou, M. et al. Proc. Natl. Acad. Sci. USA 92, 7612–7616 (1995). 5. Pulver, A.E. et al. J. Nerv. Ment. Dis. 182, 476–478 (1994). 6. Maynard, T.M. et al. Proc. Natl. Acad. Sci. USA 100, 14433–14438 (2003). 7. Shifman, S. et al. Am. J. Hum. Genet. 71, 1296–1302 (2002). 8. Liu, H. et al. Proc. Natl. Acad. Sci. USA 99, 3717–3722 (2002). 9. Chen, J. et al. Am. J. Hum. Genet. 75, 807–821 (2004). 10. Bearden, C.E. et al. Am. J. Psychiatry 161, 1700–1702 (2004). 11. Mattay, V.S. et al. Proc. Natl. Acad. Sci. USA 100, 6186–6191 (2003). 12. Seamans, J.K. & Yang, C.R. Prog. Neurobiol. 74, 1–57 (2004). 13. Blasi, G. et al. J. Neurosci. 25, 5038–5045 (2005). 14. Gogos, J.A. et al. Proc. Natl. Acad. Sci. USA 95, 9991–9996 (1998). 15. Fuller, R. et al. Am. J. Psychiatry 159, 1183–1189 (2002).

1502

VOLUME 8

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


Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

[

NUMBER 11

[



ARTICLES

MPS-1 is a K+ channel b-subunit and a serine/ threonine kinase Shi-Qing Cai, Leonardo Hernandez, Yi Wang, Ki Ho Park & Federico Sesti1 We report the first example of a K1 channel b-subunit that is also a serine/threonine kinase. MPS-1 is a single–transmembrane domain protein that coassembles with voltage-gated K1 channel KVS-1 in the nervous system of the nematode Caenorhabditis elegans. Biochemical analysis shows that MPS-1 can phosphorylate KVS-1 and other substrates. Electrophysiological analysis in Chinese hamster ovary (CHO) cells demonstrates that MPS-1 activity leads to a significant decrease in the macroscopic current. Single-channel analysis and biotinylation assays indicate that MPS-1 reduces the macroscopic current by lowering the open probability of the channel. These data are consistent with a model that predicts that the MPS-1–dependent phosphorylation of KVS-1 sustains cell excitability by controlling K1 flux.

Potassium (K+) channels are an important class of integral membrane proteins that control the resting membrane potential and excitability of biological cells. K+ channels are tetramers composed of four identical subunits, packed around a water-filled pore1,2. This simple homomeric structure is sufficient to give rise to functional channels, but it is observed only sporadically in nature. Often, K+ channels require additional regulatory proteins, such as b-subunits, to fine-tune their trafficking to the plasma membrane, their location and abundance, their sensitivity to stimulation and their pharmacology. KCNE proteins are a class of b-subunits of voltage-gated K+ channels3–7 that have been conserved through evolution. These subunits are integral membrane proteins8 and, by virtue of this characteristic, are able to interact intimately with the channel9–11 to affect several functional properties12–15: namely, to control how signaling molecules modulate the channel16,17, to partner with multiple channels4,18 and to cause congenital and acquired channelopathies19–24. Signaling molecules such as protein kinases also have a central role in the modulation of K+ channels. Protein kinases are among the largest and most heterogeneous families of proteins in eukaryotes; they regulate innumerable biological processes including metabolism, transcription, cell cycle progression, cytoskeletal rearrangement and cell movement, and apoptosis and differentiation. Protein phosphorylation is also critical in intercellular communication during development, in physiological response and homeostasis, and in the functioning of the nervous and immune systems25. Even though both protein kinases and b-subunits act to alter the function of channels and, ultimately, the electrical activities of the cell, the molecular basis for the way in which they function, and the effects they produce, are fundamentally different. Accessory subunits form stable complexes with channels and therefore cause permanent modifications by passively changing the three-dimensional structure of the complex, its amino acid content, or

both. In contrast, protein kinases produce reversible alterations by actively catalyzing the incorporation, into the target protein, of phosphates that can be removed by phosphatases. In this paper we report an example of a K+ channel b-subunit that is also a serine/threonine kinase. MPS-1 was originally cloned during the process of identifying C. elegans orthologs of human KCNE proteins26. MPS-1 forms a channel complex with the voltage-gated K+ channel KVS-1 in the nervous system of the nematode6,26. When expression of the MPS-1 gene is reduced by RNA interference (RNAi), the animals show several sensory defects, thus emphasizing the important physiological roles of this K+ channel and of MPS-1 (ref. 26). Here we show that MPS-1 is able to control KVS-1 activity through independent mechanisms that stem from its dual role—as b-subunit and protein kinase. RESULTS MPS-1 has kinase activity The amino acid sequence of MPS-1 contains two characteristic protein kinase motifs: a DFG (AspPheGly) triplet, which identifies the Mg2+ binding site of ATP molecules, and a HisSerAsp (HSD) triplet that can have catalytic function (Fig. 1), which suggests that MPS-1 might be a kinase25. To test the hypothesis that MPS-1 is a kinase, we expressed histidine-tagged wild-type MPS-1 in Escherichia coli, along with a putatively inactive variant bearing a mutation that causes a D178N amino acid substitution in the DFG catalytic site. We purified the recombinant proteins by immobilized metal ion affinity chromatography and evaluated their enzymatic activity in kinase assays. In the insoluble fraction of E. coli lysates, antibodies to histidine detected the full-length protein (B35 kDa) as well as a fragment migrating at approximately the 16-kDa position (wild type, Fig. 2a; data for the D178N mutant not shown) that was still present after purification

University of Medicine and Dentistry of New Jersey, Robert Wood Johnson Medical School, Department of Physiology and Biophysics, 683 Hoes Lane, Piscataway, New Jersey 08854, USA. 1Correspondence should be addressed to F.S. ([email protected]). Received 15 August; accepted 2 September; published online 16 October 2005; doi:10.1038/nn1557


[

NUMBER 11

[

NOVEMBER 2005

1503

ARTICLES

Figure 1 MPS-1 shows characteristic protein kinase motifs. Amino acid sequence of MPS-1. The transmembrane domain and the motifs thought to be critical for catalytic function are in bold. DFG is well conserved among kinases and it chelates the Mg2+ ions of ATP. The aspartate of the HRD triplet (modified to HSD) can be catalytic by acting as a base acceptor.

(as shown in the Coomassie blue–stained blot in Fig. 2b). The antibodies did not detect any protein in the solubilized fraction (data not shown), suggesting that the fragment is a proteolytic cleavage product of MPS-1 that retains the N-terminus and the transmembrane span but lacks the catalytic domain. To test for MPS-1 kinase activity, we carried out enzymatic reactions using myelin basic protein (MBP) as a test substrate, and found that wild-type, but not D178N, MPS-1 catalyzed the incorporation of 32P into MBP (Fig. 2c). Moreover, incorporation of the phosphate could be inhibited by the generic kinase inhibitor staurosporine (10 mM) and partially reversed by calf intestinal alkaline phosphatase in imidazole-free conditions (Fig. 2d–e). The lack of activity in the mutant confirmed that the proteolytic cleavage product was enzymatically inactive and ruled out the possibility that kinases other than MPS-1 might phosphorylate MBP—a result that was not unexpected given that E. coli does not have endogenous serine/ threonine kinases. To obtain a rough estimate of the ability of MPS-1 to phosphorylate MBP, we compared its activity to that of bovine protein kinase A (PKA), whose interaction with MBP has been extensively investigated27,28. By quantifying the bands on the autoradiogram by densitometric analysis using Gel-Doc software (Bio-Rad), we estimated that MPS-1 phosphorylates MBP approximately 20 times less efficiently than does PKA (data not shown). This low activity is not unusual29,30 and could be caused by several factors, including the relative impurity of the preparation and the fact that MBP is not a physiological substrate for MPS-1. The time course of phosphorylation of MBP by MPS-1 (Fig. 2f) showed that, as with other kinases31,32, the incorporation of 32P into MBP increased during the first B15 min and then saturated. We next assessed MPS-1–mediated phosphorylation of MBP by using rabbit antibodies to phosphoserine and phosphothreonine (antipS/pT) that have been shown to produce very specific immunological

b

c

d

reactions33. The rationale for these experiments was twofold: first, to provide an independent line of evidence in support of the notion that MPS-1 is a kinase; and second, to test the reagents that were used to determine the phosphorylation of an MPS-1 substrate in the CHO cell expression system (see below). As expected, staining with anti-pS/pT showed that a significantly greater MBP phosphorylation occurred when MPS-1 was present (Fig. 3). The antibodies also detected phosphorylated residues in MPS-1, suggesting that MPS-1 might autophosphorylate—a common feature of protein kinases34. The fact that we did not observe the incorporation of 32P into MPS-1 (Fig. 2) might indicate that purified MPS-1 is phosphorylated in E. coli, a relatively frequent occurrence35. K+ channel KVS-1 is a physiological substrate for MPS-1 We next asked whether the K+ channel KVS-1 is a substrate for MPS-1, as the two proteins form stable and functional complexes in native cells6,26. To evaluate KVS-1 phosphorylation biochemically, we used a construct, KVS-1–HA, encoding KVS-1 with the hemagglutinin (HA) epitope tag attached to its C terminus. This epitope tagging had no effect on the macroscopic channel activity (data not shown). In a typical experiment, KVS-1–HA subunits were expressed, alone or with MPS-1, in CHO cells (Fig. 4a); KVS-1 was immunoprecipitated, and protein levels were evaluated by Coomassie staining and then western blotting using antibodies to HA. The same blot was then thoroughly washed and re-stained with anti-pS/pT. Although the amounts of KVS-1 protein were similar in both lanes, anti-pS/pT staining was significantly greater in the MPS-1 + KVS-1–HA lane (Fig. 4a); this effect was suppressed by preincubation with staurosporine (Fig. 4b). In a few experiments, KVS-1 alone yielded a faint band (see, for example, Fig. 4c), an effect we ascribe to the action of endogenous kinases. Deleting the entire MPS-1 C terminus, by introducing a stop sequence at position 132 (DMPS-1; Fig. 4c), or inactivating the DFG catalytic site (Fig. 4c) was sufficient to suppress the MPS-1–mediated phosphorylation of KVS-1. To exclude the possibility that the lack of phosphorylation could be due to defective coassembly, we coimmunoprecipitated KVS-1 and MPS-1 subunits tagged with c-Myc at the C terminus (the tag had no effect on the macroscopic channel activity; data not shown). As expected, both wild-type and mutant MPS-1 subunits formed stable complexes with KVS-1 in CHO cells (Fig. 4d). Taken together, these data led us to conclude that MPS-1 is the first example of a K+ channel b-subunit that is also a protein kinase. MPS-1 activity decreases the macroscopic K+ current In CHO cells, KVS-1 subunits generated robust K+ currents characterized by an A-type profile (Fig. 5a). Coexpression with MPS-1

e

f Normalized units

a MPS-1

MBP

ni

nd uc In ed du ce d

16 kDa

U


35 70 105 140 175 210 243

MPS-1 D178N

– +

+ –

MBP MPS-1 D178N Stauro

+ – – –

+ – + –

+ + – –

+ + – +

+ + – –

MBP + MPS-1 + CIAP +

+ + –

+ – +

0

5 15 Time (min)

25

1.0 0.5 0.0 0

10 20 Time (min)

Figure 2 MPS-1 phosphorylates MBP in vitro. (a) Western blot visualization, using antibodies to histidine, of recombinant histidine-tagged MPS-1 (arrow) from insoluble fractions of E. coli lysates. The lower band is an MPS-1 proteolytic cleavage. (b) Coomassie staining of purified D178N and wild-type (MPS-1) proteins. (c) Autoradiogram of phosphorylation reactions carried out in the presence of [g-32P]ATP and B3 mg of each indicated protein. (d) As in c, with and without 10 mM staurosporine included in the reaction. (e) As in c, with and without 20 units of CIAP in imidazole-free solutions. The faint band in the first lane is probably due to residual traces of imidazole, an inhibitor of CIAP. (f) Time course of MBP phosphorylation by MPS-1, as quantified by densitometric analysis (n ¼ 2).

1504

VOLUME 8

[

NUMBER 11

[


ARTICLES

a

inactivation (data not shown) in either the homomeric KVS-1 or the heteromeric KVS-1–MPS-1 channels.

b

37 kDa MPS-1

MPS-1 activity decreases the open probability MPS-1 might decrease the macroscopic current of the complex through distinct mechanisms. For instance, phosphorylation might alter the unitary attributes of the channel or, alternatively, its expression in the plasma membrane. To distinguish between these possibilities, we used the on-cell configuration of the patch clamp to study the properties of single KVS-1 channels under three conditions: alone, with wild-type MPS-1 and with DMPS-1. Consistent with macroscopic inactivation, cell depolarization evoked transient single-channel activity characterized by brief, irregular channel opening (Fig. 6a) that subsided after a few milliseconds. Ensemble histograms constructed by summing between 100 and 300 individual traces showed that the inactivating currents could be fitted to a single exponential (Fig. 6b): the time course of inactivation, t, for the KVS-1, KVS-1–MPS-1 and KVS-1–DMPS-1 channels were, respectively, 35 ± 11 ms (mean ± s.e.m.) (n ¼ 3), 17 ± 7 ms (n ¼ 3) and 21 ± 9 ms (n ¼ 2)—in good agreement with whole-cell data. The unitary slope conductance (Fig. 6c) of the KVS-1–MPS-1 and KVS-1–DMPS-1 channels (g ¼ 19 ± 3 pS and g ¼ 20 ± 2 pS, respectively) was not substantially different from that of the KVS-1 channel (g ¼ 17 ± 2 pS); this ruled out the possibility that MPS-1–dependent phosphorylation affects the permeation of K+. We next tested the possibility that MPS-1 might affect surface expression. Accordingly, we incubated cells transfected with KVS-1 alone or with KVS-1 and MPS-1 or mock transfected with a membraneimpermeant biotin analog; we then precipitated extracts using streptavidin-agarose beads and estimated the amount of protein in the gel by densitometric analysis. The results of these experiments indicated that MPS-1 neither decreased nor increased the number of KVS-1 channels in the plasma membrane (a representative gel is shown in Fig. 6d): in fact, we estimated that the ratio of KVS-1 channels to KVS-1–MPS-1 channels was 0.9 ± 0.1 (n ¼ 3; data not shown). Similarly, MPS-1 neither increased nor decreased the total amount of KVS-1 protein that was detected upon Coomassie staining (data not shown). These results led us to conclude that MPS-1–dependent phosphorylation decreases the open probability. To evaluate this quantity, we idealized single-channel transitions and estimated the dwellclosed and dwell-open times. Both open- and closed-time distributions were well fit by two exponential components. Consistent with their

19 kDa


MBP MBP

+

+

–

MBP

+

+

–

MPS-1

–

+

+

MPS-1

–

+

+

Figure 3 MPS-1 activity can be detected by antibodies to phosphoserine and phosphothreonine (anti-pS/pT). (a) Coomassie staining of protein content of typical kinase reactions with the indicated proteins. The marker is BenchMark (Invitrogen). (b) Western blot visualization of the same sample gel with anti-pS/pT.

introduced several modifications in the current (Fig. 5), including decreased macroscopic current density (Fig. 5a–b), speeded inactivation kinetics (Fig. 5d), a leftward shift in the midpoint for activation, slower recovery from inactivation and increased susceptibility to 4-aminopyridine26. To assess whether some of these effects were due to the phosphorylation of KVS-1, we electrophysiologically characterized channel complexes composed of KVS-1 alone and of KVS-1 with wild-type MPS-1, D178N or DMPS-1. In both mutants, the magnitude of the current density was restored to values similar to those seen with KVS-1 subunits (Fig. 5b). In contrast, the inactivation kinetics (Fig. 5d), voltage activation and recovery from inactivation (data not shown) of the mutants were not substantially different from those of the channels formed with wild-type subunits. To corroborate the notion that MPS-1 kinase activity modulates the magnitude of the current density, we treated with staurosporine cells that expressed either KVS-1 channels or KVS-1–MPS-1 channels. Although preincubation with 2 mM staurosporine did not modify the characteristics of the KVS-1 channels, it led to an approximately twofold increase in the magnitude of the current produced by KVS1–MPS-1 channels (Fig. 5b). When staurosporine was dialyzed into the cell from the patch pipette solution, the current increased sigmoidally for about 15 min and then saturated—probably owing to an incomplete diffusion of staurosporine in the narrowest part of the pipette (Fig. 5c). Moreover, staurosporine had no effect on inactivation kinetics (Fig. 5d), voltage activation or recovery from

a

b

c

d

IP: Anti-HA

Co-IP

Coomassie 32 kDa Anti-S/T

Anti-HA

14 kDa KVS-1 MPS-1 D178N ∆MPS-1

Anti-S/T KVS-1 MPS-1 Stauro

+ + –

+ – –

+ + –

+ + +

+ – – –

+ – – +

+ + – –

+ – + –

+ – – –

KVS-1 MPS-1 D178N ∆MPS-1

+ – – –

+ – – +

+ – + –

+ + – –

Figure 4 MPS-1 phosphorylates KVS-1 in CHO cells. (a) Coomassie stain (top) and western blot visualization of immunoprecipitates (IP, with antibody to HA; middle) of HA epitope–tagged KVS-1 subunits with MPS-1 and alone. The membrane was washed and re-stained with rabbit antibodies to phosphoserine and phosphothreonine (anti-pS/pT; bottom). MPS-1 yielded a more intense band that could not be attributed to differences in protein levels. (b) Western blot visualizations as in a, for CHO cells co-transfected with MPS-1 and KVS-1 cDNA; not incubated (left) or preincubated with 2 mM staurosporine for 45 min (right). (c) Western blot visualizations as in a for the indicated proteins. (d) Co-immunoprecipitations of KVS-1–DMPS-1 (B14 kDa band), D178N–KVS-1 and KVS-1–MPS-1 (B32-kDa bands) channels. KVS-1 channels alone did not yield any band. The band around 25 kDa is the light chain of the antibody to hemagglutinin conjugated to the matrix.


[

NUMBER 11

[

NOVEMBER 2005

1505

faster inactivation kinetics, heteromeric (that is, KVS-1–MPS-1) channels had shorter mean open times than did KVS-1 channels; in contrast, coexpression with wild-type—but not DMPS-1—resulted in a roughly two-fold increase in dwell time in the long closed state (Table 1). Thus, shorter open times coupled with longer closed times are expected to decrease the open probability of the channel and, consequently, the macroscopic current. Because of the flickering behavior of these channels, we also used nonstationary noise variance analysis36 to obtain an independent estimate of the unitary current i and the open probability po. Representative experiments show (Fig. 7) that although the estimated values of the unitary current were similar for both the KVS-1 and KVS-1–MPS-1 channel types (i ¼ 0.78 ± 0.09 pA and i ¼ 0.74 ± 0.07 pA, respectively; n ¼ 3), the estimated open probability for the KVS-1 channel was about twice that of the KVS-1–MPS-1 channel (po ¼ 0.63 ± 0.07 and po ¼ 0.30 ± 0.06, respectively). Thus we conclude that the kinase activity of MPS-1 decreases the open probability of the complex.

a

KVS-1

KVS-1 + MPS-1

KVS-1 + ∆MPS-1

a

KVS-1

MPS-1

∆MPS-1

D178N

40 pA/pF

120 mV –80 mV

25 ms

Stauro

c

200

100

d 80

1.8 Normalized current

**

1.6 τ (ms)

*

b

1.4

60

1.2

40

1.0

20

0.8

0 KV

M

S1 PS -1 D 17 8 ∆M N PS -1 KV S M -1 PS -1

Current density (pA/pF)

Figure 5 MPS-1 kinase activity decreases the macroscopic current. (a) Whole-cell macroscopic currents generated by voltage jumps from –80 mV to +120 mV in 20 mV increments (voltage protocol shown above graph for MPS-1), in CHO cells expressing KVS-1 alone or with wild-type(MPS-1), D178N or DMPS-1. (b) Macroscopic current densities calculated by normalizing the peak current at +120 mV to the cell capacitance (filled bars; n Z 20 cells for each bar). Unshaded bars indicate current densities in the presence of 2 mM staurosporine from cells preincubated for 30 min with 2 mM staurosporine (left unshaded bar: n ¼ 10 cells; right unshaded bar: n ¼ 11 cells). Error bars represent s.e.m. Asterisks indicate significant differences in current densities (*, P o 0.05; **, P o 0.01; t-test). (c) Time course of staurosporine-dependent current increase of the KVS-1–MPS-1 channels. The current was normalized to its value at t ¼ 0—the time at which the whole-cell configuration was established. Cells were not preincubated with staurosporine. n ¼ 3 cells. (d) Inactivation rates for the KVS-1 (, n ¼ 19), KVS-1–MPS-1 (., n ¼ 22), KVS-1–D178N (’, n ¼ 13) and KVS-1–DMPS-1 (m, n ¼ 12) channels and for the KVS-1 (J, n ¼ 10) and KVS-1–MPS-1, X, n ¼ 11) channels with 2-mM staurosporine. Time constants were obtained by fitting the macroscopic currents to a single exponential function: I0 + I1exp(–t/t).

0 10 20 Time (min)

80 120 Voltage (mV)

DISCUSSION We report here the identification of a K+ channel b-subunit with kinase properties. Using 32P-radiolabeling and immunochemical methods, we found that MPS-1 can phosphorylate the classic test substrate MBP in vitro. Further, another set of immunochemical tests showed that in mammalian cells, MPS-1 assembles with and phosphorylates its physiological substrate, the K+ channel KVS-1. Notably, MPS-1 does not show sequence homologies to conventional kinases, yet it has two conserved motifs, including a DFG triplet. The fact that the aspartate in the DFG site is replaced with asparagine and that staurosporine inhibits the catalytic activity of the protein suggest that MPS-1 might operate through conventional mechanisms. In fact, sequence homology is not a prerequisite for kinase function—a-kinases, a recently discovered family37 that includes channel-kinases38,39, show no detectable sequence homologies to conventional protein kinases; yet the majority of structural elements, sequence motifs and the position of key amino acid residues important for catalysis appear to be remarkably conserved39. The bifunctional nature of MPS-1 is emphasized by the observation that the inactivation or deletion of the catalytic domain

O– C– 2 pA 50 ms

b

Ensembles 1.5 pA

3.5 pA

4.0 pA

2 pA

2.5

d K

c

15 ms

2.0

1.5 120 160 Voltage (mV)

1506

M+K

50 ms

I (pA)


ARTICLES

Figure 6 MPS-1 decreases the open probability of the KVS-1 channel. Between 50 and 300 depolarizing voltage jumps, each lasting 2–5 s, were applied consecutively (with a 1-s–long interpulse interval) so as to raise the channel voltage from –80 mV to +140 mV. The capacitative transient was eliminated by subtracting out each single trace from an ensemble of traces with no activity. (a) Representative recordings of the activity in the KVS-1, KVS-1–MPS-1 and KVS-1–DMPS-1 channels, in CHO cells at +140 mV. Closed (c-) and open (o-) states are indicated. For display purposes, the data were digitally filtered at 0.5 kHz. (b) Representative ensemble histograms of the same channels at +140 mV. The histograms were constructed by averaging 100 (KVS-1), 300 (KVS-1–MPS-1) and 200 (KVS-1–DMPS-1) traces. For each channel, the data were fitted to a single exponential (t ¼ 38 ms, t ¼ 19 ms and t ¼ 26 ms, respectively). (c) i-V characteristics of KVS-1 (), KVS-1–MPS-1 (J) and KVS-1–DMPS-1 channels (n). Each data point represents the average of two or three patches. Inset, representative idealized single-channel transitions. Top to bottom, KVS-1 channels alone, KVS-1–MPS-1 and KVS-1–DMPS-1. (d) Surface expression of KVS-1 and KVS-1–MPS-1 channels in CHO cells. First lane, mock-transfected cells. Second lane, cells transfected with KVS-1–HA but not incubated with the biotin analog. Third and fourth lanes: cells transfected, respectively, with KVS-1–HA (K) and with KVS-1–HA and MPS-1 (K+M). Western blot visualization was performed with a monoclonal antibody to HA; for fluorography, chemiluminescence was performed with a secondary antibody coupled to horseradish peroxidase.

VOLUME 8

[

NUMBER 11

[


ARTICLES Table 1 First latency and time constants of open and closed states of KVS-1, KVS-1–MPS-1 and KVS-1–DMPS-1 channels Closed times

Channel

1st lat., ms

A2/(A1+A2)

to1, ms

to2, ms

A2/(A1+A2)

tc1, ms

tc1, ms

KVS-1 (n ¼ 7) MPS-1/KVS-1 (n ¼ 6)

2.7 ± 1.3 4.6 ± 2.4

0.06 ± 0.01 0.14 ± 0.06 (**)

0.6 ± 0.1 0.7 ± 0.1

5.5 ± 1.6 2.7 ± 0.2 (*)

0.36 ± 0.1 0.32 ± 0.1

3.0 ± 0.4 3.3 ± 0.8

11.4 ± 1.6 25.1 ± 5.5 (*)

DMPS-1/KVS-1 (n ¼ 3)

3.4 ± 1.4

0.11 ± 0.06

0.7 ± 0.1

3.2 ± 0.8

0.37 ± 0.1

2.1 ± 0.8

13.9 ± 4.6

method46,

Channels were studied at +140 mV in on-cell patches. Using the maximum likelihood histograms were fitted to a double exponential function: A0+A1exp(–t1/t) + A2exp(–t2/t). The number of patches is indicated in parentheses. Asterisks indicate significant differences from the KVS-1 values (*, P o 0.05; **, P o 0.02; t-test).

gives rise to a protein that retains the ability to alter some KVS-1 functional attributes. This suggests that MPS-1 controls KVS-1 function through multiple, independent mechanisms and underscores the existence of structural and functional principles common to all KCNE b-subunits7,22,26,40–42. In silico analysis seems to exclude the existence of other MPS-1–like bifunctional proteins in the nematode’s genome. The two closest MPS-1 relatives, C. elegans MPS-3 (13% identity, 31% similarity) and human KCNE4 (13% identity, 27% similarity), lack the characteristic kinase signatures in their amino acid sequences, although this does not rule out the possibility that they might have other regulatory functions. A predicted MPS-1 homolog in which the catalytic domain is conserved is found in the genome of C. briggsae (predicted protein CBG02619, 64% identity and 17% similarity). C. elegans and C. briggsae are close relatives, although they diverged evolutionarily approximately 50 million years ago. Nonetheless, the existence of an MPS-1 homolog in another species indicates that bifunctional proteins like MPS-1 may represent a potentially general mode of K+ channel regulation in invertebrates, whereas (at present) there is no evidence that such proteins might operate in mammals. KVS-1–MPS-1 channels contribute to the total current in the chemosensory neurons of C. elegans6,26. Our findings give rise to intriguing questions about the role of MPS-1 in the sensory apparatus of the animal, because phosphorylation-dependent changes in the magnitude of the KVS-1 K+ current might have a significant impact on cell signaling. The ability of neurons in C. elegans to detect the multiple cues that the animal can distinguish43–45 supports the notion that these cells must possess broad signaling capabilities. Our data would support a model predicting that MPS-1 activity favors cellular excitability through the controlled decrease or increase of the KVS-1 potassium current. Because K+ fluxes act to stabilize the cell membrane potential, and MPS-1 decreases the current passed by KVS-1, the

a

KVS-1

KVS-1 + MPS-1

phosphorylation or dephosphorylation of this channel might have a role in the mechanisms determining sensory adaptation. Questions concerning the details of these mechanisms lie ahead and will be matter for future investigations. METHODS Molecular biology. MPS-1 mutants were constructed by polymerase chain reaction (PCR). KVS-1 was epitope tagged by replacing the terminal stop codon with nucleotides encoding HA residues (YPYDVPDYA-STOP). MPS-1 was tagged by inserting the c-Myc sequence (ISMEQKLISEEDLN). The constructs were subcloned into pcI-neo vector (Promega) for expression in CHO cells and into the pET-30b vector (Invitrogen) for expression in BL21 E. coli cells. All sequences were confirmed by automated DNA sequencing. Transcripts were quantified with spectroscopy and compared with control samples that were separated by agarose gel electrophoresis and stained with ethidium bromide. Protein purification. MPS-1 in the pET-30b vector was transformed into the E. coli strain BL21(D3) pLysS. Bacteria were grown in 1 liter of LB medium at 37 1C to an OD580 of 550. Expression of recombinant MPS-1 was induced by the addition of 400 mM isopropyl b-D-thiogalactoside (IPTG). Cells were harvested by centrifugation, resuspended in BugBuster Protein Extraction Reagent (Novagen) with nuclease and incubated at room temperature for 10 min. After centrifugation at 16,000 rpm for 30 min, the pellet was resuspended in 50 mM Tris-HCl buffer, pH 8.0, containing 500 mM NaCl, 1% n-dodecyl-b-D-maltoside, 1 mM phenylmethylsulfonyl fluoride (PMSF), protease inhibitor and 10 mM imidazole. The suspension was incubated at 4 1C for 2 h and then centrifuged at 18,000 rpm for 30 min. The supernatant was loaded onto a HisBind column (Novagen) and washed with 50-mM Tris-HCl buffer, pH 8.0, 500 mM NaCl, 0.1% n-dodecyl-b-D-maltoside and 40 mM imidazole. Elution of the fusion protein was accomplished with 400 mM imidazole in the same Tris-HCl buffer solution. For the experiments with calf intestinal alkaline phosphatase (Roche), MPS-1 was washed in Amicon Ultra binding filters (Millipore) and resuspended in an imidazole-free solution.

b 600

200 pA 50 ms

50 ms

2

2

500 pA

1,000

Variance (pA2)

500 pA

Variance (pA2)


Open times

500

0

400 pA

20 ms

200 0

0 20 ms

400

500 1,000 Mean current (pA)

0

200 400 600 Mean current (pA)

Figure 7 Fluctuation analysis of KVS-1 and KVS-1–MPS-1 channels. (a) Mean currents and time-dependent variances (at +60 mV) computed from 30 consecutive sweeps filtered at 5 kHz. (b) Mean-variance plots. Data were fitted to equation (1) as follows: i ¼ 0.81 pA and N ¼ 3,000 for the KVS-1 channels; i ¼ 0.70 pA and N ¼ 3,333 for the KVS-1–MPS-1 channels. The corresponding maximal open probability calculated from equation (2) was po ¼ 0.60 for the KVS-1 channels and po ¼ 0.32 for the KVS-1–MPS-1 channels.


[

NUMBER 11

[

NOVEMBER 2005

1507


ARTICLES Kinase assays. Catalytic reactions were carried out in 50 mM Tris-HCl buffer containing 50 mM NaCl, 5 mM MgCl2, 1 mM DTT, 5 mCi [g-32P]ATP, 100 mM ATP, phosphatase inhibitor, B3 mg MPS-1 and B3 mg MBP ([g-32P]ATP was omitted in the experiments shown in Fig. 3). The reaction was incubated for 30 min at 37 1C and then stopped by the addition of SDS sample buffer (SDS sample buffer: 3.55 ml water, 1.25 ml 0.5 M TrisHCl (adjusted to pH 6.8), 2.5 ml glycerol, 2.0 ml 10% (wt/vol) SDS, 0.2 ml 0.5% (wt/vol) bromophenol blue, freshly added 0.5 ml b-mercaptoethanol). Polyclonal rabbit antibodies against phosphoserine and phosphothreonine (anti-pS/pT) were purchased from Zymed. PKA from bovine heart was purchased from Sigma.

pulses from a 80-mV holding voltage in the whole cell configuration. Mean variance plots were fitted to s2 ¼ iI

I2 N

where s2 is the variance, i is the unitary current, I is the macroscopic current and N is the number of channels in the cell. The maximal open probability, po, was calculated as: po ¼

IMax Ni

Throughout, statistical quantities are expressed as mean ± s.e.m. Immunoprecipitations and co-immunoprecipitations. CHO cells were washed with 10 ml ice-cold Dulbecco’s phosphate-buffered saline (PBS) and lysed with B2 ml ice-cold RIPA buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% IGEPAL CA-630, 0.5% (wt/vol) deoxycholate, 0.1% (wt/vol) SDS, freshly added 10 mM iodoacetamide, phosphatase and protease inhibitors) for 30 min at 4 1C. Cell lysates were centrifuged for 60 min at 4 1C and the supernatant was mixed with HA-conjugated beads (Roche) and rocked at 4 1C for 3 h. Beads were washed three times with ice-cold TBST (Tris-buffered saline, pH adjusted to 7.6 with HCl, 0.1% Tween-20) and incubated in SDS sample buffer at B90–95 1C for 15 min. For co-immunoprecipitations, cells were lysed with 1% Nonidet P-40 buffer, 50 mM Tris, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% IGEPAL CA-630, phosphatase and protease inhibitors (Calbiochem). Cell lysates were centrifuged for 60 min at 4 1C and the supernatant mixed with HA-conjugated beads (Roche) and rocked at 4 1C for 3 h. Beads were washed three times with ice-cold 1% Nonidet P-40 buffer and then incubated in SDS sample buffer at B90–95 1C for 15 min.

ACKNOWLEDGMENTS We thank L. Runnels, A. Ryazanov and J. Lenard for their comments on the manuscript and Fulvio Sesti for help with the graphics. This work was supported by grant R01GM68581-01 from the US National Institutes of Health to F.S. COMPETING INTERESTS STATEMENT The author declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

Electrophysiology. CHO cells were transiently transfected with cDNA using Superfect kit (Qiagen) and studied for 24–36 h after transfection. Data were recorded with an Axopatch 200B (Axon), a PC (Dell) and Clampex software (Axon), and filtered at fc ¼ 1 kHz and sampled at 2.5 kHz (whole cell) or fc ¼ 5 kHz and sampled at 15 kHz (single-channel and noise analysis). Bath solution was (in mM): 4 KCl, 100 NaCl, 10 HEPES (pH adjusted to 7.5 with NaOH), 1.8 CaCl2 and 1.0 MgCl2. Pipette solution (in mM): 100 KCl, 10 HEPES (pH adjusted to 7.5 with KOH), 1.0 MgCl2, 1.0 CaCl2, 10 EGTA (pH adjusted to 7.5 with KOH). Addition of 2 mM magnesium ATP (Sigma) in the pipette solution had no effect on the macroscopic currents, and therefore ATP was generally omitted. Whole-cell currents were evoked by 0.1-s voltage sweeps from a holding potential of 80 mV to +120 mV, in 20-mV increments. In the experiments with staurosporine, CHO cells were preincubated for 30 min with 2 mM staurosporine (Sigma) freshly dissolved in the medium from a 2-mM stock solution (in dimethyl sulfoxide (DMSO)) and used for no longer than 30 min. During measurements, staurosporine was maintained in the bath solution at the same concentration. The concentration of DMSO in the final dilution was 0.1%, a level at which DMSO had no effect on the current (data not shown). To dialyze staurosporine from the patch pipette, we followed the procedure used with the perforated-patch technique46. Thus, the pipette was dipped in a standard staurosporine-free solution for B2 s and then was back-filled with a staurosporine-containing solution. After a gigaohm seal was established, we waited for 20 min before establishing the whole-cell configuration. Voltage protocols for single-channel analysis consisted of a train of 100–300 single 2–5-s sweeps from a holding voltage of 80 mV to +120 mV, +140 mV or +160 mV. Single-channel activity was analyzed manually by pClamp 8 software (Axon). Dead time was calculated as Td ¼ 0.18/fc ¼ 36 ms46. For noise analysis, macroscopic currents were induced by 50 consecutive +60-mV depolarizing

1. MacKinnon, R. Determination of the subunit stoichiometry of a voltage-activated potassium channel. Nature 350, 232–235 (1991). 2. Doyle, D. et al. The structure of the potassium channel: molecular bases for K+ conduction and selectivity. Science 280, 69–77 (1998). 3. Abbott, G. & Goldstein, S. A superfamily of small potassium channel subunits: form and function of the MinK-related peptides (MiRPs). Q. Rev. Biophys. 31, 357–398 (1998). 4. McCrossan, Z.A. & Abbott, G.W. The MinK-related peptides. Neuropharmacology 47, 787–821 (2004). 5. Wang, Y., Park, K.H., Hernandez, L., Cai, S.-Q. & Sesti, F. Biophysical and biomedical aspects of KCNE potassium channel ancillary subunits. in Recent Research Developments in Biophysics Vol. 3 Part II (ed. Pandalai, S.G.) 351–363 (Transworld Research Network, Trivandrum, India, 2004). 6. Park, K.H., Hernandez, L., Cai, S.Q., Wang, Y. & Sesti, F. A Family of K+ Channel Ancillary Subunits Regulate Taste Sensitivity in Caenorhabditis elegans. J. Biol. Chem. 280, 21893–21899 (2005). 7. Anantharam, A. et al. RNA interference reveals that endogenous Xenopus MinK-related peptides govern mammalian K+ channel function in oocyte expression studies. J. Biol. Chem. 278, 11739–11745 (2003). 8. Takumi, T., Ohkubo, H. & Nakanishi, S. Cloning of a membrane protein that induces a slow voltage-gated potassium current. Science 242, 1042–1045 (1988). 9. Tai, K. & Goldstein, S. The conduction pore of a cardiac potassium channel. Nature 391, 605–608 (1998). 10. Melman, Y.F., Um, S.Y., Krumerman, A., Kagan, A. & McDonald, T.V. KCNE1 binds to the KCNQ1 pore to regulate potassium channel activity. Neuron 42, 927–937 (2004). 11. Chen, H., Sesti, F. & Goldstein, S.A. Pore- and state-dependent cadmium block of I(Ks) channels formed with MinK-55C and wild-type KCNQ1 subunits. Biophys. J. 84, 3679– 3689 (2003). 12. Sesti, F. & Goldstein, S.A. Single-channel characteristics of wild-type IKs channels and channels formed with two minK mutants that cause long QT syndrome. J. Gen. Physiol. 112, 651–663 (1998). 13. Sesti, F., Tai, K.K. & Goldstein, S.A. MinK endows the I(Ks) potassium channel pore with sensitivity to internal tetraethylammonium. Biophys. J. 79, 1369–1378 (2000). 14. Yang, Y. & Sigworth, F.J. Single-channel properties of IKs potassium channels. J. Gen. Physiol. 112, 665–678 (1998). 15. Pusch, M. Increase of the single-channel conductance of KvLQT1 potassium channels induced by the association with minK. Pflugers Arch. 437, 172–174 (1998). 16. Marx, S.O. et al. Requirement of a macromolecular signaling complex for beta adrenergic receptor modulation of the KCNQ1-KCNE1 potassium channel. Science 295, 496–499 (2002). 17. Kurokawa, J., Chen, L. & Kass, R.S. Requirement of subunit expression for cAMPmediated regulation of a heart potassium channel. Proc. Natl. Acad. Sci. USA 100, 2122–2127 (2003). 18. McCrossan, Z.A. et al. MinK-related peptide 2 modulates Kv2.1 and Kv3.1 potassium channels in mammalian brain. J Neurosci. 3;23, 8077–91 (2003). 19. Splawski, I., Tristani-Firouzi, M., Lehmann, M.H., Sanguinetti, M.C. & Keating, M.T. Mutations in the hMinK gene cause long QT syndrome and suppress IKs function. Nat. Genet. 17, 338–340 (1997). 20. Piccini, M. et al. KCNE1-like gene is deleted in AMME contiguous gene syndrome: identification and characterization of the mouse homologues. Genomics 60, 251–257 (1999). 21. Abbott, G. et al. MiRP2 forms potassium channels in skeletal muscle with Kv3.4 and is associated with periodic paralysis. Cell 104, 217–231 (2001). 22. Abbott, G. et al. MiRP1 forms IKr potassium channels with HERG and is associated with cardiac arrhythmia. Cell 97, 175–187 (1999).

1508

VOLUME 8

Membrane biotinylation. At 30 h after transfection, CHO cells were washed three times with PBS at room temperature (22–25 1C) and cell surface proteins were biotinylated by 1.0 mg/ml of the impermeant biotin analog EZ-link sulfo-NHS-Lc-biotin (Pierce) in PBS. After incubation at 4 1C for 1 h, cells were washed five times with ice-cold PBS to remove any remaining biotinylation reagent. Cells were then harvested in RIPA buffer. Lysate proteins were precipitated with streptavidin-agarose beads. The precipitated KVS-1 was detected by monoclonal antibodies to HA (Roche).

[

NUMBER 11

[



ARTICLES 23. Sesti, F. et al. A common polymorphism associated with antibiotic-induced cardiac arrhythmia. Proc. Natl. Acad. Sci. USA 97, 10613–10618 (2000). 24. Splawski, I. et al. Spectrum of mutations in long-QT syndrome genes. KVLQT1, HERG, SCN5A, KCNE1, and KCNE2. Circulation 102, 1178–1185 (2000). 25. Manning, G., Whyte, D.B., Martinez, R., Hunter, T. & Sudarsanam, S. The protein kinase complement of the human genome. Science 298, 1912–1934 (2002). 26. Bianchi, L., Kwok, S.M., Driscoll, M. & Sesti, F. A potassium channel-MiRP complex controls neurosensory function in Caenorhabditis elegans. J. Biol. Chem. 278, 12415– 12424 (2003). 27. Hwang, I-S., Kim, J-H. & Choi, M-U. Kinetic study of dephosphoryltation of Myelin Basic Protein by some protein phosphates. Bull. Korean Chem. Soc. 18, 428–432 (1997). 28. Miyamoto, E. & Kakiuchi, S. In vitro and in vivo phosphorylation of myelin basic protein by exogenous and endogenous adenosine 3¢:5¢-monophosphate-dependent protein kinases in brain. J. Biol. Chem. 249, 2769–2777 (1974). 29. Laurino, J., Colca, J., Pearson, J., DeWald, D. & McDonald, J. The in vitro phosphorylation of calmodulin by the insulin receptor tyrosine kinase. Arch. Biochem. Biophys. 265, 8–21 (1988). 30. Tonks, N., Diltz, C. & Fischer, E. CD45, an integral membrane protein tyrosine phosphatase. Characterization of enzyme activity. J. Biol. Chem. 265, 10674–10680 (1990). 31. Ryazanova, L.V., Dorovkov, M.V., Ansari, A. & Ryazanov, A.G. Characterization of the protein kinase activity of TRPM7/ChaK1, a protein kinase fused to the transient receptor potential ion channel. J. Biol. Chem. 279, 3708–3716 (2004). 32. Beeton, C.A., Chance, E.M., Foukas, L.C. & Shepherd, P.R. Comparison of the kinetic properties of the lipid- and protein-kinase activities of the p110alpha and p110beta catalytic subunits of class-Ia phosphoinositide 3-kinases. Biochem. J. 350, 353–359 (2000). 33. Rintamaki, E. et al. Phosphorylation of light-harvesting complex II and photosystem II core proteins shows different irradiance-dependent regulation in vivo. Application of phosphothreonine antibodies to analysis of thylakoid phosphoproteins. J. Biol. Chem. 272, 30476–30482 (1997).


[

NUMBER 11

[

NOVEMBER 2005

34. Zheng, B. et al. The mPer2 gene encodes a functional component of the mammalian circadian clock. Nature 400, 169–173 (1999). 35. Du, P. et al. Phosphorylation of serine residues in histidine-tag sequences attached to recombinant protein kinases: a cause of heterogeneity in mass and complications in function. Protein Expr. Purif. published online 7 June 2005 (doi:10.1016/ j.pep.2005.04.018). 36. Sigworth, F.J. The variance of sodium current fluctuations at the node of Ranvier. J. Physiol. (Lond.) 307, 97–129 (1980). 37. Ryazanov, A.G., Pavur, K.S. & Dorovkov, M.V. Alpha-kinases: a new class of protein kinases with a novel catalytic domain. Curr. Biol. 9, R43–R45 (1999). 38. Runnels, L.W., Yue, L. & Clapham, D.E. TRP-PLIK, a bifunctional protein with kinase and ion channel activities. Science 291, 1043–1047 (2001). 39. Drennan, D. & Ryazanov, A.G. Alpha-kinases: analysis of the family and comparison with conventional protein kinases. Prog. Biophys. Mol. Biol. 85, 1–32 (2004). 40. Barhanin, J. et al. K(V)LQT1 and lsK (minK) proteins associate to form the I(Ks) cardiac potassium current. Nature 384, 78–80 (1996). 41. Schroeder, B. et al. A constitutively open potassium channel formed by KCNQ1 and KCNE3. Nature 403, 196–199 (2000). 42. Melman, Y.F., Domenech, A., de la Luna, S. & McDonald, T.V. Structural determinants of KvLQT1 control by the KCNE family of proteins. J. Biol. Chem. 276, 6439–6444 (2001). 43. Ward, S. Chemotaxis by the nematode Caenorhabditis elegans: identification of attractants and analysis of the response by use of mutants. Proc. Natl. Acad. Sci. USA 70, 817–821 (1973). 44. Bargmann, C. & Mori, I. in C. elegans II (eds. Riddle, D.L., Blumenthal, T., Meyer, B.T. & Priess, J.R.) Ch. 25 717–37 (Cold Spring Harbor Laboratory Press, Cold Spring Harbor, New York, 1997). 45. Bargmann, C.I. & Horvitz, H.R. Chemosensory neurons with overlapping functions direct chemotaxis to multiple chemicals in C. elegans. Neuron 7, 729–742 (1991). 46. Sakmann, B. & Neher, E. (eds) Single-Channel Recording (Plenum Press, New York and London, 1995).

1509


ARTICLES

Lbx1 and Tlx3 are opposing switches in determining GABAergic versus glutamatergic transmitter phenotypes Leping Cheng1,2,5, Omar Abdel Samad1,2, Yi Xu1,2, Rumiko Mizuguchi3, Ping Luo1,2,5, Senji Shirasawa4, Martyn Goulding3 & Qiufu Ma1,2 Most neurons in vertebrates make a developmental choice between two principal neurotransmitter phenotypes (glutamatergic versus GABAergic). Here we show that the homeobox gene Lbx1 determines a GABAergic cell fate in the dorsal spinal cord at early embryonic stages. In Lbx1–/– mice, the presumptive GABAergic neurons are transformed into glutamatergic cells. Furthermore, overexpression of Lbx1 in the chick spinal cord is sufficient to induce GABAergic differentiation. Paradoxically, Lbx1 is also expressed in glutamatergic neurons. We previously reported that the homeobox genes Tlx1 and Tlx3 determine glutamatergic cell fate. Here we show that impaired glutamatergic differentiation, observed in Tlx3–/– mice, is restored in Tlx3–/– Lbx1–/– mice. These genetic studies suggest that Lbx1 expression defines a basal GABAergic differentiation state, and Tlx3 acts to antagonize Lbx1 to promote glutamatergic differentiation.

Neurons in the dorsal spinal cord form a network that integrates and transmits diverse types of somatic sensory information, such as pain, touch, itching and temperature1,2. Dorsal horn neurons can be grouped as either excitatory neurons that utilize glutamate as their transmitter or inhibitory interneurons that use GABA and/or glycine3–6. It is generally believed that a balance between excitation and inhibition within this dorsal horn neuron network ‘gates’ the relaying of the somatic sensory information to the CNS5–10. With respect to pain, disinhibition in this network resulting from the damage to the peripheral nerves or to the spinal cord often leads to excess excitation and chronic pain disorders, the management of which remains a major medical problem11. The molecular mechanisms that govern the fate choices between the excitatory and inhibitory cell fates in the dorsal horn are only beginning to be understood12. Previously, we reported that the Tlx-class homeobox genes Tlx1 and Tlx3 act as genetic switches that select a glutamatergic over a GABAergic transmitter phenotype in the dorsal spinal cord13, and a similar binary decision between glutamatergic and GABAergic neurons has been demonstrated in the developing forebrain14. Nonetheless, several key questions remain. First, it is unclear whether the Tlx genes act directly or indirectly to promote glutamatergic differentiation. Second, the opposing switches that select a GABAergic over a glutamatergic cell fate have not been characterized. Although several genes have been implicated in controlling GABAergic differentiation, including the Dlx-class homeobox genes in the forebrain, the basic helix-loop-helix gene Ptf1a in the cerebellum and the paired class homeobox gene Pax2 in the dorsal spinal cord, mutations of Pax2 and the Dlx genes do not cause a switch from a GABAergic to a glutamatergic cell fate13,15,16.

In the dorsal neural tube, neurogenesis in the dorsal neural tube is marked by the emergence of two major classes of neurons that are defined by their differential reliance on Lbx1, a homeobox gene, and Olig3, a basic helix-loop-helix class transcription factor gene17–20. Class A neurons develop from the most dorsal neural precursors that express Olig3, they do not express Lbx1 and they settle in the deep lamina of the dorsal horn. Class B neurons develop from Olig3-negative precursors, express Lbx1, migrate dorsally and occupy the superficial lamina (Fig. 1a)17,18,20. Class B neurons are further divided into two subgroups on the basis of their complementary expression of the homeobox genes Pax2 and Tlx313,17,18,20. We have recently shown that at embryonic day 13.5 (E13.5), Pax2+ and Tlx3+ populations represent GABAergic and glutamatergic neurons, respectively13. Lbx1 is therefore expressed in both excitatory and inhibitory neurons at early embryonic stages (Supplementary Fig. 1). In this study, we show that the expression of Lbx1 in class B neurons defines a basal GABAergic differentiation state for dorsal horn neurons. Tlx3 antagonizes Lbx1, which in turn allows a subset of Lbx1+ cells to differentiate into glutamatergic neurons. RESULTS Lbx1 controls GABAergic differentiation Lbx1 deficiency leads to a progressive loss of dorsal horn neurons starting at E13.5 (ref. 17; by E14.5, GABAergic neurons are markedly reduced in these mice18). To determine if Lbx1 is required to specify the GABAergic transmitter phenotype, we reexamined the Lbx1 null phenotype at stages E11.5 and E12.5, the stages before increased cell death is first detected17. GABAergic neuron markers are markedly

1Dana-Farber Cancer Institute and 2Department of Neurobiology, Harvard Medical School, 1 Jimmy Fund Way, Boston, Massachusetts 02115, USA. 3Molecular Neurobiology Laboratory, The Salk Institute for Biological Studies, 10010 North Torrey Pines Rd., La Jolla, California 92037, USA. 4Department of Pathology, International Medical Center of Japan, 1-21-1 Toyama, Shinjuku-ku, Tokyo 162-8655, Japan. 5Present address: Institute of Biochemistry and Cell Biology, Shanghai Institute for Biological Sciences, Chinese Academy of Sciences, 320 Yueyang Road, Shanghai 200031, China. Correspondence should be addressed to Q.M. ([email protected]).

Received 25 August; accepted 19 September; published online 23 October 2005; corrected after print 21 November 2005; doi:10.1038/nn1569

1510

VOLUME 8

[

NUMBER 11

[


ARTICLES

a

Figure 1 Lbx1 expression and its requirement for GABAergic differentiation. (a) Broad Lbx1 expression was detected by in situ hybridization in the superficial E13.5 wild-type dorsal horn. (b–e) In situ hybridization with indicated probes on transverse sections through E12.5 wild-type and Lbx1–/– dorsal spinal cord. Note a loss of Gad1 and Viaat expression in the dorsal (arrows) but not the ventral (arrowheads) spinal cord. Scale bars: 50 mm.

Lbx1

E13.5

b

c

+/+

Lbx1 −/−


Gad1

E12.5

d

e Viaat

corresponded well with the net reduction of Gad1+ neurons (n ¼ 260, see above), suggesting that dorsal neurons switch their neurotransmitter fate. Several observations further supported a transformation of prospective GABAergic neurons into glutamatergic cells in the Lbx1–/– dorsal horn. In E12.5 wild-type dorsal spinal cord, GABAergic neurons coexpressed the transcription factors Pax2 and Lhx1/5 (Fig. 2b)13,17,18. The expression of Pax2, which is required for GABAergic differentiation13, was markedly reduced in the E12.5 Lbx1–/– dorsal horn neurons at the lumbar level17. However, Lhx1/5 expression persisted in E12.5 Lbx1–/– dorsal spinal cord (Fig. 2c), and the majority of Lhx1/5+ neurons in the dorsal lateral area coexpressed VGLUT2 (Fig. 2c), which was in sharp contrast to the nonoverlapping expression of these two genes in wild-type embryos (Fig. 2c). Combined with the net increase of VGLUT2+ neurons, these findings suggested that in the absence of Lbx1, the presumable GABAergic neurons were transformed into glutamatergic cells at E12.5. Lbx1 thus acts to promote a GABAergic over a glutamatergic cell fate.

E12.5

reduced in Lbx1–/– dorsal spinal cord at E11.5 and E12.5, including Gad1 (also known as GAD67) and Gad2 (also known as Gad65; Fig. 1b,c and data not shown), which encode the enzymes necessary for GABA synthesis21, and Viaat (Fig. 1d,e), which encodes the transporter that packages GABA into synaptic vesicles22. The number of Gad1+ neurons per entire transverse section through E12.5 lumbar spinal cord is significantly lower in Lbx1–/– mice (304 ± 13) than in wild-type mice (564 ± 26; P o 0.001). These data suggest that Lbx1 is required for the initial specification of dorsal spinal GABAergic neurons at E11.5 and E12.5.

Lbx1 action is independent of Tlx genes The dual functions of Lbx1 raised the question of how glutamatergic neurons emerge from Lbx1-expressing cells in the dorsal horn. We previously showed that Tlx3 and Tlx1 determine a glutamatergic over a GABAergic neuron cell fate in the dorsal horn13. We therefore asked whether the expansion of VGLUT2+ neurons that was observed in the Lbx1–/– dorsal horn was caused by the derepression of Tlx genes in prospective GABAergic neurons. Notably, no expansion of Tlx

a

VGLUT2 Lbx1 −/−

+/+

b

E13.5 wild-type

E12.5

Lbx1 suppresses glutamatergic differentiation As Lbx1 is also expressed in Tlx3+ glutamatergic neurons (Supplementary Fig. 1), we assessed whether Lbx1 has an independent role in controlling glutamatergic differentiation by examining VGLUT2 expression in the Lbx1–/– dorsal horn. VGLUT2 encodes the transporter that packages glutamate into excitatory synaptic vesicles; thus, it is a prospective marker for glutamatergic neurons in the embryonic dorsal spinal cord13,23. Notably, rather than seeing a reduction of VGLUT2 expression, we observed a marked increase (Fig. 2a) in dorsal VGLUT2+ neurons per transverse section through E12.5 caudal spinal cord (536 ± 29 in wild-type mice versus 760 ± 9 in Lbx1–/– mice, P o 0.001). Indeed, the net increase of VGLUT2+ neurons (n ¼ 224)

Pax2

Lhx1/5

Merged

c

Lhx1/5, VGLUT2 +/+

Lbx1 −/−

E12.5

Figure 2 Expansion of glutamatergic neurons in Lbx1–/– dorsal spinal cord. (a) In situ hybridization with the VGLUT2 probe on transverse sections through E12.5 wild-type (+/+) and Lbx1–/– dorsal spinal cord. (b) Double immunostaining with Pax2 antibody and Lhx1/5 antibody in E12.5 wild-type dorsal spinal cord. Yellow color on the merged image indicates colocalization. (c) Double immunostaining of VGLUT2 mRNA and Lhx1/5 protein on transverse sections through the dorsal lateral areas of E12.5 wild-type (+/+) and Lbx1–/– dorsal horn. Asterisks (*) indicate the nuclei. Scale bars: 50 mm.


[

NUMBER 11

[

NOVEMBER 2005

1511

ARTICLES

a

Tlx3 −/−

Lbx1

+/+


E12.5

b

Tlx3, Lhx1/5 +/+

Lbx1−/−

E12.5

expression was observed in Lbx1–/– dorsal horn (Fig. 3a). Furthermore, in contrast to the expansion of VGLUT2 expression in Lhx1/5+ neurons (Fig. 2c), Tlx3 and Lhx1/5 expression remained intermingled in both wild-type and Lbx1–/– dorsal horns (Fig. 3b), implying that Lbx1 suppresses glutamatergic differentiation through a pathway independent of Tlx genes. Tlx3 antagonizes Lbx1 Thus far, we have shown that (i) Lbx1 and Tlx3 are coexpressed in glutamatergic neurons, (ii) Lbx1 and Tlx3 acts as opposing switches in determining a glutamatergic versus a GABAergic cell fate and (iii) in the absence of Lbx1, the hidden glutamatergic differentiation program emerges in prospective GABAergic neurons without the involvement of Tlx3 (Figs. 2,3). Given these observations, we postulated that one way for Tlx3 to promote a glutamatergic cell fate would be to antagonize Lbx1 function (for a schematic description, see Supplementary Fig. 2). A key prediction for this epitasis would be that the impaired glutamatergic differentiation observed in Tlx–/– dorsal horn should be restored in Lbx1 and Tlx compound deficient (Lbx1–/–Tlx–/–) embryos. To test this model, we generated and analyzed the phenotypes of Lbx1–/–Tlx–/– mice. In doing this, we took advantage of the finding that redundancy between Tlx3 and Tlx1 is restricted to the rostral spinal cord13,24. At the hindlimb level, where only Tlx3 is expressed in the dorsal neural tube24, glutamatergic differentiation was nearly completely abolished in Tlx3 single-mutant dorsal horn (Fig. 4a,b). We therefore made Lbx1–/–

Figure 3 Lbx1 suppresses glutamatergic differentiation through a Tlx3independent pathway. (a) In situ hybridization with the Tlx3 probe on transverse sections through E12.5 wild-type and Lbx1–/– dorsal spinal cord at the lumbar level. (b) Double immunostaining of Tlx3 and Lhx1/5 on transverse sections through E12.5 wild-type and Lbx1–/– dorsal spinal cord. Scale bars: 50 mm.

Tlx3–/– embryos and analyzed the phenotypes at the hindlimb level. We detected extensive VGLUT2 expression in Lbx1–/–Tlx3–/– dorsal horn at E14.5 (Fig. 4c), in marked contrast to the elimination of VGLUT2 expression in Tlx3–/– dorsal horn (Fig. 4b). A portion of VGLUT2+ neurons should be transformed from presumptive GABAergic neurons; this does indeed occur in the absence of Lbx1 through a pathway independent of Tlx3 (Figs. 2,3). To analyze the fate of the presumptive Tlx3-dependent glutamatergic cells, we took advantage of the finding that expression of Lmx1b, a homeobox gene that colocalizes with Tlx3 in wild-type glutamatergic neurons13, can still be detected by in situ hybridization in E14.5 Lbx1–/–Tlx3–/– embryos (Fig. 4d). To determine if mutant Lmx1b+ neurons were glutamatergic, we counted the numbers of cells expressing Lmx1b, VGLUT2 or Stmn2. (Stmn2, also known as SCG10, is a panneuronal marker25 used for determining total neuron numbers.) We chose the most dorsolateral region (Fig. 4d) because of the extensive expression of Lmx1b. We found that 78% and 93% of neurons in this area expressed Lmx1b and VGLUT2, respectively. Accordingly, at least 71% of Lmx1b+ neurons coexpressed VGLUT2 (78 – (100 – 93) ¼ 71), suggesting that specification of the glutamatergic transmitter phenotype proceeded normally in Lbx1–/–Tlx3–/– dorsal horn, thus supporting the model that Tlx3 acts primarily to antagonize Lbx1 and thereby promote an excitatory cell fate. In Tlx mutants, the prospective glutamatergic neurons are transformed into GABAergic neurons13. To determine if Tlx suppresses GABAergic differentiation by antagonizing Lbx1, we examined GABAergic neuronal markers in the wild-type, Tlx3–/–, Lbx1–/– and Lbx1–/–Tlx3–/– dorsal horn. We found that the expression of Pax2, Gad1/2 and Viaat was expanded in E12.5 Tlx3–/–dorsal horn at the lumbar level (Fig. 5 and data not shown), as previously reported13, but it was markedly reduced in either the Lbx1–/– or Lbx1–/–Tlx3–/– dorsal spinal cord (Fig. 5 and data not shown). The phenotypes were virtually identical between Lbx1–/– single and Lbx1–/–Tlx3–/– compound mutants in caudal E12.5 dorsal spinal cord, suggesting that in the absence of Lbx1, loss of Tlx3 is no longer able to transform the prospective

a

b

+/+

Tlx3 −/−

c

Tlx3 −/−Lbx1 −/−

E14.5

Figure 4 Tlx3 antagonizes Lbx1 to promote glutamatergic differentiation. (a–c) In situ hybridization was performed with the VGLUT2 probe on transverse sections through the dorsal spinal cord of E14.5 embryos at the caudal lumbar level (with indicated genotypes). (d) Counting the number of Lmx1b+VGLUT2+ and Stmmn+ neurons. Adjacent sections through E14.5 Lbx1–/–Tlx3–/– caudal lumbar spinal cord were subjected to in situ hybridization with VGLUT2, Lmx1b and Stmmn2 probes. The pan-neuronal marker Stmmn2 was used to estimate total neuronal cell numbers and to calculate the percentages of cells expressing VGLUT2 or Lmx1b. Scale bars: 50 mm.

1512

d

Tlx3 −/−Lbx1 −/− Lmx1b

Numbers of positive cells in Lbx1−/−Tlx3−/− mutants (boxed areas) +

VGLUT2

Stmmn2

288 344 368

Lmx1b + VGLUT2 + Stmmn2

% of Lmx1b+ cells: 78 +

% of VGLUT2 cells: 93

VOLUME 8

[

NUMBER 11

[


ARTICLES glutamatergic neurons into GABAergic cells. Tlx3 thus acts to antagonize Lbx1 to both promote glutamatergic and suppress GABAergic differentiation (Supplementary Fig. 2).

−/−

Tlx3

Lbx1−/−

Tlx3 −/−Lbx1 −/−

Cell fate switch by Lbx1 overexpression To evaluate if Tlx3 or its downstream targets act to antagonize or sequester Lbx1 protein function, we examined if overexpression of Lbx1 in the developing chick spinal cord was able to overcome Tlx3-mediated inhibition. We electroporated stage 13–14 chick brachial neural tubes with a human Lbx1 expression construct and analyzed the electroporated neural tubes 2 d later (E4.5) (Fig. 6a–c). In E4.5 dorsal neural tube (equivalent to E11.5 mouse dorsal neural tube), six classes of dorsal horn neurons (DI1-6) are formed along the dorsoventral axis, including Tlx3+ DI3 and DI5, and Pax2+ DI4 and DI6 (Supplementary Figure 5 Tlx3 antagonizes Lbx1 to suppress GABAergic differentiation. In situ hybridization with Fig. 1)17,18,24,26–29. We found that these early- indicated probes on transverse sections through the dorsal spinal cord of E12.5 embryos at the caudal born Tlx3+ and Pax2+ cells differentiated as lumbar level (with indicated genotypes). Scale bars: 50 mm. glutamatergic and GABAergic neurons, respectively (Supplementary Fig. 1). Accordingly, expression of VGLUT2 and Gad1 on the nonelectroporated side glutamatergic differentiation13. It is noteworthy that the onset of Lbx1 typically appeared in alternating stripes along the dorsoventral axis and Pax2 expression in E11.5 dorsal spinal cord is independent of (Fig. 6b,c). However, after ectopic expression of Lbx1 on the electro- signals released from the roof plate18, suggesting that the GABAergic porated side (Fig. 6a), there was a marked reduction of VGLUT2 cell fate might represent a ground differentiation state for dorsal horn expression (Fig. 6b). At the same time, expression of Gad1 was neurons, at least at early embryonic stages. expanded into the areas where glutamatergic differentiation normally There seem to be two distinct mechanisms acting to generate occurs, resulting in a continuous band of Gad1-expressing cells on the glutamatergic neurons in the dorsal neural tube. At early times, signals electroporated side (Fig. 6c). These findings provide evidence that secreted from the roof plate suppress Lbx1 expression in class A overexpression of Lbx1 is sufficient to switch differentiating neurons, interneurons18, all of which go on to differentiate as glutamatergic including Tlx3+ neurons, from a glutamatergic to a GABAergic cell fate. neurons (Supplementary Fig. 1). For class B neurons, Tlx3 suppresses the function of Lbx1, which in turn allows the emergence of glutamatergic neurons from a subset of Lbx1+ cells. The coexpression of Tlx3 DISCUSSION Our genetic data suggest that during the period when neurotransmitter and Lbx1 in glutamatergic neurons (Supplementary Fig. 1) also phenotypes are first established at embryonic stages E11.5 to E12.5, the suggests that Tlx3 does not repress Lbx1 expression per se. Instead, post-mitotic homeobox gene Lbx1 actively promotes a GABAergic over Tlx3 may act to suppress Lbx1 protein function. It is notable that an a glutamatergic neuron cell fate in the dorsal neural tube. Loss of overexpression of Lbx1 in chick neural tube is able to overcome Tlx3Lbx1 results in a transformation of prospective GABAergic neurons mediated suppression and thereby promote GABAergic differentiation into glutamatergic neurons at E12.5, whereas ectopic expression of in presumptive glutamatergic neurons (Fig. 6). Therefore, one possiLbx1 in the chick spinal cord is sufficient to cause the opposite bility is that Tlx3 or its downstream targets might sequester Lbx1, for switch, from glutamatergic to GABAergic. This ‘switch’ activity distin- example, through a protein-protein interaction. Because Tlx3 has an guishes Lbx1 from its downstream target Pax2 (ref. 17), which is indirect role, the positive regulators directly responsible for VGLUT2 required for GABAergic differentiation but is incapable of suppressing upregulation in glutamatergic neurons remain to be determined. Gad1


Pax2

+/+

a

Lbx1

b

c

VGLUT2

Gad1

DI1−3 DI4 DI5 DI6


[

NUMBER 11

[

NOVEMBER 2005

Figure 6 Cell fate switch by Lbx1 ectopic expression. (a–c) Transverse sections through an E4.5 chick spinal cord, with right side of the neural tubes electroporated with human Lbx1 expression construct (RCAS-Lbx1). In situ hybridization was performed with Lbx1, Gad1 and VGLUT2 as the probes. Lbx1 is expressed throughout the entire spinal cord on the electroporated side (a, arrow). On the nonelectroporated side, expression of VGLUT2 and Gad1 appears complementary (b and c, arrows). As Pax2 is associated with GABAergic neurons, the two most dorsal patches of Gad1+ neurons (c) are likely to correspond to Pax2+ DI4 and DI6 neurons, respectively. The intermediate stripe of VGLUT2+ neurons (b, ventral arrow) are likely to correspond to Tlx3+ DI5 cells, whereas the most dorsal group of VGLUT2+ cells may correspond to DI1-3 cells (b, dorsal arrow) (Supplementary Fig. 1). On the electroporated side, expression of VGLUT2 was markedly repressed (b, arrowheads versus arrows), whereas Gad1 expression was accordingly increased (c, arrowheads versus arrows).

1513


ARTICLES A recent study in the embryonic Xenopus laevis spinal cord shows that neuronal activity can regulate excitatory and inhibitory transmitter phenotypes in a homeostatic fashion30. Our findings are consistent with the concept that neurons can switch their neurotransmitter phenotype, in that mutation of Tlx3 and Lbx1, two homeobox genes that are expressed in postmitotic neurons17,18,24, are able to cause a switch between glutamatergic and GABAergic transmitter phenotypes. On the other hand, the expression of Tlx3 and Lbx1 is subject to precise temporal and spatial control17,18,24. We postulate that transmitter specification might be initially subject to a ‘hard-wired’ genetic control, through the regulation of the expression of intrinsic switch genes like Tlx3 and Lbx1, and that neuronal activity might control the plasticity of transmitter phenotypes at later developmental stages. With regard to the latter, it is noteworthy that although GABAergic differentiation is virtually eliminated in the E11.5 and E12.5 Lbx1–/– dorsal spinal cord, a group of GABAergic neurons reemerged at E13.5 and E14.5 in Lbx1–/– mice (data not shown). A progressive reduction of GABAergic neurons in the wild-type rat spinal cord at prenatal and postnatal stages has also been reported31,32, although it is not clear whether this involves an in vivo switch between GABAergic and glutamatergic neuron transmitter programs. In summary, our studies suggest that Lbx1 and Tlx act as opposing switches that specify a GABAergic versus a glutamatergic cell fate. Expression of these genes is, however, restricted primarily to the dorsal spinal cord and several hindbrain nuclei17,18, reinforcing the idea that specification of the two principal excitatory or inhibitory neurotransmitters in the vertebrate brain is not controlled by a coherent genetic program; rather, it is regulated by a set of region-specific transcriptional factors13,15,16. As excess excitation in the dorsal horn neuron network has been implicated in the development of chronic pain disorders11, the identification of a dorsal horn–specific molecular pathway that determines the neurotransmitter profile will aid in the design of new strategies for pain therapy.

were hybridized with Gad1 or VGLUT2. Positive cells containing nuclei in the entire spinal cord sections were counted. Values were presented as mean ± s.d. The differences in values were considered to be significant at P o 0.05 by Student’s t-test. Counting of VGLUT2+, Lmx1b+ and SCG10+ in Lbx–/–Tlx3–/– double-mutants (Fig. 4) was performed as follows. Three adjacent sets of sections (15 mm thickness) through E14.5 lumbar spinal cord of Lbx–/–Tlx3–/– double-mutants were subjected to in situ hybridization with the VGLUT2, Lmx1b and Stmn2 probes, respectively. Only those cells containing nuclei and in the dorsal lateral areas were counted (boxed area in Fig. 4). Stmn2 is a panneuronal marker25 and was used to estimate total neuronal cell numbers that were subsequently used for the calculation of the percentages of cells expressing VGLUT2 or Lmx1b. The dorsolateral area was selected because the areas closer to the ventricular zone may contain newly formed Lmx1b+ cells that have not yet turned on VGLUT2 expression. In ovo electroporation. The Lbx1 cDNA fragment, encoding a full length of human Lbx1 protein17, was cloned into the RCASBP chick viral expression vector37, and the resulting construct is referred to as RCAS-Lbx1. The purified plasmid DNA was resuspended at a concentration of 4 mg/ml and injected into neural tubes of stage 13–14 chick embryos. After electroporation, the embryos were allowed to grow at 37.5 1C for a further 48 h. Embryos were fixed, and frozen sections were then used for in situ hybridization. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank J. Johnson, D. Rowitch, R. Puettmann-Holgado and F. Yang for critical comments on the manuscript. We dedicate this work to the memory of S. Korsmeyer, in whose laboratory the Tlx gene mutant mice were made. Q.M. is a Claudia Adams Barr Scholar and a Pew Scholar in Biomedical Sciences. This work is supported by grants from the US National Institutes of Health to Q.M. and M.G. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

METHODS Animals. The generation of Tlx3 mutant mice and Lbx1 mutants has been described previously33,34. The morning that vaginal plugs were observed was considered as E0.5. Genotyping was performed as described previously24,33. All animal procedures are contained in protocols reviewed and approved by the Animal Care Committees at the Dana-Farber Cancer Institute (DFCI), Harvard Medical School.

Cell counting. For counting of Gad1+ and VGLUT2+ neurons on sections through E12.5 wild-type and Lbx1–/– spinal cord (Figs. 1,2), three independent sets of transverse sections (15 mm thickness) through the caudal spinal cord

1. Gardner, E.J., Martin, J.H. & Jessell, T. The bodily senses. in Principles of Neural Science (eds. Kandel, E.R., Schwartz, J.H. & Jessell, T.M.) 430–450 (McGraw-Hill, 2000). 2. Craig, A.D. Pain mechanisms: labeled lines versus convergence in central processing. Annu. Rev. Neurosci. 26, 1–30 (2003). 3. Polgar, E., Fowler, J.H., McGill, M.M. & Todd, A.J. The types of neuron which contain protein kinase C gamma in rat spinal cord. Brain Res. 833, 71–80 (1999). 4. Azkue, J.J. et al. Glutamate-like immunoreactivity in ascending spinofugal afferents to the rat periaqueductal grey. Brain Res. 790, 74–81 (1998). 5. Lu, Y. & Perl, E.R. A specific inhibitory pathway between substantia gelatinosa neurons receiving direct C-fiber input. J. Neurosci. 23, 8752–8758 (2003). 6. Lu, Y. & Perl, E.R. Modular organization of excitatory circuits between neurons of the spinal superficial dorsal horn (laminae I and II). J. Neurosci. 25, 3900–3907 (2005). 7. Melzack, R. & Wall, P.D. Pain mechanisms: a new theory. Science 150, 971–979 (1965). 8. Malcangio, M. & Bowery, N.G. GABA and its receptors in the spinal cord. Trends Pharmacol. Sci. 17, 457–462 (1996). 9. Kerchner, G.A., Wang, G.D., Qiu, C.S., Huettner, J.E. & Zhuo, M. Direct presynaptic regulation of GABA/glycine release by kainate receptors in the dorsal horn: an ionotropic mechanism. Neuron 32, 477–488 (2001). 10. Dickenson, A.H. Gate control theory of pain stands the test of time. Br. J. Anaesth. 88, 755–757 (2002). 11. Scholz, J. & Woolf, C.J. Can we conquer pain? Nat. Neurosci. 5 (suppl.), 1062–1067 (2002). 12. Fitzgerald, M. The development of nociceptive circuits. Nat. Rev. Neurosci. 6, 507–520 (2005). 13. Cheng, L. et al. Tlx3 and Tlx1 are post-mitotic selector genes determining glutamatergic over GABAergic cell fates. Nat. Neurosci. 7, 510–517 (2004). 14. Schuurmans, C. et al. Sequential phases of cortical specification involve Neurogenindependent and -independent pathways. EMBO J. 23, 2892–2902 (2004). 15. Panganiban, G. & Rubenstein, J.L. Developmental functions of the Distal-less/Dlx homeobox genes. Development 129, 4371–4386 (2002). 16. Hoshino, M. et al. Ptf1a, a bHLH transcriptional gene, defines GABAergic neuronal fates in cerebellum. Neuron 47, 201–213 (2005).

1514

VOLUME 8

In situ hybridization and immunostaining. Section in situ hybridization was performed as described previously35. Detailed protocols are available upon request. The following mouse in situ probes have described previously13, including mouse Gad1/GAD67, Gad2/GAD65, VGLUT2, Pax2 and chick VGLUT2 and Gad1. Other probes include Tlx3 (ref. 24), Lmx1b (ref. 36) and Stmn2 (also known as SCG10)25. The Lbx1 probe (0.7 kb) was amplified by PCR from the cDNA template prepared from newborn mouse brains. Double color in situ hybridization (Supplementary Fig. 1) was performed as described previously24. For double staining of a cytoplasmic mRNA and a nuclear protein (Fig. 2c, Supplementary Fig. 1), in situ hybridization was performed first, followed by immunostaining with anti-Lhx1/5 (Developmental Studies Hybridoma Bank) or anti-Pax2 (Zymed Laborotories). The in situ signals were photographed under trans-luminescent light and converted into pseudo–red fluorescent color, and Lhx1/5 protein was detected with Alexa 488–conjugated secondary antibodies (Molecular Probes). Standard double immunostaining was also performed with mouse anti-Lhx1/5 antibody (Developmental Studies Hybridoma Bank) plus rabbit anti-Tlx3 (a gift of C. Birchmeier, Max-Delbruck-Center for Molecular Medicine, Germany) or anti-Pax2 (Zymed Laborotories).

[

NUMBER 11

[



ARTICLES 17. Gross, M.K., Dottori, M. & Goulding, M. Lbx1 specifies somatosensory association interneurons in the dorsal spinal cord. Neuron 34, 535–549 (2002). 18. Muller, T. et al. The homeodomain factor Lbx1 distinguishes two major programs of neuronal differentiation in the dorsal spinal cord. Neuron 34, 551–562 (2002). 19. Kruger, M., Schafer, K. & Braun, T. The homeobox containing gene Lbx1 is required for correct dorsal-ventral patterning of the neural tube. J. Neurochem. 82, 774–782 (2002). 20. Muller, T. et al. The bHLH factor Olig3 coordinates the specification of dorsal neurons in the spinal cord. Genes Dev. 19, 733–743 (2005). 21. Erlander, M.G., Tillakaratne, N.J., Feldblum, S., Patel, N. & Tobin, A.J. Two genes encode distinct glutamate decarboxylases. Neuron 7, 91–100 (1991). 22. McIntire, S.L., Reimer, R.J., Schuske, K., Edwards, R.H. & Jorgensen, E.M. Identification and characterization of the vesicular GABA transporter. Nature 389, 870–876 (1997). 23. Fremeau, R.T. Jr. et al. The expression of vesicular glutamate transporters defines two classes of excitatory synapse. Neuron 31, 247–260 (2001). 24. Qian, Y., Shirasawa, S., Chen, C.L., Cheng, L. & Ma, Q. Proper development of relay somatic sensory neurons and D2/D4 interneurons requires homeobox genes Rnx/Tlx-3 and Tlx-1. Genes Dev. 16, 1220–1233 (2002). 25. Stein, R., Mori, N., Matthews, K., Lo, L.-C. & Anderson, D.J. The NGF-inducible SCG10 mRNA encodes a novel membrane-bound protein present in growth cones and abundant in developing neurons. Neuron 1, 463–476 (1988). 26. Logan, C., Wingate, R.J., McKay, I.J. & Lumsden, A. Tlx-1 and Tlx-3 homeobox gene expression in cranial sensory ganglia and hindbrain of the chick embryo: markers of patterned connectivity. J. Neurosci. 18, 5389–5402 (1998).


[

NUMBER 11

[

NOVEMBER 2005

27. Caspary, T. & Anderson, K.V. Patterning cell types in the dorsal spinal cord: what the mouse mutants say. Nat. Rev. Neurosci. 4, 289–297 (2003). 28. Goulding, M., Lanuza, G., Sapir, T. & Narayan, S. The formation of sensorimotor circuits. Curr. Opin. Neurobiol. 12, 508–515 (2002). 29. Helms, A.W. & Johnson, J.E. Specification of dorsal spinal cord interneurons. Curr. Opin. Neurobiol. 13, 42–49 (2003). 30. Borodinsky, L.N. et al. Activity-dependent homeostatic specification of transmitter expression in embryonic neurons. Nature 429, 523–530 (2004). 31. Schaffner, A.E., Behar, T., Nadi, S., Smallwood, V. & Barker, J.L. Quantitative analysis of transient GABA expression in embryonic and early postnatal rat spinal cord neurons. Brain Res. Dev. Brain Res. 72, 265–276 (1993). 32. Somogyi, R., Wen, X., Ma, W. & Barker, J.L. Developmental kinetics of GAD family mRNAs parallel neurogenesis in the rat spinal cord. J. Neurosci. 15, 2575–2591 (1995). 33. Gross, M.K. et al. Lbx1 is required for muscle precursor migration along a lateral pathway into the limb. Development 127, 413–424 (2000). 34. Shirasawa, S. et al. Rnx deficiency results in congenital central hypoventilation. Nat. Genet. 24, 287–290 (2000). 35. Birren, S.J., Lo, L.C. & Anderson, D.J. Sympathetic neurons undergo a developmental switch in trophic dependence. Development 119, 597–610 (1993). 36. Chen, Z.F. et al. The paired homeodomain protein DRG11 is required for the projection of cutaneous sensory afferent fibers to the dorsal spinal cord. Neuron 31, 59–73 (2001). 37. Morgan, B.A. & Fekete, D.M. Manipulating gene expression with replication-competent retroviruses. in Methods in Avian Embryology Vol. 51 (ed. Bronner-Fraser, M.E.) 185–218 (Academic, San Diego, 1996).

1515

E R R ATA

Erratum: Lbx1 and Tlx3 are opposing switches in determining GABAergic versus glutamatergic transmitter phenotypes Leping Cheng, Omar Abdel Samad, Yi Xu, Rumiko Mizuguchi, Ping Luo, Senji Shirasawa, Martyn Goulding & Qiufu Ma Nat. Neurosci. 8, 1510–1515 (2005)


This article contained a misspelling. Lhx1/2 should have read Lhx1/5 throughout the text.

Erratum: Why pictures look right when viewed from the wrong place Dhanraj Vishwanath, Ahna R Girshick & Martin S Banks Nat. Neurosci. 8, 1401–1410 (2005) On page 1402, the first two sentences of the second full paragraph in the second column were omitted. The paragraph should have begun as follows: “An alternative explanation, the local-slant hypothesis, suggests that location of the CoP is not recovered. Instead, the observed invariance is due to an adjustment of the retinal-image shape based on measurements of the local slant of the picture surface at the point of interest. This hypothesis does not require estimates of the location of or distance to the CoP.”

NATURE NEUROSCIENCE


ARTICLES

The transmembrane semaphorin Sema6A controls cerebellar granule cell migration Ge´raldine Kerjan1, Jackie Dolan2, Cećile Haumaitre3, Sylvie Schneider-Maunoury3, Hajime Fujisawa4, Kevin J Mitchell2 & Alain Che´dotal1 The transmembrane semaphorin protein Sema6A is broadly expressed in the developing nervous system. Sema6A repels several classes of developing axons in vitro and contributes to thalamocortical axon guidance in vivo. Here we show that during cerebellum development, Sema6A is selectively expressed by postmitotic granule cells during their tangential migration in the deep external granule cell layer, but not during their radial migration. In Sema6A-deficient mice, many granule cells remain ectopic in the molecular layer where they differentiate and are contacted by mossy fibers. The analysis of ectopic granule cell morphology in Sema6a2/2 mice, and of granule cell migration and neurite outgrowth in cerebellar explants, suggests that Sema6A controls the initiation of granule cell radial migration, probably through a modulation of nuclear and/or soma translocation. Finally, the analysis of mouse chimeras suggests that this function of Sema6A is primarily non-cell-autonomous.

In the vertebrate central nervous system (CNS), cerebellar granule cell precursors generated in the embryonic rhombic lips migrate tangentially over the cerebellar plate to form the external granular layer (EGL)1. In rodents, this secondary germinative zone produces granule cells only after birth2. In the deeper part of the EGL, postmitotic granule cells migrate tangentially in all directions2,3 and undergo a series of profound morphological changes4. They first extend a process horizontally, which is soon followed by a second one extending in the opposite direction, before a third process develops perpendicular to the surface. This precedes the inward radial migration of granule cell bodies, along Bergmann glia fibers5 to the internal granular layer (IGL, Fig. 1a). The horizontal processes become the granule cell axons known as parallel fibers. The radial migration of granule cells has been extensively studied (see ref. 5 for a review), but much less is known about the molecules that control their tangential migration or the switch from tangential to radial migration5. Semaphorins are secreted or membrane-bound proteins that control axon guidance and cell migration6. Many, if not all, transmembrane semaphorins are expressed in the developing CNS, but little is known of their functions in vivo. Class 6 semaphorins comprise four proteins, Sema6A–Sema6D, that are closely related to invertebrate transmembrane semaphorins6. Class 6 semaphorins can collapse various classes of axons7–9; this process seems to be mediated by receptors of the plexin-A family9,10. Class 6 semaphorins and the related insect class 1 semaphorins6,11 are also capable of bidirectional signaling, apparently acting as receptors in certain contexts10,12. Recently, a mouse line deficient in Sema6A was obtained using a gene-trapping strategy13,14. Sema6A-deficient mice have abnormalities in several axonal tracts,

including thalamocortical projections13. Here we show that in the developing cerebellum, Sema6A may control the initiation of granule cell radial migration, probably by influencing nucleus or soma translocation in a non-cell-autonomous manner. RESULTS Sema6A is transiently expressed in the deeper EGL Sema6A messenger RNAs (mRNAs) were detected in the EGL of the E15–E17 embryonic mouse cerebellum as previously described (data not shown and ref. 15). In the postnatal cerebellar cortex, Sema6A transcripts were found in the deeper EGL and in the IGL (Fig. 1b). To localize Sema6A protein, we used monoclonal and polyclonal antibodies directed against ectodomains of the human or mouse Sema6A (amino acids 19–649). On western blots of cell extracts collected from COS7 cells transfected with plasmids encoding full-length Sema6A (see Methods), these antibodies recognized a band at the expected molecular weight (120 kDa) but did not recognize other class 6 semaphorins (data not shown). To confirm the specificity of Sema6A antibodies, we labeled sections from postnatal day (P) 10 Sema6a/ mice13. These mutant mice carry a gene-trap insertion in the Sema6A gene that results in the fusion of a portion of the extracellular region of Sema6A to the reporter gene TM--geo (see Methods). Such fusion proteins are sequestered intracellularly, probably in the endoplasmic reticulum, reliably producing phenotypic null alleles14. In the cerebellum of Sema6a/mice, Sema6A immunostaining differs from that in wild-type mice. Subcellular Sema6A-positive dots were observed in the cerebellar cortex (Fig. 1c), most likely corresponding to vesicles of the endoplasmic

1Centre National de la Recherche Scientifique UMR7102, Universite ´ de Paris 6, Case 12, 9 Quai Saint-Bernard, 75005 Paris, France. 2Smurfit Institute of Genetics, Trinity College Dublin, Dublin 2, Ireland. 3Centre National de la Recherche Scientifique UMR7622, Universite´ de Paris 6, Batiment C, Case 24, 9 Quai Saint-Bernard, 75005 Paris, France. 4The 21st Century Center of Excellence Program, Division of Biological Science, Nagoya University Graduate School of Science, Chikusa-ku, Nagoya 464–8602, Japan. Correspondence should be addressed to A.C. ([email protected]).

Received 11 July; accepted 31 August; published online 2 October 2005; doi:10.1038/nn1555

1516

VOLUME 8

[

NUMBER 11

[


ARTICLES

a

Sema6a

b

c

Sema6A+Hoechst

EGL

1

2

3 P10

IGL

P10

4

d

f

Sema6A+CaBP+Hoechst

g

β-gal

Sema6A+CaBP

Sema6A

P10

P0

e

h

Sema6A+CaBP+Hoechst

i

Sema6A+BrdU

Sema6A+Hoechst

j

P10

ML

P21

EGL

P10


5

IGL

PGL

reticulum. Confocal microscopy showed that Sema6A-positive dots coexpressed b-galactosidase (Fig. 1d,e) as expected for a fusion protein. From birth to P10, Sema6A immunoreactivity was observed in the deeper EGL as shown on sections double stained with the Purkinje cell marker calbindin D28k16 (Fig. 1f–h and data not shown). At P10, BrdU-positive proliferating granule cell precursors in the upper EGL were Sema6A negative (Fig. 1i). Also, radially migrating granule cells in the molecular layer and postmigratory granule cells in the IGL were not immunoreactive to Sema6A, although they expressed Sema6A mRNA, indicating a precise translational regulation of Sema6A expression. At P21, when granule cell proliferation and migration had stopped17, Sema6A immunoreactivity was no longer detected in the molecular layer or IGL (Fig. 1j). Arrested granule cell migration in Sema6A-deficient mice To determine the function of Sema6A in granule cell development, we analyzed mice genetically engineered to be deficient in Sema6A13. Sema6a/ mice are viable and fertile and do not show any substantial behavioral defects (data not shown). The gross anatomy of the cerebellum from Sema6a/ mice was indistinguishable from that of the wild type (data not shown). In sections from wild-type mice and Sema6a+/ stained with cresyl violet, the adult cerebellar cortex had its typical regular organization (Fig. 2a,b). In Sema6A-deficient mice (Fig. 2c,d), Purkinje cells and molecular layer interneurons showed normal distribution and density, but the size of the IGL was reduced by 10%. Moreover, in all folia, numerous cells with a small, dense nucleus were found in the molecular layer, sometimes clustered at the cerebellar surface (Fig. 2d). To


ML

[

NUMBER 11

[

NOVEMBER 2005

Figure 1 Sema6A is expressed in tangentially migrating granule cells. (a) Phases of granule cell development. (1) Granule cell precursors proliferate in the outer portion of the external granule cell layer (EGL). (2) Postmitotic granule cells in the deeper EGL extend two horizontal processes and migrate tangentially. They next send a third process perpendicular (3) and migrate radially through the molecular layer and Purkinje cell layer (grey circles) along radial glia (4) to settle in the internal granule cell layer or IGL (5). (b) At P10, Sema6a mRNAs are expressed in the deeper EGL and in the IGL. (c) Cerebellum section from P10 Sema6Aa/ mouse. In contrast with the wild-type (see h), Sema6A immuno-positive dots are observed throughout the EGL, molecular layer and IGL. (d,e) 1-mm confocal image of the cerebellar cortex of a P10 Sema6a/ mouse, double labeled with antibodies to Sema6A (in e) and anti-bgalactosidase (in d) antibodies, showing that Sema6A and b-gal are co-expressed, most likely in small intracytoplasmic vesicles. (f) At P0, Sema6A expression is restricted to the deeper EGL (arrowheads) above CaBP-positive Purkinje cells (arrows). (g,h) P10 cerebellum section immunostained with antibody to Sema6A and CaBP with Hoechst counterstaining. Sema6A is still restricted to the deeper EGL. (i) Sema6A immunoreactive granule cells (asterisks) are immediately adjacent to the upper EGL containing proliferating BrdU-labeled granule cell precursors (arrowheads). (j) At P21, Sema6A is not expressed in the cerebellar cortex. Scale bars, 50 mm (b,c,f,j), 500 mm (g), 35 mm (d,e,h,i).

determine whether they corresponded to granule cells, sections from adult cerebellum were labeled with antibody to the granule cell–specific a6 GABAA receptor subunit (a6)18. In all cases, Sema6a+/ mice were identical to wild-type mice. In the wild type, a6 is expressed only by granule cells that have reached the IGL (Fig. 2e,g and ref. 18). In contrast, in Sema6a/ mice, a large proportion of a6-positive neurons was found either within the molecular layer or aggregated in clusters under the pial surface (Fig. 2f,h). These ectopic cells also expressed Ca2+/calmodulin-dependent protein kinase IV (CaMKIV), another marker of mature postmigratory granule cells19 (data not shown; discussed below). Quantification of CaMKIV-positive granule cells indicated that about 40% remained in the molecular layer. The same results were obtained using Sema6A knockout mice 1 year of age (data not shown), indicating that the migration is not simply delayed and that the ectopic granule cells survive. The differentiation of Purkinje cells—the postsynaptic partner of granule cells—as visualized with antibodies to CaBP or zebrin II (ref. 20) or with BDA labeling (see Methods), did not seem to be affected (Fig. 2g,h and data not shown). The number of molecular-layer interneurons, as determined by Hoechst staining and parvalbumin immunostaining, was also comparable to that in the wild type (Fig. 2g,h and data not shown). To further study the granule cell migration defect in Sema6Adeficient mice, we performed pulse labeling of migrating granule cells with BrdU (see Methods). At 108 h after a single BrdU injection, about 90% of BrdU labeled cells were found in the IGL in wild-type mice (Supplementary Fig. 1 online), whereas 40% of the BrdU-labeled cells were still in the molecular layer in Sema6Aa/ mice (Supplementary

1517

Fig. 1). Overall, these results suggest that granule cell migration within the EGL or from the EGL is perturbed in Sema6a/ mice. Granule cell proliferation, differentiation and survival Although Sema6A is already expressed during the first phase of tangential migration of granule cell precursors, in situ hybridization with the granule cell precursor markers Math1 (also known as Atoh1 and HATH1) and Barhl1 (ref. 21) showed that the EGL was indistinguishable in newborn Sema6a+/ and Sema6a/ mice (Supplementary Fig. 2 online). This suggests that the early migration of granule cell precursors is normal. We quantified granule cell proliferation in P0, P10 and P21 cerebella using BrdU (see Methods) and antibodies to phospho-histone-H3. In Sema6A-deficient mice, the number of proliferating granule cells (± s.e.m.; errors are stated as s.e.m. throughout) in the upper EGL

a

b

Sema6a +/–

c

Sema6a –/–

d

Sema6a –/–

e

f

P10 Dcx+α6

h

IGL

i

Sema6a –/–

k

l PCL

P10

GFAP

Sema6a +/–

1518

j

P16

g

P10

TAG-1

Sema6a +/–

ML

P0

Math1

Sema6a –/–

L1

P8 Caspase-3 act.

Sema6a +/–

Cresyl violet

WT

a

Sema6a –/–

WT

b

ML

c

Sema6a –/–

d ML

PCL

PCL

Sema6a –/–

WT

e

f

ML IGL

ML

α6

IGL

g

h

CaBP+α6+Hoechst

Figure 2 Abnormal granule cell migration in Sema6A-deficient mice. (a–d) Cresyl violet staining of cerebellum sections from adult wild-type (a,b) or Sema6a/ (c,d) mice. In the wild-type (wt) cerebellum, the border between the molecular and Purkinje cell layers is sharp (arrow in a). In contrast, in Sema6a/ mice this border is irregular and fuzzy (arrow in c; see also d). Moreover, cell density is higher in the molecular layer (ML) of Sema6a/. The density of molecular layer interneurons (arrowhead in b and d) is similar, but there are many cells with small and dense nuclei (arrows in d) dispersed in the molecular layer of Sema6a/ mice or accumulated near the pial surface. (e,f) Adult cerebellum sections immunostained for a6. In Sema6a+/, all a6-immunoreactive granule cells are found in the IGL and none in the molecular layer (ML), whereas in Sema6a/, many ectopic a6-positive granule cells are found in the ML and near the pial surface (arrowheads in f). (g,h) Adult cerebellum section double stained for CaBP and a6 with Hoechst counterstaining. In Sema6a+/, the molecular and Purkinje cell layers do not contain a6-immunoreactive cells. The nuclei of stellate and basket cells can be observed with Hoechst in the molecular layer (arrowheads in g). In Sema6a/ mice, numerous a6-positive granule cells are found in the molecular layer (arrows in h). Interneurons are also observed (arrowheads in h). Scale bars, 450 mm (a,c), 45 mm (b,d), 135 mm (e,f), 35 mm (g,h).

P10


ARTICLES

(0.025 ± 0.005 cells per mm of EGL length at P0 and 0.069 ± 0.003 cells per mm at P10) was similar to that in heterozygous mice (0.024 ± 0.001 cells per mm of EGL length at P0 and 0.058 ± 0.005 cells per mm at P10; data not shown). Moreover, by P21, no proliferating cells were detected in the molecular layer in Sema6A-deficient mice, demonstrating that the timing and level of granule cell proliferation were also normal. In the mouse cerebellum, the peak of apoptosis occurs around P8–P10 (ref. 22). To measure the level of apoptosis in the developing cerebellum, we used antibody to activated caspase-3 (ref. 23). No significant (P 40.05) difference was observed in P8 Sema6A-deficient mice (Fig. 3a,b). During their differentiation, granule cells are known to sequentially express different genes, such as the transcription factors Math1 and Pax6 (ref. 21). The expression patterns of these two genes were normal in Sema6A-deficient mice (Fig. 3c,d and data not shown). Likewise, the expression patterns of TAG-1 and L1, two molecules involved in granule cell radial migration24,25, were normal (Fig. 3e–h).

Figure 3 Normal apoptosis, differentiation and radial glia organization in Sema6A knockouts. (a–l) Sagittal cerebellum sections from Sema6a+/ (a,c,e,g,i,k) or Sema6a/ (b,d,f,h,j,l) mice were immunostained with antibodies against activated caspase-3 (a,b), L1 (e,f), doublecortin (Dcx; i,j), TAG-1 (g,h) or GFAP (k,l) or hybridized with Math1 riboprobes (c,d) and counterstained with Hoechst (a–j). At P8, the number of caspase-3–labeled cells in the EGL (arrowheads in a and b) is similar in Sema6a+/ (3.9 ± 0.1 cells per mm of EGL length) and Sema6a/ (3.8 ± 0.5 cells/ per mm of EGL length). In c and d, Math1 signal has been artificially colored in red using Photoshop and superimposed on the Hoechst labeling (in blue). There is no difference in math1 expression between Sema6a/ and Sema6a+/. At P10, L1 ( e,f) and TAG-1 (g,h) expression patterns are identical in Sema6a+/ and Sema6a/. In i and j, sections from P16 cerebellum are double stained for Dcx and a6. In both Sema6a+/ and Sema6a/, Dcx is expressed in the deep EGL (arrow in i and j) and in perpendicular processes of radially migrating granule cells in the molecular layer (arrowheads in i and j). In Sema6a/, many a6 positive cells are found in the molecular layer. At P10, the radial organization of GFAP-positive Bergmann glia cells (arrows in k and l) is similar in Sema6a+/ and Sema6a/. Scale bars, 115 mm (a,b), 55 mm (c,d), 30 mm (e–l).

VOLUME 8

[

NUMBER 11

[


ARTICLES Sema6a –/–

WT

c

b

d

BDA+CaBP

ML

ML

IGL

IGL α6

VGLUT2

e Sema6a –/–


BDA

a

VGLUT2+α6

f

g

ML

PCL IGL

Doublecortin expression in radially migrating granule cells was also identical26 (Fig. 3i,j). Finally, ectopic granule cells still expressed markers of mature granule cells such as a6 and CaMKIV (Fig. 3i,j; and data not shown). Granule cell migration defects in Sema6a/ mice could be due to abnormal Bergmann glia. This seemed unlikely, however, as Sema6A is expressed in tangentially migrating granule cells in the EGL, but not in radially migrating granule cells or in Bergmann glia fibers. Immunostaining of the P10 cerebellum by glial fibrillary acidic protein (GFAP) confirmed that the number, disposition and morphology of Bergmann glia fibers in Sema6a/ mice were similar to those in Sema6a+/ mice (Fig. 3k,l). Thus the defects in migration seemed to be intrinsic to granule cells. In wild-type mice, granule cells in the IGL establish synaptic contact with mossy fibers27. The large mossy fiber terminals are called ‘rosettes’. To label mossy fibers, we injected BDA into the adult cerebellum of wild-type and Sema6A-deficient mice (Fig. 4a–d). In the wild type, BDA-labeled mossy fibers were confined to the IGL (Fig. 4a,b). In contrast, in Sema6A-deficient mice, BDA-labeled mossy fibers were also observed in the molecular layer where they formed typical rosettes (Fig. 4c,d). In addition, rosettes can be visualized using immunocytochemistry for the vesicular glutamate transporter VGLUT2 (ref. 28; Fig. 4e). VGLUT2 staining also labels climbing fibers (Fig. 4e), but the size and morphology of the two types of terminals are clearly distinguishable. VGLUT2-positive rosettes were found in the molecular layer (Fig. 4e) close to ectopic a6-immunopositive granule cells (Fig. 4e–g). Overall, these results strongly suggest that granule cell differentiation and maturation are normal in Sema6A-deficient mice. Normal parallel fiber extension and granule cell polarity To further understand the role of Sema6A in granule cell migration, we labeled individual granule cells and their axons (the parallel fibers) with BDA. All parallel fibers extend along the mediolateral axis of the


[

NUMBER 11

[

NOVEMBER 2005

Figure 4 Ectopic granule cells are contacted by mossy fibers in Sema6a/. (a–d) Wild-type (a,b) and Sema6a/ (c,d) adult mice were injected into the cerebellum with BDA. Some sections (b,d) were also immunostained with antibody to CABP to label Purkinje cells. The dashed line in a and c delineates the pial surface. In wild-type (a,b), all large mossy fiber terminals (or rosettes) are located in the internal granular layer (IGL; arrows in a and arrowheads in b). No rosette is observed in the molecular layer (ML). By contrast, in Sema6a/ (see c,d), many rosettes are detected in the ML (arrows in c and arrowheads in d), in addition to the IGL. (e–g) 1-mm confocal images of VGLUT2 and a6 double immunostaining on Sema6a/ cerebellum. VGLUT2 labels large mossy fiber rosette (arrowheads in e and g) and punctate climbing fiber terminals (arrow in e). Mossy fiber rosettes contact ectopic a6-positive granule cells (arrowheads in f and g). Scale bars, 100 mm (a,c), 35 mm (b,d), 45 mm (e–g).

cerebellar cortex29. In adult wild-type mice, BDA injection in the molecular layer labeled parallel fibers running across the entire length of the injected folia and granule cell bodies in the IGL (Fig. 5a). In Sema6A-deficient mice, BDA also labeled parallel fibers throughout the injected lobule, but BDA-labeled granule cells were observed in the molecular layer in addition to the IGL (Fig. 5b). At higher magnification, the morphology of ectopic granule cell was clearly visible (Fig. 5c–h). Two main categories of cells were observed. The first type was found throughout the molecular layer (Fig. 5c,e,h). These cells extended a process resembling mature granule cell dendrites toward the IGL, and additional dendritic processes emerged from the cell body. These processes were immunoreactive for a6, which is expressed only on granule cell bodies and dendrites18 (Fig. 5f–h). These ectopic granule cells had a regular T shape3, extending a thinner process toward the pial surface (not immunoreactive for a6) from which the parallel fiber emerged. The length of this upward process was comparable for all ectopic granule cells. The second type of cell was found near the pial surface within the a6-immunoreactive clusters (see Fig. 2f). These cells had a rather bipolar shape (Fig. 5d,e) with parallel fibers emerging directly from the cell body, but they also had dendritic processes, one of which was oriented toward the IGL. Abnormal in vitro migration of Sema6a/ granule cells To determine whether Sema6A controls granule cell neurite outgrowth, we seeded dissociated granule cells from the cerebellum of P5 wild-type mice onto Sema6A-expressing COS7 cells and onto mock-transfected cells. Cultures were labeled after 24 h, using phalloidin and b-tubulin immunostaining. In Sema6A-expressing cells, the total neuritic length (length of the longest neurite) and the number of branch points were not significantly different (P 4 0.5) from those in the controls (Supplementary Fig. 3 online). Thus, exogenous Sema6A does not influence the elongation of granule cell neurites. We next compared the growth of cerebellar granule cell neurites in wild-type and Sema6a/ mice. We found that neurite outgrowth from Sema6a/ granule cells was normal. The total neuritic length and the length of the longest neurite (137.3 ± 6.9 mm and 100.3 ± 4.8 mm, respectively; n ¼ 127) were not significantly different (P 4 0.5) from those in the wild type (126.3 ± 6.1 mm and 97.4 ± 4.9 mm, respectively; n ¼ 179). The number of branch points was also similar (1.75 ± 0.04 mm for Sema6a/ neurons compared to 1.70 ± 0.03 mm for wild type, P 4 0.5; data not shown). To assess cell migration more directly, we cultured EGL explants on polylysine and laminin for between 18 h and 4.5 d (ref. 30). In these cultures, granule cells migrate away from the explant in the absence of any radial glia and follow the exact sequence of in vivo differentiation. The number of migrating granule cells, quantified using Hoechst staining and phalloidin (Fig. 6), in cultures derived from EGL explants

1519

ARTICLES WT

b

b

c

d

e

f

d

e

f

g

h

Distance from the explant (µm)

g

BDA

α6

α6

36 h

BDA

c

β-tubulin

108 h

BDA+Hoechst

a

IGL

Sema6a –/–


Hoechst

BDA+Hoechst

ML

Sema6a –/–

Phalloidin

Sema6a +/–

a

*

0–50

*

50–100 100–150 150–200 200–250 250–300 0

Figure 5 In vivo labeling of ectopic granule cells and parallel fibers in Sema6a/. (a–e,g,h) Cerebellum sections of wild-type and Sema6a/ adult mice injected with BDA; in a and b, with Hoechst counterstaining. (a,b) In wild-type (wt; see a) and Sema6a/ (see b), BDA labels parallel fibers (arrows in a and b) in the molecular layer (ML) extending across the mediolateral axis of the injected folia. However, whereas retrogradely labeled granule cell bodies in wt are exclusively found in the IGL, in Sema6a/, many are observed in the ML (arrowheads). Panels c–e,g and h are confocal images of ectopic granule cells in the upper part of the molecular layer of Sema6a/ mice. Many ectopic granule cells have a typical morphology with a T-shape axon (arrow in c and h) and hook-ended dendrites (arrowheads in c–e and h). Other ectopic granule cells close to the pial surface have a more bipolar morphology (arrows in d and e). These dendrites are also a6immunoreactive (arrows in g and h), whereas the axon is not (arrowhead in h). (f) In adult Sema6a/ cerebellum, most ectopic granule cells extend in the ML a6-positive dendrites (short arrows) oriented toward the IGL. Scale bars, 75 mm (a,b), 10 mm (c,e), 20 mm (d,g,h), 35 mm (f).

10% 20% 30%

Migration rate

Figure 6 Abnormal in vitro migration of Sema6A-deficient granule cells. (a–f) Migration and neurite outgrowth from P4 EGL microexplants after 36 h (a,b,e,f) or 108 h (c,d). Hoechst staining revealed that after 36 h, there is a significantly lower number of migrating granule cells in Sema6a/ (b) compared to Sema6a+/ explants (a). After 108 h in vitro, phalloidin staining shows that migrating granule cells form large aggregates with Sema6a+/ explants (arrows in c), whereas the aggregates are much smaller with Sema6a/ explants (arrows in d). The number of aggregates is also lower. After 36 h in vitro, Tuj-1 immunostaining reveals that the number of neurites cutting a line at 150 mm from the explant) was similar in Sema6a+/ (see e) (367.4 ± 25.3 neurites/mm; n ¼ 24) and Sema6a/ (see f) explants (379.6 ± 26.3 neurites/mm; n ¼ 10). (g) Quantification of the distribution of migrating granule cells around EGL explants in Sema6a/ and Sema6a+/ (see Methods). The asterisk indicates a significant difference between the two (one-tailed t-test; *P o 0.05). Scale bars, 225 mm (a–d), 55 mm (e,f).

from Sema6A-deficient mice was considerably reduced after 36 h in vitro (Fig. 6a,b,g). After 36 h, the maximum migration distance (the external limit of the region that comprised 90% of the granule cells) was also significantly lower (P ¼ 0.02) for the Sema6a/ granule cells (127.9 ± 12.4 mm; n ¼ 5) than for the Sema6a+/ granule cells (189 ± 17.1 mm; n ¼ 6) . After 4.5 d in culture, most granule cells from wild-type mice aggregated, forming large clusters (Fig. 6c). With explants from Sema6a/ cerebellum, aggregates formed but their number and size were smaller (Fig. 6d). The timing of TAG-1 and MAP2 expression was similar in wild-type and Sema6a/ EGL cultures, showing that the absence of Sema6A does not affect granule cell maturation (data not shown). Notably, the number of processes growing from EGL explants (Fig. 6e,f) was similar in Sema6a+/ and Sema6a/. Finally, wild-type EGL explants were cultured on Sema6A substrate (n ¼ 23) or in a medium containing Sema6A-Fc (n ¼ 27; see Methods) for between 36 and 96 h. In both conditions, the number of

migrating granule cells and granule cell clusters were unchanged (Supplementary Fig. 3). These results suggested that, in the absence of Sema6A, granule cells could extend normal processes but that the movements of the nucleus, soma, or both were perturbed. To confirm this, we performed timelapse videomicroscopy on EGL explants after 24–48 h in cultures (Fig. 7a,b). In control explants (n ¼ 3 explants; two independent experiments), migrating cells rapidly extended a long leading process bearing typical growth cones away from the explant. A large enlargement at the rear contained the nucleus (Supplementary Video 1 online and Fig. 7a). Time-lapse videomicroscopy showed that nuclear movement was saltatory and was preceded by the appearance of an elongated swelling ahead of the nucleus, such as has recently been described for cortical interneurons31 (Supplementary Video 2 online and Fig. 7a). Although the nucleus was often stationary, the overall movement of the cell body was always in the opposite direction from the explant core. In

1520

VOLUME 8

[

NUMBER 11

[


ARTICLES 1:40

2:30

3:20

4:10

5:00

Sema6a –/–

b Sema6a –/–

c

d WT

Sema6a –/–

e

f

g

h

Sema6a –/–

Sema6a –/–

Sema6a –/–

WT

WT

Sema6a –/–

(Fig. 7i,j). Finally, the actin cytoskeleton labeled with phalloidin appeared similar in wild-type and Sema6A-deficient granule cells (Fig. 7k,l).

i

j

k

l

γ-tubulin

Phalloidin

explants from Sema6A-deficient mice (n ¼ 8 explants; three independent experiments), long leading processes with very dynamic growth cones were also observed, suggesting that neurite extension was normal (Fig. 7b and Supplementary Video 3 online). However, the forward movement of the nucleus was perturbed. Most cell bodies appeared to oscillate without moving substantially. Furthermore, cells migrating toward the explant were always observed (Supplementary Videos 3–5 and Fig. 7b). Finally, in the vicinity of the explants, cell bodies seemed more tightly aggregated than in the wild type. To further characterize this abnormal migration, EGL cultures at 36 h were fixed and labeled with markers of the cytoskeleton and centrosome that are important components of the migration machinery31,32. The overall disorganization of the cohort of migrating granule cells was clearly observed using a-tubulin staining (Fig. 7c,d). However, the perinuclear cage of microtubules32 appeared similar in wild-type and Sema6A-deficient cells (Fig. 7e–h). The centrosome could also be observed using antibodies to g-tubulin32; it had a comparable position in the wild-type and Sema6A-deficient cells

Figure 8 Non-cell-autonomous function of Sema6A in migrating granule cells. (a–h) Cerebellum sections from GFP; Sema6a+/:wt (in a) and GFP; Sema6a/:wt (in b–h) mouse chimeras labeled with antibody to GFP (a–d,g,h) and CaMKIV (a,b,f,h), or a6 (c,e). In GFP; Sema6a+/:wt chimera (see a), all CaMKIV- and GFP-positive granule cells (arrowheads) are in the IGL. Some Purkinje cells also express GFP (arrow). In contrast, in GFP; Sema6a/:wt chimera (see b), many CaMKIV-positive ectopic granule cells are observed in the molecular layer (arrowheads). (c–h) 4-mm confocal images of cerebellum sections from GFP; Sema6a/:wt chimera labeled with a6 (in c and e), CaMKIV (f,h) and GFP. Many a6- or CaMKIV-positive granule cells are located in ML. Only a subset of these ectopic granule cells are GFP positive (and thus Sema6a/; arrowheads), and the majority are GFP negative (and thus of wild-type origin; arrows). Scale bars, 75 mm (a,b), 35 mm (c–h).


[

NUMBER 11

[

NOVEMBER 2005

Non-cell-autonomous function of Sema6A in granule cells Sema6D and insect class I semaphorins act as ligands and receptors10,12. To determine whether Sema6A functions as a receptor or as a ligand in migrating granule cells, we generated mouse chimeras. Sema6A mutant mice were first crossed with mice expressing green fluorescent protein (GFP) attached to the chicken Actb (b-actin) promoter (Actb-GFP mice)33 to allow the cells of each genotype to be distinguished (see Methods). GFP;Sema6a/morulae were then aggregated to wild-type (‘wt’) morulae to obtain GFP;Sema6a/:wt chimeric mice (n ¼ 2). As a control, we also obtained GFP;Sema6a+/:wt chimeras (n ¼ 5). The proportion of Sema6A-deficient cells, as determined on the basis of GFP expression, was between 30% and 40% (n ¼ 2). GFP was detected in all types of cerebellar cells such as Bergmann glia and Purkinje cells (Fig. 8 and data not shown). In GFP;Sema6a+/:wt chimeras, no

GFP; Sema6a +/–:WT

GFP; Sema6a –/–:WT

a

b

CaMKIV+GFP

β-tub+Hoechst

WT

α-tubulin


a

Figure 7 Abnormal granule cell movement in absence of Sema6A. (a,b) Time-lapse imaging of granule cell migration from P5 EGL microexplants cultured for 24 h before imaging. Explants are located on the top of the frame (interval between pictures is 50 min). In wild-type (see a), granule cells migrate opposite to the explant. They transiently show an elongated swelling just ahead of the nucleus (arrow). In contrast, most Sema6a/ granule cells (see b) remains stationary (asterisk) or move back toward the explant (arrow). (c–h) a-tubulin immunostaining of wild-type (c,e) or Sema6a/ (d,f–h) EGL microexplants cultured for 36 h (with Hoechst counterstaining). Explants are on the left side of the frame for c and d and in the bottom for e–h. Many migrating Sema6A/ granule cells (arrows in d) appear misoriented compared to wild type (arrows in c). However, the perinuclear cage of microtubule labeled with antibody to a-tubulin appears similar in granule cells from wild-type (see e) and Sema6A-deficient (f–h) mice. (i,j) g-tubulin immunostaining of the centrosome in granule cells migrating from EGL explants. In granule cells derived from both wild-type (i) and Sema6a/ mice (j), the centrosome is positioned ahead of the nucleus. (k,l) Phalloidin-FITC staining of the actin cytoskeleton in granule cells migrating from EGL explants. No difference is observed between wild-type (i) and Sema6a/ (j). Scale bars, 25 mm (a,b,k,l), 30 mm (c,d), 6 mm (e–h), 15 mm (i,j).

α6+GFP +Hoechst

c

α6

GFP

d

e

CaMKIV

f

GFP

g

CAMKIV +GFP

h

GFP; Sema6a –/–:WT

0:50

WT

0:00

1521


ARTICLES

DISCUSSION It has been proposed that each phase of granule cell development is controlled by both extrinsic factors and an intrinsic developmental program34. Some factors acting on granule cells within the EGL have begun to be identified. Thus, the bHLH transcription factor Math1 is necessary for lineage determination of granule cells35. In the outer EGL, the proliferation of granule cell progenitors requires Purkinje cell– derived Sonic hedgehog36. The passage from the outer EGL to the deeper EGL involves the chemokine stromal-derived factor-1 and EphB2-ephrinB2 reverse signaling37. The initiation of granule cell axon elongation in the deeper EGL is regulated by interaction between the serine/threonine kinase Unc51.1, SynGAP and syntenin38. Finally, many extracellular cues—such as glutamate39, neuregulin40 and BDNF41—modulate the radial migration of granule cells and may cooperate with Sema6A to trigger migration. However, Sema6A function differs from these factors in that proliferation, migration to the deeper EGL and parallel fiber extension are normal in Sema6a/ mice. Our data, together with previous anatomical2,4 and time-lapse analyses3, show that Sema6A is expressed by tangentially migrating granule cells in the deep EGL and is turned off when they initiate radial migration. Sema6A function seems to be specifically required for the initiation of radial migration. We favor a model (Supplementary Fig. 4 online) in which Sema6A regulates interactions between granule cells, rather than with Bergmann glia. First, Sema6A is no longer expressed by granule cells when this interaction occurs and is never expressed by Bergmann glia, whose morphology appears normal in Sema6a/ mutants. Second, granule cell migration from EGL explants in vitro, which occurs without Bergmann glia, is abnormal in absence of Sema6A. Notably, Sema6A does not affect granule cell neurite outgrowth in vitro and in vivo, which might have been its expected role. Our observations suggest, rather, that Sema6A is specifically involved in the early steps of nuclear and soma translocation within the radially oriented granule cell processes (Supplementary Fig. 4). Thus, Sema6A function differs from that of other known molecules expressed in the deeper EGL (such as Unc51.1 and Pax6) that principally control parallel fiber extension. Drosophila Sema1a, which is closely related to class 6 semaphorins, seems capable of acting as a receptor12, and Sema6D can act as a receptor for plexin-A1 (ref. 10). However, the analysis of Sema6A chimeras reveals that Sema6A function in the EGL is primarily noncell-autonomous. This does not rule out a possible receptor function for Sema6A at certain time-points or in other neurons. The best candidate receptors for Sema6A on granule cells are type A plexins9,10, one of which—plexin-A4—mediates Sema6A repulsive activity on sympathetic axons. However, plexin-A4 mRNA was not detected in the developing cerebellar cortex (unpublished data). Sema6A does not seem to be cleaved and thus may work as a contact repellent, providing spatial information for underlying cells in the deeper EGL—in particular, for radially migrating cells; thus Sema6A,

together with gradients of secreted factors37, may serve to orient granule cell migration away from the EGL (Supplementary Fig. 4). This is consistent with the failure of many granule cells to migrate away from the EGL in Sema6A mutants and with the observations, from time-lapse videomicroscopy, that many granule cells from Sema6Amutant explants migrate in the wrong direction (towards the explant). Accordingly, the perturbation of this spatial information may also explain the presence of many ectopic wild-type granule cells in Sema6a/:wt chimeras. However, the extinction of Sema6A expression with the initiation of radial migration also suggests that the spatiotemporal control of Sema6A expression provides a carefully regulated paracrine signal that may initiate radial migration or arrest tangential migration. An alternative hypothesis is that Sema6A primarily controls the level of adhesion between tangentially migrating cells and that the defects in the initiation of radial migration are secondary to increased adhesion between granule cells in the absence of Sema6A function. Sema1A has been shown to affect defasciculation of motor axons at discrete choice points42 by countering the attractive activity of homophilic proteins. Notably, in Sema6a/ mice, many ectopic granule cells are clustered near the pial surface, and the motility of granule cell bodies is also markedly reduced in EGL cultures. Thus, Sema6A could function as a de-adhesive molecule in tangentially migrating granule cells, facilitating, through contact repulsion, granule cell movement in the EGL and the response to another signal to initiate radial migration. However, in EGL explant cultures, the Sema6A ectodomain does not seem to influence granule cell migration and clustering. Our time-lapse imaging experiments showed that the primary consequence of Sema6A deficiency is a defect in the translocation of the granule cell soma or nucleus. Oriented nuclear translocation, or nucleokinesis, has been described in most migrating neurons and in other cell types31,32,43. Migrating neurons are highly polarized, with cytoplasmic organelles such as the centrosome and the Golgi apparatus at their leading edge, and the nucleus at their rear. We could not detect any obvious defects in the centrosome or cytoskeleton in Sema6a/ granule cells, suggesting that their abnormal migration may not result from a major disorganization of the migration machinery. It will be important to determine whether the expression or function of the centrosomal components is altered, and in particular those of Par6a, which regulates cytoskeletal dynamics and nucleokinesis in radially migrating granule cells32. Many other questions remain unanswered. Sema6A mRNA is expressed in granule cells in the IGL, although these cells are not immunoreactive with Sema6A. This shows that the expression of Sema6A protein is developmentally regulated and subject to precise translational modifications, post-translational modifications or both. The defects in initiation and orientation of granule cell migration suggest that the regulation of Sema6A protein expression provides a precise spatiotemporal signal. In addition, the observation that about 60% of granule cells are still able to reach the IGL suggests either that granule cells are heterogeneous or that other molecules act together with Sema6A to control the transition from tangential to radial migration. Sema6A-deficient mice also provide a good model for studying other aspects of cerebellar development. During development, mossy fibers never invade the molecular layer and remain below the Purkinje cells44. In vitro experiments have also suggested that mossy fiber growth is regulated by target-derived stop signals produced by granule cells, and by inhibitory signals produced by Purkinje cells or present in the molecular layer16. In Sema6a/ mice, ectopic granule cells in the molecular layer differentiate and receive mossy fiber innervation. Thus,

1522

VOLUME 8

ectopic a6- or CaMKIV-positive cells were detected in the molecular layer (Fig. 8a). In contrast, many ectopic CaMKIV- and a6-positive granule cells were detected in the molecular layer of GFP;Sema6a/:wt (Fig. 8b,c). Among GFP-expressing granule cells, about 30% were ectopic. Confocal microscopy analysis revealed that only a minority— about 30–40%—of the CaMKIV- and a6-positive ectopic cells were also GFP positive and thus deficient in Sema6A (Fig. 8c–h). These cells were dispersed throughout the molecular layer. The remaining a6-positive ectopic cells were GFP negative, and thus of wild-type origin; like the GFP-positive cells, these were homogeneously distributed in the molecular layer.

[

NUMBER 11

[



ARTICLES the unusual cellular environment does not preclude an intrinsic differentiation program. In addition to, and together with, other previous examples of mossy fiber contacts observed on ectopic granule cells in wild-type mice or as a result of 6-hydroxydopamine treatments45,46, this result shows that there are no inhibitors that prevent mossy fibers from entering the molecular layer. Rather, the surprising capacity of mossy fibers to find and contact granule cells far from the IGL suggests that granule cells secrete long-range chemoattractants for mossy fibers. METHODS Animals. Swiss and C57/Bl6 mice (Janvier) were used for expression studies. The Sema6A-deficient line has been previously described13: briefly, a cassette encoding CD4 transmembrane domain–b-galactosidase–neomycin phosphotransferase (TM-b-geo) and human placental alkaline phosphatase (PLAP), separated by an internal ribosome entry site (IRES), was inserted in the 17th intron of Sema6A14. The remaining N-terminal portion of the Sema6A protein up to amino acid 623 (and thus lacking the transmembrane and cytoplasmic domains) was fused to b-galactosidase and trapped in the endoplasmic reticulum (see Results). GFP;Sema6a mice were obtained by crossing Sema6A-deficient mice with transgenic mice expressing GFP under the chicken b-actin promoter33. P0–P5 mice were anesthetized on ice and, after P5, by inhalation of isofluorane Foren (Abbott). The day of birth corresponds to P0. All animal procedures were carried out in accordance with institutional guidelines. Genotyping of Sema6A-deficient mice. DNA were prepared from tails using the Red Extract-N-Amp Tissue PCR kit (Sigma). The Sema6A genotype was determined by PCR using the following primers: 5¢-GAG ATG CAC AGC TAA CTT CTG GTG-3¢ (wild-type allele forward primer), 5¢-TTG AAG CCT GCT CTT AGT GGC TCC-3¢ (reverse primer) and 5¢-GCT ACC GGC TAA AAC TTG AGA CCT-3¢ (mutant allele reverse primer), which amplified a 1.43-kb product for the wild-type allele and a 990-kb product for the mutant allele. Immunocytochemistry. Brains were collected as has been previously described47. Brain sections were incubated with antibodies against hSema6A (1:200 ; R&D Systems), mSema6A (1:200; R&D Systems), bgalactosidase (1:1,000; Cappel), GABAA receptor a6 subunit (1:1,000; Chemicon), CaMKIV (1:500), TAG-1 (TG1; 1:3000), L1 (1:100; Chemicon), Pax6 (1:1,000; Chemicon), CaBP (1:1,000; SWANT), parvalbumin (1:1,000; Sigma), zebrin II Q113 (1:500), GFAP (1:400; Chemicon), VGLUT2 (1:3,000; Chemicon), phosphohistone-H3 (1:1,000; Cell Signaling), activated caspase-3 (1:250; Cell Signaling), Dcx (1:1,000; Chemicon), fibronectin (1:500; Sigma), GFP (1:300; Molecular Probes), GFP (1:200; US Biological) or NCL-Ki67 (1:1,000; Novocastra) followed by species-specific secondary antibodies (Jackson ImmunoResearch). Sections were counterstained with Hoechst 33258 (10 mg/ml, Sigma), mounted in Mowiol (Calbiochem) and examined with a fluorescence microscope (DMR, Leica) or a fluorescence confocal microscope (DM IRBE, Leica). Deconvolution was performed by the 3D deconvolution module from Metamorph 6.2r2 software (Universal Imaging Corp.). BrdU injections and staining. P7 and P10 mice were injected intraperitoneally with BrdU (Sigma; 15 mg/ml, 50 mg per kg of body weight) diluted in a saline solution. Animals were perfused 3 h or 108 h after injection. Brain sections were incubated with a rat antibody to BrdU (1:100; Harlan). In situ hybridization. Antisense riboprobes were labeled as described previously47 by in vitro transcription of cDNAs encoding mouse Math1 (gift from M. Wassef48), Barlh1 (gift from F. Qiu21) or Sema6a15. In situ hybridization was done as described47. BDA injections. Adult mice were anesthetized with ketamine (Imalgene 500, 146 mg/kg; Merial) and xylazine (Rompun 2%, 7.4 mg/kg; Bayer), after which 1 ml of 10% biotin-dextran (BDA; 10,000 MW; Molecular Probes) was injected into the cerebellum with a glass micropipet. Mice were killed 48 h later. BDA was revealed with CY3-conjugated streptavidin (1:400; Jackson Immunoresearch).


[

NUMBER 11

[

NOVEMBER 2005

Generation of chimeric animals. To obtain chimeras between GFP;Sema6a/ and wild-type cells, morulae from GFP;Sema6a/ GFP;Sema6a/ crosses were aggregated in vitro with morulae of wild-type mice. As controls, we aggregated morulae from wild-type mice with those from GFP;Sema6a/ GFP;Sema6a+/+ crosses. Morula aggregation was performed as has been previously described49, with a few modifications. Briefly, after removal of the zona pellucida, two morulae of appropriate genotypes were put in contact in a drop of 100 mg/ml phytohemagglutinin PHA-P (Sigma) in PBS, until they were physically attached (about 1 min). The embryo pairs were then transferred into depression wells containing M16 medium (Sigma). After overnight culture at 37 1C in 7% CO2, blastocysts resulting from aggregations were transferred into pseudopregnant C57Bl/6/CBA females. Dissociated granule cell cultures. Granule cells were purified as has been previously described50 and plated onto polylysine (0.2 mg/ml; Sigma) and laminin (20 mg/ml; Sigma), onto COS7 cells, or onto COS7 cells transfected with full-length Sema6A, in culture medium (BME (Invitrogen), 0.45% glucose, 10% horse serum, 5% fetal bovine serum (Eurobio)). Cultures were fixed after 24 or 48 h with 4% paraformaldehyde (PFA), 0.33 M sucrose and immunostained with mouse antibody to b-tubulin (Tuj-1: 1:1,000; Eurogentec). Microexplants culture. Microexplants cultures of P4 or P5 mice were prepared as described previously30. Cultures were fixed after 18, 36, 72, 96 or 108 h with 4% PFA, 0.33 M sucrose. To test Sema6A activity, EGL explants were cultured on polylysine (0.2 mg ml1; Sigma), laminin (10 mg/ml or 20 mg/ml) and recombinant Sema6-Fc (R&D Systems; 1,000 mg/ml) or in culture medium containing 1,000 ng/ml of Sema6A-Fc preclustered with antibody to human IgG1 (Jackson ImmunoResearch). Cultures were stained with phalloidin (Sigma) or immunostained with mouse antibody to b-tubulin or mouse antibody to a-tubulin (1:1,000; Sigma) and counterstained with Hoechst 33342 (10 mg/ml, Sigma). For centrosome stainings, cultures were fixed 10 min in 20 1C methanol and immunostained with rabbit antibody to g-tubulin (1:500; Sigma). Migration and outgrowth analysis. Analysis was performed with Metamorph. To measure migration rates, we delimited concentric areas of 50-mm width at increasing distances from the explant border. The number of Hoechst-labeled pixels within each area was counted and then expressed as a percentage of the total number of pixels. To evaluate the overall rate of neuronal migration, the total number of Hoechst-labeled pixels surrounding each half of the explant was counted. The number of clusters (containing a minimum of five cells) surrounding the explant was also counted. Maximal neurite length were estimated by measuring the three longest Tuj-1–positive neurites of each explant. Neuritic length was estimated by laying out a circle containing approximately 90% of the Tuj-1–positive neurites. We also counted the total number of Tuj-1–positive neurites within the 150-mm perimeter around the explant border. Time-lapse videomicroscopy. EGL microexplants were cultured 24–48 h as described below, in Petri dishes equipped with glass coverslips. Before imaging, 20 mM HEPES was added to the culture medium. Pictures were acquired with an inverted microscope (DM IRBE, Leica) equipped with a Micromax CDD Princeton camera driven with Metamorph Software. Pictures were captured every 5 min using a 20 dry objective equipped with phase optics, in a 37 1C chamber. Hoechst 33342 was added to culture medium (1:106; Sigma) to visualize the nucleus in live cells. Cresyl violet staining. Sections were colored 7 min in a 1% cresyl violet and 1% thionine solution and then differentiated in 80% ethanol and acetic acid. Western blotting. COS7 cells were transfected with Lipofectamine2000 (Invitrogen). Cell extracts were collected 48 h later, separated by SDS-PAGE and western blotted by standard methods. Proteins were immunodetected with antibodies to mSema6A (1:500; R&D Systems), hSema6A (1:250; R&D Systems), hSema6A (1:500; R&D Systems) or myc E910 (1:200; Santa Cruz), followed by horseradish peroxidase–linked secondary antibodies and ECL kit (Amersham).

1523

ARTICLES Statistical analysis. For all statistical analysis, the significance was calculated by ANOVA (Statview, Abacus Concepts).

1. Wingate, R.J. The rhombic lip and early cerebellar development. Curr. Opin. Neurobiol. 11, 82–88 (2001). 2. Ryder, E.F. & Cepko, C.L. Migration patterns of clonally related granule cells and their progenitors in the developing chick cerebellum. Neuron 12, 1011–1028 (1994). 3. Komuro, H., Yacubova, E. & Rakic, P. Mode and tempo of tangential cell migration in the cerebellar external granular layer. J. Neurosci. 21, 527–540 (2001). 4. Ramon y Cajal, S. Histologie du syste`me nerveux de l’homme et des verte´bre´s (Maloine, Paris, 1911). 5. Yacubova, E. & Komuro, H. Cellular and molecular mechanisms of cerebellar granule cell migration. Cell Biochem. Biophys. 37, 213–234 (2003). 6. Fiore, R. & Puschel, A.W. The function of semaphorins during nervous system development. Front. Biosci. 8, s484–s499 (2003). 7. Qu, X. et al. Identification, characterization, and functional study of the two novel human members of the semaphorin gene family. J. Biol. Chem. 277, 35574–35585 (2002). 8. Kikuchi, K. et al. Cloning and characterization of a novel class VI semaphorin, semaphorin Y. Mol. Cell. Neurosci. 13, 9–23 (1999). 9. Suto, F. et al. Plexin-a4 mediates axon-repulsive activities of both secreted and transmembrane semaphorins and plays roles in nerve fiber guidance. J. Neurosci. 25, 3628–3637 (2005). 10. Toyofuku, T. et al. Guidance of myocardial patterning in cardiac development by Sema6D reverse signalling. Nat. Cell. Biol. (2004). 11. Winberg, M.L. et al. Plexin A is a neuronal semaphorin receptor that controls axon guidance. Cell 95, 903–916 (1998). 12. Godenschwege, T.A., Hu, H., Shan-Crofts, X., Goodman, C.S. & Murphey, R.K. Bidirectional signaling by Semaphorin 1a during central synapse formation in Drosophila. Nat. Neurosci. 5, 1294–1301 (2002). 13. Leighton, P.A. et al. Defining brain wiring patterns and mechanisms through gene trapping in mice. Nature 410, 174–179 (2001). 14. Mitchell, K.J. et al. Functional analysis of secreted and transmembrane proteins critical to mouse development. Nat. Genet. 28, 241–249 (2001). 15. Zhou, L. et al. Cloning and expression of a novel murine semaphorin with structural similarity to insect semaphorin I. Mol. Cell. Neurosci. 9, 26–41 (1997). 16. Rabacchi, S.A. et al. Collapsin-1/semaphorin-III/D is regulated developmentally in Purkinje cells and collapses pontocerebellar mossy fiber neuronal growth cones. J. Neurosci. 19, 4437–4448 (1999). 17. Fujita, S., Shimada, M. & Nakamura, T. H3-thymidine autoradiographic studies on the cell proliferation and differentiation in the external and the internal granular layers of the mouse cerebellum. J. Comp. Neurol. 128, 191–208 (1966). 18. Nusser, Z., Sieghart, W. & Somogyi, P. Segregation of different GABAA receptors to synaptic and extrasynaptic membranes of cerebellar granule cells. J. Neurosci. 18, 1693–1703 (1998). 19. Sakagami, H., Umemiya, M., Kobayashi, T., Saito, S. & Kondo, H. Immunological evidence that the beta isoform of Ca2+/calmodulin-dependent protein kinase IV is a cerebellar granule cell-specific product of the CaM kinase IV gene. Eur. J. Neurosci. 11, 2531–2536 (1999).

20. Ahn, A.H., Dziennis, S., Hawkes, R. & Herrup, K. The cloning of zebrin II reveals its identity with aldolase C. Development 120, 2081–2090 (1994). 21. Li, S., Qiu, F., Xu, A., Price, S.M. & Xiang, M. Barhl1 regulates migration and survival of cerebellar granule cells by controlling expression of the neurotrophin-3 gene. J. Neurosci. 24, 3104–3114 (2004). 22. Wood, K.A., Dipasquale, B. & Youle, R.J. In situ labeling of granule cells for apoptosisassociated DNA fragmentation reveals different mechanisms of cell loss in developing cerebellum. Neuron 11, 621–632 (1993). 23. Matsunaga, E. et al. RGM and its receptor neogenin regulate neuronal survival. Nat. Cell. Biol. 6, 749–755 (2004). 24. Kyriakopoulou, K., de Diego, I., Wassef, M. & Karagogeos, D. A combination of chain and neurophilic migration involving the adhesion molecule TAG-1 in the caudal medulla. Development 129, 287–296 (2002). 25. Lindner, J., Rathjen, F.G. & Schachner, M. L1 mono- and polyclonal antibodies modify cell migration in early postnatal mouse cerebellum. Nature 305, 427–430 (1983). 26. Gleeson, J.G., Lin, P.T., Flanagan, L.A. & Walsh, C.A. Doublecortin is a microtubuleassociated protein and is expressed widely by migrating neurons. Neuron 23, 257–271 (1999). 27. Chan-Palay, V. Arrested granule cells and their synapses with mossy fibers in the molecular layer of the cerebellar cortex. Z. Anat. Entwicklungsgesch. 139, 11–20 (1972). 28. Hioki, H. et al. Differential distribution of vesicular glutamate transporters in the rat cerebellar cortex. Neuroscience 117, 1–6 (2003). 29. Soha, J.M., Kim, S., Crandall, J.E. & Vogel, M.W. Rapid growth of parallel fibers in the cerebella of normal and staggerer mutant mice. J. Comp. Neurol. 389, 642–654 (1997). 30. Nagata, I. & Nakatsuji, N. Granule cell behavior on laminin in cerebellar microexplant cultures. Brain Res. Dev. Brain Res. 52, 63–73 (1990). 31. Bellion, A., Baudoin, J.P., Alvarez, C., Bornens, M. & Metin, C. Nucleokinesis in tangentially migrating neurons comprises two alternating phases: forward migration of the Golgi/centrosome associated with centrosome splitting and myosin contraction at the rear. J. Neurosci. 25, 5691–5699 (2005). 32. Solecki, D.J., Model, L., Gaetz, J., Kapoor, T.M. & Hatten, M.E. Par6alpha signaling controls glial-guided neuronal migration. Nat. Neurosci. 7, 1195–1203 (2004). 33. Hadjantonakis, A.K., Gertsenstein, M., Ikawa, M., Okabe, M. & Nagy, A. Generating green fluorescent mice by germline transmission of green fluorescent ES cells. Mech. Dev. 76, 79–90 (1998). 34. Yacubova, E. & Komuro, H. Intrinsic program for migration of cerebellar granule cells in vitro. J. Neurosci. 22, 5966–5981 (2002). 35. Jensen, P., Smeyne, R. & Goldowitz, D. Analysis of cerebellar development in math1 null embryos and chimeras. J. Neurosci. 24, 2202–2211 (2004). 36. Wechsler-Reya, R.J. & Scott, M.P. Control of neuronal precursor proliferation in the cerebellum by Sonic Hedgehog. Neuron 22, 103–114 (1999). 37. Lu, Q., Sun, E.E., Klein, R.S. & Flanagan, J.G. Ephrin-B reverse signaling is mediated by a novel PDZ-RGS protein and selectively inhibits G protein-coupled chemoattraction. Cell 105, 69–79 (2001). 38. Tomoda, T., Kim, J.H., Zhan, C. & Hatten, M.E. Role of Unc51.1 and its binding partners in CNS axon outgrowth. Genes Dev. 18, 541–558 (2004). 39. Komuro, H. & Rakic, P. Modulation of neuronal migration by NMDA receptors. Science 260, 95–97 (1993). 40. Rio, C., Rieff, H.I., Qi, P., Khurana, T.S. & Corfas, G. Neuregulin and erbB receptors play a critical role in neuronal migration. Neuron 19, 39–50 (1997). 41. Borghesani, P.R. et al. BDNF stimulates migration of cerebellar granule cells. Development 129, 1435–1442 (2002). 42. Yu, H.H., Huang, A.S. & Kolodkin, A.L. Semaphorin-1a acts in concert with the cell adhesion molecules fasciclin II and connectin to regulate axon fasciculation in Drosophila. Genetics 156, 723–731 (2000). 43. Gomes, E.R., Jani, S. & Gundersen, G.G. Nuclear movement regulated by Cdc42, MRCK, myosin, and actin flow establishes MTOC polarization in migrating cells. Cell 121, 451–463 (2005). 44. Arsenio Nunes, M.L. & Sotelo, C. Development of the spinocerebellar system in the postnatal rat. J. Comp. Neurol. 237, 291–306 (1985). 45. Sievers, J., Mangold, U. & Berry, M. 6-OHDA-induced ectopia of external granule cells in the subarachnoid space covering the cerebellum. III. Morphology and synaptic organization of ectopic cerebellar neurons: a scanning and transmission electron microscopic study. J. Comp. Neurol. 232, 319–330 (1985). 46. Berciano, M.T. & Lafarga, M. Colony-forming ectopic granule cells in the cerebellar primary fissure of normal adult rats: a morphologic and morphometric study. Brain Res. 439, 169–178 (1988). 47. Marillat, V. et al. Spatiotemporal expression patterns of slit and robo genes in the rat brain. J. Comp. Neurol. 442, 130–155 (2002). 48. Louvi, A., Alexandre, P., Metin, C., Wurst, W. & Wassef, M. The isthmic neuroepithelium is essential for cerebellar midline fusion. Development 130, 5319–5330 (2003). 49. Hogan, B., Beddington, R.S., Costantini, F. & Lacy, E. Manipulating the Mouse Embryo: A Laboratory Manual (Cold Spring Harbor Laboratory Press, Cold Spring Harbor, New York, 1994). 50. Hatten, M.E. Neuronal regulation of astroglial morphology and proliferation in vitro. J. Cell Biol. 100, 384–396 (1985).

1524

VOLUME 8


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

ACKNOWLEDGMENTS We thank C. Sotelo for his constant support and comments on the manuscript; H. Sakagami and H. Kondo (Tohoku University) for anti-b isoform of CaMKIV antibody; R. Hawkes (University of Calgary, Canada) for Zebrin II Q113 antibody; D. Karagogeos (University of Crete, Greece) for TAG-1 antibody; F. Qiu (UMDNJ, Piscataway, USA) for Barhl1 cDNA; M. Wassef (ENS, Paris, France) for Math1 cDNA; K. Skurka and D. Rottkamp for help in localizing the insertion site in the Sema6a gene trap allele; R. Schwartzmann and V. Georget (Service d’Imagerie IFR83, Universite´ Paris 6, France) for their help with confocal and videomicroscopy studies; and M. Wassef, C. Lebrand, C. Me´tin and J.P. Baudoin for discussions. A.C. is supported by the Fondation pour la Recherche sur le Cerveau (FRC), the Schlumberger Foundation and the Association pour la Recherche sur le Cancer (ARC). K.J.M is a Science Foundation Ireland (SFI) Investigator. This work was supported by SFI grant 01/F.1/B006 to K.J.M; and grants from the 21st Century COE Program, and from the Core Research for Evolutional Science and Technology (CREST) of the Japan Science and Technology Agency (JST) to H.F. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

[

NUMBER 11

[



ARTICLES

TARP c-8 controls hippocampal AMPA receptor number, distribution and synaptic plasticity Nathalie Rouach1,4,5, Keith Byrd2,4, Ronald S Petralia3, Guillermo M Elias1, Hillel Adesnik1, Susumu Tomita2, Siavash Karimzadegan1, Colin Kealey1, David S Bredt2 & Roger A Nicoll1,2 Synaptic plasticity involves activity-dependent trafficking of AMPA-type glutamate receptors. Numerous cytoplasmic scaffolding proteins are postulated to control AMPA receptor trafficking, but the detailed mechanisms remain unclear. Here, we show that the transmembrane AMPA receptor regulatory protein (TARP) c-8, which is preferentially expressed in the mouse hippocampus, is important for AMPA receptor protein levels and extrasynaptic surface expression. By controlling the number of AMPA receptors, c-8 is also important in long-term potentiation, but not long-term depression. This study establishes c-8 as a critical protein for basal AMPA receptor expression and localization at extrasynaptic sites in the hippocampus and raises the possibility that TARPdependent control of AMPA receptors during synapse development and plasticity may be widespread.

Glutamate mediates most excitatory synaptic transmission in the CNS by activating AMPA- and NMDA-type ionotropic neurotransmitter receptors. AMPA receptors (AMPARs) mediate much of the momentto-moment transmission, whereas NMDA receptor (NMDAR) activation initiates both long-term potentiation (LTP) and long-term depression (LTD). Synaptic AMPARs, in contrast to NMDARs, are highly mobile, and activity-dependent recruitment of synaptic AMPARs underlies aspects of synaptic plasticity1–3. The mechanisms underlying AMPAR trafficking have received considerable attention. AMPARs are heterotetramers derived from subunits GluR1–GluR4 (refs. 4–6), and the cytoplasmic tails of AMPAR subunits interact with numerous postsynaptic scaffolding proteins7–10. Many of these proteins, including GRIP/ABP, PICK1 and SAP97, contain PDZ domains, which are protein-protein interaction motifs that bind to the C-terminal tails of specific receptors. Other cytoplasmic proteins, including NSF, AP2 and protein 4.1, bind to more proximal sites in the C-terminal domains of GluR subunits. A number of these interactions, especially those with GluR2, have been implicated in hippocampal synaptic plasticity; however, mice lacking GluR2 and GluR3 have normal LTP and LTD11. The first transmembrane protein found to interact with AMPA receptors, stargazin, which is mutated in stargazer mice12, controls the synaptic targeting and functioning of AMPARs in cerebellar granule cells by multiple mechanisms13–15. Stargazin interacts with AMPARs early in the synthetic pathway, ensures their proper maturation and promotes their surface expression16. Stargazin also binds to the scaffolding protein PSD-95, and this interaction translocates surface AMPARs to the synapse15,17. The defects in AMPAR trafficking in stargazer mice seem to be restricted to cerebellar granule cells, as

synaptic transmission in hippocampus is normal14. Among a large family of stargazin-related proteins18,19, a subset (g-3, g-4, g-8), the TARPs, can rescue the AMPAR defects in stargazer cerebellar granule cells16. g-8 has high sequence homology with stargazin18,19; however, g-8 also has two unique domains in the C-terminal tail20. TARPs are differentially distributed in the brain16, raising the possibility that they either function similarly in different brain areas or have distinct properties. However, no direct evidence exists demonstrating a critical role for TARPs in AMPAR trafficking outside of the cerebellum. To address these issues we investigated the role of the TARP g-8, which is preferentially expressed in the hippocampus16. Overexpressing g-8 in hippocampal pyramidal cells caused a threefold enhancement in extrasynaptic AMPAR-mediated currents, but no change in synaptic AMPAR currents. Mice deficient in g-8 protein showed a substantial loss and mislocalization of hippocampal GluR1 and GluR2/3 proteins. These mice also showed a differential regulation of functional AMPARs in extrasynaptic and synaptic pools: extrasynaptic receptors recorded from somatic outside-out patches were essentially depleted, whereas synaptic pools were modestly impaired. Finally, g-8/ (also known as Cacng8/) mice had impaired synaptic plasticity. This establishes g-8 as a critical protein for AMPAR expression and distribution in the hippocampus. RESULTS c-8 expression alters functional distribution of AMPARs To examine the effects of g-8 in hippocampal pyramidal neurons, we coexpressed g-8 and green fluorescent protein (GFP) via an internal ribosome entry site (g-8-IRES-GFP) in organotypic slice cultures using a Semliki Forest virus. Simultaneous recordings from GFP-labeled and

1Department of Cellular and Molecular Pharmacology and 2Department of Physiology, University of California, San Francisco, California 94143, USA. 3Laboratory of Neurochemistry, National Institute on Deafness and Other Communication Disorders, National Institutes of Health, Bethesda, Maryland 20892, USA. 4These authors contributed equally to this work. 5Present address: Laboratory of Neuropharmacology, Colle`ge de France, 11 place Marcelin Berthelot, 75005 Paris, France. Correspondence should be addressed to R.A.N. ([email protected]).

Received 19 May; accepted 29 August; published online 9 October 2005; doi:10.1038/nn1551


[

NUMBER 11

[

NOVEMBER 2005

1525

ARTICLES

GFP = γ-8–overexpressing

GluR2/3 total

GluR2 surface

neighboring control neurons compared the size of synaptic AMPAR and NMDAR currents. Recordings from a number of pairs demonstrated that g-8 overexpression did not influence synaptic transmission (mean peak amplitude IAMPA: control, 49 ± 7 pA; infected, 44 ± 7 pA, n ¼ 16 pairs; mean peak amplitude INMDA: control, 24 ± 5 pA; infected, 26 ± 6 pA, n ¼ 14 pairs; Fig. 1a,b). By contrast, g-8 caused a threefold enhancement of AMPA-mediated responses in outside-out somatic membrane patches (control, 560 ± 100 pA, n ¼ 9; infected, 1,712 ± 407 pA, n ¼ 9; Fig. 1c,d). As excitatory synapses do not contact the soma of pyramidal cells, this finding indicates that g-8 increases the number of extrasynaptic AMPARs. Is the enhancement in surface AMPARs that results from overexpression of g-8 attributable to the delivery of an intracellular pool of preformed AMPARs, or does g-8 promote the synthesis of Figure 2 Targeted disruption of g-8 decreases level of AMPAR proteins. (a) Schematic representation of g-8 protein, genomic locus, targeting vector and targeted allele. Homologous recombination disrupts transmembrane domains 2 and 3 (TM2, TM3). PDZ-BD: PDZ binding domain. E2: exon 2; E3: exon 3. (b) Southern blot analysis of SpeI-digested genomic DNA using the probe in a demonstrates proper targeting. (c) Quantitative immunoblot analysis of hippocampal extracts from g-8+/+, g-8+/ and g-8/ mice confirms that the g-8 protein is not present in g-8/ mice. The levels of AMPAR proteins GluR1 and GluR2/3 are reduced in g-8+/ and g-8/mice. The amount of NMDAR subunit NR1 is unaffected. Similar tubulin (Tub) levels confirm equivalent loading. (d) AMPAR protein levels are reduced in g-8+/ and g-8/ mice (summary of three experiments). Error bars represent s.e.m. (e) Northern blotting indicates that deleting g-8 does not affect mRNA levels for GluR1, GluR2/3 or PSD-95. (f) In the hippocampus of g-8/ mice, a large fraction of GluR2/3 remains immature and sensitive to EndoH glycosidase (a longer exposure of this western blot shows a small amount of EndoH-resistant GluR2/3 protein). Glycosylation is removed by the nonspecific N-glycosidase, PNGaseF.

1526

c-8 disruption alters AMPAR expression and distribution We disrupted the g-8 gene by homologous recombination in embryonic stem cells (Fig. 2a). Southern blot analysis of SpeI-digested genomic DNA using the probe schematically depicted in Figure 2a was used to ensure proper targeting (Fig. 2b). Crossing g-8+/ mice showed that g-8/ mice were born in mendelian ratios, and no gross behavioral or anatomical defects were apparent. Western blot analysis (Fig. 2c,d)

a

γ-8

TM 1

N

TM 2

TM 3

TM 4

PDZ-BD C

25 AA E2

E3 p1

Spe 1

Spe 1

p2

neo Spe 1

Spe 1

1 kb

p3

neo

b

+/+

+/–

c

+/–

+/+

d

–/–

γ-8 GluR1 4.5 GluR2/3

2.5

Probe

p4

Spe 1

Spe 1

Spe 1

125

–/– +/–

100

NR1

75

*

*

50 25 0

Tub

* *

*

*

1

γ-8

e

+/+

–/–

+/+

–/–

+/+

–/–

f e

Control

as

0

100

GluR2/3

G

50

Control cells

3

e

0

AMPARs that are then delivered to the surface? To examine this issue, we overexpressed g-8 in dissociated hippocampal neurons and compared the level of surface and total GluR2 levels in g-8–overexpressing cells and neighboring uninfected cells. We found that the total GluR2 levels were unchanged in g-8–overexpressing cells (96 ± 8% of control cells, n ¼ 42 pairs of neurons from four experiments). However, the ratio of surface-to-total GluR2 levels in the soma of g-8 overexpressing cells was higher than in neighboring control cells (166 ± 14% of control cells, n ¼ 42 pairs of neurons from four experiments; Fig. 1e). This suggests that overexpressing g-8 may enhance the delivery of a preexisting intracellular pool of AMPARs to the plasma membrane.

R

0

–500

N

50

–1,000

Endo H –

+

–

+

PN

100

–1,500

1

I NMDA (pA)

2/

–100

Control cells

R

–50

R

0

lu

* –2,000

G

d

–50

Figure 1 g-8 expression increases extrasynaptic but not synaptic AMPARs. (a) Sample traces of evoked EPSCs recorded simultaneously from a pair of CA1 pyramidal cells at membrane potentials of –70 mV and +40 mV. The recordings are from a g-8-IRES-GFP–infected cell (g-8) and a neighboring uninfected cell (control). Scale bars: 10 pA, 50 ms. (b) Distribution of EPSC amplitudes in pairs of g-8–infected cells and control cells shows that g-8 has no effect on either synaptic AMPAR or NMDAR EPSCs (P ¼ 0.25, n ¼ 16 pairs and P ¼ 0.63, n ¼ 14 pairs, respectively). Large filled circles represent the average value for all cells. (c) Sample traces of AMPA-evoked currents (2 s, 500 mM AMPA + 100 mM cyclothiazide) from outside-out somatic patches in control and g-8–infected pyramidal cells. Scale bar, 200 pA, 2 s. (d) g-8 overexpression markedly increases (B300%) extrasynaptic AMPAR responses (P ¼ 0.001, n ¼ 9 for control and infected cells). (e) Immunofluorescent labeling of rat hippocampal cultured neurons overexpressing g-8 (GFP-expressing neuron at left) and control neurons (non–GFP expressing neuron at right) shows that in g-8–overexpressing cells, total somatic GluR2 levels were unchanged, but surface GluR2 levels were increased.

γ8

γ-8

lu

Control

–100

I AMPA peak amplitude (pA)

γ-8–infected cells

c

I AMPA (pA)

0

γ-8–infected cells


b

γ-8

G

Control

Protein level (% of +/+)

a

4.7 GluR1

VOLUME 8

[

GluR2

NUMBER 11

PSD-95

[

+/+

–/–


ARTICLES

a

+/+

b

–/–

+/+ Sr

Cx


γ-8

Cp

–/–

Py

Cb

Hp

GluR1

GluR2/3

NR1

Figure 3 g-8 mediates subcellular targeting of hippocampal AMPARs. Immunohistochemical staining of sagittal brain (a) and hippocampal (b) sections from g-8+/+ and g-8/ mice. (a,b) Immunostaining shows diffuse localization of g-8 protein in hippocampus (Hp), cerebral cortex (Cx) and corpus striatum (Cp) in g-8+/+ mice, all of which is absent in g-8/ mice. The distributions of GluR1 and GluR2/3 are altered selectively in the hippocampus of g-8/ mice. In g-8+/+ mice, these AMPAR subunits are diffusely localized in the neuropil of all layers in hippocampus and are less concentrated in the pyramidal cell bodies (Py). In contrast, GluR1 and GluR2/3 expression is reduced in the neuropil and concentrated in the pyramidal cell bodies of g-8/ mice. The localization of the NMDAR subunit NR1 remains unchanged in the hippocampus of g-8/ mice. Sr: stratum radiatum; Py, pyramidal cell. (c) Immunogold labeling of hippocampal spines shows differential reductions of GluR1 in g-8/ mice. Synapses from g-8/ hippocampi have less than one-third the amount of synaptic GluR1 (arrows) than do g-8+/+ synapses. Extrasynaptic GluR1 (arrowhead) are reduced by B95%. The absence of g-8 also reduces the amount of cytoplasmic GluR1 (asterisk) by B80%. Scale bar: 100 nm.

confirmed that g-8 protein was absent in g-8/ mice. We did not find any change in the level of expression of other TARPs (data not shown). Quantitative western blot analysis also showed that GluR1 and GluR2/3 protein levels in the hippocampus were reduced by B30% in g-8+/ and B85% in g-8/ mice (Fig. 2c,d). NMDA receptor (NR1) and tubulin levels were unchanged. Northern blot analysis of hippocampus (Fig. 2e) demonstrated that mRNA for GluR1, GluR2 and PSD-95 occurred in normal amounts in g-8/ mice (normalization to g-8+/+: 1 ± 0.12 for GluR1, 1.06 ± 0.11 for GluR2, 1.09 ± 0.07 for PSD-95; n ¼ 3), implying that g-8 does not affect the transcription of AMPARs and that the loss of g-8 is likely to destabilize GluR1 and GluR2/3 proteins. Many membrane proteins are regulated by quality control mechanisms for exit from the endoplasmic reticulum21. Glutamate receptors receive high mannose glycosylation in the endoplasmic reticulum and are later modified with more complex sugars in the Golgi22,23. The glycosylation patterns can be distinguished with endoglycoside H (EndoH), which digests the immature high mannose sugars, and PNGaseF, which removes all N-linked carbohydrates. In contrast to g-8+/+ mice, much of the GluR2/3 from the hippocampus of g-8/ mice was sensitive to EndoH (Fig. 2f), suggesting a large pool of immature receptors held in the endoplasmic reticulum. The cleaved


[

NUMBER 11

[

NOVEMBER 2005

c

+/+

–/–

*

band of receptors comigrated with the completely deglycosylated receptors obtained by treatment with PNGaseF. Immunohistochemical staining (Fig. 3a,b) also showed that levels of GluR1 and GluR2/3 were greatly reduced in g-8/ mice and that the remaining receptors were mislocalized as they were redistributed from the dendrites to the cell body (Fig. 3a,b). This loss of AMPARs was selective, as the NR1 subunit of NMDARs (Fig. 3a,b) and PSD-95 (data not shown) were unaltered. To further characterize AMPAR distribution at the subcellular level, we performed immunogold electron microscopy of GluR1 (Fig. 3c and Supplementary Fig. 1). The synaptic receptor density was decreased by approximately 67% in g-8/ mice (g-8+/+, 1.77 ± 0.3, n ¼ 86; g-8/, 0.56 ± 0.1, n ¼ 129, P ¼ 0.0003). Furthermore, the number of extrasynaptic receptors was reduced by B95% (g-8+/+, 0.63 ± 0.14, n ¼ 85; g-8/, 0.04 ± 0.02, n ¼ 128, P ¼ 0.0001), and cytoplasmic receptors were reduced in number by B80% (g-8+/+, 0.96 ± 0.18, n ¼ 85; g-8/, 0.14 ± 0.04, n ¼ 128, P ¼ 0.00002). The pattern of staining of GluR2/3 in g-8/ mice is markedly similar to that reported for GluR1/ (also known as GluRA/) mice24. As AMPAR expression is dependent on the presence of g-8, we wondered if the amount and distribution of g-8 might be altered in the GluR1/ mice. The level of g-8 protein measured with western

1527

ARTICLES

a

b +/+

d

–/–

+/+

–/–

NMDA

AMPA

–0.20 –0.15 –0.10 –0.05 0.00 0.05

0.10

0.15

0.20

0.25

0.30

1.0

Cumulative probability

–0.25

e AMPA/NMDA ratio (+ 40 mV)

fEPSP slope (mV/ms)

c

–0.30

1.0 0.8

*

0.6 0.4 0.2 0.0

0.8 0.6 0.4 +/+ –/–

0.2 0.0

+/+

+/–

–/–

0

–10

Fiber volley amplitude (mV)

f

+/+ –/–

1.2

0.8

0.4

–40

–50

1.2

80 1.0

60

Rectification index

Normalized IAMPA amplitude (% I –70 mV)

100

1.6

–30

i 120

g

–20

mEPSC amplitude (pA)

h

PPF


–0.35

40 20 0 –20 –40 –60 –80

0.8 0.6 0.4 0.2

–100 0.0

+/+

–/–

–120 –80

0.0 –60

–40

–20

0

20

40

60

80

+/+

–/–

Vm

Figure 4 CA1 synaptic AMPAR-mediated responses are impaired in g-8/ mice. (a) Input-output curves for basal synaptic transmission in hippocampal slices. As illustrated in the sample traces and the graph below, for each input (fiber volley, see arrow), the output (fEPSP) is reduced by B35% in g-8/ slices (P r 0.01, n ¼ 22 for g-8+/+; n ¼ 27 for g-8/). Scale bar, 0.1 mV, 10 ms. (b,c) Ratio of AMPA to NMDA current is reduced by B30% in CA1 pyramidal cells from g-8/ mice (P r 0.01, n ¼ 21 for g-8+/+; n ¼ 20 for g-8+/; n ¼ 35 for g-8/). Sample NMDA and AMPA EPSCs from representative cells are shown in b above the respective bars. Calibration: g-8+/+, 12 pA, 50 ms; g-8+/, 16 pA, 50 ms; g-8/, 14 pA, 50 ms. (d,e) mEPSCs from g-8/ CA1 pyramidal cells have reduced amplitude but unchanged frequency. Sample traces are shown in d. Scale bar, 10 pA, 0.5 s. e shows cumulative frequency distribution of mEPSC amplitudes (P r 0.01, Kolmogorov-Smirnov test, n ¼ 14 for g-8+/+; n ¼ 24 for g-8/). (f,g) Paired-pulse facilitation (PPF) does not differ between g-8/ (n ¼ 19) and g-8+/+ (n ¼ 19) cells. f shows sample traces above the graph. Scale bar, 25 pA, 25 ms. (h) Current-voltage (I/V) plots of evoked AMPA EPSCs show no difference between g-8/ (n ¼ 20) and g-8+/+ (n ¼ 14) cells. (i) Rectification indices are similar in both genotypes (n ¼ 14 for g-8+/+; n ¼ 20 for g-8/). Error bars in all graphs represent s.e.m.

Synaptic and extrasynaptic AMPARs are impaired in c-8/ mice To assess the strength of synaptic transmission, we compared the size of the presynaptic fiber volley (input) with the slope of the EPSP (output) in stratum radiatum (Fig. 4a) and found that synaptic transmission was reduced by B35% in g-8/ mice (Fig. 4a). To determine whether this reduction in excitatory synaptic transmission was specific to AMPARs, we compared the AMPAR and NMDAR components of the EPSC (Fig. 4b). The AMPA/NMDA ratio was reduced by approximately 30% in g-8/ mice (Fig. 4c), indicating that the defect in synaptic transmission in g-8/ mice was limited to the AMPAR component (AMPA/NMDA ratio: g-8+/+, 0.85 ± 0.07, n ¼ 21; g-8+/, 0.88 ± 0.09, n ¼ 20; g-8/, 0.65 ± 0.05, n ¼ 35). As a further test, we

compared the stimulus strength used to generate NMDAR EPSCs to the amplitude of these EPSCs in g-8+/+ and g-8/ cells. The size of the stimulus required to generate similar-size EPSCs in g-8+/+ and g-8/ mice was not significantly different (g-8+/+, INMDA ¼ 43.6 ± 8.3 pA, stimulus size ¼ 2.7 ± 0.4 arbitrary units, ratio of INMDA/stimulus size ¼ 22 ± 4.4, n ¼ 21; g-8/, INMDA ¼ 60.3 ± 9.1 pA, stimulus size ¼ 3.4 ± 0.3 a.u., ratio of INMDA/stimulus size ¼ 25.4 ± 6, n ¼ 35). The lack of change in the NMDAR EPSC argues against a change in transmitter release. We also found a significant reduction in the size of the AMPAR mEPSCs (Fig. 4d,e; mEPSC mean peak amplitude: g-8+/+, 18.1 ± 1.3 pA, n ¼ 14; g-8+/, 15.9 ± 1.5 pA, n ¼ 18; g-8/, 11.9 ± 0.8 pA, n ¼ 24). Paired-pulse facilitation, a sensitive measure of changes in the probability of transmitter release, was unchanged (g-8+/+, 1.59 ± 0.08, n ¼ 19; g-8+/, 1.61 ± 0.09, n ¼ 14; and g-8/ 1.46 ± 0.04, n ¼ 19; Fig. 4f,g). Furthermore, the frequency of mEPSCs, a measure of the probability of transmitter release, was unchanged (1.9 ± 0.4 Hz, n ¼ 14 in g-8+/+, 2.4 ± 0.4 Hz, n ¼ 18 in g-8+/ and 1.7 ± 0.5 Hz, n ¼ 24 in g-8/ cells). Finally, we examined the current/voltage relationship of AMPAR EPSCs to determine if the biophysical property of the

1528

VOLUME 8

blotting was slightly reduced, but not nearly as much as the decrease of GluR1 in g-8/ mice (data not shown). Immunohistochemical staining of the GluR1/ hippocampus showed that the remaining GluR2/3 was redistributed to the cell body layer as reported previously24, but there was no clear change in the pattern of staining for g-8 (data not shown).

[

NUMBER 11

[


receptors was altered in g-8/ cells (Fig. 4h). We did not find any change in the rectification index (Fig. 4i; RI ¼ 1.02 ± 0.04, n ¼ 14 for g-8+/+, RI ¼ 1.01 ± 0.06, n ¼ 11 for g-8+/ cells and RI ¼ 1.01 ± 0.05, n ¼ 20 for g-8/ cells), suggesting no change in AMPAR subunit composition. Notably, in all experiments performed, we did not observe any difference in basal synaptic transmission between g-8+/+ and g-8+/ mice. Are the defects in synaptic AMPAR-mediated transmission restricted to CA1 Schaffer collateral synapses? To address this issue, we turned to the CA3 mossy fiber synapse and compared the size of the fiber volley with the AMPAR field EPSP in g-8+/+ and g-8/ mice (Fig. 5a). As with the synapses in CA1, the AMPAR EPSP was clearly reduced. Although the excitatory synapses in the CA1 region of the hippocampus are typical of most cortical excitatory synapses in that AMPARs and NMDARs are colocalized in the postsynaptic density, hippocampal mossy fiber synapses also express kainate receptors (KARs)25,26. Does g-8 have a role in trafficking KARs? After first comparing the size of the fiber volley with the AMPAR EPSP field (Fig. 5b), as previously described, we then added the AMPAR-selective antagonist GYKI 53655 and stimulated the mossy fibers repetitively to evoke a KAR-mediated field EPSP (fEPSP; Fig. 5c). We then compared the size of the kainate fEPSP with the original fiber volley. We did not detect any difference between g-8+/+ and g-8/ mice (Fig. 5d). Although the defect in synaptic transmission in g-8/ mice was significant, the magnitude of the defect was modest compared with the marked loss of AMPAR protein. To address this issue, we recorded in the presence of tetrodotoxin whole-cell responses to bath application of AMPA, which activates AMPARs throughout the cell, including somatic and dendritic extrasynaptic AMPARs, as well as synaptic AMPARs. We found a marked reduction in the size of currents in g-8+/ mice (392 ± 57 pA, n ¼ 9) and g-8/ mice (156 ± 25 pA, n ¼ 15) compared with g-8+/+ mice (674 ± 84 pA, n ¼ 14; Fig. 6a). This result suggests that the pool of extrasynaptic receptors is greatly reduced. This possibility was directly tested by pulling outside-out patches from the soma of hippocampal pyramidal cells. Indeed, there was a 90% decrease in AMPAR-mediated responses from g-8/ mice (Fig. 6b,c). Furthermore, as with the whole-cell responses, the responses in g-8+/ mice (366.8 ± 70.3 pA, n ¼ 20) were intermediate between g-8+/+ (601.3 ± 72.3 pA, n ¼ 12) and g-8/ mice (52.2 ± 9.5 pA, n ¼ 17), indicating that the number of extrasynaptic AMPARs is tightly coupled to the number of g-8 molecules. Although the extrasynaptic responses were obtained from somatic membrane, the immunogold studies suggest that the loss of extrasynaptic receptors also occurs throughout the dendrites. We also examined the responses


[

NUMBER 11

[

NOVEMBER 2005

a

–0.5 fEPSP slope (mV/ms)

Figure 5 CA3 synaptic AMPAR, but not KAR, responses are impaired in g-8/ mice. (a) Input-output curves for basal CA3 mossy fiber synaptic transmission in hippocampal slices from g-8+/+ and g-8/ mice. The slope of fEPSPs evoked (0.05 Hz) by a range of stimulus intensities is plotted against the amplitude of the corresponding fiber volley. As illustrated, for each input (fiber volley), the output (fEPSP) is reduced by B50% in slices from g-8/ mice (P r 0.01, n ¼ 4 for g-8+/+; n ¼ 4 for g-8/). AMPAR-mediated mossy fiber fEPSPs (b) were evoked by single stimulation (0.2 Hz) in the presence of AP5 (50 mM) and picrotoxin (100 mM) and were abolished by GYKI 53655 (100 mM), whereas KAR-mediated mossy fiber fEPSPs (c) were subsequently evoked by repetitive stimulation (six stimuli at 30 Hz given every 5 s) and were abolished by CNQX (50 mM). As illustrated in the sample traces (b,c) and the summary below (d), for a given fiber volley, the AMPAR fEPSP amplitude is reduced by B50% in slices from g-8/ mice, whereas the KAR fEPSP is unaltered (P r 0.01, n ¼ 12 for g-8+/+; n ¼ 9 for g-8/). Scale bars, 0.1 mV, 10 ms (AMPAR fEPSP) and 0.05 mV, 50 ms (KAR fEPSP). Error bars in all graphs represent s.e.m.

+/+ –/–

–0.4 –0.3 –0.2 –0.1 0.0 0.05

0.10 0.15 0.20 0.25 Fiber volley amplitude (mV)

b

0.30

c

d

6 5 fEPSP/fiber volley


ARTICLES

+/+ –/–

4

*

3 2 1 0 AMPA

Kainate

to NMDA application with both whole-cell recording and outside-out patches and found no difference between g-8+/+ and g-8/ mice (whole-cell recording: g-8+/+, 1.0 ± 0.1 nA, n ¼ 5; g-8/, 1.1 ± 0.1 nA, n ¼ 5; outside-out patches: g-8+/+, 31.6 ± 0.6 pA, n ¼ 10; g-8/, 31.0 ± 0.4 pA, n ¼ 8). All cell types analyzed in the hippocampus of g-8/ mice, including pyramidal cells in CA1, interneurons in stratum radiatum and granule cells in dentate gyrus, showed a marked alteration in extrasynaptic AMPAR responses (Fig. 6d,e; interneurons: g-8+/+, 254 ± 83 pA, n ¼ 7; g-8/, 36 ± 14 pA, n ¼ 6; granule cells: g-8+/+, 302 ± 53 pA, n ¼ 7; g-8/, 96 ± 29 pA, n ¼ 6). In contrast, striatal medium spiny neurons had normal extrasynaptic AMPA responses (Fig. 6d,e; g-8+/+, 126 ± 22 pA, n ¼ 7; g-8/, 123 ± 30 pA, n ¼ 6), indicating that the effect of g-8 deletion is specific to the hippocampus. Is the trafficking of synaptic AMPARs in the g-8/ mice due to other TARPs or to a TARP-independent mechanism? To begin to address this question, we have generated double g-2/g-8/ knockout mice. These mice showed a more severe reduction (B50%) in synaptic transmission (AMPA/NMDA ratio g-2+/+g-8+/+, 0.87 ± 0.06, n ¼ 41; g-8/, 0.65 ± 0.05, n ¼ 35; g-2/g-8/, 0.44 ± 0.06, n ¼ 9), supporting the proposal that at least some of the remaining AMPAR-mediated synaptic transmission in the g-8/ mice is due to other hippocampal TARPs. Long-term potentiation is impaired in c-8/ mice What might be the consequences of the loss of g-8 on synaptic plasticity? In a blind fashion, we compared the magnitude of LTP induced by brief tetanic stimulation (two 1-s, 100-Hz tetani separated by 20 s), using field potential recordings in the CA1 area of the hippocampus. LTP in g-8/ mice was reduced by 75% compared with LTP in g-8+/+ mice (fEPSP slope 30–40 min after tetanization:

1529

ARTICLES

a I AMPA amplitude (pA)

0 –100 –200 –300 –400 –500

+/+ +/– –/–

–600


–700 –800 0

2

4

6

8

10 12 14 16 18 20 22 24 26 28 30

Time (min)

b


c

presynaptically27,28. Mossy fiber LTP was unaltered in g-8/ mice (g-8+/+, 121 ± 5%, n ¼ 6; g-8/, 125 ± 7%, n ¼ 9). We also examined long-term depression (LTD) in CA1 hippocampal pyramidal cells from g-8/ mice. We carried out whole-cell recording and induced LTD by 3-min pairing of 5-Hz synaptic stimulation with depolarization of the cell to –40 mV. We did not find any difference in the magnitude of LTD between g-8+/+ and g-8/ cells (Fig. 7c).

–700 –600 –500

*

–400 –300 –200

*

–100 0

+/+

+/–

–/–

d


e

Figure 6 Extrasynaptic AMPAR-mediated responses are severely reduced in hippocampus of g-8/ mice. (a) Whole-cell currents evoked by bath application of AMPA (10 min, 100 nM + 0.5 mM tetrodotoxin + 10 mM cyclothiazide) were strongly reduced in g-8/ pyramidal cells (g-8/, n ¼ 15 decreasing to 12; g-8+/ n ¼ 9; g-8/ n ¼ 16 decreasing to 11). (b,c) AMPA-evoked currents (2 s, 500 mM AMPA + 100 mM cyclothiazide) from outside-out somatic patches decreased markedly (90%) in the number of extrasynaptic AMPARs in g-8/ hippocampal pyramidal cells. Responses in g-8+/ cells (n ¼ 20) are intermediate between g-8+/+ (n ¼ 12) and g-8/ cells (n ¼ 17). Sample currents are shown above the respective bars. Calibration, 100 pA, 2s. (d,e) Hippocampal interneurons in stratum radiatum and granule cells in dentate gyrus from g-8/ mice also show a marked alteration in extrasynaptic AMPAR-mediated responses (interneurons: g-8+/+, n ¼ 7; g-8/, n ¼ 6; granule cells: g-8+/+, n ¼ 7; g-8/, n ¼ 6). However, g-8/ medium spiny neurons recorded from the striatum have normal extrasynaptic AMPAR-mediated responses (g-8+/+, n ¼ 7; g-8/, n ¼ 6). Scale bar, 100 pA, 2 s. Error bars in all graphs represent s.e.m.

–350 –300 –250 –200

DISCUSSION These results indicate that TARP g-8 regulates AMPAR protein levels and extrasynaptic surface expression. By controlling the number of AMPA receptors, g-8 has an important role in LTP, which requires the recruitment of AMPARs to synapses. These findings extend previous work on stargazer in cerebellar granule cells by showing that a protein homologous to stargazin strongly regulates AMPAR expression and function in the hippocampus. g-8 has properties distinct from stargazin, as it preferentially controls hippocampal extrasynaptic AMPAR pools. These findings were unexpected, considering the stargazer mouse phenotype, which shows only modest reduction in cerebellar AMPAR levels and an absence of both synaptic and extrasynaptic AMPARs.

g-8+/+, 140 ± 12%, n ¼ 10; g-8/, 109 ± 4%, n ¼ 17; Fig. 7a). Because synaptic AMPAR responses are reduced in g-8/ mice, it was possible that unblocking of NMDARs was reduced compared with control slices. To circumvent this, recordings were made with cesium-filled electrodes to facilitate depolarizing the cell, and LTP was induced by 1-min pairing of 2-Hz synaptic stimulation with depolarization of the cell to 0 mV. The defect in LTP was still apparent in these experiments (EPSC amplitude 30–40 min after pairing, g-8+/+, 165 ± 14%, n ¼ 10 versus g-8/, 118 ± 11%, n ¼ 14; Fig. 7b), indicating that the defect in LTP in g-8/ mice is downstream of NMDAR activation. Notably, in all the experiments performed, we did not observe any difference in LTP between g-8+/+ and g-8+/ mice. As a control for the effect of deleting g-8 on NMDAR-dependent LTP, we also examined mossy fiber LTP in CA3, which is independent of NMDARs and is expressed

Loss of AMPAR protein in the c-8/ hippocampus The role of g-8 in AMPAR function in hippocampus showed notable differences from the role of stargazin in the cerebellum. Stargazer mice have a B20% reduction in GluR2/3 levels in cerebellum16. On the other hand, the primary defect in the g-8/ mouse was the loss of AMPAR proteins in the hippocampus (85%). It is noteworthy that this massive loss makes it difficult to determine to what extent the defect in synaptic and extrasynaptic AMPAR-mediated currents actually reflects compromised supply pools of receptors, compromised trafficking mechanisms, or both. Indeed, in addition to the regulation of AMPAR expression, g-8 may also contribute to the increased delivery of AMPARs to the membrane from an intracellular pool, as suggested by the acute enhancement in the number of surface AMPARs induced by g-8 overexpression, independent of total AMPAR protein levels. How might assembly with g-8 regulate AMPAR expression? The normalcy of GluR1 and GluR2 mRNA levels in g-8/ hippocampus suggests that the loss of AMPAR proteins may be due to enhanced receptor degradation, although one cannot exclude altered protein synthesis. However, the absence of stargazin induces the unfolded protein response in cerebellar granule cells29, supporting the idea that TARPs help stabilize AMPAR proteins and prevent their degradation by the proteasome. This by itself could increase the probability of AMPAR membrane insertion. One cannot exclude the possibility that g-8 might

1530

VOLUME 8

–150

*

–100

*

–50 0

+/+ –/– Granule cells

+/+ –/– +/+ –/– Interneurons Medium spiny neurons

Hippocampus

Striatum

[

NUMBER 11

[


ARTICLES +/+

b

+/+ –/–

200

150

100

–/–

+/+

c

Normalized EPSC amplitude

Normalized fEPSP slope


–/–

Normalized EPSC amplitude

+/+

a

200

150

100

–/–

140 120 100 80 60 40 20 0

0

10

20

30

40

50

0

10

Time (min)

20

30

Time (min)

40

50

0

10

20

30

40

50

Time (min)

Figure 7 Impairment of LTP, but not LTD, in hippocampal CA1 synapses of g-8/ mice. (a) Extracellular recordings of fEPSPs in slices before and after tetanic stimulation of Schaffer collaterals (arrow, two trains of 100 Hz for 1 s, 20 s apart). LTP was produced in g-8+/+ slices (n ¼ 10) and was significantly (P r 0.05) reduced in g-8/ slices (n ¼ 17). Sample traces represent averaged field potentials before and 30–40 min after tetanization. Scale bar, 0.1 mV, 10 ms. Data were obtained in 100 mM picrotoxin and 4 mM Ca2+ and Mg2+. (b) Whole-cell recordings from CA1 pyramidal cells before and after a pairing protocol (arrow, 2 Hz, 0 mV for 1 min.). LTP was induced in g-8+/+ cells (n ¼ 14 decreasing to 9) but was severely decreased in g-8/ cells (P r 0.01, n ¼ 22 decreasing to 11). Sample EPSCs recorded before and 30–40 min after pairing are shown above. Scale bar for b and c: 50 pA, 50 ms. (c) Whole-cell recordings from CA1 pyramidal cells before and after a pairing protocol (bar, 5 Hz, 40 mV for 3 min). LTD was induced, and no difference was found between g-8+/+ (n ¼ 6) and g-8/ cells (n ¼ 7). Sample EPSCs recorded before and 30–40 min after pairing are shown above.

also stabilize extrasynaptic and synaptic AMPARs through reduced basal receptor endocytosis, although this seems unlikely for activitydependent synaptic receptor endocytosis, as LTD is not modified in g-8/ mice. The immature endoplasmic reticulum–type glycosylation of AMPARs in the g-8/ hippocampus also suggests that g-8 has a role in AMPAR trafficking early in the biosynthetic pathway. Indeed, incompletely folded or assembled proteins cannot exit the endoplasmic reticulum because of a stringent quality control system30. Lack of g-8 may expose endoplasmic reticulum retention signals on AMPAR subunits, as has been shown for the regulated trafficking of ATPsensitive K+ channel complexes31. Future studies will be required to determine the detailed molecular mechanisms for g-8 regulation of AMPAR expression. Deletion of the GluR1 subunit24 redistributes GluR2/3 in a manner similar to what we found here for both GluR1 and GluR2/3 in the g-8/ mice. However, in contrast to the deletion of g-8, GluR1/ mice show modest changes in the expression of GluR2, GluR3 and GluR4 (refs. 24,32). The amount of g-8 protein is only slightly reduced, and its distribution not obviously affected in the GluR1/ mice. We suggest that in the GluR1/ mice there is a great excess of g-8 compared with the remaining GluR2/3 receptors. Thus, although GluR2/3, which is presumably associated with g-8, is mislocalized, this represents a small fraction of the total g-8 pool. This is likely to indicate that g-8 does not require association with AMPARs for its processing and dendritic trafficking. This is reminiscent of other ionic channel non-pore-forming auxiliary subunits, such as the b1b and b2a subunit of L-type calcium channels, which can be normally expressed and trafficked to the membrane in the absence of the pore-forming subunit a1 (refs. 33–35). Presumably, normal trafficking of these auxiliary subunits is made possible by the lack of both a retention signal from these proteins and an export signal from other proteins. Differential delivery of limited AMPARs to the synapse Functional AMPARs are essentially absent from extrasynaptic membranes and synapses in cerebellar granule cells of stargazer mice13–15,


[

NUMBER 11

[

NOVEMBER 2005

whereas AMPAR synaptic transmission is reduced by only B35% in the hippocampus of g-8/ mice, despite the marked decrease in AMPAR expression. Whereas stargazin seems to be the only TARP expressed in cerebellar granule cells, hippocampal pyramidal cells express significant amounts of other TARPs (g-2, g-3 and g-4)15,16. This redundancy could mask some functions of g-8 that we did not demonstrate in the present study, such as a direct role in synaptic trafficking. Indeed, it is possible that AMPAR trafficking by other TARPs accounts for the remaining currents in hippocampal g-8/ pyramidal cells. g-2/g-8/ mice show a more severe reduction (B50%) in synaptic transmission, supporting the proposal that other hippocampal TARPs contribute to the synaptic targeting of at least part of the remaining AMPARs in the g-8/ mice. Alternatively, it is possible that some residual TARP-independent trafficking of AMPARs occurs in hippocampal pyramidal cells. Whereas the synaptic AMPAR response was reduced by only B35%, the extrasynaptic responses were reduced 90% in the g-8/ mice. As extrasynaptic AMPARs represent the majority of surface AMPARs, we cannot exclude that the massive decrease in extrasynaptic receptors may in part be secondary to the diminished pool of AMPARs in the cell. In addition, extrasynaptic AMPAR responses are significantly reduced in the g-8+/ mice and greatly enhanced in neurons overexpressing g-8. This enhancement occurrs in the absence of obvious changes in synaptic AMPARs. This could be due to a limited number of g-8–interacting proteins (such as membrane-associated guanylate kinases (MAGUKs)), necessary for the synaptic insertion of AMPARs, or to a sufficient level of endogenous g-8 to support AMPAR synaptic targeting. These results, which are similar to the results of overexpression of g-2 (ref. 17), indicate that the number of extrasynaptic AMPARs is tightly regulated by the levels of g-8. On the other hand, synaptic AMPARs are largely immune from the marked loss of cytoplasmic and extrasynaptic AMPARs, indicating that pyramidal cells selectively maintain AMPAR-mediated synaptic transmission with very few receptors. GluR1/ mice show a similar loss of extrasynaptic AMPARs with modest36 or no24 decreases in

1531

ARTICLES


synaptic transmission. We propose that the strong reduction in the total number of AMPARs observed in both the GluR1/ and the g-8/ mice is responsible for this common phenotype. Indeed, synaptic AMPARs, which represent a small proportion of surface AMPARs, can presumably be maintained by the relatively normal levels of other AMPAR subunits (GluR2, GluR3, GluR4) in the GluR1/ mice and by the few remaining AMPAR subunits (GluR1, GluR2, GluR3, GluR4) that may be associated with other TARPs in the g-8/ mice. LTP is impaired in the c-8/ mouse The defect in LTP in g-8/ mice, which occurs downstream of NMDAR activation, has a number of possible explanations. An intriguing possibility is that g-8 has a direct role in LTP. In support of this conclusion we have recently found that the intracellular C termini of stargazin and other TARPs have multiple phosphorylation sites for CaMKII/protein kinase C (PKC) and that expressing a construct in which the serines are mutated to alanines also blocks LTP37. Another possibility is that during LTP, AMPARs are recruited to the synapse either from the extrasynaptic pool by lateral diffusion or from an intracellular pool, and the loss of AMPARs in g-8/ mice contributes to the defect in LTP. A similar interpretation could explain the loss of LTP in the GluR1/ mice24. Unique role for TARPs in regulating ionotropic receptors Numerous studies have addressed the mechanisms underlying synaptic AMPAR trafficking1–3,7–10, but a number of issues remain poorly understood. The proposed mechanisms include phosphorylation of the C-terminal region of AMPARs and protein-protein interactions with cytoplasmic scaffolding proteins. Although both mechanisms have been implicated in LTP and LTD, many questions persist. Mice lacking the phosphorylation sites on GluR1 that are involved in synaptic plasticity show normal basal synaptic transmission38. This suggests that the activity-dependent trafficking of AMPARs differs from basal receptor delivery. Mice lacking both GluR2 and GluR3 have normal LTP and LTD, but effects on baseline transmission are less clear11. That is, the input/output curve for field EPSPs is reduced, but the amplitudes of AMPAR-mediated mEPSCs are normal11. Further studies are required to reconcile discrepancies between the mouse knockout studies and the large body of literature showing that subunit-specific AMPAR protein interactions with previously identified cytoplasmic proteins are essential in hippocampal receptor trafficking. The present results demonstrate a role for g-8 in maintaining normal levels and distribution of AMPAR protein in hippocampal neurons. The fact that g-8 overexpression greatly increases the number of surface extrasynaptic receptors, independent of total AMPAR protein levels, suggests that g-8 has an additional role in AMPAR trafficking after receptor maturation that is distinct from maintaining the levels of AMPAR protein. However, we cannot exclude the possibility that the potential stabilization of AMPAR subunits by g-8 may increase the probability of their membrane insertion. A general theme derived from studies on voltage-sensitive ion channels indicates that non-poreforming auxiliary subunits are important for controlling the formation, stabilization, trafficking and gating of the pore forming subunit39,40. g-8, as well as g-2, are the first proteins found to have an analogous role for any ionotropic receptor41. This raises the intriguing possibility that auxiliary proteins may control the expression of other ionotropic receptors. Alternatively, because of the central role AMPAR trafficking plays in plasticity, TARPs may have specifically evolved to assist in the dynamic behavior of this class of receptor. It will be of interest in future studies to determine the trafficking mechanism responsible for

1532

controlling the remaining AMPARs in the g-8/ mice and to determine how TARP-dependent trafficking and the various other proposed trafficking mechanisms work in concert to govern the number of AMPARs present at the synapse. METHODS Antibodies. The rabbit polyclonal antibody to g-8 has been characterized previously16. A list of the commercial antibodies used in this study is available in the Supplementary Methods. Knockout mice. Experimental procedures were in accordance with the animal welfare guidelines of the University of California, San Francisco. To generate knockout mice, embryonic stem cells were injected into mouse blastocytes, and the embryos were implanted into surrogate mothers. Chimeric mice were identified by coat color and mated to C57/B16 mice. Further details on how the knockout mice were made are available in Supplementary Methods. Northern and Southern analysis. Hippocampal RNAs from mice were purified using TRIZOL (Invitrogen). Equal amounts of total RNA were separated on a 1% agarose, 1.3% paraformaldehyde gel followed by transfer to nylon membranes and ultraviolet crosslinking. Further details are available in the Supplementary Methods. Immunoblotting. Hippocampi from mouse siblings (P35–P45) were homogenized in 320 mM sucrose buffer and then sonicated in 2% final SDS. Equal amounts of protein were separated on 8% polyacrylamide gels followed by transfer to PVDF membranes. Proteins were detected by immunoblotting using the HRP-ECL kit from Amersham. Further details are available in the Supplementary Methods. Immunohistochemistry. Anesthetized mice were transcardially perfused with phosphate-buffered saline (PBS) followed by 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4) for 5 minutes. Further details on the preparation of brain sections are available in the Supplementary Methods. Postembedding immunogold labeling. For postembedding immunogold labeling, parasagittal sections of the hippocampus were cryoprotected, frozen, infiltrated with Lowicryl HM 20 and polymerized with ultraviolet light. Thin sections were immunolabelled using 10 nm immunogold (Ted Pella). Details on how the data were scored are available in the Supplementary Methods. Immunofluorescent assay of GluR2 protein. Staining for GluR2 was performed on rat hippocampal dissociated neuronal cultures 2–3 weeks after plating. Further details of the staining of surface and intracellular GluR2 is provided in the Supplementary Methods. Electrophysiology and viral infection. Rat hippocampal slice cultures17 and acute mouse hippocampal slices42 were prepared and recorded from as described previously. Further details are available in the Supplementary Methods. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank Y.-X. Wang for help in the immunogold studies, P. Seeburg for the GluRA/ (GluR1/) mice and A. Tzingounis for discussion and reading the paper. This research was supported by grants (to D.S.B. and R.A.N) from the National Institutes of Health (NIH), the Howard Hughes Medical Institute Research Resources Program (to D.S.B.) and the Human Frontier Research Program (to D.S.B.). R.S.P. is supported by the Intramural Research Program of the NIH/NIDCD (National Institute on Deafness and Other Communication Disorders). R.A.N. is a member of the Keck Center for Integrative Neuroscience and the Silvo Conte Center for Neuroscience Research. D.S.B. is an established investigator for the American Heart Association. N.R. is supported by a fellowship from the International Human Frontier Science Program Organization. K.B. and S.T. are supported by postdoctoral fellowships from NIH. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.

VOLUME 8

[

NUMBER 11

[


ARTICLES


Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. Malinow, R. & Malenka, R.C. AMPA receptor trafficking and synaptic plasticity. Annu. Rev. Neurosci. 25, 103–126 (2002). 2. Song, I. & Huganir, R.L. Regulation of AMPA receptors during synaptic plasticity. Trends Neurosci. 25, 578–588 (2002). 3. Bredt, D.S. & Nicoll, R.A. AMPA receptor trafficking at excitatory synapses. Neuron 40, 361–379 (2003). 4. Dingledine, R., Borges, K., Bowie, D. & Traynelis, S.F. The glutamate receptor ion channels. Pharmacol. Rev. 51, 7–61 (1999). 5. Hollmann, M. & Heinemann, S. Cloned glutamate receptors. Annu. Rev. Neurosci. 17, 31–108 (1994). 6. Wisden, W. & Seeburg, P.H. Mammalian ionotropic glutamate receptors. Curr. Opin. Neurobiol. 3, 291–298 (1993). 7. Barry, M.F. & Ziff, E.B. Receptor trafficking and the plasticity of excitatory synapses. Curr. Opin. Neurobiol. 12, 279–286 (2002). 8. Garner, C.C., Nash, J. & Huganir, R.L. PDZ domains in synapse assembly and signalling. Trends Cell Biol. 10, 274–280 (2000). 9. Scannevin, R.H. & Huganir, R.L. Postsynaptic organization and regulation of excitatory synapses. Nat. Rev. Neurosci. 1, 133–141 (2000). 10. Sheng, M. Molecular organization of the postsynaptic specialization. Proc. Natl. Acad. Sci. USA 98, 7058–7061 (2001). 11. Meng, Y., Zhang, Y. & Jia, Z. Synaptic transmission and plasticity in the absence of AMPA glutamate receptor GluR2 and GluR3. Neuron 39, 163–176 (2003). 12. Letts, V.A. et al. The mouse stargazer gene encodes a neuronal Ca2+-channel gamma subunit. Nat. Genet. 19, 340–347 (1998). 13. Chen, L., Bao, S., Qiao, X. & Thompson, R.F. Impaired cerebellar synapse maturation in waggler, a mutant mouse with a disrupted neuronal calcium channel gamma subunit. Proc. Natl. Acad. Sci. USA 96, 12132–12137 (1999). 14. Hashimoto, K. et al. Impairment of AMPA receptor function in cerebellar granule cells of ataxic mutant mouse stargazer. J. Neurosci. 19, 6027–6036 (1999). 15. Chen, L. et al. Stargazin regulates synaptic targeting of AMPA receptors by two distinct mechanisms. Nature 408, 936–943 (2000). 16. Tomita, S. et al. Functional studies and distribution define a family of transmembrane AMPA receptor regulatory proteins. J. Cell Biol. 161, 805–816 (2003). 17. Schnell, E. et al. Direct interactions between PSD-95 and stargazin control synaptic AMPA receptor number. Proc. Natl. Acad. Sci. USA 99, 13902–13907 (2002). 18. Klugbauer, N. et al. A family of gamma-like calcium channel subunits. FEBS Lett. 470, 189–197 (2000). 19. Burgess, D.L., Gefrides, L.A., Foreman, P.J. & Noebels, J.L. A cluster of three novel Ca2+ channel gamma subunit genes on chromosome 19q13.4: evolution and expression profile of the gamma subunit gene family. Genomics 71, 339–350 (2001). 20. Chu, P.J., Robertson, H.M. & Best, P.M. Calcium channel gamma subunits provide insights into the evolution of this gene family. Gene 280, 37–48 (2001). 21. Hurtley, S.M. & Helenius, A. Protein oligomerization in the endoplasmic reticulum. Annu. Rev. Cell Biol. 5, 277–307 (1989).


[

NUMBER 11

[

NOVEMBER 2005

22. Hollmann, M., Maron, C. & Heinemann, S. N-glycosylation site tagging suggests a three transmembrane domain topology for the glutamate receptor GluR1. Neuron 13, 1331–1343 (1994). 23. Sans, N. et al. Synapse-associated protein 97 selectively associates with a subset of AMPA receptors early in their biosynthetic pathway. J. Neurosci. 21, 7506–7516 (2001). 24. Zamanillo, D. et al. Importance of AMPA receptors for hippocampal synaptic plasticity but not for spatial learning. Science 284, 1805–1811 (1999). 25. Castillo, P.E., Malenka, R.C. & Nicoll, R.A. Kainate receptors mediate a slow postsynaptic current in hippocampal CA3 neurons. Nature 388, 182–186 (1997). 26. Vignes, M. & Collingridge, G.L. The synaptic activation of kainate receptors. Nature 388, 179–182 (1997). 27. Nicoll, R.A. & Malenka, R.C. Contrasting properties of two forms of long-term potentiation in the hippocampus. Nature 377, 115–118 (1995). 28. Henze, D.A., Urban, N.N. & Barrionuevo, G. The multifarious hippocampal mossy fiber pathway: a review. Neuroscience 98, 407–427 (2000). 29. Vandenberghe, W., Nicoll, R.A. & Bredt, D.S. Interaction with the unfolded protein response reveals a role for stargazin in biosynthetic AMPA receptor transport. J. Neurosci. 25, 1095–1102 (2005). 30. Ellgaard, L. & Helenius, A. Quality control in the endoplasmic reticulum. Nat. Rev. Mol. Cell Biol. 4, 181–191 (2003). 31. Zerangue, N., Schwappach, B., Jan, Y.N. & Jan, L.Y. A new ER trafficking signal regulates the subunit stoichiometry of plasma membrane K(ATP) channels. Neuron 22, 537–548 (1999). 32. Jensen, V. et al. A juvenile form of postsynaptic hippocampal long-term potentiation in mice deficient for the AMPA receptor subunit GluR-A. J. Physiol. (Lond.) 553, 843–856 (2003). 33. Chien, A.J. et al. Roles of a membrane-localized beta subunit in the formation and targeting of functional L-type Ca2+ channels. J. Biol. Chem. 270, 30036–30044 (1995). 34. Brice, N.L. et al. Importance of the different beta subunits in the membrane expression of the alpha1A and alpha2 calcium channel subunits: studies using a depolarizationsensitive alpha1A antibody. Eur. J. Neurosci. 9, 749–759 (1997). 35. Bogdanov, Y. et al. Acidic motif responsible for plasma membrane association of the voltage-dependent calcium channel beta1b subunit. Eur. J. Neurosci. 12, 894–902 (2000). 36. Andrasfalvy, B.K., Smith, M.A., Borchardt, T., Sprengel, R. & Magee, J.C. Impaired regulation of synaptic strength in hippocampal neurons from GluR1-deficient mice. J. Physiol. (Lond.) 552, 35–45 (2003). 37. Tomita, S., Stein, V., Stocker, T.J., Nicoll, R.A. & Bredt, D.S. Bidirectional synaptic plasticity regulated by phosphorylation of stargazin-like TARPs. Neuron 45, 269–277 (2005). 38. Lee, H.K. et al. Phosphorylation of the AMPA receptor GluR1 subunit is required for synaptic plasticity and retention of spatial memory. Cell 112, 631–643 (2003). 39. Catterall, W.A. Structure and regulation of voltage-gated Ca2+ channels. Annu. Rev. Cell Dev. Biol. 16, 521–555 (2000). 40. Arikkath, J. & Campbell, K.P. Auxiliary subunits: essential components of the voltagegated calcium channel complex. Curr. Opin. Neurobiol. 13, 298–307 (2003). 41. Tomita, S. et al. Stargazin modulates AMPA receptor gating and trafficking by distinct domains. Nature 435, 1052–1058 (2005). 42. Fremeau, R.T. et al. Vesicular glutamate transporters 1 and 2 target to functionally distinct synaptic release sites. Science 304, 1815–1819 (2004).

1533


ARTICLES

Cbln1 is essential for synaptic integrity and plasticity in the cerebellum Hirokazu Hirai1,3,4, Zhen Pang1,3,4, Dashi Bao1, Taisuke Miyazaki2, Leyi Li1, Eriko Miura2, Jennifer Parris1, Yongqi Rong1, Masahiko Watanabe2, Michisuke Yuzaki1,3 & James I Morgan1 Cbln1 is a cerebellum-specific protein of previously unknown function that is structurally related to the C1q and tumor necrosis factor families of proteins. We show that Cbln1 is a glycoprotein secreted from cerebellar granule cells that is essential for three processes in cerebellar Purkinje cells: the matching and maintenance of pre- and postsynaptic elements at parallel fiber–Purkinje cell synapses, the establishment of the proper pattern of climbing fiber–Purkinje cell innervation, and induction of long-term depression at parallel fiber–Purkinje cell synapses. Notably, the phenotype of cbln1-null mice mimics loss-of-function mutations in the orphan glutamate receptor, GluRd2, a gene selectively expressed in Purkinje neurons. Therefore, Cbln1 secreted from presynaptic granule cells may be a component of a transneuronal signaling pathway that controls synaptic structure and plasticity.

The identification of molecules that influence synapse formation and function has importance for our understanding of the mechanisms contributing to both neurodevelopment and information storage and processing in the nervous system. Moreover, such molecules could potentially contribute to the pathogenesis of neurological and neuropsychiatric disorders and might represent previously unknown therapeutic targets or entities. Here we elucidate the properties of Cbln1, a secreted protein that is essential for synapse structure and function in the cerebellum. Cbln1 is the prototype for a family of four brain-specific proteins (Cbln1–Cbln4) of unknown function that was first identified by virtue of its harboring a naturally occurring 16-amino-acid peptide, cerebellin1. Both cbln1 and cbln3 mRNAs are selectively expressed in the cerebellum, and their levels increase in parallel with synaptogenesis2. As the cerebellin peptide is enriched in synaptosomes3 and has been reported to undergo release upon depolarization4, it was initially suggested to be a neuromodulator. However, Cbln1 lacks the typical structural features of a neuropeptide precursor; rather, it belongs to the C1q and tumor necrosis factor (TNF) family4. The members of this family are frequently secreted as trimers that serve diverse roles in intercellular communication. Indeed, TNFa has been implicated in neuronal plasticity5. We show here that Cbln1 is secreted from cerebellar granule cells as a glycoprotein that controls synaptic plasticity and synapse integrity of Purkinje cells. RESULTS Although cbln1 mRNA is enriched in adult cerebellum1, the source of Cbln1 is unknown. In situ hybridization showed that, like cbln3 (ref. 2),

cbln1 mRNA was localized to the internal granule cell layer (Fig. 1a,b). Owing to the limited resolution of in situ hybridization, we could not conclusively exclude some expression of cbln1 mRNA within Purkinje cells, which is notable for three reasons: the cerebellin peptide is enriched in synaptosomes3, anti-cerebellin antibodies react with Purkinje cell dendritic spines6 and the deficits described below in cbln1-null mice involve Purkinje cells. Therefore, we generated transgenic mice in which b-galactosidase was driven from the mouse cbln1 promoter. Analysis of cbln1-lacZ mice demonstrated expression of the reporter in granule cells but not Purkinje cells (Fig. 1c,d). Using in situ hybridization and cbln1-lacZ mice, we did not observe cbln1 expression in the inferior olive, which sends climbing fiber (CF) inputs to Purkinje cells (Supplementary Fig. 1). Thus, Cbln1 may be secreted from granule cells and interact with postsynaptic structures of Purkinje cells. To establish whether Cbln1 was secreted, we performed immunoblotting. Immunoblotting of culture medium of cerebellar granule cells identified a protein of approximately 35 kDa, larger than the predicted molecular weight for Cbln1 (B20 kDa; Fig. 1e). An immunoreactive protein of equivalent mass was also detected in medium from HEK 293 cells expressing recombinant Cbln1 (Fig. 1f) but not empty vector (data not shown). Moreover, an immunoreactive band with the same pattern of migration as secreted Cbln1 was detected in cerebellar extracts (Fig. 1g) indicating that it also exists in vivo. Treatment of secreted recombinant Cbln1 with glycosidases converted the 35 kDa band into two products of lesser mass (Fig. 1f), suggesting glycosylation. Furthermore, treatment of cerebellar lysates with glycosidases eliminated the 35 kDa band and generated bands

1Department

of Developmental Neurobiology, St. Jude Children’s Research Hospital, 332 North Lauderdale Street, MS 323, Memphis, Tennessee 38105-2794, USA. of Anatomy, Hokkaido University School of Medicine, Sapporo 060-8638, Japan. 3Current addresses: Advanced Science Research Center, Kanazawa University, PRESTO, Japan Science and Technology Agency, Kanazawa 920-8640, Japan (H.H.); Roche Palo Alto, 3431 Hillview Ave., R2-101 Palo Alto, California 943041320 USA (Z.P.) and Department of Physiology, Keio University School of Medicine, Tokyo 160-8582, Japan (M.Y.) 4These authors contributed equally to this work. Correspondence should be addressed to M.Y. ([email protected]) or J.M. ([email protected]). 2Department

Received 13 July 2005; accepted 21 September 2050; published online 23 October 2005; doi:10.1038/nn1576

1534

VOLUME 8

[

NUMBER 11

[


a

b

c

d

e

Cerebellar culture medium

f

g Cerebellar

293 cell medium Cbln1

Glycosidases –

Figure 1 Cbln1 is a glycoprotein secreted from cerebellar granule cells. (a,b) In situ hybridization of cbln1 in mouse cerebellum. Note high expression of cbln1 within the internal granule cell layer and lack of overt hybridization to Purkinje cells. (c,d) b-galactosidase staining of granule cells but not Purkinje neurons (arrows in d) in cerebellum of a cbln1-lacZ transgenic mouse. Scale bars: a, 1 mm; b, 150 mm; c, 250 mm; d, 20 mm. (e) Immunoblotting of medium from primary cerebellar cultures indicates that Cbln1 undergoes post-translational modification and is secreted. (f) Treatment of recombinant secreted Cbln1 with glycosidases (+) yields proteins of lower molecular masses, indicating glycosylation. The Cbln1(N23,79Q) mutant, in which both asparagine residues predicted to undergo N-linked glycosylation are changed to glutamine residues, has a molecular mass equivalent to that of Cbln1 after digestion with glycosidases. (g) Immunoblotting of a mouse cerebellar extract shows a protein with the same mobility as secreted Cbln1. Treatment of the extract with glycosidases (+) produces proteins of lower molecular weights, consistent with those observed for secreted Cbln1.

lysates

N23,79Q

+

–

+

–

kDa 36 28 20 6

with masses equivalent to those generated from secreted recombinant proteins (Fig. 1g). To confirm glycosylation, we mutated two asparagines (Asn23 and Asn79) that are predicted sites of N-linked glycosylation to glutamine residues. Cbln1(N23,79Q) was still secreted, but its molecular mass was reduced to that of the smaller of the two bands detected after glycosidase treatment (Fig. 1f). This suggests that Cbln1 is glycosylated at these asparagine residues and that the intermediate band of Cbln1-like immunoreactivity observed after glycosidase treatment represents a partial digestion product (Fig. 1f). Immunoblot analysis of adult cerebellar lysates detected not only a protein of equivalent mass to secreted Cbln1 but also bands with molecular masses smaller than the mature protein (Fig. 1g). Whereas the 35 kDa band was sensitive to glycosidase treatment, many of the

Figure 2 Generation of cbln1 null mice. (a) Gene targeting strategy. The organization of the wildtype28 and recombined cbln1 alleles are depicted in relation to the targeting vector. Boxes indicate the three exons of cbln1. Correctly targeted clones were identified by digestion with XbaI and ApaI and Southern blotting with probes a (external) and b (internal). Genotyping was performed by PCR. Primers were cbln1KO-5¢ (GTGGAGCTGCTGCTGT TGGGGACTG), corresponding to nucleotides 349-373 of cbln1 (AF164680) and cbln1KO-3¢ (CCGCCTTCGAAGCTCTCCCTGTC) that is antisense to nucleotides 650-673 of cbln1, as well as a neospecific primer, cbln1-neo (GAATGGGCTGACCGCT TCCTCGTG). The wild-type and targeted alleles gave rise to 0.31-kb and 0.68-kb fragments, respectively. Primer locations indicated by short horizontal arrows. A, ApaI; B, BamHI; E, EcoRI; H, HindIII; N, NotI; X, XbaI. Mice lacking both cbln1 alleles show loss of cbln1 mRNA on Northern blots (b), an absence of cerebellin peptide by radioimmunoassay (c) and loss of Cbln1 protein on immunoblots of whole cerebellar extracts (d). Mice heterozygous for cbln1 have intermediate levels of mRNA (b) and cerebellin peptide (c).


[

lower molecular weight species were not (Fig. 1g). Some of these proteins could represent Cbln1 in the initial stages of trafficking through the secretion pathway; however, others are smaller than the 20 kDa precursor and must be degradation products. These smaller species may be produced during the process that liberates the cerebellin peptide from Cbln1. To establish its function, cbln1 was eliminated from the genome of mice (Fig. 2a). Cerebella from homozygous cbln1-null mice showed a loss of cbln1 mRNA (Fig. 2b) and cerebellin peptide (Fig. 2c) and protein (Fig. 2d). Heterozygous mice had intermediate levels of mRNA (Fig. 2b) and peptide (Fig. 2c). These data also establish that Cbln1 is the source of cerebellin. cbln1–/– mice were ataxic, having markedly impaired performance on the accelerating rotarod (Fig. 3a, Supplementary Fig. 2) and a wide waddling gait (Fig. 3b). Impaired coordination was detected at postnatal day 18 (P18) and persisted throughout life, but unlike many strains of ataxic mice, the cbln1-null animals had no marked tremor. Otherwise, the mice were fertile, with no overt anatomical abnormalities and normal life spans. The cerebella of cbln1-null mice had normal foliation and laminated cortical structure (Fig. 4a,b). All principal neuronal types were present in the cerebella of cbln1-null mice

N

HA

a

X neo

Targeting vector

tk

X E

N

HA

A A

B

Wild-type allele 8.9 kb HA

N

X

E A A b 3.9 kb

b

NUMBER 11

B

neo

Disrupted allele

c

Radioimmunoassay of cerebellin peptide levels

a

d

Anti-Cbln1 +/+

+/+ +/– –/–

28 S cbln1

18 S

Actin

[

NOVEMBER 2005

Cerebellin content (pmol/mg tissue ± s.e.m. (n))


ARTICLES

0.5

cbln1

n=8

cbln1–/– 120

0.4

kDa

92 55

0.3

36 0.2 29

n=4 0.1 0

cbln1+/+

cbln1+/–

n=8

20

cbln1–/–

6

1535

ARTICLES

a

cbln1 +/+

b

+/+

cbln1 –/–

+/– 7

Rotarod scores

L5

R3

n=7

6


R6

–/–

R5

L3 L4

5 n=8 4

R4

R2

3

R3a

L3 L2

n=6

2

L2

1 R1 0 0

1

2

3 4 5 Time (d)

6

7

8

L1

R3 R2

L1 R1

Figure 3 Ataxic phenotype of cbln1-null mice. (a) Comparison of the performance of cbln1+/+ (circles), cbln1+/– (triangles) and cbln1–/– (squares) mice from the same litters and matched genders on a standardized accelerating rotarod test over a course of four trials spanning 7 d (n ¼ number of individual mice in each cohort). Whereas cbln1-null mice never reached criterion (score of 6) and never improved their scores, mice of the other two genotypes reached criterion by the second trial. Error bars represent s.e.m. (b) Representative gait analyses of 2-month-old littermate cbln1+/+ and cbln1–/– mice ascending an inclined plane. Front paws are marked with red and hind paws with blue nontoxic water-soluble paint. L: left paw; R: right paw; numbers indicate step count. Note irregular, shortened gait and skips in cbln1-null mice.

(Fig. 4c,d), and Purkinje cells, the sole output neuron of the cerebellar cortex, appeared grossly normal (Fig. 4e–h). However, as in many strains of ataxic mice, sporadic pyknotic nuclei were observed within the internal granule cell layer of adult cbln1-null mice after the onset of locomotor deficits. At 1 month of age, the cerebellum and granule cell layers of cbln1-null mice showed a trend towards reduced volume, although this did not reach statistical significance (Supplementary Table 1). As a result of the slightly reduced thickness of the molecular layer, longitudinal outgrowth of Purkinje cell dendrites was decreased, but dendritic spines in the mutant mouse were aligned along the tertiary dendrites as densely as those in wild-type mice (Fig. 4g,h). Immunostaining for aldolase C/zebrin II showed characteristic bands of Purkinje cells7 that were indistinguishable in cbln1-null mice and wild-type littermates (data not shown), indicating no major disruption of cerebellar regionalization. As there were no neuroanatomical anomalies that could account for the ataxic behavior, we undertook an electrophysiological analysis to evaluate synaptic interactions and plasticity in the cerebella of cbln1null mice. Purkinje cells receive two excitatory inputs, parallel fibers (PFs) from granule cells and CFs from the inferior olivary nuclei. We Figure 4 Neuroanatomical analysis of cbln1–/– mice. Cresyl violet–stained sagittal sections of 1-month-old wild-type (a,c) and cbln1–/– (b,d) littermates show no overt neuroanatomical anomalies other than occasional pyknotic cells in the internal granule cell layer of the knockout mice. The volumes of the cerebellum and internal granular cell layer showed a slight reduction in the cbln1–/– mice, but this did not attain statistical significance (Supplementary Table 1). Calbindin 28K immunostaining of wild-type (e) and cbln1–/– (f) Purkinje cells revealed no overt differences in the positioning, number and overall dendritic morphology between the two genotypes. Dendritic spines in cbln1–/– mice (h, box in f) were aligned along the tertiary dendrites as densely as those in wild-type mice (g, box in e). As with the granule cell layer, there was a slight reduction in the thickness of the molecular layer in cbln1–/– mice. Scale bars: a, 500 mm; e, 20 mm; g, 5 mm.

1536

examined excitatory postsynaptic currents (EPSCs) in response to PF or CF stimulation in P21–P30 and P60–P68 slice preparations by whole-cell patch-clamping (Fig. 5a). Passive membrane properties of Purkinje and granule cells were normal (Supplementary Table 2). However, PF-EPSC amplitudes in cbln1–/– Purkinje cells were consistently smaller than those in cbln1+/+ Purkinje cells (Fig. 5b,c), indicating impaired PF–Purkinje cell synaptic function. We examined the pattern of PF innervation using double immunohistochemical labeling of the vesicular glutamate transporter 1 (VGluT1), a marker of PF terminals8, and calbindin, a marker of Purkinje cells. As in wild-type mice, VGluT1-positive PF terminals were distributed all along Cbln1–/– Purkinje cell dendrites (Fig. 5d,e). However, an electron microscope analysis showed that the number of PF–Purkinje cell synapses was markedly reduced in cbln1–/– cerebella (Fig. 6a,b). Purkinje cells in cbln1–/– cerebella had unique ‘naked’ spines that contained postsynaptic density (PSD)–like condensations but lacked presynaptic contact (Fig. 6a,b). To confirm that the naked spines were not artifacts of the plane of section, we examined 80-nm serial sections (Fig. 6c–f). Approximately 50% of Purkinje cell spines on proximal dendrites (42 mm in caliber) and 80% of spines on distal dendrites (typical spiny branchlets) in cbln1–/– cerebella were free of synaptic contacts, whereas we did not observe any naked spines in cbln1+/+ cerebella (Fig. 6i). However, total spine density on Purkinje cells per 100 mm2 did not differ significantly between strains (cbln1–/–, 45 ± 8; cbln1+/+, 34 ± 4, n ¼ 3 mice for each genotype, P ¼ 0.29). Although the spine density might be overestimated in cbln1–/– cerebellum because of its slightly reduced surface area, this establishes that Purkinje cell dendritic spines are formed in cbln1–/– mice but that synaptic contact with PFs is severely impaired. The remaining PF–Purkinje cell synapses in cbln1–/– mice showed another specific abnormality: the PSDs were frequently longer than the active zone. When we counted the number of synapses in which the edges of the active zone and the PSD were more than 100 nm apart, 65 ± 9% (n ¼ 3 mice) of PF synapses were mismatched in cbln1–/– Purkinje cells, whereas none were mismatched in cbln1+/+ Purkinje

cbln1+/+

cbln1–/–

a

b

c

d

e

f

g

VOLUME 8

[

NUMBER 11

h

[


ARTICLES

CF


GC

d

cbln1

600 400 200 0 0

100 200 Stimulus intensity (µA)

e

+/+

Calb

cbln1 VGluT1

cbln1

–/–

Overlay

Calb

800 600 400 200 0

0

Figure 6 Appearance of free spines and abnormal PF–Purkinje cell synapses in cbln1–/– mice. (a,b) Electron micrographs of the molecular layer in wild-type (a) and cbln1–/– (b) cerebella at 3 months of age. Purkinje cell spines are marked by asterisks. In the cbln1–/– cerebellum, numerous Purkinje cell spines that possess PSD-like condensation are evident, but they lack synaptic contacts with PF terminals. (c–f) Serial electron micrographs of a representative cbln1–/– cerebellum showing a Purkinje cell spine (labeled by an asterisk) that is completely free of presynaptic contact. (g,h) The length of the PSD of cbln1–/– (h) Purkinje cells is greater than that of wild-type cells (g). Arrowheads indicate the outer edge of the PSD. (i,j) Quantitative analysis of the mean percentage of ‘naked’ spines lacking synaptic contacts (i) and the mean number of asymmetrical synapses per 100 mm2 of the molecular layer (j). The s.e.m. is indicated; n ¼ 3 mice for each genotype (i,j). Scale bars, 500 nm (a) and 200 nm (f,g).

[

NUMBER 11

[

100 Stimulus intensity (µA)

cbln1 –/– VGluT1

cells. We next measured the largest PSD length of given PF–Purkinje cell synapses from serial profiles of individual synapses. The mean maximal length of PSDs in cbln1–/– Purkinje cells was 1.4 times larger than that in wild-type Purkinje cells (Fig. 6g,h,j). This could reflect a compensatory response to the reduced number of PF-Purkinje cell synapses in cbln1–/– cerebella. Indeed, compared with the severe reduction in the number of PF-Purkinje cell synapses (Fig. 6i), amplitudes of PF-EPSCs were less affected in cbln1–/– cerebella (Fig. 5b–c). There were no obvious abnormalities in the distribution and density of synaptic vesicles in the presynaptic terminals (Supplementary Fig. 3). Similarly, PF-EPSCs demonstrated paired-pulse facilitation, which reflects presynaptic functions, at interpulse intervals between 10 and 400 ms (data not shown). These results indicate that CFs were not recruited by our stimulation protocol and suggest that Cbln1 has a role in matching the PSD with the presynaptic element at PF–Purkinje cell synapses. Immature Purkinje cells are innervated by multiple CFs9, and redundant CFs are eliminated during development until a one-toone relationship is established by the end of the third postnatal week. As reported previously10,11, single CF-EPSCs are elicited in more than 90% of wild-type Purkinje cells from P21–P30 mice (Fig. 7a,b), indicating establishment of single CF innervation. In contrast, only 53% of cbln1–/– Purkinje cells attained a one-to-one relationship with CFs, a deficit that was still evident at P60 through P68 (Fig. 7c). Thus,


Figure 5 Reduced efficiency of PF–Purkinje cell synaptic transmission in cbln1–/– mice. (a) Purkinje cells (PC) were whole-cell voltage clamped by a patch electrode. Stimulation was given to parallel fibers (PF), axon bundles of granule cells (GC) or climbing fibers (CF). (b,c) Amplitudes of PF-EPSCs in cbln1–/– Purkinje cells. In response to increasing stimulus intensities (20-mA increments), PF-EPSCs were elicited in Purkinje cells from nine wild-type (n ¼ 22 cells, open circles) and six cbln1–/– (n ¼ 18 cells, closed circles) mice between P21 and P30 (b), or from three wild-type (n ¼ 16 cells, open circles) and seven cbln1–/– (n ¼ 15 cells, closed circles) mice between P60 and P68 (c). Insets show representative PF-EPSC traces. Each point represents the mean ± s.e.m. *P o 0.05; **P o 0.01. P values were obtained by one-way ANOVA. (d,e) Reduced numbers of PF terminals on cbln1–/– Purkinje cells. PF terminals in wild-type and cbln1–/– cerebellar sections were visualized with antibodies to VGluT1. Purkinje cells were stained with antibodies to calbindin (Calb). Scale bar: 20 mm.

P60–P68

NOVEMBER 2005

200

Overlay

establishment of single CF innervation is impaired in cbln1–/– Purkinje cells for at least 8 to 10 weeks after birth. Whereas most CF-EPSCs in cbln1–/– Purkinje cells had a rapid rise time, similar to those in wild-type cells, CF-EPSCs with small amplitude in cbln1–/– Purkinje cells were often associated with a markedly slower rise time (Fig. 7a, arrows in right traces; Supplementary Table 3). When stimulus intensity was increased, this slow component was masked by large EPSCs with a rapid rise time. The EPSCs with a slow rise time showed paired-pulse depression at interpulse

a

b * *

* *

c

*

*

d *

g

*

f

*

*

*

j *

*

*

*

*

e

i

h

*

*

80

**

30

***

60 40 20 0

+/+ –/– +/+ –/– Proximal Distal

Mismatched spine (%)

PC

800

+/+

Naked spine (%)

PF

c 1,000

P21–P30

PF-EPSC amplitude (pA)

b PF-EPSC amplitude (pA)

a

* **

20

10

0

+/+ –/– +/+ –/– Proximal Distal

1537

ARTICLES

b

100

Percentage of cells


cbln1–/–

cbln1+/+

c

P21–P30

80

–/– n = 103 60 40 20 0

Calb VGluT2

100

+/+ n = 81

80

cbln1+/+

P60–P68 +/+ n = 32 –/– n = 35

60 40 20 0

1 2 3 4 Number of steps

Figure 7 Multiple CF innervation of Purkinje cells in cbln1–/– mice. (a) Impaired developmental regression of CF innervation in cbln1–/– mice. EPSCs were recorded from P23 Purkinje cells by stimulation of CFs in the granule cell layer with membrane potentials held at –10 mV. EPSCs in wild-type Purkinje cells appeared as all-or-none traces (left), whereas EPSCs in cbln1–/– cells appeared as two (center) or three (right) discrete steps, indicating innervation by two or three CFs, respectively. The smaller EPSCs in the right traces have slower rise times (arrowheads). Scale bars, 20 ms and 500 pA. (b,c) Histograms of discrete steps of CF-EPSCs in Purkinje cells from wildtype and cbln1–/– mice at P21–P30. (b) and P60–P68 (c). In b, results from 81 Purkinje cells from 17 wild-type mice and 103 Purkinje cells from 17 cbln1–/– mice at P21–P30. In c, 32 Purkinje cells from 3 wild-type mice and 35 Purkinje cells from 4 cbln1–/– mice at P60–P68. (d–i) Increase in the distribution and number of CF terminals in cbln1–/– mice. Sections of wildtype (d,e,f) and cbln1–/– (g,h,i) cerebella were double immunolabeled for VGluT2 (red) and calbindin (green). The pial surface is indicated by the dotted line. CF terminals were mainly associated with shaft dendrites in wildtype cerebella (f), whereas many CF terminals terminated at distal spiny branchlets in cbln1–/– cerebella (i). Scale bars, 20 mm.

cbln1–/–

Percentage of cells

a

VGluT2 Calb VGluT2

1 2 3 4 Number of steps

cbln1–/–

VGluT2

intervals between 10 and 3,000 ms (data not shown), a hallmark of CF-EPSCs. As the EPSC rise time measured at the soma is influenced by the electrotonic distance between the activated synapses and the soma, it is possible that the CF-EPSCs with slow rise times were attributable to abnormal CF innervation of distal dendrites of cbln1–/– Purkinje cells8. Purkinje cell dendrites have two separate domains: proximal regions innervated by CFs and distal regions innervated by PFs. To examine CF innervation of Purkinje cells in cbln1–/– cerebella, sections were immunostained for vesicular glutamate transporter 2 (VGluT2), which is predominantly expressed in CF terminals8. The density of VGluT2immunopositive puncta was larger in cbln1–/– cerebella than in wildtype cerebella (P o 0.0001; Fig. 7d–i). Moreover, CF terminals penetrated 87 ± 1% (n ¼ 3 mice) of the molecular layer thickness in cbln1–/– cerebella (Fig. 7g,h), whereas the depth reached by the most distal terminal of CFs in wild-type cerebella was significantly less (79 ± 3%, n ¼ 3 mice, P ¼ 0.03; Fig. 7d,e). CF terminals were mainly associated with shaft dendrites in wild-type cerebella8 (Fig. 7f), whereas they terminated at both shaft dendrites and distal spiny branchlets in cbln1–/– cerebella (Fig. 7i). These results may account for the slow rise time of CF-EPSCs in cbln1–/– Purkinje cells (Fig. 7a). Therefore, Cbln1 is essential for the establishment or maintenance of PF-Purkinje cell synapses, but its absence also leads to multiple innervation by CFs, some of which invade the ‘parallel fiber domain’ of distal dendrites of Purkinje cells.

Simultaneous activation of PFs and CFs induces long-term depression (LTD), a form of synaptic plasticity thought to underlie motor coordination and information storage12. Therefore, we examined whether LTD could be induced in cbln1–/– mice (P21–P30) when no prominent loss of granule cells was observed. To exclude any effect from abnormal CF innervation, which serves to evoke Ca2+ spikes essential for LTD induction12, we replaced CF stimulation with direct depolarization of Purkinje cells. In this protocol, regenerative Ca2+ spikes in the out-of-clamp dendritic region increase the Ca2+ concentration in Purkinje cells13. There was no significant difference in the number of Ca2+ spikes seen in Purkinje cells of cbln1+/+ (134 ± 21 spikes, n ¼ 8 Purkinje cells) and cbln1–/– mice (154 ± 15 spikes, n ¼ 9 Purkinje cells). In wild-type Purkinje cells, this protocol induced LTD of PF-EPSCs (Fig. 8a,c), whereas it did not induce LTD in cbln1–/– Purkinje cells (Fig. 8b,d). The failure to induce LTD was not attributable to reduced initial PF-EPSC amplitudes in cbln1–/– Purkinje cells because LTD could not be elicited even when the initial PF-EPSC amplitude was similar to that in wild-type Purkinje cells (Fig. 8a,b). A major form of cerebellar LTD is believed to occur solely in postsynaptic Purkinje cells, probably by increased endocytosis of postsynaptic AMPA receptors14,15. Therefore, Cbln1 is likely to have an essential role in postsynaptic signaling pathways involved in synaptic plasticity at PF–Purkinje cell synapses. As decreased inhibitory synapses in the cerebellum can contribute to cerebellar dysfunction and ataxia16 we examined the pattern of inhibitory innervation by immunohistochemical labeling of the vesicular GABA transporter (VGAT), a marker of glycine- and GABA-containing terminals. Soma and dendrites of cbln1–/– Purkinje cells were surrounded by VGAT-positive terminals in a similar manner to wild-type cells (Supplementary Fig. 4). Total VGAT-positive puncta density per 100 mm2 in the molecular layer was slightly increased in cbln1–/– cerebella (cbln1+/+, 14 ± 3; cbln1–/–, 22 ± 6, n ¼ 3 mice for each genotype, P ¼ 0.30), probably reflecting a slight reduction in molecular layer area. The ratio of Purkinje cells forming pinceau (that is, clustered basket cell axons) around the initial segment of Purkinje cell axons in cbln1–/– cerebella was similar to that in wild-type cerebella (Supplementary Fig. 4). Finally, there was no difference between wild-type and cbln1–/– mice in the number of puncta double-immunoreactive to both VGAT and calbindin on neurons of the deep cerebellar nuclei (P ¼ 0.67), which represent Purkinje cell inputs (Supplementary Fig. 4). Thus, there were no obvious abnormalities in the inhibitory circuits in cbln1–/– cerebella other than a moderate increase of VGAT-positive puncta in the molecular layer.

1538

VOLUME 8

d

f

e

g

h

i

[

NUMBER 11

[


ARTICLES accompanied by moderately reduced cerebellar size22,24. In addition, both show innerva1. Before 2. After tion by multiple CFs23,24 invading the distal 100 pA 100 pA dendritic territory of Purkinje cells8, and both 20 ms 20 ms show impaired LTD21. Notably, when expres200 200 sion of GluRd2 increases in the proximal 150 150 dendrites after decreased neuronal activity, 100 100 the PF territory reciprocally expands into 50 50 proximal dendrites26. GluRd2-null mice also 0 0 –10 0 10 20 30 40 –10 0 10 20 30 40 share the more unsual aspects of the cbln1-null Time (min) Time (min) phenotype. Thus, in adult GluRd2-null cerec cbln1+/+ d cbln1–/– bellum, approximately 40% of spines on distal 120 120 n = 10 dendrites of Purkinje cells are naked, and they 100 100 do not show signs of PF sprouting22. In n = 9 80 80 addition, in the remaining PF synapses, a 60 60 mismatch is frequently observed between the 40 40 –10 0 10 20 30 40 –10 0 10 20 30 40 lengths of the PSD and apposing presynaptic Time (min) Time (min) active zone. These results suggest that GluRd2, –/– which is in the PF–Purkinje cell postsynapses, Figure 8 LTD is deficient in Purkinje cells of cbln1 mice. (a,b) After obtaining a stable baseline and Cbln1, which is expressed in granule cell response for 15 min, we applied an LTD-inducing stimulus at time 0. Representative results of the time course of EPSC amplitude of nine Purkinje cells from seven wild-type (a) and ten Purkinje cells from presynapses, engage in a common signaling six cbln1–/– (b) mice and the averaged (± s.e.m.) results for each (wild-type in c, cbln1–/– in d) are pathway or process critical for synapse formashown. The amplitudes of PF-EPSCs were normalized to baseline values, which were the average tion, maintenance and plasticity. responses over the 5 min just before conjunctive stimulation. Inset traces are EPSCs just before (1) and Despite these striking similarities in pheno40 min after (2) the conjunctive stimulation. types, there are differences between Cbln1-null and GluRd2-null cerebella. For example, naked spines on distal dendrites of Purkinje cells are more frequently DISCUSSION Cbln1 is a glycoprotein secreted from granule cells that is essential for observed in cbln1-null mice (78 ± 3%; Fig. 6i) than in GluRd2-null22 three processes in Purkinje cells (Supplementary Fig. 5): the matching (37 ± 5%) mice. In contrast, CF terminals reach more distal regions of of pre- and post-synaptic elements at PF–Purkinje cell synapses, the Purkinje cell dendrites in GluRd2-null mice than they do in cbln1-null establishment of proper CF–Purkinje cell innervation patterns and the mice; CF terminals penetrate 95 ± 0.4% of the molecular layer thickness induction of LTD at PF–Purkinje cell synapses. Although many ataxic in GluRd2-null cerebella8, whereas they penetrate 87 ± 1% of the way in strains of mice show impairment of LTD, Cbln1 is unique in that it is cbln1–/– cerebella (Fig. 7). Notably, mice that lack both GluRd2 and produced in the presynaptic neuron, and its absence does not result in Cbln1 do not show an additive phenotype but rather are similar to mice any gross morphological abnormalities in the cerebellum. Together lacking only GluRd2 (Supplementary Fig. 6 and data not shown). This with the finding of mismatched PSDs at PF–Purkinje cell synapses in genetic interaction implies that the two gene products function in a cbln1–/– Purkinje cells, this suggests that Cbln1 participates in a common mechanism or pathway rather than acting entirely independently. It is unlikely that Cbln1 and GluRd2 physically associate with transneuronal signaling pathway. Whereas mismatching of presynaptic and postsynaptic structures is a one another, but their signaling pathways may converge on some rare phenotype, ‘naked’ spines are observed in a number of settings. For common downstream mechanism. As other members of the Cbln example, they occur when PFs are reduced by irradiation, and they are family and GluRd1 (a relative of GluRd2)27 are expressed in other brain present in nodding17 and tottering18 mice. However, in these animals the regions, this transneuronal process may represent a previously remaining PFs undergo sprouting, probably as a compensatory unknown, more widespread mechanism controlling synaptic structure response, and the number of PF synaptic contacts increases over time. and neuronal function. Therefore, further studies are warranted to Such compensatory responses do not seem to occur in cbln1-null mice, investigate their signaling pathways and whether these contribute to as even in adults, 80% of Purkinje cell spines on distal dendrites are human diseases of the nervous system. naked. Although it is unclear whether Cbln1 is essential for PF–Purkinje cell synapse formation or maintenance or both, these findings suggest METHODS that there may at least be an ongoing requirement for its presence. Cbln1-lacZ transgenic mice. A bacteriophage P1–based mouse genomic DNA Sustained innervation of single Purkinje cells by multiple CFs has library was screened by PCR (Genome Systems) using cbln1-specific primers. been observed in animals that have genetic mutations affecting synaptic An 11.57-kb HindIII-BamHI fragment containing the entire cbln1 gene was transmission11,19 or reduced granule cell numbers20. However, in these isolated and subcloned into pGEM 11-Zf (Promega). A unique Sal1 restriction mice, surplus CFs do not normally innervate distal dendrites. There- site was created at the third codon of cbln1 permitting the in-frame insertion fore, the invasion of CFs into PF territories in cbln1-null mice is a rare of a lacZ cDNA. The construct retains all introns and exons28of cbln1 as well as B5 kb of 5¢ sequence and B2 kb of downstream sequence . Transgenic mice phenotype and indicates a role for Cbln1 in establishing input-specific were produced by standard pronuclear injection. domains along Purkinje cell dendrites. Notably, all ultrastructural, cellular and physiological abnormalities in cbln1-null mice are shared Targeting of the cbln1 allele. The targeting vector was constructed from the by mice that lack the gene encoding orphan d2 glutamate receptors same 11.57-kb HindIII-BamHI genomic fragment described above. A neomycin (GluRd2)8,21–24, which is expressed selectively in Purkinje cells25. For resistance cassette (neo) was inserted into the first BstXI site downstream of the example, GluRd2- and cbln1-null mice are ataxic and have grossly start codon, and a Herpes simplex thymidine kinase (tk) cassette was inserted normal cerebellar anatomy except for sporadic pyknotic granule cells next to the EcoRI site downstream of the stop codon. The BstXI site was blunt-

a

b

cbln1+/+

cbln1–/–


2. After

1+2

EPSC amplitude (%)

EPSC amplitude (pA)

EPSC amplitude (pA)

1. Before

EPSC amplitude (%)


1+2

[

NUMBER 11

[

NOVEMBER 2005

1539


ARTICLES ended before ligation with neo, resulting in the replacement of four nucleotides from the third nucleotide of codon 28 to codon 29. In the final construct, the left arm is a 5.82-kb HindIII-BstXI fragment, and the right arm is a 2.4-kb BstXIEcoRI fragment. The construct was linearized with HindIII, electroporated into AB2.2 prime embryonic stem (ES) cells (Stratagene) and selected with G418 and FIAU. Genotyping was performed by PCR. Primers were cbln1KO-5¢ (GTGGAGCTGCTGCTGTTGGGGACTG), corresponding to nucleotides 349–373 of cbln1 (AF164680) and cbln1KO-3¢ (CCGCCTTCGAAGCTCTCCCT GTC) that is antisense to nucleotides 650–673 of cbln1, as well as a neo-specific primer, cbln1-neo (GAATGGGCTGACCGCTTCCTCGTG). Four independent chimeras derived from three different ES clones underwent germline transmission and generated founder strains used to characterize the cbln1-null phenotype. Mice were maintained at St. Jude Children’s Research Hospital and had access to food and water ad libitum. Investigational procedures conformed to all applicable federal rules and guidelines and were approved by the Institutional Animal Care and Use Committee. Cell culture. Primary dissociated cerebellar cultures were prepared from dayof-birth ICR (Institute of Cancer Research) mice29. After 2 weeks in vitro, culture medium and cell lysate were subjected to immunoblotting for Cbln1. Human embryonic kidney (HEK) 293T cells were cultured using standard techniques. Northern blotting. Total RNA from mouse cerebellum was extracted using RNAzol B (Tel-Test) and hybridized to a 32P-labelled probe (cbln1 bases 147–471) as described2. Blots were subsequently stripped and rehybridized to a 32P-labelled probe to b-actin. Immunoblotting. Culture medium, cell lysates and cerebellar extracts were run on SDS-PAGE, electrotransferred and immunoblotted for the V5 epitope or Cbln1 using a mouse anti-V5 antiserum (Invitrogen) or a rabbit anti-Cbln1 antiserum, respectively. The rabbit anti-Cbln1 antiserum was generated to the C1q globular domain of Cbln1 (Rockland) and does not cross-react with any of the other three Cbln family members and does not recognize the cerebellin peptide sequence (data not shown). Specificity of the anti-Cbln1 antiserum was established using western blots (Fig. 2d). Bound immunoglobulin was detected with the ECL chemiluminescence system (Amersham). In vitro Cbln1 expression. A full-length mouse cbln1 cDNA was subcloned into the BamHI and XbaI sites of pcDNA3.1V5-His (Invitrogen). PCR-based mutagenesis was used to convert Asn23 and Asn79 of Cbln1 to glutamine; the construct was cloned into the pcDNA3.1V5-His vector and is termed Cbln1(N23,79Q). Recombinant DNAs were purified using QIAgen Midi-Prep kits (QIAgen) before transfection into HEK 293 cells using the Fugene 6 reagent (Roche). Media and cell lysates were collected for analysis 48 h after transfection. Deglycosylation. Cerebellar extracts or conditioned medium were subjected to N- and O-linked deglycosylation using the Glycopro deglycosylation kit (Prozyme). To 100 mg protein in 30 ml was added 10 ml of 5 reaction buffer (0.25 M sodium phosphate, pH 7.0) and 2.5ml denaturation solution (2% SDS, 1 M b-mercaptoethanol) and the mixture heated to 100 1C for 5 min. After cooling, 2.5 ml of 15% Triton X-100 was added, followed by 1 ml each of PNGase F, sialidase A, endo-O-glycosidase, b(1–4)galactosidase and glucosaminidase (enzymes were omitted from control reactions). The mixture was incubated at 37 1C for 3 h and the reaction terminated by adding an equal volume of 2 sample buffer and boiling for 5 min. Radioimmunoassay. Individual cerebella were homogenized in 6 M guanidine hydrochloride, and cerebellin peptide was purified by reverse phase chromatography30. The concentration of cerebellin was determined with a published radioimmunoassay30. Histological methods. Mice were perfused transcardially with buffered 4% paraformaldehyde and the cerebellum removed and postfixed as described2. For double immunolabeling of presynaptic PF or CF terminals and postsynaptic Purkinje cells, microslicer sections were incubated overnight with a mixture of rabbit anti-calbindin antiserum (1:10,000) and either guinea pig antiserum to VGluT1 (to detect PFs; 0.5 mg/ml), VGluT2 (to detect CFs; 0.5 mg/ml) or VGAT (to detect inhibitory interneuron terminals). Sections were incubated for

1540

1 h with Cy3- or FITC-labeled species-specific secondary antibodies (1:200). For aldolase C/zebrin II staining, sections were incubated with mouse antizebrin II (1:100) and bound antibody detected using Alexa 488–conjugated goat anti-mouse immunoglobulin (Molecular Probes; 1:200). A rabbit antiserum to PEP-19 (the product of the pcp4 gene) was used to stain cerebellar Purkinje cells as described31. For in situ hybridization, a 33P-labelled riboprobe (cbln1 bases 147–471) was used following the identical procedures described for cbln3 (ref. 2). Histochemistry for b-galactosidase was as described previously32. For quantitative determinations of volume, cryostat sections (20 mm) from 1-month-old wild-type and cbln1–/– littermates mice of the same gender were cut at 100 mm intervals in the sagittal plane and stained with cresyl violet. Volumes of the cerebellum and internal granule cell layer were calculated from area measurements made using the Bioquant Nova Image Analysis System (R&M Biometrics). Due to variations in dissections, neither the flocculus nor paraflocculus were included in the volume estimates. Electron microscopy. Parasagittal sections (50 mm thick) through the cerebellar vermis were incubated overnight in streptavidin-horseradish peroxidase diluted with 0.1 M phosphate buffer containing 0.5% Tween 20, and the bound complexes were visualized with diaminobenzidine. Sections were postfixed for 30 min with 1% osmium tetroxide in 0.1 M phosphate buffer, block-stained overnight with 1% aqueous uranyl acetate solution, dehydrated in graded alcohols and embedded in Epon 812 (Wako Pure Chemical Industries). Electron micrographs were taken at an original magnification of 5,000 and printed at a final magnification of 10,000. Electrophysiology. Sagittal slices (200 mm thick) of cerebellum were prepared and whole-cell patch-clamp recordings from Purkinje cells were made at room temperature (24 1C) as described33. The slices were continuously superfused with an extracellular solution containing 124 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 1.5 mM MgSO4, 2 mM CaCl2, 26 mM NaHCO3, 20 mM glucose and 0.1 mM picrotoxin (pH 7.35, 310 mOsm kg–1, 24 1C). Patch pipettes had a resistance of 3–4 MO in the intracellular solution containing 135 mM cesium D-gluconate, 15 mM CsCl, 1 mM MgCl2, 10 mM HEPES and 5 mM EGTA (pH 7.3). The series resistance was less than 10 MO and was not compensated. Square pulses (duration, 10 ms; amplitude, 20–220 mA) were applied to a stimulus glass electrode (tip diameter, 5–10 mm) filled with 140 mM NaCl and 10 mM HEPES (pH 7.35) via an isolation unit (Axon Instruments). Purkinje cells were clamped at –80 mV for recording PF-EPSCs and at –10 mV for recording CF-EPSCs. An EPC-7 amplifier (HEKA Instruments) and pCLAMP8 software (Axon Instruments) were used for recording and data analysis. Signals were filtered at 2 kHz and digitized at 4 kHz (Digidata 1320A, Axon Instruments). After stable PF-EPSCs were observed for 15 min, LTD was induced by conjunctive stimulation that consisted of 30 single PF stimuli together with a depolarizing pulse that lasted for 50 ms from a holding potential of –70 mV to +20 mV. A hyperpolarizing pulse (–10 mV, 50 ms) was applied 420 ms before each PF stimulus to monitor the access resistance. If the resistance changed more than 20%, the record was discarded. PF-EPSCs were monitored every 10 s by a glass electrode placed in the molecular layer (pulse width, 10 ms; strength, 20–100 mA) approximately 100 mm away from the pia. Statistical analysis. Student’s t-test was used unless otherwise stated. Accession codes. GenBank: probe to b-actin, AA138737; probe to cbln1 bases 147–471, 164680. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank M. Mishina for the GluRd2–/– mice, R. Hawkes for the Zebrin II antiserum and T. Torashima for technical assistance. Supported in part by US National Institutes of Health grants ES10772, NS040361, NS040749, NS042828 (J.M.), the Toray Science and Technology Grant (M.Y.), Japanese Grants-in-Aid for Scientific Research (M.Y.), Cancer Center Support Core Grant CA 21765 (J.M., M.Y.) and the American Lebanese Syrian Associated Charities (J.M., M.Y.). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.

VOLUME 8

[

NUMBER 11

[


ARTICLES


Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Urade, Y., Oberdick, J., Molinar-Rode, R. & Morgan, J.I. Precerebellin is a cerebellumspecific protein with similarity to the globular domain of complement C1q B chain. Proc. Natl. Acad. Sci. USA 88, 1069–1073 (1991). 2. Pang, Z., Zuo, J. & Morgan, J.I. Cbln3, a novel member of the precerebellin family that binds specifically to Cbln1. J. Neurosci. 20, 6333–6339 (2000). 3. Slemmon, J.R., Blacher, R., Danho, W., Hempstead, J.L. & Morgan, J.I. Isolation and Sequencing of Two Cerebellum-Specific Peptides. Proc. Natl. Acad. Sci. USA 81, 6866– 6870 (1984). 4. Kishore, U. et al. C1q and tumor necrosis factor superfamily: modularity and versatility. Trends Immunol. 25, 551–561 (2004). 5. Beattie, E.C. et al. Control of synaptic strength by glial TNFalpha. Science 295, 2282– 2285 (2002). 6. Mugnaini, E., Dahl, A.L. & Morgan, J.I. Cerebellin is a postsynaptic neuropeptide. Synapse 2, 125–138 (1988). 7. Hawkes, R. & Herrup, K. Aldolase C/zebrin II and the regionalization of the cerebellum. J. Mol. Neurosci. 6, 147–158 (1995). 8. Ichikawa, R. et al. Distal extension of climbing fiber territory and multiple innervation caused by aberrant wiring to adjacent spiny branchlets in cerebellar Purkinje cells lacking glutamate receptor delta 2. J. Neurosci. 22, 8487–8503 (2002). 9. Crepel, F., Mariani, J. & Delhaye-Bouchaud, N. Evidence for a multiple innervation of Purkinje cells by climbing fibers in the immature rat cerebellum. J. Neurobiol. 7, 567–578 (1976). 10. Kano, M. et al. Impaired synapse elimination during cerebellar development in PKC gamma mutant mice. Cell 83, 1223–1231 (1995). 11. Kano, M. et al. Persistent multiple climbing fiber innervation of cerebellar Purkinje cells in mice lacking mGluR1. Neuron 18, 71–79 (1997). 12. Ito, M. Long-term depression. Annu. Rev. Neurosci. 12, 85–102 (1989). 13. Lev-Ram, V., Jiang, T., Wood, J., Lawrence, D.S. & Tsien, R.Y. Synergies and coincidence requirements between NO, cGMP, and Ca2+ in the induction of cerebellar long-term depression. Neuron 18, 1025–1038 (1997). 14. Matsuda, S., Launey, T., Mikawa, S. & Hirai, H. Disruption of AMPA receptor GluR2 clusters following long-term depression induction in cerebellar Purkinje neurons. EMBO J. 19, 2765–2774 (2000). 15. Wang, Y.T. & Linden, D.J. Expression of cerebellar long-term depression requires postsynaptic clathrin-mediated endocytosis. Neuron 25, 635–647 (2000). 16. Watanabe, D. et al. Ablation of cerebellar Golgi cells disrupts synaptic integration involving GABA inhibition and NMDA receptor activation in motor coordination. Cell 95, 17–27 (1998). 17. Sotelo, C. Cerebellar synaptogenesis: what we can learn from mutant mice. J. Exp. Biol. 153, 225–249 (1990).


[

NUMBER 11

[

NOVEMBER 2005

18. Rhyu, I.J., Abbott, L.C., Walker, D.B. & Sotelo, C. An ultrastructural study of granule cell/ Purkinje cell synapses in tottering (tg/tg), leaner (tg(la)/tg(la)) and compound heterozygous tottering/leaner (tg/tg(la)) mice. Neuroscience 90, 717–728 (1999). 19. Chen, C. et al. Impaired motor coordination correlates with persistent multiple climbing fiber innervation in PKC gamma mutant mice. Cell 83, 1233–1242 (1995). 20. Benoit, P., Delhaye-Bouchaud, N., Changeux, J.P. & Mariani, J. Stability of multiple innervation of Purkinje cells by climbing fibers in the agranular cerebellum of old rats Xirradiated at birth. Brain Res. 316, 310–313 (1984). 21. Kashiwabuchi, N. et al. Impairment of motor coordination, Purkinje cell synapse formation, and cerebellar long-term depression in GluR delta 2 mutant mice. Cell 81, 245–252 (1995). 22. Kurihara, H. et al. Impaired parallel fiber-Purkinje cell synapse stabilization during cerebellar development of mutant mice lacking the glutamate receptor delta2 subunit. J. Neurosci. 17, 9613–9623 (1997). 23. Hashimoto, K. et al. Roles of glutamate receptor delta 2 subunit (GluRdelta 2) and metabotropic glutamate receptor subtype 1 (mGluR1) in climbing fiber synapse elimination during postnatal cerebellar development. J. Neurosci. 21, 9701–9712 (2001). 24. Lalouette, A., Lohof, A., Sotelo, C., Guenet, J. & Mariani, J. Neurobiological effects of a null mutation depend on genetic context: comparison between two hotfoot alleles of the delta-2 ionotropic glutamate receptor. Neuroscience 105, 443–455 (2001). 25. Yuzaki, M. The delta2 glutamate receptor: 10 years later. Neurosci. Res. 46, 11–22 (2003). 26. Morando, L., Cesa, R., Rasetti, R., Harvey, R. & Strata, P. Role of glutamate delta -2 receptors in activity-dependent competition between heterologous afferent fibers. Proc. Natl. Acad. Sci. USA 98, 9954–9959 (2001). 27. Lomeli, H. et al. The rat delta-1 and delta-2 subunits extend the excitatory amino acid receptor family. FEBS Lett. 315, 318–322 (1993). 28. Kavety, B., Jenkins, N.A., Fletcher, C.F., Copeland, N.G. & Morgan, J.I. Genomic structure and mapping of precerebellin and a precerebellin-related gene. Brain Res. Mol. Brain Res. 27, 152–156 (1994). 29. Furuya, S., Makino, A. & Hirabayashi, Y. An improved method for culturing cerebellar Purkinje cells with differentiated dendrites under a mixed monolayer setting. Brain Res. Brain Res. Protoc. 3, 192–198 (1998). 30. Morgan, J.I. et al. Cerebellin and related postsynaptic peptides in the brain of normal and neurodevelopmentally mutant vertebrates. Synapse 2, 117–124 (1988). 31. Ziai, M.R., Sangameswaran, L., Hempstead, J.L., Danho, W. & Morgan, J.I. An immunochemical analysis of the distribution of a brain-specific polypeptide, PEP-19. J. Neurochem. 51, 1771–1776 (1988). 32. Oberdick, J., Smeyne, R.J., Mann, J.R., Zackson, S. & Morgan, J.I. A promoter that drives transgene expression in cerebellar Purkinje and retinal bipolar neurons. Science 248, 223–226 (1990). 33. Kohda, K., Wang, Y. & Yuzaki, M. Mutation of a glutamate receptor motif reveals its role in gating and delta2 receptor channel properties. Nat. Neurosci. 3, 315–322 (2000).

1541


ARTICLES

Activity-dependent decrease of excitability in rat hippocampal neurons through increases in Ih Yuan Fan1,4, Desdemona Fricker2,4, Darrin H Brager1, Xixi Chen1, Hui-Chen Lu3, Raymond A Chitwood1 & Daniel Johnston1 Hippocampal long-term potentiation (LTP) induced by theta-burst pairing of Schaffer collateral inputs and postsynaptic firing is associated with localized increases in synaptic strength and dendritic excitability. Using the same protocol, we now demonstrate a decrease in cellular excitability that was blocked by the h-channel blocker ZD7288. This decrease was also induced by postsynaptic theta-burst firing alone, yet it was blocked by NMDA receptor antagonists, postsynaptic Ca21 chelation, low concentrations of tetrodotoxin, x-conotoxin MVIIC, calcium/calmodulin-dependent protein kinase II (CaMKII) inhibitors and a protein synthesis inhibitor. Increasing network activity with high extracellular K1 caused a similar reduction of cellular excitability and an increase in h-channel HCN1 protein. We propose that backpropagating action potentials open glutamate-bound NMDA receptors, resulting in an increase in Ih and a decrease in overall excitability. The occurrence of such a reduction in cellular excitability in parallel with synaptic potentiation would be a negative feedback mechanism to normalize neuronal output firing and thus promote network stability.

Hippocampal LTP is a form of neuronal plasticity that is thought to underlie forms of learning and memory1. It generally refers to the persistent enhancement of the excitatory postsynaptic potential (EPSP) after appropriate synaptic stimulation. Highly localized synaptic changes enable this plasticity and result in an increased synaptic strength. Potentiated EPSPs often generate action potentials more effectively than would be expected from the increase in EPSP amplitude alone, suggesting a higher efficiency of EPSP-to-spike coupling after LTP induction2,3. Several factors may contribute to this enhanced excitability, including changes in synaptic inhibition and in somatic or dendritic membrane conductances4. A localized decrease in dendritic A-type K+ currents after LTP induction may also contribute to the enhanced excitability from the potentiated synaptic input5. Activity-dependent modifications of intrinsic electrical properties are common in the developing brain6 and are increasingly recognized as important in neuronal plasticity4,7,8. They occur in cell culture systems9,10, in cerebellar neurons after tetanic stimulation11 and in brain slices after synaptic activation or LTP5,12–16. Changes in postsynaptic membrane properties can affect synaptic signal integration globally or can be restricted to a potentiated synapse. It is possible that the intracellular Ca2+ signal associated with LTP induction, which causes activation of different protein kinases17, might initiate signaling cascades that also regulate postsynaptic ionic conductances. We began this study to examine whether synaptic LTP is accompanied by changes in intrinsic excitability that are global in nature and are not restricted to the potentiated input. We used a specific LTP

induction protocol consisting of the pairing of small Schaffer collateral synaptic potentials with postsynaptic action potentials at theta frequencies. We observed a dampening of overall neuronal excitability after this LTP induction that seemed to result from increased Ih. This change, however, could be induced only with postsynaptic backpropagating action potentials (b-APs), but it still required NMDA receptors, spontaneous synaptic potentials, Ca2+ influx, CaMKII activity and protein synthesis. We also found that enhancing NMDA receptor– mediated synaptic transmission with high extracellular K+ induced a similar reduction in cellular excitability and a significant upregulation of the HCN1 subunit protein. We propose that such a negative feedback mechanism will tend to normalize output firing of a neuron and thus promote network stability. RESULTS LTP induction and associated decrease in excitability LTP was induced using a theta-burst pairing (TBP) protocol5,18,19, which consisted of subthreshold synaptic stimulation paired with backpropagating action potentials at theta frequency. This procedure was chosen for its robust induction of LTP as well as for its physiological relevance19–21 (the recording and stimulating configuration is shown in Fig. 1a). EPSPs were evoked in stratum radiatum (120–150 mm from soma) in close proximity (o30 mm) to the dendrite. TBP induced strong LTP that developed over time (Fig. 1b), yielding a B175% increase in EPSP initial slope, as measured at 30 min after the pairing stimulation. Blocking NMDA receptors with AP5 (50 mM) plus

1Center for Learning and Memory, University of Texas at Austin, 1 University Station, C7000, Austin, Texas 78712, USA. 2Institut National de la Sante ´ et de la Recherche Me´dicale (INSERM) U739–Cortex et Epilepsie, Faculte´ de Me´decine Pitie´-Salpeˆtrie`re, 105 Boulevard de I’Hoˆpital, 75013 Paris, France. 3Cain Foundation Laboratories, 4 Department of Pediatrics, Division of Neurology and Developmental Neuroscience, Baylor College of Medicine, Houston, Texas 77030, USA. These authors contributed equally to this work. Correspondence should be addressed to D.J. ([email protected]).

Received 3 August; accepted 16 September; published online 23 October 2005; corrected after print 14 December 2005; doi:10.1038/nn1568

1542

VOLUME 8

[

NUMBER 11

[


ARTICLES

0

d

Input resistance (M)

c –63 mV pre TBP

5 mV 200 ms

–63 mV post TBP

f

Control AP5/MK-801

60 40 20 0

10 20 Time (min)

30

40

Normalized input resistance

Input resistance (MΩ)

e

75

10 20 Time (min)

30

40

Pre-TBP Post-TBP

*

50 25 0

Control

AP5/MK-801

(700–750 ms) current injections (Fig. 1c–f). NMDA receptor antagonists blocked this change in input resistance, and, notably, the time course of the change in input resistance mirrored the change in EPSP slope (Fig. 1e,f). No significant changes in series resistance were observed during these experiments (monitored separately).

Control AP5/MK-801 1.00 0.75 0.50 0.25 1.0

1.5 2.0 2.5 3.0 3.5 Normalized EPSP slope

Reduced excitability accompanies synaptic potentiation The decrease in input resistance that accompanies LTP indicated a possible decrease in intrinsic cellular excitability. We examined excitability using step current injections from the soma before and after LTP induction (Fig. 2). In keeping with the observed change in input resistance, we found that cellular excitability was indeed reduced after LTP. More current was necessary to reach firing threshold and to obtain the same number of spikes as before LTP induction (Fig. 2a).

4.0

MK-801 (10 mM) prevented the change (Fig. 1b). Similar to previous reports5,18,19, these results demonstrate that TBP-induced LTP is largely, but not completely, dependent on NMDA receptor activation. The potentiation of the synaptic response was accompanied by a significant decrease in the input resistance (RN) as measured by long

b

LTP

Number of spikes

10 8

Pre-TBP 10 min post-TBP 20 min 30 min

6

d

10

* *

4 2

*

0 150

5 mV 50 ms

c

Control

Number of spikes

a

20 mV 200 ms

200 250 I injection (pA)

8 6 4 2 0

300

150 20 mV 200 ms

e

Pre-TBP

t = 0 min t = 10 min t = 20 min t = 30 min


f

–44

0 eT Af BP t in er du L ct TP io n

(mV)

After LTP induction

–48

Pr

–52 2s

V threshold

0.16

* 0

***

40 30 20 10 0

Background spikes Evoked spikes

–44 –46 –48

0.08

Pr eT Af BP in ter du L ct TP io n

–52

Spontaneous Spikes per second

Evoked 100

–48

TBP-induced change in V threshold (mV)

–44 Probability %

(mV)

–42

300

50

P

X3

l

0

Post


1

LT

Pre

2

tro

20 mV 200 ms

200 ms

3

on

Post

C

Pre ~120 µm

Figure 1 TBP-induced LTP was accompanied by a decrease in input resistance (RN). (a) A TBP protocol was used to induce LTP. The trace illustrates a representative response to a train of TBP. The locations of the recording and stimulating electrodes are shown at left. (b) Time course and magnitude of synaptic potentiation. TBP (indicated by filled triangle) resulted in a time-dependent and sustained increase in EPSP slope that was largely suppressed in the presence of AP5 (50 mM) and MK-801 (10 mM). (c) Representative voltage deflections in response to a series of step current injections (–100–100 pA) before and 30 min after LTP. (d) Summary data showing a significant decrease in input resistance after TBP in control saline that was blocked by AP5 and MK-801. *, P o 0.05. (e) Time course of RN changes after TBP, from control and AP5/MK-801 experiments. (f) Relationship between the RN and EPSP slope after TBP, in control and AP5/MK-801 experiments. There was a strong linear relationship between the decrease in RN and the increase in EPSP slope in control saline that was abolished in the presence of AP5/MK-801.

Control AP5/MK-801

4

Change in Ithreshold (%)

b

Theta-burst pairing (TBP)

Normalized EPSP slope

a

Pre-TBP

After LTP induction

2 1 0 –1 –2 Spontaneous

Evoked

Figure 2 TBP-LTP was associated with a persistent decrease in excitability. (a) TBP resulted in a significant decrease in excitability. The decrease in excitability was significant by 10 min post-TBP. Action potential from potentiated EPSP shown at the left is truncated for clarification (arrow). (b) There was no significant change in excitability without TBP *, P o 0.05. (c) Summary graph showing current threshold increase by 41.2 ± 5.7% after TBP and almost no change (6.7 ± 4.9%) in control recordings (***, P o 0.001). (d) Representative traces of 16-s recordings before and after LTP induction. Membrane potential was kept at threshold level, such that the evoked EPSP initiated action potentials in 50% of trials in control conditions (upper traces). Stimulations are indicated by filled triangle. Spontaneous background EPSPs impinging on the cell induced firing between stimulations. After LTP induction (lower traces), the potentiated EPSP triggered firing in all trials. However, background synaptic activity initiated action potentials much less effectively than before LTP. (e) The probability to initiate spikes from the evoked EPSP was set close to 50% in control conditions (pre-TBP), whereas after LTP induction, spikes were initiated at each stimulation (100%). The background synaptic activity led to spike firing at a frequency of 0.133 ± 0.032 s–1, which decreased to 0.026 ± 0.012 background spikes s–1 after LTP induction (*, P o 0.05). (f) Voltage threshold of background spikes remained stable over time (top plot). Threshold of spikes arising from the potentiated EPSP was significantly lowered (lower plot).


[

NUMBER 11

[

NOVEMBER 2005

1543

Background synaptic inputs become less efficient Our data imply that potentiation of synaptic inputs is accompanied by a reduced excitability of CA1 pyramidal neurons. Although this was evident in responses to somatic current injections, it was not clear whether excitability was also decreased at dendritic sites near excitatory connections. If so, nonpotentiated excitatory synaptic events might initiate spike firing less effectively after LTP. To test this hypothesis, we examined the input-output relation of naive inputs to the postsynaptic cell. We recorded long stretches of spontaneous background synaptic events that were impinging on a CA1 pyramidal cell and then induced LTP (Fig. 2d). Evoked and spontaneous EPSPs were recorded at a holding potential where about 50% of the stimulations initiated spikes. Spontaneous action potentials were generated in six of eight cells tested, at a rate of around 0.2 spikes/s. After LTP induction, the potentiated synapses increased in efficacy to produce action potentials in 100% of responses to stimulation. In contrast to this (and consistent with our hypothesis), the number of spontaneous EPSPs that produced action potentials was reduced to o50% of control in all six cells (measured 20 min after LTP induction; Fig. 2e). EPSP-to-spike coupling of unpotentiated inputs was thus less efficient. The change in the spontaneous and evoked events that leads to action potential generation is shown in Supplementary Figure 1 online. We were concerned that this observation might result from an artifactual rundown of neuronal excitability caused by washout of intracellular components. In particular, the sodium channel activity

a

could have been reduced over time. If so, the voltage threshold for action potential initiation would be expected to change. We measured threshold as the voltage where the first derivative of the action potential waveform exceeded 10 V s–1 (dV/dt 4 10 V s–1) for both evoked and spontaneous spikes (Fig. 2f). Action potentials arising from spontaneous EPSPs had a stable voltage threshold over time (DVthreshold ¼ 0.2 ± 0.4 mV), whereas the spike threshold from the potentiated EPSPs decreased (DVthreshold ¼ –1.47 ± 0.45 mV; P o 0.01). In two experiments when our TBP protocol did not induce LTP, the rate of background spikes that were due to spontaneous EPSPs remained stable or even increased slightly (data not shown), arguing against a loss of excitability because of the recording configuration or the length of the recording. Mechanisms underlying changes in excitability Modulation of several ionic currents might underlie these changes in intrinsic cellular excitability. Our first indication of their identity came from the decrease in input resistance after TBP-LTP induction (Fig. 1). Also, somatic recordings with step current injections often showed more prominent ‘sag’ upon hyperpolarization together with an enhanced rebound depolarization (Supplementary Fig. 2). Coupled to a small but significant membrane depolarization after TBP-LTP induction (Vm increased by 4 ± 1 mV; P o 0.05, n ¼ 6), these changes pointed to an increase in the cationic Ih current22–24. We therefore tested the effects of the Ih blocker ZD7288 (20 mM) on responses to step current injections after TBP-LTP induction (Fig. 3a). This manipulation rescued the probability of action potential generation to an amount that was similar to the probability during the control period (n ¼ 7). In addition, the presence of ZD7288 during the entire recording period completely prevented any significant changes in excitability (119.8 ± 14.4 MO versus 131.0 ± 12.8 MO; n ¼ 7), without affecting TBP-LTP induction or magnitude (262 ± 30% versus 307 ± 34% with ZD7288; n ¼ 7; P ¼ 0.28; Fig. 3b–d). Identical results were obtained with ZD7288 included in the recording pipette, indicating that ZD7288 had an intracellular site of action and that its effect was postsynaptic25.

b

Pre-TBP

LTP

5 mV 20 ms Post-ZD7288

c 5 mV 20 ms Pre-TBP

LTP

10

Pre-TBP 20 min post TBP 30 min post TBP

d

8 6 4 2

TBP TBP + ZD7288

1.0 Cumulative probability

Input-output curves were shifted to the right, whereas in control recordings very little change in excitability was observed (Fig. 2b). Current threshold, which is defined as the minimum current injection needed to evoke a spike, was stable over time in control conditions but increased significantly after LTP induction (Fig. 2c). The current threshold increased gradually over time (Fig. 2a), similar to that observed for input resistance (Fig. 1e). There was no change in voltage threshold, however, as measured from the first action potential initiated by a threshold current stimulus (before TBP, –49.1 ± 0.5 mV; after TBP, –50.0 ± 0.8 mV; P 4 0.05, n ¼ 6).

Number of spikes

0.8 0.6 0.4 0.2

0 20 mV 200 ms

100

20 mV 200 ms

200

300

0

I injection (pA)

e

100 200 300 400 EPSP slope (%)

TBP

f

Figure 3 Blockade of h-channels reversed and occluded the decrease in excitability TBP + ZD7288 1.5 without affecting LTP of EPSPs. (a) Representative traces showing potentiation ** 2 TBP of EPSP by TBP (upper traces) and the decrease in excitability (lower traces). 1 Subsequent application of the h-channel blocker ZD7288 reversed the decrease ** in excitability but not the EPSP potentiation. (b) Presence of ZD7288 throughout *** 1 the experiment prevented the decrease in excitability but did not prevent LTP. 2 TBP (c) Summary data showing the stability of input-output firing property after TBP in the 1 + ZD7288 presence of ZD7288. (d) Cumulative distribution of EPSP potentiation induced by 0.5 TBP with and without ZD7288, respectively. No significant difference was observed. 1 mV 20 ms (e) After TBP, the 10–90% decay time of evoked EPSPs was decreased, and this change was prevented by ZD7288. Note the increase in the voltage undershoot after TBP in trace 2 (upper traces). (f) Summary graph of changes in EPSP decay, current threshold and input resistance after LTP induction using TBP, in the absence or presence of ZD7288. The 10–90% EPSP decay time ratio decreased, and current threshold ratio increased significantly after TBP, as compared with experiments in the presence of ZD7288 (**, P o 0.005; ***, P o 0.0001, respectively). Input resistance decreased after LTP induction using the TBP protocol, whereas it remained stable in the presence of ZD7288 (**, P o 0.01).

1544

VOLUME 8

[

NUMBER 11

[

re

I si npu st t an ce

d

ol

sh

re

th

I

de 10– ca 90 y % tim e

LTP induction/baseline


ARTICLES


ARTICLES

b

Theta-burst firing (TBF)

c

Ic

30 min post-TBF 20 mV 200 ms 1nA ×3

Post-ZD7288

5 mV 200 ms


Pre-TBF

120 100

30 min post-TBF

80 60

***

40 20 0

Post-ZD7288 20 mV 200 ms

16 14 12 10 8 6 4 2 0

Pre-TBF 30 min post TBF Post-ZD7288

100

200 300 Iinjection (pA)

400

Figure 4 Theta-burst firing (TBF) alone sufficed to produce a decrease in excitability. Pre-TBF (a) TBF protocol used to elicit postsynaptic firing. As indicated, TBF was the postsynaptic 30 min post TBF 120 component of TBP shown in Figure 1a. The upper trace was an example of the voltage Pre-TBF Post-ZD7288 response to a train of TBF. (b) Representative traces showing the voltage response to 30 min post TBF 100 Post-ZD7288 subthreshold step current injections, obtained before TBF, 30 min after TBF and after 80 subsequent ZD7288 bath application. (c) Summary graph of the RN profile obtained under 60 three sequential conditions. RN was significantly decreased 30 min after TBF. Subsequent 40 bath application of ZD7288 reversed RN. (***, P o 0.0001). (d) Sample recordings of action potential elicited by suprathreshold current step before TBF, 30 min after TBF and 2 mV 20 ** after ZD7288 application, respectively. (e) Input-output firing properties before TBF, 200 ms 0 30 min after TBF and after ZD7288 application. More current was needed to evoke action potentials after TBF, resulting in a rightward shift in the input-output relationship. (f) Example traces of simulated EPSP (aEPSP) by injecting a train of five a function currents at 20 Hz. The first aEPSP peaks were scaled for comparison. Note the progressive decline of the aEPSP amplitude after TBF and the shortened duration (arrow). The temporal summation of aEPSP was reduced after TBF and reversed by ZD7288. ZD7288 also eliminated the voltage undershoot. (g) Summary graph of the temporal summation ratio obtained under three sequential conditions (**, P o 0.01).

f

An upregulation of Ih after TBP stimulation might also be expected to accelerate the decay of EPSPs and to increase the following undershoot. We quantified these changes by comparing the 10–90% decay time of the EPSP before and after potentiation. For LTP induced by the TBP protocol, the decay time ratio was less than 1, indicating a faster decay of the potentiated EPSP. In the presence of ZD7288, however, the potentiated EPSP returned to baseline more slowly, with a prolonged decay time (Fig. 3e,f). Taken together, these results indicate that an increase in Ih after the TBP-LTP induction may be responsible for a global reduction of cellular excitability. This finding was surprising to us, as previous studies have not documented an h-current–dependent decrease in input resistance of the postsynaptic cell. We therefore carried out experiments in which a more typical 100-Hz, 1.5-s train of Schaffer collateral stimulation was used. There was no change in input resistance associated with LTP (Supplementary Fig. 3), indicating that the input resistance change was dependent on the LTP induction paradigm. Taken together with the decrease in input resistance, these data indicate that an increase in Ih is associated with TBP-LTP. Mechanism for the increase in Ih Ih was upregulated after LTP induction using the TBP protocol but not with more standard 100-Hz stimulus procedures. One striking difference between these procedures is the number of postsynaptic action potentials that occur during the induction procedures. Typically, few, if any, action potentials were triggered during the 1.5-s, 100-Hz train, whereas an action potential occurred with each EPSP during the TBP protocol for a total of 150 during the procedure. Perhaps the number of action potentials is the critical difference. We therefore asked whether postsynaptic firing alone could upregulate Ih. The TBP protocol consists of postsynaptic action potentials paired with EPSPs in a theta-burst pattern, so we examined the effects of triggering action potentials with intracellular current injection in the same theta-burst pattern but without synaptic stimulation. We called this theta-burst firing or TBF (Fig. 4a). We also observed a qualitatively


[

NUMBER 11

[

NOVEMBER 2005

g

α EPSP summation ratio (%)


200 ms

e

140

Pre-TBF

Vm

d

Pre-TBF 30 min post TBF Post-ZD7288

Number of spikes

a

similar (but smaller) decrease in input resistance (53.3 ± 3.8 MO versus 75.0 ± 5.6 MO; n ¼ 18; P o 0.0001) after TBF, which was rescued by subsequent bath application of ZD7288 (117.2 ± 8.6 MO; n ¼ 12) (Fig. 4b,c). The resting membrane potential was depolarized by 6.8 ± 0.6 mV at 30 min after TBF (n ¼ 18). As with TBP, the decrease in input resistance was accompanied by a decrease in excitability as measured by current injection to the soma that was reversed by the h-channel blocker ZD7288 (Fig. 4d,e). Ih is known to reduce temporal summation of synaptic input22,26. Temporal summation was determined by applying a brief train of current injections modeled by a function to the soma to mimic a train of EPSPs. The ratio of amplitudes of the last and of the first aEPSP was used as an index of temporal summation. This temporal summation ratio was significantly reduced after the TBF (14.41 ± 5.81% versus 37.41 ± 9.38%; n ¼ 6; P o 0.01) but was reversed after the application of ZD7288 (87.32 ± 17.28%; n ¼ 4; Fig. 4f,g). We asked whether ZD7288 could prevent the change in input resistance after TBF, rather than simply reversing it by delivering TBF in the presence of ZD7288 in the bath. Excitability was not reduced in these experiments (RN: 109.2 ± 10.7 MO versus 110.7 ± 10.7 MO, n ¼ 9; temporal summation ratio: 66.81 ± 3.34% versus 65.07 ± 7.29%; n ¼ 4; Supplementary Fig. 4). Furthermore, there was no significant difference in the input resistance of the neurons treated with ZD7288 before TBF and those treated with ZD7288 at 30 min after TBF (110.7 ± 10.7 MO versus 117.2 ± 8.6 MO; n ¼ 9 and 12, respectively). Increase in Ih requires Ca2+ influx through NMDA receptors These results indicate that postsynaptic burst firing at theta frequencies can trigger an upregulation of Ih. Such postsynaptic firing increases [Ca2+]i through activation of voltage-gated Ca2+ channels and/or glutamate-bound NMDA receptors19. We tested directly whether Ca2+ is involved in the TBF-induced upregulation of Ih by loading the Ca2+ chelator BAPTA (20 mM) into the postsynaptic neuron by way of the whole-cell patch pipette. We found that BAPTA completely prevented the change in excitability after TBF (RN: 82.8 ± 12.3 MO

1545

ARTICLES

60 40

10 5

* 20

* l N

Q

iC

X

P5

+ e

-A

re nu

,L

na

l

ip

Ky

N

N

im

im

od

od

ip

2

0

2

P5 -A ,L D

na re nu

40

te

0

60

N

20

30 min post-TBF

in

***

40

Pre-TBF

80

400

C

*

60


D

80

te

0

h

30 min post-TBF


Pre-TBF

100

iC

g

N

10

15

100

X

20

Pre-TBF 30 min post-TBF

+

30

20

30 min post-TBF

e

40

30 min post TBF

0

0 50

Pre-TBF

25

20

Q

2 mV 200 ms

20 mV 200 ms

80

N

f Pre-TBF 30 min post-TBF

Pre-TBF

in

e

100

Ky


30 min post TBF

d

C

5 mV 200 ms

c



Pre-TBF

Pre-TBF 30 min post TBF

Number of spikes

b


a

Figure 5 Decreased excitability required postsynaptic Ca2+ and NMDA-receptor activation. (a) Example traces of the voltage responses to subthreshold step current injections obtained before and 30 min after TBF, in the presence of intracellular BAPTA (20 mM). (b) Summary graph illustrating the stability of RN after TBF in the presence of BAPTA. (c) Sample recordings of action potential elicited by suprathreshold step current injection before and 30 min after TBF, in the presence of BAPTA. (d) Input-output firing relationship remained stable after TBF for 30 min in the presence of BAPTA. (e) Example traces of aEPSP before and 30 min after TBF, in the presence of BAPTA. (f) Summary graph showing the stability of temporal summation ratio before and 30 min after TBF, in the presence of BAPTA. (g) Summary graph of RN profile before and after TBF, in the presence of broad-spectrum ionotropic glutamate receptor blocker kynurenic acid (1 mM), NMDA receptor blocker AP5 (50 mM), AMPA receptor blocker CNQX (20 mM), and L-type and T-type VDCC blockers nimodipine (10 mM) and NiCl2 (50 mM), respectively. Neither CNQX nor nimodipine and NiCl2 could block TBF-induced change in RN. (***, P o 0.001; *, P o 0.05.) (h) Summary graph of temporal summation profile before and after TBF, in the presence of the blockers. Neither CNQX nor nimodipine and NiCl2 could block TBF-induced change in temporal summation.

versus 75.3 ± 7.5 MO, n ¼ 10; temporal summation ratio: 34.08 ± 4.16% versus 34.75 ± 3.38%, n ¼ 5; Fig. 5a–f). Because voltage-gated Ca2+ channels provide a major source for the action potential–induced rise in postsynaptic Ca2+ (ref. 27), we tested whether L- and T-type Ca2+ channels were involved. Notably, bath application of nimodipine and Ni2+ did not prevent the TBF-induced change in excitability (RN: 68.4 ± 7.9 MO versus 87.5 ± 9.4 MO; temporal summation ratio: 22.60 ± 3.81% versus 38.70 ± 7.78%; n ¼ 6; P o 0.05; Fig. 5g,h). Alternative sources of Ca2+ influx include Ca2+-permeable AMPA and NMDA receptors. Several factors indicate that AMPA and/or NMDA receptors may have been involved. First, there clearly was spontaneous excitatory synaptic transmission during recording, as indicated by randomly occurring small EPSPs. Second, the Ca2+/Mg2+ ratio in our external solution was 2:1, thereby favoring NMDA receptor activation because of reduced Mg2+ block. Third, the percentage of input resistance reduction induced by TBF was smaller than that induced by TBP (TBP, 40 ± 4%; TBF, 26 ± 5%). The degree of activation of AMPA and/or NMDA receptors seems to be the only substantial difference between the TBP and TBF stimulation paradigms. As a first test of whether ionotropic glutamate receptors are involved in the modulation of intrinsic excitability, we included the broadspectrum glutamate antagonist kynurenic acid (1 mM) in the bath. No decrease in input resistance after TBF was found under this condition (RN: 91.9 ± 8.2 MO versus 78.4 ± 8.4 MO; P o 0.01; temporal summation ratio: 50.95 ± 6.14% versus 29.53 ± 5.66%; P o 0.01; n ¼ 6; Fig. 5g,h). Further, bath application of the NMDA receptor antagonist AP5 (50 mM) in combination with the AMPA receptor antagonist CNQX (20 mM) blocked the TBF-induced decrease in excitability

(RN: 82.0 ± 10.5 MO versus 73.5 ± 10.1 MO; P ¼ 0.32; temporal summation ratio: 43.30 ± 6.94% versus 28.90 ± 4.55%; P o 0.05; n ¼ 7; data not shown). To determine each receptor’s contribution, we added AP5 (50 mM) or CNQX (20 mM) separately to the bath while delivering TBF. AP5 alone sufficed to block the change (RN: 76.1 ± 6.4 MO versus 76.0 ± 7.5 MO; P ¼ 0.97; temporal summation ratio: 50.19 ± 4.92% versus 39.28 ± 3.74%; P ¼ 0.12; n ¼ 6; Fig. 5g,h), whereas CNQX alone did not prevent it (RN: 46.8 ± 4.0 MO versus 71.6 ± 6.1 MO; P o 0.001; temporal summation ratio: 7.52 ± 3.00% versus 19.19 ± 2.67%; P o 0.05; n ¼ 5; Fig. 5g,h). These experiments indicate that during both TBP and TBF, Ca2+ may enter primarily through activated NMDA receptors to initiate the change in input resistance. B-APs are required for TBF-induced changes in excitability Action potentials triggered in the soma backpropagate into the dendrites and can amplify Ca2+ entry through NMDA receptors19,28. We tested the hypothesis that backpropagating action potentials are required for TBF-induced changes in excitability by bath applying low concentrations of tetrodotoxin (TTX). At low concentrations, TTX can selectively block the backpropagation of action potentials into dendrites29,30. In our experiments, 10 nM TTX did not affect the amplitude of action potentials measured in soma (108.8 ± 1.8 mV versus 107.7 ± 1.4 mV; n ¼ 9) but did produce a small (5.1 ± 0.9 mV) increase in the somatic firing threshold (Fig. 6a,b). The efficacy of 10 nM TTX to block backpropagation of action potentials was verified by Ca2+ imaging experiments under identical recording conditions except for the addition of the calcium indicator dye bis-fura-2 (100 mM) in the internal solution. In these experiments, 10 nM TTX significantly reduced the backpropagating action potential-induced

1546

VOLUME 8

[

NUMBER 11

[


ARTICLES

10 nM TTX Washout

b

c

d

10 nM TTX Washout

170 µm

Control TTX

%∆F/F

100 80 60 40 20 0 80 µm

Superimposed

*

8 6

**

4 2

20 µm

AP amplitude (mV)

0 6% ∆F/F 80 mV 50 ms

f

60 µm 120 µm 180 µm Distance from soma


80 60 40 20 0

g

Pre-TBF Figure 6 Backpropagating action potentials were necessary for a TBFSpontaneous 30 min post-TBF induced decrease in excitability. (a) Representative traces of action potential 60 Baseline 10 nM TTX recorded in the presence of a low concentration of TTX (10 nM) and after 50 washout of TTX. Note the slight increase in somatic firing threshold in TTX –65 mV 40 (arrows). (b) Group data showing that low TTX did not affect the amplitude 30 of action potential, measured from baseline membrane potential to action potential peak. (c) Left: fluorescence image of the main trunk of the apical 20 dendrite of a bis-fura-2–filled CA1 pyramidal neuron. Right: changes in Ca2+ 1 mV 10 500 ms in response to a backpropagating action potential before (top and center 0 traces) and after (bottom trace) low TTX application were measured as a percentage of DF/F simultaneously along the apical trunk. The color of the traces corresponds to the colored boxes in the fluorescence image. (d) Summary data of the blocking effect of low TTX on backpropagating action potential, as indicated by the reduction of F/F values at different locations along the apical trunk. (e) Summary plot of RN profile before and after TBF in the presence of low TTX. RN remained stable after TBF. (f) Summary plot showing the stability of temporal summation ratio before and after TBF in the presence of low TTX. (g) Sample recordings showing that low TTX did not substantially affect spontaneous EPSPs. α EPSP summation ratio (%)


100

10

120

20 mV 20 ms

12

e

Control 10 nM TTX


a

Ca2+ signals in the dendrites that were induced by backpropagating action potentials, which is consistent with a reduction in backpropagating action potential amplitude (Fig. 6c,d). In the presence of 10 nM TTX, there was no significant change in input resistance or excitability after TBF (RN: 72.8 ± 8.8 MO versus 78.3 ± 4.9 MO; temporal summation ratio: 38.34 ± 6.13% versus 31.80 ± 5.72%; n ¼ 8; Fig. 6e,f), which is consistent with the idea that the backpropagation of action potentials during TBF is necessary to upregulate Ih. Spontaneous synaptic activities were not substantially altered by 10 nM TTX (Fig. 6g). We hypothesized that the backpropagation of action potentials in dendrites opens glutamate-bound NMDA receptors and thus initiates changes in input resistance and excitability after TBF. The existence of NMDA receptors that are bound with glutamate should depend on the spontaneous release of transmitter. To test this possibility, we bath applied o-conotoxin (CTX) MVIIC (10 mM), a blocker of N- and P/Q-type Ca2+ channels, to suppress transmitter release triggered by spontaneous action potentials. As expected, both spontaneous and evoked EPSPs were reduced by the application of CTX (n ¼ 4; Fig. 7a–c). We also found that the change in input resistance and temporal summation after the TBF was prevented (RN: 78.4 ± 10.1 MO versus 70.9 ± 11.7 MO; temporal summation ratio: 30.00 ± 14.20% versus 18.40 ± 8.60%; n ¼ 4 and 3, respectively, Fig. 7d–f). Background synaptic inputs become less efficient after TBF The results presented so far indicate a reduced excitability of CA1 pyramidal neurons after TBF. Although this was evident in response to somatic current injections, it was not clear how this might affect the integration of naturally occurring EPSPs. To test this, the input-output relationship of the spontaneous EPSPs to the postsynaptic cell was examined in a similar manner to that shown in Figure 2d. Specifically, long stretches (10 s) of spontaneous background synaptic inputs impinging on a postsynaptic neuron were


[

NUMBER 11

[

NOVEMBER 2005

recorded at a constant holding potential before and 30 min after TBF. The holding potential of a neuron was typically set, before TBF, at an arbitrary depolarized level at which the cell fired at a rate of about 1 Hz (Supplementary Fig. 5). This manipulation showed the summation of the spontaneous EPSPs that led to action potential firing. After TBF, the number of spontaneous action potentials produced by spontaneous EPSPs was reduced significantly (15.3 ± 11.1% of baseline; P o 0.01, n ¼ 6). Thus, after TBF, the efficacy of background EPSPs to evoke action potentials was reduced, consistent with a global reduction in excitability. Further examination of the evoked EPSP profile indicated that the half-width was significantly reduced after TBF (29.56 ± 1.52 ms versus 35.72 ± 1.79 ms, percentage of reduction, 17.0 ± 2.9%; P o 0.01, n ¼ 6). This is in agreement with the reduction in temporal summation assessed with a function current injection into the soma after TBF (see Fig. 4f). Increase in Ih requires CaMKII and protein synthesis In a final series of experiments, we explored the signal transduction pathways involved in the TBF-induced upregulation of Ih. Bath application of the mitogen-activated protein kinase (MAPK) inhibitor U0126 (10 mM) did not block the TBF-induced change in excitability (P o 0.05, n ¼ 4; Fig. 8a,b). The CaMKII inhibitors KN-62 (3 mM) and KN-93 (10 mM) applied for at least 20 min before delivering the TBF prevented the change in excitability (n ¼ 4 and 5, respectively; Fig. 8a,b). KN-92 (10 mM), the inactive analog of KN-93, had no effect on the TBF-induced decrease in excitability (n ¼ 6; Fig. 8a,b). KN-62 and KN-93 are similar molecules, both interfering with the binding of CaM to CaMKII. We therefore verified the role of CaMKII using a mechanistically different reagent, CaMKII inhibitory peptide 281–309 (AIP), which instead inhibits the autophosphorylation of CaMKII. Intracellular application of this peptide (10 mM) prevented the TBF-induced change in excitability (n ¼ 4; Fig. 8a,b). These experiments strongly suggest an essential role for CaMKII activation

1547

ARTICLES

5 mV 50 ms

Figure 7 Spontaneous glutamate release was required for TBF-induced decrease in excitability. (a) Sample recordings showing the blocking effect of o-conotoxin MVIIC on spontaneous EPSPs. (b) Sample recordings showing the blocking effect of o-conotoxin MVIIC on evoked EPSPs (arrow) without any effect on the membrane input resistance. (c) Summary data showing that there was 91.6 ± 4.4% block of evoked EPSP by o-conotoxin MVIIC. (d) Sample trace showing the voltage deflection in response to subthreshold current steps before and after TBF, in the presence of o-conotoxin MVIIC. (e) Summary plot of RN profile before and after TBF, in the presence of o-conotoxin MVIIC. RN remained stable after TBF. (f) Summary plot showing the stability of temporal summation ratio before and after TBF, in the presence of o-conotoxin MVIIC.

a

1548

40 20

5 mV 200 ms

0

e 100

f

Pre-TBF 30 min post TBF

80 60 40 20 0

b

80

*

60

* 40 20 0 U0126 KN-62 KN-93 AIP

c

e Figure 8 CaMKII and de novo protein synthesis were required for the TBPand TBF-induced decrease in excitability. (a) Blocking CaMKII activation prevented the TBF-induced decrease in RN (P o 0.05). (b) Blocking CaMKII activation prevented the TBF-induced decrease in temporal summation ratio (P o 0.05). (c) Pretreatment with anisomycin (20 mM) blocked the TBFinduced decrease in RN and aEPSP summation ratio. (d) Pretreatment with anisomycin did not prevent TBP-induced LTP. (e) Anisomycin pretreatment blocked the TBP-induced decrease in RN. (f) Anisomycin disrupted the linear relationship between RN reduction and EPSP slope potentiation after TBP.

60


100 Input resistance (MΩ)

Upregulation of HCN1 subunits by extracellular high K+ The finding that TBF-induced upregulation of Ih could be blocked by anisomycin prompted us to test whether new Ih channel protein production might be involved. Because it is technically difficult to address this issue on the single-cell level, a compatible induction procedure that produces effects in many cells was needed. We found that a 5-min bath application of 10 mM KCl (in ACSF) caused a marked increase in background spontaneous synaptic activity (Supplementary Fig. 6). Notably, at 30 min after the initial episode of high K+ treatment, we observed a significant decrease in neuronal input resistance (39.3 ± 6.0%, n ¼ 3), a 10.8 ± 1.0 mV depolarization of

80

Pre-TBF 60

30 min post-TBF

50 40 30 20 10 0

resting potential and an apparent increase in voltage sag (Supplementary Fig. 6). These results resemble those observed in TBP/TBF experiments, suggesting a possible common mechanism. In a separate set of experiments, the amount of HCN1 and HCN2 proteins in the CA1 region were examined between control slices and KCl-treated slices using immunoblotting analysis. A significant increase in HCN1, but not HCN2, protein level was detected in the KCl-treated groups (HCN1: 118.0 ± 6.3% of control; P o 0.05, n ¼ 11 pairs; HCN2: 101.1 ± 12.4% of control, n ¼ 8 pairs, Supplementary Fig. 7). These results suggest that HCN1 and HCN2 subunits, two main components of the native Ih channels in the hippocampal CA1 cells31, may be differentially modulated by enhanced synaptic activities. Further, NMDA receptor blockers (AP5 + MK801) significantly suppressed

Input resistance (MΩ) α EPSP summation ratio (%)

and autophosphorylation in mediating the reduction of excitability after TBF. It is not known whether the persistent change in excitability requires de novo protein synthesis. We therefore performed additional experiments in which the brain slices were incubated in artificial cerebrospinal fluid (ACSF) containing anisomycin (20 mM), a specific protein translational inhibitor, for at least 40 min before recording and stimulation. We did not find any significant change in excitability after TBF (RN: 78.3 ± 10.3 MO versus 77.4 ± 5.3 MO; temporal summation ratio: 29.40 ± 6.71% versus 28.65 ± 2.76%; n ¼ 9, Fig. 8c), suggesting that protein translation was necessary to sustain a persistent change in Ih. This result prompted us to ask whether the LTP induced with the TBP protocol might also be dependent on protein synthesis. In a similar series of experiments with anisomycin but now with the TBP protocol, we found that there was no effect on LTP, but the change in input resistance and excitability was blocked (Fig. 8d–f). These results argue that protein synthesis is required for the upregulation of Ih but not for this form of early LTP.

100


1 mV 500 ms

Baseline 10 µM ω-CTX MVIIC



–65 mV


120


Evoked

75

75 50 25 0

Input resistance

α EPSP

*

Control

VOLUME 8

[

30 20

*

*

10 0 U0126 KN-62 KN-93 AIP

KN-92

Control 20 µM anisomycin

6 4 2 0

10

20 Time (min)

30

40

f

25 0

40

0

Baseline Post-TBP

50

50

d

Baseline Post-TBF

100


60

KN-92

Normalized EPSP slope

10 µM ω-CTX MVIIC

d

Baseline ω-CTX MVIIC

Normalized input resistance

Baseline

c Normalized evoked EPSP amplitude (%)

b

Spontaneous


a

Anisomycin

NUMBER 11

[

Control 20 µM anisomycin 1.00 0.75 0.50 0.25 1.5

2.0

2.5 3.0 3.5 4.0 4.5 Normalized EPSP slope

5.0


ARTICLES


the level of HCN1 increase after high K+ treatment (percentage of suppression: 34.2 ± 14.1%; P o 0.05, n ¼ 12 pairs), reinforcing the idea that an increase in NMDA activation can lead to an increase in HCN1 protein level. DISCUSSION Although activity-dependent changes in neural circuits have been traditionally assumed to be mediated by synaptic plasticity, increasing evidence suggests that neuronal intrinsic excitability may also be modulated by activity4,8,32. Such modulation is common in the developing brain6 and occurs during behavioral learning tasks in adult animals33. Adding to the growing knowledge about this type of plasticity, we demonstrate here an NMDA receptor-dependent decrease in input resistance and overall excitability after LTP induction using a theta-burst pairing protocol. This change is consistent with an upregulation of Ih and leads to a long-lasting decrease in global intrinsic excitability that is temporally correlated with the synaptic enhancement. Our data, however, also show that Ih upregulation is independent of LTP per se, and instead results from the postsynaptic burst firing at theta frequencies and the level of evoked and/or spontaneous synaptic input. The Ih conductance is evidently a dynamically regulated homeostatic effector for neuronal excitability. Recently, several other groups have reported changes in excitability in hippocampal and cortical pyramidal neurons after the induction of LTP8,12,14–16. In particular, a localized decrease in Ih (ref. 12) and a decrease in action potential threshold and Na+ channel activation voltage14 have been reported with LTP induction. A similar increase in excitability was also found in layer V pyramidal neurons in the visual cortex8. Although these groups did not report the more global increase in Ih reported here, there are several important differences in experimental procedures that may account for this. For example, these groups used relatively young animals (B2–3 weeks old in those experiments, versus B5–7 weeks old in our experiments). The age of experimental animals is likely to be an important factor when investigating Ih, because of its late postnatal development34–36. In particular, it has been shown that a mature HCN1–4 mRNA expression profile is not established until the end of the third postnatal week in rat hippocampus35. Another factor may be recording temperature. Our experiments were done at B34 1C, rather than room temperature. The level of spontaneous synaptic activity and the amplitudes and shapes of backpropagating action potentials are known to differ significantly between room-temperature and higher-temperature conditions37 (L. Yuan, X. Chen, C. Bernard and D.J. unpublished data). Also, recordings by one group were performed in the presence of synaptic blockers8. We have presented evidence here that even in the absence of synaptic stimulation, NMDA receptors are required for the upregulation of Ih. Finally, small differences in the induction protocols among the studies may also have contributed to the differing results. It should be noted, however, that the change in voltage threshold we observed in the potentiated input (Fig. 2f) may be explained by the change in spike threshold and Na+ channel properties reported previously8,14. Ih upregulation reduces overall excitability EPSP duration, input-output firing properties and temporal summation of CA1 pyramidal neurons are altered after TBP or TBF. We propose that these changes depend on upregulation of Ih. The primary consequence of this change is to increase resting membrane conductance. Additionally, increased Ih will speed the decay rate of EPSPs and reduce temporal summation38,39, reduce the depolarization produced by a given excitatory input and reduce EPSP efficacy to initiate action potential firing. Notably, a new anticonvulsant drug, lamotrigine,


[

NUMBER 11

[

NOVEMBER 2005

achieves its therapeutic effect in part by increasing the Ih conductance22,40, thereby reducing the responses of CA1 pyramidal neurons to all their inputs. Mechanisms for upregulation of Ih The upregulation of Ih depends on postsynaptic theta-burst firing concurrent with postsynaptic NMDA receptor activation. The latter may result from either spontaneously released glutamate (as occurred in the TBF condition) and/or evoked release (as in the TBP condition). Action potentials initiated near the soma can actively backpropagate into the dendritic arbor of pyramidal neurons27, providing strong postsynaptic depolarization and subsequent Ca2+ influx41. Cytoplasmic free Ca2+ regulates many critical cellular functions, including neurotransmitter release, gene transcription and channel modulation42. We have shown that Ca2+ influx through NMDA receptors—but not through Ca2+-permeable AMPA receptors, T-type or L-type voltagegated Ca2+ channels—is critical for the upregulation of Ih. In our experiments, backpropagating action potentials during TBP and TBF provide sufficient depolarization to activate NMDA receptors, bypassing the need for AMPA receptor activation. Consistent with the finding that NMDA but not AMPA receptor activation is necessary for Ih upregulation, the extent of input resistance decrease in LTP is larger than that in TBF. This can be explained by the fact that there is more glutamate release from evoked (TBP) versus spontaneous (TBF) synaptic activity. A similar coupling between enhanced synaptic glutamatergic inputs and upregulation of Ih has also been described in CA1 neurons13. In addition to NMDA receptors, however, those authors also proposed a role for AMPA receptors and voltage-gated Ca2+ channels. The difference may be due to the TTX (1 mM), which was continuously present in their experiments, eliminating the possibility of action potential generation and backpropagation. How can a transient intracellular Ca2+ rise due to NMDA receptor activation lead to a persistent modification of Ih conductance? There are several Ca2+-dependent protein kinases (PKA, PKC and CaMKII); our data indicate that activation of CaMKII is important in the modulation of excitability. CaMKII is likely to be a target of NMDAdependent increases in postsynaptic Ca2+, because it is highly enriched in the postsynaptic densities (PSD) of hippocampal excitatory synapses43. Upon activation of NMDA receptors, CaMKII translocates from the cytosol to PSD44. CaMKII then autophosphorylates on Thr286, rendering the enzyme calcium independent. Persistent activation of CaMKII is also thought to strengthen synaptic transmission (that is, LTP) through direct phosphorylation of AMPA receptors and addition of AMPA receptors to synapses45. It remains to be determined whether CaMKII directly modifies h-channel subunits, activates downstream signaling molecules or indirectly activates protein synthesis machinery that leads to an increase in surface expression of new h-channels. Although there has not been any study directly testing the activity-dependent trafficking of h-channels in neuronal plasticity, surface expression of Ih has recently been shown to be regulated through protein-protein interaction at the HCN C terminus by proteins that are known to be important in vesicle trafficking46 (A.E. Cuadra et al., Soc. Neurosci. Abstr. 52.1, 2004). It will be interesting to test whether inhibiting membranous fusion machinery can block the effects of TBP and TBF. The marked and long-lasting reduction of input resistance observed in our study could result from a change in Ih activation kinetics, from insertion of spare or newly synthesized h-channel into the membrane, or both. In this context, it is intriguing that long-lasting upregulation of Ih requires protein synthesis. Indeed, the protein levels of HCN1 (the main isoform of h-channels in CA1 pyramidal neurons) can be

1549

ARTICLES


significantly upregulated within 30 min by increasing NMDA receptor activation with high K+, as indicated by our western blot data. Owing to its dependence on NMDA receptor activation, the induction of the Ih upregulation would seem to be a synaptic event occurring in the dendrites. However, the site of its expression is not clear. The observed reduction in excitability could result from an increase in Ih conductance in the perisomatic area, or in a combination of perisomatic and dendritic regions. Functional implications of regulation of excitability LTP was first postulated in Donald Hebb’s original learning rule, describing the strengthening of a synapse in response to correlated activity47. To maintain modifiable synapses of a neuronal network within their optimal working range, the positive feedback mechanism of Hebbian plasticity must be constrained by regulatory processes. One proposed homeostatic mechanism suggests that the synaptic modification threshold, at which LTP switches to LTD, is dynamically regulated by the level of postsynaptic activity48. With a sliding modification threshold dependent on the average firing rate, this model addresses two problems of Hebbian plasticity: it both introduces competition between synapses and also acts to normalize synaptic weights. As we have shown, strengthening some synapses does not necessarily increase postsynaptic firing. The enhancement of Ih changes dendritic integration so that synaptic potentiation may be accompanied by a reduction of overall spike output of the neuron. This contributes to maintaining neuronal output at a similar mean level after LTP, independent of synaptic normalization. This plasticity of intrinsic cellular excitability may permit a global homeostatic regulation of activity, preventing hyperexcitability and extending the computational capabilities of a neuron. Postsynaptic h-channel–dependent dampening of excitability together with enhanced presynaptic neuronal excitability after correlated presynaptic and postsynaptic spiking49 may act synergistically to enhance synapse specificity. The increased precision of spike timing after synaptic potentiation may promote the ability of pyramidal cells to detect coincident EPSPs50. It will be interesting to test this prediction experimentally as well as to examine how it contributes to learning and memory in the behaving animal. METHODS Electrophysiology. Transverse hippocampal slices (350 mM) were prepared from 5- to 7-week-old male Sprague-Dawley rats using standard procedures5 that were approved by the Baylor College of Medicine Institutional Animal Research Committee. Recordings were made at 31–34 1C. The external recording solution contained 125 mM NaCl, 2.5 mM KCl, 2 mM CaCl2, 1 mM MgCl2, 1.25 mM NaH2PO2, 25 mM NaHCO3 and 10 mM dextrose, bubbled with 95% O2/5% CO2. For LTP experiments, bicuculline (10 mM) and picrotoxin (10 mM) were added to the bath to block GABAA receptor–mediated synaptic inhibition. For whole-cell recording, the electrodes were filled with an internal solution containing 120 mM potassium gluconate, 20 mM KCl, 2 mM MgCl2, 10 mM HEPES, 4 mM Na2ATP, 0.3 mM Tris-GTP and 14 mM phosphocreatine (pH 7.3, adjusted with KOH). All drugs were made from stock solutions in either water or DMSO (final concentration of DMSO, r0.1%) accordingly. ZD7288; D,L-AP5; MK-801; nimodipine; CNQX; bicuculline; picrotoxin and TTX were obtained from Tocris Cookson. Kynurenic acid, BAPTA and anisomycin were obtained from Sigma. KN-62, KN-93, KN-92 and CaMKII inhibitory peptide 281–309 were obtained from Calbiochem. U0126 was obtained from Promega. Patch pipettes (4–8 MO) were pulled from borosilicate glass of external diameter 2.0 mm (Garner) using a Flaming/Brown electrode puller (model P-97, Sutter Instruments) to produce a uniform tip size of B1 mm in diameter. Hippocampal CA1 pyramidal neurons were visualized with a Zeiss Axioskop 2 equipped with 60 objective, using infrared video microscopy and differential interference contrast optics. Whole-cell current clamp recordings were

1550

made using a Dagan BVC-700 amplifier (Dagan Instruments) in active ‘bridge’ mode. Signals were low-pass analog filtered at 3 KHz, digitized with an ITC-18 (Instrutech) computer interface and custom acquisition software (IgorPro, Wavemetrics) at 5–20 KHz, and stored on the computer hard disk for offline analysis. Series resistance was constantly monitored, and cells in which the series resistance exceeded 30 MO were discarded. Only cells with a resting membrane potential of –60 mV or more hyperpolarized were used for further recording. A tungsten bipolar stimulating electrode was placed in stratum radiatum (120–150 mm away from the soma) to elicit EPSPs. Experiments where EPSP slope was not stable during the 5-min baseline period were discarded. Backpropagating action potentials were elicited by direct somatic current injection (2 ms, 1–2 nA). Simulated EPSPs were modeled by an a function of the form I ¼ Imax (t/a) e–at (ref. 22), a ¼ 0.1. LTP was induced using the TBP procedure previously described5,19, unless otherwise stated. Briefly, EPSPs were paired with single backpropagating action potentials timed so that action potentials occurred at the peak of the EPSP. A single burst contained five pairs delivered at 100 Hz and ten bursts were delivered at 5 Hz per sweep. Three sweeps were delivered at 10-s intervals for a total of 30 bursts (150 backpropagating action potential–EPSP pairs). For 100 Hz LTP, the protocol consisted of 150 synaptic stimuli at 100 Hz. For TBF experiments, the protocol was identical to TBP, except that there was no synaptic stimulation. Calcium imaging. For Ca2+ imaging experiments, the internal solution contained the fluorescent calcium indicator bis-fura-2 (100 mM). After establishment of whole-cell configuration, the dye was allowed to dialyze into the cell for at least 20 min and reach equilibrium. Somatic current-clamp recordings were then performed in combination of Ca2+ imaging. The imaging and electrical recordings were controlled through the same Igor software (IgorPro, Wavemetrics) and synchronized with each other. Ca2+ signals were recorded at 60 Hz. The dye was excited with 380 nm light via a Xenon lamp and a 380/13 excitation filter. The emission light was collected through a 495-nm high-pass filter with a Quantix cooled charge-coupled device (CCD) camera (Roper Scientific). Relative changes in internal calcium concentration ([Ca2+]in) were quantified as changes in DF/F, where F is fluorescence intensity before stimulation (after subtracting autofluorescence) and DF is the change from this value during firing activity (corrected for bleaching during the run). An increase of [Ca2+]in was reflected by a decrease of DF/F, which was flipped in the figures for display. High extracellular K+ stimulation and western blots. Hippocampal slices (350 mm thick) from the same rats were evenly divided into two groups according to experimental purposes. For high-K+ treatment, slices were incubated for 5 min in artificial cerebrospinal fluid (ACSF) containing 10 mM KCl. The treated slices were then placed in normal ACSF (the same as external recording solution) for B30 min. In pharmacological block experiments, D,L-APV (50 mM) and MK-801 (10 mM) were delivered together with 10 mM KCl. The hippocampal CA1 region of both groups was then subdissected in ice-cold saline and immediately frozen on dry ice. CA1 tissue was homogenized, and 30 mg of total protein for each lane was separated on SDS-PAGE gels and electrophoretically transferred to nitrocellulose membranes. The following antibodies were used: anti-actin (rabbit polyclonal antibody, Sigma), HCN-1 and HCN-2 (rabbit polyclonal antibodies, Chemicon) at 1:750 dilutions. The signals were visualized with horseradish peroxidase conjugated goat anti-rabbit secondary antibody (Pierce Biotechnology) at 1:10,000 dilutions. Signals were detected and quantified with enhanced chemiluminescent reactions (Pierce) and densitometric quantification with US National Institutes of Health (NIH) Image software (Scion). Bands corresponding to the full-length HCN1 (117 kDa) and HCN2 (97 kDa) subunits were densitized using NIH software and normalized to actin (42 kDa) immunoreactivity. Data analysis. All data analysis was performed using Igor software (IgorPro, Wavemetrics) and StatView (SAS Institute) for statistic comparisons. EPSP strength was measured by a linear fit to the initial slope of 3–4 averaged EPSPs. Magnitude of LTP was determined as the percentage of baseline in EPSP slope 30 min after TBP. Input resistance measurement was made over the steady-state

VOLUME 8

[

NUMBER 11

[



ARTICLES voltage amplitudes (the last 50 ms of the responses to a family of 700-ms current steps). Input resistance was determined by the slope of the linear regression line through the I-V plot (constructed by plotting the amplitude of the steady-state voltage against the corresponding current injection amplitude (–100 pA to +100 pA, 700–750 ms at 20- or 25-pA intervals)). Input-output curves were constructed by plotting the numbers of action potentials against the amplitude of the current injections (+100 pA to +400 pA, for 500–700 ms at 50-pA intervals). Temporal summation ratio was measured as the amplitude of the fifth aEPSP relative to first in a 20-Hz train of 5 aEPSPs ((aEPSP5– aEPSP1)/ aEPSP1; ref. 22). Group data were reported as mean ± s.e.m. Statistical analysis was performed using ANOVA, Student’s t-test, MannWhitney U test accordingly, and differences were considered significant if P o 0.05. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank R. Gray, N. Poolos and A. Frick for discussions and help with data acquisition and analysis software. We also thank W.S. She for technical help with western blotting. This work was supported by US NIH grants MH48432, MH44754 and NS37444 (D.J.) and NS48884 and AHA0465158Y (H.-C.L.); a fellowship from the American Heart Association (Y.F.); and a fellowship from the North Atlantic Treaty Organization (D.F.). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. Bliss, T.V. & Collingridge, G.L. A synaptic model of memory: long-term potentiation in the hippocampus. Nature 361, 31–39 (1993). 2. Andersen, P., Sundberg, S.H., Sveen, O., Swann, J.W. & Wigstrom, H. Possible mechanisms for long-lasting potentiation of synaptic transmission in hippocampal slices from guinea-pigs. J. Physiol. (Lond.) 302, 463–482 (1980). 3. Bliss, T.V. & Lomo, T. Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path. J. Physiol. (Lond.) 232, 331–356 (1973). 4. Zhang, W. & Linden, D.J. The other side of the engram: experience-driven changes in neuronal intrinsic excitability. Nat. Rev. Neurosci. 4, 885–900 (2003). 5. Frick, A., Magee, J. & Johnston, D. LTP is accompanied by an enhanced local excitability of pyramidal neuron dendrites. Nat. Neurosci. 7, 126–135 (2004). 6. Spitzer, N.C. A developmental handshake: neuronal control of ionic currents and their control of neuronal differentiation. J. Neurobiol. 22, 659–673 (1991). 7. Turrigiano, G.G. & Nelson, S.B. Homeostatic plasticity in the developing nervous system. Nat. Rev. Neurosci. 5, 97–107 (2004). 8. Cudmore, R.H. & Turrigiano, G.G. Long-term potentiation of intrinsic excitability in LV visual cortical neurons. J. Neurophysiol. 92, 341–348 (2004). 9. Desai, N.S., Rutherford, L.C. & Turrigiano, G.G. Plasticity in the intrinsic excitability of cortical pyramidal neurons. Nat. Neurosci. 2, 515–520 (1999). 10. Turrigiano, G., Abbott, L.F. & Marder, E. Activity-dependent changes in the intrinsic properties of cultured neurons. Science 264, 974–977 (1994). 11. Aizenman, C.D. & Linden, D.J. Rapid, synaptically driven increases in the intrinsic excitability of cerebellar deep nuclear neurons. Nat. Neurosci. 3, 109–111 (2000). 12. Wang, Z., Xu, N.L., Wu, C.P., Duan, S. & Poo, M.M. Bidirectional changes in spatial dendritic integration accompanying long-term synaptic modifications. Neuron 37, 463– 472 (2003). 13. van Welie, I., van Hooft, J.A. & Wadman, W.J. Homeostatic scaling of neuronal excitability by synaptic modulation of somatic hyperpolarization-activated Ih channels. Proc. Natl. Acad. Sci. USA 101, 5123–5128 (2004). 14. Xu, J., Kang, N., Jiang, L., Nedergaard, M. & Kang, J. Activity-dependent long-term potentiation of intrinsic excitability in hippocampal CA1 pyramidal neurons. J. Neurosci. 25, 1750–1760 (2005). 15. Daoudal, G., Hanada, Y. & Debanne, D. Bidirectional plasticity of excitatory postsynaptic potential (EPSP)-spike coupling in CA1 hippocampal pyramidal neurons. Proc. Natl. Acad. Sci. USA 99, 14512–14517 (2002). 16. Marder, C.P. & Buonomano, D.V. Timing and balance of inhibition enhance the effect of long-term potentiation on cell firing. J. Neurosci. 24, 8873–8884 (2004). 17. Roberson, E.D., English, J.D. & Sweatt, J.D. A biochemist’s view of long-term potentiation. Learn. Mem. 3, 1–24 (1996). 18. Watanabe, S., Hoffman, D.A., Migliore, M. & Johnston, D. Dendritic K+ channels contribute to spike-timing dependent long-term potentiation in hippocampal pyramidal neurons. Proc. Natl. Acad. Sci. USA 99, 8366–8371 (2002).


[

NUMBER 11

[

NOVEMBER 2005

19. Magee, J.C. & Johnston, D. A synaptically controlled, associative signal for Hebbian plasticity in hippocampal neurons. Science 275, 209–213 (1997). 20. Buzsaki, G. Theta oscillations in the hippocampus. Neuron 33, 325–340 (2002). 21. Thomas, M.J., Watabe, A.M., Moody, T.D., Makhinson, M. & O’Dell, T.J. Postsynaptic complex spike bursting enables the induction of LTP by theta frequency synaptic stimulation. J. Neurosci. 18, 7118–7126 (1998). 22. Poolos, N.P., Migliore, M. & Johnston, D. Pharmacological upregulation of h-channels reduces the excitability of pyramidal neuron dendrites. Nat. Neurosci. 5, 767–774 (2002). 23. Shah, M.M., Anderson, A.E., Leung, V., Lin, X. & Johnston, D. Seizure-induced plasticity of h channels in entorhinal cortical layer III pyramidal neurons. Neuron 44, 495–508 (2004). 24. Maccaferri, G., Mangoni, M., Lazzari, A. & DiFrancesco, D. Properties of the hyperpolarization-activated current in rat hippocampal CA1 pyramidal cells. J. Neurophysiol. 69, 2129–2136 (1993). 25. Shin, K.S., Rothberg, B.S. & Yellen, G. Blocker state dependence and trapping in hyperpolarization-activated cation channels: evidence for an intracellular activation gate. J. Gen. Physiol. 117, 91–101 (2001). 26. Magee, J.C. Dendritic hyperpolarization-activated currents modify the integrative properties of hippocampal CA1 pyramidal neurons. J. Neurosci. 18, 7613–7624 (1998). 27. Johnston, D., Magee, J.C., Colbert, C.M. & Christie, B.R. Active properties of neuronal dendrites. Annu. Rev. Neurosci. 19, 165–186 (1996). 28. Schiller, J., Schiller, Y. & Clapham, D.E. NMDA receptors amplify calcium influx into dendritic spines during associative pre- and postsynaptic activation. Nat. Neurosci. 1, 114–118 (1998). 29. Mackenzie, P.J. & Murphy, T.H. High safety factor for action potential conduction along axons but not dendrites of cultured hippocampal and cortical neurons. J. Neurophysiol. 80, 2089–2101 (1998). 30. Magee, J.C. & Carruth, M. Dendritic voltage-gated ion channels regulate the action potential firing mode of hippocampal CA1 pyramidal neurons. J. Neurophysiol. 82, 1895–1901 (1999). 31. Robinson, R.B. & Siegelbaum, S.A. Hyperpolarization-activated cation currents: from molecules to physiological function. Annu. Rev. Physiol. 65, 453–480 (2003). 32. Nolan, M.F. et al. A behavioral role for dendritic integration: HCN1 channels constrain spatial memory and plasticity at inputs to distal dendrites of CA1 pyramidal neurons. Cell 119, 719–732 (2004). 33. Saar, D. & Barkai, E. Long-term modifications in intrinsic neuronal properties and rule learning in rats. Eur. J. Neurosci. 17, 2727–2734 (2003). 34. Vasilyev, D.V. & Barish, M.E. Postnatal development of the hyperpolarization-activated excitatory current Ih in mouse hippocampal pyramidal neurons. J. Neurosci. 22, 8992– 9004 (2002). 35. Bender, R.A. et al. Differential and age-dependent expression of hyperpolarizationactivated, cyclic nucleotide-gated cation channel isoforms 1–4 suggests evolving roles in the developing rat hippocampus. Neuroscience 106, 689–698 (2001). 36. Zhu, J.J. Maturation of layer 5 neocortical pyramidal neurons: amplifying salient layer 1 and layer 4 inputs by Ca2+ action potentials in adult rat tuft dendrites. J. Physiol. (Lond.) 526, 571–587 (2000). 37. Yuan, L.L., Adams, J.P., Swank, M., Sweatt, J.D. & Johnston, D. Protein kinase modulation of dendritic K+ channels in hippocampus involves a mitogen-activated protein kinase pathway. J. Neurosci. 22, 4860–4868 (2002). 38. Magee, J.C. Dendritic Ih normalizes temporal summation in hippocampal CA1 neurons. Nat. Neurosci. 2, 508 (1999). 39. Berger, T., Larkum, M.E. & Luscher, H.R. High I(h) channel density in the distal apical dendrite of layer V pyramidal cells increases bidirectional attenuation of EPSPs. J. Neurophysiol. 85, 855–868 (2001). 40. Berger, T. & Luscher, H.R. Associative somatodendritic interaction in layer V pyramidal neurons is not affected by the antiepileptic drug lamotrigine. Eur. J. Neurosci. 20, 1688–1693 (2004). 41. Frick, A., Magee, J., Koester, H.J., Migliore, M. & Johnston, D. Normalization of Ca2+ signals by small oblique dendrites of CA1 pyramidal neurons. J. Neurosci. 23, 3243– 3250 (2003). 42. Berridge, M.J. Neuronal calcium signaling. Neuron 21, 13–26 (1998). 43. Erondu, N.E. & Kennedy, M.B. Regional distribution of type II Ca2+/calmodulindependent protein kinase in rat brain. J. Neurosci. 5, 3270–3277 (1985). 44. Shen, K. & Meyer, T. Dynamic control of CaMKII translocation and localization in hippocampal neurons by NMDA receptor stimulation. Science 284, 162–166 (1999). 45. Lisman, J., Schulman, H. & Cline, H. The molecular basis of CaMKII function in synaptic and behavioural memory. Nat. Rev. Neurosci. 3, 175–190 (2002). 46. Santoro, B., Wainger, B.J. & Siegelbaum, S.A. Regulation of HCN channel surface expression by a novel C-terminal protein-protein interaction. J. Neurosci. 24, 10750– 10762 (2004). 47. Bi, G. & Poo, M. Synaptic modification by correlated activity: Hebb’s postulate revisited. Annu. Rev. Neurosci. 24, 139–166 (2001). 48. Bienenstock, E.L., Cooper, L.N. & Munro, P.W. Theory for the development of neuron selectivity: orientation specificity and binocular interaction in visual cortex. J. Neurosci. 2, 32–48 (1982). 49. Ganguly, K., Kiss, L. & Poo, M. Enhancement of presynaptic neuronal excitability by correlated presynaptic and postsynaptic spiking. Nat. Neurosci. 3, 1018–1026 (2000). 50. Pouille, F. & Scanziani, M. Enforcement of temporal fidelity in pyramidal cells by somatic feed-forward inhibition. Science 293, 1159–1163 (2001).

1551

CO R R I G E N DA

Corrigendum: Protocadherin Celsr3 is crucial in axonal tract development


Fadel Tissir, Isabelle Bar, Yves Jossin, & André M Goffinet Nat. 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 tract Yaobo Liu, Jun Shi, Chin-Chun Lu, Zheng-Bei Wang, Anna I Lyuksyutova, Xuejun Song & Yimin Zou Nat. 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 Ih Yuan Fan, Desdemona Fricker, Darrin H Brager, Xixi Chen, Hui-Chen Lu, Raymond A Chitwood & Daniel Johnston Nat. 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.

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 1 | JANUARY 2006

147


ARTICLES

Fine-scale specificity of cortical networks depends on inhibitory cell type and connectivity Yumiko Yoshimura1,2 & Edward M Callaway1 Excitatory cortical neurons form fine-scale networks of precisely interconnected neurons. Here we tested whether inhibitory cortical neurons in rat visual cortex might also be connected with fine-scale specificity. Using paired intracellular recordings and cross-correlation analyses of photostimulation-evoked synaptic currents, we found that fast-spiking interneurons preferentially connected to neighboring pyramids that provided them with reciprocal excitation. Furthermore, they shared common fine-scale excitatory input with neighboring pyramidal neurons only when the two cells were reciprocally connected, and not when there was no connection or a one-way, inhibitory-to-excitatory connection. Adapting inhibitory neurons shared little or no common input with neighboring pyramids, regardless of their direct connectivity. We conclude that inhibitory connections and also excitatory connections to inhibitory neurons can both be precise on a fine scale. Furthermore, fine-scale specificity depends on the type of inhibitory neuron and on direct connectivity between neighboring pyramidal-inhibitory neuron pairs.

The precision of cortical connections is manifested at many different levels of organization. The cerebral cortex is parceled into functionally and anatomically distinct areas, each of which connects to distinct subsets of the other areas1. And within each area, specific connections create and respect laminar and columnar functional architecture2–8. Within any one of these laminar or columnar modules there are numerous inhibitory and excitatory neuron types whose dendritic and axonal arbors are intimately entangled. Yet, even at this level of organization, connections remain specific such that axonal arbors that terminate in a given region selectively connect to some cell types while avoiding others9–21. Thus, neighboring neurons of different types receive input from different cortical layers. Neighboring neurons of the same type, however, tend to receive local input from the same cortical layers. Most recently, it has been found that even neighboring neurons of the same anatomical type can, nevertheless, receive input from different sources22,23. This ‘fine-scale’ specificity has been demonstrated using methods that allow differences in input to neighboring neurons to be probed with high resolution. For example, by simultaneously recording from (and stimulating) three or more neighboring layer 5 pyramidal neurons, it was found that when two neurons are connected to each other, the probability that they will share input from a third neuron is higher than expected in a randomly connected network22. In previous work, we combined simultaneous recordings from pairs of layer 2/3 pyramidal neurons with laser scanning photostimulation23. This stimulation method evokes action potentials in a small, spatially confined population of neurons and can be repeated rapidly at many different stimulation sites. As the population of neurons that is stimulated fires action potentials relatively asynchronously, cross-correlation analyses

of the synaptic currents recorded in neuron pairs can be used to determine whether they share common input on a fine spatial scale. This study showed that fine-scale specificity of inputs to layer 2/3 pyramidal neurons depends on the laminar source of the excitatory input and differs for excitatory versus inhibitory inputs. Layer 2/3 pyramids share common input from layer 4 and from within layer 2/3 only in the minority of cases in which they are directly connected to each other. But they share common inhibitory input and excitatory input from layer 5 regardless of their direct connectivity. The excitatory neurons within layers 4 and 2/3 therefore create preferentially connected subnetworks embedded within the laminar and columnar functional architecture. In this study, we investigated whether these fine-scale excitatory subnetworks might have precise relationships to inhibitory neurons. Can inhibitory connections be specific on a fine scale? Is there fine-scale specificity of excitatory connections to inhibitory neurons, and does it depend on the inhibitory cell type? Does the direct connectivity between neighboring inhibitory and excitatory neurons depend on whether they are part of the same fine-scale excitatory subnetwork? We found that fast-spiking inhibitory neurons in cortical layer 2/3 connect preferentially to neighboring pyramidal neurons that provide them with reciprocal excitation. We also found that these reciprocally connected fast-spiking interneuron–pyramidal cell pairs shared common fine-scale excitatory input and thus belonged to the same fine-scale subnetworks. Other cell pairs did not share common fine-scale excitatory input: these pairs included fast-spiking interneuron–pyramidal cell pairs with one-way or no connections and all pairs involving adapting neurons. These findings have implications for understanding how connections from different types

1Systems Neurobiology Laboratories, The Salk Institute for Biological Studies, 10010 North Torrey Pines Road, La Jolla, California 92037 USA. 2Department of Visual Neuroscience, Research Institute of Environmental Medicine, Nagoya University, Furo-cho, Chikusa-ku, Nagoya 464-8601, Japan. Correspondence should be addressed to E.M.C. ([email protected]).


1552

VOLUME 8

[

NUMBER 11

[


ARTICLES

a

b

50 µm

20 mV 50 ms 20 mV 50 ms

d

FS → P

P → FS

20 mV

50 pA 20 ms

e AD → P

P → AD

20 mV

20 mV

20 mV

50 pA 20 ms

10 pA 20 ms

f

10 pA 20 ms

IN → P one-way

Spike-evoked IPSCs

150 100 50 0

FS → P AD → P

P < 0.0001

2 1 0

75

(1)

25

(22)

(104)

FS&P

AD&P

2

40 30 20 10 0

FS → P AD → P

P → FS P → AD

0

RESULTS The methods used in this study were identical to those described previously23. Whole-cell recordings were made simultaneously from two neighboring layer 2/3 neurons in rat visual cortex slices. We then tested the two neurons to determine whether they were connected to each other by stimulating one of the neurons and measuring postsynaptic currents in the other. We then used focal uncaging of glutamate (photostimulation) to generate action potentials in small, spatially confined populations of neurons. Control experiments showed that only neurons within about 50 mm from the stimulation site fire action potentials23. We stimulated hundreds of sites spanning all cortical layers while recording excitatory postsynaptic currents (EPSCs) during voltage clamp. We then conducted cross-correlation analyses on the EPSCs recorded from the two cells. Shifted correlograms (from nonmatched stimulation trials) were subtracted from matched correlograms to determine the number of synchronous EPSCs that could be attributed to common input to the two neurons rather than stimulus-evoked time locking of action potentials in different presynaptic neurons23,24. These ‘shared’ EPSCs were then expressed as a percentage of all evoked EPSCs to calculate the ‘correlation probability’ (CP). The CP represents the proportion of all evoked EPSCs that were attributable to common input from the same presynaptic neurons. CPs were calculated separately for EPSCs evoked by photostimulation at sites in each of the cortical layers. These studies differ from previous studies only in that instead of targeting intracellular recordings to pairs of neighboring pyramidal

[

NUMBER 11

[

P → IN one-way Reciprocal P < 0.05

1

of inhibitory neurons mediate interactions within and across excitatory subnetworks.


No connection Pyramid → IN one-way IN → pyramid one-way Reciprocal

(13)

50

0

(5) (17) (12)

50

Amplitude (pA)

200

(7)

Spike-evoked EPSCs

Reciprocal 3

P < 0.001

Rise time (ms)

250

g

100

Rise time (ms)

c

Connection probability (%)

20 mV 50 ms

20 mV 50 ms

Amplitude (pA)


50 µm

NOVEMBER 2005

Figure 1 Direct connections between layer 2/3 inhibitory neuron–pyramidal neuron pairs. (a,b) Anatomical reconstructions and intrinsic firing properties for neuron pairs consisting of a simultaneously recorded pyramidal neuron (red dendrites) and inhibitory neuron (blue dendrites and grey axons). Intrinsic firing properties indicated that the inhibitory neuron in a was fastspiking (FS) and the inhibitory neuron in b was adapting (AD). Direct connectivity between neuron pairs was assessed by recording postsynaptic currents during voltage clamp in one cell while evoking action potentials during current clamp recording in the other cell (c,d). (c) Results from a reciprocally connected pyramidal neuron–FS interneuron pair. (d) Results for a pyramidal neuron–AD interneuron pair. Recordings from pyramids are red and from inhibitory neurons are blue. (e) Percentage of cell pairs of each type that were either not connected, connected one way from the interneuron to the pyramid, connected one way from the pyramid to the interneuron, or reciprocally connected. Numbers of cell pairs are indicated in parentheses. (f) Comparison of amplitudes and rise times (mean ± s.e.m.) of inhibitory postsynaptic currents (IPSCs) measured in pyramids after stimulation of FS interneurons or AD interneurons for pairs with one-way versus reciprocal connections. (g) Comparison of amplitudes and rise times (mean ± s.e.m) of EPSCs measured in FS interneurons or AD interneurons after stimulation of pyramids, for pairs with one-way versus reciprocal connections.

P → FS P → AD

neurons, differential interference contrast (DIC) optics were used to target recordings to one pyramidal neuron and one inhibitory interneuron (Fig. 1a,b). Inhibitory neurons were classified as either fast-spiking or adapting on the basis of their intrinsic firing properties in response to intracellular current injection (Fig. 1a,b; see Methods)18,25,26. ‘Fast-spiking’ interneurons are identified by the lack of spike rate adaptation, and many previous studies have shown that these are parvalbumin-expressing basket cells27. Adapting interneurons potentially include any of several other distinct cell types, but they are clearly distinguished from fast-spiking inhibitory neurons in terms of their morphology, physiology, gene expression and connectivity18,25–28. Fast-spiking interneuron–pyramid pairs Direct connections were tested between 43 layer 2/3 neuron pairs consisting of one pyramidal neuron and one fast-spiking inhibitory neuron (Fig. 1). (Photostimulation data were collected for 23 of these same cell pairs; see below.) For 51% of these cell pairs (22 of 43) there were no detectable connections between cells in either the inhibitory-toexcitatory or the excitatory-to-inhibitory direction. Most of the connections that were present in the remaining 21 cell pairs were inhibitory. In particular, fast-spiking inhibitory neurons had a 47% chance (20 of 43) of being connected to each neighboring pyramidal neuron, whereas a connection from the pyramidal neuron onto its fast-spiking neighbor was detected in only 19% of pairs (8 of 43). Notably, however, when there was an excitatory connection from the pyramid to the fast-spiking interneuron, there was almost always (7 of 8 pairs or 88%) a reciprocal connection from the inhibitory neuron to the pyramid. The probability of an inhibitory connection onto the pyramid was, therefore, highly dependent on whether the same cell pair also had an

1553

Pyramidal

FS

L2/3 1

L2/3

Matched Shifted CP = 0.19

50

L4

3

3

0

100 200 pA

Number of events

20 pA 20 ms

CP = 0.32

20

20

100 µm 250 500 750 pA

40

0 –40 –20 0 20 40

L5

L5

0

Number of events

0 –40 –20 0 20 40

L4

2

2

CP = 0.06

10 0

–40–20 0 20 40

b

FS

Pyramidal L2/3

1 L2/3

Number of events

Time (ms)

1

excitatory connection from the pyramid to the inhibitory neuron. In cell pairs without an excitatory connection, the probability that the inhibitory neuron would connect to the pyramid was only 37% (13 of 35 cases with no excitatory connection), but the probability of an inhibitory connection increased significantly, to 88% (7 of 8 cases with an excitatory connection) in cell pairs with an excitatory connection (P ¼ 0.017; Fisher Exact Test). These observations indicate that connections between fast-spiking interneurons and neighboring pyramidal neurons are highly specific and not random. Fast-spiking inhibitory neurons connect relatively promiscuously to their neighboring pyramids, yet they still maintain a strong preference for the minority of pyramidal neurons that provide their direct excitatory input. Further specificity was also reflected in the strength of the connections from fast-spiking interneurons. For the cases in which the connection from the interneuron to the pyramid was not reciprocated, the IPSC amplitudes averaged only 54.0 ± 6.0 pA (mean ± s.e.m.; Fig. 1f). When there was a reciprocal excitatory connection, however, the IPSC amplitudes were more than threefold larger—a significant increase (172.2 ± 37.9 pA; P o 0.001). Thus, not only are fast-spiking interneurons more likely to connect to neighboring pyramidal cells that reciprocate their connections, but these connections are also far stronger than unreciprocated connections.

1

100

Number of events

a

40


30 20 10 0 –40 –20 0 20 40

L4 2

L4

2

Number of events

Figure 2 Cross-correlation analyses of photostimulation-evoked EPSCs simultaneously recorded in adjacent layer 2/3 neuron pairs consisting of one pyramidal neuron and one fast-spiking inhibitory neuron (FS). (a) Reciprocally connected pair. (b) Pair with a one-way inhibitory connection. Plots at left of each panel show for each cell (FS or pyramidal) reconstructions of the locations of photostimulation sites (colored squares) relative to the locations of laminar borders and cell bodies of recorded neurons (triangles represent pyramids; circles represent inhibitory neurons). The color of each square indicates the sum of amplitudes of EPSCs that were observed after photostimulation at that site, as indicated by the colored scales beneath each plot. Traces at center of panels show example voltageclamp recordings for stimulation sites indicated by large numbered squares. Simultaneous recordings from the FS cell (blue) and the pyramidal cell (red) are shown for these representative stimulation sites. Short horizontal lines above each trace indicate the onset of photostimulation. Histograms to the far right of each panel show matched and shifted correlograms computed from data collected after stimulation in each layer. The corresponding CPs computed from these analyses are also indicated.

3

3

L5

100 µm 0

250 500 750 pA

0

100

200 pA

20 pA 20 ms

30

CP = 0.06

20 10 0 –40 –20 0 20 40

L5 Number of events


ARTICLES

15

CP = 0.03

10 5 0 –40 –20 0 20 40

Time (ms)

The size and kinetics of synaptic currents measured from direct connections between cell pairs also differed for adapting versus fastspiking interneurons. Unlike fast-spiking interneuron–pyramid pairs, for pairs involving adapting interneurons, there was no significant difference in IPSC amplitudes between pairs with one-way versus reciprocal connections (Fig. 1f). And the synaptic currents for fastspiking interneuron–pyramid pairs had significantly faster kinetics than those for pairs involving adapting interneurons. The rise times for IPSCs from fast-spiking interneurons were significantly shorter (1.7 ± 0.04 ms) than those from adapting interneurons (2.6 ± 0.03 ms; P o 0.0001), probably owing to the more proximal location of synapses from fast-spiking basket cells relative to those from some types of adapting interneurons27. The rise times for EPSCs onto fastspiking interneurons were also significantly shorter (0.95 ± 0.10 ms) than those onto adapting interneurons (1.5 ± 0.13 ms; P o 0.05), probably owing to the presence of glutamate receptors with faster kinetics at the synapses on fast-spiking interneurons28.

Adapting interneuron–pyramid pairs We tested direct connectivity between 138 neuron pairs consisting of one adapting inhibitory neuron and one pyramidal neuron (Fig. 1). (Photostimulation data were collected for 38 of these same cell pairs; see below). An excitatory connection from the pyramid to the interneuron was detected in 12% (17/138) of pairs, and an inhibitory connection from the interneuron to the pyramid was detected in 16% (22/138) of pairs. Reciprocal connections were rare, occurring in only 3.6% (5/138) of pairs. Unlike connections involving fast-spiking interneurons, for adapting interneurons, connections in one direction did not depend significantly on the presence of connections in the other direction. In particular, for the subset of 17 pairs in which there was a connection from the pyramid to the interneuron, it was reciprocated in 29% (5/17) of cases. This proportion does not vary significantly from the proportion of inhibitory connections in the pairs with no excitatory connection (14%, 17/121, P 4 0.1 Fisher Exact Test). Thus, adapting interneurons do not show the same preference as fast-spiking interneurons to connect preferentially to pyramids that provide them with excitatory input.

Shared input to FS interneuron–pyramid pairs depends on connectivity As detailed above, fast-spiking (FS) interneurons connected preferentially to neighboring pyramidal neurons that provided them with reciprocal excitatory connections. Here we show that these same reciprocally connected cell pairs shared excitatory input from the same neurons in layers 2/3 and 4, whereas unconnected pairs and pairs with one-way inhibitory connections did not. We combined laser scanning photostimulation with paired whole-cell recordings to assess fine-scale specificity of connections (Fig. 2). Figure 2a illustrates an example for a cell pair consisting of a fast-spiking interneuron and a reciprocally connected pyramidal neuron. Figure 2b shows an example

1554

VOLUME 8

[

NUMBER 11

[


ARTICLES

FS ←→ P

0.0 –0.1

Figure 3 Correlation probabilities for EPSCs measured simultaneously in fast-spiking interneuron–pyramidal neuron pairs. Correlation probabilities are shown separately for photostimulation sites in each cortical layer and for cell pairs that either were not connected (squares), had a one-way inhibitory connection (inverted triangles) or were reciprocally connected (circles). Cell pairs with one-way excitatory connections were not sampled using photostimulation. Mean values for each group are indicated by horizontal lines. Brackets highlight statistically significant differences between groups (*, P o 0.005; **, P o 0.0005; Mann-Whitney test).

from a cell pair consisting of a fast-spiking neuron whose connection to the pyramidal neuron was not reciprocated. In both cases, photostimulation was used to generate action potentials at hundreds of sites across the cortical layers. Examples of photostimulation-evoked EPSCs for all four neurons from the example cell pairs are illustrated in Figure 2. Cross-correlation analyses show that some EPSCs occur nearly synchronously within the two simultaneously recorded neurons (see ‘zero’ bin in histograms to the right of Fig. 2), whereas some EPSCs do not (non-zero bins). Of those EPSCs that do occur synchronously, some result from common input to the two cells: that is, a single stimulated neuron that fired an action potential was connected to both recorded neurons. Other EPSCs occur nearly synchronously because

AD1

Pyramidal

30

L2/3

Number of events

a

L2/3 1

1

Number of events

2

L5

0

50 100 150 pA

Number of events

100 µm

50 25

20 pA 20 ms

L2/3 L2/3

CP = 0.01

3

Time (ms)

L5

400 pA


[

NUMBER 11

CP = 0.00

20 10

3

100 µm 0

100 200 pA

0

50 100 150 pA

20 ms

30

CP = 0.03

20 10 0 –40 –20 0 20 40

Time (ms)

20

20 pA 20 ms 200

30

40

L5

0

10 0 –40 –20 0 20 40


2

100 µm 500 1,000 pA 0

L5

20 pA

L4

3

30 20

0 –40–20 0 20 40

L4

3


0 –40 –20 0 20 40

10

60

1

2

L4

2

40

L5

20

Pyramidal 1

1

L4 2

Time (ms)

c

L2/3 1

CP = 0.11

0 –40 –20 0 20 40

AD2

L2/3

10

30

3

100 200 pA

Pyramidal

0 –40 –20 0 20 40

L5

0

AD1

0 –40 –20 0 20 40

L4

3

b

20

L4 2


Number of events

–0.1

0.1

Number of events

0.0

0.2

two different stimulated neurons, each connecting to a different recorded neuron, fired action potentials nearly synchronously. To determine how many of the synchronous EPSCs were due to common input, as opposed to synchronous action potentials in neurons providing non-shared inputs, we computed shifted correlograms and subtracted them from correlograms based on matched stimulation trials (see Methods)24. Previously published control experiments demonstrate that shifted correlograms accurately estimate the correlation of spike times in presynaptic, photostimulated neurons (see supplementary material in ref. 23). The remaining EPSCs in the zero bin can be attributed to shared input to the two recorded neurons. These remaining EPSCs were then compared with the total number of evoked EPSCs to calculate the CP; that is, the proportion of evoked EPSCs that come from inputs shared by both cells. For example, a CP of 0.19 from stimulation sites in layer 2/3 for the reciprocally connected cell pair in Figure 2a indicates that these two cells share 19% of their layer 2/3 excitatory inputs. For the reciprocally connected cell pair (Fig. 2a), there were sharp peaks in the zero bin of the matched correlograms for stimulation sites in both layer 2/3 and layer 4. The CPs for this cell pair were 0.19 and 0.32 for layer 2/3 and layer 4, respectively. For layer 5 stimulation sites, there was not a sharp peak in the zero bin of the matched correlogram, and the CP was 0.06. For the cell pair with a one-way inhibitory-toexcitatory connection (Fig. 2b), the results from the cross-correlation analyses were very different: there were not sharp peaks in the zero bin of the matched correlograms for stimulation sites in any of the cortical layers. Thus, for this cell pair, CPs for all layers were relatively small (0.04 for layer 2/3, 0.06 for layer 4 and 0.03 for layer 5). The trends illustrated for these two example cell pairs were representative of all 23 fast-spiking–pyramid pairs for which photostimulation data were collected. Reciprocally connected cell pairs shared more

Number of events

0.1

0.3

FS ←→ P

0.2

co No nn t . FS → P

–0.1

0.3

0.4


0.0

Correlation probability

0.1

L5

*

FS ←→ P


*

0.2

L4

0.4

Number of events


*

**

0.3



L2/3 0.4

[

NOVEMBER 2005

Figure 4 Cross-correlation analyses of photostimulation-evoked EPSCs simultaneously recorded in adjacent layer 2/3 neuron pairs consisting of one pyramidal neuron and one adapting inhibitory neuron (AD). (a) Data from a reciprocally connected cell pair including a type 1 adapting inhibitory neuron (AD1). (b) Data from a cell pair including an AD1 inhibitory neuron with a one-way inhibitory connection onto the pyramidal neuron. (c) Data from a cell pair including a type 2 adapting inhibitory neuron (AD2) with a one-way inhibitory connection onto the pyramidal neuron. All plots are as in Figure 2.

1555

0.1 0.0 –0.1


0.3 0.2 0.1 0.0 –0.1

0.4

L5

0.3 0.2 0.1 0.0 –0.1 co No nn t AD . → P P→ AD AD ←→ P

0.2

L4


0.3

0.4

co No nn t AD . → P P→ AD AD ←→ P

L2/3


0.4

co No n t AD n. → P P→ AD AD ←→ P


ARTICLES

Figure 5 Correlation probabilities for EPSCs measured simultaneously in adapting interneuron–pyramidal neuron pairs. Correlation probabilities are shown separately for photostimulation sites in each cortical layer and for cell pairs that either were not connected (squares), had a one-way inhibitory connection (inverted triangles), had a one-way excitatory connection (triangles) or were reciprocally connected (circles). Filled symbols indicate cell pairs including type 1 adapting interneurons, and open symbols indicate cell pairs including type 2 adapting interneurons. Mean values for each group are indicated by horizontal lines.

and the layer stimulated (Fig. 5). Average CPs never exceeded 0.10 for any group. For stimulation sites in layer 2/3, the CPs of pairs including AD1 versus AD2 interneurons were indistinguishable (0.04 ± 0.01, n ¼ 24 pairs versus 0.05 ± 0.01, n ¼ 12). Therefore, we combined data for AD1 and AD2 neurons for subsequent analyses. Analyzed according to direct connectivity between neuron pairs, CPs for photostimulation sites in layer 2/3 were 0.04 ± 0.01 (mean ± s.e.m., n ¼ 14 cell pairs) for unconnected cell pairs, 0.05 ± 0.01 (n ¼ 9) for pairs with one-way inhibitory connections, 0.03 ± 0.02 (n ¼ 8) for pairs with oneway excitatory connections, and 0.05 ± 0.01 (n ¼ 5) for pairs with reciprocal connections. For stimulation sites in layer 4 the corresponding CPs were –0.008 ± 0.01, n ¼ 9; 0.03 ± 0.01, n ¼ 8; 0.06 ± 0.05, n ¼ 4; and 0.06 ± 0.03, n ¼ 4). For stimulation sites in layer 5 the CPs were 0.04 ± 0.04, n ¼ 8; 0.08 ± 0.04, n ¼ 7; 0.04 ± 0.04, n ¼ 4; and 0.03 ± 0.02, n ¼ 4.

Adapting interneuron–pyramid pairs share little common input We collected photostimulation data from 38 adapting interneuron– pyramid pairs and found that, unlike pairs involving fast-spiking interneurons, these cell pairs did not share much common excitatory input, regardless of their direct connectivity. We separated adapting interneurons into two groups on the basis of the relative strength of excitatory input from different cortical layers18. Analysis of results from this data set (data not shown) confirmed previous published observations18, indicating that the laminar organization of local excitatory inputs to adapting interneurons distinguishes two nonoverlapping populations: type 1 adapting interneurons (AD1; 26 of 38 cells), which received more balanced input from layers 2/3, 4 and 5 (Fig. 4a,b), and type 2 adapting interneurons (AD2; 12 of 38 cells), which received the great majority of their excitatory input from layer 2/3 and very little (if any) from layer 4 or layer 5 (Fig. 4c). Therefore, for cell pairs including AD2 neurons, we calculated CPs only for stimulation sites in layer 2/3. For adapting interneuron–pyramidal cell pairs, there were never sharp peaks in the zero bins of the matched correlograms, regardless of the adapting interneuron type, connectivity to the pyramidal neuron or the layer stimulated (see representative cross-correlation analyses of photostimulation-evoked EPSCs in Fig. 4). We plotted CPs according to the type of direct connectivity between the recorded neurons (unconnected, one-way inhibitory, one-way excitatory, reciprocal)

DISCUSSION We have shown that both inhibitory connections and excitatory connections to inhibitory neurons can be precise on a fine scale (Supplementary Fig. 1), using analysis of direct connections measured during paired recordings and cross-correlation analyses of photostimulation-evoked EPSCs. The fine-scale organization of connections between inhibitory and excitatory neurons depends, however, both on the inhibitory cell type and the organization of other connections within the network. Analyses of direct connectivity during paired recordings showed that fast-spiking inhibitory neurons in layer 2/3 connect to their neighboring pyramids with a moderately high probability (47%). But this probability is highly dependent on whether the pyramid makes an excitatory connection to the fast-spiking interneuron. Usually the pyramid does not connect to the fast-spiking interneuron, and in these cases there is a one-way inhibitory connection in only 37% of the cell pairs. But when the pyramidal neuron does connect to the fastspiking interneuron, the connection is nearly always (seven of eight pairs) reciprocated. In view of the fact that some connections are cut during preparation of brain slices, it could be that the true probability of reciprocity is closer to 100%. Not only was the probability of an inhibitory connection increased by more than twofold when there was an excitatory connection, but in the reciprocally connected cases, the IPSCs were more than threefold larger than in the case of one-way connections. Together these differences make the inhibition from fastspiking interneurons strongly biased (by about sixfold) toward the same layer 2/3 pyramidal neurons that provide them with excitatory input. A previously published study of connections between 243 fastspiking interneuron–pyramidal neuron pairs showed no difference in the probability of inhibitory connections regardless of whether a cell pair had an excitatory connection29. Furthermore, in that study, excitatory connections were observed in more than half of the cell pairs tested (exact numbers were not reported), far more than the rate of 19% that we observe. These differences could be due to the use of recordings from a different cortical area, as data were combined for recordings made from ‘visual and somatosensory neocortical areas’ in that study. We believe it is more likely, however, that the difference is due to the immature state of the cortex in that study (postnatal days 14– 16, compared with postnatal days 21–26 in our study). Age-dependent refinement of the specificity of synaptic connections has been extensively documented in many systems, including the visual cortex30. Not only do fast-spiking inhibitory neurons connect preferentially to the same layer 2/3 pyramids that excite them, but these reciprocally connected pairs are also part of the same subnetworks of excitatory neurons (Supplementary Fig. 1). Our cross-correlation analyses of

1556

VOLUME 8

common input than did both unconnected cell pairs and cell pairs with one-way inhibitory connections (Fig. 3). For stimulation sites in layer 2/3, CPs for reciprocally connected cell pairs were high, averaging 0.17 ± 0.02 (mean ± s.e.m., n ¼ 7 pairs), whereas the CPs were significantly lower for unconnected pairs (CP ¼ 0.05 ± 0.01, n ¼ 8, P o 0.0005, Mann-Whitney U test) and for pairs with one-way inhibitory connections (CP ¼ 0.01 ± 0.03, n ¼ 8, P o 0.005). Results for layer 4 stimulation sites were similar to those for layer 2/3. CPs were high for reciprocally connected cell pairs (CP ¼ 0.22 ± 0.03, n ¼ 7) and significantly lower for unconnected pairs (CP ¼ 0.06 ± 0.02, n ¼ 8, P o 0.005) and for pairs with one-way inhibitory connections (CP ¼ 0.05 ± 0.02, n ¼ 8, P o 0.005). For stimulation sites in layer 5, CPs were calculated only when both cells in the pair received significant evoked excitatory input from that layer (see Methods). For stimulation sites in layer 5, the CPs for reciprocally connected cell pairs were larger than for other cell pairs (CP ¼ 0.14 ± 0.04, n ¼ 6 versus 0.05 ± 0.04, n ¼ 6 for unconnected pairs and 0.07 ± 0.04, n ¼ 5 for one-way inhibitory pairs), but the difference was not statistically significant (P 4 0.09).

[

NUMBER 11

[



ARTICLES photostimulation-evoked EPSCs show that reciprocally connected fastspiking interneuron–pyramid pairs share common excitatory input from layer 4 and from within layer 2/3. Therefore, not only does the fine-scale specificity of excitatory connections create preferentially connected subnetworks of excitatory neurons23, but these subnetworks also preferentially select the same fast-spiking interneurons that provide them with reciprocal inhibition. In contrast to results for fast-spiking interneurons, we observed no evidence for fine-scale specificity of connections involving adapting interneurons. For paired recordings involving an adapting interneuron and a pyramid, reciprocal connections were not observed with greater frequency than expected from chance coincidence on the basis of the probability of one-way connections. Moreover, cross-correlation analyses of photostimulation-evoked EPSCs in these same cell pairs indicated that they share little common excitatory input, regardless of whether they are directly connected. Nevertheless, our adapting interneurons probably included multiple inhibitory neuron types that could not be distinguished with the methods we used, and there are probably some relatively rare inhibitory cell types that were not sampled in our experiments. Thus, given the small sample size in some of our adapting interneuron groups (such as only five reciprocally connected pairs), it is possible that a subset of adapting interneuron types does actually connect to other neurons with fine-scale specificity. It is very unlikely that the systematic differences in CPs that we observe between neuron pairs, on the basis of either their direct connectivity or on the type of inhibitory neuron, can be attributed to effects of cutting during brain slice preparation. This is because the recorded neurons were sufficiently deep in the brain slices such that their recipient dendritic arbors were preserved, regardless of cell type. Cutting might, however, have reduced the extent of the differences in CP that we observe to be correlated with connectivity between fastspiking inhibitory neuron–pyramidal neuron pairs. This is because cutting could have affected the direct connectivity between pairs such that pairs that were reciprocally connected before cutting had only a one-way connection or no direct connection at all after cutting. But these previously connected pairs would still share common input resulting in a high value for CP. This potential artifact could explain the higher CP values for a small subset of pairs with one-way connections (Fig. 3). We have previously reported fine-scale specificity of excitatory connections to pyramidal neurons in rat visual cortex23. Pairs of neighboring layer 2/3 pyramidal neurons that were directly connected to each other shared common excitatory input from the same presynaptic neurons in layer 4 and within layer 2/3, whereas unconnected pairs shared little common excitatory input from these layers. For these same cell pairs, however, excitatory input from layer 5 and photostimulation-evoked inhibitory connections from layer 2/3 and layer 4 followed different rules. The cell pairs shared common input from these sources regardless of whether they were directly interconnected. These observations suggested that excitatory connections from layer 4 to layer 2/3 and within layer 2/3 are precise on a fine scale, but inhibitory connections do not respect the specificity of these fine scale subnetworks. The present study, however, demonstrates that inhibitory connections from fast-spiking inhibitory neurons can be specific on a fine scale. The likely reason for this difference between the present study and our previous report is that photostimulation-evoked IPSCs almost certainly reflect inputs from multiple inhibitory neuron types that are activated indiscriminately during photostimulation23. In particular, our analyses of the rise times of photostimulation-evoked IPSCs measured in the previous study, compared with those from paired


[

NUMBER 11

[

NOVEMBER 2005

cell stimulation in the present study, suggest that the lack of specificity reflected a contribution of IPSCs from adapting interneurons that masked the specificity of connections involving fast-spiking interneurons. In the present study we have found that the rise times of IPSCs (onto pyramids) that came from fast-spiking interneurons were significantly shorter (1.7 ms) than the IPSCs that came from adapting interneurons (2.6 ms; Fig. 1f). In the previously published study23, the rise times of IPSCs evoked in layer 2/3 pyramids by photostimulation were, on average, intermediate in duration (2.0 ms). Another factor that is likely to have contributed to the lack of specificity of inhibitory input observed in our previous study is that there are also some promiscuous connections from fast-spiking interneurons that connect to pyramidal neurons outside of their own fine-scale subnetwork (Supplementary Fig. 1). These promiscuous connections are, however, only about one-sixth the strength of within-network inhibition from fast-spiking interneurons (see above). In classical terms, the inhibitory connections that we observe from layer 2/3 interneurons onto layer 2/3 pyramids can be considered as providing feed-forward, feedback or lateral inhibition (or combinations of all three). The nature of these connections depends on the sources of excitatory input to the inhibitory neurons that in turn provide inhibition to the layer 2/3 pyramids. Here we refer to these types of connections with reference to the laminar flow of information within the cortex as well as to the fine-scale excitatory subnetworks. Because layer 4 provides the dominant feed-forward excitation to both pyramidal and fast-spiking inhibitory neurons in layer 2/3 (ref. 18), we consider it to be a source of feed-forward excitation5,6. If an inhibitory neuron in layer 2/3 connects to its neighboring pyramid and these cells share common excitatory input from layer 4, then the inhibitory neuron contributes feed-forward inhibition. We classify excitatory connections between layer 2/3 pyramids as ‘recurrent’31,32, and when layer 2/3 neurons excite the same interneurons that in turn inhibit them, this is feedback inhibition. Finally, we consider lateral inhibition with respect to the fine-scale excitatory subnetworks. Inhibition that impinges on the subnetwork but that is driven by excitation from outside the subnetwork is ‘lateral’ with respect to these subnetworks. In terms of these definitions, our findings show that fast-spiking interneurons provide all three types of inhibition to their neighboring pyramids, but lateral inhibition from this cell type is much weaker than inhibition within the subnetwork. The presence of feed-forward inhibition is apparent from the shared excitatory input from layer 4 for reciprocally connected cell pairs, and similarly, shared input from within layer 2/3 is evidence of feedback inhibition. Lateral inhibition is apparent from the observation that layer 2/3 pyramidal neurons receive inhibition from fast-spiking interneurons to which they do not provide excitation and with which they do not share common excitatory input from layer 4 or within layer 2/3. But these nonreciprocated, lateral inhibitory connections are about sixfold weaker than the feedforward and feedback inhibition within the subnetwork (see above). In contrast to inhibition from fast-spiking interneurons, inhibition from adapting neurons is most consistent with lateral forms of inhibition. These inhibitory cells do not belong to the same fine-scale subnetworks as their neighboring pyramidal neurons, regardless of whether they inhibit them directly. Thus, they cannot provide either strong feed-forward or strong feedback inhibition. It is also important to consider, however, that layer 2/3 adapting interneurons also differ from fast-spiking interneurons in their sources of excitatory input. They typically receive their strongest excitation from sources other than layer 4 (ref. 18) and can also receive strong excitation from outside the local cortical region12. Thus, the inhibition from adapting interneurons onto

1557


ARTICLES layer 2/3 pyramids is not as easily classified as that from fast-spiking interneurons and may contribute to other more subtle functions. Feed-forward and/or feedback inhibition that also has a lateral component has been suggested as an important element of cortical processing in several experimental and theoretical studies32–35. Our results show that this organization is present with respect to fine-scale neuronal subnetworks and differs according to inhibitory cell type. The results also provide quantitative measures of the relationships between these inhibitory components. The presence of distinct types of cortical inhibitory neurons, both in the hippocampus and in the cerebral cortex, suggests that each type may have a unique role in the cortical network36,37. The unique properties of fast-spiking interneurons28 as well as studies of their functional interactions with excitatory neurons38,39 have suggested that they are important in precisely regulating the timing of cortical activity, particularly at high frequencies. In particular, the combination of gap junction and synaptic coupling between fast-spiking interneurons10,40 results in entrainment at gamma frequencies39, and the in vivo spike timing of fast-spiking, parvalbumin-positive basket cells is strongly correlated with local field potentials during gamma frequency and high-frequency ripple oscillations38. Our observation that reciprocally connected fast-spiking interneuron–pyramid pairs share common excitatory input on a fine scale suggests that these networks are also likely to regulate the precise timing of activity within fine-scale cortical subnetworks. In this context, fine-scale subnetworks may contribute to the synchronization of activity within particular neuronal subpopulations. METHODS The methods used for both the collection and analyses of data in this study were the same as those described previously23, except that recordings were made from pairs of neurons consisting of one pyramidal neuron and one inhibitory neuron instead of two pyramidal neurons. All experimental procedures involving live animals were approved by the Salk Institute Animal Care and Use Committee. Slice preparation, photostimulation and recordings. A vibratome was used to cut 300-mM-thick coronal brain slices from the primary visual cortex of postnatal day (P) 21–26 Long-Evans rats. Slices were cut in artificial cerebral spinal fluid (ACSF; composition in mM: 124 NaCl, 5 KCl, 1.25 KH2PO4, 1.3 MgSO4, 3.2 CaCl2, 26 NaHCO3 and 10 glucose) and stored in an interface chamber at B34 1C for at least 1 h until they were transferred to a roomtemperature (20–24 1C) recording chamber containing ACSF with 60–80 mM ‘caged’ glutamate (g-(a-carboxy-2-nitrobenzyl)ester, trifluoroacetate, L-glutamic acid). An infrared Olympus DIC microscope with a 40, 0.8 NA water-immersion lens was used to visualize and target recording electrodes to pairs of layer 2/3 pyramidal neurons with somata separated by, on average, about 50 mM for whole-cell recordings. The distances (mean ± s.e.m.) between recorded cells were 49.0 ± 3.7 mm for fast-spiking interneuron–pyramid pairs that were not connected; 57.5 ± 7.7 mm for fast-spiking interneuron–pyramid pairs with one-way inhibitory connections; 45.0 ± 8.3 mm for fast-spiking interneuron–pyramid pairs with reciprocal connections; 42.7 ± 5.1 mm for adapting interneuron–pyramid pairs that were not connected; 43.7 ± 6.6 mm for adapting interneuron–pyramid pairs with one-way inhibitory connections; 47.6 ± 7.1 mm for adapting interneuron–pyramid pairs with one-way excitatory connections; and 52.7 ± 6.5 mm for adapting interneuron–pyramid pairs with reciprocal connections. Cell bodies of recorded neurons were at least 50 mm from the surface of the slice. Glass recording electrodes (4–6 MO resistance) were filled with an intracellular solution consisting of 130 mM potassium gluconate, 6 mM KCl, 2 mM MgCl2, 0.2 mM EGTA, 10 mM HEPES, 2.5 mM Na2ATP, 0.5 mM Na2GTP, 10 mM potassium phosphocreatine and 0.3% biocytin. pH was adjusted to 7.25 with KOH. All intracellular recordings had access resistances less than 20 MO. In all paired recordings, connections between neuron pairs were assessed by injecting current to evoke action potentials in one of the cells recorded in current-clamp while testing for

1558

PSCs during voltage-clamp recording in the other cell (Fig. 1c,d). For detection of excitatory input, inhibitory cells were voltage clamped at 65 mV. For detection of inhibitory input, pyramidal cells were voltage clamped at 0 mV. For each pair, connections were tested for at least 50 trials generating single action potentials in the presynaptic neuron, for each direction. When connections were not detected with this procedure, they were also tested by stimulating in trains of 4–5 action potentials at 50 Hz to induce possible potentiation of weak connections. Inhibitory neurons were classified as either adapting (Fig. 1b) or ‘fast spiking’ (nonadapting; Fig. 1a) on the basis of temporal patterns of action potentials evoked by intracellular current injection18,25,26. Also, identification of inhibitory neurons versus pyramidal neurons was always confirmed on the basis of biocytin staining of the intracellularly labeled neurons. Photostimulation was achieved by uncaging glutamate with 10-ms duration flashes of ultraviolet light from an argon-ion laser focused through the 40 microscope objective. This results in generation of action potentials only in neurons with cell bodies within 100 mM and usually less than 50 mM from the site of uncaging23. Photostimulation-evoked excitatory postsynaptic currents were measured from recorded neurons in voltage-clamp mode with the holding potentials at –65 mV to isolate EPSCs. Spontaneous EPSCs were also recorded in interleaved trials with no stimulation. Photostimulation-evoked IPSCs were not measured. Data analysis. Maps of photostimulation sites were aligned to laminar borders in fixed and stained tissue (Figs. 2 and 4) and each site was assigned a laminar identity. Sites within 50 mM of laminar borders were discarded from further analyses in order to limit the number of evoked synaptic currents arising from neurons with cell bodies possibly outside the stimulated layer. The electrical recordings from trials with and without (‘control’) photostimulation were analyzed using peak analysis software from Synaptosoft and other custom software developed by the authors. The times of onset and amplitudes of all EPSCs or IPSCs occurring within 150 ms of stimulation were marked. Rise times of PSCs were calculated as the time from 10% to 90% of the amplitude from baseline to peak. Analyses of the laminar sources and strengths of excitatory and inhibitory input to layer 2/3 pyramidal neurons (data not shown) gave results indistinguishable from those described previously18 and no systematic differences that correlated with results from cross-correlation analyses. Cross-correlograms of EPSCs were computed for each pair of simultaneously recorded layer 2/3 neurons with separate correlograms being computed for stimulation sites from each cortical layer. For some cells certain layers provided weak or no input to recorded neurons preventing evoked EPSCs from being clearly distinguished from spontaneous EPSCs. In these cases, correlograms were not computed for the corresponding layers (see Results). Correlograms were also computed for spontaneous EPSCs. Cross-correlation data were binned into histograms with 4-ms bins, with the central bin including values of 0 ± 2 ms. Data from the stimulation trials (from the same layer) were also used to create shifted correlograms for each layer and cell pair24. To compute correlation probability (CP), the shifted correlogram was subtracted from the unshifted correlogram for the corresponding layer, and then the value in the central bin was divided by the average for the two cells of the estimated total number of evoked EPSCs observed for all trials in the relevant layer. The average number of evoked EPSCs was calculated as the total number of measured EPSCs for cell A minus the expected number of spontaneous EPSCs for that cell, plus the same value calculated for cell B, divided by two. Correlation probabilities for spontaneous EPSCs were invariably small (0.003 ± 0.003 for fast-spiking interneuron–pyramid pairs; 0.017 ± 0.003 for adapting interneuron–pyramid pairs) and thus have a negligible influence on the CP values calculated for evoked EPSCs. Statistical analysis. Statistical comparisons between correlation probabilities from different groups of neuron pairs (Fig. 3) used the non-parametric MannWhitney U test because there was no information available about the normality of these distributions. Comparisons of the proportions of cell pairs with direct inhibitory connections for pairs with excitatory connections versus those without excitatory connections used the Fisher exact test, which is designed specifically for this type of data set and yields an exact measurement of p. Values reported are two-tailed.

VOLUME 8

[

NUMBER 11

[


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

ACKNOWLEDGMENTS Supported by grants from the US National Institutes of Health (MH063912, EY010742) and from the Japanese Ministry of Education, Culture, Science, Sports and Technology (17023026, 17500208). We thank Y. Komatsu and H. Sato for discussions.


COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Felleman, D.J. & Van Essen, D.C. Distributed hierarchical processing in the primate cerebral cortex. Cereb. Cortex 1, 1–47 (1991). 2. Fitzpatrick, D. The functional organization of local circuits in visual cortex: insights from the study of tree shrew striate cortex. Cereb. Cortex 6, 329–341 (1996). 3. Martin, K.A.C. in Cerebral Cortex, Vol. 2. (eds. Jones, E.G. & Peters, A.) 241–284 (Plenum, New York, 1984). 4. Mooser, F., Bosking, W.H. & Fitzpatrick, D. A morphological basis for orientation tuning in primary visual cortex. Nat. Neurosci. 7, 872–879 (2004). 5. Callaway, E.M. Local circuits in primary visual cortex of the macaque monkey. Annu. Rev. Neurosci. 21, 47–74 (1998). 6. Gilbert, C.D. Microcircuitry of the visual cortex. Annu. Rev. Neurosci. 6, 217–247 (1983). 7. Gilbert, C.D. & Wiesel, T.N. Columnar specificity of intrinsic horizontal and corticocortical connections in cat visual cortex. J. Neurosci. 9, 2432–2442 (1989). 8. Lund, J.S. Anatomical organization of macaque monkey striate visual cortex. Annu. Rev. Neurosci. 11, 253–288 (1988). 9. Agmon, A. & Connors, B.W. Correlation between intrinsic firing patterns and thalamocortical synaptic responses of neurons in mouse barrel cortex. J. Neurosci. 12, 319–329 (1992). 10. Gibson, J.R., Beierlein, M. & Connors, B.W. Two networks of electrically coupled inhibitory neurons in neocortex. Nature 402, 75–79 (1999). 11. Gonchar, Y. & Burkhalter, A. Connectivity of GABAergic calretinin-immunoreactive neurons in rat primary visual cortex. Cereb. Cortex 9, 683–696 (1999). 12. Gonchar, Y. & Burkhalter, A. Distinct GABAergic targets of feedforward and feedback connections between lower and higher areas of rat visual cortex. J. Neurosci. 23, 10904–10912 (2003). 13. Meskenaite, V. Calretinin-immunoreactive local circuit neurons in area 17 of the cynomolgus monkey, Macaca fascicularis. J. Comp. Neurol. 379, 113–132 (1997). 14. Staiger, J.F. et al. Innervation of interneurons immunoreactive for VIP by intrinsically bursting pyramidal cells and fast-spiking interneurons in infragranular layers of juvenile rat neocortex. Eur. J. Neurosci. 16, 11–20 (2002). 15. Yabuta, N.H., Sawatari, A. & Callaway, E.M. Two functional channels from primary visual cortex to dorsal visual cortical areas. Science 292, 297–300 (2001). 16. Sawatari, A. & Callaway, E.M. Diversity and cell type specificity of local excitatory connections to neurons in layer 3B of monkey primary visual cortex. Neuron 25, 459–471 (2000).


[

NUMBER 11

[

NOVEMBER 2005

17. Briggs, F. & Callaway, E.M. Layer-specific input to distinct cell types in layer 6 of monkey primary visual cortex. J. Neurosci. 21, 3600–3608 (2001). 18. Dantzker, J.L. & Callaway, E.M. Laminar sources of synaptic input to cortical inhibitory interneurons and pyramidal neurons. Nat. Neurosci. 3, 701–707 (2000). 19. Shepherd, G.M., Stepanyants, A., Bureau, I., Chklovskii, D. & Svoboda, K. Geometric and functional organization of cortical circuits. Nat. Neurosci (2005). 20. Schubert, D. et al. Layer-specific intracolumnar and transcolumnar functional connectivity of layer V pyramidal cells in rat barrel cortex. J. Neurosci. 21, 3580–3592 (2001). 21. Schubert, D., Kotter, R., Zilles, K., Luhmann, H.J. & Staiger, J.F. Cell type-specific circuits of cortical layer IV spiny neurons. J. Neurosci. 23, 2961–2970 (2003). 22. Song, S., Sjostrom, P.J., Reigl, M., Nelson, S. & Chklovskii, D.B. Highly nonrandom features of synaptic connectivity in local cortical circuits. PLoS Biol. 3, e68 (2005). 23. Yoshimura, Y., Dantzker, J.L. & Callaway, E.M. Excitatory cortical neurons form finescale functional networks. Nature 433, 868–873 (2005). 24. Aertsen, A.M., Gerstein, G.L., Habib, M.K. & Palm, G. Dynamics of neuronal firing correlation: modulation of ‘effective connectivity’. J. Neurophysiol. 61, 900–917 (1989). 25. Kawaguchi, Y. Groupings of nonpyramidal and pyramidal cells with specific physiological and morphological characteristics in rat frontal cortex. J. Neurophysiol. 69, 416–431 (1993). 26. Connors, B.W. & Gutnick, M.J. Intrinsic firing patterns of diverse neocortical neurons. Trends Neurosci. 13, 99–104 (1990). 27. Kawaguchi, Y. & Kondo, S. Parvalbumin, somatostatin and cholecystokinin as chemical markers for specific GABAergic interneuron types in the rat frontal cortex. J. Neurocytol. 31, 277–287 (2002). 28. Blatow, M., Caputi, A. & Monyer, H. Molecular diversity of neocortical GABAergic interneurones. J. Physiol. (Lond.) 562, 99–105 (2005). 29. Holmgren, C., Harkany, T., Svennenfors, B. & Zilberter, Y. Pyramidal cell communication within local networks in layer 2/3 of rat neocortex. J. Physiol. (Lond.) 551, 139–153 (2003). 30. Katz, L.C. & Shatz, C.J. Synaptic activity and the construction of cortical circuits. Science 274, 1133–1138 (1996). 31. Martin, K.A. Microcircuits in visual cortex. Curr. Opin. Neurobiol. 12, 418–425 (2002). 32. Douglas, R.J., Koch, C., Mahowald, M., Martin, K.A. & Suarez, H.H. Recurrent excitation in neocortical circuits. Science 269, 981–985 (1995). 33. Dragoi, V. & Sur, M. Dynamic properties of recurrent inhibition in primary visual cortex: contrast and orientation dependence of contextual effects. J. Neurophysiol. 83, 1019–1030 (2000). 34. Swadlow, H.A. Fast-spike interneurons and feedforward inhibition in awake sensory neocortex. Cereb. Cortex 13, 25–32 (2003). 35. Lauritzen, T.Z. & Miller, K.D. Different roles for simple-cell and complex-cell inhibition in V1. J. Neurosci. 23, 10201–10213 (2003). 36. Somogyi, P., Tamas, G., Lujan, R. & Buhl, E.H. Salient features of synaptic organisation in the cerebral cortex. Brain Res. Brain Res. Rev. 26, 113–135 (1998). 37. Somogyi, P. & Klausberger, T. Defined types of cortical interneurone structure space and spike timing in the hippocampus. J. Physiol. (Lond.) 562, 9–26 (2005). 38. Klausberger, T. et al. Brain-state- and cell-type-specific firing of hippocampal interneurons in vivo. Nature 421, 844–848 (2003). 39. Tamas, G., Buhl, E.H., Lorincz, A. & Somogyi, P. Proximally targeted GABAergic synapses and gap junctions synchronize cortical interneurons. Nat. Neurosci. 3, 366–371 (2000). 40. Galarreta, M. & Hestrin, S. A network of fast-spiking cells in the neocortex connected by electrical synapses. Nature 402, 72–75 (1999).

1559


ARTICLES

Induction of sharp wave–ripple complexes in vitro and reorganization of hippocampal networks Christoph J Behrens1, Leander P van den Boom1, Livia de Hoz1, Alon Friedman1,2 & Uwe Heinemann1 Hippocampal sharp wave–ripple complexes (SPW-Rs) occur during slow-wave sleep and behavioral immobility and are thought to represent stored information that is transferred to the neocortex during memory consolidation. Here we show that stimuli that induce long-term potentiation (LTP), a neurophysiological correlate of learning and memory, can lead to the generation of SPW-Rs in rat hippocampal slices. The induced SPW-Rs have properties that are identical to spontaneously generated SPW-Rs: they originate in CA3, propagate to CA1 and subiculum and require AMPA/kainate receptors. Their induction is dependent on NMDA receptors and involves changes in interactions between clusters of neurons in the CA3 network. Their expression is blocked by low-frequency stimulation but not by NMDA receptor antagonists. These data indicate that induction of LTP in the recurrent CA3 network may facilitate the generation of SPW-Rs.

After initial encoding, memory traces require a process of consolidation to become long lasting1. This process is believed to involve interactions between the hippocampus and neocortical cell assemblies and to result in the stabilization of memory traces across the cortical mantle2. It has been hypothesized that consolidation is dependent on hippocampal network oscillations, such as SPW-Rs, which are observed in vivo during slow-wave sleep3,4 and behavioral immobility4,5. SPW-Rs are fast (140–200 Hz) oscillations in field potential recordings that are superimposed on a slow field potential transient, where the ripples result from synchronized discharges of hippocampal CA3 cells. The mechanisms underlying the induction of SPW-Rs remain to be elucidated. It is generally accepted that during learning, synaptic plasticity results in changes in the strength of the connections between neurons within the recruited population. LTP6 is a long-lasting change in the responses of single neuronal synapses that can be induced by highfrequency stimulation (HFS) and theta burst stimulation (TBS) patterns both in vivo and in vitro6,7. LTP is considered to be a good cellular model of learning and memory8,9. In the present study, we explored whether physiological processes that result in synaptic plasticity in the hippocampus can also lead to the generation of SPW-Rs. We specifically tested the hypothesis that protocols that induce LTP in the CA3 region can result in the subsequent generation of SPW-Rs in this area such that strengthening of synaptic coupling becomes associated with a tendency of these neurons to fire synchronously as it occurs during SPW-R. We used an in vitro preparation to activate CA3 from different stimulation sites with recurrent HFS or TBS. Recordings were made in areas CA3, CA1 and the subiculum. The physiological and pharmacological characteristics of the induced SPW-Rs were then compared with

those of SPW-Rs that occurred spontaneously in ventral hippocampal slices. In a subset of experiments, the activity of individual CA3 pyramidal cells was recorded at the same time as the induction of the SPW-Rs. Thus we determined not only the contribution of a single cell to the synchronized network activity during SPW-Rs but also whether this contribution changed during SPW-R induction. The main findings of this series of in vitro experiments are that SPW-Rs can be induced with stimulation protocols that are known to induce LTP in vivo and in vitro10–12 and that the induction is accompanied by synaptic reorganization within the recurrent network of area CA3. RESULTS Repeated HFS induces SPW-Rs Our main finding is that repeated application of HFS trains (three tetani of 40 pulses at 100 Hz; 40-s interval; repeated up to 15 times every 5 min) to the stratum radiatum of area CA1 (Fig. 1a), a protocol that successfully induced LTP (Fig. 1b), led to the generation of SPW-Rs in areas CA3 and CA1 (Fig. 1c). Such stimuli reliably induced SPW-Rs when repeated three to five times (Fig. 1d). The induced SPW-Rs persisted for at least 2 h. Similar persistence was also observed when SPW-Rs were induced with 7 instead of 15 stimulation trains (n ¼ 6; data not shown). The HFS-induced SPW-Rs, when recorded in the stratum pyramidale or stratum oriens, consisted of 30- to 80-ms positive field potential transients (sharp waves, SPWs), superimposed by a series of small population spikes (ripples) with a frequency of about 180 Hz, as shown by power spectrum analysis and the interval between the first and second peak in auto-correlation analysis, which indicated a mean interspike interval of 5.8 ms (Fig. 2a–c). Although the frequency of ripples, once induced, did not

1Institute for Neurophysiology, Charite ´-Universita¨tsmedizin Berlin, Tucholskystrasse 2, 10117 Berlin, Germany. 2Department of Neurosurgery, Zlotowski Centre for Neuroscience, Ben-Gurion University of the Negev and Soroka Medical Center, Beer-Sheva, 84105 Israel. Correspondence should be addressed to U.H. ([email protected]).

Received 11 July; accepted 21 September; published online 16 October 2005; doi:10.1038/nn1571

1560

VOLUME 8

[

NUMBER 11

[


ARTICLES Figure 1 Recurrent stimulation induces SPW-Rs CA3 CA3 in area CA3. (a) Placement of recording and 500 stimulating electrodes in areas CA3 and CA1. Black circles, interneurons; triangles, pyramidal HFS CA1 cells; white circles, excitatory connections; 300 T terminals, inhibitory connections. The flash indicates stimulation site in stratum radiatum of area CA1. Stimulation consisted of 400-ms trains 100 with 100 Hz (pulse duration, 0.1 ms) repeated 3rd 5th 15th three times with 40-s intervals every 5 min, 0 repeated up to 15 times. Stimulus intensity was 0 20 40 60 CA1 Time (min) adjusted to evoke 70% of maximal response in area CA3 and CA1. (b) LTP in CA3 induced 2 mV CA3 by CA1 stratum radiatum HFS (open circles, 12 CA1 first population spike; filled circles, second * 4 20 ms 200 8 population spike; n ¼ 6) as indicated by nd rd th th th th th 2 * 2 3 4 5 6 7 15 4 a change in the normalized amplitude of the 170 0 population spike (D PS). Inset: sample recordings 0 CA1 12 before (above) and after (below) induction of LTP. CA3 * * 4 The first antidromic population spike remained 8 200 2 unaltered after induction of LTP. (c) Top: sample 4 CA3 170 recordings of SPW-Rs corresponding to time 0 rd th th 0 rd th th 3 5 15 3 5 15 points indicated below, at expanded time scale 3rd 5th 15th (open arrows) Bottom: simultaneous recordings in HFS stratum pyramidale of area CA1 and CA3 during the induction protocol from the second to the fifteenth stimulus train application (black arrows). Note appearance of first SPW-Rs between third and fourth stimulus train, with a further increase in amplitude upon recurrent stimulation. Calibration bars, 1 min and 2 mV. (d) Left to right: changes in ripple frequency, amplitude of SPW-Rs and incidence of SPW-Rs in CA3 (bottom) and CA1 (top) after the third, fifth and fifteenth stimulus train repetition. *P o 0.05.

b

∆ PS (%)

a

change significantly with further stimulation, the amplitude and incidence of occurrence of the induced SPW-Rs increased with further repetition of three to five stimulus series but then remained unaltered (Fig. 1d). Although the stimulation protocol that was used may seem to be rather unphysiological, the applied stimuli induced submaximal responses in areas CA1 and CA3. During the stimulus train, CA3 pyramidal cells were depolarized by 17.9 ± 0.6 mV (n ¼ 51; eight slices) and [K+]o, a reliable measure of overall neuronal activity, increased by 1.2 ± 0.1 mM (n ¼ 81; eight slices), which is comparable to rises in [K+]o that are observed after physiological sensory stimulation in other parts of the brain13. In addition, the more physiological TBS was also able to induce SPW-Rs (see below). The stimuli induced field potential transients, typically with two population spikes (Fig. 1b). The first represented an antidromic population spike, as it persisted after application of CNQX or after lowering of the extracellular Ca2+ concentration (data not shown). The second population spike corresponded to the population response that was evoked by the induced excitatory postsynaptic potentials (EPSPs). After HFS application, the first population spike remained unaltered, whereas the second population spike was potentiated (average increase in response was about 140%; Fig. 1b). In very ventral portions of the rat hippocampus14 and in mouse hippocampus15, SPW-Rs can occur spontaneously some time (1–2 h) after slice preparation (Supplementary Fig. 1 online). Therefore, it may be argued that stimulus-induced SPW-Rs result from the facilitated appearance of spontaneous SPW-Rs. In slices that did not develop spontaneous SPW-Rs within 4 h after preparation, however, we could still induce SPW-Rs by HFS. We also studied whether SPW-Rs could be observed in dorsal hippocampal slices (obtained by sagittal or coronal sectioning) that did not produce spontaneous SPW-Rs. This was indeed the case. We were able to induce SPW-Rs using the same stimulus protocol in both sagittal and coronal slices (Fig. 2d–f) with properties similar to those observed in horizontal slices of ventral hippocampus (Fig. 2g).


[

NUMBER 11

[

NOVEMBER 2005

Incidence (min–1)

Amplitude (mV)

d

Ripple frequency (Hz)


c

Site of origin and propagation of HFS-induced SPW-Rs As described for spontaneous SPW-Rs in vivo and in vitro, the SPW-R positive field potential transients that were recorded in stratum pyramidale of the CA3 and CA1 areas were associated with a negative field potential transient in the corresponding stratum radiatum (Fig. 3a). The origin of the SPW-Rs was confirmed in a set of experiments (n ¼ 6) in which cutting the connections between CA3 and CA1 after the fifteenth stimulus repetition resulted in the complete disappearance of CA1 SPW-Rs with persistence of those in area CA3 (Fig. 3b). In intact slice preparations, the evoked SPW-Rs, like spontaneous ones5, first appeared in area CA3 and propagated to area CA1 (latency, 4.2 ± 0.46 ms; n ¼ 10) and the subiculum (latency, 8.5 ± 0.6 ms; n ¼ 10) as shown by cross-correlation analysis (Fig. 3c). Because we recorded spontaneous SPW-Rs in some unstimulated ventral slices, we could compare the pharmacological properties of induced and spontaneous SPW-Rs. We found that spontaneous SPW-Rs were also generated in area CA3. Although stimulus-induced SPW-Rs had rather uniform amplitude and duration, the amplitude (0.2–0.8 mV) and duration (20–60 ms) of spontaneous SPW-Rs were more variable. The ripple frequency (177.2 ± 3.3 Hz; n ¼ 17), however, was comparable (P ¼ 0.28; Supplementary Fig. 1). As previously shown for spontaneous SPW-Rs15,16, we found that both blockade of AMPA/kainate receptors by CNQX (20 mM) and the disruption of gap junctions by carbenoxolone (200 mM) suppressed the expression of stimulus-induced SPW-Rs (Fig. 4a) and spontaneous SPW-Rs. This suggests dependence on excitatory synaptic coupling and involvement of gap junctions in synchronization. Conversely, neither HFS-induced (n ¼ 5; data not shown) nor spontaneous SPW-Rs (50 mM; n ¼ 5; Supplementary Fig. 1) were blocked by D-AP5, a competitive NMDA receptor antagonist. MK-801, a noncompetitive antagonist, also failed to block spontaneous SPW-Rs (50 mM; n ¼ 5). As previously shown for spontaneous SPW-Rs17, MK-801 and D-AP5 increased the incidence of stimulus-induced SPW-Rs by 16.4 ± 2.8% and 18.2 ± 4.9%, respectively.

1561

ARTICLES

b

Filtered 40–400 Hz

c

Auto-correlation

2 mV

0

–1.0 –20

0 Time (ms)

d

Horizontal

e

Sagittal

f

3rd

5th

7th

5th

0

100 200 300 Frequency (Hz)

200

100

0 2 mV

5 min

CA3

b

CA1

SR

Because of the Mg2+ sensitivity of NMDA receptors, we tested whether a reduction in extracellular Mg2+ concentration could affect induction of SPW-Rs. When the Mg2+ concentration was lowered to a more physiological concentration as observed in rat cerebrospinal fluid (close to 1 mM21), both TBS and HFS led to the appearance of SPW-Rs after the first stimulus train (n ¼ 7 and 4, respectively; Fig. 5a,c). Induction was also observed after orthodromic activation of area CA3 by stimulation of the stratum moleculare of the dentate gyrus and the mossy fiber pathway. Dentate gyrus stimulation was done in the infrapyramidal blade (n ¼ 6), the crest (n ¼ 8) and the suprapyramidal blade (n ¼ 4). Both stimulation paradigms, HFS and TBS, reliably induced SPW-R (Fig. 5b,d) as well as LTP (Fig. 5e,f) in area CA3. The induced SPW-Rs saturated after the second stimulus repetition (Fig. 5g), and their induction was also reversibly blocked by D-AP5 (50 mM; n ¼ 4). Ripple frequencies were comparable to those under a high Mg2+ concentration during both stimulus conditions (Fig. 5h). Stimulation of the mossy fiber pathway was done within area CA3c in coronal slices using stimulus intensities that were suprathreshold for the induction of population spikes. Mossy fiber stimulation typically increased the amplitude of evoked field potentials by about 300% when changing stimulus frequency from 0.1 Hz to 1 Hz. Repeated stimulation induced SPW-Rs (201.9 ± 7.3 Hz) after

c

CA1

CA1

2 mV

CA3

Cut Sub SP

CA3 CA1

CA3 2 mV

DG

50 ms

CA3

CA1 2 mV

CA1 Sub

CA3 SO

Figure 2 Sample recording of HFS-induced SPW-Rs. (a) Representative field potential recording of HFS-induced SPW-Rs recorded in stratum pyramidale of area CA3 (bottom). The 40–400 Hz band-pass filtered data (top) shows superimposed fast oscillatory activity (ripple). (b) Averaged auto-correlations with shaded s.e.m. of HFS-induced SPW-Rs (n ¼ 28; five slices) showing peaks at –5.8 ms and 5.8 ms (172 Hz). (c) Averaged power spectra with superimposed s.e.m. of HFS-induced SPW-Rs showing a prominent peak at B180 Hz (n ¼ 15; three slices, different from recordings in b). (d–f) Left: schematic drawing of tissue preparation for different slice orientations. Middle: induction of SPW-Rs in (d) horizontal, (e) sagittal and (f) coronal slices. SPW-Rs appear after the third stimulus application to stratum radiatum in area CA1. Calibration bar, 2 mV. Right: representative recording of induced SPW-Rs after the fifth stimulus repetition. (g) Average ripple frequency according to slice orientations: horizontal (197 ± 6.5 Hz; n ¼14), sagittal (177.5 ± 6.2 Hz; n ¼ 6) and coronal (200.4 ± 7.7 Hz; n ¼ 5).

50 ms

Parallels between LTP and SPW-R induction Stimulus-induced LTP in the associational network of area CA310 and in area CA111,18 has been shown to be sensitive to NMDA receptor antagonists. To test whether in our experimental paradigm the induction of SPW-Rs also depends on activation of NMDA receptors, we applied the noncompetitive NMDA receptor antagonist MK-801 or the competitive NMDA receptor antagonist D-AP5 before onset of stimulation. Under these conditions, SPW-Rs could not be induced (Fig. 4b). When D-AP5 was washed out before a new set of stimulations, SPW-Rs could then be induced (n ¼ 2; data not shown). Low-frequency stimulation (LFS) paradigms that induce long-term depression (LTD) of synaptic coupling (for review, see refs. 19,20) were able to reverse LTP that was induced by either HFS (data not shown) or TBS (Fig. 4c). The same depotentiation protocol (LFS; 900 pulses at 1 Hz) also reduced the incidence of stimulus-induced SPW-Rs (Fig. 4d). Under control conditions, the incidence of SPW-Rs/min after the seventh HFS was 7.9 ± 0.3 (n ¼ 7); the first LFS (n ¼ 7; after the seventh HFS) reduced the incidence to 0.7 ± 0.4, and a second LFS applied 10 min later reduced this to 0.2 ± 0.1. After a third LFS, SPW-Rs were completely abolished. When SPW-Rs had been suppressed by recurrent LFS, one or two HFSs could reestablish the SPW-Rs (n ¼ 4; data not shown). Laminar profile Amplitude (mV) 1 2 –1 0

0

g

Coronal

a

0.3

Frequency (Hz)


0.1 s

20

0.6

on Sa tal gi t C tal or on al

CA3

Power (10–3 mV2)

Correlation coefficient

1.0

Power spectra

H or iz

a

50 ms

50 ms

10 s

Figure 3 Origin of HFS-induced SPW-Rs. (a) Laminar profile of SPW-Rs in areas CA3 and CA1. Left: averaged amplitudes of SPW field potential transients in area CA3 (open circles) and CA1 (open triangles; n ¼ 20 from four slices). Middle and right: simultaneous sample recordings of induced SPW-R in stratum radiatum (SR), stratum pyramidale (SP) and stratum oriens (SO) of areas CA3 and CA1. (b) Left: schematic drawing illustrating cut between area CA3 and CA1 after induction of SPW-R (dashed line). Right: recordings of SPW-Rs in CA1 and CA3 after separating these two areas. Calibration bar, 2 mV. (c) Simultaneous recording of SPW-Rs from areas CA3 and CA1 and from subiculum to indicate propagation.

1562

VOLUME 8

[

NUMBER 11

[


ARTICLES

b CBX CNQX

8

12 SPW-Rs (min–1)

4 0 40 60 Time (min)

CTL MK-801 APV

4

80

2 mV

0

d

20 ms TBS

LFS

500 300 100

20

12 SPW-Rs (min–1)

20

FP CA3

∆ PS (%)

40 60 80 100 120 Time (min) LFS

8 CTL LFS

4

Figure 4 Stimulation dependence and pharmacological properties of SPW-Rs. (a) CNQX (20 mM) and carbenoxolone (CBX; 200 mM) reduced SPW-Rs. Average reduction was 100% for CNQX (n ¼ 5) and 89.6 ± 0.5% for carbenoxolone (n ¼ 5). (b) Evolvement of SPW-Rs during recurrent stimulation (HFS, arrow; control, CTL; n ¼ 10) was prevented by NMDA receptor antagonists D-AP5 and MK-801 (both 50 mM; n ¼ 5 each). (c) LTP in area CA3 that was induced by CA1 stratum radiatum TBS and subsequent depotentiation that was induced by LFS (1 Hz; 900 s; n ¼ 3), plotted as changes in population spike amplitude (D PS). Inset: Sample field potential (FP) recordings before and after induction of LTP, and after depotentiation. The amplitude of the first antidromically induced population spikes remains unaltered. (d) HFS-induced SPW-Rs (control, CTL; n ¼ 10) suppressed by LFS (n ¼ 7) to 0.7 ± 0.4 (first) and 0.2 ± 0.1 (second) SPW-Rs/min; full suppression occurred after the third LFS.

0 0

20

40 60 80 100 Time (min)

0

20

40 60 Time (min)

80

the second stimulus repetition and LTP of about 250% (n ¼ 4; Supplementary Fig. 2). We also tested whether stimulation with lower frequencies that were insufficient to induce LTP could lead to the generation of SPW-Rs. CA3 stratum radiatum stimulation with 5 Hz (n ¼ 3), 10 Hz (n ¼ 5) or 20 Hz (n ¼ 7) induced neither LTP nor SPW-Rs. Stimulation with 50 Hz (n ¼ 3) or 100 Hz (n ¼ 4), however, led to induction of both LTP and SPW-Rs (Supplementary Fig. 2). SPW-Rs associate with compound EPSPs and/or IPSPs in CA3 In subsequent experiments, we studied single-cell activity during SPWRs in area CA3, the region of SPW-R generation. We carried out simultaneous extra- and intracellular recordings with electrode tips separated by 70–300 mm. We analyzed recordings from 42 neurons with stable resting membrane potentials (less than –60 mV) and input resistance (40.5 ± 1.5 MO) and overshooting action potentials. Synchronously with stimulus-induced generation of SPW-Rs, neurons responded with an EPSP–inhibitory postsynaptic potential (IPSP) sequence, an IPSP-EPSP sequence or prominent IPSPs with no evidence of EPSPs (Fig. 6a–d), but never with isolated EPSPs. This indicates activation of inhibitory interneurons during SPW-Rs. EPSPs were determined by their capacity to evoke action potentials either at resting potential or on membrane depolarization from the resting potential, and in fact they often triggered one to four action potentials. The amplitudes of compound EPSPs varied between 6.4 and 18.2 mV (11.5 ± 0.2 mV; n ¼ 64; eight slices). In cells with prominent IPSPs, when the cell was depolarized to near threshold by intracellular current injection, firing was interrupted during SPW-Rs. Overall,

54.8% of cells responded with prominent EPSPs, whereas 45.2% responded with prominent IPSPs during the SPW-Rs. Because SPW-Rs increased in amplitude and frequency during recurrent stimulation, we hypothesized that an increasing number of CA3 pyramidal cells is recruited into SPW-Rs as a result of the activity within this associative network. To test our hypothesis, we followed changes in intracellular responses during SPW-R development in 24 CA3 pyramidal cells that were recorded for more than 45 min from the beginning of the stimulation. We identified four different patterns. First, in 10 of 24 cells (41.7%), we noted a gradual increase in EPSP amplitude and in the number of evoked action potentials in parallel to the extracellularly recorded increase in SPW-R amplitude (type A cell; Fig. 6a for sample cell). Second, in nine cells (37.5%), the increase in SPW-R amplitude was associated with a gradual increase in IPSP amplitude, which blocked action potential firing during SPW-Rs (type B cell; Fig. 6b). Third, in four cells (16.7%), we observed a gradual reduction in the amplitude of the early IPSP and a concomitant increase in the amplitude of the following EPSP, which

a


[

NUMBER 11

[

NOVEMBER 2005

2nd

3rd

DG stimulation

HFS 1st

4th

2nd

3rd

4th

2 mV

c

d

∆ PS (%)

500

HFSSM

f

500

300

0

TBSSM

300 100

100

g

1 min

TBS

TBS

e Figure 5 Facilitated SPW-R induction in the presence of a physiological concentration of extracellular Mg2+. (a) SPW-R induction by HFS of stratum radiatum in area CA1 in 1.2 mM Mg2+. (b) SPW-R induction by HFS of stratum moleculare in dentate gyrus (DG). (c) SPW-R induction by TBS of stratum radiatum in area CA1. (d) SPW-R induction by TBS of stratum moleculare in dentate gyrus. Calibration bars in d apply to all. (e) LTP in CA3 induced by HFS applied to stratum moleculare (SM) in dentate gyrus (n ¼ 8), indicated as DPS. (f) LTP in CA3 induced by TBS applied to stratum moleculare in dentate gyrus (n ¼ 5), indicated as DPS. (g) Evolvement of SPW-Rs (SPW-Rs/min; mean values ± s.e.m.) induced by CA1 (stratum radiatum, SR) and dentate gyrus (stratum moleculare) stimulation. (h) Average ripple frequencies: HFS CA1, 189.1 ± 3.9 Hz (n ¼ 4); HFS dentate gyrus, 194.3 ± 4.2 Hz (n ¼ 8); TBS CA1, 198.2 ± 5.1 Hz (n ¼ 7); TBS dentate gyrus, 201.3 ± 5.3 Hz (n ¼ 5).

b

CA1 stimulation HFS 1st

0

20 40 60 Time (min)

0

80

TBSSR HFSSR

16 SPW-Rs (min–1)


8

0 0

c

HFS

∆ PS (%)

SPW-Rs (min–1)

12

12

TBSSM HFSSM

8 4 0

h

10 15 20 Time (min)

20 40 60 Time (min)

80

TBS

200

100

0 5

0

HFS Frequency (Hz)

a

CA1

DG

CA1

DG

1563

ARTICLES

a

4th

b

7th

3rd

7th

c

d 3rd

6th

4th

6th

2 mV 50 ms 20 mV

e

f

120 100 80 60

SPW-R EPSP

40 3rd

4th

5th

6th

120 Change (%)

Change (%)


1s

100 80 60

SPW-R IPSP

40 7th

Figure 6 Different types of changes in neuronal behavior during evolvement of SPW-Rs. (a–d) Sample recordings of changes in neuronal behavior during induction of SPW-Rs in four CA3 pyramidal cells (top, SPW-R; bottom, intracellular recordings). (a) Early during development of SPW-Rs, the cell shows a short compound EPSP superimposed by one action potential. After the seventh train, the compound EPSP has increased in amplitude and duration and evokes four action potentials that are synchronous with some of the ripple deflections. (b) Cell with a small EPSP followed by a pronounced IPSP, which increases in amplitude and duration on further stimulation as indicated for sample recording after the seventh stimulation with no cell firing during SPW-Rs. Inset: during cell depolarization by current injection to evoke firing, action potential generation was interrupted during the SPW-R. Calibration bar in inset: 20 mV (top); 2 mV (bottom). (c) Cell that initially presents with an IPSP followed by an EPSP and later (after sixth stimulation train) with an augmented EPSP superimposed by two action potentials. (d) Cell where the EPSP amplitude decreases during the evolvement of the SPW-R. Calibrations in d also apply to all other unlabeled records. (e,f) Changes in (e) EPSP amplitude in cells showing type A behavior and (f) IPSP amplitude in cells having type B behavior with recurrent stimulation plotted together with SPW-R changes; n ¼ 2 cells in both plots. Postsynaptic potentials and SPW-Rs were normalized to 100% at end of stimulation period. Mean and s.e.m. calculated from last ten SPW-R and PSPs preceding next stimulus train.

3rd

4th

5th

6th

7th

We thus assume that selected synaptic connections between cells in the recurrent CA3 network become strengthened, in a manner that is dependent on NMDA receptors, with a subset of cells becoming linked and firing synchronously in a SPW-R. The participating network should then be input dependent. We therefore stimulated two different sites within area CA1, one of which induced action potentials (proximal) and one of which induced a delayed small EPSP (distal; Fig. 7b). We then induced SPW-Rs (Fig. 7c) by stimulating stratum radiatum in CA1 close to area CA3 (proximal). Once SPW-R activity was induced, we changed the stimulation site to distal area CA1. When we applied HFS to this site, it resulted in a shortening of the SPW-Rs and a reduction in the number of ripples per SPW-R from 5.3 ± 0.2 (n ¼ 158) to 2.5 ± 0.1 (n ¼ 93) without major changes in ripple frequency. The activity of a given cell that was recorded from intracellularly during SPW-Rs also changed (Fig. 7d). During SPW-Rs that were induced by proximal stimulation, the cell had EPSPs and superimposed action potentials; during SPW-Rs after distal stimulation, the cell ceased to fire action potentials. This pattern was reversible. Returning to stimulation at the proximal site resulted in recovery of the original SPW-R activity pattern, with an increase in the average number of ripples back to 5.1 ± 0.2 (n ¼ 70) and the re-establishment of the intracellularly recorded EPSPs and superimposed action potentials.

resulted in action potential generation during SPW-Rs (Fig. 6c). Fourth, the remaining neuron (4.2%) showed an increasing IPSP and a subsequently declining EPSP amplitude, which resulted in a blockade of action potential generation during SPW-Rs (Fig. 6d). As there was some variation in the intracellularly recorded signals from SPW-R to SPW-R, we analyzed in more detail the change in EPSPs in type A cells and the change in IPSPs in type B cells. We averaged the SPW-R amplitude of the ten SPW-Rs that preceded a given stimulus train and compared this average to the averaged EPSPs or IPSPs, respectively (Fig. 6e,f). Thus, in type A cells, the increase in SPW-R amplitudes from stimulus train to stimulus train was accompanied by a saturating increase in the intracellularly recorded compound EPSP amplitudes (Fig. 6e). In type B cells, however, the increase in SPW-R amplitude was associated with an increase in IPSP amplitude during evolvement of SPW-Rs (Fig. 6f). This analysis indicates that both the excitatory and inhibitory connectivity between cells in the CA3 region changes markedly and progressively during the evolution of the SPW-R. That SPW-Rs represent synchronized network activity to which many neurons may contribute is further supported by the following findings. In several experiments, we recorded extracellular action potentials that were synchronous with the SPW-Rs as well as with the intracellular action potentials generated by a type A cell. The intracellular action potentials were synchronous with the ripples and with a subset of the extracellularly recorded action potentials (Fig. 7a). Hyperpolarization of the single cell prevented action potential generation without significant effect on the evolved SPW-Rs. Whereas four to nine ripples were superimposed on SPWs, only 1.5 ± 0.2 action potentials (range, 1–4; n ¼ 431; seven slices) were superimposed on accompanying compound EPSPs. The minimal interval between two action potentials in a given cell corresponds to a maximum frequency of 94 Hz, which is well below the ripple frequency. Nevertheless, action potentials occurred within narrow time intervals before and after the negative peak of the ripples. The absolute value of this latency was found to be 0.7 ± 0.02 ms (n ¼ 431).

DISCUSSION SPW-Rs were induced with stimulation protocols known to induce LTP in vivo and in vitro10–12. The emergence and persistence of stimulusinduced SPW-Rs showed several parallels with changes in synaptic plasticity that result from the induction of LTP. Stimulation of commissural and associational fibers as well as activation of mossy fibers with protocols that induce LTP in area CA3 led to the induction of SPW-R. Conversely, stimulation protocols that do not induce LTP (stimulation with frequencies below 50 Hz) failed to induce SPW-Rs. Like LTP, the SPW-R induction depended on NMDA receptor activation and was abolished by LFS. The induction and expression, like LTP, was input dependent. Once SPW-Rs were established, as is the case with saturated LTP6,22,23, there was no further increase in amplitude or frequency upon ongoing stimulus repetition, and exposure to NMDA receptor antagonists did not block the induced SPW-Rs. Overall, this pattern of results indicates that SPW-R induction may represent a

1564

VOLUME 8

[

NUMBER 11

[


ARTICLES

a CA3

i.c.

2 mV

–61 mV

–66 mV 20 mV 50 ms


b

Proximal CA1

Distal CA1

CA3

4 mV

i.c.

50 mV 20 ms

c

Proximal CA1

Distal CA1

Proximal CA1 2 mV

CA3

50 mV

i.c. 10 s

d 2 mV

CA3

10 ms

i.c. 20 mV

100 ms

process of stimulus-dependent network reorganization in which NMDA receptor activation determines which of the participating synapses are strengthened, thus permitting self-organization of neuronal activity into SPW-Rs. Notably, spontaneous SPW-Rs have been observed in horizontal slices from ventral portions of the hippocampus of rats14,16 and of mice15, but not in parasagittal or coronal dorsal rat hippocampal slices. This fact may relate to differences in synaptic coupling24,25 and neuronal excitability along the longitudinal axis of the hippocampus. Also, it is well established that the ventral hippocampus is more prone to the generation of seizures than is the dorsal hippocampus26 while being less sensitive to ischemic damage27. That we were able to induce SPW-Rs in all types of sections indicates not only that our protocol is not merely facilitating a phenomenon that would spontaneously emerge anyhow but also that the longitudinally extended recurrent collaterals (coronal slices) and those running along the lamella (horizontal and parasagittal slices) are equally efficient for the generation of SPW-Rs, a finding that is relevant for the transmission and consolidation of information within the hippocampus. This indicates also that if sufficient recurrently connected circuits are available, population events underlying the SPW-R will self-organize. Here we show that in circuits that do not present with sufficient excitatory connectivity for such activity, SPW-Rs can be induced by mechanisms that strengthen synaptic connectivity similar to what has been described for LTP. Interictal discharges, like SPW-Rs, result from the synchronous activity of many neurons. Several pieces of evidence indicate that what we record may not represent interictal discharges. Although interictal responses are homogeneous within the affected network, SPW-Rs that were recorded simultaneously at different sites within area CA3 varied considerably in shape. Also, during interictal discharges, hippocampal principal cells, as well as many interneurons, have paroxysmal depolarization shifts with a unitary amplitude of 30–40 mV and a duration of more than 80 ms superimposed by bursts


Figure 7 Stimulus-specific cell response in area CA3 during HFS-induced SPW-R activity. (a) Top, SPW-R; bottom, intracellular recording (i.c.). Left: field potential recording of evoked SPW-Rs in area CA3 showing five extracellular action potentials in synchrony with ripples and a recruited CA3 pyramidal cell at resting membrane potential with a series of small EPSPs leading to the generation of two action potentials (truncated in this trace) that are synchronous with ripples. Right: analogous to left but during hyperpolarization of the same pyramidal cell by hyperpolarizing current injection by about 5 mV. Small EPSPs are in synchrony with ripple activity in field potential recording (bottom). (b) Evoked field potential (top) and intracellular response (bottom) to paired stimulation from stratum radiatum in area CA1 close to area CA2 (proximal, left) and from a stimulation site close to subiculum (distal, right). Note direct activation of CA3 cell as indicated by antidromic action potential generation in response to stimulation only from proximal CA1. (c) Effects of changing stimulation sites (top, extracellular recording; bottom, intracellular recording). (d) Sample recordings of SPW-Rs corresponding to c (top, extracellular recording; bottom, intracellular recording; inset, SPW-R on expanded time scale). Note EPSP and IPSP sequence with two superimposed action potentials after stimulation of proximal CA1 (left), changes in SPW-Rs and the intracellular response after stimulating distal CA1 (middle) and recovery of original SPW-Rs and intracellular response after switching stimulation site back to proximal CA1 (right).

[

NUMBER 11

[

NOVEMBER 2005

of action potentials (often more than ten)28,29 unlike stimulus-induced SPW-Rs, during which a single cell generated on average 1.5 action potentials. In addition, our study shows that almost half of the participating pyramidal cells had inhibitory potentials during SPW-Rs. Finally, although in vivo and in vitro spontaneous SPW-Rs had ripple frequencies at or below 200 Hz, bursts in chronic human epileptic tissue and rodent models of epilepsy are often accompanied by higher frequencies of up to 500 Hz30. We consistently found that intracellularly recorded responses occurred in synchrony with the ripples, a finding that supports the notion that SPW-Rs result from synchronized neuronal activity. The precision of spike firing in CA3 during stimulus-induced EPSPs is rather coarse, with variations in spike induction in the 10-ms range31,32. Here we found synchrony in the submillisecond range. This high synchrony suggests that the transmission underlying these phenomena is not just chemical. In the present study, the induction of SPW-Rs was indeed abolished by a gap junction blocker, indicating that electrical transmission may be involved in the observed high synchrony. This is in agreement with previous evidence of the presence of electrical coupling in the hippocampus33 and its contribution to fast oscillatory activity34,35. Single CA3 pyramidal neurons responded with either prominent EPSPs (54.8%) or prominent IPSPs (45.2%) during the induction of SPW-Rs. The increase in amplitude that was observed during the initial stages of SPW-R induction was paralleled by changes in the synaptic strengths of participating individual cells. Stimulus-dependent increases in the efficacy of synaptic coupling between pyramidal cells have been previously observed36, but these were found in association with reduced synaptic inhibition. In our experimental conditions, the increase in spiking output relative to baseline (induction of SPW-Rs) was associated with increased inhibition. This might be a mechanism to limit the (potentially dangerous) increase in excitability of the network. Indeed, after a certain number of stimulus repetitions, there was no further increase in incidence and amplitude of SPW-Rs, indicating saturation of such network activity on a new level of balanced excitation and inhibition. Processes that lead to some maintenance of overall excitability in the network have been described in several systems37–42. Such homeostatic mechanisms may become activated during the induction of SPW-Rs, thus preventing the transition from

1565


ARTICLES SPW-Rs to epileptiform discharges. Our findings indicate that synaptic mechanisms may also contribute to a limitation in the overall increase of network activity. Whether information is permanently stored in the hippocampus is controversial43,44, but it is generally agreed that the hippocampus is involved in the stabilization of memories across cortical structures, a process known as memory consolidation. SPW-Rs have been suggested to represent stored information that is replayed and transferred to the cortex during quiet waking and slow-wave sleep3. That SPW-Rs might be involved in the generation of synaptic changes downstream is suggested by the fact that LTP can be induced by stimulation with ripple-like activity patterns45. SPW-Rs themselves seem to be sensitive to changes in synaptic plasticity, however. The incidence of SPW-Rs, for example, is increased after learning46, as well as after sleep deprivation47. Also, HFS of the commissural and associational network in vivo increases the incidence and amplitude of SPW-Rs48. Our finding that SPW-Rs can be induced by standard LTP protocols, and that their incidence is diminished by LFS, is thus in line with the idea that learning-induced synaptic plasticity is associated with the generation of SPW-Rs and potentially with memory consolidation. METHODS Slice preparation. Wistar rats (aged 5–8 weeks; 160–250 g) of either sex were decapitated under deep ether anesthesia, and their brains were rapidly removed and washed in cold artificial cerebrospinal fluid (ACSF) containing 129 mM NaCl, 21 mM NaHCO3, 3 mM KCl, 1.6 mM CaCl2, 1.8 mM MgSO4, 1.25 mM NaH2PO4 and 10 mM glucose (saturated with 95% O2/5% CO2). In a set of additional experiments, the MgSO4 concentration was lowered to 1.2 mM. Animal procedures were approved by local animal ethics committees. Horizontal hippocampal slices (400 mm; at bregma –4.7 to –7.3 mm) at an angle of 121 in the fronto-occipital direction were obtained using a vibratome (752 M Vibroslice; Cambden Instruments) in cooled and oxygenated ACSF. The slices were immediately transferred to an interface chamber and were perfused with carbogenated ACSF (36 ± 0.5 1C; flow rate, 1.6 ml/min; pH 7.4). Slices with spontaneous SPW-Rs were usually obtained between –6.3 and –7.3 mm from the bregma. Dissection between CA3 and CA1 was made with a micro cutter (Fine Science Tools). Dorsal hippocampus was studied using parasagittal and coronal slices. Recordings. Extracellular field potentials were recorded from stratum pyramidale of CA3b, CA1, and, where stated, from subiculum, with microelectrodes filled with 154 mM NaCl (5–10 MO). For detection of changes in extracellular potassium concentration ([K+]o), double-barreled ion-sensitive microelectrodes were manufactured and calibrated as described previously49. Ionsensitive microelectrodes with a sensitivity of 59 ± 2 mV to a tenfold increase in K+ were used. Activity-induced increases in [K+]o were calculated from recorded potential changes by a modified Nernst equation49. Intracellular electrodes were pulled from borosilicate glass (o.d.,1.2 mm) and were filled with 2 M K+-acetate (70–90 MO). Recordings were amplified by a homemade amplifier and a SEC 05L (NPI Instruments), low-pass filtered at 1 kHz, digitized at 5 kHz for extracellular and 10 kHz for intracellular recordings and stored on computer disk using a CED 1401 interface (Cambridge Electronic Design).

placed in stratum radiatum in CA3c. Field potential facilitation was examined by switching single-pulse stimulation from 0.1 Hz to 1 Hz. Slices were stimulated using a stimulus intensity (1.8–3 V) that was submaximal (70%) of that required to evoke a maximal amplitude population spike in response to single-pulse stimulation in CA3 and CA1. Therefore, in CA3 the amplitude of the second population spike (population response evoked by the induced EPSPs) was calculated. During stimulation, increases in [K+]o were recorded. LTP was induced in CA3 by applying either single TBS or three trains of HFS (400 ms; 100 Hz; 40-s interval). LTD was induced by 1-Hz stimulation for 900 s (repeated one to three times; 10-min interval). Data analysis. The oscillation power of the raw data was determined from the power spectrum as the integral between 0 and 400 Hz (Spike 2 software, Cambridge Electronic Design). For ripple detection, raw data were filtered with a Spike 2 software band-pass filter of 40–400 Hz (threshold: 4–6 times the s.d. of eventless baseline noise). For sharp wave (SPW) detection, recordings were low-pass filtered at 20 Hz. The temporal precision between action potentials and ripple activity was calculated by 40–400 Hz band-pass filtering of field potential raw data. The time interval between the peak of the action potential and the nearest negative peak of the ripple was determined by use of custommade software (H. Siegmund, Institute for Neurophysiology, Charite´). Autoand cross-correlations were calculated within 500-ms epochs. Dominant frequency was determined from the interval between the first and second peak in auto-correlograms. For the calculation of signal delay between different areas, raw data were low-pass filtered, and latency was determined by the positive peak in the average cross-correlation analysis. If not denoted otherwise, data were reported as mean ± s.e.m. Statistical significance was determined using a one-way ANOVA or the Wilcoxon test using Origin software (version 6; Microcal Software). P o 0.05 was considered to be significant. Drugs. All drugs were dissolved in ACSF and applied by continuous bath perfusion. These included 20 mM CNQX (6-cyano-7-nitro-quinoxaline-2, 3-dione), 50 mM MK-801, 200 mM carbenoxolone and 50 mM D-AP5 (D-(–)-2-amino-5-phosphonopentanoic acid). All drugs were purchased from Sigma-Aldrich Chemie except MK-801 (Tocris/Biotrend). For the analysis of pharmacological effects, the incidence of SPW-Rs and ripple frequency were determined under control conditions, during drug application and after washout, where possible. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS This research was supported by the Sonderforschungsbereich 515. We are grateful for discussions with M.J. Gutnick and D. Schmitz, and for technical assistance and the development of data analysis tools by H. Siegmund and H.J. Gabriel. We acknowledge participation of N. Maggio in some of the recordings in dorsal hippocampal slices. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

Stimulation protocols. SPW-Rs were induced by tetanic stimulation (three trains of 40 pulses, 10-ms interval, 100-ms duration, 40-s intertrain interval; repeated up to 15 times every 5 min) or by TBS (12 trains of four pulses at 100 Hz, 10-ms interval, 100-ms duration, 200-ms intertrain interval) with a bipolar platinum electrode (50 mm, 140–200 mm tip separation) placed in stratum radiatum of CA1. In additional experiments, the stimulation site was shifted to distal stratum radiatum near subiculum, to stratum moleculare of the dentate gyrus, to the hilus or to stratum radiatum in CA3. In further experiments, tetanic stimulation was applied to stratum radiatum in CA3 using frequencies of 5, 10, 20, 50 and 100 Hz. The mossy fiber pathway was stimulated in coronal slices using a monopolar platinum electrode (20 mm)

1. Squire, L.R. Memory and the hippocampus: a synthesis from findings with rats, monkeys, and humans. Psychol. Rev. 99, 195–221 (1992). 2. Wiltgen, B.J., Brown, R.A., Talton, L.E. & Silva, A.J. New circuits for old memories: the role of the neocortex in consolidation. Neuron 44, 101–108 (2004). 3. Buzsaki, G. Two-stage model of memory trace formation: a role for ‘‘noisy’’ brain states. Neuroscience 31, 551–570 (1989). 4. Buzsa´ki, G. Memory consolidation during sleep: a neurophysiological perspective. J. Sleep Res. 7, 17–23 (1998). 5. Buzsaki, G., Leung, L.W. & Vanderwolf, C.H. Cellular bases of hippocampal EEG in the behaving rat. Brain Res. 287, 139–171 (1983). 6. Bliss, T.V.P. & Lømo, T. Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path. J. Physiol. (Lond.) 232, 331–356 (1973). 7. Dunwiddie, T. & Lynch, G. Long-term potentiation and depression of synaptic responses in the rat hippocampus: Localization and frequency dependency. J. Physiol. (Lond.) 276, 353–367 (1978). 8. Martin, S.J. & Morris, R.G. New life in an old idea: the synaptic plasticity and memory hypothesis revisited. Hippocampus 12, 609–636 (2002).

1566

VOLUME 8

[

NUMBER 11

[



ARTICLES 9. Kandel, E.R. The molecular biology of memory storage: a dialogue between genes and synapses. Science 294, 1030–1038 (2001). 10. Harris, E.W., Ganong, A.H. & Cotman, C.W. Long-term potentiation in the hippocampus involves activation of N-methyl-D-aspartate receptors. Brain Res. 323, 132–137 (1984). 11. Bliss, T.V.P. & Collingridge, G.L. A synaptic model of memory: Long-term potentiation in the hippocampus. Nature 361, 31–39 (1993). 12. Zalutsky, R.A. & Nicoll, R.A. Comparison of two forms of long-term potentiation in single hippocampal neurons. Science 248, 1619–1624 (1990). 13. Kann, O., Schuchmann, S., Buchheim, K. & Heinemann, U. Coupling of neuronal activity and mitochondrial metabolism as revealed by nad(p)h fluorescence signals in organotypic hippocampal slice cultures of the rat. Neuroscience 119, 87–100 (2003). 14. Papatheodoropoulos, C. & Kostopoulos, G. Spontaneous, low frequency (approximately 2–3 Hz) field activity generated in rat ventral hippocampal slices perfused with normal medium. Brain Res. Bull. 57, 187–193 (2002). 15. Maier, N., Nimmrich, V. & Draguhn, A. Cellular and network mechanisms underlying spontaneous sharp wave-ripple complexes in mouse hippocampal slices. J. Physiol. (Lond.) 550, 873–887 (2003). 16. Colgin, L.L., Kubota, D., Jia, Y., Rex, C.S. & Lynch, G. Long-term potentiation is impaired in rat hippocampal slices that produce spontaneous sharp waves. J. Physiol. (Lond.) 558, 953–961 (2004). 17. Colgin, L.L., Jia, Y., Sabatier, J.M. & Lynch, G. Blockade of NMDA receptors enhances spontaneous sharp waves in rat hippocampal slices. Neurosci. Lett. 385, 46–51 (2005). 18. Collingridge, G.L., Kehl, S.J. & McLennan, H. Excitatory amino acids in synaptic transmission in the Schaffer collateral-commissural pathway of the rat hippocampus. J. Physiol. (Lond.) 334, 33–46 (1983). 19. Bear, M.F. & Abraham, W.C. Long-term depression in hippocampus. Annu. Rev. Neurosci. 19, 437–462 (1996). 20. Malenka, R.C. & Bear, M.F. LTP and LTD: an embarrassment of riches. Neuron 44, 5–21 (2004). 21. Windmuller, O. et al. Ion changes in spreading ischaemia induce rat middle cerebral artery constriction in the absence of NO. Brain 128, 2042–2051 (2005). 22. Dudek, S.M. & Bear, M.F. Bidirectional long-term modification of synaptic effectiveness in the adult and immature hippocampus. J. Neurosci. 13, 2910–2918 (1993). 23. Moser, M.B. & Moser, E.I. Pretraining and the function of hippocampal long-term potentiation. Neuron 26, 559–561 (2000). 24. Amaral, D.G. & Witter, M.P. The three-dimensional organization of the hippocampal formation: A review of anatomical data. Neuroscience 31, 571–591 (1989). 25. Amaral, D.G. & Witter, M.P. The hippocampal formation. in The Rat Nervous System (ed. Paxinos, G.) 443–494 (Academic, New York, 1995). 26. Bragdon, A.C., Taylor, D.M. & Wilson, W.A. Potassium-induced epileptiform activity in area CA3 varies markedly along the septotemporal axis of the rat hippocampus. Brain Res. 378, 169–173 (1986). 27. Ashton, D., Van Reempts, J., Haseldonckx, M. & Willems, R. Dorsal-ventral gradient in vulnerability of CA1 hippocampus to ischemia: A combined histological and electrophysiological study. Brain Res. 487, 368–372 (1989). 28. Wong, R.K.S., Traub, R.D. & Miles, R. Cellular basis of neuronal synchrony in epilepsy. Adv. Neurol. 44, 583–592 (1986).


[

NUMBER 11

[

NOVEMBER 2005

29. Johnston, D. & Brown, T.H. The synaptic nature of the paroxysmal depolarizing shift in hippocampal neurons. Ann. Neurol. 16 (Suppl.), S65–S71 (1984). 30. Bragin, A., Engel, J., Jr., Wilson, C.L., Fried, I. & Mathern, G.W. Hippocampal and entorhinal cortex high-frequency oscillations (100–500 Hz) in human epileptic brain and in kainic acid-treated rats with chronic seizures. Epilepsia 40, 127–137 (1999). 31. Fricker, D. & Miles, R. EPSP amplification and the precision of spike timing in hippocampal neurons. Neuron 28, 559–569 (2000). 32. Axmacher, N. & Miles, R. Intrinsic cellular currents and the temporal precision of EPSPaction potential coupling in CA1 pyramidal cells. J. Physiol. (Lond.) 555, 713–725 (2004). 33. Dudek, F.E., Snow, R.W. & Taylor, C.P. Role of electrical interactions in synchronization of epileptiform bursts. Adv. Neurol. 44, 593–617 (1986). 34. Draguhn, A., Traub, R.D., Schmitz, D. & Jefferys, J.G.R. Electrical coupling underlies high-frequency oscillations in the hippocampus in vitro. Nature 394, 189–192 (1998). 35. Schmitz, D. et al. Axo-axonal coupling: A novel mechanism for ultrafast neuronal communication. Neuron 31, 831–840 (2001). 36. Miles, R. & Wong, R.K.S. Latent synaptic pathways revealed after tetanic stimulation in the hippocampus. Nature 329, 724–726 (1987). 37. Hirase, H., Leinekugel, X., Czurko, A., Csicsvari, J. & Buzsaki, G. Firing rates of hippocampal neurons are preserved during subsequent sleep episodes and modified by novel awake experience. Proc. Natl. Acad. Sci. USA 98, 9386–9390 (2001). 38. King, C., Henze, D.A., Leinekugel, X. & Buzsaki, G. Hebbian modification of a hippocampal population pattern in the rat. J. Physiol. (Lond.) 521, 159–167 (1999). 39. Turrigiano, G.G. & Nelson, S.B. Hebb and homeostasis in neuronal plasticity. Curr. Opin. Neurobiol. 10, 358–364 (2000). 40. Buzsaki, G. et al. Homeostatic maintenance of neuronal excitability by burst discharges in vivo. Cereb. Cortex 12, 893–899 (2002). 41. Dragoi, G., Harris, K.D. & Buzsaki, G. Place representation within hippocampal networks is modified by long-term potentiation. Neuron 39, 843–853 (2003). 42. Royer, S. & Pare, D. Conservation of total synaptic weight through balanced synaptic depression and potentiation. Nature 422, 518–522 (2003). 43. Nadel, L. & Moscovitch, M. Hippocampal contributions to cortical plasticity. Neuropharmacology 37, 431–439 (1998). 44. Squire, L.R. & Alvarez, P. Retrograde amnesia and memory consolidation: a neurobiological perspective. Curr. Opin. Neurobiol. 5, 169–177 (1995). 45. Buzsa´ki, G., Haas, H.L. & Anderson, E.G. Long-term potentiation induced by physiologically relevant stimulus patterns. Brain Res. 435, 331–333 (1987). 46. Kudrimoti, H.S., Barnes, C.A. & McNaughton, B.L. Reactivation of hippocampal cell assemblies: effects of behavioral state, experience, and EEG dynamics. J. Neurosci. 19, 4090–4101 (1999). 47. Ponomarenko, A.A., Lin, J.S., Selbach, O. & Haas, H.L. Temporal pattern of hippocampal high-frequency oscillations during sleep after stimulant-evoked waking. Neuroscience 121, 759–769 (2003). 48. Buzsaki, G. Long-term changes of hippocampal sharp-waves following high frequency afferent activation. Brain Res. 300, 179–182 (1984). 49. Heinemann, U. & Arens, J. Production and calibration of ion-sensitive microelectrodes. in Practical Electrophysiological Methods: a Guide for In Vitro Studies in Vertebrate Neurobiology (eds. Grantyn, R. & Kettenmann, H.) 206–212 (Wiley-Liss, New York, 1992).

1567


ARTICLES

Encoding a temporally structured stimulus with a temporally structured neural representation Stacey L Brown, Joby Joseph & Mark Stopfer Sensory neural systems use spatiotemporal coding mechanisms to represent stimuli. These time-varying response patterns sometimes outlast the stimulus. Can the temporal structure of a stimulus interfere with, or even disrupt, the spatiotemporal structure of the neural representation? We investigated this potential confound in the locust olfactory system. When odors were presented in trains of nearly overlapping pulses, responses of first-order interneurons (projection neurons) changed reliably, and often markedly, with pulse position as responses to one pulse interfered with subsequent responses. However, using the responses of an ensemble of projection neurons, we could accurately classify the odorants as well as characterize the temporal properties of the stimulus. Further, we found that second-order follower neurons showed firing patterns consistent with the information in the projection-neuron ensemble. Thus, ensemble-based spatiotemporal coding could disambiguate complex and potentially confounding temporally structured sensory stimuli and thereby provide an invariant response to a stimulus presented in various ways.

Visual1–4, auditory5–7, tactile8–10, and olfactory11–15 senses all use temporal coding mechanisms16 for which the timing, rather than just the rate, of action potentials is important for the neural representation of the stimulus17,18. If a neural response pattern significantly outlasts the stimulus, new stimuli might arrive before the response to a previous stimulus is complete19. This situation is particularly interesting in olfaction; in principal neurons, odor pulses can elicit response patterns that endure beyond the pulse’s offset11,14 and, further, natural odor plumes can have repeated, rapid and nearly overlapping encounters with olfactory receptors20. Given these potential confounds, how can neural systems use spatiotemporal representations to encode and decode temporally structured stimuli? In the locust, projection neurons in the antennal lobe—the analogs of vertebrate mitral cells21—respond to odors with complex spiking patterns that consist of epochs of excitation, inhibition and quiescence. These patterns vary with odor identity and concentration, and can greatly outlast the odor receptor neurons’ encounter with the odor11,14. The antennal lobe’s network of excitatory projection neurons and inhibitory local neurons generates these firing patterns when driven by input from olfactory receptor neurons22,23. These spiking patterns, distributed broadly across the projection neuron population, are parsed into brief segments by odor-evoked oscillations and are read by downstream follower neurons (the Kenyon cells) which receive convergent input from many projection neurons15,24 (R.A. Jortner and G. Laurent, Soc. Neurosci. Abstr. 412, 21, 2004). The oscillating projection neuron ensemble, through feed-forward inhibition, generates brief integration windows in the Kenyon cells24. Kenyon cells respond to odors with very sparse spiking, often demonstrating great specificity with respect to odors and even particular concentrations of odors25,26.

Here we examine how the locust olfactory system responds to very short (100-ms) repeated odor pulses; the timing was chosen to approximate that observed in natural odor plumes20. In a turbulent environment, odor is carried in an intermittent fashion in the form of filaments of odor-laden air that vary in concentration, duration and frequency with which they appear. These factors are modulated by wind speed, amount of turbulence, delivery mechanism and distance from the odor source20,27. We selected stimulus parameters on the basis of odor plume measurements made outdoors, in which filaments were observed to encounter a sensor in a series of bursts that averaged about 100 ms in duration and arrived, on average, at approximately 500-ms intervals20. We made intracellular recordings from projection neurons and extracellular tetrode recordings from projection neurons and Kenyon cells in adult locusts. We delivered brief (100-ms) odor pulses in trains of 3 or 10 pulses; inter-pulse intervals ranged from 500 to 1,250 ms but were constant within a train. For each pulse pattern, we delivered blocks of 10 trials (15–30 s inter-trial interval), with the blocks given in random order. We used a variety of odorants and concentrations (see Methods). Our results showed that ensemble-based coding mechanisms can disambiguate complex and potentially confounding temporally structured sensory stimuli, thus providing an invariant response to a stimulus presented in different ways while preserving information about the stimulus timing. RESULTS Projection neuron responses vary with odor pulse pattern For most odor–projection neuron combinations, the number of spikes elicited by odor pulses changed reliably, and sometimes substantially, with pulse position, because lengthy responses to one pulse interfered

National Institute of Child Health and Human Development, US National Institutes of Health, Building 35, Room 3A-102, Bethesda, Maryland 20892, USA. Correspondence should be addressed to M.S. ([email protected]). Received 15 August; accepted 19 September; published online 16 October 2005; doi:10.1038/nn1559

1568

VOLUME 8

[

NUMBER 11

[


ARTICLES 0.1 HEX

1 OCT

Figure 1 The responses of projection neurons change with odor and inter-pulse intervals. Rapid trains of odor pulses evoked responses that interfered with one another (see text). Example of a projection neuron (intracellular recording), responding to three odors (columns) presented in four patterns (rows). Black bars on abscissa: odor presentation times. Gray histograms, responses to ten consecutive trials exemplified by the intracellular trace; gray lines, simultaneous EAG recordings indicating afference from the antenna; note that the responses of the projection neurons may outlast this afference. Top row, single 100-ms pulse; second row, train of three 100-ms pulses with a 1,250-ms IPI; third row, train of three 100-ms pulses with a 500-ms IPI; bottom row, single 1,000 ms pulse. Scale bar, 10 mV. 0.1 HEX, diluted hexanol; 1 OCT, octanol; 0.1 T2H, diluted trans-2-hexen-1-ol.

0.1 T2H

Frequency (Hz)

80 40 0

80 40 0

80 40 0 –2

0

2

4

6

0

2 Time (s)

4

6

0

with responses to subsequent pulses. We recorded intracellularly from 211 projection neurons (in 92 experiments) and extracellularly (with tetrodes) from 117 projection neurons (in 14 experiments) and 67 Kenyon cells (in 10 experiments). In a typical intracellular recording (Fig. 1), the projection neuron responded reliably to a single, brief pulse of diluted (0.1) hexanol (see Methods) with a quick burst of spikes which was followed by another, smaller burst and a lengthy hyperpolarization. The projection neuron also spiked reliably in response to each of the pulses in a three-pulse train (with inter-pulse intervals, IPIs, of 500 ms and 750 ms). Note that the electroantennogram (EAG) response decreased in amplitude over the pulse trains, probably indicating olfactory receptor adaptation; this recovered within seconds, before the next trial. For another odorant (1 octanol), single pulses elicited only inhibition, whereas trains of three pulses Cell 8 0.01 HXA 1 pulse

6

reliably elicited spikes for the second and third pulses (but not the first). For 0.1 trans-2-hexanol, single pulses and trains with a 750-ms IPI elicited reliable spiking, whereas trains with the 500-ms IPI elicited reliable spiking only for the first and third pulses (but not the second). Responses to trains of odor pulses often included strong and reliable bursts of spikes after the last pulse; these ‘after-responses’ became longer and more intense as the duration of individual odor pulses increased (see, for example, Fig. 1, bottom, 1-s odor pulse). The timing and intensity of excitatory and inhibitory odor response components in this projection neuron reflected multiple underlying mechanisms (see Discussion) and determined how closely the bursts of spikes could track each separate odor pulse within a train. We obtained similar results from extracellular recordings made simultaneously from multiple projection neurons (Fig. 2). Different

2

3

4

5

6

7

8

9

0.01 1

3 pulses, 1,250 ms IPI

HXA

10 pulses, 500 ms IPI 4 Time (s)

4

OCT

3 pulses, 500 ms IPI

0

2

Cell 1

Trial


80 40 0

0.01

8 1

Figure 2 The responses of projection neurons to odors and the extent of interference between overlapping HEX responses patterns vary greatly depending on the odor 0.01 and the cell. Spike-time raster plots show simultaneous extracellular recordings from nine projection neurons, in 1 response to four odor-pulse delivery patterns (ten trials each) for four odorants, each at two concentrations. Inset, enlarged example of a particular combination of odor and GER 0.01 projection neuron. Gray bars indicate odor delivery times. Stimuli were delivered in random order but are organized here in the sequence shown in the inset. Note the variety 1 of responses observed in different projection neurons and to different odors and concentrations.


[

NUMBER 11

[

NOVEMBER 2005

1569

ARTICLES Figure 3 Over time, the responses of projection neurons varied more across odors and concentrations than across trials. The average Euclidian distance (see text and Methods) for every 50-ms snapshot of the projection-neuron ensemble response is shown for each of the odor pulse patterns tested (indicated by red bars on the abscissa). In the ensemble response over time, repeated trials of the same odor and concentration elicited small variations from the baseline (red line), whereas trials with changes in concentration of the same odor (green) or changes of odors at the same concentrations (blue) elicited, respectively, larger variations from the baseline and from each other. From top: single 100-ms pulse; three pulses with a 500-ms IPI; three pulses with a 1,250-ms IPI; ten pulses with a 500-ms IPI. Error bars, s.d.

100

50

0

Euclidean distance (arbitrary units)

50

and 3% to all but the last pulse. Thus, for most projection neurons, the temporal structure of the odor interfered with the temporal structure of its neural representation.

0 100

50

0 100

50

0 –2

0

2

4

6

8

Time (s)

projection neurons responded in different ways to the same stimulus pulses. In some neurons, spikes elicited by some odors faithfully tracked every pulse in a train (see, for example, the response of cell 8 to 0.01 hexanal, Fig. 2 inset); however, this was atypical, and such a projection neuron responded differently when the animal was presented with a different odor (see cell 8’s response to 1 OCT). None of the 117 projection neurons in our extracellular set were able to reliably track the pulse timing of all odor-concentration combinations. Of the 936 odor–projection neuron combinations in our set, 719 elicited spikes. For most of these combinations, the number of spikes generated varied significantly (P o 0.05; see Methods) with pulse position: 59% showed significant changes over three pulses with a 500-ms IPI, and 68% showed significant changes over ten pulses with a 500-ms IPI. In about one-third of the cases in which spikes were elicited, the changes were substantial. For example, 10% of the odor– projection neuron combinations resulted in spiking in response to only the first pulse; 10% to all but the first pulse; 7% to only the last pulse;

Analyzing responses of an ensemble of projection neurons Multiple projection neurons converge onto their follower neurons (which include Kenyon cells; R.A.J. and G.L., Soc. Neurosci. Abstr. 412, 21, 2005), and odors are represented by spiking activity distributed across ensembles of projection neurons11,14. We considered that, despite the interference-induced variability observed in the responses of individual projection neurons, invariant properties of the stimulus might emerge at the ensemble level. Thus, we pooled our set of extracellularly recorded projection neurons so as to approximate a portion of the projection-neuron population found in the antennal lobe (see Methods) and then examined the set’s information content. We sought to analyze the projection-neuron ensemble response as it evolved during and after each stimulus presentation. Therefore, we binned the firing patterns of each projection neuron into brief (50-ms or 100-ms) time segments and then constructed, for each time segment, a separate response vector consisting of the binned firing of each of the 117 projection neurons26. Thus, each vector had 117 elements and represented a ‘snapshot’ of the transient coactivity of projection neurons in the ensemble at a given point in time. This technique allowed us to make rigorous comparisons between responses. The vectors could be treated as points within a 117-dimensional response space; the Euclidean distance between these points provided a measure of the similarity of the response patterns. By considering a series of vectors representing consecutive points in time, we were able to compare responses as they evolved. We found that (as is already evident from the examples in Figs. 1 and 2) the responses were reliable (Fig. 3): (i) repeated trials with the same odor and concentration elicited

a

3

b

4 3

2

2 1 1

Figure 4 Cross-correlations indicate that the response of the projectionneuron ensemble evolves gradually over the duration of the response. (a–d) For each delivery pattern of the 1 hexanol odorant, the cross-correlation coefficients (P o 0.001; see Methods) were calculated between consecutive 50-ms bins of the ensemble response. The cross-correlations obtained from other odorants were similar. Comparing the response to a single pulse with itself over time shows a correlated region between 150- and 300-ms wide, indicating that the subset of projection neurons active during the response evolves gradually. Also, responses to any given pulse in a train were highly correlated with other responses to pulses in that train. Red bars, odor pulse times. (a) Single 100-ms pulse; (b) three pulses with a 500-ms IPI; (c) three pulses with a 1,250-ms IPI; (d) ten pulses with a 500-ms IPI. Note the different time scales.

1570

0

0 0

c

1

2

3

0

1

2

3

4

d

5

8

1

4 Time (s)


100

6

3

0.8 0.6

4

2

0.4 2

1

0.2 0

0

0 0

VOLUME 8

1

[

2 3 Time (s)

4

NUMBER 11

5

[

0

2

4

6

8


ARTICLES

a

b Single pulse 4

Three pulses 4

100-ms duration

0

0 5

2 2

0

0 2

5 0

0 2

4 5

c

4 5

d

Three pulses 4 LLE 3

Ten pulses 4

500-ms IPI

2

2

0 2 2

500-ms IPI

0 5 0 LLE 2

2

2 2

0 LLE 1

5 0 2

4 5 GER

0

OCT

4 5

HEX

HXA

responses that were similar to one another; (ii) changing the odor concentration elicited response patterns that quickly began to differ from baseline and also from patterns elicited by different concentrations; and (iii) changing the odor identity elicited response patterns that were even further from baseline and from each other. The patterns of activity elicited by the odorants changed gradually during and after the stimulus for any given coactivity snapshot, ensemble responses were highly and significantly correlated with the responses obtained from snapshots before and after it as is evident in the correlation plots (Fig. 4a). Further, responses to multiple odor pulses were highly and significantly correlated with each other (Fig. 4b–d). Notably, relatively abrupt changes in correlation occurred during the pulse trains, as new responses appeared to truncate ongoing responses (Fig. 4b–d).

patterns of the projection-neuron ensemble; thus they illustrate the fact that gradually changing subgroups of projection neurons within the ensemble were transiently coactive during and after the odor presentation. Trajectories representing repeated trials were superimposable, which indicates that ensemble responses were reliable and reproducible. The trajectories elicited by different odorants looped through separate regions of the response space because different odors transiently activated different subgroups of projection neurons within the ensemble. Also, we found that each odor pulse within a train, regardless of the IPI, elicited a trajectory that looped through and re-circled the same odor-specific regions of response space as did the single odor pulse (Fig. 5b–d). The response of the projection-neuron ensemble, considered piecewise, was odor specific and invariant with respect to the odor-pulse pattern.

100 75 50

Visualizing ensemble responses to odor pulses To visualize the gradually evolving ensemble responses, we first reduced the dimensionality of the snapshot vectors using local linear embedding (LLE), a technique suitable for high-dimensional, nonlinear and gradually changing data26,28 (see Methods). Then, we plotted the first three dimensions so as to obtain a series of points that, when joined in sequence, formed a trajectory representing the response over time to each odor presentation (Fig. 5). Single, brief pulses of odorant elicited simple response trajectories looping away from, and then back to, the rest point (Fig. 5a). These trajectories were constructed by sampling, in brief time bins, the firing

25 0 100 75 50 Percentage correct


2 2

1,250-ms IPI

2

2

Figure 5 Visualization of the projection-neuron ensemble responses over time reveals invariant odor-specific trajectories, regardless of odor delivery pattern. (a–d) Different odor delivery patterns; colors indicate different odorants. The trajectories, describing the evolution of the ensemble response over time, return to the same areas of response space over repeated trials and remain in the same orientation, regardless of odor delivery pattern. The longer IPI allows the trajectory to return to the spontaneous state after each pulse; however, the shorter IPI causes the trajectories to loop away from the spontaneous state within the odor space. Three-dimensional embedding using LLE was computed for each time point (100-ms bins), for 5 s from the onset of the first odor pulse, averaged over three trials. Consecutive time points were joined together to form distinct trajectories for each odor response. Short black lines indicate the trajectory point 0.5 s after the onset of the stimulus; gray arrows indicate the direction of the trajectory over time. GER, geraniol; OCT, octanol; HEX, hexanol; HXA, hexanal. For clarity, only responses to the higher concentration of each odor are shown here; the response trajectories for the lower concentration were shorter but occupied the same manifolds as did the higher ones. (High- and low-concentration response trajectories are shown together in Supplementary Fig. 2.)

25 0 100 75 50 25

Figure 6 Ensemble responses are well classified, regardless of odor delivery pattern. Any given 50-ms template could effectively classify the odors. Blue line, result of an unsupervised classification algorithm applied separately to every 50-ms time bin for all the different odor delivery patterns (red bars on abscissa). Chance level, 12.5%. Classification success exceeded 80% throughout, independent of the pulse sequence. Red and green lines indicate classification performance obtained by applying a single 50-ms bin ‘template.’ Arrows indicate the templates (red, from the single pulse; green, from the second in the train of ten pulses). Classification success far exceeded chance for all delivery patterns and registered a peak at the time of each odor pulse.


[

NUMBER 11

[

NOVEMBER 2005

0 100 75 50 25 0 –2

0

2

4 Time (s)

6

8

10

1571

The existence of odor-specific, transiently coactive subgroups of projection neurons (as revealed by the distance, correlation and trajectory analyses) suggested that the responses should contain sufficient information about the odorants to allow the stimulus to be effectively classified—even given only brief snapshots of time. Indeed, we found that a simple, unsupervised classification algorithm, using only the eight most significant dimensions in the dataset (see Methods), could effectively distinguish the eight odor-concentration combinations from one another; using any given 50-ms time bin during the odor response, the algorithm was successful at classifying the odors (80–100% classification success), regardless of the pulse pattern (Fig. 6). Thus, the information available in any brief snapshot of the ensemble response is sufficient to classify the odors, independent of their temporal presentation characteristics.

a6 Time (s)

Correlation (P < 0.001) 1 0.6

a

KC 1

b

1,000 ms pulse

1,250 ms IPI

c

500 ms IPI

d

500 ms IPI

4 2

0.3 0

0

e Time (s)

100 60 P < 0.002 Chance

f

g

h

6

Template classification

30 12.5

Classification success (%)

4 2 0 0

2

4

6

0

2

6 0 Time (s)

4

2

4

6

0

2

4

6

Figure 7 Correlation and classification analyses show that responses to single odor pulses are comparable to responses during odor pulse trains. (a–h) Correlation (a–d) and classification (e–h) analyses, comparing responses to single pulses (ordinate) to responses to trains of three or ten pulses (abscissa) over time, show common response features regardless of the odor presentation pattern. The common features (subgroups of transiently coactive projection neurons) allow for high levels of successful response classification, particularly during the onsets and offsets of responses. Correlations shown are significant (P o 0.001). Classification probabilities were determined as described in Methods and in Supplementary Figure 5. Red bars on axes represent odor pulse times.

Recognizing time-varying features of the odor stimulus However, the evolving trajectories indicate gradual, continuous changes in the composition of odor-specific sets of transiently coactive projection neurons. This raises interesting questions: (i) if odor recognition requires the matching of a transient stimulus-elicited activity pattern with a stored template, do many different templates—each representing different points along the trajectory—need to be stored?; and (ii) does a template representing one odor-presentation pattern match the projectionneuron ensemble activity created by a different presentation pattern of the same odorant? We found that projection-neuron ensemble responses evoked by one presentation pattern were highly and significantly correlated with the

KC 2

KC 3

Odor 1

responses evoked by the same odorant presented in a different pattern (Fig. 7a–d). This suggests that cross-classification should be successful. We then used our projection-neuron activity vectors as templates, each template representing a 50-ms snapshot of the ensemble activity evoked by different odors presented in different patterns. We tried to classify responses evoked by one presentation pattern using responses evoked by the other patterns (Fig. 7e–h). The resulting template classification diagram for single, brief (100-ms) odor pulses reveals an evolving series of useful templates (Fig. 7e)—no single time bin was optimally effective during and after the stimulus. However, because the ensemble response patterns evolved gradually, any given 50-ms template could effectively classify much longer (150–300 ms-long) portions of the response. When we applied the template sequence formed from the single pulse to the responses evoked by a widely spaced (1,250-ms IPI) train of three pulses, a similar effect emerged (Fig. 7f): odors were successfully classified throughout the responses—although with the greatest success during the first part of each response. When the single-pulse template was applied to a three-pulse train with a shorter (500-ms) IPI, classification was again successful throughout the responses, but this time the best classification occurred at the beginning and the end of

Odor 2

0

2

4

6

8

0

2

4

6

2

2

1

1

0

0

2

2

1

1

0

0

2

4

6

0 0

2

4

6

8 Time (s)

1572

8

Time (s)

b Frequency (Hz)


ARTICLES

0

2

4

6

8

8

Figure 8 Kenyon cells fire sparsely throughout the odor responses. (a) Examples of Kenyon-cell responses to different odors and presentation patterns. Individual Kenyon cells revealed preferences for odors and response times. KC 1 and KC 2 were recorded simultaneously; odor 1: 1 octanol; odor 2: trans-2-hexen-1-ol. For KC 3, odor 1: 1 octanol; odor 2: 1 hexanol. KC 1 responded to most presentations of one odorant regardless of presentation pattern, but did not respond to a different odorant. KC 2 responded reliably but exclusively to the onset of an odor pulse or train for one odorant, but did not respond to other odorants. Finally, KC 3 responded to both odorants, although with differently timed responses for each, firing throughout odor 1, but only during the ‘after-response’ of odor 2. (b) Histograms of the firing frequencies, over time, of all 67 examined Kenyon cells, shown for each odor delivery pattern. Spiking occurred throughout the odor response but was most frequent at the beginning.

VOLUME 8

[

NUMBER 11

[



ARTICLES each pulse in the series (Fig. 7g). The arrival of each new pulse seemed to truncate the period of classification success, because the onset responses differed from the offset responses; however, the last pulse of the train elicited a lengthy offset response that matched the offset of the single pulse. The response to a ten-pulse train with a 500-ms IPI effectively matched the response to the single pulse, mainly during the train’s onset but also at other times during each of the ten pulses (Fig. 7h). We obtained similar results from all possible crossclassification matches (data not shown). Thus, a single stored 50-ms template was sufficient for recognizing the ensemble-wide activity patterns evoked by an odorant, regardless of the odorant’s presentation pattern (Fig. 6); by tracking the templatematches over time, we were able to obtain a description of the timevarying features of the stimulus (its odor, concentration and timing pattern). Multiple templates, if available, could serve to distinguish and recognize the onsets of pulses of a given odor. Responses of downstream neurons to odor pulses Our analysis of the ensemble response was based on the integration properties of the projection-neuron followers, the Kenyon cells. Thus, our analysis leads to predictions for Kenyon cell behavior. Because different subsets of projection neurons were coactivated by different odors and because these subsets converge onto Kenyon cells, we expected the responses in some Kenyon cells to be odor specific. Further, because the projection-neuron subsets that were coactive during the brief Kenyon cell integration window continuously changed during the odor response, we expected that Kenyon cells would respond to different odors at different times during this evolution. Also, because the ensemble response patterns elicited by trains of odor pulses tended to repeat with each pulse (that is, the trajectories re-circled portions of the same response space), we expected that the Kenyon cells would respond at particular times during the odor trains. Our recordings from Kenyon cells confirmed these predictions (Fig. 8), showing that individual Kenyon cells can respond with some specificity to different odors and to different temporal features of an odor presentation. Taken collectively, the data show that the Kenyon cells responded throughout the ensemble response but most prominently during response onset. DISCUSSION We examined how a temporally structured stimulus is encoded by a temporally structured distributed neural representation. The responses of individual first-order interneurons (projection neurons in the locust antennal lobe) indicated that the two types of temporal structures significantly interfered with each other. However, this potential confound was resolved by considering the responses of these neurons in the context of their ensemble activity; this activity was sufficiently informative to allow the accurate detection and classification of the invariant properties of the stimulus and its temporal structure. Second-order ‘decoder’ interneurons (Kenyon cells) responded as predicted on the basis of the information available in the first-order ensemble response. Spatiotemporal coding mechanisms—including those subserving vision1–4, audition5–7, touch8–10 and olfaction11–15—are common in neural systems. In most cases, neural responses are brief (although some neurons, such as those in inferior temporal cortex, show sustained firing following brief visual stimuli1) and seem to track the stimuli closely. The potential for confounds between temporally structured stimuli and their neural representations seems greatest in olfaction, where the response duration in principal neurons often outlasts the odor stimuli11,14,15. Olfactory receptor neurons in insects can follow rapid odor-pulse trains29–32 precisely; in particular, macroglomerular projection


[

NUMBER 11

[

NOVEMBER 2005

neurons, specialized for pheromone detection, have been shown to track the timing of pheromone plumes reliably33–35. Some insects, for example, seem to use this information when moving toward pheromone sources36–38. However, non-pheromonal afferent activity that is relatively stimulus locked is reformatted by the circuitry of the antennal lobe into complex spatiotemporal activity patterns distributed across the population of projection neurons11,15,22,26,39. These activity patterns, which vary with—and thus contain information about— odor identity and concentration (Figs. 1 and 2), are further parsed into a series of gradually evolving snapshots, roughly 50 ms long, by an oscillatory synchronization mechanism driven by the circuitry of the antennal lobe24. Projection neurons provide the only pathway for olfactory information from the antennal lobe to reach other neural targets, including the Kenyon cells. The locust antennal lobe contains about 830 projection neurons19; recent results suggest that more than 100 of these converge onto each of about 50,000 Kenyon cells (R.A.J. and G.L., Soc. Neurosci. Abstr. 412, 21, 2004). Odor identity information has been shown to be broadly distributed across the ensemble of projection cells11,24,26. (In the locust, the extent to which this convergence is genetically or otherwise specified is not known; by comparison, genetic labeling techniques have shown that Drosophila have some wellspecified connectivity40,41.) Our analysis adhered to these features of olfactory anatomy and physiology; we used techniques that permitted us to examine the odorelicited responses in an ensemble of more than 100 projection neurons in a series of brief time bins (Supplementary Fig. 1). Odorants evoked responses in most projection neurons (Fig. 2). Measures of the interresponse Euclidean distance (Fig. 3) and cross-correlation (Fig. 4) indicated (i) that the responses evolved gradually during and after the stimulus, (ii) that different odor concentrations evoked responses that differed from each other to some extent, and (iii) that different odorants evoked responses that differed to a greater extent. These responses were reliable over repeated trials and could be visualized as trajectories looping through a high-dimensional representation space (Fig. 5, Supplementary Fig. 2). These results confirm our earlier work26. Odor stimuli tend to repeat in nature because of iterative olfactory behaviors such as sniffing42 and antennal sweeping43 and also because of turbulence, which can deliver an odor to receptors as a series of transient encounters with odor filaments20. We found that, most of the time, the long, complex projection-neuron responses to each of the odor pulses within a train interfered with each other (Fig. 1). Yet we found no projection neurons that closely tracked trains of all odors, and thus no evidence for a specialized ‘channel’ for stimulus timing as exists in auditory systems. Nonetheless, when our set of 117 projection neurons was considered together, responses to multiple and rapid pulses were sufficiently correlated with one another (Fig. 7) to permit the odors to be accurately classified, regardless of the odor-pulse presentation pattern (Figs. 6 and 7). Indeed, trajectory representations of the ensemble response show largely overlapping, re-circling patterns corresponding to each odor pulse in a train (Fig. 5). Any given 50-ms time bin contained enough information to successfully serve as a template to classify odors (Fig. 6). Further, we found that single 50-ms templates taken from one presentation pattern could successfully classify responses to other presentation patterns (Fig. 6); thus, the olfactory system need not memorize every permutation of odor and presentation pattern in order to recognize a particular odor in the future. Additionally, by describing brief portions of odor responses, such as the response onset (as in Fig. 6), templates can characterize the temporal properties of the odor stimulus.

1573


ARTICLES When the IPI was brief, each new ensemble response truncated the previous one, as is evident in the correlation analyses (Figs. 4 and 7) and trajectory representations (Fig. 5). This truncation left the early portion of each response largely intact. Although useful templates could be drawn from any time in the response, the onset of the response to each pulse in the train provided the most reliable across-pulse and across-pattern templates (Fig. 6). Notably, the most effective templates occurred about 300 ms after the ensemble response onset (Supplementary Fig. 3)—the earliest time at which maximal odoridentifying information is available from projection-neuron responses (see Fig. 6d in ref. 26). Thus, the circuitry of the antennal lobe seems to be largely ‘reset’ by each new stimulus pulse. As a result, regardless of the presentation pattern, each odor-specific ensemble response begins similarly and then follows a similar trajectory. We observed up to 68% variability in the responses to each combination of projection neuron and odor; these responses were calculated cumulatively over either three or ten pulses. However, the response variability between any one pulse and any other was less than 68%. For example, for the series of three pulses with a 500-ms IPI, the percentages of neurons that changed their responses significantly (P o 0.05; see Methods) between the first and second, first and third, and second and third pulses were 26.67%, 30.1% and 22.3%, respectively. In addition, of the projection neurons that changed their responses, only 20.62% of the neurons changed their responses across all pulse pairs. Thus, considered only pair-wise (that is, between any given pulse and another), most individual projection neurons had similar responses, thus providing a foundation for extracting invariance. It is critical to note, however, that the subset of projection neurons that responded consistently for one pair of pulses was not the same as the subset that responded consistently for a different pair. For example, only 54.8% of the projection neurons that significantly changed their responses between the first and second pulses also changed their responses between the first and third pulses. Our analysis shows that the problem of response classification can be solved by the convergence of many projection neurons onto Kenyon cells. The Kenyon cells each examine a large ensemble of projection neurons, which—by virtue of its large size—will contain enough projection neurons that respond consistently to a given stimulus. The remaining highly variable responses, which could dominate a smaller ensemble (Supplementary Fig. 4), would be insufficient to prevent successful classification. How large should a projection-neuron ensemble be if responses are to be classified successfully? If we assume that the connectivity between projection neurons and Kenyon cells is imprecisely specified, Kenyon cells cannot know which subset of the projection neurons will provide faithful, stimulus-tracking responses for any given odor and pulse pattern. With our set of 117 projection neurons, we achieved classification success far better than chance (Fig. 6). We also found, not surprisingly, that response classification deteriorated when we used successively smaller subsets of projection neurons for our analysis (Supplementary Fig. 4; see also Fig. 6e in ref. 26). Classification with smaller projection-neuron ensembles might be possible with more precisely specified mapping between projection neurons and Kenyon cells, experience-dependent plasticity or both. We analyzed brief (50-ms) time bins of activity because Kenyon cells seem to sample the projection-neuron ensemble on this time scale24. It is not known whether neurons that succeed Kenyon cells in the olfactory pathway are able to integrate olfactory information over longer durations (as was predicted in a previous study44). Not surprisingly, we found that longer templates, formed by concatenating up to 20 of the 50-ms templates, marginally improved odor classification, but

once the template duration exceeded the IPI of the odor train, classification deteriorated (data not shown). Thus, if the IPI is unpredictable or variable, as is the case in most natural odor plumes, brief samples of activity may be of the most general utility. A system designed to use such samples would function well when it encounters rapid trains of brief odor pulses (such as, for instance, when an odor source is distant). When odor samples are encountered at longer and more widely spaced intervals, the same system could continue to extract and memorize the additional, partially redundant information available in the longer responses—consistent with behavioral results showing that animals require more time to solve difficult olfactory tasks45. In Kenyon cells, the responses to odors were consistent with the information detectable by the convergence of multiple projection neurons: as previously described26, Kenyon cells fired in response to particular odors and concentrations, and at particular times. Over the trains of odor pulses, individual Kenyon cells responded reliably at particular times to the pulses (Fig. 8a, KC 1,2). If a Kenyon cell responded to more than one test odorant, the response time could be different for different odorants (Fig. 8a, KC 3). This timing preference probably results from the transient coactivity of subgroups of projection neurons that converge onto a given Kenyon cell; through such convergence, Kenyon cells correspond to particular points along the response trajectory of the projection-neuron ensemble. As a group, Kenyon cells responded during and briefly following odor presentations, with small activity peaks for each pulse; notably, however, most responses occurred at the onset of each odor train (Fig. 8b). When the inter-pulse interval was brief (that is, 500 ms), Kenyon cells responded most often to the first pulse but continued to respond throughout the train. Given our relatively sparse sampling of the large Kenyon-cell population, it remains an open and interesting question why the first pulse elicited the strongest responses. There is no behavioral evidence to suggest that locusts use information about the temporal presentation characteristics of odors. Our results, however, suggest that such timing information is available within the olfactory response. We speculate that animals would benefit from detecting and recognizing characteristic timing features of an odor plume, such as its onset, continued presence and offset; we predict that one could, for example, train animals to respond behaviorally to specific, rewarded temporal features in a complex odor sequence. Multiple factors seem to underlie the changes in the response patterns of projection neurons and Kenyon cells over the course of a rapid train of odor pulses. In projection neurons, these factors include the cumulative superimposition of long-lasting, odor-driven excitatory and inhibitory inputs from afferent, projection and local neurons within the antennal lobe. These factors constrain the firing patterns of projection neurons and thus constitute the mechanism underlying the partial circuit reset triggered by each odor pulse. In addition, there are at least two forms of plasticity: rapid but short-lived adaptation in odor receptors (detected as decreasing EAG amplitude, Fig. 1); and gradual but long-lasting ‘fast learning’ within the lobe46,47. Kenyon cells are additionally influenced by feed-forward inhibition from the lateral horn24. We are presently exploring these mechanisms and their interactions in detail (S.L.B. and M.S., unpublished data).

1574

VOLUME 8

METHODS Animals. Experiments were performed on 116 intact adult locusts (Schistocerca americana) of both sexes from our crowded colony. Animals were immobilized and stabilized with wax; one antenna was secured intact and the other was removed and used to record an electroantennogram (see below). The brain was exposed, desheathed and superfused with locust saline as previously described11.

[

NUMBER 11

[



ARTICLES Odor stimulation. Dried and activated carbon-filtered air (0.75 liter min–1) flowed continuously across the antenna through a Teflon tube (6.35 mm inner diameter) placed perpendicular to and within 4 mm of the intact antenna. A large vacuum funnel was placed 10 cm behind the antenna so as to quickly remove odorants. Twenty milliliters of each liquid odorant—either neat or diluted in mineral oil (J.T. Baker)—were placed in 60-ml glass bottles; the odors, drawn from the headspace above these odorants, were puffed by a pneumatic picopump (WPI) into the continuously flowing air stream, thus further diluting the odorant. The timing of odor release was controlled by a Master-8 stimulus generator (A.M.P.I.) or by a custom computer program. The odorants used in the intracellular recordings were 1-hexanol, 1-octanol, hexanal (Fluka), geraniol (Sigma), trans-2-hexen-1-ol and 2-heptanone (Aldrich) as well as extracts of strawberry, cinnamon, peach, lime (Balducci’s), wintergreen and Sambuca (Wagner’s), used neat or dilutions of 10:1 and 100:1. For the extracellular recordings, 1-hexanol, geraniol, 1-octanol and hexanal were used neat and at 100:1 dilutions (denoted throughout the text as 1 and 0.01). Electrophysiology. Electroantennogram recordings (EAGs) were made by inserting chlorided silver wire electrodes (0.127 mm diameter, WPI) into the cut ends of an isolated antenna. Wires were secured with small drops of wax and fed into a DC amplifier (Brownlee Precision). Intracellular recordings were made from neurons in the antennal lobe using sharp glass micropipettes (outer diameter 1.0 mm; Warner Instruments); these had been pulled with a Sutter P97 horizontal puller (Sutter Instruments) and filled with 0.5 M potassium acetate and 5% neurobiotin (Vector Laboratories) to yield resistances of 80–230 MO. The data were digitally acquired at a sampling rate of 5 kHz (LabView software; PCI-6602 DAQ and PCI-MIO-16E-4 hardware, National Instruments) and stored on a PC hard drive; they were then analyzed off-line using MATLAB (MathWorks). Multiunit recordings from the projection neurons were made using 16-channel, 4 4 silicon probes (NeuroNexus Technologies) and from the Kenyon cells using custom-made twisted wire tetrodes24 with a 16-channel DC amplifier (Biology Electronics Shop, Caltech). The data were digitally sampled at 15 kHz. Multiunit projection-neuron and Kenyon-cell spike sorting was achieved offline by a four-wire, whole-waveform algorithm (Spike-O-Matic48) implemented in Igor (Wavemetrics). Spike sorting was conservative: we analyzed only those clusters that were unambiguously defined and clearly separated from one another throughout the experiment. The criteria we used to select these clusters included the following: (i) nearest cluster projections had to lie at least 5 standard deviations (s.d.) apart, (ii) no more than 2% of the ISIs were under 20 ms and (iii) the waveform s.d. could not exceed 5%. These criteria are described elsewhere48. Analysis. All analyses were performed using custom programs in MATLAB. To determine whether the odor-elicited spike numbers changed significantly with pulse position, we used the following algorithm: (i) we evaluated whether a particular combination of odor and projection neuron elicited spikes—defined as an increase over the baseline firing rate by 6.5 s.d., in any post-stimulus 50-ms bin (this procedure yielded response detection that closely matched an observer’s judgments); (ii) we then performed a two-way analysis of variance (ANOVA) on the spike counts obtained, over ten repeated trials, during the 500 ms after each pulse. To analyze the response of the projection-neuron ensemble (117 projection neurons) over time, we constructed a series of 117-dimensional vectors, each representing the ensemble’s firing during a single 50- or 100-ms time bin; each vector element consisted of the number of spikes in a single projection neuron during the time bin. We refer to these vectors as snapshots, as they represent the ensemble response at a moment in time. Beginning at 2 s before the first odor pulse, we collected data for a total of 15 s; then, to measure the Euclidean distance between ensemble response vectors, these data were binned into 50-ms time bins. The Euclidean distance for one group (that is, all trials, odors or concentrations) is given by the average across all pairs within the group. For two groups, the distance is given by the average across all pairs from the two groups. The cross-correlations between the response vectors were computed for each combination of time bins in the ensemble response as follows: if x is the 117-dimensional vector representing the response to a given odor, then for the


[

NUMBER 11

[

NOVEMBER 2005

ith trial, x(i), the cross-correlation between the responses at times k and l (that is, xk(i) and xl(i)) is calculated as 10 X i¼1

! xTk ðiÞxl ðiÞ =

10 X i¼1

! xTk ðiÞxk ðiÞ

10 X

!!1=2 xTl ðiÞxl ðiÞ

i¼1

We then averaged across odors to get the cross-correlation at the (k,l)th point in the plot. Estimates were made using ten trials, and only correlations with significance of P o 0.001 (t-test) are shown. For nonlinear dimensionality reduction with locally linear embedding (LLE)28, we used code from S. Roweis (http://www.cs.toronto.edu/~roweis/ lle). The LLE method first finds a linear transformation, invariant to rotation and scaling, that represents each point in terms of each of its neighbors. Then a representation with reduced dimensionality is calculated by estimating a new set of points that satisfies the above transformation with respect to all the points. When applied to our data, principal component analysis (PCA) yielded results that were qualitatively similar to those from LLE, but—because our high-dimensional data were not well described by linear functions—LLE characterized and separated the response trajectories more effectively. The input to LLE consisted of the 117-dimensional snapshot vectors, each 100-ms wide and averaged over three trials. Qualitatively similar results were obtained using 50-ms time bins. LLE was calculated for the four different pulse sequences jointly and later plotted on separate, matching axes. LLE gave qualitatively similar results for a wide range of neighborhood values (between 10 and 20), indicating the genuine presence of low-dimensional manifolds that are well characterized by this analysis. For the classification analysis, we used the 50-ms time bins. The high dimensionality of our dataset could permit spurious over-classification of odors; to avoid this, we first reduced the dimensionality of our dataset using PCA. For classification based on templates, we applied the k-means function in MATLAB to the first eight dimensions in our dataset. Our choice of eight dimensions (which together contributed 32% of the variance) was based on standard practice: we identified the elbow in a scree plot of our eigenvalues49. Neighbors were decided on the basis of the Euclidean distance. For templatebased classification, the templates were the centroids of ten repeated trials. To estimate the confidence level of our template-based classification, a histogram of the percentage of template-based classification was calculated from ensemble activity at times before the odors were presented; the probability of obtaining 430% classification success in the absence of a stimulus was o0.002 (Supplementary Fig. 5). We pooled projection neurons sampled from 14 experiments to approximate a single, large (technically unfeasible) recording from a single animal. This raises the possibility of including in our set multiple examples of the same projection neuron from different animals. However, an inspection of the odor responses of our 117 projection neurons suggested that no duplicates existed. Further, we performed a 10,000-iteration Monte Carlo simulation of our sampling process with the same numbers of projection neurons (between 3 and 17) and animals (14) as in our data. Results showed a non-zero, but very small, probability for the duplication of any projection neuron (see Supplementary Fig. 6; for a discussion of this issue, see ref. 26). Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We are grateful to members of the Stopfer lab for helpful discussions. This work was funded by an intramural grant from the National Institutes of Health, the National Institute of Child Health and Human Development. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. Fuster, J.M. & Jervey, J.P. Neuronal firing in the inferotemporal cortex of the monkey in a visual memory task. J. Neurosci. 2, 361–375 (1982). 2. Richmond, B.J., Optican, L.M., Podell, M. & Spitzer, H. Temporal encoding of twodimensional patterns by single units in primate inferior temporal cortex. I. Response characteristics. J. Neurophysiol. 57, 132–146 (1987).

1575


ARTICLES 3. McClurkin, J.W., Optican, L.M., Richmond, B.J. & Gawne, T.J. Concurrent processing and complexity of temporally encoded neuronal messages in visual perception. Science 253, 675–677 (1991). 4. Meister, M. Multineuronal codes in retinal signaling. Proc. Natl. Acad. Sci. USA 93, 609–614 (1996). 5. Nagarajan, S.S. et al. Representation of spectral and temporal envelope of twitter vocalizations in common marmoset primary auditory cortex. J. Neurophysiol. 87, 1723– 1737 (2002). 6. Gehr, D.D., Komiya, H. & Eggermont, J.J. Neuronal responses in cat primary auditory cortex to natural and altered species-specific calls. Hear. Res. 150, 27–42 (2000). 7. Machens, C.K. et al. Representation of acoustic communication signals by insect auditory receptor neurons. J. Neurosci. 21, 3215–3227 (2001). 8. Jones, L.M., Depireux, D.A., Simons, D.J. & Keller, A. Robust temporal coding in the trigeminal system. Science 304, 1986–1989 (2004). 9. Arabzadeh, E., Zorzin, E. & Diamond, M.E. Neuronal encoding of texture in the whisker sensory pathway. PLoS Biol. 3, E17 (2005). 10. Jones, L.M., Lee, S., Trageser, J.C., Simons, D.J. & Keller, A. Precise temporal responses in whisker trigeminal neurons. J. Neurophysiol. 92, 665–668 (2004). 11. Laurent, G. & Davidowitz, H. Encoding of olfactory information with oscillating neural assemblies. Science 265, 1872–1875 (1994). 12. Hamilton, K.A. & Kauer, J.S. Responses of mitral/tufted cells to orthodromic and antidromic electrical stimulation in the olfactory bulb of the tiger salamander. J. Neurophysiol. 59, 1736–1755 (1988). 13. Meredith, M. Patterned response to odor in mammalian olfactory bulb: the influence of intensity. J. Neurophysiol. 56, 572–597 (1986). 14. Laurent, G., Wehr, M. & Davidowitz, H. Temporal representations of odors in an olfactory network. J. Neurosci. 16, 3837–3847 (1996). 15. Wehr, M. & Laurent, G. Odour encoding by temporal sequences of firing in oscillating neural assemblies. Nature 384, 162–166 (1996). 16. Lestienne, R. Spike timing, synchronization and information processing on the sensory side of the central nervous system. Prog. Neurobiol. 65, 545–591 (2001). 17. Theunissen, F. & Miller, J.P. Temporal encoding in nervous systems: a rigorous definition. J. Comput. Neurosci. 2, 149–162 (1995). 18. Laurent, G., MacLeod, K., Stopfer, M. & Wehr, M. Dynamic representation of odours by oscillating neural assemblies. Entomol. Exp. Appl. 91, 7–18 (1999). 19. Laurent, G. A systems perspective on early olfactory coding. Science 286, 723–728 (1999). 20. Murlis, J. & Jones, C.D. Fine-scale structure of odor plumes in relation to insect orientation to distant pheromone and other attractant sources. Physiol. Entomol. 6, 71–86 (1981). 21. Hildebrand, J.G. & Shepherd, G.M. Mechanisms of olfactory discrimination: Converging evidence for common principles across phyla. Annu. Rev. Neurosci. 20, 595–631 (1997). 22. Wehr, M. & Laurent, G. Relationship between afferent and central temporal patterns in the locust olfactory system. J. Neurosci. 19, 381–390 (1999). 23. Bazhenov, M. et al. Model of cellular and network mechanisms for odor-evoked temporal patterning in the locust antennal lobe. Neuron 30, 569–581 (2001). 24. Perez-Orive, J. et al. Oscillations and sparsening of odor representations in the mushroom body. Science 297, 359–365 (2002). 25. Laurent, G. & Naraghi, M. Odorant-induced oscillations in the mushroom bodies of the locust. J. Neurosci. 14, 2993–3004 (1994). 26. Stopfer, M., Jayaraman, V. & Laurent, G. Intensity versus identity coding in an olfactory system. Neuron 39, 991–1004 (2003).

27. Murlis, J., Elkinton, J.S. & Carde, R.T. Odor plumes and how insects use them. Annu. Rev. Entomol. 37, 505–532 (1992). 28. Roweis, S.T. & Saul, L.K. Nonlinear dimensionality reduction by locally linear embedding. Science 290, 2323–2326 (2000). 29. Bau, J., Justus, K.A. & Carde, R.T. Antennal resolution of pulsed pheromone plumes in three moth species. J. Insect Physiol. 48, 433–442 (2002). 30. Barrozo, R.B. & Kaissling, K.E. Repetitive stimulation of olfactory receptor cells in female silkmoths Bombyx mori L. J. Insect Physiol. 48, 825–834 (2002). 31. Marion-Poll, F. & Tobin, T.R. Temporal coding of pheromone pulses and trains in Manduca sexta. J. Comp. Physiol. A 171, 505–512 (1992). 32. Lemon, W. & Getz, W. Temporal resolution of general odor pulses by olfactory sensory neurons in American cockroaches. J. Exp. Biol. 200, 1809–1819 (1997). 33. Lei, H. & Hansson, B.S. Central processing of pulsed pheromone signals by antennal lobel neurons in the male moth Agrotis segetum. J. Neurophysiol. 81, 1113–1122 (1999). 34. Heinbockel, T., Christensen, T.A. & Hildebrand, J.G. Temporal tuning of odor responses in pheromone-responsive projection neurons in the brain of the sphinx moth Manduca sexta. J. Comp. Neurol. 409, 1–12 (1999). 35. Vickers, N.J., Christensen, T.A., Baker, T.C. & Hildebrand, J.G. Odour-plume dynamics influence the brain’s olfactory code. Nature 410, 466–470 (2001). 36. Vickers, N.J., Christensen, T.A. & Hildebrand, J.G. Combinatorial odor discrimination in the brain: attractive and antagonist odor blends are represented in distinct combinations of uniquely identifiable glomeruli. J. Comp. Neurol. 400, 35–56 (1998). 37. Willis, M.A. & Avondet, J.L. Odor-modulated orientation in walking male cockroaches Periplaneta americana, and the effects of odor plumes of different structure. J. Exp. Biol. 208, 721–735 (2005). 38. Justus, K.A. & Carde, R.T. Flight behaviour of males of two moths, Cadra cautella and Pectinophora gossypiella, in homogeneous clouds of pheromone. Physiol. Entomol. 27, 67–75 (2002). 39. Lemon, W.C. & Getz, W.M. Rate code input produces temporal code output from cockroach antennal lobes. Biosystems 58, 151–158 (2000). 40. Jefferis, G.S. et al. Developmental origin of wiring specificity in the olfactory system of Drosophila. Development 131, 117–130 (2004). 41. Ramaekers, A. et al. Glomerular maps without cellular redundancy at successive levels of the Drosophila larval olfactory circuit. Curr. Biol. 15, 982–992 (2005). 42. Fontanini, A., Spano, P. & Bower, J.M. Ketamine-xylazine-induced slow (o1.5 Hz) oscillations in the rat piriform (olfactory) cortex are functionally correlated with respiration. J. Neurosci. 23, 7993–8001 (2003). 43. Okada, J. & Toh, Y. Spatio-temporal patterns of antennal movements in the searching cockroach. J. Exp. Biol. 207, 3693–3706 (2004). 44. Nowotny, T., Rabinovich, M.I., Huerta, R. & Abarbanel, H.D. Decoding temporal information through slow lateral excitation in the olfactory system of insects. J. Comput. Neurosci. 15, 271–281 (2003). 45. Abraham, N.M. et al. Maintaining accuracy at the expense of speed: stimulus similarity defines odor discrimination time in mice. Neuron 44, 865–876 (2004). 46. Bazhenov, M., Stopfer, M., Sejnowski, T.J. & Laurent, G. Fast odor learning improves reliability of odor responses in the locust antennal lobe. Neuron 46, 483–492 (2005). 47. Stopfer, M. & Laurent, G. Short-term memory in olfactory network dynamics. Nature 402, 664–668 (1999). 48. Pouzat, C., Mazor, O. & Laurent, G. Using noise signature to optimize spike-sorting and to assess neuronal classification quality. J. Neurosci. Methods 122, 43–57 (2002). 49. Joliffe, I.T. Principal Component Analysis (Springer-Verlag, New York, 1986).

1576

VOLUME 8

[

NUMBER 11

[



ARTICLES

Drosophila melanogaster homolog of Down syndrome critical region 1 is critical for mitochondrial function Karen T Chang & Kyung-Tai Min Mitochondrial dysfunction has emerged as a common theme that underlies numerous neurological disorders, including Down syndrome. Down syndrome cultures and tissues show mitochondrial damage such as impaired mitochondrial enzyme activities, defective mitochondrial DNA repairs and accumulation of toxic free radicals, but the cause of mitochondrial dysfunction remains elusive. Here we demonstrate that the Drosophila melanogaster homolog of human Down syndrome critical region gene 1 (DSCR1), nebula (also known as sarah, sra), has a crucial role in the maintenance of mitochondrial function and integrity. We report that nebula protein is located in the mitochondria. An alteration in the abundance of nebula affects mitochondrial enzyme activities, mitochondrial DNA content, and the number and size of mitochondria. Furthermore, nebula interacts with the ADP/ATP translocator and influences its activity. These results identify nebula/DSCR1 as a regulator of mitochondrial function and integrity and further suggest that an increased level of DSCR1 may contribute to the mitochondrial dysfunction seen in Down syndrome.

The maintenance of mitochondrial function and integrity is crucial for normal cell physiology, particularly for cells with high energy demand. Neurons, for example, rely on the ATP generated by mitochondria through oxidative phosphorylation for functions such as synaptic transmission and channel activity1. In addition to oxidative phosphorylation, mitochondria mediate cell death and survival by integrating cellular stress signals, and they are the main source of reactive oxygen species (ROS), a cause of oxidative stress2. Consequently, mutations in mitochondrial DNA (mtDNA) and nuclear DNA that affect mitochondrial respiration, morphology and mtDNA content have been associated with a wide range of neuromuscular and neurological disorders including Down syndrome3–7. Altered mitochondrial enzyme activities, elevated ROS and defective mtDNA repair after oxidative damage have been reported in Down syndrome cultures and tissues5,8–10, but the underlying cause of defective mitochondrial respiration is still unclear. The Down syndrome critical region 1 gene (DSCR1), located in the region 21q22.1–q22.2 of chromosome 21, is overexpressed in brain from human fetuses with Down syndrome11,12. DSCR1 is highly conserved across species and belongs to a family of proteins called calcipressins, which bind and inhibit calcineurin12–16. Using D. melanogaster as a model system, we previously demonstrated that the D. melanogaster homolog of DSCR1, nebula, is required for learning and long-term memory through its regulation of the calcineurin-mediated signaling pathway. We found that flies with nebula loss-of-function and overexpression show defective learning, suggesting that a delicate balance in the abundance of nebula is crucial for learning12. In addition, studies done in cell culture suggest that

DSCR1 may have functions involving the regulation of oxidative stress response17–19. Given the importance of mitochondria in oxidative stress response and neurological disorders, we hypothesized that DSCR1 may mediate mitochondrial-dependent processes. Using the previously established D. melanogaster model system, we demonstrate here that maintaining the correct abundance of nebula is crucial for normal mitochondrial respiration, generation of ROS, maintenance of mtDNA content, and the number and size of mitochondria. Furthermore, nebula is located in the mitochondria and interacts with a mitochondrial protein—the ADP/ATP translocator (ANT)— and influences its activity. Consistent with our findings, brain tissues from fetuses with trisomy 21 showing DSCR1 overexpression also show reduced mtDNA content. These results indicate that nebula and DSCR1 are important in the maintenance of mitochondrial function and integrity. RESULTS nebula mutants have altered mitochondrial metabolism To determine whether nebula mediates mitochondrial-dependent processes, we examined several aspects of mitochondrial metabolism in the nebula mutants. First, we determined the abundance of ATP in the heads of the nebula hypomorphic mutant (nla1) and transgenic flies overexpressing nebula in neurons driven by the pan-neuronal ElavGal4 (Elav) driver (Elav/+;nlat1/+)12. Both fly lines showed a substantially lower abundance of ATP as compared with control (CS) fly heads (Fig. 1a), indicating that alteration in the amount of nebula affects the overall ATP level.

Neurogenetics Branch, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, Maryland 20892, USA. Correspondence should be addressed to K.-T.M. ([email protected]). Received 15 August; accepted 14 September; published online 16 October 2005; doi:10.1038/nn1564


[

NUMBER 11

[

NOVEMBER 2005

1577

ARTICLES

a

Relative ROS level

ATP (µmol/µg protein)

Control

30

*

20

nla1

* 10 0

*

1.5 1.0 0.5 0

Control

*

nla1 Elav/+;nlat1/+

Control

nla1 Elav/+;nlat1/+

Head

Body

mito tRNA Co I Co III Cyt B

1.2 1.0 0.8

*

0.6

e

COX activity

f

SDH activity

*

0.4

Control

0.2 0

Control

nla1 Elav/+;nlat1/+

nla1

1.2

30 Change in activity relative to control (%)

d

Relative mtDNA content

Elav/+;nla t1/+



c

b 40

10 0

0.8 0.6 0.4

COX SDH

–10 –20 –30

1.0

*

20

*

* nla1/+ nla1 Elav/+;nlat1/+

Elav/+;nlat1/+

0.2 0

Control

nla1 Elav/+;nlat1/+

Figure 1 The nebula mutant flies show altered mitochondrial function. (a) ATP abundance in the head extracts of control, homozygous nebula loss-of-function mutant (nla1) and transgenic flies overexpressing nebula in neurons (Elav/+;nlat1/+). *P o 0.05. n ¼ 9 experiments done in triplicate. (b) ROS abundance in the brains of different fly lines detected by dihydroethidium. (c) Quantification of the relative ROS level normalized to the control brain. *P o 0.05. n ¼ 3 experiments each done using three brains from each group mounted in parallel. (d) Quantitative RT-PCR results were obtained from DNA isolated from either the head or the body of each fly line. Values were normalized to CS flies (control) and Gapdh2 was used as internal control. mito tRNA, primers specific for mitochondrial tRNAs; Co I and Co III, primers for cytochrome c oxidase subunits I and III, respectively; Cyt B, primers for cytochrome b. *P o 0.05. n ¼ 3 experiments done in triplicate. (e) Representative COX and SDH activity stainings. The dark red color by the retina is eye pigment that sometimes remains after the washes, but does not affect the outcome of the staining. (f) Quantification of changes in COX and SDH activities relative to the control. *P o 0.001. n ¼ 4 independent experiments done with multiple fly heads from each group. Values represent mean ± s.e.m.

Second, as perturbations in ATP level and oxidative phosphorylation have been shown to increase the production of ROS2, we analyzed the abundance of ROS in the nebula mutants. Dihydroethidium staining of dissected fly brains revealed that reduced expression and overexpression of nebula resulted in elevated ROS production (Fig. 1b,c), consistent with the observed depletion in ATP level. Third, because elevated ROS often affects mtDNA content3,7, we determined whether the nebula mutants have altered mtDNA content using quantitative real-time polymerase chain reaction (RT-PCR). The nla1 mutant and flies overexpressing nebula showed a significant reduction in the amount of mtDNA specifically in the head, but not in the body (Fig. 1d). These results are consistent with the expression pattern of nebula (Supplementary Fig. 1 online) and further imply that nebula is important for mtDNA maintenance. Fourth, to understand whether alteration in the abundance of nebula influences the activities of enzymes involved in oxidative phosphorylation, we determined the activity of cytochrome c oxidase (COX, complex IV) and succinate dehydrogenase (SDH, complex II). We used histochemical methods to specifically monitor brain regions with altered mitochondrial enzyme activities. The COX activity of nla1 flies was substantially lower than that seen in the control fly throughout the brain, and the overall COX activity in the flies overexpressing

nebula was also lower (Fig. 1e). Quantification of the overall staining intensity throughout the brain revealed that COX activity seen in the nebula loss-of-function flies was 20.1% ± 1.1% (P o 0.001) lower than that of the control, whereas nebula overexpression caused a 14.3% ± 1.7% (P o 0.001) decrease in COX activity (Fig. 1e,f). The nla1 flies showed considerably greater SDH activity in the brain than controls. The most pronounced increases were in the lamina, where the photoreceptor axons enter to synapse with the lamina neurons (arrow), and, to a lesser extent, in the central neuropil (Fig. 1e). Flies overexpressing nebula, however, showed SDH activity comparable with that in the control fly (Fig. 1e). Quantification of the overall intensity confirmed these findings, with nla1 flies showing a 22.9% ± 1.5% (P o 0.001) increase in SDH activity, and nebula-overexpressing flies showing no significant change (6.8% ± 4.0%; P 4 0.1). The nla1 heterozygotes did not show apparent alteration in either SDH (9.4% ± 6.5%; P 4 0.2) or COX activity (4.9% ± 1.8%; P 4 0.05) compared with controls, consistent with the near-normal abundance of nebula transcripts12 and protein (Supplementary Fig. 1). Complex II is encoded entirely by the nuclear genome20, and thus its activity may correlate with the number of mitochondria rather than mitochondrial function. The increase in SDH, and decrease in COX, activities may hence indicate that nla1 flies have more dysfunctional mitochondria in the brain,

1578

VOLUME 8

[

NUMBER 11

[


3

**

2

*

1

0.10

a1 / nl nla 6 a t2 2 t5 C /+;n (re /nla la 1 sc t2 El ue ;nla 1 av ) /+ ;n la t1 /+

*

*

*

0

a1 nl

Control (old)

*

0.05

Ac

e

**

0.15

nl

nl

Control

on

b

Act5C/nlat1;nla1 (rescue)

tro l a 1/

+

0

nla1

0.20

a1 /n Ac nla t2 la 62 t5 C /+;n (re /nla la 1 sc t2 El ue ;nla 1 av ) /+ ;n la t1 /+

4

+

5

0.25

nl

R7

*

0.30

tro l a 1/

R8

L4

*

6

C


L3 R6

*

on

L2

R5

2

L1 R2

7 per µm PR axon

R4

R3

R1

Number of mitochondria

Lamina cartridge

0.35

a1

8

0.40

nl

d

nl

c

C

a

Cross-sectional area of mitochondria (µm2)

ARTICLES

nla1 (young)

nla1 (old)

Figure 2 The nebula mutants show a profound increase in the number of smaller mitochondria. (a) Lamina region examined and a schematic of the lamina cartridge depicting the lamina (L1–L4) and photoreceptor axons (R1–R8). (b) Cross-section of the lamina cartridge (top) and of a single photoreceptor axon (bottom). (c) The number of mitochondria normalized to the area of photoreceptor (PR) axon. Only electron micrographs showing the cross-section of lamina cartridge were analyzed. (d) Cross-sectional area of mitochondria detected in the photoreceptor axons. For c and d, values represent mean ± s.e.m. *P o 0.05 compared with control. **P o 0.05 compared with either nla1, nla1/nla62 or nlat2/+;nla1 fly lines. (e) Representative electron micrographs of the central brain neuropil of a 45-d-old CS fly (control, old), 1-d-old nebula loss-of-function fly (nla1, young) and 45-d-old nla1 fly (nla1, old). More mitochondria (arrows) are sometimes observed in the neuropil of young nla1 flies than the control fly, and the axons of old nla1 flies are often seen packed with mitochondria. For b and e scale bar corresponds to 1 mm.

especially in the lamina. These results suggest that a fine balance in the abundance of nebula is important for normal mitochondrial function and integrity. Abnormal number and size of mitochondria in nebula To further examine whether the increase in SDH activity seen in nla1 flies represents a change in the number of mitochondria, we examined the fly brain using electron microscopy. We specifically examined the lamina region, as this region showed the most markedly altered SDH activity. Figure 2a depicts the lamina region, and a schematic of the cross-section through the lamina cartridge. The photoreceptor axons of nla1 flies showed an increase in the number of mitochondria that are substantially smaller in size, filling most of the space in the photoreceptor axons (Fig. 2b and Supplementary Fig. 2 online). This increase in the number of mitochondria is likely not due to abnormal distribution of mitochondria within the photoreceptor neuron or excessive transport from the soma, as mitochondria were clearly seen in the soma of photoreceptor neurons in nla1 flies (data not shown). In addition, the morphology of mitochondria seemed normal, with visible cristae (Fig. 2b, bottom). We also noted that this increase in the number of smaller mitochondria was observed in young nla1 flies and persisted in old flies. To better quantify changes in the number of mitochondria, we determined the number of mitochondria per area of photoreceptor axon. To minimize error in our calculation, only electron micrographs with all R1–R6 axons visible in the cross-section through the lamina cartridge were used in the analysis. Quantification of electron micrographs revealed that nla1 flies have 12 times more


[

NUMBER 11

[

NOVEMBER 2005

mitochondria per area of photoreceptor axon, but an 82% reduction in the cross-sectional area of mitochondria compared with control flies (Fig. 2c,d). This increase in the number of mitochondria is consistent with the increased SDH staining. We also analyzed another allele of the nebula mutant, nla1/nla62. The nla62 homozygotes are not viable, whereas nla1/nla62 flies show a 68.3 ± 1.2% decrease (P o 0.001) in the abundance of nebula transcript compared with wild-type flies, and a significant decrease in the abundance of nebula protein (Supplementary Fig. 1). Consistently, nla1/nla62 also showed similar changes in mitochondrial number and cross-sectional area. The nla1 heterozygotes, by contrast, have a normal number of mitochondria (Fig. 2c,d). Flies overexpressing nebula in neurons (Elav/+;nlat1/+) also showed a moderate increase in the number of mitochondria and a decrease in mitochondria cross-sectional area (Fig. 2c,d), although there was no significant change in SDH activity. This difference is likely observed because electron microscopy has a higher resolution in determining the number of mitochondria. In addition, we analyzed the central brain neuropil region of the nla1 flies. An increase in mitochondrial number was indeed detected in the neuropil of young nla1 flies, but the phenotype was more severe and more frequently seen in older flies (Fig. 2e). In the old nla1 flies, sometimes the axons were seen packed with mitochondria, forming clumps of mitochondria (Fig. 2e). To verify that changes in the number of mitochondria are indeed caused by mutation in nebula, we rescued the phenotype by overexpressing nebula in the nebula loss-of-function background using the previously generated transgenic UAS-nebula fly line (nlat2) and the Act5C-Gal4 (Act5C) driver (Act5C/nlat2;nla1)12. Ubiquitous expression

1579

of nebula under the control of the Act5C driver in the nla1 background (Act5C/nlat2;nla1) caused a significant decrease in the number of mitochondria and an increase in mitochondrial size compared with nla1 flies and flies carrying the nlat2 transgene in nla1 background without the Act5C driver (nlat2/+;nla1), confirming that nebula is indeed responsible for the observed mitochondrial phenotypes (Fig. 2b–d). Note that only partial rescue of the mitochondrial phenotypes was seen in Act5C/nlat2;nla1 flies; it is likely that the exact level of nebula expression is vital for maintaining the number and size of mitochondria. Nebula is present in the mitochondria Having established that nebula affects several aspects of mitochondrial function and morphology, we next determined whether nebula is present in the mitochondria. Western blot analyses and immunohistochemical staining of the fly brain using an antibody to the nebula protein indicated that nebula is highly expressed in neurons (Supplementary Fig. 1). Subcellular fractionation of wild-type fly heads showed that nebula is localized to the mitochondria as well as

b

Nebula

Mito Tracker

Merge

Nuc

Nuc+AP

20 µg 2.2 µg

Cyto Cyto+AP

a

Cyto Mito Nuc

Figure 3 Nebula protein is found in the mitochondria and is mainly localized to the matrix adjacent to the inner mitochondrial membrane. (a) Subcellular fractionation of wild-type fly head extract. Nebula was found in the mitochondrial (Mito) fraction, but not the cytoplasmic (Cyto) or the nuclear (Nuc) fractions, when 2.2 mg of protein was loaded, suggesting that the signal is not due to cytoplasmic or nuclear contamination. Western blots were done using antibodies to nebula (Nla) and cytochrome c (Cyt c). Cytochrome c was used as marker for mitochondria. Right, cytoplasmic and nuclear fractions treated with or without alkaline phosphatase (AP). (b) Immunostaining of primary neuronal cultures of wild-type larval brain. Nebula is found in the neurites, cytoplasm and nucleus. MitoTracker is a mitochondrial-specific marker. Nebula colocalizes with mitochondria in the neurites and cytoplasm (arrowhead). (c) Immunogold electron microscopic staining of partially purified mitochondrial sections from fly brains. Sections were labeled with nebula primary antibody and 12-nm gold-conjugated secondary antibody. As controls, mitochondria were labeled with secondary antibody only. The sections were stained with uranyl acetate so that the membranes are not stained (light gray) but the area around the membranes is stained (dark gray). Nebula is found mainly in the matrix adjacent to the inner mitochondrial membrane. Scale bar, 100 nm.

Nla

Cyt c

c Nebula

Control

to cytoplasm and nucleus (Fig. 3a). Presence of nebula in the cytoplasm is expected based on its interaction with calcineurin12, whereas nuclear localization of nebula is consistent with reports of DSCR1 in the nucleus15,21. Nebula is also found in the mitochondria, although it does not contain any of the known mitochondrial targeting signals. Detection of nebula in the mitochondrial fraction is not due to contamination from either the cytosol or the nucleus, as nebula was detected only in the mitochondrial fraction when an equal amount of protein (2.2 mg) from different fractions was loaded (Fig. 3a). Further subfractionation of mitochondria showed that nebula is located mainly in the soluble fraction rather than the membrane portion (Supplementary Fig. 3 online), suggesting that it is present in either the mitochondrial intermembrane space or the matrix. We also noticed that cytoplasmic nebula migrated as two bands on the western blot analysis, whereas nuclear and mitochondrial nebula

e

f

1580

Nebula

VOLUME 8

[

Change in activity relative to control (%)

Nebula1–120

c


d

b

ANT-V5

LacZ-V5

a

Nebula1–200

+ ANT-V5 Figure 4 Nebula interacts with ANT. (a) Nebula + Fly extract interacts with ANT in vitro. Top, specific interaction between ANT-V5 and nebula (Nla). kDa kDa Middle, input extracts used for immunoprecipi292 1 118 RRM 188 IP: anti-V5 55 55 His Calcipressin Nebula * * IP: V5 40 tation are approximately equal. LacZ-V5 and Nla 125 40 IB: Nla 33 ANT-V5 immunoprecipitation was confirmed RRM 24 1 200 55 55 Nebula1–200 His using an antibody to the V5 protein (bottom). Nla Input 40 40 IB: Nla 33 1 120 (b) Schematic diagram of histidine (His)-tagged 1–120 His 130 Nebula 24 LacZ-V5 nebula constructs. The RRM and calcipressin 55 IP: V5 40 33 IB: V5 domains were predicted using Pfam Database. ANT-V5 33 (c) Coimmunoprecipitation of V5-tagged sesB 20 Control expressed in S2 cells (ANT-V5) and His-nebula 1.2 sesB1/+ 1.2 constructs indicates that the RRM domain and 10 1.0 1.0 the calcipressin domain mediate nebula-ANT * 0.8 0.8 COX * * interaction. The arrow highlights the fact that * 0 * * 0.6 SDH 0.6 1–120 * nebula cannot interact with ANT (top). 0.4 0.4 –10 Asterisks mark the interaction between ANT-V5 0.2 0.2 and endogenous nebula in the S2 culture. IB, 0 0 –20 Control nla1 Elav/+;nlat1/+ sesB1/+ CytB mitotRNA Col Colll * immunoblot; IP, immunoprecipitation. Middle, the amounts of His-nebula constructs used in the experiment are equal. ANT-V5 immunoprecipitation was confirmed with the V5 antibody (bottom). (d) Relative ANT activity. Values were normalized to control (CS) flies and represent mean ± s.e.m. *P o 0.05. n ¼ 3 experiments done in triplicate. (e) mtDNA content in sesB1 heterozygotes (sesB1/+) as compared with control flies. n ¼ 3 independent experiments done in triplicate. (f) Quantification of the COX and SDH activities in sesB1/+ flies; n ¼ 4 independent experiments done with multiple fly heads from each group. Values represent mean ± s.e.m. *P o 0.05. Values were normalized to control (CS) flies. Relative ANT activity


ARTICLES

NUMBER 11

[


ARTICLES

Control

CB LM

sesB1/+

CB

LM

60 50

sesB1/+ nla1 sesB1/+;nla1

*

40

a

b 30

nla1 Change in SDH activity relative to control (%)

b Change in SDH activity relative to control (%)

a

* *

30 20

EP(2)0774/+

10 0 Lamina

CB

20 15 10 5

–10

CB LM

0

EP(2)0774/+; nla1

c sesB1/+;nla1

Overall

LM

nla1

EP(2)0774/+ EP(2)0774/+; nla1

40 Recovery (s)

30 20 10 0 Control

nla1/+

nla1

sesB1/+ sesB1/+;nla1

Figure 5 Nebula and sesB interact in vivo. (a) Representative SDH staining. LM, lamina; CB, central brain. (b) Quantification of changes in SDH activity as compared with control. Double mutants of sesB1/+ and nebula (sesB1/+;nla1) showed significant greater SDH activity than either nla1 or sesB1/+ flies. *P o 0.05. n ¼ 4 experiments done with multiple flies from each group. (c) Bang sensitivity. The sesB1/+;nla1 flies remained paralytic for a prolonged period after mechanical stress. The nla1 flies also showed moderate sensitivity to mechanical stress. n ¼ 20 flies. Values represent mean ± s.e.m.

appeared as a single band. Several studies have indicated that DSCR1 is a phosphoprotein22–25; we therefore examined whether nebula is also phosphorylated. Alkaline phosphatase treatment changed the pattern of cytoplasmic nebula, but not of nuclear nebula (Fig. 3a), suggesting that nebula exists in different phosphorylation states in distinct subcellular domains. To confirm the subcellular location of nebula, we stained primary neuronal cultures from wild-type larval brains with the nebula antibody and the mitochondrial-specific stain MitoTracker (Fig. 3b). Consistent with our biochemical data, we found that nebula colocalizes with mitochondria in the neurites and the cytoplasm (arrowheads) in vivo, and that nebula is present in cytoplasm and nucleus. Second, we performed immunogold electron microscopy to localize nebula within the mitochondria. Nebula is found mostly in the matrix and adjacent to the inner mitochondrial membrane and is sometimes seen in the intermembrane space (Fig. 3c), consistent with the mitochondrial subfractionation results.

Figure 6 Nebula regulates mitochondrial function independently of calcineurin. (a) SDH staining. (b) Quantification of changes in SDH activity compared with control. Mutation in CanB2 in nla1 background (EP(2)0774/+; nla1) did not rescue the increase in SDH activity seen in nla1 flies. n ¼ 3 experiments done with multiple flies from each group.

exchange of ADP/ATP between the mitochondrial matrix and the cytosol and is vital for coupling ATP synthesis to cellular energy demand27. Identification of the nebula-ANT interaction correlates with our immunogold electron microscopy results indicating that nebula is mainly found in the mitochondrial matrix adjacent to the inner mitochondrial membrane. Based on the interaction between nebula and ANT and the presence of nebula within mitochondria, we speculated that nebula might affect ANTactivity. Therefore, we directly measured ANTactivity in the nebula loss-of-function and overexpression flies by incubating mitochondria isolated from fly heads with [3H]ADP. Both nla1 and Elav/+;nlat1/+ flies have reduced ANT activity (Fig. 4d), suggesting that the proper abundance of nebula may influence ADP/ATP translocation. We also examined the D. melanogaster mutant of ANT, sesB1, for altered mitochondrial function. Because the sesB1 homozygote is semilethal, we used sesB1/+ for measuring mitochondrial function. The sesB1 heterozygotes also have reduced ANT activity (Fig. 4d), mtDNA content and COX activity, whereas SDH activity remained comparable to that of the control (Fig. 4e,f). Consistently, the number of mitochondria per area of photoreceptor axon in the lamina cartridge of sesB1/+ was normal compared with wild-type flies (sesB1/+, 0.6 ± 0.1, versus CS, 0.5 ± 0.1). We also examined a few escapers of sesB1/Y for the number of mitochondria in the lamina under electron microscopy. We

1.2

Nebula interacts with ANT and regulates its activity To understand how mutations in nebula affect mitochondrial function, we sought to identify proteins interacting with nebula by means of immunoprecipitation. We overexpressed hemagglutinin (HA)-tagged nebula in fly neurons and used an antibody to HA peptide to pull down nebula together with other possible interacting proteins. After polyacrylamide gel electrophoresis and peptide sequencing, we identified that nebula coimmunoprecipitated with sesB, a D. melanogaster homolog of human ANT26. We further confirmed the ANT-nebula interaction using D. melanogaster S2 cells overexpressing V5-tagged ANT (Fig. 4a). To better identify regions of nebula responsible for its interaction with ANT, we generated different nebula deletion constructs and examined their ability to interact with ANT-V5 (Fig. 4b,c). Coimmunoprecipitation experiments revealed that the calcipressin domain and the RNA-recognition motif (RRM) of nebula are important for its interaction with ANT (Fig. 4c). ANT, a protein located in the mitochondrial inner membrane, is responsible for the


[

NUMBER 11

[

NOVEMBER 2005

1.0 Relative mtDNA level


nla1

Central brain

25

0.8 Control 0.6 Trisomy 21 0.4

0.2 0 mtRNA

Co I

Co III

Cyt B

Figure 7 Trisomy 21 fetal brain tissues show reduced mtDNA content. mtDNA content in 22-week-old control and trisomy 21 fetal brains was determined by quantitative PCR. mtRNA, primers specific for mitochondrial tRNAs; Co I and Co III, primers for cytochrome c oxidase subunits I and III, respectively; Cyt B, primers for cytochrome b. Relative change was calculated by normalizing to the control brains. Values represent mean ± s.e.m. n ¼ 2 samples each tested in triplicate.

1581


ARTICLES

Brain from fetuses with trisomy 21 shows reduced mtDNA content We have previously shown that DSCR1 transcripts are increased by 1.5-fold in brain from fetuses with Down syndrome12. To correlate the findings obtained using our D. melanogaster model with DSCR1 overexpression in Down syndrome, we determined the mtDNA content in human brain tissue from fetuses with trisomy 21 using quantitative RT-PCR. Consistent with the results obtained from flies overexpressing nebula, the amount of mtDNA in the brain from fetuses with Down syndrome is substantially lower than in normal fetal brains (Fig. 7). These results confirm that Down syndrome is associated with mitochondrial dysfunction and further imply that DSCR1 overexpression may contribute to mitochondrial dysfunction in Down syndrome.

DISCUSSION In this study, we demonstrated that perturbation in the abundance of nebula caused several mitochondrial abnormalities, including elevated ROS, decreased ATP and mtDNA, abnormal mitochondrial enzyme activities, and changes in the number and size of mitochondria. Furthermore, we have shown that nebula is present in the mitochondria, and that it can interact with ANT and influences its activity. Based on this and previous reports, we propose that mutations in nebula affect mitochondrial function mainly by acting through ANT (Supplementary Fig. 4 online). Several lines of evidence support the idea that the mitochondrial phenotypes are a result of perturbed ANT function in the nebula mutants. First, our results show that nebula can interact with ANT in vitro and that mutations in nebula affected ANT activity in isolated mitochondria. Second, the mitochondrial phenotypes seen in the sesB mutant are comparable to those seen in the nebula mutants. Third, double mutants of nebula and sesB showed enhancement of the phenotypes associated with the individual mutations, thus supporting the notion that nebula and sesB function in the same pathway. Fourth, other studies suggest that the mammalian ANT isoform 1 (ANT1) protein is also important for mitochondrial DNA maintenance and mitochondrial integrity32–35. The above observations are consistent with the model that nebula can interact with ANT to modulate mitochondrial function and integrity. How does the decrease in ANT activity seen in the nebula mutants then cause the observed mitochondrial phenotypes? A reduction in ANTactivity can disrupt the cellular equilibrium of ADP/ATP, resulting in a block in the electron transport chain and a decrease in ATP synthesis36. Furthermore, studies have suggested that any perturbation of oxidative phosphorylation could lead to accumulation of intracellular ROS2,37. As a result, ROS may subsequently cause oxidative stress, leading to mtDNA damage and further disruption of the enzymes in the electron transport chain2,36. We have shown that flies with nebula loss-of-function or overexpression showed a similar decrease in ANT activity. It is possible that an optimum abundance of nebula is crucial for normal ANT function; thus, altering the amount of nebula protein in either direction may lead to similar affect on ANT activity. This result is similar to our previous finding that a fine balance in the abundance of nebula is required for normal brain function such as learning12. Future experiments using reconstituted ANT membrane preparations will help to determine the molecular nature of the nebula-ANT interaction. The nebula loss-of-function flies showed a different increase in the number of mitochondria compared with nebula overexpression mutant and sesB1 flies. Although deficient ANT activity may be the main cause of the mitochondrial phenotypes, participation of nebula in other physiological pathways may contribute to the different degrees of increase in mitochondrial number. The presence of nebula in different subcellular domains support the idea that nebula is a multifunctional protein. Nebula acts as an inhibitor of calcineurin in the cytosol12. Here we excluded the possible involvement of calcineurin in generating the mitochondrial phenotypes seen in the nebula mutants, as reducing calcineurin activity in the nebula mutant background did not rescue the increased SDH activity in nla1 flies (Fig. 6). We have also shown that nebula is found in the nucleus. Sequence analyses suggest that nebula and proteins in the calcipressin family may work as transcription factors because of their acidic and proline-rich domains11. Similar to DSCR1, nebula also contains an RRM domain, which is found in many RNA-binding proteins and some single-stranded, DNA-binding proteins38. It is thus plausible that nebula may regulate gene expression in the nucleus. It

1582

VOLUME 8

detected a small change in the number of mitochondria per area of photoreceptor axon (sesB1/Y, 0.8 ± 0.1, versus CS, 0.5 ± 0.1; P o 0.05) as well as a modest decrease in the cross-sectional area of mitochondria (sesB1/Y, 0.22 ± 0.02 mm2, versus CS, 0.31 ± 0.04 mm2; P o 0.05). These changes in mitochondrial function and morphology are similar to those seen in the flies overexpressing nebula. To verify that nebula and ANT interacted in vivo, we generated double mutants of sesB1 and nla1 (sesB1/+;nla1). We used SDH activity as a measure of interaction, as sesB1/+ did not show altered SDH activity. We found that sesB1/+;nla1 flies showed an enhancement in SDH activity compared with nla1 or sesB1/+ flies (Fig. 5a,b). The increase in SDH activity was seen in both the lamina (P o 0.05) and the central brain (P o 0.001), indicating that nebula and sesB interact in vivo. We also used sensitivity to mechanical stress to show genetic interaction between nebula and sesB, as sesB homozygous flies and another mutant of mitochondrial function have been shown to be bang sensitive26,28,29. To measure sensitivity to mechanical stress, flies 7–10 d old were vortexed for 10 s, and the time that it took for the flies to recover by actively trying to upright themselves was recorded. In this assay, normal flies recovered almost immediately after vortex, whereas the sesB1/+;nla1 flies remained paralytic for a prolonged period after mechanical stress (Fig. 5c). The observation that sesB1/+;nla1 flies required a longer time to recover than either sesB1/+ or nla1 flies further confirmed interaction between nebula and sesB. The nla1 flies also showed moderate sensitivity to mechanical stress (nla1, 5.9 ± 1.3 s, versus CS, 0.8 ± 0.2 s; P o 0.001), suggesting that mutation in nebula causes the flies to be more intolerant to exercise. Using biochemical and behavioral analyses, we demonstrated that nebula and sesB indeed interact in vivo to affect mitochondrial function. Role of nebula in mitochondria is calcineurin independent As we have previously shown that nebula regulates calcineurinmediated signaling, we next examined whether the elevated calcineurin activity seen in nla1 flies also contributes to the mitochondrial phenotypes12. We generated nla1 flies containing a mutation in the calcineurin B2 (CanB2) gene (EP(2)0774/+; nla1) to restore the elevated calcineurin activity in nla1 flies. D. melanogaster CanB2 is expressed in the CNS and encodes the regulatory subunit of calcineurin, which is obligatory for calcineurin activity30,31. Homozygous mutants of CanB2 are lethal, whereas heterozygous mutation of CanB2 is sufficient to decrease calcineurin activity30,31. Reducing the abundance of CaNB2 in the nla1 background did not rescue the increase in SDH activity (Fig. 6), suggesting that nebula does not act through calcineurin to regulate mitochondrial number. Furthermore, heterozygous mutants of CaNB2 (EP(2)0774/+) did not show altered SDH activity compared with the control, indicating that reducing calcineurin activity alone is also not sufficient to affect mitochondrial function.

[

NUMBER 11

[



ARTICLES will be interesting to determine whether nuclear nebula affects mitochondrial function. Our findings that nebula affects mitochondrial function may have several physiological consequences. We have previously shown that mutations in nebula resulted in altered calcineurin activity and defective learning. Although altered calcineurin activity is not likely to be the cause of mitochondrial dysfunction, we cannot exclude the possibility that defective mitochondrial respiration affects learning and memory, as ATP is essential for synaptic transmission and enzymes involved in learning and memory pathways1. Furthermore, mitochondrial dysfunction has been linked to aging and a diverse array of neuromuscular and neurological disorders, including Down syndrome and Alzheimer disease3,4,6,7. Studies have shown that tissues from individuals with Down syndrome tissues show intracellular accumulation of toxic ROS and increased lipid peroxidation, leading to neuronal apoptosis5,10 as well as impaired mitochondrial enzyme activities and oxidative damage4,8. There is also a report of abundant mitochondria in a person with Down syndrome with lymphoblast leukemia39. Similarly, in Alzheimer disease, there are consistent reports of defective mitochondrial enzyme activities and abnormal mitochondrial structure40,41. Notably, most people with Down syndrome develop the pathological and neurochemical changes of Alzheimer disease by about age 50 (refs. 42,43). Examination of postmortem brains from people with Alzheimer disease revealed that there is an increase in the level of DSCR1 transcripts44 analogous to Down syndrome tissues. Together with our finding that nebula regulates mitochondrial function and integrity, and that Down syndrome fetal brains contain reduced mtDNA content, it is intriguing to speculate that DSCR1 may contribute to the mitochondrial defects seen in both Down syndrome and Alzheimer disease. Future experiments examining ANT activity in Down syndrome and Alzheimer disease tissues will help to determine how nebula/DSCR1 upregulation affects mitochondrial function in these diseases. Furthermore, mutations in human SLC25A4 (formerly ANT1) have been linked to autosomal dominant progressive external ophthlamoplegia34–36. People with Down syndrome also have a substantially elevated incidence of ophthalmoplegic disorders42. Based on our finding that nebula interacts with ANT, it will be interesting to determine whether DSCR1 overexpression also contributes to ophthalmoplegic disorders. Down syndrome is a complex disorder caused by an imbalanced dosage of genes on all or part of chromosome 21. Although studies with mouse Down syndrome models have provided useful information in investigating the mechanisms of some of the phenotypes in Down syndrome45, understanding the genotype-to-phenotype correlations remains a major challenge. D. melanogaster, with its powerful genetics, is an ideal model system for investigating the genotype to phenotype correlation of a given gene or a group of genes. In this study, we uncovered a previously unknown role of nebula in the mitochondria. We found that nebula is crucial for mitochondrial function and mtDNA maintenance, as well as ANT activity. Our findings may provide new insights into the mechanism underlying mitochondrial dysfunction in Down syndrome. METHODS D. melanogaster stocks and material. The generation of hypomorphic nla1 mutant and transgenic flies (nlat1 and nlat2) has been described12. We generated nla62 flies by imprecise P-element excision of nla1. Sequencing analysis revealed that nla62 still contains a partial P-element and chromosomal rearrangement adjacent to the site of P-element insertion. The expression level of nla1/nla62 was determined by quantitative RT-PCR and western blot analysis. The sesB1 stock was obtained from M. Ashburner (University of Cambridge).


[

NUMBER 11

[

NOVEMBER 2005

EP(2)0774/Cyo stock was obtained from the Bloomington D. melanogaster Stock Center. Human control and trisomy 21 fetal brain tissues were obtained from the Brain and Tissue Bank at the University of Maryland, Baltimore. ATP abundance. The ATP bioluminescence HSII assay kit (Roche) was used to measure total ATP in fly heads, as described46. ATP concentrations were determined by measuring the luciferase activity using a Wallac 1450 MicroBeta counter (Perkin Elmer). ROS detection. Flies were briefly etherized and dissected in Schneider’s medium (Invitrogen). Freshly dissected fly brains were incubated with 20 mM dihydroethidium (Invitrogen) at room temperature (23 1C) for 15 min. After several washes, brains were mounted in PBS between coverslips to avoid crushing them. To minimize variations in signal intensity, brains of different fly lines were mounted on the same slide and fluorescence intensities of at least three different brains of the same fly line were averaged to give one data point. Experiments were repeated three times, totaling a minimum of nine brains from each fly line. AxioVision 4.2 was used for the analysis of fluorescence intensity (Carl Zeiss). COX and SDH activities. We used 10-mm fresh-frozen horizontal head sections. For SDH staining, fly head sections were incubated with the staining solution (5 mM EDTA, 1 mM potassium cyanide, 0.2 mM phenazine methosulfate, 50 mM succinic acid, 1.5 mM nitro blue tetrazolium (NBT) in 5 mM sodium phosphate buffer) for 10 min at room temperature. For COX activity, fly head sections were incubated with staining solution (0.1%, 3¢-diaminobenzidine (DAB), 0.1% cytochrome c, 0.02% catalase in 5 mM sodium phosphate buffer) for 20 min at room temperature. Experimental and control samples were stained for an equal amount of time and carried out in parallel. The image of each fly brain was captured using a CCD camera (Diagnostic Instruments), and the same exposure time that yielded no saturating pixel intensity was used for all images within the same experiment. The average staining intensity of each fly brain was determined using AxioVision 4.2 (Carl Zeiss). Electron microscopy. Fly heads were fixed overnight at 4 1C in 2% paraformaldehyde plus 2% glutaraldehyde in 0.1 N cacodylate buffer, postfixed in 1% osmium tetroxide at room temperature, dehydrated in an ethanol series and embedded in Epon 812. Ultrathin sections were examined with a JEOL 200 CX transmission electron microscope at 80 kV. Quantification of the number of mitochondria per area of photoreceptor axon was done using the AxioVision 4.2 program (Carl Zeiss). To reduce variability, only cross-sections of the lamina cartridge showing six photoreceptor axons were used for calculation. Student’s t-test was used to assess statistical significance. Immunogold electron microscopy. Partially purified mitochondria were isolated from Canton-S fly heads as described above. The mitochondrial pellet was fixed in 4% paraformaldehyde at 4 1C overnight and then cryoprotected. We cut 70-nm ultrathin cryosections at –100 1C and placed on Formvar and carbon-coated nickel grids. Immunolabeling was performed by slight modifications of the Tokuyasu technique47, using antibody to nebula at 1:1,000 overnight at 4 1C, and 12-nm gold-conjugated donkey antibody to rabbit at 1:20 for 1 h at room temperature. For a negative control, primary antibody was omitted in the preparation. Subcellular fractionation and alkaline phosphatase treatment. Heads of wildtype (CS) flies were decapitated and collected on ice. Cytoplasmic, nuclear and mitochondrial fractions were isolated by differential centrifugation using the Mitochondria Isolation Kit (Sigma). All steps were carried out at 4 1C. Briefly, flies were homogenized in extraction buffer, and nuclei and debris were removed by low-speed centrifugation at 1,000g for 5 min. The mitochondrial-enriched fraction was obtained by pelleting mitochondria at 7,500 g for 10 min, and the supernatant contains the cytoplasmic fraction. The nuclear fraction was washed three times and spun at 1,000g to remove contaminating debris. To obtain a more purified heavy mitochondrial fraction for western blot analysis, the mitochondrial-enriched fraction was spun at 1,000g to remove contaminants, and mitochondria were further pelleted at 4,000g and washed two additional times. The cytoplasmic fraction was cleared by spinning the supernatant at 14,000 r.p.m. for 15 min. Alkaline phosphatase treatment was

1583

ARTICLES


performed by incubating 20 mg of protein with 5 mg protein per unit of alkaline phosphatase for 40 min at 37 1C.

1. Mattson, M.P. & Liu, D. Energetics and oxidative stress in synaptic plasticity and neurodegenerative disorders. Neuromolecular Med. 2, 215–231 (2002). 2. Balaban, R.S., Nemoto, S. & Finkel, T. Mitochondria, oxidants, and aging. Cell 120, 483–495 (2005). 3. Wallace, D.C. Mitochondrial diseases in man and mouse. Science 283, 1482–1488 (1999). 4. Arbuzova, S., Hutchin, T. & Cuckle, H. Mitochondrial dysfunction and Down’s syndrome. Bioessays 24, 681–684 (2002). 5. Busciglio, J. & Yankner, B.A. Apoptosis and increased generation of reactive oxygen species in Down’s syndrome neurons in vitro. Nature 378, 776–779 (1995). 6. Busciglio, J. et al. Altered metabolism of the amyloid beta precursor protein is associated with mitochondrial dysfunction in Down’s syndrome. Neuron 33, 677–688 (2002). 7. Zeviani, M. & Carelli, V. Mitochondrial disorders. Curr. Opin. Neurol. 16, 585–594 (2003). 8. Prince, J., Jia, S., Bave, U., Anneren, G. & Oreland, L. Mitochondrial enzyme deficiencies in Down’s syndrome. J. Neural Transm. Park. Dis. Dement. Sect. 8, 171–181 (1994). 9. Druzhyna, N., Nair, R.G., LeDoux, S.P. & Wilson, G.L. Defective repair of oxidative damage in mitochondrial DNA in Down’s syndrome. Mutat. Res. 409, 81–89 (1998). 10. Capone, G. et al. Evidence for increased mitochondrial superoxide production in Down syndrome. Life Sci. 70, 2885–2895 (2002).

11. Fuentes, J.J. et al. A new human gene from the Down syndrome critical region encodes a proline-rich protein highly expressed in fetal brain and heart. Hum. Mol. Genet. 4, 1935–1944 (1995). 12. Chang, K.T., Shi, Y.J. & Min, K.T. The Drosophila homolog of Down’s syndrome critical region 1 gene regulates learning: implications for mental retardation. Proc. Natl. Acad. Sci. USA 100, 15794–15799 (2003). 13. Fuentes, J.J. et al. DSCR1, overexpressed in Down syndrome, is an inhibitor of calcineurin-mediated signaling pathways. Hum. Mol. Genet. 9, 1681–1690 (2000). 14. Kingsbury, T.J. & Cunningham, K.W. A conserved family of calcineurin regulators. Genes Dev. 14, 1595–1604 (2000). 15. Rothermel, B. et al. A protein encoded within the Down syndrome critical region is enriched in striated muscles and inhibits calcineurin signaling. J. Biol. Chem. 275, 8719–8725 (2000). 16. Ryeom, S., Greenwald, R.J., Sharpe, A.H. & McKeon, F. The threshold pattern of calcineurin-dependent gene expression is altered by loss of the endogenous inhibitor calcipressin. Nat. Immunol. 4, 874–881 (2003). 17. Ermak, G., Harris, C.D. & Davies, K.J. The DSCR1 (Adapt78) isoform 1 protein calcipressin 1 inhibits calcineurin and protects against acute calcium-mediated stress damage, including transient oxidative stress. FASEB J. 16, 814–824 (2002). 18. Crawford, D.R. et al. Hamster adapt78 mRNA is a Down syndrome critical region homologue that is inducible by oxidative stress. Arch. Biochem. Biophys. 342, 6–12 (1997). 19. Leahy, K.P. & Crawford, D.R. adapt78 protects cells against stress damage and suppresses cell growth. Arch. Biochem. Biophys. 379, 221–228 (2000). 20. Lewis, D.L., Farr, C.L. & Kaguni, L.S. Drosophila melanogaster mitochondrial DNA: completion of the nucleotide sequence and evolutionary comparisons. Insect Mol. Biol. 4, 263–278 (1995). 21. Pfister, S.C., Machado-Santelli, G.M., Han, S.W. & Henrique-Silva, F. Mutational analyses of the signals involved in the subcellular location of DSCR1. BMC Cell Biol. 3, 24 (2002). 22. Genesca, L. et al. Phosphorylation of calcipressin 1 increases its ability to inhibit calcineurin and decreases calcipressin half-life. Biochem. J. 374, 567–575 (2003). 23. Lin, H.Y. et al. Oxidative and calcium stress regulate DSCR1 (Adapt78/MCIP1) protein. Free Radic. Biol. Med. 35, 528–539 (2003). 24. Hilioti, Z. et al. GSK-3 kinases enhance calcineurin signaling by phosphorylation of RCNs. Genes Dev. 18, 35–47 (2004). 25. Vega, R.B., Yang, J., Rothermel, B.A., Bassel-Duby, R. & Williams, R.S. Multiple domains of MCIP1 contribute to inhibition of calcineurin activity. J. Biol. Chem. 277, 30401–30407 (2002). 26. Zhang, Y.Q. et al. stress sensitive B encodes an adenine nucleotide translocase in Drosophila melanogaster. Genetics 153, 891–903 (1999). 27. Klingenber, M. in The Enzymes of Biological Membranes (ed. Martonosi, A.N.) (Plenum, New York, 1985). 28. Homyk, T., Jr. Behavioral mutations of Drosophila melanogaster. II. Behavioral analysis and locus mapping. Genetics 87, 105–128 (1977). 29. Royden, C.S., Pirrotta, V. & Jan, L.Y. The tko locus, site of a behavioral mutation in D. melanogaster, codes for a protein homologous to prokaryotic ribosomal protein S12. Cell 51, 165–173 (1987). 30. Sullivan, K.M. & Rubin, G.M. The Ca(2+)-calmodulin-activated protein phosphatase calcineurin negatively regulates EGF receptor signaling in Drosophila development. Genetics 161, 183–193 (2002). 31. Gajewski, K. et al. Requirement of the calcineurin subunit gene canB2 for indirect flight muscle formation in Drosophila. Proc. Natl. Acad. Sci. USA 100, 1040–1045 (2003). 32. Graham, B.H. et al. A mouse model for mitochondrial myopathy and cardiomyopathy resulting from a deficiency in the heart/muscle isoform of the adenine nucleotide translocator. Nat. Genet. 16, 226–234 (1997). 33. Esposito, L.A., Melov, S., Panov, A., Cottrell, B.A. & Wallace, D.C. Mitochondrial disease in mouse results in increased oxidative stress. Proc. Natl. Acad. Sci. USA 96, 4820– 4825 (1999). 34. Kaukonen, J. et al. Role of adenine nucleotide translocator 1 in mtDNA maintenance. Science 289, 782–785 (2000). 35. Biousse, V., Pardue, M.T., Wallace, D.C. & Newman, N.J. The eyes of mitomouse: mouse models of mitochondrial disease. J. Neuroophthalmol. 22, 279–285 (2002). 36. Wallace, D.C. Mouse models for mitochondrial disease. Am. J. Med. Genet. 106, 71–93 (2001). 37. St-Pierre, J., Buckingham, J.A., Roebuck, S.J. & Brand, M.D. Topology of superoxide production from different sites in the mitochondrial electron transport chain. J. Biol. Chem. 277, 44784–44790 (2002). 38. Strippoli, P., Lenzi, L., Petrini, M., Carinci, P. & Zannotti, M. A new gene family including DSCR1 (Down Syndrome Candidate Region 1) and ZAKI-4: characterization from yeast to human and identification of DSCR1-like 2, a novel human member (DSCR1L2). Genomics 64, 252–263 (2000). 39. Hodson, D., Gatward, G. & Erber, W. Azurophilic granules in acute lymphoblastic leukaemia resulting from abundant mitochondria. Br. J. Haematol. 125, 265 (2004). 40. Hirai, K. et al. Mitochondrial abnormalities in Alzheimer’s disease. J. Neurosci. 21, 3017–3023 (2001). 41. Wong-Riley, M. et al. Cytochrome oxidase in Alzheimer’s disease: biochemical, histochemical, and immunohistochemical analyses of the visual and other systems. Vision Res. 37, 3593–3608 (1997).

1584

VOLUME 8

D. melanogaster larval brain culture and immunohistochemistry. Primary dissociated larval brain cultures were prepared as described48. Cells were maintained for 3 d at room temperature before being used for immunohistochemistry. Cells were incubated with 100 nM MitoTracker Red CMXRox (Invitrogen) for 30 min at room temperature and washed with culture medium. Cells were then fixed in 4% paraformaldehyde for 20 min at room temperature, washed with PBS + 0.1% Triton X-100 (PBST) and blocked with 5% normal donkey serum in PBST. Rabbit polyclonal nebula antibody was used at 1:2,000 and FITC-conjugated donkey anti-rabbit secondary antibody was used at 1:200 (Jackson ImmunoResearch). ANT activity. Mitochondrial-enriched fractions were prepared from each fly line as described above, but were washed an additional two times by spinning at 7,500g. The amount of mitochondria was determined by quantifying the amount of mitochondrial protein using the Bradford protein assay. ANT activity was measured by slight modification of the inhibitor-stop method49. Briefly, 4 mg of mitochondria from each fly line was incubated with 1 mCi of 3H-ADP for 1 min at 4 1C. The reaction was then stopped by 10 mM atractyloside, an ANT inhibitor, and washed three times to remove residual [3H]ADP. Mitochondria were lysed with 0.1 M NaOH and spotted onto filter paper for quantification using a Wallac 1450 MicroBeta counter (Perkin Elmer). Bang-sensitivity assay. Bang sensitivity was determined by a slight modification of a described method50. Individual flies that were between 7 and 10 d old were transferred to empty vials and vortexed for 10 s at the highest sitting using a Daigger Vortex Genie 2 (A. Daigger & Co.). The time that it took for each fly to recover by actively kicking and trying to right itself was recorded. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank the EM Facility, Protein/Peptide Sequencing Facility and Light Imaging Facility at NINDS, NIH for assistance; EM Facility at Johns Hopkins University for help with immunogold electron microscopy; M. Ashburner for the sesB1 stock; R. Garesse for the ATPb synthase antibody; the Brain and Tissue Bank at the University of Maryland, Baltimore, for fetal brain tissues; H. Nash and K. Fischbeck for critical reading of the manuscript; and Y. Shi for technical assistance. This work was supported by an intramural fund from NINDS, NIH, and funds from the March of Dimes Birth Defects Foundation to K.-T.M. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

[

NUMBER 11

[



ARTICLES 42. Roizen, N.J. & Patterson, D. Down’s syndrome. Lancet 361, 1281–1289 (2003). 43. Menendez, M. Down syndrome, Alzheimer’s disease and seizures. Brain Dev. 27, 246– 252 (2005). 44. Ermak, G., Morgan, T.E. & Davies, K.J. Chronic overexpression of the calcineurin inhibitory gene DSCR1 (Adapt78) is associated with Alzheimer’s disease. J. Biol. Chem. 276, 38787–38794 (2001). 45. Gardiner, K. et al. Report on the ‘Expert workshop on the biology of chromosome 21: towards gene-phenotype correlations in Down syndrome’, held June 11–14, 2004, Washington D.C. Cytogenet. Genome Res. 108, 269–77 (2005). 46. Cvejic, S., Zhu, Z., Felice, S.J., Berman, Y. & Huang, X.Y. The endogenous ligand Stunted of the GPCR Methuselah extends lifespan in Drosophila. Nat. Cell Biol. 6, 540– 546 (2004).


[

NUMBER 11

[

NOVEMBER 2005

47. Tokuyasu, K.T. Immunochemistry on ultrathin frozen sections. Histochem. J. 12, 381– 403 (1980). 48. Wu, C.F., Suzuki, N. & Poo, M.M. Dissociated neurons from normal and mutant Drosophila larval central nervous system in cell culture. J. Neurosci. 3, 1888–1899 (1983). 49. Duan, J. & Karmazyn, M. Relationship between oxidative phosphorylation and adenine nucleotide translocase activity of two populations of cardiac mitochondria and mechanical recovery of ischemic hearts following reperfusion. Can. J. Physiol. Pharmacol. 67, 704–709 (1989). 50. Ganetzky, B. & Wu, C.F. Drosophila mutants with opposing effects on nerve excitability: genetic and spatial interactions in repetitive firing. J. Neurophysiol. 47, 501–514 (1982).

1585


ARTICLES

Neural correlates of binocular rivalry in the human lateral geniculate nucleus Klaus Wunderlich, Keith A Schneider & Sabine Kastner When dissimilar images are presented to the two eyes, they compete for perceptual dominance so that only one image is visible at a time while the other one is suppressed. Neural correlates of such binocular rivalry have been found at multiple stages of visual processing, including striate and extrastriate visual cortex. However, little is known about the role of subcortical processing during binocular rivalry. Here we used fMRI to measure neural activity in the human LGN while subjects viewed contrast-modulated gratings presented dichoptically. Neural activity in the LGN correlated strongly with the subjects’ reported percepts, such that activity increased when a high-contrast grating was perceived and decreased when a low-contrast grating was perceived. Our results provide evidence for a functional role of the LGN in binocular rivalry and suggest that the LGN, traditionally viewed as the gateway to the visual cortex, may be an early gatekeeper of visual awareness.

Binocular rivalry occurs when the input from the two eyes cannot be fused into a single, coherent percept. Rivalry can be induced experimentally by simultaneously presenting dissimilar stimuli to the two eyes, such as a vertical grating to one eye and a horizontal grating to the other. Rather than being perceived as a merged plaid, the two stimuli compete for perceptual dominance such that subjects perceive only one stimulus at a time while the other is suppressed from visual awareness1. Usually one stimulus predominates for several seconds, and the extent of competition between any pair of stimuli depends on stimulus properties, such as their relative contrast or spatial frequency2,3. Because the subjects’ perceptual experiences change over time while the retinal stimulus remains constant, binocular rivalry provides an intriguing paradigm to study the neural basis of visual awareness4. The neural mechanisms underlying binocular rivalry have been much debated. In monkeys trained to report their perceptual experiences during rivalry, single-cell physiology experiments have demonstrated the existence of neural correlates of binocular rivalry mainly in higher-order visual areas5. The responses of about 90% of neurons in inferior temporal cortex increase when the neuron’s preferred stimulus is perceived during rivalry, whereas only about 40% of neurons in areas V4 and MT, and even fewer in early visual areas V1 and V2, show such response enhancement6,7. On the basis of these findings, it has been concluded that binocular rivalry is mediated by competitive interactions between binocular neuronal populations representing the two stimuli at several stages of visual processing subsequent to the convergence of the input from the two eyes in V1 (the pattern-competition account). Alternatively, it has been suggested that binocular rivalry reflects competition between monocular channels either at the level of V1 or the lateral geniculate nucleus (LGN) and is mediated by mutual inhibition and reciprocal feedback suppressing the input from one

eye1,8. This interocular-competition account has recently received support from fMRI studies showing that signal fluctuations in area V1 (ref. 9) and, more importantly, in the monocular V1 neurons representing the blind spot10 are correlated with subjects’ perceptual experiences. Neural activity of monocular V1 neurons varies according to subjects’ perceptual reports, and the signal amplitudes measured during rivalry are similar to those measured during presentations of identical monocular stimuli; together, these observations suggest that rivalry is completely resolved in monocular V1 neurons. However, little is known about the role of subcortical processing stages—such as the LGN— in binocular rivalry. The LGN is the thalamic station in the projection of the visual pathway from retina to V1 (ref. 11). It is typically organized into six layers, each of which receives input from either the contralateral or the ipsilateral eye and contains a retinotopic map of the contralateral hemifield registered to those of other layers. In addition to retinal afferents, the LGN receives input from multiple sources including V1 and the thalamic reticular nucleus (TRN). Given its anatomical organization and afferent projections, the LGN has often been considered a possible site of suppression in accounts of interocular competition8,12. However, single-cell recording studies in the LGN of awake monkeys viewing rivalrous stimuli have not found evidence to support this hypothesis13. We investigated the functional role of the human LGN in binocular rivalry using fMRI in subjects viewing dichoptically presented contrastmodulated grating stimuli9. We found that fMRI signals in the LGN and V1 were strongly correlated with subjects’ perceptual experiences during binocular rivalry. The amplitude of fMRI signals increased when subjects perceived a high-contrast stimulus and decreased when they perceived a low-contrast stimulus. A similar response pattern—

Department of Psychology, Center for the Study of Brain, Mind, and Behavior, Princeton University, Princeton, New Jersey 08544, USA. Correspondence should be addressed to S.K. ([email protected]). Received 7 July; accepted 1 September; published online 23 October 2005; doi:10.1038/nn1554


[

NUMBER 11

[

NOVEMBER 2005

1595

ARTICLES

a

Rivalry Left eye Right eye

Left eye Right eye

Left eye Right eye

Left eye Right eye

Left eye Right eye

Stimulation

b

Physical alternation Left eye Right eye

Stimulation

Perception

Time (s)

c fMRI signal (%)


Perception

d

Rivalry

Physical alternation

2 0 –2 0

10

20 30 Time (s)

40

50

0

10

20 30 Time (s)

40

50

Figure 1 Experimental design and stimuli. (a) Subjects viewed red/green orthogonal sinusoidal gratings through matching filter glasses, with the higher-contrast horizontal grating visible to only one eye and the lower-contrast vertical grating to the other eye. Despite the invariant physical stimulation, subjects experienced binocular rivalry and reported switches in perception between horizontal and vertical gratings every few seconds. (b) The perceptual experience during rivalry was simulated in a physical stimulus alternation condition by presenting the green or the red grating to one eye and a uniform field to the other eye. The presentation times of alternating stimuli were identical to perceptual durations of the corresponding grating that the same subject reported during rivalry. All stimulus parameters were identical to the rivalry condition. In both experiments, subjects maintained fixation and indicated by button presses which grating was perceived. (c) Fluctuations of fMRI signals related to the perception of the high- and low-contrast gratings during binocular rivalry are evident in the raw time series of fMRI signals in the LGN from a single subject (S1). Phases during which the subject perceived high-contrast horizontal or low-contrast vertical gratings are shaded in green and red, respectively. Periods of intermittent piecemeal perception are not colored. (d) Raw time series of fMRI signals in the LGN from the same subject viewing physical stimulus alternations.

RESULTS Five subjects participated in one scanning session for the rivalry experiment, followed by one session for the physical alternation experiment. In the rivalry experiment, subjects viewed superimposed sinusoidal gratings through red or green filter glasses such that one eye viewed a high-contrast, green, horizontal grating and the other viewed a low-contrast, red, vertical grating (Fig. 1a). The gratings filled an annular aperture centered on a fixation point and reversed contrast to minimize adaptation. The orthogonal orientations of the two gratings prevented them from being fused and also induced rivalrous perceptual oscillations between them. The luminance contrasts and reversal rates of the gratings were individually optimized for each subject so as to maximize the perceptual duration of the weaker, low-contrast stimulus (Table 1). Subjects maintained fixation and reported which grating was perceived by pressing a button; periods of mixed ‘piecemeal’ percepts of the two stimuli were indicated with a third button. In the physical alternation experiment, we used sequential monocular presentations of the same grating stimuli to produce perceptions similar to, but physical stimulations different from, those in the rivalry experiment. This was achieved by presenting the low- or high-contrast grating to one eye and a uniform field to the other eye (Fig. 1b), using the identical temporal sequence of stimulus alternations reported by

the same subject in the rivalry experiment. During these physical alternations, subjects maintained fixation and pressed buttons to indicate which grating they perceived. In the LGN and V1, the amplitude of fMRI signals increases monotonically with stimulus contrast; reliable fMRI signals are typically evoked by stimuli of more than 10% contrast and signal saturation occurs with stimuli of more than 35% contrast14–16. Therefore, the different signal amplitudes evoked by low- and high-contrast stimuli can be used as a ‘neural signature’ of the LGN and V1 populations representing these stimuli—as was previously shown for physical and rivalrous alternations of contrast-modulated gratings represented in V1 (ref. 9). In the physical alternation experiment, we expected fMRI signals to increase when the high-contrast grating was shown monocularly and to decrease when the low-contrast grating was shown monocularly. Further, we reasoned that if the subjects’ perceptual experiences during rivalry were reflected in the fMRI signals, signal fluctuations similar to those obtained during the physical alternations should reflect the reported percepts, despite the unchanging retinal stimulation. We used the contrast-modulated grating paradigm in both the rivalry and physical alternation conditions (i) to replicate previous findings showing that signal fluctuations in V1 were related to perceptual experience during rivalry9,10, (ii) to investigate whether such signal fluctuations were present even earlier than V1 in the visual processing hierarchy—namely, in the LGN, (iii) to compare, within each area (that is, LGN and V1), the signal obtained during rivalry with that obtained during physical alternation, and (iv) to compare the signal obtained in one area, during rivalry or physical alternation, with its counterpart in the other area. In each scanning session, we identified regions of interest in the thalamus and visual cortex by presenting flickering checkerboard stimuli alternately to the right and left visual hemifields, while the subject maintained fixation. The checkerboards activated the right and

1596

VOLUME 8

mimicking their perceptions during rivalry—was obtained when subjects viewed sequences of non-rivalrous physical alternations of the same stimuli. Our results provide the first evidence that, in humans, neural correlates of binocular rivalry can be found even earlier than V1 in the visual processing hierarchy: that is, in the LGN. These findings support interocular-competition accounts of binocular rivalry, including models of selective suppression of eye-specific LGN layers. Further, they indicate that neural correlates of conscious perception are not confined to cortical processing.

[

NUMBER 11

[


ARTICLES Table 1 Stimulus conditions and perceptual dominance Contrast

Grating

Eye dominance

L/H (%)

Cyc/deg

M

%

S

M

%

S

S1 S2

Left Right

17/70 17/70

0.4 0.4

9.3 4.2

68 55

116 201

4.0 3.0

29 39

112 201

S3 S4

Right Right

14/70 20/70

0.5 0.4

6.1 5.1

31 49

77 306

4.9 3.5

35 34

108 309

S5

Right

20/70

0.4

4.0

59

346

2.0

28

332

Low contrast

The high-contrast grating was presented to the dominant eye. The contrast and spatial frequency of the gratings were adjusted individually to maximize the contrast difference while maintaining an adequate predominance duration. Psychophysical data for average perceptual duration (M), predominance—the percentage of time that the subject reported perceiving each of the two stimuli (%)—and number of occurrences (S) are reported for perceptions of high-contrast and low-contrast gratings. Predominance times do not total 100%; subjects perceived a piecemeal mixture of the stimuli during the remaining time.

Behavioral results In the rivalry experiment, subjects experienced vigorous perceptual alternations between the horizontal high-contrast and vertical lowcontrast gratings. For each subject, the perceptual durations of both stimuli varied randomly from 2 to 15 s and were distributed according to a gamma-shaped function (Supplementary Fig. 2), as is classically found in rivalry studies2. In accordance with such findings2, the highcontrast grating—which is perceptually more salient—was perceived for significantly longer than the low-contrast grating, with some variability among subjects (Table 1). Across all subjects, the highcontrast stimulus was perceived for 5.1 ± 0.09 s (mean ± s.e.m.) compared to 3.1 ± 0.09 s for the low-contrast stimulus (P r 0.001; Supplementary Fig. 2). On average, subjects reported about 160 perceptual switches between the gratings; piecemeal perception occurred 3–34% of the time (Table 1). fMRI results: binocular rivalry fMRI signals in the LGN and V1 fluctuated while subjects perceived the rivalrous grating stimuli. The amplitude of the signals increased when

Figure 2 fMRI signals during binocular rivalry and physical stimulus alternations in the LGN and V1 (group analysis). (a,b) Data from (a) the LGN and (b) V1 of five subjects were combined across left and right hemispheres. Neural activity was averaged across all occurrences of perceptual switches from the low-contrast to the high-contrast grating (black curve) and across those from the high-contrast to the low-contrast grating (gray curve). The responses were time-locked to each subject’s manual response, as indicated by the black vertical line at time point 0, and are shown within a relative time window of –4 to +9 s. All events were normalized, so that responses at time point 0 started at a value of 0% signal change. The vertical bar on each curve indicates one standard error of the mean. Asterisks indicate significant differences between data points of the two curves (one-tailed t-test, *P o 0.05; **P o 0.01; ***P o 0.001). Left, results from rivalry scans. Right, results from physical stimulus alternation scans. Neural activity increased when subjects perceived the high-contrast stimulus and decreased when they perceived the low-contrast stimulus during rivalry conditions. A similar response pattern was found when subjects viewed physical alternations of the same gratings.


[

NUMBER 11

[

NOVEMBER 2005

subjects reported perceiving the high-contrast grating and decreased when they reported perceiving the low-contrast stimulus. These signal modulations can be seen in the raw time-series of the fMRI signals of single subjects. For example, shortly after subject S1 reported a perceptual switch from the low-contrast to the high-contrast grating, there was a sharp increase in the fMRI signal of the LGN (periods shaded in green, Fig. 1c). When the subject’s perceptual experience changed to the low-contrast grating (shown shaded in red), the fMRI signals tended to decrease. Periods of piecemeal perception are not colored and were rare for this subject. To analyze the fMRI time series obtained in the rivalry experiment in relation to subjects’ behavioral responses, an event-related analysis was performed separately for the LGN and V1 of each subject. Mean fMRI signals were derived by averaging the fMRI time series across all events of a reported switch to the high-contrast grating and, separately, across all events of a reported switch to the low-contrast grating. The events were time-locked to the subjects’ manual responses and spanned a period of 4 s before, and 9 s, after each response. These mean signals were then averaged across subjects and are presented as group data (n ¼ 5) for the LGN and for V1 (Fig. 2). Although both gratings were constantly present during rivalry, the amplitude of the fMRI signals in

a

LGN Alternation

Rivalry

**** * ** ** ** ** fMRI signal (%)

left LGN and V1 (Supplementary Fig. 1). The locations of the functional LGN activations were consistent across subjects and across experiments and were in close correspondence to the anatomical locations of the LGN. The activated LGN volume, averaged over all subjects and all experiments, was 190 mm3, similar to those observed in previous studies15,17,18. Activations in area V1 were identified based on anatomical or retinotopic mapping criteria.

** ** ** ** * *****

0.1 0 –0.1 0

5 10 0 Time from percept change (s)

5

10

low-to-high contrast high-to-low contrast

b

V1

Rivalry

Alternation

** *** *** *** fMRI signal (%)


Subject

High contrast

*** *** *** *** *

0.2

0

–0.2 0

5 10 0 Time from percept change (s)

5

10

1597

LGN Rivalry

V1 Alternation

****

******

0 –0.3

***

0.2

0.2

0

0 –0.2

–0.2

*****

0.2

*******

0 S3 –0.2

0.1

S4

Alternation

******

0.4

***** S2

Rivalry

******** 0.8

0.4 S1 0.2 0 –0.2

fMRI signal (%)

Figure 3 fMRI signals during binocular rivalry and physical stimulus alternations in the LGN and V1 (single subjects). Mean fMRI time series obtained while subjects (S1–S5) perceived the high-contrast grating (black) or low-contrast grating (gray) during binocular rivalry or physical stimulus alternations in the LGN and V1. For each subject, the fMRI signal increased in the LGN, and similarly in V1, after transitions to the high-contrast stimulus and decreased after transitions to the low-contrast stimulus. In the rivalry condition, differences between the high- and low-contrast fMRI time series were statistically significant for each individual subject, in both the LGN and V1, for at least the data point at the peak value of the curves. Each asterisk indicates a significant difference (one-tailed t-test; *P o 0.05) between a single point on the high-contrast time series and its counterpart on the lowcontrast time series. Within subjects, the shape of the curves is markedly similar in the LGN and V1. Time series for physical alternations scans show a similar pattern as compared to those from the rivalry scans. Error bars, s.e.m. at the data point with the most significant response difference between the green and red curve in each panel. Other conventions are as in Figure 2.

***

**

*

*

*

****

0.2 0 –0.2

******

** *

0.1 0 –0.1

0

**

***

both the LGN and V1 increased shortly after switches to the percept of the high-contrast grating (black lines) and decreased when the percept changed to the low-contrast grating (gray lines). The peak or trough of the hemodynamic signals occurred 3–6 s after the perceptual switches. Activity related to the percept of the high-contrast grating was significantly different from that related to the percept of the lowcontrast grating for five data points in the LGN panel (Fig. 2a) and for four data points in the V1 panel (Fig. 2b; one-tailed t-test, P r 0.05 or below). Despite individual differences between subjects, the basic response patterns associated with the perceptual reports of high- and low-contrast gratings were present for each subject; moreover, for each subject, the two response patterns were significantly different from one another, at least at the point of peak response (one-tailed t-test, P r 0.05, Fig. 3). The averaged fMRI activity associated with perceptual switches between high- and low-contrast stimuli showed a very similar pattern in the LGN and V1, as is apparent from both the group analysis (Fig. 2) and the single-subject analysis (Fig. 3). For individual subjects, there was a strong correlation between the amplitudes of the fMRI signal in the LGN and V1 (r ¼ 0.92, P r 0.03, Supplementary Fig. 3). In agreement with previous studies15,17,18, the amplitude of the signal was smaller in the LGN than in V1. Notably, the fluctuations in the V1 signal—reflecting the subject’s perceptual experiences—in the rivalry condition confirm previous findings9,10. Our findings of similar fluctuations in the LGN signal extend these studies by demonstrating that the LGN is the first visual processing stage at which neural correlates of binocular rivalry can be observed. To further examine the correlation between perception and the fMRI signal, we investigated whether the perceptual duration of each stimulus predominance period (which varied among subjects from 2 to 15 s) were reflected in the fMRI signals obtained on single trials. The perceptual events were sorted into four time categories (2–3 s, 3–5 s, 5–7 s and 4 7s), and the fMRI signals were averaged separately for each category. The mean fMRI time series for the group of subjects was plotted as a function of the perceptual duration of the stimulus in the LGN (Fig. 4a). Because subjects rarely experienced the low-contrast stimulus for longer than 7 s, the fMRI signal for this time category is shown only for the high-contrast stimulus. It is evident that as the duration of the percept increased, the amplitude and dispersion of the fMRI signal increased. The averaged time series of the fMRI signal was fit to a Gaussian; in both the LGN and V1, the area under this curve was linearly correlated with perceptual duration (r ¼ 0.98, P o 0.0001, Fig. 4b). This tight coupling between perceptual duration and the magnitude of the

Figure 4 Effect of perceptual duration on fMRI signals. (a) fMRI time series averaged across subjects and time-locked to the subjects’ manual responses are shown for the LGN as a function of perceptual duration. Dashed gray lines, 2–3 s; solid black lines, 3–5 s; solid gray lines, 5–7 s; dotted black line, 47 s (only shown for transitions to the high-contrast grating). The amplitude and duration of fMRI signals increased with increasing duration of the percept. (b) The time series data shown in a were fit to a Gaussian function. The area under these fitted curves was linearly correlated with the perceptual duration in the LGN (r ¼ 0.98, P r 0.02). A similar correlation was observed with fMRI responses measured during stimulus alternations in the LGN and during both conditions in area V1. The abscissa of each dot corresponds to the average duration of trials in the corresponding perceptual duration category. Positive values indicate perceptual durations of the highcontrast grating; negative values indicate those of the low-contrast grating.

1598

VOLUME 8

–0.1 0.1

S5

** 0.1 0 –0.1

0 –0.1 0

5

10

0

5

10

0

5

10

0

5

10

Time from percept change (s)

fMRI signal, in both the LGN and V1, suggests that both these structures form a neural circuit that is closely linked to visual awareness during binocular rivalry. fMRI results: physical stimulus alternations If the signal fluctuations measured during the binocular rivalry experiment do indeed reflect the responses of the neuronal population underlying the rivalrous perception of the high- and low-contrast stimuli, then mimicking these rivalrous perceptions by using physical alternations of the same stimuli should yield similar results. As expected from the contrast response functions of the LGN and V1

0.4 0.2

b

2–3 s 3–5 s 5–7 s >7s

Area under Gaussian fit

a fMRI signal (%)


ARTICLES

0 –0.2 –0.4

0 5 Time from percept change (s)

[

NUMBER 11

[

10

2

LGN rivalry LGN alternation V1 rivalry V1 alternation

0

–2 –10 0 10 Perceptual duration (s)


ARTICLES

LGN V1

MRiv / MPhys

1.5

1.0

0.5


0

S1

S2

S3

S4

S5

Figure 5 Comparison of fMRI signals during binocular rivalry and physical stimulus alternations in the LGN and V1. For each subject, the difference in signal amplitudes evoked by the high- and low-contrast gratings during rivalry or physical alternations was computed in both LGN and V1. The ratio of the signal differences obtained during rivalry to those obtained during physical alternations is plotted for each subject and area. The fractions of fMRI signals evoked during rivalry and physical alternations were similar in the LGN (dark bar) and in V1 (light bar). Vertical bars, s.e.m.

(refs. 14–16), fMRI signals increased when the high-contrast grating was presented and decreased when the low-contrast grating was shown. This pattern of response, markedly similar to that obtained during rivalry, was observed in the raw time-series (shown for the LGN in Fig. 1d), the group data (shown for the LGN in Fig. 2a and for V1 in Fig. 2b) and the single subject data (shown for the LGN and V1 in Fig. 3). Previous studies have quantitatively compared the magnitude of the signal modulations obtained during rivalry with those obtained during physical stimulus alternations9,10,19. It has been reasoned that if the difference in signal amplitudes evoked by the two physical stimuli is similar to that evoked by the rivalrous stimuli, the invisible stimulus in the rivalry condition must be completely suppressed. We computed a ‘suppression index’ as follows: for each subject, we first determined the difference in signal amplitudes evoked by the high- and low-contrast stimuli, separately for the rivalrous condition and the physical alternation condition; we then calculated the ratio of the difference in the rivalry condition to that in the physical alternation condition (Fig. 5). An index value (that is, ratio) of 1 indicates equal differences in signal magnitude in the two conditions and can be interpreted as complete suppression of the invisible stimulus during rivalry. Index values between 0 and 1 indicate a smaller amplitude difference during rivalry than during physical alternation and may be interpreted as partial suppression. In the LGN, the level of suppression varied among our subjects as indicated by index values ranging from 0.5 to 1.3; for each subject we observed similar levels of suppression in the LGN and V1 (Supplementary Table 1). Three of the five subjects (S1, S2 and S5) had index values within one s.e.m. of 1, indicating a complete suppression of the competing input during rivalry at the level of the LGN and V1. The other two subjects (S3 and S4) had smaller index values, suggesting only partial suppression. Notably, subjects S1, S2 and S5 reported piecemeal perception rarely (for, respectively, only 3%, 6% and 13% of the time), whereas subjects S3 and S4 reported it more frequently (for 34% and 17% of the time, respectively). Less complete suppression might yield more frequent piecemeal perception, and it is possible that suboptimal viewing conditions and less stable percepts during rivalry contributed to the weaker signal amplitudes (and hence lower suppression indices) in these subjects. However, given the small number of subjects that were tested in this study, more evidence will be needed to support such an idea. Overall, the amount of piecemeal perception was loosely correlated with the suppression index (r ¼ 0.78, P ¼ 0.11).


[

NUMBER 11

[

NOVEMBER 2005

DISCUSSION By demonstrating systematic fluctuations in the fMRI signals associated with the subjects’ perceptual experiences during binocular rivalry, we showed that, in humans, neural activity correlates with visual awareness as early as in the LGN. When superimposed orthogonal, contrast-modulated gratings were viewed dichoptically, fMRI signals in the LGN and V1 increased when subjects reported perceiving the high-contrast grating and decreased when subjects reported perceiving the low-contrast grating. The signal fluctuations observed during rivalry were similar to those evoked by physical alternations of the same monocular stimuli. Across subjects, the signal amplitudes evoked during rivalry were between 50% and 130% of those evoked by the physical alternations; further, the signal ratios were similar in the LGN and V1 for each subject. Because the input to the monocular LGN layers was unchanged during the perceptual oscillations during binocular rivalry, the modulation in LGN activity must be attributed to interactions within the nucleus or to modulatory inputs from other brain regions, such as feedback from V1. Notably, in both the LGN and V1, the magnitude and dispersion of fMRI signals evoked during rivalry were correlated with the duration of the subjects’ perceptual experience, suggesting that neural activity at the earliest stages of visual processing reflects both the content and the duration of the percept and is therefore closely linked to visual awareness during binocular rivalry. Previous neuroimaging studies of binocular rivalry have found correlations between the subjects’ perceptual experiences and neural activity in V1 (refs. 9,20,21), including activity in the monocular representation of the blind spot10. Our results confirm these findings by demonstrating similar correlations between the fMRI signals in area V1 and subjects’ perceptual states; further, we extend these findings by demonstrating that neural correlates related to perceptual experiences during binocular rivalry exist even earlier, at a subcortical processing stage—the LGN of the thalamus. Notably, the latter finding provides physiological evidence in support of accounts of interocular competition that assume inhibitory interactions between monocular channels before binocular convergence1,8,12. Advocates of these accounts have considered the LGN as a possible site at which the invisible stimulus is suppressed during binocular rivalry. Neurons in the LGN are exclusively monocular, with inputs from each eye segregated into separate layers. These adjacent laminae form an ideal substrate for inhibitory interactions between the two eyes; such an interaction would allow the signal from one eye to be selectively suppressed. Binocular interactions, predominantly inhibitory ones, have been widely reported in the LGN of both the monkey22–24 and the cat LGN25–30 and might provide a neural substrate for producing rivalry. These inhibitory interactions may be mediated by several anatomical pathways11, including interneurons extending between LGN layers, corticogeniculate feedback from striate cortex (which comprises about 30% of the input to the LGN) and modulatory input from the TRN (which provides another 30% of the modulatory LGN input). One possibility is that feedback from the binocular neurons in layer 6 of V1 (refs. 31,32) to the monocular LGN layers could provide a descending control signal, indicating whether stimuli have been binocularly fused and regulating the strength of the inhibitory network8. The importance of feedback from V1 in controlling the observed LGN activity cannot be overemphasized. With the current temporal resolution of fMRI, it is not possible to determine whether the LGN controls V1 activity or merely inherits, through feedback, the binocular resolution (that is, the complete suppression of the input from one eye) that might take place in V1 or a higher cortical area. Another possibility is that the TRN (which receives inputs from V1, several extrastriate areas and the pulvinar), may serve as a node where

1599


ARTICLES several cortical areas and thalamic nuclei of the visual system can interact so as to exert additional control on the LGN by modulating thalamocortical transmissions to LGN neurons via inhibitory connections33. It should be noted that these possibilities are not mutually exclusive. To summarize, therefore, on the basis of its anatomy and the organization of its retinal and cortical feedback input, the LGN seems to be in an ideal position to play an important functional role in binocular rivalry—as our present findings suggest. The present and previous findings, that neural correlates of binocular rivalry exist at the earliest stages of visual processing9,10,20,21, contradict results from single-cell physiology studies. In monkeys trained to report perceptual switches during binocular rivalry, the percentage of neurons whose firing rates correlate with the monkeys’ perceptual experiences progressively increases across a hierarchy of cortical visual areas5–7. When their preferred stimulus is perceived during binocular rivalry, the vast majority of neurons in higher-order visual areas show increased activity; in contrast, only a small percentage of (almost exclusively binocular) neurons in early visual cortex do so. Most notably, although an early study reported the existence of a neural correlate of rivalry in the LGN of anesthetized cats30, later studies with anesthetized cats25 and awake monkeys13 were unable to confirm that the LGN was indeed involved in binocular rivalry. The apparent discrepancies between single-cell recording and functional brainimaging studies have been discussed elsewhere9—in terms of interspecies differences, eye movement confounds and differences between the blood oxygen level–dependent (BOLD) signal and single-unit activity—and will not be repeated here. In our view, it is possible that measures of neural activity at the population level (such as in fMRI), rather than at the single-cell level, may be better suited for uncovering large-scale modulatory activity. Small modulatory effects that cannot be reliably found by measuring neural activity at the singleor multi-unit level may be revealed when summed across large populations of neurons. Such a notion is supported by the finding that the BOLD signal correlates better with local field potentials, which reflect the synaptic input to an area, than with single- or multi-unit activity34. For example, modulatory inputs may have little effects on the spiking rate of single units but will still evoke strong responses in the BOLD signal. Thus, the discrepancies between previous electrophysiological and fMRI studies of binocular rivalry may be explained by sub-threshold modulations that are not reflected in the spiking output of neurons. Owing to the spatial resolution limits of our fMRI technique, we were not able to image the individual layers of the LGN; however, comparing the fMRI signal in the LGN measured during rivalry with that measured during physical alternations of the same monocular stimuli may provide a measure of the degree of suppression among the layers. It has been reasoned that if binocular rivalry were fully resolved, one would predict similar magnitudes of signal fluctuation for perceived changes during rivalry as for physical stimulus changes: this would indicate that the input from the invisible stimulus was completely suppressed9,10,19. In the LGN, this would be instantiated as a suppression of activity in the eye-specific layers. Previous studies have reported that in V1, signal amplitude measured during rivalry is 50–85% of that measured during physical alternation9, suggesting a partial suppression of the competing input; alternatively, in monocular V1 neurons, equal responses have been measured during rivalry and physical alternation10, suggesting a complete suppression. Our results confirm both these findings: in three subjects, suppression was essentially complete, as evidenced by equal signal amplitudes during the rivalry and physical alternation conditions; by contrast, in the other two subjects, signal amplitudes were smaller during rivalry than

during physical alternation, suggesting that only partial suppression occurred. Notably, both these last two subjects perceived a high proportion of piecemeal blends, which may indicate suboptimal viewing conditions and percepts that were, overall, less stable during rivalry. It is also possible that individual differences in response criteria contributed to this variability among subjects. Such factors may have led to weaker signals during the rivalry condition, as compared to the unambiguous physical alternation condition, and may account for the differences between individual subjects that were found here and in previous studies9. While interpreting our findings, we need to consider alternative possibilities for how signal fluctuations in the LGN and V1 might be related to subjects’ perceptual experiences. As neural activity in the LGN is considerably modulated by visual attention17, one possibility is that subjects paid more attention to the high- than to the low-contrast stimulus, and that the observed signal fluctuations were therefore caused by attentional switches. This interpretation is not satisfying, however, because the attentional demands of the task did not differ between the stimuli. If anything, the lower-contrast grating demanded more volitional attention because its predominance duration tended to be shorter. Switches to the high-contrast stimulus might initially capture attention, but the activity we observed was sustained over several seconds and was closely linked to the perceptual duration reported by the subjects. Another possibility is that the neural activity was confounded by systematic differences in eye movement patterns as subjects viewed the vertical or horizontal gratings. We were not able to measure eye movements in the MRI scanner because subjects wore filter glasses during the experiments and these obscured their eyes. However, our control experiment outside the scanner indicated that there were no differences in eye movement patterns for the two grating stimuli. Thus it seems unlikely that our findings could be sufficiently explained in terms of attentional modulation of neural activity or eye movement confounds. Our study showed that neural activity that is closely linked to the duration and content of conscious perception is not confined to cortical processing as previously thought35,36, but occurs even at the thalamic level. Much remains to be learned about the complex thalamic circuitry that subserves conscious perception in the LGN. From our study, we conclude that the LGN seems to be the first stage in visual information processing at which the neural correlates of visual awareness during binocular rivalry can be found. Our findings further suggest the need to revise the traditional view of the LGN as a mere gateway to the visual cortex. The LGN may, in fact, participate in a network of widely distributed cortical and subcortical brain systems, serving as an early gatekeeper of visual awareness.

1600

VOLUME 8

METHODS Subjects, visual stimuli and tasks. Five healthy subjects (3 male; 22–36 years old; normal or corrected-to-normal visual acuity) gave written informed consent for participation in the study, which was approved by the Institutional Review Panel of Princeton University. All subjects received training before the scanning sessions to ensure that subjects were able to report their perceptual experiences during binocular rivalry. The rivalrous stimulus consisted of a pair of superimposed horizontal and vertical sinusoidal gratings (0.4–0.5 cpd) that were presented within an annulus (1.8–5.41) centered at the fixation point. When these were viewed through a red filter glass by one eye and through a green filter glass by the other eye, only the horizontal grating was visible to the dominant eye and the vertical grating to the other eye (Fig. 1a). The two gratings also differed in color and luminance contrast. The vertical red grating was presented at 14–20% contrast and at a mean luminance of 0.5 cd m–2 when viewed through the matching filter. The horizontal green grating was presented at 70% contrast and at a mean

[

NUMBER 11

[



ARTICLES luminance of 0.6 cd m–2. Each grating reversed contrast at a frequency of 1.1– 1.4 Hz to prevent adaptation. Stimulus contrasts and reversal rates were individually adjusted for each subject so as to maximize the perceptual duration of the weaker stimulus (Table 1). Subjects were instructed to maintain fixation and to report their perceptual experience by pressing one of three buttons corresponding to the vertical red grating, the horizontal green grating or phases of an unstable piecemeal blend of the two. The perceptual experience during rivalry was simulated in a physical stimulus alternation condition through the presentation of the green or the red grating to one eye, and a uniform field to the other eye (Fig. 1b). The presentation times of alternating stimuli were identical to the perceptual durations of the corresponding grating that the same subject reported during rivalry. To mimic the smooth transitions that were perceived during rivalry, stimuli were sinusoidally faded into each other over a 1-s period. Contrasts, colors and mean luminances were identical to those in the rivalry condition. As in the rivalry condition, subjects maintained fixation and indicated which grating was viewed by pressing buttons. Neural representations of the peripheral annulus in the LGN and area V1 were localized by presenting a flickering checkerboard stimulus (contrast reversing at 8 Hz) in blocks of 16 s, alternating between the right and left hemifield (Supplementary Fig. 1)17. Subjects were instructed to maintain fixation during these presentations. Data acquisition and analysis. Subjects participated in one or two scanning sessions for the rivalry experiment and an additional session for the physical alternation experiment. Data were acquired with a 3-T head scanner (Allegra, Siemens) using a standard head coil. Functional images were taken with a gradient echo, echoplanar sequence (TR ¼ 1 s for rivalry and physical alternation scans, and 2 s for localizer scans; flip angle ¼ 641 for rivalry and physical alternation scans, and 901 for localizer scans; TE ¼ 30 ms; 64 64 matrix). Sixteen axial slices (3-mm thickness, in-plane resolution 3 3 mm2) covering the thalamus and visual cortex were acquired in six series of 272 volumes each for the rivalry and physical alternation scans and six series of 128 volumes for the localizer scans. A high-resolution anatomical scan of the whole brain (MPRAGE sequence; TR ¼ 2.5 s; TE ¼ 4.3 ms; flip angle ¼ 81; 256 256 matrix; 1 mm3 resolution) was acquired in the same session to align the functional images. Data were analyzed using AFNI (http://afni.nimh.nih.gov/afni). The functional images were motion-corrected to the image acquired closest in time to the anatomical scan and normalized to percent signal change by dividing the time series by its mean intensity. Regions of interest (ROIs) in the LGN and V1 were defined based on activations obtained in the localizer scans. A squarewave function reflecting the contrast between left and right visual-hemifield stimulations was convolved with a gamma-variate function37 to generate an idealized response function; this function was used as a regressor of interest in a multiple regression in the framework of the general linear model38. Additional regressors were included to account for variance that is due to baseline shifts between time series, linear drifts within time series and head motion. Statistical maps were thresholded at P o 0.01 and overlaid on anatomical scans. LGN activations were identified from contiguous voxels in the anatomical location of the LGN15,17,18. V1 activations were identified on the basis of their location in the calcarine sulcus and on retinotopic mapping using standard procedures in three subjects18,39. Data from the LGN and V1 were combined across hemispheres. Subjects who did not show bilateral LGN activation were excluded from the study. Event-related fMRI time-series analyses were carried out on all activated voxels within a given ROI. Linear and quadratic signal trends were removed and the time series were low-pass filtered through a convolution with a threepoint-width Hamming window. Mean time series of fMRI signals were calculated separately for switches from the high- to the low-contrast grating and vice versa, by averaging across all events during which subjects reported a perceptual switch within a restricted window of 4 s to +9 s relative to the manual response. All events were normalized, so that responses at time point zero started at a value of 0% signal change. Perceptual durations of less than 2 s were excluded from this analysis because they elicited fMRI signals too weak to be distinguished from noise. Differences in mean fMRI signals during switches from low to high contrast and from high to low contrast were tested for


[

NUMBER 11

[

NOVEMBER 2005

significance, with a two-sample, one-tailed t-test, at each of nine data points following the behavioral response. The subjects’ perceptual experiences and the resulting fMRI activity were analyzed by grouping all single trials into one of four categories of perceptual duration (2–3 s; 3–5 s; 5–7 s; 4 7 s). The mean fMRI signals were fit via nonlinear least-squares to a Gaussian, ðtmÞ2 =2 s2 f ðtÞ ¼ ae , and parameterized as the area under the curve, pffiffiffiffiffi A ¼ as 2p. A suppression index was defined by first determining the difference between the response amplitudes obtained for the two rivalrous stimuli and for the two physical alternation stimuli, and then calculating the ratio between the differences. An index value of 1 indicates fluctuations of equal magnitude during the rivalry and physical alternation conditions and could be interpreted as a complete suppression of the invisible stimulus during rivalry. Index values between 0 and 1 indicate that smaller amplitudes occur during rivalry than during physical alternations and could be interpreted as a partial suppression. Eye movement control. Given that the two rivalrous stimuli were gratings perpendicular to each other, we considered the possibility that the two stimuli elicited different patterns of eye movements, thereby confounding the results obtained in the LGN and V1. Because the filter glasses obscured the subjects’ eyes during scanning, it was not possible to monitor eye movements directly during these experiments. Instead, we carried out a behavioral control experiment outside the scanner by monitoring eye movements in all five subjects using an infrared eye-tracking device (ASL Model 5000 control unit and standard Model 504 remote optics, Applied Science Laboratories) while they viewed 20 alternating 8-s blocks of the vertical and horizontal grating stimuli without wearing filter glasses. We observed no significant differences in the mean or standard deviation of eye position, or mean eye velocity in either the vertical or horizontal direction (paired two-tailed t-test, P 4 0.05), indicating that there were no obvious differences in fixation or eye movements when subjects viewed the two gratings. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank K. Weiner for help with manuscript preparation. This study was supported by NIH grants R01MH-64043, P50MH-62196 and T32 MH065214. K.W. was also supported by the German National Academic Foundation and the German Academic Exchange Service. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Blake, R. A neural theory of binocular rivalry. Psychol. Rev. 96, 145–167 (1989). 2. Levelt, W.J. Binocular brightness averaging and contour information. Br. J. Psychol. 56, 1–13 (1965). 3. Mueller, T.J. & Blake, R. A fresh look at the temporal dynamics of binocular rivalry. Biol. Cybern. 61, 223–232 (1989). 4. Crick, F. & Koch, C. Consciousness and neuroscience. Cereb. Cortex 8, 97–107 (1998). 5. Sheinberg, D.L. & Logothetis, N.K. The role of temporal cortical areas in perceptual organization. Proc. Natl. Acad. Sci. USA 94, 3408–3413 (1997). 6. Logothetis, N.K. & Schall, J.D. Neuronal correlates of subjective visual perception. Science 245, 761–763 (1989). 7. Leopold, D.A. & Logothetis, N.K. Activity changes in early visual cortex reflect monkeys’ percepts during binocular rivalry. Nature 379, 549–553 (1996). 8. Lehky, S.R. & Blake, R. Organization of binocular pathways: modeling and data related to rivalry. Neural Comput. 3, 44–53 (1991). 9. Polonsky, A., Blake, R., Braun, J. & Heeger, D.J. Neuronal activity in human primary visual cortex correlates with perception during binocular rivalry. Nat. Neurosci. 3, 1153–1159 (2000). 10. Tong, F. & Engel, S.A. Interocular rivalry revealed in the human cortical blind-spot representation. Nature 411, 195–199 (2001). 11. Sherman, S.M. & Guillery, R.W. Exploring the Thalamus (Academic Press, San Diego, 2001). 12. Lehky, S.R. An astable multivibrator model of binocular rivalry. Perception 17, 215–228 (1988). 13. Lehky, S.R. & Maunsell, J.H. No binocular rivalry in the LGN of alert macaque monkeys. Vision Res. 36, 1225–1234 (1996).

1601


ARTICLES 14. Boynton, G.M., Engel, S.A., Glover, G.H. & Heeger, D.J. Linear systems analysis of functional magnetic resonance imaging in human V1. J. Neurosci. 16, 4207–4221 (1996). 15. Kastner, S. et al. Functional imaging of the human lateral geniculate nucleus and pulvinar. J. Neurophysiol. 91, 438–448 (2004). 16. Schneider, K.A. & Kastner, S. Visual responses of the human superior colliculus: a highresolution fMRI study. J. Neurophysiol. 94, 2491–2503 (2005). 17. O’Connor, D.H., Fukui, M.M., Pinsk, M.A. & Kastner, S. Attention modulates responses in the human lateral geniculate nucleus. Nat. Neurosci. 5, 1203–1209 (2002). 18. Schneider, K.A., Richter, M.C. & Kastner, S. Retinotopic organization and functional subdivisions of the human lateral geniculate nucleus: a high-resolution functional magnetic resonance imaging study. J. Neurosci. 24, 8975–8985 (2004). 19. Tong, F., Nakayama, K., Vaughan, J.T. & Kanwisher, N. Binocular rivalry and visual awareness in human extrastriate cortex. Neuron 21, 753–759 (1998). 20. Lee, S.H. & Blake, R. V1 activity is reduced during binocular rivalry. J. Vis. 2, 618–626 (2002). 21. Lee, S.H., Blake, R. & Heeger, D.J. Traveling waves of activity in primary visual cortex during binocular rivalry. Nat. Neurosci. 8, 22–23 (2005). 22. Rodieck, R.W. & Dreher, B. Visual suppression from nondominant eye in the lateral geniculate nucleus: a comparison of cat and monkey. Exp. Brain Res. 35, 465–477 (1979). 23. Marrocco, R.T. & McClurkin, J.W. Binocular interaction in the lateral geniculate nucleus of the monkey. Brain Res. 168, 633–637 (1979). 24. Schroeder, C.E., Tenke, C.E., Arezzo, J.C. & Vaughan, H.G., Jr. Binocularity in the lateral geniculate nucleus of the alert macaque. Brain Res. 521, 303–310 (1990). 25. Sengpiel, F., Blakemore, C. & Harrad, R. Interocular suppression in the primary visual cortex: a possible neural basis of binocular rivalry. Vision Res. 35, 179–195 (1995).

26. Pape, H.C. & Eysel, U.T. Binocular interactions in the lateral geniculate nucleus of the cat: GABAergic inhibition reduced by dominant afferent activity. Exp. Brain Res. 61, 265–271 (1986). 27. Sanderson, K.J., Bishop, P.O. & Darian-Smith, I. The properties of the binocular receptive fields of lateral geniculate neurons. Exp. Brain Res. 13, 178–207 (1971). 28. Schmielau, F. & Singer, W. The role of visual cortex for binocular interactions in the cat lateral geniculate nucleus. Brain Res. 120, 354–361 (1977). 29. Singer, W. Inhibitory binocular interaction in the lateral geniculate body of the cat. Brain Res. 18, 165–170 (1970). 30. Varela, F.J. & Singer, W. Neuronal dynamics in the visual corticothalamic pathway revealed through binocular rivalry. Exp. Brain Res. 66, 10–20 (1987). 31. Livingstone, M.S. & Hubel, D.H. Psychophysical evidence for separate channels for the perception of form, color, movement, and depth. J. Neurosci. 7, 3416–3468 (1987). 32. Lund, J.S. & Boothe, R.G. Interlaminar connections and pyramidal neuron organisation in the visual cortex, area 17, of the Macaque monkey. J. Comp. Neurol. 159, 305–334 (1975). 33. Guillery, R.W., Feig, S.L. & Lozsadi, D.A. Paying attention to the thalamic reticular nucleus. Trends Neurosci. 21, 28–32 (1998). 34. Logothetis, N.K., Guggenberger, H., Peled, S. & Pauls, J. Functional imaging of the monkey brain. Nat. Neurosci. 2, 555–562 (1999). 35. Crick, F. & Koch, C. Are we aware of neural activity in primary visual cortex? Nature 375, 121–123 (1995). 36. Lumer, E.D., Friston, K.J. & Rees, G. Neural correlates of perceptual rivalry in the human brain. Science 280, 1930–1934 (1998). 37. Cohen, M.S. Parametric analysis of fMRI data using linear systems methods. Neuroimage 6, 93–103 (1997). 38. Friston, K.J. et al. Statistical parametric maps in functional imaging: a general linear approach. Hum. Brain Mapp. 2, 189–210 (1995). 39. Sereno, M.I. et al. Borders of multiple visual areas in humans revealed by functional magnetic resonance imaging. Science 268, 889–893 (1995).

1602

VOLUME 8

[

NUMBER 11

[



ARTICLES

Transcriptional and behavioral interaction between 22q11.2 orthologs modulates schizophrenia-related phenotypes in mice Marta Paterlini1,2, Stanislav S Zakharenko3,9, Wen-Sung Lai1,2, Jie Qin4, Hui Zhang5, Jun Mukai1, Koen G C Westphal6, Berend Olivier6, David Sulzer5, Paul Pavlidis4, Steven A Siegelbaum3,7,8, Maria Karayiorgou2 & Joseph A Gogos1,3 Microdeletions of 22q11.2 represent one of the highest known genetic risk factors for schizophrenia. It is likely that more than one gene contributes to the marked risk associated with this locus. Two of the candidate risk genes encode the enzymes proline dehydrogenase (PRODH) and catechol-O-methyltransferase (COMT), which modulate the levels of a putative neuromodulator (L-proline) and the neurotransmitter dopamine, respectively. Mice that model the state of PRODH deficiency observed in humans with schizophrenia show increased neurotransmitter release at glutamatergic synapses as well as deficits in associative learning and response to psychomimetic drugs. Transcriptional profiling and pharmacological manipulations identified a transcriptional and behavioral interaction between the Prodh and Comt genes that is likely to represent a homeostatic response to enhanced dopaminergic signaling in the frontal cortex. This interaction modulates a number of schizophrenia-related phenotypes, providing a framework for understanding the high disease risk associated with this locus, the expression of the phenotype, or both.

An increased frequency of microdeletions at the 22q11.2 locus has been found in individuals with schizophrenia1. Two independent systematic approaches2–4 and several candidate gene studies have identified candidate schizophrenia susceptibility genes from the 22q11.2 region. Both systematic approaches, using different methodologies to analyze almost all the genes in this locus, provided convergent evidence for an involvement of the gene encoding the mitochondrial enzyme proline dehydrogenase (PRODH). A previous study2 provided evidence that a haplotypic variant of the gene is preferentially transmitted in individuals with schizophrenia, a finding recently replicated in a large-scale family study5. The earlier study2 also found that incidences of rare variants of PRODH affecting highly conserved amino acids and causing drastic reduction in enzymatic activity6 are elevated to various degrees in individuals with schizophrenia—a finding since replicated in an independent set of studies4,7. Other genes in the 22q11.2 region, including the ZDHHC8 gene and the gene encoding catechol-Omethyltransferase (COMT), have also been implicated by systematic3,8 and candidate gene approaches9,10, respectively. Taken together, these studies suggest that 22q11.2 microdeletion-associated schizophrenia

may have the characteristics of a contiguous gene syndrome, in which more than one gene contributes to the marked increase in disease risk. Given the limitations of genetic association studies and the restricted knowledge of the functional impact of human genetic variation, we used studies in animal models as a powerful means for examining the function and genetic interactions of candidate schizophrenia susceptibility genes from this locus. RESULTS Establishment of a genetic mouse model Previously11, in the Pro/Re hyperprolinemic mouse strain, we identified a mutation in the mouse ortholog of the human PRODH gene that introduces a premature termination (E453X) and results in a reduction in enzymatic activity. Preliminary analysis indicated that these Prodhknockdown mice have regional neurochemical alterations in the brain accompanied by a deficit in sensorimotor gating, similar to that seen in individuals with schizophrenia and other neuropsychiatric disorders11. To minimize the influence of genetic background, we introduced the E453X mutation into the 129/SvEv strain through backcrossing for ten

1Department

of Physiology and Cellular Biophysics, Columbia University College of Physicians and Surgeons, 701 West 168th Street, New York, New York 10032, USA. Neurogenetics Laboratory, Rockefeller University, 1230 York Avenue, New York, New York 10021, USA. 3Center for Neurobiology and Behavior, Columbia University, 722 West 168th Street, New York, New York 10032, USA. 4Genome Center and Department of Biomedical Informatics, Columbia University College of Physicians and Surgeons, 1150 St. Nicholas Avenue, New York, New York 10032, USA. 5 Departments of Neurology and Psychiatry, Columbia University College of Physicians and Surgeons, 701 West 168th Street, New York, New York 10032, USA. 6Department of Pharmacology, University of Utrecht, Sorbonnelaan 16, 3584 CA Utrecht, The Netherlands. 7Howard Hughes Medical Institute, Columbia University, 722 West 168th Street, New York, New York 10032, USA. 8Department of Pharmacology, Columbia University, 722 West 168th Street, New York, New York 10032, USA. 9Present address: Department of Developmental Neurobiology, St. Jude Children’s Research Hospital, 332 North Lauderdale Street, Memphis, Tennessee 38105, USA. Correspondence should be addressed to M.K. ([email protected]) or J.A.G. ([email protected]). 2Human


1586

VOLUME 8

[

NUMBER 11

[


Prodh deficiency alters glutamatergic transmission Evidence that L-proline is selectively accumulated by a brain-specific, high-affinity transporter localized exclusively in a subset of glutamatergic synapses suggests that L-proline may modulate transmission at glutamate synapses12. High concentrations of exogenous L-proline can activate NMDA and AMPA receptors13, and more physiological concentrations potentiate excitatory transmission at hippocampal synapses between CA3 and CA1 pyramidal neurons14. Nevertheless, there is no evidence that PRODH affects basic glutamate-mediated synaptic transmission. To examine the role of PRODH, we compared the synaptic input-output relation at the well-characterized glutamatergic CA3–CA1 synapses in acute hippocampal slices from Prodh-deficient mice and their wild-type littermates. Recordings of the extracellular field EPSP (fEPSP) showed that the loss of Prodh caused a significant enhancement in synaptic transmission over a wide range of stimulus intensities (Fig. 1a), similar to results previously reported with exogenous application of L-proline14. However, unlike the results observed with acute application of L-proline, where there was no enhancement in presynaptic function, an analysis of paired-pulse stimulation suggested that glutamate release is increased in Prodh-deficient mice. Thus, we found that the paired-pulse ratio (PPR) measured at all interpulse intervals ranging from 20 ms to 1,000 ms was significantly decreased in the mutant mice compared to their wild-type littermates (Fig. 1b). Notably, paired pulses with a 20-ms interval yielded facilitation in wildtype mice but depression in the Prodh-deficient mice. The decrease in PPR is often a hallmark of an increase in the probability of transmitter release during the first pulse, which can lead to depression of release during the second pulse or to an occlusion of the normal increase in the probability of release. It has recently been shown that long-term potentiation (LTP) at CA3–CA1 hippocampal synapses induced by 200-Hz or theta-burst stimulation protocols enhances presynaptic function by increasing the probability of transmitter release15. We therefore explored the effect of Prodh deficiency on the induction of 200-Hz LTP at CA3–CA1 synapses. In wild-type mice, 200-Hz tetanic stimulation enhanced the fEPSP to180.3 ± 14.8% (mean ± s.e.m.; n ¼ 7) of its initial level, when


[

NUMBER 11

[

NOVEMBER 2005

fEPSP2 slope/fEPSP1 slope

b 3 2 1 0 30 10 20 Stimulation (V)

0

c

150

100 20 30 40 Time (min)

1.2 1.0 0.8 0

200 400 600 800 1,000 1,200

d

200

10

wt kd

1.4

Interpulse interval (ms)

200 Hz

0

1.6

40

fEPSP2 slope/fEPSP1 slope

generations. Measurements of serum proline levels confirmed the previously described increase in L-proline levels in mice homozygous for the mutation. Serum levels of L-proline in the homozygous mutant mice range from 300 to 600 mmol liter–1 as compared to less than 100 mmol liter–1 in their wild-type littermates11. These levels are comparable to those observed in some individuals with the 22q11.2 microdeletion and in heterozygous carriers of PRODH rare variants but are well below the levels observed in individuals with hyperprolinemia type I, a rare condition often accompanied by epilepsy and mental retardation4. Therefore, both in terms of the nature of the mutation and the ensuing increase in L-proline levels, homozygous Prodh-knockdown mice represent an accurate animal model not only of a susceptibility gene but also of a susceptibility allele.

a fEPSP slope (Vs–1)

Figure 1 Electrophysiological characterization of Prodh-deficient mice. (a) Mean fEPSP slope measured as a function of stimulation intensity in slices from Prodh-deficient mice (n ¼ 15 slices) and their wild-type littermates (n ¼ 15 slices). (b) Paired-pulse ratio (fEPSP2 slope to fEPSP1 slope) versus the interpulse interval length for slices from Prodh-deficient mice (n ¼ 50 slices) and wild-type littermates (n ¼ 35 slices). (c,d) Mean fEPSP slope (c) and paired-pulse ratio with a 50-ms interpulse interval (d) as a function of time before and after tetanization at 200 Hz (arrow) for slices from Prodh-deficient mice (open squares, n ¼ 11 slices) and wild-type littermates (closed circles, n ¼ 7 slices). All data show mean ± s.e.m. wt, wild-type littermate control mice; kd, homozygous Prodh-knockdown mice.

Normalized fEPSP


ARTICLES

50

60

1.6 1.4 1.2 1.0 0.8 0

10

20 30 40 Time (min)

50

60

measured 40 min after tetanization (Fig. 1c). In contrast, the magnitude of 200-Hz LTP in Prodh-deficient mice was only 139.9 ± 6.6% (n ¼ 11), significantly lower than that in their wild-type littermates (P o 0.05, Kolmogorov-Smirnov test). In support of the view that 200-Hz LTP enhances presynaptic function, we found that the increase in the fEPSP in wild-type mice was accompanied by a significant decrease in the PPR (measured at a 50-ms interpulse interval; Fig. 1d). Thus, 40 min after induction of LTP, the PPR decreased to 90.5 ± 1.4% (n ¼ 7) of its initial value (1.35 ± 0.03, n ¼ 7; P o 0.001, paired t-test). In contrast, induction of 200-Hz LTP caused no change in the PPR in Prodh-deficient mice. Before induction of LTP, the PPR in Prodh-deficient mice was 1.13 ± 0.02 (50-ms interval, n ¼ 11), which was significantly lower than that in their wild-type littermates (P o 0.001, Kolmogorov-Smirnov test). Forty minutes after the induction of LTP in Prodh-deficient mice, the PPR remained at 99.0 ± 2.6% of its initial level (n ¼ 11; P 4 0.05, paired t-test). In summary, Prodh deficiency seems to increase the initial probability of glutamate release at CA3–CA1 synapses that leads to a decrease in the dynamic range of plastic presynaptic modifications at glutamatergic synapses, inhibiting paired-pulse facilitation and LTP. Behavioral analysis Mice treated chronically with NMDA receptor antagonists, such as dizocilpine (MK801) and phencyclidine (PCP), represent one of the most widely used pharmacological models of schizophrenia. NMDA receptor antagonists increase glutamate release16, suggesting that this class of drugs may act primarily via secondary activation of nonNMDA receptor glutamatergic neurotransmission16. It has been shown that persistent glutamatergic dysfunction in these models leads to secondary dysregulation of frontal dopaminergic neurotransmission and hypersensitivity to the locomotor effects of amphetamine17, as well as to performance decrement in some cognitive tasks18–20. We asked whether a similar pattern of deficits develop as a result of Prodh deficiency and the ensuing increase in glutamate release. Response to psychomimetic drugs We evaluated baseline locomotor activity and the locomotor response to two psychomimetic drugs, MK801 (an NMDA receptor antagonist) and D-amphetamine (which causes dopamine release), by monitoring drug-induced locomotor responses. In an open-field test, we detected strong deficits in baseline locomotor activity of Prodh-deficient mice compared to that of their wild-type littermates (Fig. 2a). In the

1587

c

b

Post- to pre-injection activity ratio

450 400 350 300 250 200 150 100 50 0 min

Total distance

**

** *** 5

10

15

* 20

* 25

30

* 1.4

n.s.

*

1.2

*

0.8 0.6 0.4 0.2 0 Vehicle

MK801 MK801 (0.25 mg kg–1) (0.4 mg kg–1)

50 Total distance

500

*

40

* *

30 20

*

10

400 300 200 100 0

* *

Saline

20 25 30 35 40 45

1,200 800 400

–1

–1

Amphetamine (2.5 mg kg ) Amphetamine (7.5 mg kg ) Cued 1.0

*

0.4 0.3 0.2 0.1

Percentage freezing

0.7 0.5

* wt

0.8

kd

0.6 0.4 0.2

0

presence of a difference in baseline locomotor activity, we used the ratio of the activity during the half-hour after injection of a test drug to the activity during the half-hour before injection as an index of a drug’s locomotor activating effects. Consistent with previous reports16, both in wild-type control and mutant mice, the administration of MK801 or D-amphetamine stimulated locomotor activity in a dose-dependent manner, as compared to saline administration which did not stimulate activity (Fig. 2b,c). However, at the higher drug dosage, Prodh-deficient mice seemed to be less responsive to MK801 than their wild-type littermates were (Fig. 2b). Such a response may indicate an effect of the increase in basal glutamate release that occludes any additional effects of MK801 on glutamate efflux, a developmental desensitization of the MK801modulated glutamatergic pathways21, or both. By contrast, Prodh deficiency led to a significant potentiation of D-amphetamine–induced locomotor activity, especially at the dose of 7.5 mg per kg body weight (Fig. 2c). This pattern of enhanced amphetamine-induced locomotor activity is similar to the one described after persistent glutamatergic dysfunction in pharmacological models17 and is reminiscent of the increased susceptibility to the disorganizing effects of D-amphetamine22 observed in individuals with schizophrenia.

20 25 30 35 40 45

20 25 30 35 40 45

Contextual 0.6

20 25 30 35 40 45

1,600

0

0 min 20 25 30 35 40 45

d

*

1.0

Total distance

a


Figure 2 Behavioral characterization of Prodhdeficient mice. (a) Total distance traveled is significantly decreased in Prodh-deficient mice (overall P ¼ 0.0006). The figure shows the time course of the effect of the mutation (E453X) over 30 min, grouped in six 5-min intervals. Results are from an open-field test probing spontaneous exploratory behavior under mildly stressful conditions (novel environment and light). *P o 0.05; **P o 0.01; ***P o 0.001. (b) Locomotoractivating effect of MK801 on Prodh-deficient mice and wild-type littermate control mice. Mice were injected i.p. with vehicle, drug at a low dose (0.25 mg per kg body weight) or drug at a high dose (0.4 mg per kg body weight). *P o 0.05. (c) D-Amphetamine–induced activity in Prodhdeficient mice and wild-type littermate control mice. y-axis represents ratios of post-treatment to pre-treatment values of the effect of the mutation over 5-min intervals (counting from the time of D-amphetamine administration, 15 min before the animals were placed in the open-field apparatus). Mice were injected i.p. with vehicle or with drug at 2.5 or 7.5 mg per kg body weight. Far right, the total distance traveled (absolute values) by Prodh-deficient mice and wild-type littermate control mice over the 30-min period, before (top) and after (bottom) administration of 7.5 mg of D-amphetamine per kg body weight. *P o 0.05. (d) Associative learning and memory in Prodhdeficient mice. Results from both contextual and cued tests are shown. All data show mean ± s.e.m. wt, wild-type littermate control mice; kd, homozygous Prodh-knockdown mice. *P o 0.05.

Percentage freezing


ARTICLES

0

Pre-CS

CS

hippocampus23—tone-dependent freezing 48 h after training was significantly reduced in the Prodh-deficient mice compared to wildtype littermate control mice (P o 0.05; Fig. 2d). In contrast, no significant differences (P 4 0.05) were found between groups in the absence of the tone. We also used the T-maze delayed-alternation task to examine whether low Prodh activity was associated with changes in spatial working memory, which is thought to be modulated by the frontal regions of the mouse neocortex24. Prodh-deficient mice learned the 5-s–delay T-maze task during five consecutive training days and performed as well as their wild-type littermates. Mutants also did not show any difference in their T-maze working memory test (data not shown; also see below). Because the pattern of the behavioral profile of the mutant mice is probably determined by the primary synaptic deficit as well as by the emergence of brain region–specific compensations, normal working memory performance may indicate the emergence of cortical compensations in the Prodh-deficient mice.

Cognitive tasks First, we used the Pavlovian conditioned-fear protocol to assess associative learning and memory in Prodh-deficient mice. In the contextual fear-conditioning test—which is both amygdala and hippocampus dependent—Prodh-deficient mice froze significantly less than did wild-type mice when, 24 h after training, they were returned to the context in which they had received the shock (P o 0.05; Fig. 2d). In the cued version of the test—which requires the amygdala but not the

Expression profiling in the frontal cortex The effect of a mutation on synaptic transmission and associated pathways, as well as the development of any compensatory changes, can be reflected at the level of transcriptome25. We reasoned that transcriptional profiling in the brain of a mouse model for a schizophrenia susceptibility allele might provide an unbiased evaluation of the transcriptional programs altered by the disruption of the gene; such alterations would reflect either causal downstream effects of the mutation or reactive changes. We focused on the frontal cortex, on the basis of clues provided by our initial behavioral analysis as well as of numerous studies suggesting functional and structural pathology of this brain region in schizophrenia.

1588

VOLUME 8

[

NUMBER 11

[


ARTICLES

a

b

~2.34 mm from bregma 0.25 mm

c

wt

wt 6 5 4

~1.42 mm from bregma 0.5 mm

3

kd

2

kd

Gene Ontology (GO) term

GO id

P-value

Eukaryotic 48S initiation complex Eukaryotic 43S preinitiation complex Regulation of translational initiation Protein processing

GO:0016283 GO:0016282 GO:0006446 GO:0016485

<10–6 <10–6 <10–6 <10–6

NADH dehydrogenase activity Hydrogen ion transporter activity Inorganic cation transporter activity Oxidative phosphorylation Tricarboxylic acid cycle

GO:0003954 GO:0015078 GO:0015077 GO:0006119 GO:0006099

<10–6 0.00000143 0.00000457 0.00003466 0.00007285

Neurotransmitter secretion Neurotransmitter levels Synaptic vesicle transport

GO:0007269 GO:0001505 GO:0016181

0.00000355 0.00000681 0.00006976


1 0 wt

d

kd wt kd 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1449183_at 1429906_at 1454886_x_at 1444973_at 1452423_at 1451830_a_at 1445537_at 1432415_at 1421136_at 1433884_at 1431812_a_at 1434582_at 1455766_at 1442707_at 1448807_at 1417746_at 1418436_at 1417702_a_at 1415766_at 1447863_s_at 1440834_at 1420418_at 1444455_at 1416561_at 1456406_at 1419246_s_at

2.002E-5 6.228E-4 0.001203 0.001546 0.00338 0.003486 0.007766 0.009047 2.211E-4 7.403E-4 7.518E-4 0.001122 0.001551 0.00155 0.002005 0.002071 0.004798 0.004904 0.005717 0.005933 0.006112 0.006817 0.007048 0.007419 0.007637 0.008669

COMT catechol-O-methyltransferase NRXN3 neurexin III TRIM9 tripartite motif protein 9 KCNMA1 potassium large conductance calcium-activated channel, subfamily M, alpha member 1 PCLO piccolo (presynaptic cytomatrix protein) SPNB2 spectrin beta 2 SRGAP2 SLIT-ROBO Rho GTPase activating protein 2 RAB3C RAB3C, member RAS oncogene family EDN3 endothelin 3 SYT1 synaptotagmin 1 SLC6A9 solute carrier family 6 (neurotransmitter transporter, glycine), member 9 D14ERTD171E DNA segment, Chr 14, ERATO Doi 171, expressed -1.51948043 GABRA1 gamma-aminobutyric acid (GABA-A) receptro, subunit alpha 1 CAMK2A calcium/calmodulin-dependent protein kinase II alpha HRH3 histamine receptor H3 CPLX1 complexin 1 STX7 syntaxin 7 HNMT histamine N-methyltransferase SEC22L1 SEC22 vesicle trafficking protein-like1 (S. cerevisiae) NR4A2nuclear receptor subfamily 4, group A, member 2 SLC5A10 solute carrier family 5 (sodium/glucose contransporter), member 10 SYT2 synaptotagmin 2 CACNA1A calcium channel, voltage-dependent, P/Q type, alpha 1A subunit GAD1 glutamic acid decarboxylase 1 NRXN2 neurexin II RAB14 RAB14, member RAS oncogene family

Figure 3 RNA expression profiling in the frontal cortex of Prodh-deficient mice. (a) Representative Nissl-stained coronal section, from Prodh-deficient mice and wild-type littermate control mice, of frontal cortex localized to approximately 2.34 mm rostral to bregma and medial prefrontal cortex localized to approximately 1.42 mm rostral to bregma. (b) TUNEL assay in the frontal cortex of postnatal day 8 (P8) Prodh-deficient mice (E453X) and wild-type littermate control mice. Graphs represent mean ± s.e.m. of positive cells per section. (c) Top-scoring Gene Ontology (GO) terms listed with the corresponding P-value and GO identification numbers. This approach provides a statistical measure of the significance of the co-occurrence of genes with related function. (d) Matrix visualizing Affymetrix microarray data of the top-scoring, differentially expressed ‘neurotransmitter release and regulation’ genes. Genes are ranked in order of increasing template-match P-value and are listed separately as up- or down-regulated. Each gene is visualized as a row of colored squares, with one square for each sample. The color indicates the relative expression of the gene, with lighter colors indicating lower levels of expression. Annotations for each gene and the t-test P-value are shown at the right of the figure. The Comt gene shows an upregulation of about 70% (P ¼ 2.0 10–5). wt, wild-type littermate control mice; kd, homozygous Prodh-knockdown mice.

In control experiments in our mouse model, histological analysis (Fig. 3a), cell counting (data not shown) and cell death assessment in the developing cortex (Fig. 3b) did not reveal any gross anatomical abnormalities. We used ten independent RNA samples from each genotype to compare and contrast gene expression profiles in the frontal cortex of 8-week-old Prodh-deficient mice. We used t-tests on Affymetrix GeneChip data analyzed by Robust Multi-array Average (RMA)26,27 to identify genes whose expression differed between mutant mice and their wild-type littermate controls. We observed changes in the levels of a few hundred transcripts in response to the mutation (false discovery rate o 0.05; M.P., P.P., M.K. and J.A.G., unpublished data). For the purpose of this study, we clustered the annotated genes into groups by biological function and applied a statistical analysis using ErmineJ28. On the basis of strict criteria, expression of 12 sets of genes (Gene Ontology (GO) terms) was significantly altered (Fig. 3c), implicating only three major processes: (i) translational initiation and protein processing, (ii) mitochondrial function and metabolism and (iii) presynaptic neurotransmitter release and regulation. Most notably, one of the top upregulated genes in the entire dataset and the top upregulated gene in the third group was Comt (P ¼ 0.00002) (Fig. 3d), which also maps within the 22q11.2 locus and is a candidate schizophrenia susceptibility gene. We prioritized Comt for further investigation because our expression analysis suggested a previously unsuspected interaction between Prodh and Comt that could, in principle, modulate the risk associated with the 22q11.2-related psychiatric phenotypes, their expression or both.


[

NUMBER 11

[

NOVEMBER 2005

Confirmation of Comt upregulation Quantitative reverse transcription polymerase chain reaction (RTPCR) experiments confirmed the upregulation of Comt mRNA in an independent cohort of frontal cortices (B25% increase, P ¼ 0.004; Fig. 4a). Western blots in the frontal cortices of adult Prodh-deficient mice and wild-type littermate control mice indicated a highly reproducible increase (P ¼ 0.0001) in the levels of both the soluble and the membrane-associated form of the Comt protein (Fig. 4b). In rodent cortex, Comt is primarily localized in large pyramidal neurons29. Nonquantitative immunocytochemical analysis using an antibody to Comt confirmed both the cellular distribution pattern as well as the increase in the Comt protein levels observed in the western blots (Fig. 4c). Developmental analysis following the time course of change in Comt levels suggested that the observed increase was established at around the third week of life (B30% increase, P ¼ 0.01; Fig. 4b), a period of active synapse formation and stabilization in the frontal cortex that coincides with a robust postnatal peak in Prodh gene expression and a two- to threefold decrease (compared to adult levels) in synaptic L-proline uptake30,31. Degradation by Comt is likely to be a key mechanism for regulating the synaptic action of dopamine in the frontal cortex, where the dopamine transporter is expressed at very low levels, but not in the striatum, where the dopamine transporter is abundantly expressed. Consistent with this view, immunoblot analysis did not reveal a significant increase of Comt in the striatum of Prodhdeficient mice (Fig. 4b); therefore, the upregulation we observed selectively in the frontal cortex, a major recipient of dopaminergic input, may signify a response to increased local dopaminergic

1589

ARTICLES Figure 4 Validation of RNA expression profiling of 1 2 Comt in Prodh-deficient mice. (a) Validation of *** 1.4 RNA expression profiling of Comt using Taqman ** 1.2 real-time RT-PCR. Comt shows an upregulation of * 1.0 about 25% (P ¼ 0.004). Data (threshold cycle) wt 1 0.8 are standardized to b-actin and shown both as kd 0.6 values for each of the four or five independent experiments (black bars) and as mean ± s.e.m. of 0.4 all experiments performed for each genotype (gray 0.2 bars). **P o 0.01. (b) COMT levels in the frontal 0 0 0 1234 12345 FCTX STR FCTX cortex during left, early postnatal development P0 P3 P8 P20 Adult and right, in the adult frontal cortex and striatum of Prodh-deficient mice (n ¼ 20) and wild-type wt kd FCTX STR littermate controls (n ¼ 10). COMT levels are 28 kDa Vehicle MK801 Vehicle MK801 FCTX also statistically unchanged in the striatum 24 kDa 28 kDa (P ¼ 0.083). We observed small increases in this 24 kDa 28 kDa STR 24 kDa brain region occasionally, but not reproducibly ** Vehicle between experiments. Mean ± s.e.m. optical 1.4 1.4 wt kd 0.1 mm densities are shown, reflecting the normalized MK801 1.2 1.2 Comt levels. *P o 0.05; **P o 0.001. 1.0 1.0 (c) Representative immunoblots and 0.8 0.8 immunostained sections from the frontal cortex 0.6 0.6 and striatum of Prodh-deficient mice and wild0.4 0.4 0.2 0.2 type littermate control mice. (d) Effect of 0 0 subchronic administration of the non-competitive NMDA antagonist MK801 on Comt expression in frontal cortex and striatum of 129/SvEv wild-type mice. Data were standardized by b-actin and are given as mean ± s.e.m. of normalized optical densities for 10 mice per group. wt, wild-type littermate control mice; kd, homozygous Prodh-knockdown mice; FCTX, frontal cortex; STR, striatum. COMT levels

b


COMT levels

a

c

d

Dopaminergic dysregulation in the frontal cortex We asked whether the observed upregulation of Comt levels signifies a more generalized dysregulation of cortical dopamine turnover and transmission. First, we determined levels of both total dopamine and extracellular dopamine in both cortex and striatum. We did not observe any significant sustained differences in basal total dopamine levels in either the frontal cortex or striatum of Prodh-deficient mice as compared to their wild-type littermates (levels of 3,4-dihydroxyphenylacetic acid (DOPAC), homovanillic acid (HVA), norepinephrine and hydroxytryptamine (5-HT) were also unchanged; Fig. 5a). We also monitored extracellular levels of dopamine using in vivo microdialysis. Baseline levels did not differ between the two genotypes in either cortex or striatum. After acute systemic D-amphetamine administration (2.5 and 7.5 mg per kg body weight intraperitoneally (i.p.)), extracellular dopamine increased from the baseline levels in both genotypes and both brain regions (Fig. 5b,c). Notably, Prodh deficiency significantly (P o 0.05) potentiated cortical (Fig. 5b), but not striatal (Fig. 5c), dopamine overflow induced by administration of 7.5 mg of D-amphetamine per kg body weight; this was in good correlation with the response pattern observed in locomotor assays. Second, we examined whether, in addition to Comt upregulation, other changes that would also result in dampening of dopaminergic

transmission develop in the frontal cortex of Prodh-deficient mice. We analyzed the levels of dopamine receptors and of several signaling molecules shown to participate in or modulate dopamine action32–34 (Fig. 6). We assessed protein levels in western blots from cortical extracts (dopamine receptor D1 (DRD1) was assayed by quantitative RT-PCR because antibodies with acceptable mouse specificity in western blots were not available). We observed significant changes, reproducible in three experiments, in the levels of the following: (i) DRD1 (B35% downregulation, P ¼ 0.02); (ii) the dopamine- and cAMP-regulated phosphoprotein DARPP-32, also known as PPP1R1B (B30% downregulation, P ¼ 0.02); and (iii) protein phosphatase 3, also known as calcineurin (all three catalytic subunits, PPP3CA, PPP3CB and PPP3CC; about 33% upregulation, P ¼ 0.001). Because the antibody to calcineurin recognizes all three catalytic subunits, we performed a follow-up analysis with subunit-specific antibodies. A specific, quantifiable signal was observed only for PPP3CC (calcineurin gamma catalytic subunit), revealing a significant increase (P ¼ 0.03) and suggesting that an increase in the levels of this subunit accounts, at least partially, for the signal observed with the non-discriminant antibody. Notably, dopamine acting through DRD1 results in a phosphorylation of DARPP-32 mediated by protein kinase A (PKA), causing inhibition of phosphatase 1. Calcineurin acts as the principal mediator of the dephosphorylation-dependent inactivation of DARPP-32 (ref. 32; see also Fig. 6). Because DARPP-32 phosphorylation is enhanced by DRD1 activation and inhibited by calcineurin, the net effect of the changes we observed in the cortex of Prodhdeficient mice—namely, an increase in COMT and calcineurin and a decrease in DRD1 and DARPP-32—should result in a synergistic decrease in dopamine-mediated protein phosphorylation and signaling. Notably, a chronic MK801 treatment protocol was previously shown to result in a similar decrease of DRD1 levels in the cortex of rats35. By contrast, levels of dopamine receptor D2 (DRD2, which is predominantly localized in subcortical areas), as well as of downstream DRD2 effectors33,34, remained statistically unchanged (P 4 0.05) (Fig. 6).

1590

VOLUME 8

hypersensitivity. Notably, subchronic treatment of 129/SvEv wild-type mice with MK801 (one daily injection of 0.25 mg per kg body weight for 2 weeks), followed by a comparison of Comt levels in the frontal cortex and striatum of treated and untreated animals, revealed a regionally selective pattern of Comt expression which phenocopied that of the Prodh-deficient mice. That is, Comt levels increased in frontal cortex (by roughly 35%, P ¼ 0.01) but not in striatum (P ¼ 0.4; Fig. 4d). This observation, along with the finding that upregulation of Comt levels (unique among all 22q11 orthologs) is part of a coordinated change in expression of genes regulating neurotransmitter action, strongly suggests that it is the effect of Prodh deficiency on synaptic transmission that underlies the increase of cortical Comt levels.

[

NUMBER 11

[


ARTICLES

200 100 0

60

120

Amphetamine (2.5 mg

5

240

2 1

600 500 400 300 200 100

0 HVA

NE

60

120

Amphetamine (7.5 mg

5-HT

240

500 400 300 200 100 0 60

100

120

Time (min)

0

60

120

240

300

kg–1)

600 500 400 300 200 100 0

–60

0

60

120

240

300

Amphetamine (7.5 mg kg–1)

600

0

200

Amphetamine (2.5 mg

550 500 450 400 350 300 250 200 150 100 50 0

kg–1)

700

–60

300

0

300

Percent change DA (AUC values)

DOPAC

Percent change DA

DA

0

400

–60

0 –60

500

kg–1) Percent change DA (AUC values)

3

600

300

700

4

Percent change DA

300

–60

Percent change DA

ng per mg of protein

400

0

6


500

Percent change DA

kd

600

Vehicle

c

550 500 450 400 350 300 250 200 150 100 50 0

240

300

550 500 450 400 350 300 250 200 150 100 50 0

* Percent change DA

wt

Vehicle 700 Percent change DA (AUC values)

b Percent change DA

a

600 500 400 300 200 100 0 –60

0

60 120 Time (min)

240

300

Figure 5 Dopaminergic dysregulation in the frontal cortex of Prodh-deficient mice. (a) HPLC analysis of dopamine (DA), DOPAC, HVA, norepinephrine (NE) and 5-HT (ng per mg of protein) in the frontal cortex of Prodh-deficient mice and wild-type littermate control mice (mean ± s.e.m., n ¼ 5). (b,c) Monitoring the extracellular levels of dopamine using in vivo microdialysis. Microdialysis probes were placed in both prefrontal cortex (b) and dorsal striatum (c) and dopamine levels were measured at baseline after administration of saline as well as after acute systemic administration of D-amphetamine (at 2.5 and 7.5 mg per kg body weight i.p.) (mean ± s.e.m.). wt, wild-type littermate control mice; kd, homozygous Prodh-knockdown mice; AUC, area under curve. *P o 0.05.

Behavioral interaction between Prodh and Comt We next extended our molecular studies to behavioral and pharmacological analysis to ask whether the epistatic interaction between Prodh and Comt, which we observed at the transcriptional level, correlates with epistatic interactions at the behavioral level. If Comt upregulation develops as a homeostatic response to, rather than primary cause of, local dopaminergic hypersensitivity, inhibition of Comt activity should exaggerate rather than attenuate observed behavioral deficits and may induce additional deficits. It was not practical to address this issue by generating double heterozygous and double homozygous mouse mutants for both Prodh and Comt because of the physical proximity of the two genes and the fact that heterozygous Prodh mutant mice have near normal L-proline levels. Instead we used tolcapone, a reversible Comt inhibitor that crosses the blood-brain barrier36, to examine the effects of reduced Comt activity in three schizophrenia-related mouse behaviors that are critically influenced by the level and pattern of cortical dopamine transmission: namely, sensitivity to the locomotor effects of D-amphetamine37,38, working memory as assayed by the T-maze test39 and sensorimotor gating as assayed by the PPI test40. First, we compared the effect of tolcapone on the locomotor effects of low-dose D-amphetamine (which does not stimulate locomotor activity when injected alone) in wild-type and homozygous Prodh-knockdown mice. We used locomotor activity as a convenient behavioral trait to assess because of the good correlation between D-amphetamine–induced cortical dopamine release and response in locomotor assays and because of several previous rodent studies demonstrating that locomotor activity is profoundly influenced by the level and pattern of cortical dopamine transmission17,41. Prodhdeficient mice and wild-type littermate control mice received injections of saline, tolcapone (30 mg per kg body weight), D-amphetamine (1 mg


[

NUMBER 11

[

NOVEMBER 2005

per kg body weight) or both drugs together. Although the locomotor stimulating effect (as compared to the saline vehicle) of individual administration of tolcapone or low-dose D-amphetamine was not significantly different in either genotype, tolcapone significantly potentiated the effect of low-dose D-amphetamine in Prodh-deficient mice (P ¼ 0.04), but not in wild-type littermate mice (Fig. 7a). Second, we compared working memory performance in a delayedalternation test (T-maze), a cognitive task critically dependent on synaptically released dopamine in prefrontal cortex, in the wild-type and homozygous Prodh-knockdown mice. As mentioned previously, at baseline, Prodh-deficient mice do not have deficits in their spatial working memory, probably owing to the emergence of compensation. One day after they reached the testing criteria (training accuracy 470% for two consecutive days), mice were challenged with vehicle, the DRD1 agonist SKF-38393 (5 mg per kg body weight subcutaneously), tolcapone (30 mg per kg body weight i.p.) or a combination of SKF38393 and tolcapone. After the administration of either tolcapone alone or combined tolcapone-SKF-38393 injections, we found a significant difference in working memory performance between the two genotypes. Specifically, under conditions of Comt inhibition, mutant but not wildtype mice consistently made more working memory errors in the Tmaze than did vehicle-treated controls (P o 0.05) (Fig. 7b). SKF38393 had no effect on the error rate in our assay on either genotype. Third, we compared the efficiency of sensorimotor gating as assayed by the PPI test. Prodh-deficient mice and wild-type littermate control mice were treated with saline or tolcapone (30 mg per kg body weight), 20 min before testing (Fig. 7c). Under conditions of Comt inhibition, the mutant mice consistently demonstrated lower PPI as compared to vehicle-treated mutants (P o 0.05). No such effect was observed in wild-type control mice (Fig. 7c).

1591

In summary, Comt inhibition had minimal or no effect on all three tested behaviors in wild-type mice. However, it invariably either exaggerated or induced behavioral deficits in Prodh-deficient mice. This pattern of nonadditivity in the behavioral contributions of these two genes is a hallmark of epistasis, providing strong evidence of an interaction between the two genes.

100 90 80 70 60 50 40 30 20 10

*

DRD1

2.25 2.00 1.75 1.50 1.25 1.00 0.75 0.50 0.25

*

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

DARPP-32

PPP3CA PPP3CB PPP3CC (–)

DRD1

PKA

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

**

*

1.2

1.2

1.2

1.0

1.0

1.0

0.8

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

PPP3CC

PPP3R1

0.2

PP1

PKA

Calcineurin PP1

DARPP-32

Dopamine-

(–) wt (+) DISCUSSION dependent DA effects on The genetic complexity of common psychia(–) kd behavior AKT GSK3β tric disorders has been repeatedly inferred DRD2 (–) (–) from the pattern of inheritance and research3.0 1.2 ers’ inability to identify consistent linkage 1.6 1.6 1.6 1.6 1.6 1.0 2.5 signals. Epistasis among susceptibility genes 1.2 1.2 1.2 1.2 1.2 0.8 2.0 lies at the core of this complexity but its 0.6 1.5 0.8 0.8 0.8 0.8 0.8 0.4 1.0 biological basis remains elusive. In this con0.4 0.4 0.4 0.4 0.4 0.2 0.5 text, perhaps the most important insight provided by our analysis is the clear demonGSK3β DRD2 AKT1 AKT2 AKT3 PIPR3K PKC stration of epistatic interaction between the Prodh and Comt genes at the level of tran- Figure 6 Dopamine-related signaling molecules in the frontal cortex of Prodh-deficient mice. The figure shows levels of dopamine receptors as well as of several signaling molecules that participate in or scription and behavior, which is likely to modulate dopamine action. Relations among some of these molecules are summarized in the diagram. represent a homeostatic response to enhanced Data were normalized to b-actin and represent mean ± s.e.m. of normalized optical densities for between dopaminergic signaling in the frontal cortex 18 and 20 mice per group and three independent experiments. DRD1 levels were compared using that emerges as a result of Prodh deficiency. TaqMan real-time RT-PCR in ten mutant and ten wild-type mice in three independent experiments. The fact that interaction between these two Data were standardized by b-actin and given as mean ± s.e.m. wt, wild-type littermate control mice; genes modulates a number of schizophrenia- kd, homozygous Prodh-knockdown mice. *P o 0.05; **P o 0.01. related phenotypes in mice provides a framework for understanding the high risk for schizophrenia associated with microdeletions of the 22q11.2 locus, interactions among unlinked loci that produce impaired synaptic which encompasses both the PRODH and COMT genes. If COMT function or impaired development of homeostatic response may also upregulation is indeed one of the mechanisms used to control cortical account for the epistatic component of the genetic risk of psychiatric dopaminergic hypersensitivity, then individuals with schizophrenia disorders in general. Identification, in the brains of Prodh-deficient who have a 22q11.2 deletion are at a particular disadvantage because mice, of the upregulation of genes encoding calcineurin subunits— they are deficient for both genes and perhaps unable to compensate including PPP3CC, a strong candidate schizophrenia susceptibility efficiently, through COMT, for the cortical dopaminergic hyperactivity gene42—supports this idea; it also suggests that variation in the induced by PRODH deficiency. Therefore, as already implied by genetic PPP3CC gene may increase the risk of the disease by impairing association studies, 22q11.2-associated schizophrenia may have the compensation for an overactive dopaminergic transmission caused characteristics of a contiguous gene syndrome: deficiency in more than by other primary deficits, including PRODH deficiency. one gene contributes to the condition by impairing both synaptic Although the emergence of cortical dopaminergic dysregulation lies function and the development of compensatory response. Such syner- at the basis of the observed epistasis, its cause has yet to be determined. gistic interaction among two physically linked genes could in principle One possibility, supported by the results presented here, is that in early lead to the high disease risk associated with this locus, modulate the postnatal life, PRODH deficiency affects the state of cortical dopamiexpression of the phenotype or both. Similar patterns of genetic nergic transmission and signaling indirectly, through interference with

2 0

– –

+ –

– +

+ Tolcapone + Amphetamine

4 2 0

– –

+ –

– +

+ Tolcapone + Amphetamine

– – + + – + – +

4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0

c

* * wt kd

– – + + – + – +

Tolcapone SKF38393

100 80 60 40 20 0

Percentage PPI

4

6

3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0

Percentage PPI

6

8

b 4.5 4.0 5-s errors

8

*

5-s errors

10

10


a Post- to pre-injection activity ratio


ARTICLES

78 82 90 78 82 90 –

–

–

+

+

+

100 80 60 40 20 0

*

*

78 82 90 78 82 90 –

–

–

+

+

+

PPI (db) Tolcapone

Figure 7 Epistatic interaction between Prodh and Comt at the behavioral level. (a) Effect of tolcapone on D-amphetamine–induced locomotor activity. (b) Effect of tolcapone on working memory performance in a delayed-alternation test (T-maze). Effect of SKF38393, a DRD1 agonist, was also assayed as a control. There were no tolcapone-induced effects in latency and time to complete the task. In control experiments, we found that tolcapone treatment had no effect on locomotor activity of either genotype (data not shown). Results for 20-s inter-run delays are not shown because the accuracy of T-maze choices were close to chance for this delay interval (the number of 20-s delay errors in ten trials was 4.243 ± 0.553 for wild-type mice and 4.571 ± 0.528 for Prodh-deficient mice), suggesting mouse-strain–specific difficulties in mnemonic maintenance. (c) Effect of tolcapone on sensorimotor gating assayed by the PPI test. Mice received injections of saline or tolcapone 20 min before the test. All data are mean ± s.e.m. wt, wild-type littermate control mice; kd, homozygous Prodhknockdown mice. *P o 0.05.

1592

VOLUME 8

[

NUMBER 11

[



ARTICLES glutamatergic pathways. This possibility is also supported by extensive literature on pharmacological rodent models of schizophrenia in which chronic administration of drugs such as MK801 or PCP that dysregulate glutamatergic transmission results in a sustained hyper-responsivity of dopaminergic neurotransmission17,43. Indeed, the pattern of enhanced amphetamine-induced cortical dopamine release and locomotor activity is very similar to the one observed after chronic treatment with PCP in rodents17; notably, the behavioral effects of such drugs are also modulated by COMT levels and are exaggerated in Comt-null mice (M.P., M.K. and J.A.G., unpublished data). Of course, we cannot formally exclude the possibility that PRODH deficiency directly affects dopaminergic transmission (either within the frontal cortex or in other brain areas that supply dopaminergic afferents); neither can we ignore the possibility that the emergence of cortical dopaminergic dysregulation may be indirectly related to prominent changes in cortical expression of genes regulating protein synthesis or processing and mitochondrial function. It is also unclear how a deficiency in Prodh causes an enhancement of glutamatergic synaptic transmission. This enhancement is probably due to an increase in probability of glutamate release given the observed change in paired-pulse facilitation, a marker of presynaptic function. Furthermore, we found that 200-Hz LTP at these synapses is significantly decreased. As this form of LTP is due in part to an increase in glutamate release15, the reduction in LTP may be explained by an occlusion of this presynaptic component by the enhanced release resulting from the loss of Prodh activity. As our experiments were performed in the absence of any blockers of GABA receptors, it is also possible that a reduction in inhibitory synaptic transmission may contribute to the enhancement in the EPSP. However, the finding that the proline transporter is localized to a subset of glutamatergic terminals12 and is not present in GABAergic synaptic terminals suggests that the primary change may involve glutamatergic synapses. Despite the insights into 22q11.2-associated schizophrenia provided by the present analysis, several issues remain unresolved. What percent of the risk or what aspects of the phenotype in individuals with schizophrenia can be explained by this interaction? Do additional genes in the region that may contribute to the risk exert their effects by modulating this interaction? Or do they act independently of PRODH? These questions notwithstanding, our model offers testable predictions that can now be examined in individuals with 22q11.2 microdeletions. METHODS All animal procedures have been performed according to protocols approved by appropriate Animal Care and Use Committees established by Columbia University, Rockefeller University and the University of Utrecht, under federal and state regulations. Prodh-deficient mice. The Prodh-deficient mice have been described in detail elsewhere11. Electrophysiology. Hippocampal slices were prepared from Prodh-deficient and wild-type male mice, between 8 and 13 weeks of age. All experiments were performed in a blind manner as described previously15 and as detailed in Supplementary Methods. RNA isolation and probe preparation for the microarray hybridization. We dissected a total of 20 frontal cortices from male littermate mice 8 weeks of age: 10 were wild-type mice and 10 were homozygous Prodh-knockdown mutants; these were processed using standard protocols recommended by Affymetrix (http://www.affymetrix.com/products/arrays/specific/mgu74.affx). For hybridization, cRNA was fragmented and exposed to Affymetrix Mouse genome 430 2.0 array set chips (which probe expression of 39,000 full-length genes and


[

NUMBER 11

[

NOVEMBER 2005

expressed sequence tag (EST) clusters from the Unigene database maintained by the National Center for Biotechnology Information). After hybridization, microarrays were washed and scanned (Agilent). Microarray data analysis. Initial microarray images were analyzed with Affymetrix Microarray Suite version 5 to extract intensity values for each probe. Expression values for each probe set were then calculated using the RMA27 approach, run using default settings. Standard Student’s t-tests were used to identify genes showing group-dependent expression changes. We used the method described in a previous study44 to determine the false discovery rate at different P-value thresholds for gene selection. Quantitative real-time RT-PCR. Real-time RT-PCR analysis was performed on an ABI Prism 7900 sequence-detection system (PE Applied Biosystems) as detailed in Supplementary Methods. TUNEL. Terminal deoxynucleotidyl transferase-mediated deoxyuridine triphosphate nick end-labeling assay was performed 8 d after birth, in triplicate, as detailed in Supplementary Methods. Gross brain morphology. Morphometric analysis was performed on four male mice 10 to 16 weeks of age as detailed in Supplementary Methods. Protein extraction and western blots. We prepared protein extracts from frontal cortex and striatum of male mice 6 to 8 weeks of age of both genotypes (20 Prodh-mutant and 18 wild-type littermate control mice). In assays involving chronic MK801 treatment, mice received daily injections with vehicle or drug (MK801 at 0.25 mg per kg body weight) for 14 d and were killed 1 h after the last drug treatment. Extract preparation and western blot assays were performed as detailed in Supplementary Methods. HPLC analysis of total content of neurotransmitters and their metabolites. Five pairs of mice were analyzed for tissue content of dopamine, DOPAC, HVA, norepinephrine and 5-HT, as detailed in Supplementary Methods. In vivo microdialysis. Forty male Prodh-deficient mice were tested together with wild-type littermate controls. Copper microdialysis probes (Microbiotech) were placed in both the dorsal striatum and prefrontal cortex according to the stereotaxic atlas of the mouse brain45. Microdialysis was performed as detailed in Supplementary Methods. Behavioral testing procedures. Behavioral testing was performed as described previously8,11 and as detailed in Supplementary Methods. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS The authors acknowledge C. Frazier and M. Sribour for technical support and assistance with the mouse colony, J. Chan for help with the behavioral analysis, M. Fazzini for help with the immunocytochemistry and the Sloan-Kettering Genomics Core Laboratory (A. Viale, Director) for help with expression profiling. This research was supported in part by the US National Institutes of Health (grant MH67068 to M.K. and J.A.G. and grant DA07418 to D.S.) and by the New York Academy of Sciences (J.A.G.). J.A.G. is also an EJLB Scholar, a Vicente Young Investigator of the National Alliance for Research on Schizophrenia and Depression (NARSAD) and the recipient of a McKnight Brain Disorders Award. S.S.Z. is a recipient of the NARSAD Young Investigator award and the Hereditary Disease Foundation postdoctoral fellowship. M.P. is supported in part by Telethon, Italy (fellowship no. GFP02011). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. Karayiorgou, M. et al. Schizophrenia susceptibility associated with interstitial deletions of chromosome 22q11. Proc. Natl. Acad. Sci. USA 92, 7612–7616 (1995). 2. Liu, H. et al. Genetic variation at the 22q11 PRODH2/DGCR6 locus presents an unusual pattern and increases susceptibility to schizophrenia. Proc. Natl. Acad. Sci. USA 99, 3717–3722 (2002).

1593


ARTICLES 3. Liu, H. et al. Genetic variation in the 22q11 locus and susceptibility to schizophrenia. Proc. Natl. Acad. Sci. USA 99, 16859–16864 (2002). 4. Jacquet, H. et al. PRODH mutations and hyperprolinemia in a subset of schizophrenic patients. Hum. Mol. Genet. 11, 2243–2249 (2002). 5. Li, T. et al. Evidence for association between novel polymorphisms in the PRODH gene and schizophrenia in a Chinese population. Am. J. Med. Genet. B Neuropsychiatr. Genet. 129, 13–15 (2004). 6. Bender, H.U. et al. Functional consequences of PRODH missense mutations. Am. J. Hum. Genet. 76, 409–420 (2005). 7. Jacquet, H. et al. Hyperprolinemia is a risk factor for schizoaffective disorder. Mol. Psychiatry 10, 479–485 (2005). 8. Mukai, J. et al. Evidence that the gene encoding ZDHHC8 contributes to the risk of schizophrenia. Nat. Genet. 36, 725–731 (2004). 9. Shifman, S. et al. A highly significant association between a COMT haplotype and schizophrenia. Am. J. Hum. Genet. 71, 1296–1302 (2002). 10. Egan, M.F. et al. Effect of COMT Val108/158 Met genotype on frontal lobe function and risk for schizophrenia. Proc. Natl. Acad. Sci. USA 98, 6917–6922 (2001). 11. Gogos, J.A. et al. The gene encoding proline dehydrogenase modulates sensorimotor gating in mice. Nat. Genet. 21, 434–439 (1999). 12. Renick, S.E. et al. The mammalian brain high-affinity L-proline transporter is enriched preferentially in synaptic vesicles in a subpopulation of excitatory nerve terminals in rat forebrain. J. Neurosci. 19, 21–33 (1999). 13. Henzi, V., Reichling, D.B., Helm, S.W. & MacDermott, A.B. L-proline activates glutamate and glycine receptors in cultured rat dorsal horn neurons. Mol. Pharmacol. 41, 793–801 (1992). 14. Cohen, S.M. & Nadler, J.V. Proline-induced potentiation of glutamate transmission. Brain Res. 761, 271–282 (1997). 15. Zakharenko, S.S. et al. Presynaptic BDNF required for a presynaptic but not postsynaptic component of LTP at hippocampal CA3-CA1 synapses. Neuron 39, 975–990 (2003). 16. Moghaddam, B. & Adams, B.W. Reversal of phencyclidine effects by a group II metabotropic glutamate receptor agonist in rats. Science 281, 1349–1352 (1998). 17. Balla, A., Sershen, H., Serra, M., Koneru, R. & Javitt, D.C. Subchronic continuous phencyclidine administration potentiates D-amphetamine-induced frontal cortex dopamine release. Neuropsychopharmacology 28, 34–44 (2003). 18. O’Donnell, J., Stemmelin, J., Nitta, A., Brouillette, J. & Quirion, R. Gene expression profiling following chronic NMDA receptor blockade-induced learning deficits in rats. Synapse 50, 171–180 (2003). 19. Jentsch, J.D., Tran, A., Le, D., Youngren, K.D. & Roth, R.H. Subchronic phencyclidine administration reduces mesoprefrontal dopamine utilization and impairs prefrontal cortical-dependent cognition in the rat. Neuropsychopharmacology 17, 92–99 (1997). 20. Stefani, M.R. & Moghaddam, B. Effects of repeated treatment with amphetamine or phencyclidine on working memory in the rat. Behav. Brain Res. 134, 267–274 (2002). 21. Latysheva, N.V. & Rayevsky, K.S. Chronic neonatal N-methyl-D-aspartate receptor blockade induces learning deficits and transient hypoactivity in young rats. Prog. Neuropsychopharmacol. Biol. Psychiatry 27, 787–794 (2003). 22. Szeszko, P.R., Bilder, R.M., Dunlop, J.A., Walder, D.J. & Lieberman, J.A. Longitudinal assessment of methylphenidate effects on oral word production and symptoms in firstepisode schizophrenia at acute and stabilized phases. Biol. Psychiatry 45, 680–686 (1999). 23. Phillips, R.G. & LeDoux, J.E. Differential contribution of amygdala and hippocampus to cued and contextual fear conditioning. Behav. Neurosci. 106, 274–285 (1992). 24. Lalonde, R. The neurobiological basis of spontaneous alternation. Neurosci. Biobehav. Rev. 26, 91–104 (2002).

25. Anholt, R.R. et al. The genetic architecture of odor-guided behavior in Drosophila: epistasis and the transcriptome. Nat. Genet. 35, 180–184 (2003). 26. Bolstad, B.M., Irizarry, R.A., Astrand, M. & Speed, T.P. A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics 19, 185–193 (2003). 27. Irizarry, R.A. et al. Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res. 31, e15 (2003). 28. Pavlidis, P., Lewis, D.P. & Noble, W.S. Exploring gene expression data with class scores. Pac. Symp. Biocomput. 474–485 (2002). 29. Matsumoto, M. et al. Catechol O-methyltransferase mRNA expression in human and rat brain: evidence for a role in cortical neuronal function. Neuroscience 116, 127–137 (2003). 30. Maynard, T.M. et al. A comprehensive analysis of 22q11 gene expression in the developing and adult brain. Proc. Natl. Acad. Sci. USA 100, 14433–14438 (2003). 31. Cohen, S.M. & Nadler, J.V. Sodium-dependent proline and glutamate uptake by hippocampal synaptosomes during postnatal development. Brain Res. Dev. Brain Res. 100, 230–233 (1997). 32. Greengard, P., Allen, P.B. & Nairn, A.C. Beyond the dopamine receptor: The DARPP-32/ protein phosphatase-1 cascade. Neuron 23, 435–447 (1999). 33. Emamian, E.S., Hall, D., Birnbaum, M.J., Karayiorgou, M. & Gogos, J.A. Convergent evidence for impaired AKT1-GSK3b signaling in schizophrenia. Nat. Genet. 36, 131–137 (2004). 34. Beaulieu, J.M. et al. Lithium antagonizes dopamine-dependent behaviors mediated by an AKT/glycogen synthase kinase 3 signaling cascade. Proc. Natl. Acad. Sci. USA 101, 5099–5104 (2004). 35. Healy, D.J. & Meador-Woodruff, J.H. Differential regulation, by MK801, of dopamine receptor gene expression in rat nigrostriatal and mesocorticolimbic systems. Brain Res. 708, 38–44 (1996). 36. Kaakkola, S., Gordin, A. & Ma¨nnisto¨, P.T. General properties and clinical possibilities of new selective inhibitors of catechol O-methyltransferase. Gen. Pharmacol. 25, 813–824 (1994). 37. Yui, K. et al. Neurobiological basis of relapse prediction in stimulant-induced psychosis and schizophrenia: the role of sensitization. Mol. Psychiatry 4, 512–523 (1999). 38. Laruelle, M. The role of endogenous sensitization in the pathophysiology of schizophrenia: implications from recent brain imaging studies. Brain Res. Brain Res. Rev. 31, 371–384 (2000). 39. Castner, S.A., Goldman-Rakic, P.S. & Williams, G.V. Animal models of working memory: insights for targeting cognitive dysfunction in schizophrenia. Psychopharmacology (Berl.) 174, 111–125 (2004). 40. Geyer, M.A., Krebs-Thomson, K., Braff, D.L. & Swerdlow, N.R. Pharmacological studies of prepulse inhibition models of sensorimotor gating deficits in schizophrenia: a decade in review. Psychopharmacology (Berl.) 156, 117–154 (2001). 41. Vezina, P., Blanc, G., Glowinski, J. & Tassin, J.P. Opposed behavioural outputs of increased dopamine transmission in prefrontocortical and subcortical areas: a role for the cortical D-1 dopamine receptor. Eur. J. Neurosci. 3, 1001–1007 (1991). 42. Gerber, D.J. et al. Evidence for association of schizophrenia with genetic variation in the 8p21.3 gene, PPP3CC, encoding the calcineurin gamma subunit. Proc. Natl. Acad. Sci. USA 100, 8993–8998 (2003). 43. Seeman, P. Dopamine receptors and the dopamine hypothesis of schizophrenia. Synapse 1, 133–152 (1987). 44. Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. R. Stat. Soc. Ser. B 57, 289–300 (1995). 45. Franklin, K.B.J. & Paxinos, G. The Mouse Brain in Stereotaxic Coordinates (Academic Press, New York, 1997).

1594

VOLUME 8

[

NUMBER 11

[



ARTICLES

Neural basis and recovery of spatial attention deficits in spatial neglect Maurizio Corbetta1–4, Michelle J Kincade5, Chris Lewis2, Abraham Z Snyder1,3 & Ayelet Sapir1 The syndrome of spatial neglect is typically associated with focal injury to the temporoparietal or ventral frontal cortex. This syndrome shows spontaneous partial recovery, but the neural basis of both spatial neglect and its recovery is largely unknown. We show that spatial attention deficits in neglect (rightward bias and reorienting) after right frontal damage correlate with abnormal activation of structurally intact dorsal and ventral parietal regions that mediate related attentional operations in the normal brain. Furthermore, recovery of these attention deficits correlates with the restoration and rebalancing of activity within these regions. These results support a model of recovery based on the re-weighting of activity within a distributed neuronal architecture, and they show that behavioral deficits depend not only on structural changes at the locus of injury, but also on physiological changes in distant but functionally related brain areas.

Injury to a brain area causes behavioral deficits that are thought to reflect the local dysfunction of neurons at the site of injury. This logic (the local injury hypothesis) has been used for over 150 years by physicians to localize lesions in the brain. Neuropsychologists have built on the same logic to show the independence of mental processes (for example, see ref. 1). However, as originally pointed out by Hughlings Jackson, the localization of normal functions (or mental operations) may or may not correspond to the localization of behavioral deficits. A lesion may cause dysfunction in other nodes of a functional brain network2,3, impairing processes other than those mediated by neurons at the site of injury (the distributed injury hypothesis). Accordingly, recovery of function may depend on the restoration and rebalancing of activity in structurally normal, but functionally impaired, nodes of a task-relevant network. Here, we test whether the distributed injury hypothesis applies to spatial neglect, one of the main attentional syndromes following injury to the human brain. Spatial neglect occurs in about 25–30% of all stroke-affected individuals (an estimated 3–5 million a year, worldwide)4,5. It is a complex syndrome characterized by a failure to attend to, look at and respond to stimuli (objects, food, people) located on the side of space or of the body opposite to the side affected by a brain lesion6–8. This spatial bias coexists with difficulties in maintaining alertness and detecting targets that are not lateralized to one side of space and has been linked to (non-spatial) deficits in attentional capacity (spatial and temporal) and impaired vigilance9–11. Over 90% of individuals with spatial neglect have right hemisphere injury and neglect of the left side of space or body. The most frequent sites of damage are right inferior parietal, ventral frontal12,13 and superior temporal cortex14, along with subcortical nuclei12,15.

Although the contribution of different regions to the different processing deficits in neglect is unclear (but see refs. 16 and 17), it is currently assumed that these regions serve as specialized nodes of a network that mediates spatial attention, visuomotor behavior (eye-hand coordination) and vigilance6,8. Notably, the lesion anatomy of spatial neglect does not closely match the pattern of brain activation associated with spatial attention and visuomotor behavior. When subjects direct attention, eye movements or hand movements to visual objects—tasks on which individuals with neglect show a rightward bias—parietal and frontal regions are activated that are more dorsal than those anatomically damaged in neglect (Fig. 1a). These regions form a bilateral dorsal frontoparietal network (Fig. 1a, blue regions) that governs spatial attention and visuomotor control (eye-hand movement)18–21, contains visuotopic maps of contralateral space22,23 and is involved in goal-directed stimulus and response selection24. This network is a plausible, neural substrate of spatial biases in neglect. The location of anatomical damage and its right hemisphere lateralization more closely matches a set of ventral temporoparietal and frontal regions related to the detection of salient sensory events18,25,26(Fig. 1a). These regions form a ventral attention network that redirects the dorsal network to novel and behaviorally relevant stimuli, especially when these are unattended24 (Fig. 1a, orange regions). Damage to these ventral regions may directly mediate deficits in ‘non-spatial’ processes such as vigilance or (attentional capacity10) as well as in attentional reorienting. We hypothesize that spatial attention deficits in neglect arise from the structural or functional dysfunction of both dorsal and ventral attention networks. A stroke in ventral cortex (either frontal or parietal) should interfere with attentional reorienting. Moreover, as the ventral

1Departments of Neurology, 2Radiology and 3Anatomy and Neurobiology, Washington University, 660 South Euclid Avenue, St. Louis, Missouri 63110, USA. 4The Rehabilitation Institute of St. Louis, 4444 Duncan Avenue, St. Louis, Missouri 63108, USA. 5Department of Psychology, Washington University, 660 South Euclid Avenue, St. Louis, Missouri 63110, USA. Correspondence should be addressed to M.C. ([email protected]).



[

NUMBER 11

[

NOVEMBER 2005

1603

ARTICLES

a

Figure 1 Functional-anatomical model of attention. (a) Dorsal (blue, topdown) and ventral (orange, stimulus-driven) regions of the human attention system. Black, hypothetical cortical lesion in ventral frontal, insular and perisylvian cortex, causing spatial neglect. (b) Anatomical model of attention and changes in relative activation after acute damage to right ventral frontal cortex. Areas in blue (dorsal system) mediate top-down stimulus-response selection and bias the activity in visual cortex. Areas in orange (ventral system) mediate stimulus-driven reorienting. The shading in light blue and red indicate, respectively, relative decreases and increases in functional activity. IPS-SPL, intraparietal sulcus–superior parietal lobule; FEF, human frontal eye field; VFC, ventral frontal cortex; TPJ, temporoparietal junction; IFG, inferior frontal gyrus; STG, superior temporal gyrus; MFG, middle frontal gyrus; IPL, inferior parietal lobule.

FEF

IPS-SPL

Ventral cortex damaged


Visual cortex

TPJ (IPL-STG)

VFC (IFg–MFg)

Top-down control

b L FEF

R FEF Stimulus– response selection

L VFC

Structural damage

neuropsychological and computerized tasks which assessed the presence of spatial or body neglect, anosognosia, vigilance, spatial attention and reaching deficits. Circuit breaker

Dynamic imbalance L TPJ

Stimulus-driven control

Relative decrease

Functional damage

Relative increase

network normally sends the dorsal network a ‘circuit-breaking’ signal during target detection, a ventral lesion should also decrease activity of the (ipsilateral) right dorsal network (Fig. 1b). The resulting hemispheric imbalance could produce a rightward spatial bias in visual processing. A final prediction is that the recovery of these attentional deficits is associated with a normalization of activity in both dorsal and ventral attention networks. Previous work shows that the recovery of neglect is associated with the restoration of normal activity in ipsilateral subcortical nuclei after frontal damage in monkeys27 or in right hemisphere regions after cortical-subcortical damage in humans17,28,29. However, no study to date has measured functional task-evoked brain activity in a relatively numerous and anatomically homogenous group of individuals with spatial neglect during both acute and chronic stages of recovery and related brain activity to behavioral performance. Here we show that spatial attention deficits in neglect after right frontal damage correlate with abnormal functional activation of structurally intact regions of the dorsal and ventral attention networks and that recovery of these deficits correlates with the normalization of activity within these regions.

Anatomy The majority of subjects (63%, or 7 of the 11) had lesions centered in the perisylvian region, including superior temporal gyrus (STG), frontal operculum, insula and putamen (Fig. 2a,d). The temporoparietal junction (TPJ), including the supramarginal gyrus (SMG) and underlying white matter, was damaged in 45% of subjects (5 of 11; Fig. 2a,b), whereas no subject had lesions that extended into dorsal posterior parietal cortex (specifically, intraparietal sulcus, IPS) or frontal cortex (specifically, frontal eye field, FEF) (Fig. 2c). One subject had a lesion in the parahippocampal gyrus, but otherwise the visual cortex was completely spared. On average this group was representative of the most common lesion sites in neglect15. The location of TPJ damage matched the location of maximal damage after middle cerebral artery strokes13. Behavior Clinically, from the acute to the chronic stage of recovery, all subjects improved on traditional measures of spatial neglect (Supplementary Table 1). Whole-brain functional magnetic resonance imaging (fMRI) of the blood oxygenation level–dependent (BOLD) signal, an indirect

a

c

SFSPrCeS

IPS SPL

d b

0.6

IFG

Insula

SMG

RESULTS To test the above predictions, we performed a prospective longitudinal study of individuals with spatial neglect (n ¼ 11) following unilateral strokes. Subjects were enrolled on the basis of the presence of extinction to double simultaneous stimulation, omission of targets during visual search or evidence of clinical neglect in activities of daily living within the first week of their stroke (see inclusion criteria in Supplementary Methods online). All subjects underwent standard rehabilitation for at least 3 months after stroke. They were tested at the acute (B4 weeks, mean ± s.d. ¼ 32 ± 22.8 days) and chronic stages of recovery (B39 weeks, mean ± s.d. ¼ 39 ± 11.5 weeks) using a battery of

Figure 2 Lesion anatomy. (a) Atlas brain; right hemisphere, anatomical average of individual lesions. Color scale indicates percentage of subjects with lesion overlapping a specific voxel. Red-yellow areas, 50–70% overlap; yellow-green areas, 30–50% overlap; purple-blue areas, o10% overlap. Dashed lines indicate sections at the level of (b) the TPJ (coronal view; SMG, supramarginal gyrus), (c) dorsal frontoparietal regions (transverse view; SFS-PrCeS, superior frontal sulcus-precentral sulcus: that is, locations of FEF and IPS-SPL) and (d) ventral frontal and insular cortex (transverse view; IFG, inferior frontal gyrus).

1604

VOLUME 8

0

[

NUMBER 11

[


ARTICLES right TPJ (Fig. 4a). These maps are not directly comparable to those in the stroke650 620 affected subjects but provide a qualitative 625 600 baseline for comparison. 600 580 575 In the neglect group, at 4 weeks after stroke 560 550 (Fig. 4b), a significant alteration was evident 540 525 520 in the activation pattern. In the damaged right 500 500 475 hemisphere, large portions of occipital visual 480 450 cortex, posterior parietal cortex (especially IPS 460 425 Ipsilesional Valid Contralesional Invalid and superior parietal lobule (SPL)) and dorTarget validity Visual field solateral prefrontal cortex (DLPFC) showed weak or no task-related activity, even though Figure 3 Behavioral results. (a) Recovery in contralesional visual field. (b) Recovery in reorienting to these regions were anatomically intact. In the invalid targets. left hemisphere, there was decreased activity in occipital visual cortex and prefrontal cortex noninvasive indicator of neuronal activity, was acquired at 4 weeks and but robust activation in parietal cortex and sensory motor cortex 39 weeks after stroke. Subjects were scanned while performing a Posner (SMCX; Fig. 4b). At 39 weeks, a strong reactivation occurred in visual orienting task used to define dorsal and ventral frontoparietal many right hemisphere regions but also in many left hemisphere attention networks in normal observers24 and previously used to study regions (Fig. 4c). spatial neglect30. Subjects viewed a central arrow cue that covertly directed their attention to a left or right location on a computer screen. Dorsal parietal cortex (IPS-SPL) After a random delay, a target (an asterisk) was briefly flashed at one of To determine which brain regions changed their level of activation from the two locations. On 75% of the trials the target was presented at the 4 to 39 weeks, we carried out a random-effect voxel-wise ANOVA using location indicated by the cue (valid), whereas on 25% of the trials it was the MR frame (frames 1–8) and stage (acute or chronic) as factors. One presented at the opposite location (invalid). Subjects pressed a key with notable pattern was observed in dorsal parietal cortex, the posterior their right hand as soon as they detected the target, and accuracy and core of the dorsal attention network (Fig. 5a,b). In the right hemireaction times were measured. Activity induced by the presentation of sphere, dorsal parietal cortex (specifically, IPS-SPL) was not active at the cue stimulus was not separated from activity induced by the the acute stage but strongly reactivated at the chronic stage (pIPS-SPL presentation of the target. All subjects were tested before scanning to 23, –73, 51, P ¼ 0.0001; ventral IPS (vIPS)-precuneus 14, –76, 36, establish that they could see the stimuli, maintain accurate fixation on a P ¼ 0.005). This reactivation was independent of the visual field in large majority of trials (490%) and carry out the task. Eye movements which the target was presented. In contrast, in the left hemisphere, dorsal parietal cortex activity was stronger at the acute than at the were not recorded in the scanner. The behavioral data were analyzed with a three-way analysis of chronic stage (SPL –21, –60, 58, P ¼ 0.008; IPS –26, –54, 24 F7,70 ¼ variance (ANOVA), using stage (acute or chronic), visual field (left or 3.54, P ¼ 0.003). Dorsal parietal cortex was the only brain region that right) and cue validity (valid or invalid) as factors. Overall, subjects showed this interhemispheric ‘push-pull’ pattern from the acute to the detected more targets at the chronic than acute stage (87.7% versus 81.1%; F1,10 ¼ 6.46, P o 0.05), in the ipsilesional (right) than SMCX contralesional (left) visual field (87% versus 82%; F1,10 ¼ 8.35, a FEF IPS-SPL P ¼ 0.01) and when the target was validly cued rather than invalidly R TPJ cued (87% versus 81%; F1,10 ¼ 6.62, P ¼ 0.03). Similar effects were -SMG DLPFC -STG found for the reaction time to detect targets. Recovery was indexed by two measures. First, there was a significant Young Insula adults decrement in the rightward processing bias, as shown by a greater IFG Visual cortex 0.0 10.0 L 10.0 0.0 R improvement in reaction time to targets in the contralesional (left) rather than the ipsilesional (right) visual field (two-way ANOVA of b stage visual field; F 1,10 ¼ 4.77, P ¼ 0.053; Fig. 3a). Second, there was a significant improvement in attentional reorienting, expressed as an improvement in the hit rate and reaction time for detecting invalidly Neglect acute cued rather than validly cued targets (hit rates for acute valid and acute 0.0 5.5 0.0 7.0 invalid were 87% and 76%, respectively; hit rates for chronic valid and chronic invalid were 88% and 87%, respectively; F1,10 ¼ 14.35, IPS-SPL c P ¼ 0.004; reaction time: F1,10 ¼ 4.79, P ¼ 0.053, Fig. 3b). Rightward R TPJ DLPFC -STG DLPFC bias and impaired attentional reorienting (or the ‘‘disengage deficit’’; ref. 30) are robust measures of the spatial impairment in neglect and Neglect correlate with the severity of, and recovery from, spatial neglect as chronic assessed by more traditional measures31. 0.0 7.5 Visual

a

700

Acute Chronic


b

660 640

Acute Chronic

Reaction times (s)

Reaction times (ms)

675

cortex

Functional MRI The normal pattern of brain activation on the Posner task, as shown previously18 for a group of young adults, involves large swaths of occipital, parietal, temporal and frontal cortex bilaterally, except in


[

NUMBER 11

[

NOVEMBER 2005

0.0

5.5

Figure 4 Functional maps of the Posner task. (a) In young adult observers. (b,c) In subjects with neglect at (b) the acute and (c) the chronic stage. Anatomical abbreviations as in previous figures. DLPFC, dorsolateral prefrontal cortex; SMCX, sensory-motor cortex.

1605

ARTICLES L DLPFC

a

R DLPFC

R IPS-SPL

L SPL

b

L SPL (–21, –60, 58)

R IPS-SPL (23, –73, 51)

Percentage signal change

0.5

0.6

0.4

0.4

0.3

0.2

0.2 0.1

0

0

–0.2

–0.1

4

Left L SPL

20

–0.4

0.3

0.1 0

0.2

–0.1

0.1

–0.2 0 0

5

10

15

5

20

10

15

20

10

15

20

5

10

15

20

Chronic left VF Chronic right VF

Acute left VF Acute right VF

Time (s)

5

R DLPFC (37, 43, 28)

0.2

Hit Miss

–0.2

0

0.5 0.4 0.3 0.2 0.1 0 –0.1 –0.2 0

0.3

0.4

0

15

L DLPFC (–46, 20, 39)

0.5

–0.1

10

0.4

L SPL (–15, –74, 48)

0.6

5

Time (s)

Right

R STG


c

0

–0.2

4

–4

Figure 5 BOLD correlates of rightward bias in parietal cortex. (a) BOLD changes from acute to chronic stage in parietal and frontal cortex (thresholded at z ¼ 2.5, P o 0.01 uncorrected), superimposed on three-dimensional atlas brain with average lesion embedded. The dashed black line indicates slice through SPL in panel c. (b) BOLD signal time courses for targets in left and right visual fields at acute and chronic stages. (c) BOLD signals for missed targets in left SPL. Graph, BOLD time course for hits and misses in left SPL.

chronic stage. For example, prefrontal cortex reactivated bilaterally at the chronic stage (left DLPFC: –46, 20, 39, P ¼ 0.009; right DLPFC: 37, 43, 28, P ¼ 0.001; Fig. 5a,b); left FEF activity did not change, whereas adjacent clusters in right FEF showed opposite patterns (right precentral gyrus: 31, –15, 61, P ¼ 0.02 acute 4 chronic; right precentral gyrus: 40, –10, 43, P ¼ 0.01 chronic 4 acute). See Supplementary Table 2 for a complete list of coordinates. To confirm that dorsal posterior parietal cortex was the site of activity imbalance, we carried out a regional ANOVA using regions of interest (ROIs) from the young adult group in IPS (anterior and posterior) and FEF (medial and lateral), regions previously shown to be involved in controlling spatial attention18. This analysis confirmed an imbalance in dorsal parietal cortex but not in the FEF (Supplementary Fig. 1). The interhemispheric push-pull pattern in dorsal parietal cortex is consistent with the hypothesis that the lateralized (rightward) bias in

b y = –76 R IPS/SPL

Visual cortex

neglect is caused by a left hemisphere–orienting mechanism that is relatively hyperactive32. If left parietal cortex hyperactivity mediates the rightward spatial bias, then greater activity in left SPL should correlate with a greater number of missed targets; this invariably occurred in the left visual space. A voxel-wise ANOVA identified several left hemisphere regions active for missed targets; one of the most significant regions was the left SPL (–15, –79, 40; Fig. 5c). In a second analysis, we directly compared hit and miss trials in a voxel-wise ANOVA. Once again we found significant effects in left SPL (graph in Fig. 5c) where the response was significantly stronger for miss than for hit trials (F7,70 ¼ 3.85, P ¼ 0.001), especially at the acute stage (response stage MR frame; F7,70 ¼ 3.16, P ¼ 0.006). Finally, we found a positive significant correlation between rightward bias and left SPL activity (r2 ¼ 0.36, P ¼ 0.051) at the chronic stage (see Supplementary Fig. 2), confirming that hyperactivity in this region correlated with poor orienting toward the left visual field.

L 0.4

0.35

0.35

0.3

0.3

0.25

0.25

0.2

0.2

0.15

0.15

0.1

0.1

0.05

0.05 0

0 –0.05 –0.1

0

c

R

0.4

5

10

15

Time (s)

20

–0.05 –0.1

R visual cortex Young adults (n = 13)

0.4


a


Acute Chronic

0

5

10

15

20

Time (s)

0.3

Left VF Right VF

0.2 0.1 0 –0.1 0

5

10

d Figure 6 BOLD correlates of rightward bias in visual cortex. (a) BOLD changes from acute to chronic stage in visual cortex (thresholded at z ¼ 2.5, P o 0.01 uncorrected). Coronal slice at y ¼ –76. (b) Retinotopic ROIs selected in young adults: green, ventral; purple, dorsal. BOLD signal time courses from left and right ventral ROIs in subjects with neglect. (c,d) BOLD signal time course, in right ventral visual cortex, for left and right visual field targets (c) in young adults and (d) in neglect group at acute and chronic stages.

1606

0.4

Acute left VF Acute right VF Chronic left VF Chronic right VF

15

20

Time (s)

–0.2



–4

R visual cortex Neglect (n = 11)

0.3 0.2 0.1 0 0

5

10

15

20

Time (s)

VOLUME 8

[

NUMBER 11

[


ARTICLES

Temporoparietal junction The second index of behavioral recovery was the improved ability to reorient to unattended locations. This function is known to correlate with the recovery of spatial neglect31 and is specifically associated with damage to the STG16. The TPJ region was defined in our laboratory as the clusters of activation in SMG and STG that show a differential response to unattended (invalidly cued) versus attended (validly cued) visual targets (Fig. 1). This region was damaged in 5 of 11 subjects (Fig. 2a). We observed some reactivation in the ventral part of the TPJ from the acute to the chronic stage (right STG: 63, –44, 21, P o 0.001; right parietal operculum: 57, –35, 35, P o 0.01; see Supplementary Table 2),


[

NUMBER 11

[

NOVEMBER 2005

L SMG (–32, –57, 44)

03

R Pcu (17, –65, 44)

04

34

Left

R STG (56, –45, 16)

0.4

L SMG 0.4 0.2 0

0

5

10 15 Time (s)

20



b

R Pcu

0.2

0

0

5

0.1 0

5

–0.1

10

15

20



0.3

L STG

0.2

0

Time (s)

L SMG

0.3 –0.1 –0.5 200

400

600

800

0.2

600

RT (ms)

800


L STG

400

20

0.1 0

0

5

10 Time (s)

0.5

R Pcu

0.1 –0.3 300

500

700

RT (ms)

–0.2 –0.6

15

R STG

RT (ms) 0.6

20

Acute invalid Chronic invalid


0.7

15

0.2

–0.1

Acute valid Chronic valid

c

10 Time (s)

–0.2

0.3

37

Right

L STG (–62, –44, 13)


Visual cortex Neural models of attention24,33 postulate that posterior parietal cortex interacts with visual cortex for the selection of relevant objects. Our hypothesis predicted that activity in visual cortex should mirror the push-pull interhemispheric pattern observed in posterior parietal cortex. Many regions in visual cortex showed significant changes in task-related BOLD activity from the acute to the chronic stage (Fig. 6a). To test this hypothesis, dorsal and ventral retinotopic ROIs were selected from visual cortex in the young adult group (see Supplementary Methods), and signal time courses were extracted from these ROIs in the neglect group at the acute and chronic stages. In both ventral and dorsal retinotopic ROIs, we observed a relative imbalance at the acute stage with more activity in the left than right hemisphere, and a rebalancing at the chronic stage with a reactivation of the right hemisphere (graph in Fig. 6b). This was confirmed by a significant interaction of stage (acute or chronic) hemisphere (left or right) time in ventral visual cortex (three-way ANOVA; F7,70 ¼ 2.62 P ¼ 0.02) and stage hemisphere visual field time in dorsal visual cortex (four-way ANOVA; F7,70 ¼ 2.59, P ¼ 0.02). In right dorsal occipital cortex, the reactivation was larger for targets in the left (contralesional) than in the right (ipsilesional) visual field. All results were confirmed when using only hit trials. We also observed a disruption of spatially selective responses in right visual cortex. In the young adult group (Fig. 6c), targets in the contralateral visual field evoked stronger responses than did targets in the ipsilateral visual field (F7.84 ¼ 3.18, P ¼ 0.0049), especially in the right hemisphere (F1,12 ¼ 6.4, P ¼ 0.03). In the neglect group (Fig. 6d), a normal lateralization was observed in left visual cortex, whereas in right visual cortex, targets in the left (contralesional) visual field evoked significantly less activity than did those in the right (ipsilesional) visual field (three-way ANOVA: MR frame hemisphere visual field; F1,10 ¼ 4.83, P ¼ 0.05). When compared to the young adult group, this inversion was significant at the acute stage (four-way ANOVA: hemisphere visual field MR frame group; F 7,154 ¼ 2.38, P ¼ 0.02) but not at the chronic stage, even though the time course of the BOLD signal was not qualitatively different (Fig. 6d). Correlation analyses showed only marginal correlations between left or right visual cortex activity and measures of rightward bias (all comparisons, 0.05 o P o 0.10).

a



Figure 7 BOLD correlates of attentional reorienting. (a) Regions involved in the recovery of reorienting. ANOVA (stage validity MR frame) interaction map (expressed as z, thresholded at z ¼ 2.5, P o 0.01 uncorrected). (b) BOLD signal time courses for valid and invalid trials at acute and chronic stages, averaged over left and right visual fields. (c) Correlation across subjects between the magnitude of the BOLD signal and reaction, on invalid trials at the acute stage. Pcu, precuneus.

0.6

R STG

0.2 –0.2 –0.6

400

600

800

RT (ms)

but the degree of reactivation depended on the presence of anatomical damage (Supplementary Fig. 3). To identify regions whose activity varied as a function of both the stage of recovery and attentional reorienting, as indexed by target validity, we ran a voxel-wise ANOVA with MR frame, stage (acute or chronic) and target validity (valid or invalid) as factors. We identified several regions that showed an interaction of stage validity time, including left and right STG (ventral part of TPJ), but also dorsal regions such as the right precuneus and left IPS (Fig. 7a and Supplementary Table 3). This reorienting network in stroke-affected subjects largely overlapped with that recruited in normal subjects under the same conditions. Time-course analysis indicated that the interaction was carried by a weaker and delayed response at the acute stage, especially for invalid targets (Fig. 7b). This interaction was very clear in right TPJ when the time-course analysis compared subjects with and without anatomical damage to this region (Supplementary Fig. 3). The modulation of right TPJ by both the stage of recovery and attentional reorienting (that is, target validity) was confirmed by replicating the stage validity time interaction in a regional ANOVA, in which the ROI was independently selected from the

1607


ARTICLES young adult group. Only the hit trials were included to avoid contamination from error related signals. Finally, we observed, in TPJ and other regions of the reorienting network, a specific correlation at the acute stage between the magnitude of the BOLD signal and the reaction time to invalidly cued targets (left STG: –62, –44, 13, r2 ¼ 0.73, P o 0.001; right STG: 56, –45, 16, r2 ¼ 0.39, P ¼ 0.04; right precuneus: 13, –43, 60, r2 ¼ 0.54, P ¼ 0.01; Fig. 7c). No relationship was found for validly cued targets except in left STG (r2 ¼ 0.49, P ¼ 0.02), eliminating an effect of time on task as an explanation for the correlation. None of these effects interacted with the visual field of the target (all comparisons, P 4 0.05). DISCUSSION We showed that attentional deficits in spatial neglect did not depend just on neuronal dysfunction at the site of injury, but were mediated by the combined structural and functional dysfunction of two interacting frontoparietal attention networks. The recovery of spatial attention deficits, accordingly, correlates with the reactivation and rebalancing of normal activity within these networks. BOLD signals in human stroke model There is growing evidence that BOLD signals may be abnormal in individuals with stroke; hence an important issue is whether our findings might be artifactual. Mechanisms linking local neuronal activity to local hemodynamic changes (blood flow or BOLD)— so-called neurovascular coupling—may be impaired after a stroke, even in the unaffected hemisphere34–36. Although these findings suggest caution in relating BOLD-fMRI signals to neuronal changes in strokeaffected individuals, they cannot explain the current findings. First, changes in the BOLD response during recovery showed a strong correlation with performance. Second, although in many areas recovery was associated with larger BOLD responses, in other areas recovery induced an attenuation of a relatively hyperactive response. Third, none of the strokes in our sample were lacunar—the type associated with artifactual decrement in task-evoked BOLD response35. Finally, most of the results we report occurred in areas that were distant from the core of the lesion where time-dependent changes have been reported36. Rightward bias and reorienting We found different neural correlates for two separate spatial attention deficits and their recovery: rightward spatial bias and the reorienting deficit. The rightward spatial bias, a relative impairment in detecting targets in the left visual field, was associated with a relative functional imbalance at the acute stage in dorsal parietal cortex (IPS-SPL) and visual cortex. At the chronic stage, activity in these regions rebalanced in parallel with behavioral recovery. Our interpretation is that the decreased BOLD responses at the acute stage in dorsal parietal cortex reflects the lack of an excitatory ‘circuitbreaking’ stimulus-driven signal from injured ventral areas during target detection (Fig. 1). Under normal conditions this signal reorients the dorsal system to relevant events, but after VFC damage, its absence induces a relative deactivation of ipsilateral (right) dorsal parietal cortex. The resulting functional imbalance, at the acute stage, in dorsal parietal and visual cortices is manifested as a relative hyperactivation on the left and relative deactivation on the right (dynamic imbalance, Fig. 1). This imbalance and the rebalancing that occurs over time are consistent with competitive (possibly cross-inhibitory37) interactions between oppositely lateralized orienting mechanisms for directing attention and visual representations, as previously hypothesized32. The results of our experiment provide evidence for these competitive

1608

interactions, localize the site of competition to dorsal parietal cortex and show a clear functional interaction between dorsal parietal and visual cortices. The relationship between activity changes in dorsal parietal cortex and the rightward spatial bias was supported by two independent analyses. There was a relatively higher response in left SPL in subjects who detected fewer targets in the left visual field and in those who responded more slowly to left, as compared to right, visual field targets. Notably, the shape of the BOLD response, in left SPL, to missed targets was sustained and outlasted the response to detected targets (Fig. 5), suggesting that the orienting bias in SPL was tonic and endogenous. Tonic oculomotor biases that are independent of visual stimulation have been described in neglect38. The response in right visual cortex, especially at the acute stage, not only was decreased but also did not show a normal lateralization—that is, a stronger response to contralateral (left) than to ipsilateral (right) visual targets (Fig. 6). One interpretation of these results is that unbalanced top-down modulation from dorsal parietal cortex decreased both stimulus-evoked responses and spatial selectivity of visual neurons, weakening the relative salience of stimuli presented in the left visual field. That is, the rightward spatial bias may reflect both abnormal orienting mediated by imbalanced IPS-SPL activity and abnormal sensory processing of stimuli in the left visual field. However, as there was no clear relationship between BOLD response in visual cortex and rightward bias, the role of visual neurons in mediating the rightward bias will require further tests, such as the separation of cueand target-related activity or the correlation, trial by trial, of brain activity and behavioral performance. A second key deficit in spatial neglect is the inability to reorient to behaviorally relevant stimuli presented at unattended locations: the socalled disengage deficit30. Our subjects with neglect showed good recovery of reorienting, with faster and more accurate responses over time to unattended targets. Previous work correlated stimulus-driven reorienting with a right hemisphere–dominant ventral and dorsal network, including TPJ24. Here, we found that reorienting deficits and their recovery also correlated with functional changes in a similar network. In subjects with lesions restricted to ventral frontal cortex and related subcortical structures (Supplementary Figs. 2 and 3), right TPJ reactivated from the acute to the chronic stage (Supplementary Fig. 3), and this reactivation was modulated by whether the target was attended or unattended (Fig. 7). For targets presented at unattended locations, BOLD signals in right and left STG (a subregion of TPJ) were delayed at the acute stage as compared to the chronic stage (Fig. 7 and Supplementary Fig. 3). Moreover, at the acute stage, subjects with stronger STG activity responded more slowly to targets at unattended locations. One interpretation is that signals in the TPJ indexed the time it takes to reorient to a novel location of interest, a process that was delayed at the acute stage. A new anatomical model of spatial neglect These results provide a new framework for thinking about the pathophysiology of spatial neglect and reconcile functional neuroimaging results with the anatomy of neglect. Ventral lesions in frontal or temporoparietal cortex12–14 cause dysfunction of dorsal parietal areas that seem to mediate a rightward bias during spatial attention. However, isolated damage to these dorsal areas does not typically cause neglect, even though it can produce deficits of eye movements, attention and visuomotor hand coordination39,40. Therefore, damage (functional or structural) to both dorsal and ventral attention networks is necessary for neglect to occur. This result rules out the possibility that spatial neglect results from the critical dysfunction of one brain area14.

VOLUME 8

[

NUMBER 11

[



ARTICLES The TPJ region is crucial because it provides a signal that marks sensory events of interest for the dorsal system, especially when they are unattended. Damage to TPJ produces two effects that contribute to neglect. First, it decreases the overall detection capacity—that is, the capacity across the visual field11. Second, it biases competitive interactions between orienting mechanisms in dorsal parietal cortex32. Therefore, a stimulus in the left visual field will be at a disadvantage, as compared to a stimulus in the right visual field, on two counts: (i) a decreased stimulus-driven capture resulting from damage of right TPJ and (ii) a top-down bias against exploring leftward locations, owing to imbalanced orienting mechanisms in left and right IPS. The idea that non-spatial processing deficits contribute to spatial neglect and may exacerbate spatial biases has been suggested before9,10, but we provide a new anatomical framework in which to think about these interactions. The right hemisphere dominance of spatial neglect has previously been explained by theories that emphasize hemispheric asymmetries in spatial maps, with right parietal cortex coding for both sides of space and left parietal cortex coding predominantly for the contralateral (right) space6,8,41. However, recent studies, in normal observers, that mapped visuotopic responses in frontal and parietal cortices have not revealed any hemispheric asymmetry in spatial representations or orienting signals22,23. In contrast, there is compelling evidence for a right hemisphere–dominant ventral attention system, including TPJ (reviewed in ref. 24). Therefore, our current hypothesis is that the higher frequency of left-sided neglect is a function of the right hemisphere dominance of non-spatial processes mediated by right TPJ, coupled with their physiological impact on ipsilateral spatial processes mediated by IPS- SPL. Implications regarding mechanisms of recovery of function These results show that a neurological deficit after focal brain injury does not reflect only local dysfunction at the site of injury, but also is determined by the distributed impairment of connected neural systems that are structurally intact2,3. This dysfunction may be reflected neurally—not just by diaschisis at rest, as shown in previous studies17,27,28—but also by deactivation, hyperactivity or interhemispheric imbalance during task processing, as shown here. Although this distributed impairment principle has been demonstrated here for spatial neglect, it is likely to apply to other deficits such as aphasia or sensory-motor deficits, and thus have widespread implications for the fields of neuropsychology and neurology. For example, the localization of specific neuropsychological syndromes on the basis of anatomical information should be re-examined by combining both anatomical and functional information. That a behavioral deficit reflects a distributed dysfunction does not imply that different nodes of a functional network do not perform specialized operations. The notion of distributed injury is neutral with respect to the issue of whether cognitive operations in the intact brain are carried out in specialized nodes (one-to-one mapping) or over many nodes (one-to-many), or whether different operations are mapped to the same node as a function of task demands (many-toone). Nonetheless, in our case the evidence strongly indicates a relative specialization of different nodes—as in the case of IPS-SPL for directing attention or TPJ for reorienting attention. The notion of competition between hemispheres and the negative influence of activity in the intact hemisphere is emerging as an important principle at the systems level to understand recovery of function, not only in spatial neglect, but also in studies of motor and language recovery42,43. Modulation of these competitive interactions either by increasing the excitability of the ipsilesional cortex or by


[

NUMBER 11

[

NOVEMBER 2005

decreasing the excitability of the intact cortex should have a beneficial effect44. For example, we predict that in individuals with chronic neglect who show a persistent rightward bias despite extensive rehabilitation, there should be persistent left SPL hyperactivation; reducing that hyperactivity should be beneficial. This hypothesis has been tested with some success in TMS studies that have broadly targeted the left parietal cortex of individuals with neglect45. Our results suggest a more specific site where TMS treatment might have a favorable therapeutic effect. METHODS Participants were eleven patients (mean age 60 years; 8 male) with right frontoparietal stroke and clinical neglect. Both behavioral testing and fMRI were conducted first in the acute stage (3–4 week post-stroke) and then at the chronic stage (46 months post-stroke). Individual lesions were segmented (in atlas space) using a supervised fuzzy class-means procedure on the basis of co-registered T1- and T2-weighted structural data acquired in the chronic stage. The Posner task was implemented during fMRI as previously described18,24. Details of functional scanning procedures (sequence parameters and data analysis techniques) are as previously described18,21. Additional technical details are given in the Supplementary Methods. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank G.L. Shulman for discussions and comments. Supported by the J. S. McDonnell Foundation, the J. S. McDonnell Center for Higher Brain Function and the National Institute of Neurological Disorders. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Caramazza, A. Some aspects of language processing revealed through the analysis of acquired aphasia: the lexical system. Annu. Rev. Neurosci. 11, 395–421 (1988). 2. Jackson, J.H. Evolution and dissolution of the nervous system. Br. Med. J. 1, 591, 660–703 (1884). 3. von Monakow, C. Lokalisation der hirnfunktionen [Localization of brain functions]. J. Psychol. Neurol. 17, 185–200 (1911). 4. Pedersen, P.M., Jorgensen, H.S., Nakayama, H., Raaschou, H.O. & Olsen, T.S. Hemineglect in acute stroke—incidence and prognostic implications. The Copenhagen stroke study. Am. J. Phys. Med. Rehabil. 76, 122–127 (1997). 5. Appelros, P., Karlsson, G.M., Seiger, A. & Nydevik, I. Neglect and anosognosia after firstever stroke: incidence and relationship to disability. J. Rehabil. Med. 34, 215–220 (2002). 6. Heilman, K.M., Bowers, D., Valenstein, E. & Watson, R.T. in Neurophysiological and Neuropsychological Aspects of Spatial Neglect (ed. Jeannerod, M.) 115–150 (NorthHolland, Amsterdam, The Netherlands, 1987). 7. Halligan, P.W. & Marshall, J.C. Toward a principled explanation of unilateral neglect. Special issue: the cognitive neuropsychology of attention. Cogn. Neuropsychol. 11, 167–206 (1994). 8. Mesulam, M.M. Spatial attention and neglect: parietal, frontal and cingulate contributions to the mental representation and attentional targeting of salient extrapersonal events. Phil. Trans. R. Soc. Lond. B 354, 1325–1346 (1999). 9. Robertson, I.H., Mattingley, J.B., Rorden, C. & Driver, J. Phasic alerting of neglect patients overcomes their spatial deficit in visual awareness. Nature 395, 169–172 (1998). 10. Husain, M. & Rorden, C. Non-spatially lateralized mechanisms in hemispatial neglect. Nat. Rev. Neurosci. 4, 26–36 (2003). 11. Peers, P.V. et al. Attentional functions of parietal and frontal cortex. Cereb. Cortex (2005). 12. Vallar, G. & Perani, D. in Neurophysiological and Neuropsychological Aspects of Spatial Neglect (ed Jeannerod, M.) 235–258 (North-Holland, Amsterdam, The Netherlands, 1987). 13. Mort, D.J. et al. The anatomy of visual neglect. Brain 126, 1986–1997 (2003). 14. Karnath, H.O., Ferber, S. & Himmelbach, M. Spatial awareness is a function of the temporal not the posterior parietal lobe. Nature 411, 950–953 (2001). 15. Karnath, H.O., Fruhmann Berger, M., Kuker, W. & Rorden, C. The anatomy of spatial neglect based on voxelwise statistical analysis: a study of 140 patients. Cereb. Cortex 14, 1164–1172 (2004). 16. Friedrich, F.J., Egly, R., Rafal, R.D. & Beck, D. Spatial attention deficits in humans: a comparison of superior parietal and temporal-parietal junction lesions. Neuropsychology 12, 193–207 (1998).

1609


ARTICLES 17. Hillis, A.E. et al. Anatomy of spatial attention: insights from perfusion imaging and hemispatial neglect in acute stroke. J. Neurosci. 25, 3161–3167 (2005). 18. Corbetta, M., Kincade, J.M., Ollinger, J.M., McAvoy, M.P. & Shulman, G.L. Voluntary orienting is dissociated from target detection in human posterior parietal cortex. Nat. Neurosci. 3, 292–297 (2000). 19. Connolly, J.D., Goodale, M.A., Menon, R.S. & Munoz, D.P. Human fMRI evidence for the neural correlates of preparatory set. Nat. Neurosci. 5, 1345–1352 (2002). 20. Astafiev, S.V. et al. Functional organization of human intraparietal and frontal cortex for attending, looking, and pointing. J. Neurosci. 23, 4689–4699 (2003). 21. Kincade, J.M., Abrams, R.A., Astafiev, S.V., Shulman, G.L. & Corbetta, M. An eventrelated functional magnetic resonance imaging study of voluntary and stimulus-driven orienting of attention. J. Neurosci. 25, 4593–4604 (2005). 22. Sereno, M.I., Pitzalis, S. & Martinez, A. Mapping of contralateral space in retinotopic coordinates by a parietal cortical area in humans. Science 294, 1350–1354 (2001). 23. Silver, M.A., Ress, D. & Heeger, D.J. Topographic maps of visual spatial attention in human parietal cortex. J. Neurophysiol. 94, 1358–1371 (2005). 24. Corbetta, M. & Shulman, G.L. Control of goal-directed and stimulus-driven attention in the brain. Nat. Rev. Neurosci. 3, 201–215 (2002). 25. Downar, J., Crawley, A.P., Mikulis, D.J. & Davis, K.D. A multimodal cortical network for the detection of changes in the sensory environment. Nat. Neurosci. 3, 277–283 (2000). 26. Macaluso, E., Frith, C.D. & Driver, J. Supramodal effects of covert spatial orienting triggered by visual or tactile events. J. Cogn. Neurosci. 14, 389–401 (2002). 27. Deuel, R.K. & Collins, R.C. The functional anatomy of frontal lobe neglect in the monkey: behavioral and quantitative 2-deoxyglucose studies. Ann. Neurol. 15, 521–529 (1984). 28. Vallar, G. et al. Recovery from aphasia and neglect after subcortical stroke: neuropsychological and cerebral perfusion study. J. Neurol. Neurosurg. Psychiatry 51, 1269– 1276 (1988). 29. Pizzamiglio, L. Recovery of neglect after right hemispheric damage: H215O positron emission tomographic activation study. Arch. Neurol. 55, 561–568 (1998). 30. Posner, M.I., Walker, J.A., Friedrich, F.J. & Rafal, R.D. Effects of parietal injury on covert orienting of attention. J. Neurosci. 4, 1863–1874 (1984). 31. Morrow, L.A. & Ratcliff, G. The disengagement of covert attention and the neglect syndrome. Psychobiology 16, 261–269 (1988).

32. Kinsbourne, M. in Hemi-inattention and Hemispheric Specialization (eds. Weinstein, E.A. & Friedland, R.L.) 41–52 (Raven Press, New York, 1977). 33. Kastner, S. & Ungerleider, L.G. Mechanisms of visual attention in the human cortex. Annu. Rev. Neurosci. 23, 315–341 (2000). 34. Rossini, P.M. et al. Does cerebrovascular disease affect the coupling between neuronal activity and local haemodynamics? Brain 127, 99–110 (2004). 35. Pineiro, R., Pendlebury, S., Johansen-Berg, H. & Matthews, P.M. Altered hemodynamic responses in patients after subcortical stroke measured by functional MRI. Stroke 33, 103–109 (2002). 36. Binkofski, F. & Seitz, R.J. Modulation of the BOLD-response in early recovery from sensorimotor stroke. Neurology 63, 1223–1229 (2004). 37. Luck, S.J., Hillyard, S.A., Mangun, G.R. & Gazzaniga, M.S. Independent hemispheric attentional systems mediate visual search in split-brain patients. Nature 342, 543–545 (1989). 38. Hornak, J. Ocular exploration in the dark by patients with visual neglect. Neuropsychologia 30, 547–552 (1992). 39. Lynch, J.C. & McLaren, J.W. Deficits of visual attention and saccadic eye movements after lesions of parietooccipital cortex in monkeys. J. Neurophysiol. 61, 74–90 (1989). 40. Perenin, M.T. & Vighetto, A. Optic ataxia: a specific disruption in visuomotor mechanisms. I. Different aspects of the deficit in reaching for objects. Brain 111, 643–674 (1988). 41. Pouget, A. & Driver, J. Relating unilateral neglect to the neural coding of space. Curr. Opin. Neurobiol. 10, 242–249 (2000). 42. Murase, N., Duque, J., Mazzocchio, R. & Cohen, L.G. Influence of interhemispheric interactions on motor function in chronic stroke. Ann. Neurol. 55, 400–409 (2004). 43. Heiss, W.D., Kessler, J., Thiel, A., Ghaemi, M. & Karbe, H. Differential capacity of left and right hemispheric areas for compensation of post-stroke aphasia. Ann. Neurol. 45, 430–438 (1999). 44. Naeser, M.A. et al. Improved picture naming in chronic aphasia after TMS to part of right Broca’s area: an open-protocol study. Brain Lang. 93, 95–105 (2005). 45. Brighina, F. et al. 1 Hz repetitive transcranial magnetic stimulation of the unaffected hemisphere ameliorates contralesional visuospatial neglect in humans. Neurosci. Lett. 336, 131–133 (2003).

1610

VOLUME 8

[

NUMBER 11

[



ARTICLES

Perceptions of moral character modulate the neural systems of reward during the trust game M R Delgado1, R H Frank2 & E A Phelps1,3 Studies of reward learning have implicated the striatum as part of a neural circuit that guides and adjusts future behavior on the basis of reward feedback. Here we investigate whether prior social and moral information about potential trading partners affects this neural circuitry. Participants made risky choices about whether to trust hypothetical trading partners after having read vivid descriptions of life events indicating praiseworthy, neutral or suspect moral character. Despite equivalent reinforcement rates for all partners, participants were persistently more likely to make risky choices with the ‘good’ partner. As expected from previous studies, activation of the caudate nucleus differentiated between positive and negative feedback, but only for the ‘neutral’ partner. Notably, it did not do so for the ‘good’ partner and did so only weakly for the ‘bad’ partner, suggesting that prior social and moral perceptions can diminish reliance on feedback mechanisms in the neural circuitry of trial-and-error reward learning.

The human striatum has been implicated as a critical structure in trialand-error feedback processing and reward learning. In particular, the caudate nucleus, a structure linked to learning and memory in both animals1–3 and humans4–6, has been shown to have a role in processing affective feedback7,8, with responses in this region varying according to properties such as valence and magnitude9,10. It has been shown that activation in the human caudate nucleus is modulated as a function of trial-and-error learning with feedback5,6,11,12. Activation in this region in response to reward feedback diminishes as, over time, cues begin to predict the correct action and outcome, thus making feedback less informative11. This has led to the hypothesis that the caudate nucleus may serve as a key component of an ‘actor-critic’ model processing the contingent behavior that led to the feedback, with the purpose of guiding future actions13–15. Recently, this role for the caudate in feedback processing has extended to social interactions in the economic domain using a repeated-interaction ‘trust’ game in which participants learned, through trial and error, whether their partners were trustworthy. As expected, activity in the caudate decreased over trials as feedback from the partners became more predictable16. Our study is motivated by the observation that not all choices are made on the basis of trial-and-error learning involving material rewards. Moral beliefs, for example, have been shown to influence economic choices. In one study, participants were more likely to cooperate in one-shot prisoner’s dilemmas if they perceived their partners as cooperative after a brief period of informal conversation17. There is also evidence that many individuals are willing to work for lower salaries if they believe their employer’s mission is morally praiseworthy18. To examine the influence of social and moral learning on choice, we used an iterated version of the trust game involving two-person

interactions in which mutually beneficial outcomes are more likely if partners are trustworthy and perceive one another as such19,20 (Fig. 1a). Participants could either keep $1 on a given trial or transfer it to a partner, in which case the partner would receive $3. The partner could either keep the entire $3 or give half of it back (‘sharing’). In previous experiments involving economic games, participants learn which partners are trustworthy through trial and error16,21,22. In our version of the game, participants were instructed that they would be playing with three fictional partners and were given detailed, vivid descriptions of each partner’s life events that indicated either praiseworthy, neutral or suspect moral character. Participants were told that the partners’ responses might or might not be consistent with the descriptions given. Presentation of the trust trials were intermixed with ‘lottery’ trials, in which participants merely decided whether they wanted to play or not play a lottery for a chance to earn $1.50 reward. Previous investigations of the reward circuitry and error-based learning15,18,23,24 suggest three possible hypotheses. One is that no differences will be observed because rational economic decision makers will view information about the moral characteristics of partners as uninformative. Alternatively, information about moral characteristics might create expectations, and failed predictions that are based on those expectations might lead to an increase in the magnitude of responses in the brain circuitry associated with trial-and-error learning. A third possibility is that the bias induced by information about moral characteristics will modulate the brain mechanisms associated with trial-and-error learning, making participants less reliant on feedback or willing to discount such information. The present results suggest that moral beliefs can affect economic decision making, as a result of modulations in the human caudate nucleus that may influence the adjustment of choices based on trial-and-error feedback.

1Department of Psychology and 3Center for Neural Science, 6 Washington Place, New York University, New York, New York 10003, USA. 2Johnson School of Management, Cornell University, Ithaca, New York 14853, USA. Correspondence should be addressed to E.A.P. ([email protected]).



[

NUMBER 11

[

NOVEMBER 2005

1611

ARTICLES Decision phase

Trust trial

a

Figure 1 Experimental design and behavioral results. (a) A trial in the game was divided into a decision phase and an outcome phase and was played with one of three possible partners (good, bad or neutral) or against the computer (lottery). (b) Trustworthiness ratings, pre- and postexperiment (± s.e.m.). Participants’ perception of moral character influenced trust ratings before experiment, which was adjusted after trial and error. (c) Behavioral choices during experimental sessions (± s.e.m.). Overall choices suggest participants were affected by the moral character of the fictional partner: rates of cooperation or sharing were higher when playing with the good partner, but not with the other partners.

Outcome phase Chris has decided to share the money

Chris has decided to keep the money

Chris Thompson Keep $1.00

Give $3.00 You have decided to keep the money

Lottery trial

You have not won the lottery

No

Yes You have decided not to play the lottery

3s

12 s

1s

12 s

Decision

Inter-trial interval

Outcome

Inter-trial interval

b Trust ratings pre- and post-experimental session

c

Overall behavioral choices

"How trustworthy do you feel this player is (1–7)" 7

0.8 0.7

5

Time of rating

4

Pre Post

3 2 1 0

Choice ratio (%)

6

Ratings


You have won the lottery

$

0.6 0.5 0.4 0.3 0.2 0.1

Good

Bad

Neutral

0

Good

Bad

RESULTS Behavioral results After reading each fictional partner’s biography, 12 participants gave trustworthiness ratings before each experimental session. These ratings confirmed that the manipulation affected the perceptions of trustworthiness (Fig. 1b). This was also reflected in choices: participants made more ‘share’ (that is, transfer of money) than ‘keep’ decisions when playing with the good partner (t11 ¼ 2.46, P o 0.05); however, no such difference was observed when participants played with the bad (t11 ¼ –1.42, P ¼ 0.18) or neutral (t11 ¼ –0.05, P ¼ 0.96) partners (Fig. 1c). In addition, when comparing share decisions between partners (good versus bad), participants made more share decisions overall when playing with the good partner than with the bad (t11 ¼ 3.26, P o 0.01) or neutral (t11 ¼ 2.0, P ¼ 0.07) partners. Finally, using reaction time data acquired during the experimental session, we observed that participants were faster to share when playing with the good partner compared to the bad (t11 ¼ –3.73, P o 0.005) and neutral partners (t11 ¼ –1.89, P ¼ 0.08). Thus, biasing the participants according to the perceived moral characteristics of partners influenced decision-making overall. As the experiment progressed, however, participants learned in some ways that the three partners were not responding quite in the manner suggested by their descriptions. For example, trustworthiness ratings acquired post-scanning differed from those acquired before scanning (Fig. 1b), as indicated in a two-way analysis of variance (ANOVA) by an interaction between partner and time of rating (F2,22 ¼ 11.78, P o 0.0001). Further, after the experiment, we asked participants what percentage of time they believed each partner shared with them (good: mean ¼ 44.58, s.d. ¼ 22.10; bad: mean ¼ 35.00, s.d. ¼ 18.71; neutral: mean ¼ 38.33, s.d. ¼ 19.11). Although the mean scores ordered according to acquired bias (that is, good 4 neutral 4 bad), no significant differences were observed when comparing the two extreme partners (good versus bad: t11 ¼ 1.01, P ¼ 0.34). However, in spite of this explicit knowledge of similar response patterns, an investigation of early (first two of eight runs, containing roughly six trials per partner) and late decisions (last two runs) showed that participants made more

1612

share decisions when playing with the good partner than with the bad partner, on both early (t11 ¼ 3.99, P o 0.005) and late Decision (t11 ¼ 2.18, P o 0.05) trials (see SupplemenKeep Share tary Fig. 1). This suggests that explicit knowledge of feedback probabilities may not be completely predictive of choice behavior for partners of different perceived moral character. Neutral To account for any asymmetry or special characteristics in partner biographies that may influence the results, we also conducted a separate trust game experiment with a separate set of participants and a new set of biographies, also depicted as morally good, bad and neutral (see Supplementary Note). This additional behavioral study replicated the previous results (Supplementary Fig. 2), further suggesting that moral and social characteristics can influence decision-making in an economic game. Neuroimaging results We conducted analyses of functional magnetic resonance imaging (fMRI) data separately for the outcome and decision phases. During Table 1 Brain areas activated during the outcome phase Talairach coordinates Region of activation

Voxels

Laterality

x

y

z

17 15

L R

–24 13

–5 8

59 1

Precuneus (7) Middle frontal gyrus (10/46)

116 43

R L

21 –42

–79 36

44 5

Caudate Putamen

222 32

R R

15 22

20 17

7 2

Caudate Ventral striatum

707 12

R R

13 16

8 0

1 –7

Inferior temporal gyrus (19) Orbitofrontal gyrus (11)

14 11

R R

55 29

–65 27

–2 –11

Parahippocampus gyrus Negative 4 positive feedback

11

L

–42

–34

–16

Insular cortex

14

R

54

8

13

Positive 4 negative feedback Medial frontal gyrus (6) Medial frontal gyrus (6)

The contrast of positive versus negative feedback yielded several regions defined by strength of effect (P o 0.001) and size (10 or more voxels). The reverse contrast yielded activation in one region. Brodmann’s areas are depicted in parentheses (under the Region of activation column) when applicable. Cluster size (number of voxels) and laterality (right or left hemisphere) are also given. The stereotaxic coordinates of the peak of the activation are given according to Talairach space. The region of interest discussed in the paper is in bold.

VOLUME 8

[

NUMBER 11

[


ARTICLES Table 2 Brain areas activated during the decision phase

Table 3 Brain areas activated during the decision phase contrasting bias-incongruent choices

Talairach coordinates Talairach coordinates Region of activation

Voxels

Laterality

x

y

z Laterality

x

y

Voxels 79

L

–29

–17

68

67 24

R L

2 –50

1 –20

57 55

R R

7 4

–73 8

44 47

35

L

–4

17

38

20 21

R R

2 49

–84 –32

32 25

Inferior parietal cortex (40) Middle occipital gyrus (19)

14 20

L L

–49 –21

–37 –88

20 17

Insular/frontal cortex Insular cortex

17 40

L L

–63 –58

–6 5

10 7

Insular cortex Insular cortex

21 69

L R

–47 49

14 20

3 2

119 64

L R

–31 6

–67 –78

–22 –25

32

L

–66

–34

26

Precentral gyrus (4/6)

Insular cortex Lingual gyrus (18)

43 15

L R

–47 6

18 –70

10 –13

Medial frontal gyrus (6) Precentral gyrus (4)

112 26

L L

–29 –36

–16 –79

3 –2

Precuneus (7) Cingulate cortex (32/6)

57 12

Ventral striatum

26

R

21

6

–4

Cingulate cortex (32)

Fusiform gyrus (19) Fusiform gyrus (19)

34 27

R R

12 22

–64 –67

–7 –11

Cuneus (18/19) Inferior parietal cortex (40)

Keep 4 share decisions Perirhinal cortex

20

L

–35

1

–25

Perirhinal cortex

45

R

32

–6

–28

The contrast of share versus keep decisions yielded several regions defined by strength of effect (P o 0.001) and size (10 or more voxels). The reverse contrast yielded activation in a few regions. Specific annotations in the table have been previously described. The region of interest discussed in the paper is in bold.

the outcome phase, we conducted a random-effects general linear model (GLM) analysis using each condition (good, bad, neutral and lottery outcome trials) and associated outcome (positive and negative feedback) as predictors. We generated statistical maps contrasting positive and negative feedback. Positive feedback referred to a monetary gain of $1.50 because the partner either cooperated (that is, ‘shared back’ the money) or the participant hit the lottery; negative feedback referred to no monetary gain ($0) either because the partner defected (that is, kept the money) or the participant did not hit the lottery. During the decision phase, we conducted a random-effects GLM analysis using each condition (good, bad and neutral decision trials) and the associated decision (share and keep) as predictors (for lottery trials, play decisions were also used as a predictor). We generated statistical maps (Tables 1–3) contrasting share and keep decisions. From contrasts generated during the outcome and decision phases, regions of interest (ROIs) in the striatum were identified based on peak activity, and a priori analyses for individual conditions were performed on each ROI using mean beta weights for each predictor. Outcome phase During the outcome phase, the contrast between positive (monetary gain) and negative (monetary loss) feedback yielded activation in the striatum, which was most robust in the ventral caudate nucleus (Fig. 2a and Table 1). We defined an ROI centered on the peak of this caudate activation. We then conducted additional analyses to investigate the effects of social and moral bias on feedback processing within this striatum ROI. First, we conducted a repeated-measures ANOVA using

Cerebellum Cerebellum

participants’ mean beta weights. This analysis probed an interaction between moral character (good and bad versus neutral) and outcome (positive versus negative feedback). A significant interaction was observed (F1,10 ¼ 5.13, P o 0.05), suggesting that perceived moral character influenced the underlying neural activity involved in feedback processing (Fig. 2b). To further explore the source of this interaction, we conducted two ANOVAs comparing the good and bad partner separately with the neutral partner. There was a significant interaction in the response to positive and negative feedback between the good and neutral partner (F1,10 ¼ 5.8, P o 0.05; Fig. 2c). Although a similar pattern was observed for the bad partner, this interaction did not reach significance (F1,8 ¼ 0.49, P ¼ 0.51; Fig. 2d).

a

b

Neutral × Moral (Good & Bad)

15 –4.40 8.00

5 0 –5 –10

[

NUMBER 11

[

NOVEMBER 2005

–15

P < 0.001

15

Outcome phase: Neutral × Good

10

Type of Condition Neutral Good Bad

5 0 –5 –10 –15 Share

Keep Partner decision

d Parameter estimate (β)

c

Type of Condition Neutral Moral (Good & Bad)

10

t(11)


Outcome phase:

–8.00

4.40

Figure 2 Caudate nucleus activation during outcome phase. (a) When contrasting positive and negative outcomes were collapsed across all conditions, activity was observed in the ventral portion of caudate nucleus (x, y, z ¼ 13, 8, 1). (b) Repeated-measures ANOVAs using mean beta weights extracted from caudate nucleus ROI showed a significant interaction between moral character (good and bad versus neutral) and outcome (positive versus negative feedback). Separate ANOVAs investigated the source of the interaction by comparing the good and bad partner separately with the neutral partner. (c,d) A significant interaction between moral character and outcome was observed when comparing the neutral partner (c) with the good partner but (d) not with the bad.

z

The contrast of decisions that were incongruent with behavioral bias (share with bad partner and keep with good partner versus the alternative choices) yielded several regions defined by strength of effect (P o 0.001) and size (10 or more voxels). Specific annotations in the table have been previously described. Region of interests discussed in the paper are in bold.

Parameter estimate (β)

Putamen Inferior occipital gyrus (18)

Parameter estimate (β)


Region of activation Share 4 keep decisions Inferior parietal cortex (40)

Share Keep Partner decision

Outcome phase: Neutral × Bad

15 10 5 0 –5 –10 –15

Share

Keep Partner decision

1613

ARTICLES Decision phase During the decision phase, activation of the ventral striatum (nucleus accumbens and ventral putamen) was observed when contrasting share and keep decisions (Table 2 and Fig. 4). We did not observe activity in the caudate nucleus, in agreement with the T5 T6 T7 T5 T6 T7 T1 T2 T3 T4 T1 T2 T3 T4 idea that although this area may be more Time (scans) Time (scans) involved with processing feedback to guide Partner decision / Lottery outcome future behavior13,14, ventral portions of the Share / Win Keep / Lose striatum may be more important for making predictions14,26 and anticipating the outc 0.25 d Outcome phase: 'bad' partner Outcome phase: 'good' partner 0.25 comes of risky decisions27,28. We used mean 0.2 0.2 0.15 0.15 beta weights extracted from this ROI in an 0.1 0.1 ANOVA similar to the one described in the 0.05 0.05 0 0 outcome phase to probe an interaction –0.05 –0.05 –0.1 –0.1 between moral character (good and bad ver–0.15 –0.15 sus neutral) and decision (share versus keep). –0.2 –0.2 T5 T6 T7 T5 T6 T7 T1 T2 T3 T4 T1 T2 T3 T4 No interaction was observed with this analysis Time (scans) Time (scans) (F1,11 ¼ 0.22, P ¼ 0.65). However, exploratory analysis of each individual condition sugFigure 3 Time course of activation (± s.e.m.) of caudate nucleus ROI during the outcome phase for each condition. Time points are referred to as T1–T7 and correspond to 2 s each. (a–c) Time course data gested that some differences may exist. For shows differential responses between positive and negative feedback during lottery trials, (b) during example, when subjects were playing with the interactions with the neutral partner and (c) with the bad partner, to a lesser extent. (d) No differences bad partner, mean beta weights for share were observed with the good partner. decisions were significantly higher than those for keep decisions (t11 ¼ 2.64, P o 0.05), and lottery play decisions (t11 ¼ 2.00, Additional post-hoc investigations of the mean beta weights in the P o 0.05). For the neutral partner, we observed a trend both when ventral caudate ROI for each of the four individual conditions showed comparing beta weights for share and keep decisions (t11 ¼ 1.71, that positive outcomes were significantly different from negative out- P ¼ 0.06) and when comparing them for share and lottery play decisions comes for lottery (t11 ¼ 3.22, P o 0.005) and neutral partner (t10 ¼ (t11 ¼ 1.70, P ¼ 0.06). We found no differences when participants were 5.03, P o 0.0005) trials, consistent with previous studies showing this playing the good partner in either comparison (share keep: region differentiates between positive and negative outcomes8,25 t10 ¼ 1.29, P ¼ 0.11; share lottery: t11 ¼ 0.40, P ¼ 0.35). Although (Fig. 3). Less differentiation was observed for trials with the bad the individual conditions analysis supports the idea that prior moral partner (t9 ¼ 1.57, P ¼ 0.08). As expected from the interaction analysis, and social beliefs can influence the reward circuitry, the results must be no significant differences were observed for trials involving the good interpreted cautiously because of the lack of overall interaction between partner (t11 ¼ 1.34, P ¼ 0.11). An additional analysis on percent signal moral character and decision. Behavioral results showing that participants were more likely to change values showed that any difference observed between positive and negative feedback was maximal at 6–9 s after feedback delivery, share when playing with the good partner and keep when playing with consistent with previous studies of reward processing8,25 (see Supple- the bad partner are consistent with the induced-bias manipulation. Given this, we posited that areas involved in cognitive control and mentary Note).

a

a


Outcome phase: 'neutral' partner

0.25 0.2 0.15 0.1 0.05 0 –0.05 –0.1 –0.15 –0.2


Percent signal change Percent signal change

b

–8.00

c

Decision phase: 'neutral' partner

Decision phase: 'bad' partner

–4.40 8.00

0.3

Player decision Share Keep

0.2 0.1 0 –0.1 –0.2 T1

4.40 t(11)

Right ventral striatum (x, y, z = 21, 6, –4) Percent signal change

Right ventral striatum (x, y, z = 21, 6, –4) Percent signal change

T2

T5 T3 T4 Time (scans)

T6

0.3

Player decision

0.2 0.1

Share Keep

0 –0.1 –0.2

T7

T1

T2


T6

T7

P < 0.001

d Figure 4 Ventral striatum activation during the decision phase. (a) When contrasting share and keep decisions collapsed across all conditions, activity was observed in the ventral striatum (x, y, z ¼ 21, 6, –4). (b,c) Time course of activation (± s.e.m.) for all three partners shows differences between ‘share’ and ‘keep’ decisions observed during interactions with the (b) neutral and (c) bad partners. (d) No differences were observed with the good partner, according to exploratory mean beta weights comparisons.

1614

VOLUME 8

[

Decision phase: 'good' partner Percent signal change


b

Outcome phase: lottery trial

0.25 0.2 0.15 0.1 0.05 0 –0.05 –0.1 –0.15 –0.2

Right ventral striatum (x, y, z = 21, 6, –4)

0.3

Player decision

0.2 0.1

Share Keep

0 –0.1 –0.2

NUMBER 11

T1

[

T2


T6

T7


ARTICLES

a

b

–8.00

c

Decision phase: 'neutral' partner

–4.40 8.00

4.40

0.5 0.4 0.3 0.2 0.1 0 –0.1 –0.2


T1

T2


T6

T7

Player decision

0.5 0.4 0.3 0.2 0.1 0 –0.1 –0.2

Share Keep

T1

d

T2


T6

T7

Decision phase: 'good' partner Cingulate cortex: BA 32 (x, y, z = –4, 17, 38)

Figure 5 Cingulate cortex activation during the decision phase. (a) Activation of cingulate and insular cortex. Contrast of bias-incongruent decisions (that is, share with the bad partner and keep with the good partner). (b–d) Time course of activation (± s.e.m.) for cingulate cortex (Brodmann’s area 32; x, y, z ¼ –4, 17, 38). No differences between ‘share’ and ‘keep’ decisions are observed for (b) the neutral partner, but such differences are observed for (c) good and (d) bad partners, consistent with bias-incongruent behavioral decisions.

conflict evaluation would be recruited more strongly when participants made bias-incongruent choices (that is, keep with the good partner and share with the bad partner) compared with the alternative decision (Table 3). As predicted, the contrast of bias-incongruent decisions yielded activation in the cingulate and bilateral insular cortex (Fig. 5), regions previously implicated in cognitive control, conflict monitoring and fairness of decisions29–31. For the cingulate cortex, we used the extracted mean beta weights in a repeated-measures ANOVA with moral character (good, bad or neutral) and decisions (share and keep) as factors. We found an interaction between moral character and decision (F2,20 ¼ 4.92, P o 0.05), supporting the idea that the cingulate cortex is involved in resolving moral conflicts32. DISCUSSION In the current study, participants played a trust game with three hypothetical partners depicted as being of good, bad or neutral moral character. The perception of moral characteristics biased preexperimental self-ratings of trust and behavioral choices as participants chose to be more cooperative with the morally good partner. In addition, this bias was successful in modulating neural structures underlying reward and feedback learning. Specifically, activation in the caudate nucleus was observed during the outcome phase for the neutral partners and lottery trials, showing differential responses between positive and negative feedback. These results are consistent with the suggestion that the caudate nucleus is processing feedback information5,7,8, especially when the feedback is behaviorally relevant33,34, with the purpose of learning and adapting choices through trial and error13,14. The neutral partner results are also consistent with previously observed activation in the caudate nucleus when contrasting reciprocated and unreciprocated cooperation during an economic game in which participants had no prior knowledge about their trading partners35. In contrast, the differential response in the caudate nucleus was either not as robust (bad partner) or nonexistent (good partner) when participants had an expectation about the trial on the basis of the partner’s perceived moral character. This was especially interesting during good partner interactions, where the number of share choices remained larger than in other conditions, even though frequent violations of expectations occurred. In a rational choice framework23, the finding that feedback processing in the human caudate nucleus and subsequent behavior are altered by perceptions of social and moral characteristics might seem puzzling, because individuals are assumed to always be alert to the possibility that exchange partners will behave opportunistically. In Bayesian terms, the


[

NUMBER 11

[

NOVEMBER 2005



t(11)

P > 0.001

Decision phase: 'bad' partner Cingulate cortex: BA 32 (x, y, z = –4, 17, 38)



Cingulate cortex: BA 32 (x, y, z = –4, 17, 38)

0.5 0.4 0.3 0.2 0.1 0 –0.1 –0.2


T1

T2


T6

T7

character profiles may create a prior belief but feedback should also adjust this prior belief to reflect new evidence. The results suggest that the good profile not only creates a prior belief, but also disrupts the regular encoding of evidence, or learning, from surprising keep outcomes. Further, the lack of differential responses between positive and negative outcomes when playing with the morally good partner stands in contrast to error-prediction learning during trial-and-error association tasks24,36. This learning hypothesis would predict a sharp decrease in the feedback response following violations of expectations, a finding that has been observed in different parts of the human striatum37,38, more recently in the caudate nucleus39. Although the coding of a teaching signal in the caudate nucleus, reflected during feedback processing, was apparent when subjects were playing with the morally neutral partner and, to a lesser extent, with the bad partner, it was not observed during trials with the good partner. Participants instead seemed either less reliant on feedback information or discounted this information. These results indicate that perceptions of moral characteristics can influence choices and the neural mechanisms involved in feedback processing in trial-and-error learning. During the decision phase, activation was observed in the more ventral portions of the striatum when contrasting risky share decisions with keep choices. This is consistent with the idea that although the dorsal striatum is more involved in processing feedback contingent on choices13,14, the human ventral striatum has been primarily linked with a role in making predictions14,26 and anticipating the outcomes of risky decisions27,28. In this study, the riskier decisions involved the bad and neutral trials, both of whose mean beta scores suggested a differential response between share and keep. This activation was not merely due to anticipation of a reward, as mean beta scores for share decisions were also higher than those for lottery play in which participants also anticipated a potential gain of the same monetary value. As in the outcome phase, no differential responses were observed for the good partner, perhaps owing to a smaller expectation of risk, further supporting the idea that moral and social perceptions can influence choices. An alternative hypothesis, however, is that participants had a reward reaction to the presentation of the morally good partner, irrespective of decision. This could occur for two potential reasons. First, in previous studies of economics games, activation of the ventral striatum has been reported during cooperation21 and in response to pictures of previous cooperators40, suggesting that participants possibly experienced the presentation of the good partner, or the idea of transferring funds to him, as

1615


ARTICLES studies are necessary to understand why certain behaviors are affected by a sense of moral obligation. Self-interest alone cannot explain, for example, why people would leave tips in a restaurant they will never visit again. The present study suggests that moral and social perceptions can modulate neural mechanisms associated with feedback and reward processing and cognitive control, thereby influencing our day-to-day choices.

rewarding in itself. A second possibility is that the reward response elicited by the good partner is owing to the participant’s experience of possible revenge with a keep decision, because this partner violated trust on half the trials41. Although it is unclear what specifically the ventral striatum is responding to (that is, anticipation or prediction of risky decision or reward response owing to presentation of face or opportunity for revenge), reward-related activation in this region may also be modulated by moral and social perceptions. Our manipulation of the perception of moral characteristics was evident in the behavioral choices, as participants chose to share more with the morally good partner and keep more with the morally bad partner. This bias suggests that cognitive mechanisms involved in conflict processing are recruited when subjects are making a biasincongruent choice, such as choosing to keep when playing the good partner. Such contrast revealed activation in a number of areas (Table 3), including the following: the cingulate cortex, previously linked to conflict monitoring29, cognitive control29,31 and resolving difficult moral conflict that may involve personal violations of values32; and bilateral insular cortex, implicated in general uncertainty and arousal responses42 and fairness of decisions30. Taken together, these results suggest that although moral information can modulate rewardand feedback-based learning systems in the human brain, the neural circuitry involved in conflict monitoring processes remain unaltered, as participants’ analysis of difficult moral decisions are consistent with previous conflict monitoring studies. In the current experiment, participants’ choice behavior in an iterated version of the trust game was influenced by a previous bias regarding moral character. The data suggest that a signal linked to trial-and-error feedback processing, which serves to adapt and optimize choices, was observed in the caudate nucleus during conditions where no information (that is, lottery trials) or little relevant information (that is, neutral partner trials) was available. However, the availability of prior information about moral character diminished the differential signal observed in the caudate. This led participants to discount feedback information and not adapt their choices accordingly, despite showing declarative evidence of learning (as indicated by participant’s pre- and post-experiment trustworthiness self-ratings). What remain unanswered are the following questions: (i) what neural mechanisms are linked to this modulation of the teaching signal observed in the caudate nucleus, and (ii) what neural mechanisms are underlying the learning as expressed by changes in the explicit ratings of trustworthiness? Although the current study cannot provide conclusive answers to these questions, it is clear from both the human and animal literature that there are multiple, interacting systems linked to learning and memory2,43–46 and that different tasks rely on these systems to varying degrees. In particular, it has been suggested that the caudate feedback learning mechanism may be less involved in tasks where declarative, hippocampal-dependent knowledge can be used to guide behavior2,44. This raises the possibility that the lack of observed differential activity in the striatum, as well as the declarative learning shown at the end of the experiment (pre- and post-experiment self-ratings), are hippocampally mediated. We did not observe activation in the hippocampus in our analysis at the thresholds selected for significance. This is perhaps due to the fact that the design used was not optimized to observe hippocampal-dependent learning throughout the task. Additional experiments will be needed to further investigate this potential interaction between the caudate feedback learning system and the hippocampus-declarative memory system during economic interactions. Moral assessment is a complex domain that needs to be considered when investigating human interactions and decision-making. Future

Behavioral analysis. During the behavioral session, we analyzed data pertaining to reaction time and choice (‘keep’ or ‘share’ decision). We used paired t-tests to compare the percentage of time share and keep decisions were made, for each partner individually and also compared decisions between partners. We conducted similar analyses to compare reaction time data. Finally, to determine

1616

VOLUME 8

METHODS Participants and procedure. Fourteen right-handed volunteers participated in this study (8 male, 6 female; average age: mean ¼ 26.64, s.d. ¼ 4.11). One participant was removed from further analysis because of scanner malfunction, and another because of a failure to comprehend instructions (as assessed through post-experimental forms, behavioral results and self-assessment). Data acquired from the 12 remaining participants was included in the analysis. Participants responded to posted advertisements and gave informed consent according to the New York University’s Committee of Activities Involving Human Subjects. Participants were instructed that they were playing a trust game and that they would be playing the three hypothetical partners described in three separate biographies (see Supplementary Note). Each partner was introduced with a picture (white, neutral, male faces taken from the NimStim Face Stimulus Set; N. Tottenham, A. Borscheid, K. Ellertsen, D.J. Marcus & C.A. Nelson, Cogn. Neurosci. Soc. Abstr. 2002) and a biography describing his characteristics and a recent noteworthy event. Two of the biographies were constructed so as to depict partners with exaggerated moral qualities: one praiseworthy (an English graduate student and volunteer inner-city teacher who rescued a friend from a fire during a crowded concert), and the other suspect (a business graduate student who had been arrested for trying to sell tiles of the space shuttle Columbia on an Internet auction site). The third biography described an engineering graduate student who narrowly missed a doomed flight, but contained no information relevant for assessing his moral character. Participants were instructed that the fictional partners may or may not play according to their described personality or moral characteristics. In fact, each partner had the same 50% reinforcement rate. Each trial proceeded as follows (Fig. 1). The trial was divided into a decision phase and an outcome phase. During the decision phase, participants viewed the name and face of the partner and the options to keep or share, for 3 s followed by a 12-s interval. During the outcome phase, one of three possible outcomes was displayed, indicating the following: (i) the participant chose to keep, (ii) the partner chose to keep or (iii) the partner chose to share. The feedback was presented for 1 s, followed by a 12-s interval. Participants were told they would also play trials involving a lottery game, intermixed with trust trials. In the lottery, participants viewed a dollar sign and chose between ‘No’ and ‘Yes’ for a chance to gain $1.50. There was no penalty for losing the lottery. Thus, as expected, ‘No’ responses were never recorded. There were 96 interleaved trials, divided into 8 runs of 12 trials each. Participants played with each partner 24 times and played 24 lottery trials. Before the scanning session, participants filled out a seven-point Likert-scale questionnaire, rating the perceived trustworthiness of the three partners. The same questionnaire and additional assessments of learning were administered following the scanning session. After the experimental session was complete, participants were debriefed and paid according to performance (the monetary sum acquired in four out of eight blocks of trials, chosen randomly). Although trial order was predetermined, outcome and feedback were contingent on performance and varied between participants. There were two counterbalanced trial orders, and all three pictures used to represent the partners’ faces were counterbalanced across the study. Trials in which a response was not made in time carried a monetary penalty of $1.00 and were excluded from further analysis. Stimulus presentation and behavioral data acquisition were controlled by a Macintosh computer with PsyScope software47.

[

NUMBER 11

[


ARTICLES


how participants adapted their choices as the session progressed, we performed an analysis of early decisions (trials included in the first two fMRI runs, roughly six per partner) and late decisions (last two fMRI runs), using paired t-tests. For the questionnaire data (acquired using a seven-point scale), we conducted a repeated-measures ANOVA using participants as a random factor (n ¼ 12), with type of partner (good, bad and neutral) and time of rating (preand post-experiment) as within-subject factors. We also used paired t-tests to analyze the final post-experiment question regarding the percentage of time participants believed each partner shared with them. fMRI acquisition and analysis. We used a 3-Tesla Siemens Allegra scanner to collect structural (T1-weighted MPRAGE: 256 256 matrix; FOV ¼ 256 mm; 176 1-mm sagittal slices) and functional images (single-shot gradient echo EPI sequence; TR ¼ 2000 ms; TE ¼ 25 ms; FOV ¼ 192 cm; flip angle ¼ 801; matrix ¼ 64 64; slice thickness ¼ 3 mm). Forty contiguous oblique-axial slices (3 3 3 mm3 voxels) parallel to the anterior commissure–posterior commissure (AC-PC) line were obtained. fMRI data was analyzed using Brain Voyager software. Preprocessing included motion correction (six-parameter, threedimensional motion correction), spatial smoothing (4-mm FWHM), voxelwise linear detrending, high-pass filtering of frequencies (3 cycles per time course) and normalization to Talairach stereotaxic space48. We performed random-effects analyses on the functional data for the decision and outcome phase separately. For the outcome phase, we defined a general linear model (GLM) that included eight regressors: two outcomes (positive or negative) for each of four situations (good, bad, neutral or lottery). We conducted a similar GLM during the decision phase that included seven regressors: two decisions (keep or share) for each of three partners (good, bad or neutral), and a lottery play trial. Statistical maps were created using a threshold of P o 0.001 with a cluster threshold of 10 voxels49. The primary contrasts of interest was the differential response between all positive and negative feedback during the outcome phase, and the differential response between all share and keep choices during the decision phase. A secondary analysis contrasted bias-incongruent (for example, keep with good partner) and bias-congruent choices during the decision phase. These overall contrasts yielded several regions, including striatum ROIs. We derived statistics from each functional ROI after identifying the peak of activation and surrounding voxels encompassing 10 mm3. During the outcome phase, a striatum ROI was present in the right hemisphere and contained two separate peaks. The largest peak of activation was observed in the ventral head of the caudate nucleus (x, y, z ¼ 13, 8, 1), which was similar in location to other recent reward-learning studies11,12. We then used mean beta weights extracted from this ROI in a repeated-measures ANOVA that probed the interaction between moral character (good and bad versus neutral) and outcome (positive versus negative feedback). To explore the source of the interaction, we conducted two additional ANOVAs comparing the good and bad partner separately with the neutral partner. We further investigated these ROIs using mean beta weights for each predictor, which were compared using one-tailed paired t-tests. Finally, we plotted time-series data for target ROIs averaged over trials for each individual playing partner and the lottery. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS The authors wish to acknowledge K. Nearing for assistance and useful discussion, J. Pearson and E. Neaville for technical assistance and S. Ravizza for informative discussion and constructive criticism. This work was supported by the US National Institute of Mental Health, the James S. McDonnell Foundation, the Beatrice and Samuel A. Seaver Foundation and Cornell’s S.C. Johnson School of Management. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. White, N.M. Mnemonic functions of the basal ganglia. Curr. Opin. Neurobiol. 7, 164– 169 (1997).


[

NUMBER 11

[

NOVEMBER 2005

2. Packard, M.G. & Knowlton, B.J. Learning and memory functions of the basal ganglia. Annu. Rev. Neurosci. 25, 563–593 (2002). 3. Takikawa, Y., Kawagoe, R. & Hikosaka, O. Reward-dependent spatial selectivity of anticipatory activity in monkey caudate neurons. J. Neurophysiol. 87, 508–515 (2002). 4. Shohamy, D. et al. Cortico-striatal contributions to feedback-based learning: converging data from neuroimaging and neuropsychology. Brain 127, 851–859 (2004). 5. Poldrack, R.A. et al. Interactive memory systems in the human brain. Nature 414, 546– 550 (2001). 6. Filoteo, J.V. et al. Cortical and subcortical brain regions involved in rule-based category learning. Neuroreport 16, 111–115 (2005). 7. Elliott, R., Sahakian, B.J., Michael, A., Paykel, E.S. & Dolan, R.J. Abnormal neural response to feedback on planning and guessing tasks in patients with unipolar depression. Psychol. Med. 28, 559–571 (1998). 8. Delgado, M.R., Nystrom, L.E., Fissell, C., Noll, D.C. & Fiez, J.A. Tracking the hemodynamic responses to reward and punishment in the striatum. J. Neurophysiol. 84, 3072– 3077 (2000). 9. Cromwell, H.C. & Schultz, W. Effects of expectations for different reward magnitudes on neuronal activity in primate striatum. J. Neurophysiol. 89, 2823–2838 (2003). 10. Delgado, M.R., Locke, H.M., Stenger, V.A. & Fiez, J.A. Dorsal striatum responses to reward and punishment: effects of valence and magnitude manipulations. Cogn. Affect. Behav. Neurosci. 3, 27–38 (2003). 11. Delgado, M.R., Miller, M.M., Inati, S. & Phelps, E.A. An fMRI study of reward-related probability learning. Neuroimage 24, 862–873 (2005). 12. Haruno, M. et al. A neural correlate of reward-based behavioral learning in caudate nucleus: a functional magnetic resonance imaging study of a stochastic decision task. J. Neurosci. 24, 1660–1665 (2004). 13. Tricomi, E.M., Delgado, M.R. & Fiez, J.A. Modulation of caudate activity by action contingency. Neuron 41, 281–292 (2004). 14. O’Doherty, J. et al. Dissociable roles of ventral and dorsal striatum in instrumental conditioning. Science 304, 452–454 (2004). 15. Barto, A.G. Adaptive critics and the basal ganglia. in Models of Information Processing in the Basal Ganglia (eds. Houk, J.C., Davis, J.L. & Beiser, D.G.) 215–232 (MIT Press, Cambridge, Massachusetts, 1995). 16. King-Casas, B. et al. Getting to know you: reputation and trust in a two-person economic exchange. Science 308, 78–83 (2005). 17. Frank, R.H., Gilovich, T. & Regan, D. The evolution of one-shot cooperation. Ethol. Sociobiol. 14, 247–256 (1993). 18. Frank, R.H. What Price the Moral High Ground? (Princeton Univ. Press, Princeton, New Jersey, 2004). 19. Camerer, C.F. & Weigelt, K. Experimental tests of a sequential equilibrium reputation model. Econometrica 56, 1–36 (1988). 20. Berg, J., Dickhaut, J. & McCabe, K. Trust, reciprocity, and social history. Games Econ. Behav. 10, 122–142 (1995). 21. Rilling, J. et al. A neural basis for social cooperation. Neuron 35, 395–405 (2002). 22. Glimcher, P.W. & Rustichini, A. Neuroeconomics: the consilience of brain and decision. Science 306, 447–452 (2004). 23. Coleman, J.S. Foundations of Social Theory (Belknap Press, Cambridge, Massachusetts, 1990). 24. Schultz, W. & Dickinson, A. Neuronal coding of prediction errors. Annu. Rev. Neurosci. 23, 473–500 (2000). 25. Delgado, M.R., Stenger, V.A. & Fiez, J.A. Motivation-dependent responses in the human caudate nucleus. Cereb. Cortex 14, 1022–1030 (2004). 26. O’Doherty, J.P. Reward representations and reward-related learning in the human brain: insights from neuroimaging. Curr. Opin. Neurobiol. 14, 769–776 (2004). 27. Knutson, B., Adams, C.M., Fong, G.W. & Hommer, D. Anticipation of increasing monetary reward selectively recruits nucleus accumbens. J. Neurosci. 21, RC159 (2001). 28. Breiter, H.C., Aharon, I., Kahneman, D., Dale, A. & Shizgal, P. Functional imaging of neural responses to expectancy and experience of monetary gains and losses. Neuron 30, 619–639 (2001). 29. Carter, C.S. et al. Anterior cingulate cortex, error detection, and the online monitoring of performance. Science 280, 747–749 (1998). 30. Sanfey, A.G., Rilling, J.K., Aronson, J.A., Nystrom, L.E. & Cohen, J.D. The neural basis of economic decision-making in the ultimatum game. Science 300, 1755–1758 (2003). 31. Ridderinkhof, K.R., Ullsperger, M., Crone, E.A. & Nieuwenhuis, S. The role of the medial frontal cortex in cognitive control. Science 306, 443–447 (2004). 32. Greene, J.D., Nystrom, L.E., Engell, A.D., Darley, J.M. & Cohen, J.D. The neural bases of cognitive conflict and control in moral judgment. Neuron 44, 389–400 (2004). 33. Zink, C.F., Pagnoni, G., Martin-Skurski, M.E., Chappelow, J.C. & Berns, G.S. Human striatal responses to monetary reward depend on saliency. Neuron 42, 509–517 (2004). 34. Elliott, R., Newman, J.L., Longe, O.A. & William Deakin, J.F. Instrumental responding for rewards is associated with enhanced neuronal response in subcortical reward systems. Neuroimage 21, 984–990 (2004). 35. Rilling, J.K., Sanfey, A.G., Aronson, J.A., Nystrom, L.E. & Cohen, J.D. Opposing BOLD responses to reciprocated and unreciprocated altruism in putative reward pathways. Neuroreport 15, 2539–2543 (2004). 36. Schultz, W., Dayan, P. & Montague, P.R. A neural substrate of prediction and reward. Science 275, 1593–1599 (1997). 37. 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). 38. O’Doherty, J.P., Dayan, P., Friston, K., Critchley, H. & Dolan, R.J. Temporal difference models and reward-related learning in the human brain. Neuron 38, 329–337 (2003).

1617


ARTICLES 39. Davidson, M.C. et al. Differential cingulate and caudate activation following unexpected nonrewarding stimuli. Neuroimage 23, 1039–1045 (2004). 40. Singer, T., Kiebel, S.J., Winston, J.S., Dolan, R.J. & Frith, C.D. Brain responses to the acquired moral status of faces. Neuron 41, 653–662 (2004). 41. de Quervain, D.J. et al. The neural basis of altruistic punishment. Science 305, 1254–1258 (2004). 42. Critchley, H.D., Mathias, C.J. & Dolan, R.J. Neural activity in the human brain relating to uncertainty and arousal during anticipation. Neuron 29, 537–545 (2001). 43. Gabrieli, J.D. Cognitive neuroscience of human memory. Annu. Rev. Psychol. 49, 87–115 (1998). 44. Squire, L.R., Knowlton, B. & Musen, G. The structure and organization of memory. Annu. Rev. Psychol. 44, 453–495 (1993).

45. Poldrack, R.A. & Packard, M.G. Competition among multiple memory systems: converging evidence from animal and human brain studies. Neuropsychologia 41, 245–251 (2003). 46. Zola-Morgan, S. & Squire, L.R. Neuroanatomy of memory. Annu. Rev. Neurosci. 16, 547–563 (1993). 47. Macwhinney, B., Cohen, J. & Provost, J. The PsyScope experiment-building system. Spat. Vis. 11, 99–101 (1997). 48. Talairach, J. & Tournoux, P. Co-planar Stereotaxic Atlas of the Human Brain: An Approach to Medical Cerebral Imaging (Thieme Medical Publishers, New York, 1988). 49. Forman, S.D. et al. Improved assessment of significant activation in functional magnetic resonance imaging (fMRI): use of a cluster-size threshold. Magn. Reson. Med. 33, 636– 647 (1995).

1618

VOLUME 8

[

NUMBER 11

[



TECHNICAL REPORT

A hybrid approach to measuring electrical activity in genetically specified neurons Baron Chanda1,4, Rikard Blunck1,4, Leonardo C Faria2, Felix E Schweizer3, Istvan Mody2 & Francisco Bezanilla1

The development of genetically encoded fluorescent voltage probes is essential to image electrical activity from neuronal populations. Previous green fluorescent protein (GFP)-based probes have had limited success in recording electrical activity of neurons because of their low sensitivity and poor temporal resolution. Here we describe a hybrid approach that combines a genetically encoded fluorescent probe (membrane-anchored enhanced GFP) with dipicrylamine, a synthetic voltage-sensing molecule that partitions into the plasma membrane. The movement of the synthetic voltage sensor is translated via fluorescence resonance energy transfer (FRET) into a large fluorescence signal (up to 34% change per 100 mV) with a fast response and recovery time (0.5 ms). Using this twocomponent approach, we were able to optically record action potentials from neuronal cell lines and trains of action potentials from primary cultured neurons. This hybrid approach may form the basis for a new generation of protein-based voltage probes. To understand the mechanisms underlying information processing by the central nervous system, it is necessary to track the activity of an ensemble of neurons. Although electrophysiological techniques remain the method of choice, they are unsuitable for recording from more than a few neurons at a time. Optical imaging using potentiometric and Ca2+-sensitive dyes has been used to monitor electrical activity of neuronal populations with single-neuron resolution1–4. These dyes, unlike electrophysiological methods, have high spatial resolution in addition to high temporal resolution. As a result, activity even at the level of individual synapses can be imaged noninvasively1,5. Nevertheless, a significant limitation of the dye-based methods is the indiscriminate staining of both neuronal and non-neuronal cell types, which reduces the signal-to-background ratios considerably. Moreover, to dissect the circuit elements of a neuronal network, it is important to target the fluorescent probes to a genetically distinct class of neurons. One particularly attractive approach is to use genetically encoded fluorescent probes to monitor neuronal activity6,7. To detect action potentials, a number of GFP-based sensors have been constructed by inserting a GFP molecule into various parts of voltage-gated Na+ or K+

channels. A reporter protein called FlaSH was generated by inserting the GFP into the C terminus of the S6 helix of the shaker K+ channel8, but the kinetics of the FlaSH response are quite slow, as they correlate mostly with C-type inactivation. In the SPARC protein, a GFP molecule was inserted between the linkers of domains I and II of the rat skeletal muscle Na+ channel9. A third type of reporter protein termed VSFP1 was generated by attaching a cyan and a yellow fluorescent protein in tandem to a truncated K+ channel10. Probes based on voltage-gated ion channels have some significant shortcomings. Voltage-gated ion channels undergo conformational changes over a narrow voltage range, which restricts the response range of these attached probes. In addition, these probes typically have modest fractional fluorescence changes since the GFP chromophore is buried within a beta barrel structure. Finally, many of these GFP-tagged, voltage-gated ion channels tend to be trapped in internal compartments, thereby increasing the background fluorescence. A variety of two-component voltage sensors based on FRET have been described previously, from fixed fluorescent donors to translocating oxonol acceptors11. The measured fluorescence ratio changes vary from 34% per 100 mV (t ¼ 0.38 ms) using fluorescent lectins11 to 0.3% using transmembrane GFP (R.W. Friedrich et al., Soc. Neurosci. Abstr., 293.11, 1999). This technique successfully imaged activity from multiple individual neurons in leech nerve cords during locomotor activity, though to achieve dye penetration in this more intact preparation, a slower-translocating oxonol with a time constant of 0.4 s was used12. In comparison, the time constant of a widely used one-component dye like di-8-ANEPPS is about 2 ms at room temperature, with signals up to 22% per 100 mV13,14. In this paper, we describe a two-component system that is genetically encoded and enables the imaging of fast membrane potential changes in cells and neurons. This hybrid voltage sensor (hVOS) consists of a synthetic voltage-sensing molecule, dipicrylamine (DPA), and a genetically encoded fluorophore, farnesylated enhanced GFP (eGFP-F). DPA, which translocates from one side of the membrane to the other in response to potential changes, can also absorb energy from eGFP via FRET, even though it is nonfluorescent. Therefore, fluorescence from eGFP, which is attached to the inner leaflet of the plasma membrane through a farnesylation tag15, becomes a genetically encoded optical readout for membrane potential changes. This hybrid approach gives a

1Departments of Physiology and Anesthesiology, 2Department of Neurology and 3Department of Neurobiology, David Geffen School of Medicine at the University of California Los Angeles (UCLA), Los Angeles, California 90095, USA. 4These authors contributed equally to this work. Correspondence should be addressed to F.B. ([email protected]).



[

NUMBER 11

[

NOVEMBER 2005

1619

TECHNICAL REPORT

b

Relative intensity

1.0

α-CD8

0.8

CD8

0.6 Depolarization

0.4

Hyperpolarization

0.2 0.0


c

440

eGFP-F

480

520 560 Wavelength (nm)

eGFP-F

600

e

d 10% ∆F/F 10 ms

Normalized fluorescence

a

f

1.0 0.8

50 ms

0.6 0.4 0.2 –200 mV

–100 mV

0 mV

100 mV

0.0 –150 –100

–50

0

50

100

150

Voltage (mV)

Figure 1 The hVOS system in HEK 293 cells. (a) Overlap (shaded gray) of eGFP emission (black) and DPA absorption (light gray) leads to resonance energy transfer. (b) Upper panel, labeled anti-CD8 antibody bound to CD8 in the plasma membrane. In the hyperpolarized state, DPA molecules (orange dumbbell) accumulate on the outer leaflet of the membrane, dimming the fluorescence of antibody-bound dye as a result of FRET. Upon depolarization, translocation of DPA to the inner leaflet reduces FRET, which results in an increased dye fluorescence. Lower panel, eGFP-F molecules are anchored to the cytoplasmic face of the membrane. Hence, depolarization in the presence of acceptor DPA molecules quenches the eGFP fluorescence. (c) Confocal section of HEK293 cells expressing eGFP-F shows very little fluorescence from internal membranes. Scale bar, 20 mm. (d) Fluorescence response (shown as colored increments) of hVOS on voltage pulses from 120 to +120 mV in steps of 20 mV from a holding potential of 0 mV in patch-clamped HEK293 cells. These measurements, which are single sweeps, were carried out at room temperature in the presence of 6 mM DPA. (e) Normalized fluorescence voltage characteristic of hVOS in HEK293 cells. (f) Fluorescence response of hVOS in HEK293 cells on a voltage ramp from 200 mV to +100 mV (0.86 mV/ms).

large fractional fluorescence change (maximal DF/F ¼ 34% per 100 mV) with a fast response time (time constant of 500 ms). We demonstrate that the hVOS probe can be used to measure trains of action potential spikes in primary neurons and is therefore a promising alternative to currently available probes for imaging neuronal activity.

Upon depolarization, DPA molecules translocate from outside to inside with a time constant of several hundred microseconds. As a consequence, the distance between the antibody and the DPA increases, resulting in unquenching of the antibody fluorescence. Conversely, when the donor fluorophore is anchored to the inner leaflet, as with eGFP-F, depolarization results in quenching of donor fluorescence. eGFP-F was generated by fusing a 20-amino acid farnesylation sequence derived from c-Ha-Ras protein to the C-terminal end of eGFP. This sequence provides farnesylation and palmitoylation signals that target the host protein to the inner leaflet of the plasma membrane.

RESULTS A genetically encoded two-component system for voltage sensing Our dual-component system for voltage sensing is based on changes in FRET between a synthetic voltage-sensing molecule and a genetically encoded fluorescent reporter. In our system, the voltage sensor is a charged hydrophobic molecule—DPA—that tends to partition into the lipid membrane close to the lipid water interface. DPA is a nonfluorescent absorber with an absorption maximum of 420 nm and has a significant spectral overlap with the emission of green fluorescent probes (Fig. 1a). DPA can therefore be excited through energy transfer by a green fluorescent label in close proximity; the calculated R0 for an eGFP-DPA energy transfer pair is 37 A˚. The extent of energy transfer for this pair can only be followed by measuring donor emission because DPA is fluorescently silent. Depending on the membrane potential, the negatively charged DPA molecules are distributed between the outer and inner leaflet of the plasma membrane. Changing the membrane potential results in rapid redistribution of DPA between the two membrane leaflets. This relocation changes the average distance between DPA molecules and an immobile reporter that is restricted to one leaflet of the membrane, thus changing the FRET efficiency. The effect on FRET for the two different reporter groups due to potential change is schematically shown in Figure 1b. The reporter moiety is a fluorescently labeled (Alexa-488) antibody to CD8. At resting membrane potential, most of the DPA molecules are closer to the outer leaflet of the membrane, so the emission of the antibody fluorophore is quenched.

hVOS signals are highly sensitive to voltage changes In general, two-component systems, on which the hVOS is based, show a much higher sensitivity than the traditional potentiometric dyes, in which a single chromophore interacts with the membrane field. Since the hVOS is based on FRET changes between an immobile donor and a mobile acceptor, its sensitivity depends on the distance between the donor and the acceptor, as well as on their R0, the distance of 50% transfer efficiency, which can be estimated from the spectral overlap (Supplementary Note online). We used eGFP-F, which is a known plasma membrane marker, as the immobile donor. A confocal section of HEK293 cells transfected with eGFP-F showed that most of the fluorescence was localized in the plasma membrane (Fig. 1c). The fluorescence was confined almost exclusively to the plasma membrane, with very little or no eGFP-F fluorescence from internal membranes. Under voltage-clamped conditions, single fluorescence sweeps of the hVOS system in HEK293 cells showed a large stepwise response (Fig. 1d). The maximum fractional fluorescence change (DF/F) measured for a 100-mV pulse (from 0 to 100 mV) was 34%, making the measured fluorescence signal one of the largest among both fast and slow response dyes. Widely used potentiometric dyes, such as

1620

VOLUME 8

[

NUMBER 11

[


TECHNICAL REPORT

tON (ms)

DF/F

hVOS FlaSH

0.53 10.0

5–34% 2.2%a

SPARC VSFP1

0.8–7 0.5% 0.7 1.8%

Cutoff frequency

Dynamic range (mV)

Basis

0.67 kHz 150 +50 eGFP-F + DPA 0.02 kHz 80 to +50 50 to +30 Shaker-GFP/CFP Z0.2 kHz ?

80 o140

10 +200

SkM1-GFP DKv2.1-YFP-CFP

Shown are the characteristics of known genetically encoded fluorescent voltage sensing probes9,10,30. tON gives the time constant of the fluorescence response to a step pulse. DF/F is the maximal normalized fluorescence change per 100 mV membrane potential change. Cutoff frequency indicates the corner frequency, up to which the system responds with 90% of the maximal change. The dynamic range marks the boundaries of the fluorescence response. The voltage sensors are based on constructs listed in the ‘basis’ column. aThis value was 5% for long depolarizations.

di-8-ANNEPS, which is used to follow fast potential changes, show fractional changes in the range of 5–20% per 100 mV in cells and hemispherical bilayers14,16. The steady-state fluorescence–voltage relationship could be fitted to a single Boltzmann curve with V1/2 ¼ 56.3 mV and z ¼ 0.57 (Fig. 1e). This is consistent with a two-state model in which the majority of the DPA molecules are distributed between the outer and the inner leaflet of the membrane. The apparent valence estimated from the Boltzmann fits of the fluorescence data was only 0.5 elementary charges, e0, whereas the apparent valence from gating current measurements was approximately 1e0 (ref. 17). This is not unexpected, since the fluorescence response in FRET has a nonlinear dependence on distance. It is also likely that the charged group in the DPA molecule moves across the full extent of the electric field, whereas the larger chromophoric group moves across only a fraction of the electric field. The voltage dependence of the fluorescence change over the entire voltage range was evoked by a slow ramp between 200 mV to +100 mV (Fig. 1f). The fluorescence signals saturated at potentials below 180 mV and above +50 mV. Thus, the hVOS system has a large dynamic voltage range and has an almost linear response over the entire physiologically relevant voltage range. Speed and frequency response of the hVOS In the neuron, an action potential is typically completed within a few milliseconds. To be able to follow this fast electrical activity, it is necessary for the reporter system to respond quickly and with high Figure 2 Speed and recovery of hVOS. (a) Temporal resolution of hVOS as determined from fluorescence response (gray, noisy) to a voltage pulse from 60 mV to +40 mV in HEK293 cells. The trace was well fit to a monoexponential decay (black, smooth) and the DF/F was 20% per 100 mV for this cell in the presence of 6 mM DPA. All data shown in this figure are single sweeps. (b) Fluorescence response of hVOS (bottom, gray) to a train of voltage pulses (top, black) from 60 mV to +40 mV in HEK293 cells. The duration of the voltage pulses was 0.93 ms and the repetition frequencies were 416 Hz (top), 179 Hz (center) and 88 Hz (bottom). The data were filtered at 5 kHz. (c) Frequency response plot of hVOS obtained by varying the interpulse interval while keeping the pulse duration constant (0.93 ms).

fidelity to changes in membrane potential. A monoexponential fit of the time course of fluorescence signal in response to a 100-mV voltage pulse gave a time constant of 0.54 ms (Fig. 2a). This is comparable to other voltage-sensitive reporter proteins such as VSFP1 (t ¼ 0.8 ms) and SPARC (t ¼ 1.5 ms), but much faster than FlaSH (t ¼ 10–20 ms) (Table 1). The response speed of hVOS is rate-limited by the speed of the DPA moving through the plasma membrane and therefore may depend on membrane parameters like fluidity and lipid order. In squid axons, the speed of DPA translocation, estimated from gating current measurements, has been reported to be at least twofold faster than measured here18. Therefore, the speed of the response in the hVOS system may vary within certain bounds with different cell types and the physicochemical state of the membranes. To test the ability of the dye system to follow trains of action potentials, we recorded the fluorescence response of hVOS by applying pulses of an amplitude (60 to +40 mV) and duration (0.93 ms) matching those of a typical action potential with a varying interpulse interval (Fig. 2b). Our data indicate that shortening the interpulse interval to less than 1 ms attenuates the fluorescence signal. These data sets are plotted as a frequency response curve of hVOS (Fig. 2c). The cutoff frequency estimated by fitting the data was 667 Hz. For spike trains discharging at frequencies higher than 667 Hz, the amplitude diminished according to the fourth power of the input frequency. At frequencies higher than the cutoff, the hybrid voltage sensor will act as an integrator and the entire train will be detected as a single event. Even up to train frequencies of 750 Hz, the fluorescence still changed 90% of the maximal DF/F as an integrated event. In most excitable tissues, the trains of action potentials have maximal frequencies up to 250 Hz with action potential durations of 1–2 ms19. They should therefore be detectable by the hVOS system. Toxicity of hVOS As DPA is a synthetic chromophore in the hVOS system, it is likely to have some phototoxicity20. In addition, it may have some intrinsic toxicity in neurons and brain slice preparations which may severely limit its use. We tested the phototoxicity of hVOS by repetitively pulsing the transfected HEK293 cells to a test potential in presence of DPA and while exposed to light. We used an increase in leak current as a measure of cell viability. In fluorescence recordings during cumulative exposure to light, the amplitude of the signal remained unchanged for

a

b 5% ∆F/F

40 mV –60 mV

2 ms

40 mV

c

–60 mV 0.20 0.18

10% ∆F/F

0.16

∆F/F per 100 mV


Table 1 Comparison chart of genetically encoded voltage sensors

5 ms

0.14

40 mV

0.12 0.10

–60 mV

0.08 0.06 0.04 0.1

0.5

1

frequency (kHZ)


[

NUMBER 11

[

NOVEMBER 2005

1621

TECHNICAL REPORT

a

b

0.45 0.40

Normalized leak current

46 s

73 s

–40 mV

–40 mV

0.30 0.25 0.20 0.15 0.10 0.05

–100 mV 10 ms

0.00 0

c

200 400 600 Exposure time (s)

800

d Control DPA

DPA

Superimposed

4 µM

1.5

Normalized fEPSP


120 s

0.35

1.0 10 µM

0.5 4 µM 10 µM 20 µM

0.0

with reduced axonal excitability due to an increased capacitive load induced by high DPA concentrations. Thus, by keeping the concentration at or lower than 4 mM and the total light exposure to less than 200 s, we can minimize the side effects of DPA on neuronal recordings.

1 mV 10 ms 20 µM

–10

0

10 Time (min)

20

30

durations of up to 120 s of exposure (Fig. 3a). The plot of leak current versus cumulative light exposure shows very little phototoxic effect of DPA on cells for exposures of up to 100 s (Fig. 3b). Exposure beyond 200 s resulted in exponential reduction of cell viability, setting upper bounds for total light exposure. Even if DPA is not significantly phototoxic, it is expected to increase the membrane capacitance, which may result in inhibition of action potentials. In addition, DPA is a charged hydrophobic ion and thus may have other toxic effects, either on action potentials or on synaptic transmission in live central neurons. We measured the amplitude of extracellularly recorded population spikes evoked by Schaffer collateral or commissural fiber stimulation in acute adult mouse hippocampal slices at 35 1C. The field EPSPs (fEPSPs) and population spikes were unaffected by 4 mM DPA, a concentration similar to that used for measurements in cultured cells (Fig. 3c,d). At 10 mM, DPA slightly enhanced the slope of the fEPSPs but left population spikes unaffected. A significant reduction in the evoked responses was observed when the DPA concentration was raised from 10 mM to 20 mM. This is consistent

Figure 4 Recording of action potentials with hVOS in GT1 cells. (a,b) Expression of eGFP-F in GT1 cells. (a) Confocal section of GT1 cells expressing eGFP-F. (b) Overlay of multiple confocal sections of cultured GT1 cells. Scale bars in a,b, 10 mm. (c,d) Fluorescence responses of hVOS on action potentials in GT1 cell culture in current-clamped cells measured at 32 1C with 2 mM of DPA. Cells were held at a resting potential of 100 mV and pulsed to 50 mV, where the action potential (black voltage trace) started to develop autonomously. The fluorescence response (gray trace) showed a normalized fluorescence change of 3% for 58 mV measured from the foot of the spike. The fluorescence trace was digitally filtered at 100 Hz after acquisition. (d) The trace was recorded with an expanded time resolution and digitally filtered at 200 Hz after acquisition.

1622

Figure 3 Toxicity of hVOS system. (a,b) To test phototoxicity, we exposed HEK293 cells transfected with eGFP-F in the presence of DPA (420 mM) with very high light intensities and determined the leak current as a measure of cell viability. In a, single sweeps of fluorescence response after different cumulative exposure are shown. Leak current with increasing excitation light exposure is shown in b. (c) Normalized CA1 field EPSP slope at the indicated DPA concentrations. The bar indicates the time DPA was added to the perfusion chamber. The values are normalized to the average slope of the fEPSPs recorded during the 10 min prior to DPA perfusion. Each concentration is shown as mean ± s.e.m. from 8 experiments carried out at 35 1C. (d) Effect of DPA on the shape of the population spikes recorded in the CA1 pyramidal cell layer. The traces depict population spikes evoked in a control ACSF and in the presence of the indicated concentration of DPA. The right panels show superposition of the two traces.

Recording of action potentials in neurons The ultimate test of an optical voltage reporting system is its ability to follow an action potential in neurons or neuronal cell lines. We initially transfected the neuronal cell line GT1 with eGFP-F21. A confocal section of GT1 cells transfected with eGFP-F and an overlay of all the sections are shown in Figure 4a,b. The eGFP-F was distributed mainly throughout the plasma membrane, although a significantly larger fraction of eGFP-F labeling is seen in the internal membranes in GT1 cell lines than in HEK293 cells. To stimulate action potentials in GT1 cells, the cells were depolarized to 50 mV by injecting a positive current to cells current-clamped at 90 mV (Fig. 4c,d). The time course of the measured fluorescence spike followed the time course of the action potential. The DF/F of the action potential (48 mV 10 mV) was 3% (DF/F ¼ 5.2% per 100 mV). For the measurement of action potentials, the DPA concentration was 2.2 mM instead of 6 mM

a

c

20 mV

–100 mV 2% ∆F/F 50 ms

b

d

50 ms

100 mV

3% ∆F/F

VOLUME 8

[

NUMBER 11

[


system noise. We took the average of eight individual traces by aligning the peak of the action potential spike, which clearly improved the signal-to-noise ratio from 1.3 to 4.9 (Fig. 6c). Thus, small spikes should also be easily detectable in a low-noise recording system, such as in confocal spot detection or 2-photon microscopy. A histogram showing a distribution of different neuronal spikes versus the measured signal over background fluorescence revealed that in 70% of the measured signals, the DF/F was within 3–6% per 100 mV change in potential (Fig. 6d).

Figure 5 Primary neurons transfected with farnesylated eGFP. Overlay of two confocal sections of eGFP-F transfected neurons from rat hippocampus. Note that the label is mainly confined to the plasma membrane, as in HEK293 cells, but considerable fluorescence is visible in the internal membranes in primary neurons. Scale bar, 50 mm.

used in voltage-clamp conditions, which may be the primary reason for the diminished signal. Increasing the DPA concentration adds capacitive load on the surface membrane, which results in aborted action potentials. Increasing the temperature alleviates this problem, however, because the kinetics of the Na+ current will be faster, which should result in completion of the action potential. The measurements described here were carried out at 32 1C. Increased temperature and higher Na+ channel density in neurons may allow the cells to fire even in presence of higher DPA concentrations. The critical test for every optical detection system for neuronal activity is to record action potentials directly in a preparation from native neurons. Therefore, we transfected 7-d-old primary rat hippocampal cultures with eGFP-F DNA (Fig. 5). In the presence of 3-mM DPA, electrical activity and fluorescence changes were recorded from transfected neurons (Fig. 6a). The action potentials were resolved in the fluorescence traces with high time resolution. The mean DF/F of the fluorescence change of hVOS in primary neurons was 4.2% per 100 mV change and was comparable to that measured in GT-1 cells. The background instrument noise was 1%, so an action potential spike of 50 mV was easily visible (Fig. 6b). With decreasing spike amplitudes (25–30 mV), however, the signals were difficult to distinguish from

a

b

25 ms

50 mV

A modified hVOS with fluorescently labeled antibody The hVOS system described here is not restricted to the combination of DPA with eGFP-F. In fact, any immobile membrane-anchored molecule whose emission spectrum overlaps with DPA absorption can be used as an energy transfer donor to DPA. Whereas eGFP-F has to be expressed using cell-specific promoters, cell-specific membrane antibodies are readily available for many neuronal cell types. We transfected HEK293 cells with CD8 and labeled them with anti-CD8 antibody conjugated to Alexa488. The distribution of the fluorescent antibody in the HEK293 cells shows surface labeling as well as some internal labeling, as would be expected when the antibody recycles to internal compartments (Fig. 7a). Voltage change ranging from 200 mV to 100 mV (from a holding potential of 150 mV) causes a fast fluorescence signal that correlates with translocation of DPA (Fig. 7b). The steady-state fluorescence voltage relationship (Fig. 7c) could be fitted to a single Boltzmann curve with V1/2 ¼ 51 mV and z ¼ 0.58. The smaller fractional fluorescence changes (DF/F ¼ 1.3% per 100 mV) may be due to the longer distance between DPA and the antibody compared to the DPA/eGFP-F pair. As FRET has a nonlinear dependence on distance, an energy transfer pair at distances much larger than R0 will not show any distance-dependent fluorescence change. A possible advantage of using a fluorescent antibody is that one is not restricted to genetically encoded dyes. The fluorescence readout can easily be tuned to measure at a different wavelength, for different colorcoded antibodies (Fig. 7a,b) against cell-specific molecules. For example, the DPA absorption spectrum also has a significant overlap with red emission (R0 ¼ 20 A˚), so red-labeled antibodies as well as redshifted GFP derivatives can also be used to monitor membrane potential changes. The activity of identified subpopulations of cells within a single neuronal circuit can thus be monitored.

c

10 ms

25 mV

20 ms

50 mV

d Number of neuronal spikes


TECHNICAL REPORT

14 12 10 8 6 4 2 0

2

3

4 5 6 7 ∆F/F per 100 mV

8

9

Figure 6 Optical recordings of electrical activity in primary neuronal cultures. Electrical and optical activity was recorded from primary neuronal cultures derived from rat hippocampus. The neurons were transfected with eGFP-F and the recordings were done under current-clamp conditions in the presence of 3-mM DPA at 32 1C. (a) Optical response of the hVOS system to single action potentials elicited by current injection (gray, optical response; black, voltage). The fluorescence was filtered at 250 Hz. (b) Optical response of the hVOS system to a series of spontaneous action potentials (colors and filter as in a). (c) Averaging single sweeps improve the signal-to-noise ratio. Optical response of the hVOS system to eight individual action potentials filtered at 500 Hz (lower traces) and the averaged optical response (center trace, blue). The corresponding voltage spike is shown in red. Note that the second subthreshold depolarization is detectable in the optical recordings. (d) Distribution of the normalized fluorescence change per 100 mV (n ¼ 42, DF/F (mean) ¼ 4.2% per 100 mV, s.d. ¼ 1.8).


[

NUMBER 11

[

NOVEMBER 2005

1623

TECHNICAL REPORT

a

b 1% ∆F/F


c

Normalized fluorescence

10 ms

1.0 0.8 0.6 0.4 0.2 0.0 –300 –200 –100 0 100 Voltage (mV)

200

300

Figure 7 Fluorescence recordings of membrane potential using DPA in conjunction with fluorescently labeled anti-CD8 antibodies. (a) Confocal section through CD8-expressing HEK293 cell culture labeled with anti-CD8Alexa647 antibody. Scale bars, 20 mm. (b) Fluorescence response of CD8expressing HEK293 cells labeled with anti-CD8-Alexa488 in the presence of DPA. Cells were pulsed from 150 mV to potentials ranging from 200 mV to +100 mV in steps of 20 mV. The fluorescence was filtered at 5 kHz. The data shown are single sweeps. (c) Fluorescence voltage characteristics for the data shown in b. The data were fitted to a single-component Boltzmann curve, with V1/2 ¼ 51 mV and z ¼ 0.58.

activity. The relative permeability coefficient of DPA with respect to potassium (PDPA/PK) is estimated to be 100. At 5 mM DPA, its contribution to the equilibrium resting potential is less than 2 mV. Thus, at concentrations used for optical measurements, the DPA does not shift the equilibrium potentials significantly. Nevertheless, the addition of DPA increases the capacitive load on the membranes, which may lead to abortive firing or subthreshold activity in the neurons. The calculated increase in membrane charge is 8,300e0 (DPA molecules) per mm2 at a concentration of 2 mM, which corresponds to a 70% increase in membrane capacitance. It is worth noting that a similar increase in membrane capacitance is also produced by ion channel–based voltage indicators. In this case, every ion channel that inserts into the membrane adds up to 12–13e0, compared to 1e0 for each DPA molecule. A typical expression level of voltage-sensitive ion channels used in previous demonstrations of genetically encoded probes adds approximately 23,000 charges per mm2—a threefold larger charge per unit area than that produced by DPA22. More pertinently, the capacitive load added to the membrane by the charged DPA molecules is minimized because of its shallow voltage dependence, which distributes the charge movement over a 150-mV voltage range. In comparison, most of the charges in an ion channel move over a 40-mV voltage range, adding a substantial capacitive load close to the threshold of an action potential. Recordings of field potentials in hippocampal slices (at 35 1C) indicate that the effect of increased capacitance (or some unknown toxicity) is manifested only at concentrations five times higher than those used to measure action potential spikes in neurons.

DISCUSSION In order to image electrical activity in neurons, a putative genetically encoded voltage-sensing fluorescent probe has to fulfill several requirements. The probe should not only have a fast response time, but it should also have a short refractory period and a large linear fluorescence response, especially in the voltage range of a typical action potential. And finally, at working concentrations, the probe components must not be toxic to the organism or tissue under investigation. The hybrid approach combining a genetically encoded fluorescent reporter protein (eGFP-F) with a synthetic voltage-sensing molecule fulfils many of the above criteria. The hVOS system shows robust fluorescence changes up to 34% DF/F per 100 mV with an apparent valence of 0.5e0 estimated from Boltzmann fits to the fluorescencevoltage relationship. As a result of the shallow fluorescence voltage curves, the hVOS system has a large dynamic range (150 to +50 mV), which spans the entire voltage range seen during neuronal activity. The response time constant of the hVOS system is 0.53 ms, which is sufficiently fast to record repetitive electrical activity in neuronal populations (Fig. 6). Although the use of a nonfluorescent acceptor like DPA does not allow measurements in ratiometric mode, there are a number of advantages to estimating voltage change by monitoring just donor fluorescence (see Supplementary Note). These include low background fluorescence due to the lack of direct excitation of the acceptor and the choice of a wide range of chromophores to optimize the magnitude of the signal. Furthermore, the actual gain in signal-tonoise ratios due to ratiometric measurements for FRET change are seen only in a limited range of conditions (Supplementary Note). The use of DPA as a permeable ion may raise the concern that it could influence the membrane potential. Previously it has been shown that 1-h incubation in the presence of 10-mM DPA does not affect the resting membrane potential of the squid giant axon18. The contribution of a permeable ion to the equilibrium potential can be calculated using the Goldman-Hodgkin-Katz equation using its permeability and

The genetic versus hybrid approach to voltage sensing The hybrid approach, which combines genetic and synthetic components, represents a major departure in probe design compared to other protein-based voltage probes. The performance of the hybrid system compares favorably with other genetically encoded voltage probes (Table 1). Previous genetically encoded fluorescent voltage sensors were based on the voltage sensor of voltage-sensitive ion channels. VSFP1 and hVOS both cover the entire membrane potential range of neuronal activity. In contrast, FlaSH and SPARC have steep voltage dependencies, which will result in a digital response upon crossing a threshold, thus being especially useful to amplify small signals. The large dynamic range seen in VSFP1 and in hVOS, on the other hand, enables quantification of the size of voltage response. Most importantly, using the hybrid approach, we were able to optically measure an action potential in a neuron using a genetically encoded probe. Another important reason for the increased fluorescence response of the hybrid system compared to the other genetically encoded probes may be the use of farnesylated GFP. For a typical transmembrane protein, the surface-to-internal distribution is 40–60% on the surface23. Therefore, any GFP-fused transmembrane protein will significantly contribute to background fluorescence owing to the proteins recycling through internal membranes which do not see any membrane potential changes. eGFP-F, on the other hand, is synthesized as a soluble protein and is post-translationally modified by farnesyl and palmitoyl transferase enzymes. As a result, most of the eGFP-F remains primarily localized in the plasma membrane (Fig. 1c) where it contributes to optical readouts of the membrane potential.

1624

VOLUME 8

Outlook The real power of genetically encoded fluorescent voltage sensors is the possibility of imaging from a specific population of neurons by expressing the eGFP-F molecule under the control of a cell-specific promoter. Various proteins, including GFP constructs, have been

[

NUMBER 11

[



TECHNICAL REPORT successfully expressed in specific types of neurons24,25. An alternative approach would be to use fluorescently labeled antibodies instead of eGFP-F against cell-specific markers, as demonstrated here using antibody to CD8. Though the larger distance to the membrane reduces the size of the signals, the use of readily available antibodies might save a lot of effort. In summary, the hybrid approach reported here appears to be a promising step forward in our quest to image electrical activity from specific neuronal populations, and it opens the door to an new generation of genetically encoded probes. METHODS Chemicals and solutions. All chemicals were purchased from Sigma-Aldrich or VWR if not stated otherwise. The external solution for HEK293 cells consisted of 158 mM NaCl, 20 mM HEPES, 3 mM KCl and 1 mM CaCl2, with pH adjusted to 7.4 with NaOH. The internal solution contained 135 mM KCl, 20 mM KF, 20 mM Hepes, 3 mM NaCl, 1 mM MgCl2 and 2 mM EGTA, with pH adjusted to 7.4 with KOH. For current-clamp measurements in GT1 cells, the external solution contained 126 mM NaCl, 10 mM glucose, 2 mM MgCl2, 2 mM CaCl2, 2.5 mM KCl, 1.25 mM NaH2PO4 (pH adjusted to 7.2 with NaOH); the internal solution contained 120 mM KMeSO4, 10 mM KCl, 5 mM NaCl, 10 mM Hepes, 2 mM MgCl2, 0.1 mM EGTA, 2 mM Mg-ATP and 0.5 mM Na-GTP (pH adjusted to 7.2 with KOH). Molecular biology. The pEGFP-F vector was purchased from Clontech, then it was amplified and used without further modification. The vector includes an eGFP containing the 20-amino acid farnesylation signal from the c-Ha-Ras protein, targeting the eGFP to the plasma membrane15. A phi-H3 vector containing CD8 at the XhoI site was transiently transfected in HEK293 cells for the anti-CD8 antibody experiments. Cell culture and transfection. HEK293 cells were cultured following standard protocols in DMEM supplemented with 10% FBS, 100 U/ml penicillin and 100 mg/ml streptomycin at 37 1C and 5% CO2. GT1 cells were cultured using DMEM/F12 (containing equal proportions of DMEM and Ham’s F12) supplemented with 10% FBS, 5% horse serum, 100 U/ml penicillin and 100 mg/ml streptomycin at 37 1C and 5% CO2. Twelve hours before transfection, the cells were seeded on the coverslip of the recording chamber (f ¼ 10 mm) in a density of 105 cells per chamber. The chamber was prepared by drilling a hole in the middle of a 35-mm Petri dish and gluing a No.1 coverslip onto the hole. The coverslip was subsequently treated with poly-L-lysine and laminin (both from Sigma-Aldrich). The cells were transfected with 0.2–0.5 mg DNA mixed with 1 ml of Lipofectamine 2000 (Invitrogen) using standard procedures. Primary hippocampal neurons were cultured from newborn rat pups on Matrigel-covered coverslips as previously described26. After 7–10 days in vitro, neurons were transfected using calcium phosphate and imaged 12–48 h afterwards. Hippocampal slice recordings. Coronal hippocampal slices were obtained from adult C57Bl6 mice using standard techniques27. Briefly, mice were anesthetized with halothane and decapitated. The brains were cooled to 4 1C then rapidly removed, and whole brain slices (350 mm thick) were cut in the horizontal plane on a VT100S vibroslicer (Leica Microsystems). Slices were incubated in artificial cerebrospinal fluid (ACSF: 126 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 2.0 mM CaCl2, 2.0 mM MgCl2, 26 mM NaHCO3 and 10 mM D-glucose) for at least 1 h in a storage chamber at 32 1C and were then transferred to a recording chamber and continually perfused with the latter ACSF (2 ml/min, 35 1C) in an atmosphere of humidified 95% O2 and 5% CO2 Extracellular field excitatory postsynaptic potentials (fEPSPs) and population spikes were recorded as previously described28 in the CA1 pyramidal cell layer following stimulation in the stratum radiatum. Antibody labeling. For antibody binding, cells were incubated for 30 min in PBS with 5% donkey serum, and then for 2 h in PBS with 0.5% donkey serum and 1 mg/ml anti-CD8 Alexa488 antibody at 4 1C under steady shaking. The cells were washed twice in PBS to remove excess antibody.


[

NUMBER 11

[

NOVEMBER 2005

Electrophysiology, optical recordings and data analysis. The simultaneous optical and electrophysiological recordings were performed on a setup described earlier29, with a few small modifications. A standard patch-clamp setup was mounted on an inverted microscope (Zeiss IM35). The fluorescence was excited using a blue (470-nm) LED (Lumileds) or a mercury lamp with a low-noise power supply. The light was focused so that the light excites only one cell. The excitation light as well as the collected fluorescence were filtered using a FITC filter set (Ex: HQ480/40x; Di: Q505LP; Em: HQ535/50m, Chroma Technology). The light was collected using a high oil-immersion objective (NA 1.25, Olympus) and detected by a Photomax 200 amplifier (Dagan) with a cooled APD detector headstage. The data were acquired and stored on a PC using software developed in house. Trains of voltage pulses of different duration and frequency were generated by a function generator, and could be added via a simple circuitry to the command voltage during the duration of a TTL signal given by the recording software. The frequency response curve was fitted to a 4-pole low-pass filter with a corner frequency of 0.67 kHz: , Aðf Þ ¼ A0

2n !1=2 f 1+ fC

with A0 ¼ 0.19, the steady state normalized fluorescence change; f ¼ applied frequency; fC ¼ 0.67 kHz, the corner frequency of the filter; and n ¼ 4, order of the filter. All experiments were done in whole-cell patch-clamp configuration. For the measurements of the action potentials, we switched to current-clamp after establishing the whole cell patch. Voltage and current protocols were applied as indicated. A 20-mM stock solution of DPA in DMSO was prepared fresh from powder every day and diluted to a final concentration of either 2 mM or 6 mM in the external recording solution. Incorporation of DPA into the membrane was monitored by observing the increase in the capacitive transient. Electrophysiology and fluorescence data were acquired as described previously29. Data were analyzed using in-house analysis software, Excel (Microsoft) and Origin (Microcal). Confocal imaging. Confocal images of the cells expressing eGFP-F or CD8, labeled with anti-CD8, were imaged using a commercial confocal microscope (Olympus Fluoview/IX70). Series of sections through the cells along the optical axis were recorded, to show the distribution of the fluorescence in the cell. The cells were prepared as they were for electrophysiological recordings. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank W. Hubbell (UCLA) for the gift of DPA and A. Charles (UCLA) for the GT1 cells. Thanks to F. Chow for her assistance with primary neuronal cultures and transfections. We also thank T. Otis and M. Pratap for preparing viruses for slice transfections and for access to their setup. This work was supported by grants from the US National Institutes of Health (GM30376 to F.B., NS30549 to I.M., NS41317 to F.E.S.), the American Heart Association (0225006Y and 0535214N to B.C.) and Deutschen Forschungsgemeinschaft (BL538-1/1 to R.B.). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. Salzberg, B.M., Grinvald, A., Cohen, L.B., Davila, H.V. & Ross, W.N. Optical recording of neuronal activity in an invertebrate central nervous system: simultaneous monitoring of several neurons. J. Neurophysiol. 40, 1281–1291 (1977). 2. Djurisic, M. et al. Optical monitoring of neural activity using voltage-sensitive dyes. Methods Enzymol. 361, 423–451 (2003). 3. Ikegaya, Y. et al. Synfire chains and cortical songs: temporal modules of cortical activity. Science 304, 559–564 (2004). 4. Lieke, E.E., Frostig, R.D., Arieli, A., Ts’o, D.Y., Hildesheim, R. & Grinvald, A. Optical imaging of cortical activity: real-time imaging using extrinsic dye-signals and high resolution imaging based on slow intrinsic-signals. Annu. Rev. Physiol. 51, 543–559 (1989).

1625


TECHNICAL REPORT 5. Grinvald, A., Salzberg, B.M., Lev-Ram, V. & Hildesheim, R. Optical recording of synaptic potentials from processes of single neurons using intracellular potentiometric dyes. Biophys. J. 51, 643–651 (1987). 6. Miyawaki, A. Fluorescence imaging of physiological activity in complex systems using GFP-based probes. Curr. Opin. Neurobiol. 13, 591–596 (2003). 7. Miesenbock, G. Genetic methods for illuminating the function of neural circuits. Curr. Opin. Neurobiol. 14, 395–402 (2004). 8. Siegel, M.S. & Isacoff, E.Y. A genetically encoded optical probe of membrane voltage. Neuron 19, 735–741 (1997). 9. Ataka, K. & Pieribone, V.A. A genetically targetable fluorescent probe of channel gating with rapid kinetics. Biophys. J. 82, 509–516 (2002). 10. Sakai, R., Repunte-Canonigo, V., Raj, C.D. & Knopfel, T. Design and characterization of a DNA-encoded, voltage-sensitive fluorescent protein. Eur. J. Neurosci. 13, 2314–2318 (2001). 11. Gonzalez, J.E. & Tsien, R.Y. Voltage sensing by fluorescence resonance energy transfer in single cells. Biophys. J. 69, 1272–1280 (1995). 12. Cacciatore, T.W. et al. Identification of neural circuits by imaging coherent electrical activity with FRET-based dyes. Neuron 23, 449–459 (1999). 13. Bedlack, R.S., Jr., Wei, M.D., Fox, S.H., Gross, E. & Loew, L.M. Distinct electric potentials in soma and neurite membranes. Neuron 13, 1187–1193 (1994). 14. Rohr, S. & Salzberg, B.M. Multiple site optical recording of transmembrane voltage (MSORTV) in patterned growth heart cell cultures: assessing electrical behavior, with microsecond resolution, on a cellular and subcellular scale. Biophys. J. 67, 1301–1315 (1994). 15. Jiang, W. & Hunter, T. Analysis of cell-cycle profiles in transfected cells using a membrane-targeted GFP. Biotechniques 24, 349–354 (1998). 16. Loew, L.M. et al. A naphthyl analog of the aminostyryl pyridinium class of potentiometric membrane dyes shows consistent sensitivity in a variety of tissue, cell, and model membrane preparations. J. Membr. Biol. 130, 1–10 (1992). 17. Chanda, B., Asamoah, O.K., Blunck, R., Roux, B. & Bezanilla, F. Gating charge displacement in voltage-gated ion channels involves limited transmembrane movement. Nature 436, 852–856 (2005). 18. Fernandez, J.M., Taylor, R.E. & Bezanilla, F. Induced capacitance in the squid giant axon. Lipophilic ion displacement currents. J. Gen. Physiol. 82, 331–346 (1983).

19. de Ruyter van Steveninck , RR., Lewen, G.D., Strong, S.P., Koberle, R. & Bialek, W. Reproducibility and variability in neural spike trains. Science 275, 1805–1808 (1997). 20. Kalyanaraman, B., Feix, J.B., Sieber, F., Thomas, J.P. & Girotti, A.W. Photodynamic action of merocyanine 540 on artificial and natural cell membranes: involvement of singlet molecular oxygen. Proc. Natl. Acad. Sci. USA 84, 2999–3003 (1987). 21. Hales, T.G., Sanderson, M.J. & Charles, A.C. GABA has excitatory actions on GnRHsecreting immortalized hypothalamic (GT1–7) neurons. Neuroendocrinology 59, 297–308 (1994). 22. Bezanilla, F., Perozo, E. & : Stefani, E. Gating of Shaker K+ channels: II. The components of gating currents and a model of channel activation. Biophys. J. 66, 1011–1021 (1994). 23. Salter-Cid, L. et al. Transferrin receptor is negatively modulated by the hemochromatosis protein HFE: implications for cellular iron homeostasis. Proc. Natl. Acad. Sci. USA 96, 5434–5439 (1999). 24. Metzger, F. et al. Transgenic mice expressing a pH and Cl- sensing yellow-fluorescent protein under the control of a potassium channel promoter. Eur. J. Neurosci. 15, 40–50 (2002). 25. Oliva, A.A., Jr., Jiang, M., Lam, T., Smith, K.L. & Swann, J.W. Novel hippocampal interneuronal subtypes identified using transgenic mice that express green fluorescent protein in GABAergic interneurons. J. Neurosci. 20, 3354–3368 (2000). 26. Sippy, T., Cruz-Martin, A., Jeromin, A. & Schweizer, F.E. Acute changes in short-term plasticity at synapses with elevated levels of neuronal calcium sensor-1. Nat. Neurosci. 6, 1031–1038 (2003). 27. Gordey, M., Mekmanee, L. & Mody, I. Altered effects of ethanol in NR2A(DeltaC/DeltaC) mice expressing C-terminally truncated NR2A subunit of NMDA receptor. Neuroscience 105, 987–997 (2001). 28. Stell, B.M., Brickley, S.G., Tang, C.Y., Farrant, M. & Mody, I. Neuroactive steroids reduce neuronal excitability by selectively enhancing tonic inhibition mediated by delta subunit-containing GABAA receptors. Proc. Natl. Acad. Sci. USA 100, 14439–14444 (2003). 29. Blunck, R., Starace, D.M., Correa, A.M. & Bezanilla, F. Detecting rearrangements of shaker and NaChBac in real-time with fluorescence spectroscopy in patch-clamped mammalian cells. Biophys. J. 86, 3966–3980 (2004). 30. Knopfel, T., Tomita, K., Shimazaki, R. & Sakai, R. Optical recordings of membrane potential using genetically targeted voltage-sensitive fluorescent proteins. Methods 30, 42–48 (2003).

1626

VOLUME 8

[

NUMBER 11

[


Nature Neuroscience November 2005.pdf

Recommend Documents