www.nature.com/natureneuroscience

EDITORIAL OFFICE neurosci@natureny.com345 Park Avenue South, New York, NY 10010-1707Tel: (212) 726 9319, Fax: (212) 696 0978Editor: Sandra Aamodt Senior Editor: Kalyani Narasimhan Associate Editors: Cara Allen, I-han Chou, Annette Markus Assistant Editor: Charvy NarainCopy Editor: Dorothy MooreProduction Editor: Ivelisse RoblesAssistant Production Editor: Elizabeth A. MelchorCover Design: Erin BoyleEditorial Assistant: Jessica Chen

MANAGEMENT OFFICESNPG New York345 Park Avenue South, New York, NY 10010-1707Tel: (212) 726 9200, Fax: (212) 696 9006Publisher: Beatrice RenaultExecutive Editor: Linda MillerChief Technology Officer: Howard RatnerGlobal Head of Advertising: Fabien SavenayHead of Nature Research & Reviews Marketing: Sara GirardMarketing Manager: Naomi MulgraveProduction Coordinator: Allyson SkinnerAssociate Director, New Technology: Timo HannayAssociate Director, Content Systems: Joe LandolfiWeb Production Editor: Edwina Hay

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

NPG TokyoMG Ichigaya Bldg. 5F, 19-1 Haraikatamachi, Shinjuku-ku, Tokyo 162-0841Tel: 81 3 3267 8751, Fax: 81 3 3267 8746Asia-Pacific Publisher: Antoine E. BocquetManager: Koichi NakamuraSenior Marketing Manager: Peter YoshiharaAsia-Pacific Sales Director: Kate YoneyamaAsia-Pacific Sales Manager: Rinoko Asami

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

NATUREJOBS naturejobs@natureny.com (US/Canada) naturejobs@nature.com (Europe) advertising@naturejpn.com (Japan)Publisher: Ben Crowe, Tel: (212) 726 9245, Fax: (212) 696 9482US Sales Manager: Peter Bless, Tel: (212) 726 9248, Fax: (212) 696 9482Japan Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746naturejobs Sales Director: Nevin Bayoumi, Tel: 44 207 843 4961, Fax: 44 207 843 4996

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

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

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

i

VOLUME 8 NUMBER 5 MAY 2005

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature Publishing Group, a trading name of Nature America Inc. located at 345 Park Avenue South, New York, NY 10010-1707. Periodicals postage paid at New York, NY and additional mailing post offices. Editorial Office: 345 Park Avenue South, New York, NY 10010-1707. Tel: (212) 726 9321, Fax: (212) 696 0978. Annual subscription rates: USA/Canada: US$199 (personal), US$1,240 (institution). Canada add 7% GST #104911595RT001; Euro-zone: 289 (personal), 1,279 (institution); Rest of world (excluding China, Japan, Korea): 175 (personal), 775 (institution); Japan: Contact Nature Japan K.K., MG Ichigaya Building 5F, 19-1 Haraikatamachi, Shinjuku-ku, Tokyo 162-0841. Tel: 81 (03) 3267 8751, Fax: 81 (03) 3267 8746. POSTMASTER: Send address changes to Nature Neuroscience, Subscriptions Department, 303 Park Avenue South #1280, New York, NY 10010-3601. Authorization to photocopy material for internal or personal use, or internal or personal use of specific clients, is granted by Nature Publishing Group to libraries and others registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided the relevant copyright fee is paid direct to CCC, 222 Rosewood Drive, Danvers, MA 01923, USA. Identification code for Nature Neuroscience: 1097-6256/04. Back issues: US$45, Canada add 7% for GST. CPC PUB AGREEMENT #40032744. Printed by Publishers Press, Inc., Lebanon Junction, KY, USA. Copyright 2005 Nature Publishing Group. Printed in USA.

E D I TO R I A L535 The perils of public debate

CO R R E S P O N D E N C E537 Is prefrontal white matter enlargement a human evolutionary specialization?

B O O K R E V I E W

539 Principles of Brain Evolutionby Georg F StriedterReviewed by Jon H Kaas

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

541 Imaging orientation selectivity: decoding conscious perception in V1Geoffrey M Boynton see also pp 679 and 686

542 Finding the G spot on fusion machineryJane Sullivan see also p 597 and Nat. Neurosci. 8, p 421

544 Axon formation: fate versus growthHui Jiang & Yi Rao see also p 606

546 Song learning and sleepDaniel Margoliash

548 Trafficking in emotionsDan Ehninger, Anna Matynia & Alcino J Silva

550 SK channels: a new twist to synaptic plasticityKalyani Narasimhan see also pp 635 and 642

Genetic control of prefrontal and midbrain function

(p 594)

Obesity and related metabolic disorders are on the rise worldwide. To effectively combat this new epidemic,

we need a detailed understanding of the mechanisms that regulate

energy intake and expenditure. In this issue we present four review articles,

a perspective and a commentary highlighting current progress in the neurobiology of feeding regulation,

energy metabolism and obesity. This special focus is sponsored by the

Obesity Research Task Force of the National Institutes of Health. Cover

image: Venus from Willendorf, ca. 25,00022,000 BCE.

Naturhistorisches Museum, Vienna, Austria.

(pp 551589)

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

iii

VOLUME 8 NUMBER 5 MAY 2005

NATURE NEUROSCIENCE

INTRODUCTION: FEEDING REGUL ATION AND OBESIT Y

551 Neurobiology of obesity Annette Markus

SPONSORS FOREWORD: FEEDING REGULATION AND OBESITY

552 Obesity on the brainAllen Spiegel, Elizabeth Nabel, Nora Volkow, Story Landis & T K Li

COM M E N TA RY: F E E D I N G R E G U L AT I O N A N D O B E S I T Y

555 How can drug addiction help us understand obesity?Nora D Volkow & Roy A Wise

P E R S P E C T I V E : F E E D I N G R E G U L AT I O N A N D O B E S I T Y

561 The hardship of obesity: a soft-wired hypothalamusTamas L Horvath

R E V I E W S : F E E D I N G R E G U L AT I O N A N D O B E S I T Y

566 Molecular and anatomical determinants of central leptin resistanceHeike Mnzberg & Martin G Myers, Jr

571 Anatomy and regulation of the central melanocortin systemRoger D Cone

579 Hypothalamic sensing of fatty acidsTony K T Lam, Gary J Schwartz & Luciano Rossetti

585 Endocannabinoid control of food intake and energy balanceVincenzo Di Marzo & Isabel Matias

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

591 Neurons in macaque area V4 acquire directional tuning after adaptation to motion stimuliA S Tolias, G A Keliris, S M Smirnakis & N K Logothetis

594 Midbrain dopamine and prefrontal function in humans: interaction and modulation by COMT genotypeA Meyer-Lindenberg, P D Kohn, B Kolachana, S Kippenhan, A McInerney-Leo, R Nussbaum, D R Weinberger & K F Berman

A mechanism for axon specification(pp 544 and 606)

STAT signaling controls astrogliogenesis

(p 616)

SK channels regulate synaptic plasticity

(pp 550, 635 and 642)

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

v

VOLUME 8 NUMBER 5 MAY 2005

NATURE NEUROSCIENCE

-Synuclein toxicity in a fly model of Parkinson disease

(p 657)

Decoding subjective perception with fMRI

(pp 541, 679 and 686)

A R T I C L E S

597 G acts at the C terminus of SNAP-25 to mediate presynaptic inhibitionT Gerachshenko, T Blackmer, E-J Yoon, C Bartleson, H E Hamm & S Alford see also p 542

606 Asymmetric membrane ganglioside sialidase activity specifies axonal fateJ Santos Da Silva, T Hasegawa, T Miyagi, C G Dotti & J Abad-Rodriguez see also p 544

616 A positive autoregulatory loop of Jak-STAT signaling controls the onset of astrogliogenesisF He, W Ge, K Martinowich, S Becker-Catania, V Coskun, W Zhu, H Wu, D Castro, F Guillemot, G Fan, J de Vellis & Y E Sun

626 PI(4,5)P2 regulates the activation and desensitization of TRPM8 channels through the TRP domainT Rohcs, C M B Lopes, I Michailidis & D E Logothetis

635 SK channels regulate excitatory synaptic transmission and plasticity in the lateral amygdalaE S L Faber, A J Delaney & P Sah see also p 550

642 SK channels and NMDA receptors form a Ca2+-mediated feedback loop in dendritic spinesT J Ngo-Anh, B L Bloodgood, M Lin, B L Sabatini, J Maylie & J P Adelman see also p 550

650 Fast delayed rectifier potassium current is required for circadian neural activityJ N Itri, S Michel, M J Vansteensel, J H Meijer & C S Colwell

657 -Synuclein phosphorylation controls neurotoxicity and inclusion formation in a Drosophila model of Parkinson diseaseL Chen & M B Feany

664 The MAPK pathway and Egr-1 mediate stress-related behavioral effects of glucocorticoidsJ-M Revest, F Di Blasi, P Kitchener, F Roug-Pont, A Desmedt, M Turiault, F Tronche & P V Piazza

673 Electroreceptor neuron dynamics shape information transmissionM J Chacron, L Maler & J Bastian

679 Decoding the visual and subjective contents of the human brainY Kamitani & F Tong see also p 541

686 Predicting the orientation of invisible stimuli from activity in human primary visual cortexJ-D Haynes & G Rees see also p 541

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 DSee back pages.

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 5 | MAY 2005 535

E D I TO R I A L

The recent legal and political battles over assisted feeding for Terri Schiavo, a severely brain-damaged woman, thrust the neurologi-cal terms persistent vegetative state and minimally conscious into everyday discourse. Amid the tangle of medical facts and mis-conceptions and the blurring of ethical and scientific questions, one theme stood out clearly: the misuse and misunderstanding of science. For science to effectively inform such debates, scientists and the public need to develop a more nuanced understanding of what science can and cannot be expected to contribute to public policy discussions.

The scientific perspective is only one piece of the puzzle, and scien-tists should recognize the legitimate role of ethical principles, including those based in religion, in public debates. We can get very gummed up very fast if we fail to separate and identify the elements in these controversies that are matters of fact from those that are matters of principle, says University of Pennsylvania neuroscientist Martha Farah. Even if it were possible to prove beyond doubt that Terri Schiavo had no sense of awareness and no hope of recovery, that would not provide a single, correct answer for the ethical question of whether to remove her feeding tube. Similarly, no scientific argument levied on either side of highly politicized debates about human embryonic stem cell research will determine whether destruction of embryos for biomedical research is ethical. Science cannot answer such questions, and it cannot replace the moral values on which they depend.

However, most people do base their position on such issues in part on their understanding of the facts, particularly when they are asked to weigh two opposing principles. What looks at first like a difference in ethical principles may really be a difference in understanding about the empirical facts, says Farah. Scientists have a special ability to assess scientific evidence, and we have a responsibility to disseminate this information in a way that allows the public to make informed decisions. In Terri Schiavos case, much confusion centered on whether or not she was aware or had any chance for meaningful recovery. Yet the diagnosis that she was in a persistent vegetative state, without consciousness, was made with a higher degree of certainty than was probably apparent to much of the public.

Some of the responsibility for such misunderstandings lies with jour-nalists, who have an incentive to stir up controversy to attract readers. Nicholas Schiff, a Cornell physician and neuroscientist who studies severely brain-damaged patients, notes that he and his colleague were unable to convince a New York Times editor to change a misleading headline about their research on minimally conscious patients before the story went to print. The headline (New research suggests that many vegetative patients are more conscious than previously supposedand might eventually be curable. A whole new way of thinking about pull-ing the plug) illustrates the confusion surrounding the facts of this case. Schiff and a small number of scientists have worked to clarify the boundary between persistent vegetative state and minimally conscious state, with a long-term goal of developing more sophisticated methods

to evaluate patients in terms of their likelihood to recover or be mis-diagnosed. Such confusion in the press undermines these researchers efforts to reduce the uncertainties in treating patients with disorders of consciousness.

Even with the most careful reporting, of course, science cannot pro-vide absolute certainty. However, scientists can often provide a good estimate of the uncertainty. We do not understand the full neural basis of consciousness, but patients like Schiavo, who have suffered a global block of blood supply to the brain, sustain an overwhelming and per-manent loss of the corticothalamic networks that seem to be required for consciousness. Likewise, it is unclear whether adult stem cells will ultimately present as much potential for therapeutic use as embryonic cells, but most scientists agree that the uncertainty is still far too great to justify dismissing human embryonic stem cell research. We must take care to explain that uncertainties are inherent in science, to pres-ent them honestly and to help determine which ones are particularly relevant to ethical choices. Although we can rarely provide the decisive answer that the public may wish to hear, we can provide information about likelihoods and how we interpret them, which may help the pub-lic weigh the strengths and weaknesses of the scientific evidence used to support different viewpoints.

Frustration with the misuse and misunderstanding of science in pub-lic debate may tempt scientists to stop participating in these discus-sions, butas the stem cell controversy illustrateswe cannot afford this luxury. Instead, we should improve our ability to communicate effectively with the public by recognizing the ethical issues involved and by listening to the questions that people want us to answer, rather than presuming we know best what the public needs to hear. We should avoid speaking in utilitarian terms or implicitly placing science above other forms of knowledge; either is likely to make the public tune us out. In addition to providing information about specific issues, we need to help the public understand the scientific process, including the role of empirical (versus anecdotal) evidence in resolving conflicting views. We might also establish direct contacts with the public through lectures, discussions or websites rather than relying on intermediaries such as journalists and politicians, who may get the science wrong or present it in a biased way.

Incentives to manipulate scientific findings for political ends will always exist, but scientists can and should make it much more difficult for opportunists to misuse their words and their work. We already have a great advantage in this regard: the respect of the public. People in the US consistently give scientists a high vote of confidence, second only to physicians in most years1. In contrast, few express confidence in politi-cians or the press. More thoughtful communication will be the key to maintaining this confidence and putting it to good use.

1. Davis, J.A., Smith, T.W. & Marsden, P.V. General Social Surveys: 19722002 Cumulative Codebook (National Opinion Research Center at the University of Chicago, Chicago, Illinois, USA, 2003).

The perils of public debate

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 5 | MAY 2005 537

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

Is prefrontal white matter enlargement a human evolutionary specialization?To the Editor:Using a comparative volumetric analysis of MRI scans of brains from 11 primate spe-cies (including monkeys, apes and humans), Schoenemann and colleagues1 claim that the prefrontal white matter of humans is enlarged compared to that of other primates. This would suggest that the evolution of human cognitive capacities mediated by prefrontal circuitry relies on enhanced interconnectiv-ity. Problems with the authors definition of the prefrontal sector and assessment of human deviations from scaling trends, how-ever, suggest that these conclusions are not well supported.

Unlike the frontal cortex, which can be reli-ably demarcated by the central sulcus2, the borders of prefrontal cortex cannot be iden-tified based on gross anatomy. Schoenemann and colleagues devised a proxy measure of the prefrontal sector comprising all tissue lying in coronal slices anterior to the genu of the corpus callosum. To justify application of this segmentation scheme across phylogeny, the authors referred to cytoarchitectural maps in select species and concluded that their pre-frontal proxy will underestimate values only in humans, not in other primates. However, they did not refer to existing cytoarchitectural

maps of great apes3,4, which suggest that their prefrontal measurement underestimates these species as well. The application of a topologi-cal border that does not bear a realistic cor-respondence to the histological parcellation of the prefrontal cortex is problematic. Because their proxy measure of prefrontal cortex underestimates the actual value to varying degrees among primates, their findings should be treated with caution.

Notwithstanding problems in their ana-tomical definition of prefrontal, even if their segmentation is accepted, it is not certain that their reported data support the conclusion that human prefrontal white matter is dispro-portionately enlarged. They found that human prefrontal white matter is 41% greater than that predicted for a nonhuman primate with the same non-prefrontal white matter volume. However, the slope of such scaling relation-ships is notoriously sensitive to the taxonomic composition of the reference group used to calculate the line5,6. We fit a separate regres-sion line to the great ape data and used it to predict the individual human values. Based on our analysis, human prefrontal white matter is only 12% 11 (mean s.d.; n = 12) greater than allometric expectations (Fig. 1a). This suggests that humans have, at most, a moder-

ate increase in prefrontal white matter volume when taking into account their phylogenetic affinities with great apes. Because great apes are the closest relatives of humans, any unique human neural specialization should be plainly apparent in comparison with this group.

Whether the volume of human prefrontal white matter departs from more general scal-ing trends for interconnectivity can be exam-ined in another way. It is well established that overall neocortical white matter increases disproportionately compared with gray mat-ter in mammals7, either because of increased demands for connections across areas or because of increases in axon diameter. Thus, the strongest case for exceptional enlargement of prefrontal white matter would be to observe human values larger than those predicted for a primate of similar prefrontal gray matter size. We tested this idea and found that humans have only 2% 28 more prefrontal white matter than expected for a primate of similar prefrontal gray matter volume (Fig. 1b). On the basis of a prediction calculated from only great ape data, the human values are actually 17% 35 lower than expected. Given these results, it is difficult to make a strong claim for the evolution of specialized human enlarge-ment of prefrontal white matter beyond

Figure 1 The allometric scaling relationship of prefrontal white matter volume. In both plots, the least-squares regression lines are calculated based on mean values from nonhuman species, and data points are shown for reference. (a) Separate lines are fit to data from great apes (dashed line: y = 1.242x 2.344; r2 = 0.777, P = 0.119) and other nonhuman primates (solid line: y = 1.049x 1.340; r2 = 0.959, P < 0.001). (b) The line fit to all nonhuman primates is shown (y = 0.976x 0.049; r2 = 0.977, P < 0.001). The line based only on great ape data is y = 1.110x 0.632; r2 = 0.734, P = 0.143.

Log prefrontal gray matter volume

orferp goLlatnw

ho v retta

m etie

mul

2.4 2.8 3.2 3.6 4.0 4.4 4.8 5.22.4

2.8

3.2

3.6

4.0

4.4

4.8

5.2a b

Humans

Great apes

Other primates

Log non-prefrontal white matter volume

orferp goLlatnw

hov retta

m etie

mul

3.6 4.0 4.4 4.8 5.2 5.6 6.02.4

2.8

3.2

3.6

4.0

4.4

4.8

5.2

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

538 VOLUME 8 | NUMBER 5 | MAY 2005 NATURE NEUROSCIENCE

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

simple allometric scaling to maintain func-tional interconnectedness at a larger overall brain size. Specialization of cortical neuron types8 and elevated gene expression associated with metabolism and synaptic plasticity9 in humans suggest that subtle modifications of architecture, function and connectivity10 may have been critical in the evolution of human cognitive capacities.

Chet C Sherwood1, Ralph L Holloway2,3, Katerina Semendeferi4 & Patrick R Hof3,5

1Department of Anthropology and School of Biomedical Sciences, Kent State University, Kent, Ohio 44242, USA. 2Department of Anthropology, Columbia University, New York, New York 10025, USA. 3New York Consortium in Evolutionary Primatology, New York, New York, USA. 4Department of Anthropology, University of California, San Diego, La Jolla, California 92093, USA. 5Department of Neuroscience, Mount Sinai School of Medicine, New York, New York 10029, USA.e-mail: csherwoo@kent.edu

1. Schoenemann, P.T., Sheehan, M. J. & Glotzer, L. D. Nat. Neurosci. 8, 242252 (2005).

2. Semendeferi, K., Lu, A., Schenker, N. & Damasio, H. Nat. Neurosci. 5, 272276 (2002).

3. Bailey, P., von Bonin, G. & McCulloch, W.S. The Isocortex of the Chimpanzee (Univ. of Illinois Press, Urbana, Illinois, 1950).

4. Mauss, T. J. Psychol. Neurol. 18, 410467 (1911).5. Holloway, R.L. & Post, D.G. in Primate Brain Evolution:

Methods and= Concepts (eds. Armstrong, E. & Falk, D.) 5776 (Plenum, New York, 1982).

6. Harvey, P.H. & Krebs, J.R. Science 249, 140146 (1990).

7. Zhang, K. & Sejnowski, T.J. Proc. Natl. Acad. Sci. USA 97, 56215626 (2000).

8. Nimchinsky, E.A. et al. Proc. Natl. Acad. Sci. USA 96, 52685273 (1999).

9. Cceres, M. et al. Proc. Natl. Acad. Sci. USA 100, 1303013035 (2003).

10. Holloway, R.L. Am. J. Phys. Anthropol. 118, 399401 (2002).

Schoenemann et al. reply:Sherwood et al. find the prefrontal volume proxy we used problematic, even though it has been widely used in the neuropsychological literature for many years14, and it (or vari-ants) have also been applied to non-human primates5,6. Sherwood et al. believe this proxy specifically underestimates the size of ape val-ues, yet the frontispiece of one of the sources they cite7 clearly shows any underestimation is minor compared to that found in humans.

Taking refs. 7 and 8 together, we see that the degree of underestimation using this method increases as one gets closer to humans. An image highlighting the approximate degree of underestimation based on cytoarchitec-tural maps7,8 is posted on our web site, so interested readers may judge for themselves (http://www.sas.upenn.edu/~ptschoen/Pics/prefrontal-delineation.jpg).

Furthermore, using a proxy for prefron-tal volume on MRI data is exactly what Semendeferi et al. have reported in this same journal9. Their proxy was total frontal volume minus precentral gyrus volume, which also does not follow cytoarchitectural boundaries but leads to a varying degree of overestimation of prefrontal size across species. Nevertheless, the authors argue their data ...goes against the large relative differences in the prefrontal cor-tex between humans and great apes reported in previous publications... (p. 274) and ... should prove useful until more definitive data become available... (p. 273). Our proxy is no less valid; it simply focuses on more anterior regions of the frontal cortex. Together, these studies suggest that as one looks at increasingly anterior regions, humans seem increasingly disproportionate. Comparing Figures 2 and 3 from ref. 9 to our Figures 2 and 4 makes this abundantly clear.

Sherwood et al. believe the strongest case for specialized enlargement of prefrontal white matter would be to show that it is dis-proportionate relative to prefrontal gray mat-ter. On the contrary, given that the role of the prefrontal cortex includes executive oversight of posterior regions, the interesting question is how extensively it interconnects relative to non-prefrontal regions. Our data show that the distribution of white matter is peculiar in humans, even though it scales with prefrontal gray matter.

Sherwood et al. also argue that great apes alone are the only valid comparison group. The problem is that only four data points can be used to estimate this relationship, thereby vastly reducing confidence in the regression prediction. (Humans would have to be more than 950% larger than predicted in order to be significantly larger.) Thus, it is an open ques-tion whether humans have more prefrontal

white matter with respect to non-prefrontal white matter than great ape data predict, but it is not an open question regarding primates as a whole (at least from our data).

How humans differ from primates as a whole, versus how they differ from great apes alone, are really two different, equally impor-tant questions. Brodmanns original data show that chimpanzees have 56% more prefrontal surface area than predicted from non-pre-frontal surface area. This, combined with our data suggesting an increased slope within great apes, may suggest that prefrontal elaboration accelerated in great apes.

The most interesting question is what all these patterns mean behaviorally. It is important to recognize that both behavioral selection and developmental constraint expla-nations exist for allometric scaling. Showing that allometry statistically explains some pattern does not indicate that it is therefore behaviorally irrelevant.

Semendeferi et al. note, In a previous study, we found that the relative volume of white mat-ter underlying prefrontal association cortices is larger in humans than in great apes9. We believe our study is consistent with this statement.

P Thomas Schoenemann, L Daniel Glotzer & Michael J Sheehan

Department of Anthropology, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA.e-mail: ptschoen@sas.upenn.edu

1. Zipursky, R.B., Lim, K.O., Sullivan, E.V., Brown, B.W. & Pfefferbaum, A. Arch. Gen. Psychiatry 49, 195205 (1992).

2. Sax, K.W. et al. Am. J. Psychiatry 156, 139141 (1999).

3. Laakso, M.P. et al. Psychiatry Res. 114, 95102 (2002).

4. Raz, N. et al. Cereb. Cortex; published online 9 February 2005 .

5. Lyons, D.M., Afarian, H., Schatzberg, A.F., Sawyer-Glover, A. & Moseley, M.E. Behav. Brain Res. 136, 5159 (2002).

6. McBride, T., Arnold, S.E. & Gur, R.C. Brain Behav. Evol. 54, 159166 (1999).

7. Bailey, P., von Bonin, G. & McCulloch, W.S. The Isocortex of the Chimpanzee. (Univ. of Illinois Press, Urbana, Illinois, 1950).

8. Brodmann, K. Anat. Anz. 41 (suppl.), 157216 (1912).

9. Semendeferi, K., Lu, A., Schenker, N. & Damasio, H. Nat. Neurosci. 5, 272276 (2002).

10. Holloway, R.L. Am. J. Phys. Anthropol. 118, 399401 (2002).

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 5 | MAY 2005 541

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

Imaging orientation selectivity: decoding conscious perception in V1Geoffrey M Boynton

In V1, neurons preferring similar orientations are grouped in columns too small to be resolved by conventional fMRI. Two studies circumvent this limitation by using algorithms to recognize patterns of activation across a large area. This new trick allows the authors to distinguish responses to different orientations in human V1 and to study its contribution to conscious perception.

Functional MRI is currently the best non-inva-sive tool for measuring human brain activity at below-centimeter resolution1. This spatial resolution is ideal for detecting and studying entire cortical maps, such as those in primary visual cortex (V1), but it is still a long way from measuring the responses of individual neurons. Electrophysiological recording and optical imaging in animals have shown that V1 neurons preferring similar orientations form columns about 500 m across. Measuring the response from these homogenous clusters would be an important step toward increas-ing the spatial resolution of functional imag-ing, but even these have been too fine to be routinely resolved by fMRIuntil now. In this issue, two studies2,3 show that it is possible to estimate the orientation of a stimulus from the pattern of fMRI responses it produces in V1. This enables us for the first time to study how this fundamental form of visual information is represented in human cortex.

Over the past decade, the spatial resolution of fMRI has been gradually improving through technical advances such as increased magnetic field strength, better receiving coils and more reliable gradients and amplifiers. But direct mea-surements below the resolution of a millimeter can be obtained only with massive amounts of signal averaging from a carefully selected group of subjects4. Attempts to study orientation selec-tivity with fMRI using indirect methods such as adaptation have also met with difficulty5. In the two current papers2,3, the authors took an alternative approach to measuring orienta-tion selectivity through a clever data analysis

tion found the orientation that was most likely to have induced a given pattern of responses.

Because fMRI responses are noisy, many voxels must be incorporated to obtain reason-

trick. Remarkably, both groups used traditional high-field (3-T) fMRI data acquisition meth-ods, which means that evidence of signals at the columnar level may already be available in all of our existing fMRI data sets.

How did they measure such a fine spa-tial structure without special equipment? In a simulated orientation pinwheel map (Fig. 1a), different colors indicate the prefer-ence of orientation columns on a 9 9 mm region of the cortical surface6. Each of the nine 3 3-mm fMRI voxels, shown as black squares, contains a broad range of orienta-tion preferences. On closer inspection, how-ever, some voxels contain more columns of one orientation preference than another (Fig. 1b). Although all voxels respond to all orientations, voxels clearly have a variable response across orientations. This variability is evidence of high spatial frequency informa-tion, even if the measurement tool is sampling at a lower frequency, a phenomenon known as aliasing in the signal processing literature.

How can these weak biases in fMRI responses be used to predict the orientation of a subse-quently viewed stimulus? The trick is to first associate a range of test stimulus orientations with their patterns of fMRI responses. There are a variety of ways of doing this. Kamitani and Tong2 use a linear support vector machine that creates classifiers for each stimulus orien-tation by summing weighted responses across voxels, obtaining optimal weights during a training period. When oriented stimuli were presented after training, the response for each classifier to the fMRI image was calculated, and the actual stimulus was estimated from the classifier with the largest response estimates. Haynes and Rees3 used a linear discriminant analysis method in which a Bayesian calcula-

Geoffrey M. Boynton is at the Salk Institute, 10010

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

email: boynton@salk.edu

Figure 1 Patterns of orientation-selective responses measured with fMRI. (a) Synthetic orientation tuning data generated by band-pass filtering random orientation values6. The black squares represent 3 3 mm fMRI voxels. (b) Histograms showing the proportion of selectivity inside each voxel to each of the eight orientations shown below. This shows how different stimulus orientations produce slightly different patterns of responses in V1. Algorithms such as those used in the current studies2,3 can estimate from these responses the orientation of a subsequently presented stimulus.

3 m

m

a

b

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

542 VOLUME 8 | NUMBER 5 | MAY 2005 NATURE NEUROSCIENCE

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

able accuracy. Kamitani and Tong2 show that in V1, the pattern of responses across a few hundred voxels can predict with nearly per-fect accuracy which orientation was shown. Haynes and Rees3 found an impressive average accuracy of 80% from the response to a single two-second stimulus presentation.

It is not known how far we can go with this method. If too many columns fit inside the aver-age voxel, then voxels should have nearly equal responses to all orientations. On the other hand, increasing voxel size dramatically increases the reliability of the signal from each voxel, so results may not be highly dependent on voxel size. This is not the first time that this method has been used to measure stimulus selectivity at a sub-voxel level. Haxby and colleagues7 used a similar method to show that the pattern of fMRI responses in the ventral temporal cortex could predict the category of an object (such as faces, cats, houses and shoes) that was being shown. It remains to be seen if this method will be suc-cessful at studying columnar structures in other sensory areas, such as those in auditory cortex, or even motor areas of the human brain.

Although orientation selectivity is ubiqui-tous in mammalian V1, the ability to exam-ine it in humans provides the opportunity to study the role of this area in conscious visual perception. For example, if V1 is so early in the processing stream, does orienta-tion selectively occur automatically, or can it be affected by the will of the subject? If it does occur automatically, does it always lead to a conscious visual percept? Both papers went well beyond verifying that orientation-selectivity can be measured in human V1 and have given important insights into these fundamental questions.

Kamitani and Tong2 were able to predict the orientation that a subject was thinking about. Subjects were instructed to attend to one of two orthogonal orientations forming a plaid stimulus. The physical stimulus did not

change across trials; only the instructions to the subjects did. The authors found that they could predict which orientation the subject was attending with 80% accuracy on a trial-by-trial basis. This suggests that attending to one ori-entation and ignoring the other changed the pattern of fMRI responses enough to look like only the attended orientation was presented. This is similar to the effects seen in single-neuron recordings when multiple stimuli are presented within the receptive field of a cell8,9. But this is the first such evidence in humans, and the first in the primary visual cortex.

Haynes and Rees3, on the other hand, were able to predict the orientation of a stimulus that subjects could not see. Rapidly alternating a stimulus of co-oriented lines with a multi-oriented masking stimulus renders the ori-ented stimulus invisible10. Subjects can clearly see the alternation between the oriented and the masking stimulus, but cannot determine the angle of the oriented stimulus. However, Haynes and Rees3 used the fMRI response to this alternating stimulus to predict the masked stimulus even though the subjects could not tell which orientation was being shown.

These two studies have interesting implica-tions about the role of V1 in consciousness. Being just two synapses away from the eye, V1 is usually considered an early visual area. Early visual areas tend to represent properties of the physical stimulus, whereas visual areas later in the processing stream seem to hold our con-scious percept, or our brains interpretation of the stimulus5,11. The finding by Haynes and Rees3 is consistent with this idea, and supports the theory12 that we are not consciously aware of all of the processing going on in V1.

But is V1 a passive feed-forward image pro-cessing machine that is unaffected by what the observer was thinking or doing? It seems not. Allocating attention to a particular location in space (without moving the eyes) can affect fMRI and electrophysiological responses in V1

(refs. 1315). Kamitani and Tong2 show that allocation to the feature of an attended stimulus can affect V1 responses as well. Far from being unavailable to consciousness, V1 responses can be examined as if part of a mind-reading exer-cise on their subjects, the authors suggest2.

These two papers show that V1 appears to be neither at the beginning nor at the end of visual processing. Considering the array of feedback connections from higher visual areas to V1 and the feed-forward loops back through the genic-ulate to V1, perhaps it is unwise to consider any cortical visual area as early. Instead, the distinc-tion between early and late processing may all be in the timing. It is likely that the response to the invisible stimulus by Haynes and Rees3 occurs early in the temporal response to the stimulus, and the modulations of the orienta-tion-selective response by attention found by Kamitani and Tong2 occur 100200 ms later, as has been seen for spatial attention14. The answer may already be there in the data, waiting for another clever algorithm to tease it out.

1. Engel, S.A., Glover, G.H. & Wandell, B.A. Cereb. Cortex 7, 181192 (1997).

2. Kamitani, Y. & Tong, F. Nat. Neurosci. 8, 679685 (2005).

3. Haynes, J. & Rees, G. Nat. Neurosci. 8, 686691 (2005).

4. Cheng, K., Waggoner, R.A. & Tanaka, K. Neuron 32, 359374 (2001).

5. Boynton, G.M. & Finney, E.M. J. Neurosci. 23, 87818787 (2003).

6. Rojer, A.S. & Schwartz, E.L. Biol. Cybern. 62, 381391 (1990).

7. Haxby, J.V. et al. Science 293, 24252430 (2001).8. Moran, J. & Desimone, R. Science 229, 782784

(1985).9. Reynolds, J.H., Chelazzi, L. & Desimone, R.

J. Neurosci. 19, 17361753. (1999).10. Macknik, S.L. & Livingstone, M.S. Nat. Neurosci. 1,

144149 (1998).11. Treue, S. Trends Neurosci. 24, 295300 (2001).12. Crick, F. & Koch, C. Nature 375, 121123 (1995).13. Gandhi, S.P., Heeger, D.J. & Boynton, G.M. Proc. Natl.

Acad. Sci. USA 96, 33143319 (1999).14. Martinez, A. et al. Nat. Neurosci. 2, 364369

(1999).15. Somers, D.C., Dale, A.M., Seiffert, A.E. & Tootell,

R.B. Proc. Natl. Acad. Sci. USA 96, 16631668 (1999).

Jane Sullivan is at the Department of Physiology

and Biophysics, University of Washington School

of Medicine, Box 357290, Seattle, Washington

98195, USA.

e-mail: jmsull@u.washington.edu

Finding the G spot on fusion machineryJane Sullivan

Activation of G proteincoupled receptors can inhibit secretion of neurotransmittters and hormones. Two recent reports in Nature Neuroscience show that this inhibition is due to G binding to SNAP-25, directly blocking the vesicle fusion machinery.

Activation of presynaptic G proteincoupled receptors (GPCRs) by ligands such as GABA, glutamate, serotonin or adenosine is a power-ful negative feedback mechanism for modu-lating transmission at synapses throughout

the brain1,2. For example, glutamate released from hippocampal mossy fibers during high-frequency trains of action potentials activates presynaptic G proteincoupled metabotropic glutamate receptors and inhibits subsequent

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 5 | MAY 2005 543

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

transmitter release3. G protein and sub-units that are liberated upon ligand binding to GPCRs can reduce neurotransmitter and hor-mone secretion through at least three mecha-nisms: inhibition of calcium channels that control evoked release, activation of potas-sium channels that suppress excitation and an apparently direct inhibition of the fusion machinery1,2. A direct effect of G proteins on fusion machinery had long been suspected, because activation of many different types of GPCRs reduces the frequency of spontaneous neurotransmitter release from neurons under conditions where there is little or no calcium entry through voltage-dependent channels4.

Solid evidence for a direct effect of G pro-teins on exocytosis was recently provided by the groups of Tom Martin, Simon Alford and Heidi Hamm5: injection of G subunits into lamprey reticulospinal neurons caused inhibi-tion of evoked release without any measurable

change in presynaptic calcium transients, but the mechanism underlying this direct inhibi-tion of fusion machinery was not resolved. These groups have teamed up again, and in two Nature Neuroscience papersone published last month6 and one in this issue7they show that this direct inhibition is due to G bind-ing to SNAP-25, one of the three proteins that form the SNARE complex mediating vesicle fusion6,7. In PC12 cells that release catechol-amines from dense core vesicles6, as well as in lamprey reticulospinal neurons that release glu-tamate from small synaptic vesicles7, treatment with botulinum toxin A (BoNT A) significantly reduces or abolishes G-mediated inhibition of release. Because BoNT A cleaves the last nine amino acids of SNAP-25, this region is strongly implicated as the binding site for G.

Consistent with the physiology, biochemi-cal experiments show that G binds both to individual SNAP-25 molecules and to

trimeric SNARE complexes made up of SNAP-25, syntaxin and VAMP (also known as synaptobrevin)6,7. Biochemical experi-ments also show that G binding to SNARE complexes is significantly reduced by BoNT A treatment6,7. Although it is possible that BoNT A cleaves some other protein that is the relevant binding partner of G subunits (and the binding of G to SNAREs is an irrelevant coincidence) or that BoNT A cleav-age of SNAP-25 changes the configuration of some other protein that contains the real G binding site, the most likely explanation is that G binding to the C terminus of SNAP-25 inhibits SNARE-mediated exocytosis.

How might G binding to the C terminus of SNAP-25 inhibit release? Although potentially it could interfere with the interaction of any pro-tein whose association with SNAP-25 mediates vesicle fusion, one particularly appealing possi-bility is that it disrupts the binding of SNAP-25 to synaptotagmin, the putative sensor control-ling calcium-triggered (that is, evoked) release of neurotransmitter. Synaptotagmin I binds to SNAP-25 and SNARE complexes in response to calcium8,9. Like G, synaptotagmin binds to the C terminus of SNAP-25; cleavage of the last nine amino acids of SNAP-25 with BoNT A significantly reduces calcium-dependent binding of synaptotagmin to either SNAP-25 or SNARE complexes, whereas cleavage of the last 26 amino acids of SNAP-25 with botulinum toxin E (BoNT E) abolishes calcium-depen-dent synaptotagmin binding to SNAP-25 and reduces calcium-dependent binding to SNARE complexes by more than 70% (refs. 8, 9). G and synaptotagmin compete with one another for binding to SNARE complexes, but only in the presence of calcium6. Synaptotagmin does not bind efficiently to SNARE complexes with-out calcium8, but G does6. These findings suggest that G could bind to SNAP-25 under the nominally calcium-free conditions of a rest-ing presynaptic terminal and inhibit subsequent calcium-dependent binding of synaptotagmin to the SNARE complex (Fig. 1).

The interaction between synaptotagmin and SNAP-25 seems to be important for transmis-sion, because cleavage by BoNT A (which dimin-ishes synaptotagmin binding) shifts the calcium dependence of release so that more calcium is required to trigger release8,10, whereas cleavage by BoNT E (which eliminates synaptotagmin binding) almost completely abolishes calcium-dependent release8,10. In addition, SNAP-25 C-terminal mutants that show significantly reduced calcium-dependent synaptotagmin binding in vitro do not support evoked release from PC12 cells9. G binding to SNAP-25 could, therefore, acutely mimic the effects of BoNT A cleavage or C-terminal mutations on release by interfering

In response to Gproteincoupled receptoractivation

Synaptotagmin BoNT A

Plasma membrane

Synaptic vesicle

Syntaxin

VAMP

SNAP-25

In response tocalcium

Synaptotagmin BoNT A

Plasma membrane

Synaptic vesicle

SNAREcomplex

a

b

Ivel

isse

Rob

les

G

Syntaxin

VAMP

SNAP-25

SNAREcomplex

Figure 1 G binding to the C terminus of SNAP-25 directly inhibits release. In response to calcium entry, synaptotagmin, a synaptic vesicle protein that is the putative calcium sensor controlling evoked neurotransmitter release, binds to the C terminus of SNAP-25 (refs. 8,9), one of three proteins in the SNARE complex, a key component of vesicle fusion machinery. Cleavage of the last nine amino acids of SNAP-25 with BoNT A reduces synaptotagmin binding8,9 and shifts the calcium dependence of release8,10. When G proteincoupled receptors are activated, G subunits are liberated and bind to SNAP-25 (ref. 6). New studies from Blackmer et al.6 and Gerachshenko et al.7 suggest that this interaction between G and SNAP-25 is responsible for direct G proteinmediated inhibition of release, perhaps by interfering with calcium-dependent binding of synaptotagmin6. Cleavage of the last nine amino acids of SNAP-25 with BoNT A eliminates G binding and abolishes G proteinmediated inhibition of release, implicating this region as the site of action6,7.

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

544 VOLUME 8 | NUMBER 5 | MAY 2005 NATURE NEUROSCIENCE

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

with the interaction between the calcium sensor (synaptotagmin) and the fusion machinery (the SNARE complex).

So, is this all there is to know about the direct effects of G proteins on release machinery? Probably not. For one thing, interference with synaptotagmin binding is unlikely to inhibit spontaneous calcium-independent releasethe initial finding that suggested a direct effect of G proteins on release machineryas spontane-ous release is unaffected by the genetic ablation or acute inactivation of synaptotagmin, even though fast action potentialevoked transmis-sion is abolished1113. In addition, G has no effect on spontaneous calcium-independent release in PC12 cells6 and no reported effect on spontaneous release in lamprey neurons. Taken together, these observations suggest that some other direct effect on release machinery might be found to inhibit spontaneous release in mammalian CNS neurons, although it is possible that G binding to the C terminus of SNAP-25 inhibits spontaneous release at mam-malian synapses through a synaptotagmin-independent mechanism that does not operate in PC12 cells. In any case, identifying the func-tion of spontaneous neurotransmitter release has only recently begun14, so we are probably quite far from understanding the importance of G proteinmediated inhibition of calcium-independent release. For now, G proteinmedi-ated effects on spontaneous release can be interpreted only as evidence of an underlying mechanism whose dependence on SNAP-25 binding remains to be determined.

G inhibition of calcium-dependent syn-aptotagmin binding to SNAP-25 is a promis-

ing candidate mechanism for reducing evoked release at mammalian CNS synapses, but testing this hypothesis will not be easy. Some of the very things that make the lamprey and PC12 cell sys-tems ideal for studying direct G inhibition of fusion machinery are features that distinguish them from mammalian neurons, which means that extrapolation must be made with some caution. One big advantage of the lamprey reticulospinal preparation is that the calcium channels responsible for evoked release from these cells are not modulated by G proteins5,15. In most mammalian neurons, though, G pro-teins inhibit the opening of N- and P/Q-types of calcium channels, which control evoked release, through a direct, membrane-delimited interaction between the calcium channels and G1,2. The permeabilized PC12 cells used by Blackmer et al.6 eliminate the need for calcium entry through calcium channels because cal-cium can be directly delivered to the release machinerybut just how much of the fusion machinery controlling slow dense core vesicle release is shared with the machinery controlling fast synaptic vesicle release is an open question. Although fusion of both dense core vesicles and synaptic vesicles relies on SNARE complexes and synaptotagmin, there are enough discrep-ancies in the details of exocytosis to suggest that not all of the molecules involved are com-mon to both systems. Determining the role of GSNAP-25 interactions at mammalian CNS synapses will require an experimental system that can bypass G proteinmediated effects on calcium channels: for example, by injecting G into the giant presynaptic terminal of the calyx of Held in rodent brainstem slices and looking

at the effects on release evoked by rapid flash photolysis of caged calcium, ideally before and after BoNT A cleavage of SNAP-25.

Specific examples of functional roles for G proteinmediated modulation of exocytosis are not yet numerous, but the multitude of dif-ferent types of GPCRs and the redundancy of mechanisms by which G protein subunits can influence release is a strong indication of the profound significance of this signaling system. By pinpointing the site of direct G interac-tion with the fusion machinery, Blackmer et al.6 and Gerachshenko et al.7 have identified yet another way that G proteins influence release and have provided us with a new target for mediating presynaptic inhibition.

1. Hille, B. Trends Neurosci. 17, 531536 (1994).2. Miller, R.J. Annu. Rev. Pharmacol. Toxicol. 38, 201

227 (1998).3. Scanziani, M., Salin, P.A., Vogt, K.E., Malenka, R.C.

& Nicoll, R.A. Nature 385, 630634 (1997).4. Thompson, S.M., Capogna, M. & Scanziani, M. Trends

Neurosci. 16, 222226 (1993).5. Blackmer, T. et al. Science 292, 293297 (2001).6. Blackmer, T. et al. Nat. Neurosci. 8, 421425

(2005).7. Gerachshenko, T. et al. Nat. Neurosci. 8, 597605

(2005).8. Gerona, R.R., Larsen, E.C., Kowalchyk, J.A. & Martin,

T F. J. Biol. Chem. 275, 63286336 (2000).9. Zhang, X., Kim-Miller, M.J., Fukuda, M., Kowalchyk,

J.A. & Martin, T.F. Neuron 34, 599611 (2002).10. Capogna, M., McKinney, R.A., OConnor, V., Gahwiler,

B.H. & Thompson, S.M. J. Neurosci. 17, 71907202 (1997).

11. Geppert, M. et al. Cell 79, 717 727 (1994).12. Marek, K.W. & Davis, G.W. Neuron 36, 805813

(2002).13. Yoshihara, M. & Littleton, J.T. Neuron 36, 897908

(2002).14. Zucker, R.S. Neuron 45, 482484 (2005).15. Takahashi, M., Freed, R., Blackmer, T. & Alford, S.

J. Physiol. (Lond.) 532, 323336 (2001).

Axon formation: fate versus growthHui Jiang & Yi Rao

Actin destabilization is an early step in specifying axon identity in young neurons. A new paper proposes a molecular mechanism for this process, but the data can also be explained by making a distinction between axon specification and axon growth.

Hui Jiang is at the Institute of Neuroscience, Shanghai Institutes of Biological Sciences, and the Graduate School, Chinese Academy of Sciences, 320 Yueyang Road, Shanghai 200031, China, and Yi Rao is in the Department of Neurology, Northwestern University Feinberg School of Medicine, 303 East Chicago Avenue, Ward 10-185, Chicago, Illinois 60611, USA.e-mail: y-rao@northwestern.edu

thought about how this happens: Da Silva and colleagues report that a membrane enzyme, plasma membrane ganglioside sialidase (PMGS) and other associated molecules are required for axon formation but do not affect dendrite formation. Here we discuss this work and raise the question of whether molecular mechanisms for axon specification and axon growth must be related.

A major model system for studying axon-dendrite specification, cultured hippocam-pal pyramidal neurons, was pioneered by

Gary Banker more than 20 years ago2. The neurites are similar to each other in the first 24 hours after culturing. These neurons undergo a critical transition between stages 2 and 3 and are considered to be polarized when the axon is distinct from the dendrites both in length and in expression of unique molecular markers. Because axons are often longer than dendrites, it is sometimes diffi-cult to distinguish between a role for a mol-ecule in axon specification and a role in axon growth. This poses a special problem when

Early in neural development, one neurite becomes differentiated from the others to form the axon. A paper in this issue1 from Carlos Dottis group provides much food for

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 5 | MAY 2005 545

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

the length of neurites is one of the major cri-teria to define an axon.

Several molecules are involved in establish-ing and maintaining neuronal polarity; when their functions are disturbed, multiple axons can form39, or the neurites appear to be nei-ther axons nor dendrites10. We have shown that reducing the activity of either of two such mol-ecules, GSK3 and PTEN, can switch the fate of a neurite from dendrite to axon8, as evidenced by expression of molecular markers and, in the case of GSK3 manipulation, the functional-ity of synaptic vesicle recycling. Such a switch of axonal and dendritic fates provides a clear definition of roles in axon-dendrite specifi-cation (Fig. 1a). Although fate switches have not been analyzed for other molecules such as Par3 and 6 and the small GTPases Rap1B and Cdc42, photographs in published papers39 are consistent with the possibility that when their activities are changed, axons form at the expense of dendrites, suggesting that they are also involved in axon-dendrite specification.

Dottis group has been searching for early events that break the symmetry of unpolarized neurons. They had previously found that actin instability is higher in one neurite growth cone of the unpolarized neuron11, that an axon can be selectively induced by locally destabilizing actin filaments in one growth cone and that multiple axons can be induced by global application of actin destabilization drugs11. Thus, actin desta-bilization is involved in axon specification.

They began the current study1 with the inter-esting observation that a membrane enzyme, PMGS, is enriched in one neurite of a morpho-logically unpolarized stage 2 neuron. This is the prospective axon, both because it has a low level of filamentous actin (F-actin), a marker for early (and mature) axons, and because PMGS is found in the axon of mature polarized neurons. PMGS overexpression reduced F-actin content, whereas PMGS inhibition by RNA interference (RNAi) or a chemical inhibitor (NeuAc2en) increased F-actin content. The authors conclude that PMGS can regulate actin stability.

When assaying the developmental function of molecules, Da Silva and colleagues used two criteria to identify axons: length (an axon being longer than 40 mm and 3 times the length of other neurites) and Tau-1 immunostaining. By these criteria, no axon formed after treatment with PMGS RNAi or the chemical inhibitor. By contrast, PMGS overexpression accelerated axon formation so that rather than forming a long axon after 48 hours in culture, as control neurons do, a neuron overexpressing PMGS formed a long axon in 24 hours. However, PMGS overexpression did not cause the for-mation of multiple axons1, which is different from the effect of actin destabilization11.

The product of PMGS, GM1, is a com-ponent of lipid microdomainsthe lipid raftsuggesting that local lipid environment can modulate the development of the actin net-work and axons. Thus, it is natural to think of

receptors localized in the lipid raft, especially those interacting with GM1 (ref. 12), such as the neurotrophin receptors. By the criteria described above, Da Silva and colleagues1 found that axon formation was accelerated by nerve growth factor (NGF), but not by two other neurotrophins (brain-derived neurotrophic factor and neurotrophin 3), and that this effect depended on PMGS activity. NGF enhanced the effect of PMGS overexpression on axon forma-tion. Biochemically, both PMGS and GM1 bind to phospho-TrkA (pTrkA), the activated form of the NGF receptor. NGF stimulation of TrkA phosphorylation was potentiated by PMGS overexpression. Subcellularly, pTrkA is localized with PMGS: it is present in one neurite of a stage 2 neuron and in the axon of a stage 3 neuron. Phosphorylated TrkA seems to be downstream of PMGS because it is mislocalized after PMGS inhibition, although the possibility of mutual dependence has not been ruled out.

Intracellular components downstream of PMGS include PI3K, the small GTPases Rac and RhoA and the complex between the RhoA kinase (ROCK) and Profilin IIa1. PMGS over-expression leads to RhoA dissociation from the membrane (and presumably its inactivation), and this depends on PI3K and Rac. PMGS over-expression causes the dissociation of ROCK from Profilin IIa and decreases Profilin IIa phos-phorylation, which favors actin polymeriza-tion13. From these data, Da Silva and colleagues propose an interesting model of axon specifica-

Figure 1 Models for molecular mechanisms of axon/dendrite specification and axon growth. (a) Molecules involved in axon/dendrite specification can be localized in one type of neurite, or their activities can be preferentially localized. There is no direct evidence for any extracellular cues or transmembrane receptors involved in making the axon/dendrite choice. The links between intracellular components are based on data from several papers. The Cdc42/Par3/Par6/aPKC complex is enriched in the axon, and PI3K, Akt, GSK-3 and CRMP2 are preferentially activated in the axon. Functional evidence suggests that PI3K activates Cdc42 through Rap1B. How the Par3 complex and Akt/GSK-3 pathways interact is not clear. (b) Regulation of axon growth by the neurotrophin/Trk pathways. PMGS locally recruits activated TrkA receptors in the growth cone of the axon, favoring axon outgrowth through PI3K, Rac activation and RhoA inhibition1. TrkB and TrkC activate PI3K more potently than TrkA9. They promote axon growth and branching through the PI3K/Akt/GSK-3/CRMP2 pathway9. Through interaction with Numb, CRMP2 also regulates L1 endocytosis and recycling in the growth cone16, which is critical for axon growth.

PI3K

PPPP

AktPP

GSK-3Par-3

Par-6aPKCGTP

Par-3Par-3Par-6Par-6

aPKCaPKCCdc42

GTP

STEFRacGTPGTPGTP

GTPGTPRap-1B

GSK-3

CRMP-2CRMP-2 CRMP-2CRMP-2 PP

Microtubule

?

RhoAGTPGTPGTP

RhoA

Actin

GDP

?

b Axon formation and growth a Choice of axon/dendrite fates TrkB,TrkC

PPPPPP

PPPPPPI3K

PPPP

AktPP

GSK-3b

GSK-3b

CRMP-2CRMP-2CRMP-2CRMP-2 P

Microtubule

PPPPPP

PPPPPP

TrkA

RacGTPGTPGTP

Rac

PI3K

GDP

RhoAGTPGTP GTP

RhoAGDP

PMGS

RhoAGTPGTPGTP

ROCK PIIaPP

Actin

GTP

CRMP-2CRMP-2

P

L1

Numb

L1 endocytosis

PTEN

P

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

546 VOLUME 8 | NUMBER 5 | MAY 2005 NATURE NEUROSCIENCE

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

tion in which spatially segregated TrkA signaling caused by PMGS and its product, GM1, leads to local inactivation of RhoA, increasing actin instability in one neurite of an unpolarized neu-ron. In this model, PMGS acts in the prospec-tive axon, but not in the dendrites, to specify one neurite to become the axon10. An alterna-tive model of axon growth is that Trk signaling and PMGS do not specify axonal fate but rather are required for axon formation and growth (Fig. 1b). Although the available data are not sufficient to prove either model definitely, there are several pieces of evidence that make the alternative model worth considering.

The axon specification model is favored by the findings that (i) TrkA and PMGS are local-ized very early to one neurite in a stage 2 unpo-larized neuron, (ii) PMGS inhibition causes the failure of axon formation and (iii) Trk activation and PMGS overexpression lead to precocious axon formation. The strongest evi-dence inconsistent with the axon specification model, but consistent with the axon growth model, is the failure of Trk activation and PMGS overexpression to cause the formation of multiple axons or to re-specify dendrites into axons. This is in contrast to treatments such as PTEN inactivation, Akt activation, GSK inactivation, Rap1B activation, Cdc42 activation, APC overexpression, CRMP2 over-expression or disinhibition or actin destabi-lization, all of which cause the formation of multiple axons39. Although Trk activation1,9 or PMGS overexpression1 can cause the for-mation of one axon per neuron precociously and can make axons grow longer, neither Trk activation nor PMGS overexpression change the choice of axonal versus dendritic fates. It is possible the PMGS overexpression did not increase its activity sufficiently so that either an enhanced PMGS level or an activating PMGS mutant could potentially convert den-drites into axons. However, until this is proven,

it remains possible that PMGS (and TrkA) may function after the specification of axonal or dendrite fate to promote axon growth.

The alternative axon growth model has not been established by existing data and also requires further experiments. If TrkA and PMGS function after the early specification step, then molecules required for selecting axon fate should be localized in the prospective axon before TrkA and PMGS localization. Rap1B and Cdc42, two molecules involved in axon-dendrite specifica-tion, are preferentially localized to the axon tip in stage 2 unpolarized neurons6. Activation of these molecules causes the formation of multiple axons6, allowing them to be functionally classi-fied as axon specification molecules (Fig. 1a). The localization of other axon specification mol-ecules such as Par3 and 6, GSK3, APC or CRMP2 have not been examined at as many time points as Rap1B, Cdc42 or TrkA and PMGS. Thus, it remains to be determined which molecule is the first to be localized to the axon. Improvement in the ability to correlate molecular localiza-tion and neurite fates with time-lapse micros-copy and the ability to follow the localization of molecular activity (rather than just the molecules themselves) will help resolve this issue.

If PMGS is involved only in axon growth, why would inhibition of PMGS activity prevent development of the axon? This can be explained by proposing that PMGS inhibition causes defective axon growth, such that the axon fails to grow long enough to express axonal markers. Because all dendritic markers are also found in the proximal part of the axon, whereas axonal markers are localized in the distal part of the axon, it might not be possible to see differential marker localization without a long neurite.

It is also possible, although not proven, that there is a link between axon growth and axon specification. In the hippocampal model, the axon is the longest process, and it develops from the fastest-growing neurite. Mechanical

manipulation that eliminates the length dif-ferences among developing neurites can reset axon-dendrite specification14, whereas mechan-ical pulling that makes a neurite grow faster can specify an axon15. Molecules that re-specify den-drites into axons do make hippocampal neurites grow faster4. Is this an epiphenomenon, or is it fundamental to axon-dendrite specification in all neurons, including those without significant length differences between axons and dendrites? An essential link between specification and growth would favor the model of Da Silva et al. If growth advantage is an epiphenomenon, one can explain the results of Da Silva and colleagues as the promotion of axon growth after PMGS overexpression; growth would be required for axon formation or for full expression of axon-specific characteristics but not for the first step in breaking the symmetry of all presum-ably equivalent neurites. Finally, although it is conceptually important to distinguish between specification and growth, it is possible that some molecules are involved in both processes.

1. Da Silva, J.S. et al., Nat. Neurosci. 8, 606615 (2005).

2. Craig, A.M. & Banker, G. Annu. Rev. Neurosci. 17, 267310 (1994).

3. Inagaki, N. et al. Nat. Neurosci. 4, 781782 (2001).4. Shi, S.H., Jan, L.Y. & Jan, Y.N. Cell 112, 6375

(2003).5. Nishimura, T. et al. Nat. Cell Biol. 6, 328334

(2004).6. Schwamborn, J.C. & Puschel, A.W. Nat. Neurosci. 7,

923929 (2004).7. Shi, S.H. et al. Curr. Biol. 14, 20252032 (2004).8. Jiang, H., Guo, W., Liang, X. & Rao, Y. Cell 120, 123

135 (2005).9. Yoshimura, T. et al. Cell 120, 137149 (2005).10. Kishi, M., Pan, Y.A., Crump, J.G. & Sanes, J.R. Science

307, 929932 (2005).11. Bradke, F. & Dotti, C.G. Science 283, 19311934

(1999).12. Simons, K. & Toomre, D. Nat. Rev. Mol. Cell. Biol. 1,

3139 (2000).13. Da Silva, J.S. et al. J. Cell Biol. 162, 12671279

(2003).14. Dotti, C.G. & Banker, G.A. Nature 330, 254256

(1987).15. Lamoureux, P. et al. J. Cell Biol. 159, 499508

(2002).

Song learning and sleepDaniel Margoliash

Song learning in juvenile birds is guided by daytime sensorimotor feedback, but nighttime sleep is also integral to song learning, reports a study in Nature, setting the stage for physiological insights into sleep-dependent learning mechanisms.

Daniel Margoliash is in the Committee on

Computational Neuroscience and the Department

of Organismal Biology and Anatomy, The University

of Chicago, 1027 East 57th Street,

Chicago, Illinois 60637, USA.

email: dan@bigbird.uchicago.edu

In humans, evidence strongly suggests that sleep is critical for multiple forms of learn-ing, although this conclusion is not univer-sally accepted. Birdsong learning is one of the few animal models that has been developed to study sleep effects on procedural (skill)

learning, but the original proposal was based on physiological data and was limited to observations of adult birds1. A recent paper by Dergnaucourt and colleagues in Nature2 has now elegantly demonstrated that sleep is critical for developmental song learning in

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 5 | MAY 2005 547

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

juvenile birds. This opens the door for com-bining physiological and behavioral studies in this system, which, up until now, has been exploited only in studies of hippocampus-dependent spatial memory in rats.

After memorizing tutor song(s) early in development, songbirds experience an extended period of gradual vocal development that requires trial-and-error practice in order for the bird to acquire adult singing patterns. Auditory feedback is essential for this development3. The standard model for birdsong posits that a sen-sory memory of tutor songs acquired early in life is the reference (template) for error assessment. How the template and error are represented in the brain and how and when the error signal is applied to modify vocal output are some of the central issues in birdsong learning.

Earlier neurophysiological observations had suggested the possible role of sleep in adult song maintenance and, by extension, in juvenile song learning. However, this had never been directly observed, and the structure, if any, of sleep-related patterns of vocal development was unknown. Dergnaucourt and colleagues2 took advantage of signal processing techniques designed to trace the developmental trajectory of individual syllables (components of song) and assess the similarity of those syllables to the corresponding tutor syllables. When they applied this approach to the entire set of vocal-izations acquired during development (circa 1 million syllables per bird), they observed a circadian pattern: there was greater variability comparing syllables before and after sleep than comparing morning and afternoon singing.

By manipulating the period of intense singing that normally occurs in the morning, either by housing birds in the laboratory for several hours in the morning or by artificially induc-ing sleep, Dergnaucourt and colleagues2 dem-onstrated that the changes in syllable patterns were specifically related to sleep and could be dissociated from a purely circadian effect.

Importantly, the birds that ultimately most faithfully copied the tutor songs were those that showed the largest fluctuations overnight. Further analysis of individual variation, which is crucial for demonstrating the specificity of the sleep effect on vocal development, might have given greater insight into how sleep affects vocal learning. The authors also demonstrated that juveniles that were raised isolated from tutor songs showed little circadian variation in song, which also helps to associate a specific mecha-nism for vocal learning with the circadian fluc-tuation of vocal patterns during development. An interaction with sleep was also observed for song deterioration in adult birds following deaf-ening, providing additional evidence that adult song maintenance and juvenile song learning rely on a common set of mechanisms4.

Thus, the immediate effect of sleep during song development is deterioration of song production, which is subsequently reversed by daytime singing. These results can be consid-ered within the context of the neurobiological mechanisms of song learning. A posterior path-way in the forebrain HVC to the robust nucleus of the arcopallium (RA) influences moment-to-moment motor control (Fig. 1). An anterior forebrain pathway with basal ganglia compo-

nents also influences the RA through its output nucleus, the lateral magnocellular nucleus of the anterior nidopallium (lMAN). Lesions of lMAN result in abnormal termination of song development in young birds5 and prevent song deterioration in adult birds6. In both cases, the system fails to express variability in singing when lMAN is removed, suggesting that the anterior forebrain pathway is crucial to song learning.

How does sleep influence this system? RA neurons, and the HVC neurons that project (exclusively) to RA, have very phasic and pre-cise activity during singing. In RA, a neuron may show 20 different burst patterns during singing, each associated with a different part of a song syllable (a component of song) with sufficient precision and reliability that each burst has a unique identity7. During singing, in HVC, each RA-projecting neuron is uniquely associated with a specific small window of time somewhere in the song, also conferring unique identity8. The activity patterns of single RA neurons recorded when the animal sings are also observed when the animal sleeps. The precision of the activity patterns under the two conditions produces visually compelling and statistically highly significant matches1. The activity of RA neurons during sleep is driven by the RA-projecting HVC neurons8, so these too presumably show the same activity dur-ing sleep and singing. The pattern of playback during sleep does not occur randomly across neurons, as observed directly with simultane-ous multiple recordings. In general, there is substantial bursting activity and tight tem-poral correlation throughout the song system that could support singing-like activation of the entire set of pathways during sleep.

This sets the stage for singing-like activity during sleep to modify motor output. In a preliminary report, small but reliable changes in the structure of RA burst patterns were observed following sleep in adult zebra finches (P.L. Rauske, A.S. Dave & D.M., Soc. Neurosci. Abstr. 381.3, 2001). Such changes are what one would predict if the highly stereotyped pat-terns of adult song were being dynamically maintained by minute adjustments of the motor output. Such small changes in behavior would be difficult to observe in the adult2, but one could presumably observe much larger changes in RA patterns over periods of sleep in juvenile birds. The challenge will be interpret-ing neuronal changes against the background of far more variable plastic singing.

In sum, these data identify the basic pro-cesses that are likely to be recruited in service of a sleep-mediated component of vocal learn-ing. They set the stage for examining what information is stored during daytime singing, how it is stored, and how it is modified during

Figure 1 Connection diagram of the song system. The motor pathway includes the HVC, which projects to the robust nucleus of the archistriatum (RA). RA projects to the tracheosyringeal portion of the hypoglossal nucleus (nXlts), which controls the syrinx or vocal organ, and to the dorsal medial subdivision of the nucleus intercollicularis nuclei (DM) that control respiratory muscles. HVC also projects to the nucleus area X, part of the anterior forebrain pathway. The anterior forebrain pathway forms a loop through the medial nucleus of the dorsolateral thalamus (DLM) and the lateral magnocellular nucleus of the anterior nidopallium (lMAN) before joining the motor pathway at the RA. Auditory inputs enter the system at the level of the HVC and nucleus interfacialis (NIf). Inset, zebra finch.

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

548 VOLUME 8 | NUMBER 5 | MAY 2005 NATURE NEUROSCIENCE

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

sleep. One possibility is that random modi-fications of RA burst patterns over periods of sleep induce an error in morning singing that is subsequently adjusted during the day. The daytime adjustment brings the network into a slightly more stable state that then con-strains the range of random modifications that are induced by the following period of sleep. From this perspective, the magnitude of the error is the critical feature of the sleep-induced change. Alternatively, the structure of the sleep-induced changes, i.e., which bursts are changed and how they are changed, carries information. In that case, there is information in the state space explored by the song system in bursting activity during sleep.

In humans, dreams with vivid imagery are associated with rapid eye movement (REM) sleep. Neural plasticity during REM sleep is thought to help consolidate memory. There are direct data to support the REM plasticity hypothesis9, but the larger body of evidence tends to be correlative, and the hypothesis remains controversial10. In any case, the dif-ferent roles of sleep stages in consolidation or other aspects of procedural learning are not well established.

REM sleep has been described in birds but is reported as occurring in extremely short inter-vals (circa 4 s) that are relatively rare. Although numerous species have been examined, virtu-

ally none of them were true (oscine) song-birds, but when oscine birds were examined, a pattern with more REM was observed11. Recent work in zebra finches confirms and extends this work. In preliminary results, sleep staging in zebra finches was reported to include slow wave sleep (SWS), intermediate stages and far more and longer periods of REM than observed in non-passerine birds (P.S. Low, S.S. Shank & D.M., Soc. Neurosci. Abstr. 769.5, 2003). There are systematic changes in the relative frequency of non-REM and REM sleep throughout the night, and there is structure to the patterns of transitions between stages. This opens the pos-sibility for examining the roles of sleep stages in song learning. The nuclear pattern of anatomi-cal organization of the song system also should facilitate the design of sleep-stage perturbation experiments that spare whole-animal sleep. This can address issues of stress and other non-specific effects of sleep deprivation.

The results of Dergnaucourt et al.2 focus on sensorimotor learning but do not exclude the possible roles of sleep, presumably in the form of memory consolidation, on sensory learning. Auditory activity in the song system is highly state dependent12. A recent playback experi-ment with juvenile zebra finches reported that auditory responses were stronger for the tutor song during the day and were stronger for the developing birds variable plastic song at

night13. This could be part of the mechanism that assesses error on the basis of auditory feed-back during the day and modifies the network on the basis of song-like activity at night.

It is uncommon, particularly from a neuroethological standpoint, for physiological obser-vations to suggest behavioral phenomena that have yet to be observed. Dergnaucourt and colleagues2 bring a welcome behavioral focus to the analysis of how sleep affects the develop-mental learning of bird song. Coupled with the strong physiological insight that the field enjoys, it seems likely that a mechanistic explanation for the actions of sleep on several forms of learning is likely to emerge.

1. Dave, A.S. & Margoliash, D. Science 290, 812816 (2000).

2. Dergnaucourt, S. et al. Nature 433, 710716 (2005).

3. Konishi, M. Z. Tierpsychol. 22, 77083 (1965).4. Nordeen, K.W. & Nordeen, E.J. Behav. Neural Biol. 57,

5866 (1992).5. Bottjer, S.W., Miesner, E.A. & Arnold, A.P. Science

224, 901903 (1984).6. Brainard, M.S. & Doupe, A.J. Nature 404, 762766

(2000).7. Yu, A.C. & Margoliash, D. Science 273, 18711875

(1996).8. Hahnloser, R.H. et al. Nature 419, 6570 (2002).9. Stickgold, R. et al. Science 294, 10521057 (2001).10. Siegel, J.M. Science 294, 10581063 (2001).11. Rattenborg, N.C. et al. PLoS Biol. 2, E212 (2004).12. Dave, A.S., Yu, A.C. & Margoliash, D. Science 282,

22502254 (1998).13. Nick, T.A. & Konishi, M. J. Neurobiol. 62, 231242

(2005).

Trafficking in emotionsDan Ehninger, Anna Matynia & Alcino J Silva

Postsynaptic receptor trafficking is associated with long-term synaptic plasticity, but whether this mechanism actually mediates learning is unclear. A new study shows that fear learning drives AMPA receptors into synapses in the lateral amygdala.

Dan Ehninger, Anna Matynia and Alcino J. Silva are

in the Departments of Neurobiology, Psychiatry and Psychology, Brain Research Institute, University

of California Los Angeles, Los Angeles, California

90095, USA.

e-mail: Silvaa@mednet.ucla.edu

Changes in synaptic strength are critical for learning and memory, but much remains to be understood about the molecular and cel-lular events that accompany learning and are required for memory. In a recent paper, Rumpel et al.1 used powerful molecular tools to demonstrate that fear conditioning in rats triggers the insertion of AMPA receptors into synapses in the lateral amygdala, caus-ing increases in synaptic strength required for emotional memory. These findings have

far-reaching implications for the role of the amygdala in fear learning and, more generally, for the role of postsynaptic receptor traffick-ing in memory.

The ability to remember pleasant and aver-sive events (emotional memory) is critical for survival, and, not surprisingly, many of the underlying mechanisms seem to be evolution-arily conserved. One of the most studied forms of emotional memory, auditory fear condition-ing, depends on a subjects ability to associate a tone with an aversive stimulus, such as an elec-tric foot shock (Fig. 1a). When re-exposed to the tone, trained subjects show fear responses, including freezing, the cessation of all but respiratory movements. Freezing is a reliable measure of fear conditioning, and the amyg-dala is important in this type of memory.

Learning is associated with changes in syn-aptic strength required for memory, as many molecular and cellular studies demonstrate2. Synapses modify their strength by changing the number of postsynaptic AMPA-type glutamate receptors3,4. For example, AMPA receptors with GluR1 subunits are driven into the syn-apse in response to plasticity-inducing stimuli (Fig. 1b), including electrical stimulation that increases synaptic strength in a hippocampal slice preparation5 and sensory input from whis-kers in rodents, which induces plasticity in the developing somatosensory cortex6. Before the study by Rumpel et al., however, it was unclear whether AMPA receptor trafficking was associ-ated with learning and memory1.

The authors applied elegant experimental tools developed in previous studies5,6 to test

20

05 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

ren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 5 | MAY 2005 549

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

the hypothesis that AMPA receptor traffick-ing underlies memory associated with fear conditioning. One of the tools was a recom-binant GluR1 subunit fused with a green fluorescent protein (GFP) gene packaged in a herpes simplex virus (HSV) vector. The GFP marker allowed the authors to identify and record from cells that had been infected with their unique construct, and the distinct biophysical properties of homomeric GluR1 AMPA receptors (greater conductance when passing inward current than when passing outward current) provided an ingenious way to determine whether these receptors had been inserted into synapses. By stimulating audi-tory fibers from the thalamus and recording from labeled (infected) neurons in the lateral amygdala in brain slices, Rumpel et al. showed that recombinant GluR1-containing receptors were clearly detectable in thalamo-amygdala synapses after auditory fear conditioning. This result is consistent with the idea that learning is associated with AMPA receptor trafficking that causes changes in synaptic strength required for memory. Importantly, recombinant GluR1 receptors were not detected in synapses of rats that had not been conditioned, confirming the behavioral specificity of the plasticity and jus-tifying their label of the vector as a plasticity

tag. Clearly, this is not the last time we will hear about these plasticity tags, as there are many obvious uses for a tool that can mark cells and synapses activated by learning!

The second molecular tool used by Rumpel et al. consisted of a C-terminal fragment of GluR1 fused with a GFP gene in an HSV vec-tor. This C-terminal fragment blocks synaptic plasticity by interfering with the synaptic inser-tion of GluR1 receptors7. Importantly, electro-physiological experiments showed that this plasticity block vector did not affect basal AMPA receptormediated transmission or dis-rupt freezing immediately after training. Thus, the plasticity block vector did not affect gen-e